Entorhinal-hippocampal interactions lead to globally coherent representations of space

Abstract

The firing maps of grid cells in the entorhinal cortex are thought to provide an efficient metric system capable of supporting spatial inference in all environments. However, whether spatial representations of grid cells are determined by local environment cues or are organized into globally coherent patterns remains undetermined. We propose a navigation model containing a path integration system in the entorhinal cortex and a cognitive map system in the hippocampus. In the path integration system, grid cell network and head direction (HD) cell network integrate movement and visual information, and form attractor states to represent the positions and head directions of the animal. In the cognitive map system, a topological map is constructed capturing the attractor states of the path integration system as nodes and the transitions between attractor states as links. On loop closure, when the animal revisits a familiar place, the topological map is calibrated to minimize odometry errors. The change of the topological map is mapped back to the path integration system, to correct the states of the grid cells and the HD cells. The proposed model was tested on iRat, a rat-like miniature robot, in a realistic maze. Experimental results showed that, after familiarization of the environment, both grid cells and HD cells develop globally coherent firing maps by map calibration and activity correction. These results demonstrate that the hippocampus and the entorhinal cortex work together to form globally coherent metric representations of the environment. The underlying mechanisms of the hippocampal-entorhinal circuit in capturing the structure of the environment from sequences of experience are critical for understanding episodic memory.

Keywords: Global representations; Grid cells; Head direction cells; Local environment anchors; Simultaneous localization and mapping.

Graphical abstract

Fig. 1

Globally coherent encoding of space requires long-term exploration. (A) Spatial representations in a complex maze. Initially, the grid cell firing pattern is determined by local sensory cues (bottom left). After experience-dependent correction during the traverse of the environment, the grid cell firing pattern becomes globally coherent (bottom right). (B) Spatial representations on a road. Before loop closure by revisiting familiar places, the grid pattern is locally anchored due to path integration errors (bottom left). After loop closures, globally coherent pattern is formed (bottom right).

Fig. 2

The software architecture of the SLAM system. Images and odometry information are provided by the sensor/bagfile node. Whether the current view is familiar or not is determined by the local view cells node. The grid cell model and the HD cell model are implemented in the Bayesian attractor network node, which performs path integration and makes decisions of loop closures. The experience map node achieves the global graph optimization of the experience map.

Fig. 3

Screenshots of SLAM system for iRat Australia dataset. (A) Neural activities of HD cells; (B) Overhead image; (C) Input visual scene (top); the local view template and the matched template (bottom); (D) Neural activities of grid cells; (E) Experience map.

Fig. 4

Map calibration in hippocampus and activity correction in MEC during loop closures are necessary for the formation of globally coherent firing maps. (A) Inconsistent representations without map calibration. Top: The summed firing map of the HD cells with north or south preference in the ring manifold. Bottom: The firing map of an example grid cell at (0,0) in the torus manifold. (B) Locally anchored representations. Top: Without firing activity correction, the summed firing rate map of the HD cells at 0 and π in the ring manifold anchors to local space, preferring particular directions at different part of the environment. Bottom: The firing map of the grid cell at (0, 0) in the torus manifold has multiple firing fields, which are anchored to a local region on a rectangular grid. (C) Globally coherent representations. Top: With firing activity correction, the summed firing rate map of the HD cells at 0 and π in the ring manifold has strong activity along the trajectories orienting towards east and west, showing global coherency across the environment. Bottom: The globally coherent grid pattern of the grid cell at (0,0) in the torus manifold has regularly spaced firing field across the whole environment. In each panel, firing rate is plotted at the locations in the experience map. The colorbar shown to the right of each panel color-codes peak firing rate with red and silent activity with blue. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 5

HD cells show globally coherent directional selectivity with firing activity correction. (A) Without firing activity correction, the distributions of the HD cell activity are not localized. Two cells with opposite preferred directions are shown in columns. Top row depicts firing rate as a function of HD. Bottom row plots firing rate histograms. (B) Within each quarter of the exploration time, The activity of the HD cell at 0 concentrates on multiple clusters. Each panel shows one fourth of the total firing activity. Red color encodes peak activity and blue encodes zero activity. Some clusters can be fitted by Gaussian functions (red solid lines). (C) With firing activity correction, the distributions of the HD cell activity follow bell-shaped distributions during the whole exploration (top). The firing rate histograms in polar coordinates reveal strong directional selectivity (bottom). (D) Within each quarter of the exploration time, the HD cell at 0 keeps its preferred direction stable. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 6

Globally coherent grid maps have high degree of symmetry. (A) Firing rate maps; (B) Autocorrelograms of the firing maps; (C) Rectangular gridness score estimation. Each row shows the computation of gridness score for grid maps without map calibration (top), without activity correction (middle) and with activity correction (bottom).

Fig. 7

With firing activity correction, grid cells maintain globally coherent grid maps. Each row shows one example grid cell for the three conditions respectively. Left column: firing rate maps of the example grid cells during the first half of exploration. Middle column: firing rate maps of the example grid cells shown in the left during the second half of exploration. Right column: the crosscorrelograms between the firing rate maps of the first half and the second half of each cell. With firing activity correction (bottom row), the crosscorrelogram of the example grid cell has peek in the center and high degree of spatial symmetry, confirming stable grid codes over time. Without loop closure or firing activity correction (two rows on the top), the central peaks of the crosscorrelograms are weak, demonstating the loss of stable or globally coherent grid patterns.

Fig. 8

Development of the firing map of a grid cell.. (A), (B), (C) and (D) show the firing map of one example grid cell for each interval of one quarter of the exploration. During exploration, the grid map develops and is maintained coherent globally by correcting localization through the feedback from cognitive map.

Fig. 9

Global coherent representations with various phases. (A) Global coherent HD representations. The firing maps of four example HD cells are shown. The cells are at 0, $\frac{\pi }{2}$, π, and $\frac{3\pi }{2}$ in the ring manifold from top to bottom. (B) Global coherent grid representations. Each panel shows the firing map of one example grid cells. From top to bottom, the grid cells are at (0, 0), $(\frac{\pi }{2},\frac{\pi }{2})$, $(\frac{3\pi }{2},\pi )$, and $(2\pi ,\frac{3\pi }{2})$ in the manifold.

Fig. 10

Global coherent grid maps with various spacings. (A), (B), and (C) show firing maps of three grid cells from three grid modules, with grid spacing 0.5m, 1.0m, and 1.5m, respectively.

Fig. 11

The interaction between the path integration system in the entorhinal cortex and the cognitive map system in the hippocampus.