REVIEW THE NEUROSCIENCE OF REMOTE SPATIAL MEMORY: A TALE OF TWO CITIES

Transcription

1 Neuroscience 149 (2007) 7 27 REVIEW THE NEUROSCIENCE OF REMOTE SPATIAL MEMORY: A TALE OF TWO CITIES H. J. SPIERS* AND E. A. MAGUIRE* Wellcome Trust Centre for Neuroimaging, Institute of Neurology, University College London, 12 Queen Square, London WC1N 3BG, UK Abstract Most of our everyday activities take place in familiar environments learned in the past which we need to constantly navigate. Despite our obvious reliance on these remote spatial memories, until quite recently relatively little was known about how they are instantiated in the human brain. Here we will consider developments in the neuropsychological and neuroimaging domains where innovative methodologies and novel analysis techniques are providing new opportunities for exploring the brain dynamics underpinning the retrieval and use of remotely learned spatial information. These advances allow three key questions to be considered anew: What brain areas in humans support the retrieval and use of remotely learned spatial information? Where in the brain are spatial memories stored? Do findings relating to remote spatial memory inform theoretical debates about memory consolidation? In particular, the hippocampus, parahippocampus, retrosplenial and parietal cortices are scrutinized, revealing new insights into their specific contributions to representing spaces and places from the past IBRO. Published by Elsevier Ltd. All rights reserved. Key words: hippocampus, retrosplenial cortex, fmri, neuropsychology, amnesia, navigation. Contents The hippocampus 8 Three theories 9 The case of patient E.P. 9 Memories of Toronto 9 Lessons from London 13 Summary 17 Beyond the hippocampus 17 Parahippocampus 19 Retrosplenial cortex 19 Parietal cortex 20 Conclusions from remote spatial memory s anatomical snapshots 21 Brain dynamics underlying active navigation in London 21 From verbal reports to navigational thoughts 21 A navigational guidance system in the human brain 23 Conclusions 24 Acknowledgments 24 References 24 *Corresponding author. Tel: ; fax: address: h.spiers@fil.ion.ucl.ac.uk (H. J. Spiers), e.maguire@ fil.ion.ucl.ac.uk (E. A. Maguire). Abbreviations: fmri, functional magnetic resonance imaging; MTL, medial temporal lobe; MTT, multiple trace theory; VR, virtual reality /07$ IBRO. Published by Elsevier Ltd. All rights reserved. doi: /j.neuroscience A quainter corner than the corner where the Doctor lived, was not to be found in London. There was no way through it, and the front window of the Doctor s lodgings commanded a pleasant little vista of street that had a congenial air of retirement on it. There were few buildings then, north of the Oxford-road, and forest-trees flourished, and wild flowers grew, and the hawthorn blossomed, in the now vanished fields. As a consequence country airs circulated in Soho with vigorous freedom, instead of languishing into the parish like stray paupers without a settlement; and there was many a good south wall, not far off, on which the peaches ripened in their season.... It was a cool spot, staid but cheerful, a wonderful place for echoes, and a very harbor from the raging streets. A Tale of Two Cities, Charles Dickens (p 96, 1859). In A Tale of Two Cities Charles Dickens describes with typical aplomb his memory for how the area of Soho in London used to be in times past. But how did he, and indeed do we accomplish such feats of memory, recalling often in accurate detail places initially experienced perhaps decades ago? Here we will focus on this question, and in so doing not only will we too tell a tale of two cities, but also of three theories, and one neuro-anatomical journey. We will begin by asking why the neuroscience of remote spatial memory in humans deserves our interest. A good deal of our everyday activities in this world takes place in familiar environments learned in the past, which we need to constantly navigate or occasionally re-visit. This ability to recall and use remotely learned spatial information is therefore critical to mobility and survival, allowing food and shelter to be re-located, and dangerous places avoided. Spatial memories have arguably a privileged role in our mnemonic repertoire, since all things learned were at one time experienced in a particular place (O Keefe and Nadel, 1978; Tulving, 1983). Thus understanding how remote spatial memories are instantiated in the brain may provide important clues about how our episodic (autobiographical) memories for personally experienced events in general are represented. With attention still firmly focused on how the neural systems supporting memories might change as memories age (e.g. Frankland and Bontempi, 2005), the last number of years has seen increased interest in how remote spatial memories, largely ignored previously, might inform this debate. Finally, unlike autobiographical memory whose existence in non-human species is still contentious (Clayton et al., 2007), findings in relation to spatial memory can be much more directly

2 8 linked across species given our common need to remember locations and navigate in large-scale space. One reason why spatial memory has lagged behind other aspects of memory, such as autobiographical memory, in the consideration of memory remoteness is because the majority of spatial memory findings have been derived from studying memories acquired relatively recently (e.g. Maguire et al., 1998; Burgess et al., 2002; Spiers et al., 2001a; Hartley et al., 2003) and often in laboratory settings using stimuli such as tabletop arrays (e.g. Milner, 1972; Bohbot et al., 1998; Abrahams et al., 1999; Holdstock et al., 2000). While much has been learned from this approach about the neural substrates of encoding and short-term retrieval (see e.g. Burgess et al., 2002), the fate of truly remote spatial memories has remained largely unknown. The focus on recent memories is not surprising given a host of problems confronting those attempting to examine any type of remote memory (Moscovitch et al., 2006). Unlike, testing new learning or the retrieval of recent memories, control over encoding is not possible. Validating subjects reports of memories can be difficult, or impossible, and individuals can differ in the amount and type of experience with the materials tested. Additional problems pertain to studying remote spatial memory (see e.g. Aguirre and D Esposito, 1999). While there are standardized tests for probing remote semantic (fact) memory and episodic (autobiographical) memory (e.g. Crovitz et al., 1992; Kopelman, 1994; Levine et al., 2002), no such tests exist for remote spatial memory. Also a wide range of tests can be applied, each emphasizing different cognitive requirements. Arguably the most intuitive and indeed ecologically valid way to probe remote spatial memory is to concentrate on knowledge acquired of real-world places and environments learned in the past. But here too, tests must be tailored to the specific environment involved, and spatial memory in this context encompasses a wide range of processes and functions. For example it is possible to test subjects sense of direction, landmark recognition, ability to judge landmark proximities and distances, give route descriptions, and also to navigate. Assessing any one of these alone might lead to an erroneous impression of a general spatial memory problem, where only a selective deficit might exist. Of all the possible tests, the hardest to conduct is in situ navigation since a real environment is not typically under experimental control, and may change from day to day (e.g. a street is blocked), creating different problems for each subject tested. Furthermore, testing navigation is complicated by the different strategies that can be employed, such as the use of allocentric map-like representations (e.g. go 300 m to the north and then go west) or the use of egocentric landmark-actions associations (e.g. go to the clock tower, then turn left) (Siegel and White, 1975; Thorndyke and Hayes-Roth, 1982). Thus, it is important to consider how such tasks are being performed by subjects. Finally, generalizing across different environments can be complicated by differences in size, density of landmarks/streets, and the complexity of their layouts. While the issues surrounding remote memory in general have been discussed of late (e.g. Moscovitch et al., 2005, 2006; Cipolotti and Bird, 2006; Squire and Bayley, 2007) (including sub-sections on spatial memory), there is no recent review wholly dedicated to remote spatial memory. A more focused consideration of this topic is timely, however, given very recent advances in both neuropsychological and neuroimaging domains. These have been facilitated by novel methodologies and analysis techniques which present new opportunities for overcoming some of the aforementioned difficulties, paving the way for possible conceptual advances. Here we will focus on real environments, using spatial memory to denote the memory for landmarks, their locations and spatial relations, and the ultimate expression of this spatial knowledge in the ability to navigate in large-scale space. Remoteness will be represented in two ways, with environments being learned initially in the distant past (decades ago) and then navigated subsequently and regularly for many years, or environments learned and navigated repeatedly in the distant past that were then vacated and only revisited many years later. The former is more common in the neuropsychological studies so far conducted. Given our interests in real large-scale space and long timescales as the means of probing remote spatial memory, by necessity the focus will be on humans (neuropsychology and neuroimaging), although we will also draw on some relevant non-human work. The questions that will percolate through our discussion are simple ones, although answers have been much more difficult to find. What brain areas in humans support the retrieval and use of remotely learned spatial information? Where in the brain are spatial memories stored? Do findings relating to remote spatial memory inform theoretical debates about memory consolidation? In addressing these questions we will journey between two cities, London (UK) and Toronto (Canada), each with its own tale to tell about remote spatial memory. We will also embark on an anatomical journey, starting at hub of the memory system, the hippocampus, then taking a posterior superior route through the parahippocampus, retrosplenial cortex to the parietal cortex. En route, the evidence for the contribution of each of these regions to remote spatial memory is evaluated, anatomical snapshots so to speak. We will then seek to understand how these and related brain regions operate together, dynamically, on a second-by-second basis, to meet the demands of navigating through the urban sprawl of a long-familiar city (Spiers and Maguire, 2006a,b, 2007a,b). THE HIPPOCAMPUS The hippocampus has occupied a central role in neurobiological theories of memory since the discovery by Scoville and Milner (1957) that bilateral removal of the medial temporal lobes (MTL) caused dense amnesia. The importance of the hippocampus in particular for long-term memory was later confirmed by a number of studies reporting amnesia following relatively selective bilateral hippocam-

3 9 pal damage (e.g. Zola-Morgan et al., 1986; Rempel- Clower et al., 1996; see Spiers et al., 2001b for a review). Anatomical definitions of the terms hippocampus can vary. Here, we define the hippocampus as the hippocampus proper (the CA fields) and the dentate gyrus, hippocampal formation as the hippocampus and the subiculum, and MTL as the hippocampal formation plus the entorhinal, perirhinal and parahippocampal cortices (Amaral, 1999). The latter collection of structures has also been referred to as the hippocampal complex (Nadel and Moscovitch, 1997; Moscovitch et al., 2005, 2006). THREE THEORIES Three theories dominate contemporary debates on the role of the hippocampus in remote spatial memory. These are the cognitive map theory (O Keefe and Nadel, 1978), standard consolidation theory (Marr, 1971; Squire, 1992; Alvarez and Squire, 1994; Frankland and Bontempi, 2005) and multiple trace theory (MTT) (Nadel and Moscovitch, 1997; Moscovitch et al., 2005, 2006). All theories agree that the initial acquisition of long-term spatial memories requires the hippocampus, but disagree about its role in remote spatial memory. Cognitive map theory postulates that the hippocampus stores an allocentric spatial representation of the environment which supports flexible navigation, such as the ability to take detours or shortcuts. The original evidence for this came from the discovery of hippocampal place cells in rodents which signal the current allocentric location of the animal in space (O Keefe and Dostrovsky, 1971). Cognitive map theory does not formally distinguish between cognitive maps learned recently from those learned long ago, and so it would seem to predict that spatial memory for and navigation in environments learned recently or remotely require the hippocampus. The standard consolidation model (Marr, 1971; Squire, 1992; Alvarez and Squire, 1994; Frankland and Bontempi, 2005) contends that the hippocampus is needed for the storage and recovery of recent memories (both spatial and nonspatial), initially linking or indexing different sensory representations stored in neocortical regions. As time passes, memory representations become consolidated at a systems-level such that traces stored in neocortical regions become strengthened and ultimately independent of the hippocampus. MTT, as currently formulated (Moscovitch et al., 2005, 2006), posits that the hippocampus is always needed to re-experience or recollect the past in vivid detail, including spatial information. However, remembering semantic or coarse information (including spatial information) only requires the hippocampus for a limited period, after which it is supported solely by neocortex. MTT therefore maintains that the hippocampus is necessary to richly re-experience or retrieve detailed (episodic) information about remotely learned environments, but not for the retrieval of coarse or semanticized information about such environments. Recently these theories have been tested by examining the remote spatial memory of patients with damage which included the hippocampi. To summarize, for such patients cognitive map theory predicts their remote spatial memories will be impaired, standard consolidation theory predicts they will be spared, and MTT predicts that only their ability to remember detailed spatial information or richly re-experience the environment will be impaired, while general semantic-like spatial information will still be intact. THE CASE OF PATIENT E.P. Patient E.P. became amnesic following extensive damage to his medial and anterior temporal lobes as a result of herpes simplex encephalitis (Teng and Squire, 1999) (see Fig. 1). He lived in an area of California for 22 years in his early life before moving away, after which he returned only occasionally (see Fig. 2A). He was tested more than 50 years later on his ability to remember aspects of this remotely learned environment. The tests involved describing routes between home and places in the area, describing routes between different locations (not home), between these locations if a main street was blocked, and imagining being in a particular orientation at certain locations and having to point toward specific landmarks. Despite being profoundly amnesic and unable to learn new environments, E.P. s performance on all tests was comparable with five control subjects who had also moved away from the area in a similar timeframe. These results were taken as evidence against the cognitive map theory and in favor of the standard consolidation theory, the conclusion being that the hippocampus and MTL do not appear to be the repository of remote spatial memories (Alvarez and Squire, 1994; Bayley et al., 2003, 2005; Squire and Bayley, 2007). The relevance of the results to MTT is less clear. It has been argued that E.P. may have been able to perform the tests on the basis of well-rehearsed spatial layouts, akin to semantic memories, which were devoid of the rich topographical details evoked during vivid recollection, and which were not tapped by the testing employed (Rosenbaum et al., 2000; see also Moscovitch et al., 2005, 2006). Therefore the results do not necessarily conflict with MTT. A more detailed investigation of this issue was conducted in Toronto where our story next leads. MEMORIES OF TORONTO Rosenbaum et al. (2000, 2005a) investigated patient K.C., who lived in a small Toronto neighborhood for 40 years (Fig. 2B). K.C. became profoundly amnesic following a closed head injury that caused widespread damage which included the hippocampi bilaterally, parahippocampal cortices, and infarction to the medial occipital region (Rosenbaum et al., 2000; see Fig. 1). Compared with four control subjects, K.C. was unimpaired on a range of spatial memory tests. Specifically he was able to describe routes between places when the most direct route was blocked, indicate directions and distances between landmarks, make proximity judgments between locations, and order landmarks in the sequence they would be passed if one navigated through the area. A similar pattern of performance was reported in another patient, S.B., who had

4 10 worked as a taxi driver in downtown Toronto (see Fig. 2C) for 40 years until he was forced to retire due to probable Alzheimer s disease and extensive temporal and occipital lobe damage (Rosenbaum et al., 2005b; see Fig. 1). Thus, findings from these cases provide further evidence to reject the cognitive map theory in that the hippocampus does not appear to be crucial for the maintenance or retrieval of all remote allocentric spatial representations. However, K.C. produced a small number of landmarks on a sketch map of his neighborhood, demonstrated impoverished geographical knowledge, and was impaired on his recognition of incidental neighborhood landmarks, indicating that his remote memory may not be entirely normal for detailed spatial information. Thus, the authors argued that the data diverge from that predicted by cognitive map or standard consolidation theory and support instead the prediction from MTT that only memory for detailed spatial information is disrupted by hippocampal damage (Rosenbaum et al., 2000; Moscovitch et al., 2005, 2006). To explore this issue further, Rosenbaum et al. (2004) used functional magnetic resonance imaging (fmri) to examine whether or not the hippocampus would be engaged when remote spatial memories of downtown Toronto were retrieved. Healthy Toronto residents with extensive experience of navigation in the city for between five and 10 years were scanned as they viewed the names of landmarks and performed tasks requiring the mental navigation of blocked routes, landmark proximity judgments, and landmark sequencing. The recognition of Toronto landmarks from photos was also examined. No increased hippocampal activity was found during any of the tasks, instead a number of other brain regions were found to be more active (discussed in more detail later, see also Fig. 3). Toronto s human inhabitants are not alone in contributing to our understanding of remote spatial memory. In a Toronto research center, there is a small village maze with several interconnecting compartments designed to test the remote spatial memory of rats (Winocur et al., 2005). One group of rats was reared in this village over several months, while another group was not. Both groups were then trained to navigate in the maze and subsequently given hippocampal lesions. Accurate navigation in the environment was found only in the rats reared there, suggesting that after extensive experience of the environment, regions other than the hippocampus come to support navigation. Another recent rodent study also suggests that regions other than the hippocampus are capable of rapidly assimilating new information about an environment when a pre-existing schema of associations has been learned through extensive experience (Tse et al., 2007). Thus, overall the evidence from neuropsychological, neuroimaging and rodent lesions studies appears to suggest that memory for environments learned long ago does not necessarily engage or require the hippocampus. Fig. 1. MRI scans of the brains of patients E.P., K.C., S.B. and T.T. Top panel shows a T2-weighted axial section of E.P. s brain, showing atrophy in anterior MTL (Teng and Squire, 1999). Panels below show T1-weighted coronal sections through the brains of K.C. (Rosenbaum et al., 2000), S.B. (Rosenbaum et al., 2005b) and T.T. (Maguire et al., 2006a), white arrows indicate the atrophic hippocampi and white arrowheads (K.C. scan) indicate atrophic parahippocampal gyri. Reproduced from the original sources with permission.

5 11 Fig. 2. Maps from the environments used to test remote spatial memory. (A) A map of patient E.P. s neighborhood in Castro-Haywood valley, California (adapted from Teng and Squire, 1999; reproduced with permission). (B) A map of patient K.C. s neighborhood in Toronto (adapted from Rosenbaum et al., 2005a; reproduced with permission). (C) A map of downtown Toronto (Canada) reproduced by permission of City Maps Inc (D) A map of central London (UK) reproduced by permission of Geographers A-Z Map Co. Ltd. Crown Copyright All rights reserved. License number Main A roads are shown in orange and yellow, non-a roads in white.

6 12 Fig. 3. Brain regions associated with remote spatial memories of Toronto. The brain regions activated in the comparison of all remote spatial memory tasks with the perceptual baseline task in the study by Rosenbaum et al. (2004). (A) A map showing the region of downtown Toronto from which tests of spatial memory were constructed. (B D) The functional maps are overlaid on the averaged anatomical scans from all participants in relevant sagittal and axial views. The right hemisphere is shown on the left side of the images. Areas of activity common across tasks included right parahippocampal gyrus (B), left retrosplenial cortex (C) and posterior right parietal/superior occipital cortex (D). The hippocampus (open circle in B) was not activated during any task. Images adapted from Rosenbaum et al. (2004) and Moscovitch et al. (2005) and reproduced with permission.

7 13 There are, however, several issues that arise from these studies which may affect this conclusion. Firstly, with regard to patient findings, there is an obvious tension between the reports of patients K.C. and S.B. on the one hand, and that of patient E.P. on the other. All are apparently unimpaired at navigation, yet in one case (E.P.) the conclusion is that the hippocampus is not needed for remote spatial memories, while on the basis of the other cases (K.C., S.B.) it is deduced that the hippocampus is implicated in some forms of remote spatial memory. Although the evidence, particularly in K.C., is suggestive of a possible dissociation between representations that are coarse/semantic and those that allow rich re-experiencing, with the hippocampus necessary for the latter, the data are not conclusive. Secondly, on the basis of the three previous cases, the cognitive map theory is rejected, in that all patients seemed able to access allocentric spatial information, and performed well on tasks requiring flexible navigation, such as detours, which are prime indicators of hippocampal-dependent cognitive mapping according to this theory. However, it has been noted that memory for the environments learned remotely and which were wellpracticed may have been transformed in the process such that the representations no longer code for detailed information, being supported instead by extra-hippocampal areas (Rosenbaum et al., 2000). The same argument has been proposed to explain the spared performance of the hippocampal rats reared in the village maze (Winocur et al., 2005). This transformation might be particularly true of environments that have a relatively simple and predictable, regular (grid-like) layout (see Fig. 2A C). The neighborhoods of all of the previous patients, E.P., K.C., and S.B., and that used in the fmri study of Rosenbaum et al. (2004) (Fig. 3) were of this type, as was the rat village maze in the study of Winocur et al. (2005). Finally, while it is reported that K.C. navigated normally in his neighborhood (Moscovitch et al., 2005), the third potential concern with previous patient and neuroimaging studies is the dearth of in-depth studies and systematic data relating to in situ navigation ability. While static, or off-line, tests can permit examination of allocentric spatial processing, and are suited for use in memory-impaired patients, additional important information might be gleaned by examining dynamic navigation in the complex real world where it typically takes place. To address these issues it is necessary to leave Toronto and travel to another city, London. LESSONS FROM LONDON Like Toronto, the city of London (UK) has functioned as a test bed for exploring the neural basis of remote spatial memory. Unlike Toronto and the Californian valley of patient E.P., London has a high number and density of roads in a very unpredictable and irregular layout, in addition to numerous complex one-way systems (see Fig. 2D). This makes it possible to examine navigation and spatial memory in a remotely-learned complex environment where there is an indisputable call on allocentric information. London also has a highly proficient group of expert navigators whose knowledge of the city can be known with accuracy, licensed London taxi drivers. In the UK, licensed London taxi drivers undergo extensive training over a period of 2 4 years known as acquiring The Knowledge. This involves learning the layout of 25,000 streets in the city, thousands of places of interest, leading to a stringent set of examinations by the Public Carriage Office in order to obtain an operating license. In addition, a highly accurate and interactive virtual reality (VR) simulation of central London, UK, is available, developed as the backdrop for commercial video game, enabling in situ navigation to be assessed in a controlled manner. The Getaway ( Sony Computer Entertainment Europe) has over 110 km (70 miles) of London s drivable roads, accurately recreated from Ordinance Survey map data, covering 50 km 2 (20 square miles) of the city center. The one-way systems, working traffic lights, the busy London traffic, and an abundance of Londoners going about their business are all included (see Fig. 4). Conveniently, one can simply navigate freely (with the usual game scenarios suspended) around the city using the game console, with a normal ground-level first person perspective, in a car of one s choice. Despite some aspects of navigating in the game being different from the real world (e.g. the lack of vestibular and proprioceptive sensation), taxi drivers confirm that it is very reminiscent of their experience of navigating in central London. In addition to these advantages, London is also the residence of patient T.T. He became profoundly amnesic following a rare form of limbic encephalitis that left him with damage primarily involving the hippocampi bilaterally, with other areas of the MTL apparently intact (Fig. 1). Crucially, prior to his illness T.T. had worked for nearly 40 years as a licensed London taxi driver. Thus it was possible to know that he initially learned London s layout nearly four decades ago, and had navigated continuously in London since that time. In addition, the standard of his premorbid navigation ability was well characterized, and enabled comparison with very appropriate control subjects, namely his similarly-qualified and experienced fellow taxi drivers (Maguire et al., 2006a). T.T. showed remarkably spared performance on static tests similar to cases E.P., K.C. and S.B. His was able to recognize London landmarks, make proximity judgments and distance estimates normally, he performed well when required to take a survey perspective able to locate places on a map of London, and his direction-pointing too was entirely spared. However, unlike the other patients, T.T. was significantly impaired at interactive in situ navigation in (virtual) London (see Fig. 5), taking on average significantly longer routes than the control taxi drivers, and often not able to reach some destinations at all (Maguire et al., 2006a). What might be the reason for the distinction between T.T. and the other patients? It may be that London s high number and density of roads, unpredictable irregular layout, or complex and numerous one-way systems require functioning hippocampi, whereas for the environments of the other patients, particularly when highly familiar, the hippocampal role is minimized (see Fig. 2). Evidence from

8 14 Fig. 4. Virtual London (UK). Example views from within the simulation of central London used to test in situ navigation. (A) A view at Trafalgar Square, (B) a view at Piccadilly Circus, (C) a view looking toward the London Eye/Millennium Wheel. These images from The Getaway are reproduced with the kind permission of Sony Computer Entertainment Europe. several fmri studies lends support to the idea that the nature of the large-scale space may have an impact on the hippocampal involvement in navigation. As discussed above, no increased hippocampal activity was observed during the mental navigation in Toronto (Rosenbaum et al., 2004). By contrast two recent fmri studies examining navigation in London have found evidence of increased hippocampal activity (Kumaran and Maguire, 2005; Spiers and Maguire, 2006a). Spiers and Maguire (2006a) had long-time residents navigate around the same VR London used with T.T., and found significantly increased activity in the hippocampus when subjects planned routes between start and destination points (see Fig. 6). In another fmri study, the hippocampus was also significantly active when subjects, who had lived in London for on average 16 years, engaged in mental navigation (Kumaran and Maguire, 2005). Thus the combined neuroimaging and neuropsychological evidence indicates that whereas navigating in London appears to involve the hippocampus, navigating in Toronto does not. The question that naturally arises is what aspect of London, or potentially any complex large-scale space, is the hippocampus necessary for supporting? While overall T.T. s navigation performance was significantly worse than control subjects, this was not an all-or-nothing result. For some routes he was error free and others grossly impaired (see Fig. 5). Perhaps the VR task, with trials over several minutes, was adversely affected by T.T. s anterograde memory deficit. However, the navigation routes which elicited impairment in TT did not take any more time than those where he was error free, did not involve more turns, or more decisions. At no point did T.T. forget what he was doing or get distracted. Thus, on the face of it, the reason for T.T. s pattern of performance was not obvious. This matter was investigated further by conducting an in-depth analysis of the route characteristics. This made it possible to rule out a number of potential influences, such as physical features of routes (e.g. length, width, number of junctions and decision points, presence of major landmarks), factors related to T.T. (e.g. memory for destinations, distribution of errors across a route), and to controls (e.g. T.T. s error-filled routes were not more difficult for them than his error-free ones, they made errors where TT did not and vice versa). In fact, of 27 variables, only one was significant in relation to T.T. s pattern of performance. If a route contained a high number of minor (non- A ) roads, then TT was impaired (see Fig. 5). Thus, he was able to navigate using main arterial or A roads (roads colored in orange/yellow on the London map in Fig. 2D), but was unable to successfully negotiate the myriad of others that comprise the complex road matrix covering the city. What is it about non-a roads that makes them hippocampal-dependent? Licensed London taxi drivers do not learn the layout of London by first learning the main artery roads, nor is this distinction made or highlighted as part of the testing. Thus, at an explicit level, this is not the framework around which their spatial representation of London is built. Perhaps non-a roads are used less frequently? When questioned about this the taxi drivers had trouble dealing with the concept of frequency of use, as they had spent 40 years navigating in London, and everything was highly familiar, both A and non-a roads alike. T.T. s performance on one particular route offers a further clue in this

9 15 regard. There were no A roads on a route through Soho (Route 3 Fig. 5) yet T.T. performed normally. The Soho area of London has many small streets with a complex and dense layout, however some of the streets are very well known and we speculate that they might have acquired A road status. This would offer support for the idea that T.T. is able to navigate using main or frequently used roads. London taxi drivers have a saying, If in doubt, follow the yellow-brick road. By this they mean if you are not sure about your route, stick to the main roads. Thus, while impossible to get a measure of frequency of road use in long-experienced London taxi drivers, the A roads may have been experienced more and, over time, acquired a more semantic-like status, becoming independent of the hippocampus. London s non-a roads are densely packed together, and may be more visually similar to each other than the A roads. Gilbert et al. (1998, 2001) found that hippocampal lesions in rats impaired the ability to remember locations with increased spatial proximity, but memory for locations further apart was preserved. This may also explain why hippocampal lesions in rodents have consistently been found to impair remote spatial memories in the watermaze; a task which requires constant updating of precise spatial positions from distal landmarks for successful performance (Clark et al., 2005; Martin et al., 2005). In a similar vein, Moscovitch et al. (2005, 2006) suggested that a schematic or coarse representation of topography can exist independently of the hippocampus. Thus, a loss of, or inability to access, fine-grained as opposed to coarse spatial representations may be another possible explanation for T.T. s deficit. The case of T.T. is unique in terms of the location of the lesions centered on the hippocampi, combined with the history of taxi driving, and the novel means of testing and analyzing navigation performance. Not only that, findings from T.T. also speak directly to the three theories of hippocampal function concerning remote memory. That T.T. could not navigate normally around (virtual) London accords with the cognitive map theory (O Keefe and Nadel, 1978). However, T.T. s exquisitely preserved performance on static tasks, some of which involve allocentric information, is problematic for this view (see also Moscovitch et al., 2006). In addition, it is not clear that a distinction between A and non-a roads can be made in terms of the allocentric information they embody or flexibility of the navigation they permit. Thus, while the cognitive map theory remains viable in its treatment of remote spatial memory, in its original form it cannot account completely for all of the findings from T.T. Alternatively, the standard theory of consolidation might explain the data if the routes T.T. was impaired at navigating were learned more recently and hence still reliant on hippocampal function. This was not the case; spatial memories for all routes were acquired in the remote past, and the A roads were not learned before the non-a roads during training. In fact, T.T. s performance is most consistent with MTT, since T.T. could navigate in London to some degree, but with a deficit in the finer detail of his spatial representation. In this view his preserved coarse representation of the main artery roads may have, over time and with frequency of use, become semanticized, and so insulated from hippocampal damage. His impoverished representation or inability to access the fine details of London s layout may have prevented his rich re-experiencing of the city, thus compromising his navigation when it depended on non-a roads. T.T. s deficit in navigation clearly shows that the hippocampus is necessary for some aspects of remote spatial memory. Does this mean that the detailed representation of London s layout is stored in the hippocampus, becoming lost to T.T. once his hippocampal tissue was destroyed? As concluded above, the findings suggest either a loss of the fine detail stored in the hippocampus, or an inability to access or act upon fine detail that might be stored in the hippocampal remnants or elsewhere. A definitive distinction between these possibilities cannot be made on the basis of T.T. alone. One further strand of London-based evidence may provide further clues. The volume of the hippocampus in non-humans is known to vary as a function of the demands placed on spatial memory (Barnea and Nottebohm, 1994; Smulders et al., 1995; Volman et al., 1997; Lee et al., 1998; Biegler et al., 2001). Similar effects have been found in humans, with licensed London taxi drivers having greater gray matter volume in their posterior hippocampus compared with non-taxi driver control subjects (Maguire et al., 2000), and also when compared with London bus drivers (Maguire et al., 2006b) (who navigate the city along a small number of fixed routes). Furthermore, the number of years navigating in London correlated with posterior hippocampal gray matter volume in the taxi drivers, with greater volume associated with more navigation experience. These structural MRI findings suggest that learning, representing and using a spatial representation of a highly complex and large-scale environment is a primary function of the hippocampus in humans such that this brain region might adapt structurally to accommodate its elaboration. It might seem somewhat paradoxical that the gray matter in posterior hippocampus increases with more navigation experience, since it might have been predicted instead that taxi drivers would have most knowledge when they are newly qualified (Terrazas and McNaughton, 2000). Taxi drivers, however, report that their mental map of London becomes more integrated and coheres over a long period as it gets more finetuned through experience of the relationships among roads and between places, You have to take in a lot of information in this game, not just at the beginning, but over the years (Dave, licensed London taxi driver of 10 years). Although further longitudinal studies are required to examine the within-subject effects of taxi driver training, an intuitively appealing conclusion from the data so far is that the posterior hippocampus stores the detailed spatial representation of London, and the more detailed this is, the larger the posterior hippocampal volume.

10 16 Fig. 5. Patient T.T. s navigation performance around London. (A) An example of one of the routes taken by patient T.T. (black line on the map) during the navigation of virtual London. The route tested was from St. Paul s Cathedral to the Bank of England. The red line shows the optimal route, taken by all 10 control subjects, also London taxi drivers, matched in age and experience to T.T. and who retired at around the time T.T. became ill. Also of note, the game was developed approximately 4 years prior to use with T.T. Thus, the London captured in the game is London as it was

11 17 SUMMARY Human neuropsychological, neuroimaging, and rodent lesion studies have examined the role of the hippocampus in the memory for very familiar remotely learned environments. The two cities of London and Toronto have been important contributors in several key studies. Results from the patient and lesions studies show that a remarkable amount of information can be retained following hippocampal and even extensive MTL damage. However, the findings overall suggest that the hippocampus in humans is necessary for facilitating navigation in places learned long ago when complex large-scale spaces are concerned and successful navigation requires access to detailed spatial representations. BEYOND THE HIPPOCAMPUS In our anatomical journey we now leave the hippocampus to consider the following question, if the hippocampus is not needed for all aspects of remote spatial memory, then what other brain areas are implicated? Both standard consolidation theory and MTT propose that neocortical regions support long-term memories, with MTT arguing these regions support semantic, non-detailed, spatial memories. It should be noted, however, that in general the rigorous neuropsychological testing of remote spatial memory as applied to T.T. and K.C. has generally yet to be directed at patients with well-characterized lesions outside the hippocampus. Similarly, for electrophysiological recordings in non-humans, while shedding light on the computations that might be performed by cells in extra-hippocampal regions, there is a dearth of studies examining responses in very familiar environments like the village maze used by Winocur et al. (2005). As with the hippocampus, a central question concerning cortical areas is whether or not they are a repository of allocentric spatial maps, or do they provide additional functions that are necessary for retrieval or manipulation of the memories? For instance, lesions to lingual gyrus have reliably been found to cause disorientation in familiar environments (Hecaen et al., 1980; Landis et al., 1986; Funakawa et al., 1994; McCarthy et al., 1996; Suzuki et al., 1996). However these deficits can, in most cases, be attributed to an inability to recognize or process visual landmarks, a problem termed landmark agnosia (Aguirre and D Esposito, 1999), thus this region is not implicated in the retrieval of remotely learned environments. Similarly, some brain regions, such as the caudate nucleus, may support the navigation of frequently used routes in familiar environments by providing a sequence of stimulus response associations (e.g. at the next junctions, turn left, right, left) (Packard and McGaugh, 1996; Hartley et al., 2003; Iaria et al., 2003; Voermans et al., 2004). While these regions may be responsible for efficient navigation on some routes in familiar environments, they are unlikely to support retrieval of the layout of remotely learned environments, or the navigation of novel or blocked routes within them. The temporal and prefrontal cortices have also been implicated in the retrieval of remote spatial memories (see e.g. Frankland and Bontempi, 2005; Squire and Bayley, 2007). Current disagreement exists, however, as to the role of the prefrontal cortex in this context. Studies of activity-dependent gene expression in mice have found increasing activity in the prefrontal cortex for remote memories (Bontempi et al., 1999; Maviel et al., 2004), and the disrupted retrieval of remote spatial memories following inactivation of the prefrontal cortex (Maviel et al., 2004). This has been interpreted as providing evidence that, with time, the prefrontal cortex becomes the main storage site for the remote spatial memories (Maviel et al., 2004). However, others (Rudy et al., 2005) have provided a plausible alternative account in which the prefrontal cortex is involved in remote memories because these memories have a weaker strength than recent memories and require more top-down monitoring at retrieval, in line with other theories of prefrontal function (e.g. Miller and Cohen, 2001). Notably, prefrontal activity has not been a consistent feature in fmri studies of remote spatial memory in either London or Toronto (Maguire et al., 1997; Rosenbaum et al., 2004; Kumaran and Maguire, 2005), and where it has been detected it can generally be related to monitoring or planning processes (Rosenbaum et al., 2004; Spiers and Maguire, 2006a, 2007b). It has also been suggested that the temporal cortex may store remote memories based on similar activity-dependent gene activation studies (Maviel et al., 2004). Regarding the left temporal cortex in humans, Rosenbaum et al. (2005b) have shown that while damage to this region can impair the naming of familiar Toronto landmarks and locations, it has little or no effect on the ability to perform remote spatial memory tests. Moreover, anterior lateral regions of the left and right temporal lobe have not been consistently activated in neuroimaging studies of remote spatial memory in either of the two cities (Maguire et al., 1997; Rosenbaum et al., 2004; Kumaran and Maguire, 2005; Spiers and Maguire, 2006a). If there is equivocal evidence for involvement of the above brain areas in remote spatial memory, what about other areas that have been strongly implicated in spatial learning and memory over shorter timescales? We now critically assess the contribution of the parahippocampus, retrosplenial and parietal cortices to remote spatial memory. approximately 2 years before T.T. became ill. This is important because it allowed T.T. to be tested in the London that he experienced, and without changes to buildings or layout that may have occurred since (or in the 2 years before) he stopped taxi-driving. The map of central London (UK) is reproduced by permission of Geographers A-Z Map Co. Ltd. Crown Copyright All rights reserved. License number (B) A graph showing T.T. s navigation performance around (virtual) London for each of the 13 routes. Note: routes were not performed in the order shown but are presented in this way to better illustrate the pattern of performance. (C) Shows the corresponding percentage of non-a roads for each route. (D) Indicates whether routes could be navigated using only A roads. See text for further details of Route 3 (Soho). Images adapted from Maguire et al. (2006a) and reproduced with permission.

12 18 Fig. 6. Brain activity time-locked to spontaneous thoughts during navigation in London. Subjects navigated to destinations in a VR simulation of central London, UK, and were subsequently interviewed with a verbal report protocol (Spiers and Maguire, 2006a). (A) Shows a section of the region of London simulated with an illustrative route overlaid. Top left shows an image from the environment at Trafalgar Square. The map of central London (UK) reproduced by permission of Geographers A-Z Map Co. Ltd. Crown Copyright All rights reserved. License number (B) The

13 19 PARAHIPPOCAMPUS The parahippocampus wraps around the lateral and inferior edges of the hippocampus. Both neuropsychological and neuroimaging studies have consistently found evidence that the parahippocampal gyrus is involved in acquiring spatial memories (Ross, 1980; Habib and Sirigu, 1987; Barrash et al., 2000; Epstein et al., 2001; Aguirre et al., 1996). In particular, many neuroimaging studies have observed activation of a region of posterior parahippocampal cortex that responds more to scenes than other visual materials, this area being commonly called the parahippocampal place area (PPA) (Epstein and Kanwisher, 1998; Epstein et al., 2001). In several models of spatial memory retrieval, the parahippocampal cortex acts to represent the allocentric spatial relationships between landmarks (Moscovitch et al., 2005, 2006) or objects (Burgess et al., 2001; Byrne et al., 2007). These representations are then passed to the hippocampus for memory indexing or binding (Moscovitch et al., 2006), or are combined to form the allocentric spatial representations seen in hippocampal place cells (Burgess et al., 2001; Byrne et al., 2007). In contrast to new learning, the parahippocampal contributions to remote spatial memory are far from clear, with some authors arguing it may be vital (Rosenbaum et al., 2000; Moscovitch et al., 2005) and others that it is not required (Aguirre and D Esposito, 1999; Teng and Squire, 1999; Epstein et al., 2001). The posterior portions of the parahippocampal cortex in patients E.P., K.C. and T.T. were relatively spared, and it might be concluded that this region could be the repository for allocentric cognitive maps. However, the deficits observed in T.T. speak against the parahippocampus as being sufficient to support navigation in a remotely learned large-scale space. Neuroimaging studies examining the retrieval of remote spatial memories in London (Maguire et al., 1997; Kumaran and Maguire, 2005; Spiers and Maguire, 2006a), Toronto (Rosenbaum et al., 2000), and Liverpool in the UK (Mayes et al., 2004), have all reported parahippocampal activations. In the case of Toronto, activations were independent of whether the task required landmark sequencing, proximity judgments or mental navigation. In London when well-known routes are recalled or planed from start to end the parahippocampus was consistently activated (Maguire et al., 1997; Kumaran and Maguire, 2005; Spiers and Maguire, 2006a). Thus, unlike the hippocampus, the parahippocampus does not appear to be strongly affected by the complexity/street density of the environment or by the particular remote spatial memory task demands. Neuropsychological studies of patients with lesions including the parahippocampus have so far generally concluded that it is not necessary for remembering remotely learned or familiar environments (Ross, 1980; Habib and Sirigu, 1987; Barrash et al., 2000; Epstein et al., 2001). However, only very limited testing via sketch maps, verbal reports and landmark recognition was performed. Furthermore, the environments tended to be old apartments, rather than cities or neighborhoods. Notably, the sketch map of a remotely learned apartment drawn by a patient tested by Epstein et al. (2001) had less information in it that the drawing by his wife. Thus, it remains an open question whether or not the parahippocampus is in any way necessary for the remote spatial memory of largescale environments. RETROSPLENIAL CORTEX Traveling in a posterior direction from the parahippocampus, to where the cortex curves in an upward arc behind the splenium of the corpus callosum, we enter the territory of the retrosplenial cortex. This region has its own distinct cytoarchitecture distinguishing it from the nearby posterior cingulate cortex, and has a strong connectivity with other regions in our anatomical journey (Vogt et al., 2001). Damage to this region can cause a severe amnesia, similar that resulting from MTL lesions (Bowers et al., 1988; Heilman et al., 1990), or a relatively selective disorientation, the latter typically after right lateralized damage (see e.g. Takahashi et al., 1997; and Maguire, 2001 for a review). In contrast to parahippocampal damage, disruption of the retrosplenial cortex in humans reliably results in disorientation deficits for both new and previously familiar environments. Retrosplenial lesions in rodents can impair spatial navigation based on allocentric cues, and do so reliably when the lesion completely encompasses the region (Vann and Aggleton, 2002, 2004). As far as we are aware the effect of such lesions on rodents remote spatial memory has not been examined. Concordant with lesion work, neuroimaging studies have consistently reported robust retrosplenial activity (often of a bilateral nature) during navigation in environments that are relatively novel (Maguire et al., 1998 Gron et al., 2000; Mellet et al., 2000; Hartley et al., 2003; Iaria et al., 2003) or familiar (Maguire et al., 1997; Rosenbaum et al., 2004; Kumaran and Maguire, 2005; Spiers and Maguire, 2006a). Considering remote spatial memories in particular, Rosenbaum et al. (2004) found that whereas the parahippocampal gyrus was equally activated by all Toronto tasks, the retrosplenial cortex route presented in (A) is shown classified into a sequence of events and epochs from the different categories of thought extracted from an example subject s verbal report. Below is shown the main thought categories with a color key. (C) A schematic example of an fmri time-series after classification of the verbal report, with the time-series segregated into different events and epochs corresponding to the different categories present. Each subject s time-series was unique. (D) Examples of findings where activity during a particular thought category was compared with matched events/epochs during coasting, the baseline condition where subjects reported navigating, but with no directed thoughts. Red box (top): increased activity in the hippocampus during initial route planning. Blue box: increased activity in retrosplenial and anterior prefrontal cortices during spontaneous route planning. Yellow box: increased activity in the retrosplenial cortex and posterior parietal cortex when expected landmarks were spotted. Green box: increased activity in a region stretching from the parahippocampal cortex to occipital cortex and also the ventrolateral prefrontal cortex during the visual inspection of the environment. The images are adapted from Spiers and Maguire (2006a; reproduced with permission), where the results from other thought categories can be found along with a movie of footage in the environment combined with a dynamic presentation of the fmri data.

14 20 was most active when subjects made proximity or distance judgements about Toronto landmarks (see Fig. 3). In addition, Spiers and Maguire (2006a) found that while both the retrosplenial cortex and parahippocampus were involved in planning routes through London, the retrosplenial cortex was also more active when landmarks one expected to see en route were subsequently observed (see Fig. 6). The specific deficits of patients with retrosplenial lesions may offer further clues as to its possible function. A typical description of retrosplenial case comes from Takahashi et al. (1997): he was driving his taxi in the same city (in which had worked for 6 years), he suddenly lost his understanding of the route to the destination. As he could quickly recognize the buildings and landscapes around him, he was able to determine his current location. However, he could not determine in which direction he should proceed he tried to return to the main office but did not know the appropriate direction to drive. The small amount of formal testing conducted with these patients has revealed an inability to draw maps of familiar environments, describe routes through them and indicate the direction between familiar landmarks (see e.g. Takahashi et al.,1997). These findings, combined with evidence that rodent retrosplenial cells can code current head direction (Chen et al., 1994), led to the suggestion that the retrosplenial cortex provides information about allocentric heading direction (Takahashi et al., 1997; Aguirre and D Esposito, 1999). Thus in the model by Moscovitch and colleagues (2006), it serves as a long-term store of the allocentric directions between landmarks, complementing a parahippocampal store of allocentric locations. Many patients recover from their post-lesional disorientation after about 8 weeks (Aguirre and D Esposito, 1999; Maguire, 2001) indicating that retrosplenial function can sometimes be compensated for with time, particularly if the lesion is right lateralized (Maguire, 2001). It may be that the long-term store of allocentric headings is initially disrupted in such patients, but residual maps residing in other regions, including possibly the left retrosplenial cortex (Maguire, 2001), come to support this function. However, the patients recovery has also been interpreted as evidence that the allocentric map, including directional information, lies elsewhere (possibly the MTL, see above) and that the retrosplenial cortex provides a manipulation or transformation function on the stored maps (Burgess et al., 2001). Hence, an alternative proposal is that the retrosplenial cortex uses its information about current heading direction to translate long-term allocentric MTL representations into egocentric parietal cortical representations, required to reconstruct a scene or imagine new perspective (Byrne et al., 2007; Hassabis and Maguire, 2007). In summary, current evidence suggests the retrosplenial cortex is important for remote spatial memory, possibly with a specific role in processing allocentric direction information between landmarks. However, it is not clear whether it supports a long-term repository of allocentric maps, or whether it has a role in manipulating information necessary for remote spatial memory recall. Based the observation of head direction cells in this region and the reported problems of patients, it has been suggested that it either stores allocentric heading directions (Moscovitch et al., 2005), or helps convert the between allocentric and egocentric representations (Burgess et al., 2001; Byrne et al., 2007). PARIETAL CORTEX Continuing our anatomical journey we reach the parietal cortex. Like the retrosplenial cortex, bilateral or right lateralized lesions of the posterior parietal cortex can disrupt new spatial learning and cause a severe disorientation in familiar environments, but generally spare landmark recognition abilities (De Renzi, 1962; Kase et al., 1977; Levine et al., 1985; Aguirre and D Esposito, 1999). Thus, it has been postulated that this region, along with the retrosplenial cortex may act as the repository of spatial maps of environments learned long ago (Teng and Squire, 1999). However, a body of evidence indicates that the parietal cortex represents spatial information in an egocentric frame of reference, rather than a map-like reference frame (Levine et al., 1985; Milner and Goodale, 1995; Aguirre and D Esposito, 1999). Thus, the posterior parietal cortex likely supports remote spatial memory by providing access to egocentric mental views of places in familiar environments. Whether the posterior parietal cortex stores longterm memories of these views or manipulates information retrieved from elsewhere is an important question for theories of remote spatial memory. The study of patients suffering from what has been called representational neglect following posterior parietal lobe lesions (often right lateralized) with spatial imagery tasks has been particularly useful in addressing this question. When asked to describe a route, the turns on the left side of the path would often be missed out, regardless of the direction of travel (Bisiach et al., 1993). A similar pattern was observed when patients from Milan were asked to describe what they would see when standing in Milan s Piazza del Duomo, when either facing toward the cathedral or facing away from it. When imagining facing the cathedral they neglected to mention the buildings to the left side of that view point; but when imagining the view facing the other direction away from the cathedral they now described all these previously neglected buildings (now on their right) and neglected to mention the buildings now on their left (Bisiach and Luzzatti, 1978). Thus, these patients retain knowledge of familiar routes and the location of buildings around a familiar square, but lose access to (or neglect) those on the left side of space when trying to retrieve the remote memories. Such problems can occur not just for large-scale space but also for small arrays of objects (see e.g. Milner and Goodale, 1995), thus this region has been implicated in a general problem with egocentric spatial processing. Studies examining the activity of cells in primate posterior parietal cortex support the notion that this region is important of egocentric spatial processing. Cells in different parietal subregions, such as the intraparietal sulcus and area 7a can represent visual information in a range of egocentric coordinates, including retina-, head-, body- and hand-cen-

15 21 tered reference frames (see Andersen and Buneo, 2002 for review). These representations can also be mixed such that it is possible for cells to code space in an eye-centered frame modulated by head-position (Andersen and Buneo, 2002). Neuroimaging studies of remote spatial memory in London and Toronto have also observed engagement of the posterior parietal cortex, as well as a more medial area that includes the precuneus (Maguire et al., 1997; Rosenbaum et al., 2004; Kumaran and Maguire, 2005; Spiers and Maguire, 2006a). It has been argued that this latter region acts as the mind s eye allowing the inspection of mental images (Fletcher et al., 1995; Burgess et al., 2001), a function it performs as part of the network including the retrosplenial cortex and MTL (Burgess et al., 2001; Moscovitch et al., 2005; Byrne et al., 2007). Other lateral parietal areas have been suggested to aid the transformation of spatial information across different reference frames, akin to the retrosplenial cortex (Burgess et al., 2001). Overall, the available evidence indicates that the parietal cortex supports remote spatial memory by processing egocentric spatial information to either remember or imagine places from particular viewpoints, rather than as acting as longterm store of remote memories. CONCLUSIONS FROM REMOTE SPATIAL MEMORY S ANATOMICAL SNAPSHOTS Our journey from the hippocampus through parahippocampus and retrosplenial cortex to the parietal cortex sadly ends without a definitive answer to the question, which regions store the remote spatial memory traces. Whereas the hippocampus may lay claim to the fine-grained spatial maps of remotely learned environments (Rosenbaum et al., 2000; Maguire et al., 2006a), the evidence that the other regions in this network act as the sole repository of the coarse remote representations remains inconclusive. Tentative evidence currently favors the retrosplenial cortex, but until careful testing of (rare) patients with selective retrosplenial lesions is conducted and more is learned about how their deficits can resolve, this conclusion should be treated cautiously. At least two possibilities exist. First, it may be that all regions (parahippocampus, retrosplenial, and parietal cortices) are necessary in conjunction to support the coarse map-like representation. Rosenbaum et al. (2004) have argued that it is the links between these structures (initially mediated by the hippocampus) that are the basis of successful performance on spatial memory tasks requiring semantic-like spatial representations. Another possibility is that other brain regions are responsible, such as the temporal cortex and prefrontal cortex, but as noted at the start of this section there is little evidence to currently support this assertion. Thus while these anatomical snapshots offer some intriguing insights into the neural substrates of remote spatial memory in humans, and have gone some way to informing extant theoretical debates, questions remain. It may be that attempting to ascertain the function of brain regions in isolation may ultimately constrain our understanding how each region operates in the context of the dynamic brain systems of which they are a part. Interestingly, it has very recently become possible to overcome the limitations of snapshots and instead start to examine how these brain regions operate together on a fine-grained time scale during navigation in a city learned in the remote past. This line of research leads once more back to the city of London. BRAIN DYNAMICS UNDERLYING ACTIVE NAVIGATION IN LONDON Navigating through a city such as London (see Fig. 2D) is a demanding task requiring the coordination of many cognitive operations. Attempting to ascertain the online neural correlates of navigating in the real world represents a huge challenge. Previous functional neuroimaging studies employing VR environments found that the spatial memory network of brain regions, including the hippocampus, parahippocampal, retrosplenial and parietal cortices, were more active during navigation (Aguirre et al., 1996; Maguire et al., 1998; Gron et al., 2000; Hartley et al., 2003; Iaria et al., 2003). However, these studies typically averaged activity over blocks of s, which fails to capture the multi-faceted and highly dynamic operation of the human navigation system. Furthermore, unlike the familiar and remotely learned places we typically live in and navigate in the real world, the VR simulations employed were deserted, simplistic and largely unfamiliar. FROM VERBAL REPORTS TO NAVIGATIONAL THOUGHTS To explore the brain regions supporting the navigation of a familiar environment Spiers and Maguire (2006a) scanned subjects with fmri as they navigated in the highly accurate virtual simulation of London described previously (see Fig. 6). During scanning, subjects (who were licensed London taxi drivers) responded to customers requests by delivering them to their required destinations, while driving a London taxi. To gain a comprehensive understanding of the navigation process on a second-by-second basis, an innovative approach was taken to reading the minds of the subjects while they navigated. Immediately post-scan and without prior warning, subjects watched a video replay of their performance and were interviewed using a retrospective verbal report protocol (Ericsson and Simon, 1980). This involved getting subjects to review their performance and report on what they had been thinking while they had been doing the task in the scanner. Subjects were able to produce very detailed accounts of what they had been thinking during the navigation, and were also clear about exactly when they had experienced particular thoughts. This enabled a complete specification of each subject s fmri time series in terms of onsets and durations of events/epochs (see Fig. 6). The precision of the timings was further tested using independent eye-tracking data acquired during the scan. Careful examination of the subjects verbal reports made it possible to create a step-by-step breakdown of how navigation unfolds in a familiar city. In brief, we initially

16 22 Fig. 7. Spatial goal coding during navigation in London. (A) An example of one of the routes taken by a subject navigating virtual London. The map is reproduced by permission of Geographers A-Z Map Co. Ltd. Crown Copyright All rights reserved. License number This route starts in Oxford Street, has Peter Street as the goal location and is indirect because of obstructed streets and the one-way system. The goal is marked with a G in a black circle and one of the many locations on the route, in this case near the goal, is marked with a * in a black circle.