Analysis of information flow during a verbal working memory task

The task

In the task, sets of consonants are presented and had to be memorized. The set size (4, 6, or 8 letters) determines WM workload. In each trial, presentation of a letter string (encoding period, 2 s) is followed by a delay (maintenance period, 3 s) where participants activate their phonological loop. After the delay, a probe letter is shown, and participants indicate whether the probe was presented during the encoding period (Boran et al., 2019a).

Analysis steps

1. Data Preprocessing

├── Appy bipolar montage in LFP channels

├── Rereference hippocampal LFP and temporal ECoG using separate iEEG contacts

|── Resample the data to 500 Hz

|── Reject large unitary artifacts in scalp EEG recordings

2. Power Spectral Density (PSD)

├── Multitaper time-frequency transformation

├── PSD computation during encoding and maintenance in hippocampal LFP, temporal ECoG and scalp EEG channels

├── Baseline against fixation

3. Functional Connectivity

├── Phase locking value (PLV) calculation between all pairs of hippocampal LFP and ECoG grid channels.

├──── Multitaper frequency transformation with 2 tapers using FFT, (frequency resolution = 1 Hz)

4. Source reconstruction of EEG using beamforming

├── Forward problem

├──── > Precomputed BEM head-model with 3 compartments (scalp,skull,brain)

├──── > EEG electrodes alignment to the scalp compartment

├──── > Determine source grid locations acoording to AAL parcel

├──── > Source model and leadfield computation

├── Inverse problem

├──── > Linearly constrained minimum variance beamformers

├──── > Reconstruction separate for fixation/encoding /maintenance

├──── > Baseline encoding/maintenance sources against fixation sources

5. Spectral Granger causality (GC)

Evaluation of GC in the range [4 30] Hz ├── > Downsample the signals to 2*Nyquist frequency = 60 Hz

├── > Frequency transformation using multitaper FFT method with 2 Hann tapers, 20 seconds of padding

├── > Non parametric spectral factorization method for GC

├── > GC between hippocampal LFP and beamforming EEG sources

6. Statistics

├── > Cluster-based nonparametric permutation tests

├── > GC differences comparison to null distribution of differences

├── > To test the statistical significance of the spatial spread of contacts with high PSD/PLV/GC we computed the spatial collinearity on grid contacts agains null distribution

├── > Significant beamforming EEG sources during encoding and maintenance are tested against fixation with non-parametric permutation t-test

├── > Group statistics to assess the direction of information flow using group cluster based permutation t-test from FieldTrip

Research Papers

[1] Persistent hippocampal neural firing and hippocampal-cortical coupling predict verbal working memory load Boran E, Klaver P, Hilfiker P, Stieglitz L, Grunwald T, Sarnthein J, Science Advances 2019, Vol.5, Issue 3, doi: https://doi.org/10.1126/sciadv.aav3687

[2] Information flows from hippocampus to auditory cortex during replay of verbal working memory items, Vasileios Dimakopoulos, Pierre Mégevand, Lennart Stieglitz, Lukas Imbach, Johannes Sarnthein, bioRxiv 2021.03.11.434989; doi: https://doi.org/10.1101/2021.03.11.434989

This code has been written by

Vasileios Dimakopoulos (Vasileios.Dimakopoulos@usz.ch)

Parts of the preprocessing and statistics have been implemented initially by

Ece Boran (Ece.Boran@usz.ch)

For questions regarding the task contact

Johannes Sarnthein (Johannes.Sarnthein@usz.ch)