Supplementary Table 1: Results of linear classifier distinguishing between UP and DOWN states based on EEG power.
Also shown are accuracies of ‘null models’ with random label permutation performed on the same data, along with statistical comparison across subjects of true accuracies vs randomly permuted accuracies (Wilcoxon signed rank tests).
Supplementary Table 2: F tests following mixed effects models comparing background EMG across conditions and blocks. Shown is feedback type (UP,DOWN) by block number interactions on each session.
Supplementary Table 3: Percentage of EMG trials retained. Shown is the percentage (of total number of collected MEP trials) that were retained per block and per session for analysis following post-processing screening of background EMG amplitude.
Supplementary Table 4: Results of mixed effects models on the percentage of retained background EMG trials during neurofeedback. The dependant variable was the percentage of trials that were retained following screening for background EMG crossing the criterion threshold. Four models were performed; One on data during all neurofeedback trials (10 levels of ‘trial’) modelling changes over the duration of training (across trials) that were modulated by trial type (UP or DOWN), demonstrating no systematic effects of MEP amplitude modulation on the quantity of trials that were retained for analysis. The three remaining models were conducted separately on the data from Session 1, Session two and Session 3, again revealing no significant modulation on any session. Mixed models employed a compound symmetry covariance matrix, fixed effects of ‘trial’ and ‘trial type’, and random effect of ‘participant’.
Supplementary Table 5: F tests following mixed effects models comparing background EMG across conditions and blocks for the retention test conducted months following initial training. Shown is feedback type (UP, DOWN) by block number two levels, baseline + retention) interactions for each of the four recorded muscles.
Supplementary Table 6: In debriefing following the experiment, all participants were asked to write down their final strategies for the UP and DOWN conditions, in their own words.
Supplementary Table 7: Outline of experimental timeline for each day of testing. Each block of data collection is shown in a separate cell, with the beginning of testing indicated by red vertical lines. Each row depicts testing carried out on separate days unless indicated otherwise (Session 3). Blocks in which MEP neurofeedback was conducted are shaded in grey. All other measurements were taken at rest. Sessions 1 and 2 (blocks 1-8) were considered as ‘training’, as auditory and small financial rewards were presented during feedback. All subsequent blocks of neurofeedback (Session three onwards) were conducted post-training, without financial rewards. Auditory feedback was removed only for the EEG session, to avoid auditory-evoked potentials. The order of UP and DOWN feedback session types was counterbalanced across participants for all sessions. Assessment of resting motor threshold (RMT) was carried out at the beginning of every day of testing. A recruitment curve measurement was performed only at the beginning of the first day in order to establish TMS intensity (as a % of RMT) that evoked MEPs of an amplitude that was 50% of the individual’s maximum before plateau. This % of RMT was used on every subsequent session to elicit MEPs during neurofeedback and for resting blocks.