Spatial Memory or Memory Space: Is that the Question?

Sean M. Montgomery
Spring, 2002

The cognitive map theory of hippocampal function began in 1971 with an electrophysiological recording study by O’Keefe & Dostrovsky. Contrary to many of the electrophysiology and behavior studies at the time, O’Keefe & Dostrovsky set about investigating hippocampal function from an observational rather than a strict experimental vantage point. The authors used an exploratory approach in the study by placing recording electrodes in the hippocampus and observing the responses of hippocampal units as the animal wandered around the cage and as the experimenters presented various olfactory, auditory, sensory, and visual cues to the animal and removed various cues from the environment. The authors found the cells to respond in a number of different ways, but remarkably, the authors found cells that would respond only when the animal was in a particular allocentric location. The response of these cells, later termed ‘place cells’, remained despite removal of cues from the environment, unless the cue removal was so radical as to cause the animal to behave as though it were in a new environment. These allocentric cells shared a great similarity to what might be expected from a neural substrate for the cognitive maps that Tolman (1948) had previously proposed.

In 1978, O’Keefe & Nadel published a book detailing their theory that there are many memory systems in the brain and that the hippocampus is one of these memory systems, which underlies spatial learning and memory. This theory was a substantial departure from many of the theories about hippocampal function that were generated from human neuropsychological studies. But, when the theory came out, it encompassed the vast majority of the data coming from both animal lesion and recording studies. This multi-level analysis, in combination with the fact that the theory was very concrete in its assertions gave it a great deal of authority over many theories of the day that were commonly plagued by hypothesis drift and operationalism (Nadel, 1991). Additionally, a spatial hypothesis for hippocampal processing could find firm grounding from an ethological perspective. Space and spatial orientation is of particular importance to the survival of an individual. If one cannot remember the spatial organization of his surroundings, including where food, safety, and danger can be found, he will soon be swatted aside by the unforgiving hand of evolution.

Since 1978 there have been a number of theories that have opposed the cognitive map theory of the hippocampal function. Many of these theories differ in the specifics of what the hippocampus actually does, but they generally agree that spatial memory is only one example of the type of processing the hippocampus can perform. One prominent theory of hippocampal function has been proposed by Eichenbaum et al. (1999), which argues that the hippocampus permits the formation of relational memories or a ‘memory space’ that allows the memories to be used in a flexible manner. Eichenbaum et al. correctly argue that the cells described by O’Keefe and Dostrovsky (1971) were influenced by more than the just the location of the animal (like whether or not the experimenter was touching the animal while the animal was in the proper location). Eichenbaum et al. argue that in reality, true place cells that O’Keefe and Nadel (1978) describe only occur in circumstances where the spatial cues are the only regularities in the experimental paradigm (e.g. Muller & Kubie, 1987). While the spatial cues are maintained, the delivery of rewards and the onset, direction, speed, and punctuation of movements are intentionally randomized to subtract out the influence of other behaviors. In fact, it has been shown by Markus et al. (1995) that while true place cells are seen when an animal randomly forages in an open field, hippocampal cells show direction selectively when the animals systematically visit a small number of reward locations in the same environment. Further evidence against a spatial hypothesis comes from McEchron & Disterhoft (1997), who show that units recorded from a rabbit hippocampus during eyeblink conditioning show activity related to conditioning cues that changes with training even though the animal’s location remains constant throughout training.

Eichenbaum et al. argue that these and other data reject the notion that the hippocampus is solely playing a role in spatial memory. Instead, they propose that the hippocampus is involved in encoding sequences of behaviorally relevant events. Over multiple exposures to partially overlapping sequences the connections between common elements may be strengthened, while the non-overlapping features may be cancelled out. After repeated exposure to these sequences, or episodes, neurons in the hippocampus come to represent the common features between the episodes. Eichenbaum et al. argue that under this conception it is possible to explain all of the current data including the place related firing of neurons in the hippocampus. The development of a place field by this notion would involve an animal experiencing multiple traversals through a particular location in the environment from various directions. As the animal makes the different traversals, the stored sequences of events will be linked through partially overlapping sensory attributes of the traversals. As these temporal sequences are laid over one another, the constant features (the spatial location in this case) will come to be represented by hippocampal neurons. Eichenbaum et al. emphasize that in this scheme it is sequences of events in time that form the organizing principle for the development of representations rather than spatial relations between objects in the physical environment. Eichenbaum et al. argue that this conception of hippocampal function would enable neurons in the hippocampus to represent any behaviorally relevant features of the environment that show stable relationships over multiple encounters.

The experiment performed by Wood et al. (1999) was designed with the intent of directly demonstrating that hippocampal neurons show activity that is specifically related to features of a task even though these features show no correlation to space. In the experiment, the authors ran rats on a variant of a delay non-match to sample task. Specifically, on each trial a cup containing scented sand was placed at one of nine locations on a raised platform. On half of the trials the scent of the platform was different from that of the previous trial and was baited with a froot loop at the bottom of the cup. On the other half of the trials the scent was the same as the previous trial and the cup was not baited. On each trial the rats approached and sniffed the cup and would either dig to the bottom of the cup or would turn away. In total the rats were run on 108 trials a day where the 9 possible odors, the 9 possible locations and the match-non-match contingencies were pseudo-randomized such that each session contained 12 presentations of each odor and each cup location.

During performance of this task hippocampal units were recorded. The authors characterized the activity of these neurons by looking at the one-second period before a response was made. Neurons were statistically characterized based on the task parameters to which the neuron showed selective responding. For example, a cell would have been considered an ‘odor’ cell if the cell fired strongly on trials in which odor 5 was present, but not on trials in which any of the other odors were present and additionally showed no selective firing with respect to the other task parameters (location, match/non-match). In order to evaluate unit responses during the animal’s approach to the cup the authors also compared the activity of the neurons in the one-second period beginning three seconds before the behavioral response to the activity of the neurons in the one-second period immediately before the behavioral response. Cells that had a significantly higher firing in one second period beginning three seconds before the response and didn’t show firing related to any of the other tested variables was considered an ‘approach’ cell. The characterization of the 127 cells recorded in the study can be seen in table 1.

From these data, Wood et al. argue that in this task the majority of neurons in the hippocampus show activity purely related to non-spatial variables of the task. This may not be an entirely just portrayal of the data because of the 36 cells that weren’t reported on it is not clear how many of these showed place activity on the rest of the platform (i.e. when the animal wasn’t approaching a cup). It is possible that all 36 of these cells were place-related. Regardless, even if only 40.2% of the cells respond to non-spatial variables in the task, I think these data indeed make a strong case against a purely spatial role for the hippocampus. The data show that when one carefully designs an experiment so that it is possible to dissociate position from other task variables, unit firing in the hippocampus can be shown to correlate with other task variables. It is interesting to note that while reward in this task had no contingency to position there were still a substantial proportion of neurons that encoded aspects of position. While this could be used to argue for a spatial bias of the rodent hippocampus, it could also be argued that space is always a constant feature of the environment and that the animal needs to remember its location so that it doesn’t fall off the platform. These data definitively demonstrate that the neurons in the hippocampus can respond to a number of non-spatial parameters that are relevant to a task at hand. This data is inconsistent with the cognitive map theory and consistent with a memory space or relational theory of hippocampal function.

Lever et al. (2002) recorded from hippocampal neurons as rats foraged for randomly dropped food in a square and a circular arena. Contrary to results in similar studies done previously, 73% of the recorded neurons (48/66) had place fields in a similar location upon initial exposure to the two different environments. To further explore these results, the authors ran a new group of rats on four to six trials a day in which the animals were placed the circle and square arenas on alternating trials to determine if the animal would begin to show differential neuronal firing of the two environments. The hippocampal units of these animals did indeed begin to differentiate the environment over many days. Of the 17 units recorded on day one, 14 of them (82%) showed similar place field location, whereas by day 21, only 7 of the 38 recorded cells (14%) had shown similar field location. The authors reported that this change took place in the absence of any systematic changes in the animals’ behavior. The authors also examined the shifting process more closely and found that the fields shifted in a number of different ways and some of them shifted quickly over a few trials, while some shifted/disappeared more gradually over several days. After several weeks of not testing the animals, the investigators put them back in both environments and found that the majority of cells in all animals showed different place-related firing in the two environments. The authors went on to investigate what the perceptual basis for the acquired environmental discrimination and whether it was transferable to new environments. Specifically they tested whether animals would discriminate inter-shape characteristics over intra-shape characteristics. To test this the authors used a square and circle from the earlier experiments in addition to a ‘morph’ arena that could be configured to dimensions that matched either the square or the circle. They tested the rats that had previously acquired the square/circle discrimination in the morph circle and morph square and trained an additional two rats in the morph circle and morph square and then exposed these animals to the originally used wooden square and wooden circle. The authors found that in all five rats, place related firing was consistent in both the wooden square and the morph square. The results of the circle transfer were mixed with two of the rats trained in the wooden circle showing differential firing patterns between the morph circle and wooden circle while the remaining animals showed similar firing patterns between the two circles.

The authors conclude that these data are consistent with cognitive map theory because animals discriminate between environments on the basis of geometrical features in the absence of differential reinforcement. However, these data are also fully consistent with the memory space theory. In this study it took many exposures to the two environments for the cellular representations to discriminate between the square and circle environments. The fact that the cellular representations largely stayed the same after the first exposure to the square and circle constructed of a new material isn’t in any way indicative that geometry is privileged in determining hippocampal representations. In fact, the memory space hypothesis can account for the fact that some cells discriminate between the two environments of the same shape and others don’t.

In sum, a critical evaluation of Wood et al. (1999) supported the memory space hypothesis of hippocampal function and rejected the cognitive map theory of hippocampal function. The results from Lever et al. (2002) similarly consistent with both theories. The results from these two studies therefore supports a memory space hypothesis of hippocampal function.

One could argue, however, that the memory space theory is too general in its conception. Especially if one considers a form of the theory in which the ‘functional relevance’ of cues determines the degree of representation these cues receive in the hippocampus, it may be impossible to nail down predictions that can be disproved. If the hippocampus represents whatever is important to the animal, it is difficult to think of a result that couldn’t be cast in the light of the memory space theory. This may come back to the hypothesis drift problem that O’Keefe and Nadel (1978) were aiming to avoid in specifying the cognitive map theory. On this basis Lynn Nadel argued, in a shared commentary with Howard Eichenbaum (Nadel and Eichenbaum, 1999), that the cognitive map theory provides a better framework for understanding hippocampal computation. Nadel argues that while many other theories purport to account for place cell data, the accounts are superficial and don’t provide a concrete framework in which grapple with the process by which information is transformed in the hippocampus.

However, in return it could be offered that there is a growing body of evidence for which the cognitive map theory cannot account. And for this reason, the cognitive map theory will inherently give an incomplete account of hippocampal function. Worse still, it’s possible that trying to understand hippocampal computation in terms of the cognitive map theory will lead to an incorrect account of hippocampal function. For example, trying to understand the firing of hippocampal units in terms of distances and angles and calculations analogous to the Pythagorean theorem may lead only to a dead end or a wrong conclusion because the premise on which the data are based is false. An account that explains all the data is going to have to move beyond purely spatial relationships and conceptions that don’t take into account non-spatial information processing from the beginning are unlikely to be more than mental exercises.

So where does this leave the field? The cognitive map theory isn’t robust enough to account for all of hippocampal function and it’s not clear that it can even be determined whether or not the memory space hypothesis can account for hippocampal function. It’s possible that this situation signals an instance in which the function of the brain is too complex to be adequately understood in terms of simple behavioral and psychological ideas. As the theory approaches an adequate level of complexity to account for the data the theory becomes too complex to posit definitive predictions. It is possible that in order to further specify function of the hippocampus it will be necessary to understand the way that the biology of the hippocampus forms the basis for information processing. There is no guarantee that uncovering the ideas in this manner will be the most efficient approach, but I think one can guarantee that explanations couched in biological terminology will have the specificity to delineate the true function of the hippocampus.

Eichenbaum, H, Dudchenko, P, Wood, E, Shapiro, M and Tanila, H (1999). The hippocampus, memory, and place cells: is it spatial memory or a memory space. Neuron 23:209-226.

Lever, C, Wills, T, Cacucci, F, Burgess, N and O’Keefe, J (2002). Long-term plasticity in the hippocampal place-cell representation of environmental geometry. Nature 416:90-94.

Markus, EJ, Qin, Y-L, Leonard, B, Skaggs, WE, McNaughton, BL and Barnes, CA (1995). Interactions between location and task affect the spatial and directional firing of hippocampal neurons. Journal of Neuroscience 15:7079-7094.

McEchron, MD and Disterhoft JF (1997). Sequence of single neuron changes in CA1 hippocampus of rabbits during acquisition of trace eyeblink conditioned responses. Journal of Neurophysiology 78:1030-1040.

Muller, RU and Kubie, JL (1987). The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. Journal of Neuroscience 7:1951-1968.

Nadel, L (1991). The hippocampus and space revisited. Hippocampus 1:221-229.

Nadel, L and Eichenbaum, H (1999). Introduction to the special issue on place cells. Hippocampus 9:341-345.

O’Keefe, J and Dostrovsky, J (1971). The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Research 34:171-175.

O’Keefe, J and Nadel, L (1978). The Hippocampus as a Cognitive Map (New York: Oxford University Press).

Tolman, EC (1948). Cognitive maps in man and animals. Psychological Review 55:189-208.