A total of 30 participants (15 young, 15 older) volunteered to participate in the study. Three of the older participants were excluded due to poor data quality. One young participant was also excluded as an outlier. To obtain equal sample sizes across groups, two young participants were pseudo-randomly selected and excluded from analyses. Finally, data of 12 young (3 male) and 12 older (2 male) participants were analyzed. All participants had normal or corrected to normal vision. Age groups were matched in years of education they received (Table 1). There was one left-handed participant in each group. Participants with head trauma, psychiatric diagnoses, or psychiatric medicine use were not included in the study. This study was approved by ethics committee in Bremen University and all participants signed an informed consent form before participating in the study.

A black rectangle on a white background was centered on the computer screen. Three different sequences of stimuli were created using this base rectangle: one for endogenous multistable motion, one for vertical exogenous motion, and one for horizontal exogenous motion (Figure 1). Stroboscopic alternative motion (SAM) was created by subsequent repeated presentation of diagonal double-dot and fixation stimuli (see Supplementary Video S1). Sequence of stimulus presentation was in the following order and repeated throughout the experiment: fixation, double-dot, fixation, double-dot. Duration of each such sequence was approximately 500 ms. A slightly modified version of the SAM rectangles was used to create the control stimulus (i.e, exogenous stimulus). Double-dot stimuli were placed at either horizontal or vertical corners instead of diagonal corners for this version (Figure 1). Motion change was externally applied in exogenous stimulus display by changing the double dot stimulus sequence (see Supplementary Video S2). A computer controlled 19″ CRT (Cathode-Ray Tube) monitor was used for stimulus presentation.

Schematic illustration of both conditions and each of the perceived motions are presented in Figure 1. The endogenous condition employed the stroboscopic alternative motion stimulus (SAM). Participants reported internally generated spontaneous reversals of the direction in motion perception by pressing a button on a custom response device. There are two visual illusions at play in this task: (1) one is the illusion of stroboscopic motion where participants perceive dots moving from one location to another and (2) the illusory alternation of the motion direction due to the diagonal positioning of the dots (Mathes et al., 2016). Participants either experience vertical or horizontal motion during continuous observation.

The exogenous condition was utilized as a control measure for the endogenous condition to ensure that participants correctly responded with a button press when perceiving a reversal of motion direction. The alteration in task design to the classical version of SAM allowed the assessment of reaction times upon externally (i.e., by stimulus change) triggered motion reversals (Başar-Eroglu et al., 2016; Mathes et al., 2016). Similar to SAM, perceived stroboscopic motion of the exogenous stimulus was either vertical or horizontal. Number of external reversals within a minute was set to eight for the exogenous stimulus in both groups; a slightly lower number compared to previous studies in younger adults (Mathes et al., 2016). This modification was undertaken due to lower reversal rates of older adults (Jalavisto, 1964) and, thus, achieve a comparable number of reversals during the endogenous and the control condition.

Participants were seated in a dimly lit, sound proof, electromagnetically shielded room. Experimenters informed participants about the endogenous SAM reversals and explained that they have to press the button whenever they experience a reversal. Participants were asked to focus to the central white dot of the stimulus throughout the experiment. Participants were trained for approximately 100 s to ensure that they understood the requirements of the two tasks. Duration of each task was 10 min, summing up to a total of 20 min.

The EEG was recorded using Ag-AgCl electrodes at F3, F4, C3, C4, P3, P4, T5, T6, O1, and O2 locations according to the 10–20 system. Earlobe electrodes were used as reference electrodes. Electrode impedances were below 5 kΩ for all electrodes. EOG was recorded from medial-upper and lateral-orbital rim of the right eye. The EEG was amplified with a Nihon Kohden (EEG-4421 G) EEG device with band limits 0.1–70 Hz. The EEG was digitized online with a sampling rate of 500 Hz. Digitized EEG was stored within a hard disc of a computer for offline EEG analysis. A 50 Hz Notch filter was applied to remove the mains interference. For the recording of EOG, the time constant 0.3 s with a low pass filter at 70 Hz was applied. All channels were displayed by a computer monitor to observe EEG activity during the recording.

Time-frequency decomposition of the EEG data was conducted via wavelet convolution in frequency domain (for details see: Cohen and Donner, 2013) utilizing custom scripts in Matlab (2018) (The MathWorks). For each participant, EEG epochs of 5 s for all conditions (reversal and non-reversal epochs) were concatenated prior to wavelet transformation and segmented again thereafter in order to reduce wavelet folding artefacts. Selected electrodes were F3, F4, C3, C4, P3, P4, O1, and O2. Then complex wavelets were constructed and scaled for each frequency by:

Where t is time, f is frequency, σ2 scales the gaussian window of the complex wavelets according to n, which is the number of wavelet cycles. Frequencies ranged from 8 to 14 Hz in 0.25 steps. Number of wavelet cycles was n = 7. Complex wavelets were normalized to have unit energy. The concatenated signals and the complex wavelets were then transformed into frequency domain using the built-in Fast Fourier transform algorithm (fft.m function in MATLAB; Frigo and Johnson, 1998) in Matlab (2018). Fast Fourier transform (FFT) is defined (Eq. 3) for vector x and n sampling points by:

Where ω=e−2πin is one of n complex roots of unity and i is the imaginary unit. Resulting power spectrums of the signal and the complex wavelets were then multiplied. Result of this multiplication was transformed into time domain using the built-in inverse FFT (iFFT) algorithm in Matlab (2018). Using the same definitions above, with vector x, n sampling points, and ω; iFFT is defined by:

The resulting complex signal was then reshaped into a format where conditions, trials, and time points are separated, identical to the signal format before concatenation. Then frequency-band specific power A was calculated for each participant at each time point, and averaged over trials in each condition by:

Where Z is the complex signal, t is time, and n is the total number of trials of a participant in a given condition.

To illustrate the changes in oscillatory activity in reversal epochs relative to non-reversal epochs, baseline normalization was conducted separately for each participant, each task, and channel. In accordance with previous studies, it was decided to take averaged values of non-reversal epochs as the baseline activity instead of assessing remote time points within the unstable epochs (Mathes et al., 2014; Başar-Eroglu et al., 2016; Mathes et al., 2016). Baseline windows started 500 ms before the epoch marker (0 ms) and ended at the zero-time point. Power values within the baseline time window were averaged across time points for each frequency, resulting in one mean value for each frequency (Cohen, 2014). Each of the baseline values were then log transformed and subtracted from each log transformed spectral estimate of averaged unstable epochs for each task and participant (Eq. 6). This procedure created baseline normalized time-frequency power values for each participant and task. Relative change in activity in unstable epochs was represented in decibels (dB):

Normalized time-frequency power values were then averaged within the specified time windows for alpha band and used in statistical analyses. The center frequency for the alpha band was set to 11 Hz for all participants, and 95% of its activity included the activity between 7.87 and 14.13 Hz (see Supplementary Figure S1). This range was defined as two standard deviations of the wavelet in the frequency domain. Given that wavelet cycle was set to 7, current center frequency (11 Hz) was the only frequency in the alpha band range that includes all of the activity in traditional alpha band (8–14 Hz).

Time windows were defined according to the button press demarcation in both the exogenous and the endogenous tasks. Therefore, time intervals before the button presses are shown by minus “−” symbol to indicate their relative time point to the button press. Two time windows were defined for endogenous condition: pre-reversal (−950 to −650 ms) and reversal (−350 to −50 ms) windows. Pre-reversal period represented the activity before the perceptual reversal. This window was contrasted to reversal window so that the reversal related decrease in alpha activity could be shown. Time windows for the exogenous task were defined by sliding pre-reversal (−850 to −550 ms) and reversal (−250 to 50 ms) windows 100 ms towards the button press to account for slight time shift of oscillatory responses between the conditions (Başar-Eroglu et al., 2016; Mathes et al., 2016). A third time window, baseline window, representing the baseline activity was extracted from the 500 ms interval preceding the artificial marker of non-reversal epochs. Activity in pre-reversal and reversal windows represented reversal-related change in alpha power relative to baseline window in decibel units (dB); whereas activity in the baseline window reflected the alpha power (μV²) during perceptual stability without baseline correction. The wavelet’s width in time was estimated as the twice of its folding time exp (−2) and found to be approximately 300 ms for 11 Hz center frequency (Torrence and Compo, 1998). Each of the time intervals interjecting those three time windows of interest were longer than 150 ms, indicating that separation of the time windows were adequate.

Group comparisons of behavioral measures and their significance levels are depicted in Figure 3. Young participants showed significantly faster reversal rates (M = 7.18 ± 1.97) compared to older adults (M = 4.83 ± 2.26) during the endogenous condition t (22) = −2.683, p = 0.014, d = 0.97. Young participants responded significantly faster to changes in motion (M = 0.58 ± 0.09) compared to older participants (M = 0.78 ± 0.22) during the exogenous condition U (N_older = 12, N_young = 12) = 33, p = 0.024. Young participants were also significantly more accurate (M = 98.55 ± 3.14) compared to older participants (M = 93.46 ± 6.28) when reporting the direction of new motion after exogenous reversals U (N_older = 12, N_young = 12) = 115, p = 0.01.

Statistical parameters of main and interaction effects are shown in Table 2. Alpha modulation from pre-reversal to reversal window at the whole cortex is depicted in Figure 4.

Overall, modulation of alpha power was not different between the groups (main effect: Group) or regions of interest (main effect: ROI). Alpha modulation was significantly lower in reversal compared to pre-reversal window (main effect: Time). A significant three-way interaction of Time × ROI × Group was found in the endogenous condition. Furthermore, linear trends of Time and ROI showed significant group differences in temporal modulation of alpha activity across different regions of interest F (1, 22) = 11.162, p = 0.003, η_p² = 0.337. Specifically, reversal-related decrease was higher at posterior ROIs for young adults but it was higher at anterior ROIs for older adults. These results are in line with planned comparisons that showed significant pre-reversal to reversal decrease in alpha at frontal t(11) = 4.759, p = 0.001, Hedge’s g_av = 0.54, central t(11) = 3.184, p = 0.009, Hedge’s g_av = 0.65, and parietal ROIs t(11) = 2.675, p = 0.022, Hedge’s g_av = 0.56 for older adults; and parietal t(11) = 3.650, p = 0.004, Hedge’s g_av = 0.63 and occipital ROIs t(11) = 6.707, p = 0.000, Hedge’s g_av = 1.31 for young adults (Figure 5). These results show that reversal-related alpha suppression was shifted towards anterior locations in older adults. Furthermore, regression analyses have shown that relative alpha power in reversal window predicts mean dwell times with frontal ROI for older and occipital ROI for young adults (Supplementary Figure S2), exclusively.

Baseline alpha power was not significantly different between the groups. However, it was significantly different between ROIs F(3, 66) = 3.503, p = 0.038, η_p² = 0.137 (Figure 5). Post-hoc comparisons have shown that frontal baseline alpha power was significantly lower than parietal baseline alpha t(23) = 3.778, p = 0.001, Hedge’s g_av = 0.36. Frontal alpha was also smaller compared to central and occipital ROIs, however, these comparisons did not reach significance after correcting for multiple comparisons (p > 0.004).

Alpha power in the reversal window was significantly lower than the activity in the pre-reversal window (main effect: Time; see Table 2). Furthermore, the magnitude of decrease in alpha was increased linearly from frontal to occipital areas (interaction: Time x ROI) as shown by linear trend contrasts of Time and ROI F(1, 22) = 7.715, p = 0.011, η_p² = 0.260. However, this contrast did not reach significance after Bonferroni correction (p > 0.004). Planned comparisons have shown significant parietal t(11) = 2.267, p = 0.045, Hedge’s g_av = 0.64 and occipital t(11) = 2.293, p = 0.012, Hedge’s g_av = 0.63 decrease in alpha activity for young adults and significant decrease at occipital ROI for older adults t(11) = 2.268, p = 0.044, Hedge’s g_av = 0.74 (Figure 5).

Baseline alpha activity was not different between the groups. There was a marginally significant effect of ROI on baseline alpha activity F(3, 66) = 3.583, p = 0.049, η_p² = 0.140 (Figure 5). Post-hoc comparisons showed that frontal baseline activity was smaller than parietal baseline activity t(23) = 3.428, p = 0.002, Hedge’s g_av = 0.39. The rest of the comparisons did not reach significance after correction (p > 0.004).

Healthy aging is related to decreased cognitive performance in several domains. Performance impairments are usually reported in the form of slowed reaction times (Salthouse, 1996; Schmiedt-Fehr et al., 2011) and degree of slowing increases as task difficulty increases (Grady, 1998; Anstey and Smith, 1999; Schmiedt-Fehr et al., 2009). Aging also decreases perceptual reversals (Jalavisto, 1964) even when eye movements and pupil size are controlled for Aydin et al. (2013) or multistable stimuli are auditory (Kondo and Kochiyama, 2018). Our findings have replicated the age-related decrease of perceptual reversals. Aging was also shown to slow the reaction times to stimulus triggered reversals in the exogenous task, replicating the general slowing of reaction times in aging.

It may be argued that age-related decrease in reversal rates is secondary to age-related slowing and decreased accuracy rates. However, post-hoc correlation analyses did not show such relationship between reversal rates, reaction times, and omissions (see Supplementary Table S1). It may also be argued that older participants did not understood the task and hesitated to press buttons during endogenous task. However, this assumption is not plausible for two reasons: (i) it was made sure that all participants have seen both perceptual interpretations and - by using a control stimulus - associated correct responses to each percept before experiment begins and (ii) accuracy rates during the control task were well above %80 for older group, eliminating the possibility of task not being understood by older adults. Therefore, we argue that the age difference in reversal rates stems from age-related changes in perceptual processing and not other changes such as age-related slowing.

Alpha responses are involved in sensory, motor, and cognitive processes (Başar et al., 1997; Klimesch et al., 1998; Başar et al., 2000). Its sensory function is operationalized as decreased activity in modality specific area following a simple sensory stimulation (Niedermeyer, 1993; Başar and Schürrmann, 1996). Another well-established finding is the decrease in alpha activity during and/or in anticipation to various cognitive tasks (Klimesch, 1997; Babiloni et al., 2006). As Başar et al. (1997) and many other researchers pointed out, there is no single and general pattern of alpha activity that correlates with all the different cognitive processes (Klimesch, 1999; Başar-Eroglu et al., 2016). Instead, there are distinct and distributed alpha networks that respond to specific events in different alpha sub-bands (Klimesch, 1999; İşoǧlu-Alkaç and Strüber, 2006). Here, we aimed to identify the age-related changes in alpha activity which is thought to be related to perceptual maintenance and destabilization of multistable percepts.

Alpha activity displays a reliable pattern in response to perceptual reversals in multistable perception tasks. Studies conducted on healthy young adults have repeatedly shown gradual decrease in alpha activity at posterior areas that starts approximately 1 s before the motor response that indicate the reversals (İşoğlu-Alkaç et al., 2000; Strüber and Herrmann, 2002; Başar-Eroglu et al., 2016). This was considered to represent the destabilization of activity in neural populations due to adaptation. This interpretation is rooted in findings of studies with different multistable stimuli (İşoğlu-Alkaç et al., 2000), data analysis methods (Mathes et al., 2010), and brain imaging techniques (Strüber and Herrmann, 2002). Our findings replicated the gradual decrease in parieto-occipital alpha activity during endogenous reversals in young adults. However, older adults showed a topographically distinct alpha response; occipital alpha response was diminished and instead it shifted to frontal, central, and parietal areas. Furthermore, this was reflected in regression analyses such that mean dwell times were predicted by frontal alpha in older and occipital alpha in young adults (Supplementary Figure S2). These supported our hypotheses regarding the diminished occipital response and compensatory alpha activity in older adults. Analysis of baseline alpha amplitudes did not show between-group differences in alpha activity that could have affected this topographical shift. Therefore, these findings indicate an age-related change in task driven alpha responses that are independent of baseline alpha activity.

A similar age-related shift in alpha topography was reported in previous literature: Kolev et al. (2002) employed a simple visual stimulation task and reported increased alpha (7–15 Hz) amplitude and phase-locking at anterior areas of middle aged adults (mean age = 53.6 ± 2.2) compared to younger adults. Furthermore, amplitude and phase-locking at occipital location were decreased. Similar findings were obtained for auditory stimulation, anterior alpha enhancement and phase-locking was increased in middle aged adults (Yordanova et al., 1998). Kolev et al. (2002) argued that sensory information within older brains is transferred to associative areas to compensate for reduced functionality of sensory brain areas. These studies show the same topographical pattern observed in this study even though the signal decomposition and data analysis methods (e.g., digital filtering, root mean squared-enhancement factor, phase-locking) are different than the ones applied in this study (e.g., Morlet wavelet convolution, baseline normalized time-frequency power). This indicates to a general age-related cortical reorganization of alpha networks in sensory and perceptual processing.

Studies on different animal species also show age-related functional decrement of neurons in the visual cortex. Signal to noise ratio of old V1 cells are decreased in rats (Ding et al., 2017), cats (Hua et al., 2006) and monkeys (Schmolesky et al., 2000; Yang et al., 2009). Shared conclusion of these studies was that aging decreased stimulus selectivity of older neurons due to reduced ability to retrieve sensory signals because of the noisy ongoing activity. In fact, human perceptual discrimination studies support this interpretation. A simple perceptual decision making study reported age-related slowing of reaction times both in visual and auditory modalities (Laurienti et al., 2006). Another study showed decreased perceptual discrimination accuracy in older adults, again, both in auditory and visual modalities (O'Brien et al., 2020). Literature indicates that decreased efficacy of sensory areas prolong or, with increased task demands, limit the perceptual discrimination capabilities in older animals, including humans. It appears that whole brain networks compensate for this inefficacy by recruiting additional alpha networks across the cortex.

Current findings might hint to a reason that explains decreased reversal rates in older adults. Traditionally, reversals thought to occur when the adaptation of percept-specific neural populations reach a certain threshold (Kohler, 1940; Orbach et al., 1963; Leopold and Logothetis, 1999; Kornmeier and Bach, 2012). As mentioned before, studies have considered reversal-related alpha response to be an index of adaptation process. This means that recruitment of additional alpha networks increases the number of neural populations that need to reach the adaptation threshold before a reversal can occur. Therefore, neural adaptation will take longer for older adults because compensatory alpha recruitment increases the number of neural populations to be activated. We argue that increased number of active alpha networks might have prolonged the adaptation process and dwell times in older adults, resulting in lower reversal rates.

The most apparent difference in alpha response between exogenous and endogenous reversals is the time course of activity: decrease in former is steeper and shorter compared to the latter (Strüber and Herrmann, 2002; Mathes et al., 2010). Strüber and Herrmann (2002) argued that alpha activity preceding exogenous reversals is primarily a stimulus-driven response. This interpretation was in line with the results of Başar-Eroglu et al. (2016) where the difference in alpha response between healthy controls and schizophrenia patients was apparent during endogenous but not during exogenous reversals. In line, no impairment of stimulus-driven processing itself was expected in patients, but an impairment of internal processing of perceptual information maintaining and switching between endogenous stimulus perceptions. Similarly, current results displayed no difference in alpha response during exogenous reversals for the aging brain. However, aging is known to change topography of sensory alpha response (Yordanova et al., 1998; Kolev et al., 2002), which seem to contradict with the current results for the exogenous stimulus change. Schmiedt-Fehr et al. (2009) also dealt with a similar contradiction in a Go/NoGo paradigm and argued that involvement of higher-order processes might have improved the sensory function of alpha networks. We could argue that top-down responses to exogenous reversals in older adults might have eliminated the need for compensatory alpha activation. Future studies should investigate the effect of varying degrees of stimulus strength on top-down activation and its relationship to compensatory alpha responses to further explore this issue. Alternatively, choosing 11 Hz as the center frequency and not dividing alpha into sub-bands might have masked the group differences in anterior alpha responses in the current study.

There was an unequal number of males and females within the sample. Therefore, our findings and interpretations should be evaluated under this constraint. Furthermore, aging and its effects on functional neural activity is a slow and gradual process. This study investigated young adults and older adults exclusively. However, including middle aged and older adults with even more advanced ages (+80 years of age) could have shed light into the gradual effects of aging on brain oscillations. Future studies should consider the limitation of severely decreased reversal rates after 80 years of age and design experiments accordingly.

Previous studies reported a shift in the alpha frequency peak with advancing age. There are also studies that specified that low and mid alpha ranges are the most involved during multistable perception. We have not used individual alpha frequencies or analyzed sub-bands of alpha activity in this study. Furthermore, we argue that this could have been a reason for the lack of between-group difference in the exogenous task. Future studies should delineate the effect of individual alpha frequencies and sub-bands of alpha band to examine if they alter age-related differences in endogenous and exogenous tasks.

Multistable perception requires the collaboration of top-down and bottom-up processes which are represented in multiple frequency bands across the selectively distributed oscillatory networks. Therefore, investigating the interaction of multiple frequency bands can provide important insights that cannot be captured by analysis of a single oscillatory activity.