Sensory Drive Modifies Brain Dynamics and the Temporal Integration Window

In recent decades, there is accumulating evidence from behavior and neurobiology that perception is not continuous, as it may seem, but is, in fact, discrete or at least periodic. In other words, perception fluctuates (VanRullen, 2018). Support for this argument can be found on both the perceptual and neurophysiological levels. Perceptually, our senses are constantly bombarded by stimuli, overwhelmingly more than we can fully process. Through mechanisms of attention, we select portions of the sensory input that are most relevant to our goals (Nobre & van Ede, 2023; Beck & Kastner, 2009; Treue, 2003). Previous findings from our group and others suggest that ongoing perception is not uniform over time. Specifically, after a flash, the likelihood of detecting subtle stimuli and the underlying neural activity fluctuate rhythmically (Re, Karvat, & Landau, 2023; Fiebelkorn & Kastner, 2019; Helfrich, Breska, & Knight, 2019; Re, Inbar, Richter, & Landau, 2019; Landau, Schreyer, van Pelt, & Fries, 2015; Landau & Fries, 2012).

At the neural level, reports of oscillations in neurophysiological signals date back to the earliest EEG recordings in which alpha and beta oscillations were first described (Berger, 1929). These reports led theoreticians to suggest the prominent alpha rhythm (8–12 Hz) as the physiological temporal substrate for scanning the environment (Wiener, 1948; Pitts & McCulloch, 1947). This suggestion was put into a theoretical framework, according to which perception is divided into discrete moments, each encompassing one alpha cycle (VanRullen, 2018). Consequently, when two stimuli arrive during the same cycle, it is predicted that they will integrate into a single percept, whereas when they arrive at different cycles, they will be perceived as segregated. This theoretical viewpoint gives rise to a prediction that the individual temporal integration window (TIW), and consequently, the temporal segregation threshold, depends on the length of the alpha cycle. That is, for individuals with a higher peak frequency within the alpha band, and thus shorter cycles, stimuli closer in time would be perceived as separate (VanRullen, 2016).

Several lines of evidence support the idea that the TIW is linked to the length of the alpha cycle. A number of studies measured the visual TIW in the context of cross-modal temporal integration, using the sound-induced flash illusion (Shams, Kamitani, & Shimojo, 2000); presentation of a single flash together with two beeps generates the illusion of “fission” into two flashes, as long as both beeps are presented during the visual TIW. Using this method, TIWs spanning 50–150 msec were found (Buergers & Noppeney, 2022; Venskus & Hughes, 2021; Cecere, Rees, & Romei, 2015). These TIWs fit with the notion that the alpha cycle might be relevant for this process. Furthermore, the TIW can be lengthened or shortened by stimulating the brain in frequencies slower or faster than the individual alpha frequency (IAF), respectively (Venskus et al., 2021; Cecere et al., 2015), and correlations were found between the IAF and illusion magnitude (Noguchi, 2022; Keil & Senkowski, 2017) or TIW length (Venskus et al., 2021; Cecere et al., 2015). However, a recent study challenged this IAF to TIW link, reporting no correlation between spontaneous prestimulus alpha peak frequency and the perceptual integration window (Buergers & Noppeney, 2022).

A more direct measure of the visual TIW is the two-flash fusion (2FF) threshold. This threshold is measured in an experimental setup where two visual stimuli are presented successively, and the minimal ISI required for the participant to detect the two stimuli, rather than perceiving them as one, is defined as the threshold. Studies over the past decades revealed that both state-related parameters and stimulus parameters affect the subjective 2FF threshold. For example, observers' state parameters include general arousal (measured by skin conductance; Venables, 1963a); influence of drugs (Kopell, Noble, & Silverman, 1965); individual differences in the aperiodic component of the EEG spectrum (Deodato & Melcher, 2023); mood, anxiety, and age (Melcher, Lapomarda, & Deodato, 2023); and between-trial differences in alpha frequency, power, and phase (Drewes, Muschter, Zhu, & Melcher, 2022; Baumgarten, Schnitzler, & Lange, 2015; Coffin & Ganz, 1977). The influencing stimulus parameters include the location of the flashes relative to foveal focus (with lower thresholds for peripheral stimulation; Yeshurun & Levy, 2003) as well as the size, duration, ISI range, and luminance of the flashes, with prolonged and brighter flashes associated with shorter thresholds (Table 1). The size and luminance of the surrounding background of the flashes also contribute to the variability in the subjective 2FF threshold, as brighter and bigger surroundings are associated with lower thresholds (Foley, 1956, 1961). Importantly, it was recently shown that although presentation of the flashes in the dark is associated with a correlation between the IAF and the 2FF threshold, this correlation breaks when the flashes are surrounded by an annulus of light (Gray & Emmanouil, 2020). Gray and Emmanouil (2020) also tested whether flickering the surrounding illumination at 8.3 or 12.5 Hz (just slower and just faster than typical alpha) can influence 2FF thresholds and their correlation with IAF. Their results indicated that 2FF thresholds are not influenced by alpha-band flicker frequency, suggesting that IAF was not entrained by visual flicker.

In the current study, we utilized a different approach to manipulate the surround stimulus and investigate its role in the subjective 2FF threshold and its correlation with cortical oscillations. To manipulate oscillations, we used gratings with high or low contrast. This method is known to reliably induce internally generated gamma oscillations and reduce alpha, in a controllable manner. Moreover, presenting gratings at different contrasts has been shown to have an effect on the magnitude of alpha reduction and gamma increase, as well as on gamma peak frequencies (Scheeringa et al., 2011; Koch, Werner, Steinbrink, Fries, & Obrig, 2009; Fries, Scheeringa, & Oostenveld, 2008; Hoogenboom, Schoffelen, Oostenveld, Parkes, & Fries, 2006). This enabled the measurement of the effect of the surrounding stimulus both on the behavioral thresholds and on physiological signatures, such as brain waves in a wide range of frequencies recorded with EEG.

Perceptual, and consequently neuronal, preflash states were experimentally manipulated by presenting a large grating before and throughout the two-flash stimuli. In each trial, a grating was presented, which was either high contrast or low contrast, creating different degrees of visual-input drive. The two-flash stimuli were then presented through an aperture in the grating, and we tested the influence of the stimulus surrounding on the TIW and on neuronal states, in both the group and individual participant levels. Specifically, we examined whether the grating contrast affected the 2FF threshold (as a proxy of the TIW). We also explored the neural impact of this context manipulation by investigating oscillatory responses in a wide range of frequencies as well as the IAF at rest.

We found that, first, and in agreement with Gray and Emmanouil (2020), when the flashes are presented within a grating annulus, the gap-detection threshold is shorter than when the flashes are presented without the annulus. Second, high-contrast surrounding stimuli yielded higher gamma power and shorter TIWs compared to low-contrast stimuli. This finding suggests a link between oscillations and temporal segregation. Third, and quite surprisingly, in some of the participants, the presentation of the surrounding grating was also accompanied by increased alpha power. Because power fluctuations can result from nonrhythmic processes (Fransen, van Ede, & Maris, 2015), and to allow closer investigation of the link between alpha oscillations and the TIW, we developed a method to measure the sustainability of oscillations (termed within-trial phase locking, WTPL). Using this novel method, we found sustained alpha activity in about half of the participants. Finally, we found that the individual differences in the emergence of alpha, but not individual peak frequencies, tended to relate to detection threshold.

Forty-two individuals (30 women, average age = 22.9 [SD = 2.6] years) participated in the experiment. Participants were recruited from the university community and were compensated for their time with either money (€10 per hour) or class credit. The sample size was based on previous work (Hülsdünker & Mierau, 2021; Grandy et al., 2013). During the experimental session, participants performed two experiments: a temporal bisection task reported elsewhere (Ofir & Landau, 2022) and a two-flash task reported here for the first time. The session order (bisection/two-flash) was counterbalanced between participants. All procedures were approved by the institutional review board. Participants were excluded because of technical issues during recordings (three participants). In addition, we used the slope of the psychometric curve as a measure of internal noise and excluded participants with a slope lower than 0.13, as suggested by Deodato and Melcher (2024). Two participants were rejected because of this criterion. To note, these participants had gap-detection thresholds longer than 2 SDs above the group average. In total, the data of 37 participants (27 women) were analyzed.

The experiment was coded using Psychtoolbox (Brainard, 1997) in MATLAB 2019b (The MathWorks) and ran with a BenQ XL2420Z monitor with a refresh rate of 144 Hz. A red fixation point was displayed at the center of the screen and was visible throughout the entire experiment, except breaks. A trial began with the presentation of a square wave grating annulus around the center of the screen on a gray (RGB = 128,128,128) background (Figure 1). The grating had spatial frequency of three cycles per visual angle degree and one of two levels of contrast: 100% (ranging from RGB = 0,0,0 [black] to RGB = 255,255,255 [white]) or 50% (ranging from RGB= 64,64,64 to RGB = 191,191,191). The outer diameter of the annulus extended 8° of the visual field; and the inner diameter, 1°. For the flash presentation to start, participants had to fixate within a 1.5° radius of the fixation dot for 1 sec. After a continuous second of fixation, the flash sequence began. In “two-flash” trials, a 1° white (RGB = 255,255,255) disc filling the internal circle of the annulus (first flash) was presented for 14 msec (i.e., two frames of the screen), after which it disappeared for a variable duration and reappeared for additional 14 msec (second flash). In “one-flash” trials, a white disc appeared for a variable duration that equaled the total duration (from the onset of the first flash to the offset of the second flash) of the different “two-flash” trials. This was done to ensure that total duration could not be used as a cue for gap existence (Samaha & Postle, 2015). After the offset of the last flash, participants responded with a key stroke whether they saw one or two flashes. The grating remained on the screen until participants made their response. Before the task, participants completed a brief practice using two trials with 42-msec-long gaps and two trials with no gaps to familiarize themselves with the task.

For the first 23 participants, we used eight levels of gap duration (7, 14, 21, 28, 35, 42, 49, and 56 msec), which meant that the flashes of “one-flash” trials lasted [35, 42, 49, 56, 63, 70, 77, 84] msec. Each combination (gap duration, gap existence, and grating contrast) was presented 12 times, so that each participant completed 192 trials. For the rest of the sample (14 participants), we removed the two longest gaps. Durations 7–28 were presented 36 times each, and Durations 35–42 were presented 24 times each, keeping the total number of trials 192. For all participants, we varied the orientation (tilt of 45° clockwise or counterclockwise) and phase (0°, 90°, or 180°) of the grating between trials. Because different experimental parameters aside from IAF and background contrast can influence the 2FF threshold (Table 1), we tested another group of participants (n = 30) without surround gratings, to ensure the thresholds are comparable to similar studies in the literature. This group was tested only behaviorally, that is, without EEG recordings.

To assess the threshold of each participant, we fit logistic psychometric functions to the behavioral responses (i.e., the probability of “two-flashes detected”) of individual participants to low- and high-contrast trials separately using the Palamedes toolbox (Prins & Kingdom, 2018). Threshold and slope were free parameters, whereas guess and lapse were fixed at 0.02. The thresholds were defined as the gap duration for which participants were predicted to report seeing two flashes 50% of the times. In addition, and to allow direct comparison to previous studies, we also calculated thresholds defined as the gap corresponding to the mean of the psychometric curve fitted to percent correct response (usually ∼75%; Gray & Emmanouil, 2020; Samaha & Postle, 2015).

We recorded the EEG of participants while they performed the task described above, using a g.GAMMAcap (gTec) and a g.HIamp amplifier (gTec). The cap had 62 active electrodes distributed over the scalp (ground located at the location of AFz), positioned according to the extended 10–20 system, with the addition of two active earlobe electrodes. In addition, we recorded the horizontal EOG using passive electrodes placed at the outer canthi of both eyes and the vertical EOG using electrodes placed above and below the left eye. The EEG was continuously sampled at 512 Hz. During the experiment, we monitored the eye position using an infrared EyeLink camera (SR Research), sampling at 1000 Hz. The EyeLink signal and the EEG signal were time-aligned and stored for offline analysis using a Simulink model (The MathWorks).

We performed all preprocessing steps using the Fieldtrip toolbox (Oostenveld, Fries, Maris, & Schoffelen, 2011) in the MATLAB environment. First, the EEG data were referenced offline to the average of the earlobe electrodes. Then, data were inspected visually [using the function ft_databrowser()], and bad channels of each participant, on average 0.13 ± 0.07, were rejected. In addition, channels T7, T8, F9, and F10, which tended to be noisy in many participants, were rejected from all participants. Rejected channels were replaced with a spherical spline interpolation [using the “spline” method in ft_channelrepair()]. Scalp muscle artifacts were detected as epochs in which the amplitude of the bandpass-filtered signal at 100–120 Hz at all channels exceeded a z score of 30, with the function ft_artifact_zvalue(). Finally, spatio-temporal components containing eye blinks and saccades were detected and removed using independent component analysis, with the “runica” algorithm in ft_componentanalysis(). We preprocessed the session as a whole (i.e., without dividing into trials). Division into trials was done for each of the subsequent analyses separately.

To compute the time–frequency representation (TFR) of each participant, we used the recently introduced “superlet” transformation (Moca, Bârzan, Nagy-Dăbâcan, & Mureșan, 2021). This transformation offers a better control of the trade-off between frequency and time resolutions, which is necessary because of the Heisenberg–Gabor uncertainty principle, according to which it is impossible to specify exactly both frequency and time of the occurrence of a wave. By averaging a set of wavelets per time point and frequency, a better time resolution is achieved while keeping a good frequency resolution, compared to conventional methods such as single-wavelet and sliding-window Fourier transforms. Furthermore, the number of wavelets in the set (i.e., the order) per frequency can be chosen adaptively, to compensate the increasing wavelet bandwidth with increasing frequency, while maintaining a reasonable trial length and computation time (for details, see Moca et al., 2021).

To allow long enough data per trial to calculate two-cycle-wide wavelets at 1.25 Hz, we windowed the EEG data at [−2, 2.75] sec relative to grating onset. Then, per contrast, we calculated power with the routine ft_freqanalysis() with the “superlet” method, and the following parameters: frequencies of interest of 1.25–30 Hz in steps of 1.25 Hz and 32.5–100 Hz in steps of 2.5 Hz (Scheeringa et al., 2011), time of interest of −1 to 1.75 sec relative to grating onset, the order of the superlet increased linearly from 1 at 1.25 Hz to 40 at 100 Hz, and the length (in cycles) of the first wavelet in the set in each frequency was set to 2 and increased multiplicatively with the order. The power was calculated separately for high- and low-contrast trials, for each electrode, time point, frequency, and trial, and then averaged over trials. Then, we baseline corrected, using the mean power at the epoch [−0.8 to 0.2] relative to grating onset as baseline, and expressed the corrected values in decibel.

To facilitate Time × Frequency analysis and presentation (Figure 2A and C), we aimed to focus on electrodes showing the strongest response to grating presentation. Therefore, we searched for a cluster of electrodes exhibiting significantly higher gamma power (35–80 Hz; p < .05) at high-contrast trials compared to low-contrast trials. For each participant, gamma power was evaluated for the epoch of 0.2–1 sec relative to grating onset, that is, after the evoked response has waned, but before the presentation of the first flash. The participant-level data were then subjected to a group-level cluster analysis (see Statistical Analysis). One cluster, containing the channels PO3, POz, and PO4 (Figure 2D), fulfilled this requirement and was used in all further analyses.

To estimate the power spectra of the response to the grating (Figure 2E), we calculated the event-related spectral perturbation (ERSP) by dividing the poststimulus power spectrum by the prestimulus spectrum, which is collapsed over time points (Makeig, Debener, Onton, & Delorme, 2004). For each trial, we defined the window [0.3, 1] sec relative to grating onset as “poststimulus” and the window [−0.7, 0] as “prestimulus.” Then, we estimated the power in channels PO3, POz, and PO4 by padding each period with zeros to the nearest power of two samples (i.e., 1 sec), Hann windowing, and Fourier transforming. Finally, we expressed the ratio between poststimulus and prestimulus in each trial in decibels and averaged over channels.

We estimated phases by windowing the data from channels PO3, POz, and PO4 at [−1, 2] sec relative to grating onset and transforming the data to the frequency domain using five-cycle-wide Morlet wavelets centered at the frequencies of 1–100 Hz with 1-Hz steps. We used the obtained Fourier-spectra complex values for both intertrial coherence (ITC) and WTPL.

The ITC, also known as intertrial phase-locking factor, was taken as the mean (over trials) of the phases in each point on the Time × Frequency plane (Delorme & Makeig, 2004; Tallon-Baudry, Bertrand, Delpuech, & Pernier, 1996). Specifically, we normalized each Fourier value to a unit value by dividing by the amplitude, summed all values over trials, and divided by the number of trials.

Phase-locked fluctuations in power can result from both genuine oscillatory activity and nonoscillatory time-locked responses (van Diepen & Mazaheri, 2018). To evaluate the rhythmicity of the signal, that is, how sustained the oscillation is for at least one period (Shin, Law, Tsutsui, Moore, & Jones, 2017; Fransen et al., 2015), we developed the WTPL measure. In each frequency, time point, and trial, we computed the difference between the phase in the time of interest (ϕ₀), one cycle beforehand (ϕ₋₁) and one cycle after (ϕ₁). Then, we calculated the mean resultant vector length of the two differences (see equations and further details in Box 1, WTPL). Because some frequencies (e.g., alpha) tend to have higher overall lagged coherence than others (Rayson et al., 2022; Fransen et al., 2015), we corrected the WTPL values by subtracting the mean of the epoch [−0.5, 0] relative to the grating onset in each frequency. See further details in Box 1 and MATLAB code for implementation of this method in Appendix 1.

To estimate IAF at rest, we recorded a 2-min session where participants were awake with their eyes closed. The power spectrum densities in all channels throughout this session were estimated using the Welch method with 2-sec-wide Hann windows and 50% overlap. The IAF was chosen as the frequency of peak power in the range of 6–14 Hz. For all participants, the electrode with peak power was above the parietal or parieto-occipital areas. The mean ± SEM of all IAFs was 10.3 ± 0.17 Hz (range = 7.3–13 Hz).

To estimate individual alpha and gamma frequencies during the task, we first calculated the ERSP as described above, averaged over channels PO3, POz, and PO4. We averaged the ERSP over all trials, high-contrast trials and low-contrast trials. The individual frequencies in each condition were defined as the frequency with maximal power in the range of 7–13 Hz (for alpha) or 35–80 Hz (for gamma) per condition.

All statistical analyses were performed in MATLAB with the Statistics and Machine Learning, Fieldtrip, and CircStat toolboxes (Oostenveld et al., 2011; Berens, 2009). Data throughout this report are presented as mean ± SEM unless otherwise stated. The statistical tests and corrections for multiple comparisons used for each analysis are presented in Table 2. Significance level was set at α = .05. When multiple comparisons involved two dimensions (i.e., Time × Frequency), we first computed the t test at each data point (compared to either the mean per frequency in the baseline epoch or 0; see Table 2). Then, we clustered neighboring points exceeding a significance level of α = .05 and summed the t values of each cluster. This sum served as the value for comparison in the cluster-level statistics, in which we shuffled the labels (e.g., trial or baseline) and took the sum of the largest cluster. We repeated this step 1000 times, and clusters in the original data with summed t values larger than 97.5% of the shuffled distribution were considered significant. This analysis was implemented either with the Fieldtrip function ft_freqanalysis() with the “montecarlo” method (Oostenveld et al., 2011; Maris & Oostenveld, 2007) or with a routine contributed by Gerber (2023). Before using t tests as the parametric measure (i.e., without a following nonparametric cluster distribution test), we tested the distribution for normality using the Kolmogorov–Smirnov test.

We asked participants to report whether they detected one or two flashes of light presented either over a fixation point (Group 1) or at the center of a square wave grating annulus (Group 2). Between trials, we manipulated three critical variables: (1) the number of flashes (1 or 2), (2) the duration of the gap in two-flash trials (7–42 msec, in steps of 7 msec) or of the flash in one-flash trials (35–70 msec, to match the total duration of the two-flash trials), and (3) the contrast of the grating in the surrounding annulus (for Group 2, 50% or 100%; see Figure 1A and Methods). For each participant, we fitted a logistic psychometric function (Prins & Kingdom, 2018) to the percentage of reported two flashes per gap duration, treating one-flash trials as a 0-msec gap (Appendix Figure 1). Overall, the degree of preflash sensory drive inversely correlated with the TIW, indicated by a shorter 2FF threshold when the surrounding contrast was higher (Figure 1B; high [100%] contrast: 13.7 ± 0.82 msec, low [50%] contrast: 16.5 ± 0.74 msec), paired t(36) = 5.53, p = 2.9 × 10⁻⁶, Cohen's d (effect size): large, 1.01. Thirty of 37 participants (83%) had a shorter threshold for the high-contrast condition (Figure 1C). The threshold of participants from Group 1 (no grating annuli) was longer than that of participants from Group 2 [ANOVA: F(2, 101) = 47.71, p < 2.59 × 10⁻¹⁵, no grating different from both grating contrasts with p < 1 × 10⁻⁸, Tukey–Kramer post hoc analysis]. To note, this threshold dependency on the surround stimulus was not evident when considering percent correct responses rather than percent reported “two flashes” (high contrast: 15.87 ± 0.99 msec, low contrast: 15.79 ± 0.74 msec), t(36) = 0.09, p > .9. Together, these results suggest that the surround gratings change the criterion, but not the sensitivity, to two flashes.

Previous studies showed that strong sensory drive can modify brain dynamics. Specifically, a continuous increase in power of narrow-band gamma (45–65 Hz) is observed when participants watch high-contrast gratings (Scheeringa et al., 2011; Fries et al., 2008; Hoogenboom et al., 2006). Therefore, we next asked whether our manipulation of the grating contrast affected preflash oscillations in gamma and other bands and if these controlled brain-dynamics modifications are related to the effect of sensory drive on the TIW. We calculated the TFR with the recently introduced “superlet” method (Moca et al., 2021) and found increased gamma power after the onset of the grating in both high- and low-contrast trials (Figure 2A and B, top). However, only for high-contrast trials, this increase was significant in a cluster-based permutation test (Figure 2E), and the high-contrast gamma power at 45–65 Hz was significantly higher than in low-contrast trials throughout the presentation of the grating (Figure 2C; black contour indicates significant [p < .05] clusters). The high- to low-contrast difference was strongest in parieto-occipital channels (Figure 2D, top).

Investigation of the lower frequencies (Figure 2A and B, bottom) revealed a significant, short-lasting cluster of increased power at 3–15 Hz immediately after the grating onset that can be attributed to the stimulus-evoked potential. Expectedly, this response was stronger in high-contrast trials compared to low-contrast trials (Figure 2C, bottom). Finally, note that, 500 msec after grating onset, an increase in alpha (∼10-Hz) power can be seen. This increase is evident in both high- and low-contrast trials. However, the observed power increase did not reach significance in the group-level permutation test.

Alpha activity is often taken to represent the inhibition of unattended stimuli (Cooper, Croft, Dominey, Burgess, & Gruzelier, 2003), so it is often not characterized alongside the grating-induced continuous gamma, thought to reflect stimulus processing. This raises an intriguing question: Do individual participants exhibit significantly increased alpha power locked to the attended stimulus? This exploration is motivated by our goal of linking variability in the integration window of individuals to neurophysiological dynamics.

A major goal of this study was to investigate the neural dynamics that account for individual differences in temporal integration and segregation. Therefore, we next characterized spectral signatures at the single-participant level (Figure 3). In agreement with the group-level analysis, we found that more participants had significantly increased gamma (45–65 Hz) activity in high-contrast trials, compared to low-contrast trials (Figure 3C and D, red). In addition, ∼20% of participants had clusters with significantly increased alpha band (8–12 Hz) activity in both high- and low-contrast trials, whereas a similar number of participants exhibited significantly decreased power in this band (compare red and blue traces in Figure 3C and D).

The different spectral signatures between participants raise the question whether different power levels indeed reflect differences in underlying oscillatory processes. To answer this question, we need a tool sensitive enough to characterize the presence of sustained oscillations in the single-participant level. We hypothesized that an oscillatory process should be sustained (Fransen et al., 2015) and developed a tool to measure this sustainment over time within single trials, termed WTPL (see Box 1 for further details). A visual inspection of the single-participant EEG response locked to the grating onset revealed clear, phase-locked, and sustained oscillations in some participants (Figure 4). The WTPL allows testing the sustainability of oscillations in a statistically defined manner at the single-participant level. We found a significant (p < .05, permutation test) increase in alpha sustainment in the time window of 0.5–1 sec after grating onset for 21 of 37 participants (57%). Each of the 14 participants who had significantly increased alpha power (measured by ERSP) also had significantly increased alpha sustainment (measured by WTPL). Interestingly, and unlike the power analysis, WTPL revealed an increase in alpha oscillations at the group level. The significant cluster extended from 400 msec after grating onset and at least until the presentation of the first flash (Figure 4 and Box 1). Importantly, this alpha activity is phase-locked within and between trials and can be separated from the transient (0–300 msec) time-locked average response to the grating onset (Figure 4A and Box 1).

Individual participants performed the same two-flash task and differed in the existence of sustained alpha activity. This raises the question of whether, in addition to the potential influence of IAF on TIW (see Introduction and below), the existence of sustained alpha is related to TIW. To test this hypothesis, we split participants according to whether they displayed significantly increased WTPL in the alpha band (7–13 Hz) between 0.5 and 1 sec after grating onset (Figure 6A). The 21 (57%) participants who exhibited increased visually induced sustained alpha (VISA) tended to have longer gap-detection thresholds than the 16 participants who did not, t(35) = 1.82, p = .078, Cohen's d = 0.603 (medium).

As mentioned earlier, the IAF was suggested to influence the differences in TIW between individuals (Samaha & Postle, 2015). The IAF can be measured either globally as the peak in the power spectral density profiles of participants during a session of rest (i.e., awake with eyes closed) or specifically during task-related intervals (i.e., VISA). In Figure 7, we present the correlations between both measures of IAF and the gap-detection threshold of all trials, or split into high-contrast and low-contrast trials. In agreement with a recent report (Buergers & Noppeney, 2022), we did not find significant correlations between the IAF and gap-detection thresholds, even when splitting participants according to significant and nonsignificant VISA (Appendix Figure 5). Likewise, we found no difference between the peak power and frequency of alpha during high- and low-contrast trials [peak cycle: t(35) = 0.26, p = .8, peak power: t(35) = 0.32, p = .75; Appendix Figure 5]. As reported above, such difference in power does exist in the gamma band (Figure 2), t(35) = 3.19, p = .003, and we found also a significantly different gamma peak cycle lengths between high- and low-contrast trials (Appendix Figure 5), t(35) = 2.72, p = .01. Because the cycles of peak gamma frequencies were found in the range of the gap-detection threshold (15–25 msec), we tested for correlations between individual peak gamma frequencies and individual thresholds. However, no significant correlations were found in these tests.

In the present study, we tested the effect of the context (i.e., surrounding grating annulus) on the 2FF threshold and its electrophysiological basis. At the perceptual level, we found a strong dependence of the TIW on the grating: First, when the two flashes appeared inside the grating, TIWs were short (15–20 msec) relative to when the flashes appeared without surround gratings (27 msec), and second, in trials with a high-contrast grating, the 2FF threshold was shorter relative to trials with a low-contrast grating. At the neurophysiological level, we found increased power in the alpha band during the relatively long (∼1 sec) presentation of the stimulus. To ensure that the power fluctuations are related to genuine rhythmic activity, we developed a method sensitive to the per-trial sustainability of the signal. We found that about half of the participants exhibited significantly increased alpha activity, termed here VISA, that appeared approximately 500 msec after the grating's onset and persisted at least until the flash stimuli were presented. Finally, as predicted from the literature, a persistent narrow gamma-band increase was measured in response to the prolonged display of the grating (Scheeringa et al., 2011; Fries et al., 2008; Hoogenboom et al., 2006).

Our behavioral findings, that in high-contrast trials, the integration windows are shorter, are in line with previous physiological work (Kohn, 2007). Specifically, studies have found that the integration time of lateral geniculate nucleus neurons shortens with increasing contrast levels (Mante, Bonin, & Carandini, 2008). It was suggested that the visual persistence, that is, manifestation of a stimulus beyond its physical duration, is a psychophysical result of this integration time of neurons (Purcell & Stewart, 1971). This dependency can explain the wide range of thresholds found under different stimulation conditions (Bowen, Markell, & Schoon, 1980; Purcell & Stewart, 1971; Utial & Hieronymus, 1970; Lewis, 1967, 1968; Kietzman, 1967; Mahneke, 1958). However, in our study, the contrast level of the surrounding gratings affected the TIW of the center flashes, which had constant luminance. We suggest that this effect may be explained within the framework of lateral inhibition; until the visual persistence of the first flash has dampened, the presentation of the second flash will be integrated with the first. Because the activation of receptors in the visual system elicits inhibition from their neighbors (e.g., center-on/surround-off), the stronger the surround activation (e.g., by higher contrast grating), the stronger the associated inhibition. Stronger inhibition requires less time to bring the system to the prestimulus state, that is, shorter persistence and shorter TIW (Purcell & Stewart, 1971). For flash stimuli, Drewes and colleagues (2022) have shown that when the first stimulus is strong (full contrast), the performance in the 2FF task is significantly better than when the first stimulus is in near-threshold contrast. At the surround stimuli level, in trials with high surround contrast, we can expect a stronger lateral inhibition and shorter persistence of the flash, resulting in shorter TIWs. Support for this idea comes from the higher false-alarm rate in the high-contrast relative to low-contrast trials (Figure 1B), which may point to a stronger tendency of the system to parcellate the stimulus into chunks. We elaborate more on the role of lateral inhibition and visual persistence in temporal segregation in our companion paper published in this issue (Karvat & Landau, 2024).

The sustained and phase-locked alpha activity we report here adds to a growing list of functions associated with the alpha rhythm. Prestimulus alpha has been related to attention as an inhibitory force (Foxe & Snyder, 2011; Hanslmayr, Gross, Klimesch, & Shapiro, 2011); alpha activity, as seen also in our results, is suppressed by stimulus onset (Wang, Megla, & Woodman, 2021; Chaumon & Busch, 2014); finally, alpha can also be induced by stimulation (Lozano-Soldevilla & VanRullen, 2019; VanRullen & Macdonald, 2012). Although the previous studies found alpha enhancement when white noise luminance sequences were presented, we show that static stimuli can also enhance alpha-band activity. Additional studies are required to shed light on how this enhancement depends on stimulation features (duration, contrast, luminance, spatial structure, etc.) and on their relation to inhibitory roles associated with alpha-band activity. In the context of the current special focus on alpha dynamics and temporal processing, the sustained and phase-locked alpha activity also provides an elegant way to control the phase of ongoing alpha as well as a tool to study the relation of alpha phase and perception in a controlled fashion.

The lack of correlations between individual alpha-peak frequency and 2FF thresholds we presented here supports prima facie the rejection of the IAF-to-TIW hypothesis. However, does it mean that the theory put forth by Samaha and Postle (2015) is void? Not necessarily. A good scientific theory offers a hypothesis that can be disproved experimentally. Null results are then an invitation to amend and refine, and vice versa. So where can we go from here?

First, the effect might be related more to variations of alpha parameters in the within-participant level than to differences between participants. For example, the within-participant differences in the alpha phase at the moment of stimulus presentation in each trial are likely to have a bigger effect on the temporal binding window than the (relatively small) interparticipant differences in frequency (Baumgarten et al., 2015; Busch, Dubois, & VanRullen, 2009; Mathewson, Gratton, Fabiani, Beck, & Ro, 2009; Varela, Toro, John, & Schwartz, 1981; but see Ruzzoli, Torralba, Morís Fernández, & Soto-Faraco, 2019). The relatively short 15- to 20-msec thresholds reported here provide a rationale for the importance of intertrial phase differences; if temporal integration occurs over a period encompassing one eighth to one fifth of the alpha cycle length, the two stimuli should arrive at specific parts of the wave (i.e., phase), not only on the same cycle, for alpha to have an influence. Second, the effect of oscillatory cycle length on temporal resolution can be manifested in other frequency bands in addition to alpha. For example, the 15- to 20-msec thresholds reported here fit well to the cycle duration of the gamma frequencies. In our paradigm, a preflash visual-grating stimulation around the fixation point induced an increase in power in both alpha and gamma.

Taken together, our results indicate that the two-flash threshold is influenced by the strength of the surrounding sensory drive. The sensory drive is also known to affect both the lateral inhibition and oscillations in the gamma and alpha bands. The gamma-band oscillations are affected by mechanisms of activation-induced pyramidal interneuron network gamma (Whittington, Traub, Kopell, Ermentrout, & Buhl, 2000; Traub, Whittington, Stanford, & Jefferys, 1996). Furthermore, alpha-band effects are discussed below and further detailed in Karvat and Landau (2024). Together, these findings suggest that the influence of the surround stimulus on the 2FF threshold can be explained by the complex effect of oscillations on the persistence of the first flash.

Next, we want to point out some limitations in the current study and suggest trajectories for future studies. First, this study was designed to investigate the effect of brain states on thresholds. As such, the number of trials around the threshold was relatively low. This limited our ability to apply signal detection theory methods as well as to compare alpha phase between trial outcomes, which rely on sufficient amount of detected as well as undetected trials. It should be noted that, in EEG, we record dipolar fields, and because of the morphology of the cortex and specifically the existence of sulci and gyri, directly estimating the exact phase is not trivial (van Diepen & Mazaheri, 2018). In this regard, the finding of sustained and phase-locked visually induced alpha can pave the way for future studies. For example, because we have evidence for a “phase reset” at the time of grating presentation, the hypothesis of phase-related TIW at the time of flash presentation can be tested by jittering the duration between grating and flash onsets and by focusing more on ISIs around the threshold.

Second, in our EEG recordings, some participants exhibited significant alpha activity during stimulus presentation, but others did not. This might be an interesting finding resulting, for example, from different participants using different strategies, implemented by the presence or absence of alpha. On the other hand, it could also be that the detection or none thereof of alpha modulation resulted from the placing of the electrodes on the scalp and different dipole orientations between participants (van Diepen & Mazaheri, 2018). In addition, we report here the rather puzzling existence of alpha (which is thought to suppress sensory input) on channels responding to visual stimuli, while the stimulus is presented. However, the spatial resolution of noninvasive EEG is limited and does not allow us to determine if this alpha response indeed coexists with the input at the local network level. It will be interesting to revisit these issues in future studies using methods with better spatial resolution such as intracranial recordings or high-density M/EEG combined with source estimation analyses.

Finally, in the present study, we treat all oscillations in the range of 8–12 Hz as “alpha.” However, there is accumulating evidence suggesting the existence of at least two “alphas.” Spatial propagation analysis has revealed two distinct alpha-band oscillations based on the direction of traveling waves (Pang, Alamia, & VanRullen, 2020; Alamia & VanRullen, 2019). Although visual perception increases forward alpha waves (i.e., occipital to frontal) during visual stimulation, backward waves (frontal to occipital) increase when the stimulus is off. A recent study (Alamia, Terral, D'ambra, & VanRullen, 2023) found that the backward (top–down) waves are correlated with the overall alpha power and the attentional and inhibitory effects (Händel, Haarmeier, & Jensen, 2011; Jensen & Mazaheri, 2010). Forward (bottom–up) alpha waves, on the other hand, are directly related to the sensory input (similar to the VISA reported here and in Lozano-Soldevilla & VanRullen, 2019; VanRullen & Macdonald, 2012). Therefore, it would be beneficial to include wave propagation analysis methods in future studies of oscillatory effects on the temporal binding window. The possible theoretical implications of the existence of two alphas are further discussed in our companion paper published in this issue (Karvat & Landau, 2024).

To summarize, in this work, we set out to investigate the contextual effects of oscillatory brain activity on the temporal resolution of visual perception. We found that contextual stimulus drive had the expected effect on oscillatory brain activity and a compelling effect on the behavioral estimation of the TIW. We did not find direct evidence for previously documented effects of alpha frequency on the TIW. However, we did characterize another alpha-related individual difference that may relate to temporal cognition, namely, the VISA, which could prove useful in future studies on the role of alpha oscillations in temporal segregation. The report contributes a unique approach to the study of temporal segregation in the brain. We present noninvasive means to control and manipulate fluctuations in visual cortex excitability to understand the neurobiological makeup of temporal segregation of different visual events. Together with the novel analytical approach presented here (WTPL), this approach can inspire future studies aimed to further elucidate the role of alpha on temporal resolution.

We thank Noa Itzhaki and Segev Feinstein for their assistance in data acquisition. We also thank Dr. Flor Kusnir for providing insightful remarks on previous drafts of this article.

Corresponding authors: Ayelet N. Landau, Department of Psychology, Hebrew University of Jerusalem, Mt. Scopus, Jerusalem, 9190501, Israel, or via e-mail: [email protected] (Golan Karvat) or [email protected] (Ayelet N. Landau).

Most of the analyses in this manuscript were based on the freely available MATLAB toolbox Fieldtrip. The code and example data can be found here: https://github.com/laaanchic/WTPL. Data will be available upon request by email to the lead author.

Golan Karvat: Data curation; Formal analysis; Investigation; Methodology; Software; Visualization; Writing—Original draft; Writing—Review & editing. Nir Ofir: Conceptualization; Data curation; Investigation; Methodology; Project administration; Software; Validation; Writing—Original draft; Writing—Review & editing. Ayelet N. Landau: Conceptualization; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Supervision; Validation; Visualization; Writing—Original draft; Writing—Review & editing.

The Brain Attention and Time Lab (PI: A. N. L.) is supported by the James McDonnell Scholar Award in Understanding Human Cognition, grant number: ISF grant 958/16. This project has received funding from the European Research Council under the European Union's Horizon 2020 research and innovation programme, grant number: 852387.

Retrospective analysis of the citations in every article published in this journal from 2010 to 2021 reveals a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .407, W(oman)/M = .32, M/W = .115, and W/W = .159, the comparable proportions for the articles that these authorship teams cited were M/M = .549, W/M = .257, M/W = .109, and W/W = .085 (Postle and Fulvio, JoCN, 34:1, pp. 1–3). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article's gender citation balance. The authors of this article report its proportions of citations by gender category to be M/M = .595, W/M = .243, M/W = .122, and W/W = .041.