The role of motor cortex in motor sequence execution depends on demands for flexibility

Abstract

The role of the motor cortex in executing motor sequences is widely debated, with studies supporting disparate views. Here we probe the degree to which the motor cortex's engagement depends on task demands, specifically whether its role differs for highly practiced, or 'automatic', sequences versus flexible sequences informed by external cues. To test this, we trained rats to generate three-element motor sequences either by overtraining them on a single sequence or by having them follow instructive visual cues. Lesioning motor cortex showed that it is necessary for flexible cue-driven motor sequences but dispensable for single automatic behaviors trained in isolation. However, when an automatic motor sequence was practiced alongside the flexible task, it became motor cortex dependent, suggesting that an automatic motor sequence fails to consolidate subcortically when the same sequence is produced also in a flexible context. A simple neural network model recapitulated these results and offered a circuit-level explanation. Our results critically delineate the role of the motor cortex in motor sequence execution, describing the conditions under which it is engaged and the functions it fulfills, thus reconciling seemingly conflicting views about motor cortex's role in motor sequence generation.

Extended Data Figure 1:. Histology from cohorts of MC lesioned rats.

A-B. Outlines of MC lesion boundaries of rats imaged with micro-CT. White lines denote AP and ML from bregma, dashed lines are spaced every 1mm. A. MC lesions of cohort of rats trained on the combined task (CUE, WM, and AUTO). Shown are outlines from n=5/7 rats; two rats were imaged via Nissl stain. B. MC lesions of cohort of rats trained only on the automatic sessions (n=6).

Extended Data Figure 2:. CUE and WM performance following contralateral lesion.

Fraction of successful trials pre and post a unilateral lesion to the contralateral hemisphere in the CUE and WM task. **P<0.01, ***P<0.001, two tailed t-test.

Extended Data Figure 3:. Individual lever presses are species-typical and unaffected by the MC lesion.

A. Average movement trajectories (scaled) for the left (L), center (C), and right (R) lever presses for all animals (n=7). Each line is a different rat. Top row is the horizontal (left) and vertical (right) trajectories pre-lesion, and bottom row is the trajectories post-lesion. B. Mean (left) and max (right) forelimb speed over single lever presses, before and after the lesion. Lines indicate individual rats. C. Correlation of the mean forelimb trajectory (horizontal and vertical) during the single lever press, across levers (L, C, or R), ordinal positions (1st, 2nd, or 3rd), and rats (n=7). Each dot indicates a different comparison. Comparisons are made across mean forelimb trajectories pre-lesion, post-lesion, and between pre-lesion and post-lesion trajectories.

Extended Data Figure 4:. Sensory- and working-memory guided performance kinematics resembles performance early in training.

A-B. Performance from the 1000 trials at the start of training, immediately pre-lesion, and immediately post bilateral lesion for the (A) trial duration, and (B) horizontal movement speed. C. Kinematic traces from one example rat early in learning, before the lesion, and after the lesion. D. Average trial-to-trial correlation of forelimb trajectories for a single sequence, averaged across all rats. One of seven rats had no videos captured in early learning for trajectory analysis.

Extended Data Figure 5:. Lever presses occur in discrete positions and error mode distributions.

A. Spatial distributions of an example rat’s nose for rewarded/unrewarded sequences, sampled pre- and post-lesion. B. Same as A, but the nose location is sampled only during the lever press. C. The variability of the nose position, quantified by computing the entropy of the spatial distribution across 2000 trials, for rewarded and unrewarded presses (dark/light shades) and pre-/post-lesion (red/blue), averaged across rats (n=5). Shaded area indicates s.e.m. D. Proportion of error trials classified as ‘motor errors’ for the different CUE, WM, and AUTO-only tasks, across lesion conditions. P>0.05, two-tailed t-test.

Extended Data Figure 6:. Automatic task performance in combined cohort does not recover after one month of retraining.

A. Performance in the AUTO task from the 1st week of training (early), 7 days before the lesion (pre-lesion), 7 days after the lesion (post-lesion), and 1 month following lesion (late) in the combined task (green, n=7) and AUTO-only (purple, n-6) cohorts. B. Average performance, measured through the fraction of successful trials, from time conditions (pre, post, late) across rats, represented as individual lines. C-E. Kinematic metrics plotted in the week before lesion (pre), the week after lesion (post), and a month following lesion (late). C. Trial time. D. Trial speed. E. Forelimb trajectory correlation.. *P<0.05, **P<0.01, ***P<0.001, two-sided paired (within cohort) or unpaired (across cohorts) t-test.

Extended Data Figure 7:. Pre-lesion training metrics do not differ across the combined task and AUTO only cohorts.

A. Rats across both cohorts (combined task (FT) – green and overtrained only (AO) – purple) perform a similar number of trials per session before the MC lesion. Dots represent individual rat averages, and bars are grand averages. B. Both combined and AUTO-only cohorts reach expert AUTO performance (see Methods) in a similar number of training trials. C-D. Both cohorts train for a similar number of total trials (c) and sessions (d) on the AUTO sequence before the lesion.

Figure 1:. Addressing the role of motor cortex in the execution of automatic and flexible motor sequences.

A. Simplified schematic of the motor control system considered in this study, and our hypotheses about how it is employed in response to different challenges. Subcortical controllers can generate a wide range of species-typical motor elements (left) and highly practiced, i.e., ‘automatic’, motor sequences assembled from these (center) without motor cortex. However, we posit that ordering the same motor elements into different sequences informed by sensory inputs or working-memory (i.e. ‘flexible’ motor sequences) depends on motor cortex (right). B. ‘Piano-task’ paradigm to probe the role of motor cortex in generating automatic and flexible motor sequences. Rats are rewarded for performing a prescribed three-element sequence of lever-presses on each trial. C. In the ‘automatic’ (AUTO) task, the same three-element sequence is rewarded for the duration of the experiment. D. In the ‘flexible’ task, a different three-element sequence is rewarded on a given block of six trials. The prescribed sequence in a block is randomly chosen and signaled by visual cues for the first three trials (CUE condition) and then by remembering the sequence for the remaining three trials (working memory or WM condition; see Methods).

Figure 2:. Automatic motor sequence execution is robust to motor cortex lesion.

A. Performance on the automatic task – in which the same sequence of lever presses is rewarded throughout. Data from the first week of training (early), the week before bilateral motor cortex lesions (pre-lesion), and the first week of training after recovery (post-lesion), averaged over the population (n=6, error bars are s.e.m.). Lines denote individual rats. Stars denote whether performance on a given day is significantly different from average pre-lesion performance. B. Average performance of rats (n=6) in the first week of training (early), the week before (pre) and days 3–7 following (post) motor cortex lesion, accounting for surgery-related recovery. Lines indicate individual rats. Post-lesion performance is significantly above chance levels (p=0.0014, one-sampled two-sided t-test, where chance is defined as performing a random three-lever sequence consisting of press and orient movements) C. Trial times averaged over 1000 trials before and after bilateral lesions (see Methods). D. As in C for trial speeds. E. The movement kinematics of the active forelimb on eight example trials with similar durations overlayed and compared before and after motor cortex lesions in an example rat. Blue arrows indicate the time of the first lever press, and vertical bars indicate 100 pixels (see supplementary videos) or approximately 3.5 cm. F. Similarity in forelimb movement trajectories measured through the average trial-to-trial trajectory correlations, before lesion, after lesion, and across the lesion conditions. Trajectories are local-linearly warped to the lever taps. *P<0.05, ***P<0.001, two-sided paired t-test.

Figure 3:. Flexible motor sequence execution depends on motor cortex.

A. Performance of cue-guided (CUE) and working memory-guided (WM) motor sequences in the week before (pre-lesion), the week after (post-lesion), and 1 month following (late) bilateral lesion. Shown is the fraction of successful trials, averaged over rats (n=7; error bars are s.e.m.). Stars denote whether performance is significantly different on a given day, relative to average performance in the week before lesion, for each condition. Chance performance is defined as a 8.33% success rate, equivalent to guessing at the one in 12 sequences. Thin lines denote individual rats. B. Average performance over the week before (pre) and on days 3–7 after (post) motor cortex lesion, and, to account for experience-dependent recovery, one month after lesion (late). Lines indicate individual rats. Post-lesion performance is not significantly different from chancel levels (p=0.0725 and p=0.2637 for CUE and WM respectively, one-sampled two-sided t-test) C. Variability in the errors, as quantified through the Shannon entropy, for each condition (CUE, WM) on 1000 trials from before (pre), 3–7 days after (post), and 1 month after (late) lesion. D. Duration (trial time) between the 1st and 3rd lever presses for 1000 trials before, 3–7 days after, and 1 month after lesion. E. Same as D for trial speeds. F. Horizontal and vertical kinematic traces of the dominant forelimb for 8 correct trials of the same sequence, overlayed, from one example rat. Blue arrows denote the time of the first lever press, and vertical bars indicate 100 pixels (see Supplementary video 1 and 2) or approximately 3.5 cm. G. Trial-to-trial correlation averages for successful trajectories of the same sequence, time warped to the lever presses, of the dominant forelimb (both horizontal and vertical components) of all rats, before lesion, after lesion, and across lesion conditions. *P<0.05, **P<0.01, ***P<0.001, two-sided paired t-test.

Figure 4:. Interference across tasks renders automatic motor sequences motor cortex dependent.

A. Comparison of the performance in automatic (AUTO) task sessions for rats trained on the combination of the automatic and flexible tasks (‘Combined’; n=7, green lines, same cohort as Fig. 3) versus rats trained on the automatic task alone (‘AUTO-only’; n=6, purple, from Fig. 2a, error bars are s.e.m.), in the week before and after bilateral motor cortex lesions. Stars denote whether performance is significantly different on a given day relative to average performance in the week before lesion for each cohort. Thin lines denote individual rats. Data from the AUTO-only cohort is replotted from Fig. 2. B. Average performance in the week before and 3–7 days following lesion in AUTO trials. Lines indicate individual rats. Post-lesion ‘combined’ performance was not significantly better than chance performance, defined as (p=0.1659, one sampled two-sided t-test). C. Trial time plotted in the 1000 trials before (pre) and after (post) lesion. D. As in C for trial speed. E. Horizontal and vertical position of the dominant forelimb on 8 example trials, sampled before and after the lesion, from one rat in each cohort (combined – green, AUTO-only – purple). Blue arrows denote the time of the first lever press, and vertical bars indicate 100 pixels or approximately 3.5cm. F. Average trial-to-trial forelimb correlations (both horizontal and vertical positions), time warped to the lever presses, of all rats before lesion, after lesion, and across lesion conditions. *P<0.05, **P<0.01, ***P<0.001, two-sided t-test.

Figure 5:. A neural network model reproduces the experimental results and predicts interference between automatic and flexible task conditions.

A: Schematic illustrating the architecture of our neural network model. In this model, a motor cortex module (MC) receives cue inputs and projects via output z_MC to a downstream module that we equate with the dorsolateral striatum (DLS; see text for the rationale behind this interpretation). Both MC and DLS modules learn to interact with a brainstem/spinal cord control module (BS) via outputs y_MC and y_DLS to control the total motor output y_total. The BS module in turn sends efference/proprioceptive feedback to the MC and DLS modules. B. Example trajectories on a simulated version of the piano-task following model training in either CUE, AUTO-only, or AUTO-combined tasks. The network controls the velocity of a ‘forelimb’ and must move it into three circular regions (representing ‘lever presses’) in the correct sequential order (in this example, ‘right-center-left’ for all tasks). C. Learning rules for the MC and DLS modules (see text for details). y_target is the target trajectory for the current trial, used to train the MC module. This target signal is not intended to be biologically realistic but rather is an abstract way to capture the propensity of learning in MC to improve task performance; f_BS indicates the transformation applied by the brainstem module; θ refers to the weights of each module; and r indicates whether the trial was rewarded or not (i.e., r=1 for a correct trial and r=0 for an incorrect trial). D. Measure of engagement of the MC module throughout training, for the CUE AUTO-only, and AUTO-combined tasks, averaged over n=20 training runs. E. Effects on task performance of ‘lesioning’ the MC module (i.e., clamping its outputs z_MC and y_MC to zero). Lines indicate individual performance over n=20 runs. F. Mean manipulandum trajectories, pre- and post- MC module removal, for the overtrained sequence ‘right-center-left’. G. Difference (normalized root mean squared error) between the average manipulandum trajectories before and after removing the MC module (n=20 for each training condition). H. The motor circuit we consider for sequence production. I. Hypothetical roles for motor cortex (red) and subcortex (blue) in the production of discrete motor sequences. For a single automatic sequence (left), the mapping between past and future movements is unambiguous, meaning that motor efference/history is sufficient to specify the progression of the motor sequence, something that can be done subcortically. For flexible sequences (center), the transition between elements is inherently ambiguous and cannot be specified simply by mapping past to future actions (e.g., action A can transition to either action B or C depending on the sequence). Thus, when challenged to produce an automatic sequence in the context of combined training on both tasks (right), concurrent demands for flexibility interferes with a rote mapping between past to future actions, and hence prevents subcortical consolidation of the automatic motor sequence, making it dependent on inputs from motor cortex.