Dynamics of brain-muscle networks reveal effects of age and somatosensory function on gait

Abstract

Walking is a complex motor activity that requires coordinated interactions between the sensory and motor systems. We used mobile EEG and EMG to investigate the brain-muscle networks involved in gait control during overground walking in young people, older people, and individuals with Parkinson's disease. Dynamic interactions between the sensorimotor cortices and eight leg muscles within a gait cycle were assessed using multivariate analysis. We identified three distinct brain-muscle networks during a gait cycle. These networks include a bilateral network, a left-lateralized network activated during the left swing phase, and a right-lateralized network active during the right swing. The trajectories of these networks are contracted in older adults, indicating a reduction in neuromuscular connectivity with age. Individuals with the impaired tactile sensitivity of the foot showed a selective enhancement of the bilateral network, possibly reflecting a compensation strategy to maintain gait stability. These findings provide a parsimonious description of interindividual differences in neuromuscular connectivity during gait.

Keywords: Age; Neuroscience; Techniques in neuroscience.

Graphical abstract

Figure 1

Time-frequency coherence between EMG and EEG channels during overground walking in healthy young people (A–D) Coherence is shown for a pair of muscles within the same leg, i.e., between the left Soleus muscle and the left lateral Gastrocnemius (A), between two muscles of each leg, i.e., between the left Soleus and right lateral Gastrocnemius (B), between the motor cortex and a contralateral muscle, i.e., between the left Soleus muscle and the right motor cortex (C), and between the left and right motor cortices (D). Coherence values are thresholded: average coherence values below the 95% CI are set to zero (white). The x axis shows the time in seconds relative to heel strike (t = 0) of the left foot and the y axis the frequencies in Hz. Black vertical lines indicate the footswitch events (lhs, left heel strike; rto, right toe-off).

Figure 2

Network weights and activations obtained using orthogonal non-negative matrix factorization of inter-/corticomuscular coherence (A–C) Coherence spectra of all EEG and EMG channel combinations, participants, conditions (left, right) and time points of the gait cycle are decomposed into three components (networks). Each network (A, B, C) is characterized by the weights (adjacency matrices) and activations time course throughout the gait cycle. The adjacency matrices show the average weights across participants for each frequency band separately (alpha, low beta, high beta, gamma). These weights give the edges between the 10 nodes of each network. The temporal activation patterns of each component are disaggregated by group (young, old, PD). LHS, left heel strike; LTO, left toe-off; MCl/MCr, motor cortex left/right; GMl/GMr, gastrocnemius medialis left/right; GLl/GLr, gastrocnemius lateralis left/right; RHS, right heel strike; RTO, right toe-off; SOLl/SOLr, soleus left/right; TAl/TAr, tibialis anterior left/right.

Figure 3

Topologies of binarized multilayer networks (A–C) Topologies displayed over human body show the minimally connected networks (thresholded edge weights; threshold was 0.145 for network 1 (A), 0.278 for network 2 (B), and 0.30 for network 3 (C)). The layers of each network (alpha, low beta, high beta, gamma) are displayed as separate columns and with different color coding (connections at alpha (green), low beta (orange), high beta (blue) and gamma (pink) frequencies). The right-hand side panels show multilayer strength centrality for each network estimated using non-thresholded weights. Each bar graph shows the strength for each node (muscles or cortical sites) of each multilayer network. The higher the strength of a node, the stronger the connectivity with other nodes of the network. MCl/MCr, motor cortex left/right; GMl/GMr, gastrocnemius medialis left/right; GLl/GLr, gastrocnemius lateralis left/right; SOLl/SOLr, soleus left/right; TAl/TAr, tibialis anterior left/right.

Figure 4

State-space embedding of the trajectories of the temporal activations of the three extracted networks (A) Different lines depict the group-average trajectories (Young, Old, PD). In addition to the trajectories in the 3D subspace, the projections on each of the three axis is shown. (B) The root-mean-square deviation (RMSD) of the group-average trajectories of young vs. old, old vs. PD, and young vs. PD. Grey-patches show the 95% confidence interval estimated based on 1000 bootstrapped samples.

Figure 5

Principal component analysis and group effects of network activations For both principal components (PC1 and PC2) we show the network activations throughout the gait cycle (left column) and the eigenvector coefficients per group (right column). Colored dots show individual data of each participant, black horizontal lines show the group mean, gray boxes show the SEM. PC, principal component; λ, explained variance; LHS, left heel strike; RTO, right toe-off; RHS, right heel strike; LTO, left toe-off.