A dual mechanism underlying retroactive shifts of auditory spatial attention: dissociating target- and distractor-related modulations of alpha lateralization

Attention can be allocated to mental representations to select information from working memory. To date, it remains ambiguous whether such retroactive shifts of attention involve the inhibition of irrelevant information or the prioritization of relevant information. Investigating asymmetries in posterior alpha-band oscillations during an auditory retroactive cueing task, we aimed at differentiating those mechanisms. Participants were cued to attend two out of three sounds in an upcoming sound array. Importantly, the resulting working memory representation contained one laterally and one centrally presented item. A centrally presented retro-cue then indicated the lateral, the central, or both items as further relevant for the task (comparing the cued item(s) to a memory probe). Time–frequency analysis revealed opposing patterns of alpha lateralization depending on target eccentricity: A contralateral decrease in alpha power in target lateral trials indicated the involvement of target prioritization. A contralateral increase in alpha power when the central item remained relevant (distractor lateral trials) suggested the de-prioritization of irrelevant information. No lateralization was observed when both items remained relevant, supporting the notion that auditory alpha lateralization is restricted to situations in which spatial information is task-relevant. Altogether, the data demonstrate that retroactive attentional deployment involves excitatory and inhibitory control mechanisms.


Control analyses: Lateral eye movements prior to and following retro-cue presentation
Previous related experiments from the visual domain raised concerns about alpha power asymmetries being potentially confounded by lateral shifts in gaze position1,2. Figure S1 shows that the frontal event-related potentials (ERP) contain a minor fixation offset towards the lateralized item in the sound array prior to retro-cue onset as well as following the retrocue. Since gaze position has been shown to affect the perceived sound eccentricity3, lateral saccadic eye movements may have likewise affected the electrophysiological correlates of attentional orienting towards perceived sound locations. We performed several control analyses to rule out such contamination of our data: First of all, we calculated correlations between single-trial indices of lateral saccadic eye movements and posterior alpha power asymmetries. The ipsilateral minus contralateral difference in ERP amplitude at fronto-lateral channels F9/10 served as a measure of singletrial lateral saccadic eye movements. That is, for left-sided targets or distractors (depending on condition), ERP amplitudes at F9 minus F10 were subtracted, whereas for right-sided targets and distractors ERP amplitudes at F10 minus F9 were subtracted. Note that electrode positions F9 and F10 correspond to the most frontal channels in our EEG setup and are thus comparable to typical hEOG channel locations. Specifically, the ERP asymmetries were measured in the 200 ms preceding retro-cue onset as well as in-between 700 -1300 ms post retro-cue onset (i.e., the same interval used for statistical analysis of alpha lateralization), accounting for lateral eye movements prior to and following retro-cue onset, respectively.
Hemispheric asymmetries in the posterior alpha frequency band were analogously computed by calculating the lateralization index, as described in the methods section of the manuscript. On a single-trial level, this was done separately for right-sided and left-sided targets/distractor. Single-trial alpha asymmetries were assessed in the same frequency range, time interval and electrode cluster used for the main analysis (i.e., [8][9][10][11][12][13] Hz, electrode cluster: PO7/8, P7/8, P5/6, PO3/4, time window: 700 -1300 ms post retro-cue onset).
Pairwise Spearman's Rho or Pearson correlation coefficients (depending on normality properties of the data) were then calculated for each subject and each condition (i.e., target left, target right, distractor left, distractor right) as well as the two time intervals (pre-and post-retro-cue onset). After Fisher-Z transforming the correlation coefficients, one-sample ttests were conducted in order to test for a statistically reliable relation between lateral eye movements and alpha lateralization in each condition. The resulting p-values were FDRcorrected for multiple comparisons4 (corrected p-values are denoted as padj). The scatter plots in figures S2 -S5 illustrate that there was no apparent relationship between the two measures.
The analysis confirmed this, revealing that in both time intervals, the single-subject correlation coefficients were not significantly different from zero, neither for target lateral nor for distractor lateral trials (all t < .01, p > .9, padj < 2.16, BFs < .24). Note that we did not perform this control analysis for neutral trials, since we did not observe any significant lateralization of alpha power in that condition.
In addition, since the above-mentioned correlative approach relies on the presence of null findings (i.e., a non-significant correlation between lateral saccadic eye movements and alpha lateralization), we also ran two repeated-measures analyses of covariance (ANCOVA), including the within-subject factor retro-cue type (distractor lateral vs. target lateral) and a covariate to account for the impact of lateral eye movements prior to and after the cue, respectively. That is, the first ANCOVA included the ipsilateral minus contralateral portions of the average ERP asymmetry across target-lateral and distractor-lateral conditions prior to retro-cue onset as a covariate (electrodes F9/10). This parameter did not include the neutral retro-cue condition. The second ANCOVA included the ERP asymmetry difference between distractor lateral and target lateral trials after retro-cue presentation (700 -1300 ms post retrocue onset) as a covariate. P-values were corrected for multiple comparisons across the two ANCOVAs using FDR-correction4.

Bilateral alpha power desynchronization as a measure of cognitive task demands
The line plots in Fig. 5 (a-c) clearly illustrate that there is a bilateral suppression of alpha power following retro-cue onset. Here, this desynchronization of alpha power appears to be more pronounced in distractor lateral trials than in target lateral or neutral trials. To statistically assess differences in alpha desynchronization between conditions, we performed an additional one-way repeated measures ANOVA, including the factor retro-cue type and bilateral, baseline-corrected alpha power as a dependent variable. Baseline-corrected ERSPs were computed, using Morlet wavelet convolution as described in the method section, but a spectral baseline was extracted for each frequency (-300 to 0 relative to pre-cue onset). Mean alpha power (8)(9)(10)(11)(12)(13) was computed for each subject and the three conditions (i.e., target lateral, distractor lateral, neutral) at a posterior electrode cluster (PO7/8, P7/8, P5/6, and PO3/4) in-between 700 to 1300 ms post retro-cue onset (i.e., using the same parameters as for alpha lateralization). Such desynchronized alpha activity, resulting in low levels of alpha power (i.e., small amplitudes), has been associated with states of high excitability5 and is commonly interpreted as a mechanism reflecting functional engagement and information processing6. Accordingly, the event-related desynchronization of alpha power has been associated with stimulus processing (as opposed to 'idling')7, increased working memory load8,9, and greater semantic elaboration10. In line with an interpretation as a signature of cognitive processing demands, distractor lateral trials in the current study presented the acoustically most challenging spatial condition, because the to-be-attended (central) sound was originally embedded by twoneighboring sounds. Thus, one may speculate that the representation generated at encoding is likely to be of lower quality than that of the lateral sound stimuli and may thus require more attentional resources to be re-focused within working memory. To follow up on this, we