Abstract

Predictions strongly influence perception. However, the neurophysiological processes that implement predictions remain underexplored. It has been proposed that high- and low-frequency neuronal oscillations act as carriers of sensory evidence and top-down predictions, respectively (von Stein and Sarnthein 2000; Bastos et al. 2012). However, evidence for the latter hypothesis remains scarce. In particular, it remains to be shown whether slow prestimulus alpha oscillations in task-relevant brain regions are stronger in the presence of predictions, whether they influence early categorization processes, and whether this interplay indeed boosts perception. Here, we directly address these questions by manipulating subjects' prior expectations about the identity of visually presented letters while collecting magnetoencephalographic recordings. We find that predictions lead to increased prestimulus alpha oscillations in a multisensory network representing grapheme/phoneme associations. Furthermore, alpha power interacts with stimulus degradation and top-down expectations to predict visibility ratings, and correlates with the amplitude of early sensory components (P1/N1m complex), suggesting a role in the selective amplification of predicted information. Our results thus indicate that low-frequency alpha oscillations can serve as a mechanism to carry and test sensory predictions about letters.

Introduction

Everyday life and laboratory experiments suggest that perception is strongly influenced by previous knowledge and expectations (Series and Seitz 2013). This is reflected in a variety of perceptual effects such as the “light-from-above prior,” which makes us perceive some convex spheres as concave because of our lifelong experience that light comes from above. Starting with Helmholtz (1867), such observations have led to the conceptualization of perception as an inferential process, in which sensory input is integrated with prior knowledge, the result of which reaches awareness. In line with a strong influence of expectations on perception, recent studies have shown that short-term priors (expectations) also affect perception, improving the sampling of sensory information, lowering visibility thresholds, and speeding up access to awareness (Melloni et al. 2011; Moca et al. 2011; Aru et al. 2012). This occurs not only in vision, but also in other senses, for example, audition (Sivonen et al. 2006; Arnal and Giraud 2012).

Despite behavioral and neuroimaging evidence for the influence of expectations on perception (Series and Seitz 2013), little is known about the underlying neuronal mechanisms. Previous studies have shown that internal states that require top-down control such as attention, mental imagery, divergent thinking, or the retention of information in working memory are all accompanied by increased oscillatory power in the alpha frequency band (Bastiaansen et al. 2002; Jensen et al. 2002; Cooper et al. 2003; Raghavachari et al. 2006; Benedek et al. 2011; Johnson et al. 2011; Jauk et al. 2012). Increased alpha oscillations have also been observed during states of expectancy in cat visual cortex (Chatila et al. 1992), and during top-down attention in macaque inferotemporal (IT) cortex (Mo et al. 2011). Furthermore, expectations affect interareal synchronization in the alpha band (Freunberger, Klimesch, et al. 2008), with phase lags and laminar distributions compatible with a top-down effect (Bernasconi and Konig 1999; von Stein and Sarnthein 2000; von Stein et al. 2000). This has led to the proposal that top-down processing is implemented by low-frequency oscillations (von Stein et al. 2000; Wang 2010; Arnal and Giraud 2012; Bastos et al. 2012). Several observations support this idea: First, alpha power increases before stimulus onset in macaque IT cortex when animals attend to visual stimuli, which facilitates visual processing (Mo et al. 2011). Second, alpha coherence has been observed between deep layers of association cortex and superficial layers of primary visual cortex in cat visual cortex under conditions of expectancy, in line with the known anatomical sources and targets of cortical feedback projections (von Stein et al. 2000). More recently, van Kerkoerle et al. (2014) have observed a similar pattern of connectivity in the alpha band in macaque visual cortex when animals were performing a demanding attention task. Finally, Granger Causality analyses of low-frequency oscillations in humans further substantiate a privileged role of alpha oscillations in top-down processing (Fontolan et al. 2014). Taken together, this makes alpha oscillations a viable candidate for enacting sensory predictions.

It has been proposed that alpha oscillations exert their top-down effects through targeted inhibition. Namely, alpha oscillations are thought to reflect inhibitory processes operating under top-down control that direct cortical activation and thereby the flow of information (Klimesch 2012). Specifically, alpha oscillations may impose phasic inhibition in task-relevant areas, which increases the effective signal-to-noise ratio (SNR) by selecting highly excitable, task-relevant neurons, while simultaneously silencing task-irrelevant neurons with lower levels of excitation (Klimesch 2011, 2012). This could promote the online maintenance and amplification of expected content and simultaneously the suppression of unexpected inputs.

Once inputs arrive in cortex, they can be tested against sensory predictions, which are reflected in the amplitude of the P1/N1 event-related component (Hopf et al. 2002; Dambacher et al. 2009). Alpha oscillations may also play a critical role in this process, as several studies have documented a tight relationship between alpha oscillations and the P1/N1 complex. In particular, it has been shown that the P1/N1 complex has a dominant frequency in the alpha range; that alpha-band oscillations propagate as a traveling wave (Fellinger et al. 2012); that alpha power predicts P1 latency (Freunberger, Holler, et al. 2008); that phase alignment of alpha oscillations predicts P1/N1 latencies and amplitudes (Gruber et al. 2005); and that N1 amplitude increases when stimuli are presented during the positive-going alpha cycle (Barry et al. 2003). Finally, evoked alpha oscillations in the P1/N1 time window have been shown to influence early perceptual categorization and detection performance (Palva et al. 2005; Klimesch 2011). Hence, ongoing alpha oscillations might be centrally involved not only in the implementation but also in the testing of predictions.

Based on a mechanism originally proposed to explain the effects of anticipatory top-down corollary discharge signals in macaque visual cortex (Ito et al. 2011) and the well-established idea that alpha oscillations reflect inhibition (Klimesch 2011), we suggest that following the presentation of a stimulus, alpha-band inhibition is reduced such that highly excitable, pre-activated neurons are quickly and synchronously released from inhibition, leading to stronger synchronization and thus an increased evoked response. In contrast, neurons that are not pre-activated and less excited take longer and more variable time to recover from inhibition, which decreases their impact on downstream areas and results in a diminished evoked response. This mechanism could, in principle, account for the observed relationship between alpha oscillations and the amplitude of the P1/N1 complex (Makeig et al. 2002; Gruber et al. 2005; Klimesch 2011).

In this study, we investigate the previously underexplored role of alpha oscillations in top-down expectations by combining magnetoencephalography (MEG) with a task in which we manipulated sensory evidence for and the predictability of letters. This allows us to test several open questions about alpha oscillations, namely whether prestimulus alpha oscillations in task-relevant brain regions are stronger in the presence of predictions, whether they influence early categorization processes as reflected by the P1/N1 complex, and whether this interplay indeed boosts perception. We find that predictions lead to increased prestimulus alpha oscillations in a multisensory network representing grapheme/phoneme associations. Furthermore, alpha power interacts with stimulus degradation and top-down expectations to predict behavioral visibility ratings, and correlates with the amplitude of early evoked components, that is, P1/N1m complex, suggesting a role in the selective amplification of predicted information. Taken together, this suggests that alpha oscillations can serve as a mechanism to carry and test sensory predictions.

Materials and Methods

Subjects

Twenty-five healthy volunteers (9 males, mean age 25 years, range 18–39 years) participated in this study. Twenty-two were right-handed as assessed by the Edinburgh Handedness Inventory (Oldfield 1971), all reported normal or corrected-to-normal visual acuity, no history of neurological or psychiatric disorders, and gave written informed consent before participation, in accordance with the Declaration of Helsinki. Subjects were compensated with 15€ per hour for their participation in the study.

Task and Stimuli

The aim of our study was to measure the effect of perceptual expectations on the threshold of perception. To that end, we presented blocks of 11 consecutive stimuli, all containing the same token. We refer to this as sequence (Fig. 1A). During a sequence, we first increased the visibility of the stimuli by increasing the sensory evidence from trials 1 to 6, up to a point of maximal visibility (“ascending sequence”). Following that, the order of stimulus presentation was reversed and consequently visibility was decreased (“descending sequence”). Note that the same stimuli were used for the ascending and the descending flank of the sequence; hence, they only differed in the history of stimulation. Increasing the visibility of the stimuli in the ascending sequence allowed us to induce target-specific expectations, which were created as soon as the target became visible. In turn, when reducing the visibility of the following items in the descending sequence, the effect of these expectations on perception can be measured as increments in visibility that exceed the visibility afforded by stimulus evidence alone. This is referred to as perceptual gain or hysteresis (Kleinschmidt et al. 2002). There was no temporal cue that subjects could use to predict the start or the end of a sequence, as the intersequence interval was of the same duration as the intertrial interval. Also, the gain in visibility due to the presence of stimulus-specific expectation (see Results) further masks predictability within and between sequences, as it breaks the strict linearity between stimuli stemming from stimulus information alone. The experiment consisted of 6–7 runs, each containing 19 sequences of 11 stimuli (∼1254 trials per subject), with self-paced breaks between runs.

Figure 1.

Experimental design and behavioral results. (A) Example stimuli used in the study. Stimuli were presented in the upper left quadrant at 4° eccentricity. Bottom-up information was manipulated by parametrically varying the ratio of dot density between the figure and the background. Six levels of degradation were used. To create top-down effects, we increased and subsequently decreased the visibility of an initially hidden letter. As stimulus recognition occurs during the rising phase (ascending sequence), target-specific expectations are generated. By reducing the visibility of the following items (descending sequence), these expectations (or priors) interact with the sensory information of the stimuli. (B) Behavioral results. The x-axis shows the 6 stimulus degradation levels. The y-axis indicates the percentage of seen responses. The ascending sequence is shown in black, and the descending sequence is shown in gray. Error bars indicate SEM. Note the horizontal shift of the psychometric function for the descending sequence, which carries the top-down information. The inset shows the average visibility threshold (inflection point) per sequence obtained by fitting a sigmoid function.

Figure 1.

Experimental design and behavioral results. (A) Example stimuli used in the study. Stimuli were presented in the upper left quadrant at 4° eccentricity. Bottom-up information was manipulated by parametrically varying the ratio of dot density between the figure and the background. Six levels of degradation were used. To create top-down effects, we increased and subsequently decreased the visibility of an initially hidden letter. As stimulus recognition occurs during the rising phase (ascending sequence), target-specific expectations are generated. By reducing the visibility of the following items (descending sequence), these expectations (or priors) interact with the sensory information of the stimuli. (B) Behavioral results. The x-axis shows the 6 stimulus degradation levels. The y-axis indicates the percentage of seen responses. The ascending sequence is shown in black, and the descending sequence is shown in gray. Error bars indicate SEM. Note the horizontal shift of the psychometric function for the descending sequence, which carries the top-down information. The inset shows the average visibility threshold (inflection point) per sequence obtained by fitting a sigmoid function.

A detailed description of the experimental procedures can be found in Melloni et al. (2011). In short, each trial started with the presentation of a red fixation square on a gray background (1.5–2 s). Next, the stimulus appeared for 0.5 s, immediately followed by the presentation of a question mark which prompted participants to rate the subjective visibility by a button press. The trial terminated with the subject's response. To assess visibility, we used the Perceptual Awareness Scale (PAS; Overgaard et al. 2006), which has been shown to be a highly sensitive measure for subjective awareness (Sandberg et al. 2010). The PAS maps the visibility of a stimulus onto 1 of 4 alternatives: “no experience” of the stimulus, “brief glimpse” of the stimulus but no identification, “almost clear perception” and thus identification of the stimulus, and “clear experience” of the stimulus. Before the experiment started, the experimenter extensively discussed the operationalization of the PAS with the subjects. This was done to assure consistency across participants in terms of their criterion. We emphasized the distinction between responses 2 and 3, which demarcates no recognition from explicit recognition of the stimulus. Furthermore, in order to assess the reliability of the subjects' judgments, we introduced trials in which no stimulus was present (degradation level 1), which should lead to low visibility ratings, and trials in which the stimulus was clearly visible (degradation level 6), which should lead to high visibility ratings. Importantly, signal absent and signal present trials were embedded in each sequence, thus allowing for a reliable estimate of visibility ratings for each sequence.

We have previously shown that, in this experimental paradigm, gains in visibility directly relate to stimulus-specific expectations and not to unspecific factors, such as temporal order effect or response bias (Melloni et al. 2011): First, when the temporal order is preserved but stimulus-specific expectations are completely obliterated by varying stimulus identify on every trial, there is no gain in visibility. Second, when a predictable sequence is interrupted by an unpredicted stimulus, there is no gain invisibility for that item, which rules out unspecific biases to respond with higher ratings as a consequence of temporal sequence effects or overconfidence.

Stimuli were presented on a transparent screen with a gray background located 53 cm in front of the subjects. An LCD projector (60 Hz refresh rate) located outside the magnetically shielded room was used to project the stimuli onto a screen inside the MEG cabin via 2 front-silvered mirrors. The stimulus set consisted of 43 highly familiar targets, comprised of 26 letters of the Roman alphabet, 10 numbers, and 7 symbols (for a list of specific tokens used in the study, see Supplementary Material) embedded in a random noise background (Fig. 1A). The total display subtended 13.134 × 20.626 degrees visual angle (dva). The target stimulus (2.182 dva) was presented in the upper left quadrant at 4 dva eccentricity. The visibility of the target stimulus was modulated parametrically by changing the dot density of the target shape while keeping the background dot density constant. By linearly decreasing the dot density of the target, we created 6 different stimulus degradation levels. The lowest level corresponds to a uniform field of random dots, whereas the highest level corresponds to a clearly visible stimulus. The same set of stimuli with their 6 degradation levels was used in all participants. To avoid low-level sensory adaptation, we jittered the precise location of the target stimuli, and of the individual dots, which were randomly distributed for every image. All stimuli were created using Matlab (The MathWorks). Presentation software (Version 10.3) was used for stimulus presentation and response collection.

Data Acquisition

Ongoing brain activity was recorded using whole-head MEG with 275 axial gradiometers (Omega 2005, VSM MedTech Ltd, Coquitlam, BC, Canada) in a magnetically shielded room with dimmed light. Brain signals were sampled at 600 Hz with a hardware antialiasing filter at 150 Hz in a synthetic third order axial gradiometer configuration (Data Acquisition Software Version 5.4.0, VSM MedTech Ltd). Head localization relative to the MEG helmet gradiometer was measured before and after each run using coils that were placed at both the nasion and the preauricular points. Before the start of each run, subjects were instructed to adjust their head position to the one originally measured at the start of the experiment. Head position was stabilized using cushions placed lateral to the subject's head. Runs with head movements exceeding 5 mm were excluded from data analysis. We monitored eye movements and blinks during the recordings with 2 pairs of electrooculogram (EOG) electrodes. Behavioral responses were recorded using a fiber-optic response device (Lumitouch, Photon Control).

Individual high-resolution structural magnetic resonance images (MRIs) were acquired on a 3-T Siemens Allegra scanner (Siemens, Erlangen, Germany) using a T1-weigthed magnetization-prepared rapid acquisition gradient-echo (MPRAGE) sequence (160 slices; time repetition 2300 ms; time echo 3.93 ms; flip angle 12°; field of view 256 mm; voxel size 1 × 1 × 1 mm). For the alignment of the MEG and MRI data, we marked the position of both the nasion and the preauricular points with vitamin E capsules.

Data Analysis

Behavioral Analyses

Visibility ratings were recoded into 2 categories: “unseen” stimuli (responses 1 and 2), and “seen” stimuli (responses 3 and 4), in line with the fact that PAS = 3 responses require subjects to identify the stimuli. We then calculated the percentage of visible responses for each degradation level (1–6) and sequence order (ascending and descending). The resulting data were analyzed by means of a Greenhouse–Geisser-corrected repeated-measures analysis of variance (rmANOVA) with factors degradation level and sequence order. Additionally, sigmoidal fits were performed separately for the ascending and descending sequences on the 6 degradation levels using the following generalized sigmoid function: f(x) = 1/(1 + ea(xφ)) + b, where a defines the slope (steepness) of the sigmoid; b, the offset; and φ, the horizontal shift of the function (the inflection point). Before fitting, the values representing the percentage of seen stimuli were normalized to the interval [0–1]. We were mainly interested in the parameter φ, the inflection point. When expectations influence visual perception, the inflection point for the descending sequence should shift leftwards on the horizontal axis, reflecting that the threshold of perception is reached at a higher degradation level (i.e., less sensory evidence) in the descending sequence compared with the ascending sequence. The inflection point was compared between the 2 sequences by means of a paired t-test. Corresponding analyses were conducted on the slope values.

To investigate whether the mere repetition of a stimulus also had an influence on visibility ratings, we took advantage of the fact that target stimuli were presented 3 times throughout the experiment for the ascending and descending sequences, respectively. The percentage of seen responses was computed for the first presentation of the stimuli, the first and the second repetition as a function degradation level, and analyzed in an rmANOVA with factors repetition (3 levels), and degradation level (5 levels), separately for the ascending and descending sequences.

MEG Data Analyses

MEG data were analyzed using the Fieldtrip toolbox (Oostenveld et al. 2011; version 2008-12-08, http://fieldtrip.fcdonders.nl/) in Matlab R2007b (The MathWorks).

Event-Related Field Analyses

Before calculating event-related fields (ERFs), the continuous data were bandpass-filtered (Butterworth filter, fourth order, 0.3–20 Hz). Epochs were then segmented relative to stimulus onset, from −0.7 s before to 0.7 s after stimulus onset. An automatic artifact detection algorithm, as implemented in Fieldtrip, was used to exclude epochs contaminated with eye blinks, muscle activity, or sensor jump artifacts. After artifact rejection, 78.4% of the trials were retained for the ERF analysis. ERFs were baseline-corrected using an interval of −0.5 to −0.1 s prior to stimulus onset. We calculated planar gradients of the MEG field distribution, which simplifies the interpretation of sensor-level data as sensors showing maximal activity are more likely to be located above the actual sources (Bastiaansen and Knosche 2000). All analyses were conducted over 266 MEG sensors, after discarding 9 defect sensors. Epochs were then sorted according to the order in the sequence, which resulted in 2 conditions: Ascending and descending sequences. The ascending sequence consisted of trials from stimuli 1 to 5 that were “not expected,” whereas the descending sequence consisted of trials from stimuli 7 to 11 that due to the position in the sequence were “expected.”

Spectral Analyses (Time–Frequency Responses)

For spectral analyses, we extracted epochs from 1 s before to 1 s after stimulus onset. For each trial, signals were analyzed in the time–frequency domain by convolution with a complex Gaussian Morlet wavelet with a bandwidth parameter of M = f0/σf = 3, that is, wavelets of approximately 3 cycles length. We chose M = 3 to attain high time resolution, as our main interest resided in characterizing the time course of alpha oscillations, and in particular to ascertain that the effect observed in the baseline was confined to the time period of interest. To avoid moving average effects in the time–frequency decomposition and a violation of the “admissibility criterion” arising for wavelets with low-cycle numbers (Sinkkonen et al. 1995), we used the original full Morlet wavelet formulation (Sadowsky 1996) and not the simplified wavelet standardly implemented in Fieldtrip.

We investigated the power of frequencies between 2 and 20 Hz, with a focus on alpha-band activity (8–12 Hz) in the period of −0.5 s before to 1 s after the stimulus onset. We obtained the amplitude of the horizontal and vertical component of the planar gradient and then calculated their sum of squares using Pythagoras' rule to obtain spectral power in planar gradient representation. As we were interested in prestimulus baseline modulations as a function of expectations, we did not perform any baseline correction. Epochs were sorted into ascending and descending sequences. Finally, we calculated the mean of the planar gradient power estimate per condition per subject.

Statistical Analyses

Statistical analyses were performed on the sensor and source level as well as on evoked and oscillatory activity (ERF and time–frequency responses, respectively), using nonparametric cluster-based permutation tests (Maris and Oostenveld 2007). This method identifies clusters of significant differences over space, time, and/or frequency that show the same effect, effectively dealing with the multiple comparison problems (MCPs) while also accounting for the inherent correlation structure of the data. For the ERF analysis, we focused on an early time window of 0–0.25 s after stimulus onset, the time at which the P1m–N1m component should be maximal (Gruber et al. 2005; Fellinger et al. 2012). To investigate pre- and poststimulus differences in alpha power across experimental conditions, we averaged over a frequency range of 8–12 Hz, and compared conditions for a time window of −0.5 s before stimulus onset to 0.3 s after stimulus onset across all 266 sensors. For each sensor and time point of interest, a paired t-test was computed, comparing the ascending with the descending sequence. To correct for multiple comparisons, we used a cluster-based statistical analysis (temporo-spatial adjacency) with the criterion set to P < 0.05. The sum of the t-values within a cluster was used for cluster-level statistics. The permutation distribution was approximated by drawing 5000 random permutations of the observed data, that is, by shuffling the labels across conditions. This reference distribution was used to evaluate the statistics of the actual data. The obtained P-values represent the probability under the null hypothesis (no difference between the conditions) of observing a maximum (minimum) cluster-level statistic that is larger (smaller) than the observed cluster-level statistics.

In the “perception as inference” framework, the contribution of prior knowledge scales with the strength of the sensory evidence, such that the weaker the sensory evidence the higher the contribution of the prior (Vilares and Kording 2011). Thus, to further investigate this possibility, alpha power (8–12 Hz) over a window from 0.5 s before stimulus onset to 0.3 s after stimulus onset across all 266 sensors was split between the ascending and descending sequences and further between high and low degradation levels. For strongly degraded stimuli, we averaged alpha power over levels 1 + 2 + 3 and 11 + 10 + 9 for ascending and descending sequences, respectively; for weakly degraded stimuli, we averaged alpha power over conditions 4 + 5 + 6 and 8 + 7 + 6 for ascending and descending sequences, respectively. We compared ascending and descending sequences for highly degraded stimuli and for low degraded stimuli separately by means of paired t-tests, and used cluster-based statistical analysis to correct for multiple comparisons. To investigate the interaction between the factors sequence and degradation level, we further contrasted the difference between ascending and descending sequences for high and low degradation levels.

Single-Trial Analyses

We also performed regression analyses to assess whether prestimulus alpha power predicted visibility on a trial-by-trial level. To this end, we built a logistic regression model, predicting binarized visibility ratings (i.e., seen/unseen) from the natural logarithm of prestimulus alpha power in a −0.4 to −0.25 s time window, stimulus level (1–5), ascending versus descending sequence order, and their interactions. Alpha power was z-scored before entering the analyses. Outliers were excluded based on Anscombe residuals ≥abs(3.1). This led to a rejection of <1% of the data. Goodness of fit of the model was assessed by comparison to an intercept-only model and by using a Hosmer–Lemeshow test with a correction for large sample size (Paul et al. 2013).

Alpha Power-ERF Correlations

We investigated the relationship between alpha power and ERF as a function of expectations by correlating the sensor-level differences in alpha power between expected and non-expected trials with the corresponding sensor-level differences in the P1 m/N1m amplitude. Alpha power values were selected from the prestimulus time window (−0.4 to −0.25 s) in which we observed significant differences between ascending and descending sequences (see Results). In this time window, alpha power values were averaged for the ascending and descending condition, respectively, and then subtracted. The resulting values were correlated with the average difference in ERF activity per subject. To gain insights into the spatiotemporal relationship between prestimulus alpha oscillations and the event-related response, Pearson's correlations were calculated in 6 overlapping ERF time windows, ranging from 0.12 to 0.155 s (6 time windows of 10 ms each, moving in 5 ms steps) based on the latency of the P1m/N1m components. The correlation analysis was done separately for 12 left occipital sensors, resulting in 1 × 6 × 12 = 72 tests, and corrected for multiple comparisons by means of the false discovery rate (FDR; Genovese et al. 2002) with q = 0.05.

Source Reconstruction of Alpha Power

A frequency domain Beamformer (Gross et al. 2001) was used to localize the sources of oscillatory activity in the alpha band. This approach utilizes an adaptive spatial filter that enables the estimation of power in the frequency of interest at any location in the brain. Source analysis techniques such as Beamforming require the calculation of a forward model, for which the following computations were done: First, we created a cortical mesh out of an MRI template brain in the MNI space and warped this mesh to fit an individual structural MRI (Mattout et al. 2007). This step consisted in the inverse transformation of the template mesh by means of commonly used spatial normalization methods (Ashburner and Friston 2005). The cortical mesh had a resolution of 1 cm and comprised of 6783 vertices. Second, the individual cortical meshes were co-registered with the subjects' MEG sensor positions by alignment of the fiducial points. A “single-shell” model (Hamalainen and Sarvas 1987, 1989; Nolte 2003) was used as a volume conductor. The result of the forward model computation, which is based on Maxwell's equations, is an N × M lead field matrix, with N describing the number of MEG sensors and M defining the number of mesh vertices. Second, we calculated cross-spectral density matrices for each condition and specific time window using multitapers on the single-trial axial gradient data in the alpha band (center frequency 9 Hz and spectral smoothing 4 Hz). To localize the significant effects for alpha power found on the sensor level, we calculated the cross-spectral density matrices across the ascending and descending sequences (common spatial filter) for the time windows during which we found differences between “expected” and “not expected” stimuli over occipital sensors (prestimulus period: −0.4 to −0.25 s; poststimulus period: 0.05–to 0.15 s). To gain a better spatial resolution and a similar window length for both periods, we extended the time window of the prestimulus period by 0.05 s and the poststimulus period by 0.1 s, resulting in 0.2 s windows of −0.45 to −0.25 s and −0.05 to 0.15 s. Finally, we applied the Beamformer using a regularization of λ = 5% to counterbalance the short-duration time window.

The resulting source activities of the 2 conditions were separately compared over the pre- and poststimulus time windows over the whole brain, using the aforementioned permutation test (5000 iterations). Results were cluster-corrected for multiple comparisons at P < 0.05 (two-tailed). We located the MNI space coordinates of the resulting clusters (peak voxels) by means of the Harvard–Oxford cortical structural atlas (Desikan et al. 2006) and the Juelich histological atlas (Eickhoff et al. 2007), both provided in FSL (http://www.fmrib.ox.ac.uk/fsl/data/atlas-descriptions.html).

Eye Blink and Eye Movement Control Analysis

While subjects were instructed to maintain fixation throughout the experiment, eye blinks and eye movements can still occur occasionally. To rule that the neural effects could be explained by a differential pattern of eye movements or blinks between the ascending and descending sequences, we compared the EOG traces (2 EOG channels, 1 for vertical and 1 for horizontal eye movements) during the baseline period (−0.5 to 0 s) as well as in the peristimulus period (0–0.3 s) between the ascending and descending sequences by means of paired t-tests and used cluster-based statistical analysis to correct for multiple comparisons.

Results

Sensory Expectations Lower the Threshold of Visibility

Figure 1B shows the mean psychometric function of visibility for the ascending and descending sequences. As expected, visibility increased as a function of the degradation level, with a higher percentage of seen responses for stimuli that are less degraded (F1.959,47.009 = 305.014, P < 0.0001). This replicates previous results showing a sigmoidal curve relating stimulus evidence to perceptual thresholds (Del Cul et al. 2007; Melloni et al. 2011). Importantly, we observed a clear effect of expectations on the threshold of visibility, which is evident by the higher percentage of seen responses when subjects had a clear expectation about the upcoming stimuli (descending sequence 57%), than when they did not (ascending sequence 43%; F1,24 = 119.742, P = 1E−10). The effect was clearly noticeable by the horizontal shift in the psychometric function, which reflects a strong reduction in threshold values, indicating that participants needed less stimulus evidence to achieve the same visibility threshold in the presence of a clear expectation. This result was confirmed when comparing the threshold values (inflection point) obtained by the sigmoidal fit (t(24) = 8.837, P = 5.1E−9). While the visibility threshold was at degradation level 3.82 during the ascending sequence, the threshold was reached at degradation level 2.91 during the descending sequence (perceptual gain of ∼1 degradation level). This change in threshold was not accompanied by a difference in slope (t(24) = 1.704, P = 1). Finally, we observed that the effect of expectations interacts with degradation level (F2.221,53.311 = 19.421; P = 2.0E−7), indicating that a minimal amount of stimulus evidence is needed for expectations to exert an effect on visibility. This is evident at degradation level 1, for which the percentage of seen responses is the same regardless of whether expectations are present or absent, respectively (t(24) = −2.049, P = 0.260, Bonferroni-corrected). A control analysis further established that the observed gain in visibility appears to be specifically related to the development of stimulus-specific expectations and not to the mere repetition of the stimuli, as no comparable gain in visibility was observed when contrasting repetitions of the same stimulus throughout the experiment as a function of degradation level for the ascending or descending sequence, respectively (main effect of repetition: all P > 0.1; interaction repetition × degradation level: all P > 0.3; see Supplementary Fig. 1).

Alpha Oscillations as Carriers of Sensory Predictions

Having established a clear behavioral effect of expectations on the psychometric visibility function, we turned to our main question of interest, that is, whether alpha oscillations play a functional role in the implementation of content-based predictions. As can be seen in Figure 2A, strong and sustained alpha activity was observed before stimulus onset, which was accompanied by characteristic occipital alpha desynchronization after stimulus onset. In addition, alpha amplitude is higher during the baseline period for the descending sequences (Fig. 2B), when subjects had a clear expectation about the identity of the letter, than for the ascending sequence, when subjects had not yet seen the stimuli and thus did not have a clear expectation of the upcoming stimuli and perception relied more on sensory evidence. None of these effects could be explained by eye movements or blinks, as there was no significant difference in EOG activity between the sequences pre- or poststimulus (P > 0.05, cluster-corrected for MCP).

Figure 2.

Time course of alpha power. (A) Time–frequency plot obtained by averaging over all stimulus conditions and MEG sensors. X-axis denotes time. The white line indicates stimulus onset. Y-axis denotes frequency. (B) Time–frequency plots for an example left occipital sensor for the ascending (left plot) and descending (right plot) sequence. Note the strong alpha power during the baseline time period and the alpha desynchronization after stimulus presentation. The reduction in alpha power temporally coincides with the P1m–N1m component.

Figure 2.

Time course of alpha power. (A) Time–frequency plot obtained by averaging over all stimulus conditions and MEG sensors. X-axis denotes time. The white line indicates stimulus onset. Y-axis denotes frequency. (B) Time–frequency plots for an example left occipital sensor for the ascending (left plot) and descending (right plot) sequence. Note the strong alpha power during the baseline time period and the alpha desynchronization after stimulus presentation. The reduction in alpha power temporally coincides with the P1m–N1m component.

To test the prediction that alpha oscillations enact prior information, and thus should be higher when expectations are present—which should already be reflected during the baseline period—we compared alpha power (8–12 Hz) over an extended time window, from −0.5 s before stimulus onset to 0.3 s poststimulus between the descending (“expected”) and ascending sequence (“not expected”). Figure 3A depicts the difference in alpha power between the ascending and descending sequence for all MEG sensors. In line with our predictions, we observed a significant alpha cluster (cluster correction permutation test) over left occipital electrodes, that is, ipsilateral to the stimulus but consistent with the lateralization of letter and language processing, when subjects could predict the stimulus identity. This effect was present in a specific time window between −0.4 and −0.25 s before stimulus onset. Note that these trials only differ in the history of stimulation: subjects have a clear expectation in the descending sequence, which is not yet fully established in the ascending sequence. Also, as the observed differences in alpha power occur during the baseline period, they likely reflect an effect of expectation and not differences in visual stimulation. Moreover, we observed stronger prestimulus differences in alpha power between ascending and descending sequences when the stimulus evidence was low than when it was high, in line with a “perception as inference” account in which prior expectations exert stronger influences when stimulus evidence is scarce (Fig. 4). Significantly, the stronger alpha modulations for highly degraded than for low degraded stimuli exhibit a comparable topography and time windows of effects to the one observed when contrasting ascending and descending sequences.

Figure 3.

Alpha power results. (A) Alpha power (8–12 Hz) differences between the descending and ascending sequence for all MEG sensors. The x-axis represents time, the y-axis, the left (L) and right (R) MEG sensors located in C, central; F, frontal; O, occipital; P, parietal; T, temporal; Z, centro-occipital regions. The color scale indicates the strength of alpha power in Tesla. Time 0, highlighted by the white line, corresponds to the onset of stimulus presentation. The circled areas contain the significant time-sensor pairs (all effects P < 0.05, cluster-corrected for MCP). (B) Alpha power topographies of the pre- (−0.5 to −0.25 s) and post- (0–0.15 s) stimulus time period for the ascending and descending sequences (first and second head) and their difference (third head). Significant sensors are highlighted by black dots. (C) Time course of the difference in alpha power between the ascending and descending sequences for left (red) and right (blue) occipital and temporal sensors. X-axis represents time; y-axis represents alpha power in Tesla. Left occipital sensors show 2 peaks, one during the baseline period (−0.45 to −0.25 s), and another after stimulus presentation (0.05–0.15 s). Note the apparent phase opposition between left and right occipital sensors. (D) Results of the single-trial analysis. Prestimulus alpha power predicted seen versus unseen responses depending on both the level of stimulus degradation and whether expectations were present or absent, as evidenced by a significant three-way interaction (Wald's χ2(1) = 4.67, P = 0.0306). The probability to see a letter was higher for low degradation levels (x-axis, levels 1–5) and when an expectation was formed (gray lines). High prestimulus alpha power was associated with a higher probability to see the letter when stimulus evidence was weak (solid lines: +2SD alpha, dashed lines −2SD alpha), especially in the descending sequence, that is, when expectations were present.

Figure 3.

Alpha power results. (A) Alpha power (8–12 Hz) differences between the descending and ascending sequence for all MEG sensors. The x-axis represents time, the y-axis, the left (L) and right (R) MEG sensors located in C, central; F, frontal; O, occipital; P, parietal; T, temporal; Z, centro-occipital regions. The color scale indicates the strength of alpha power in Tesla. Time 0, highlighted by the white line, corresponds to the onset of stimulus presentation. The circled areas contain the significant time-sensor pairs (all effects P < 0.05, cluster-corrected for MCP). (B) Alpha power topographies of the pre- (−0.5 to −0.25 s) and post- (0–0.15 s) stimulus time period for the ascending and descending sequences (first and second head) and their difference (third head). Significant sensors are highlighted by black dots. (C) Time course of the difference in alpha power between the ascending and descending sequences for left (red) and right (blue) occipital and temporal sensors. X-axis represents time; y-axis represents alpha power in Tesla. Left occipital sensors show 2 peaks, one during the baseline period (−0.45 to −0.25 s), and another after stimulus presentation (0.05–0.15 s). Note the apparent phase opposition between left and right occipital sensors. (D) Results of the single-trial analysis. Prestimulus alpha power predicted seen versus unseen responses depending on both the level of stimulus degradation and whether expectations were present or absent, as evidenced by a significant three-way interaction (Wald's χ2(1) = 4.67, P = 0.0306). The probability to see a letter was higher for low degradation levels (x-axis, levels 1–5) and when an expectation was formed (gray lines). High prestimulus alpha power was associated with a higher probability to see the letter when stimulus evidence was weak (solid lines: +2SD alpha, dashed lines −2SD alpha), especially in the descending sequence, that is, when expectations were present.

Figure 4.

Alpha power differences between the descending and ascending sequence for (A) highly degraded stimuli (average of levels 11+10 + 9 and 1 + 2 + 3, respectively) and (B) low degraded stimuli (average of levels 8 + 7 + 6 and 4 + 5 + 6, respectively). Topographies of the effects averaged over the whole analysis window (−0.5 to 0.3 s poststimulus). The color scale indicates the direction and strength of the differences in t-values; no negative P-values were observed. (C) Interaction effect, that is, the difference in contrast (A) and (B). All effects are significant at P < 0.05, cluster-corrected for MCP. The interaction reveals a stronger modulation of alpha oscillations for highly degraded stimuli with a selective right occipital topography (in accordance with the stimuli being presented in the left visual field), which is most evident during the baseline period and the time window of the early evoked sensory components.

Figure 4.

Alpha power differences between the descending and ascending sequence for (A) highly degraded stimuli (average of levels 11+10 + 9 and 1 + 2 + 3, respectively) and (B) low degraded stimuli (average of levels 8 + 7 + 6 and 4 + 5 + 6, respectively). Topographies of the effects averaged over the whole analysis window (−0.5 to 0.3 s poststimulus). The color scale indicates the direction and strength of the differences in t-values; no negative P-values were observed. (C) Interaction effect, that is, the difference in contrast (A) and (B). All effects are significant at P < 0.05, cluster-corrected for MCP. The interaction reveals a stronger modulation of alpha oscillations for highly degraded stimuli with a selective right occipital topography (in accordance with the stimuli being presented in the left visual field), which is most evident during the baseline period and the time window of the early evoked sensory components.

Next, we used single-trial logistic regression to assess whether prestimulus alpha power (−0.4 to −0.25 s) predicted visibility (binarized into seen/unseen) and how this interacted with the presence of expectations. To this end, we built a regression model with prestimulus alpha power, stimulus level (1–5), ascending versus descending sequence order, and their interactions with visibility ratings on the basis of single-trial data of all subjects. Alpha power was the natural logarithm of the alpha power averaged over the sensors found to be significant for the contrast ascending versus descending in the group analysis, but the results were robust to the precise choice of sensors (see Supplementary Material). This model yielded a good fit to the data (Hosmer–Lemeshow test, HL(1800) = 1589.9775, P > 0.05; adjusted McFadden R2 = 0.4649) and outperformed an intercept-only model (likelihood ratio test, LR(7) = 9885.21, P < 0.01, Table 1). Importantly, we found a significant three-way interaction between alpha power, stimulus level, and sequence order (Wald's χ2(1) = 4.67, P = 0.0306; likelihood ratio test against a model without the three-way interaction term, LR(1) = 4.6935, P = 0.0303). As can be seen in Figure 3D, the probability to see a letter was generally higher for low degradation levels; predictions carried by prestimulus alpha oscillations increased the probability of perceiving the letter when sensory evidence was reduced, and particularly so in the descending sequence when a clear prediction had been formed.

Table 1

Results logistic regression

Predictor Beta SE Wald's χ2 df P-value Odds ratio 
Constant −6.6482 0.1491 1987.9332 <0.01 N/A 
Alpha power 0.566 0.1443 15.4277 <0.01 1.7622 
Stimulus level 1.7114 0.0386 1963.3248 <0.01 5.5370 
Ascending/descending 1.5616 0.1878 69.1335 <0.01 4.7666 
Alpha × stimulus level −0.1550 0.0376 17.0329 <0.01 0.8564 
Alpha × asc./desc. 0.2909 0.1770 2.7021 >0.10 1.3377 
Stimulus level × asc./desc. −0.0018 0.0533 0.0011 >0.90 0.9982 
Alpha × stim. level × asc./desc. −0.1082 0.0500 4.6777 <0.05 0.8975 
Predictor Beta SE Wald's χ2 df P-value Odds ratio 
Constant −6.6482 0.1491 1987.9332 <0.01 N/A 
Alpha power 0.566 0.1443 15.4277 <0.01 1.7622 
Stimulus level 1.7114 0.0386 1963.3248 <0.01 5.5370 
Ascending/descending 1.5616 0.1878 69.1335 <0.01 4.7666 
Alpha × stimulus level −0.1550 0.0376 17.0329 <0.01 0.8564 
Alpha × asc./desc. 0.2909 0.1770 2.7021 >0.10 1.3377 
Stimulus level × asc./desc. −0.0018 0.0533 0.0011 >0.90 0.9982 
Alpha × stim. level × asc./desc. −0.1082 0.0500 4.6777 <0.05 0.8975 

Alpha Oscillations Relate to Stimulus-Specific Expectations and Are Not Explained by Mere Passage of Time

To test whether the increases in prestimulus alpha oscillations were due to stimulus-specific expectations or to unspecific effects such as the mere passage of time or a buildup of adaptation, we performed several control analyses. If increases in alpha power were due to such unspecific effects, alpha power should increase from trial to trial despite the fact that stimulus identity, and thus the basis of the perceptual expectation, changes from trial to trial. To investigate this possibility, we contrasted the difference in alpha power between the ascending and descending sequences within a trial (i.e., when stimulus identity is preserved between the 2 phases and thus allows for the buildup of a stimulus-specific prediction), with the difference in alpha power in the descending phase of a trial and the ascending phase of the following trial (see Supplementary Fig. 2). As stimulus identity changes between trials, a new prediction needs to be formed. If alpha power merely increases with time, the difference in alpha power between trials should be larger than or equal to the difference than within a trial. In contrast, if alpha power increases are due to stimulus-specific expectations, the difference in alpha power between trials should be smaller or of the opposite sign than the difference in alpha power within a trial. Indeed, we find that within-trial difference in prestimulus alpha power exceed between-trial differences in prestimulus alpha power (main effect of trial order, F1,24 = 6.957, P = 0.014), which suggests that prestimulus alpha effects are specific to the expected stimulus.

This conclusion was further supported by an analysis investigating differences in alpha power as a function of degradation level and the presence/absence of expectations. We found that alpha power increased steadily during the ascending sequence within a trial, but then plateaued once an expectation had been formed (degradation level × expectation interaction, F3.567,1.043E4 = 1.865E−3, P < 0.0001; see Supplementary Fig. 3). An account based on the mere passage of time or continuously increasing adaptation levels would have predicted that alpha continues to increase in the descending phase, which was not the case.

Taken together, these analyses show that alpha power has a time course that is consistent with a gradual buildup of stimulus-specific expectations, and accordingly breaks down when stimulus identity changes between trials.

Alpha Oscillations Are Involved in Testing Sensory Predictions

Our next step was to investigate how ongoing predictions are tested against incoming evidence, and whether alpha oscillations also play a role in this process. Left occipital alpha power rebounded in the poststimulus period around the time of the P1m–N1m component (0.05–0.15 s). In contrast to the waxing and waning of alpha power observed over occipital sensors, we found a sustained increase in alpha power from −0.5 pre- to 0.15 s poststimulus over left temporal sensors (Fig. 3AC, all effects P < 0.05, cluster-corrected for MCP). We reasoned that if alpha oscillations play a role in carrying and testing predictions within a cortical circuit, then one would expect a strong correlation between the alpha topographies before and after stimulus onset (Fig. 3B), reflecting that both phenomena arise from the same underlying oscillatory ensemble. To test this prediction, we compared the topographical distribution of alpha-band power of the expectation-related effect before and after stimulus onset (Fig. 3B), and found that both topographies were practically identical (spatial correlation: r = 0.7723, P < 0.0001). This indeed suggests that alpha oscillations carrying predictions prestimulus are also involved in testing these predictions once input arrives.

We then went onto investigate amplitude differences in the P1/N1m complex between expected and non-expected trials, as the P1/N1m complex has been related to conscious detection (Palva et al. 2005) and has been shown to be influenced by expectations (Melloni et al. 2011). Moreover, the P1/N1 complex has been related to early visual-semantic categorization process and recognition, in which bottom-up information is integrated with prior knowledge (Klimesch 2011), a process that is essential for the testing of predictions. Notably, it has been argued that alpha oscillations play a crucial role in the generation of the P1/N1m component (Gruber et al. 2005; Fellinger et al. 2012), offering a link between our previous finding on alpha oscillations, in which we observed increased alpha power for trials with expectations poststimulus onset, consistent with the typical time window of the P1/N1m component. In line with this prediction, and as shown in Figure 5AC, we found higher amplitude in the early evoked components, P1/N1 complex for “expectation” than for “no expectation” trials, with a similar topography (left occipito-temporal sensors) and latency (0.125–0.155 s poststimulus) as the previously documented effects on alpha. Accordingly, we observed a very high spatial correlation of the topographical distribution of the expectation effect for poststimulus alpha power and the P1/N1m ERF (r = 0.518, P < 0.0001). To further test whether the pre- and poststimulus differences in neural activity for the alpha oscillations and early evoked (P1/N1m) component, which correspond to content-related expectations, were part of a common process, we correlated the differences in prestimulus alpha power and the differences in amplitude of the evoked component in the P1/N1 complex across subjects. This analysis revealed that larger expectation-driven effects in prestimulus alpha power were consistently accompanied by large differences in P1/N1m amplitude poststimulus (all P < 0.0051, for r-values see Fig. 5D). Interestingly, we observed that the maximum correlation showed a propagation in time and space, running from anterior to posterior occipital sensors (Fig. 5D, rightmost panel). This resembles a traveling wave as it has previously been described for the EEG P1/N1 complex (Fellinger et al. 2012).

Figure 5.

Event-related magnetic field results. (A) ERF differences between the descending and ascending sequence for all MEG sensors. The x-axis represents time after stimulus onset. The color scale indicates magnetic field activity in Tesla. The circled areas contain the significant time-sensor pairs (all effects P < 0.05, cluster-corrected for MCP). (B) ERF topographies of the significant time period (0.125–0.155 s) for both sequences (first and second topography) and their difference (third topography). (C) Example ERF time courses of 2 significant sensors. Gray lines mark the time period during which the difference is significant. (D) Results of the correlation analysis. Left plot: Pearson's correlation across subjects between prestimulus alpha power and ERF over occipital sensors for the differences between descending and ascending conditions, respectively. The x-axis shows 6 ERF time windows; the ERF difference in each of them was correlated with the alpha baseline effect (−0.45 to −0.25 s). The y-axis shows individual MEG sensors. The color scale indicates the strength of the correlation masked by FDR-corrected significance (P < 0.05). Right image: ERF difference topography between 0.12 and 0.15 s and enlarged snapshot with the significant sensors of the correlation.

Figure 5.

Event-related magnetic field results. (A) ERF differences between the descending and ascending sequence for all MEG sensors. The x-axis represents time after stimulus onset. The color scale indicates magnetic field activity in Tesla. The circled areas contain the significant time-sensor pairs (all effects P < 0.05, cluster-corrected for MCP). (B) ERF topographies of the significant time period (0.125–0.155 s) for both sequences (first and second topography) and their difference (third topography). (C) Example ERF time courses of 2 significant sensors. Gray lines mark the time period during which the difference is significant. (D) Results of the correlation analysis. Left plot: Pearson's correlation across subjects between prestimulus alpha power and ERF over occipital sensors for the differences between descending and ascending conditions, respectively. The x-axis shows 6 ERF time windows; the ERF difference in each of them was correlated with the alpha baseline effect (−0.45 to −0.25 s). The y-axis shows individual MEG sensors. The color scale indicates the strength of the correlation masked by FDR-corrected significance (P < 0.05). Right image: ERF difference topography between 0.12 and 0.15 s and enlarged snapshot with the significant sensors of the correlation.

Neural Sources of Sensory Predictions

Finally, we performed Beamformer source analysis to gain insights into the cortical areas involved in the representation of the prior. Comparisons between “expected” and “not expected” trials in the prestimulus period during which predictions are kept online (−0.45 to −0.25 s; Fig. 6, upper row) revealed enhanced alpha power in the left temporal cortex, with peaks of activity in anterior and posterior superior temporal gyrus (STG), extending into Heschl's gyrus/planum temporale (HG/PT). Furthermore, we found significant effects in the left superior parietal lobule (SPL) and the right superior frontal gyrus (SFG; see Table 2 for MNI coordinates). Notably, these areas overlap with a set of areas that have been found in an fMRI study investigating the cortical network involved in top-down letter processing in an illusory letter detection task (Liu et al. 2010). Importantly, increases in alpha oscillations were observed in auditory cortex, a result which resonates with previous studies revealing a distributed supramodal (multisensory) representation of the concept of a letter (Raij et al. 2000; van Atteveldt et al. 2004).

Table 2

Peak MNI coordinates of the sources corresponding to the prestimulus period (−0.45 to −0.25 s)

Region Peak MNI coordinate [X, Y, ZP-value 
Prestimulus period (−450 to −250 ms) 
 Anterior superior temporal gyrus [−60, −10, 0] <0.05, corrected 
 Posterior superior temporal gyrus [−70, −20, 0] <0.05, corrected 
 Heschl's gyrus/planum temporale [−50, −10, 0] <0.05, corrected 
 Left superior parietal lobe [−20, −60, 50] <0.05, corrected 
 Right superior frontal gyrus [20, 10, 50] <0.05, corrected 
Region Peak MNI coordinate [X, Y, ZP-value 
Prestimulus period (−450 to −250 ms) 
 Anterior superior temporal gyrus [−60, −10, 0] <0.05, corrected 
 Posterior superior temporal gyrus [−70, −20, 0] <0.05, corrected 
 Heschl's gyrus/planum temporale [−50, −10, 0] <0.05, corrected 
 Left superior parietal lobe [−20, −60, 50] <0.05, corrected 
 Right superior frontal gyrus [20, 10, 50] <0.05, corrected 
Figure 6.

Alpha power Beamformer source reconstruction results. Results were obtained by contrasting the descending with the ascending condition for the pre- (−0.45 to −0.25 s) and post- (−0.05 to 0.15 s) stimulus time period. Sources are masked by significance (P < 0.05, MCP-corrected). The color scale reflects t-values. Sources of alpha activity during the baseline period were found in the left temporal cortex with peaks in anterior and posterior STG, HG/PT, left SPL, and right SFG. Sources of alpha activity during the stimulation period were found in anterior STG and TP, left superior parietal, inferior, and superior frontal lobes, and left early visual cortex extending into the VWFA. For MNI coordinates of peak activity, see Tables 2–3. Note the striking similarities between the sources found for the baseline and stimulation period. STG: superior temporal gyrus; SPL: superior parietal lobe; SFG: superior frontal gyrus; TP: temporal pole; VWFA: visual word form area; HG/PT: Heschl's gyrus/planum temporale.

Figure 6.

Alpha power Beamformer source reconstruction results. Results were obtained by contrasting the descending with the ascending condition for the pre- (−0.45 to −0.25 s) and post- (−0.05 to 0.15 s) stimulus time period. Sources are masked by significance (P < 0.05, MCP-corrected). The color scale reflects t-values. Sources of alpha activity during the baseline period were found in the left temporal cortex with peaks in anterior and posterior STG, HG/PT, left SPL, and right SFG. Sources of alpha activity during the stimulation period were found in anterior STG and TP, left superior parietal, inferior, and superior frontal lobes, and left early visual cortex extending into the VWFA. For MNI coordinates of peak activity, see Tables 2–3. Note the striking similarities between the sources found for the baseline and stimulation period. STG: superior temporal gyrus; SPL: superior parietal lobe; SFG: superior frontal gyrus; TP: temporal pole; VWFA: visual word form area; HG/PT: Heschl's gyrus/planum temporale.

A similar set of areas was observed when investigating the differences in alpha activity between expected and unexpected trials in the poststimulus period (−0.05 to 0.15 s; Fig. 6, lower row). The peak activity in left temporal areas was located more anteriorly in the temporal pole (TP) and STG, as well as in the left superior parietal, left inferior frontal, right superior frontal, and left early visual cortices, extending into the visual word form area (VWFA; see Table 3 for MNI coordinates).

Table 3

Peak MNI coordinates of the sources corresponding to the poststimulus period (−0.05 to 0.15 s)

Region Peak MNI coordinate [X, Y, ZP-value 
Poststimulus period (−50 to 150 ms) 
 Temporal pole [−50, 10, −10] <0.05, corrected 
 Superior temporal gyrus [−60, 0, −10] <0.05, corrected 
 Left superior parietal lobe [−30, −60, 60] <0.05, corrected 
 Left inferior frontal lobe [−50, 10, 0] <0.05, corrected 
 Right superior frontal lobe [30, 0, 70] <0.05, corrected 
 Left visual cortex [−10, −70, −10] <0.05, corrected 
Region Peak MNI coordinate [X, Y, ZP-value 
Poststimulus period (−50 to 150 ms) 
 Temporal pole [−50, 10, −10] <0.05, corrected 
 Superior temporal gyrus [−60, 0, −10] <0.05, corrected 
 Left superior parietal lobe [−30, −60, 60] <0.05, corrected 
 Left inferior frontal lobe [−50, 10, 0] <0.05, corrected 
 Right superior frontal lobe [30, 0, 70] <0.05, corrected 
 Left visual cortex [−10, −70, −10] <0.05, corrected 

Discussion

We investigated how expectations influence perception, and how this is mediated by alpha oscillations. In line with previous studies (Melloni et al. 2011; Moca et al. 2011; Aru et al. 2012), we observed that expectations lower psychophysical thresholds. These behavioral effects were accompanied by stronger prestimulus alpha oscillations in areas responsible for the supramodal representation of letters when stimuli were expected. We also observed higher alpha power and early evoked amplitude responses of the P1/N1 complex (and a high correlation between them) after stimulus onset. The P1/N1 complex is thought to arise from ongoing alpha oscillations (Gruber et al. 2005; Fellinger et al. 2012) and to reflect the influence of top-down processes on bottom-up drive (Hopf et al. 2002; Chaumon et al. 2008; Dambacher et al. 2009; Melloni et al. 2011). Control analyses and previous studies (Melloni et al. 2011; Moca et al. 2011) have established that these behavioral and neural effects are due to stimulus-specific expectations, and not the mere passage of time, changes in criterion, and/or mere repetition of stimuli. Taken together, this suggests that alpha oscillations in visual and auditory areas do not only implement expectations about letters before stimulus onset, but are also involved in matching these expectations with incoming information.

Alpha Oscillations Carry and Test Sensory Expectations

Alpha oscillations have been extensively investigated in the context of attention and have often been attributed an inhibitory role, for example, in filtering task-irrelevant information (Jensen and Mazaheri 2010). For instance, unattended stimuli are accompanied by higher alpha oscillations than attended stimuli (Banerjee et al. 2011). Nonetheless, alpha oscillations also play an active role in other forms of top-down modulation of information processing (von Stein et al. 2000; Palva and Palva 2007; Johnson et al. 2011; Klimesch 2012; van Kerkoerle et al. 2014), specifically in the implementation of expectations. For example, alpha oscillations entrain to temporally predictable stimulus sequences, amplifying the response to temporally expected stimuli (Rohenkohl and Nobre 2011; Premereur et al. 2012), and play a critical role in giving rise to illusory perception of visual motion that is thought to be caused by internally generated predictions (Sanders et al. 2014). We extend these findings by showing that alpha oscillations can also implement content-based predictions. Consistent with the hypothesis that alpha oscillations enable access to the knowledge system (Klimesch 2012), and that they favor expectation-driven operations over external sensory input (Cooper et al. 2003), we found higher prestimulus alpha oscillations in a supramodal network representing letters when stimuli were predictable, which correlated with visibility on a trial-by-trial basis. These results mesh well with a recent study investigating the effects of prior expectations in motor preparation (de Lange et al. 2013), which also showed that alpha oscillations over occipital sensors were modulated by expectations. In that study, however, predicted and unpredicted cues were not physically matched leaving the question open whether alpha oscillations reflected stimulus differences or expectations. Our study alleviates this concern, as sensory input was identical for expected and unexpected stimuli, strongly echoing the idea that alpha oscillations can reflect prior expectations about the identity of the upcoming target.

Previous studies have reported not only positive correlations between prestimulus alpha power and behavioral performance (Linkenkaer-Hansen et al. 2004; Babiloni et al. 2006; Becker et al. 2011; Mo et al. 2011) similar to what we found here, but also negative correlations (Hanslmayr et al. 2007; van Dijk et al. 2008), which appear at odds with a role of alpha oscillations in exerting facilitatory effects on perception through implementing sensory priors. A recent biophysical model of prestimulus alpha oscillations (Lundqvist et al. 2013) reconciles both directions of effects by relating them to specific neuronal populations. In this model, increasing alpha power by increasing activity of inhibitory population leads to decreases in detection rates as it renders network dynamics more stable and further removed from the threshold of excitation, making it more difficult to activate cell assemblies. In contrast, increasing alpha power by increasing excitation of pyramidal cells leads to improved detection rates due to a less stable network state that is closer to threshold and thus facilitates activation by weak inputs. The latter mechanism is consistent with the idea that alpha oscillations increase the SNR in task-relevant networks by precisely timed selection of highly excitable neurons while simultaneously silencing neurons with lower levels of excitation (Klimesch 2011, 2012), thereby promoting the online maintenance of expected stimuli.

We also found evidence for an involvement of alpha oscillations in testing predictions against sensory evidence, as increases in alpha oscillations had a timing and spatial profile consistent with the N1m, which is thought to reflect early stimulus recognition. Several pieces of evidence suggest that expecting a letter and testing this prediction are indexed by the same underlying alpha oscillation: First, the spatial distribution of effects in the alpha band before and after stimulus onset was very similar. This also held true for the spatial profile of the P1/N1m effect and alpha oscillations during stimulus presentation. Second, prestimulus alpha oscillations predicted P1/N1m amplitude differences as a function of expectations. These results are consistent with the hypothesis that prestimulus alpha oscillations are closely linked to the event-related processes indexed by the P1/N1 complex. Our results suggest that perceptual expectations boost early recognition processes, as indexed by the amplitude of the P1/N1m, possibly reflecting privileged access to the knowledge system for predicted stimuli. We have previously shown that expectations increase visibility by improving the sampling of visual information, as measured by the pattern of eye movements (Moca et al. 2011). This finding is consistent with the idea that subjects hold a clear prior that guides overt but also covert search behavior and mediates recognition. Presumably, it is this early access to visual-semantic information, which is guided by the prior information that functionally relates to the evoked alpha oscillations and modulates the early waveforms of the visual ERP.

Multisensory Cortices Store Top-Down Priors for Letters

Source analysis of alpha oscillations carrying expectations in the prestimulus period revealed a mainly left lateralized network, encompassing STG, middle temporal gyrus (MTG), auditory cortex (HG and PT), SPL, and SFG. These results mesh well with a previous fMRI study investigating top-down representations of letters (Liu et al. 2010), which found SPL and MTG activity when subjects detected illusory letters in noise; and also with a study which found activation in left posterior superior temporal areas during visual imagery of letters (Raij 1999). Exploiting the temporal resolution of MEG, we can now show that these areas are already active and represent prior information before stimulus onset. Interestingly, the areas carrying and testing predictions about letters were not exclusively visual. Instead, we found several areas implicated in audiovisual integration. In particular, SPL is involved in phonological coding during the translation of individual letters into sounds (Joseph et al. 2003, 2006), whereas STG is implicated in the audiovisual integration of uni- and bimodally presented phonemic and graphemic forms (Raij et al. 2000; van Atteveldt et al. 2004), thereby most directly reflecting the supramodal representation of letters (Blomert 2011). In contrast, HG/PT is usually not activated by visually presented letters but shows audiovisual congruency effects, which has been interpreted as feedback from STG (Calvert et al. 2000; van Atteveldt et al. 2004). Taken together, the activation of multisensory and auditory areas before stimulus onset in the presence of a content-specific expectation suggests that sensory priors may be enacted in the form of phonemes or audiovisual objects.

Such a cross-modal effect could be due to the primacy of spoken over written language (Mattingly 1972; van Atteveldt et al. 2009). Also, letters convey little semantic content but mostly phonological information, which might place stress on their phonological aspect. Moreover, this format of representation may be more efficient than visual storage in the current task, as the location of the letter was jittered from trial to trial. This may have encouraged using an auditory representation that generalizes across locations.

Testing Top-Down Expectations

The network in which top-down expectations were tested against incoming evidence largely overlapped with the areas carrying expectations in the prestimulus period. Additionally, we found an effect of prior expectations in the left inferior frontal gyrus pars opercularis (posterior IFG, Broca's area, and BA44), VWFA, and left V1/V2. Furthermore, activity in the temporal lobe shifted more anteriorly to include anterior STG and the TP.

Activation of the IFG, VWFA, and early visual cortex is consistent with the literature on letter processing and reading. In particular, IFG is implicated in sub- and nonlexical processing of letters (van Atteveldt et al. 2007; Vinckier et al. 2007; Liu et al. 2010) and words (Herbster et al. 1997; Poldrack et al. 1999; Heim et al. 2005), and has been suggested to play a critical role in grapheme-to-phoneme conversion (Fiez and Petersen 1998). The VWFA is considered a core region in the visual processing of graphemes (Dehaene and Cohen 2011) and has been suggested to integrate top-down predictions derived from high-level associations such as speech sounds (Price and Devlin 2011). The role of early visual areas in letter processing is less well understood, but several studies have documented activity in V1 during reading (Szwed et al. 2011, 2014) and visual imagery of letters (Raij 1999).

Finally, the shift of activity from more posterior regions in the temporal lobe during the prestimulus period to more anterior regions during the poststimulus period may reflect higher-level processing in the ventral speech-processing stream (Hickok and Poeppel 2007; DeWitt and Rauschecker 2012). For example, anterior temporal regions are more active for intelligible spoken and written narratives (Spitsyna et al. 2006), or comparing spoken vowels to noise (Obleser et al. 2006). This implies that anterior temporal cortex represents semantic and not sensory information (Lee and Noppeney 2011). Our results suggest that reaching this higher-order processing step is facilitated when stimuli are predictable.

In summary, we propose that alpha oscillations integrate prestimulus expectations with sensory evidence, which allows reaching higher stages of processing and boosts visibility. Our results thus indicate that low-frequency alpha oscillations can serve as a mechanism both to carry and to test sensory predictions.

Authors' Contributions

L.M. design and conceived the experiment. A.M. collected data. A.M., C.M.S., and L.M. analyzed the data. M.W. contributed analytical tools. C.M.S., W.S., and L.M. wrote the paper.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

This work was supported by the Max Planck Society, a LOEWE grant “Neuronale Koordination Forschungsschwerpunkt Frankfurt” (NeFF) (to M.W., W.S., and L.M.), a Marie Curie International Outgoing Fellowship within the 7th European Community Framework Programme (to L.M.), and a Human Frontier Science Program Long-term Fellowship (LT001118/2012-L, to C.M.S.).

Notes

We thank Ulla Gebauer and Stefan Westermann for help with data acquisition and the members of the EEGlab of the Department of Neurophysiology at the Max Planck Institute for Brain Research for insightful discussions. We are indebted to Saskia Haegens, Timo van Kerkoerle, Tessa van Leeuwen, and 3 anonymous reviewers for insightful comments on the manuscript. Conflict of Interest: None declared.

References

Arnal
LH
,
Giraud
AL
.
2012
.
Cortical oscillations and sensory predictions
.
Trends Cogn Sci
 .
16
:
390
398
.
Aru
J
,
Axmacher
N
,
Do Lam
AT
,
Fell
J
,
Elger
CE
,
Singer
W
,
Melloni
L
.
2012
.
Local category-specific gamma band responses in the visual cortex do not reflect conscious perception
.
J Neurosci
 .
32
:
14909
14914
.
Ashburner
J
,
Friston
KJ
.
2005
.
Unified segmentation
.
Neuroimage
 .
26
:
839
851
.
Babiloni
C
,
Vecchio
F
,
Bultrini
A
,
Luca Romani
G
,
Rossini
PM
.
2006
.
Pre- and poststimulus alpha rhythms are related to conscious visual perception: a high-resolution EEG study
.
Cereb Cortex
 .
16
:
1690
1700
.
Banerjee
S
,
Snyder
AC
,
Molholm
S
,
Foxe
JJ
.
2011
.
Oscillatory alpha-band mechanisms and the deployment of spatial attention to anticipated auditory and visual target locations: supramodal or sensory-specific control mechanisms?
J Neurosci
 .
31
:
9923
9932
.
Barry
RJ
,
de Pascalis
V
,
Hodder
D
,
Clarke
AR
,
Johnstone
SJ
.
2003
.
Preferred EEG brain states at stimulus onset in a fixed interstimulus interval auditory oddball task, and their effects on ERP components
.
Int J Psychophysiol
 .
47
:
187
198
.
Bastiaansen
MC
,
Knosche
TR
.
2000
.
Tangential derivative mapping of axial MEG applied to event-related desynchronization research
.
Clin Neurophysiol
 .
111
:
1300
1305
.
Bastiaansen
MC
,
Posthuma
D
,
Groot
PF
,
de Geus
EJ
.
2002
.
Event-related alpha and theta responses in a visuo-spatial working memory task
.
Clin Neurophysiol
 .
113
:
1882
1893
.
Bastos
AM
,
Usrey
WM
,
Adams
RA
,
Mangun
GR
,
Fries
P
,
Friston
KJ
.
2012
.
Canonical microcircuits for predictive coding
.
Neuron
 .
76
:
695
711
.
Becker
R
,
Reinacher
M
,
Freyer
F
,
Villringer
A
,
Ritter
P
.
2011
.
How ongoing neuronal oscillations account for evoked fMRI variability
.
J Neurosci
 .
31
:
11016
11027
.
Benedek
M
,
Bergner
S
,
Konen
T
,
Fink
A
,
Neubauer
AC
.
2011
.
EEG alpha synchronization is related to top-down processing in convergent and divergent thinking
.
Neuropsychologia
 .
49
:
3505
3511
.
Bernasconi
C
,
Konig
P
.
1999
.
On the directionality of cortical interactions studied by structural analysis of electrophysiological recordings
.
Biol Cybern
 .
81
:
199
210
.
Blomert
L
.
2011
.
The neural signature of orthographic-phonological binding in successful and failing reading development
.
Neuroimage
 .
57
:
695
703
.
Calvert
GA
,
Campbell
R
,
Brammer
MJ
.
2000
.
Evidence from functional magnetic resonance imaging of crossmodal binding in the human heteromodal cortex
.
Curr Biol
 .
10
:
649
657
.
Chatila
M
,
Milleret
C
,
Buser
P
,
Rougeul
A
.
1992
.
A 10 Hz “alpha-like” rhythm in the visual cortex of the waking cat
.
Electroencephalogr Clin Neurophysiol
 .
83
:
217
222
.
Chaumon
M
,
Drouet
V
,
Tallon-Baudry
C
.
2008
.
Unconscious associative memory affects visual processing before 100 ms
.
J Vis
 .
8
3
:
1
10
.
Cooper
NR
,
Croft
RJ
,
Dominey
SJ
,
Burgess
AP
,
Gruzelier
JH
.
2003
.
Paradox lost? Exploring the role of alpha oscillations during externally vs. internally directed attention and the implications for idling and inhibition hypotheses
.
Int J Psychophysiol
 .
47
:
65
74
.
Dambacher
M
,
Rolfs
M
,
Gollner
K
,
Kliegl
R
,
Jacobs
AM
.
2009
.
Event-related potentials reveal rapid verification of predicted visual input
.
PLoS ONE
 .
4
:
e5047
.
Dehaene
S
,
Cohen
L
.
2011
.
The unique role of the visual word form area in reading
.
Trends Cogn Sci
 .
15
:
254
262
.
de Lange
FP
,
Rahnev
DA
,
Donner
TH
,
Lau
H
.
2013
.
Prestimulus oscillatory activity over motor cortex reflects perceptual expectations
.
J Neurosci
 .
33
:
1400
1410
.
Del Cul
A
,
Baillet
S
,
Dehaene
S
.
2007
.
Brain dynamics underlying the nonlinear threshold for access to consciousness
.
PLoS Biol
 .
5
:
e260
.
Desikan
RS
,
Segonne
F
,
Fischl
B
,
Quinn
BT
,
Dickerson
BC
,
Blacker
D
,
Buckner
RL
,
Dale
AM
,
Maguire
RP
,
Hyman
BT
et al
.
2006
.
An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest
.
Neuroimage
 .
31
:
968
980
.
DeWitt
I
,
Rauschecker
JP
.
2012
.
Phoneme and word recognition in the auditory ventral stream
.
Proc Natl Acad Sci USA
 .
109
:
E505
E514
.
Eickhoff
SB
,
Paus
T
,
Caspers
S
,
Grosbras
MH
,
Evans
AC
,
Zilles
K
,
Amunts
K
.
2007
.
Assignment of functional activations to probabilistic cytoarchitectonic areas revisited
.
Neuroimage
 .
36
:
511
521
.
Fellinger
R
,
Gruber
W
,
Zauner
A
,
Freunberger
R
,
Klimesch
W
.
2012
.
Evoked traveling alpha waves predict visual-semantic categorization-speed
.
Neuroimage
 .
59
:
3379
3388
.
Fiez
JA
,
Petersen
SE
.
1998
.
Neuroimaging studies of word reading
.
Proc Natl Acad Sci USA
 .
95
:
914
921
.
Fontolan
L
,
Morillon
B
,
Liegeois-Chauvel
C
,
Giraud
AL
.
2014
.
The contribution of frequency-specific activity to hierarchical information processing in the human auditory cortex
.
Nat Commun
 .
5
:
4694
.
Freunberger
R
,
Holler
Y
,
Griesmayr
B
,
Gruber
W
,
Sauseng
P
,
Klimesch
W
.
2008
.
Functional similarities between the P1 component and alpha oscillations
.
Eur J Neurosci
 .
27
:
2330
2340
.
Freunberger
R
,
Klimesch
W
,
Griesmayr
B
,
Sauseng
P
,
Gruber
W
.
2008
.
Alpha phase coupling reflects object recognition
.
Neuroimage
 .
42
:
928
935
.
Genovese
CR
,
Lazar
NA
,
Nichols
T
.
2002
.
Thresholding of statistical maps in functional neuroimaging using the false discovery rate
.
Neuroimage
 .
15
:
870
878
.
Gross
J
,
Kujala
J
,
Hamalainen
M
,
Timmermann
L
,
Schnitzler
A
,
Salmelin
R
.
2001
.
Dynamic imaging of coherent sources: studying neural interactions in the human brain
.
Proc Natl Acad Sci USA
 .
98
:
694
699
.
Gruber
WR
,
Klimesch
W
,
Sauseng
P
,
Doppelmayr
M
.
2005
.
Alpha phase synchronization predicts P1 and N1 latency and amplitude size
.
Cereb Cortex
 .
15
:
371
377
.
Hamalainen
MS
,
Sarvas
J
.
1987
.
Feasibility of the homogeneous head model in the interpretation of neuromagnetic fields
.
Phys Med Biol
 .
32
:
91
97
.
Hamalainen
MS
,
Sarvas
J
.
1989
.
Realistic conductivity geometry model of the human head for interpretation of neuromagnetic data
.
IEEE Trans Biomed Eng
 .
36
:
165
171
.
Hanslmayr
S
,
Aslan
A
,
Staudigl
T
,
Klimesch
W
,
Herrmann
CS
,
Bauml
KH
.
2007
.
Prestimulus oscillations predict visual perception performance between and within subjects
.
Neuroimage
 .
37
:
1465
1473
.
Heim
S
,
Alter
K
,
Ischebeck
AK
,
Amunts
K
,
Eickhoff
SB
,
Mohlberg
H
,
Zilles
K
,
von Cramon
DY
,
Friederici
AD
.
2005
.
The role of the left Brodmann's areas 44 and 45 in reading words and pseudowords
.
Brain Res Cogn Brain Res
 .
25
:
982
993
.
Helmholtz
H
.
1867
.
Handbuch der physiologischen Optik
 .
Leipzig
:
Leopold Voss
.
Herbster
AN
,
Mintun
MA
,
Nebes
RD
,
Becker
JT
.
1997
.
Regional cerebral blood flow during word and nonword reading
.
Hum Brain Mapp
 .
5
:
84
92
.
Hickok
G
,
Poeppel
D
.
2007
.
The cortical organization of speech processing
.
Nat Rev Neurosci
 .
8
:
393
402
.
Hopf
JM
,
Vogel
E
,
Woodman
G
,
Heinze
HJ
,
Luck
SJ
.
2002
.
Localizing visual discrimination processes in time and space
.
J Neurophysiol
 .
88
:
2088
2095
.
Ito
J
,
Maldonado
P
,
Singer
W
,
Grun
S
.
2011
.
Saccade-related modulations of neuronal excitability support synchrony of visually elicited spikes
.
Cereb Cortex
 .
21
:
2482
2497
.
Jauk
E
,
Benedek
M
,
Neubauer
AC
.
2012
.
Tackling creativity at its roots: evidence for different patterns of EEG alpha activity related to convergent and divergent modes of task processing
.
Int J Psychophysiol
 .
84
:
219
225
.
Jensen
O
,
Gelfand
J
,
Kounios
J
,
Lisman
JE
.
2002
.
Oscillations in the alpha band (9–12 Hz) increase with memory load during retention in a short-term memory task
.
Cereb Cortex
 .
12
:
877
882
.
Jensen
O
,
Mazaheri
A
.
2010
.
Shaping functional architecture by oscillatory alpha activity: gating by inhibition
.
Front Hum Neurosci
 .
4
:
186
.
Johnson
JS
,
Sutterer
DW
,
Acheson
DJ
,
Lewis-Peacock
JA
,
Postle
BR
.
2011
.
Increased alpha-band power during the retention of shapes and shape-location associations in visual short-term memory
.
Front Psychol
 .
2
:
128
.
Joseph
JE
,
Cerullo
MA
,
Farley
AB
,
Steinmetz
NA
,
Mier
CR
.
2006
.
fMRI correlates of cortical specialization and generalization for letter processing
.
Neuroimage
 .
32
:
806
820
.
Joseph
JE
,
Gathers
AD
,
Piper
GA
.
2003
.
Shared and dissociated cortical regions for object and letter processing
.
Brain Res Cogn Brain Res
 .
17
:
56
67
.
Kleinschmidt
A
,
Buchel
C
,
Hutton
C
,
Friston
KJ
,
Frackowiak
RS
.
2002
.
The neural structures expressing perceptual hysteresis in visual letter recognition
.
Neuron
 .
34
:
659
666
.
Klimesch
W
.
2011
.
Evoked alpha and early access to the knowledge system: the P1 inhibition timing hypothesis
.
Brain Res
 .
1408
:
52
71
.
Klimesch
W
.
2012
.
Alpha-band oscillations, attention, and controlled access to stored information
.
Trends Cogn Sci
 .
16
:
606
617
.
Lee
H
,
Noppeney
U
.
2011
.
Physical and perceptual factors shape the neural mechanisms that integrate audiovisual signals in speech comprehension
.
J Neurosci
 .
31
:
11338
11350
.
Linkenkaer-Hansen
K
,
Nikulin
VV
,
Palva
S
,
Ilmoniemi
RJ
,
Palva
JM
.
2004
.
Prestimulus oscillations enhance psychophysical performance in humans
.
J Neurosci
 .
24
:
10186
10190
.
Liu
J
,
Li
J
,
Zhang
H
,
Rieth
CA
,
Huber
DE
,
Li
W
,
Lee
K
,
Tian
J
.
2010
.
Neural correlates of top-down letter processing
.
Neuropsychologia
 .
48
:
636
641
.
Lundqvist
M
,
Herman
P
,
Lansner
A
.
2013
.
Effect of prestimulus alpha power, phase, and synchronization on stimulus detection rates in a biophysical attractor network model
.
J Neurosci
 .
33
:
11817
11824
.
Makeig
S
,
Westerfield
M
,
Jung
TP
,
Enghoff
S
,
Townsend
J
,
Courchesne
E
,
Sejnowski
TJ
.
2002
.
Dynamic brain sources of visual evoked responses
.
Science
 .
295
:
690
694
.
Maris
E
,
Oostenveld
R
.
2007
.
Nonparametric statistical testing of EEG- and MEG-data
.
J Neurosci Methods
 .
164
:
177
190
.
Mattingly
IG
.
1972
.
Reading, the linguistic process, and linguistic awareness
. In:
Kavanagh
JF
,
Mattingly
IG
, editors.
Language by Ear and by Eye: The Relationship between Speech and Reading
 . pp.
133
147
.
Cambridge, MA
:
MIT Press
.
Mattout
J
,
Henson
RN
,
Friston
KJ
.
2007
.
Canonical source reconstruction for MEG
.
Comput Intell Neurosci
 .
2007
:
67613
. .
Melloni
L
,
Schwiedrzik
CM
,
Muller
N
,
Rodriguez
E
,
Singer
W
.
2011
.
Expectations change the signatures and timing of electrophysiological correlates of perceptual awareness
.
J Neurosci
 .
31
:
1386
1396
.
Mo
J
,
Schroeder
CE
,
Ding
M
.
2011
.
Attentional modulation of alpha oscillations in macaque inferotemporal cortex
.
J Neurosci
 .
31
:
878
882
.
Moca
VV
,
Tincas
I
,
Melloni
L
,
Muresan
RC
.
2011
.
Visual exploration and object recognition by lattice deformation
.
PLoS ONE
 .
6
:
e22831
.
Nolte
G
.
2003
.
The magnetic lead field theorem in the quasi-static approximation and its use for magnetoencephalography forward calculation in realistic volume conductors
.
Phys Med Biol
 .
48
:
3637
3652
.
Obleser
J
,
Boecker
H
,
Drzezga
A
,
Haslinger
B
,
Hennenlotter
A
,
Roettinger
M
,
Eulitz
C
,
Rauschecker
JP
.
2006
.
Vowel sound extraction in anterior superior temporal cortex
.
Hum Brain Mapp
 .
27
:
562
571
.
Oldfield
RC
.
1971
.
The assessment and analysis of handedness: the Edinburgh inventory
.
Neuropsychologia
 .
9
:
97
113
.
Oostenveld
R
,
Fries
P
,
Maris
E
,
Schoffelen
JM
.
2011
.
FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data
.
Comput Intell Neurosci
 .
2011
:
156869
.
Overgaard
M
,
Rote
J
,
Mouridsen
K
,
Ramsoy
TZ
.
2006
.
Is conscious perception gradual or dichotomous? A comparison of report methodologies during a visual task
.
Conscious Cogn
 .
15
:
700
708
.
Palva
S
,
Linkenkaer-Hansen
K
,
Naatanen
R
,
Palva
JM
.
2005
.
Early neural correlates of conscious somatosensory perception
.
J Neurosci
 .
25
:
5248
5258
.
Palva
S
,
Palva
JM
.
2007
.
New vistas for alpha-frequency band oscillations
.
Trends Neurosci
 .
30
:
150
158
.
Paul
P
,
Pennell
ML
,
Lemeshow
S
.
2013
.
Standardizing the power of the Hosmer-Lemeshow goodness of fit test in large data sets
.
Stat Med
 .
32
:
67
80
.
Poldrack
RA
,
Wagner
AD
,
Prull
MW
,
Desmond
JE
,
Glover
GH
,
Gabrieli
JD
.
1999
.
Functional specialization for semantic and phonological processing in the left inferior prefrontal cortex
.
Neuroimage
 .
10
:
15
35
.
Premereur
E
,
Vanduffel
W
,
Janssen
P
.
2012
.
Local field potential activity associated with temporal expectations in the macaque lateral intraparietal area
.
J Cogn Neurosci
 .
24
:
1314
1330
.
Price
CJ
,
Devlin
JT
.
2011
.
The interactive account of ventral occipitotemporal contributions to reading
.
Trends Cogn Sci
 .
15
:
246
253
.
Raghavachari
S
,
Lisman
JE
,
Tully
M
,
Madsen
JR
,
Bromfield
EB
,
Kahana
MJ
.
2006
.
Theta oscillations in human cortex during a working-memory task: evidence for local generators
.
J Neurophysiol
 .
95
:
1630
1638
.
Raij
T
.
1999
.
Patterns of brain activity during visual imagery of letters
.
J Cogn Neurosci
 .
11
:
282
299
.
Raij
T
,
Uutela
K
,
Hari
R
.
2000
.
Audiovisual integration of letters in the human brain
.
Neuron
 .
28
:
617
625
.
Rohenkohl
G
,
Nobre
AC
.
2011
.
Alpha oscillations related to anticipatory attention follow temporal expectations
.
J Neurosci
 .
31
:
14076
14084
.
Sadowsky
J
.
1996
.
Investigation of signal characteristics using the continuous wavelet transforms
.
Johns Hopkins APL Technical Digest
 .
17
:
258
269
.
Sandberg
K
,
Timmermans
B
,
Overgaard
M
,
Cleeremans
A
.
2010
.
Measuring consciousness: is one measure better than the other?
Conscious Cogn
 .
19
:
1069
1078
.
Sanders
LL
,
Auksztulewicz
R
,
Hohlefeld
FU
,
Busch
NA
,
Sterzer
P
.
2014
.
The influence of spontaneous brain oscillations on apparent motion perception
.
Neuroimage
 .
102 Pt 2
:
241
248
.
Series
P
,
Seitz
AR
.
2013
.
Learning what to expect (in visual perception)
.
Front Hum Neurosci
 .
7
:
668
.
Sinkkonen
J
,
Tiitinen
H
,
Naatanen
R
.
1995
.
Gabor filters: an informative way for analysing event-related brain activity
.
J Neurosci Methods
 .
56
:
99
104
.
Sivonen
P
,
Maess
B
,
Lattner
S
,
Friederici
AD
.
2006
.
Phonemic restoration in a sentence context: evidence from early and late ERP effects
.
Brain Res
 .
1121
:
177
189
.
Spitsyna
G
,
Warren
JE
,
Scott
SK
,
Turkheimer
FE
,
Wise
RJ
.
2006
.
Converging language streams in the human temporal lobe
.
J Neurosci
 .
26
:
7328
7336
.
Szwed
M
,
Dehaene
S
,
Kleinschmidt
A
,
Eger
E
,
Valabregue
R
,
Amadon
A
,
Cohen
L
.
2011
.
Specialization for written words over objects in the visual cortex
.
Neuroimage
 .
56
:
330
344
.
Szwed
M
,
Qiao
E
,
Jobert
A
,
Dehaene
S
,
Cohen
L
.
2014
.
Effects of literacy in early visual and occipitotemporal areas of Chinese and French readers
.
J Cogn Neurosci
 .
26
:
459
475
.
van Atteveldt
N
,
Formisano
E
,
Goebel
R
,
Blomert
L
.
2004
.
Integration of letters and speech sounds in the human brain
.
Neuron
 .
43
:
271
282
.
van Atteveldt
N
,
Roebroeck
A
,
Goebel
R
.
2009
.
Interaction of speech and script in human auditory cortex: insights from neuro-imaging and effective connectivity
.
Hear Res
 .
258
:
152
164
.
van Atteveldt
NM
,
Formisano
E
,
Goebel
R
,
Blomert
L
.
2007
.
Top-down task effects overrule automatic multisensory responses to letter-sound pairs in auditory association cortex
.
Neuroimage
 .
36
:
1345
1360
.
van Dijk
H
,
Schoffelen
JM
,
Oostenveld
R
,
Jensen
O
.
2008
.
Prestimulus oscillatory activity in the alpha band predicts visual discrimination ability
.
J Neurosci
 .
28
:
1816
1823
.
van Kerkoerle
T
,
Self
MW
,
Dagnino
B
,
Gariel-Mathis
MA
,
Poort
J
,
van der Togt
C
,
Roelfsema
PR
.
2014
.
Alpha and gamma oscillations characterize feedback and feedforward processing in monkey visual cortex
.
Proc Natl Acad Sci USA
 .
111
:
14332
14341
.
Vilares
I
,
Kording
K
.
2011
.
Bayesian models: the structure of the world, uncertainty, behavior, and the brain
.
Ann N Y Acad Sci
 .
1224
:
22
39
.
Vinckier
F
,
Dehaene
S
,
Jobert
A
,
Dubus
JP
,
Sigman
M
,
Cohen
L
.
2007
.
Hierarchical coding of letter strings in the ventral stream: dissecting the inner organization of the visual word-form system
.
Neuron
 .
55
:
143
156
.
von Stein
A
,
Chiang
C
,
Konig
P
.
2000
.
Top-down processing mediated by interareal synchronization
.
Proc Natl Acad Sci USA
 .
97
:
14748
14753
.
von Stein
A
,
Sarnthein
J
.
2000
.
Different frequencies for different scales of cortical integration: from local gamma to long range alpha/theta synchronization
.
Int J Psychophysiol
 .
38
:
301
313
.
Wang
XJ
.
2010
.
Neurophysiological and computational principles of cortical rhythms in cognition
.
Physiol Rev
 .
90
:
1195
1268
.