This book begins by exploring navigational processes in a variety of species ranging from insects to humans, outlining both the phenomenology and possible neural bases. Studies of animals and humans have proved to be mutually illuminating, and have begun to reveal the computational and neural principles that underlie many kinds of spatial behaviour.
We start with one of the simplest forms of navigation, a faculty known as dead reckoning (originally a nautical term) or path integration (a biological one). These terms refer to the capacity of a great variety of animals to return directly to the starting point of their journey (usually a nest or home base of some sort) after a circuitous outward trip. Animals can ‘home’ very rapidly, at any arbitrary time after they began their journey, and this has led biologists to speculate that the representation of the direction and (probably) distance home must be constantly updated, using information about the movements the animal has made, rather than being calculated de novo on the basis of currently visible landmarks. This enables homing to be fast and is therefore adaptive, especially for species that have a fixed home and are prey rather than predators, and animals can do it without reference to external landmarks or odour trails. They must therefore maintain a continuously stored record of the distance(s) and direction(s) since departure. The computations involved are nevertheless not trivial, and are equivalent to performing trigonometric operations on the incoming data. Finding out how nervous systems do this may give us some insights into how the brain, in general, goes about performing mathematical computations.
This capacity to path integrate evolved a long time ago, and even relatively simple creatures like ants and spiders can path integrate over enormous (to them) distances. The opening chapter, by Wehner and Srinivasan, reviews what is known about path integration in these creatures. Not only is this an intriguing scientific question in its own right, but the study of how arthropods have solved the path integration problem can point mammalian biologists to the kinds of processes they might expect to find in rodent and even human brains. It is clear that the basic inputs to the path integration system—namely, the distances travelled in which directions—must be the same in all species. Similarly, the drawbacks of the process—such as its tendency to accumulate errors unless reset periodically with visual, or other, ‘fixes’—are also common to all species. Because insects are easy to keep and study, and because their brains are so small, it has been possible to discover the internal workings of their path integrators to a degree that mammalian neurobiologists can only envy.
Wehner and Srinivasan show that insects use a variety of information sources, such as the direction of the sun or the polarization patterns and spectral gradients in the sky, to determine their heading. Somewhat remarkably, given the small size of their brains, they are able to compensate for the time of day so that the sun always provides a constant indicator of direction. In this way, an insect setting out in the morning and returning in the afternoon does not become lost as a result of the sun’s having moved across the sky. The other aspect (p.4) of path integration involves determining distance travelled in a particular direction. Do insects have internal odometers? A number of experiments suggest that they do, and that the odometer can, amazingly, compensate for undulations in the terrain, so that only the straight-line distance is estimated (Wohlgemuth et al. 2001).
Mammals can also path integrate, some with exceptional ease. Accumulating evidence (reviewed in Chapters 2, 3, and 7 by Wallace et al., Etienne, and Wang and Spelke, respectively) suggests that mammalian path integration is intimately bound up with the representation of the animal’s whole surroundings, rather than simply the path home. Wallace et al., in Chapter 2, take an ethological approach to the study of path integration by examining a laboratory homing task, on a holeboard maze, that mimics (in a simplified way) the natural foraging conditions of wild rats. This enables close observation of the kinds of behaviours that the animals exhibit when foraging or homing. A dissociation is apparent between outward trips, when the animal is searching for food, and homing trips when it takes a more or less direct route back to the ‘home’ hole. Since this straight-line homing occurs equally well in the absence of visual or other localizing information, the animals must rely on path integration, and in fact Wallace et al. show that they probably rely mainly on path integration to home in the light too. This faculty is disrupted by lesions of the fornix (a major source of input to and output from the hippocampus) and hippocampus, suggesting that path integration may be mediated by the same structure(s) implicated in the representation of space. Interestingly, lesions to the vestibular system, which provides inertial information regarding motion in the three dimensions, also disrupts the ability of rats to home across the holeboard. Wallace et al. put forward the intriguing suggestion that the hippocampus has a specialized role in integrating vestibular signals with information coming from other sensory sources.
Ethological studies like the above have shown that path integration seems to form a continuous undercurrent to spatial behaviour. Etienne returns to this point in Chapter 3, and explores how path integration acts in concert with other navigational processes. For path integration to be effective in the real world, the brain must have some means of integrating this motion-derived signal, which is more or less independent of the outside world (except for the Earth’s inertial frame of reference) with sensory information emanating from static cues such as landmarks. This is partly because path integration is a somewhat imprecise process, prone to accumulating errors unless ‘reset’ periodically, and partly so that if it is interrupted for some reason then the animal can use the features of the environment (assuming it has encountered them before) to become oriented again. Etienne examines the interplay between motion cues and other sensory cues, and shows that information from the various sensory modalities seems to be ordered hierarchically, with vision dominating over olfaction, and both dominating over motion cues. Such a hierarchy is obviously adaptive, since it seems to be organized on the basis of the reliability of the various information sources, and stands as a testament to the ways in which animals can use multiple information sources in a highly opportunistic way.
Etienne then turns to the way in which location-based cues (mainly visual) and path integration can cooperate in orienting the animal. Just as for sailors dead reckoning their way across the oceans, the opportunity to take a visual ‘fix’ can remove any errors that may have accumulated, and restore the animal’s estimate of location. Both behavioural and physiological studies have been revealing here, and show that the path integration signal is reset after the animal has caught sight of known visual cues. This then raises an interesting question: how can an array of visual landmarks restore a drifted path integration signal unless the animal has a stored (p.5) record of where they are in relation to each other? In other words, there must be some interaction between the allocentric representation of the objects—which seems, on the face of it, rather like a map—and the path integration signal. And similarly, if the animal makes a movement in the absence of a constant visual input from the surrounding landmarks, then its estimate of location must be updated using path integration, and this in turn helps recognition of the landmarks when they become visible again. There thus seems to be an ongoing and reciprocal relationship between the spatial representation and path integration.
Chapter 4, by Collett et al., moves on from path integration to look at the use of non-spatial environmental information in navigation. This chapter focuses on insects, which turn out to have a remarkably elaborate spatial representation, and offers insights into the study of spatial behaviour of relevance to mammalian biologists too. Indeed, context is a concept that recurs in the chapter by Wiltgen and Fanselow, and again in the chapter by Anderson et al. and it seems increasingly apparent that it must form a crucial backdrop to any representation of place. Potential contextual cues, in the case of insects, include the visual panorama, the distance travelled along a route, the cues encountered along the way, time of day, and motivational state. Collett et al. show that insects are quite sophisticated in their capacity to use context to modulate their spatial behaviour, using these cues to disambiguate other spatial features such as landmarks, to enable context-specific behaviours, and to retrieve memories.
How do contextual cues interact with other, spatial cues? The bee experiments described by Collett et al. have strong parallels with the place cell experiments described by Anderson et al. in Chapter 15. In both cases, it seems that sets of cues become bound together as ‘stimulus configurations’, so that an animal (or place cell) can make a response to a combination of stimuli that it could not make if it only received inputs from the stimulus elements themselves. These kinds of experiments, aimed at ‘pulling apart’ the contextual and spatial representations, provide an inroad into the underlying neurobiology and point us towards the kinds of mechanism we might expect to see operating at a neural level.
The issue of how non-spatial and spatial information interact to guide behaviour is, in general, attracting increasing attention from neurobiologists. It is all very well for an animal to have a sophisticated representation of the environment, but what is this representation for? Ultimately, of course, it is to help the animal behave adaptively in its environment, seeking out resources and avoiding harm. In this light, Wiltgen and Fanselow in Chapter 5 explore the interplay between the spatial representation and the circuitry underlying the expression of fear responses. They review research into this circuitry, of which a crucial component is the amygdala, and then look at how the spatial representation interacts with it so that animal can learn to fear a place in which it has experienced an aversive event. Intriguingly, animals only learn to fear a place if they are given time to experience the place before the to-be-feared event occurs—a finding that would not have been predicted on the basis of other learning studies, in which pre-exposure to a stimulus lessens the amount of learning about it that subsequently occurs. Fanselow suggested a number of years ago (Fanselow, 1986) that this might be because animals need time to assemble a representation of the environment, or ‘context’, before they can proceed to learn things about it. On the basis of other research reviewed in this volume we might expect the hippocampus to be involved somehow in the ‘place’ aspect of this learning, and evidence suggests that this is indeed so. A number of studies have implicated the hippocampus in contextual fear conditioning (Kim and Fanselow 1992; Phillips and LeDoux 1992), supporting the idea that the hippocampus is the structure where the environmental stimuli are assembled into a representation of (p.6) spatial context. How this occurs is an issue that is discussed by Wiltgen and Fanselow, and returned to in Chapter 15 by Anderson et al. In particular, Wiltgen and Fanselow suggest that the representation of context involves a network of structures in and around the hippocampus, which can be broadly divided into a dorsal system, involved in forming compound (‘configural’) representations of the spatial aspects of the context, and a ventral system which combines this information with emotional and motivational information. Together, these systems can bring together what an animal knows about its location with what it knows about what happened there. Not only does this allow it to execute an appropriate fear response when finding itself in that place in future, but such a system may form the underpinnings, as discussed in Chapters 8 and 16, of mammalian episodic memory.
The next chapters in the book turn to the nature of the spatial relationship itself, dealing with the thorny question of whether this representation takes the form of an allocentric cognitive map, in the Tolman/O’Keefe-Nadel sense, or whether it is something less map-like and more egocentric. Healy et al. in Chapter 6 take an ethological perspective and discuss the advantages and disadvantages that having a cognitive map would confer upon an animal. The principal advantage is that such a map would allow the planning of novel routes through the environment, such as when a usual path is blocked (necessitating a detour), a previous blockade is relieved (permitting a short-cut), or the animal is traversing an area it has never previously visited. Beginning with the seminal experiments of Tolman (1948), studies on a variety of species, both in the wild and in the laboratory, have looked for evidence of such behaviours. However, these studies, as Healy et al. point out, are hard to conduct (particularly in the field, where tracking an animal for the entire length of its journey may be unfeasible) and hard to interpret. In the laboratory, most features of the environment are visible from both the start points and end points of a journey, so even if animals have cognitive maps, they may not use them to navigate under these conditions. Healy et al. conclude that a reasonable jury must still be undecided on the question of whether map use can be ruled in or out. Complex spatial behaviour may involve a map, but equally, it may arise from the flexible and opportunistic use of other, simpler kinds of stimuli and behaviours.
Wang and Spelke in Chapter 7 agree that the evidence for cognitive maps is inconclusive, and suggest that even in humans, the representation of space is more ephemeral and flexible than previously thought. They examine three human spatial competences: path integration, scene recognition, and reorientation. To begin, they review evidence that humans path integrate continuously, and that this process works best when the subject physically moves through space. They discuss data suggesting that during path integration, although the representations of the positions of objects in the environment are updated on the basis of information generated by the subject’s movements each object’s position is updated independently of the others. This suggests that the objects are represented with respect to the subject rather than to each other. This somewhat surprising finding suggests that even in humans, the fundamental representation of space (at least as constructed via path integration) may be egocentric, rather than allocentric as we tend to suppose.
Turning to scene recognition, Wang and Spelke explore the existence of view-dependent representations in a wide variety of species from insects to humans. This kind of representation involves a visual matching between the visible panorama and a representation stored in the animal’s memory. It does not require the incorporation of any explicitly spatial information (such as distance or direction) and, as such, is relatively low in its computational requirements. Again, in humans, it seems that this kind of representation is updated automatically during movement, so that a change in viewpoint is more easily processed if the (p.7) subject moved around the environment than if the environment moved around the subject. Once more, this points to the likelihood that the representation of the objects contained in a view is egocentric rather than allocentric (since recognition of them is disrupted if they move with respect to the viewer, even if they maintain the same relationship to each other). This kind of viewpoint dependence of spatial processing is returned to in the following chapter, by Hartley et al., who have a different, albeit related, perspective.
Reorientation, the third of Wang and Spelke’s three spatial competences, is the process by which an animal becomes oriented again when path integration has been disrupted for some reason. As such, this process depends on the perception of cues in the environment and the matching of these to some internal representation of location. Wang and Spelke discuss evidence that, in both rats and young humans, the fundamental process of reorientation is governed by the extended surface layout (in other words, the shape of the boundaries defining the environment). This is interesting because, as we shall see in the Part II of this book, place cells are also strongly influenced by the surface layout of the environment (O’Keefe and Burgess 1996), and less so or not at all by objects within it (Cressant et al. 1997). Cheng (1986) suggested that an animal’s use of geometry to reorient itself is an example of a ‘module’—a particular mental faculty that has evolved to handle a restricted kind of information (by contrast with a multi-purpose processor which could, in theory, operate on any kind of information). The question of whether cognitive systems are modularized in this way has greatly exercised behavioural psychologists, and Wang and Spelke use reorientation as a system with which to explore this issue. Fodor (1983) outlined a set of criteria that define a module, and Wang and Spelke examine three of these: domain specificity, task specificity, and encapsulation, to see whether reorientation might be a module, concluding that the process does indeed seem to meet these criteria. This chapter ends with the important point that in this area, not only do animal studies illuminate the human condition, but experiments on humans are contributing to our understanding of animal navigation.
Hartley et al., in the concluding chapter of Part I, also examine spatial navigation in humans, but they go a step further and argue that such a system also forms the basis for the human capacity for episodic memory. By contrast with Wang and Spelke, they argue that the human representation of space is likely to be allocentric rather than (solely) egocentric, for the simple reason that egocentric representations require constant updating and would tend to accumulate errors. They suggest that representations of salient features of the environment, such as objects, are stored in long-term memory, and can be used to reorient the animal when it is faced with some of these features seen from an arbitrary viewpoint. Using data gleaned from studies of a hippocampally damaged patient, Jon (Vargha-Khadem et al. 1997), they argue that the role of the hippocampus may be to allow the representations of objects to be ‘rotated’ in a way that brings them in line with the current visual scene. They further suggest that the locations of the objects may themselves be stored egocentrically rather than allocentrically, a proposal that offers a point of contact with Wang and Spelke’s view. An allocentric spatial representation could thus be built from a set of egocentric ‘snapshots’, each taken from a particular viewpoint, and joined together by the place cell network. This intriguing proposal may explain the hitherto puzzling finding that hippocampal cells recorded from primates seem more responsive to the visual panorama than to the spatial location of the animal per se (Rolls and O’Mara 1995), as well as a variety of other findings in both the behavioural and physiological domains.
While single-unit studies are difficult or impossible to perform in humans (for obvious ethical reasons), the development of functional imaging techniques, particularly in conjunction (p.8) with emerging virtual reality technologies, have made it possible to expose human subjects to artificial navigation experiences, and visualize which areas of the brain become active. Hartley et al. review some of these studies and outline the network of structures in the hippocampal region that seem to support human navigation. They then turn to the question of how this network might function in storing memory for events that have occurred. Using the same functional imaging and virtual reality techniques as in the navigational studies, they explored the effect of introducing events into their subjects’ virtual environments, using the resulting data to outline a possible anatomical substrate. Interestingly, not only does the episodic memory circuitry have much in common with the navigational circuitry, but there is evidence that, at least in humans, these two functions have become somewhat lateralized. This supports the possibility, first suggested by O’Keefe and Nadel (1978), that evolution has built our episodic memory system upon the underlying scaffolding of a spatial representation.
To summarize, then, the chapters in the first half of this book examine the architecture of the spatial representation, examining not only how different spatial behaviours are manifest in animals, but how they interact with other kinds of behaviour, and what kinds of neural circuitry might underlie such behaviours. In mammals, the hippocampus emerges as the prime candidate for a spatial representation of some kind, map-like or not. These chapters thus set the scene for Part II, in which we explore, at a neural level, the details of the hippocampal representation of space.
Cressant, A., Muller, R. U., and Poucet, B. (1997). Failure of centrally placed objects to control the firing fields of hippocampal place cells. J Neurosci, 17, 2531–42.
Fanselow, M. S. (1986). Associative vs. topographical accounts of the immediate shock freezing deficit in rats: implications for the response selection rules governing species-specific defensive reactions. Learn Motiv, 17, 16–39.
Fodor, J. (1983). The modularity of mind. MIT Press, Cambridge, Massachusetts.
Kim, J. J. and Fanselow, M. S. (1992). Modality-specific retrograde amnesia of fear. Science, 256, 675–7.
O’Keefe, J. and Burgess, N. (1996). Geometric determinants of the place fields of hippocampal neurons. Nature, 381, 425–8.
O’Keefe, J. and Nadel, L. (1978). The hippocampus as a cognitive map. Clarendon Press, Oxford.
Phillips, R. G. and LeDoux, J. E. (1992). Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav Neurosci, 106, 274–85.
Rolls, E. T. and O’Mara, S. M. (1995). View-responsive neurons in the primate hippocampal complex. Hippocampus, 5, 409–24.
Tolman, E. C. (1948). Cognitive maps in rats and men. Psychol Rev, 40, 40–60.
Vargha-Khadem, F., Gadian, D. G., Watkins, K. E., Connelly, A., Van Paesschen, W., and Mishkin, M. (1997). Differential effects of early hippocampal pathology on episodic and semantic memory. Science, 277, 376–80.
Wohlgemuth, S., Ronacher, B., and Wehner, R. (2001). Ant odometry in the third dimension. Nature, 411, 795–8.