The topography of memory

Article metrics

For more than 40 years neuroscientists have explored the cerebral cortex with microelectrodes, recording the electrical activity of single neurons while ‘tickling’ them with different stimuli. Early investigators found that the cortical areas responsible for initial processing of sensory information yielded their secrets relatively easily1,2 — for each area, a small set of simple sensory features that activate the neurons could be identified. Many of these features have since been found to have a systematic anatomical organization; they are mapped onto the surface of the cortex, and the topographies for different sensory features are interleaved within each cortical area.

Since those early successes, the feature-coding properties of other cortical areas, including late stages of sensory3 and cognitive4 processing, have also been characterized. Yet there has been limited success in describing a topography for those cortical areas5, suggesting that higher and more complex aspects of perception and cognition might not be systematically mapped. On page 610 of this issue, however, Hampson and colleagues6 reveal a functional organization for the hippocampus, one of the highest cortical-processing areas in the brain.

Although the hippocampus is crucial for memory, the nature of its contribution is controversial. Human studies point to critical hippocampal function in ‘declarative memory’7 (our ability to record personal experiences and weave these episodic memories into our knowledge of the world about us). By contrast, animal studies indicate that the hippocampus is dedicated to ‘spatial mapping’. Animals with hippocampal damage show severe and selective impairments in spatial memory8 (their ability to learn and express knowledge of spatial relations among cues in the environment). Animal studies have also revealed the existence of hippocampal ‘place cells’ — neurons that fire when the animal is at a particular location, as though signalling occupancy of a coordinate in a spatial map9.

Recent studies offer the beginnings of a reconciliation between the ‘declarative’ and ‘spatial’ views of hippocampal processing. For example, rats with hippocampal damage have impaired learning, and they cannot express knowledge about relations between distinct experiences in nonspatial memory10. It has also become clear that hippocampal neurons are activated in association with specific nonspatial cues, as well as with cognitive events and behaviours that occur at different times in many places11. These observations suggest the hippocampus is fundamental in registering and linking episodic memories into larger networks for both spatial and nonspatial memory12. In a network of spatial memories, the critical links may be remembered places that a person visited at different times. In a nonspatial memory network, the links may be particular objects, people or events that are common to (and hence can bridge) distinct episodic memories that include those items. But how would such memory networks be organized within the architecture of the hippocampus?

Hampson et al.6 now outline, for the first time, a functional organization of the hippocampus — a sketch of a memory network on the surface of the hippocampus. The authors monitored the activity of hippocampal neurons in rats, first when they performed a simple task, and then when they remembered the episode in a subsequent test. On each trial the animal initially pressed a ‘sample’ lever, presented in one of two positions in a chamber. The rat maintained this memory for several seconds, then demonstrated it by choosing the other lever when both were presented in a ‘nonmatch’ phase of the trial.

To make these recordings, Hampson et al. used a unique electrode array that allowed them to monitor many cells at known distances apart in the hippocampus (Fig. 1). In each animal the activity of some neurons was associated with the position of the lever being pressed, regardless of whether this occurred during the sample or nonmatch trial phase. Other cells fired during just one of the trial phases, independently of the lever's position. Yet other cells fired in association with the various combinations of lever position and trial phase (for example, left–match), or with several events that made up a specific type of trial (for example, right–sample then left–nonmatch). All of which means that hippocampal neuronal activity represents both the relevant aspects of space and the relevant nonspatial features of the task, consistent with the mixture of spatial and nonspatial coding observed in other situations11.

Figure 1: ‘Unfolding’ of a functional organization in the hippocampus.

(Figure provided by R. E Hampson and S. A. Deadwyler.)

a, Pairs of hippocampal lamellae — anatomical segments within which CA3 and CA1 are closely connected. Each pair of lamellae represents a single lever position (black or white for left or right) and a set of trial phases (sample and nonmatch). b, Normal appearance of the dorsal hippocampal regions CA1 and CA3, and the sites where Hampson et al.6 made recordings with the electrode array. c, Unfolding of the hippocampus, revealing the systematic functional architecture.

By combining the data from several animals, Hampson et al. found a set of regular anatomical patterns (Fig. 1). Lever-position codings are segregated, such that alternating 0.6–0.8-mm cross-sectional segments of the hippocampus contain clusters of ‘left’ or ‘right’ lever-position codings. Trial-phase codings are also segregated in alternating 0.2–0.4-mm cross-sectional clusters of ‘sample’ and ‘nonmatch’ responses. The two topographies are interleaved, such that each position cluster contains clusters for both trial phases. Furthermore, in this mapping, function follows form — the clusterings of position and trial-phase specificity follow the known anatomical organization in which neurons are more closely interconnected within cross-sectional segments (called lamellae) than they are across them13. This correlation between anatomical and functional organization resembles the common anatomical and functional ‘columns’ of well-connected, like-responsive cells in most areas of the cerebral cortex1,2.

Two other aspects of Hampson and colleagues' observations should be emphasized. First, the representation of space in this functional organization is not a detailed Cartesian mapping, as envisaged within the dedicated spatial-mapping view of hippocampal function9. The authors' code involves only two positions — not a continuous set of spatial coordinates. Perhaps the hippocampus discriminates space only at the resolution required to remember where important events occurred.

Second, the full scope of declarative memory probably cannot be reduced to features of working memory such as ‘position’ and ’trial phase’. Yet the topographies for these features fill the hippocampus, and there are no empty slots for other features of memory. The same hippocampal cells can encode different — and seemingly unrelated — features when an animal is exposed to different situations, even in the same environment, and the scope of hippocampal coding12 goes beyond the memory task studied by Hampson et al.. So functional organization of the hippocampus probably contains a different set of interleaved topographies for each of several memory networks.

Finally, the authors offer a further clue that may relate the organization of these networks to their operation. Hippocampal neurons encoding combinations of the position and trial phases are found on the borders of the position and trial-phase codings. Perhaps these cells represent events unique to particular types of trial episode. The lever-position and trial-phase cells, by contrast, encode events that are common to, and might link, the representations of different episodes. A memory network based on these linked episodic codings could mediate a rat's ability to remember previous trials based on current events. Of course, this is the critical memory demand in the task. But the same kind of functional organization could mediate the linking of episodic memories — and access to them through current cues — across many domains of memory in humans as well as in animals.


  1. 1

    Mountcastle, V. B. J. Neurophysiol. 20, 408–434 (1957).

  2. 2

    Hubel, D. H. & Wiesel, T. N. J. Physiol. 160, 106–154 (1962).

  3. 3

    Gross, C. G., Roche-Miranda, D. B. & Bender D. B. J. Neurophysiol. 35, 96–111 (1972).

  4. 4

    Fuster, J. M., Bauer, R. H. & Jervey, J. P. Exp. Neurol. 77, 679–694 (1982).

  5. 5

    Tanaka, K. Science 262, 685–688 (1993).

  6. 6

    Hampson, R. E., Simeral, J. D. & Deadwyler, S. A. Nature 402, 610–614 (1999).

  7. 7

    Cohen, N. J. & Squire L. R. Science 210, 207–210 (1980).

  8. 8

    Morris, R. G. M., Garrud, P., Rawlins, J. P. & O'Keefe, J. Nature 297, 681–683 (1982).

  9. 9

    O'Keefe, J. A. Exper. Neurol. 51, 78–109 (1976).

  10. 10

    Bunsey, M. & Eichenbaum, H. Nature 379, 255–257 (1996).

  11. 11

    Wood, E. R., Dudchenko, P. A. & Eichenbaum, H. Nature 397, 613–616 (1999).

  12. 12

    Eichenbaum, H., Dudchenko, P., Wood, E., Shapiro, M. & Tanila, H. Neuron 23, 1–20 (1999).

  13. 13

    Andersen, P., Bliss, T. V. P. & Skrede, K. K. Exp. Brain. Res. 13, 222–238 (1971).

Download references

Author information

Correspondence to Howard Eichenbaum.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Eichenbaum, H. The topography of memory. Nature 402, 597–598 (1999) doi:10.1038/45117

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.