Coregistration of heading to visual cues in retrosplenial cortex

Abstract

Spatial cognition depends on an accurate representation of orientation within an environment. Head direction cells in distributed brain regions receive a range of sensory inputs, but visual input is particularly important for aligning their responses to environmental landmarks. To investigate how population-level heading responses are aligned to visual input, we recorded from retrosplenial cortex (RSC) of head-fixed mice in a moving environment using two-photon calcium imaging. We show that RSC neurons are tuned to the animal's relative orientation in the environment, even in the absence of head movement. Next, we found that RSC receives functionally distinct projections from visual and thalamic areas and contains several functional classes of neurons. While some functional classes mirror RSC inputs, a newly discovered class coregisters visual and thalamic signals. Finally, decoding analyses reveal unique contributions to heading from each class. Our results suggest an RSC circuit for anchoring heading representations to environmental visual landmarks.

Fig. 1. Individual neurons in the RSC exhibit heading responses in the absence of physical head movement.

A Schematic of the imaging set-up. Mice are head fixed in a floating chamber and allowed to voluntarily move the chamber floor using their paws, allowing control of their position and orientation within the chamber. Tracking information from the chamber is synchronized to the collection of the two-photon image acquisition. B Left: Example widefield image of the entire cortical window with the location of the imaged plane in dysgranular RSC indicated with a red box. Scale bar = 1 mm. Right: Example imaging plane in RSC using two-photon microscopy (n = 4 imaging fields over 3 mice) Scale bar = 200 µm. C Top: Neural responses are first extracted as changes in fluorescence (∆F/F), then spikes are inferred from the calcium traces. Gray traces show the ∆F/F trace and the green trace shows the inferred spikes. Inset: Zoomed section of the blue box on the left. Bottom: Example inferred spikes from a population of cells in a single recording. D Example chamber tracking data from a single experiment, showing X position, Y position, speed, and heading angle. E Example rate maps (top) and heading direction tuning curves (bottom) from neurons recorded from RSC. Note that neurons exhibit selectivity to heading, but not to allocentric position in the floating chamber. F Schematic of the controlled rotation set-up. A motorized wheel is added to control the rotation of the floating cage. The mouse is head-fixed in the center of the chamber and the X-Y translation of the chamber is restricted by guide bearings. The red arrows show the rotation of the wheel and chamber. G Left: Example of a single neuron’s response to rotation in the chamber. The orientation of the chamber (“heading”) is plotted against time, and the activity represented by the line color as indicated in the inset colorbar. The neuron consistently responds at the same heading across trials. Right: Average tuning curve across all trials, showing the elevated response of the neuron at a specific heading. Inset: The same tuning curve plotted in polar coordinates. H Tuning curves of all responsive neurons, sorted by cross-validated angle of peak response, showing that the preferred headings of the population span the entire environment.

Fig. 2. Neurons in the RSC retain similar tuning in head-rotation or chamber-rotation conditions.

A Example of the adapted head-rotation set-up for comparisons between head-rotation and chamber rotation experiments. In addition to the motor-driven wheel that can rotate the chamber while the head is immobilized, there is a separate rotation collar that can be used to rotate the mouse over the stationary chamber. Inset: Zoomed in image of the rotation collar assembly. The rotation collar contains an x-y translator to shift the axis of rotation to the center of the imaging field. A belt-drive connects the collar to a motor for controlled rotation. B Schematic of the two experimental conditions. Either the chamber or the head rotates independently, shown with a red arrow, while the other remains stationary, shown with a dashed gray capped line. In both cases, this causes the visual cue card to move across the visual field in the same direction. C Example of derotation process for image timeseries. Each of the three raw images is rotated to match the template on the right. Colored borders show the position and rotation of each individual image. The final template shown on the right is after all the images have been registered for the recording (n = 5 imaging fields over 4 mice). D Example of three matched ROIs from the recordings. Purple denotes greater brightness in the chamber rotation recording, whereas green denotes greater brightness in the head-rotation recording. White denotes equal brightness and suggests good alignment of the ROIs across recordings. HR: Heading rotation, CR: chamber rotation. E Three example tuning curves from pairs of ROIs showing similar tuning preferences. The tuning curve in the head rotation condition is shown in green, whereas the tuning curve in the chamber rotation condition is shown in purple. F Histogram showing the difference in peak heading preference across recordings for matched ROIs. The zero-centered peak in the data (green) indicates that cells remained tuned to the similar heading angles independent of head- or chamber-rotation. Shuffled distribution is shown in gray. ***p = 2.5 × 10⁻⁴, V-test for circular nonuniformity against 0°.

Fig. 3. The heading network coherently remaps when visual cues are changed.

A Schematic showing coherent remapping across two different contexts. Because each individual neuron’s preferred heading is mapped as a relative offset to visual landmarks, the relationships between neuron offsets would be expected to be preserved across contexts, leading to a coherent population shift. B Three example neurons from the same session showing a unified shift in preferred heading from context A to context B. Tuning curve from context A is shown in green, while tuning curve from context B is shown in purple. Arrow shows the phase offset between contexts. C Histogram showing that neurons coherently remap across conditions. For each recording session, a neuron’s coherence with the population was calculated as the difference between the neuron’s degree of remapping and averaged degree of remapping across the remainder of the population (see “Methods”). The peak around 0 suggests in the real data (purple) shows that individual neurons are shifting preferred directions in a similar amount to the rest of the population in each recording. Shuffled distribution is shown in gray. ***p = 4.9 × 10⁻⁶, V-test for circular nonuniformity against 0°.

Fig. 4. Imaging projections from ADN and visual cortex reveal distinctly tuned responses sent to RSC.

A Top: Example of injection sites in separate cohorts of mice. AAV-Syn-GCaMP7b was injected into either ADN or AM/PM of WT mice. Middle: Example imaging field in RSC of axon terminals originating from ADN (n = 17 imaging fields over 4 mice). Bottom: Example ∆F/F traces and inferred spikes for imaged ADN terminals it RSC. B Example schematic of experimental conditions. The single white cue card has been replaced with paired symmetric cues on the floating chamber, which is driven by a motorized wheel. Neurons from RSC are recorded across light-on and light-off phases. C Schematic of expected axonal responses. Left: Axons terminals encoding heading independent of visual cues are expected to have unimodal tuning curves across both conditions that are not affected by the light condition. Right: Axons terminals responding to visual cues are expected to show a single peak in a single cue condition, but dual peaks in the paired symmetric cue condition. In both cases, the responses will be reduced in the light-off condition. D Responses of ADN axons across recordings. Top: Heat maps in the light-on (left) and light-off (right) condition. Each row represents the aligned response of a single axon, and axons are organized by similarity to a group mean in the light-on condition. Bottom: Single axon response in light-on and light-off conditions (left). Averaged response across all ADN axons, showing an aligned unimodal peak in both light-on and light-off conditions (right). Traces shown are mean ± s.e.m. E Same as (D), but for visual cortex axons. Averaged response shows a bimodal peak in the light-on condition, and no response in the light-off condition.

Fig. 5. Unsupervised clustering of neural responses in the RSC reveals distinct functional classes.

A Heat maps of the responses of all somas imaged in RSC. Each row corresponds to a neuron in the light-on and light-off condition, and responses are ordered by their preferred direction. Light-off responses are ordered based on their light-on position. B Left: Example of a two-term Gaussian fit on an aligned and normalized tuning curve. The outputs of the two-term Gaussian for light-on and light-off conditions are used for clustering. Right: Scatter plot of clustered responses in a reduced dimension (LDA space). Histograms along the edges show the distribution of values along each dimension. Cluster numbers are denoted by color. Note that LDA plot is only for visualization, clustering was performed in full 8-dimensional space (see “Methods”). C Top: Heat maps showing the aligned responses of the heading neurons (cluster 1), ordered by similarity to a cluster averaged response. Bottom left: Example tuning curve in light-on and light-off conditions for a single cell. Bottom right: Averaged responses of the neurons in the heading cluster in light-on and light-off conditions. Traces are shown as mean ± s.e.m. D Same as (C), but for landmark neurons (cluster 2). E Same as (C), but for alignment neurons (cluster 3). F Histogram showing light-on and light-off flip scores for the heading cluster. p = 0.67, F-test. G Same as (F), but for landmark neurons (cluster 2). p = 3.1 × 10⁻¹⁶¹, F-test. H Same as (F), but for alignment neurons (cluster 3). p = 2.0 × 10⁻⁴², F-test. I Location of axonal data on the reduced dimensional space. Gray dots are the positions of RSC soma. Red dots are ADN axons and blue dots are visual axons. Colored circles represent the 95% CI ellipses surrounding each cluster, with colors as above. J Schematic of standard rotational experiment versus visual cue only rotation. K Light-on and light-off tuning curves of matched neurons in the standard rotation (top) and visual cue only rotation (bottom) for each cell cluster.

Fig. 6. Decoding analysis shows independent contributions of each functional subclass to the overall heading representation in RSC.

A Pseudopopulation decoding of neural responses across trials. Top: Predicted heading plotted for each 100 ms time bin (blue dots) overlaid on actual heading (black dotted line) for each trial, transitioning from the light-on (yellow bar) to the light-off (gray bar) condition. The transition line between conditions is shown as a vertical gray dashed line. Bottom: Decoder error (gray dots) and the average decoder accuracy for each trial (green dots and line). The dashed green line denotes chance level for decoder accuracy (0.10). Decoder accuracy is plotted as mean ± s.e.m. Bootstrapped t-test (see “Methods”), n = 1000 bootstrapped samples. p = 7.6 × 10⁻³. B Top: Decoder error (gray dots) and decoder accuracy (green dots and line) for decoder using only heading cells across conditions (left). p = 0.11 Comparison of decoder error between heading class cells and other cell classes (T: total, H: heading, L: landmark, A: alignment) showing differences in performance across classes (right). Red shading indicates that the heading class outperforms the compared class; blue shading indicates the opposite. Middle: Same as top, but for landmark cells only. p = 8.0 × 10⁻⁴. Bottom: Same as top, but for alignment cells only. p = 0.54 Decoder accuracy is plotted as mean ± s.e.m. Bootstrapped t-test (see “Methods”), n = 1000 bootstrapped samples. C Schematic of circuit for integration of visual information into the HD network, indicating changes in circuit representation during light-on condition (left), light-off condition (middle), and back in the light-on transition (right). Top: The blue feedback line describes putative feedback from the RSC to the ADN, sending adjusted heading information. Bottom: Population likelihood curves are shown in dark blue, with each cell class (heading, landmark, and alignment) shown in red, blue, and purple, respectively. Blue arrows above the likelihood curves denote the predicted internal heading at each time point. Black arrows below the likelihood curves denote the physical head direction at each time point. Dotted lines denote the previous position of the likelihood curve. n.s.: not significant, *p < 0.5, **p < 0.01, ***p < 0.001. Linear mixed-effects model.