Context-dependent limb movement encoding in neuronal populations of motor cortex

Neuronal networks of the mammalian motor cortex (M1) are important for dexterous control of limb joints. Yet it remains unclear how encoding of joint movement in M1 depends on varying environmental contexts. Using calcium imaging we measured neuronal activity in layer 2/3 of the M1 forelimb region while mice grasped regularly or irregularly spaced ladder rungs during locomotion. We found that population coding of forelimb joint movements is sparse and varies according to the flexibility demanded from individual joints in the regular and irregular context, even for equivalent grasping actions across conditions. This context-dependence of M1 encoding emerged during task learning, fostering higher precision of grasping actions, but broke apart upon silencing of projections from secondary motor cortex (M2). These findings suggest that M1 exploits information from M2 to adapt encoding of joint movements to the flexibility demands of distinct familiar contexts, thereby increasing the accuracy of motor output.


Supplementary Figure 1. Forelimb performance score during motor learning.
(a) Evolution of the forelimb performance score for individual mice (n = 7) during learning of skilled locomotion and silencing of M2-M1-projections on the regular pattern (cyan). m1 -m7 refers to mouse 1 to 7, n1-n9 to neuronal network 1 to 9; (b) Evolution of the forelimb performance score for the same mice during learning of skilled locomotion and silencing of M2-M1-projections on the irregular pattern (magenta); based on the performance score we divided motor learning during both conditions into 'naive' (days 1-4), 'learning' (days 5-8) and 'expert' phase (days 9-12, performance score at saturating level); in mice 1 to 5 (1 st subset of experiments), calcium imaging was only applied during the expert phase; in mice 6 (n6 and n7) and 7 (n8 and n9), which correspond to the 2 nd subset of experiments, we included an additional expert phase with silencing of M2-M1 projections ('Expert M2-M1-silenced'), and calcium imaging was performed during all phases. Rating according to the forelimb performance score: 0 = Total miss; 1 = Deep slip; 2 = Slight slip; 3 = Replacement; 4 = Correction; 5 = Partial placement; 6 = Correct placement. (a) Naive training phase: Prediction histograms: One 30% test set was randomly selected from 500 shuffled datasets (white for regular and irregular) or 500 30% test sets were randomly selected from the one true data set (cyan for regular, magenta for irregular). The 2 distributions of 500 predictions in each recorded neuronal network (n6-n9, number in square brackets below corresponds to the cell count of each neuronal network) during the naive phase are compared by calculating the area under the curve (AUC) of a ROC-analysis. AUCvalues ≥ 0.9 in the respective neuronal network are regarded as significant encoding. (b) Same analyses for the learning phase. Note that high encoding in the shuffled distribution can arise as a consequence of the random shuffling and additionally by random selection of the test set. Still, true and shuffled distributions are in some cases completely separated. (a) Original decoding in the regular context (cyan) and decoding in the regular context if the prediction model was created from the same neuronal network in the irregular condition (purple; 70% training sets on the irregular condition, 30% test sets in the regular condition). Prediction power significantly decreases for finger movements. (b) Original decoding in the irregular context (magenta) and decoding in the irregular context if the prediction model was created from the same neuronal network in the regular condition (purple; 70% training sets in the regular condition, 30% test sets in the irregular condition). Prediction power significantly decreases for all joints. Asterisks indicate P<0.05 (paired t-test, P-value adjusted according to Holm-Bonferroni).

Supplementary Figure 7: Relationship of encoding with potential confounding factors.
(a) Left panels: Histograms of joint angle (JA) differences between all twin grasps for regular and irregular. Distributions show differences of z-scored JAs, and respective mean (solid red line) and s.d. (dashed grey lines) are visible in each plot; mean and s.d. of twin grasp differences, when JAs were not z-scored, are shown as text inlays (Δabs); Freq = frequency. Right panels: Comparison between differences of z-scored JAs, z-scored joint angle speeds (ASp) and z-scored joint angle accelerations (AA), when twin grasps are regarded; for ΔJA, ΔASp and ΔAA, mean±s.d. is shown. Note that joint angles, their speeds and accelerations were controlled similarly for each joint by the twin grasp pruning. (b) Relationship of encoding differences (ΔPCC) with mean JA differences, mean ASp differences as well as mean AA differences for each joint during twin grasps. (c) Relationship between true encoding differences and modelled encoding differences when a single variable such as general task difficulty would have affected encoding differences between the regular and irregular condition similarly across joints, separately for the whole data set (left panel) and for twin grasps (right panel). Increased task difficulty was modelled by multiplying the baseline encoding on the regular wheel with arbitrary positive factors between 0 and 10 (the regression analysis is thereby not affected by the value of the respective factor). b, c: Linear regression with clustered standard error (robust), cluster variable = neuronal network).

Supplementary Figure 8: Kinematic features during learning and M2-M1-silencing. (a)
Histograms of maximal reaching distance (RD) during each grasp, grasping duration (GD), mean grasping speed during paw reaching/retraction (positive/negative values), mean grasping acceleration during reaching/retraction (positive/negative values) and running speed during the three learning phases as well as during silencing of M2-M1-projections in the expert phase (Exp. M2-M1-sil.) when mice performed skilled locomotion on the regular wheel (cyan). Same conventions as in Fig. 1b,d. (b) Same analyses when mice performed skilled locomotion on the irregular wheel (magenta). Note that both across phases and conditions, the distributions of basic kinematic features are similar. Data pooled across animals of the second experimental series (m6 and m7).

Supplementary Figure 9. Forepaw movement within the horizontal plane. (a)
In the first three animals, we recorded ten runs with a camera placed underneath the animals to track their forepaw movement within the horizontal plane. Both for the regular (cyan dots) and irregular (magenta dots) condition, forepaw movement in the medio-lateral direction (m-l) occurred to a negligible degree when compared to the posterior-anterior (p-a) direction. (b) During locomotion on the regular (cyan) and irregular (magenta) wheel, standard deviations of tracking dots in the m-l direction (0.04-0.07 cm) are significantly lower than in the p-a direction (0.56-0.75 cm). Asterisk indicates P<0.01; paired t-test; n = 3.

Supplementary Figure 10: Encoding of joint angles with activity of individual cells.
(a) Four-dimensional plot showing the prediction of forelimb joint angles from the activity of individual cells, based on the Pearson correlation coefficient (PCC) between real and predicted joint angle traces in the test sets of the cross-validation procedure. Prediction values from all neurons of the 9 recorded neuronal networks are shown for the regular pattern. For each cell, the prediction for the shoulder and finger angles can be concluded from the spatial location in the diagram while the prediction for elbow and wrist can be derived from the color code in the upper right corner. (b) same conventions as in (a) but for the irregular condition. Each symbol represents data of a different neuronal network; n1 -n9 refers to neuronal networks 1 to 9. Note that negative correlation means that the relationship of this neuron to the respective joint was opposite between training and test set. Therefore, the cross-validated correlation values of cells that significantly encode a joint in the complete dataset, are positive in this plot.