Hybrid dedicated and distributed coding in PMd/M1 provides separation and interaction of bilateral arm signals

Abstract

Pronounced activity is observed in both hemispheres of the motor cortex during preparation and execution of unimanual movements. The organizational principles of bi-hemispheric signals and the functions they serve throughout motor planning remain unclear. Using an instructed-delay reaching task in monkeys, we identified two components in population responses spanning PMd and M1. A "dedicated" component, which segregated activity at the level of individual units, emerged in PMd during preparation. It was most prominent following movement when M1 became strongly engaged, and principally involved the contralateral hemisphere. In contrast to recent reports, these dedicated signals solely accounted for divergence of arm-specific neural subspaces. The other "distributed" component mixed signals for each arm within units, and the subspace containing it did not discriminate between arms at any stage. The statistics of the population response suggest two functional aspects of the cortical network: one that spans both hemispheres for supporting preparatory and ongoing processes, and another that is predominantly housed in the contralateral hemisphere and specifies unilateral output.

Fig 1. Behavior.

(A) Monkeys reached to one of six virtual targets, indicated by grey spheres in the cartoon. During the task these would be invisible until one appeared to instruct the reach. (B) Trials consisted of 3 phases. Each trial was initiated by placing both hands in start targets and remaining still for 500ms (Rest phase). A small target then appeared at the location of the future reach in a color that indicated which hand to use. The monkey remained still during cue presentation for 0.5–1.5s (Instruct phase). The start target for the reaching hand then disappeared while the reach target enlarged to cue movement (Move phase). (C) Hand assignments followed a blocked schedule. (D) Distributions of reaction times (top row) and reach durations (bottom row) for each monkey, hand, and target. Left hand reaches in yellow, right in purple. Horizontal black bars show means, red bars show medians. (E) Speed profiles during left- or right-hand trials. Both reaching and stationary hands are plotted in each, although stationary speeds are near 0 and hardly visible. Vertical red lines indicate threshold crossing to mark movement onset. Monkey O main, monkey W inset. Mean +/- standard deviation.

Fig 2. Neural recordings.

(A) MRI-based volume renderings of the skull and target brain regions. Top panel shows the arrangement of the two chambers. Two bottom rows show segmented brain regions within the cranial window of each chamber, for each monkey. Region boundaries were assigned based on [28]. Red—somatosensory cortex; blue—primary motor cortex (M1); pink—dorsal premotor cortex (PMd); green—ventral premotor cortex; white—frontal eye field. CS—central sulcus; SPCD—superior pre-central dimple; ArS—arcuate sulcus. Grey ellipses indicate regions sampled by recordings. (B) Interlaminar recordings were obtained using V- and S- probes (Plexon, Inc., Dallas, TX) with 24–32 electrodes aligned perpendicular to the cortical surface. Example waveforms were all simultaneously recorded from a single probe. (C) MRI coronal slice, monkey O. 3mm black bar is approximately equal to the distance spanned by electrodes on 32-channel probes. Same landmark labels as in (A).

Fig 3. Firing rate traces of example neurons and population means.

(A-C) Trial-averaged firing rates for three neurons from the left hemisphere. Each line color represents a different target according to the color-coding in the top right. The time windows used to represent each phase in the analysis are indicated by the horizontal bars at the top, and the modulation strength values for each phase are included as annotations (M). Traces display mean +/- SEM. (A) An M1 unit exclusively modulated during ipsilateral movements. Instruct arm preference of -0.84, Move arm preference -0.99. (B) A PMd unit with both Instruct and Move phase modulation for both arms. Instruct arm preference of 0.56, Move arm preference 0.11. (C) A PMd unit with modest contralateral modulation during the Instruct phase and strong contralateral modulation during movement, but no modulation on ipsilateral trials. Instruct arm preference of 0.63, Move arm preference 0.97. (D,E) Mean firing rates for PMd and M1 populations, +/- SEM. The full distributions of modulation and arm preference values for the two populations are provided in Fig 4. These means were calculated over all units and targets; as such, the means reflect the net excitation-inhibition, which is not the same as the modulation strength.

Fig 4. An increasing number of arm-dedicated units emerge with each task phase.

(A) Cumulative distribution of single-unit modulation during each phase, arm. Left panel PMd, right panel M1. Large values cut off by plot: monkey O Contra Move [134(PMd), 133(PMd), 104(PMd)], Ipsi Move [234(M1), 181(M1), 130(M1)]; monkey W Contra Move [125(M1)]. Monkey O main, monkey W inset. (B) Distributions of arm preferences during each phase. Negative values are ipsi-preferring (grey background), positive values are contra-preferring (white background). Solid black vertical lines indicate the mean of each distribution, and dashed lines mark the upper quartile.

Fig 5. Neural activity is progressively consolidated within arm-specific subpopulations.

(A) Modulation for the preferred arm plotted against arm preference, for all units in each brain area and task phase. Log-linear best fit lines are displayed in red. Inset figures belong to Monkey W. (B) Slopes of regression lines fit to data from (A), independently for ipsi- and contra-preferring sub-populations. Mean +/- bootstrapped 95% confidence interval. (C-E) For the Move phase in monkey O, cumulative modulation plotted against arm preference, i.e. each point indicates the proportion of modulation accounted for by all units with arm preference values to the left of the indexed position. Positive values on the x-axis indicate contra-preferring, and negative values indicate ipsi-preferring. Shaded error bars indicate bootstrapped standard error. See S2 Fig for cumulative modulation plots for both animals in each phase. (C) The full spectrum of arm preferences is shown. Shaded backgrounds indicate three partitions: Contra-dedicated [0.4, 1] and Ipsi-dedicated [-1, -0.4] in white, and Neutral [-0.3, 0.3] in grey. (D) Cumulative modulation within contra-dedicated regime. (E) Same as (D), but ipsi-dedicated. Note inverted axis. (F) The proportion of modulation within each partition from (C) during ipsi- or contralateral movements. Note that the total modulation is significantly lower for ipsilateral movements, particularly for Monkey W, and these data are only displayed as proportions. Mean +/- bootstrapped 95% confidence interval.

Fig 6. Population activity reorganizes and diverges for the two limbs throughout planning.

(A) Dimensionality of the PCA subspace estimated as the number of components that minimizes the cross-validated reconstruction error of the full-dimensional neural data. Mean +/- standard error across datasets. (B,C) Heat maps indicate alignment of 4-dimensional PCA subspaces between all pairs of timepoints across the Instruct and Move phases of the task, averaged across sessions. (B) Compares subspaces across time for movements of the same arm. Three blocks forming along the diagonal indicate three distinct subspaces: a pre-instruction Rest space, a post-instruction Instruct space, and a peri-movement Move space. (C) Compares subspaces across time for movements of opposite arms. Prior to instruction there is a moderate alignment of the subspaces for each limb, however, the two subspaces diverge around 100ms post instruction. (D) Summary of the data in (B,C). Mean +/- standard deviation across datasets.

Fig 7. Separation of arm-specific subspaces relies upon unit-level segregation.

(A-C) Single session example of a PCA model trained to capture bi-hemispheric activity during left arm movements. Held-out testing data for 82 simultaneously recorded units were used. (A) Cumulative proportion of variance accounted for across the top 10 principal components. (B) For each component, the ratio of the explained variance between the two limbs. (C) For each component circled in red in 7B, the absolute values of the coefficient weights are plotted against the corresponding unit’s arm preference. Top row represents components 1–3; bottom row represents components 4–6. Positive arm preference values indicate right arm preferring units. (D) The component variance ratio for the two arms plotted against a coefficient-weighted average of the arm preferences for each unit in that component. Datapoints represent the top 5 principal components of left or right arm trained models across all sessions. Separate models for each phase are plotted in each column. Because these models include activity from both hemispheres, hands are referred to as “left” and “right” as opposed to “ipsi” and “contra”. Pearson correlation coefficient for each dataset is displayed in the red box. Top row monkey O, bottom row monkey W.

Fig 8. Behaviorally specific information exists within a subspace that captures bilateral activity.

(A) Illustration of the population partitioning approach. Each unit is represented as a pie-chart displaying the relative modulation during left and right arm trials. Most units in the left hemisphere are more strongly modulated during right arm movements (mostly purple pie-charts), yet some prefer left arm movements (mostly yellow pie-charts). Regardless as to which hemisphere each unit is in, the population may be subdivided into left and right arm preferring sub-populations. On the extreme that all information about each arm is contained within dedicated sub-populations, this simple division will fully segregate the signals such that movements of the non-preferred arm cannot be classified. (B) Modulation as a function of time, taken as the mean over all units during trials of their preferred or non-preferred arm, +/- standard error. Horizontal bars at the top indicate the phase windows used in analysis. (C) Target classification accuracy using LDA for movements of the preferred arm. Models are trained on each time point and tested on each time point to provide high temporal resolution and inform cross-phase generalization of the classifier. Plots are averaged over all sessions (13 Monkey O, large plots, 7 Monkey W, small plots) and both sub-populations (left-preferring, right-preferring). (D) Same as (C), but for non-preferred arm movements. (E) Summary data of (C,D) for monkey O, top panel, and monkey W, bottom panel. Mean +/- standard deviation across datasets. (F) Ratio of the variance captured in the distributed subspace for the two limbs.