Fig 1.
Single-subject power and coherence in the lower limb.
Raw EMG is shown for three sample trials (A,D,G). The grey boxes indicate the cued contraction phase of the task, and the vertical blue lines represent the two fast Fourier transform windows during the hold phase. The spectral plots show relative power (B,E,H) and coherence with EDB (C,F). The 15-30Hz beta-band is designated by the green boxes, and the dotted horizontal lines indicate the significance level for coherence. In most subjects, power and coherence spectra peaked in the 15-30Hz band.
Fig 2.
Average coherence spectra are shown for MG-EDB and TA-EDB in the lower limb (A,C) and for EDC-FDI and FDS-FDI in the upper limb (B,D). The dotted horizontal lines represent the significance level for average coherence, and the green boxes indicate the 15-30Hz beta-band. Significant average coherence was present in the 15-30Hz band for each muscle pair. Typically, coherence demonstrated a peak or an inflexion within this window and a further one around 9-12Hz whilst dropping off at higher frequencies.
Fig 3.
Correlation between coherence and age.
For each muscle pair, average 15-30Hz coherence is plotted against age. There was no significant correlation between coherence and age in any muscle pair (Spearman’s ρ). Linear regression models are shown as the blue dashed lines, and significance levels for average 15-30Hz coherence (for L = 200) are represented by the horizontal dotted lines.
Fig 4.
Coherence by decade of age with corresponding variable kernel density estimates and summary statistics.
Each decade is illustrated in a different colour. The stairstep curves show the distribution of average 15-30Hz coherence for subjects within a given decade, with the smooth curves showing corresponding density estimates (A,B,E,F). Quartiles derived from density estimates are shown by the box plots with additional horizontal lines indicating 5th and 95th centiles, overlain on a dot plot of individual coherence values (C,D,G,H). Coherence did not vary significantly with age (Kruskal-Wallis test). Significance levels for average 15-30Hz coherence (for L = 200) are shown as the vertical or horizontal dotted lines.
Fig 5.
Coherence across all ages with normal and variable kernel density models.
Since coherence did not vary with age, we pooled coherence readings across all ages into a single dataset. The stairstep curves illustrate the distribution of average 15-30Hz coherence for subjects of all ages. The data were modelled with a normal distribution (blue) and variable kernel density estimation (red). The density estimation model achieved a closer fit throughout, whilst still smoothing out some of the variability of the data. Significance levels for average 15-30Hz coherence (for L = 200) are shown as the vertical dotted lines.
Fig 6.
Z-scores for intrasession differences in coherence.
In each subject, the recording session was divided into two epochs for which separate 15-30Hz coherence values were calculated. Single-subject Z-scores quantify the difference between both coherence values in each individual and their distribution is shown by the stairstep curves; under the null hypothesis, they should follow a standard normal distribution (blue curve). The mean compound Z-score for all subjects (; vertical dotted lines) was not significant in any muscle pair (PZ; blue box showing range of ±1.96). However, in all muscle pairs the variance of individual Z-scores (Var) was significantly greater than unity as estimated by Monte-Carlo simulations (PMC).
Fig 7.
Difference between variances of intersession and intrasession Z-scores.
Each recording session from experiment 2 was split into two epochs for which separate 15-30Hz coherence values were computed. Single-subject Z-scores were calculated for differences in coherence between both halves of the same session (‘intrasession’) and for differences in coherence between corresponding halves of both sessions (‘intersession’). The difference of the variances of intersession and intrasession Z-scores was then computed (ΔVar; vertical dotted lines). For each muscle pair, the null distribution was estimated using Monte-Carlo simulations (green histograms). The probability of the observed value occurring under the null hypothesis was calculated with reference to the estimated null distribution (PMC). Intersession variance exceeded intrasession variance in all muscle pairs, with the differences reaching significance in TA-EDB and EDC-FDI.