Human local field potentials in motor and non-motor brain areas encode upcoming movement direction

Abstract

Limb movement direction can be inferred from local field potentials in motor cortex during movement execution. Yet, it remains unclear to what extent intended hand movements can be predicted from brain activity recorded during movement planning. Here, we set out to probe the directional-tuning of oscillatory features during motor planning and execution, using a machine learning framework on multi-site local field potentials (LFPs) in humans. We recorded intracranial EEG data from implanted epilepsy patients as they performed a four-direction delayed center-out motor task. Fronto-parietal LFP low-frequency power predicted hand-movement direction during planning while execution was largely mediated by higher frequency power and low-frequency phase in motor areas. By contrast, Phase-Amplitude Coupling showed uniform modulations across directions. Finally, multivariate classification led to an increase in overall decoding accuracy (>80%). The novel insights revealed here extend our understanding of the role of neural oscillations in encoding motor plans.

Fig. 1. Representation of intracranial implantation and brain coverage across subjects projected on a standard 3D MNI brain and experimental design.

a Top, front and right views of the depth-electrode recording sites. b Top, left and right views of the number of recording sites that contribute to each vertex (i.e. spatial density). c SEEG locations per subject d Experimental design of the delayed center-out motor task. After a 1 s rest period (rest, −1000 to 0 ms) a first cue (Cue 1) instructed subjects to prepare to move their hand (motor planning, 0–1500 ms). Next, a go signal (Cue 2) appeared prompting participants to execute the movement (motor execution, 1500–3000 ms).

Fig. 2. Relative power modulations per directions (up/right/down/left) relative to baseline ([−750, −250] at rest) for a premotor SEEG site.

a Time–frequency representation. b Singe-trial high gamma [60, 200 Hz] power modulation.

Fig. 3. Phase modulations per directions (up/right/down/left) for a premotor SEEG site.

a Phase-locking factor across trials. b Singe-trial very low-frequency phase (VLFC, [0.1, 1.5 Hz]) modulation.

Fig. 4. Phase-amplitude coupling modulations per direction (up/right/down/left) during planning [0, 1500 ms] and execution [1500, 3000 ms] windows for a premotor iEEG recording site.

Comodulograms representing PAC as a function of frequency for phase and amplitude during planning (a) and execution (b). Singe-trial PAC modulations, per direction, for delta [2, 4 Hz], theta [5, 7 Hz] and alpha [8,13 Hz] phase coupling with high-gamma [60, 200 Hz] amplitude for planning c and execution d windows.

Fig. 5. 4-directions decoding of intended and executed limb movements using power, phase and PAC features over several frequency bands, Power features are presented within alpha (α), beta (β) and high-gamma (high-γ) bands and VLFC ([0.1, 1.5 Hz]) phase.

Each column summarized SEEG sites that present significant decodings during the entire planning or execution period. Non-significant areas are presented in gray (p < 0.05 after correction for multiple comparisons using maximum statistics through SEEG sites, time, and frequencies).

Fig. 6. Sorted density of significant timings across region of interest (ROI) and power features.

This density was obtained using single feature that presented at least three consecutive significant decodings after correction for multiple comparisons (p < 0.05 corrected for SEEG sites, features and time using maximum statistics).

Fig. 7. Time-resolved 4-directional tuning task-induced power and phase modulations.

(up: red; right; brown: down: blue; left: green) and associated decoding accuracies (purple) using an LDA with a 10 times 10-fold cross-validation on three SEEG sites. The power is computed every 50 ms using a 700 ms window. The two vertical lines at 0 and 1500 ms, respectively, represent the onset of the planning phase (Cue 1) and the execution phase (Go signal, Cue 2). The horizontal black plain line represents the theoretical chance level (4-classes, 25%) and the red dotted line represents the significance level computed from permutations at p < 0.05 after correction for multiple comparisons through time points using maximum statistics, a and b alpha power [8, 13 Hz] for two electrode contacts located in the posterior middle frontal gyrus (pMFG), c high-gamma power [60, 200 Hz] of a posterior pre-SMA electrode contact, d VLFC phase [0.1, 1.5 Hz] of the same posterior pre-SMA site. Shaded areas represent the SEM.

Fig. 8. Temporal generalization (TG) using power features on three distinct SEEG sites.

The vertical and horizontal lines at 0 and 1500 ms stand respectively for Cue 1 and Cue 2. White contoured zones delimit statistically significant decodings at p < 0.01 (binomial test) after Bonferroni correction through time. No decoding are performed on the diagonal, a TG in a pMFG site using alpha [8, 13 Hz] power, b and c TG in two distinct SEEG sites located in the posterior pre-SMA using high-gamma [60, 200 Hz] power, d TG of the three combined sites (alpha pMFG + high-gamma posterior pre-SMA).

Fig. 9. Decoding results of the multi-features procedure.

a Time-resolved decoding accuracy and associated deviation using the MF selection for the subject S1. Cue 1 and Cue 2 are represented with two solid gray lines. Blue lines indicate the maximum decodings reached respectively during the planning and execution periods. The horizontal solid gray line represents the theoretical chance level of 25% and the solid red line is the corrected decoding accuracy (p < 0.05 corrected using maximum statistics across time points) obtained by randomly shuffling the label vector (permutations). Shaded areas represent the SEM. b Most selected features during planning (blue bars) and execution (red bars). For each barplot, the y-axis show the number of times a feature was selected (occurrence), and the x-axis shows the feature type (power, phase and PAC) as well as the name of the frequency band. c Best decoding accuracies per subject for intention and execution. The solid black line represents the theoretical chance level of a 4-class classification problem (25%) and the dotted red line (~40%), the statistical chance level at p < 0.05 (corrected using maximum statistics across subjects). d Most recurrent selected power features during the multi-features procedure as a function of Brodmann area. For each power frequency band and for each Brodmann area, we subtracted the number of features selected during preparation from those during execution. Thus, blue and red colors mean that a feature has been selected more times respectively during the preparation and execution of the movement (specificity). In the same way, the white color means that as many features have been selected for both conditions while black rectangles stand for no selected features.