Behavioral analysis showed WM response time (RT) to increase as memory load increased (two items: ‘motion’ = 908 ms, s.e.m. = 58 ms, ‘color’ = 906 ms, s.e.m. = 64 ms; four items: ‘remember both’ = 1019 ms, s.e.m. = 60 ms; P < 0.005). There was no significant difference between color and motion tasks in RT (P > 0.05), and no significant differences between any of the tasks in accuracy (‘motion’ = 78%, s.e.m. = 2%; ‘color’ = 74%, s.e.m. = 2%; ‘remember both’ = 73%, s.e.m. = 2%; P > 0.05).
To evaluate the relationship between neural measures associated with top-down modulation (attending and ignoring) and indicators of WM performance, trials from the ‘motion’ and ‘color’ tasks for each participant were split by fast and slow responses across the median RT. This approach is consistent with many other studies that have used RT as a sensitive indicator of behavioral performance (e.g. Stroop, 1935
; Posner et al., 1978
). Indeed, fast RT trials were associated with enhanced WM accuracy in both ‘motion’ (fast = 82%, s.e.m. = 3%; slow = 73%, s.e.m. = 3%; P
< 0.005) and ‘color’ tasks (fast = 81%, s.e.m. = 2%; slow = 67%, s.e.m. = 2%; P
< 0.001). Subsequent analyses of ERP data are separately evaluated for all trials (average performance), high-performance trials (i.e., fast RT) and low-performance trials (i.e., slow RT).
Event-Related Potentials during Stimulus Encoding
To identify neural markers of top-down modulation during the encoding period of each task, we compared peak amplitude measures of event-related potential (ERP) waveforms time-locked to relevant and irrelevant stimuli during the two-item WM tasks. This analysis focused on two peaks, the P1 (a positive deflection, approximately 100 ms post-stimulus onset) and the N1 (a negative deflection, approximately 170 ms post-stimulus onset), which reflect early stages of visual processing and have previously been shown to be modulated by different types of attention (Rugg et al., 1987
; Hillyard et al., 1998
; Valdes-Sosa et al., 1998
). displays ERPs to attended and ignored stimuli from posterior electrodes. Attentional modulation at the P1 and N1 is apparent in parietal and occipital regions. Therefore, average peak amplitude measures from the electrodes of interest for the P1 and N1 were submitted to an ANOVA with stimulus (color / motion) and condition (attend / ignore) as factors.
ERP timeseries from select posterior-occipital electrodes when viewing A, motion or B, color stimuli. Note: electrode locations are an approximation.
The P1 amplitude displayed a main effect for both stimulus (F(1,18) = 13.74, P < 0.005) and condition (F(1,18) = 6.08, P < 0.05), indicating motion yields a larger P1 than color (motion: 3.21 µV, s.e.m. = 0.30 µV; color: 1.83 µV, s.e.m. = 0.15 µV) and that the amplitude of the P1 was modulated by attention (attend = 2.72 µV, s.e.m. = 0.29 µV; ignore = 2.32 µV, s.e.m. = 0.23 µV). Additionally, a stimulus by condition interaction (F(1,18) = 9.42, P < 0.01) was observed. Post-hoc analysis identified significant attentional modulation (attend > ignore) of the P1 amplitude for motion stimuli (, P < 0.01), while no such modulation was observed for color (P > 0.05). Taken together, these results indicate that the P1 elicited by motion stimuli is modulated by attention.
Figure 3 Attentional modulation at electrode of interest. A, ERP waveform for attended (solid line) and ignored (dashed line) motion stimuli. Attentional modulation is observed at the P1. B, Comparison of the P1 modulation index (difference between attended and (more ...)
The N1 amplitude displayed a main effect for condition (F(1,18) = 8.57, P < 0.01), such that attending to the stimuli elicits a larger (i.e. more negative) amplitude than ignoring the stimuli (attend = −4.90 µV, s.e.m. = 0.41 µV; ignore = −4.51 µV, s.e.m. = 0.42 µV). Moreover, a stimulus by condition interaction (F(1,18) = 15.50; P < 0.005) was also observed. Post-hoc analysis indicated that the N1 amplitude to color stimuli was modulated by attention (, P < 0.001), whereas motion stimuli did not display attentional modulation (P > 0.05). Thus, the N1 elicited by color stimuli is modulated by attention and the color amplitude drove the observed main effect.
Overall, the P1 amplitude to motion stimuli was modulated by attention whereas the N1 to color stimuli displayed attentional modulation. Therefore, further ERP analysis utilized these measures as functional markers of top-down modulation during WM encoding.
To identify if there as an impact of WM load on markers of early attentional modulation, paired t-tests were conducted to compare the ‘remember both’ (four-item)_task with the two-item tasks (P1 for motion and N1 for color). Results indicate that stimuli from the four-item task elicited similar ERPs as stimuli from the two-item tasks for the same relevant dimension (P1 for motion and N1 for color; P > 0.05).
To evaluate if the degree of top-down modulation differed across low- and high-performance trials in the two-item tasks, independent evaluations of these trial subtypes were conducted. Comparable to the behavioral analysis, high- and low-performance ERPs were calculated based on the median RT split from all artifact-free trials. Peak P1 and N1 amplitudes from motion and color stimuli, respectively, were submitted to an ANOVA with condition (attend / ignore) and performance (high / low) as factors.
The P1 evoked by motion stimuli yielded main effects for both condition (F(1,18) = 9.25, P < 0.01) and performance (F(1,18) = 7.09, P < 0.05), such that attentional modulation (attend > ignore) was observed (attend = 3.56 µV, s.e.m. = 0.32 µV; ignore = 2.78 µV, s.e.m. = 0.32 µV) and low-performance trials produced a larger P1 amplitude compared to high-performance trials (low = 3.42 µV, s.e.m. = 0.31 µV; high = 2.92 µV, s.e.m. = 0.34 µV). Additionally, a condition by performance interaction was observed (F(1,18) = 4.81, P < 0.05). Post-hoc analysis revealed significant modulation (attend > ignore) during high-performance trials (light gray bars, , P < 0.005), but not during low-performance trials (dark gray bars, ; P > 0.05). Furthermore, paired t-tests on the magnitude of modulation (attend – ignore) identified significantly reduced attentional modulation in low-performance relative to high-performance trials (P < 0.05). Therefore, high WM performance for motion stimuli is associated with P1 attentional modulation, whereas a lack of modulation at the P1 is associated with low WM performance.
The N1 evoked by color stimuli displayed a main effect for condition (F(1,18) = 7.61, P < 0.05), such that attentional modulation (attend > ignore, i.e. more negative) was observed (attend = −4.58 µV, s.e.m. = 0.41 µV; ignore = −3.77 µV, s.e.m. = 0.36 µV). Moreover, a condition by performance interaction was observed (F(1,18) = 5.21, P < 0.05). Similar to the P1 for motion stimuli, post-hoc analysis of the color-evoked N1 amplitude revealed significant modulation (attend > ignore) during high-performance trials (light gray bars, , P < 0.005), but not during low-performance trials (dark gray bars, ; P > 0.05). Furthermore, paired t-tests on the magnitude of modulation (attend – ignore) identified significantly reduced attentional modulation in low-performance relative to high-performance trials (P < 0.05). Thus, high WM performance for color stimuli is associated with N1 attentional modulation, whereas a lack of modulation at the N1 is associated with low WM performance.
These results suggest that top-down modulation of the P1 for motion stimuli and the N1 for color stimuli reflects subsequent WM performance after the delay period. However, this analysis did not identify whether the absence of modulation observed during the low-performance trials was the result of a lack of attention to relevant stimuli, failure to ignore irrelevant stimuli, or a combination of both. To assess why top-down modulation differs with performance, additional post-hoc analyses on the ERPs from the low- and high-performance trials directly compared both relevant and irrelevant stimuli from the two-item tasks.
Analysis revealed a marked difference based on stimulus relevance. When participants viewed relevant stimuli, no difference existed in the ERPs between low- and high-performance trials, for either motion or color stimuli (; P > 0.05). However, when participants viewed irrelevant stimuli, the ERPs were significantly different across the performance split, such that neural markers of attentional modulation were increased in low-performance trials (; motion & color; P < 0.05). Moreover, the peak ERP amplitude to irrelevant stimuli during low-performance trials did not differ significantly from those associated with relevant stimuli during the low-, or even high-performance trials (dashed black line, ). This data suggests that during low-performance trials, participants allocated as much attention to irrelevant stimuli as they did on trials when that same stimulus was task-relevant. Taken together, these results indicate that WM performance is not associated with fluctuations in attention to relevant stimuli during encoding, but rather with fluctuations across trials in adequately ignoring irrelevant stimuli. This result is supported by ERP analyses comparing low- and high-performance trials from the ‘remember both’ task, in which all stimuli were relevant. The analysis revealed no significant difference between the P1 for attended motion stimuli or the N1 for attended color stimuli across performance subtypes (P > 0.05), thus supporting the conclusion that variation in attention to relevant stimuli during encoding was not associated with WM performance level.
Figure 4 ERP comparisons of low-performance (dark gray line) and high-performance (light gray line) trials for attended and ignored stimuli. Inset bar graphs compare designated peak ERP measures between low- and high-performance trials. A, No differences observed (more ...)
Time-Frequency Analysis during WM Maintenance
In the previous section, we dissociated neural signatures of attending and ignoring visual stimuli during encoding and revealed their differential impact on subsequent WM performance. However, we are also poised to explore the neural consequences of improper attentional allocation during encoding on WM maintenance activity. Given that low-performance trials are associated with a failure to ignore irrelevant stimuli, we hypothesized that during these trials participants encoded more items into memory than during the high-performance trials, resulting in interference that subsequently diminished recognition performance. To evaluate this, an initial time-frequency analysis of the delay period was conducted to determine whether performance differences exist in the alpha (8–12 Hz), beta (12–20 Hz) or gamma (20–50 Hz) frequency ranges.
For each frequency band, average values from consecutive 500 ms time-frequency windows with 50% overlap were extracted across the delay period beginning 250 ms after the final stimulus offset to prevent the confound of including a stimulus offset response. Average values were submitted to an ANOVA with stimulus (color / motion), performance (high / low), time-window and electrode as factors. Five electrodes distributed about the scalp were selected for this analysis: FCZ, PZ, P5, P6 and IZ. shows desynchronized activity (decreased activity in blue) in the alpha and beta-bands during the first 1500 ms of the delay period, whereas subsequent activity is synchronized (increased activity in red). To explore changes independently within these regions of synchronized and desynchronized activity, the alpha and beta-bands were subjected to two separate ANOVAs: one with time-windows spanning 250–1500 ms and one with time-windows spanning 1500–4000 ms of the delay period. The gamma-band displayed no such temporal structure; therefore, the analyzed time-windows spanned the entire delay period.
Figure 5 Time-frequency maps from five electrodes of interest during the delay period. The upper maps are from the ‘motion’ task during A, high-performance trials and B, low-performance trials. The lower maps are from the ‘color’ (more ...)
Desynchronized alpha activity early in the delay period displayed no main effects or interactions. However, synchronized alpha activity later in the delay period exhibited main effects for both time (F(8,144) = 5.36, P < 0.05) and electrode (F(4,72) = 4.68, P < 0.005), such that the alpha activity peaks between 1750–2250 ms and was greatest at electrode P6.
Desynchronized beta activity early in the delay period displayed main effects for time (F(3,54) = 12.98, P < 0.001) and electrode (F(4,72) = 4.14, P < 0.05). The decrease in activity was greatest early in the delay period (250–750 ms) and was most pronounced at electrode P6. Furthermore, a performance by time interaction was observed (F(3,54) = 6.45, P < 0.005). Post-hoc analysis indicated that the performance difference was greatest at the earliest time-window (250–750 ms) such that low-performance trials (mean z-score: −0.57, s.e.m. = 0.04) were more desynchronized than the high-performance trials (mean z-score: −0.31, s.e.m. = 0.04; P < 0.005). Synchronized beta activity displayed main effects for time (F(8,144) = 13.63, P < 0.001), electrode (F(4,72) = 6.73, P < 0.001) and stimulus (F(1,18) = 4.91, P < 0.05). The synchronized beta activity peaked between 1500–2000 ms, was greatest at electrode P6 and was more pronounced during the delay period of the ‘color’ task.
Analysis of the gamma-band revealed a main effect for time (F(13,234) = 9.86, P < 0.001), such that it was most desynchronized at the beginning of the delay period and displayed peak synchronization between 1250–1750 ms into the delay period. Additionally, a time by electrode interaction (F(52,936) = 2.91, P < 0.005) was observed.
Taken together, these results show that performance differences only exist early in the delay period as reflected by beta-band desynchronization. To determine if low-performance trials from ‘motion’ and ‘color’ tasks (encode two items) were comparable to ‘remember both’ (encode all four items) during the WM maintenance period, further analysis of the delay period focused on the early desynchronized beta band activity (black boxes, ). Because no stimulus main effect or interaction was observed, subsequent analysis of the spectral activity during the delay period was averaged across motion and color data. Thus, comparisons between performance and WM load utilized an ANOVA with electrode and type (high performance two-item task, low performance two-item task and average performance four-item task) as factors.
Analysis of the desynchronized beta activity resulted in main effects for electrode (F(4,72) = 8.47, P < 0.001) and type (F(2,36) = 4.64, P < 0.05). Post-hoc analysis indicated that beta desynchronization was greater at the lateral electrodes (mean z-scores: P5 = −0.55, s.e.m. = 0.06; P6 = −0.60, s.e.m. = 0.05) compared to the midline electrodes (mean z-scores: FCZ = −0.42, s.e.m. = 0.07; PZ = −0.41, s.e.m. = 0.07; IZ = −0.37, s.e.m. = 0.07; P < 0.05). Furthermore, the high WM load task (, black bar) yields more beta desynchronization compared to the high-performing two-item task (, light gray bar), whereas no difference is observed between the high WM load and the low-performing two-item task (, dark gray bar). Overall, these results suggest that attending to two relevant stimuli and failing to ignore two irrelevant stimuli (i.e. two-item tasks, low-performance) yields WM maintenance activity measures that are comparable to maintaining all four items in WM. This data is paralleled by behavioral analyses that revealed the RT of low-performance, two-item trials (‘motion’ = 1,084 ms, s.e.m. = 76 ms; ‘color’ = 1,120 ms, s.e.m. = 94 ms) were not significantly different from the mean RT of the four-item task (‘remember both’ = 1,025 ms, s.e.m. = 57 ms; P > 0.05), whereas high-performance trials were significantly faster (‘motion’ = 713 ms, s.e.m. = 34 ms; ‘color’ = 692 ms, s.e.m. = 28 ms, ‘remember both’ = 1,025 ms, s.e.m. = 57 ms; P < 0.05). Taken together, these data suggest that failure to ignore irrelevant information reduces WM performance by increasing the memory load during WM maintenance.
Figure 6 Average beta-band (12–20 Hz) activity between 250–750 ms into the delay period. Greater beta-desynchronization is observed during the four-item task (black bar) and low-performance two-item task (dark gray bar) relative to the high-performance (more ...)
Event-Related Potentials during Probe Stimulus
To identify if WM performance may also be influenced by attentional changes at retrieval, ERPs to the probe stimuli were analyzed. No P1 or N1 differences between low- and high-performance trials on either ‘motion’ or ‘color’ tasks were observed in response to the probe stimulus (P > 0.05, all comparisons). This suggests that variation in WM performance across trials was due to improper allocation of attention during the encoding stage and not during WM retrieval.