Abstract
Adaptive behaviors require the ability to resolve conflicting information caused by the processing of incompatible sensory inputs. Prominent theories of attention have posited that early selective attention helps mitigate cognitive interference caused by conflicting sensory information by facilitating the processing of task-relevant sensory inputs and filtering out behaviorally irrelevant information. Surprisingly, many recent studies that investigated the role of early selective attention on conflict mitigation have failed to provide positive evidence. Here, we examined changes in the selectivity of early visuospatial attention in male and female human subjects performing an attention-cueing Eriksen flanker task, where they discriminated the shape of a visual target surrounded by congruent or incongruent distractors. We used the inverted encoding model to reconstruct spatial representations of visual selective attention from the topographical patterns of amplitude modulations in alpha band oscillations in scalp EEG (∼8–12 Hz). We found that the fidelity of the alpha-based spatial reconstruction was significantly higher in the incongruent compared with the congruent condition. Importantly, these conflict-related modulations in the reconstruction fidelity occurred at a much earlier time window than those of the lateralized posterior event-related potentials associated with target selection and distractor suppression processes, as well as conflict-related modulations in the frontocentral negative-going wave and midline-frontal theta oscillations (∼3–7 Hz), thought to track executive control functions. Taken together, our data suggest that conflict resolution is supported by the cascade of neural processes underlying early selective visuospatial attention and frontal executive functions that unfold over time.
Significance Statement
The ability to resolve conflict is essential for adaptive behaviors. Here, we utilized a model-based approach to examine conflict-related changes in spatial representations of visuospatial attention from alpha band oscillations (∼8–12 Hz) in EEG obtained from human participants performing an attention-cueing Eriksen flanker task. We observed increased fidelity of neural representations of early visuospatial attention immediately after stimulus onset in trials with incongruent compared with congruent distractors, which occurred more rapidly than overall changes in parietal and frontal activity. This finding implicates the role of early selective visuospatial attention in mitigating cognitive interference and emphasizes that attention mechanisms involve multiple selection processes that evolve over time.
Introduction
Selective attention is essential for navigating complex environments like busy urban streets, where abundant and conflicting sensory inputs pose challenges to the limited capacity of the human brain (e.g., when road signs are misplaced or pedestrians and drivers violate traffic rules). Organisms must efficiently allocate cognitive resources to select behaviorally relevant stimuli (i.e., targets), filter out irrelevant sensory information (i.e., distractors), and inhibit inappropriate responses. Thus, failure to control selective attention effectively can lead to errors.
Competing computational models, termed single- and dual-process models, have sought to explain how selective attention handles the processing of conflicting sensory inputs. Single-process models, drawing from Broadbent's “early selection” and “zoom lens” theories (Broadbent, 1958; Neisser, 1976; Posner et al., 1980; Johnston and Dark, 1982; C. W. Eriksen and St. James, 1986; Cave and Bichot, 1999), propose a unitary selection process with a graded attentional spotlight applied early in sensory processing (J. Cohen et al., 1992; Spieler et al., 2000; Heitz and Engle, 2007; Liu et al., 2009; Yu et al., 2009; White et al., 2011). Under this computational framework, selective attention prioritizes behaviorally relevant inputs during the earliest stages of sensory information processing, resulting in enhanced target-related and reduced distractor-related sensory information passed through later processing stages, ultimately aiding conflict mitigation. Such a process is thought to be controlled by the frontoparietal regions thought to exert top-down control over neural populations in early visual areas, significantly impacting early sensory processing (Kastner et al., 1999; Corbetta et al., 2000; Hopfinger et al., 2000; M. M. Müller et al., 2003; McMains and Somers, 2005; Shulman et al., 2010; Itthipuripat et al., 2014b; Bezdek et al., 2015; Phangwiwat et al., 2024).
Alternatively, dual-process models propose that selective attention involves at least two processing stages—early and late (Hübner et al., 2010; Hübner and Töbel, 2012). Under this view, attentional selection relies predominantly on “late selection” mechanisms, suggesting that target selection occurs only after receiving full semantic analyses of all sensory inputs—behaviorally relevant or not (cf., Moray, 1959; Deutsch and Deutsch, 1963; Shiffrin and Schneider, 1977; Duncan, 1980; Yantis and Johnston, 1990). Thus, top-down attentional modulations on target- and distractor-related sensory responses should be similar across different levels of conflicting information.
Previous computational work has provided support for both single- (J. Cohen et al., 1992; Spieler et al., 2000; Heitz and Engle, 2007; Liu et al., 2009; Yu et al., 2009; White et al., 2011) and dual-process models (Hübner et al., 2010; Hübner and Töbel, 2012). These studies typically employed variants of the Eriksen flanker task where subjects identified a target surrounded by congruent or incongruent distractors in order to measure the spatial distribution of visual attention (C. W. Eriksen and Hoffman, 1973; B. A. Eriksen and Eriksen, 1974; C. W. Eriksen and Yeh, 1985; C. W. Eriksen and St. James, 1986). A recent study used an adapted version of the Eriksen flanker task to investigate whether conflict between targets and distractors influences the spatial gradient of selective attention (Servant and Logan, 2019). In this study, subjects performed a secondary probe detection task after responding to the main flanker task. Contrary to single-process models, they found that stimulus incongruency did not alter the shape of visuospatial attention (Servant and Logan, 2019). However, the timing of the probe, introduced after the response execution of the main task, might have limited the detection of early covert attention shifts.
In alignment with the behavioral findings reported by Servant and Logan (2019), a previous EEG study by Derosiere et al. (2018) provided evidence against single-process models. They observed relatively larger distractor-related steady-state visually evoked responses (SSVEPs) in incongruent compared with congruent trials in the Eriksen flanker task, suggesting “less” fine-tuned visuospatial attention under conflict expectation (Derosiere et al., 2018). Our recent study, employing a similar SSVEP approach with the color-naming Stroop task, another well-known conflict task, found no conflict-related SSVEP modulations (Itthipuripat et al., 2019a). Using multivariate analyses on EEG data measured across auditory and multisensory discrimination tasks, Nuiten et al. (2021) found that neural activity patterns associated with the spatial location and content-specific information of a sensory stimulus could be detected even when it was completely unattended. Additionally, these neural representations remained unchanged across differing levels of cognitive interference (Nuiten et al., 2021). Together, these EEG results challenge the notion that early selective attention can mitigate behavioral interference driven by conflicting sensory inputs.
Although the aforementioned EEG studies challenged single-process models, they did not directly assess top-down processes related to changes in the spatial scope of early selective attention. Another line of research links alpha band EEG oscillations (∼8–12 Hz) to the top-down control of early visuospatial attention (Foxe et al., 1998; Fries et al., 2001; Sauseng et al., 2005; Rihs et al., 2007; Gould et al., 2011; Dombrowe and Hilgetag, 2014; C. Wang et al., 2016; Ikkai et al., 2016; Itthipuripat et al., 2019b; Gundlach et al., 2020; Keefe and Störmer, 2021; Li et al., 2021; Alamia et al., 2023; Gundlach et al., 2023). Recent studies using a multivariate image reconstruction method, known as the inverted encoding model (IEM), have also shown that topographical patterns of alpha band oscillations contain precise information regarding the locus, size, and fidelity of visuospatial attention (Foster et al., 2017, 2021; Foster and Awh, 2019; Feldmann-Wüstefeld and Awh, 2020; Feldmann-Wüstefeld, 2021; Sutterer et al., 2021).
To test if conflicting sensory information could shape neural representations of early selective visuospatial attention, we adopted the IEM-based approach to analyze EEG data obtained from human subjects performing an adapted version of the attention-cueing Eriksen flanker task. Here, we examined conflict-related modulations of spatial selectivity and the timing of alpha-based neural representations of visuospatial attention in relation to changes in amplitudes and timing of event-related parietal and frontal activity. Note that in the context of the current study, conflict is defined as a mismatch in the identity of incoming sensory inputs. In many circumstances, stimulus inputs are also linked to different response outputs. Thus, conflicting sensory information could, in turn, lead to a mismatch between sensory inputs and their response mapping. Note that while different types of conflicts could have different impacts on the low-level sensory and attentional processes as well as higher-level sensorimotor and cognitive control processes, the present study did not aim to dissociate how different types of conflicts (e.g., stimulus–stimulus and stimulus–response conflicts) might mediate these brain processes.
Overall, we found that conflicting sensory information shaped the fidelity of the alpha-based representations of visuospatial attention in a manner where attentional focus was relatively more precisely tuned to the target location. Moreover, these conflict-related changes in the alpha-based spatial reconstructions occurred very early, from ∼0 to 379 ms after stimulus onset, which was much earlier than the posterior and frontal event-related potentials (ERPs) thought to track target selection (i.e., the N2-posterior-contralateral or N2pc component from ∼197 to 455 ms), distractor suppression (i.e., the distractor positivity or Pd component from ∼102 to 246 ms), and conflict monitoring and executive control functions, respectively—i.e., the frontocentral negative-polarity incongruency wave or the FNinc component from ∼266–668 ms relative to stimulus onset and the midline-frontal theta oscillations (EEG oscillations at ∼3–7 Hz) from approximately −563 to 314 ms relative to response onset.
Together, our results suggest that conflicting sensory stimuli dynamically shape the spatial selectivity of early visual attention, emphasizing the important role of early selective attention in conflict resolution. Moreover, the cascade of conflict-related neural modulations from the parietal to frontal cortex supports a recent “diachronic” view that describes attentional selectivity as a process that unfolds over time (Zivony and Eimer, 2022). Overall, our findings are partially consistent with both single- and dual-process models and suggest that predictions these models make about “early selection” and “late selection” mechanisms are not mutually exclusive.
Materials and Methods
Subjects
We recruited 33 male and female adult participants (17 female, 3 left-handed) with ages ranging from 18 to 36 years (mean age, 25.0 years; SD = 5.6 years) from local communities in and nearby King Mongkut's University of Technology Thonburi (KMUTT), Bangkok, Thailand. All participants had normal or corrected-to-normal color vision and no history of neurological or psychiatric disorders. Prior to their participation, all participants provided written informed consent in accordance with the local Institutional Review Board (IRB) at KMUTT. The research protocol was conducted in accordance with the Declaration of Helsinki. The sample size was within the typical range used in prior studies that examined the roles of early selective attention in conflict resolution and those that used the IEM and other multivariate decoding methods to unravel the spatially specific information from alpha band activity and visual evoked potentials (VEPs) in EEG (Foster et al., 2017, 2021; Bae and Luck, 2018; Derosiere et al., 2018; Foster and Awh, 2019; Itthipuripat et al., 2019a; Servant and Logan, 2019; Feldmann-Wüstefeld and Awh, 2020; Feldmann-Wüstefeld, 2021; Nuiten et al., 2021; Sutterer et al., 2021).
Stimuli and tasks
Stimuli and tasks were presented using the Psychophysics Toolbox (Brainard and Vision, 1997; Pelli and Vision, 1997) and MATLAB 2020 (MathWorks) on a personal computer with the Windows 10 operating system. Participants were seated 80 cm from the gray background LCD monitor (300 cd/m2, 165 Hz refresh rate) in a dimly lit and quiet room. They performed an attention-cueing Eriksen flanker task while their EEG and behavioral responses were simultaneously recorded (see Fig. 1a,b for an illustration of the task and trial types).
An adapted version of the attention-cueing Eriksen flanker task and behavioral results. a, An example of a trial sequence began with an arrow cue at the fixation pointing to one of the 12 possible target locations, followed by a circular array of 12 visual stimuli rendered in red or blue. Participants had to discriminate the shape of a visual target where the arrow was pointing (either a diamond or an hourglass). They received feedback at the end of each trial, telling them if they responded correctly, incorrectly, or too slowly. b, Trial types. In each trial, a target stimulus was surrounded by either congruent or incongruent adjacent distractors (i.e., flankers) whose shape matched or mismatched the shape of the target, respectively. Fifty percent of the trials contained a salient distractor whose color was different from the rest of the visual stimuli in the circular array. The other 50% of the trials contained no salient distractors. c, Hit rates were significantly reduced in the incongruent compared with the congruent conditions as well as in trials with versus those without distractors. d, Response times in correct trials (i.e., correct RTs) were increased in the incongruent compared with the congruent conditions. There was no significant effect of distractor saliency on correct RTs. Error bars represent the within-subject standard error of the mean (SEM).
In this task, participants discriminated the shape of a visual target embedded in a circular array comprising 12 diamond- and hourglass-shaped stimuli presented equidistantly from one another at an eccentricity of 7.937° visual angle. Each visual stimulus had a 2.864° visual angle in length and 2.793° visual angle in height. At the beginning of each trial, an arrow cue (size, 2.435° × 2.793° visual angle) appeared behind a black fixation point (radius, 1.432° visual angle) at the center of the computer screen. The arrow cue pointed to one of the 12 possible target locations (0°, 30°, 60°, …, 330° with equal probabilities). The arrow cue stayed on the screen for 150 ms, followed by a blank period of 400–600 ms, and the circular stimulus array that appeared for 1,550 ms. The stimulus presentation was then followed by another blank period of 500 ms, and the response feedback was displayed for 300 ms. The response feedback was presented as a diagram picture indicating whether the response was correct, incorrect, or too slow. Intertrial intervals (ITIs) were pseudorandomly drawn from a uniform distribution of 600–1,000 ms.
The shape of the target (either a diamond or hourglass) was pseudorandomly assigned on a trial-by-trial basis. In 50% of the trials, two distractors adjacent to the cued target (i.e., flankers or surrounding distractors) had the same shape as the target (congruent trials). In the other 50%, the shape of these flankers differed from the shape of the target (incongruent trials). In 50% of these congruent and incongruent trials, all 12 visual stimuli in the circular array were rendered in the same color (either red with RGB of [255 0 0] or blue with RGB of [0 0 255]). In the other 50% of the congruent and incongruent trials, one of the 12 stimuli was a color singleton, whose color differed from the rest of the visual stimuli in the circular array (salient trials). The location of the color singleton was manipulated independently from the cued location (all 12 locations were equally likely). Overall, there were four trial types based on stimulus congruency and saliency (nonsalient congruent, nonsalient incongruent, salient congruent, and salient incongruent) × 12 cued locations (0°, 30°, 60°, …, 330° polar angle). For the nonsalient trials, each of the congruent and incongruent conditions at each of the 12 cued locations had 12 repeats. For the salient congruent and incongruent trials, the 12 cued locations were fully crossed with the 12 salient locations (each had one repeat). This resulted in 576 trials in total. The trial order was pseudorandomized to prevent subjects from predicting the cued and salient locations as well as trial types based on stimulus congruency and saliency. The experiment consisted of 16 blocks of 36 trials each, lasting ∼3 min per block. The entire experiment, including behavioral training, EEG preparation, and breaks, lasted ∼2.5 h. Participants received a monetary compensation of 500 Thai Baht for their participation.
At the beginning of the experiment, subjects were instructed to fixate at the fixation point at the center of the computer screen and covertly shift their attention to the cued location to discriminate the shape of a visual target as quickly and accurately as possible. They were instructed to minimize blinks, eye movements, head movements, and jaw movements while performing the task. Responses were made by pressing one of two buttons on a keyboard using their right index and middle fingers to indicate whether the target was a “diamond” (J button) or an “hourglass” (K button), respectively. The response deadline was adjusted on a block-by-block basis so that the overall mean accuracy across all experimental conditions was ∼75%.
Behavioral analysis
Hit rates for responses made before the deadline and mean response times in those trials (i.e., correct RTs) were computed separately for congruent and incongruent trials with and without salient distractors. Note that we did not include the salient target condition in all analyses because only 4.17% of trials contained salient targets, which were insufficient for behavioral and EEG analyses. Next, we used two-way repeated-measures ANOVAs to examine the main effects of stimulus congruency and distractor saliency, as well as their interaction, on hit rates and correct RTs.
EEG recording, preprocessing, and analysis
We recorded continuous EEG using a 64-channel ActiveTwo system with eight additional external electrodes (Biosemi Instrumentation) at a sampling rate of 512 Hz. These included a pair of electrodes on the left and right mastoids for reference, a pair of electrodes near the outer canthi of the left and right eyes for monitoring horizontal eye movements, and two pairs of electrodes above and below the left and right eyes for monitoring blinks and vertical eye movements. We online referenced the EEG signals to the CMS-DRL electrode and maintained the data offsets in all channels below 20 mV, which is a standard criterion for this active electrode system.
The EEG data were preprocessed and analyzed using EEGLab v2020.0 (Delorme and Makeig, 2004) and custom MATLAB scripts. First, we rereferenced the continuous EEG data to the data recorded from the left and right mastoid electrodes (using the algebraic mean of the two reference electrodes) and applied 0.25 Hz high-pass and 55 Hz low-pass Butterworth filters (third order). Then, we epoched the continuous EEG data from 1,850 ms before to 5,150 ms after cue onset and performed independent component analysis (ICA) to remove eyeblinks, bad channels, and muscle artifacts (Makeig et al., 1996). Finally, we rejected trials confounded by saccades, leftover muscle activity, drifts, and other artifacts via visual inspection and threshold rejections, resulting in the removal of 23.90 ± 7.89% (SD) of trials across 33 subjects.
IEM analysis
In order to examine the influence of conflict and distractor saliency on the selectivity of early visuospatial attention, we employed the IEM to reconstruct spatial representations of attended locations based on changes in the amplitudes of alpha band oscillations relative to a precue baseline. Alpha band oscillations have been shown to accurately track the precise locations of top-down visuospatial attention (Foster et al., 2017, 2021; Foster and Awh, 2019; Feldmann-Wüstefeld and Awh, 2020; Feldmann-Wüstefeld, 2021; Sutterer et al., 2021).
To begin, we conducted a wavelet analysis on the artifact-corrected EEG data, filtering the EEG data on each trial with Gaussian filters centered at 8–12 Hz with a fractional bandwidth of 0.2 Hz (Grossmann and Morlet, 1984; Canolty et al., 2007; Roach and Mathalon, 2008; Itthipuripat et al., 2013a). This process produced wavelet coefficients for each central frequency from −1,850 to 5,150 ms relative to cue onset across all 64 electrodes. Subsequently, we computed the absolute values of these wavelet coefficients to obtain the amplitudes of the alpha band oscillations. Changes in alpha amplitudes relative to the baseline activity were calculated by subtracting the averaged data from −350 to −50 ms before cue onset from the data at each time point for each trial and electrode.
These cue-locked data were then used to reconstruct alpha-based spatial representations associated with the attention cues. Additionally, to examine alpha-based spatial representations during the stimulus period and around response execution, we time locked the baseline-corrected alpha data to stimulus and response onsets. Note that we used the same precue baseline for both stimulus-locked and response-locked alpha data. The cue-locked, stimulus-locked, and response-locked alpha data were subsequently subjected to the IEM, as described in detail below. Note that these different time-locking analyses enabled us to explore the temporal dynamics of alpha-based spatial representations and those of their conflict-related modulations. This would help us distinguish whether conflict-induced modulations of alpha-based spatial reconstructions were more closely related to cue-, stimulus-, or response-related processes (A. Chen et al., 2011; Zhao et al., 2015; Itthipuripat et al., 2019a; Opitz et al., 2020; Haciahmet et al., 2021, 2023).
We predicted precisely tuned alpha-based spatial representations to emerge after cue onset and to be sustained throughout the stimulus period. However, we expected no conflict-related (or saliency-related) modulations of the alpha-based spatial representations during the cue period, as the target and distractor stimuli had not yet appeared on the computer screen. Importantly, if conflicting sensory stimuli could indeed trigger changes in early selective attention, we hypothesized that changes in alpha-based spatial representations would emerge immediately after stimulus onset. Alternatively, there might be no conflict-related modulations in alpha-based spatial representations at all, or such modulations might occur later around response onset. This would suggest that changes in early selective attention might not be the primary mechanism aiding in the mitigation of conflicting sensory processing.
To reconstruct spatial representations of top-down visuospatial attention, we employed the IEM (Brouwer and Heeger, 2009, 2011; Sprague and Serences, 2013), which learned the relationship between the topographical patterns of changes in alpha amplitudes and the cue/target locations (Foster et al., 2017, 2021; Foster and Awh, 2019; Feldmann-Wüstefeld and Awh, 2020; Feldmann-Wüstefeld, 2021; Sutterer et al., 2021). The IEM operates under the assumption that changes in alpha amplitudes measured at each electrode reflect the weighted sum of different neuronal populations tuned to 12 cue/target locations arranged in a circular fashion, where adjacent locations are equally spaced by 30° from one another (i.e., 0°, 30°, 60°, …, 330° polar angle). Accordingly, these 12 cue/target locations were captured into 12 response channels.
To describe the tuning of each target location, we constructed a basis function using a half-sinusoid with the 7th power, as shown in the following equation:
For each modeling iteration, the data were divided into independent sets of training and test data. To achieve this, we pseudorandomly sampled half of the trials for each target location, stimulus congruency, and distractor saliency condition to train the IEM, reserving the other half of the data for testing the model. It is important to note that we balanced the dataset to ensure an equal number of trials across all target locations and experimental conditions.
During training, we used the training dataset to calculate weights that estimated the relative contributions of the 12 channel response functions to the observed changes in alpha amplitudes at each electrode and each time point, employing the following general linear model:
Next, we solved for the weight matrix (W) using ordinary least-squares linear regression with the following equation:
We quantified the alpha-based spatial reconstructions using a fidelity index drawn from trigonometry (see a similar method in Rademaker et al., 2019). This fidelity index was computed from the mean of single-trial alpha-based spatial reconstructions in each experimental condition and subject. To do this, we extracted the channel response at each location and projected this vector onto the cued of location (i.e., the reconstruction center or 0°) via the following equation:
To assess whether the alpha-based reconstructions were selectively tuned to the cue/target locations, we conducted paired t tests to determine if the fidelity measures, collapsed across all experimental conditions, were significantly above zero at each time point. To investigate the impact of stimulus congruency, we collapsed the data across trials with and without salient distractors and then compared fidelity values between the congruent and incongruent conditions. Similarly, to assess the effect of distractor saliency, we collapsed the data across the congruent and incongruent conditions and compared fidelity values between trials with and without salient distractors. For all analyses, we corrected for multiple comparisons across three time windows: −500 to 500 ms relative to cue onset, 0–500 ms relative to stimulus onset, and −977 to 391 ms relative to response onset, using the cluster-based permutation method (Maris and Oostenveld, 2007). It is important to note that we collapsed the data across distractor saliency and stimulus congruency for these analyses to avoid underpowering the IEM and statistical results due to the relatively smaller number of trials per location per condition in our study compared with previous studies (Foster et al., 2017; Feldmann-Wüstefeld and Awh, 2020). Due to this limitation, we did not examine the interaction between stimulus congruency and distractor saliency.
For the alpha-based IEM, we conducted three additional controlled analyses to ensure that our main findings—i.e., conflict-related increases in the fidelity of the alpha-based spatial representations—(1) were not influenced by the saliency of the incongruent distractors, (2) did not depend on our choice of baseline correction methods, and (3) were consistent across a set of basis functions with varied tuning width.
First, it was possible that the observed conflict-related modulations of alpha-based spatial representations were driven by the interaction between stimulus congruency and distractor saliency—particularly, the incongruent salient distractors could have driven all the conflict-related effects we observed. To control for this potential confound, we excluded trials where the surrounding distractors were salient from the IEM analysis (from both the training and testing datasets) to assess if similar results could still be observed.
Second, our primary alpha-based IEM analysis used alpha amplitudes with trial-by-trial baseline correction. We employed this approach to minimize confounding effects from random noise and potential carryover effects from previous trials and unrelated cognitive processes during the ITI. However, some may raise concerns regarding single-trial baseline normalization as it might be susceptible to outliers (M. X. Cohen, 2014). Therefore, we conducted an additional analysis where we subtracted the averaged baseline across trials in individual experimental conditions (instead of trial-by-trial) to assess whether the main results were dependent on our choice of baseline correction methods.
Finally, in the main IEM analysis, the basis set of the model was a half-sinusoid with the 7th power (Eq. 1). To ensure that our findings were not contingent on the specific tuning width of the basis set, we systematically increased the power of the cosine function from 7 to 9, 11, 15, 19, and 25 to obtain narrower basis sets. We then used these altered basis sets to train the IEM (Extended Data Fig. 2-4, top panels).
In addition to the alpha-based IEM analysis, we conducted an additional IEM analysis on the visually evoked responses (VEP) in order to test if stimulus congruency and distractor saliency modulated the early sensory processing of visual stimuli. To do so, we performed similar IEM and statistical analyses as we described above, except that we instead used the artifact-corrected EEG data to train and test the model instead of the alpha band activity. Specifically, we sorted artifact-corrected EEG data into different bins based on cue/target locations, stimulus congruency, and distractor saliency. We then subtracted out the baseline activity from the EEG data: from −200 to 0 ms relative to the cue onset for the cue-, target-, and response-locked data. For each modeling iteration, we pseudorandomly sampled these EEG data into the training and test sets (1:1). These baseline-subtracted EEG data were then subjected to the IEM using the same modeling and statistical methods used for alpha-based reconstructions (1) to examine the significance of the reconstruction fidelity above zero and (2) to test for the main effects of stimulus congruency and distractor saliency on the reconstruction fidelity (using the same data collapsing method as conducted in the alpha-based IEM analysis described above). Here, multiple comparisons across all time points were corrected using the cluster-based permutation method (Maris and Oostenveld, 2007).
Lastly, we examined if alpha band activity and VEPs contained spatially specific information related to salient distractors. To do so, we used the similar IEM methods described above on the stimulus-locked and response-locked alpha and VEP data. However, the model was trained and tested using the salient distractor locations instead of the cue or target locations. To test the robustness of the reconstruction fidelity of the distractor-related representations based on the alpha and VEP data, we used paired t tests to examine if the fidelity measure collapsed across congruent and incongruent conditions and was significantly above zero for each time point. In addition, we tested if stimulus congruence affected the fidelity of distractor-related representations using another set of paired t tests to test the difference in the fidelity values between the congruent and incongruent conditions. In all tests, multiple comparisons across time points were corrected using the cluster-based permutation method (Maris and Oostenveld, 2007).
ERP analysis
In addition to IEM analyses, we examined modulations of classic ERPs, including the N2-posterior-contralateral (N2pc) component, the distractor positivity (Pd) component, and the frontocentral negative-polarity incongruency (FNinc) wave, thought to track different neural processes underlying target selection, distractor suppression, and cognitive control mechanisms, respectively.
The N2pc component
The N2pc component is the lateralized negative-going potential contralateral to the target location. It is thought to reflect neural activity in the occipital and parietal areas that underlies target selection processes, typically observed from ∼200 to 300 ms after the target onset (Luck and Hillyard, 1994; Eimer, 1996; Woodman and Luck, 1999, 2003; Hopf et al., 2000; Eimer and Mazza, 2005; Hickey et al., 2006; Robitaille and Jolicœur, 2006; Brisson and Jolicœur, 2007; Dell’Acqua et al., 2007a, 2007b; Eimer and Kiss, 2007; Kiss et al., 2007, 2008; Mazza et al., 2007; Itthipuripat et al., 2015). To obtain the N2pc component, we first sorted trials based on the target locations, i.e., left and right of the central fixation. Next, we averaged the artifact-corrected stimulus-locked EEG signals to obtain the ERPs associated with the left and right targets. Then, we computed the amplitude difference between the contralateral and ipsilateral ERP signals, resulting in the N2pc differences elicited by left and right targets, respectively. Here, we averaged the data across two sets of posterior, posterior-occipital, and occipital electrodes: O1, PO3, PO7, and P7 for the left hemisphere and O2, PO4, PO8, and P8 for the right hemisphere.
Following this step, we collapsed the N2pc data by averaging the data across trials where targets were presented on the left and right of the central fixation. It is important to note that for the N2pc analysis, we “excluded” trials where targets were presented at or adjacent to the vertical meridian because they could not elicit lateralized responses. This resulted in ∼110 trials on average after artifact correction for the N2pc data without salient distractors (6 target locations at or adjacent to the horizontal meridian × 24 repeats × ∼76% artifact-corrected data).
For trials with salient distractors, we only “included” trials where the distractors were placed at or adjacent to the vertical meridian to minimize potential confounds from lateralized responses related to the distractors. This ∼55 trials on average after artifact correction for the N2pc data with salient distractors (6 target locations at or adjacent to the horizontal meridian × 6 distractor locations at or adjacent to the vertical meridian × 2 repeats × ∼76% artifact-corrected data). We also omitted trials where targets were salient because the number of trials was not sufficient for performing ERP analyses (1/25 of the entire trials).
To test the significance of the laterality of the N2pc responses, we used paired t tests to quantify whether the amplitude differences between the contralateral and ipsilateral ERP signals, collapsed across all experimental conditions, for each time point were less than zero. Next, we investigated if stimulus congruency and distractor saliency modulated the N2pc amplitude differences. To examine the main effect of stimulus congruency, we collapsed the N2pc data across trials with and without salient distractors and conducted paired t tests to compare the amplitude differences between the congruent and incongruent conditions. Similarly, to assess the main effect of distractor saliency, we collapsed the N2pc data across the congruent and incongruent conditions and conducted paired t tests to compare the data across trials with and without salient distractors. For all tests, multiple comparisons were corrected across times using the cluster-based permutation method (Maris and Oostenveld, 2007).
The Pd component
Next, we examined the distractor positivity (Pd) component, which is a lateralized positive-going potential contralateral to the distractor location, commonly thought to track the distractor suppression processes (Hickey et al., 2009; Sawaki and Luck, 2010; Feldmann-Wüstefeld and Schubö, 2013; Sawaki and Luck, 2013; Burra and Kerzel, 2014; Gaspelin and Luck, 2018a,b, 2019; Chelazzi et al., 2019; van Moorselaar and Slagter, 2019; Feldmann-Wüstefeld et al., 2021; Luck et al., 2021). The temporal window of the Pd component varies depending on the timing of distractor suppression processes and is often found ranging from ∼150 to 400 ms after the distractor onset (Hickey et al., 2009; Feldmann-Wüstefeld and Schubö, 2013; Weaver et al., 2017; B. Wang et al., 2019; Feldmann-Wüstefeld, 2021; van Moorselaar, 2021). A recent nonhuman primate study has suggested the Pd component reflects top-down signals generated from the prefrontal cortex that project onto the upstream areas in the extrastriate visual cortex, which help suppress task-irrelevant information from salient distractors (Cosman et al., 2018).
To obtain the Pd component, we sorted trials based on the distractor locations (left and right) to the central fixation, excluding trials where distractors were at or adjacent to the vertical meridians. We then calculated the difference between the contralateral and ipsilateral ERP signals from the same sets of electrodes as the N2pc analysis. This resulted in the Pd responses elicited by left and right distractors, which were then collapsed by averaging the signals between trials where distractors were located on the left and right of the central fixation. We did these data sorting, averaging, subtracting, and collapsing steps separately for congruent and incongruent trials. Note that for the distractor analysis, we only “included” trials where targets appeared at or adjacent to the vertical meridian to reduce confounds from lateralized responses related to the targets. This resulted in ∼55 trials on average after artifact rejection for the Pd data (6 distractor locations at or adjacent to the horizontal meridian × 6 target locations at or adjacent to the vertical meridian × 2 repeats × ∼76% artifact-corrected data).
To test the significance of the laterality of the Pd responses, we used paired t tests to examine if the amplitude differences between the contralateral and ipsilateral ERP signals for each time point where we observed the Pd component were greater than zero. Next, we used another set of paired t tests to examine the effect of stimulus congruency on the Pd amplitude. For all tests, multiple comparisons were corrected across time points using the cluster-based permutation method (Maris and Oostenveld, 2007). Note that we use the same temporal window as the N2pc window so that we can directly compare the timing of the N2pc and the Pd components.
The FNinc component
Next, we examined the stimulus congruency and distractor saliency effects on the FNinc component—a neuromarker for the frontal executive control function thought to support conflict monitoring and resolution. The FNinc component is typically observed ∼300–500 ms after the stimulus onset and peaks at the midline-frontal and central electrodes (Appelbaum et al., 2009, 2012, 2014). Due to its timing, topographic patterns, and source localization, studies have suggested that the FNinc component reflects conflict-related signals generated from areas within the prefrontal cortex, including the anterior cingulate cortex (West and Alain, 1999; Liotti et al., 2000; van Veen et al., 2001; van Veen and Carter, 2002; Atkinson et al., 2003; Hanslmayr et al., 2008; Appelbaum et al., 2009; Larson et al., 2009).
To obtain the FNinc data, we averaged the ERP data across the midline-frontal, frontocentral, central, and centroparietal electrodes (i.e., Fz, FCz, Cz, and CPz, respectively). We then time locked the FNinc data to stimulus and response onsets to further explore the temporal dynamics of conflict-related modulations of the FNinc components around the time of stimulus- and response-related processes. Next, we tested the main effects of stimulus congruency and distractor saliency on the FNinc amplitudes. For the main effect of stimulus congruency, we first collapsed the FNinc data across trials with and without salient distractors and used paired t tests to compare the FNinc amplitudes across the congruent and incongruent conditions. Similarly, for the main effect of distractor saliency, we collapsed the data across the congruent and incongruent conditions and employed paired t tests to compare the data across trials with and without salient distractors. For all tests, multiple comparisons were corrected across time points using the cluster-based permutation method (Maris and Oostenveld, 2007).
Theta band oscillation analysis
Lastly, we examined conflict- and saliency-related modulations of the midline-frontal theta band oscillations (slow EEG oscillations at ∼3–7 Hz), thought to reflect the overall increase in activity in frontal executive areas and long-range communication within the fronto-basal-ganglia network supporting behavioral tasks requiring multiple attributes of cognitive control function (cf. D’Esposito et al., 1995; Carter et al., 1998; M. Botvinick et al., 1999; M. M. Botvinick et al., 2001, 2004; Curtis and Esposito, 2003; Kane and Engle, 2003; Aron et al., 2004, 2014; Ridderinkhof et al., 2004; Cavanagh et al., 2011, 2012a,b; Itthipuripat et al., 2013b; Cavanagh and Frank, 2014; Wessel and Aron, 2017).
To do this, we first conducted a wavelet analysis, filtering the artifact-corrected EEG data trial by trial using Gaussian filters centered at 3–7 Hz with a fractional bandwidth of 0.2 Hz. This yielded wavelet coefficients for each central frequency from −1,850 to 5,150 ms relative to the cue onset. Next, we obtained the amplitudes of the theta band oscillations by taking the absolute values of the wavelet coefficients. We then calculated changes in theta amplitudes relative to baseline activity by subtracting the data averaged from −350 to −50 ms before the cue onset from the data at each time point for each trial. Following this step, we realigned the data to stimulus and response onsets to explore the temporal dynamics of theta band modulations. Here, we averaged the data across a set of midline-frontal electrodes where the amplitude changes compared with precue baseline activity, averaged across experimental conditions, were maximal: AFz, F1, Fz, F2, FCz. Note that we used the averaged values of all experimental conditions to prevent bias in selecting electrodes based on their conflict- or saliency-related modulations. Next, we tested the main effects of stimulus congruency and distractor saliency on the theta amplitudes. For stimulus congruency, we collapsed the theta data across trials with and without salient distractors and used paired t tests to compare the theta amplitudes across the congruent and incongruent conditions. For distractor saliency, we collapsed the data across the congruent and incongruent conditions and used paired t tests to compare the data across trials with and without salient distractors. For all tests, multiple comparisons were corrected across time points using the cluster-based permutation method (Maris and Oostenveld, 2007).
Results
Behavioral results
As illustrated in Figure 1, hit rates for behavioral responses made before the response deadlines were significantly lower in the incongruent compared with the congruent condition, resulting in a significant main effect of stimulus congruency on hit rates (F(1,32) = 273.84; p < 0.001). In addition, salient distractors produced significantly poorer behavioral performance, leading to a significant main effect of distractor saliency on hit rates (F(1,32) = 6.81; p = 0.0137). That said, stimulus congruency and distractor saliency did not interact (F(1,32) = 0.4; p = 0.5337). We also found that subjects were significantly slower in the incongruent condition compared with the congruent condition, resulting in a significant main effect of stimulus congruency on RTs for correct responses (i.e., correct RTs; (F(1,32) = 81.59; p < 0.001). However, we found no significant main effect of distractor saliency (F(1,32) = 0.04; p = 0.8477) and no significant interaction between distractor saliency and stimulus congruency on correct RTs (F(1,32) = 0; p = 0.9905).
Overall, the behavioral results suggested that incompatible sensory information induced cognitive interference, resulting in decreased hit rates and increased RTs. In addition, salient distractors had a slight but significant effect on hit rates but did not affect the speed of performing the behavioral task. This small saliency effect on hit rates could be due to the fact that we adjusted response deadlines for each subject to equate the difficulty level across subjects. That said, we believe our manipulation of distractor saliency was effective because we observed significant Pd results (see below), thought to reflect top-down signals from the frontoparietal attentional network to suppress the presence of salience distractors (Hickey et al., 2009; Sawaki and Luck, 2010, 2013; Feldmann-Wüstefeld and Schubö, 2013; Burra and Kerzel, 2014; Gaspelin and Luck, 2018a,b; Chelazzi et al., 2019; Gaspelin and Luck, 2019; van Moorselaar and Slagter, 2019; Feldmann-Wüstefeld, 2021; Luck et al., 2021). Related to this, previous research has also shown that primates could learn to suppress distractors, which results in null effects of salient distractors on behavioral performance but increases in the Pd amplitude (Cosman et al., 2018). Thus, small behavioral effects were expected if distractor suppression mechanisms were engaged.
IEM results
Stimulus incongruency enhanced the fidelity of alpha-based spatial representations of selective visual attention
Next, we examined the spatially selective representations of visual attention based on the alpha band activity in the EEG data. To reconstruct the spatial presentations of top-down visuospatial attention, we employed the IEM, which learned the relationship between the topographical patterns of alpha band activity and the cue/target locations (see Materials and Methods; also see Foster et al., 2017, 2021; Foster and Awh, 2019; Feldmann-Wüstefeld and Awh, 2020; Feldmann-Wüstefeld, 2021; Sutterer et al., 2021). Prior to the modeling steps, we time locked the alpha band activity to the cue onset and baseline-corrected it from −350 to −50 ms. Additionally, we time locked the baseline-corrected alpha band activity to the target and response onsets. Then, we used the cue-, target-, and response-locked data to train and test the IEM model from −500 to 500 ms from cue onset, 0–500 ms from target onset, and from −977 to 391 ms from response onset, respectively. These analyses generated alpha-based spatial reconstructions aligned with cue, stimulus, and response onsets, respectively (Figs. 2, 3). By separating the IEM analysis into these distinct time periods, we aimed to pinpoint whether the impact of stimulus incongruency on the fidelity of alpha-based reconstructions occurred around stimulus or response onset.
Alpha-based reconstructions of visuospatial attention from the IEM analysis. Note that these results were obtained from the IEM model that used changes in trial-by-trial alpha amplitudes relative to precue baseline activity. a, Cue- and target-locked alpha-based reconstructions for the congruent and incongruent conditions as well as the difference in reconstruction activity between the two conditions. b, Same as a but the data were sorted into trials with and without salient distractors. c, d, Response-locked alpha-based reconstructions are sorted by stimulus congruency and distractor saliency, respectively. The bottom panels of all subfigures illustrate reconstruction fidelity of the alpha-based reconstructions. Shaded areas in the bottom panels represent the within-subject SEM of the fidelity values. Black * signs on the top of the fidelity curves represent significant increases in reconstruction fidelity relative to zero. Red * signs in a indicate significant main effects of stimulus congruency on reconstruction fidelity (all tests were corrected for multiple comparisons using cluster-based permutation). n.s., nonsignificant. See Extended Data Figures 2-1–2-4 for the results from auxiliary analyses, which ensured that the main results were not confounded by our experimental design and choices of analysis.
Figure 2-1
Same as Figures 2a-b but the amplitudes of alpha band oscillations used for the IEM analysis were obtained from the Guassian wavelet functions with shorter time-domain standard deviations (∼45-67 ms compared to ∼156-234 ms to the main analysis). This allows more appropriate comparisons between time courses of the IEM and ERP results shown in Figures 5-6. (a) Cue- and target-locked alpha-based reconstructions for the congruent and incongruent conditions as well as the difference in reconstruction activity between the two conditions. (b) Same as (a) but the data were sorted into trials with and without salient distractors. The bottom panels of all sub-figures illustrate reconstruction fidelity of the alpha-based reconstructions. Shaded areas in the bottom panels represent the within-subject SEM of the fidelity values. Black * signs on the top of the fidelity curves represent significant increases in reconstruction fidelity relative to zero. Red * signs in (a) indicate significant main effects of stimulus congruency on reconstruction fidelity (all tests were corrected for multiple comparisons using cluster-based permutation). n.s. = non-significant. Download Figure 2-1, TIF file.
Figure 2-2
(a-b) Cue- and target-locked alpha-based spatial representations, in which trials with salient flankers were removed from the IEM analysis. This auxiliary analysis was conducted to ensure that the incongruency-induced increase in the reconstruction fidelity shown in Figure 2 was not confounded by the interaction between stimulus congruency and distractor saliency at the flanker locations. (c-d) Cue- and target-locked spatial reconstructions based on changes in alpha amplitudes normalized by the pre-cue baseline averaged across trials in individual trial types. This analysis was performed to confirm that the main results shown in Figure 2 did not depend on a certain baseline correction method (also see Extended Data Figure 2-3, which suggested neglectable confounds from outliers in the baseline-corrected alpha band activity across individual subjects). The bottom panels of all sub-figures illustrate reconstruction fidelity of the alpha-based reconstructions. Black * signs on top of the fidelity curves represent significant increases in reconstruction fidelity relative to zero. Red * signs in (a) and (c) show significant main effects of stimulus congruency on reconstruction fidelity (passing the significance threshold of 0.05, corrected for multiple comparisons using the cluster-based permutation method). Shaded areas in the bottom panels represent the within-subject SEM of the fidelity values. Download Figure 2-2, TIF file.
Figure 2-3
Examining the influence of outliers in the baseline-corrected alpha band activity across individual subjects. (a) The mean (blue traces) and median values (red traces) of the baseline-corrected alpha data were closely similar to each other in all subjects, reflecting a relatively small influence of outliers in our dataset. (b) The absolute difference between the mean and the median values (magenta traces) was also much lower than the standard deviation of the trial-by-trial alpha signals (cyan traces). Together, these investigations suggest that the baseline-corrected alpha band activity and the alpha-based IEM analysis should be minimally influenced by unrelated noises or outliers. Download Figure 2-3, TIF file.
Figure 2-4
Target-related alpha-based reconstruction results using the IEM method with channel responses of varying degrees of tuning width. (a) We manipulated the width of the channel responses by varying the power of the cosine function from 7 to 25. As the power value increased, the tuning width of response channels decreased. (b) We found consistent results, where the fidelity of the alpha-based reconstructions averaged across ∼0–439 ms after stimulus onset (the significant window in the main analysis in Figure 2) was significantly higher in the incongruent than the congruent condition (p’s = 0.0067-0.0008, Bonferroni corrected). Download Figure 2-4, TIF file.
Spatial reconstructions based on visually evoked potentials (VEPs). a, b, Cue- and target-locked VEP-based reconstructions. c, d, Response-locked VEP-based reconstructions. The bottom panels represent the reconstruction fidelity of the VEP data. Black * signs on top of the fidelity curves indicate significant increases in reconstruction fidelity relative to zero. Red * signs represent significant main effects of distractor saliency on reconstruction fidelity. Multiple statistical comparisons were corrected using the cluster-based permutation method. Shaded areas represent the within-subject SEM of the fidelity values. “n.s.” indicates nonsignificant results, meaning they did not pass the corrected threshold of 0.05 based on the cluster-based permutation method.
For the cue-locked data, we observed precisely tuned alpha-based spatial representations emerging from ∼287 to 500 ms after cue onset (Fig. 2a,b). The reconstruction fidelity, collapsed across all stimulus congruency and distractor saliency conditions, was significantly above zero during this time window (the sum of t values: t_sum(32) = 382.4727; p = 0.0173, corrected using the cluster-based permutation method). As predicted, we did not observe any significant main effects of stimulus congruency or distractor saliency on the fidelity of cue-related spatial representations (no main effect of stimulus congruency: p's ≥ 0.6343; no main effect of distractor saliency: p's ≥ 0.3783; all tests did not pass the corrected threshold of 0.05 based on the cluster-based permutation method). These null results were expected since the target and distractor stimuli had not yet appeared on the screen. The absence of significant conflict- and salience-related modulations during the cue period ensures that any effect observed during the stimulus period was not merely due to random fluctuations in covert spatial attention prior to stimulus onset.
Moving to the stimulus-locked alpha-based spatial reconstructions, we observed precisely tuned spatial representations from ∼0 to 379 ms after target onset, reflecting sustained attentional focus extended from the cue period (as seen in the cue-locked data) to the stimulus period. The reconstruction fidelity, collapsed across all stimulus congruency and distractor saliency conditions, was significantly above zero during this time window (t_sum(32) = 541.7832; p = 0.0077; corrected using the cluster-based permutation method). Importantly, we observed significant increases in the fidelity of alpha-based spatial reconstructions in the incongruent compared with the congruent conditions. Specifically, the stimulus-locked reconstructions in the incongruent condition were more precisely tuned to the target locations than those in the congruent condition from ∼0 to 439 ms after target onset (t_sum(32) = 602.2067; p = 0.0062; corrected using the cluster-based permutation method). However, there was no significant main effect of distractor saliency on the fidelity of stimulus-locked alpha-based spatial reconstructions (p's ≥ 0.5947, not passing the corrected threshold of 0.05 based on the cluster-based permutation method).
In contrast to the cue- and target-locked data, we found no selective tuning of alpha-based reconstructions from the response-locked data (Fig. 2c,d; p's ≥ 0.7326; not passing the corrected threshold of 0.05 based on the cluster-based permutation method). Moreover, there were no main effects of stimulus congruency or distractor saliency on the response-locked alpha-based reconstructions (p's ≥ 0.2298; not passing the corrected threshold of 0.05 based on the cluster-based permutation method).
Since we used Gaussian wavelet filters with a fractional bandwidth of 0.2 in the main alpha-based IEM analysis, this resulted in a substantial amount of temporal smoothing when estimating the amplitudes of alpha band oscillations (time-domain standard deviations of the filters ranged from ∼156 to 234 ms with a restricted range of frequency-domain standard deviations between ∼0.68 and 1.02 Hz). While this ensured that the visuospatial reconstruction was based on alpha band oscillations with minimal influence from neighboring frequencies, the temporal resolution of this time frequency made it difficult to compare the time courses of the alpha-based IEM results with the ERP results (see the following sections). Thus, we conducted an additional analysis where we increased the fractional bandwidth of the Gaussian wavelet functions to 0.7, resulting in time-domain standard deviations ranging from ∼45 to 67 ms and frequency-domain standard deviations ranging from ∼2.37 to 3.56 Hz. Note that we did not increase the fractional bandwidth further because of potential confounds from spectral leakages with increasing frequency-domain standard deviation. As illustrated in Extended Data Figure 2-1, we found similar results, where there was a significant incongruency effect on the fidelity of alpha-based reconstructions from ∼0 to 469 ms after stimulus onset (t_sum(32) = 719.2696; p = 0.0010, corrected using the cluster-based permutation method) without significant saliency-related modulations (p's ≥ 0.4643; not passing the corrected threshold of 0.05 based on the cluster-based permutation method).
Overall, the alpha-based IEM results suggest that the incompatibility between task-relevant and task-irrelevant sensory stimuli shapes the selectivity of early visuospatial attention, enhancing spatial selectivity toward target locations and away from surrounding distractor locations. The early significant time window of the target-locked data indicates that this heightened fidelity of attentional representations occurs rapidly near target onset, supporting predictions of single-process models based on “early selection” and “zoom lens” theories (Broadbent, 1958; Neisser, 1976; Posner et al., 1980; Johnston and Dark, 1982; C. W. Eriksen and St. James, 1986; J. Cohen et al., 1992; Cave and Bichot, 1999; Spieler et al., 2000; Heitz and Engle, 2007; Liu et al., 2009; Yu et al., 2009; White et al., 2011).
Auxiliary analyses of alpha-based spatial representations of selective visual attention
To ensure that the observed effect of stimulus congruency on the fidelity of the stimulus-locked alpha-based spatial representations reported in the previous section (Fig. 2) was not influenced by confounding factors from our experimental design and choices of analysis, we conducted three auxiliary IEM analyses (Extended Data Figs. 2-2–2-4).
First, we excluded trials where one of the surrounding distractors (i.e., flankers) had a salient color to control for potential interactions between stimulus congruency and distractor saliency. Consistent with the main result, we observed a significant effect of stimulus congruency on the fidelity of the alpha-based reconstruction from ∼0 to 328 ms relative to stimulus onset (t_sum(32) = 526.5542 and p = 0.0053; corrected using the cluster-based permutation method) without a significant effect of distractor saliency (p's ≥ 0.3964; not passing the corrected threshold of 0.05 based on the cluster-based permutation method; Extended Data Figs. 2-2a, 2-2b). This result suggests that the conflict-related modulations of the alpha-based spatial representations were not driven by the saliency of the flanker stimuli.
Second, in the main results, the alpha data underwent baseline correction on a trial-by-trial basis before training and testing the IEM model (Fig. 2). We chose this method over using nonbaselined data to mitigate potential confounding effects arising from trial-by-trial fluctuations of alpha activity during the ITI or lingering neural representations associated with attentional allocations in preceding trials. This baseline correction step is particularly crucial given the experimental design, which involves the selective processing of incompatible sensory information. Additionally, subjects received response feedback in each trial, potentially heightening attentional engagement following feedback. It is important to note that the trial-by-trial baseline method should be applied with caution due to susceptibility to unrelated noises. To ensure that our baseline-corrected alpha data were minimally confounded by outliers, we compared the median and mean values of alpha amplitudes across all electrodes in all subjects (Extended Data Fig. 2-3). We found that these values were highly similar, and the absolute differences between the median and mean values were much lower than ±1 standard deviation of trial-by-trial signals in all electrodes and in all subjects. Furthermore, to further mitigate potential artifact impacts on the baseline-corrected data, we employed an alternative baseline correction method. Here, the mean values of the precue baseline alpha activity across trials in each experimental group were subtracted from the data in the corresponding group (Extended Data Fig. 2-2c,d). Consistent with the main results, we observed that the incongruent condition significantly increased the fidelity of the alpha-based reconstruction compared with the congruent condition (0–500 ms relative to target onset; t_sum(32) = 625.2192; p = 0.0066; corrected using the cluster-based permutation method). Additionally, we found no significant main effect of distractor saliency on the alpha-based reconstruction (p's ≥ 0.8081; not surpassing the corrected threshold of 0.05 based on the cluster-based permutation method).
Lastly, we investigated whether the impact of stimulus congruency on the fidelity of the alpha-based reconstruction depended on specific sets of response channels within a particular tuning width used for training the IEM. To explore this, we manipulated the bandwidth of the response channels by adjusting the power of the cosine function that defines their shape, ranging from 7 to 25 (Eq. 5; Extended Data Fig. 2-4). As the power of the cosine function increased, the response channels became narrower in tuning width. Our findings consistently showed that the incongruent condition led to increased fidelity of the alpha-based representation averaged across ∼0–439 ms after target onset (the significant window identified in the main analysis depicted in Fig. 2), across all variations of IEMs with response channels of different tuning widths (t(32)'s = 2.9011–3.7121; p's = 0.0067–0.0008; Bonferroni corrected). This suggests that the conflict-related modulations of the alpha-based spatial representations were not contingent upon the tuning width for channel responses in the IEM.
Conflict did not modulate spatial representations based on visual evoked responses (VEPs)
Next, we examined conflict-related modulations of the spatial reconstructions based on the topographic patterns of the VEPs. We performed this analysis to examine if the increased fidelity of the alpha-based spatial reconstructions reported in the earlier section in turn led to an enhancement in the spatial selectivity of early sensory-evoked responses related to the stimulus array itself. If sensory enhancement was necessary to mitigate conflict in this task, we expected to see an increase in the fidelity of the VEP-based spatial reconstruction to the target location in the incongruent compared with the congruent conditions. Alternatively, there might not be any conflict-related modulations of early sensory responses or even enhanced distractor-related sensory activity as reported by prior studies (cf. Derosiere et al., 2018; Itthipuripat et al., 2019a; Nuiten et al., 2021).
First, we found that both cue- and target-locked VEPs contained location-specific information related to the cue and the target stimuli (Fig. 3). For the cue-locked data, we observed significant increases in the fidelity of VEP-based spatial reconstructions above zero from ∼183 to 500 ms after cue onset (the data collapsed across all stimulus congruency and distractor saliency conditions: t_sum(32) = 969.1839; p < 0.001; corrected using the cluster-based permutation method). For the target-locked data, we observed significant increases in reconstruction fidelity relative to zero extended from 0 to 500 ms after stimulus onset (t_sum(32) = 1,581.1057; p < 0.001; corrected using the cluster-based permutation method). These results suggest that the VEP data contained location-specific information for both cue and target stimuli. However, unlike the alpha-based reconstruction results, we observed no significant conflict-related modulations of the spatial reconstructions (cue-locked data: p's ≥ 0.0671; stimulus-locked data: p's ≥ 0.7902; all tests did not surpass the significance threshold of 0.05 based on the cluster-based permutation method).
Although there was no main effect of distractor saliency on the cue-locked data (p's ≥ 0.4983; not passing the cluster-based corrected threshold), there was a significant reduction in the fidelity in the stimulus-locked VEP-based reconstructions in trials compared with those without salient distractors from ∼167 to 207 ms after stimulus onset (t_sum(32) = 64.8562 and p = 0.0144 corrected using the cluster-based permutation method), which overlapped with the Pd window (see the following section for the Pd data and Fig. 5c). For the response-locked data, we did not observe any significant increase in the fidelity of the VEP-based spatial reconstructions above baseline (p's ≥ 0.9383) or any significant modulation driven by stimulus congruency or by distractor saliency before the response set (p's ≥ 0.1710; all tests did not surpass the significance threshold of 0.05 based on the cluster-based permutation method).
Together, these VEP-based reconstruction results suggested that conflict-induced changes in attentional selection might not occur at the level of early processing of incoming sensory inputs. Instead, conflict increased the strength of the top-down attentional focus as reflected by the increased fidelity of alpha-based reconstructions, and this conflict-induced change in selective attention may instead facilitate selection processes that occurred later in the postperceptual decision-related processes. On the other hand, salient distractors could interfere with target-related representations based on the VEP data, resulting in poorer performance in trials with salient distractors (Fig. 1c).
No significant spatially tuned representations of salient distractors based on alpha band activity and VEPs
In addition, we employed the IEM method to reconstruct spatial representations of salient distractors based on the topographic patterns of alpha band activity and VEPs. We found no significant changes in the fidelity of distractor-related spatial representations relative to zero from either alpha band oscillations or VEPs (Fig. 4; alpha-based reconstructions: p's ≥ 0.7116; VEP-based reconstructions: p's ≥ 0.9333; all tests did not surpass the significance threshold of 0.05 based on the cluster-based permutation method). Furthermore, stimulus congruency did not affect the distractor-related representations based on alpha band oscillations or VEPs (alpha-based reconstructions: p's ≥ 0.5308; VEP-based reconstructions: p's ≥ 0.3813; all tests did not pass the corrected threshold based on the cluster-based permutation method).
Distractor-related representations based on alpha band activity and VEPs. a, Alpha-based distractor-related representations in the congruent and incongruent conditions. b, VEP-based distractor-related representations plotted in similar configurations as a. Bottom panels of all subfigures show reconstruction fidelity. Shaded areas represent within-subject SEM of the fidelity values. n.s., nonsignificant (failing to pass the significance threshold of 0.05 based on the cluster-based permutation method). Overall, we did not observe significant spatial tuning for the distractor-related reconstructions based on either alpha or VEP data.
ERP results
In addition to the multivariate image reconstruction analyses (i.e., the IEM method), we examined other classic ERP markers, including the lateralized posterior negative potential (i.e., the N2pc component), the lateralized posterior positive potential (i.e., the Pd component), and the midline-frontal negative potential (i.e., the FNinc component), known to track target-related attentional selection, distractor suppression, and conflict monitoring processes, respectively (see details below).
The target-related lateralized negative potential: the N2pc component
To obtain the N2pc component, we computed the differences between the ERPs measured on the posterior parietal occipital electrodes contralateral and those ipsilateral to the target locations (Luck and Hillyard, 1994; Eimer, 1996; Woodman and Luck, 1999, 2003; Hopf et al., 2000; Kiss et al., 2008). Note that for the N2pc analysis, the targets were placed either on the left or the right of the fixation and not in the middle of the screen. For trials with salient distractors, we only selected a subset of trials where the salient distractors appeared at the vertical meridian to minimize confounds from the Pd component associated with the distractors. We found significant lateralization of the N2pc component from ∼197 to 455 ms after target onset (t_sum(32) = −518.9191; p = 0.0169; corrected using the cluster-based permutation method; Fig. 5a,b). That said, there were no significant main effects of stimulus congruence and distractor saliency on the N2pc amplitudes (p's ≥ 0.3310; all tests did not surpass the significance threshold of 0.05, corrected based on the cluster-based permutation).
The lateralized posterior-occipital ERPs. a, b, The N2-posterior-contralateral (N2pc) component between the congruent and incongruent conditions and between trials with and without salient distractors, respectively. The N2pc responses were sorted relative to the targets that were presented either on the left or the right side of the screen. For trials with salient distractors, we only included trials where the salient distractors were presented at and adjacent to the vertical meridian. Black bars in the bottom of the ERP traces represent significant laterality of the N2pc amplitude difference between the contralateral minus ipsilateral posterior occipital electrodes (corrected based on the cluster-based permutation method). Shaded areas represent within-subject SEM of the N2pc amplitudes. c, The distractor positivity (Pd) component across the congruent and incongruent trials where salient distractors were presented on the left or right side of the screen and targets were presented at the vertical meridian. Note that we observed robust N2pc and Pd components from ∼197–455 and ∼102–246 ms after cue onset, respectively, all of which emerged later than the incongruency-related modulations of the alpha-based reconstructions of selective attention (Fig. 2 and Extended Data Fig. 2-1 for the IEM results with higher temporal precision). Black bars on top of the ERP traces represent significant lateralization of the Pd amplitudes between the contralateral minus ipsilateral posterior occipital electrodes (corrected based on the cluster-based permutation method). Shaded areas represent within-subject SEM of the ERP amplitudes. Bottom panels illustrate topographic maps showing the N2Pc and Pd amplitude difference averaged across significant windows of ∼197–455 and ∼102–246 ms, respectively.
The distractor-related lateralized positive potential: the Pd component
To obtain the Pd component, we computed the lateralized differences between the ERPs contralateral and those ipsilateral to the salient distractors. Note that we chose only trials where the salient distractors appeared on the left and the right of the computer screen that were paired with nonsalient targets placed at the vertical meridian. This trial selection was performed to reduce confounds from the target-related N2pc component, which had the opposite polarity to the Pd component. We found significant laterality of the Pd component from ∼102 to 246 ms post distractor onset (t_sum(32) = 239.7350 and p = 0.0136; corrected using the cluster-based permutation method; Fig. 5c). However, we found no significant difference in the Pd amplitudes between the incongruent compared with the congruent trials (p's ≥ 0.5746; again not passing the corrected threshold based on the cluster-based permutation). Overall, these Pd results suggest that suppressive attentional control mechanisms were engaged by the presence of salience distractors irrespective of stimulus congruency.
Note that because we had a limited number of trials (i.e., ∼55 trials on average with salient distractors on average after artifact correction) for the N2pc and Pd analyses, the null effects of congruency on the N2pc and Pd amplitudes should be considered with caution.
The frontocentral negative-polarity incongruency wave: the FNinc component
Next, we examined the effects of congruency and distractor saliency on the amplitudes of the FNinc component (Fig. 6). Consistent with past studies, we found a significant main effect of congruency on the FNinc component from ∼266 to 668 ms after stimulus onset and from ∼−21 to 275 ms relative to response onset (Appelbaum et al., 2009, 2012, 2014). Particularly, the amplitudes of the FNinc component were higher (more negative) in the incongruent compared with the congruent conditions during these time windows (poststimulus: t_sum(32) = 1,365.9674, p = 0.0011; around response onset: t_sum(32) = 965.3963, p < 0.00; corrected using the cluster-based permutation method). In contrast to the congruency effect, there was no significant main effect of distractor saliency on the FNinc amplitudes (poststimulus: p's ≥ 0.4558; around response onset: p's ≥ 0.6961; not surpassing the corrected threshold of 0.05 based on the cluster-based permutation test). Overall, the significant effects of stimulus congruency on the FNinc amplitudes frontal executive function play a crucial role in monitoring and resolving conflicts arising from the incompatibility between sensory inputs.
The frontocentral negative-polarity incongruency wave (i.e., the FNinc component). a, b, Stimulus-locked ERP traces showing the FNinc components between the congruent and incongruent conditions and between trials with and without salient distractors, respectively. c, d, Response-locked data with the same sets of experimental comparisons. Black * signs on top of the ERP traces represent significant main effects of stimulus congruency of the FNinc amplitudes (corrected based on the cluster-based permutation method). Shaded areas represent the within-subject SEM of the FNinc amplitudes. Bottom panels of all subfigures show topographic maps of the FNinc amplitudes and the differences between labeled experimental conditions, averaged from ∼266 to 668 ms after stimulus onset and from ∼−21 to 275 ms relative to response onset for the stimulus-locked and response-locked data respectively.
Theta band activity
Lastly, we examined midline-frontal theta band oscillations (i.e., 3–7 Hz) as another neural index for frontal executive control function. This theta band activity is commonly used to monitor the recruitment of frontal brain areas during tasks requiring cognitive control processes (cf. D’Esposito et al., 1995; Carter et al., 1998; M. Botvinick et al., 1999; M. M. Botvinick et al., 2001, 2004; Curtis and Esposito, 2003; Kane and Engle, 2003; Aron et al., 2004, 2014; Ridderinkhof et al., 2004; Cavanagh et al., 2011, 2012a,b; Itthipuripat et al., 2013b; Cavanagh and Frank, 2014; Wessel and Aron, 2017). Additionally, theta band oscillations are believed to reflect long-range communication within the fronto-basal-ganglia loop, which serves domain-general cognitive control functions such as working memory, response inhibition, expectation, executive function, and conflict monitoring (Cavanagh et al., 2011, 2012a,b; Itthipuripat et al., 2013b; Rungratsameetaweemana et al., 2018). As depicted in Figure 7, we observed a marginal effect of stimulus congruency on stimulus-locked theta band amplitudes from ∼29 to 896 ms (p's ≥ 0.0953; failing to reach the corrected threshold of 0.05 based on the cluster-based permutation method). However, in the response-locked data, theta band amplitudes increased significantly for incongruent compared with congruent conditions from −563 to 314 ms relative to response onset (t_sum(32) = 1,148.6032; p = 0.0247; corrected using the cluster-based permutation method). Furthermore, there was no significant main effect of distractor saliency on theta band amplitudes for either stimulus- or response-locked data (p's = 1; all tests failed to reach the corrected threshold of 0.05 based on the cluster-based permutation method). In summary, these results suggest that the frontal-basal-ganglia network also plays an essential role in mitigating conflicts near the time of response execution.
The midline-frontal theta band activity. a, b, Stimulus-locked theta amplitudes between the congruent and incongruent conditions and between trials with and without salient distractors, respectively. c, d, Response-locked data was plotted with the same sets of experimental comparisons. Black * signs on top of the traces of theta band activity represent significant main effects of stimulus congruency (corrected based on the cluster-based permutation method). Shaded areas represent within-subject SEM of the theta amplitudes. Bottom panels of all subfigures show topographic maps of the theta amplitudes and the differences between labeled experimental conditions, averaged from ∼29 to 896 ms after stimulus onset and from approximately −563 to 314 ms relative to response onset for the stimulus-locked and response-locked data, where we found marginal and significant conflict-related modulations, respectively.
Discussion
The present study evaluated the role of early selective visuospatial attention in supporting conflict resolution. To do so, we employed the IEM to reconstruct the spatially selective representations of visual attention using the topographic patterns of alpha band activity measured from human subjects performing the attention-cueing Ericksen flanker task. We found a significant increase in the fidelity of the alpha-based spatial representations in the incongruent relative to the congruent conditions. Particularly, the conflict-related increase in spatial selectivity was driven by alpha-based spatial representations becoming relatively more fine-tuned to the target location in the incongruent conditions compared with the congruent conditions. Importantly, this conflict-related modulation occurred rapidly right after stimulus onset (∼0–379 ms poststimulus onset). Such an early time course of this conflict-related modulation supports key predictions of single-process models, which emphasize the role of “early selection” and “zoom lens” mechanisms in mitigating conflicting sensory information (J. Cohen et al., 1992; Heitz and Engle, 2007; Liu et al., 2009; Spieler et al., 2000; Yu et al., 2009; White et al., 2011; also see Broadbent, 1958; Neisser, 1976; Posner et al., 1980; Johnston and Dark, 1982; C. W. Eriksen and St. James, 1986; Cave and Bichot, 1999).
Our main results reported here could also be interpreted in light of Lavie's load theory, which posits that behavioral tasks requiring significant effort utilize a greater portion of limited resources, thereby diminishing interference from distractors (Lavie and Tsal, 1994; Lavie, 1995). According to this framework, the task load in the Ericksen flanker task in the current study, especially in the incongruent condition, might be relatively high because we observed a small (but significant) behavioral effect of salient distractors (Fig. 1c). However, this interpretation should be considered with caution because we did not systematically manipulate load in the current design.
Past studies that measured conflict-related modulations of early sensory responses in a wide range of behavioral tasks requiring conflict resolution have reported null or negative findings against the predictions of single-process models (Derosiere et al., 2018; Itthipuripat et al., 2019a; Nuiten et al., 2021). Consistent with these previous references, we actually did not observe conflict-related modulations of the spatial reconstructions based on the topographical pattern of the VEP activity in our study. Collectively, these findings suggest that changes in the selectivity of visual attention induced by incompatible sensory stimuli might not occur at this early stage of sensory information processing.
Although null reconstruction results based on the VEPs were observed in the present study, we found a robust conflict-induced increase in the fidelity of the spatial reconstruction based on the topographic patterns of alpha band oscillations. Previous studies have shown that cue-related reduction in alpha band activity in the contralateral posterior occipital electrodes is related to the allocation of visuospatial attention even when visual stimuli are not present (Sauseng et al., 2005; Kelly et al., 2009; Foxe and Snyder, 2011; Rohenkohl and Nobre, 2011; Bosman et al., 2012; Foster et al., 2016, 2017; Green et al., 2017; Foster and Awh, 2019; Itthipuripat et al., 2019a, 2023). Studies using multivariate model-based approaches have also demonstrated that alpha band oscillations in scalp-EEG contain spatially specific information related to the locus of visuospatial attention (Foster et al., 2017, 2021; Bae and Luck, 2018; Foster and Awh, 2019; Feldmann-Wüstefeld, 2021; Sutterer et al., 2021). Moreover, EEG studies using the IEM approach have shown that the spatial scope of visuospatial attention indexed by the alpha-based reconstruction is modulated by the spatial certainty of attentional cues (Voytek et al., 2017; Feldmann-Wüstefeld and Awh, 2020). Specifically, when attention is directed toward a narrower region of space, the spatial scope of attention is narrowed down in a manner similar to a “zoom-lens” (Voytek et al., 2017; Feldmann-Wüstefeld and Awh, 2020). Our present study adds that target and distractor incongruence is another key factor that shapes the spatial scope of top-down visual attention and suggests that top-down visuospatial attention plays an essential role in mitigating cognitive interference.
Interestingly, fMRI studies have shown that hemodynamic responses measured in early visual areas exhibit similar patterns of attentional modulations as those of alpha band oscillations in the EEG. First, allocating visuospatial attention has been shown to modulate fMRI activity in the retinotopically organized regions of the visual cortex in a spatially specific manner, and this is true in the absence of visual stimuli, suggesting that this type of attentional modulation reflects top-down signals from the frontoparietal regions to upstream areas in the visual cortex (Kastner et al., 1999; Buracas and Boynton, 2007; Murray, 2008; Pestilli et al., 2011; Hara et al., 2014; Sprague et al., 2018; Itthipuripat et al., 2019a,b; Foster et al., 2021). Past studies have also found that cue uncertainty modulated the magnitude and the spatial spread of fMRI activity measured in early visual areas (Herrmann et al., 2010; Pestilli et al., 2011; Itthipuripat et al., 2014b). Moreover, studies that measured attention effects on neural activity as a function of stimulus intensity have found that the neural contrast response functions (CRFs) based on alpha amplitude reduction and fMRI activity in visual cortex undergo similar additive shifts where significant attentional modulations were observed at the baseline-offset of the CRFs (Buracas and Boynton, 2007; Murray, 2008; Pestilli et al., 2011; Hara et al., 2014; Sprague et al., 2018; Itthipuripat et al., 2019a, 2023; Foster et al., 2021). These results stand in contrast to the modulatory patterns of SSVEP and VEP data, which exhibit multiplicative response gain and contrast gain patterns, suggesting that visuospatial attention and stimulus inputs interact to modulate the gain of these sensory evoked responses (Mangun and Hillyard, 1991; Morgan et al., 1996; Hillyard and Anllo-Vento, 1998; M. M. Müller et al., 1998; di Russo et al., 2001; Han and Kim, 2007; J. Wang and Wade, 2011; Itthipuripat et al., 2014a,b, 2017, 2018, 2019b, 2023; Foster et al., 2021; Sawetsuttipan et al., 2023).
At first glance, the null VEP results and the positive conflict effects on the alpha-based reconstructions might seem contradictory because increased top-down attention should lead to increased attentional gain modulations of early sensory responses. That said, this is not always the case, as some previous studies have reported that attending to task-relevant stimuli does not necessarily induce attentional gain modulations of early sensory responses in all contexts. Particularly, the effects of attention on sensory gain modulations have been shown to depend on many factors, including the physical properties of visual stimuli (e.g., size and contrast), strategies to deploy attention, perceptual difficulty, spatial uncertainty, and learning duration (Motter, 1993; Handy and Mangun, 2000; Handy et al., 2001; Boudreau et al., 2006; Y. Chen et al., 2008; Prinzmetal et al., 2009; Reynolds and Heeger, 2009; Sundberg et al., 2009; Herrmann et al., 2010; Andersen et al., 2012; Itthipuripat et al., 2014a,b, 2017, 2023; Khayat and Martinez-Trujillo, 2015; Zhang et al., 2016; Sawetsuttipan et al., 2023). In addition, selective attention could enhance the efficiency of perceptual decision-making processes by influencing postperceptual decision or response selection processes (Palmer et al., 1993; Palmer, 1995; Eckstein et al., 2002; Pestilli et al., 2011; Hara et al., 2014). Therefore, even though there might not be changes in early sensory processing, changes in top-down visuospatial attention could lead to more efficient postperceptual decision or response selection processes (cf. Palmer et al., 1993; Palmer, 1995; Eckstein et al., 2002; Pestilli et al., 2011; Hara et al., 2014).
While there was no significant conflict-related modulation of VEP-based target-related representations, we found a significant reduction in the fidelity of the VEP-based target-related representations induced by the presence of salient distractors. This reduced fidelity of target-related representations presumably led to poorer behavioral performance in trials with salient distractors compared with those without salient distractors. Interestingly, this reduced fidelity of target-related representations occurred at the similar time window as that of the Pd component (∼167–207 and ∼102–246 ms after stimulus onset, respectively). This suggests that such interference in target-related processing may trigger distractor suppression mechanisms in order to minimize the impact of salient distractor interference on discrimination performance.
Unlike the target-related reconstruction results, we observed no significant increase in the fidelity of distractor-related representations above zero for either the alpha and the VEP data. This null result stands in contrast to a recent finding where distractor locations could be predicted by the pattern of slow-wave EEG oscillations using the IEM approach (Feldmann-Wüstefeld, 2021). We assumed that this could be due to the difference in behavioral tasks. Particularly, Feldmann-Wüstefeld and others used a visual search task where subjects had to actively search for a target in the presence or absence of a salient distractor (Feldmann-Wüstefeld, 2021). Unlike our study, the cue and the distractor appeared in the stimulus array at the same time, thus subjects had to actively search for the target while trying to suppress the salient distractor (Feldmann-Wüstefeld, 2021). As a result, salient distractors had robust impacts on the speed of discrimination performance (Feldmann-Wüstefeld, 2021). In contrast, the cue in our task appeared 500–800 ms before the stimulus array containing a target and distractors. Thus, the spatial extent of attention in our task might be relatively narrower than that in the visual search task, and subjects could prepare to allocate their attention to the target location even before the stimulus array appeared. Accordingly, this highly focused endogenous visuospatial attention may pit against the exogenous effects of salient distractors, resulting in relatively less reliable distractor-related representations and no significant effects of salient distractors on RTs. Also, the fact that we observed no significant distractor-related spatial tuning could be due to distractor suppression in our study operating in a more proactive manner than that in the visual search task because subjects could prepare to engage distractor suppression mechanisms right after they saw the cue. This results in a much earlier time window for the Pd component in our study compared with that reported in the previous visual search study (∼102–246 vs ∼255–354 ms after stimulus onset, respectively; compare the present study and Feldmann-Wüstefeld, 2021).
Note that we did not find a significant effect of stimulus congruency on the Pd amplitude, whereas there was a robust stimulus congruency effect without any distractor saliency effect on the FNinc component and the midline-frontal theta band activity, which emerged later in time (∼102–246 ms poststimulus onset for the Pd component vs ∼266–668 ms poststimulus onset for the FNinc component and ∼−563 to 314 ms relative to response onset for the theta band activity). These Pd and FNinc results suggest that conflict-related processes induced by stimulus congruency operate independently of distractor suppression mechanisms. In addition, even though we observed the robust N2pc components around the expected time window (∼197–455 ms after stimulus onset), we did not find any difference in the N2pc amplitudes between the incongruent and congruent conditions or between trials with and without salient distractors. This could be due to a lack of fine-grained spatial resolutions in the univariate ERP analysis of the N2pc data for capturing conflict-related and silency-induced changes in the spatial distributions of target selection processes. The null conflict-related effect on the N2pc data could also be due to the enhanced fidelity of top-down visuospatial attention as indexed by the alpha-based reconstructions in the incongruent compared with the congruent conditions that worked against target selection processes that happened later in time.
Notably, the conflict-related modulations of the alpha-based reconstructions in the current study occurred very early, i.e., ∼0–379 ms after stimulus onset and there was no significant modulation around response onset. This explains why the previous behavioral study, which measured the spatial distributions of attention in the Eriksen flanker task using probe stimuli that appeared after motor responses, did not find any conflict-related modulation in their probe responses (i.e., their probes appeared too late; Servant and Logan, 2019). Importantly, the conflict-related modulation of the alpha-based reconstructions in the present study emerged before the temporal windows where we observed the N2pc, Pd, and FNinc components, thought to index target selection, distractor suppression, and executive control functions, respectively.
In the main IEM analysis (Fig. 2), we used Gaussian wavelet functions with small frequency-domain standard deviations (∼0.68–1.02 Hz) to restrict the time-frequency analysis to the alpha range. However, this resulted in relatively large time-domain standard deviations (∼156–234 ms), which overlapped with the significant windows of the ERP results. Thus, we conducted an auxiliary analysis employing Gaussian wavelet functions with relatively larger frequency-domain standard deviations (∼0.68–1.02 Hz) to reduce the time-domain standard deviation to ∼45–67 ms (Extended Data Fig. 2-1). With this analysis, we found consistent results, where a significant incongruency effect on the fidelity of alpha-based reconstructions was observed from ∼0 to 469 ms after stimulus onset. Note that for both the main and auxiliary IEM analyses, we observed conflict-related modulations at 0 ms after target onset. We think this is likely due to the temporal smoothing of the alpha amplitude estimation based on the time-domain standard deviations of Gaussian wavelet functions. Accordingly, the observed difference at 0 ms could be influenced by neural modulations occurring at least ∼67–156 ms after stimulus onset. That said, the new analysis, where we reduced the time-domain standard deviations of the Gaussian wavelet filters to ∼45–67 ms, ensured that the conflict-related modulations of the alpha-based visuospatial representations indeed occurred earlier than the amplitude modulations of the N2pc, Pd, and FNinc components (∼197–455, ∼102–246, and ∼266–668 ms, respectively). In conclusion, our results support the prediction of single-process models, which emphasizes the role of top-down visuospatial attention in mitigating cognitive interference.
While the conflict-related modulations of alpha-based representations of visuospatial attention provided strong support for single-process models, these results did not necessarily refute all key predictions made by dual-process models. It is true that in the most extreme cases, dual-process models undermine the role of “early selection” mechanisms and emphasize the importance of “late selection” mechanisms in conflict mitigation (Hübner et al., 2010; Hübner and Töbel, 2012). In our present study, even though early selective attention could help mitigate conflicting sensory information, we still observed conflict-related modulations of the FNinc component and midline-frontal theta band activity, which occurred in later time courses and are thought to support cognitive control functions of the frontal executive network (Appelbaum et al., 2009, 2012, 2014; Cavanagh et al., 2011, 2012a,b; Itthipuripat et al., 2013b, 2019a; Cavanagh and Frank, 2014; Wessel and Aron, 2017; Rungratsameetaweemana et al., 2018). These results suggest that both “early selection” and “late selection” mechanisms are needed for conflict mitigation, at least in the context of our present study.
Importantly, the distinct temporal dynamics of our alpha-based reconstruction, ERP, and midline-frontal theta results suggest that conflict resolution is supported by a cascade of multiple attentional and cognitive processes that unfold over time. Furthermore, the significant window of the conflict-related modulations of the alpha-based visuospatial representations extended to ∼379–469 ms, suggesting that the conflict-related modulation of the alpha-based representations was not only a rapid but also sustained process that occurred in parallel with target selection, distractor suppression, and cognitive control processes. This observation aligns with a recently proposed “diachronic” framework, which places less emphasis on directly comparing “early selection” and “late selection” mechanisms. Instead, it underscores the concept that attentional selection involves various sensory, cognitive, and decision-making processes, each with distinct temporal characteristics, working together to efficiently prioritize sensory information (Zivony and Eimer, 2021). According to the “diachronic” account, the selectivity of early visuospatial attention and modes of attentional selection mechanisms (early vs late) may vary based on task demands, cognitive inference, spatial uncertainty, and strategies (LaBerge, 1983; C. W. Eriksen and Yeh, 1985; C. W. Eriksen and St. James, 1986; Castiello and Umiltà, 1990; Heinze et al., 1994; McCormick and Jolicoeur, 1994; Barriopedro and Botella, 1998; de Fockert et al., 2001; N. G. Müller et al., 2003; Lavie et al., 2004; Lehle and Hübner, 2008; Y. Chen et al., 2008; Pestilli et al., 2011; Itthipuripat et al., 2014a,b, 2017, 2018; Feldmann-Wüstefeld and Awh, 2020; Feldmann-Wüstefeld, 2021; Sawetsuttipan et al., 2023). Future studies could adopt our experimental design and EEG-based modeling approach to explore how these different factors mediate the contributions of these attentional selection mechanisms at mitigating conflicting sensory information.
Lastly, given our experimental design where there was only a 1-on-1 mapping between stimulus inputs and response outputs, it is hard to disentangle the effects of stimulus–stimulus and stimulus–response conflicts on the observed patterns of behavioral and neural data. Our stimulus-locked and response-locked EEG analyses revealed a significant result for the conflict-related modulation in alpha-based representations only in the stimulus-locked analysis. However, the FNinc component and theta band oscillations results were also significant near response onset. These results suggest that these different neural indexes track different attentional selection mechanisms that are more or less related to early and later stages of sensory information processing. Future studies could adopt the 2-on-1 stimulus–response mapping in order to dissociate the influences of stimulus–stimulus and stimulus–response conflicts on early and late attentional selection mechanisms using similar experimental and analytic approaches (i.e., by comparing between trials with incompatible stimulus and response, those with incompatible stimulus only, and those with compatible stimulus and response; cf. B. A. Eriksen and Eriksen, 1974).
In summary, our current research offers neural evidence showing the significant contribution of top-down early visuospatial attention in mitigating cognitive interference. Furthermore, the diverse temporal patterns observed in our neural data indicate that conflict resolution relies on multiple attentional and cognitive processes. These include the initial operation of early top-down selective attention in alpha band frequencies, along with subsequent processes such as target selection, distractor suppression, and the engagement of frontal executive control functions, all unfolding over time.
Footnotes
This work was supported by the National Research Council of Thailand (NRCT; fiscal years 2021–2024) to S.I., C.C. and I.C., Thailand Science Research and Innovation (TSRI; fiscal years 2021–2024: FRB670016/64, FRB660073/0164, FRB650048/0164, and FRB640008) to S.I. and K.L., the Program Management Unit for Human Resources and Institutional Development, Research and Innovation (fiscal years 2023–2024: B46G670083 and B44G660093) to S.I., C.C. and I.C., Asahi Glass Foundation (AGF) to S.I., Research & Innovation for Sustainability Center, Magnolia Quality Development Corporation Limited to S.I., C.C. and K.L., KMUTT Partnering initiative, King Mongkut’s University of Technology Thonburi (KMUTT) to S.I., National Science and Technology Development Agency to S.I., and KMUTT’s Frontier Research Unit Grant for Neuroscience Center for Research and Innovation to S.I. and K.L.
The authors declare no competing financial interests.
- Correspondence should be addressed to Sirawaj Itthipuripat at itthipuripat.sirawaj{at}gmail.com or Chaipat Chunharas at chaipat.c{at}chula.ac.th.
