Motor cortical output for skilled forelimb movement is selectively distributed across projection neuron classes

Abstract

The interaction of descending neocortical outputs and subcortical premotor circuits is critical for shaping skilled movements. Two broad classes of motor cortical output projection neurons provide input to many subcortical motor areas: pyramidal tract (PT) neurons, which project throughout the neuraxis, and intratelencephalic (IT) neurons, which project within the cortex and subcortical striatum. It is unclear whether these classes are functionally in series or whether each class carries distinct components of descending motor control signals. Here, we combine large-scale neural recordings across all layers of motor cortex with cell type-specific perturbations to study cortically dependent mouse motor behaviors: kinematically variable manipulation of a joystick and a kinematically precise reach-to-grasp. We find that striatum-projecting IT neuron activity preferentially represents amplitude, whereas pons-projecting PT neurons preferentially represent the variable direction of forelimb movements. Thus, separable components of descending motor cortical commands are distributed across motor cortical projection cell classes.

Fig. 1.. Distributed task-related neural activity in a variable-amplitude operant task.

(A) Schematic of behavioral task and Neuropixels probe recordings from mouse forelimb MCtx (MCtx^FL). (B) Outward joystick velocity and lick rate aligned to threshold crossing for 10 sessions (six mice). Shaded area indicates the SEM. (C) Movement amplitude as a function of threshold. Probability of initiating a movement of the correct amplitude. (D and E) Distribution of movement amplitude and direction. (F) Closed-loop inactivation of MCtx^FL in VGAT-ChR2 mice (500-ms duration; yellow bar) triggered on movement initiation. Joystick velocity on control (black) and inactivated (yellow) trials. (G) Cumulative probability of movement initiation for control (black) and open-loop inactivation (yellow) trials. N = 3 mice, two sessions per mouse. (H) Spike density functions of neural activity aligned to movement onset and binned (50-μm bins) according to recording depth. Plot at right shows the number of units per bin. Task-related neural activity was widely distributed across the depth of recordings. The bottom plot shows the mean activity across all units for movements that were rewarded (black) versus comparable magnitude movements (matched median) that were unrewarded (cyan) and lacked the later reward consumption–related modulation.

Fig. 2.. Inhomogeneous representation of movement kinematics across corticostriatal depths.

(A) Top: Probe tracks reconstructed in standard brain reference coordinates (see Materials and Methods). Bottom: Labeling of pons-projecting PT neurons (green) and the probe tract (red) imaged in a cleared hemibrain (see figs. S2 and S3). Scale bar, 1 mm. (B) Decoded (blue) versus observed (gray) movement profiles for 20 rewarded (concatenated, permuted order) joystick movements (see Materials and Methods). (C) Cross-validation decoding performance compared to shuffle control (Pearson correlation; see Materials and Methods). (D and E) Targeted dimensionality reduction (see Materials and Methods) identified two orthogonal dimensions of population activity that encoded amplitude (D) and direction (E). For each: Left: Example session mean projection of movement aligned activity on AMP (D) or DIR (E) dimension as function of movement amplitude (D) or direction (E) (mean value for quintile shown in color legend) quintiles. Right: Integrated perimovement modulation of activity (loading) along AMP (D) and DIR (E) dimensions as a function of the average amplitude (D) and direction (E) of movement (offset normalized). Individual sessions: gray; mean: black. (F) Neural correlates as a function of depth. For each depth bin (250 μm), movement-aligned activity (left), relative contribution to decoder performance (middle) (see Materials and Methods), and tuning of AMP and DIR activity dimensions to movement amplitude and direction quintiles (right). *P < 0.05.

Fig. 3.. Prevalence of motor command–like activity in IT neurons.

(A) Normalized activity before and during optical silencing of pons-projecting PT⁺ neurons. (B) Same as (A) for STR-projecting IT⁺ neurons. (C) Mean firing rate change during optical tagging illumination is plotted as a function of the inferred recording depth (x axis). Filled points indicate any significant modulation (not just the more stringent tagged subset; see Methods and Fig. S12). Latency to half-maximal firing rate change is indicated with color bar. (D) Activity of an example PT⁺ neuron to laser (60 trials; 594 nm) during optotagging (left) and aligned to movement start during task performance (right). More examples are shown in fig. S7. (E) Same as (D) for an example IT⁺ neuron. (F and G) Filtered, raw voltage traces showing spike activity of example PT⁺ and IT⁺ units with the amplitude of joystick movement plotted above. (H) Normalized means ± SEM activity of PT⁺, IT⁺ neural populations aligned to joystick threshold crossing (Reach threshold), and reward delivery. The mean activity of the rest (untagged) of MCtx^FL is plotted in dotted curves for comparison. (I) The cumulative distributions of the peak activity are plotted for PT⁺ and IT⁺ neural populations. Distributions of the rest (untagged) of MCtx^FL are plotted in dotted lines for comparison.

Fig. 4.. Projection cell classes are preferentially tuned to the amplitude or direction of forelimb movement kinematics.

(A) Left: Tuning of individual units to amplitude (x axis) and direction (y axis) plotted as the slope of their weighted [AMP and DIR weights (ω_a and ω_d, respectively)] z-scored activity as a function of amplitude (units of millimeter) and direction (units of radian) quintiles. Full population of recorded units: gray. Optotagged, putative Sim1⁺ corticopontine PT (PT⁺; blue) and Tlx3⁺ corticostriatal IT (IT⁺; green) units are highlighted. Right: To compare preferential tuning across groups, we compared the difference in tuning along AMP and DIR dimensions. Populations were significantly different (KW test, P < 0.001). (B) Cross-validation performance of naive Bayes classifiers trained to predict movement amplitude tertile using all units (black) or optotagged populations (PT⁺ and IT⁺) or STR units as inferred from anatomical position. (C) Contributions to committee decoder performance (see Materials and Methods) for separate neural populations from recording sessions in Sim1-cre mice. Populations identified by optotagging (PT⁺) or inferred from anatomical position are plotted; ***P < 0.001 and **P < 0.01.

Fig. 5.. Calcium imaging shows cell type–specific differences in forelimb movement correlates.

(A) Two-photon calcium imaging schematic. (B) Top left: Example histology from each mouse line. Scale bar, 100 μm. Top right: Example imaging field of view. Bottom rows: Green traces are inferred spike rates of randomly selected IT neurons aligned to behavioral variables. A.U., arbitrary units. (C) Mean z-scored activity traces aligned to reward delivery for every imaged neuron (ROI) in the dataset. Left: PT (Sim1⁺). Right: IT (Tlx3⁺). ROIs are sorted by sign of movement-related modulation and time of peak modulation. The top row shows the average normalized joystick speed and lick rate. Scale: −0.5 (blue) to 0.5 (red) z. (D) Mean activity for each cell type. The shaded area is SEM. (E) Cumulative proportion of maximal activity for each ROI (analogous to Fig. 3I). (F) Normalized inferred spike rate for individual units with positive or negative PC1 loadings is plotted. Colored dots on the left reflect the cell type. For principal components analysis (PCA), PT units were randomly subsampled to match the size of the IT population. Individual PCs, and additional example units, are provided in fig. S8. (G) Histogram of unit weights on PC1 for IT and PT neurons.

Fig. 6.. Differential effects of cell type–specific optogenetic inactivation on forelimb movement kinematics.

(A) Schematic of closed-loop inactivation paradigm. (B) Difference in movement-aligned activity between control trials and laser inactivation trials for identified corticopontine PT⁺ neurons. (C) Difference in movement-aligned activity between control trials and laser inactivation trials for identified layer 5 corticostriatal IT⁺ neurons. (D to I) Behavioral effects of inactivation on movement amplitude and speed (D to G) and direction (H and I) were examined for inactivation of layer 5 IT neurons (D and E) and layer 5 corticopontine PT neurons (F and G). For each: Left: Means ± SEM reach amplitude/speed of unperturbed control trials (black) and perturbed inactivation (color) trials. Right: Mean reach amplitude/speed of unperturbed control (black dots) and inactivation trials (colored dots) for individual sessions. (H) For an example session in Sim1-FLInChR, mouse trajectories were reliably biased in direction on inactivation trials relative to control trials. Traces show mean trajectories with tangent vectors indicating speed (length) and direction of movement (angle). (I) Population data showing x component of movement trajectory as a function of time for inactivation trials (IT, green; PT, blue) compared to control trials (black). The shaded area is SEM. ***P < 0.001, **P < 0.01, and *P < 0.05. n.s., not significant.

Fig. 7.. MCtx^FL IT neurons are necessary for execution of forelimb movements in a reach-to-grasp task.

(A) Schematic of inactivation paradigm in reach-to-grasp task (55). Laser was triggered in randomly selected trials (~27%) at the cue onset. (B) Top: Representative PT neuronal response to optogenetic inactivation with GtACR2. Lasers were delivered for 1 or 2 s in interleaved trials. Bottom: An IT neuronal response to inactivation. (C) Identified PT⁺ units (62 from four mice) and IT⁺ units (40 from four mice) displayed comparable responses to optical silencing. (D) Left: Inactivation of IT neurons in the contralateral (left) hemisphere of the reaching arm (right) blocked initiation and successful execution of reach-to-grasp presented as an ethogram of a representative session with each behavioral component automatically labeled by JABBA (72). Right: Fraction of successful performance in control versus IT inactivation trials for all sessions (n = 4 mice, eight sessions). ***P < 0.001. (E) Left: Inactivation of PT neurons had no effect on task performance in a representative session. Right: Fraction of successful performance in control versus PT inactivation trials for all sessions (n = 4 mice, eight sessions).

Fig. 8.. Inactivation of PT and IT neurons oppositely affect striatal activity.

(A) 3D visualization of complete single-neuron reconstructions (3) for 10 representative single-cell reconstructions of layer 5 PT (top) and IT (bottom) anatomical classes from the mouselight.janelia.org database shows partially overlapping projections to the recorded region of dSTR. (B) For all units from dSTR, we computed the difference between movement-aligned, z-scored time histogram for control trials and perturbation trials in which either Sim1⁺ corticopontine PT neurons (red) or layer 5 Tlx3⁺ IT neurons (blue) were inactivated during movement. (C and D) Populations of units with low average firing rates (also broad spike widths on average) were used to assess whether modulation of dSTR activity was consistent with changes in medium spiny projection neuron activity during PT (C) or IT (D) inactivation trials. Control trials (black) reveal clear movement-aligned modulation of activity in these populations and opposing changes during inactivation.