Event‐related potential patterns of selective attention modulated by perceptual load

Abstract Introduction A high perceptual load can effectively prevent attention from being drawn to irrelevant stimuli; however, the neural pattern underlying this process remains unclear. Methods This study adopted a perceptual load paradigm to examine the temporal processes of attentional modulation by incorporating conditions of perceptual load, distractor‐target compatibility, and eccentricity. Results The behavioral results showed that a high perceptual load significantly reduced attentional distraction caused by peripheral distractors. The event‐related potential results further revealed that shorter P2 latencies were observed for peripheral distractors than for central distractors under a high perceptual load and that a suppressed compatibility effect with increasing load was reflected by the P3 component. Conclusion These findings suggested that (1) P2 and P3 components effectively captured different sides of attentional processing modulated by load (i.e., the filter processing of the object and the overall attentional resource allocation) and (2) response patterns of selective attention modulated by perceptual load were influenced by eccentricity. Our electrophysiological evidence confirmed the behavioral findings, indicating the neural mechanisms of attentional modulation.

Specifically, under a low perceptual load, the resources remaining after target task-related processing overflow automatically, and then distractors are processed. Under a high perceptual load, the processing of the target stimulus exhausts all attentional resources, and available resources are lacking for processing distractors. Recently, the level of the perceptual load has become an important factor to consider in selective attentional processing (Geden et al., 2018;Neokleous et al., 2016;Tyndall et al., 2018).
Functional magnetic resonance imaging (fMRI) studies showed a load-dependent selective attention mechanism that was enhanced by target-related brain activity as the perceptual load increased accompanied by a decrease in distractor-related activity, which was under the control of the frontoparietal network associated with spatially directed attention (Pinsk et al., 2004;Rees et al., 1997). Electroencephalography (EEG) results showed that early event-related potential (ERP) components could reflect selective attentional responses involving a perceptual load, suggesting that the perceptual load had already influenced information flow during the initial stages of visual cortical processing (Fu et al., 2010;Handy & Mangun, 2000;Handy et al., 2001).
However, a recent systematic evaluation suggested that ERP effects of load-attenuating distractor processing were more reliable at a slightly later stage than the mixed results in the early attention selection process (Brockhoff et al., 2022). For example, deeper features increased the anterior P2 amplitudes, which might reflect the process of visual feature detection and attentional selection (Cao et al., 2020;Correll et al., 2006;Nikolaev et al., 2008). It was further suggested that the P2 component was sensitive to attention processing and load, with larger amplitudes observed in the processing of emotional distractors under low loads (Doallo et al., 2006). In parallel, the parietal P3 component was used as a valid measure of voluntary attention resource allocation and indicated effective responses to the demand for attentional resources under a perceptual load (Harris et al., 2019). In light of previous efforts, focusing on the activity of the later stages (i.e., P2, P3) appeared to be a promising way to better understand the neural mechanisms in attentional processing.
In addition to the objective recording of experiments, selfobservation contributes to a more comprehensive understanding of attention. A scale that appears to do this is the Mindful Attention Awareness Scale (MAAS; Brown & Ryan, 2003), which measures the perceived degree of attentiveness in different contexts or is a stable psychological trait when used outside of attentional training situations (Brown et al., 2015(Brown et al., , 2011. The MAAS is considered a prospective predictor of not only self-reported attention-related errors but also of a behavioral measure of the ability to sustain conscious awareness of attention (Cheyne et al., 2006). Studies exploring the relationship between the MAAS and attentional control showed that individuals diagnosed with attention-deficit hyperactivity disorder who were deficient in attentional control skills scored lower on the MAAS than controls. Of note, the degree of attention lapses was negatively correlated with MAAS scores, implying that higher MAAS scores may be associated with better attentional control (Keith et al., 2017). Thus, we included the MAAS as a method of evaluating concentration ability from a first-person perspective, with high total scores reflecting a strong ability to focus attention.
This study aimed to assess neural responses at a later stage during attentional processes modulated by different perceptual loads.
In addition, we included eccentricity as a potential factor and analyzed the neural response patterns with which it was involved (Beck & Lavie, 2005). We hypothesized that attention would be modulated by perceptual load and that the P2 and P3 components would have specific representations in attentional processing. Specifically, attention to nontargets may behaviorally decrease (measured by compatibility effects) as load increases (Lavie, 2005(Lavie, , 2010. In terms of neural responses, the latency or amplitude of the P2 component may vary over the load and be potentially influenced by eccentricity, and the P3 component may tend to respond more to the overall allocation of attentional resources.

Participants
Thirty-one right-handed males from the University of Electronic Science and Technology of China were recruited (mean age ± SD = 22.29 ± 1.81 years). The sample size was derived by G*Power (1 group, 4 measurements, α = 0.05, and a 0.5 correlation among repeated measures) using a medium effect size (0.25) (Kang, 2021), suggesting that 24 subjects would achieve an acceptable statistical power (0.8) (Faul et al., 2009

Apparatus and stimuli
Based on previous studies, our experimental design was a combination of the perceptual load paradigm and EEG (Green & Bavelier, 2003Proksch & Bavelier, 2002 As shown in Figure 1a, one trial contained three types of items (the target, a filler, and a distractor) in white against a black background.
Each item consisted of a separate set of graphs: the target set included F I G U R E 1 (a) Perceptual load stimuli. Each graphical trial consisted of three items: the target, a distractor, and a filler. The perceptual load was determined by the number of fillers, eccentricity was determined by the distractor location, and compatibility was determined by the relationship between the target and the distractor. (b) The experimental procedure.
a square and a diamond; the distractor set was composed of a square, a diamond, and a slender ellipse; and the filler set consisted of a pentagon, an upside-down pentagon, a lateral trapezoid, and a triangle that was pointed either up or down. Only one target and one distractor were displayed each time, and the number of fillers was randomly set to 0, 1, 3, or 5, while the target and fillers were displayed in six identically sized circular frames to facilitate localization (Maylor & Lavie, 1998).
The distractor appeared in one of four possible locations depending on the eccentricity, which was central (0.5 • to the right or left of the fixation icon) or peripheral (4.2 • to the right or left of the fixation icon).
The average size of the target and the filler was 0.6 • vertically and 0.4 • horizontally. In addition, aiming to elicit the same amount of activity in the primary visual cortex for both eccentricity conditions, the size of the central distractor was 0.3 • vertically and 0.2 • horizontally, and the peripheral distractor was 0.9 • vertically and 0.5 • horizontally, adjusted by the cortical magnification factor (Rovamo & Virsu, 1979).
Six circular frames were placed in a ring around the fixation icon at a distance of 2.1 • , the same as the distance between the center of each two adjacent circular frames.
Regarding the experimental design, the compatibility depended on the relationship between the distractor and the target: when the distractor was an ellipse, the relationship was natural; when the distractor was the same as the target, the relationship was compatible; and when these two were opposite, the relationship was incompatible. These three conditions were randomly presented in equal numbers within each block. We measured the degree of attentional distraction by compatibility effects, defined as the extent to which responseincompatible distractors interfered with task performance relative to response-compatible distractors, that is, incompatible minus compatible mean reaction times (RTs). Except for the target within a specific frame, the number of fillers in the remaining circular frames corresponded to the low-load condition (zero or one filler) or high-load condition (three or five fillers). According to the position where the distractor appeared relative to the inside or outside of the large circle formed by six circular frames, eccentricity was classified as either a central position (inside the circle, bilateral) or a peripheral position (outside the circle, bilateral).
For targets, distractors, and fillers, their parameters, including the shapes, positions, and relative placements, were counterbalanced for each load level and presented randomly within each block. With the position (6) and shape control (2) of the target, 48 frames per block were suitable to demonstrate the full balance of the perceptual load (4), distractor compatibility (3), and distractor eccentricity (4).

Procedure
Each trial started with a 500-ms fixed cross in the center of the screen, followed by a 100-ms graphic presentation where subjects were required to identify the shape of the target object as soon as it appeared. After 2 s, a feedback interface, in which participants were to indicate their responses, including correct responses, incorrect responses, or untimely responses, was presented in text form for 500 ms (see Figure 1b). Subjects were to judge the shape of the target object as it was presented each time and ignore irrelevant stimuli with the right index and middle finger placed on two keys that they would use to indicate their responses (key 1 on the numeric keypad corresponded to the square target, and key 2 corresponded to the diamond target). The participant's goal in the task was to respond as quickly and accurately as possible and to achieve a 90% correct rate.
Subjects initially performed the practice phase (two blocks, 96 trials), and after they understood the task requirements, they were able to complete the formal experiment (12 blocks, 576 trials). In the formal experiment, subjects received information about the RT and accuracy of the previous block at the end of each block, at which point they decided whether to take a short break to gain sufficient energy.

EEG acquisition and preprocessing
The behavioral and EEG data were collected simultaneously in a dark, quiet, and electromagnetically shielded room. According to the international 10-20 system, the 32-channel EEG signals were recorded by using a Biosemi ActiveTwo system (BioSemi, Amsterdam, Netherlands) and digitized at a sampling rate of 2048 Hz. The online filter band was 0.16-100 Hz, the ground was replaced by the driven right leg (DRL) passive electrode, and electrode Fz was set as the reference channel.
To obtain reliable results, a series of EEG preprocessing procedures were adopted, including 0.1-30 Hz IIR bandpass filtering; resampling to 512 Hz; referencing to "infinity" zero (Tian & Yao, 2013;Yao, 2001); segmenting from −200 to 600 ms (time-locked to stimulus onset) followed by a 200-ms baseline correction; removing eye-movement and blink artifacts based on the independent component analysis (ICA) method (the rejection rate of each subject was kept below 5%), which had been adopted by a series of studies with stable results (Koelstra et al., 2009;Koroma et al., 2020); and removing trials with artifacts (moving window peak-to-peak method with a 200 msec moving window, a 100 msec window step, and a 65 μV threshold, resulted in a rejection of an average 5.2% of trials).

Data analysis
The following analysis was based on previous studies (Green & Bavelier, 2003Lavie, 2010). First, trials with incorrect responses were excluded. RTs (greater than 300 ms and less than 1800 ms) and error rates were included separately in a three-factor within-subject analysis of variance (ANOVA) as a function of perceptual load (low vs. high), compatibility (compatible vs. incompatible), and eccentricity (central vs. peripheral). Natural distractors were not included in the current analysis as they were not the main focus of this study (Green & Bavelier, 2003.
Second, the mean ERPs for different conditions were obtained by averaging corresponding trials across subjects. We selected P2 at electrode Fz and P3 at electrode Pz as the main components based on previous studies and EEG topography (Brockhoff et al., 2022;Carlson, 2021;Polich, 2012). For latency, we adopted the "50% area latency" measure. This method was shown to work effectively for large components such as P3 waves, by averaging voltages of individual data points within a specified window and determining the specific point that divided the area under the curve into a 50% fraction (Kiesel et al., 2008;Luck & Hillyard, 1990). Then, we averaged the voltages at each point within a given time window to obtain more stable ERP amplitudes (Gan et al., 2020;Martens et al., 2006 were estimated to reveal the strength of critical null effects by measuring the degree to which an alternative hypothesis was supported by the data relative to a null hypothesis.

Reaction times
The results revealed a significant main effect of perceptual load  Table S1). None of the other two-way interactions (i.e., load and compatibility; eccentricity and load) were significant (all ps > .108). Of note, a perceptual load × compatibility  (Table S1). However, no load-compatibility interaction was found [F < 1, BF 10 = .128], suggesting that increasing load resulted in no meaningful reduction in the compatibility effect of central distractors (Figure 2b; Table S1).

Error rates
Error rates revealed significant main effects of compatibility   Table S2.

P3
The results for P3 latencies revealed no significant main effects (compatibility: p = .535; eccentricity: p = .647; load: p = .466) or interaction More descriptive information on the P3 component is provided in Table S3.

F I G U R E 3 P2 data (left side) for central distractors (a) and peripheral distractors (b) under differential load levels at the Fz electrode with corresponding topographic maps (right side). (c)
The peripheral trials reflected a significantly shorter latency in the high-load condition than in the low-load condition; however, this outcome was not observed in central trials. Topographic maps represented the average amplitude consistent with the P2 time window from 170 to 270 ms. Error bars refer to the SEM, **p < .01.

Correlation between ERP and behavioral parameters
P3 amplitudes were positively correlated with MAAS scores (r 31 = .418, p = .019, Figure 5a) and negatively correlated with error rates under high-load conditions (r 31 = -0.366, p = .043, Figure 5b) while not detected at low-load conditions (p = .180). Regarding the P3 latency, there was a trend of a positive correlation between P3 compatibility effects and behavioral compatibility effects (p = .204, Figure 5c). Furthermore, the exploratory analysis revealed that this trend was sensitive to the perceptual load level, with a significant strong positive correlation specific to high-load conditions between P3 compatibility effects and behavioral compatibility effects (r 31 = 0.415, p = .020, Figure 5d) compared to low-load conditions (p = .321). There F I G U R E 4 P3 data (left side) and corresponding topographic maps (right side) with load and compatibility conditions for central distractors (a) and peripheral distractors (b) at electrode Pz. (c) Estimated marginal means of P3 latency by load and compatibility. (d) As the perceptual load increased, incompatible latencies were ahead of compatible latencies, reflecting the change in compatibility effects under load manipulation. P3 amplitude for load (e) and eccentricity (f). Both peripheral distractors (vs. central distractors) and high load (vs. low load) induced a significant increase in P3 amplitude. Topographic maps represented the average amplitude consistent with the P3 time window from 230 to 330 ms. Error bars refer to the SEM, **p < .01.

F I G U R E 5
Plots showing the correlation of individual participants' event-related potential (ERP) measures with their behavior according to the relationship between P3 amplitude and Mindful Attention Awareness Scale (MAAS) score (a), P3 amplitude and error rate under high-load conditions (b), compatibility effects between P3 latencies and behavior (c), and compatibility effects between P3 latencies and behavior under high-load conditions (d). (a) and (b) show that the P3 amplitude is related to the individual's current level of attention and task performance under high-load conditions. (c) and (d) indicate that the compatibility effects, a measure of the number of attentional resources, have the same trend between response time and P3 latencies, being more pronounced with higher attentional demands. was no significant correlation between P3 compatibility effects and MAAS scores (p = .733) or error rates (p = .055).

DISCUSSION
This study focused on electrophysiological markers of attentional modulation of visual processing and revealed the distinct roles of P2 and P3 in the modulation of attentional processes by perceptual load.
Additionally, this study identified load-modulated differences in selective attention associated with eccentricity and corresponding neural response patterns.
From a behavioral perspective, we verified that the attention resources differed in how they were modulated by perceptual load depending on different eccentricity rates (Green & Bavelier, 2006 (Beck & Lavie, 2005). On this basis, we controlled the stimulus size to make the central and peripheral distractors equivalent in terms of retinal visual acuity under varied loading levels (Daniel & Whitteridge, 1961) and, interestingly, found that distractors in the central and peripheral areas produced different attentional trends with load changes. Specifically, as the load increased, the attentional distraction elicited by peripheral distractors decreased significantly compared to the change in the compatibility effects of central distractors that were modulated by the load, suggesting that the advantages of foveal attention were robust regardless of the load level and that peripheral attention was manipulable with respect to perceptual load. It was possible that since the attentional resources available to humans are limited, the tighter attentional resources that developed as perceptual load increased might preferentially support central attentional processing due to the attentional priority of the fovea itself (Calvo & Castillo, 2005). As a result, the attentional resources for central stimuli were relatively abundant, along with weakened attentional processing in the peripheral visual field. An alternative viewpoint suggested that the salience of the distractor may influence the allocation of focal attention (Schubö, 2009). The differences in the compatibility effect in the peripheral high-load condition may be due to a reduction in distractor salience compared to that in the low-load condition, where distractors may be more salient and therefore more likely to induce a spatial shift of attention. If the distractor salience and a spatial shift of attention played a major role, the central distractor was less likely to be attended than the peripheral distractor because it was differentially downsized according to the cortical magnification factor (central size: 0.3 • vertically and 0.2 • horizontally; peripheral size: 0.9 • vertically and 0.5 • horizontally, see Figure 1a). However, we found that the compatibility effect in the central high-load condition was significantly larger than that in the peripheral high-load condition [t(30) = 2.55, p = .016; central high-load: 16.54 ms ± 7.06, peripheral high-load: −5.25 ms ± 8.13]. Overall, these findings were more prone to reflect changes in the distribution of attentional resources under high and low loads.
To the best of our knowledge, the present study was the first to combine the perceptual load paradigm and ERP measurements to examine attentional processing under central/peripheral distractor manipulations. We chose perceptual load as the main comparison viewpoint to explore the relevant neural patterns under load variation, which was further identified by distractor eccentricity. The anterior P2 was shown to reflect a transdimensional feature detection process and some levels of higher order processing (Federmeier et al., 2005;Luck & Hillyard, 1994), and it has recently been proposed as a more integrated and extensive real-time attentional marker (i.e., "gate-keeper," GK) (Perrone-Bertolotti et al., 2020). This GK performance in late attentional selection occurs within 250 ms, responding to all incoming stimuli and holding the necessary information (Sawaki & Luck, 2010). It appears from the results that the P2 affected by the load manipulation was probably a marker of late attentional filtering; in particular, latency in the high-load condition was significantly earlier than that in the lowload condition. For an exploratory analysis, we analyzed P2 latencies of this effect by moving the acquisition point from Fz to electrodes F3 and F4 over the left/right inferior frontal sulcus (IFS), a location identified as a responsive brain region for GK via intracerebral EEG (Perrone-Bertolotti et al., 2020). We found the same GK responses within 250 ms, as well as the main effects of load similar to Fz (see Figures S1 and S2), suggesting a relatively stable and faster response to high loads and unbiased lateralization of the broader GK mechanism distributed in prefrontal regions as detected by scalp EEG. The "GK" is suggested to have an integrative mechanism that determines the final impact of the incoming stimulus on the global brain through bottom-up (stimulus-driven features) and top-down (expectations associated with the ongoing cognitive task) information. In our study, the mechanism of faster detection in the high-load condition by the GK was helpful and efficient for subsequent controls and the eventual allocation of attentional resources.
In addition, we found similar responses to task-related processes, as different distractor-target relationships (compatible and incompatible) induced the P2 component, which was consistent with the actions that should be taken as an efficient "GK," as it would handle all salient stimuli (potential threats should remain noticed) to save cognitive resources for task-related items. Interestingly, there was an interaction between eccentricity and perceptual load, whereby P2 latency was shorter in the peripheral high-load condition than in the peripheral low-load condition with no considerable change in central distractors. This was in agreement with the direction of our behavioral results, probably a consequence of the combination of both the perception of the optic rod cells in the periphery of the retina and the tendency to focus on the central visual field. However, this finding should be interpreted with caution, as unnoticed changes may have an effect on the ERP signal. We did the material balance to the best of our ability, but we could not completely rule out a possible material effect because of the need to include different amounts of filler items for separating low and high loads, and this outcome needs to be substantiated through replication.
Brain imaging studies using fMRI implicated an execution function network involved in the top-down control of attention (Giesbrecht et al., 2003;Witt et al., 2021), and several studies have further explored attentional processing through the parietal P3 as an electrophysiological indicator (Gan et al., 2020;Lin et al., 2018). Our results mainly showed that the compatibility effect of P3 latency decreased significantly as perceptual load increased and electrophysiologically confirmed a load-modulated attention pattern similar to behavior (Lavie, 1995(Lavie, , 2005(Lavie, , 2010Lavie et al., 2004). As revealed in previous studies, P3 was highly sensitive to the processing capacity of attentional resources required in the task (Herrmann & Knight, 2001;Polich, 2012), and P3 latencies were found to be closely linked to RTs, with earlier latencies indicating faster response execution and shorter response times (Doucet & Stelmack, 1999;Pfefferbaum et al., 1980). Furthermore, the parietal lobe was shown to play the role of a hub node in target selection to guide visual attention, and P3 could depict the deeper mechanism of this time-varying brain network, reflecting topdown attentional control through the difference between the response to the target and a distractor (Bisley & Goldberg, 2010;Li et al., 2016).
Thus, what we observed by activation of this area may reflect the overall processing of attentional resources during higher cognitive processing as a result of the assessment of the current state; this reduced compatibility effects of P3 in attentional processing between high and low loads reflected the subsequent behavioral effects on performance. In addition, we observed that high-load conditions (vs. low load) and peripheral distractors (vs. central) significantly induced greater P3 magnitudes, revealing that higher amplitudes of both high-load and peripheral conditions may result from a greater synchronization of brain resources (Jo et al., 2016). On the other hand, a potential explanation from an attentional resources perspective for the parietal P3 component may be the fact that its amplitude is related to the intensity of attentional processing, suggested by previous findings that P3 amplitudes depend on the capacity for processing task-relevant stimuli and the increasing consolidation of attentional resources for taskinduced potential amplitudes as resource demands grow (Kok, 2001;Martens et al., 2006). Consistent with expectations, P3 elegantly complements behavioral measures of attentional resources, demonstrating that the attentional resource differences between high-and low-load conditions were reflected by both neural and behavioral operationalizations, as well as providing a potential explanation for behavioral differences in eccentricity.
We then related the observed electrophysiological changes in attention (i.e., P3) to behavioral performance. Studies revealed that people with high MAAS scores, more specifically, those with an advantage in voluntary attentional control, had an enhanced attentional processing intensity reflected by higher P3 amplitudes (Brown et al., 2015;Quaglia et al., 2016). We observed an analogous outcome, that is, a significant positive correlation between P3 amplitudes and MAAS scores.
This is an interesting finding because it not only reveals that the ability to focus attention is more likely to correlate with the intensity of attentional processing regardless of load manipulations but also suggests the possibility that the attentional processing of task-relevant stimuli can be enhanced or assessed through purposeful daily training that improves one's ability to control attention. In addition, there was a significant negative correlation between the P3 amplitude and the error rate in high-load conditions. It seems that individuals with higher P3 amplitudes in attentional processing were better adapted to the current task due to their likely stronger processing of task goals and greater synchronization of brain resources, especially when their attentional resources were challenged (i.e., high load), leading to better task performance overall (Gan et al., 2020). This trend was sensitive to load manipulations, possibly because the high-load condition was more attentional-resource consuming and distraction processing was less effective, in which case the error rates may better characterize the task processing performance of individuals (for a comparable explanation of P3 being sensitive to increasing load, see Bidet-Caulet et al., 2015). Interestingly, we found a positive correlation between the compatibility effect of P3 latency and behavior on a time scale, primarily driven by the performance in the high-load condition. The similar pattern of this trend being more pronounced when attentional demands increase again reflects the sensitivity of P3 as a signature of attentional capture to load manipulations, which may be an aspect to be considered in subsequent studies. Overall, in our study, the parietal lobe, as a hub node in the executive function network, provided a possible EEG measure of attention, known as P3, which effectively reflected processing outcomes in conditions with higher cognitive domains where the top-down mechanisms modulated the bottomup mechanisms, consistent with the behavioral findings of attention modulation.
There were several limitations in the current study. First, although we tried to avoid or balance as many potentially irrelevant variables as possible, the impact of this balance on ERPs was difficult to assess (e.g., the balance of cortical representations between peripheral and central distractors, additional interference caused by filler items). Hence, we interpreted the relevant results with caution and expect more sup-port from future studies. Second, we focused more on the temporal dimension of modulation in the later selective processing and explored correspondences between behavior and ERP findings. Future studies are encouraged to include other approaches, such as rhythmic oscillations and functional brain connectivity to address attentional load modulation from different perspectives. Third, our study considered male participants to eliminate potential confounding effects. To generalize these findings, we suggest that future studies consider a broader range of population attributes (e.g., females, attention deficit). In addition, given that life experiences, such as gaming, may have varying effects on attention, future research could consider these potential factors and explore whether different levels of proficiency in the use of cognitive resources may respond differently to load manipulation.
Fourth, future studies could apply the paradigm of this experiment and further improve it according to their own experimental purposes, for example, by maximizing the number of load manipulations and increasing the number of trials per condition. Fifth, the present study revealed the influences of load on attentional resources and EEG markers. However, attention may include diverse attentional subprocesses, and it would be enlightening to explore the effects of load on different attentional mechanisms (e.g., natural condition) combined with more dimensional techniques, such as fMRI.

CONCLUSION
This study explored the processing of selective attention modulated by perceptual load during EEG recordings, revealed the specific atten-

CONFLICT OF INTEREST STATEMENT
All authors claim that there are no conflicts of interest.

DATA AVAILABILITY STATEMENT
Some or all data or code generated or used are available from the corresponding author on reasonable request.