Double Dissociation of Function Within the ...

Behavioral Neuroscience 1999, Vol. 113, No. 6,1170-1188

Copyright 1999 by the American Psychological Association, Inc. 0735-7044/99/S3.00

Double Dissociation of Function Within the Hippocampus: A Comparison of Dorsal, Ventral, and Complete Hippocampal Cytotoxic Lesions B. K. Yee

D. M. Bannerman University of Oxford

University of Hong Kong

M. A. Good

M. J. Heupel, S. D. Iversen, and J. N. P. Rawlins

Cardiff University of Wales

University of Oxford

Rats with complete cytotoxic hippocampal lesions exhibited spatial memory impairments in both the water maze and elevated T maze. They were hyperactive in photocell cages; swam faster in the water maze; and were less efficient on a nonspatial, differential reinforcement of low rates (DRL) task. Performance on both spatial tasks was also impaired by selective dorsal but not ventral lesions; swim speed was increased by ventral but not dorsal lesions. Both partial lesions caused a comparable reduction in DRL efficiency, although these effects were smaller than those of complete lesions. Neither partial lesion induced hyperactivity when rats were tested in photocell cages, although both complete and ventral lesion groups showed increased activity after footshock in other studies (Richmond et al., 1999). These results demonstrate possible functional dissociations along the septotemporal axis of the hippocampus.

There have been numerous attempts to formulate a concise, unitary theory of hippocampal function that can account for the vast amounts of relevant experimental data. The most significant division between the various accounts of hippocampal function is probably that between spatial (O'Keefe & Nadel, 1978) and nonspatial theories (e.g., Eichenbaum, Otto, & Cohen, 1992; Olton, Becker, & Handelmann, 1979; Rawlins, 1985; Rudy & Sutherland, 1995). Each theory attempts to account for essentially the same data set, though each emphasizes particular aspects of the data rather than others, and each identifies quite different psychological deficits as underlying the same lesion-induced behavioral deficits. For example, exclusively spatial theorists attempt to show how performance on overtly nonspatial tasks may depend on some covert spatial bias. Conversely, proponents of nonspatial accounts attempt to show how D. M. Bannerman, M. J. Heupel, S. D. Iversen, and J. N. P. Rawlins, Department of Experimental Psychology, University of Oxford, Oxford, England; B. K. Yee, Neuroscience Research Centre, Department of Anatomy, University of Hong Kong, Hong Kong, People's Republic of China; M. A. Good, School of Psychology, Cardiff University of Wales, Cardiff, Wales. This work was supported by a Wellcome Programme Grant (UK 039129/Z/93), a Wellcome Projects Grant (UK 054143/Z/98), a grant from the Biotechnology and Biological Sciences Research Council (UK S05720), and by the Schizophrenia Research Fund. We thank Greg Daubney for assistance with the histology and Robert Hiorns for statistical advice. Correspondence concerning this article should be addressed to D. M. Bannerman or J. N. P. Rawlins, Department of Experimental Psychology, University of Oxford, South Parks Road, Oxford OX1 3UD England. Electronic mail may be sent to david.bannerman@ psy.ox.ac.uk or [email protected].

1170

performance on overtly spatial tasks depends on some covert, intrinsically nonspatial memory function. What all these theories have in common, however, is that they attempt to explain all the available data in terms of a single concept. The behavioral syndrome associated with complete lesions of the hippocampus encompasses a wide range of apparently diverse behavioral effects, although it is worth noting that both the number and diversity of these effects have been reduced as a result of the development and widespread use of more selective, fiber-sparing excitotoxic lesions (see Jarrard, 1993). Rats with excitotoxic hippocampal lesions are hyperactive and hyperresponsive, and they exhibit profound and reliable deficits on both spatial and nonspatial memory tasks. For example, complete cytotoxic hippocampal lesions impair both spatial working and spatial reference memory on a variety of maze tasks using both appetitive and aversive forms of motivation (Jarrard, 1983; Morris, Schenk, Tweedie, & Jarrard, 1990). Similarly, complete hippocampal lesions disrupt performance on both appetitive and aversive contextual learning paradigms (Honey & Good, 1993; Kim & Fanselow, 1992; but see also Richmond et al., 1999). Both spatial and contextual theories of hippocampal function bear an intuitive similarity in that they require the hippocampus to form and utilize a complex representation of the environment, encompassing a large and diffuse number of cues from all the sensory modalities (see Hirsh, 1974; Honey & Good, 1993; Kim & Fanselow, 1992; O'Keefe & Nadel, 1978). It is not immediately obvious, however, how a spatial or contextual theory of hippocampal function can account for the behavioral effects of hippocampal lesions on overtly nonspatial tasks. For example, complete cytotoxic hippocampal lesions have been shown to disrupt efficient performance

DOUBLE DISSOCIATION WITHIN THE HIPPOCAMPUS

on a differential reinforcement of low rates of responding (DRL) task (Sinden, Rawlins, Gray, & Jarrard, 1986): Efficient performance requires the rat to suppress responding (lever pressing) until some mimimum time has elapsed. Complete hippocampal lesions produce a robust DRL impairment despite the lack of an obvious spatial component to this task. Furthermore, Bunsey and Eichenbaum (1996) have shown that, in contrast to sham-operated controls, rats with hippocampal lesions fail to demonstrate the properties of associativity and transitivity after acquisition of a nonspatial, olfactory stimulus-stimulus association. More recently, Honey, Watt, and Good (1998) have demonstrated that hippocampal lesions eliminate the normal response to an associative mismatch between pairs of discrete auditory and visual cues. These findings suggest that the hippocampus has a role in processing both nonspatial and spatial information. Recent studies by Moser, Moser, and Andersen (1993) and by Moser, Moser, Forrest, Andersen, and Morris (1995) shed a different light on this issue, suggesting that attempts to formulate a unitary hypothesis of hippocampal function may be inappropriate. Moser et al. (1995) examined the effects of bilateral, symmetrical hippocampal lesions on the acquisition of a standard spatial reference memory task in the Morris water maze. Lesions were systematically varied in size from 20% to 100% of total hippocampal volume, extending from either the septal or temporal pole of the hippocampus. By using the excitotoxin ibotenic acid, the authors were able to destroy cell bodies in a given area while leaving vasculature and fibers of passage essentially intact and functioning. Consequently, excitotoxic lesions of the dorsal hippocampus should spare the rostral connections of the ventral hippocampus. The study showed that although spatial learning remained essentially intact with as little as 26% of the dorsal hippocampus still present, lesions of the septal pole, sparing as much as 60% of the ventral hippocampus, resulted in a robust water-maze impairment. These results suggest that whereas the dorsal hippocampus appears to be essential for spatial learning, the ventral hippocampus does not appear to be essential. This is consistent with previous electrophysiological data indicating not only that there is a higher proportion of place cells in the dorsal hippocampus, but also that these place cells have more refined place fields than their counterparts in the ventral hippocampus (Jung, Wiener, & McNaughton, 1994). These findings raise the possibility that the hippocampus may in fact have more than one function, and that perhaps these functions are differentially distributed along the longitudinal, septotemporal axis of the hippocampus. This hypothesis is further supported by anatomical data that demonstrate a differentiation in terms of hippocampal connectivity along this axis. In brief, the major input to the hippocampus from the sensory cortices projects primarily to the dorsal two thirds of the hippocampus, via the association cortex and the entorhinal and perirhinal cortices (Room & Groenewegen, 1986; Ruth, Collier, & Routtenberg, 1982; Witter & Groenewegen, 1984; Witter, Van Housen, & Amaral, 1989). The fact that the dorsal hippocampus receives already highly processed information from all the sensory modalities is entirely consistent with a role in spatial learning. In contrast,

1171

the ventral hippocampus is more closely associated with subcortical areas such as the hypothalamus (Kohler et al., 1986; Kohler, Swanson, Haglund, & Wu, 1985). Despite the anatomical and electrophysiological support for a preferential role for the dorsal hippocampus in spatial learning, there are other possible explanations of the Moser et al. (1995) study. Indeed, one interpretation of this result is that dorsal hippocampal lesions are simply more disruptive than ventral lesions per se. Accordingly, dorsal lesions will produce larger behavioral effects whatever the experimental paradigm, irrespective of the nature of the behavior. It may be the case that ventral hippocampal lesions are similar to, but simply less effective than, dorsal lesions, as indeed one might suspect on the basis of the dorsal and ventral single-unit recording studies (Jung et al., 1993). The recent demonstration that dorsal but not ventral cytotoxic hippocampal lesions impaired spatial working memory on the T maze does not address this concern, even though both partial lesions independently affected a motivational discrimination task (Hock & Bunsey, 1998). The two behavioral tasks may simply have differed in their sensitivity to the extent of hippocampal damage. To be absolutely confident that the dorsal hippocampus has a preferential role in spatial learning it is necessary to demonstrate a double dissociation between the effects of dorsal and ventral lesions on two separate and independent measures of behavior. Of course, obtaining a double dissociation between the effects of dorsal and ventral lesions relies on the availability of a behavior that is specifically sensitive to cell loss in the ventral hippocampus. To this end, previous studies examining the effects of hippocampal lesions on locomotor activity may provide a clue. Rats with complete cytotoxic hippocampal lesions exhibit increases in both spontaneous (Whishaw & Jarrard, 1995) and amphetamine-induced locomotor activity (Wilkinson et al., 1993). This increase in drug-induced activity is associated with a concomitant increase in extracellular dopamine levels in the nucleus accumbens. It has also been shown that cytotoxic lesions of the ventral hippocampus increase both spontaneous and amphetamine-induced locomotor activity (Lipska, Jaskiw, Chrapusta, Karoum, & Weinberger, 1992; Sams-Dodd, Lipska, & Weinberger, 1997) and that the increases in activity after lesions of the ventral hippocampus in both neonatal and adult rats are associated with changes in dopamine and dopamine metabolite levels in the nucleus accumbens (Lipska, et al., 1992; Lipska, Jaskiw, & Weinberger, 1993). In contrast, selective cytotoxic lesions of the dorsal hippocampus exhibit neither an increase in amphetamine-induced locomotion nor any change in dopamine turnover (Lipska et al., 1991). Strikingly, these behavioral observations appear to be directly opposite to those obtained for spatial learning in the water maze, and may, therefore, provide an opportunity to reveal a double dissociation. Dorsal hippocampal lesions impair spatial learning in the water maze but have no effect on locomotor activity levels. In contrast, lesions of the ventral hippocampus increase both spontaneous and amphetamine-induced locomotor activity but leave spatial learning in the water maze intact. However, these two independent findings have yet to be demonstrated in the same experiment

1172

BANNERMAN ET AL.

with the same set of animals. A within-subject, double dissociation would provide a powerful argument in favor of multiple, independent, and dissociable functions existing within the hippocampus. The aim of this study was therefore to compare the performance of rats with similar-sized dorsal and ventral hippocampal lesions to rats with complete bilateral hippocampal lesions and to sham-operated controls. The dorsal and ventral lesions were each intended to ablate 55-60% of the total hippocampal volume starting from the septal or temporal pole, respectively. They were thus intended to overlap slightly. The derivation of the stereotaxic coordinates required is described in the Method section. The experiments compared the four groups of rats across a series of behavioral tests, all of which have previously been shown to be sensitive to complete hippocampal cell loss. The present study had four aims. The first was to repeat the study of Moser et al. (1995) and establish whether our dorsal lesions were sufficient to disrupt spatial reference memory in the water maze while our ventral lesions left performance intact. The second was to assess the lesions' effect on another spatial learning task with different sensorimotor and motivational demands, which required working memory rather than reference memory. We wished to determine whether the same pattern of similarities between dorsal hippocampal lesions and complete hippocampal lesions would be found under these different test conditions. The third was to assess the performance of the various lesion groups on a nonspatial, hippocampal-dependent task: We studied DRL performance to determine whether a task with no overt spatial memory component would reveal the same pattern of lesion-dependent dysfunction as the navigational tasks referred to above. The fourth was to determine the effects of the lesions on both spontaneous and drug-induced locomotor activity. Previous work by Lipska et al. (1991, 1992) predicted that the effects of partial lesions on locomotor activity levels should be directly opposite to those obtained for spatial learning. If so, then this would demonstrate that the effects of the different partial lesions could not be subsumed within an existing, unitary account of hippocampal function. The four groups of rats were therefore trained on a standard spatial reference memory task in the Morris water maze, to allow comparison with the previous study of Moser et al. (1995). Spatial working memory was investigated with a nonmatching to place (NMTP) task on the elevated T maze. Nonspatial, hippocampal-dependent performance was assessed in an operant DRL task. Finally, levels of both spontaneous and amphetamine-induced locomotor activity were determined in photocell activity cages, allowing comparison with the previous work of Lipska et al. (1991,1992). In a companion study (Richmond et al., 1999), the contribution of the dorsal and ventral hippocampus to contextual learning was examined. Using similar groups of lesioned rats, Richmond et al. assessed the effects of dorsal, ventral, and complete lesions on both conditioned contextual freezing and conditioned tone freezing.

Method Derivation of Stereotaxic Coordinates The issue of comparability with regard to dorsal and ventral lesion size was addressed as follows: A standard rat stereotaxic atlas (Paxinos & Watson, 1986) was scanned, and the area of the hippocampus in each successive section was measured (see Figure 1A; the fimbria, the fornix, and the alveus were excluded from the areas measured). By interpolating between each section, it was possible to build up a cumulative measure of hippocampal volume that could then be related to a specific set of stereotaxic coordinates (see Figure IB). The process was repeated for both coronal and horizontal sections. This permits the determination of the stereotaxic coordinates for the border between lesions that destroy only the dorsal 50% or only the ventral 50% of the hippocampus. It proved necessary to perform the calculation for both coronal and horizontal sections because the atlas's horizontal sections do not extend through the most superficial horizontal sections of the hippocampus. They do, however, extend far enough to allow the identification of the boundary depth at which 50% of the hippocampal volume lies above and 50% lies below. Multiple microinjections of Af-methyl-D-aspartic acid (NMDA) were then made at the relevant stereotaxic coordinates to produce equivalent-sized lesions of the dorsal and ventral hippocampus. The intended lesion sizes were approximately 55-60% of the total hippocampal volume, extending just beyond the boundary point and thus overlapping slightly. Pilot lesions indicated that this was indeed the case, and that multiple, discrete injections of NMDA at the stereotaxic coordinates outlined in Table 1 resulted in near complete dorsal or ventral hippocampal lesions while completely sparing the subiculum (see Figure 2).

Subjects Male Lister hooded rats (n = 39; Harlan OLAC Ltd, Oxon, UK) served as subjects. The rats were housed in pairs with ad-lib access to food and water unless otherwise specified. A 12-hr light-dark cycle was maintained (lights on from 0700 to 1900), with all behavioral testing conducted during the light phase. The rats were experimentally naive before the start of this experiment and weighed 315^-15 g at the time of surgery. The experiment comprised four groups of rats. The first group received bilateral complete cytotoxic lesions of the hippocampus proper (n = 9; Group C). A second group of rats were given bilateral lesions of the dorsal 60% of the hippocampus proper (n = 10; Group D). The third group of rats received bilateral lesions of the ventral 60% of the hippocampus proper (n = 10; Group V). A final group of rats received sham operations (n = 10; Group S).

Surgery All rats were anesthetised with tribromoethanol (Avertin, 0.29 g/kg ip) and placed in the stereotaxic frame (Kopf Instruments, Clark Electromedical, Reading, UK) with the head level between bregma and lambda. An incision of the scalp was made along the midline, and the appropriate portion of the bone overlying the neocortex was removed. The procedure used to lesion the hippocampus was similar to that used by Jarrard (1989) and identical to that described by Richmond et al. (1999). Briefly, rats received multiple injections of NMDA (10 mg/ml, Sigma Chemical, Poole, UK) dissolved in phosphate buffered saline (pH 7.4). Injections (0.0250.10 ul) were made over approximately 30-60 s with a 5-|al

1173


50,000-,

3

4

5

6

7

8

9

1

0

DV (mm below bregma at skull level)

B 100-1

80-

60-

40-

20-

4

5

6

7

8

9

10

DV (mm below bregma at skull level) Figure 1. Determination of stereotaxic coordinates for partial hippocampal lesions. A: The area (in pixels) of the hippocampus including the dentate gyrus, at successive horizontal sections in Paxinos and Watson (1986), from Plate 89 (DV = -9.1 mm from bregma) to Plate 113 (DV = -3.1 mm from bregma). B: The percentage of cumulative hippocampal volume at successive horizontal sections. microsyringe (Scientific Glass Engineering, Milton Keynes, UK) mounted on the stereotaxic frame (see Table 1). On completion of each infusion, the syringe needle was left in position for a further 60-120 s to allow for diffusion of the neurotoxin away from the injection site. For the sham-operated rats (Group S) a similar procedure was adopted, with the exception that no excitotoxin was injected and the needle track was limited to the overlying cortex. On completion of the surgery, all rats were sutured and a topical antibiotic powder (P.E.P. 2% powder, Intervet Laboratories, Cambridge, UK) was sprinkled over the wound. All rats also received a subcutaneous injection of antibiotic (Baytril 2.5%; Bayer Ltd,

Newbury, UK). After surgery, rats were allowed at least 2 weeks to recover before the start of behavioral testing.

Experiment 1: Acquisition of a DRL Schedule Apparatus. Behavioral testing on the DRL task took place in eight Campden Instruments CI 460 operant test chambers (Campden Instruments, Loughborough, UK) that were controlled by a NOVA 4 computer programmed in ACT-N (Millenson, 1971). The rats' reward was 3 s of access to 0.1 ml of 10% (wt/vol) sucrose solution that was delivered into the food magazine by a dipper.

BANNERMAN ET AL.

1174

Table 1 Stereotaxic Coordinates for Cytotoxic Lesions of the Hippocampus AP

ML

DV

Vol. (ul)

-2.4 -2.8 -3.2 -3.2 -3.2 -3.6 -4.4 -4.4 -4.4 -4.4 -5.5

Complete hippocampal lesion (Group C) -3.3 ±1.0 -3.3 ±1.8 ±1.4 -3.3 ±1.4 -2.6 -3.1 ±3.0 -3.1 ±3.5 -3.3 ±2.8 -2.3 ±2.8 -4.2 ±4.0 -3.3 ±4.0 -2.3 ±4.0 -5.2 ±4.8 -4.2 ±4.8 -7.3 ±4.0 -4.2 ±4.0 ±4.0 -3.5 -5.6 ±5.0 ±5.0 -4.9 Dorsal hippocampal lesion (Group D) -3.3 ±1.0 ±1.8 -3.3 ±1.4 -3.3 ±1.4 -2.6 -3.1 ±3.0 ±3.5 -3.1 -3.3 ±2.8 ±2.8 -2.3 -3.3 ±4.0 ±4.0 -2.3 ±4.1 -3.5

0.075 0.075 0.050 0.050 0.100 0.075 0.050 0.050 0.050 0.050 0.025

-4.4 -4.9 -4.9 -5.2 -5.2 -5.5 -5.5

Ventral hippocampal lesion (Group V) ±4.0 -4.0 -5.2 ±4.8 -4.2 ±4.8 -7.3 ±4.0 -4.2 ±4.0 ±5.0 -5.6 -4.9 ±5.0

0.025 0.075 0.050 0.100 0.075 0.100 0.075

-2.4 -2.8 -3.2 -3.2 -3.2 -3.6 -4.4 -4.4 -4.4 -4.4 -4.4 -4.9 -4.9 -5.2 -5.2 -5.2 -5.5 -5.5

0.075 0.075 0.050 0.050 0.100 0.075 0.050 0.050 0.025 0.050 0.050 0.075 0.050 0.100 0.075 0.050 0.100 0.075

Note. Anterior-posterior (AP) and midline (ML) coordinates are in millimeters from bregma; dorsoventral (DV) coordinates are in millimeters from brain surface. Procedure. The rats were weighed and handled daily after surgery. Food deprivation began approximately 2 weeks postsurgery. The rats were given a measured amount of food after testing each day, so that their weight was reduced to 85% of the initial free-feeding value, and thereafter increased by 1% per week. The rats were allocated to one of the eight operant chambers, counterbalanced with respect to lesion group. On the 4th day of food deprivation, the rats were habituated to the operant chambers. The rats were allowed to explore the chambers and to collect reinforcers that were delivered, on average, every 16s (the magazine tray door was held open on this day). For the next 8 days, the rats were trained to press the left lever. There were 5 days of training on a variable-interval, 16-s schedule, followed by 3 days fixed-ratio (FR) training (1 day of FR2, and then 2 days of FR4). The next day, DRL-18 training began. The houselight was on throughout the session. All responses made 18 s or more after the previous response or delivery of a reinforcer were rewarded with a 3-s

presentation of sucrose and illumination of the magazine tray. Any response made before 18s had elapsed was unrewarded and reset the DRL requirement. Each training session lasted for 30 min and ended with the offset of the houselight. All rats received one training session per day for 25 consecutive days.

Experiment 2: Spatial Reference Memory in the Morris Water Maze Apparatus. Spatial learning was assessed in an open field water maze (Morris, 1981,1984) consisting of a large circular tank (diameter 2.0 m, depth 0.6 m) containing water at a temperature of 25 ± 1 °C and a depth of 0.3 m. The rats' task was to escape from the water by locating a hidden escape platform (diameter 10 cm) submerged 1-1.5 cm below the surface. The water was made opaque by the addition of 4 pints of semiskimmed milk, which not only prevented the subjects from seeing the platform but also allowed efficient tracking of swim paths. The pool was located on an elevated platform 60 cm above the floor in the center of a brightly lit room (5 m X 8 m) containing various prominent cues (e.g., wall posters, electrical fittings on the wall, a large screen behind which the experimenter sat). The room was diffusely illuminated by four 500-W floodlights positioned in the 4 corners of the room and aimed at the ceiling. The swim paths taken by the rats were monitored by a video camera mounted in the ceiling. The resulting video signal was relayed to a video recorder, allowing both on- and off-line analysis, and from there to an image analyzer (HVS VP112, HVS Image, Hampton, UK). The x and y coordinates of each rat's position were sampled in real time at 10 Hz by an Acorn computer (using a program called "Watermaze") and stored on disk. This program provides measures of latency, path length, swim speed, et cetera during acquisition, and the distribution of time spent in defined regions of the pool during the transfer test (e.g., quadrant, within a set distance from the side wall, etc.). Procedure. All behavioral testing was conducted by an experimenter unaware of the rats' group assignments. All rats were trained over 6 days to find an escape platform hidden at a fixed location. There were 4 training trials per day. The platform was located in the center of either the northwest or southeast quadrant of the pool, and the number of rats trained to each platform location was counterbalanced within each group. The rats were placed into the pool facing the side wall at one of eight start locations (arbitrarily designated north, south, east, west, northeast, northwest, southeast, and southwest; chosen randomly across trials) and allowed to swim until they found the platform or for a maximum of 120 s. Any rat that failed to find the platform within 120 s was guided to its location by the experimenter. The rats were then allowed to remain on the platform for 30 s before commencing the next trial. On completion of behavioral testing, the rats were dried and then returned to their home cages. On the 7th day of water-maze testing, a transfer test (or probe trial) was conducted to determine the extent to which the rats had learned about the location of the platform. The platform was removed from the pool, and the rat was allowed to swim freely for 60 s. The percentage of time that subjects spent in each quadrant of the maze was recorded. Other details of the training and testing procedure were identical to those described by Morris et al. (1990). Behavioral testing was conducted 2-3 months postsurgery.

Experiment 3: Spatial Learning on the Elevated T Maze Rats were once again put on a restricted feeding schedule and maintained at 85% of their free-feeding weight during testing on the elevated T maze. All behavioral testing was conducted by an


1175

Figure 2. A photomicrograph of a coronal section (corresponding approximately to Plate 40 from Paxinos & Watson, 1986) showing cell loss 7 days after a unilateral dorsal hippocampal lesion on the left-hand side and a unilateral ventral hippocampal lesion on the right-hand side (Pilot Rat 3).

experimenter unaware of rats' group assignments. Behavioral testing began 7-8 months postsurgery. Apparatus. Spatial learning was assessed with an elevated wooden T maze, which consisted of a start arm (80 cm long, 10 cm wide) and two identical goal arms (60 cm long, 10 cm wide) surrounded by a 1-cm-high raised edge. A metal food well (diameter 3 cm) was located at a point 3 cm from the end of each goal arm. The maze was located 1 m above the floor in a well-lit laboratory that contained prominent distal extramaze cues (animal caging along one wall, wall posters on 2 out of 4 walls, laboratory equipment in one corner). Preliminary training. During food deprivation, the rats were weighed and extensively handled by the experimenter on a daily basis. When the subjects had reached 85% of their free-feeding weight, preliminary training for the T maze study commenced. The training took place over a 5-day period. For the first 3 days, the rats were placed on the T maze in cage-mate pairs. They were placed at the bottom of the long arm and left to explore freely and collect food rewards for 10 min. During Days 4 and 5 of pretraining, the subjects were placed individually at the beginning of the T maze and allowed to explore freely for 5 min. By this stage, all the rats were eating from the food wells at the ends of the arms. Spontaneous rewarded alternation. Before commencing testing on discrete trial working memory (see below), spontaneous alternation was assessed. For the next 6 days, two reward pellets were placed in the food wells at the end of each arm. Each rat was placed at the beginning of the T maze and allowed the choice of either arm. However, as soon as the rat had eaten from one arm, it was removed and the choice of right or left arm was recorded. There were six trials completed by each subject per day, with an intertrial interval (ITI) of approximately 10 min. From the sequence of left and right choices made by the rat, the number of alternations (a maximum of five per session) was calculated. Spatial NMTP task. On Day 12 of T-maze testing, the rats were given six trials, during which they were forced into either the right arm or the left arm by a large wooden block that prevented entry

into one of the arms at the choice point. The order of left-right choices was determined by a pseudorandom sequence, so that each rat received three right trials and three left trials during this session. On Day 13 of testing, the rats then began NMTP testing. Each trial consisted of a sample run and a choice run. On the sample run, the rats were forced either left or right by the presence of the wooden block, according to a pseudorandom sequence (with three left and three right turns per session and with no more than two consecutive turns in the same direction), and were able to consume a one-pellet reward. The block was then removed, and the rat was placed facing the experimenter at the start of the choice arm and allowed a free choice of either arm. The rat was rewarded with two pellets for choosing the arm previously unvisited (i.e., for alternating). The subject was deemed to have made a choice when all four feet had crossed a line drawn on the outside of the maze at a point 7 cm from the start of the goal arms. After an incorrect response, the rat was immediately removed from the maze and returned to its holding cage. Rats were run one trial at a time with an ITI of approximately 10 minutes, for a total of six trials per session. The rats were tested for a total of 10 sessions, resulting in a total of 60 trials.

Experiment 4: Locomotor Activity Testing Apparatus. Locomotor activity was measured in a set of specially designed hanging wire cages (Modular Systems and Developments Co. Ltd., Hereford, UK) in which two horizontal photocell beams were located along the long axis of each cage (1.5 cm above the floor and 13 cm apart). Activity data was collected on an Acorn Archimedes RISC PC 600 computer by specialized software (Arachnid Activity Monitor, 1989, Paul Fray Ltd, Cambridge, UK), which provides a record of the number of beam breaks at both the front and the back of the cage independently, the total number of beam breaks, and the number of crossovers. Procedure. The first assessment of spontaneous locomotor activity was conducted 2 months postsurgery, at the end of DRL

1176

BANNERMAN ET AL.

testing. The rats remained on a restricted feeding schedule and were maintained at 85% of their free-feeding weight. The rats were placed individually into the wire activity cages and tested in the dark for 2 hr. The total number of photocell beam breaks was recorded during twenty-four 5-min time bins. The rats were tested again in the locomotor activity cages 7 months later (9 months postsurgery). On completion of T-maze testing, the rats returned to a free-feeding schedule, which was maintained for at least 3 weeks before activity testing. The rats were habituated to the activity cages during two 2-hr sessions conducted 24 hr apart. During these habituation sessions, the spontaneous locomotor activity of the subjects was assessed. The rats were placed individually into the wire activity cages and tested in the dark as before. The total number of photocell beam breaks was recorded during twenty-four 5-min time bins. On the 3rd day of locomotor activity testing, the rats were again habituated to the testing cages, but for 1 hour only. They were then removed from their cages and given intraperitoneal injections of either saline (0.9% [wt/vol]; 1 ml/kg) or amphetamine (Sigma, 1 mg/kg dissolved in saline). They were immediately returned to their cages, and testing resumed in the dark for a further 2 hours (twenty-four 5-min bins). Four days later, the rats were tested for a second time in exactly the same way with the exception that each subject received the opposite injection condition (amphetamine or saline). The sequence of injections was fully counterbalanced within each group. Data collection and analysis. During DRL schedule training in Experiment 1, several performance measures were recorded, including (a) the total number of reinforcers earned, (b) the total number of responses made, and (c) the number of responses made in each of a series of 3-s time bins. Any response emitted within 3 s of the previous response or reinforcer delivery was allocated to Bin 1. Responses elicited between 3 and 6 s were placed in Bin 2, and so on. A total of 25 such bins was collected. Any response that occurred after 72 s was placed in Bin 25. The data were then analyzed in terms of the percentage efficiency for each rat in each session (calculated as the total number of reinforcers collected divided by the total number of responses emitted). Both these forms of analysis generate one number per subject per session. Separate analyses were conducted on the data obtained from the first 10 sessions and the last 4 sessions, to investigate DRL performance both during acquisition and at steady state, respectively. Performance at asymptote was also analyzed according to the temporal distribution of responses (see above). The data obtained from the last 4 sessions of DRL testing were assessed and allocated into 3-s time bins (1-25) as described above. Because nearly all the responses were emitted during the first 24 s after the previous response (or reinforcer delivery), only data from the first eight 3-s bins were analysed. These data were then used to calculate the probability of emitting a response during a particular 3-s time bin, given that the rat had paused in its responding for at least the length of time elapsed before that bin was started (inter-response times per opportunity [IRT/OP]; Anger, 1956). The probability is, therefore, calculated by dividing the number of responses made within a particular time bin by the total number of responses made in that, and subsequent, time bins. For example, if a rat had paused for at least 15s before responding on three occasions, and if on one of those occasions the rat responded before 18s had elapsed but on the other two occasions it responded after 18s, then the IRT/OP value for the 15-18-s time bin would be 0.33 (or 33%). All behavioral measures were subjected to analysis of variance (ANOVA). Statistically significant differences were further examined with analysis of simple main effects or Duncan's post hoc comparisons.

Histology. At the end of behavioral testing, the rats were injected with 200 mg/kg sodium pentobarbital and perfused transcardially with physiological saline followed by 10% formolsaline. Their brains were then removed and placed in formol-saline solution. The brains were then placed in a sucrose-formalin solution for 24 hr, frozen, and sectioned (50 um) horizontally. All brain sections were then stained with cresyl violet.

Results Histology Histological evaluation was carried out by an experimenter unaware of the rats' behavioral performance. The extent of the brain damage produced by the various lesions is described with reference to the horizontal sections in Paxinos and Watson (1986; Plates 87-113). Subjects were accepted into the behavioral analysis providing they fulfilled two principal criteria within the constraints of their respective lesion groups. First, histology had to show almost complete destruction of the pyramidal cells in the CA1-CA4 subfields and of the granule cells in the dentate gyrus. Second, there had to be no concomitant damage to extrahippocampal structures such as the subiculum and entorhinal cortex. There was little evidence of any damage to the overlying cortex in any of the rats (see Figure 2). Examination of the histology from rats with complete hippocampal lesions (Group C) revealed that the lesion was highly reproducible and highly selective (see Figure 3, L-N). There was very little variation from one brain to the next. Cell loss was restricted almost exclusively to the hippocampal subfields, with little if any damage to adjacent structures such as the subiculum or entorhinal cortex and no damage beyond these structures. There was complete loss of pyramidal and granule cells in the dorsal part of Ammon's horn and the dentate gyrus, respectively (see Figure 3L). At midhippocampal levels, there was some very restricted sparing, involving the most posterior portion of the CA1 subfield and the dentate gyrus, at the apex of the hippocampus as it starts to curve downward (minimal sparing extending from Plates 108-102; see Figure 3M). Although this represented only a very limited number of intact cells, sparing was consistent across all rats. The lesion then became complete once more as it extended into the ventral hippocampus (see Figure 3N), until the most ventral tip of the hippocampus, at which point some sparing once again became apparent at Plate 91 (range; Plates 91-93), mainly involving the most posterior portion of the dentate gyrus. All rats from Group C were included in the behavioral analysis (n — 9). The lesions in both Groups D and V resembled very closely the respective lesioned areas in Group C (see Figure 3). As in previous studies (Moser et al., 1995), the transition from healthy to damaged tissue was sharp and well defined, suggesting that the density of neurons within the intact tissue remained relatively unchanged. As with Group C, the lesions in Group D were highly reproducable and highly selective (see Figure 3, D-G). As before, there was almost complete loss of pyramidal and granule cells in the dorsal part of Ammon's horn and dentate gyrus, respectively, with no


Figure 3. Photomicrographs of horizontal sections showing the typical extent of cell loss in the hippocampus in representative rats from each of the four experimental groups. Plate references are approximate and correspond to Paxinos and Watson (1986). A-C: Sections from a sham-operated rat at (A) the level of the dorsal hippocampus (Plate 113), (B) midhippocampal level (Plate 106), and (C) the level of the ventral hippocampus (Plate 97). D-G: Sections from a dorsal hippocampal-lesioned rat at (D) the most superficial level of the dorsal hippocampus, (E) the mid-dorsal hippocampal level (Plate 111), (F) the first point of significant sparing (Plate 109), and (G) the first point at which there was no significant damage (Plate 107). H-K: Sections from a ventral hippocampal-lesioned rat at (H) the point at which damage first became apparent (Plate 108), (I) the first point of significant damage to the ventral hippocampus (Plate 105), (J) the point at which cell loss first became complete (Plate 102), and (K) the temporal pole, demonstrating some sparing of the most ventral tip of the hippocampus (Plate 97). L-N: Sections from a complete hippocampal-lesioned rat at (L) the most superficial level of the dorsal hippocampus, (M) midhippocampal level (Plate 106), and (N) the level of the ventral hippocampus (Plate 97).

1177

1178

BANNERMAN ET AL.

discernable damage to the adjacent subiculum (see Figure 3, D-E). As the lesion extended ventrally, the first significant sparing typically became apparent at Plate 108 (range, 109-107; see Figure 3F). By Plate 107 (range, 107-106) the lesion had become insignificant (see Figure 3G). The remaining portion of the ventral hippocampus was completely unaffected by the lesion. All subjects in Group D fulfilled the relevant criteria and were included in the behavioral analysis (n = 10). Similarly, the lesions in Group V resembled very closely the ventral portion of the complete hippocampal lesions in Group C (see Figure 3, H-K). There was no evidence of any damage to the dorsal hippocampus or subiculum. The first signs of a lesion typically became apparent at Plate 107 (range, 109-105; see Figure 3H), with significant cell loss present by Plate 104 (range, 106-101; see Figure 31). At this point most of the pyramidal and granule cells were destroyed. The lesion was complete (removing all pyramidal and granule cells) by Plate 100 (range, 103-98) and continued with little, if any, evidence of intact cells into the ventral hippocampus (see Figure 3J). As with Group C, there was a small amount of sparing at the most ventral tip of the hippocampus from about Plate 91 (range, 93-90; see Figure 3K). Again, there was no evidence of any damage to either the entorhinal cortex or subiculum. One subject, despite exhibiting substantial damage, had extensive unilateral sparing of the CA1 subfield. Indeed, at no point along the septotemporal axis did the lesion attain completeness. This rat was removed from the behavioral analysis, resulting in a final group size for Group V of n = 9. The above information, coupled with Figure 1, allows us to estimate the volume of tissue removed in both of the partial lesion conditions. In Group D, significant sparing became apparent from Plate 108, and the size of the lesion was insignificant by Plate 107. Figure 1 indicates that this corresponds to a dorsal lesion that removes approximately 55% of the hippocampus. For Group V, the lesions first became apparent at Plate 107, were significant in size by Plate 104, and were complete, in terms of removing all pyramidal and granule cells, by Plate 100. Reference to Figure 1 suggests a lesion size of between 25% and 45% of the total hippocampal volume, with our initial estimation being that 35-40% of the tissue was typically removed. This clearly appears to be a much smaller lesion than intended. It is possible, however, that this figure represents a substantial underestimation of the size of the ventral lesion. The experimental subjects in this study were finally killed by perfusion approximately 10 months after surgery, by which time the lesioned tissue of the ventral hippocampus had completely collapsed away, leading to substantial distortion of the brain. This may have resulted in displacement of the dorsal hippocampus into more ventral areas, and thus led to an underestimation of the size of the lesions in Group V. To cater for this possibility, three additional rats (from the same cohort as those in the main experiment) were prepared according to the surgical methods detailed above. One rat received a unilateral lesion of the dorsal hippocampus, and a second rat received a unilateral lesion of the ventral hippocampus. A third rat received a unilateral dorsal hippocampal lesion on the left hand side of the brain and a ventral

hippocampal lesion on the right hand side of the brain. These three rats were killed by perfusion just 1 week after surgery, long before the likely onset of tissue collapse and any subsequent distortion of the brain. The brains from the first 2 rats were sectioned horizontally and stained with cresyl violet as before. Examination of the histological material obtained from the rat with the unilateral dorsal hippocampal lesion revealed that the lesion was nearly complete at Plate 108, which provides the first evidence of significant sparing. The lesion was insignificant in size by Plate 107. This matches perfectly with the pattern of cell loss obtained from experimental Group D, suggesting that any distortion of the brains in the experimental series did not affect our assessment of the dorsal hippocampal lesions. In contrast, a comparison of the histology from the unilateral ventral hippocampal lesion with the brains obtained from experimental Group V indicates that there may indeed have been a significant distortion of the brains taken from the experimental series. Histology obtained 7 days postsurgery showed that the lesion was apparent from Plate 110, significant in size from Plate 108, and complete by Plate 104. In comparison, in experimental Group V, the lesions first became apparent at Plate 107, were significant in size by Plate 104, and were complete by Plate 100. This supports our hypothesis that there may have been some distortion of the brains from the experimental series, possibly because the remaining dorsal hippocampal tissue may have sunk a small distance after the collapse of the lesioned tissue below. This would have led to an underestimation of the extent of the lesions in experimental Group V. If one, therefore, takes the histological data obtained 7 days postsurgery and refers to Figure 1, then the resulting estimate of the size of the ventral lesion was between 45% and 50% of the total hippocampal volume. The brain from the 3rd rat was sectioned coronally and stained with cresyl violet as before. Inspection of Figure 2 clearly shows the extent of the cell loss in both the dorsal and ventral hippocampus, and the presence of a small degree of overlap (5-10%) between the two lesions at the midhippocampal level as intended.

Experiment 1: Acquisition of a DRL Schedule During DRL testing, all rats readily acquired foodrewarded lever pressing. Inspection of the percentage-ofefficiency data over the 25 sessions of DRL testing showed that Group S rats were the most efficient responders, whereas rats in Group C were the least efficient (see Figure 4). Groups D and V both performed at a level of efficiency part way between that exhibited by Group S and Group C. An ANOVA showed that there was a significant main effect of group, F(3, 34) = 16.28, p < .0001, a significant main effect of session, F(24, 816) = 6.03; p < .0001, but no Group X Session interaction, F(72, 816) = 1.21, p > .10. Subsequent Duncan's pairwise comparisons confirmed that the rats in Group S indeed responded more efficiently than all of the other three groups (p < .01 for all three comparisons). Furthermore, there was also a significant difference


1179

25-

20-

15-

g 10a

a.

5-

10

15

20

25

Day Figure 4. Experiment 1: The effects of complete and partial hippocampal (H) lesions on percentage of efficiency of responding during the acquisition of a differential reinforcement of low rates of responding schedule. The error bar corresponds to two standard errors of the difference of the mean.

between Groups C and V (p < .01). There were no significant differences between Groups D and V or between Groups D and C (p > .05). It was our a priori intention to compare the performance of the four groups during both the period of acquisition and when the rats had reached asymptotic performance. Consequently, two further ANOVAs were performed on data from the first 10 sessions and the last 4 sessions of DRL testing. Analysis of the first 10 sessions showed that there was not only a significant main effect of group, F(3, 34) = 18.62, p < .0001, but also a significant main effect of session, F(9, 306) = 3.50, p < .0005, and a Group X Session interaction, F(27, 306) = 1.99, p < .005. Subsequent analysis of simple main effects revealed that there was significant acquisition for both Group S, F(9, 306) = 6.45, p < .001, and Group V, F(9, 306) = 2.13, p < .05, but not for Group D or Group C (F < I , p > .20 for both groups). This observation was further supported by subsequent analysis of the significant main effect of group with Duncan's pairwise comparisons, which confirmed that Group S acquired the task more rapidly than any of the other three groups (p < .01 for all 3 comparisons). In addition, pairwise comparisons also indicated significant differences between Group V and Group C (p < .01) and between Group V and Group D (p < .05). There was, however, no difference between Group C and Group D during the acquisition phase. Analysis of the percentage-of-efficiency data during the last four sessions of DRL testing revealed that there was no effect of session during this period, suggesting that performance had reached asymptote (F < 1, p > .20). There was also no Group X Session interaction (F < l,p> .20). There was still, however, a significant main effect of group, F(3,

34) = 6.18, p < .005. Further analysis (Duncan's pairwise comparisons) showed that the performance of Group S was still significantly more efficient than the other three groups (Group C atp < .01, Groups D and V at/? < .05). There was also a significant difference between Group V and Group C (p < .05) but not between Groups V and D or Groups D and C (p > .05). Examination of the temporal distribution of the responses made during the last four sessions of DRL testing showed clear evidence of a timing curve in all four groups of rats. There appeared to be a greater overall level of responding in Group C relative to the other three groups. An ANOVA confirmed these impressions. There was a significant main effect of group, F(3,34) = 735, p < .001, a significant main effect of bin, F(6, 204) = 102.95, p < .0001, and a significant Group X Bin interaction, F(18, 204) = 6.98,^ < .0001. Subsequent analysis of simple main effects showed a significant effect of time bin for all four groups of rats (p < .001), thus providing statistical evidence for the presence of timing curves. Duncan's pairwise comparisons confirmed that rats in Group C showed a greater overall level of responding than did rats in either Group S (p < .01), Group D, or Group V (both/?s < .05). There were no other between-group differences.

Experiment 2: Spatial Reference Memory in the Morris Water Maze Acquisition. All rats swam well in the pool and learned to use the platform as a means of escape from the water. There was no evidence of any sensorimotor impairment in any of the rats. During spatial training, all rats spent

BANNERMAN ET AL.

1180

progressively less time at or near the side walls and showed a gradual decrease in the latency to escape from the water (see Figure 5A). An ANOVA conducted for the escape latencies over the 24 trials of water-maze training revealed a significant main effect of group, F(3, 34) = 7.96,p < .0005, a significant main effect of trial, F(23, 782) = 26.49, p < .0001, but no significant Group X Trial interaction, F(69, 782) = 1.07, p > .20. Further examination of the overall group effect by Duncan's post hoc pairwise comparisons revealed that the rats in Group V were escaping from the pool significantly faster than rats in the other three groups (Group V vs. Group C and Group V vs. Group D alp < .01; Group V vs. Group S at p < .05). There were no other significant group differences. Examination of acquisition in terms of distance traveled to the escape platform revealed a slightly different pattern of results. All four groups took progressively shorter swim paths to the escape platform as training proceeded (see Figure 5B). It is also clear, however, that rats in Groups C and D traveled greater distances to locate the platform than did rats in Groups S and V. An ANOVA again revealed a significant main effect of group, F(3, 34) = 9.49, p < .0005,

a significant main effect of trial, F(23, 782) = 25.88, p < .0001, but no Group X Trial interaction, F(69, 782) = 1.08, p > .20. Post hoc Duncan's pairwise comparisons indicated that rats in both Groups C and D were impaired relative to both Group S (p < .05) and Group V (p < .01). There were no other significant group differences despite the indication from Figure 5B that rats in Group V performed even better than rats in Group S at the intermediate training stage. This latter observation was so surprising that we conducted a further ANOVA comparing distance traveled in just Groups S and V. This yielded a significant main effect of lesion, F(l, 17) = 10.75, p < .005. Even correcting for multiple post hoc testing, this is a striking result that warrants further investigation. Examination of swim speeds during acquisition revealed a further consequence of ventral lesions that must be taken into account. Rats in both Groups C and V swam consistently faster than rats in Groups S and D (see Figure 5C). An ANOVA showed a significant main effect of group, F(3, 34) = 5.31, p < .005, a significant main effect of trial, F(23, 782) = 3.99, p < .0001, and no Group X Trial interaction (F < 1, p > .20). Duncan's pairwise comparisons demon-

0.45 -i

120100-

0.40 -

T i

3 80-

0.35 -

>> g

60-

j

40-

_§ '^ X

'

Sham

T 1

T 1

T 1

C o m p l e t e H DorsalH V e n t r a l H

20-

Group

0 0

3 Day

B

D 401

30-

« "a

20-

3 Day

Sham

CompleteH

DorsalH

Group

Figure 5. Experiment 2: The effects of complete and partial hippocampal (H) lesions on the acquisition of a standard spatial reference memory task in the Morris water maze. A: Mean escape latencies (±SEM) during each day of acquisition. B: Mean distance traveled to the escape platform (±SEM) during each day of acquisition. C: Mean swim speed (±SEM) averaged across the 6 days of acquisition training. D: The mean percentage of time (±SEM) spent in the four quadrants of the pool during the first 30 s of the transfer test conducted at the end of spatial training. The platform had previously been located in the training quadrant (solid bar) during acquisition.

VentralH


strated that the rats in both Groups C and V swam significantly faster than rats in both Groups S (p < .01) and Group D (p < .05). There were no other group differences. This suggests that an increased swim speed provides a major contribution to the reduced escape latencies observed with ventral lesions. However, we cannot preclude the possibility that rats in Group V swam not only faster but also more accurately. Transfer test. On the 7th day of water-maze testing, the platform was removed from the pool and the rats were allowed to swim freely for 60 s. Analysis of the first 30 s of this transfer test revealed that rats in Group S (52.1%) and Group V (58.0%) spent a considerable proportion of their time searching in the quadrant of the pool in which the platform had previously been located (see Figure 5D). In contrast, rats from both Groups C (24.1%) and D (30.7%) exhibited a more diffuse pattern of searching, with much less spatial bias toward the former training quadrant. An ANOVA revealed a significant main effect of quadrant, F(2, 102) = 16.73, p < .01, and a significant Group X Quadrant interaction, F(6, 102) = 4.49, p < .01. A subsequent analysis of simple main effects indicated that there was a significant effect of quadrant for Group S, F(2, 102) = 12.58, p < .01, and Group V, F(2, 102) = 16.71, p < .01. In contrast, there was no quadrant effect for the rats in Group C (F< l,p>.20)orGroupD(F< \,p> .20). This analysis of simple main effects also showed that there was a significant group difference in the amount of time spent in the training quadrant, F(3, 102) = 12.24, p < .001, but not in any of the other three quadrants. A subsequent ANOVA comparing the time spent only in the training quadrant revealed a significant overall effect of group, F(3, 34) = 8.67, p < .0005. Subsequent Duncan's pairwise comparisons revealed that the rats in Group C were spending significantly less time in the training quadrant compared with rats from either Group S or Group V (p < .01 for both comparisons). Likewise, the rats in group D showed less spatial bias toward the former training quadrant than did rats from either Group S (p < .05) or Group V (p < .01). Duncan's pairwise comparisons revealed no other significant group differences.

1181

differences between Group C and Group S (p < .01) and between Group C and Group V (p < .05). There were no other significant between-group differences. The 10 sessions of NMTP testing (six trials per session) were combined so that each rat contributed 1 score out of 60 to the analysis. Inspection of the data showed that the performance of the four groups during spatial working memory testing on the elevated T maze was similar to that observed for spatial reference memory in the Morris water maze (see Figure 6). Rats in both Group S (92%) and Group V (84%) performed well on the spatial working memory task. In contrast, rats in Group C (59%) and Group D (70%) displayed much lower levels of performance. An ANOVA revealed a significant main effect of group, F(3, 34) = 17.35, p < .0001. Duncan's pairwise comparisons confirmed that rats in Group D were significantly impaired relative to both Group S and Group V (both at/? < .01) and that rats in Group C were impaired relative to Group S, Group V (both alp < .01), and Group D (p < .05). There were no other significant between-group differences. Experiment 4: Locomotor Activity Testing The first locomotor activity test session was conducted 2 months postsurgery, immediately after the completion of DRL testing (Experiment 1), while the rats were still food deprived. Rats in Group C demonstrated considerably higher levels of spontaneous activity than either Group S or Groups D and V (see Figure 7A). Levels of activity in Groups D and V and in Group S were comparable. This was confirmed by subjecting the number of beam breaks made by each rat to an ANOVA, which revealed a significant main effect of group, F(3,34) = 10.85,p < .0001, a main effect of time bin, F(23, 782) = 89.60, p < .0001, but no Group X Time Bin interaction, F(69, 782) = 1.10,;? > .20. Further analysis of the main effect of group confirmed that Group C was significantly more active than the other three groups (p < .01). There were no significant differences between 10090-

Experiment 3: Spatial Learning on the Elevated T Maze All rats performed well on the maze and avidly consumed the food rewards. During spontaneous alternation testing, the sequence of left and right choices made by each subject was recorded, and the number of alternations was calculated. The six sessions of spontaneous alternation testing were combined so that each rat contributed a score with a maximum of 30 (6 X a maximum of 5 alterations per rat per session). Rats in Group S demonstrated a natural tendency to alternate (71%). This propensity was greatly reduced in Group C (47%). Rats in Groups D and V showed an intermediate level of alternation (Group D, 56%; Group V, 65%). These data were subjected to an ANOVA, which revealed a significant overall effect of group, F(3, 34) = 3.92, p < .05. Duncan's pairwise comparisons revealed

T 1

t so-

T i

1 70t,

OJ

C. C 60"

T X

1 50-

rr

Sham

CompleteH DorsalH

VentralH

Group Figure 6. Experiment 3: The effects of complete and partial hippocampal (H) lesions on the performance of a nonmatching to place task on the elevated T maze. Mean percentage of correct responses (±SEM) averaged across 60 trials.

BANNERMAN ET AL.

1182

Groups S, D, and V. An identical pattern of results was obtained when the number of crossovers made by each rat was analyzed (data not shown). Spontaneous locomotor activity was assessed again 7 months later (the rats then had ad-lib access to food) and the same pattern of results was obtained. An ANOVA of the number of beam breaks again revealed a significant main effect of group, F(3, 34) = 8.10,;? < .0005, which was due

to the hyperactivity of Group C relative to the other three groups (p < .01). There were no differences between the other three groups. The ANOVA also showed a significant effect of time bin, F(23, 782) = 90.39, p < .0001, and a marginally significant Group X Time Bin interaction, F(69, 782) = 1.31, p = .05. Analysis of crossovers revealed an identical pattern of results (data not shown). When the rats were tested again under identical conditions 24 hr later, a

350^ 300'x' c o 250CJ •5

200150-

o 'o £ o u o

100-

50-

10

20

30

40

50 60 70 80 Time (minutes)

90

100

110

120

B 3,500-,

Sham

CompleteH

DorsalH

VentralH

Group Figure 7. Experiment 4: The effects of complete and partial hippocampal (H) lesions on spontaneous and amphetamine-induced locomotor activity in photocell activity cages. A: Spontaneous locomotor activity conducted 2 months postsurgery. Data were collected in 5-min time bins and expressed as the mean number of beam breaks per time bin (±SEM). B: The locomotor response after an intraperitoneal injection of either saline (1 ml/kg) or amphetamine (1 mg/kg in 1 ml/kg saline). Rats were habituated to the activity cages for 1 hr before injection. Data are expressed as the total number of beam breaks made during the 2-hr test session (±SEM) after either saline (open bars) or amphetamine (solid bars).


similar result was obtained in terms of crossovers, although, in terms of total beam breaks, the between-group differences did not quite reach statistical significance (data not shown). The two drug test sessions were conducted as part of a counterbalanced design, therefore, the data from the two sessions were combined. Examination of the data obtained for total beam breaks showed that amphetamine treatment produced a substantial increase in locomotor activity in all four groups of rats, with by far the greatest increase evident in Group C (see Figure 7B). The magnitude of the response to amphetamine was comparable in Groups S, D, and V. For each rat, the data for the twenty-four 5-min postinjection time bins were taken for both the saline and amphetamine conditions and subjected to an ANOVA. This confirmed that there was a significant overall effect of drug condition, F(l, 34) = 239.18, p < .0001, a significant overall effect of group, F(3, 34) = 4.20, p < .05, and a Group X Drug Condition interaction, F(3, 34) = 3.58, p < .05. Subsequent analysis of simple main effects revealed that there was a significant between-group difference for the amphetamine condition, F(3, 62) = 7.13, p < .001, but not for the saline condition (F .20). The ANOVA also showed an overall effect of time bin, F(23, 782) = 90.63, p < .0001, a Group X Time Bin interaction, F(69, 782) = 1.97, p < .0001, a Drug Condition X Time Bin interaction, F(23, 782) = 15.08, p < .0001, and an overall Group X Drug Condition X Time Bin interaction, F(69, 782) = 2.03, p < .0001. A second subsequent ANOVA, including just the data from the amphetamine condition, revealed a significant overall effect of group, F(3, 34) = 3.80,p < .05. There was also a significant effect of time bin, F(23,782) = 44.95, p < .0001, and a Group X Time Bin interaction, F(69, 782) = 2.03, p < .0001. Further examination of the significant main effect of group revealed that the amphetamine response exhibited by rats in Group C was indeed significantly greater than that observed in either Group S, Group V (p < .05) or Group D (p < .01). There were no significant differences between Groups S, D, and V in terms of the locomotor response to amphetamine. Analysis of crossovers again revealed an identical pattern of results (data not shown). Discussion The present experiments compared the performance of rats with dorsal and ventral partial hippocampal lesions to rats with complete lesions and to sham operated controls. All the behavioral tasks have previously been found to be sensitive to hippocampal damage. As expected, complete cytotoxic hippocampal lesions (a) impaired the acquisition of a standard spatial reference memory task in the Morris water maze (Morris et al., 1990), (b) disrupted spatial working memory performance during NMTP testing on the elevated T maze (Rawlins & Olton, 1982), (c) reduced efficiency on a DRL operant task (Sinden et al., 1986), and (d) increased both spontaneous and amphetamine-induced locomotor activity levels (Whishaw & Jarrard, 1995; Wilkinson etal, 1993). - Dorsal hippocampal lesions impaired spatial learning in the water maze, both during acquisition (escape latencies

1183

and distances to the platform), and in transfer test performance, whereas ventral hippocampal lesions left performance intact, in agreement with the previous study of Moser et al. (1995). Moreover, dorsal but not ventral hippocampal lesions impaired NMTP performance on the elevated T maze, in agreement with Hock and Bunsey (1998), although this effect was not quite as large as that observed for complete lesions. Hence, as expected on these tasks, dorsal hippocampal lesions produced effects similar to those of complete hippocampal lesions. Ventral lesions did not, and in fact left performance statistically at or above control levels. Different patterns of results were seen in the other tasks. In DRL, both partial lesion groups were less efficient than controls, although neither group was as impaired as the complete lesion group. Neither of the partial lesions affected spontaneous or amphetamine-induced locomotion in the photocell cages. This latter result contrasts with those of a previous study by Lipska et al. (1992) showing that lesions of the ventral hippocampus increase locomotion; in the present study only the complete lesion led to increased levels of activity. Finally, ventral but not dorsal lesions produced a robust increase in swim speeds during water-maze training. The complete lesion group showed an identical increase. These results thus provide not only a varying pattern of equivalence between dorsal, ventral, and complete cytotoxic hippocampal lesions, but also, importantly, a within-subjects double dissociation between the effects of dorsal and ventral hippocampal lesions. Lesion Location and Lesion Size In principle, there could be two, quite separate accounts of the differential effects of partial hippocampal lesions. The first would be simply in terms of lesion size: The bigger the lesion, the more pronounced the behavioral effect. The second would be in terms of lesion location. By this account, two lesions of equal size but at different locations would have two different effects. The present pattern of results requires reference to both accounts; neither alone can account for all the data. The results from the DRL study (Experiment 1) require no more than an explanation in terms of lesion size. The complete lesion produced the biggest deficit; the partial lesions produced a smaller but still significant deficit; and, over the last 4 days of testing, the smallest lesion group (the ventral group) differed not only from the controls but also from the complete hippocampal group. Hence, lesion location appeared to be unimportant, echoing the conclusions of Sinden et al. (1986). Results from the photocell activity cages (Experiment 4) may also be explicable in terms of lesion size; but if so, there is a clear lesion threshold effect that was not apparent in the DRL study. Alternatively, dorsal and ventral hippocampal lesions may have different, independent effects on locomotion, both of which must be present for hyperactivity to be observed in this test. The present data are consistent with either account. In contrast, it has already been convincingly demonstrated

1184

BANNERMAN ET AL.

that lesion size per se is not the critical variable in accounting for the disruption of spatial reference memory induced by hippocampal damage (Moser et al., 1995). A lesion of the septal pole, sparing 60% of the ventral hippocampus, impaired water-maze performance, whereas a lesion of the temporal pole, sparing no more than 26% of the dorsal hippocampus, left spatial learning essentially intact (Moser et al.). The spatial impairment observed in Experiment 2 is entirely consistent with an account based on lesion location. Path lengths and transfer test performance never differed between dorsal and complete lesion groups (though the present results cannot in themselves rule out a contribution from overall hippocampal lesion size, given that the ventral lesions were the most restricted of all). Although a similar pattern of results was obtained for NMTP on the elevated T maze (Experiment 3), in this case lesion size appeared to contribute as well as lesion location. The rats with dorsal lesions were significantly impaired with respect to the control and ventral lesion groups, but at the same time, were significantly less impaired than rats with complete lesions. The increased swim speed observed during water-maze training in rats with ventral hippocampal damage permits no ambiguity of interpretation. This can only be attributed to lesion location rather than lesion size, because the large dorsal lesion left swim speed unaltered. Moreover, the effect of a complete lesion was essentially identical to that of the smaller, selective ventral lesion. The present demonstration of a ventral hippocampal lesion effect in the absence of any dorsal hippocampal lesion effect provides a clear double dissociation of functions within the hippocampus. Task Sensitivity: An Effect Size Approach One response to the observation that the relative efficacies of our different lesions varied as a function of behavioral test paradigm might be to suggest that this arises simply because the tasks differ in their sensitivity to hippocampal dysfunction. By this view, the absence of any ventral hippocampal lesion effect on spatial memory might simply be because those tasks are not as sensitive to hippocampal dysfunction as are other tasks (e.g., DRL). Hence, a ventral hippocampal lesion effect is detectable in the latter but not the former tasks. There are two ways to address this issue. The first is to quantify the actual sensitivity of the tasks we deployed. We did this by calculating the effect size (Cohen, 1977) for each of the tasks in which we saw a main effect of lesion. We conducted additional ANOVAs, comparing only the complete hippocampal group and controls, and derived a population standard deviation from the mean square error term for the between-subjects layer of the analysis. From this we calculated the effect size by dividing the difference between the two means by the standard deviation (see Table 2). This revealed that the most sensitive measure of complete hippocampal damage was derived from the T maze study (on which there was no significant effect of ventral hippocampal damage). For comparison, the clear

Table 2 Effect Sizes for Sham-Operated Versus Complete Hippocampal-Lesioned Rats Behavioral parameter

Effect size

T maze/nonmatching to place Water maze/% time in training quadrant DHL/efficiency Spontaneous locomotor activity (2 months) Spontaneous locomotor activity (9 months) Water maze/swim speed Water maze/distance to escape platform Amphetamine-induced locomotor activity Water maze/latency to escape platform

2.01 1.51 0.59 0.42 0.35 0.33 0.24 0.18 0.13

Note. Effect sizes are calculated according to methods from Cohen (1977). DRL = differential reinforcement of low rates of responding.

effect of ventral hippocampal damage that we saw on the swim speed measure was observed in a task with only a moderate sensitivity to complete hippocampal damage. It is clear that the presence or absence of a significant effect of ventral hippocampal damage did not rely on task sensitivity to hippocampal damage per se. Likewise, the dorsal lesion significantly increased distance to escape during water-maze training while leaving swim speed and locomotor activity measures unaffected, even though these were more sensitive to complete hippocampal damage (see Table 2). Hence the nature of the task, and not the general sensitivity of the task, determined the presence or absence of both dorsal and ventral hippocampal lesion effects. The second is to take account of the clear double dissociation, visible within a single task, between the effects of dorsal and ventral partial lesions. In the water maze, the dorsal lesion impaired measures of spatial memory, whereas the ventral lesion did not; but in the same task, the ventral lesion led to increases in swim speed, whereas the dorsal lesion did not. This pattern of results can only be explained by reference to functional specialization within the hippocampus; no single task sensitivity parameter can accommodate these results. Taken in conjunction with the parallel observations reported by Richmond et al. (1999), we are forced to conclude that dorsal and ventral hippocampal lesions affect at least two, separate, psychologically distinct processes. Complete hippocampal lesions appear to affect both, as though these effects can be additive. Partial Hippocampal Lesion Effects on Spatial Learning Dorsal hippocampal lesions caused substantial impairments of both spatial reference memory (Experiment 2) and spatial working memory (Experiment 3) but were ineffective in other paradigms that were clearly sensitive to complete hippocampal lesions (locomotor activity in Experiment 4 and the swim-speed measure in Experiment 2). There were no clear differences in spatial performance between the dorsal and complete lesion groups in the water maze. The dorsal lesion group did, however, perform significantly


better than the complete lesion group on the T maze. Thus, the T maze more readily demonstrated residual function in rats with incomplete hippocampal damage than did the water maze; this is not to say that, with different specific testing protocols, the water maze would not be expected to yield similar results. These results support suggestions that the dorsal hippocampus has a specialized role in spatial learning (Hock & Bunsey, 1998; Moser et al., 1995), consistent with both electrophysiological and anatomical findings (Jung et al., 1994; Room & Groenewegen, 1986; Ruth et al., 1982; Witter & Groenewegen, 1984; Witter et al., 1989). It was equally clear that ventral lesions produced no clear spatial deficit. Indeed, there was some evidence of enhanced spatial performance in some measures in the water maze (see also Richmond et al., 1999), but it seems premature to claim that this lesion actually improves spatial memory. First, previous studies using similar lesions have reported no such effect, either in the water maze (Moser et al., 1995), or on the T maze (Hock & Bunsey, 1998). Second, in our own T maze study (Experiment 3) the ventral-lesioned group performed, if anything, slightly worse than controls (see Figure 6). It therefore seems possible that any apparent improvement in water-maze performance may have resulted from the undoubted increase in swim speed produced by ventral hippocampal damage. This might even indirectly improve accuracy: More rapid contact with the platform on early training trials might lead to more rapid spatial acquisition relative to controls, if spatial memory per se is unaffected by the lesion. The Effects of Hippocampal Damage on Locomotor Activity Rats with complete cytotoxic hippocampal lesions reliably demonstrated hyperactivity. These subjects showed increased levels of spontaneous activity on a number of separate occasions, ranging from 2-9 months postsurgery. They also exhibited an increased locomotor response to amphetamine and swam significantly faster during watermaze training relative to sham-operated rats. Furthermore, in the companion study by Richmond et al. (1999), rats with complete hippocampal lesions showed dramatically higher activity levels than did controls after receiving shock in an operant chamber. The observation that rats with complete hippocampal lesions are hyperactive is an extremely robust and well documented phenomenon (eg. Whishaw & Jarrard, 1995; Wilkinson et al., 1993). Neither dorsal nor ventral lesioned rats reliably demonstrated hyperactivity in the photocell activity cages relative to controls, either with or without amphetamine; the complete lesion group was consistently more active than both the partial groups. The absence in the present study of either any increase in spontaneous activity or any enhancement of the locomotor response after amphetamine in rats with ventral hippocampal lesions appears to be at odds with previous work, which has shown that ventral hippocampal lesions increase both spontaneous and amphetamine-induced activ-

1185

ity levels in both neonatal (Lipska et al., 1993; Sams-Dodd et al., 1997) and adult rats (Lipska et al., 1992). The reason for this discrepancy is not immediately obvious, although several possible accounts exist. First, the lesions differed. Lipska et al.'s lesions were larger than our own, in terms of both intra- and extrahippocampal damage. Their intrahippocampal damage may be more comparable to our complete lesion, which increased both spontaneous and amphetamine-induced locomotor activity. As noted above, there may be a clear threshold in terms of lesion size for observing hyperactivity. Hence, a larger ventral lesion could result in more widespread hyperactivity, including an enhanced locomotor response to amphetamine (but it should be noted that the current ventral lesion was sufficiently large to enhance swim speed during water-maze training and to increase locomotor activity postshock; see Richmond et al., 1999). In addition, extrahippocampal damage, in particular to the subiculum, may also have contributed to the hyperactivity that Lipska et al. (1992) observed. An alternative explanation is that the experimental protocols differed in several potentially important ways. They differed in the time between lesioning and testing (