Place Cells, Remapping and Memory
Bob Muller, close friend and collaborator, died two weeks ago. I met Bob in the early 1980s. I was a post-doc, learning to record from single neurons in Jim Ranck's lab at Downstate Medical Center in Brooklyn. Bob was a young faculty member who worked down the hall. Although Bob was doing esoteric work, studying the physics of single channels in membranes, his early graduate work had been in brain-behavior relations and he wanted to return to the study of behavior.
Jim Ranck and I were studying hippocampal place cells, neurons that "fire" when a rat walks through a restricted part of its environment. Although John O'Keefe had discovered hippocampal place cells about 10 years earlier1, the scientific community was skeptical. Nothing like them had been seen, their existence did not fit then-current theories, and they were difficult to document. However, place cells were remarkable to watch in person. The video below from the 1990s resembles what Bob saw when he visited the Ranck lab. Place cells seemed a magical connection between brain cells, cognition and behavior. Bob approached Jim and me about collaborating.
(This video was recorded in the early 1990s. Loud pops are the action-potential spikes of a place cell whose firing field is on the upper left part of the floor.)
For the next four years we worked closely; Bob set up a computer to collect and analyze data (techniques new to this field) while I prepared rats and made recordings. Together we designed, executed and analyzed experiments. Bob's brilliance, combined with an unusual intellectual background, led to a creative and careful approach. The results, published in 19872, were early computer-based recording-and-analysis of place cells, establishing some of the fundamental properties. These papers also contained the discovery that place-cell maps are completely different in two environments: a phenomenon we later called "remapping"3.
Single-Cell Recording in the Hippocampus
Both place cells and place-cell remapping depend on a scientist's ability to record the firing of single hippocampal neurons during normal behavior. Advances in recording techniques in the 1960s led to the ability to record single cells without interfering with an animal’s behavior. In 1971 John O’Keefe, working with John Dostrovsky, was the first to record and describe Hippocampal Place cells.
The figures below illustrates single-neuron recording techniques. Small wires are implanted in a rat’s brain; after recovery from surgery, the electrodes are advanced and monitored. When an electrode is close to a neuron, it can record the extra-cellular current produced by the neuron’s action potentials. Examples of oscilloscope traces from a single neuron are in the figure on the right. If the wire is firmly in position, the action potentials from a single neuron can be recorded for a day or longer. With a computer data-acquisition system, the timing of a neuron’s spikes combined with information about the location of the animal (rat or mouse) can be captured.
Place Cells and the Cognitive Map
A firing-rate map is method for describing the activity of a place cell averaged over a recording session. In the figure below, the color of a pixel represents the average firing rate while the rat’s head was deteced in that pixel averaged over the 15-minute recoding session. The yellow background is where the cell did not fire while darker colors are higher firing rates. This is a typical place cell. It has a single firing field, in this case at the 10:30 location, as seen from above.
Imagine hundreds of place cells, covering all locations of accessible space. This concept led OKeefe and Nadel to propose that the hippocampus acted as a “Cogntive Map” that represented environmental space and guided efficient navigation in rats and, presumably, humans1. A map for an environment can be described by its topological properties of neurons: pairs of cells can have:
- identical firing fields
- overlapping firing fields
- non-overlapping firing fields
In the spatial domain this is obvious: We can look down on the map and see which pairs of cells have firing fields with these properties. But, since rats can only move at a certain speed, these spatial properties convert to temporal properties. When considering pairs of place cells:
- pairs with identical or overlapping firing fields will tend to fire at the same time.
- pairs with non-overlapping firing fields will never fire together.
This is illustrated in the figures below: we see a pair of place cells with overlapping fields at the top. Below the maps is a “cross correlation histogram” that depicts the correlated firing between the firing of the two neurons. This plot is made by considering each spike from neuron 1 as a “trigger” at time zero (the center vertical line) and plotting the frequency of spikes from neuron 2 at time-offsets from the trigger — plus or minus 5 seconds. It is clear that the spikes from neuron 2 occur near-in-time to the trigger spikes from neuron 1. The peak has a width of about 1 second, roughly the time it takes the rat to cross the overlapped regions of the firing fields.
The figure below is from a recording session when the rat was in a triangular enclosure. In this case the firing fields of two cells do not overlap. Below the rate maps is a spike cross-correlation plot, similar to the one in the figure above. We see that during this session the two cells rarely fire together. This makes sense, since, when the rat is crossing one of the firing fields, only one cell fires.
In brief, we can think of the “map” of a session in terms of space (the spatial relations of firing fields) and time (the tendency for pairs of cells to fire together or not). Since the speed of rats is restricted, these are essentially equivalent. An important concept is that the map is entirely in the brain. In this description, a map is defined by the relation among hippocampal neurons, not by the relationships between neurons and the environment. The linkage to the environment is critical, but does not define the map.
The temporal relations are important for two reasons. First, neurons in the brain do not know about space directly, but they know about time. Neurons can code the timing relations of the neurons that project to it, but not the spatial relations. In other words, within the brain, the map is a timing map that encodes the temporal overlap between cell pairs. Second, the synapses connecting pairs of neurons can strengthen or weaken depending on the timing of the pre- and post-synaptic neurons. If the neurons fire together, as when their firing fields overlap, the synapses connecting the two neurons are likely to strengthen3. This is a likely mechanism for an animal to create a memory for the in a novel environment. That is, in a learned map of a familiar environment, many of the synapses connecting cells with overlapping firing fields would be strong. This concept is related to the storage and retrieval of memories, a topic that will be covered at the end of this article.
How are different environments represented in the hippocampus? When an animal goes from one familiar environment to a second familiar environment, how will the spatial maps respond? Are there rules and regularities? This is a question Muller and I addressed in one of the 1987 papers. Rats were trained in 4 environments: a small cylindrical enclosure, a small rectangular enclosure, a large cylindrical enclosure and a large rectangular enclosure. We then ran 15-minute recording sessions in each. We considered two models of possible results, illustrated in the figures to the right. In the ‘stretched map’ model, the neighborliness of all cell pairs would be maintained. That is, a pair of cells with overlapping firing fields in one environment would have overlapping fields in the second, and a pair with non-overlapping firing fields in one environment would have non-overlapping fields in the second. When recording one cell at a time, if we know a cell's firing field in one environment we should be able to predict the location of the firing field in a other environments. A second possible outcome is predicted by a "remapped" or "scrambled" model. If the cells had overlapping firing field in the first environment, they may, or may not have overlapping fields in the second; if the cells have non-overlapping firing fields in the first environment, they may or may not have overlapping firing fields in the second. Knowing the location of a cell's firing field in one environment would be of no value in predicting field location in another environment.
The results supported the “remapping" ("scambled”) model. In the 1987 study we were only able to record one-cell-at-a-time, but the results were clear. Knowing the location of a cell’s firing field on one environment was of no predictive value about the location of the cell's firing field in the second environement. Although this was not directly observed, this meant that, for pairs of cells, firing fields would scramble. In addition, we found that knowing the location of a cell’s firing field in one environment did not even predict the existence of a firing field in the second environment. We were amazed to see that neurons could be robust place cells in one environment and dead-off in another.
The figures to the right and below illustrate these results. In the figure to the right a single place cell was recorded in the cylinder and rectangular enclosure, each with a white cue card along the right wall (dark line). The neuron had a clear firing field in the 11:00 location in the small cylinder across from the cue card. In the rectangular enclosure the cell's firing field appeared to be in a random location with respect to the first — in this case, directly in front of the cue card. When recording other neurons from this rat, we could not come up with predictive rules. The cells' field locations appeared to scramble from one environment to another.
The clearest refutation of the 'stretched' model was the observation that place cells can "turn off" in specific environments. That is, a neuron could be a perfect place cell in one environment and virtually silent in another. The figure below illustrates this for one cell. This neuron had robust place fields in the cylinder and the rectangle. As we saw in other neurons, there was no relation, or predictive rule for this transform. When the rat was tested in the large cylinder and then the large rectangle, this neuron became virtually silent firing fewer than 20 spikes during these two sessions. One might think that the neuron was "lost", but when the rat was subsequently returned to the small cylinder, and then the small rectangle, robust spatial firing returned. Environment-specific silence was a common finding; a given place cell would "turn off" in about 40% of environments.
In 1991 we described two features of “remapping”3:
- If a cell has a firing fields in two environments, knowing the location of the firing field in the first environment will not predict the location of the firing field in the second environment.
- If a cell has a firing field in one environment, there is no predicable indication that the cell will have a firing field in the second environment
Although this is a strong operational definition, a critical feature of remapping is that the topology is scrambled. Across environments, pairs of cells can have their firing fields go from overlapping to non-overlapping and vice versa. By the early 1990s technology had advanced permitting simultaneous recordings from many neurons (present limit is about 100). Using improved techniques we were able to record from neuron pairs and directly observe scrambling. Cells 1 and 2 in the figure below were recorded simultaneously across two environments. (same cells as depicted above; figure re-arranged). The cell pair has overlapping firing fields in the cylinder and non-overlapping firing fields in the triangular enclosure. The topological map these cells were part of in the cylinder was clearly scrambled in the triangular enclosure. This spatial scrambling is reflected in the temporal firing patterns between the two cells. In the cylinder, these two cells fire together, in the triangle they never do. Temporal scrambling is illustrated in the two cross-correlation plots. The finding that the maps “scramble” when going from one environment to another is remapping. That is, there is a unique map in each environment.
Remapping is the way the hippocampus represents different environments. The hippocampal place cell system appears to operate at two scales: at the fine scale, individual place cells represent discrete locations; at a larger scale "maps" of place cells represent discrete environments 4.
What types of environmental changes are sufficient to induce remapping? Bob Muller and I found that changing the shape of an enclosure invariably induces full remapping. Subsequent work suggest that more modest changes, such as changing the the color of a large cue on the wall from white to black induces a slow but reliable remapping. There are indications that hippocampal maps may be effected by the reward contingencies in an environment, that maps may vary on a 24-hour cycle, and that slow map changes may occur over a 30-day span 5. In brief, the hippocampal map of space is sensitive to non-spatial factors. Understanding these factors will be essential in decoding the relationship between hippocampal maps and memory.
Making and Maintaining Maps
In the next section I'll discuss connections between place cell maps and memory. Before making the direct connection, its useful to establish parallels between the formation and maintenance of place cell maps and the formation and maintenance of memory. Several dramatic parallels have been established. A critical feature of memory is that it is acquired through experience. What of place cell maps? Are they present on first exposure to an environment or are they acquired? The initial demonstration that place cell maps are "learned" was the Bostock et al study, cited above. Rats were trained in a cylinder with a white cue card. On testing day, a rat (with its recorded place cell) was introduced to the same cylinder, but with a black cue replacing the white cue. The finding was that place cells remapped, but that remapping took time, ranging from minutes to days3. A second critical finding was reported by Cliff Kentros, working with Bob Muller and colleagues from Eric Kandel's lab. Cliff's experiment involved injecting rats with an NMDA blocker during an exposure to a novel environment. This manipulation is known to disrupt LTP and memory formation. They found that place cell maps were set up in the environment, but did not last when tested the next day6. This parallels studies on memory. Finally, a recent paper by Barry et al (working in Bob Muller's lab) demonstrated that infusions of zeta inhibiting protein (ZIP) into the hippocampus, a manipulation that disrupts spatial memory, caused remapping of an otherwise stable place cell representation7. In brief, hippocampal maps are learned, and mechanisms that disrupt memory disrupt place cell maps.
Maps and Memory
Two threads link the hippocampus to cognition and behavior. First, partially described in this article, is the hippocampal role in representing space — the so-called "cognitive map". The second thread is the role of the hippocampus in memory encoding and retrieval. This was dramatically revealed in the study of amnestic patient HM. There is a likely link between these two functions.
Psychologists have long argued that memories are stored and retrieved by a type of indexing system. Location may be the key index; that is, we index an autobiographical memory by its location. As a personal experiment, think of a place, then think of events that occurred in that place. For most of us, a single place cue will trigger recall of a set of specific memories that occurred in that location.
We suggest that the each hippocampal map is a discrete indexing cue. For the rat or human, retrieving a map of an environment is part of the process of retrieving specific memories that occurred in that environment. The multiple maps in the hippocampus, revealied by the observation of "remapping", suggests an ability to store and retrieve a vast number of memories.
Edvard Moser in an email describing the contributions of Bob Muller, summarizes the link between hippocampal maps, remapping and memory:
But most importantly, embedded in these (1987) papers was the discovery of place-cell remapping. This is a fundamental property of hippocampal neurons, which shows that hippocampal cells can participate in multiple seemingly independent representations, i.e. exactly what is needed for a system that can store large quantities of independent memories - it links place cells to memory.
Update, Oct 7: Place Cell Maps and Specific Memories. (responding to questions from twitter). The notion presented above is not that there is a different place cell map for each memory; rather that a set of episodic memories, sharing a spatial frame, are linked to (and encode by) a map. One can think of a specific episode as a path along a map — a time series of place cell activations. Considering what makes a map, one can see that a path can only occur on a specific map, since, at each transition along the path, "neighboring" place cells are activated ... and the neighborliness of the sets of cell pairs along a path is unique to a single map. Replay refers to the re-activation of a series of place cell activations that originally occurred as the rat moved along a path. A series of studies published over the past 20 years demonstrate that replay is a common occurrence, in a variety of forms. There is not space here to review these studies; an incomplete list of scientists studying replay includes Matt Wilson, Bruce MacNaughton, Gyorgi Buzsaki, David Redish, Loren Frank and George Dragoi. A recent review: Hippocampal Replay in the awake state (Carr et al, 2011)8 .
Postscript. This is intended as an introduction and overview to hippocampal remapping. For a more comprehensive treatment, including descriptions of global remapping, rate remapping and partial remapping see Colgin et al: Understanding Memory Through Hippocampal Remapping 9
References and Notes:
1 O'Keefe and Dostrovsky, The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Research, 1971 Nov;34(1):171-5.
O’Keefe, J., & Nadel, L. (1978). The Hippocampus as a Cognitive Map – Oxford Press 1978, 296. full text online.
2 Muller, R. U., & Kubie, J. L. (1987). The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. The Journal of neuroscience, 7(7), 1951–1968.
Muller, R. U., Kubie, J. L., & Ranck, J. B. (1987). Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. The Journal of neuroscience, 7(7), 1935–1950.
3 Kubie, J. L., & Muller, R. U. (1991). Multiple representations in the hippocampus. Hippocampus, 1(3), 240–242.
Bostock, E., Muller, R. U., & Kubie, J. L. (1991). Experience-dependent modifications of hippocampal place cell firing. Hippocampus, 1(2), 193–205.
4 The times window for synaptic strengthening is narrow, about 50 milliseconds. If a pair of neurons has overlapping spikes in the time scale depicted (10 seconds) there is no guarrantee the two cell will fire together within the narrow time window needed for synaptic changes.
5 Ziv, Y., Burns, L. D., Cocker, E. D., Hamel, E. O., Ghosh, K. K., Kitch, L. J., et al. (2013). Long-term dynamics of CA1 hippocampal place codes. Nature Neuroscience. Mar;16(3):264-6.
Mankin, E. A., Sparks, F. T., Slayyeh, B., Sutherland, R. J., Leutgeb, S., & Leutgeb, J. K. (2012). Neuronal code for extended time in the hippocampus. PNAC, 109(47), 19462–19467.
6 Kentros et al1998 Abolition of long-term stability of new hippocampal place cell maps by NMDA receptor blockade Science 280(5372): 2121-6, Jun 26
7 Barry et al (2012) Inhibition of protein kinase Mζ disrupts the stable spatial discharge of hippocampal place cells in a familiar environment. Oct 3;32(40):13753-62.
8 Carr, M. F., Jadhav, S. P., & Frank, L. M. (2011). Hippocampal replay in the awake state: a potential substrate for memory consolidation and retrieval. Nature Neuroscience, 14(2), 147–153. doi:10.1038/nn.2732
9 Colgin, Moser and Moser, (2008) Trends in Neurosciences, 31(9), 469-477
by John Kubie
The Society for Neuroscience and its partners are not responsible for the opinions and information posted on this page. Terms & conditions