Sustained selective attention to chromatic information enhances visuocortical gain at the population level

Prior work in selective attention research has shown that colour‐selective attention enhances neural activity in visuocortical areas sensitive to the attended colour while suppressing activity in areas sensitive to ignored colours. However, it is currently unclear whether this effect is limited to attending to specific colour hues or extends to chromatic information more broadly. To investigate this question, we used steady‐state visual evoked potentials (ssVEPs) frequency tagging to quantify participants' visuocortical responses to specific elements embedded in arrays of flickering, randomly moving mid‐complex patterns. Participants were instructed to attend to either coloured or greyscale patterns while ignoring the others. We found that attending to either coloured or greyscale patterns produced robust increases in ssVEP amplitudes both compared to ignored stimuli and to baseline. There was however no evidence of suppressed responses to ignored patterns. These findings demonstrate that attentional selection based on the presence or absence of chromatic information prompts selectively enhanced visuocortical processing but this selective amplification is not accompanied by suppression of unattended stimuli. Findings are consistent with theoretical notions that predict strong competition between specific exemplars within a given feature dimension, such as red or green, but weak competition between broadly defined stimulus categories, such as chromatic versus non‐chromatic.

. A substantial body of previous work has shown that the visuocortical responses to target features are globally enhanced, even when these features are spatially distributed across the field of view (Adamian et al., 2017;Andersen et al., 2011;Bartsch et al., 2018;Chapman & Störmer, 2021;Saenz et al., 2002;Snyder & Foxe, 2010;A. L. White & Carrasco, 2011).Its independence of spatial attention makes feature-based attention more efficient than serial scanning strategies (i.e., moving the "attentional spotlight") when behaving in the context of larger, cluttered visual scenes.Given its fundamental nature in cognition, the neural mechanisms of featurebased attention have been studied extensively.The Feature Similarity Gain model (Maunsell & Treue, 2006) has posited that the selection of an attended feature involves selectively enhancing the gain of visuocortical neurons that are tuned to the attended feature (Gledhill et al., 2015;Schoenfeld et al., 2007;Treue & Martínez Trujillo, 1999).Based on the predictions of the Feature Similarity Gain model, a number of studies in humans had observers respond to target stimuli based on their specific colour.As expected, colour-selective attention has been shown to improve task performance while also prompting enhanced visuocortical responses to a target colour (Anllo-Vento et al., 1998;Corbetta et al., 1990;Ernst et al., 2013;McMains et al., 2007;Müller et al., 2006;Schoenfeld et al., 2007;Zhou & Desimone, 2011).Moreover, reduced brain responses to distractors or taskirrelevant colour stimuli suggest attentional suppression processes, such as lateral inhibitory interactions, under some circumstances (Andersen & Müller, 2010;Forschack et al., 2017;Gundlach et al., 2022;Müller et al., 2018;Störmer & Alvarez, 2014).
Many studies of colour-selective attention use Random Dot Kinematograms (RDKs).Here, participants are instructed to focus attention on arrays of randomly moving, simple geometric target stimuli (e.g., small squares) defined by a single distinct colour hue.To eliminate confounds with spatial attention, these arrays are typically spatially overlapping with an array of competing stimuli with differing hues (Andersen et al., 2008(Andersen et al., , 2009;;Andersen & Müller, 2010;Bartsch et al., 2018;Müller et al., 2006Müller et al., , 2018;;A. L. White et al., 2015).Only a few studies have also examined simultaneous attention to two colours or distraction by multiple colours (Andersen et al., 2013;Beck et al., 2012;Chapman & Störmer, 2022;Martinovic et al., 2018;Störmer & Alvarez, 2014).As such, prior work has focused on selection and competition between specific exemplars within a feature dimension such as a specific hue (Andersen et al., 2008(Andersen et al., , 2009) ) or orientation (Thigpen et al., 2019).Thus, it is currently not known to what extent visuocortical amplification and competition/suppression are observed (a) when processing stimuli of higher geometrical complexity and (b) when attending not to a specific hue but to the broadly defined dimension of chromaticity.These questions are important to address because naturalistic vision involves attention to complex, dynamic stimulus arrays (Roelfsema, 2006;Seijdel et al., 2021) and because colour saturation may serve as a cue in many real-world attention scenarios, in addition to the specific hue of a tobe-attended object (Engmann et al., 2009;Nuthmann & Malcolm, 2016;T. E. White et al., 2017).
Here, we leveraged the properties of steady-state visual evoked potential (ssVEP) frequency tagging, to examine the selective amplification of chromaticity as well as competitive interactions between attended and unattended stimuli.The ssVEP is an oscillatory response at the neural population level to periodically modulated visual stimuli, typically recorded by means of scalp EEG (Norcia et al., 2015;Regan, 1966;Vialatte et al., 2010;Wieser et al., 2016).The strength of these responses can be readily quantified in the frequency domain, as the spectral amplitude has the same fundamental frequency as the modulation of the driving stimulus.The ssVEP approach provides two major advantages for the investigation of attentional processes.First, given that the ssVEP is evoked by continuous stimulus modulation, the effects of attention can be assessed over the entire time course of stimulus presentation.Second, responses to multiple stimuli (e.g., targets vs. distractors) simultaneously presented at different driving frequencies can be separated in the frequency domain, making ssVEPs a useful tool for assessing the effects of stimulus competition.
The present study used ssVEPs to investigate mechanisms of sustained selective attention to chromatic information in dynamic stimulus arrays.To this end, we presented spatially overlapping and randomly moving arrays of mid-complex pattern stimuli suitable for prompting recurrent processing among visual areas (Roelfsema, 2006;Seijdel et al., 2021).Importantly, participants were cued to selectively attend to target events in a subset of pattern stimuli after a pre-cue baseline period.More precisely, participants had to attend to patterns based on either the presence or absence of chromatic information (i.e., colour versus greyscale pictures), while ignoring the other stimulus set (i.e., distractors).We found that attention to chromatic information (or greyscale) increased neural gain as reflected in higher ssVEP amplitudes relative to a pre-cue baseline level and the to-be-ignored distractors but found no evidence of suppressed responses to distractors.

| Sample
A total of N = 24 undergraduate students at the University of Florida participated for course credit (age: M = 18.8, SD = 1.2 years; gender: 14 x female, 10 x male; reported ethnicity: 14 x White, 6 x White/Hispanic, 3 x Asian, 1 x Native Hawaiian/Pacific Islander/White).All participants reported having no colour vision deficiencies.The study was conducted in accordance with the Declaration of Helsinki and was approved by the local ethics committees.

| Paradigm, stimuli and procedure
Participants underwent a sustained attention task with a baseline phase and a cued selective attention phase in each trial.For the entire duration of a trial, a central grey square (RGB: 127, 127, 127; 1.0 Â 1.0 visual angle), as well as two spatially mixed picture arrays (four colour and greyscale pictures, respectively), were presented against a black background on a Display++ LCD screen (120 Hz refresh rate; Cambridge Research Systems, Rochester, UK).In each trial, eight pictures were randomly drawn from a set of 20 pictures without replacement.Each of the 20 pictures was a cut-out from a larger scene and was available in colour and in greyscale.In other words, there were 20 colour pictures and 20 greyscale pictures with the same 20 motives and the same motive was not used more than once per trial, even across colour/ greyscale arrays.Pictures were presented at a visual angle of 1.5 Â 1.5 and were distributed across a circular area of 10.2 diameter.The two picture arrays were luminance-flickered by modulating the alpha channel with a square-wave function between 0 and 1 (50% on/off; Panitz, Gundlach, et al., 2023) at 8.57 Hz and 15 Hz, respectively, counterbalanced across trials.Between screen frames (i.e., every 8.33 ms), each picture randomly moved by À0.07 , 0 or +0.07 on the horizontal and vertical axis, respectively.Pictures never overlapped with each other or with the central square.
The baseline phase lasted 1400, 1867 or 2333 ms, during which a cross (tick length: 0.85 ) was presented.A variable baseline length was chosen in order to avoid participants learning to preemptively switch to the cued attention task (see next paragraph).During this phase, participants were instructed to fixate the cross and count the number of target events, increases in tick length (+10% for 200 ms).Events could occur between 0 and 2 times per trial with a minimal onset latency of 300 ms after the trial start or a preceding event.
At the beginning of the cued attention phase, the central cross was replaced either by the letter 'C' (for colour) or 'G' (for greyscale) and instructed participants to attend to all pictures of the cued array.This trial phase lasted 3733, 4200 or 4667 ms, depending on the baseline length (total trial duration was always 6067 ms).Participants were instructed to keep their fixation on the central grey square and to count transient luminance changes in the attended picture array while ignoring changes in the other array.In the case of a target event, one picture flickered for 600 ms at 25% of the normal alpha value.There could be 0 to 3 luminance events in the cued and/or uncued array, respectively, with random picture selection for each event.There was a minimum onset latency of 800 ms after the start of the cue phase and after preceding luminance events.
After the stimulus presentation, there was a small delay (0.5 to 1.5 seconds with a white fixation cross) before the rating screen.Here, participants had 4 seconds to indicate on a slider scale the total number of counted target events, combining baseline and luminance events (scale range: 0 to 6; maximum number of events possible in a trial: 5).There was another delay with fixation cross between the rating and the next trial.The total duration of the delays and the rating was randomly drawn from a cut-off exponential distribution (min: 6.5 s, max: 16.4 s, mean: 8 s).There were a total of 192 trials: 140 trials without any event, 16 trials with baseline events only, 18 trials with luminance events only and 18 trials with both baseline and luminance events.From the 36 trials with luminance events, 12 trials had events only in the attended array, only in the to-be-ignored array or in both, respectively.Trial order was randomized and participants had a break after blocks of 64 trials that they could end themselves by a button press.Stimulus display and trial structure are shown in Figure 1.
Prior to the actual experiment, participants underwent automated instructions and training trials.Training ended when participants either had five correct responses over the previous six trials or after a total of 20 trials.Participants were given the opportunity to repeat the instructions and the training and could ask the experimenter for additional explanation if necessary.

| EEG recording and preprocessing
We recorded EEG using 129-channel HydroCel Geodesic Sensor nets and an EGI Net Amps 300 amplifier with an online lowpass filter at 131 Hz (-3 dB, 3rd order Sinc).Cz was used as the online reference and data was digitized at 500 Hz.Offline EEG processing was performed in emegs 2.8 running in MATLAB 2021b (MathWorks, Natick, MA, USA).Here, we applied highpass (À1 dB at 3 Hz, 3rd order Butterworth) and lowpass filters (À3 dB at 33 Hz, 12th order Butterworth) before segmenting data relative to the attention cue onset (À1600 to 4000 ms).Trials containing luminance events were excluded from analyses.Eye artefacts were detected and corrected using the automated algorithm by Schlögl and colleagues (Schlögl et al., 2007).Bad channels and trials were identified using the SCADS algorithm (statistical control for artefacts in dense array EEG/MEG Studies; Junghöfer et al., 2000) implemented in emegs.The SCADS algorithm identifies problematic data based on three parameters, computed for each channel and trial: absolute maximum, standard deviation and maximal absolute gradient (i.e., sample-to-sample amplitude changes).Channels in the Cz-referenced EEG were flagged as bad if, for at least one of the parameters, the channel's median was more than five standard deviations above the median of all channels' medians.Bad channels were interpolated using weighted spherical splines, with an average of 9.0 bad channels per participant.Subsequently, data were referenced against the average.For each trial, depending on the number and spatial distribution of remaining bad channels, channels with artefacts were interpolated or the trial was rejected entirely (see Junghöfer et al., 2000;Peyk et al., 2011 for details).Within each participant, trials were baseline-corrected (À200 to 0 ms relative to cueonset) and averaged across trials separately for the four trial types: cueing colour at 8.57 Hz, cueing colour at 15 Hz, cueing greyscale at 8.57 Hz and cueing greyscale at 15 Hz driving frequency.Each trial type also contained data for the distractor pictures at the other driving frequency, i.e., uncued greyscale distractors at 15 Hz were assessed in the same trials as colour targets at 8.57 Hz.On average, there were between 29.0 and 29.7 usable trials per trial type.Finally, data were transformed to current source density (CSD) before Hilbert transformation.

| Hilbert transformation
Each participant's average EEG signal in the time domain was multiplied with a squared-cosine window (ramping up/down over 400 ms) and bandpass-filtered with a 12th-order Butterworth filter (-3 dB at driving frequency +/À 0.7 Hz).In order to obtain spectral F I G U R E 1 Trial structure.During the entire trial, there was a static grey square in the centre of the screen and two picture arrays of four colour and greyscale pictures, respectively.Colour and greyscale pictures flickered at 8.57 and 15 Hz, respectively (counterbalanced across trials) and moved randomly.In the baseline phase, participants were instructed to count how often the cross in the central grey square increased in size for a moment.In the cued attention phase, the cross was replaced by a letter, instructing participants to count temporary luminance changes in either the colour ('C') or the greyscale ('G') pictures while ignoring the other array.After the cued attention phase, there was a fixation cross for 0.5 to 1.5 seconds before participants had 4 seconds to indicate the total number of target events, that is the changes in cross size and luminance changes in the attended array together.The following ITI was between 1 and 11.9 seconds long.Stimuli are not to scale.amplitudes at driving frequencies over time, bandpassfiltered data was Hilbert-transformed, resulting in eight different time courses for each participant, one for each of the combinations of cued vs. uncued, colour vs. greyscale and 8.57 Hz vs. 15 Hz.Hilbert amplitudes were normalized by dividing by the average amplitude in the time window from À1000 to À200 ms relative to cue onset and transformed into percent change relative to baseline.

| Statistical analyses
We used R (R Core Team, 2018) in the RStudio (RStudio Team, 2016) environment.For statistical testing, Hilbert amplitudes at Oz were averaged for the time window from 500 to 3500 ms post-cue.The onset of the time window was based on previous studies (Andersen & Müller, 2010;Gundlach et al., 2022;Vieweg & Müller, 2020).We computed analyses of variance (ANOVA), both frequentist (anova_test from the rstatix package; Kassambra, 2021) and Bayesian (generalTestBF from the BayesFactor package; Morey & Rouder, 2018), with a 2 Â 2 Â 2 repeated-measurement design, including the factors Cue (cued vs. uncued), Chromaticity (colour vs. greyscale) and Driving Frequency (8.57Hz vs. 15 Hz).For the frequentist ANOVA, we used Type 3 sum of squares.For the Bayesian ANOVA, we computed Bayes factors for all potential combinations of main effects and interactions, including the participant intercepts and effect slopes (i.e., interactions between the Participant factor with all main effects and two-way interactions of the Cue, Chromaticity and Driving Frequency factors) as random terms in each model following recent recommendations (van den Bergh et al., 2022).The null model included the overall intercept as well as the random effect terms.We assigned equal prior probabilities to each model for subsequent model comparisons.We used 10,000 iterations for the Monte Carlo sampling of the g parameters in the generalTestBF function and set scaling factors for effect priors to 1/2 for fixed effects and 1 for random effects (both defaults of the function).We also computed Inclusion Bayes Factors (BF Inc ) with the bf_inclusion function of the bayestestR package (Makowski et al., 2019).Here, for each ANOVA effect, all models including this effect are compared to all models without the effect.This allows to quantify evidence for single main and interaction effects rather than entire models (Hinne et al., 2020).Partial eta-square (η 2 p ) was used as a standardized effect size.
For more detailed result patterns, we conducted frequentist (t_test function from the rstatix package) and Bayesian (ttestBF function from the BayesFactor package) one-sided t-tests of ssVEP amplitudes against the baseline (cued > baseline, uncued < baseline) and between Cue conditions (cued > uncued) for each combination of chromaticity and driving frequency cell.Resulting pvalues in the cells were corrected using the (Bonferroni-) Holm method (Holm, 1979).The BayesFactor package's default scale parameter for the Cauchy distribution was used (sqrt[2]/2).To compute Bayes Factors reflecting the directed hypotheses, the intervals compared to the point null were set to [0, Infinite] (cued > uncued, cued > baseline) and to [-Infinite, 0] (uncued < baseline).
Cohen's d was computed as the mean of the pairwise difference divided by the difference's standard deviation (cohens_d from rstatix).
In addition to analyses on the a priori time window, we conducted sample-wise comparisons of cued and uncued conditions, averaging across colour and driving frequency conditions.Significance was determined using cluster-based permutation testing (Bullmore et al., 1999;Gundlach et al., 2022;Maris & Oostenveld, 2007).In the first step, for each sample, t-values were computed and thresholded ( p < .05,one-sided).Clusters of connected significant samples were identified and t-values were summed up for each one.A critical t-sum threshold was determined using a reference distribution obtained from permutation testing.In each of 1000 iterations, we randomly assigned the condition labels to the data for each participant, computed sample-wise t-tests and extracted the largest t-sum value.From the resulting distribution, we selected the 95th percentile as critical cluster statistic.

| Task performance
The median percentage of correct responses across participants was 80.5% (range: 22.4% -98.4%).Participants had a median of one trial without response (range: 0-32).In order to compare miss rates in the two task phases, we counted how often participants reported less than the actual number of presented target events in trials that contained only baseline events (Md = 12.5%) and in trials that contained luminance target events but no baseline events (Md = 20.8%).Moreover, we ruled out the possibility that participants always reported 0 target events as a general strategy (which is the correct answer in 72.0% of the trials): in trials with at least one target event, participants reported 0 events only in Md = 4.3% of the cases (with participants ranging from 0.0% to 30.4%).Additional analyses on performance data are provided in the Supporting Information (Figures S1 and S2).

| Steady-state visual evoked potential
The Cue x Chromaticity x Driving Frequency ANOVA resulted in a significant main effect of Cue (F [1,23] = 8.6, p = .007,η p 2 = .272,BF Inc = 4.29), driven by higher ssVEP amplitudes to cued compared to uncued pictures.Moreover, there was a trend for higher ssVEP amplitudes in the 8.57 compared to the 15 Hz band (main effect Driving Frequency: F Neither the main effect of Chromaticity (F [1,23] = 1.6, p = .223,η p 2 = .064,BF Inc = 0.414), nor the Chromaticity x Driving Frequency interaction were significant (F [1,23] = 0.3, p = .562,η p 2 = .015,BF Inc = 0.317).Permutation tests showed a significant difference between cued and uncued stimuli that started at 1182 ms post-cue and persisted throughout the time window of interest.Following up on the main effect of Cue, we tested the average amplitudes of all cued and all uncued conditions against baseline levels, respectively.t-tests revealed that ssVEP amplitudes to cued stimuli were significantly increased relative to baseline (t[23] = 4.10, p < .001,d = 0.84, BF = 145.5).Meanwhile, we found no evidence for attentional suppression as amplitudes to uncued pictures were not below baseline level (t[23] F I G U R E 2 (A) Time course for the main effect of Cue on Hilbert amplitudes and standard error at Oz, averaged across colour and greyscale pictures as well as driving frequencies.Amplitudes were transformed into percentage change relative to baseline (À1000 to À200 ms).Grey shaded area indicates time window of interest for statistical analyses.The blue line above the x-axis indicates significant between cued stimuli and baseline as determined by cluster-based permutation testing.The red line indicates significant differences between cued and uncued stimuli.(B) Average topographies for all cued conditions, uncued conditions and the difference, showing mean amplitudes from 500 to 3500 ms post-cue (highlighted in grey in panel A).Oz is highlighted in white.
= 1.51, p = .928d = 0.31, BF = 0.095).Figure 2 shows the time course of spectral amplitudes at Oz as well as topographies for responses to cued versus uncued stimuli.Permutation tests showed that ssVEP amplitudes to cued stimuli relative to baseline were significantly increased from 676 ms post-cue and throughout the time window of interest.Meanwhile, there were no significant clusters for the comparison of uncued pictures to baseline.

| Control analyses
We repeated the Cue x Colour x Driving Frequency ANOVA after excluding participants with less than 50% correct responses (n = 3).Original ANOVA results were replicated in that we found a significant main effect of Cue (F [1,20] = 6.6, p = .018,η p 2 = .249,BF Inc = 2.11) and a trending main effect of Driving Frequency (F [1,20] = 3.0, p = .100,η p 2 = .130,BF Inc = 0.707) but no other significant effects (all F < 1.7, p > .207,η p 2 < .079, We also conducted a control analysis to rule out the possibility that higher ssVEP amplitudes to cued pictures were not driven by participants directing attention but merely by the fact that participants fixated their eye gaze on the cued pictures.As we did not record eyetracking data, we leveraged the EOG data to collect evidence for or against this potential confound.The detailed method and results can be found in the Supporting Information (Figures S3 and S4 and Table S1).
F I G U R E 3 Distributions of ssVEP amplitude change relative to baseline for individuals (circles) and condition means (diamonds).
Briefly, we hypothesized that a fixation strategy would be reflected in continuous gaze shifting and thereby sustained increased EOG activity during the cued attention phase relative to the baseline task.This was because (a) participants would need to constantly jump between pictures which (b) are randomly moving.While we identified significant increases in EOG activity in some participants, these were not related to the differential effect of cued vs. uncued pictures on ssVEP amplitudes.We therefore ruled out fixation of cued pictures as a confound.

| DISCUSSION
The present study investigated how visuocortical responses are modulated in participants attending to stimuli based on their chromaticity.We recorded frequency-tagged ssVEPs in response to randomly moving pictures that either contained chromatic information (colour) or not (greyscale) while participants were instructed to attend to one stimulus type and ignore the other.We found that ssVEP amplitudes to cued stimuli were increased relative to distractor stimuli, regardless of whether they contained chromatic information or not.Comparing amplitudes to baseline levels revealed that this effect was driven by increased neural gain in response to to-be-attended pictures but not by suppressed responses to to-be-ignored pictures.
In this study, we found that cueing stimuli based on the presence or absence of colour evoked increased ssVEP amplitudes compared to uncued stimuli and baseline.Our results suggest that neural populations in the human visual cortex increase their gain when attending to chromatic information.Prior studies have demonstrated that individuals can selectively attend to more than one hue, but attention was typically limited to a maximum of two hues that were specifically instructed (Beck et al., 2012;Martinovic et al., 2018;Störmer & Alvarez, 2014).In our study, participants were asked to attend to the presence or absence of chromatic information.The ssVEP response we observed may be more than the sum of its parts, that is, more than the summed activity of all neurons selective to different colours.For example, previous research has shown that the sum of ssVEPs amplitudes to two simultaneously attended hues was larger when the hues were closer in colour space (Martinovic et al., 2018).This finding is consistent with lack of suppressive interactions or lack of lateral inhibitory processes between certain hues.It is currently unknown how lateral inhibition processes scale when the entire range of hues is attended, and further research is needed to explore this.Prior work explicitly combining spatial and feature-based attention using fMRI has likewise found strong evidence for feature-selective enhancement in cortical areas specialized for attended features, when switching between multiple colour and motion direction targets (McMains et al., 2007).Such enhancementas reflected by increased ssVEP amplitudeshas been mainly linked to increases in visuocortical gain (Di Russo et al., 2001;Kim et al., 2007;Lauritzen et al., 2010;Song & Keil, 2013) although smaller contributions might also come from stronger phase synchronization (Kim et al., 2007).Notably, there was no evidence for suppression of the unattended stimulus or dimension.Expanding these findings, the present study demonstrates that the neural gain in the human visual cortex can be increased in response to attending the presence or absence of chromatic information across the colour spectrum.
When analyzing ssVEPs to to-be-ignored stimuli, we found no evidence of distractor suppression as amplitudes did not fall significantly below baseline levels.Such suppression has been reported in previous studies on colour-selective attention (Andersen & Müller, 2010;Forschack et al., 2017;Gundlach et al., 2022;Müller et al., 2018;Vieweg & Müller, 2020).The current lack of suppression effects suggests that the task at hand did not induce sufficient competition between the cued and the uncued stimulus arrays.First, this might be because colourful and grey pictures are not suited to create within-dimension competition.According to the Feature Similarity Gain model, suppression of a feature will only occur if this feature is on the same dimension as the target feature (i.e., attending to blue will lead to a suppression of responses to red but not to square shape).Although colour and greyscale pictures could theoretically be mapped on the same physical feature dimension, namely saturation (Schiller & Gegenfurtner, 2016;Stuart et al., 2014), the lack of suppression in uncued pictures (in the context of the Feature Similarity Gain Model) suggests that this mapping does not translate to the neural level.In order to test whether stimuli with versus stimuli without chromatic information are processed as different dimensions (e.g., of "colouredness" versus "greyness") and can create between-dimension competition, future studies might use paradigms that have been the basis of the Dimensional Weighting Account (cf.Liesefeld et al., 2019).A second explanation for the lack of suppression below baseline levels in the current study might be insufficient spatial overlap between to-be-attended and to-be-ignored stimuli.Suppression of ignored RDKs based on their colour appears to require spatial overlap (Forschack et al., 2017;Müller et al., 2018) and probably only occurs when both targets and distractors are presented in the same receptive fields (Desimone & Duncan, 1995;Kastner et al., 2001;Kastner & Ungerleider, 2001).While we aimed at reducing the effectiveness of spatial attentional strategies by mixing and randomly moving the pictures, the larger picture size and restriction to not overlap with each other may have prevented targets and distractors to fall into the same respective fields for longer time periods.Future studies could let pictures overlap with each other to collect evidence for or against this hypothesis.
Picture stimuli were not only multi-coloured (and multi-shaded in the case of greyscale pictures) but also represented mid-complex patterns (see Figure 1) instead of uniform surfaces as used in previous work.Similar stimuli were used in studies by Kastner and colleagues (e.g., Kastner et al., 1998, 1999, 2001).Even though these studies focused on the processing of visual stimuli under spatial competition, they may have implications for the present study.Kastner and colleagues found extrastriate areas involved in attention allocation to the pattern stimuli, including V4.This area in the ventral visual cortex is important for selective feature extraction and attentional feature selection in general (Roe et al., 2012) and should be more responsive to pattern stimuli than to unicoloured shapes.Moreover, subsets of neurons in V4 play a major role specifically in colour perception (Conway, 2014) and colour-selective attention (McMains et al., 2007;Schoenfeld et al., 2007).While neural activity underlying the scalp ssVEP is generated in early visual cortex (i.e., V1-V3; Andersen et al., 2008Andersen et al., , 2009;;Di Russo et al., 2007) and motion-sensitive areas (i.e., MT/V5; Di Russo et al., 2007), we speculate that other extrastriate areas, V4 in particular, likely are involved in its modulation by attention to chromatic information as assessed in the current study.Future research may combine simultaneous EEG and MRI recording to test the dynamics of such top-down modulation within the visual cortex.
In our investigation of ssVEP time courses, we observed that responses to cued pictures descriptively started to increase relative to uncued pictures at approximately 700 ms post-stimulus onset.This differentiation became significant at around 1200 ms and persisted throughout the trial.This onset time is later than in previous studies utilizing competing RDKs to evaluate feature-selective attention, which typically showed onsets around 300-400 ms post-cue (Andersen & Müller, 2010;Forschack et al., 2022;Gundlach et al., 2022;Vieweg & Müller, 2020).While it is conceivable that detectable effects of attention to broadband chromatic information may take longer to translate into increased neural gain, a more straightforward explanation is the more complex task design used in our study.First, we included a baseline task where attention was directed towards a central cross instead of the RDKs, requiring within-trial task switching.Meanwhile, prior studies had no baseline task (Andersen & Müller, 2010;Forschack et al., 2022;Gundlach et al., 2022) or already had participants attending the RDKs focusing on a different feature (Vieweg & Müller, 2020).Secondly, previous studies used attention cues drawn in the relevant colour (e.g., a blue fixation cross for attention-to-blue) rather than symbolic letters.Consequently, our a priori time window for attentional effects was suboptimal for this paradigm, leading to underestimation of attentional effects that began to manifest only later in the trial.Nevertheless, once attention was established, the effects on neural gain persisted throughout the trial, indicating that individuals are capable of sustained attention to chromatic information.
Several limitations should be considered in the interpretation of our results.First, our RDKs had fewer and larger elements than in previous studies, which may have influenced the effectiveness of different attention strategies.We chose the current picture size in order to ensure that participants could recognize the depicted elements.With only four elements per picture array, we cannot rule out the possibility that object-based attention strategies were used by participants.Previous studies have shown that feature binding (i.e., the integration of multiple features to perceive them as one object), may oppose topdown separation of attended and unattended features as attention spreads from relevant features to irrelevant features within the same object (Snyder et al., 2012;Snyder & Foxe, 2012).These results may not directly translate to our design with multiple, independently moving objects, each exclusively containing either relevant or irrelevant features.Nonetheless, future studies could show more pictures per array, making it harder for participants to track each picture as an individual object (Adamian & Andersen, 2022).Another limitation is that, in each trial, participants had to keep in mind the number of targets in the baseline task while counting targets in the cued attention phase, which required engagement of working memory.Although we do not think that this working memory engagement was particularly high, it is possible that it influenced attentional processing.However, even if working memory engagement did have an effect on our results, it would be constant across conditions.

| CONCLUSION
In the current study, using an RDK design with midcomplex, multi-coloured patterns, we could demonstrate that the gain of neural populations in the human visual cortex increases in response to attending the presence or absence of any chromatic information across the entire colour spectrum.Notably, we found no evidence for suppression of responses to stimuli that were ignored based on their chromaticity.Further research may reveal more details of active processes when the entire range of colour hues is attended or ignored.