Article Figures & Data
Figures
- Figure 1.
A schematic representation of the data processing and analyses pipeline. A, Neurophysiological signals were recorded from 18 participants during the McGurk task using a 64-channel EEG system. B, Prestimulus EEG data were extracted from a time window spanning −800 to 0 ms relative to stimulus onset. C, The raw prestimulus EEG data was filtered, rereferenced, and subjected to independent component analysis (ICA) for artefact (eye and muscle) removal. D, PSDs were estimated for all sensors and trials for all participants during illusory and nonillusory McGurk trial conditions. E, Parameterization of the PSDs to estimate aperiodic and periodic activity using SpecParam (earlier called FOOOF) model. From the model, periodic parameters, center frequency, peak power, and bandwidth, and aperiodic parameters, offset and exponent, were estimated for all 64 sensors, across different frequency bands: theta (4–7 Hz), alpha (8–12 Hz), beta (15–30 Hz), and gamma (31–45 Hz). The parameters from all 64 EEG sensors were classified into six sensor regions: frontal, central, parietal, occipital, left temporal, and right temporal (see topoplot). F, These periodic and aperiodic parameters were subsequently fitted as predictors in sensor region-wise logistic mixed-effects models to predict the behavioral response. To counter for intersubject variability, subject ID was put as random effect in the model. G, Bayes factor analysis was performed as a post hoc test to validate the evidence of significant predictors estimated in the regression model. H, Aperiodic adjusted power was estimated by subtracting the original PSD from aperiodic fit in linear space, a.u., arbitrary units. I, Frequency-wise logistic mixed-effects interaction models were fitted with frequency power (aperiodic adjusted), aperiodic exponent, and sensor regions as predictors. Subject ID was chosen as the random effect.
- Figure 2.
Prestimulus window relative to trial timing and behavioral results. A, Example trial depicting three video frames from a video used in this experiment and intertrial interval (fixation cross) pseudorandomly varied between 1,200 and 2,800 ms. The red box indicates the 800 ms prestimulus epoch used in this study. B, Behavioral results: i, Bar graph representing interindividual variability–propensity of McGurk effect across 18 participants expressed as the percentage of /ta/ percept during the presentation of the McGurk stimulus. Dark blue represents participants below the median percentage of illusory response (or less prone), and light blue represents participants above the median percentage of illusory response (or more prone). ii, Violin plot showing intertrial variability–percentage of /ta/ (illusory), unisensory /pa/ (auditory), and unisensory /ka/ (visual) percept during the presentation of McGurk stimulus and the congruent AV stimuli (/pa/, /ta/, and /ka/). The white dot in the center of each violin plot represents the median. Significance levels are denoted by asterisks: ****p ≤ 0.0001 and n.s. is not significant.
- Figure 3.
Grand-averaged PSDs and parameterized power distributions. A, Group-averaged PSDs computed from subject-specific spectra, averaged over all sensors and trials within each participant. Mean power spectra are shown for illusory (red) and nonillusory (blue) McGurk trials, with shaded areas representing the standard error of the mean (SEM), separately for the (a) prestimulus and (b) poststimulus periods. B, Prestimulus power distributions derived from parameterized spectra. Panel a shows grand-averaged periodic (oscillatory) power, averaged over all sensors and trials within each participant for illusory (red) and nonillusory (blue) conditions, with SEM shading. Panel b displays box-and-whisker plots overlaid with density distributions for the aperiodic parameters—offset (top) and exponent (bottom)—averaged over sensors and trials within each participant, grouped by perceptual condition (red, illusory; blue, nonillusory). All extended analysis of parametrized power distributions after perception of McGurk trials indicating interindividual (or group-level) variability is shown as Extended Data Figure 3-1 and two different model algorithms were fit on one participant prestimulus McGurk data is shown as Extended Data Figure 3-2.
- Figure 4.
Topographic distribution of t values from linear mixed-effects models for periodic and aperiodic spectral components. Topoplots depicting t values from linear mixed-effects models comparing illusory (/ta/) and nonillusory (/pa/) McGurk trials for mean periodic power across (A) theta, (B) alpha, (C) beta, and (D) gamma frequency bands, as well as aperiodic components: (E) offset and (F) exponent. The color bar indicates the corresponding t values. White-marked sensors denote statistically significant differences between conditions. Periodic power was computed by subtracting the aperiodic fit from the original (untransformed) PSDs in linear space.
- Figure 5.
Bayes factor estimated for all the prestimulus periodic (CF, PW, BW) and aperiodic (offset, exponent) parameter predictors that were fitted to a logistic mixed-effects model to predict response to upcoming McGurk stimulus. The models were fit for (A) all sensor region, (B) frontal, (C) central, (D) parietal, (E) occipital, (F) left temporal, and (G) right temporal. The Bayes factor evidence scale was as follows: extreme evidence, BF > 100; strong evidence, 10 < BF < 100; moderate evidence, 3 < BF < 10; no evidence, BF < 3. For representation purposes, exact BF values have been log base 10 transformed. Summary tables of the logistic mixed-effects models fitted across the whole brain with periodic (CF, PW, BW) and aperiodic (offset, exponent) parameter as predictors is shown as Extended Data Figure 5-1. Summary tables of the logistic mixed-effects models fitted across the frontal, central, parietal, and occipital sensors are displayed as Extended Data Figures 5-2–5-5, and Extended Data Figures 5-6 and 5-7 show summary tables of the logistic mixed-effects models fitted across the left and right temporal sensors, respectively.
- Figure 6.
Association between aperiodic exponent and mean aperiodic adjusted power in the (A) theta, (B) alpha, (C) beta, and (D) gamma frequency bands before perception of illusory (red) and nonillusory (blue) McGurk trials averaged across all channels and trials for each participant. The aperiodic slope is represented on the x-axis and aperiodic adjusted power on the y-axis. The density of observations for each association plot is indicated on the margins for illusory (red) and nonillusory (blue) trial conditions. One participant contributed zero nonillusory trials, so the nonillusory condition includes N = 17. All the extended analysis of association between aperiodic offset and mean periodic power, in the (A) theta, (B) alpha, (C) beta, and (D) gamma frequency bands before perception of illusory (red) and nonillusory (blue) McGurk trials averaged across all trials and channels for each participant is shown as Extended Data Figure 6-1.
- Figure 7.
Visualization of the relationship between prestimulus aperiodic exponent and aperiodic adjusted mean power in predicting McGurk illusory response at (A) occipital theta, (B) alpha, (C) beta, and (D) occipital and parietal gamma band power. For A, C, and D frequency band models, x-axis represents the mean power (aperiodic adjusted) in the frequency range (higher values indicate higher power) and y-axis represents the probability (in percentage) of perceiving the illusion. Note that the distinction of aperiodic exponent into shallow, moderate, and steeper facets are for visualization purposes only and the aperiodic slopes were entered as continuous predictors in all the models. Also, note that B illustrates effect of individual predictors: alpha power (left; x-axis, higher values indicate higher power) and exponent (right; x-axis, absolute higher values indicate steeper slope) for prediction to response (y-axis). The shaded region indicates 83% CI.
Tables
Sensor regions Electrodes Frontal F7, F5, F3, F1, Fz, F2, F4, F6, F8, FP1, FPz, FP2, AF3, AF4 Central FC5, FC3, FC1, FCz, FC2, FC4, FC6, C5, C3, C1, Cz, C2, C4, C6 Parietal P5, P3, P1, Pz, P2, P4, P6, CP5, CP3, CP1, CPz, CP2, CP4, CP6 Occipital PO9, O1, Oz, O2, PO10, PO7, PO5, PO3, POz, PO4, PO6, PO8 Left temporal FT7, T7, TP7, TP9, P7 Right temporal FT8, T8, TP8, TP10, P8 We classified all the 64 EEG sensors into six sensor regions covering bilateral: frontal, central, parietal, occipital, and unilateral left temporal and right temporal.
- Table 2.
Fixed-effects ANOVA (type III Wald chi-square tests) results for logistic mixed interaction effects in the (A) theta, (B) alpha, (C) beta, and (D) gamma frequency bands
Fixed-effects ANOVA results for interaction models Predictors Response Chisq DF Pr (>Chisq) p (adjusted) (A) Theta model (Intercept) 2.1002 1 0.1473 0.2695 Theta power 4.2466 1 0.0393 0.0968 Exponent 22.3972 1 <0.0001 <0.0001*** Sensor regions 2.1016 5 0.8349 0.8618 Theta power × exponent 0.3536 1 0.5521 0.6795 Theta power × sensor regions 2.3918 5 0.7927 0.8455 Exponent × sensor regions 5.4656 5 0.3617 0.5512 Theta power × exponent × sensor regions 18.8219 5 0.0021 0.0066** (B) Alpha model (Intercept) 1.9742 1 0.160 0.2695 Alpha power 25.9706 1 <0.0001 <0.0001*** Exponent 15.3560 1 <0.0001 0.0005*** Sensor regions 5.6547 5 0.3413 0.5461 Alpha power × exponent 0.3849 1 0.5350 0.6795 Alpha power × sensor regions 19.9376 5 0.0013 0.0046** Exponent × sensor regions 5.0297 5 0.4122 0.5736 Alpha power × exponent × sensor regions 9.9510 5 0.077 0.1752 (C) Beta model (Intercept) 2.1394 1 0.1436 0.2695 Beta power 12.9303 1 0.0003 0.0015** Exponent 26.6636 1 <0.0001 <0.0001*** Sensor regions 1.5859 5 0.9029 0.9030 Beta power × exponent 9.2046 1 0.0024 0.0070** Beta power × sensor regions 2.8561 5 0.7221 0.8253 Exponent × sensor regions 5.2032 5 0.3916 0.5696 Beta power × exponent × sensor regions 3.5235 5 0.6198 0.7346 (D) Gamma model (Intercept) 1.9802 1 0.1594 0.2695 Gamma power 0.4862 1 0.4856 0.6475 Exponent 18.3772 1 <0.0001 0.0001*** Sensor regions 17.0688 5 0.0044 0.0116* Gamma power × exponent 0.0775 1 0.7807 0.8455 Gamma power × sensor regions 34.9677 5 <0.0001 <0.0001*** Exponent × sensor regions 8.8578 5 0.1149 0.2451 Gamma power × exponent × sensor regions 21.5860 5 0.00063 0.0025** Each model included mean periodic power and aperiodic exponent as continuous predictors. Significant predictor interactions are highlighted in bold. p values for all fixed effects were calculated using White-corrected covariance matrices (White, 1980). p values were adjusted for multiple-comparison corrections using Benjamini–Hochberg method. Significance levels are denoted by asterisks: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, and ****p ≤ 0.0001.
Data 1
MATLAB and R codes for implementing FOOOF to parameterize power spectra and for fitting logistic mixed-effects models, respectively. Download Data 1, ZIP file.
Figure 3-1
Parametrized power distributions after perception of McGurk trials indicating inter-individual (or group-level) variability. (A) Poststimulus periodic power distributions for the illusory (red) versus non-illusory (blue) McGurk trials with SEM as shaded region. (B) Box and whisker plots with density distributions for comparisons between poststimulus aperiodic offset (above) and exponent (below), by response conditions (red – illusory, and blue – non-illusory). Download Figure 3-1, TIF file.
Figure 3-2
Tuning the SpecParam (or FOOOF) algorithm for better fit. To estimate the best model parameters, two different model algorithms were fit on one participant prestimulus McGurk data (n = 75). For model #1 (red), default fit parameters were chosen with frequency range of 1 – 45 Hz. For model #2 (blue), the frequency range chosen was 3 – 45 Hz. No significant difference between (A) aperiodic exponent estimated by the two model algorithms were observed (Z = -0.0113, p = .9910, Cohen’s D = 0.0837). (B) Visual inspection of mean absolute error (or MAE) indicated that both model algorithms fit the participant data well with majority of trials with MAE < 0.100. (C) Variance explained by model #1 (R2) was slightly higher (M = 0.98, SD = 0.04) as compared to model #2 (M = 0.97, SD = 0.02). Overall, model #1 demonstrated better goodness-of-fit (R2 and MAE), supporting our decision to use the 1–45 Hz frequency range in our study. Download Figure 3-2, TIF file.
Figure 5-1
Summary tables of the logistic mixed-effect models fitted across the whole brain with periodic (CF, PW, BW) and aperiodic (offset, exponent) parameter as predictors. Predictors that significantly predicted the response are in bold. Download Figure 5-1, TIF file.
Figure 5-2
Summary tables of the logistic mixed-effect models fitted across the frontal sensors. Predictors that significantly predicted the response are in bold. Download Figure 5-2, TIF file.
Figure 5-3
Summary tables of the logistic mixed-effect models fitted across the central sensors. Predictors that significantly predicted the response are in bold. Download Figure 5-3, TIF file.
Figure 5-4
Summary tables of the logistic mixed-effect models fitted across the parietal sensors. Predictors that significantly predicted the response are in bold. Download Figure 5-4, TIF file.
Figure 5-5
Summary tables of the logistic mixed-effect models fitted across the occipital sensors. Predictors that significantly predicted the response are in bold. Download Figure 5-5, TIF file.
Figure 5-6
Summary tables of the logistic mixed-effect models fitted across the left temporal sensors. Predictors that significantly predicted the response are in bold. Download Figure 5-6, TIF file.
Figure 5-7
Summary tables of the logistic mixed-effect models fitted across the right temporal sensors. Predictors that significantly predicted the response are in bold. Download Figure 5-7, TIF file.
Figure 6-1
Association between aperiodic offset and mean periodic power, in the (A) Theta, (B) Alpha, (C) Beta, and (D) Gamma frequency bands before perception of illusory (red) and non-illusory (blue) McGurk trials averaged across all trials and channels for each participant. The aperiodic offset is represented on the x-axis and oscillatory power on the y-axis. The density of observations for each association plot is indicated on the margins for illusory (red) and non-illusory (blue) trial parameters. Download Figure 6-1, TIF file.
In this issue
