Computational protocol: Memory Consolidation Is Linked to Spindle-Mediated Information Processing during Sleep

Similar protocols

Protocol publication

[…] Category recall was our primary measure of memory accuracy. We calculated for each participant: 1) the proportion of target categories recalled at T1 that were subsequently recalled at T2, and 2) the proportion of target categories recalled at T2 that were subsequently recalled at T3 (i.e., following a night of sleep). To avoid any ambiguity related to category memory, we excluded from our analyses any item that was incorrectly classified during the object/scene categorisation task. Across all participants, we excluded 162 items out of a possible 4600 (3.52%). Category recall scores at T1, T2 and T3 were normally distributed in both the nap and wake groups (Kolmogorov-Smirnov test, p > .05), and thus met the assumptions of analysis of variance (ANOVA). As such, the data were subjected to a 2 (TMR: Cued/Not-Cued) X 2 (Group: Nap/Wake) mixed ANOVA. The statistical significance threshold was set at p < .05. Behavioral data were analyzed with SPSS statistics 24. [...] EEG data were analyzed with MATLAB, using the FieldTrip [] (v.06/02/2017) and CircStat [] (v.1) toolboxes. The continuous sleep data were segmented into epochs from −1 s to 3 s around cue onset and subjected to a two-step artifact rejection procedure. In the first step, artifacts were automatically detected and removed based on the median ± 3.5 inter-quartile ranges of both signal amplitude and gradients (the difference between two adjacent samples) of all epochs. In the second step, the remaining epochs were manually screened via FieldTrip’s visual summary functions and epochs containing amplitude, variance or kurtosis outliers were additionally removed. For TMR-cue-locked analysis of event-related potentials (ERPs), data were high-pass filtered at 0.5 Hz and baseline-corrected with respect to the −200 ms to 0 ms window before cue onset. For time-frequency representations (TFRs), data were convolved with a 5-cycles hanning taper and spectral power was obtained from 4-30 Hz in 0.5 Hz frequency steps and 5 ms time steps. For analyses, participant-specific TFRs were converted into percent power change relative to a −300 ms to −100 ms pre-cue window. Because our TFR analysis relied on extended data windows to fit 5 cycles per frequency (e.g., 15 Hz x 5 cycles = 333 ms), a −300 ms to −100 ms baseline window was chosen to mitigate baseline contamination by post-stimulus activity while preserving proximity to cue onset. Note though that TFR comparison of old cues versus control stimuli () revealed the same significant 13-16 Hz power increase when the TFR baseline window matched that of the ERP analysis (−200 to 0 ms).For representational similarity analysis (RSA) of within- versus between-category processing, a sliding window of 200 ms (in steps of 10 ms) was applied to the 0.5 Hz high-pass filtered raw EEG data to obtain, for each trial, a series of 8-channel-by-41-time points (200 Hz/5 ms sampling rate) EEG feature vectors []. Using these feature vectors, Spearman correlations were then used to quantify, for each time point, the representational similarity across all pairwise combinations of trials, resulting in an n trials x n trials correlation matrix. This matrix is symmetrical around the diagonal, and all cells below the diagonal as well as the diagonal itself were removed. Additionally, same-adjective correlations across multiple cueing rounds were removed (that is, we excluded correlations between e.g., adjective x, cueing round 1 and adjective x, cueing round 2). Next, within-category similarity was obtained by averaging across all remaining object-object and scene-scene cells. Between-category similarity was obtained by averaging across all object-scene cells. The numbers of within-category and between-category cells were equated by randomly sub-selecting cells from the majority class in each participant. Each participant’s within-category and between-category correlation time series were Fisher z-transformed to adjust for non-normality of correlation coefficients.All ERP, TFR and RSA analyses were performed as random-effects analyses (paired-samples t tests) and corrected for multiple comparisons using FieldTrip’s nonparametric cluster-based permutation method (1000 randomizations), including channel x time (ERP), channel x time x frequency (TFR) and time (RSA) as cluster-defining features. The statistical significance threshold was set at p < .05. […]

Pipeline specifications

Software tools SPSS, FieldTrip, CircStat
Application Miscellaneous