(C) PLOS One This story was originally published by PLOS One and is unaltered. . . . . . . . . . . A closed-loop auditory stimulation approach selectively modulates alpha oscillations and sleep onset dynamics in humans [1] ['Henry Hebron', 'School Of Psychology', 'University Of Surrey', 'Guildford', 'United Kingdom', 'Surrey Sleep Research Centre', 'Uk Dementia Research Institute Care Research', 'Technology Centre', 'Imperial College London', 'The University Of Surrey'] Date: 2024-06 S1 File. Fig A. Comparison of the resultant values using different referencing schemes. The resultant was calculated in electrode Fz of the hd-EEG system in experiment 1 using 2 referencing schemes, Laplacian (as used for all analysis in the main manuscript) and right mastoid (to match the referencing scheme used in the αCLAS EEG system). Channel Fz in the hd-EEG system was adjacent to this channel in the αCLAS EEG system, and the mastoid channel in the hd-EEG system was taken from TP10. The resultant values using the right mastoid reference are higher and more similar to those reported for the αCLAS EEG system. Violin plots show the resultant for each targeted phase in experiment 1 in each reference scheme. Black lines represent each participant. Stats indicate output of paired samples t test between the resultant for each reference scheme, *** indicates p < 0.001. Table A. Data from closed-loop EEG device. Table B. Data from high-density EEG device. Fig B. Phase-locking accuracy across conditions—topography of phase-locking accuracy (average resultant) across 4 conditions for experiments 1 and 2, targeting Fz and Pz electrodes, respectively (blue circles). White marks indicate channels at which resultant >0.4 and p < 0.05. Black marks indicate channels at which resultant <0.4 and p < 0.05; p-values from FDR-corrected z-test for non-uniformity. Fig C. Power change ANOVA for experiment 1 –(A) Topography of permutation ANOVA stats for frequencies across the spectrum. Each topoplot is computed using +/- 0.2 Hz around the labelled frequency. Colours and colourbars indicate F statistics. White marks indicate cluster-corrected p-values <0.05. Fig D. Power change ANOVA for experiment 2 –(A) Topography of permutation ANOVA stats for frequencies across the spectrum. Each topoplot is computed using +/- 0.2 Hz around the labelled frequency. Colours and colourbars indicate F statistics. White marks indicate cluster-corrected p-values <0.05. Fig E. Power change ANOVA for experiment 1 –(A) Topography of permutation ANOVA stats for various frequencies in and around the alpha band. Each topoplot is computed using +/- 0.2 Hz around the labelled frequency. Colours and colourbars indicate F statistics. White marks indicate cluster-corrected p-values <0.05. Fig F. Power change ANOVA for experiment 2 –(A) Topography of permutation ANOVA stats for various frequencies in and around the alpha band. Each topoplot is computed using +/- 0.2 Hz around the labelled frequency. Colours and colourbars indicate F statistics. White marks indicate cluster-corrected p-values <0.05. Fig G. Frequency estimates from off and on periods in experiments 1 and 2. (A) Frequency at frontal ROI across 4 conditions in experiment 1 for (i) the “off” period, (ii) the “on” period, and (iii) the off period subtracted from the on period. (B) The same plots but for the parietal ROI in experiment 2. For all plots, mixed effects models were run: [frequency ~ condition + (1|Participant)]. Post hoc Wald tests were run for those which showed a statistically significant effect of condition (p < 0.05). * p < 0.05, **p < 0.01, ***p < 0.001. Bars between conditions indicate differences between conditions, asterisks above conditions indicate a significant difference from zero, as per one-sample t test. Fig H. Autocorrelation of the inter stimulus interval in experiment 1. Shown is the autocorrelation coefficient of inter-stimulus intervals across 20 lags, in which a value of 1 would indicate a perfectly periodic stimulus and lower values indicate a deviation from periodicity. The 4 different colours represent the 4 phase conditions and the error bars indicate standard error of the mean. It is clear that the stimulation we administered is far from periodic, with the autocorrelation coefficient quickly dropping to <0.1 after only 2 lags. We propose that if the sound was simply entraining itself, via the EEG, then we would see a highly periodic stimulus. This is not the case here. Fig I. Stimulation-induced connectivity changes in alpha band (experiments 1 and 2). (Ai and Ei) Topography of main effect of phase on average alpha band PLV for each channel as per ANOVA, for experiments 1 and 2, respectively. White marks indicate cluster-corrected p < 0.05. (Aii and Eii) Topography of variance between conditions of average alpha band PLI for each channel as per ANOVA, for experiments 1 and 2, respectively. White marks indicate cluster-corrected p < 0.05. (Bi and Fi) Stimulation-induced changes in alpha band PLV per condition as per t test (compared to the “off” period), lines are plotted where p < 0.01. Red lines indicate an increase in the connectivity of that channel pair, blue lines indicate a decrease. (Bii and Fii) Stimulation-induced changes in alpha band PLI per condition as per t test, lines are plotted where p < 0.01. Red lines indicate an increase in the connectivity of that channel pair, blue lines indicate a decrease. (Ci and Gi) Stimulation-induced changes in alpha band PLV per condition, collapsed across the stimulation period and across all other channels and the region of interest, in experiments 1 and 2, respectively. (Cii and Gii) Stimulation-induced changes in alpha band PLI per condition, collapsed across the stimulation period and across all other channels and the region of interest, in experiments 1 and 2, respectively. For all violin plots, mixed effects models were run: [connectivity ~ condition + (1|Participant)]. Post hoc Wald tests were run for those which showed a statistically significant effect of condition (p < 0.05). * p < 0.05, **p < 0.01, ***p < 0.001. Bars between conditions indicate differences between conditions, asterisks above conditions indicate a significant difference from zero, as per one-sample t test. (Di and Hi) Stimulation-induced changes in alpha band PLV per condition, collapsed across all other channels and the region of interest across time, in experiments 1 and 2, respectively. (Dii and Hii) Stimulation-induced changes in alpha band PLI per condition, collapsed across all other channels and the region of interest across time, in experiments 1 and 2, respectively. Fig J. Topography of permutation ANOVA stats for connectivity in each frequency band in experiment 1. (A) Phase-locking value (PLV), (B) phase lag index (PLI). White dots show significant main effect of phase-targeted, cluster-corrected p < 0.05. Fig K. Topography of permutation ANOVA stats for connectivity in each frequency band in experiment 2. (A) Phase-locking value (PLV), (B) phase lag index (PLI). White dots show significant main effect of phase-targeted, cluster-corrected p < 0.05. Fig L. Auditory evoked potential (AEP) and phase reset. (A and C) Lowest pre-stimulus alpha power octile at Fz in experiment 3 (A) and Pz in experiment 4 (C). (B and D) Highest pre-stimulus alpha power octile at Fz in experiment 3 (B) and Pz in experiment 4 (D). (A, B, C, D) (i) Broadband (1–40 Hz) AEP*; (ii) Amplitude component of alpha band (7.5–12.5 Hz) endpoint-corrected Hilbert transformed AEP*; (iii) instantaneous phase of alpha band (7.5–12.5 Hz) from endpoint-corrected Hilbert transformed AEP**. Even at Pz, in experiment 4 there was post-stimuli phase alignment when the pre-stimulus alpha power was at its lowest amplitude (Ciii). * Black marks indicate ANOVA p < 0.05. ** Black marks indicate Rayleigh test p < 0.05, heatmap shows time series of Z-statistic. Fz and Pz electrodes from the hd-EEG system. Fig M. Resultant per octile, and resultant vs. Z Stat for experiments 3 and 4. (Ai and Bi) There was a clear linear relationship between alpha power octile and stimulus onset resultant in each condition, in both experiments. This meant that the phase was most consistent between trials in octile 8, and least consistent between trials in octile 1. We considered that this likely resulted from the use of 2 independent EEG systems, since the phase-locking (ecHT) system and the hd-EEG will be in greatest agreement, regarding phase, when alpha power is high, and stimulus onset is determined only by the ecHT. We suggest that the extent of the reset should not be dependent on this onset resultant. (Aii and Bii) The average phase angle was highly consistent across octiles for both experiments. (C) Here, we tested the extent by which the phase reset was related to the resultant at stimulus onset. We took 10,000 samples of 20 trials from each condition, computing stimulus onset resultant, averaging across conditions, and plotting this against auditory-evoked Z statistic. We found a statistically significant (in experiment 1, but not experiment 2), but very weak relationship (R2 values <0.001), and hence confirmed our intuition that z-stat is not strongly dependent on onset resultant. Table C. Number of trials per phase bin (experiments 3 and 4). Phase estimates were used to sort all evoked potentials by stimulus onset phase, into ten 36° bins. N indicates number of participants. Fig N. Phase accuracy plots for the ecHT electrode (in Fz) in the 3 conditions in Experiment 5. Each line represents a participant, length of line indicates resultant (between 0 and 1). Phase accuracy is high in all 3 conditions. Phases are the same for pre-peak and sham. During sham, markers were recorded for each sound stimulus, but the volume was zero. Fig O. Effects of stimulation phase on power and frequency at Fz for the whole duration of the nap opportunity in experiment 5. Time-frequency representation of differences between stimulation conditions across the 30-min nap opportunity, as per paired t tests. Red colours indicate pre-peak>pre-trough and blue indicates pre-preak