Neuron, Vol. 35, 625���641, August 15, 2002, Copyright ���2002 by Cell Press Review The Human Hippocampus and Spatial and Episodic Memory appears to remain in humans (Abrahams et al., 1999 Maguire et al.,1996a, 1998a, 1999 Spiers et al., 2001a Vargha-Khadem et al., 1997). This latter observation is Neil Burgess,1,3 Eleanor A. Maguire,2 and John O���Keefe1 1Institute of Cognitive Neuroscience and consistent with the cognitive map theory characterization Department of Anatomy and Developmental of hippocampal function (O���Keefe and Nadel, 1978, 1979). Biology The cognitive map theory proposes that the hippo- University College London campus of rats and other animals represents their envi- 17 Queen Square ronments, locations within those environments, and London WC1N 3AR their contents, thus providing the basis for spatial mem- 2 Wellcome Department of Imaging Neuroscience ory and flexible navigation. When it comes to humans, Institute of Neurology the theory suggests a broader function for the hippo- University College London campus, based at least in part on lateralization of func- 12 Queen Square tion. The right hippocampus is still viewed as encoding London WC1N 3BG spatial relationships, but the left has the altered function United Kingdom of storing relationships between linguistic entities in the form of narratives. In addition, one or both hippocampi incorporate temporal information derived from the fron- Finding one���s way around an environment and remem- tal lobes, which serves to timestamp each individual bering the events that occur within it are crucial cogni- visit to a location, thus providing the basis for a spatio- tive abilities that have been linked to the hippocampus temporal contextual or episodic memory system. and medial temporal lobes. Our review of neuropsy- The hippocampus has also been ascribed a much chological, behavioral, and neuroimaging studies of broader role in both animals and humans, encompassing human hippocampal involvement in spatial memory episodic and spatial memory along with many other concentrates on three important concepts in this field: types of memory. Primary among these broader charac- spatial frameworks, dimensionality, and orientation terizations are ���declarative��� memory (Squire and Zola- and self-motion. We also compare variation in hippo- Morgan, 1991), and ���flexible relational��� memory (Cohen campal structure and function across and within spe- and Eichenbaum, 1993). Declarative memory refers to cies. We discuss how its spatial role relates to its all forms of conscious or explicit memory, including epi- accepted role in episodic memory. Five related studies sodic, semantic, and familiarity-based recognition, with use virtual reality to examine these two types of mem- the additional suggestion that the hippocampus plays ory in ecologically valid situations. While processing a time-limited role (i.e., being needed only for recently of spatial scenes involves the parahippocampus, the acquired information). Under the flexible-relational hy- right hippocampus appears particularly involved in pothesis, hippocampal function is closely related to de- memory for locations within an environment, with the clarative memory, including all explicit memory (Eichen- left hippocampus more involved in context-dependent baum, 1999), but also favors flexible uses of memory and episodic or autobiographical memory. relational learning (e.g., performing transitive inference Bunsey and Eichenbaum, 1996 Dusek and Eichenbaum, Background 1997). See also Aggleton and Brown (1999) for a review. Impairments of spatial and episodic memory are often Here, we focus on the involvement of the human hip- the first symptoms experienced by patients with dam- pocampus in spatial memory and review relevant neuro- age to the medial temporal lobes due to progressive psychological, behavioral, and neuroimaging studies. In pathologies such as Alzheimer���s disease (e.g. Kolb and addition, we consider how results concerning its spatial Wishaw, 1996). The medial temporal lobes and the hip- role may relate to its accepted role in episodic memory. pocampus in particular have long been implicated in the We also indicate some links to the pertinent nonhuman acquisition of new memories (Scoville and Milner, 1957), data, but do not consider this field in any detail (see with visuo-spatial memory predominantly associated Eichenbaum et al., 1999, and O���Keefe, 1999, for reviews). with the right (Smith and Milner, 1981) and verbal or In the first instance, three concepts of particular impor- narrative memory with the left (Frisk and Milner, 1990). tance to understanding spatial memory will be briefly There is now a consensus that the human hippocam- reprised and key evidence reviewed: spatial frame- pus is involved in episodic memory (Eichenbaum and works, dimensionality, and orientation and self-motion. Cohen, 2001 Kinsbourne and Wood, 1975 O���Keefe and We will then discuss how the introduction of a novel Nadel, 1978 Squire and Zola-Morgan, 1991 Vargha- methodology, namely the use of virtual reality to create Khadem et al., 1997), i.e. memory for personally experi- large-scale, controlled environments, has provided new enced events set in a spatio-temporal context (Tulving, opportunities to explore these key concepts and the 1983). Equally, there is little dispute that the hippocam- role of the hippocampus in space. In particular, a series pus in infrahumans is involved in spatial or topographical of five recent virtual reality (VR) experiments that exam- memory (Eichenbaum and Cohen, 2001 Morris et al., ined topographical and episodic memory within large- 1982 O���Keefe and Nadel, 1978), and this spatial role scale spatial contexts will be considered in detail (Bur- gess et al., 2001b King et al., 2002 Maguire et al., 1998a Spiers et al., 2001b, 2001a). We believe that experiments 3 Correspondence: n.burgess@ucl.ac.uk

Neuron 626 such as these are starting to provide convergent evi- of the elements of a scene. For example, either might suffice for recognition of a scene as familiar, but the dence from neuropsychology and neuroimaging about the role of the hippocampus in memory. latter would be needed to decide upon a novel shortcut or appreciate what the scene would look like from an- Spatial Frameworks When investigating the processing of spatial information other point of view. The need to take care in distinguish- ing between different types of spatial information is illus- in the brain, it is natural to ask what type of spatial framework is being used to represent locations and what trated by reports of patients with topographical memory deficits but preserved ability in tabletop tests of spatial is the origin or center of that framework. Frameworks can be centered on different receptor surfaces, such as or geographical knowledge (Habib and Sirigu, 1987 Mc- Carthy et al., 1996) and conversely, a patient with pre- the retina, or they can be aligned with a body part, such as the midline of the head or the trunk, or with served navigational ability but poor verbal and visual memory and poor geographical knowledge (Maguire an effector, such as the arm or the hand. All of these frameworks, since they move with the body as it moves and Cipolotti, 1998). There is some evidence that parahippocampal cortical through the environment, are collectively labeled ego- centric. In contrast, frameworks that are fixed to the areas are required for iconic representations of scenes, with hippocampus being required in addition when environment itself or to individual objects in the environ- ment are called allocentric. The locations of objects memory for locations in 3D space is required. Functional neuroimaging of recognition memory for object location within allocentric frameworks do not change as the sub- ject moves in the environment. within a 2D array (Johnsrude et al., 1999) and the percep- tion of spatial scenes (Epstein and Kanwisher, 1998) The cognitive map theory posits that the hippocam- pus specifically supports allocentric processing of and buildings (Aguirre et al., 1998) consistently activates posterior right parahippocampal gyrus, but not the hip- space in contrast to other brain regions, such as the parietal neocortex, which support egocentric pro- pocampus. Similarly, recognition-based tests of spatial memory, including the recognition of landmarks (White- cessing (O���Keefe and Nadel, 1978). This is consistent with hippocampal ���place cells��� encoding the rat���s loca- ley and Warrington, 1978) and topographical scenes (Warrington, 1996) have been associated with parahip- tion within an open environment independently of its orientation (Muller et al., 1994 O���Keefe, 1976 Wilson pocampal areas in neuropsychological studies (Bohbot et al., 1998 Habib and Sirigu, 1987). Recognition of and McNaughton, 1993) and the complementary encod- ing of the orientation, independently of location, by spatial scenes has also been found to be impaired in right temporal lobectomy patients (Baxendale et al., ���head-direction cells��� in the nearby presubiculum (see e.g., Taube et al., 1990). By contrast, neocortical repre- 1998a Pigott and Milner, 1993). Both recall and recogni- tion of object locations is impaired in epilepsy patients sentations of sensory and motor information tend to be egocentric, reflecting the fact that sensory receptors after either right parahippocampal or hippocampal le- sion (Bohbot et al., 1998). Testing of unilateral temporal and motor effectors are attached to the body. Interest- ingly, as well as encoding information relative to various lobectomy (Nunn et al., 1998, 1999 Smith and Milner, 1981, 1989) and amygdalohippocampectomy (Smith et egocentric frames of reference (see e.g., Colby and Goldberg, 1999 Hyvarinen and Poranen, 1974 Mount- al., 1995) patients has clearly implicated the right medial temporal lobe in recalling the locations of objects laid castle et al., 1975), neurons in the posterior parietal cortex also appear to support translations between dif- out in a two-dimensional array (i.e., they replace the objects on the table less accurately). Moreover, the se- ferent egocentric frames (Andersen et al., 1985) and between allocentric (room-centered) and egocentric verity of impairment correlates with the extent of right hippocampal damage. In the majority of these experi- (trunk-centered) frames (Snyder et al., 1998). The different ways in which space is processed and ments, the impairment was only apparent after delays of several minutes or more between presentation and represented impact upon neuropsychological testing in humans. In tests of memory for locations on a tabletop recall, showing that spatial perception was not impaired in these patients. or computer screen relative to which the subject does not move, the use of egocentric or allocentric pro- Orientation and the Role of Self-Motion The mental representation of space must not only con- cessing cannot be distinguished, as the two frames of reference coincide. Notions of ego- or allocentricity are tain the relative locations of objects in the environment, but also has to be orientated appropriately with respect naturally intimately linked to other factors. For example, the dynamic process of extracting spatial information to that environment. Insight into how this is achieved comes from experiments in which the subject is shown from navigating through an environment will also differ from the process of extracting that information from a objects in different locations in a symmetrical rectangu- lar room, is then disoriented (using blindfolded rotation), two-dimensional (2D) map or overhead view. Similarly, scene and landmark information stored as retinotopic and then asked to locate them. Young children (Hermer and Spelke, 1994) and adults performing verbal shadow- snapshots (e.g., presented using pictures) will differ from that stored as a result of navigating through the environ- ing (Hermer-Vazquez et al., 1999) appear to reorient themselves solely according to the geometry of the ment. Several dissociations in performance across test types that speak further to the issue of the egocentric/ room���ignoring a large, colored cue present on one wall that enables adults to orient correctly under normal cir- allocentric distinction are discussed in the next sections. Dimensionality cumstances. Interestingly, this effect is weakened when larger rooms are used (Learmonth et al., 2001). A similar A fundamental distinction exists between simple iconic representations of single objects or 2D scenes and rep- disorientation procedure indicates that adults encode the locations of objects in a room individually, but en- resentations that include knowledge of the 3D locations

Review 627 code the locations of geometric features (the corners All of these studies, like the earlier ones, implicated the right medial temporal lobe. Correlation of performance of the room) within a single, unified representation (Wang and Spelke, 2000). with the extent of damage to the hippocampus and parahippocampal pointed to a crucial role for these two Experiments suggest that self-motion produces idi- othetic signals that can be used to update the orientation structures. Further evidence for a hippocampal locus for spatial of the spatial representation of an environment. Subjects were shown an array of objects on a circular table and, memory comes from a study of memory for the location of a spot of light in a patient with selective bilateral while a curtain was lowered over the table, either the subject moved around the table to a new viewpoint or hippocampal pathology (Holdstock et al., 2000b). In this study, use of egocentric representations was encour- else the table was rotated by an equivalent amount. Subjects were then better able to recognize object loca- aged by switching off the lights and testing from the same view, while allocentric representations were en- tions after self-motion than after rotation of the table (Simons and Wang, 1998 Wang and Simons, 1999). This couraged by leaving the light on and having the subject move between presentation and recall. A marginally result implies that self-motion causes an automatic up- dating of an internal representation of locations that is greater impairment was found for the allocentric condi- tion, although this also depended on the use of a filled more accurate than our ability to deliberately perform the equivalent mental rotation. Note that this process delay and on increased variance in controls��� perfor- mance in the egocentric (dark) condition. As noted probably corresponds to updating a viewpoint within a cognitive mental model but does not necessarily rely on above, while rotation of viewpoint around an array of objects does not guarantee the use of an allocentric vestibular or proprioceptive signals, as the experimental effect can also be shown using purely visual virtual real- representation, it does at least require some equivalent mechanism of generating a novel viewpoint on egocen- ity (Christou and Bulthoff, 1999). While it is conceivable that one���s knowledge of the location of objects or the trically encoded information. Equally, switching off the lights does not prevent the use of allocentric processes. layout of a simple environment is continuously updated to compensate for every movement, this ability is un- Nonetheless, a hippocampal role in rotating or otherwise manipulating viewpoints in memory would be an inter- likely to be able to accommodate the complex move- ments, long time scales, rich and continuous stimuli, esting possibility, especially since hippocampal patients are not impaired at the related task of mental rotation and multiple choices involved in navigation in the real world. The related process of maintaining a bearing to of single objects from a fixed viewpoint (Holdstock et al., 2000b Spiers et al., 2001a). the start point of a trajectory on the basis of movement information alone (path integration) can be seen to be In summary, psychological studies indicate the presence of an automatic process that updates internal insufficient to accommodate complex movements in mammals (see e.g., Etienne et al., 1996). representations to accommodate the consequences of self-motion. Neuropsychological evidence indicates The processes underlying spatial memory have been illuminated by several recent studies of performance involvement of the medial temporal lobes (most particu- larly the hippocampus) in memory for locations after and reaction times in healthy volunteers. Subjects being asked to indicate the locations of objects from a shifted movement of the subject but does not conclusively iden- tify the nature of its involvement, with issues such as point of view show a chronometric relationship between reaction times and the size of the shift in viewpoint. This frame of reference and generalizability from 2D to 3D real world situations still unclear. relationship indicates that subjects perform an iterative mental manipulation to align the test viewpoint with the Virtual Reality Realization of the importance of self-motion in the con- encoding viewpoint (Diwadkar and McNamara, 1997). The same type of chronometric relationship holds for struction and use of spatial representations, coupled with recent technological advances and the need for imagined translations of viewpoint (Easton and Sholl, 1995). In a parallel result to that of Simons and Wang repeatability and control across subjects, has led to an increase in the experimental use of VR (see e.g., Burgess (1998), subject���s memory for spatial locations is faster and more accurate following imagined movement of and King, 2001 Maguire et al., 1999). The extent of immersion or presence felt in a VR environment, i.e., the viewpoint around an array of locations than following an equivalent imagined rotation of the array (Wraga et degree to which the user treats it as s/he does the real world and behaves in a similar manner, is obviously an al., 2000). See also Kosslyn (1994). Performance or reaction time advantages have also important concern. In many of the experiments consid- ered below, movements are simply generated using a been noted for novel viewpoints aligned with environ- mental axes or landmarks (Mou and McNamara, 2002 joystick or keypad, which is not optimal for the percep- tion of the amplitude of turning movements (Chance et Shelton and McNamara, 2001). Several recent investiga- tions of object-location memory in (pre- and postopera- al., 1998), which tend to be overestimated (Klatsky et al., 1998). Conversely, there is evidence that distances tive) unilateral temporal lobectomy patients have used movement of the subject between presentation and re- can be underestimated (Witmer and Klein, 1998). De- spite these limitations, several studies have indicated trieval of object locations displayed on a tabletop to encourage the use of allocentric processing (Abrahams good correspondence between the spatial knowledge of an environment acquired in the real world and a model et al., 1997, 1999). Related studies have looked at rota- tion of the array of locations, rotation of the subject���s of that environment in VR (Arthur et al., 1997 Regian and Yadrick, 1994 Ruddle et al., 1997 Witmer et al., view using computer generated presentations (Feigen- baum et al., 1996), or rotation and translation of blind- 1996), and VR has been used to enable an amnesic patient to learn useful routes (Brooks et al., 2000) and folded subjects (Morris et al., 1999 Worsley et al., 2001).