Dynamics of attentional allocation to targets and distractors during visual search

There is much debate about the neural mechanisms that achieve suppression of salient distracting stimuli during visual search. The proactive suppression hypothesis asserts that if exposed to the same distractors repeatedly, these stimuli are actively inhibited before attention can be shifted to them. A contrasting proposal holds that attention is initially captured by salient distractors but is subsequently withdrawn. By concurrently measuring stimulus-driven and intrinsic brain potentials in 36 healthy human participants, we obtained converging evidence against early proactive suppression of distracting input. Salient distractors triggered negative event-related potentials (N1pc/N2pc), enhanced the steady-state visual evoked potential (SSVEP) relative to non-salient (filler) stimuli, and suppressed contralateral relative to ipsilateral alpha-band amplitudes—three electrophysiological measure associated with the allocation of attention—even though these distractors did not interfere with behavioral responses to the search targets. Furthermore, these measures indicated that both stimulus-driven and goal-driven allocations of attention occurred in conjunction with one another, with the goal-driven effect enhancing and prolonging the stimulus-driven effect. These results provide a new perspective on the traditional dichotomy between bottom-up and top-down attentional allocation. Control experiments revealed that continuous marking of the locations at which the search display items were presented resulted in a dramatic and unexpected conversion of the target-elicited N2pc into a shorter-latency N1pc in association with faster reaction times to the targets. + DL vs. TLDV + TL) with a paired two-sided t-test. In addition, these windowed averages were submitted to a within-subject repeated-measures ANOVA with the factors of laterality (contralateral vs. ipsilateral) x condition (TLDV, DLTV, TL, DL). Planned post-hoc comparisons were made with two-sided paired t-tests. If alpha-band activity serves as a marker of proactive suppression of a distractor, we would expect positive values at contralateral versus ipsilateral electrode sites for the search displays with a lateral distractor.


Introduction
Confronted with a cluttered visual scene like a busy street, we scan for items of immediate relevance and try to ignore distracting stimuli that are known to be irrelevant. The fundamental question of how quickly and efficiently distracting information can be rejected to facilitate search for relevant targets is central to our understanding of how attentional resources are allocated in complex scenes. Many different experimental approaches have been taken in humans to study attentional deployment to different categories of stimuli in visual search arrays ( Eckstein, 2011 ). In studies implementing electrophysiological measures, a detected target typically elicits an enlarged negative event-related potential (ERP) over the contralateral visual cortex, termed the N1pc (latency 150-200 ms) or N2pc (200-300 ms) depending on its latency. These negative ERPs are widely considered to index the allocation of attention to the target item ( Ansorge et al., 2011, Bacigalupo and Luck, 2019, Donohue et al., 2018, Eimer, 1996, Moher and Egeth, 2012, Theeuwes et al., 2000, Theeuwes Jan. 2010. In contrast, the signal suppression hypothesis ( Sawaki and Luck, 2010 ) posits that all physically salient items automatically initiate a pre-attentive, bottom-up signal (i.e., "attend to me"), but when exposed repeatedly to these distractors, top-down guided suppression can prevent attentional capture ( Gaspelin and Luck, 2018a, Luck et al., 2021, Stilwell et al., 2022. Similarily, the contingent capture hypothesis ( Folk et al., 1992 ) holds that to-be-ignored singletons will only capture attention if they match the top-down control settings, i.e., the sought-after target feature, or when the control state is not defined ( Luck et al., 2021 ). So far, previous methods have not provided unambiguous and clear cut evidence favoring one of these competing hypotheses over the other ( Luck et al., 2021 ).
One approach to resolve this controversy would be to relate the well-known ERP signatures of target facilitation (N1pc/N2pc) and distractor suppression (Pd) to other electrophysiological measures of the selective processing of attended and unattended stimuli following the orientation and allocation of attention. Endogenous alpha-band oscillations (in the 10 Hz range) have been regarded as a neural signature of stimulus processing under the influence of attention: An increase in alpha-band amplitude (alpha synchronization) at electrodes contralateral to a stimulus (relative to the amplitude over the ipsilateral hemisphere) has been proposed to signify active suppression of that stimulus (e.g., Snyder, 2011 , Jensen andMazaheri, 2010 )), whereas a lateralized alpha-band amplitude decrease (alpha desynchronization) has been considered a signature of attentional facilitation of contralateral stimuli ( Bacigalupo and Luck, 2019, Forschack et al., 2022, Neuper et al., 2006, Sauseng et al., 2005. Another measure of stimulus processing is the steady-state visual evoked potential (SSVEP), an oscillatory electrophysiological response of the visual cortex to a flickering stimulus having the same frequency as the driving stimulus. The SSVEP amplitude is strongly modulated by attention and can reveal the time course of facilitation or suppression of the cortical response to a particular stimulus ( Andersen and Müller, 2010, Forschack et al., 2016, Gundlach et al., 2020, Gundlach et al., 2021.
In a recent study ( Forschack et al., 2022 ), we concurrently measured and analyzed lateralized ERPs (N1pc, Pd), the time courses of SSVEPs and alpha-band modulations in response to target and distractor stimuli during visual search using two-item search displays. In this study, search displays consisted of either two disks differing in color (a target of one color, a distractor of another color) or two "fillers " of a third color, all presented for brief intervals at randomized adjacent positions within four ongoing, frequency-tagged stimulus streams of gray flickering disks. Results showed that attention was selectively allocated to targets in a goal-driven manner, indexed by SSVEP amplitude increases and a greater contralateral alpha desynchronization relative to the neutral "filler " stimuli, together with a pronounced N1pc. Although distractors elicited a contralateral Pd component, SSVEP and alpha-band amplitude modulations did not indicate that attentional capture was inhibited, even though target and distractor colors were kept constant throughout the experiment. These results argue against the hypothesis that distractor processing is suppressed proactively by a pre-attentive, topdown mechanism ( Gaspelin et al., 2015, Gaspelin and Luck, 2018a, Gaspelin and Luck, 2018c, Hickey et al., 2009, Hilimire et al., 2012, Sawaki and Luck, 2010, at least in simple, 2-item search displays, but instead, that processing of distractor and fillers alike peters out after an initial capture of attention. A possible limitation of that study was that trials containing the neutral "filler " stimuli were never taskrelevant and might have triggered a pronounced disengagement of attention on these trials. Therefore, using these "filler " trials as a baseline for the evaluation of distractor suppression might have been overly conservative.
The present study utilized a similar frequency-tagging design and event-related EEG analysis approach as in Forschack et al. ( Forschack et al., 2022 ) but now with 4-item search displays consisting of color changes at all four locations together with form changes of the target and distractor, which allowed the attentional modulation of target and distractor processing to be quantified relative to the processing of irrelevant fillers that served as a "neutral " baseline within each trial. These 4-item search displays were transiently embedded in ongoing streams of four grey flickering disks and were otherwise designed to be comparable to the displays used in many previous studies that have demonstrated N2pc and Pd components during visual search ( Drisdelle and Eimer, 2021, Gaspelin and Luck, 2018a, Gaspelin and Luck, 2018b, Kerzel and Burra, 2020, Wang and Theeuwes, 2020. As with our previous study ( Forschack et al., 2022 ), a major aim was to investigate whether a distractor singleton (color pop-out) automatically captures attention (as predicted by the reactive suppression hypothesis) or will be suppressed prior to its capturing attention (predicted by the proactive suppression and the contingent capture hypotheses). To this end, we concurrently recorded and analyzed event-related potentials (ERPs), frequency tagged SSVEPs, and endogenous alpha-band activity, while participants searched for a target shape in the transiently presented search displays ( Fig. 1 ). In addition, search displays containing a target alone (with no distractor) and a distractor alone (with no target) were included to investigate the allocation of attention to the respective stimuli in the absence of competitive interactions between the two. Of particular interest was the comparison between the distractor alone condition with the condition when the distractor was presented together with a target, which provided a test of the proposition that the Pd only appears under conditions of direct competition between a target and a distractor ( Hilimire et al., 2012, Kiss et al., 2012. If the distractor singleton can be proactively suppressed in accordance with the signal suppression ( Gaspelin et al., 2015 ) and the contingent capture ( Folk et al., 1992, Luck et al., 2021 hypotheses, we would expect to find a significant decrease in SSVEP amplitude at the distractor location relative to the baseline at the onset of the search display. Such a decrease should be accompanied by greater alpha synchronization at electrodes contralateral to the location of the distractor together with a pronounced contralateral Pd component in the ERP to the display, with no preceding N1pc. A further prediction of the signal suppression hypothesis is that processing of the distractor should be suppressed below the level of the always task-irrelevant filler stimuli ( Gaspelin et al., 2015 ). If that were the case, SSVEP amplitudes should be significantly reduced in response to distractors relative to fillers after search display onset. In the same vein, contralateral alpha synchronization should be greater following the distractor compared to the filler.
Because the present approach (and that of Forschack et al. (2022) ) differed from previous visual search studies in that transient search displays were presented within a continuously flickering stream of disks, we carried out two additional control experiments to investigate whether the absence of flicker (Experiment 2) or the absence of the background disks (Experiment 3) would influence the ERPs and alpha-band modulations to the task stimuli. Thus, in Experiment 1 the four background disks all flickered continuously (see Fig. 1 ), allowing recording of SSVEPs, ERPs, and alpha-band modulations. In Experiment 2, the stimuli and task were identical, but the background disks did not flicker. In Experiment 3, the search displays and task were again identical to Experiments 1 and 2, but the background disks that marked the display items' locations were omitted. We were particularly interested in whether frequency-tagged or marked positions would affect the latencies or amplitudes of the N1pc, N2pc or Pd, and whether these background manipulations would influence the ability to effectively ignore a distractor singleton. To preview the results, the presence of the positionmarking disks, flickering or not, had a profound effect on the ERPs to the search displays. showing an exemplary trial that started with the simultaneous onset of the fixation cross and flickering of grey disks at the frequencies indicated in the last frame (not shown during the experiment). After a baseline period of 600 to 1500 ms, visual search displays of 150 ms duration were presented at the continuously flickering locations at intervals of between 2500 and 3500 ms throughout the trial that lasted about 2 min. Participants discriminated as quickly and accurately as possible the side of the dot (left/right) within the target shape, which was a green diamond. The green circles were irrelevant "fillers. " The salient distractor was a red square. Target and distractor positions were randomized and could either appear individually or together (at adjacent positions) in the displays.

Participants
The study protocol was approved by the local ethics committee (298/17-ek, Ethik-Kommission an der Medizinischen Fakultät der Universität Leipzig). Thirty-six participants (22 female, mean age: 23.5; age range: 19 to 39 years) took part in Experiment 1; 24 participants (16 female, mean age: 22.3; age range: 18 to 30 years) in the Experiment 2 (marked positions without frequency tagging), and 26 participants (19 female, mean age: 24.5; age range: 19 to 38 years) in Experiment 3 (without marked search array positions). Two data sets had to be discarded from the sample of Experiment 1 because of excessive artifacts, leaving 34 participants for the analyses. One participant had to be excluded from all analyses of Experiment 2 because of not passing the step of isoluminance adjustment (see below). One dataset was excluded from the behavioral analysis only, because no behavioral data was saved, and two EEG datasets (plus corresponding behavioral data) were excluded for excessive artifacts. In the end, 20 participants remained in Experiment 2 for the behavioral analyses, and 21 for the EEG analyses. In Experiment 3, one participant had to be excluded from all analyses because of not passing the requirement of having at least 75% correct responses. Two further datasets were excluded due to excessive artifacts in the EEG, so that 23 participants remained. Based on previous investigations of visual alpha-band activity (d ∼0.7 from ( Capilla et al., 2014 )), the Pd component (d ∼0.9 from Gaspelin and Luck (2018a) ) and SSVEP amplitudes (d ∼0.5 from Walter et al. (2013) ), power calculations assuming a beta error probability of 0.2 and an error probability of 0.05 using G * Power ( Faul et al., 2009 ) estimated that sample sizes of 34 participants would be required for experiment 1 and 19 for experiments 2 and 3. Participation was either compensated by class credits or financial reimbursement (10 € per hour). All participants had normal or corrected-to-normal vision. Before the study, participants gave written informed consent and were informed about the nature of the experiment.

Stimuli, experimental procedure, and task
Four grey disks were positioned on an imaginary circumference around a central fixation cross. In Experiment 1, the left disk flickered at a frequency of 28.8 Hz, the bottom at 19.2 Hz, the top at 22.4 Hz, and the right at 25.6 Hz. Each trial began with the simultaneous onset of the fixation cross and the grey disks flickering at all four positions for 600 to 1500 ms. At random intervals of 2500-3500 ms, the four disks either changed in color or in color and shape for 150 ms to form search displays after which the shapes turned back to four grey disks for intervals of 2500-3500 ms, followed immediately by the next search display, and so on for a trial duration of about two minutes (see Fig. 1 ). Overall, there were 26 trials, each comprising 40 randomly presented search displays. Trials were separated by short, participant-paced breaks. Thus, over the entire experiment, 1040 search displays were presented. In 480 of these displays, the target (green diamond) was randomly assigned with equal probability to one of the four possible positions together with an irrelevant singleton distractor (red square) -also assigned randomly to one out of the two adjacent positions. In another 480 displays, the target or the singleton distractor were presented at either left or right positions without an adjacent singleton distractor or target, respectively. In addition, there were 40 displays each with a single target or distractor presented vertically (top or bottom position, equally distributed across trials) in order to reduce a possible spatial bias towards the lateral stimulus positions. All positions not containing a target or distractor were filled by green disks (fillers). In all trials, all colored stimuli contained a dot that was randomly located either to the right or left of the stimulus center. Participants were instructed to search for the target shape (green diamond) and to indicate the side (left or right) at which the target dot appeared by pressing the left or right arrow button, respectively. Response hand was counterbalanced across participants and changed after 13 trials. Before the actual experiment started, participants were trained on the task as in our previous study (see Inline Supplementary Materials ; ( Forschack et al., 2022 )) and all search display stimuli underwent individual isoluminance adjustments with a grey background having the same luminance as the grey disks in order to mitigate any luminancedriven saliency effects and differences between colors. In Experiment 2, stimuli and procedures were identical to Experiment 1 except that stimuli were not flickering and the inter-search-display interval was reduced to an average of 2300 ms with a jitter of + -500 ms. Experiment 3 was identical to Experiment 2 except that the grey disks (position markers) were removed altogether (see Inline Supplementary Materials for more details).

EEG recordings and analysis
EEG was recorded from 64 Ag/AgCl electrodes mounted in an elastic cap with an ActiveTwo Amplifier (BioSemi) at a sampling rate of 512 Hz with a low-pass filter of 104 Hz and stored for later offline analysis. Recording software was ActiView for Windows (version 8.11). Two electrodes were placed horizontally at the outer canthi of both eyes and vertically above and below the right eye to measure horizontal and vertical eye movements and blinks.

General preprocessing of electrophysiological data
All analyses were performed offline. For a detailed description of the preprocessing steps, correction, and rejection procedures, please see Inline Supplementary Material.

Behavioral analysis
For all behavioral analyses, trials in which responses were faster than 400 ms or slower than 1000 ms were excluded, and the remaining trials were pooled together in the following conditions: (1) trials in which the target shape was left or right, and the singleton distractor was at the top, or the bottom position, and (2) trials in which the target was at the top or bottom position and the singleton distractor either left or right. In the following, these trials will be referred to as target lateral -distractor vertical (or TLDV) and distractor lateral -target vertical (or DLTV), respectively. Trials containing only a target at the left or right position were pooled, too, and are designated target lateral (or TL) trials. Trials with a distractor alone at the left or right position were pooled together as distractor lateral (or DL) trials.
Behavioral performance (percent correct discrimination of dot side and reaction time) was compared between the TLDV, DLTV, and TL conditions with paired two-sided t-tests and corrected for multiple comparisons by false discovery rate ( Benjamini and Hochberg, 1995 ) (see Inline Supplementary Material for more Information). Tests for equality were achieved by Bayes Factor (BF) testing employing a standard JZS prior of √ 2/2 ( Rouder et al., 2009 ) as implemented by Bart Krekelbergs (Bayes Factor Matlab package: https://github.com/klabhub/bayesFactor ). Evidence in favor of the null hypothesis (NH, i.e., equality) instead of the alternative hypothesis (AH) is indicated by a BF 01 > 3 ( Jeffreys, 1961, Raftery, 1995.

Analysis of event-related potentials
For ERP analyses, epochs of 200 ms before to 400 ms after search display onset were extracted. Artifact free epochs were averaged separately for each experimental condition. Epochs were pooled over left and right lateral stimulus presentations, and difference ERPs (contralateral minus ipsilateral) were calculated at electrodes PO7/PO8 ( Gaspelin and Luck, 2018a ). These difference ERPs were baseline corrected by the mean voltage between − 200 to 0 ms relative to search display onset. To verify the presence of the N1pc/N2pc and Pd components, these difference waveforms were tested for significance in the time window from 150 to 300 ms after display onset via running t-tests against baseline for each time point. For the correction for multiple comparisons (i.e., multiple time points), threshold-free cluster enhancement (TFCE) was applied with a cluster threshold of p = 0.05 (cluster size exponent E = 0.5, statistical intensity exponent H = 2 ( Forschack et al., 2022, Mensen and Khatami, 2013, Smith and Nichols, 2009) and 100,000 permutations. In our previous study, we identified task-related N1pc and Pd components using that procedure in a two-stimulus search display design ( Forschack et al., 2022 ).
In Experiment 1 of the present study, N1pc and Pd amplitudes were quantified as mean amplitudes of the contralateral minus ipsilateral difference waves within time windows between 155 to 185 ms and 205 to 235 ms, respectively, based on the results of the above-described permutation test. To compare Pd amplitudes between the distractor lateral and the target lateral conditions, ERP amplitudes were averaged in the later window for DLTV and DL, and TLDV and TL, respectively, and submitted to a paired two-sided t-test (DLTV + DL vs. TLDV + TL). Note that in Experiment 2 (non-flickering disks) and Experiment 3 (no disks), the N1pc latencies were shorter, so that the corresponding window was adjusted to cover 120 to 150 ms relative to search display onset.

Analysis of alpha-band amplitude time courses
To extract alpha-band activity, ongoing EEG was convolved on every trial with Gabor kernels centered at 9 to 12 Hz (steps of 0.5 Hz, + /-1.4 Hz FWHM) and subsequently averaged across trials and frequencies for epochs ranging from − 1500 to 1500 ms relative to search display onset. Differences in alpha-band amplitude were tested over the time window from − 500 to 800 ms relative to search display onset between right and left clusters of electrodes (right cluster: PO8, P10, P8; left cluster: PO7, P9, P7; see ( Bacigalupo and Luck, 2019 )) contralateral and ipsilateral to the laterally presented target or distractor to identify time points when alpha-band activity became significantly lateralized. Threshold-free cluster enhancement (TFCE) was applied to the contralateral versus ipsilateral differences with the same parameters as for the ERP time courses described above, with respect to the baseline period ( − 500 to − 200 ms relative to search display onset). Alpha-band lateralization (contralateral minus ipsilateral amplitude) was averaged over the time window from 400 to 800 ms, i.e., spanning the interval when the ERD was at its maximum, and compared between the distractor lateral and the target lateral conditions by averaging these values for DLTV and DL, and TLDV and TL, respectively, and testing the difference (DLTV + DL vs. TLDV + TL) with a paired two-sided t-test. In addition, these windowed averages were submitted to a within-subject repeatedmeasures ANOVA with the factors of laterality (contralateral vs. ipsilateral) x condition (TLDV, DLTV, TL, DL). Planned post-hoc comparisons were made with two-sided paired t-tests. If alpha-band activity serves as a marker of proactive suppression of a distractor, we would expect positive values at contralateral versus ipsilateral electrode sites for the search displays with a lateral distractor.

Analysis of SSVEP time courses
Because every stimulus position was tagged with a unique frequency, the corresponding SSVEPs allowed a time-resolved analysis of earlystage neural processing of the display item at each location. For SSVEP time course analyses, the same epochs as for alpha-band analyses were extracted, and artifact free epochs were averaged for each type of display (see Inline Supplementary Material for details). To quantify SSVEP modulations, averaged epochs for each display condition were convolved with Gabor kernels centered at the four driving frequencies. Importantly, Gabor filter width and electrode clusters were the same as in the analysis of alpha-band activity, resulting in the same temporal, spectral, and spatial resolution. Because the main focus here was on lateralized ERP and EEG activity (and not the upper and lower visual hemifields), the SSVEP analysis was confined to the left and right flickering shapes and on the same left and right electrode clusters as in the alpha-band analysis (which also included the two electrodes used for the ERP analysis). After averaging across the electrodes within each cluster, the SSVEP amplitude time courses contralateral to the target or distractor (again with respect to the − 500 to − 200 ms baseline) were averaged across the left and right target/distractor presentations. These time courses for each condition were then statistically evaluated in the time window between 0-800 ms after search display onset by successive t-tests against zero where p-values were multiple comparisons corrected by threshold-free cluster enhancement (see above). In a further analysis, SSVEP amplitudes were averaged over two successive time windows (0-400 ms and 400-800 ms) for each condition, and those values were submitted to separate 2-by-4 within-subject repeated-measures ANOVAs with the factors of stimulus (target/distractor vs. filler) x condition (TLDV, DLTV, TL, DL).
Planned post-hoc comparisons tested the difference between target lateral (TLDV + TL) and distractor lateral (DLTV + DL) by paired t-tests (see analysis of alpha-band activity above).

Results -experiment 1: flickering disks
Behavioral and electrophysiological analyses compared four stimulus conditions: lateral target presented alone (TL), lateral target together with a vertical distractor (TLDV), lateral distractor presented alone (DL), and lateral distractor together with a vertical target (DLTV).

Event-related potentials
As depicted in Fig. 2 , pronounced N1pc and Pd components were evident in the difference waves formed by subtracting the ipsilateral from the contralateral waveforms. Both components had focal scalp distributions over the contralateral occipital scalp (see topographical insets of Fig. 2 ). These components were very similar to those recorded in our previous study with flickering 2-item displays ( Forschack et al., 2022 ).

SSVEP amplitude time courses
In light of previous experimental evidence that SSVEP amplitudes were enhanced to attended stimuli ( Andersen and Müller, 2010, Forschack et al., 2022, Gundlach et al., 2020, Gundlach et al., 2021, we predicted an early and rapid amplitude increase in response to the taskrelevant targets (green diamonds). Conversely, if singleton distractors are suppressed proactively, the SSVEP amplitude should be reduced after distractor (red square) presentations. In fact, the SSVEP time-courses showed significant early amplitude increases (relative to baseline) in response to both targets and distractors in the first time window (0-400 ms, Fig. 3 ). A stimulus (target/distractor vs. filler)-by-condition-ANOVA of the average amplitude values across the 0-400 ms window revealed a significant main effect of stimulus ( F (1,33) = 32.7, p < 0.001, 2 = 0.066), while the stimulus-by-condition interaction was not significant ( F (2.86,94.38) GG = 0.96, p = 0.41, 2 = 0.005) indicating that both target and distractor amplitudes were greater than filler amplitudes. Directly contrasting target lateral and distractor lateral conditions showed that SSVEP amplitudes in the first time window did not differ significantly (TL + TLDV > DL + DLTV: t (33) = 1.14, p < 0.26, BF 01 = 3). There was no initial SSVEP amplitude increase to fillers, demonstrating that the increases to targets and distractors were not just a consequence of any stimulus presentation. These early SSVEP increases provide strong evidence that attention was immediately shifted to both targets and distractors alike.
The SSVEP amplitude increase was sustained over the interval 400-800 ms after search display onset in response to targets (relative to fillers), but it was reduced following distractors in this late time interval. A similar ANOVA as for the first time window confirmed this pattern by showing a significant interaction of stimulus and condition ( F (2.17,71.58) GG = 4.64, p = 0.011, 2 = .045). A planned comparison indicated that the SSVEP amplitude in the 400-800 ms interval was overall larger following targets than distractors (TL + TLDV > DL + DLTV: t (33) = 2.59, p < 0.02).

Desynchronization of the endogenous alpha-band activity
Ongoing alpha-band activity was desynchronized (reduced compared to baseline) following both lateral targets and distractors, and this desynchronization was significantly greater over contralateral than ipsilateral scalp sites for all conditions in the 400-800 ms interval as indicated by the cluster-based permutation test ( Fig. 4 ). The finding that alpha desynchronization was greater contralaterally following both targets and distractors is consistent with attention being shifted to both types of stimuli ( Bacigalupo and Luck, 2019, Forschack et al., 2022, Neuper et al., 2006, Sauseng et al., 2005. A planned comparison showed that this lateralized amplitude difference was greater following targets Grand mean event-related current source densities contra-and ipsilateral to the green diamond target and red distractor recorded from PO8 and PO7 for the conditions "target lateral -distractor vertical (TLDV)," "single target lateral (TL)," "distractor lateral -target vertical (DLTV)," and "single distractor lateral (DL)." An illustrative search display for each condition is shown in the diagram insets. The difference potential between contralateral and ipsilateral recording sites is shown in orange for each condition, tested against zero in the time range from 150-300 ms relative to the pre-stimulus baseline. The black bold horizontal line indicates significant lateralized potentials: ptfce < 0.05 ( p -value threshold after correction for multiple comparisons). Grey shaded patches indicate the two time windows being used for the calculation of the average N1pc and Pd amplitude. Time zero marks the onset of the visual search display. than distractors in the 400-800 ms interval (TL + TLDV > DL + DLTV; t (33) = − 3.94, p < 0.001).

Results -experiment 2: non-flickering disks
The ERP patterns of our previous ( Forschack et al., 2022 ) and present SSVEP studies were identical, and the question arises as to whether the pronounced N1pc component (and the absence of an N2pc component) observed in both studies was a consequence of flickering (frequencytagging) of the four background disks that marked the positions of the items in the search array. To investigate this possibility, frequencytagging of the disks was removed in Experiment 2 so that these position markers were permanently present in the intervals between the search displays, which were transiently presented for 150 ms as in Experiment 1. Subjects performed the identical search task as in the flicker-present Experiment 1 and viewed identical search display configurations (see Material and Methods and Inline Supplementary Material for more details).

Behavior
As in Experiment 1, the critical comparison between the conditions when the distractor was present vs absent did not result in a significant difference (see Inline Supplementary Table 1 , correct responses: TLDV vs. TL, t (19) = − 1.5; p = 0.14; BF 01 = 1.6; Reaction time: TLDV vs. TL, t (19) = 0.4, p = 0.68; BF 01 = 4. Thus, the presence of a distractor was not associated with behavioral costs. Participants were significantly better in discriminating the side of the dot when it appeared in a lateral target compared to a vertical target presentation together with a lateral distractor (TLDV vs DLTV: t (19) = 2.9; p fdr = 0.01, d = 0.88; TL vs DLTV: Fig. 3. Grand mean baseline-normalized SSVEP time-courses averaged across different stimulus conditions. The horizontal lines indicate significant differences in the SSVEP amplitude time courses with respect to the baseline before search display onset. Significant modulations are indicated by black bold lines: ptfce < 0.05 ( p -value threshold after correction for multiple comparisons for both panels). Note that the SSVEP enhancement begins at or before the onset of the search array due to the filter smearing into earlier time points. Abbreviations as in Fig. 2 . t (19) = 4.5; p fdr = 0.0008, d = 1.17). No other differences between conditions were significant (see Inline Supplementary Material ).

Event-related potentials
Pre-processing of the electrophysiological data was identical to Experiment 1. ERP time courses were extracted from a cluster of three electrodes on each side of the scalp to account for the slightly more spread out topographical distribution of the major components as compared to Experiment 1 (right: PO8/PO4/O2; left: PO7/PO5/O1). Ipsilateral and contralateral ERPs were averaged separately over each cluster. The resulting ERP waveforms were very similar to those of Experiment 1, with prominent N1pc components elicited by the targets and large Pd components elicited by the distractors. However, the latencies of the N1pc and Pd components were some 20 ms earlier compared to Experiment 1 ( Fig. 5 ).
Like the ERPs, the lateralized alpha-band activity modulations following targets and distractors were very similar to those observed in Experiment 1(see Inline Supplementary Material and Supplementary Fig.  1 ). Thus, the results of Experiment 2 demonstrated that the patterns of ERP and alpha-band modulations observed in Experiment 1 and in our previous study ( Forschack et al., 2022 ) were not dependent upon the presence of flickering stimuli.

Results -Experiment 3: no position-marking disks
The results of Experiments 1 and 2 were puzzling in that the targets elicited lateralized N1pc components instead of the later N2pc components that were observed in previous 4-item visual search studies with similar designs ( Drisdelle andEimer, 2021 , Gaspelin andLuck, 2018a ). One difference from the previous studies was the presence of disks (either flickering or not) that marked the stimulus positions in the intervals between display presentations. To investigate whether this made a difference, in Experiment 3 the grey disk position markers were removed completely, making the stimulation comparable to the previously reported 4-item visual search studies.

Behavior
As in Experiments 1 and 2, distractor presence did not result in a significantly different performance as compared to search arrays where Fig. 4. Grand mean contra-and ipsilateral baseline-corrected source current density time courses of alpha-band activity for each condition. Shaded areas indicate 95% confidence intervals relative to pre-stimulus baseline. The difference between contralateral and ipsilateral amplitudes is compared in the time range from 0-800 ms after stimulus onset. The horizontal lines indicate significant alpha-band lateralization: ptfce < 0.05 ( p -value threshold after correction for multiple comparisons for both panels). Abbreviations as in Fig. 2 . the singleton distractor was absent (see Inline Supplementary Table 2

Event-related potentials
Pre-processing of the electrophysiological data was identical to that of Experiments 1 and 2. ERP time courses were extracted from a cluster of the same three electrodes on each side of the scalp (right: PO8/PO4/O2; left: PO7/PO5/O1) and averaged as in Experiment 2 to allow for a direct comparison. As shown in Fig. 6 , the resulting ERP waveforms differed dramatically from those of Experiments 1 and 2. In particular, the N1pc was much reduced in all conditions, and the targets (both TL and TLDV) elicited a prominent contralateral negativity over the interval 180-280 ms that appeared equivalent to the N2pc reported in prior studies. In addition, the Pd elicited by distractors (both DL and DLTV) was much reduced in comparison with Experiments 1 and 2. Evidently, marking the locations of the search display items has a profound influence on the neural processing of both targets and distractors.
To allow a comparison of ERPs between Experiments 2 and 3, component amplitudes were measured over the same time windows for the two experiments: 120-150 ms after display onset for the N1pc and 205-235 ms for the Pd and N2pc. Fig. 7 shows ERP amplitudes of the two time windows of interest for all three experiments.
Direct comparisons of N1pc amplitudes between Experiments 2 and 3, in which stimuli were not flickering but positions of upcoming search arrays were either marked or not marked, confirmed that marking the stimulus positions (Experiment 2) resulted in larger N1pc amplitudes than when the position marks were omitted (Experiment 3) both for target lateral (TLDV + TL: t (42) = -2.5, p = 0.02, d = 0.75) and distractor lateral conditions (DLTV + DL: t (42) = -3.02, p < 0.01, d = 0.91). A similar comparison in the 205-235 ms window confirmed that the Pd elicited by the lateral distractors was much larger when the design included marked stimulus positions (Experiment 2) than when these Fig. 5. Grand mean event-related current source densities contra-and ipsilateral to the green diamond target and red distractor averaged over PO8/PO4/O2 and PO7/PO5/O1 sites for the conditions "target lateral -distractor vertical (TLDV)," "single target lateral (TL)," "distractor lateral -target vertical (DLTV)," and "single distractor lateral (DL)." The difference potential between contralateral and ipsilateral recording sites is shown in orange for each condition. The black bold horizontal line indicates significant lateralized potentials: ptfce < 0.05 ( p -value threshold after correction for multiple comparisons). Grey shaded bars indicate the two time windows being used for the calculation of the average N1pc and Pd amplitude. Time zero marks the onset of the visual search display. marks were omitted (Experiment 3; DLTV + DL: t (42) = 4.4, p < 0.0001, d = 1.34). The same comparison between Experiments 2 and 3 in the 205-235 ms window for the target lateral conditions was also highly significant (TLDV + TL: t (42) = 4.9, p < 0.0001, d = 1.49) due to the elicitation of a robust N2pc in Experiment 3 and only small positivities in Experiment 2.
The modulations of ongoing alpha-band activity in Experiment 3 were highly similar to those seen in Experiments 1 and 2. Alpha-band amplitudes were sharply reduced below baseline in response to all stimuli, with a greater reduction (desynchronization) at contralateral than ipsilateral recording sites, more so for targets than distractors ( Supplementary Online Material and Inline Supplementary Fig. 2 ). Thus, the results of all three experiments, despite having different physical stimu-lation properties, were very similar with respect to modulations of endogenous oscillatory activity and behavioral outcome.

Further analysis: behavioral differences between experiment 2 and experiment 3
Considering the dramatic differences in latency between the targetrelated negativities in Experiment 2 (N1pc) and Experiment 3 (N2pc) it was of interest to find out whether this ERP difference was paralleled by differences between the experiments in reaction time and percent correct target discriminations. As is evident from Inline Supplementary Tables 1 and 2 , reaction times were much faster in Experiment 2 than in Experiment 3 (583 vs. 653 ms, averaged over TL and TLDV conditions; Fig. 6. Grand mean event-related current source densities contra-and ipsilateral to the green diamond target and red distractor recorded from PO8/PO4/O2 and PO7/PO5/O1 for the conditions "target lateral -distractor vertical (TLDV)," "single target lateral (TL)," "distractor lateral -target vertical (DLTV)," and "single distractor lateral (DL)." The difference potential between contralateral and ipsilateral recording sites is shown in orange for each condition. The black bold horizontal line indicates significant lateralized potentials: ptfce < 0.05 ( p -value threshold after correction for multiple comparisons). Grey shaded bars indicate the two time windows being used for the calculation of the average N1pc and Pd amplitude as in Experiment 2. Time zero marks the onset of the visual search display. t (41) = − 4.2, p < 0.001, d = 1.27) and percent correct discriminations were higher (90.1 vs. 82.7%, t (41) = 2.7, p = 0.01, d = 0.82). A regression analysis was conducted to control for a potential effect of dot luminance on behavioral performance (see Inline Supplementary Materials ). Notably, after factoring out behavioral variance due to dot luminance differences across participants, performance differences between experiment 2 and 3 remained. Thus, marking the locations of upcoming search items resulted in markedly faster reaction times and higher correct discriminations.

Discussion
The present series of experiments investigated how attention is deployed to target and distractor stimuli during visual search using a con-tinuous electrophysiological measure (the SSVEP) that closely tracks the time course of attentional allocation to individual stimuli in a multi-stimulus display ( Andersen and Müller, 2010, Forschack et al., 2016, Forschack et al., 2022, Gundlach et al., 2020, Gundlach et al., 2021. While replicating the central finding of absent processing costs with a distractor singleton ( Cosman et al., 2018, Gaspelin and Luck, 2018a, Stilwell and Gaspelin, 2021, Wang and Theeuwes, 2020, the SSVEP modulations observed in Experiment 1 provide clear evidence against the hypothesis that processing well-learned distractors is suppressed proactively ( Gaspelin and Luck, 2018a ) (i.e., prior to attentional capture). In particular, the early (0-400 ms) SSVEP enhancement following the onset of the search display was equivalent in amplitude to both task-relevant targets and irrelevant but salient distractors. Many previous studies have shown that such increments of SSVEP Fig. 7. Lateralized current source density amplitudes averaged within the ERP component time windows for TLDV, TL, DLTV, and DL conditions of each experiment. Statistical comparisons made by paired t-tests are reported in the text. Error bars indicate the 95% confidence intervals of t-tests against zero. Asterisks indicate an amplitude value that is significantly greater than the baseline. * : p < 0.05, * * : p < 0.01, * * * : p < 0.001. amplitude closely track the time course of the deployment of attention to a continuously flickering stimulus ( Andersen and Müller, 2010, Forschack et al., 2016, Forschack et al., 2022, Garcia et al., 2013, Gulbinaite et al., 2017, Gundlach et al., 2020, Gundlach et al., 2021, Hillyard et al., 1997, Zhigalov and Jensen, 2020. Thus, the present SSVEP results indicate that both targets and distractor singletons alike capture attention initially, which is consistent with stimulus-driven theories of attention, but inconsistent with a mechanism of early distractor suppression prior to attentional capture. The concurrent finding of significantly lateralized alpha desynchronization (contralateral > ipsilateral) following the distractors is also consistent with their capturing attention ( Bacigalupo and Luck, 2019, Neuper et al., 2006, Sauseng et al., 2005, as is the pronounced N1pc elicited by the distractor presented alone ( Donohue et al., 2018, Forschack et al., 2022, Hickey et al., 2009, Hilimire et al., 2012, Tay et al., 2019. SSVEP time courses differed between distractors and targets, however, in the later (400-800 ms) interval, with targets eliciting larger amplitudes. The lateralized alpha desynchronization was also greater for targets in this interval in all three experiments. These findings are indicative of a more sustained and enhanced attentional allocation to the stimulus location that contained the targets (which required further dot discrimination), consistent with goal-driven theories of attention. The late decline of SSVEP amplitude to the distractors and concurrently reduced alpha-band lateralization may reflect a withdrawal of attention or a late suppression of distractor processing relative to the initial allocation of attention to the distractor singleton. This result is in line with the reactive suppression (or rapid disengagement) hypothesis, which posits that distractor suppression occurs late, following the initial capture of attention ( Beck et al., 2018, Liesefeld et al., 2017, Moher and Egeth, 2012, Theeuwes et al., 2000, Theeuwes Jan. 2010. It should be noted that the properties of the color pop-out distractor (red square) were kept constant throughout the experiment so as to facilitate its suppression.
In all three of the present experiments, behavioral data showed that distractor presence did not result in behavioral costs. Similar effects have been reported previously and were attributed to suppression of the distractor singleton, which requires statistical learning of the distractor features ( Bacon and Egeth, 1994, Cunningham and Egeth, 2016, Gaspelin et al., 2019, Gaspelin and Luck, 2018a, Luck et al., 2021, Moorselaar and Slagter, 2019, Stilwell et al., 2022, Stilwell and Gaspelin, 2021, Wang and Theeuwes, 2020 and results in target discrimination performance that does not differ between distractor present and absent conditions. To explain this behavioral result, previous studies argued that due to proactive suppression, the set size of a visual search display is effectively reduced to facilitate target detection ( Gaspelin and Luck, 2018c ). In line with this idea, several studies have reported a distractor presence benefit instead of behavioral costs ( Gaspelin et al., 2015, Gaspelin and Luck, 2018a, Kerzel and Burra, 2020, while others found no difference between conditions when the distractor is present or absent ( Cosman et al., 2018, Gaspelin and Luck, 2018a, Hilimire et al., 2012, Kiss et al., 2012, Stilwell and Gaspelin, 2021, Wang and Theeuwes, 2020. The present results, however, suggest that distractors do capture attention initially, in which case the effective set size would not be reduced; rather, the lack of behavioral costs in the presence of a distractor may be a consequence of a late suppression of distractor processing or withdrawal of attention as indicated by the SSVEP and alpha-band modulations described above. Experiments 2 and 3 served as controls to evaluate the influence of the continuously flickering display in the SSVEP experiment on the behavioral and electrophysiological outcomes. A comparison of the results of Experiment 1 (flickering disks) with those of Experiment 2 (non-flickering disks) revealed nearly identical patterns of behavior, N1pc amplitudes (greater for targets), Pd amplitudes (greater for distractors), and alpha-band amplitude modulations, which strongly suggests that flicker per se did not disrupt the dynamics of visual search. Comparison of Experiment 2 (non-flickering disks) with Experiment 3 (no disks), however, revealed a profound shift in the latency of the negativity associated with target processing, in effect converting an N1pc (ca 120-150 ms) to an N2pc (ca 180-280 ms), which closely resembled the N2pc previously described in similar experiments ( Bacigalupo and Luck, 2019, Drisdelle and Eimer, 2021, Gaspelin and Luck, 2018a, Kerzel and Burra, 2020, Liesefeld et al., 2021, Liesefeld et al., 2017. It appears then that continuous marking of the locations of the items in the search display is associated with an earlier contralateral negativity (N1pc), which was larger to targets but also significantly present following distractors. This effect cannot be attributed to the asymmetrical presence of color as in our previous experiment with 2-item displays ( Forschack et al., 2022 ), since all the displays here presented colored (isoluminant) stimuli in all four positions. Although the N1pc and N2pc components had very similar occipital scalp topographies, it is not clear whether they represent a common process that is delayed in the absence of position markers or separate neural operations. In either case, these negativities appear to index the orientation of attention to a lateralized stimulus, either because that stimulus is a search target or because it has salient (pop-out) properties. We suggest that position marking results in more rapid orienting of attention, perhaps because of a pre-activation of the neural ensembles corresponding to the marked locations, which enables a more rapid shifting of attention to those locations. An alternative possibility would be that the target (diamond) and distractor (square) shapes are more immediately discriminable when they emerge from the background of the place-marking gray disk, possibly due to motion transients around their edges. Consistent with these proposals, reaction times for discriminating the target features were much faster (by about 70 ms) when target locations were marked (Experiment 2) than when they were not (Experiment 3). Gaspelin and Luck (2018a ) argued that participants might adopt a singleton search strategy and then search for the most salient item in the display, resulting in attentional capture by the distractor. While a singleton search mode might appear to account for the present finding that distractors elicited an N1pc, increased SSVEP amplitude, and laterally desynchronized alpha-band activity, other evidence suggests that participants actually searched for the instructed feature combination (green diamond target) rather than for the most salient item (red square). If participants in the current Experiments 1 and 2 had been searching for the most salient stimulus (i.e., the red square distractor), the N1pc elicited by the target stimulus in the lateral position should have been delayed when the salient distractor was presented vertically (TLDV condition) relative to the target alone condition (TL), because attentional deployment to the singleton distractor should have occurred first. Similarly, in the Gaspelin and Luck study [ ( Gaspelin and Luck, 2018a ), Experiment 2] and in Experiment 3 of the current study -which was most comparable to experiment 2 of Gaspelin and Luck (2018a ) -N2pc temporal dynamicsbetween the TL and TLDV conditions were highly similar. It should also be noted that in our previous study ( Forschack et al., 2022 ), where two search items (target and distractor) only differed in color (their shapes were identical) so that participants were forced to adopt a feature search mode, an N1pc component was elicited by both lateral targets and lateral distractors as in the present study. Thus, the findings of our past and present studies are wholly consistent with a search strategy of searching for the feature combination of the target rather than for the most salient item.
The event-related modulations of alpha-band activity observed in all three experiments are in line with those of a recent study employing a similar search task ( Bacigalupo and Luck, 2019 ). As in the current design, their study had no cuing period that might have biased alpha-band activity during the pre-stimulus baseline. Relative to that baseline, these studies all observed a pronounced event-related alpha desynchronization (i.e., amplitude reduction) that was greater over the hemisphere contralateral to the critical stimuli. This pattern of alpha-band modulations is consistent with alpha-band activity being an indicator of lateralized attentional allocation or cortical excitability ( Bacigalupo and Luck, 2019, Forschack et al., 2022, Klimesch, 2012, Noonan et al., 2016, Romei et al., 2008, Sauseng et al., 2005, Thut et al., 2006. This attentional allocation was observed here both in response to targets and (to a lesser extent) to distractors, in line with the SSVEP and ERP evidence that the distractors did attract attention. If distractor processing had in fact been suppressed, the predicted pattern would have been an increase in alpha-band amplitude over the hemisphere contralateral to the distractor ( Foxe and Snyder, 2011, Händel et al., 2011, Jensen and Mazaheri, 2010, Rihs et al., 2007, Snyder and Foxe, 2010, Worden et al., 2000. The current studies extend the findings of ( Bacigalupo and Luck, 2019 ), which used transient search array onsets, in that it demonstrates that alpha-band activity measured during continuous stimulation in Experiment 1 (flickering disks) and Experiment 2 (non-flickering disks) is also reduced in amplitude with search stimuli onset. Therefore, the alpha-band results are not specific to transient onset designs but are generalizable to continuous stimulation designs.
The present results call into question the interpretation of the Pd component as an index of early, proactive distractor suppression, i.e., the suppression of distractor processing prior to capture of attention. The distractors were indeed found to elicit an enlarged Pd, but it occurred in conjunction with (1) an enhanced early SSVEP, (2) a reduced contralateral alpha-band amplitude (i.e., greater contralateral alpha desynchronization) compared to ipsilateral sites, and (3) a prominent N1pc preceding the Pd in Experiments 1 and 2. These three electrophysiological indices strongly suggest that the distractor had in fact captured attention initially, contrary to the signal suppression and the contingent capture hypothesis that assume proactive suppression prevents attentional capture by the distractor (see also ( Kerzel and Burra, 2020 )). Thus, if the Pd does reflect a process of top-down distractor suppression, as is widely assumed ( Gaspelin and Luck, 2018a, Hickey et al., 2008, Hilimire et al., 2012, Kiss et al., 2012, Luck et al., 2021, the present results provide evidence that such suppression must be imposed at a later stage of processing that follows the capture of attention by the distractor ( Sawaki et al., 2012 ).
How do the current results relate to previous electrophysiological reports suggesting distractor suppression? While some studies found the Pd onset between 100-200 ms, others reported longer Pd latencies, comparable to those observed in the current study ( Bacigalupo and Luck, 2019, Burra and Kerzel, 2013, Gaspar and McDonald, 2014, Hickey et al., 2008, Kerzel and Burra, 2020, Sawaki et al., 2012. Given their similar scalp distributions and temporal evolution, it has been proposed that N1pc/N2pc and Pd reflect a balance between enhancement and suppression of stimulus processing ( Sawaki et al., 2012 ). Interestingly, despite the latency shift of the N1pc/N2pc across the experiments, suggestive of an attentional pre-activation when the stimulus positions were marked, Pd latency remained largely unaffected. This may imply that the target-related N1pc/N2pc and the distractor-related Pd do not simply represent functionally opposing processes of the same top-down mechanism implementing attentional control ( Luck et al., 2021, Sawaki et al., 2012. The amplitude of the Pd, however, changed markedly between experiments, being much larger in Experiments 1 and 2 than in Experiment 3. This raises the possibility that the latency shift of the lateralized negativity (from N1pc to N2pc) might obscure an initial attentional capture by distractor stimuli in designs employing transient presentation of 4-item search arrays, as evidenced by a reduced overall Pd amplitude to distractors in Experiment 3 (without position markers) as compared to Experiments 1 and 2 (with position markers) despite similar alpha-band dynamics and behavioral outcome. In other words, the distractor in Experiment 3 might have captured attention, too. However, due to the delayed capture in transient ERP designs, a distractor elicited N2pc may have been canceled out by a larger Pd elicited during the same time range (see e.g., ( Gaspelin and Luck, 2018a )). Notably, this account would also explain the concurrent presence of both Pd and N2pc triggered by a singleton distractor as found in experiments 1 and 2, and previous studies Luck, 2013 , Stilwell et al., 2022 ), given that N2pc amplitude and latency are modulated by target-distractor distance ( Feldmann-Wüstefeld et al., 2021 , Kerzel andHuynh Cong, 2022 ) and stimulus saliency ( Brisson et al., 2007, Töllner et al., 2011, factors in which the aforementioned studies most certainly differ. The finding that distractors elicited a Pd whether or not a competing target was present is at odds with the assumption that the Pd resolves stimulus competition between target and distractor by suppressing distractor features. In line with this assumption, previous studies ( Hilimire et al., 2012, Kiss et al., 2012 have reported the Pd in a competition condition (e.g., DLTV) compared to a single distractor condition (e.g., DL), where it was absent. However, the designs in these studies were not easily comparable to the design of the current experiments. In the DL condition in Hilimire et al. (2012) , the distractor was presented without any further stimuli, which reduced the complexity of the display. Therefore, attentional demands and thus, the need to suppress the distractor might not be comparable between 2-item and 1-item conditions. The color singleton distractor in Kiss et al. (2012) had the same shape as the other stimuli while participants were searching for a shape singleton target. In contrast, the color singleton distractor of the current experiments had a different shape than the other three additional task-irrelevant shapes (having the same color as the target), thereby increasing the perceived local physical contrast ( Nothdurft, 1993 ), making it potentially more demanding for the participants to suppress the distractor singleton in search for the target when it is actually absent. In this sense, the Pd might rather reflect general feature disambiguation that is modulated by the local feature contrast without the requirement of the target to be present. In light of the present SSVEP results, an alternative proposal that the Pd reflects attentional capture of the context elements opposite to the distractor ( Kerzel and Burra, 2020 ) seems to be unlikely as this would have resulted in larger SSVEP filler compared to distractor amplitudes. However, distractor but not filler amplitudes were enhanced right after search display onset and were reduced to filler level in a later time window. Further studies are needed to investigate whether the Pd reported in other designs (as outlined above) always coincides with a late reduction of distractor processing as indicated by the SSVEP and whether this pattern reflects: (1) reactive disengagement of attention from the distractor's location ( Belopolsky et al., 2010, Theeuwes Jan. 2010, (2) a state where the distractor features are simply ignored while "zooming in " on the target stimulus ( Forschack et al., 2022 , Liesefeld and, (3) a learning process that these features are task-irrelevant ( Moorselaar and Slagter, 2019 ), or (4) the individual effort of shielding working memory from entering highly distracting input ( Feldmann-Wüstefeld and Vogel, 2019 ).
In conclusion, the present SSVEP, ERP, and alpha-band recordings show that attention is directed to the items in a visual search array in accordance with the tenets of both stimulus-driven (i.e., bottom-up) and goal-driven (i.e., top-down) ( Beck et al., 2018, Liesefeld et al., 2017, Moher and Egeth, 2012, Theeuwes et al., 2000, Theeuwes Jan. 2010 theories. Pop-out color distractors produced no behavioral costs but elicited a sharp and immediate increase in SSVEP amplitude, a pronounced and lateralized alpha desynchronization, and an N1pc in designs with marked positions (Experiments 1 and 2). These findings are consistent with early capture of attention by salient distractors, as espoused by the reactive suppression hypothesis ( Beck et al., 2018, Liesefeld et al., 2017, Moher and Egeth, 2012, Theeuwes et al., 2000, Theeuwes Jan. 2010 ), but inconsistent with the proactive suppression ( Gaspelin et al., 2015, Gaspelin and Luck, 2018a, Sawaki and Luck, 2010 and contingent capture ( Folk et al., 1992, Luck et al., 2021 hypothesis. A similar early increment in SSVEP amplitude was elicited by the task-relevant search targets, indicating an early allocation of attention, as predicted by goal-driven theory. This enhanced SSVEP was sustained following the targets, which required further discrimination, and tapered off following the irrelevant distractors. It appears then that both stimulus-driven and goal-driven allocations of attention can act in concert, and the two corresponding theories should be seen as complementary rather than mutually exclusive ( Awh et al., 2012 ). Surprisingly, the ERP pattern associated with the allocation of attention was highly sensitive to the marking of stimulus locations, with an early N1pc converting to an N2pc when the locations were unmarked (as in most previous studies). Evidently, marking the locations of display items allows for a more rapid allocation of attention to targets and distractors alike, perhaps due to a pre-activation of the neural ensembles encoding those locations. In any case, theoretical interpretations of the N2pc in terms of focusing attention on targets or filtering of irrelevant elements (reviewed in ( Luck, 2012 )) will need to take into account the finding that the N2pc disappears or shifts dramatically in latency when stimulus positions are marked. Future electrophysiological studies investigating the functional role of the distractor positivity (Pd) should take into account the possibility that the Pd might overlap with and cancel out the electrophysiological index of attentional capture by the distractor (i.e., the N2pc) when employing transient stimulation protocols with unmarked locations.

Statement of relevance
There is a long-standing controversy about how attention is allocated to task irrelevant but distracting stimuli in a visual scene. Stimulusdriven (bottom-up) theories emphasize that all salient stimuli will capture attention regardless of their relevance, whereas goal-driven (topdown) theories assert that task-relevant (target) stimuli are primed to capture attention, while capture by irrelevant distractors can be inhibited in a top-down manner. The present study employed electrophysiological measures to show that attention is initially captured by pop-out distractors but is subsequently withdrawn, thereby demonstrating the operation of both bottom-up and top-down selection mechanisms acting in concert during visual search. In contrast, attentional deployment is sustained for targets over a much longer period. Interestingly, the early attentional capture by distractors did not result in significant behavioral costs in responding to the search targets.

Data and code availability statement
Anonymized and preprocessed EEG data, and code to replicate the manuscript figures will be made available at osf: https://osf.io/x934v/ . Versions of third party code that was used to analyze the data are indicated in the text.

Declaration of Competing Interests
The authors declare no conflict of interest.

Data availability
Data will be made available on request.