Mapping vestibular and visual contributions to angular head velocity tuning in the cortex

Abstract

Neurons that signal the angular velocity of head movements (AHV cells) are important for processing visual and spatial information. However, it has been challenging to isolate the sensory modality that drives them and to map their cortical distribution. To address this, we develop a method that enables rotating awake, head-fixed mice under a two-photon microscope in a visual environment. Starting in layer 2/3 of the retrosplenial cortex, a key area for vision and navigation, we find that 10% of neurons report angular head velocity (AHV). Their tuning properties depend on vestibular input with a smaller contribution of vision at lower speeds. Mapping the spatial extent, we find AHV cells in all cortical areas that we explored, including motor, somatosensory, visual, and posterior parietal cortex. Notably, the vestibular and visual contributions to AHV are area dependent. Thus, many cortical circuits have access to AHV, enabling a diverse integration with sensorimotor and cognitive information.

Keywords: angular head velocity; calcium imaging; head direction cells; posterior parietal cortex; primary visual cortex; retrosplenial cortex; secondary motor cortex; spatial navigation; two-photon microscopy; vestibular and visual processing.

Graphical abstract

Figure 1

Rotation-selective neurons in the retrosplenial cortex (RSC) (A) Cartoon of a head-fixed mouse in an arena with visual cues. The mouse and visual cues can be rotated together or independently while performing two-photon microscopy. (B) Top-view schematic: The mouse is randomly rotated between three target directions (dashed lines) but only gets a water reward in one direction. The circle represents the arena wall with white cues on a black background. (C) Block diagram of the sequence of events for a single trial. (D) (Left) Confocal fluorescence image of RSC coronal brain slice with cortical layers indicated (Thy1-GCaMP6s/ GP4.3 mice). (Right) Top view of cranial window implanted on dorsal (agranular) RSC and centered over the central sinus. Black box shows the typical size and location of an imaging field of view (FOV). (E) Two-photon image sequence of layer 2/3 neurons during rotation experiment (left) and when de-rotated during postprocessing (right). (F) Example fluorescence traces (fractional change of fluorescence [DF/F]) of a clockwise (CW) and counter-clockwise (CCW) selective cell. Ticks indicate deconvolved DF/F events. Blue and red bars are CW and CCW rotation trials, respectively. Polar histograms show the DF/F events as a function of direction (the arena wall with visual cues is superimposed). (G) Logarithmic plot of the fraction of rotation-selective cells at different significance levels (average ± SEM, 1,368 cells, 4 mice, 4 FOVs). (H) Distribution of rotation-selectivity index for all rotation-selective (p ≤ 0.05) and non-selective cells (average ± SEM). (I) Average activity of all rotation-selective cells separated in CW and CCW rotations. (J) Example FOV showing the spatial distribution of CW- (blue) and CCW-selective cells (red), for p ≤ 0.05.

Figure 2

Many rotation-selective neurons are biased by head direction (A) Example of a CW-rotation-selective cell that is biased by head direction. (Top) Deconvolved DF/F events (ticks) during CW and CCW rotation trials. White background indicates angular positions that were visited during a trial, and gray background indicates angular positions that were not visited. (Middle) Event histogram for the whole session. (Bottom) Polar histograms of events (bin size = 10° and smoothened using moving average over 5 bins). Dashed vertical lines indicate the stationary positions. (B) Same as (A) for a CCW example cell. (C) Percentage of CW or CCW cells that are biased by head direction (HD; 1,368 cells, 4 mice, 4 FOVs). (D) (Top) Average response of all CW cells that are significantly biased by head direction, sorted according to their preferred head direction. (Bottom) Average response of all CW cells. (E) Same for all CCW cells.

Figure 3

A large fraction of rotation-selective neurons encode AHV (A) (Top) Experimental configuration. Mouse is rotated 180° back and forth in CW and CCW directions and at different rotation speeds. The speed profiles are colored blue (CW) and red (CCW). (Bottom) Block diagram of the sequence of events for a single trial. The speed is randomly chosen for every trial. (B) Response (average rate ± SEM) of all cells classified as CW (top) or CCW selective (bottom) and separated by rotation speed (4,928 cells, 10 mice, 14 FOVs). (C) Response (average rate ± SEM) as a function of angular velocity, for CW (blue) and CCW (red) rotations. Same data as in (B). Only the first 1 s was analyzed, which is the duration of the fastest rotation. (D) Response reliability (average ± SEM) as a function of angular velocity, for CW (blue) and CCW (red) rotations. Same data as in (B). (E) (Left) Responses of an example CW cell for a whole session (deconvolved DF/F events). Trials are sorted by angular velocity. The trial starts at time = 0. White background indicates that the mouse is rotating, and a gray background indicates that the mouse is stationary. (Right) Tuning curve showing average rate (±SEM) as a function of angular velocity. (F) Cartoon of linear non-linear Poisson (LNP) model. The model input variables rotation direction, and velocities are converted into a weight parameter. Then, they are converted by a non-linear exponential into a Poisson rate and matched to the observed data (see STAR Methods). (G) Percentage of cells (average ± SEM) classified by the LNP model as rotation selective (CW and CCW cells) or tuned to AHV, as a function of significance threshold. Pie chart shows the percentage of classified cells using p ≤ 0.05. (H) AHV tuning curve examples of three cells using the LNP model (average ± SEM). (I) The average ± SEM tuning curve of all AHV cells classified by the LNP model.

Figure 4

AHV can be decoded from neural activity in the RSC (A) Example trial sequence showing a mouse being rotated in the CW or CCW direction. The periods when the mouse is stationary are removed. Traces show the actual rotation direction and that predicted by the LNP model. For decoding, all neurons in a FOV were used regardless of whether they were rotation selective or not. (B) Same trial sequence as shown in (A) but now showing the angular head velocity (AHV). (C) Scatterplot showing the actual AHV of a trial and that predicted by the LNP model. All trials of all mice are included (16 sessions, 1,102 trials, 10 mice). The size of a bubble indicates the number of trials. The number of neurons used for decoding was 308 ± 15 (average ± SEM) per session. (D) The percentage of trials (average ± SEM) correctly decoded by the LNP model. (E) The decoding error (average ± SEM) using the LNP model as a function of actual rotation velocity. (F) The decoding error as a function of the number of random neurons used for decoding.

Figure 5

AHV tuning in the RSC depends on sensory transduction in the vestibular organs (A) Experimental configuration. (Left) Mouse is rotated 180° back and forth in CW and CCW directions and at different rotation speeds. (Middle) Same experiment repeated in the dark. (Right) Mouse is stationary, but now the wall of the arena with visual cues is rotated to simulate the visual flow experienced when the mouse is rotating. (B) (Top row) Responses of an example cell for a whole session (deconvolved DF/F events) under the three different conditions indicated in (A). Trials are sorted by velocity (see color code in Figure 3A). White background indicates that the mouse is rotating, and a gray background indicates that the mouse is stationary. (Bottom row) angular velocity tuning curves showing average rate (± SEM) as a function of rotation velocity. (C) Response (average rate ± SEM) of all cells classified as CW (top row) or CCW selective (bottom row) and separated by rotation speed. Total numbers of recorded cells are as follows: rotations in light = 4,928 cells from 10 mice, rotation darkness = 4,963 cells from 10 mice, and visual flow only = 3,192 cells from 7 mice). (D) Pie charts show the average percentage of classified cells using the LNP model in each of the three conditions indicated in (A) for p < 0.05. (E) The AHV decoding error (average ± SEM) as a function of rotation speed using the LNP model.

Figure 6

AHV is widespread in cortical circuits (A) (Left) Mouse is rotated 180° back and forth in CW and CCW directions and at different rotation speeds. Map of the mouse dorsal cortex with the major cortical areas color coded (Kirkcaldie, 2012). Trapezoid outline indicates the size of the implanted large cranial windows. Gray boxes indicate the typical size of an imaging FOV. (Middle) Trapezoid cranial window centered on the central sinus with 4 example FOVs. (Right) Example two-photon image of a FOV from secondary motor cortex (M2) and RSC. Cell bodies are outlined. (B) (Left column) The percentage of rotation-selective (top) and AHV-tuned cells (bottom) across different cortical areas (average ± SEM). The baseline indicates chance level. (Right column) Statistical comparison between cortical areas (asterisk indicates pairwise comparisons with p < 0.05, Wilcoxon-Mann-Whitney test). Numbers of cells and mice recorded per area are as follows: V1, 2,463 cells from 8 mice; V2, 4,529 cells from 10 mice; RSC, 4,928 cells from 10 mice; PPC, 1,044 cells from 4 mice; M1+M2, 2,483 cells from 7 mice; and S1, 3,356 cells from 9 mice. (C) The percentage of rotation-selective (left) and AHV-tuned cells (right) mapped onto the mouse dorsal cortex. (D) Example AHV tuning curves obtained using the LNP model from two example areas M1+M2 and V1. (E) The AHV decoding error (average ± SEM) as a function of actual rotation velocity, for each of the cortical areas, by using the LNP model. (F) The decoding error as a function of the number of random neurons used for decoding.

Figure 7

Whether AHV tuning depend on head motion or visual flow is area dependent (A) Experimental configuration. (Left) Mouse is rotated 180° back and forth in CW and CCW directions and at different rotation speeds. (Middle) Same experiment repeated in the dark. (Right) Mouse is stationary, but now the wall of the arena with visual cues is rotated to simulate the visual flow experienced when the mouse is rotating. (B) Percentage of angular velocity-tuned neurons mapped onto the mouse dorsal cortex, tested under each of the three conditions. (C) Percentage angular velocity-tuned cells across different cortical areas (average ± SEM) tested under each of the three conditions. (D) The decoding error (black, average ± SEM) and percentage of angular velocity-tuned cells (color coded) under the three different conditions. The horizontal bars above show the Mann-Whitney unpaired test p values. The dashed line shows the p = 0.05 level for the decoding error. The numbers of cells and mice recorded under light conditions are given in the legend of Figure 6. Under dark conditions, results were as follows: V1, 2,385 cells from 8 mice; V2, 4,564 cells from 10 mice; RSC, 4,963 cells from 10 mice; PPC, 1,044 cells from 4 mice; M1+M2, 2,489 cells from 7 mice; and S1: 3,398 cells from 9 mice. During wall rotations, results were as follows: V1, 1,864 cells from 6 mice; V2, 3,560 cells from 8 mice; RSC, 3,192 cells from 7 mice; PPC, 816 cells from 3 mice; M1+M2, 1,095 cells from 4 mice; and S1: 1,562 cells from 6 mice.