EEG.

EEG data were recorded from a BioSemi ActiveTwo system with 64 Ag/AgCl active electrodes (Cortech Solutions). Signals were amplified and digitized at 2,048 Hz with a 24-bit resolution and no online filter. All electrode offsets were maintained at ±20 mV. EEG analysis was conducted in MATLAB 2020b (MathWorks). Raw EEG data were down-sampled to 512 Hz and bandpass filtered between 0.1 and 50 Hz with a zero-phase shift finite impulse response filter. Data were epoched into segments beginning 0.1 s prior to cue and ending 0.8 s after probe stimuli. Eye artifacts were removed through an independent component analysis (ICA) by excluding components consistent with topographies for blinks, eye movements, and the electrooculogram time series. Noisy (or bad) electrodes were excluded from ICA, then recreated using spherical interpolation. EEGLAB functions (117) were used for ICA (binica), spherical interpolation (pop_interp), and plotting topographies (topoplot). Data were rereferenced to the average and epochs containing artifacts greater than ±75 μV were excluded.

For analysis of the ERP to target stimuli, epochs were baseline corrected to the 0.1-s pretarget period prior to averaging for the ERP. ERP amplitudes were assessed based on normalized GFP, which was calculated by first normalizing the ERP data through a Z score across all time points, electrodes, conditions, and sessions per participant. The normalized amplitudes were then squared and averaged over all electrodes. Peak P1 values were chosen as the largest local peak GFP between 50- and 150-ms post stimulus onset, the N1 was identified as the largest local peak GFP between 150 and 250 ms, and the P3 values were chosen as the largest local peak amplitude between 250- and 350-ms post stimulus onset. These temporal windows do not reflect the range of observed peak latencies, but rather serve to guide selection of ERP measures. Mean GFP was measured by averaging over a temporal window (10 ms for P1 and N1; 40 ms for P3) centered around each individual participant’s peak before statistical analysis. Source localization was conducted using the LORETA-KEY software (118). P3 data to target stimuli were submitted to standardized low-resolution electromagnetic tomography (sLORETA).

For analysis of the CNV, epochs were baseline corrected to the 0.1-s precue period prior to averaging across trials. Statistical analysis of the CNV utilized electrodes from central parietal-occipital regions (CPZ, POZ, PZ, P1, P2), in line with previous research on the use of this neural metric as an index of temporal attention (70). Mean amplitudes were calculated within a window of 0.3 to 0.1 s prior to target onset.

Spectral data (4 to 50 Hz) was extracted via complex Morlet wavelets (family ratio: f_o/σ_f = 7) applied to the epoched data. Spectral power was calculated from the wavelet coefficients by averaging the magnitude of the wavelet coefficients over trials. Spectral power was then normalized for each participant by calculating the Z score over time from data that were concatenated across all conditions and cue types. This normalization was done independently for each session (pre/post-training). Analysis focused on alpha band (8 to 12 Hz) activity during the end of the expectation period by averaging data 0.3 to 0.1 s prior to target onset. Analysis of alpha activity utilized electrodes from lateralized parietal–occipital regions by averaging over electrodes from the left (P7, P9, PO3, PO7, O1) and right (P8, P10, PO4, PO8, O2) hemispheres, in line with previous research on the use of this neural metric as an index of temporal attention (70).