1. Participants

Twelve naive right-handed volunteers with normal vision took part in the study (6 women and 6 men, mean age 29.3±9.1 years). All participants provided informed written consent and received a financial compensation for their participation. Two participants who misused the response buttons were excluded from the analyses. All procedures were approved by the local research ethics committee (Comité de Protection des Personnes Île-de-France VI, Paris, France).

2. Stimuli and procedure

The stimuli were presented via a mirror at the centre of a rear projection screen using a calibrated video projector (1024×768 pixels; refresh rate, 60 Hz) located outside the shielded recording room. The distance between subjects' eyes and the screen was 0.85 m. The subjects' head was inclined so as to favor the recordings of occipital MEG sensors [37]. The stimuli were composed of two pairs of horizontal and vertical bars (mean luminance 45.7 cd/m²; length 5.0 degrees of visual angle, dva thereafter; mean distance from the centre 3 dva) displayed on a grey background (mean luminance 23.6 cd/m²) and distributed around the central fixation point so as to form a square shape (6.8×6.8 dva) with invisible corners (Figure 1A). During a trial, the bars remained static for a short variable period (450–550 ms) and then moved for 1200 ms as follows: the horizontal bars oscillated in phase along a vertical axis at f1 = 2.3 Hz, while the vertical bars oscillated in phase along a horizontal axis at f2 = 3 Hz. This stimulus was expected to trigger the responses of direction selective cells at different harmonics of the bar motion frequencies, thus allowing the identification of neural populations responding to the vertical and horizontal motions. Motion amplitude was identical for the horizontal and vertical motion and equal to 1.2 dva.

As the perceptual binding of the component motions into a rigid moving shape is known to depend on the luminance ratio between the centre and line-ends of the bars [31], we designed four conditions, each characterized by a triangular distribution of luminance along the bars, as shown in Figure 1A. High-luminance line ends favor the perceptual segmentation into unbound oscillating bars, while lower-luminance line-ends favor their perceptual integration into a single moving shape. Preliminary behavioral experiments were conducted to choose luminance distributions yielding graded percepts: from strongly bound (condition 1, Video S1) to strongly unbound (condition 4, Video S2). Note that the mean luminance and motion distribution of the bars is identical in all conditions. These stimuli, when static, are hardly discriminated on the sole basis of their luminance distribution (Figure 1). In contrast, when the bars oscillate periodically along cardinal axes, the different stimuli elicited highly distinguishable perceptual states: a square with invisible corners moving rigidly along a Lissajou's trajectory - bound percept-, or pairs of bars moving independently along the vertical and horizontal axes - unbound percept. One observer who noticed the differences in luminance distribution was removed from the analyses. Although long observation of the present stimuli results in bistable percepts [38], [39], a procedure using physically different stimuli to favor different –bound/unbound- percepts was preferred for the following reasons: 1. In an fMRI study with similar stimuli, endogenous bistability or physical induction of perceptual fluctuations did not entail significantly different BOLD signals [38]. 2. Although bistable stimuli provide an elegant way of probing the mechanisms underlying different percepts with a single stimulus, the lack of a common base line temporally close to the stimulation, the intra- and inter-subject variability of the durations of each perceptual episode and the need to record perceptual switches on-line, which implies a motor response at any time, pose difficult challenges for the data analyses (e.g. contamination by the motor preparation and execution).

The time flow of experimental trials was as follows (Figure 1B): A fixation point was first presented on the screen for 1.5 second (t = −1.5 to t = 0). At t = 0, four static bars were displayed for a duration varying randomly between 450 and 550 milliseconds to avoid the confounding effect of stimulus and motion onset. This static phase was followed by 1.2 second of motion where both pairs of bars oscillated at fixed, although different, frequencies, along an axis orthogonal to their orientation. Finally, a response screen with three color-coded disks presented side by side was displayed to indicate which response was associated with each button of the keypad: black for a rigidly moving square –“bound percept”-, white for independent bar motions –“unbound percept”-, grey for an indecipherable percept or to signal an intrusive perceptual switch –“unclassified trials” thereafter. In order to minimize artifacts associated with motor preparation, the horizontal position of the three disks was randomly shuffled on each trial so that observers had to wait for the response screen before encoding and making their motor response. Each subject underwent 8 runs of 60 trials each (15 trials per condition) for a total of 120 trials per condition.

3. MEG recordings

Continuous magneto-encephalographic signals were collected at a sampling rate of 1250 Hz, using a whole-head MEG system with 151 axial gradiometers (CTF Systems, Port Coquitlam, British Columbia, Canada), and low-pass filtered on-line at 300 Hz. Before each run, head localization was measured with respect to the MEG sensors using marker coils that were placed at the cardinal points of the head (nasion, left and right ears). Eye movements were recorded with an ISCAN eye-tracking system (240 Hz sampling rate). We also recorded the signal of a photodiode that precisely detected when the bars appeared on the screen. This allowed us to correct for the time delays introduced by the video projector (∼24 ms) and to compute event-related magnetic fields (ERFs) precisely time-locked to the real stimulus onset.

4. Data analyses

Data were first pre-processed using both CTF and in-house software (http://cogimage.dsi.cnrs.fr/logiciels/). Trials contaminated by eye movements, blinks, or muscular artifacts were rejected off-line on visual inspection of ocular and MEG traces (as a result, 30% of the trials, evenly distributed amongst the 4 conditions, were discarded from further analyses: 27.6%, 30%, 29.5% and 31.4% for condition 1 to 4 respectively, corresponding to bound trials: 1582, unbound trials: 1436 trials; condition 1: 869 trials; condition 2: 840 trials; condition 3: 845 trials; condition 4: 823 trials). Time zero was set at the onset of motion using a photodiode signal. Global analyses were performed on all non-rejected trials independently of observers' percepts. Contrasts of MEG activity were also computed between trials classified as bound or unbound; excluding unclassified trials (8.9%). Analyses performed on averaged signals (SSVEF) time-locked to the stimulation have been presented elsewhere [16]. We here analyze the oscillating activity in the alpha, beta and gamma ranges to highlight variations induced by a bound versus an unbound percept.

1. Behavioral results

The averaged distribution of observers' reports is presented in Figure 1C as a function of the 4 contrast conditions. As it can be seen, observers mostly perceived a single moving square when line-end contrast was low and perceived disconnected moving segments when line-end contrast was high. Overall, only few trials (8.9%) were unclassified, suggesting observers were confident in their choices and reliably classified their perceptual state over the duration of a trial. It is worth noting that whereas line-end luminance increases linearly across the first three conditions, observers' judgments show a discontinuity in the bound/unbound classification between condition 2 and 3, confirming that a small change in bar-end contrast entails drastic perceptual modifications. An ANOVA (3×4 factors) conducted on these data indicated a significant interaction between perception and condition (F = 114.79, p<0.05; η2 = 0.92). Additional analyses for each percept (1 factor, 4 conditions) showed a significant effect of the conditions on the response rate for the bound (F = 81.28, p<0.05; η2 = 0.891) and unbound (F = 93.58, p<0.05; η2 = 0.908) percept but not for unclassified percepts (F = 0.93, p = 0.44).

2. MEG results

Analyses of the steady state responses evoked by the oscillatory bar motions at the fundamental (2.3, 3 Hz), first harmonics (4.6, 6 Hz) and their intermodulation products have been presented elsewhere [16]. Briefly, increased power at the 10.6 Hz intermodulation product during bound states was found in frontal sensors, while the response power at the motion related frequencies of interest did not significantly differ as a function of perception on occipital or parietal sensors.

We here focus on the activities induced by the different perceptual states in three frequency bands (alpha: 8–12 Hz; beta: 15–25 Hz; gamma: 55–85 Hz), whose limits were first identified using all the trials (see above section Data analyses).

The analysis conducted using all the trials revealed modulation of gamma activity in occipital sensors (see figure 2 in [16]). In addition, decreased activity over the left motor cortex in the beta band (15–25 Hz) was observed, as expected considering that participants reported their percepts using their right hand and prepared to respond.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. MEG results.

A. Results of the spatio-temporal clustering analysis (see data analyses). In these plots, the t-values contrasting bound as compared to unbound percepts are shown for all the sensors as a function of time for alpha, beta and gamma oscillations. No significant differences were found for alpha (p>0.138) and gamma (p>0.176) frequencies. A significant cluster corresponding to more ample oscillations for bound as compared to unbound percept were found for the Beta band (p<0.01). Three topographies of sensors corresponding to three periods (230–550 ms, 550–750 ms and 750–1050 ms) are also shown. In these topographies, the color code denotes the time during which a sensor of a significant cluster was significantly different in bound as compared to unbound percepts.

https://doi.org/10.1371/journal.pone.0095541.g002

The clustering analyses did not reveal any significant changes in the alpha and gamma power as a function of perceptual reports (alpha p>0.138; gamma p>0.176). In contrast, a sustained modulation of beta power (p<0.01) was found during motion stimulation. Figure 2 presents the t-values obtained when contrasting bound and unbound percepts for the alpha, beta and gamma frequency bands for all the MEG sensors as a function of time. As it can be seen, alpha and gamma oscillations are little modulated by perception and are not significantly different during bound and unbound percepts, in contrast with beta oscillations. Topographical maps of the significant cluster at different time intervals are also shown.

As it can be seen, the differences in beta power were overall sustained, although they smoothly developed during the time course of a trial over the scalp. Differential beta power emerged about 230 ms after motion onset over occipito-parietal sensors then developed toward centro-parietal around 550 ms, before reaching left frontal sensors about 750 ms after motion onset. Difference in beta power then dropped around 1100 ms after motion onset, slightly before the end of the visual stimulation.

The topography of all sensors belonging to the identified cluster showing significant beta modulations during motion stimulation is shown in figure 3A in which the color code denotes the time during which a sensor was significantly active. As it can be seen, activity is mostly localized on central-parietal sensors. To easily visualize the differences between alpha, beta and gamma oscillations, the time frequency plots encompassing all frequency bands of interest is shown on figure 3B. To more easily visualize the relationships between perceptual reports and beta activity, the later was normalized and plotted for each participant as a function of the experimental conditions. As it can be seen, the distribution of normalized beta power (figure 3D) closely resembles the distribution of reports of bound percepts (figure 3C). Plotting beta activity against the percentage of trials classified as bound (Figure 3E) for conditions 2 and 3 that are physically very similar (Figure 1A) further indicates that lower beta power is associated to fewer bound reports for the majority of observers (7/10), while higher beta power is associated to more frequent bound reports.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Comparisons between perceptual performance and beta power.

A: Topography of the 63 sensors with significant difference between bound and unbound trials. In this plot, the color code denotes the time during which a sensor was significantly different in bound as compared to unbound percepts. B. Time frequency plots obtained by averaging sensors (n = 63) showing increased beta power for bound trials. C, D. Comparison of behavioral data (C) and normalized beta power (D) for each observer. The doted black line represents averaged data; light grey lines represent the data for each observer. E. behavioral responses for conditions 2 and 3 plotted as a function of normalized beta power. Normalized beta power increases with perceived motion coherence for all but 3 observers. F. Results of trial-by-trial classification (see method) using beta power averaged across frequencies (17–22 Hz) and time (100–1100 ms; see text for details). Significant classification (*, p<0.05) is obtained for 7 out of 10 observers.

https://doi.org/10.1371/journal.pone.0095541.g003

2.2 Source reconstruction.

Reconstructing the sources of both the gamma and beta power for all the trials and for the differences between bound and unbound trials refined the loose localization of these activities on the scalp. As a result, shown in figure 4, we found that the sources of the gamma activity were mainly confined to the occipital lobe while the sources of the decreased beta power, found when all the trials are collapsed, lie over the left motor cortex. In contrast, the sources of the difference in beta power between trials seen as bound and unbound are mostly confined to the central-parietal cortex, although some sources around the motor cortex also show increased beta power related to the bound/unbound reports. Overall, the distributions of these sources confirm the conclusions drawn from the sensor activity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Reconstruction of the sources of gamma and beta band oscillations.

Only clusters of contiguous sources (n>50) with power greater than 50% of maximum activity are shown. A. Sources of gamma power (range 55–85 Hz) computed using all the trials are mainly located within the occipital lobe. B. Power differences between bound and unbound trials fail to reveal sources in the gamma band. C. Sources of beta power (15–25 Hz) computed using all the trials show decreased beta power over the left motor cortex, as expected given that observers used their right hand to report their perception. D. Sources of beta power computed from the differences between bound and unbound trials are mostly distributed over central-parietal cortex.

https://doi.org/10.1371/journal.pone.0095541.g004

1. Physical contrast or perceptual motion integration?

In this study, small modulations of line-end bar contrast entailed clear and salient modifications in perception, despite the overall averaged contrast being kept the same across conditions. One may thus be concerned that the modulation of beta power described here is caused by these physical modulations alone. Disentangling perceptual states from physical modifications, for instance by comparing bound and unbound reports for the same condition, is however uneasy because the number of trials reported as bound or unbound is not evenly distributed. For example, the conditions 2 and 3, although physically very similar (12.5% and 19.3% line-end contrast respectively), are mostly classified as bound and unbound, respectively, leaving few trials to dissociate the effects of physical changes and perceptual reports, such that statistical analyses are irrelevant. However, if line-end contrast per se accounted for the change in Beta power, one would expect that Beta power increases -or decreases- in proportion to increasing contrast, which is not the case. In addition, if contrast was effectively driving the cortical responses, the signals directly evoked by the oscillatory motions should also be modulated by contrast. We previously reported that the responses at the fundamental and 1^st harmonics of the oscillatory motions were not statistically different for the different conditions [16], disproving the idea that differences in physical contrast entailed significantly different evoked responses. Moreover, it is unclear why physical contrast would modulate oscillatory power specifically in the beta band and not in other frequency bands, gamma band in particular [43]. The drastic influence of small contrast differences on perceptual integration calls for non-linear effects able to shift a percept of unbound segmented moving bars into that of a bound integrated moving object. Although caused by changes in line-end contrast, such non linear effects presumably reflect cortical processing of the incoming inputs, possibly related to released end-stopped inhibition at low line-end contrast [33], [52], which in turn facilitates the neuronal communication underlying motion integration across time and space, while enhanced end-stopping for high-contrast line-ends may prevent the integration of contour and motion into a whole. Although this may indicate that beta oscillations are related to modulation of end-stopping inhibition, further studies are needed to address this issue.

2. Perceptual integration or other cognitive processes?

Several studies comparing EEG and MEG activity elicited by attended and unattended stimuli report attention dependent oscillations in the alpha and/or gamma range, or in the steady-state visually evoked responses power, SSVEP [12], [44], [53]. Can a different distribution of attention during trials seen as bound or unbound account for the present results? Although participants were to attend similarly to all conditions to perform the task, one cannot exclude that allocation of attention differed during bound, as compared to unbound, percepts. According to previous results, a percept dependent shift in attention should elicit modulations of alpha, gamma power and/or SSVEP. The analyses done to test this eventuality indicate that gamma power, although strong and reliable during visual stimulation, was independent of perceptual reports. The lack of significant differences in alpha power for the different percepts similarly argues against an effect of attention. If oscillations in these frequency bands reliably index an increased attentional load toward object-like stimuli, the lack of significant differences in alpha and gamma activities suggests that allocation of attention did not consistently differ as a function of perception in this study. Finally, previous studies using temporally modulated stimuli [54] reported attentionnal power modulations at the harmonics of the periodic stimulations. If attention was to account for the differences reported here, one would also expect similar differences in the activity evoked by the oscillatory bar motions, which was not the case [16].

Another possibility is that beta activity reflects the development of decisional processes, at stake when asked to classify stimuli. For instance, reaching a decision might be more difficult in condition 2 and 3 that are physically more similar than conditions 1 and 4 that are more different. More difficult decisions for uncertain stimuli could result in differences in beta power, as has been reported [51] in a motion detection task that required integrating motion evidence over several seconds to cross a motion threshold. In this situation, gamma power correlates with motion strength, while increased beta power appears to reflect the accumulation of evidence leading to a decision. In the present study with supra-threshold stimuli, beta band oscillations do not seem to reflect decision uncertainty, as beta power is different for the more uncertain conditions 2 and 3 and for the more certain conditions 1 and 4 (Figure 3E). In addition, observers could use an “unclassified” response button whenever they found it hard to classify their percept. Those few “unclassified” trials were evenly distributed across conditions (Figure 1) and discarded from the analyses, such that more uncertain or difficult decisions may not have contributed much to the beta modulation reported here. Finally, a decision has to be made for bound as well as unbound percepts. It is unclear why neural activity would differ much for bound and unbound decisions, except if it is related to the perceptual content rather than the decision process itself. Altogether, these considerations weaken the possibility that enhanced beta power during bound reports reflects decisional processes independently of perceptual processing.

Alternately, observers could have made their decision quickly and maintained their choice until the response screen appeared, which necessitates keeping their choice in memory until a motor response can be made. In line with previous reports relating cortical oscillations and memory [55], [56], beta activity could index this process, a view compatible with the observation that the difference in beta power reaches a maximum around 800 ms after motion onset. Although we cannot refute this interpretation at this stage, it is unclear why memory load would significantly differ for bound and unbound decisions (except if the perceptual content rather than the decision itself is kept in memory). Similarly, motor preparation and motor response are unlikely to explain the present findings because motor preparation is needed whatever the perceptual report and can only be produced after the randomized stimulus-response mapping screen was displayed, after the end of the visual stimulation.

Overall, because the cognitive processes needed to perform the task are balanced across conditions and perceptual reports, we think that the paradigm used in this study is well suited to recruiting and isolating the neural correlates of perceptual form/motion integration and limits the possibility that uncontrolled cognitive processes elicited unbalanced activity for bound and unbound reports.

3. Perceptual and beta power dynamics

If beta power index perceptual integration, the observation that the difference in beta power between bound and unbound trials emerges about 230 ms after motion onset is puzzling, as perceptual integration is expected to emerge soon after motion onset. However, psychophysical studies of the dynamics of motion integration with drifting plaids or with aperture stimuli similar to those used herein indicate that motion integration develops slowly (100–300 ms), as evidenced by dynamic shifts in perceived direction [57] or coherence [58], or directly probed through the recordings of direction selective MT cells [59]. In this study, the beta power for bound and unbound percept diverges after a delay comparable to perceptual dynamics and stabilizes after motion integration is completed and perception reaches a stable state. In this view, beta oscillations could help sustaining the perceptual outcome of a bound shape until a response is given.

The present results add to recent findings of consistent relationships between cortical oscillations and perceptual and cognitive processes [2], [3], [53], [60]. A consensual view is that oscillations in different frequency bands are spectral fingerprints [36] whose characteristics are constrained by the architecture of the neural connectivity within and between regions [36], [61]. In this regard, cortical oscillations provide insights into both the processes involved in a perceptual task and the neural substratum of the underlying dynamic ensemble [3]. Gamma oscillations (35–100 Hz) are often considered to index local sensory processes within a cortical region, owing to short range excitatory and inhibitory interactions elicited by an incoming sensory input (e.g. in layer 4), a view supported by modeling [8] as well as electrophysiological and neuroimaging studies [4], [5], [62]. In contrast, beta oscillations (15–30 Hz) are thought to reflect long-range interactions facilitating information transfer between cortical regions [3], [11], and appear to mainly originate from feedback projections in superficial and deep cortical layers [63], [64]. It has further been suggested that beta oscillations develop and maintain over time so as to sustain a “status quo” [21] during which perception is stabilized, providing matter to decision and action. The present results fit well with this scheme. Sustained gamma oscillations found over the occipital lobe when including all the trials possibly index the local interactions recruited to encode each oscillating bar independently. In line with previous reports, larger beta activity could reflect increased neuronal communication across neuronal populations coding for the different bars in different cortical sites during the encoding of a moving bound shape, which engages mechanisms of motion integration, contour completion, surface filling-in, depth ordering as well as the computation of border ownership [39], [65]–[68],[34], [51]–[54]. In line with the observation that neurons in parietal regions receive convergent projections from visual areas responding to global motion [69], [70] or to shape [71], enhanced beta oscillations could facilitate the integration of oscillating bars into a moving shape in central-parietal regions. Analyzing long-range coherence, synchronization and functional connectivity between sources showing beta modulations and sources involved in processing the stimulus (as revealed by the SSVEF analysis and alpha and gamma power in all conditions) could further reveal how beta oscillations emerge from the visual stimulus processing and contribute to perceptual decision.