MEG analysis

Following band-pass filtering (0.5–45 Hz), independent component analysis (ICA) was performed with runica algorithm in EEGLAB [41] in order to remove artifact from ocular, cardiac, and muscular sources.

Since head position within the MEG helmet can vary across participants, global field power (GFP) was utilized to provide a measure of the overall activity across all channels [42]. Individual-level GFP was computed for the congruent and incongruent conditions using trial-averaged event-related fields. A bootstrapping approach was utilized to determine time intervals of significant difference between conditions, in which the GFP signals at each time point were permuted between the congruent and incongruent conditions across all subjects [43]. Using this approach, the interval of significance was identified to be 300–500 ms and used as the window of interest in subsequent analyses, consistent with prior literature [44, 45]. The mean GFP value in this time interval was then calculated for each condition (congruent and incongruent) and participant, and compared using paired t test at the group level.

Sensor level time–frequency analysis was undertaken by convolution of the data with Morlet wavelets (6 cycles) using the continuous wavelet transform function in MATLAB (The Mathworks Inc., USA). The coefficients corresponding to 0.5–45 Hz frequency in the − 200 to 900 ms time window relative to stimulus onset were extracted, and log power was computed as the square of the absolute value of the coefficients. To better understand the event-related spectral changes, the mean log power in the baseline period (− 100 to 0 ms) was subtracted from the log power in the post-stimulus period for every trial within the frequency band. Significance was assessed using a bootstrapping approach by permuting the trial-averaged wavelet power in the congruent and incongruent conditions across participants in each frequency [43]. This entailed the calculation of T-statistic for each time point and frequency between the congruent and incongruent conditions in the 800 ms following stimulus presentation. Thereafter, 1000 permutations were undertaken and new T-statistic calculated for every permutation leading to a null distribution against which the significance of the true T-statistic was assessed (with p < 0.05 considered to be significant).

Source level analysis was performed using SPM8 (Welcome Trust Centre for Neuroimaging, UK) with the forward and inverse modeling steps elaborated in previously published work [46]. Source analysis for localizing neural generators of the semantic language process was undertaken using minimum norm estimates (MNE) to maintain consistency with prior N400 studies in MEG [24, 44]. Group constraints were employed during inversion [47], and source reconstruction was based on trial-averaged data within the entire frequency range (0.5–45 Hz) and active epoch (0–900 ms relative to stimulus presentation). Source-level contrast images were derived using data in the 0.5–45 Hz frequency range and previously identified window of 300–500 ms. Statistical modeling employed a general linear model (GLM) with T-contrasts [48].