Fig 1.
Example scene working memory trials.
In a load 2 working memory trial, two encoding images are each presented for 2 seconds followed by a 6 seconds delay (maintenance) period. After the delay, a probe image is shown for 2 seconds followed by a 5 seconds scrambled scene baseline. Some trials contain a negative probe (a) where the probe stimulus is not one of the previously presented encoding images. Other trials contain a positive probe (b) where the probe stimulus is one of the previously presented encoding images. An example average EEG trace from one subject’s 94 artifact free trials is shown (c) with trial phases labeled. Onset of the delay period is marked by a short vertical red line, while offset of the delay period is marked by a longer vertical red line.
Fig 2.
Scalp electrode montage used for EEG recordings.
Scalp electrode positions used in the recordings are shown displayed on a head model. Contour lines represent 3D voltage amplitude mapping in a single subject 2000 ms after delay period onset. Note right posterior focus.
Fig 3.
Averaged whole trial window EEG traces.
An average EEG trace is shown at each sensor (a) for a single subject with 94 trials surviving artifact detection. Electrode T7 is shown in zoomed view (b) showing superimposition of all 94 trial EEG traces from the beginning to end of the time window. Note that a negative potential is an upward deflection in these plots.
Fig 4.
Temporal spectral evolution at the sensor level.
At each sensor, the average TSE for a single subject with 94 trials surviving artifact detection is shown. Positive TSE change is visible after the onset of the delay period (vertical red line), while negative TSE change is evident before and after the onset and offset of the delay period. At each sensor, the x-axis of the TSE matrix shows time relative to the event, and the y-axis displays linearly scales with frequency. TSE amplitude intensities are color coded where blue is negative and red is positive. Note positive red change dominates during the delay period, whereas blue is most prominent during the encoding and probe phases.
Fig 5.
Average delay window TSE and significant clusters of correlation with performance across all subjects.
Average delay period TSE plot (computed using whole delay window as baseline) with no statistical significance overlay (a) shows initial negative change from 6 to 28 Hz at the start of the delay period 0 to 600 ms. Weak positive change (red) is then present from around 500ms to 2925 ms at 6–26 Hz. Stronger positive change is present in the 4–8 Hz from around 275 ms to 1650 ms. A switch to negative change (blue) is present from approximately 4125 ms to the end of the delay period in the 8–26 Hz range and from 2500 ms to 5475 ms in the 4–8 Hz range. Correlation of TSE with subject d-prime revealed three significant clusters (b) in descending order from most significant. Significant clusters are represented by darker shades of red or blue overlaid on TSE activity. Cluster 1 (c, latency 4775–6000 ms and frequency range 4–36 Hz) and Cluster 3 (d, latency 3850–4800 ms and frequency range 4–24 Hz) were negatively correlated with d-prime (Cluster 1: Cluster Value = -993.061, p = .017; Cluster 3: Cluster Value = -640.877, p = .032) and Cluster 2 (e, latency 1400–2450 ms and frequency range 4–24 Hz) was positively correlated with d-prime (Cluster Value = 665.685, p = .022). Dark black lines on the graphs (c-e) are linear regression fits and dotted lines are 95% confidence intervals.
Fig 6.
Effect of baseline window on TSE-performance correlations.
When the whole 17 seconds window from encoding to scrambled scene presentation was used as baseline in computation of TSE (a) a single cluster (latency 1400–3150 ms and frequency range 4–28 Hz) was found that positively correlated with d-prime (Cluster Value = 1109.9, p = .041). When the 4 seconds period of encoding was used as baseline in computation of TSE (b) a single cluster (latency 900–3700 ms, frequency range 4–12 Hz) was found that correlated positively with d-prime (Cluster Value = 830.1, p = .024). When the 5 seconds period of scrambled scene presentation after the probe stimulus was used as the baseline in computation of TSE (c), there were no significant clusters found that correlated with d-prime.
Fig 7.
Temporal spectral evolution at the source level.
A brain region source montage shows the average TSE for a single subject with 94 trials surviving artifact detection. Positive TSE change (red) is evident after the onset of the delay period (vertical red line), while negative TSE change (blue) is evident before and after the onset and offset of the delay period. These changes are particularly noticeable in medial frontal pole (FpM_BR), left posterior temporal lobe (TPL_BR), left parietal lobe (PL_BR), medial parietal lobe (PM_BR), right parietal lobe (PR_BR), right posterior temporal lobe (TPR_BR) and medial occipital pole (OpM_BR).
Fig 8.
Source space analysis reveals TSE-performance correlation in a right parietal region.
Source TSE analysis finds that PR_BR (a) includes a cluster of negative correlation with d-prime (blue, Cluster Value = -3223.92, p = .018) during encoding and a positive correlation with d-prime during the delay (red, Cluster Value = 10.97.58, p = .06) in a whole window analysis (b) and a cluster of positive correlation with d-prime (red, Cluster Value = 680.062, p = .021) in a smaller temporal window that included just the delay period (c). Location of the PR_BR relative to the other sources used to compute TSE is shown in a top down view (d).
Fig 9.
Variance as a function of analysis window and frequency bin.
The log10 raw amplitude variance (normalized by the N-1 samples) was computed for each subject within two windows—the whole 17 sec trial window (red line) and the 6 sec delay period (blue line)–and plotted as a function of frequency (a). Subject differences in log10 raw amplitude variance (whole trial window minus delay period window) are plotted as a function of frequency (b) and show that whole trial window variances are larger than delay period variances in the range between 6 and 40 Hz, with local peak differences occurring at 8 Hz and 22 Hz. Error bars are subject standard errors of the mean.
Fig 10.
A significant cluster of TSE-performance correlation in the scrambled period.
During the 5 second scrambled scene period after probe presentation there was a cluster of positive correlation found between TSE amplitude and d-prime (red, Cluster Value = 660.673, p = 0.017) between 900 ms and 1600 ms after the onset of the scrambled scene (a). Individual subject amplitudes expressed as change relative to the mean across the 5 sec scrambled period are plotted as red filled dots with dark line indicating linear regression fits and boundary lines indicating 95% confidence intervals (b).