Alpha-band fluctuations represent behaviorally relevant excitability changes as a consequence of top–down guided spatial attention in a probabilistic spatial cueing design Open Access

Spatial attentional selection is a key function in neural processing aiding the selection and prioritization of information relevant for a meaningful interaction with the environment. One of the fundamental outcomes of the deployment of spatial attention is behavioral benefits for stimuli presented at attended locations and behavioral costs for stimuli presented at unattended locations (Carrasco, 2011; Posner, 1980). This is seen in faster and more accurate responses, enhanced detection to weaker and less salient stimuli and higher contrast sensitivity or spatial acuity (Carrasco, 2011). Different neural measures were found to be sensitive to visual spatial attention modulation: Early components of EEG-recorded responses to briefly presented either attended or unattended visual stimuli such as the P1 or N1 were amplitude modulated by attention (Hillyard & Anllo-Vento, 1998; Hillyard et al., 1998; Luck, 1995; Luck et al., 1997, 2000; Marzecová et al., 2018; Slagter et al., 2016). The earlier findings led to the sensory gain control hypothesis of attention, posing that neural activity associated with the processing of stimuli is modulated in amplitude by attention and that this gain control may be implemented at early stages of visual processing (see Hillyard et al., 1998). The sustained nature of this attentional selection and its related enhanced stimulus processing was made explicit by employing frequency tagged stimuli that evoked Steady-State-Visual-Evoked-Potentials (SSVEPs). SSVEPs are evoked oscillatory signals that follow the frequency of the tagged stimulus, respectively (Adrian & Matthews, 1934; Norcia et al., 2015; Regan, 1989), and their generators were found in early visual areas of the human brain (Boylan et al., in press; Hillyard et al., 1997; Moratti et al., 2023; Müller, Picton, et al., 1998; Müller et al., 1997; Pastor et al., 2007). Crucially, it was found that when such flickering stimuli were spatially attended, flicker-evoked SSVEP amplitudes were increased for the attended as compared with the unattended stimulus, in line with the sensory gain control hypothesis (Di Russo et al., 2001; Kim et al., 2007; Müller & Hillyard, 2000; Müller, Picton, et al., 1998; Müller, Teder-Sälejärvi, et al., 1998).

Activity in the alpha band, measurable across various sensory regions (Haegens et al., 2015) and animal species (Buzsáki et al., 2013), is another neural measure that was found to be robustly modulated by attentional demands. In visuospatial attention, it was repeatedly found that alpha-band power was lower contralateral to the attended as compared with the unattended side (Antonov et al., 2020; Bauer et al., 2014; Bollimunta et al., 2011; Capotosto et al., 2009; Foster et al., 2017; Gundlach et al., 2020; Händel et al., 2011; Keefe & Störmer, 2021; Kelly et al., 2006; Liu et al., 2022; Lobier et al., 2018; Samaha et al., 2016; Sauseng et al., 2005; Siegel et al., 2008; Slagter et al., 2016; Sokoliuk et al., 2019; Thut et al., 2006; Voytek et al., 2017; Worden et al., 2000; Zhigalov & Jensen, 2020). Thus, compared with SSVEPs, alpha-band modulations show the opposite pattern (see Keitel et al., 2019). Besides visual parieto-occipital alpha-band activity, alpha-band activity recorded over somato-motor regions (often labeled mu-alpha activity) also showed task-related modulations. Specifically, a decrease of motor/mu-alpha amplitudes seemed to be associated with task-related motor preparatory activity (Babiloni et al., 2008; Brinkman et al., 2014, 2016; Deiber et al., 2012; Maeder et al., 2012; McFarland et al., 2000; Pfurtscheller & Neuper, 1997; Stolk et al., 2019). These results led to the idea that alpha-band activity represents a marker or even a mechanism for altering neural processing by suppressing or facilitating neural excitability in phases of high or low alpha amplitude, respectively (Jensen & Mazaheri, 2010; Klimesch et al., 2007; Mathewson et al., 2011). Thus alpha-band activity may even be relevant for instantiating attentional selection by changing the neural excitability in neural populations representing the attended stimulus or feature (Foxe & Snyder, 2011; Peylo et al., 2021).

Importantly, in recent work, we and others found that SSVEP and visual parieto-occipital alpha-band amplitude modulations did not correlate (Antonov et al., 2020; Gundlach et al., 2020; Nuttall et al., 2022; Zhigalov & Jensen, 2020), pointing toward that these two measures represent different mechanisms by which attentional selection of stimulus processing is implemented, which is well in line with the idea that attention does not represent a unitary mechanism but may be implemented at different processing levels (Buschman, 2015; Hommel et al., 2019; Luo & Maunsell, 2015; Maunsell, 2015). While these previous studies investigated the relationship between SSVEP and visual alpha amplitudes, the behavioral consequences of their attentional modulation were not addressed and remained elusive. Whereas in the auditory domain some efforts were made to elucidate the relationship between alpha-band activity, sensory signal tracking, and behavior (Tune et al., 2021), in the visual domain, this relationship—to the best of our knowledge—has not been addressed yet.

In order to understand the potentially functional role of the two neural signals, in the present study, we examined the relationship between modulations of alpha-band activity, SSVEP amplitudes, and behavior in the context of spatial attention. Crucially, it is well known that attentional states fluctuate over time (Adam & deBettencourt, 2019; Esterman & Rothlein, 2019; Guilford, 1927; Rosenberg et al., 2015; Yamashita et al., 2021). Harnessing this idea, we assumed that variations in behavior related to fluctuations in attention should be associated with fluctuations in these neural measures, if they were relevant for attentional selection. For this purpose, we used a probabilistic Posner spatial cueing task (Posner, 1980), in which SSVEP evoking flickering disks were presented on each side of the screen. Spatial attention was manipulated by a cue in the form of an arrow indicating the most likely side on which a target will be presented prompting to shift attention to the left or right side of the screen in order to be able to rapidly respond to it by pressing a button (see below). The cue either validly, invalidly, or neutrally (pointing toward both stimuli) indicated the side of the upcoming target within the flickering stream of the disks. Participants had to discriminate the position of a transiently presented darker arc (upper or lower half and left or right side) at either the left or right stimulus in each trial by pressing one of four buttons on the keyboard that were assigned to the respective position within the two disks. We instructed our subjects to respond with the left hand to left targets and with the right hand to targets on the right side. With this design, we were able to concurrently measure and analyze different neural signals on a trial-by-trial basis that had been associated with selective spatial attention in a plethora of studies: stimulus evoked SSVEPs as a marker of early sensory gain, visual and motor alpha-band activity, and behavioral performance. We expected to replicate the typical behavioral pattern with faster responses for validly compared with invalidly cued targets (Carrasco, 2011; Posner, 1980).

As outlined above, relative to precue baseline, SSVEP amplitudes should increase for the cued as compared with the uncued stimulus, visual alpha-band amplitudes should be higher ipsilateral to the cued as compared with the uncued side, and motor alpha-band amplitudes should be reduced contralateral to the cued response hand. However, in one of our previous spatial cueing studies (Gundlach et al., 2020), we found considerable variations and variance in these measures across trials. Indeed, on a substantial number of trials, the lateralization pattern for visual alpha was either reversed, that is, more alpha at electrodes contralateral to the cued side, or the differences between the cortical hemispheres were negligible. Interestingly, the particular alpha-band pattern had no influence on SSVEP amplitudes (see also below). As the previous study focused on the relationship between these neural measures, by design we did not explicitly test the influence of these varying neural signals on reaction times. In the current study, designed to test this relationship, we hypothesized that faster responses to validly cued targets should be related to higher SSVEP amplitudes and/or lower visual alpha contralateral and higher visual alpha amplitudes ipsilateral to the cued side, as well as lower motor alpha amplitudes for the response hand in the cue–target interval. In the same manner, the question arises whether these neural fluctuations in the cue–target interval are predictive of a faster shift of the attentional spotlight to the other side, and thus, of faster reaction times for invalid trials, during which subjects do not “know” that a target will occur at the uncued side. In contrast to the pattern we hypothesized for trials with a valid cue, for trials with an invalid cue, one would expect the opposite pattern of alpha-band activity in the cue–target interval: lower alpha at ipsilateral and higher alpha amplitudes at contralateral electrodes relative to the cue for faster responses. The same should be true for SSVEP amplitudes (i.e., higher for the uncued stimulus at ipsilateral sites), if they contributed to the speed of responses. For motor alpha we would expect higher motor activation (i.e., lower amplitudes) for the motor cortex, ipsilateral to the cued side. To test the contributions of these EEG measures to the variance of reaction time, we employed Bayesian multilevel models based on single-trial data (see Section 2).

This Bayesian modeling approach helped to uncover the contribution of SSVEP, visual and motor alpha-band activity to behavioral responses. As attentional modulations have been shown to affect these measures independently (Antonov et al., 2020; Gundlach et al., 2020; Nuttall et al., 2022; Zhigalov & Jensen, 2020), one may assume that they also contribute to reaction time variability in different ways. Thus, SSVEP, visual, and motor alpha-band activity may reflect different stages in a hierarchical processing architecture. Based on previous suggestions, it might well be possible that alpha-band activity reflects a gating mechanism at later processing stages (Peylo et al., 2021; Zhigalov & Jensen, 2020) that operates independently from stimulus processing at early visual stages. Thus, alpha may be more closely related to the sensory readout relevant for behavioral performance, while SSVEPs might reflect early visual representation in a narrow sense not directly affecting the speed of the behavioral output.

In total, 28 participants (18 females, mean age 23.82 years; range 18 to 34) took part in the experiment. After being informed about the nature and the general aim of the experiment, all participants gave written informed consent. Participants either received class credits or were financially reimbursed (10 € per hour). The study was designed and conducted according to the Declaration of Helsinki and the local ethics committee (298/17-ek, Ethik-Kommission an der Medizinischen Fakultät der Universität Leipzig).

Visual stimulation was created with custom scripts using the Psychophysics toolbox 3 (Brainard, 1997; Kleiner et al., 2007), Matlab R2018 (The MathWorks, Natick, MA) running in an Ubuntu environment. Stimuli were presented via a PROPixx DLP LED projector (VPixx Technologies Inc., Canada) displaying images with a resolution of 960-by-540 pixels at a refresh rate of 480 Hz, projecting an image on a screen situated 120 cm in front of the subject.

The stimulus display was comparable with a display used in a previous study (Gundlach et al., 2020) and consisted of two rings with an outer and inner diameter of 5.4° and 3.2° of visual angle, presented 7.4° left and right, and 4.7° below a centrally presented gray fixation cross (luminance 420 cd/m²) with a height and width of 0.5° of visual angle on a dark gray background (luminance ~23 cd/m², see Fig. 1). Both rings flickered between background gray and bright gray (luminance 420 cd/m²) at 24 Hz for the left stimulus and at 21.818 Hz for the right stimulus.

After the introduction to the study aim and experimental procedure, the EEG was set up and participants were subsequently seated comfortably in an electromagnetically shielded and acoustically damped chamber. Participants then ran a few training trials to practice the precue and the main task (see below), got familiarized with the stimulus material, and practiced to blink only during the intertrial interval and to fixate the fixation cross throughout the entire trial.

In a given trial, during the precue period, participants attended to the fixation cross to discriminate transient increases (press “L” with the right index finger) or decreases (press “S” with the left index finger) of either the vertical or horizontal arm by 6.7 to 33.3% of size for a duration of 100 ms. In 20% of the trials, a single precue event was randomly presented in a time window between 500 ms after flicker onset and 400 ms before cue presentation. After a jittered precue interval of 1,500 to 2,000 ms, to avoid temporal expectation effects for the cue, the fixation cross changed color of the left and/or right arm to cue the to-be-attended side and remained present for the postcue period of 3,250 ms. Following this cue, a target stimulus on one of the rings was presented in each trial. The majority (90%) of targets were presented in a time window from 1,500 to 2,500 ms after cue onset (regular trials that entered the analysis), and 10% of the targets were presented in between 300 and 1,499 ms after the cue. As targets were presented throughout the trial and even close to the cue, we expected participants to shift their attention right after the cue and maintain attention at the cued side in a sustained manner anticipating the target stimulus. The presentation of the majority of target stimuli 1,500 ms after the cue (regular trials) allowed us to analyze a time window from 500 to 1,500 ms after the cue, for which spatial attentional shifts should have been accomplished and sustained attention established (Gundlach et al., 2020; Müller, Picton, et al., 1998; Müller, Teder-Sälejärvi, et al., 1998). Importantly, given that this time window was without any target stimulus in trials that entered the analysis, we were able to analyze SSVEP and alpha-band activity that was not distorted by an evoked potential elicited by a target event.

The target stimulus was a transient 150 ms long luminance decrease (to 270 cd/m²) of an arc with a width of 8° in either the left or right ring and participants had to discriminate and respond to whether this arc was presented in the upper or lower half of the left stimulus by pressing “Q” or “S” or in the upper and lower half of the right stimulus by pressing “P” or “L” on a standard QWERTZ keyboard. Importantly, subjects responded with the hand of the target position. In other words, when a target occurred on the right side, they responded with their right hand and vice versa to allow for the analysis of motor-related alpha-band amplitudes (see below). Target positions were evenly distributed between the left and right rings and upper and lower halves. Responses were registered in a time window between 200 and 1,200 ms following target onset. As in a typical Posner paradigm (Posner, 1980), the cue indicated the appearance of the target stimulus in a nondefinite but probabilistic way. From a total of 700 trials, 140 trials were neutral, that is, the cue pointed toward both stimuli. Of the remaining 560 trials in which the cue pointed either to the left or right stimulus, 420 valid trials (75%) correctly indicated the side of the target, while in 140 invalid trials (25%), the target was presented at the side opposite to the cue. Single trials were separated by an intertrial interval of 1,000 ms and were presented in 14 blocks with self-paced breaks in between.

EEG was measured from 64 Ag/AgCl electrodes mounted in an elastic cap with an ActiveTwo Amplifier (BioSemi, Amsterdam) at a sampling rate of 512 Hz with a low-pass filter of 104 Hz and stored for later offline analysis. Two electrodes were placed vertically above and below the right eye, and two were placed horizontally at the canthi of both eyes to allow measuring vertical and horizontal eye movements and blinks.

For offline data processing and analysis, the EEGLAB toolbox (Delorme & Makeig, 2004) and custom-developed Matlab scripts (The MathWorks, Natick, MA) were used. As a first step, continuous data were resampled to 256 Hz and epoched from -2,000 to 4,250 ms relative to cue onset. This relatively large time window was selected to allow for an analysis of electrophysiological single-trial data locked to the cue and target presentation, respectively (see below)^*. In the following, we identified artifacts and contaminated data (see below) in the time window 1,000 ms before and 3,250 ms after the cue to allow for an additional pretarget window analysis (see below).

First, linear trends were removed and blinks identified via an adaptive threshold procedure run for bipolarized vertical eye channels. Horizontal eye movements were identified as amplitudes exceeding 25 µV (representing eye movements of around 2° visual angle) in the bipolarized horizontal eye channels (blinks: average number of trials discarded per subject: M = 35.536; SD = 35.860; eye movements: average number of trials discarded per subject: M = 164.429; SD = 96.653). By implementing the “statistical control of artifacts in dense array EEG/MEG studies” (Junghöfer et al., 2000), single artifact-contaminated channels were identified based on statistical parameters and were spline interpolated. In case of more than 15 noisy electrode channels per trial, the entire trial was rejected. On average, 31.000 trials (SD = 26.369) were discarded and 3.722 channels per trial (SD = 0.899) interpolated. Overall, for each subject on average 260.786 valid (SD = 64.741), 90.857 invalid (SD = 19.357), and 88.000 neutral trials (SD = 23.051) entered the analysis. Preprocessed, artifact-free data were transformed to reference-free scalp current densities (SCDs) to represent data with more distinct local signal maxima (Ferree, 2006; Kayser, 2009; Kayser & Tenke, 2006; Müller et al., 2018).

The general aim of the current study was to examine the relationship between parieto-occipital visual alpha-band, motor alpha-band activity, SSVEPs, and behavior in a visuospatial attention paradigm after the cue and before the presentation of a target. As mentioned above, only those trials were considered for analysis, in which a target was presented at least 1,500 ms after the cue (called regular trials above), catch trials were all excluded.

Button presses within a time window between 200 and 1,200 ms after target onset were considered as responses in the postcue period. The same window applied to events at the fixation cross, before the cue. Correct, incorrect, or missed responses to increases or decreases were analyzed descriptively. For targets after the cue, responses were correct if participants correctly discriminated the position (upper vs. lower half and left vs. right side). The average percentage of correct responses for each cueing condition was tested with a conventional repeated-measures ANOVA comprising the factors VALIDITY and SIDE, using the afex package (Singmann et al., 2020) in R (R Core Team, 2016). For the ANOVA, degrees of freedom were Greenhouse–Geisser (GG) corrected in case of violation of the sphericity assumption. For this ANOVA model, post-hoc pairwise comparisons between the modeled marginal mean reaction times of the different levels of the factor VALIDITY were tested via the emmeans package (Lenth, 2023). p-Values were corrected for multiple comparisons via Holm correction (Holm, 1979) and Cohen’s d-values and respective 95% confidence intervals (95%-CI) will be reported, calculated with the population standard deviation defined as the root mean square value of the residual standard deviations of all factor levels via function eff_size of the emmeans package. We expected to replicate the common finding of faster responses for target events in validly compared with invalidly cued sides with neutral cues in between (Carrasco, 2011; Posner, 1980).

For examination of the relationship between electrophysiological signals and behavior, we were particularly interested in reaction times. As a first step, mirroring the analysis of the percentage of correct response, condition-averaged reaction times were tested with a conventional repeated-measures ANOVA comprising the factors of CUE_VALIDITY and SIDE and Holm-corrected post-hoc pairwise comparisons of the marginal means.

As a first omnibus test, we tested whether we can replicate the well-known finding of an SSVEP amplitude increase, relative to the precue period, for the to-be-attended compared with the to-be-ignored side/ring (Gundlach et al., 2020; Müller & Hillyard, 2000). But of particular interest was whether or not single-trial SSVEP amplitudes and reaction times are related to each other (with higher amplitudes found in trials with lower reaction times). In order to extract single-trial SSVEPs, we used a recently developed spatial filtering approach by which the SSVEP signal is extracted from an optimally weighted sum of all EEG sensors based on Rhythmic Entrainment Source Separation (RESS) (Cohen & Gulbinaite, 2017). This procedure derives spatial RESS filters with an optimized representation of the (SSVEP-) signal over noise, individually determined for each subject and frequency of interest. For the implementation of this procedure, one spatial filter for each signal was derived for which the signal component was centered at the respective SSVEP frequency (24 and 21.818 Hz) with a bandwidth of 0.5 Hz, while noise was defined as spectral components ±1.5 Hz apart from the signal with a bandwidth of 1.5 Hz. For a more numerically stable estimation of the spatial filter estimation, the covariance matrix was regularized by adding 1% of the mean of its eigenvalues to its diagonal (Gulbinaite et al., 2019; Schettino et al., 2020). Individually filter-projected data (see Figure 3 for a topographical representation of the filter weights) were then used for extracting SSVEP amplitudes.

In a first analysis, we were interested in the general pattern of attentional modulation for trials of the different cueing conditions (valid, neutral, invalid). For this purpose, RESS-filter-projected single-trial data for each SSVEP and experimental condition were FFT transformed (zero-padding to 16,384 datapoints) from a precue (-1,000 to 0 ms before the cue) and postcue time window (500 to 1,500 ms after the cue). SSVEP amplitudes for each stimulus, condition, trial, and time window were then derived by averaging the amplitude values in a range of ±0.1 Hz around the respective SSVEP frequency and averaging across trials of each condition. As exact SSVEP peak frequencies may slightly differ between participants and SSVEP, amplitude changes within the FFT time window may lead to amplitude changes of the SSVEP frequency sidebands (Cohen & Gulbinaite, 2017), by zero-padding the data and summing amplitudes across a frequency range, we made sure to capture SSVEP amplitudes for each participant. As previously reported (Gundlach et al., 2020), precue SSVEP signals are seen to depict the representation of the unattended flickering rings in early visual areas (i.e., while attention was allocated to the fixation cross). This baseline measure allows to measure cortical facilitation through attention after the cue. To this end, percentual changes of SSVEP amplitudes from the pre- to postcue time window were calculated. These values were then collapsed across stimuli and conditions in a way that they represent pre- to postcue SSVEP amplitude modulations for the cued, uncued, or neutrally cued stimulus. Potential differences between these SSVEP modulations were then tested via a one-way repeated measures ANOVA and Holm-corrected post-hoc pairwise comparisons between modeled marginal mean SSVEP modulations for the cued, uncued, and neutrally cued rings, as described above. In addition, for each factor level, marginal mean SSVEP modulations were tested against zero to estimate pre- to postcue modulations per se (Holm corrected for multiple comparisons). Cohen’s d-values were calculated with the population standard deviation defined as the root mean square value of the residual standard deviations of all factor levels.

For the analysis of visual and motor-related alpha-band activity for the different attentional cueing conditions, the analysis rationale was similar to the one described above. The basis of this analysis formed single-trial data, for which all but the SSVEP-RESS components were back projected (Cohen & Gulbinaite, 2017). To minimize the potential contamination of alpha-band activity by subharmonics and intermodulation frequencies of the SSVEP signals (Labecki et al., 2016; Zemon & Ratliff, 1982, 1984) and allow for an independent analysis of SSVEP and alpha-band activity, the RESS component was excluded before back-projection. Single-electrode and single-trial data were FFT transformed (zero-padding to 16,384 datapoints) for the same analysis windows as for SSVEPs. Visual alpha-band amplitudes (averaged across 8 to 12 Hz) were extracted (and averaged) for each experimental condition and time window from an electrode cluster contralateral to the left (P6, P8, P10, PO4, PO8, O2, I2) and right stimuli (P5, P7, P9, PO3, PO7, O1, I1) (electrode clusters identical to Gundlach et al., 2020). As above, percentual pre- to postcue modulations were calculated and then averaged to depict visual alpha-band modulations contralateral to the cued, uncued, or neutrally cued stimulus. Of note, in previous literature cue-related modulations were often found in both hemispheres and it is still under debate whether signals are related to modulations contra- and ipsilateral to the cued side or signals contralateral to the cued and the uncued side and thus potentially related to enhanced target vs. attenuated processing of the nontarget stimulus (Capilla et al., 2014; Orf et al., 2023). In other words, while we are referring to alpha-band activity contralateral to the cued or uncued stimulus, the signal is per se ambiguous, as alpha-band activity contralateral to the cued stimulus can also be labeled as activity ipsilateral to the uncued stimulus for instance. Motor preparatory processes were analyzed in the same manner by extracting motor alpha-band activity (averaged across 9 to 14 Hz) from two electrodes contralateral to the left (C3, CP3) and right (C4, CP4) response hands, respectively. Pre- to postcue motor alpha-band modulations contralateral to the cued, uncued, or neutrally cued hand were calculated the same way as above. As for the SSVEP amplitude modulations, attention-related differences in visual and motor alpha-band amplitude modulations were tested via one-way repeated measures ANOVAs with additional Holm-corrected post hoc pairwise comparisons based on the estimated marginal means of the ANOVA models. As an important critical test, we analyzed and tested the phase coherence between electrode sites at lateralized parieto-occipital locations (see above) and C3/C4 and CP3/CP4 to make sure that alpha-band signals measured at these sites have distinct neural generators and are not mainly based on either volume conduction of the same signal or represent a single dipole projecting to central and visual leads (see Supplementary Material). Results revealed a significant difference in phase between these sites (i.e., nonzero phase lag), suggesting that alpha-band activity recorded from motor as well as parieto-occipital channels had different neural generators.

In the next step, we tested to what extent trial-by-trial fluctuations of SSVEP and alpha-band amplitudes under sustained attention were predictive to reaction times. Electrophysiological measures and reaction times were extracted on the single-trial level, as described above. We used absolute amplitude values of the postcue period from 500 to 1,500 ms after the cue instead of modulation values used in the ANOVA models above to allow for an analysis of the postcue signals unbiased by precue amplitude signals and noise.

Bayesian multilevel models were fit to predict participants’ single-trial response times by different combinations and interactions of potentially predicting factors. These factors were validity of the attentional cue (Val.), SSVEP amplitude contralateral to side of target (SSVEP), visual alpha-band amplitudes contralateral to side of target (V.alpha_contra) and ipsilateral to target side (V.alpha_ipsi), as well as motor alpha-band activity contralateral to the side of target which was identical to the response hand (M.alpha_contra). These Bayesian multilevel models were built and fit using the package brms (Bürkner, 2017) running in R (R Core Team, 2016). Right skewed reaction time distributions were modeled as being represented by a shifted log-normal distribution (Cousineau et al., 2004; Wagenmakers & Brown, 2007), see Equation 1.

with parameters $μ$ or “difficulty” specifying the mean of the log-normal distribution, $σ$ or “scale” representing the standard deviation of the log-normal distribution and $θ$ or “shift” specifying the earliest possible response. The predictors and their interactions were modeled to affect the parameter $μ$ while $σ$ and $θ$ were set to be fixed and fit to the data. Separately for our different neurophysiological predictors, the most complex model was fit assuming an interactive relationship between the factor cue validity and the respective neurophysiological measure (i.e., SSVEP, visual alpha-band and motor alpha-band amplitude). The effects of the factor cue validity were allowed to vary across the grouping factor “subjects” (see Table 1 for the notation of the different models). To increase the model convergence by running brms, continuous predictors are internally centered. Coefficient estimates are then, however, reported on the original scale, facilitating the interpretation of the coefficients and effects.

Model parameters were fit in four chains, each with 2,000 iterations including 1,000 warmup draws, adding to a total of 4,000 postwarmup draws. Model summaries were then checked for model convergence of model parameters (all parameters converged for all models presented here: Rhat values in range of 1 to 1.02). Separately for the different model families, the marginal effects of the different predictors (e.g., the slope for the modeled relationship between SSVEP amplitudes and reaction times) were then extracted and interpreted. This was done, first, by using the model with its fitted model parameters to create posterior distributions of predicted values (i.e., reaction times) that also capture the uncertainty of the model parameters across the model fits of each brms model. For this purpose, posterior predicted values that were expected for the levels of the predicting factors in the model were extracted separately for all the model draws of the respective fitted model (via function epred_draws of the tidybayes package as a vignette of the brms::posterior_epred function). The emmeans package then allowed to extract the posterior distribution of predicted median RTs separately for the different levels of the predictors of interest and their credible intervals, which were then taken to plot and examine the marginal effect of the factor cue validity on reaction times as captured in the model. Pairwise contrasts between factor levels were calculated to extract the marginal effect sizes as well as their 95% highest posterior density (HPD) intervals as a measure of model uncertainty, and to evaluate how much faster participants responded after valid as compared with invalid cues for instance. For the continuous neurophysiological predictors, the modeled slopes (and their 95% HPD intervals) were extracted via the function emmeans::emtrends, to examine how the different single-trial predictors affected single-trial reaction times. The 95% HPD was used to estimate the consistency of the effect of a certain predictor across the model draws. If the slope of the predicted relationship between one of the neurophysiological measures and reaction times was not consistently different from zero across the model draws (i.e., the 95% HPD contained the zero), we interpreted the effect of this predictor on reaction times as not consistent/substantial. This allowed us to estimate whether any of the single-trial neurophysiological measures was indeed a relevant factor for predicting reaction times in our experiment.

Additionally, to estimate how much variance was explained in the model and the respective predictors, Bayesian $R2$ values for each model were calculated to estimate the amount of variance explained by the predictions of each model relative to the predictions plus variance of the errors (Gelman et al., 2019). Predictor-specific effects were evaluated by additionally fitting two basic models and extract Bayesian $R2$ values for these: one intercept model, for which reaction time measures were modeled to differ between subjects (RT ~ 1 + (sub)) and a model including the cue validity as the only predictor (RT ~ Val. + (Val. | sub)). The latter model would correspond to the classical findings (see Posner, 1980) of longer reaction times for invalidly and shorter reaction times for validly cued targets associated with typical spatial cueing paradigms.

In addition, the same modeling and evaluation pipeline was implemented for SSVEPs, visual alpha, and motor alpha data derived from a different time window: the pretarget time window 1,000 ms before target presentation (-1,000 to 0 ms). As the target stimulus was presented at random time points after the presentation of the cue, this additional analysis allowed to estimate the impact of differences in the amplitude levels of the electrophysiological signals measured directly before target presentation in contrast to the general sustained postcue modulation described above and may have picked up behaviorally relevant modulations on a smaller time scale, that is, right before stimulus processing.

During the precue task, participants responded correctly to 81.862% (SD = 11.264), missed 7.832% (SD = 7.280), and responded incorrectly to 10.306% (SD = 9.391) of the events at the fixation cross. Therefore, hit rates suggested that participants were compliant with the precue task during which attention was focused to central fixation.

We replicated the typical pattern: On average, participants responded faster and with the highest rate of correct responses following valid cues and much slower and with the lowest hit rate following invalid cues (see Table 2 and Fig. 2). The repeated measures ANOVA for reaction times revealed a main effect for the factor of VALIDITY (F(1.628,43.943) = 20.160, p < .001, η_g² = 0.031). Holm-adjusted post hoc pairwise comparisons of the marginal means revealed lower reaction times for validly as compared with invalidly cued targets (t(54) = 6.254, p < .001, d = 0.422, 95%-CI_d = [0.245 0.600]), significant differences between valid and neutral (t(54) = 2.173, p = .034, d = 0.147, 95%-CI_d = [0.006 0.288]), and neutral and invalid reaction times (t(27) = 4.081, p < .001, d = 0.276, 95%-CI_d = [0.121 0.431]). Presentation side of the target did not affect reaction times as neither the main effect of SIDE (F(1,27) = 0.027, p = .870, η_g² < 0.001) nor the interaction VALIDITY X SIDE was significant (F(1.596,43.099) = 3.059, p = .068, η_g² = 0.003). Similarly, the analysis of correct responses revealed a main effect of VALIDITY (F(1.374,37.087) = 8.919, p = .002, η_g² = 0.040). Again, we found a significant difference in the marginal means between validly and invalidly cued (t(54) = 3.866, p < .001, d = -0.451, 95%-CI_d = [-0.715 -0.181]) and neutrally and invalidly cued targets (t(54) = 3.406, p = .003, d = -0.397, 95%-CI_d = [-0.655 -0.139]), while the difference between validly and neutrally cued targets was not significant (t(54) = 0.647, p = .647, d = -0.054, 95%-CI_d = [-0.288 0.181]). Again, there was no main effect of SIDE (F(1,27) = 0.087, p = .770, η_g² < 0.001) and no significant interaction VALIDITY X SIDE (F(1.363,36.796) = 0.18, p = .770, η_g² = 0.001).

RESS components exhibited a lateralized topographical distribution for each of the two SSVEPs, and Grand Mean FFT amplitude spectra showed distinct amplitude peaks for the respective frequencies (see Fig. 3A).

The analysis of SSVEP amplitude modulations (see Fig. 3B) revealed differences between cueing conditions (see Fig. 3) as revealed by a main effect of the factor cue in a repeated measures ANOVA (F(1.944,52.475) = 4.829, p = .013, η_g² = 0.036). Planned Holm-corrected comparisons of the marginal means revealed significantly positive SSVEP amplitude modulations for the cued side (M = 4.558 %, SE = 1.794, t(41.436) = 2.541, p = .045, d = 0.480, 95%-CI_d = [0.077 0.884]) and the neutral condition (M = 4.426 %, SE = 1.794, t(41.436) = 2.468, p = .045, d = 0.466, 95%-CI_d = [0.064 0.869]), while there was no SSVEP modulation for the uncued side relative to the precue baseline (M = 0.699 %, SE = 1.794, t(41.436) = 0.390, p = .699, d = 0.074, 95%-CI_d = [-0.308 0.456]). While SSVEP amplitude modulations for neutrally cued and cued sides did not differ significantly (M = 0.132 %, SE = 1.410, t(54) = 0.093, p = .926, d = -0.014, 95%-CI_d = [-0.312 0.284]), they both differed from the modulation for the uncued side (uncued vs. cued: M = 3.859 %, SE = 1.410, t(54) = 2.737, p = .025, d = -0.407, 95%-CI_d = [-0.725 -0.089], uncued vs. neutrally cued: M = 3.728 %, SE = 1.410, t(54) = 2.644, p = .025, d = 0.393, 95%-CI_d = [-0.709 -0.076]).

Visual alpha-band activity recorded at two lateralized occipital electrode clusters revealed cue-related amplitude modulations in a time window of 500 to 1,500 ms after the cue relative to -1,000 to 0 ms before the cue (see Fig. 4).

For visual alpha-band activity, the repeated measures ANOVA revealed a main effect of the factor cue (F(1.421,38.374) = 4.697, p = .025, η_g² = 0.039). Holm-corrected comparisons of the marginal means revealed significantly lower visual alpha amplitude modulations contralateral to the cued as compared with the uncued stimulus (M = -3.467 %, SE = 1.777, t(54) = 2.945, p = .014, d = -0.468, 95%-CI_d = [-0.811 -0.125]), while neither differences between modulations contralateral to the neutral and uncued side (M = 2.601 %, SE = 1.777, t(54) = 2.209, p = .063, d = 0.351, 95%-CI_d = [0.018 0.683]) nor contralateral to the cued and neutral side (M = -0.867 %, SE = 1.777, t(54) = 0.736, p = .465, d = -0.107, 95%-CI_d = [-0.437 0.437203]) were significant. In addition, planned contrasts revealed pre- to postcue modulations not to be different from zero (contralateral to the cued side: M = -2.297 %, SE = 1.401, t(44.085) = 1.639, p = .325, d = -0.310, 95%-CI_d = [-0.700 -0.080]; contralateral to neutrally cued side: M = -1.431 %, SE = 1.401, t(44.085) = 1.021, p = .626, d = -0.193, 95%-CI_d = [-0.577 0.192]; contralateral to uncued side: M = 1.170 %, SE = 1.401, t(44.085) = 0.835, p = .626, d = 0.158, 95%-CI_d = [-0.226 0.541]).

Motor alpha-band activity recorded at two lateralized electrodes over motor regions showed cue-related modulations (see Fig. 5) as revealed by a main effect of the factor cue in the repeated measures ANOVA (F(1.981,53.497) = 9.892, p < .001, η_g² = 0.112).

While motor alpha-band activity contralateral to uncued as well as neutrally cued targets was not modulated in amplitude after the cue, as evaluated by Holm-corrected planned comparisons of the marginal means of the ANOVA model (contralateral to uncued stimulus: M = 0.783 %, SE = 1.374, t(55.023) = 0.570, p = .999, d = 0.108, 95%-CI_d = [-0.272 0.488]; contralateral to neutrally cued stimulus: M = -0.761 %, SE = 1.374, t(55.023) = 0.554, p = .999, d = -0.105, 95%-CI_d = [-0.484 0.275]), motor alpha-band amplitude contralateral to the cued target decreased after the cue (M = -5.185 %, SE = 1.374, t(55.023) = 3.775, p = .001, d = 0.713, 95%-CI_d = [-1.139 -0.288]). Accordingly, differences between motor alpha modulations contralateral to cued and uncued (M = 5.968 %, SE = 1.393, t(54) = 4.285, p < .001, d = 0.821, 95%-CI_d = [0.376 1.266]) as well as cued and neutrally cued target (M = 4.424 %, SE = 1.393, t(54) = 3.177, p = .005, d = 0.609, 95%-CI_d = [0.190 1.027]) were significantly different, while the differences contralateral to neutrally cued and uncued target did not differ (M = 1.543 %, SE = 1.393, t(54) = 1.108, p = .273, d = 0.212, 95%-CI_d = [-0.176 0.601]).

Unsurprisingly, across all models, the marginal effects of the factor cue validity were comparable: models predicted reaction times to be shortest in trials with a valid cue and longest in trials with an invalid cue with reaction times for neutrally cued targets falling in between (see Fig. 6). In addition, for all pairwise comparisons, the 95% highest posterior density (HPD) of the posterior distributions of all model draws were different from zero and, thus, revealed reaction time differences between all cue validity levels (see Fig. 6 and Table 3).

Figure 7 depicts the effects of the respective electrophysiological measure upon reaction times. Neither for SSVEPs nor for motor alpha-band activity there was a consistent relationship captured in the respective models across model draws. Here the 95% HPD intervals of posterior distributions of the predicted slopes included zero, suggesting that there is no substantial and consistent positive (or negative) relationship between these neural measures and reaction times captured across model draws (see Table 4).

For postcue visual alpha-band amplitudes at electrodes contra- and ipsilateral to the stimulus with the target, consistent relationships across the model draws were found. Visual alpha-band amplitudes recorded contralateral to the upcoming target were positively correlated with reaction times to target stimuli in neutral and valid, but not invalid trials. The lower the contralateral alpha-amplitudes in the cue–target interval after the valid and neutral cue, the faster the participants responded. For visual ipsilateral alpha-band activity, the relation was the opposite for trials with a valid cue. Here higher postcue amplitudes were related to faster responses. For trials with invalid or neutral cues, no consistent relationship was found.

How well does the visual alpha-band model predict single-trial reaction times? For this model, a Bayesian $R2$ value of 0.1917 was extracted. In comparison, a model containing only cue validity as a predictor (and no neural measure), that is, the typical Posner effect, yielded a Bayesian $R2$ value of 0.1892, and a mere intercept model for which RT measures were modeled to differ only across participants yielded a Bayesian $R2$ value of 0.1801. While the most complex model indeed explained the RT variance the best, most of the actual variance in the RT data (80.83 %) is not explained at all.

In a second step, we looked into the relationship between neurophysiological measures right before the presentation of a target and single-trial reaction times for correct responses. This was based on the idea that potential fluctuations of neurophysiological measures right before the stimulus presentation may be more relevant for the processing of and reaction to the behaviorally relevant target.

In these models, again, the same general relationship between cue validity type and response times was captured, mirroring the results of the postcue models: participants responded faster for validly cued and much slower for invalidly cued targets with neutrally cued targets falling in between. As above, this was revealed by the fact that for all pairwise comparisons between cue validity levels, the 95% HPD of the posterior distributions of the predicted reaction time differences did not include zero and thus revealed substantial reaction time differences (see Table 5).

Figure 8 and Table 6 show the results of the predicted relationship between the respective electrophysiological measures and reaction times in this window before target onset. We found some similarities to the findings for the postcue neural measures. Again, single-trial SSVEP amplitudes did not predict reaction times across all model draws in a consistent manner. As for the postcue time window, visual alpha-band amplitudes were related to response times. However, in contrast to the results of the postcue time window, effects were only found for amplitudes at sites contralateral to the target side in this time window: For trials with a valid and neutral (but not invalid) cue, lower pretarget contralateral amplitudes were associated with faster response times. Pretarget ipsilateral visual alpha amplitudes were not associated with the actual response time. In contrast to the earlier postcue time window, motor alpha-band amplitudes recorded right before the target presentation were also a consistent predictor of actual reaction times. As for visual alpha-band activity, lower pretarget motor alpha-band amplitudes contralateral to the response hand were associated with faster responses for trials with valid or neutral but not invalid cues.

We examined the interaction between attention, different electrophysiological markers of visuospatial attention, and reaction times in a spatial probabilistic cueing task. As in previous work and as expected, we found that SSVEP, visual alpha-band, and motor alpha-band activity were modulated following a spatial attentional cue: After the onset of the cue, SSVEP amplitudes were higher for the cued and neutrally cued as compared with the uncued stimulus. Visual alpha-band amplitudes were lower contralateral to the cued as compared with uncued side, and motor alpha-band activity was lower contralateral to the cued stimulus-response hand as compared with the neutrally cued and uncued response hand. The validity of the cue affected the correctness of responses as well as reaction times to targets, with participants responding more accurately and faster to validly cued targets compared with responses in the other conditions. Beyond this cue validity-dependent modulation of response times, we found that trial-by-trial fluctuations of alpha-band activity seem to predict reaction times in the valid and neutral cue condition. In the cue–target interval, lower visual alpha-band amplitudes contralateral to the stimulus with the target were associated with shorter response times after a valid or neutral cue. Higher ipsilateral visual alpha amplitudes predicted faster reaction times for valid cues in this interval. Ipsilateral alpha was now no longer predictive of reaction times in the window before target onset, but lower visual alpha-band amplitudes at contralateral electrodes continued to be associated with faster reaction times in that time window. In addition, lower motor alpha amplitudes contralateral to the response hand were associated with faster reaction times in this window following valid and neutral cues.

Our findings are well in line with previous reports on behavioral effects in spatial cueing paradigms: for validly cued events, behavioral advantages were usually found, such as faster reaction times, fewer errors, and more correct responses, while for invalidly cued events, behavioral costs were prominent, leading to slower and less accurate responses (see Carrasco, 2011; Posner, 1980). In seminal studies to investigate neural responses in such spatial cueing designs, it was commonly found that early components of the visual event-related potential (VEP) such as the N1 or P1 component were modulated in amplitude with attention, resulting in higher amplitudes when a stimulus occurred at the cued, compared with when it occurred at the uncued side (Hillyard & Anllo-Vento, 1998; Hillyard et al., 1998; Luck, 1995; Luck et al., 1997, 2000; Marzecová et al., 2018; Slagter et al., 2016). These amplitude differences were interpreted as a top–down guided sensory gain control mechanism of attention (Hillyard et al., 1998). SSVEP amplitude modulations had been linked to changes in sensory gain as well (Di Russo et al., 2001; Kim et al., 2007; Müller & Hillyard, 2000), given that SSVEP amplitudes are enhanced for an attended stimulus compared with when this stimulus needed to be ignored (cf. Gundlach et al., 2020; Müller & Hillyard, 2000; Müller, Teder-Sälejärvi, et al., 1998; Walter et al., 2014). Interestingly, amplitude decreases for the unattended stimulus relative to a precue baseline were usually not found (Gundlach et al., 2020; Müller & Hillyard, 2000; Müller, Teder-Sälejärvi, et al., 1998; Walter et al., 2014), indicating that the to-be-ignored stimulus was not further suppressed by spatial attention.

We replicated this pattern in the present study. Of interest is the fact that we found no differences in SSVEP amplitude gain for trials in which the cue pointed to one side compared with a neutral cue, pointing to both sides. In case of a split of attentional resources to both sides, one would expect lower amplitudes in the neutral cue condition compared with when all resources were shifted to one side of the screen, if attention relied on one common resource pool. A possible explanation for the fact, that this pattern was not found, offers the so-called different hemifield advantage that hypothesizes independent pools of attentional resources for both cortical hemispheres (Alvarez & Cavanagh, 2005; Mayo & Maunsell, 2016; Störmer et al., 2014; Walter et al., 2014). In this scenario, attentional resources are not shifted between cortical hemispheres (because each hemisphere has its own), resulting in similar SSVEP amplitudes in situations when one side needs to be attended or both sides are attended simultaneously.

However, although the SSVEP amplitude pattern for time domain averaged SSVEPs obviously represents sensory gain amplification initiated by the cue, single-trial SSVEP amplitude fluctuations were not related to reaction times in any condition. Of note, this finding was mirrored in an exploratory control analysis (see Supplementary Material). Here visual alpha-band amplitudes contra and ipsilateral to the side with the upcoming target and SSVEP amplitudes contralateral to the stimulus with the target were modeled together, mirroring the same relation for visual alpha-band amplitudes as well as no consistent relationship between SSVEP amplitudes and reaction times. As this finding was unexpected and points to a missing effect of SSVEPs, we have no clear explanation why this was the case and may offer only post hoc hypotheses. One reason might be related to the SNR in single-trial SSVEP extraction. Despite the fact that we employed denoising techniques such as RESS to enhance SSVEP-SNR, specifically suited for single-trial analysis regimes (Cohen & Gulbinaite, 2017), it does not prevent nonphase-locked signals from ongoing neural oscillatory activity in the present SSVEP frequency range (such as beta-band activity) to actually contribute to the SSVEP signals. In a common trial-averaging approach to study SSVEP effects, these possible nonphase-locked contributions are averaged out (or are at least significantly reduced) when time domain signals are averaged across trials first, before signals are transferred into frequency domain. Thus, trial-averaged SSVEPs more robustly reflect sensory gain mechanisms of pure stimulus evoked activity. Single-trial frequency domain data, on the other hand, may be more contaminated by nonstimulus evoked activity, thus, potentially masking the relationship between sensory gain modulations and behavior. Yet, these single-trial SSVEP amplitudes do depict the general expected cue-related attentional modulation. One may thus hypothesize that SSVEPs may not necessarily and only under certain circumstances be predictive of behavior. Attention-related behavioral benefits could arise from the implementation of attention at (hierarchically) different stages (Buschman, 2015; Hommel et al., 2019; Luo & Maunsell, 2015; Maunsell, 2015). Signatures of attention selection at different stages may be unrelated (see for the missing relationship between SSVEPs and alpha-band activity: Antonov et al., 2020; Gundlach et al., 2020; Nuttall et al., 2022; Zhigalov & Jensen, 2020), may depict different modulation profiles for different cueing conditions as in our data (i.e., comparing the effect of cue validity on SSVEPs, and visual or motor alpha-band amplitudes), and may be differently relevant for single-trial behavioral responses (Wyart et al., 2015).

Parieto-occipital alpha-band activity in the cue–target interval showed the expected attention-related modulatory pattern as well: following the attentional cue, parieto-occipital alpha-band activity was lateralized and was lower contralateral to the cued as compared with the uncued side with amplitudes in neutrally cued trials falling somewhere in between. Such a pattern has long been reported in the visual domain (Antonov et al., 2020; Bauer et al., 2014; Bollimunta et al., 2011; Capotosto et al., 2009; Foster et al., 2017; Gould et al., 2011; Gundlach et al., 2020; Händel et al., 2011; Keefe & Störmer, 2021; Kelly et al., 2006; Liu et al., 2022; Lobier et al., 2018; Samaha et al., 2016; Sauseng et al., 2005; Siegel et al., 2008; Slagter et al., 2016; Sokoliuk et al., 2019; Thut et al., 2006; Voytek et al., 2017; Worden et al., 2000; Zhigalov & Jensen, 2020), and also in the somatosensory domain (Anderson & Ding, 2011; Forschack et al., 2017; Haegens, Händel, et al., 2011; Haegens, Nácher, et al., 2011; Haegens et al., 2012; Jones et al., 2010; van Ede et al., 2011; Wiesman & Wilson, 2020) and auditory domain (Boudewyn & Carter, 2018; Deng et al., 2019, 2020; Frey et al., 2014; Strauß et al., 2014; Tune et al., 2018, 2021; Wöstmann et al., 2016, 2019). These findings led to the idea that alpha-band activity represents a neural marker for altering the neural information flow, with some accounts focusing on a potential functional (even active) inhibitory role in stimulus processing and attention (Foxe & Snyder, 2011; Jensen & Mazaheri, 2010; Klimesch et al., 2007; Mathewson et al., 2011). However, a recent review that evaluated the evidence in favor of a general (active or causal) inhibitory role of alpha-band activity during attention found mixed results for the relationship between prestimulus alpha amplitudes and changes in stimulus-evoked neural activity (Morrow et al., 2023).

While our results for valid and, to some extent for neutral trials, in the cue–target interval can be seen as supportive for the proposed hypothesis that alpha-band modulations are predictive for behavioral responses during attention, for invalid trials this was not the case. As we outlined in the Introduction section, if alpha-band activity was generally relevant for attentional selection, it should be predictive of fast responses in the invalid condition as well. Here, one would assume that in invalid trials with relatively faster reaction times to the uncued target stimulus, the alpha-lateralization pattern should be reversed during the cue–target interval as compared with the prototypical pattern. Such trial-by-trial reversals of the alpha lateralization for the same cue are indeed measurable and can be a consequence of fluctuations, as we have found in our previous study (Gundlach et al., 2020). Other studies also reported of fluctuations in attention across time (Adam & deBettencourt, 2019; Esterman & Rothlein, 2019; Guilford, 1927; Rosenberg et al., 2015; Yamashita et al., 2021) and the actual alternation of attention between stimuli (Fiebelkorn & Kastner, 2019; Helfrich et al., 2018; VanRullen, 2018). If alpha-band amplitudes played a central role for attention, one would expect that alpha-band fluctuations depict attentional fluctuations. However, neither cue–target nor pretarget interval alpha fluctuations predicted reaction times for invalid trials. This was also true for motor alpha-band activity in these trials. Thus, alpha-band amplitudes seem unrelated to behaviorally relevant fluctuations of attention.

Interestingly, in a probabilistic cueing design, Händel et al. (2011) found that alpha lateralization in the cue–target interval was, in contrast to our findings, correlated with behavioral responses for trials with an invalid cue, but—surprisingly—not for valid trials. They explained their findings with competitive interactions that needed to be resolved for the invalid but not for the valid condition. While this interpretation is open for discussion, it is important to know that they used perceptual threshold measures as behavioral responses that may account for the differing results of their and the current study. In their study, subjects needed to indicate the direction of coherent motion events after 2 seconds interval between the presentation of a target display, consisting of two potentially task-relevant and competing stimuli, and the postcued, required response to one of the previously presented stimuli. While alpha-band activity was lateralized in the cue–target interval, there was no lateralization in the target–response time window. Obviously, our discrimination task differs, as we used immediate reaction times to a single target as the behavioral measure without any delay between target and response and without any actual distractors being presented at all. In addition, results stemmed from the correlation between alpha-band amplitudes and behavior across subjects and not from trial-by-trial variations of alpha-band amplitudes, as analyzed in our study. Whether the task differences are responsible for the contradictory results between the two studies is an open question, but the idea of a general active inhibitory role of alpha-band activity in stimulus processing may not be warranted.

Motor alpha-band amplitudes recorded contralateral to the cued side, that is, the response hand, were decreased compared with ipsilateral central electrodes. In trials with a neutral cue, there were no substantial modulation of motor alpha-band activity. For valid and neutral trials, these lower amplitudes measured in a pretarget interval were predictive of faster reaction times, implying that low motor alpha amplitudes may be associated with facilitated and faster responses. Recent work found Transcranial-Magnetic-Stimulation (TMS) evoked motor-evoked-potential (MEP) pulses to be modulated by the amplitude (and phase) of ongoing motor alpha/mu-alpha-band activity (Hussain et al., 2018; Karabanov et al., 2021; Ogata et al., 2019; Thies et al., 2018; but see Zrenner et al., 2022). Lower amplitude levels for instance led to larger MEPs. TMS protocols illustrating inhibitory neural processes found measures not to be modulated by motor alpha-band activity (Bergmann et al., 2019). These findings point toward the facilitatory role of alpha-band activity in the motor cortex representing the general excitatory state of the system. Our findings are in line with this idea of neural excitability, and the decreased alpha-band levels contralateral to the cued side may point toward preparatory processes in the motor cortex associated with the cued response.

The interpretation of motor alpha-band activity as a neural signature of neural excitability in motor cortex to prepare motor execution is well in line with recent suggestions that relative alpha-band fluctuations seem to be associated with general changes in excitability of neural populations: higher alpha-band amplitudes seem to index lower excitability levels (Iemi et al., 2022; Samaha et al., 2020; Van Diepen et al., 2019). In our experiment, these behaviorally relevant excitability changes depicted by alpha-band activity were relevant for trials with a valid or a neutral cue only, and were not found for trials with an invalid cue. Does this mean that alpha-band activity is not related to changes in attentional dynamics, attentional reallocation processes, or resource allocation at cued and uncued locations (Carrasco, 2011; Eckstein et al., 2002; Macaluso & Doricchi, 2013; Posner, 1980)? This is hard to answer solely based on our results. During the cue–target interval, each subjects’ best strategy for fast responses was to shift attention to the cued side, as the target appeared three times more often at the cued side, or split attention to both sides following a neutral cue.

If alpha-band activity was relevant for instantiating the behaviorally relevant attentional selection, one could assume alpha-band fluctuations contralateral to a stimulus to depict fluctuations in the attentional selection of this stimulus. Irrespective of the actual cue validity, alpha-band amplitudes should depict the graded attentional selection of the stimulus and lower amplitudes be indicative of faster responses. This pattern was, however, not found for invalid trials. Specifically, for invalid trials, the presentation of the target on the uncued side would require a reallocation of attention and shifting of attentional resources in order to respond to the target. In invalid trials, alpha-band amplitudes seem not to index the initial attentional selection of the uncued side or the fidelity of the attentional reallocation process. Further, in these invalid trials, the attention-related modulation of cortical excitability could only occur after the presentation of the target and the consequent shift of attention. Therefore, alpha-band amplitude fluctuations seem to be behaviorally relevant only when no spatial attentional reallocation is required and may index behaviorally excitability changes only as a consequence of top–down driven attentional selection and resource allocation. As seen in previous work, the behavioral relevance of alpha-band modulations may be related to the confidence in selecting the cued location (Pilipenko & Samaha, 2024; Samaha et al., 2017, 2020), This idea is supported by the fact that after the presentation of the target and after the reallocation of attention in invalid trials, we found for all targets in all conditions a similar behaviorally relevant motor–alpha desynchronization contralateral to the response hand, with lower amplitude values indicating higher excitability of the motor system, leading to faster responses (see Supplementary Material). If participants followed a different strategy in the sense of “preparatory” alternations of attention between the two stimuli (because subjects know that there are some invalid trials as well) and alpha-band fluctuations depicted these behaviorally relevant fluctuations in the actual attentional selection, one would expect alpha-band amplitudes to be behaviorally relevant in invalid trials as well. As we did not find this pattern, this underscores the idea that alpha-band modulations are the consequence of top–down activity in the present experiment, and it is unlikely that spontaneous alpha-band fluctuations represent a mechanism that actively alternates top–down guided allocation of attentional resources.

This interpretation is further supported by the present findings (see Supplementary Material “Relationship of Post-cue Electrophysiological Measures and Attention”), which show that single-trial visual alpha-band amplitudes are not consistently predictive of SSVEP amplitudes. Additionally, several recent studies have reported that fluctuations in the alpha band do not influence SSVEP amplitudes, and therefore, do not affect early visual stimulus processing (Antonov et al., 2020; Gundlach et al., 2020; Nuttall et al., 2022; Zhigalov & Jensen, 2020). As mentioned above, SSVEPs robustly represent top–down guided neural sensory gain control in early visual cortex (Di Russo et al., 2001; Kim et al., 2007; Müller & Hillyard, 2000; Müller, Picton, et al., 1998; Müller, Teder-Sälejärvi, et al., 1998) (see also current study), this renders the view very unlikely that attentional alteration of stimulus processing in early visual cortex is mechanistically implemented by visual alpha-band activity. To what extent visual alpha-band activity reflects a gating mechanism at later processing stages (Peylo et al., 2021; Zhigalov & Jensen, 2020) that operates independently from stimulus processing at early visual stages and may index the readout of behaviorally relevant visual information (Chaumon & Busch, 2014) is still a matter of debate. Our data are in line with the proposed idea of a gating mechanism at later stages, but would suggest this gating to be a consequence of the allocation of attention. Of interest, in contrast to the here reported lack of a relationship between alpha-band activity and early sensory processing in recent work by Iemi et al. (2019), alpha-band fluctuations were in fact predictive of early sensory processing, as the early visual C1 component evoked by transiently presented wedges of checkerboard patterns was modulated by prestimulus alpha-band amplitudes. One crucial difference to our study was that in their study, effects of alpha-band amplitudes were studied under constant attentional allocation, that is, effects of lateralized alpha-band amplitudes were measured while participants were engaged in a central task throughout the experiment, while in our study spatial attentional selection of one side was manipulated on a trial-by-trial level. In addition, in our study, alpha-band amplitude measures were derived during constant visual stimulation, while for Iemi et al. (2019), signals from prestimulation periods without visual stimulation served as predictors. Whether potentially different fluctuations of alpha-band amplitudes were captured in both experiments, leading to these differential effects, will need to be addressed in future work.

Finally, some caution must also be mentioned when addressing the behavioral variance found in such designs. The classical Posner effect in this study resulted in average reaction time differences of 29.641 ms between valid and invalid trials. At the same time, there is substantial variation in response times across participants (grand mean average reaction time = 663.872 ms; SD of mean reaction times across participants = 64.899). RTs vary even more within each subject (see Fig. 2A) with an average SD of 133.509 ms. The actual Posner effect accounted only for a small part of variation within the data. In fact, extracting Bayes-R² values for the models fitted to the single-trial reaction time data revealed that only 18.010% of variance was explained by general reaction time offsets and, thus, general differences between participants. Around 1% additional variance was explained by the factor cue validity, that is, the classical Posner effect (18.922%) and less than 1% was additionally explained by postcue alpha-band activity at contralateral electrodes (resulting in overall 19.168% of the explained variance). While effects such as the classical Posner effect have been found and replicated in various experiments and publications, about 80% of single-trial variance was not explained with our models. Recent work, however, points toward the functional significance of variance in biological systems to actually allow flexible, precise, and robust propagation of information and ultimately behavioral responses (Ecker et al., 2016; McGinley et al., 2015; Rowland et al., 2023; Schölvinck et al., 2015; Vinck et al., 2015; Waschke et al., 2021). Future efforts need to be made to find, disentangle, and characterize potential additional determinants of neural and behavioral variability and the role and profile of noise for neural computations.

Our analyses were limited to time windows which in this and previous studies revealed extensive modulations of neural activity under sustained attention. We addressed whether trial-by-trial fluctuations in neural measures for a postcue window of sustained attention are behaviorally relevant. While we also examined fluctuations on a time scale closer to the actual processing of the target stimulus, neural measures were aggregated across a time window of 1 second in all approaches. It, thus, remains an open question, how dynamics and fluctuations on smaller time scales may contribute to behavior. In addition, our design did not include a graded measure of accuracy within each trial. The relevance of trial-by-trial fluctuations of neural measures related to behavioral accuracy will need to be evaluated in future studies.

Furthermore, while our analyses revealed alpha-band activity in both hemispheres to be relevant and an enhanced alpha-lateralization pattern to be behaviorally beneficial in valid trials, the functional interpretation may be ambiguous. Alpha-band modulations, typically characterized by a decrease contralateral to the cue and an increase ipsilateral to it, have been associated with enhanced processing of the target stimulus and reduced processing of distracting information (contralateral to the uncued stimulus) (see Capilla et al., 2014). Yet, both signals may be confounded and driven by a bilateral modulation related to the processing of the target stimulus. In fact, in our design, no potentially distracting events were presented at all and a functional interpretation with regard to the attenuation of distraction information and/or the enhancement of target information are, therefore, not possible. A distinct design, such as used by Orf et al. (Orf et al., 2023), would be required for disentangling target enhancement and distractor suppression-related processes.

To summarize, we found modulations of behavior, SSVEP, visual, and motor alpha-band activity through top–down guided spatial attention in a probabilistic spatial cueing task. These findings mirror previous results and are extended by evaluating the behavioral relevance of the attention-modulated but unrelated neural measures. Here, fluctuations in visual alpha-band activity measured right after the cue and stretching until target presentation as well as fluctuations in motor alpha-band activity obtained just before target presentation were linked to response times to target stimuli in valid and neutral but not invalid trials, which conflicts with the idea of alpha-band fluctuations signify top–down guided allocation of attentional resources (Gundlach & Forschack, 2020). If alpha-band activity would have played an active role in top–down attentional deployment, one would expect larger amplitudes contralateral and/or smaller amplitudes ipsilateral to the cued hemifield to be related to the faster switching of attention to the target at the uncued side in invalid trials, and hence faster reaction times. As this was not the case, we question the active role of alpha-band fluctuations in stimulus processing in such designs. Instead, and in line with some previous studies, we suggest that behaviorally relevant alpha-band modulations are the consequence of top–down guided spatial attention, representing neural excitability in cortical areas that are activated by the attentional shift. Spontaneous fluctuations of neural excitability in these areas have an influence on stimulus processing for the cued side, but will not overwrite top–down guided shifts of attentional resources to one side as long as it is the best strategy to follow the cue for best performance.

The data and code used for the analyses and results presented in this study are available at: https://osf.io/6ygvm/.

C.G., N.F., and M.M.M. designed the study. C.G. and N.F. analyzed the data. C.G. and M.M.M. wrote the manuscript. M.M.M. acquired funding.

Work was funded by grants to M.M.M. from the Deutsche Forschungsgemeinschaft (MU 972/26-1 and MU 972/26-2). Supported by the Open Access Publishing Fund of Leipzig University.

The authors declare no conflict of interest.

We thank Maria Dotzer for her help with data recordings.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00312