A neural-level model of spatial memory and imagery

Abstract

We present a model of how neural representations of egocentric spatial experiences in parietal cortex interface with viewpoint-independent representations in medial temporal areas, via retrosplenial cortex, to enable many key aspects of spatial cognition. This account shows how previously reported neural responses (place, head-direction and grid cells, allocentric boundary- and object-vector cells, gain-field neurons) can map onto higher cognitive function in a modular way, and predicts new cell types (egocentric and head-direction-modulated boundary- and object-vector cells). The model predicts how these neural populations should interact across multiple brain regions to support spatial memory, scene construction, novelty-detection, ‘trace cells’, and mental navigation. Simulated behavior and firing rate maps are compared to experimental data, for example showing how object-vector cells allow items to be remembered within a contextual representation based on environmental boundaries, and how grid cells could update the viewpoint in imagery during planning and short-cutting by driving sequential place cell activity.

Spatial coordinate transforms linking the allocentric hippocampal and egocentric parietal primate brain systems for memory, action in space, and navigation

“…The representations that are produced in the research described here are to spatial frameworks with head‐centered coordinates, body‐centered coordinates for reaching into space, bearing direction to a landmark, and then to allocentric spatial view cells that respond to viewed location in the world independently of the place where the viewer is. Moreover, in the present approach, evidence is provided that the coordinate transform with gain modulation by head direction takes place in parietal areas such as area 7a (Snyder et al, ) and is represented in the primate posterior cingulate cortex (Dean & Platt, ; whereas in the model of Bicanski and Burgess () the retrosplenial cortex is emphasized, see their Figure , though the retrosplenial cortex is implicated in memory and navigation; Vann et al, ); holds that this implements bearing direction to a landmark in a spatial scene; and goes on to show that representations in allocentric spatial view coordinates suitable for interfacing to primate allocentric spatial view representations can be produced using gain modulation by place (which was not part of the previous model). A further difference is that in the model of Bicanski and Burgess () the representations produced seem to be of the position of objects in an allocentric space.…”

“…It is of interest to compare the present model with a recent model (Bicanski & Burgess, 2018) centered on the retrosplenial cortex. This previous model is directed primarily toward spatial memory and imagery;…”

“…The previous model (Bicanski & Burgess, 2018) also utilizes object-vector cells found in rodents (Hoydal, Skytoen, Andersson, Moser, & Moser, 2019). (Indeed, the description of what Bicanski and Burgess (2018) model is: "Perceived and imagined egocentric sensory experience is represented in the "parietal window" (PW), which consists of two neural populations -one coding for extended boundaries ("PWb neurons"), and one for discrete objects ("PWo neurons").) Moreover, that model holds that "the transformation between egocentric (parietal) and allocentric (medial temporal lobe, MTL) reference frames is performed by a gain-field circuit in retrosplenial cortex" and uses head direction.…”

“…In previous work, the importance of coordinate transforms utilizing allocentric and egocentric representations for spatial navigation in primates including humans has been described though without a formal neuronal network theory and model (Ekstrom, Huffman, & Starrett, 2017). A previous model for coordinate transforms between egocentric and allocentric coordinates holds that this is performed in the retrosplenial cortex, is described in the context of a model of spatial memory, imagery, and so forth, and depends on neurons such as allocentric boundary-vector cells found in rodents with consistent human fMRI evidence (Shine, Valdes-Herrera, Tempelmann, & Wolbers, 2019) but not shown at the neuronal level in primates, and object-vector cells, and does not model the series of stages of the highly developed primate dorsal visual system leading to the parietal cortex with coordinate transforms starting with retinal coordinates (Bicanski & Burgess, 2018;Burgess & Hartley, 2001;Byrne, Becker, & Burgess, 2007;see Section 5). Furthermore, the homology between the well-defined primate retrosplenial cortex of primates (Kobayashi & Amaral, 2003) and what is described as retrosplenial cortex in rodents is not clear, and there may be no posterior cingulate cortex in rodents (Vogt, 2009).…”

Navigating with grid and place cells in cluttered environments

“…With regard to spatial navigation, other strategies such as taxis and landmark‐based navigation (Trullier et al, ) are also known to guide an animal's behavior, and should be incorporated for a more complete navigation model. Finally, look‐ahead and mental navigation could also interact with the combined vector–place strategy proposed here (Bicanski & Burgess, ; Erdem & Hasselmo, , ). In mental navigation, simulated motion (potentially driven by mock motor efference and conveyed by grid cells; Bellmund, Deuker, Schröder, & Doeller, ; Horner, Bisby, Zotow, Bush, & Burgess, ) can be thought of as accompanied by a reinstatement of sensory representations bound to locations (via place cells) along the imagined trajectory (Bicanski & Burgess, ), and would hence be particularly useful if planning involves particular sensory aspects along the route.…”