Multisensory coding of angular head velocity in the retrosplenial cortex

Abstract

To successfully navigate the environment, animals depend on their ability to continuously track their heading direction and speed. Neurons that encode angular head velocity (AHV) are fundamental to this process, yet the contribution of various motion signals to AHV coding in the cortex remains elusive. By performing chronic single-unit recordings in the retrosplenial cortex (RSP) of the mouse and tracking the activity of individual AHV cells between freely moving and head-restrained conditions, we find that vestibular inputs dominate AHV signaling. Moreover, the addition of visual inputs onto these neurons increases the gain and signal-to-noise ratio of their tuning during active exploration. Psychophysical experiments and neural decoding further reveal that vestibular-visual integration increases the perceptual accuracy of angular self-motion and the fidelity of its representation by RSP ensembles. We conclude that while cortical AHV coding requires vestibular input, where possible, it also uses vision to optimize heading estimation during navigation.

Keywords: Angular head velocity; Head direction; Multisensory integration; Navigation; Optic flow; Retrosplenial cortex; Self-motion; Spatial orientation; Vestibular sense.

Figure 1

Extensive network of AHV cells in the RSP during free exploration (A) A video frame showing the open field arena used for exploration. Insets, video frames showing a left head turn (top three) and locomotion (bottom three) tracked using the position of ears and the body. Arrows indicate head direction (top three) and linear displacement (bottom three). Scale bar: 10 cm. (B) Top: schematic of a coronal brain section with RSP highlighted in gray, Bottom: inset shows a coronal 2P image of the Neuropixels probe track marked with DiI. Scale bar: 200 μm. (C) Schematic showing chronic recording sites in all 5 mice. Circles indicate the approximate location of cells tuned to AHV, speed, and heading direction (HD) within the dysgranular/agranular (light gray) and granular (dark gray) regions of the RSP. Numbers are distances from Bregma (in mm). All schematics and boundary outlines are based on the Allen Mouse Brain Atlas (Wang et al., 2020). (D) Tuning plots of four types of AHV cells: a unidirectional cell with positive speed correlation (top-left), a cell with opposite correlations (top-right), and two bidirectional cells with positive (bottom-left) and negative (bottom-right) speed correlations. (E) Tuning plots of two representative speed cells showing positive (left) and negative (right) correlations with locomotion speed. (F) An example HD cell in the RSP. Visual landmark was located at 180°. (G) Venn diagram showing cells tuned to HD, AHV, and linear locomotion speed (n total = 359). (H) Left: summary data (mean ± SEM, n = 5 mice, 12 recordings) showing percentage of cells that were tuned to HD, AHV, or locomotion speed. Right: pie chart represents proportion of AHV cells (n = 224) with bidirectional and unidirectional (right or left turn) tuning. See also Figures S1 and S2.

Figure 2

AHV cells maintain their tuning during restrained passive motion (A) Top: schematic of the design for passive rotation experiments. Centre and bottom: position and velocity profile of an individual rotation stimulus. (B) Tuning plots of a bidirectional AHV cell recorded initially in the open field (left) and subsequently during passive rotation (right) in darkness. Top: average trace of this unit’s spike waveform recorded under each condition. Circles and shaded area on the right show trial-averaged firing rates (12 trials) and SEM, respectively. (C) Firing rate at each velocity bin for the cell shown in (B) plotted for active versus passive conditions. ^∗ Pearson’s r >95^th percentile of the null distribution = similarly tuned. (D) Distribution of Pearson’s r of correlation between passive and active tuning curves for all tracked AHV cells in the dark (n = 90). (E) Summary data (mean ± SEM; Control: n = 10 mice, 20 recordings, 676 cells; Lesioned: n = 4 mice, 7 recordings, 313 cells) showing the percentage of cells with evoked responses (significant increase or decrease in firing rate relative to baseline) to rotation in darkness or to full-field visual flow. ^∗∗∗∗p = 2.8e-5, (ns) p = 1, Kruskal-Wallis with Dunn’s test. (F) Heatmaps show baseline-subtracted, normalized average firing rate as a function of rotation speed in darkness for controls (left) and lesioned animals (right). Each pair of rows represents an individual neuron’s response to clockwise (CW) and counter-clockwise (CCW) rotations. Neurons are sorted by the magnitude of speed correlations (Pearson’s r) averaged over the two directions. See also Figures S3–S6.

Figure 3

Vestibular and optic flow input converge onto AHV cells (A) Left: experimental design for vestibular stimulation. Traces show the position (black) and velocity (red) of the rotating platform. Centre: firing rate heatmaps of all tracked AHV cells during CW and CCW rotations in darkness. Rows show baseline-subtracted, normalized average firing rate of all neurons sorted by their direction selectivity index (area under the direction ROC curve; see STAR Methods). Right: baseline-subtracted, normalized average firing rate of all neurons as a function of rotation speed. Each pair of rows shows an individual neuron’s response to CW and CCW rotations. Neurons are sorted by the magnitude of speed correlations (Pearson’s r), averaged over the two directions. (B) Left: the rotation platform was stationary (black trace) while a surround vertical grating was rotated with the same velocity profile as the vestibular stimulus in (A) (red trace). Centre and right: same as in (A) but for the visual stimulus. (C) Left: tuning curves of an example AHV cell recorded in the open field in darkness and in light. Dashed lines represent linear fits. Right: for the same cell, firing rate difference between the first and all other velocity bins is plotted against the residual of the linear fit. (D) Population data showing linear fits for all AHV cells and for each direction in darkness and in light. Thick lines are averaged fits for either positive or negative correlations. (E) Summary data (median and 95% CI, n = 272 AHV cells) showing the overall change between dark and light conditions. ^∗∗∗∗p (slope) = 2.9e-11, ^∗∗∗∗p (Pearson’s r) = 1.2e-6, Wilcoxon signed rank test.

Figure 4

Combination of vestibular and visual stimuli improves perceptual accuracy of angular self-motion (A) Schematic of trial structure for the go/no-go speed discrimination task. The trace illustrates a generic rotation velocity profile (motion stimulus) for an individual trial. Stimulus duration = 32.2 s, inter-trial interval (ITI) = 5 s. (B) Schematic of the three experimental conditions. Mice either discriminated speed of self-motion in the dark (vestibular, left) or in presence of a static visual stimulus (vestibular + static visual, center). Under the third condition (visual, right), the rotation platform was kept stationary while the grating rotated to provide visual motion stimulation with the same speed profile used in the previous two conditions (center and bottom rows). (C) Accuracies of self-motion speed discrimination for all stimulus pairs (n = 5 mice, average of 5 blocks). Small and large circles represent individual animals and group averages, respectively. Lines are sigmoid fits. (D) As in (C) but for visual motion discrimination. (E) Mean (±SEM) discrimination accuracies of all stimulus pairs that were tested under the three conditions (20:10, 30:10, and 80:10) in all 5 mice. ^∗p = 0.014, ^∗∗∗∗p = 1.1e-5, one-way ANOVA with Holm-Sidak’s test. See also Figure S7.

Figure 5

Heterogenous vestibular-visual response properties of AHV cells (A) Left: normalized AUCs for a cell that discriminates direction of rotation in presence of visual signals (vestibular + visual) but not in the dark (vestibular). Right: normalized AUCs for speed ROC showing improved speed discrimination under the vestibular-visual condition. Transparent and solid error bars indicate 99% CIs within and above chance level, respectively. (B) Similar to (A), but for a different cell that discriminates the direction and speed of rotation only in the dark. (C) Scatterplot of AUCs for direction (left) and speed (right) discrimination. AUCs under vestibular-visual condition (rotation with static visual stimulus) are plotted against AUCs under vestibular condition (rotation in the dark). Each circle represents one tracked AHV cell (n = 120). Direction ROC curves compared the firing rates between CW and CCW rotations. Speed ROC curves compared the firing rates between the speed bin that peaked at 10°/s and successive speed bins (15°/s–80°/s peak). Grey circles show non-discriminating cells. Dark blue and orange represent, respectively, cells that discriminated direction/speed exclusively in the dark (vestibular only), or only when visual stimuli were available (vestibular + static visual only). Light blue indicates significant discrimination under both conditions. (D) Summary data (median and 95% CI) showing percentage of cells that discriminated speed of rotation (10°/s v. 80°/s) under each condition. ^∗p = 0.02, Wilcoxon signed rank test. (E) Summary data (median and 95% CI) showing percentage of cells that discriminated direction of rotation under each condition, using either the first 500 ms (left) or the entire rotation window (right, 3.5 s). ^∗∗∗p = 3e−4, (ns) p = 0.73, Wilcoxon signed rank test.

Figure 6

Combination of vestibular and visual stimuli improves decoding of angular self-motion by AHV cell populations (A) LDA direction decoding accuracy (mean + IQR) in darkness as a function of population size for controls and vestibular-lesioned animals. (B) LDA speed decoding accuracies (10°/s v. 15°/s – 80°/s) in darkness with increasing population size for controls and vestibular-lesioned animals. (C) Left: LDA decoding accuracy (mean + IQR) for direction of self-rotation (blue and orange) and visual motion (green) as a function of AHV population size. Only the initial 500 ms of stimuli was considered. Right: mean (±SEM) decoding accuracy (5 pseudo-populations, 10 – 120 pooled neurons). ^∗∗∗p (vestibular v. vestibular + visual) = 4.2e-4, ^∗∗∗p (visual v. vestibular + visual) = 4.2e-4, one-way ANOVA with Holm-Sidak’s test. (D) Left: LDA decoding accuracies for speed of self-rotation (blue and orange) and visual motion (green) using all 120 AHV cells pooled into a pseudo-population. Lines are sigmoid fits. Right: mean (±SEM) decoding accuracy from all speed pairs (5 pseudo-populations, 10 – 120 pooled neurons). ^∗∗p (vestibular v. vestibular + visual) = 0.007, ^∗∗∗p (visual v. vestibular + visual) = 0.0004, one-way ANOVA with Holm-Sidak’s test. See also Figure S8.