Mouse sensorimotor cortex reflects complex kinematic details during reaching and grasping
- Graduate Program in Computational Neuroscience, The University of Chicago, United States
- Department of Organismal Biology and Anatomy, The University of Chicago, United States
- Neuroscience Institute, The University of Chicago, United States
- NSF-Simons National Institute for Theory and Mathematics in Biology, United States
Figures
Figure 1
Mice make complex reaching and grasping movements to retrieve water rewards during head fixation.
(A) Task setup: a head-fixed animal mid-reach with representative markerless tracking. (B) Task timeline, with representative 3D triangulated forelimb ‘skeletons.’ (C) Distal digit centroid trajectories from 100 ms before lift to 100 ms after contact. Stereo viewpoint corresponds to the data image in A, left and right trial traces offset for visualization. One example session. (D) Lift reaction time distributions, left and right trials depicted separately. (E) Time from lift to paw centroid arriving within 5 mm of the spout target tip. (F) Time between lift and spout contact. Box plots in D-F show the quartiles (box and central line), whiskers extend 1.5 times the interquartile range past the quartile boundary, and more distant points (outliers) are plotted individually. (G) Number of contact events within 1 s of the water cue per trial. Data were pooled across sessions for D-G. (H) Time between cue and spout contact during optogenetic experiments. (I) Representative digit centroid trajectories for a subset of right trials during optogenetic experiments, from –100 ms before lift to 1000 ms after lift. Data from one example session.
Figure 2
Paw aperture reveals stereotyped kinematics across idiosyncratic single-trial reach-to-grasp movements.
(A) Paw aperture was computed from the paw markers using a simple geometric measure. (B) Aperture time series overlaid on the Z velocity of the paw centroid. The collect and extend events are defined by the aperture minimum and its subsequent local maximum. Representative trial. (C) Histograms of collect and extend event times for all trials, left and right trials separated. (D) Lift, collect, and extend event times depicted on digit centroid trajectories for one example mouse. Left and right trials offset in X and Z as in Figure 1C. (E) Median over trials of aperture time series aligned to the moment of maximum centroid Z velocity.
Figure 3
Mouse reaching and grasping movements involve coordination across proximal and distal joints.
(A) Schematic of Euler angle inverse kinematics procedure, depicting seven joint angles over the proximal joints; 17 distal angles not shown. (B) Time series of 5 representative joints, shown from –100 to +400 ms around lift onset (vertical dashed line). Left (blue) and right (red) trials are shown separately. The horizontal scale bar is 100 ms. The vertical scale bar is 30 degrees. Dashed horizontal lines are 0 degrees. (C) Correlation matrices of joint angles over all recorded time points. Rows are ordered descending from proximal (shoulder) to distal (metacarpal phalanges, MCP, proximal interphalanges, PIP). (D) Example loss curve of the cross-validated principal component analysis performed on joint angle and joint velocity time series. (E) Estimated dimensionalities of joint angle and velocity kinematics across mice.
Figure 4 with 1 supplement
M1-fl and S1-fl cells are heterogeneously tuned to specific movement features.
(A) Two-photon imaging fields of view for all mice aligned to the Allen Atlas common coordinate framework (CCF). Primary motor (M1) corresponds to MOp in the CCF, and S1-fl corresponds to SSp-ul. The dashed gray line indicates the outline of the forelimb primary motor cortex (MOp-ul, referred to here as M1-fl) as described in Muñoz-Castañeda et al., 2021. See Figure 4—figure supplement 1 and Methods for field of view alignment details. Right, example max-projection images for M1-fl and S1-fl for mouse 1 with magnified insets. (B) Modulation metric (see Methods) of cue-locked data for individual cells in M1-fl and S1-fl. Values displayed are the logarithm of the computed p-value, where each point represents a cell. All cells with p-values <0.025 are shown colored: left-only modulated cells are blue, right-only modulated cells are red, and cells modulated to both conditions are purple. (C) Peri-event histograms and rasters of single-trial activity from individual cells in M1-fl. Each column displays two cells locked to each trial event: cue, lift, and first-spout contact. Trials grouped within each condition by time in reach. Vertical scale bar: 10 events per second (arbitrary units). Horizontal scale bar: 100 ms. (D) As in C but for S1-fl.
Figure 4—figure supplement 1
Vibration mapping for each individual mouse.
Each image shows the widefield calcium response +250 ms after a vibrational stimulus applied to the contralateral paw. Each image was aligned with the underlying Allen common coordinate framework (CCF) by placing the broad S1-fl (SSp-ul in the Allen nomenclature) shaped activity profile within the corresponding region. Placement was then verified against veridical injection coordinates obtained during surgery (not shown). The dashed line in each map indicates the outline of the forelimb primary motor cortex as named MOp-ul in Muñoz-Castañeda et al., 2021 and referred to as M1-fl here. Each gray box indicates the placement of the imaging fields of view in M1-fl and S1-fl for each animal. Note that the window for mouse 1 was tilted anterior-posterior during widefield imaging, reducing the measured response in the anterior half of the window. This was accounted for in estimating atlas alignment.
Figure 5 with 3 supplements
Kinematic details of proximal and distal joints during reaching and grasping can be decoded from M1-fl and S1-fl population activity.
(A) Performance of decoding models predicting joint angle kinematics from RADICaL-inferred M1-fl rates. Violins depict the distribution of variance explained over cross-validation folds for each individual joint angle. (f: flexion, a: abduction/adduction, r: rotation, d: deviation, o: opposition, s: splay, MCP: metacarpal-phalanges, PIP: proximal-interphalanges; digits indicated by number). (B) Reconstructed joint angle time series for 5 representative joint angles. Black traces are original and colored are reconstructed; left trials are blue and right trials are red. A vertical dashed line indicates lift time, and a horizontal dashed line indicates 0 degrees. The vertical scale bar is 30 degrees, and the horizontal scale bar is 100 ms. (C) Forelimb skeletal postures reconstructed from decoded joint angles for representative left and right trials (trials having median variance explained). Seven representative time points from lift time to grab time are shown. Gray postures are original data, colored postures are reconstructed as in B. Decoded joint angles in each time step were applied to the neutralized posture of the original data, preserving the original lengths of inter-marker links (D). Performance of decoding models for all 5 mice. Each point is the median of the distribution of variance explained over cross-validation folds, with proximal (magenta) and distal (goldenrod) angles considered separately. The black lines indicate the median variance explained over all proximal or distal joints for an individual mouse. (E–H) As in A-D but for S1-fl data.
Figure 5—figure supplement 1
Simpler processing or decoding joint angle velocities also produce strong kinematic decoding.
(A) Joint angle decoding as in Figure 5, but from M1-fl deconvolved calcium transients smoothed with a 35 ms Gaussian kernel. Data is presented as in Figure 5D and H. (B) as in A but for S1-fl. (C) Joint velocity decoding using RADICaL rates from M1-fl. (D) Same as C but for S1-fl.
Figure 5—figure supplement 2
Decoding of proximal and distal joint angles is not due to correlations between them or correlations with distance to target.
(A) Joint angle decoding as in Figure 5 but with proximal joint angle time series regressed off of distal joints and distal joint angle time series regressed off of proximal joints. Decoding performed with RADICaL rates. Data is presented as in Figure 5D and H. (B) As in A but for S1-fl. (C) Joint angle decoding as in Figure 5 but with the distance to the spout of the paw centroid regressed off of all joint angles. (D) Same as C but for S1-fl. (E) Variance of joint angles and their variance explained when decoded using M1-fl RADICaL rates as in Figure 5. Each point has coordinates of the true kinematic variance and the variance explained for a single joint on a single cross-validation fold. (F) Same as E but for S1-fl.
Figure 5—figure supplement 3
Decoding the variation around a single target is possible, and adding lags slightly improves decoding.
(A) Joint angle decoding as in Figure 5—figure supplement 2, but for single-trial residuals obtained from subtracting off the time-varying mean over all right trials. Data is presented as in Figure 5—figure supplement 2. (B) Same as A but for S1-fl. (C) Joint angle decoding as in A but for left trials. (D) Same as C but for S1-fl. (E-F) Joint angle decoding as in Figure 5 but with additional predictors composed of causal lags at 20, 40, 60, 80, and 100 ms. (G-H) Joint angle decoding as in E-F, but including both causal and acausal lags of 20, 40, 60, 80, and 100 ms.
Figure 6 with 1 supplement
M1-fl reflects target-specific information earlier and more persistently than S1-fl.
(A) Projections of population activity from held-out single trials onto the target decoding dimension for M1-fl and S1-fl. Right trials are red and left trials are blue. Traces were normalized by the 90th percentile value across trials for left and right trials separately. (B) Distribution of threshold crossing times for projections shown in A. Black lines indicate medians. Star indicates significantly different medians at p<0.05, Wilcoxon ranksum. (C) Performance of time-generalized target decoding analysis. Each heat map depicts the performance over all model training windows (columns) and test time points (rows). Time in trial runs vertically as in A. Both the training and testing epochs were –100 (leftmost column training, bottommost row test) to +1000 ms (rightmost column training, top row test) relative to cue. Training epochs were 100 ms non-overlapping windows and testing epochs were 10 ms bins. (D) Performance of movement onset prediction for M1-fl and S1-fl. The correlation coefficient r was computed between the true and predicted reaction times (RTs) across held-out trials.
Figure 6—figure supplement 1
Target classifier performance over time.
(A) Performance over time for target classifiers from Figure 6A and B. Lines show mean performance across held-out trials, fills show standard error of the mean. A dashed line indicates chance performance. (B) Performance over time for target classifiers after regressing off joint angle kinematics from neural activity. Lines show mean performance across held-out trials, fills show standard error of the mean.
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Mouse sensorimotor cortex reflects complex kinematic details during reaching and grasping
eLife 14:RP106270.
https://doi.org/10.7554/eLife.106270.3
