Multiplexed code of navigation variables in anterior limbic areas

Abstract

The brain’s navigation system integrates multimodal cues to create a sense of position and orientation. Here we used a multimodal model to systematically assess how neurons in the anterior thalamic nuclei, retrosplenial cortex and anterior hippocampus of mice, as well as in the cingulum fiber bundle and the white matter regions surrounding the hippocampus, encode an array of navigational variables when animals forage in a circular arena. In addition to coding head direction, we found that some thalamic cells encode the animal’s allocentric position, similar to place cells. We also found that a large fraction of retrosplenial neurons, as well as some hippocampal neurons, encode the egocentric position of the arena’s boundary. We compared the multimodal model to traditional methods of head direction tuning and place field analysis, and found that the latter were inapplicable to multimodal regions such as the anterior thalamus and retrosplenial cortex. Our results draw a new picture of the signals carried and outputted by the anterior thalamus and retrosplenial cortex, offer new insights on navigational variables represented in the hippocampus and its vicinity, and emphasize the importance of using multimodal models to investigate neural coding throughout the navigation system.

The retrosplenial cortex combines internal and external cues to encode head velocity during navigation

“…S1B), and consisted of both putative excitatory (121/224) and inhibitory (103/224) cell types ( Fig. S3), suggesting that AHV representation in the RSP is widespread and more prevalent than previously reported (30,51). To distinguish the contribution of various external and internal cues to AHV signalling, we tracked the same unit from the open field arena to a body-and head-restrained paradigm ( Fig.…”

A gravity-based three-dimensional compass in the mouse brain

“…The function of presubicular HD cells likely extend beyond relaying ADN HD signals, as indicated by the presence of egocentric information in presubicular but not ADN HD cells 30 and the importance of the presubiculum for visually anchoring the HD network 31 . The RSC is involved in visual processing, often hypothesized to transform visual landmarks from an egocentric to an allocentric reference frame 32,33 , and RSC HD cells may combine HD signals with visual 34 or egocentric spatial information 35,36 . Our findings (and Kim and Maguire's study 27 ) raise the possibility that the RSC may use gravity-referenced tilt signals to transform visual signals in 3D.…”

Spatial modulation of hippocampal activity in freely moving macaques

“…One issue with the traditional analyses is that a neuron may appear to encode a particular variable when other more relevant variables are excluded, therefore masking the actual, more multiplexed coding schemes. Several recent studies have emphasized the advantages of using multimodal models to characterize mixed selectivity, which are agnostic to tuning curve shape and robust to the interdependence of encoded variables (Hardcastle et al, 2017; Laurens et al, 2019).…”

“…At the cellular level, multiple cell types have been identified to encode position, head direction, speed and other spatial variables in the rodent HF (Hafting et al, 2005; McNaughton et al, 1983; O’Keefe and Dostrovsky, 1971; Taube et al, 1990), some with 3D properties (Angelaki et al, 2020; Grieves et al, 2020), that has also been found in the bat HF (Finkelstein et al, 2015; Yartsev and Ulanovsky, 2013). Recent studies have employed multimodal models to reveal multiplexed representations in the rodent HF (Hardcastle et al, 2017; Laurens et al, 2019; Ledergerber et al, 2020). A fundamental advantage of such models is that they can correctly identify multimodal responses even for variables that are correlated and interdependent, while proved remarkably immune to pitfalls such as overfitting; whereas traditional methods often fail to quantify neurons with mixed selectivity (Laurens et al, 2019).…”

“…Recent studies have employed multimodal models to reveal multiplexed representations in the rodent HF (Hardcastle et al, 2017; Laurens et al, 2019; Ledergerber et al, 2020). A fundamental advantage of such models is that they can correctly identify multimodal responses even for variables that are correlated and interdependent, while proved remarkably immune to pitfalls such as overfitting; whereas traditional methods often fail to quantify neurons with mixed selectivity (Laurens et al, 2019).…”