Elsevier

Neurocomputing

Volume 69, Issues 10–12, June 2006, Pages 1260-1263
Neurocomputing

Place field dissociation and multiple maps in hippocampus

https://doi.org/10.1016/j.neucom.2005.12.088Get rights and content

Abstract

When two sets of visual cues move in discordant fashion, the firing fields of hippocampal place cells can dissociate (follow different sets), split (follow both sets), or remap. This conflicts with views of hippocampus as an associative memory that “completes” patterns given a noisy input. We reconcile these views by treating hippocampus as a collection of independent “maplets”. Normally these act in unison, but under the right conditions they can exhibit discordant behavior. The model also accounts for “barrier cells” [B. Rivard, Y. Li, P.-P. Lenck-Santini, B. Poucet, R.U. Muller, Representation of objects in space by two classes of hippocampal place cells, J. Gen. Physiol. 124 (2004) 9–25], and the occasional temporary cessation of firing of some place cells while others remain active.

Introduction

Attractor network models of the hippocampus usually assume that recurrent connections in CA3 are initially random with a uniform distribution [3]. As a result of experience, cells with firing fields that are nearby (on some map) fire in a correlated fashion and have their mutual connections strengthened.

The attractor model explains the robustness of place cell firing fields when visual cues are removed or the lights are turned off: the attractor dynamics are strong enough to recreate the learned activity patterns of the map even in the absence of the usual sensory cues [6]. But this account conflicts with another set of observations. Discordant motion of local vs. distal cues in a double cue rotation experiment causes place fields to dissociate. Some fields rotate with the local cues, some with the distal, some show split fields, and some remap [2], [7]. The simplest explanation for the dissociation, that individual cells are tuned to only a few visual landmarks each, is contradicted by the robustness of firing fields after cue deletion. Robustness implies that cells must either be driven by a broad array of landmarks, or have attractor dynamics strong enough to overcome any missing inputs. But strong attractors should force a unified response, preventing place field dissociation. On the other hand, if attractors are weak but cells are driven by a broad array of cues, then the majority should have roughly equal input from local and distal landmarks, making split fields more common than is actually observed.

To resolve this paradox, we propose an attractor map architecture composed of many overlapped “maplets” that are essentially independent. Place cells receive input from a broad array of landmarks, but when cues are rotated into a discordant configuration, strong attractor dynamics amplifies any disparity in local vs. distal cue input, causing each maplet to choose one set of cues to follow. When two overlaid maplets make different choices, their place fields dissociate. If a maplet switches from local to distal cues or vice versa as the rat travels through the environment, split place fields will be observed.

The architecture also explains the existence of barrier cells [5] whose fields move along with a nearby barrier when it is displaced, while other cells remain tied to the room frame. In addition, some firing fields have been observed to shut off for periods ranging from a few seconds to an hour or more and then switch back on, while others remain active continuously. Cells that drop in and out as a group may be part of the same maplet, whose activity bump collapsed and later reformed.

Section snippets

Double cue rotation

In a double cue rotation experiment, rats forage for food on a ring or plus-shaped maze whose surface is marked with local sensory cues. A set of distal cues is positioned on the curtains surrounding the maze. During probe trials, the local cues are rotated by 90 in one direction and the distal cues by 90 in the opposite direction, putting them 180 out of phase, as shown in Fig. 1.

Place field dissociation and splitting, observed in double cue rotation probe trials, are not compatible with

Barrier cells

Rivard et al. described place cells whose firing fields in a cylindrical arena translated with a movable barrier, while other cells recorded at the same time remain tied to the cylinder [5]. Barrier cells can be explained by having a small number of maplets biased toward cues associated with the removable barrier. These cells shut off when the barrier is removed, rather than switching to follow another cue set. Some barrier cells are also place-specific. These can be explained by maplets

Cells dropping out/in

During the course of a recording session we have observed place cells ceasing to fire for a period, remaining inactive when the rat entered their firing fields, but later becoming active again. During this time, other cells remained active continuously. This is difficult for a unified attractor model to explain. A maplet architecture can produce this effect if the bump collapses in one maplet and later forms again, while the bumps on the other maplets remain active and continue to generate

Discussion

Why does the rodent hippocampus have so many place cells? The usual answer is: to store many maps without interference. We suggest an alternative: the large number of place cells provides for many overlaid maplets. These allow the rat to simultaneously represent its position with respect to its environment and nearby objects within that environment (barrier cells). They also provide a representation for ambiguous situations, as in a double cue rotation, by representing alternative frames of

Acknowledgments

Supported by NIH MH59932 and NS20686. We thank Mark Fuhs for helpful discussions.

David S. Touretzky is a Research Professor of Computer Science at Carnegie Mellon University. He is interested in neural representations of spatial information, such as the “cognitive maps” found in the hippocampus. Dr. Touretzky has also done modeling of the rodent head direction system.

References (7)

There are more references available in the full text version of this article.

Cited by (0)

David S. Touretzky is a Research Professor of Computer Science at Carnegie Mellon University. He is interested in neural representations of spatial information, such as the “cognitive maps” found in the hippocampus. Dr. Touretzky has also done modeling of the rodent head direction system.

Robert U. Muller has have been fascinated by the operation of the rat hippocampus since he saw a place cell being recorded more than 20 years ago. Dr. Muller's interests are understanding place cell activity and how it is related to mechanisms of synaptic plasticity local to the hippocampus. On a larger scale, he would like to understand how place cells serve as components of the rat “navigational system,” how this system allows rats to solve spatial problems, and what it teaches about neural implementations of cognitive processes.

View full text