Prestimulus feedback connectivity biases the content of visual experiences

Abstract

Ongoing fluctuations in neural excitability and in networkwide activity patterns before stimulus onset have been proposed to underlie variability in near-threshold stimulus detection paradigms-that is, whether or not an object is perceived. Here, we investigated the impact of prestimulus neural fluctuations on the content of perception-that is, whether one or another object is perceived. We recorded neural activity with magnetoencephalography (MEG) before and while participants briefly viewed an ambiguous image, the Rubin face/vase illusion, and required them to report their perceived interpretation in each trial. Using multivariate pattern analysis, we showed robust decoding of the perceptual report during the poststimulus period. Applying source localization to the classifier weights suggested early recruitment of primary visual cortex (V1) and ∼160-ms recruitment of the category-sensitive fusiform face area (FFA). These poststimulus effects were accompanied by stronger oscillatory power in the gamma frequency band for face vs. vase reports. In prestimulus intervals, we found no differences in oscillatory power between face vs. vase reports in V1 or in FFA, indicating similar levels of neural excitability. Despite this, we found stronger connectivity between V1 and FFA before face reports for low-frequency oscillations. Specifically, the strength of prestimulus feedback connectivity (i.e., Granger causality) from FFA to V1 predicted not only the category of the upcoming percept but also the strength of poststimulus neural activity associated with the percept. Our work shows that prestimulus network states can help shape future processing in category-sensitive brain regions and in this way bias the content of visual experiences.

Keywords: MEG; connectivity; oscillations; prestimulus; visual object perception.

Fig. 1.

Poststimulus MEG data contains category-sensitive information with respect to the processing of the Rubin face/vase stimulus. (A) Temporal decoding of face vs. vase reports. * represents P = 0.0001 significance of decoding accuracy (t test vs. chance) starting at 100 ms poststimulus. (B) Unmasked activation maps resulting from the source reconstruction of the classifier weights (in arbitrary units [a.u.]), applying a procedure proposed by ref. , at different time points, suggesting temporally changing informative regions (V1 ∼100 ms and FFA ∼160 ms after stimulus onset). While the unmasked plots suggest a larger spatial spread of activity at 160 ms compared with 100 ms, applying a 95%-maximum activity threshold to extract the ROIs revealed 32 grid points (8-mm resolution) exceeding the threshold at 100 ms (V1) and 1 grid point exceeding the threshold at 160 ms (FFA). Black dots represent the grid points with maximum informative activity (V1 and FFA) in the respective time periods. (C) Time-frequency contrast in V1 (face vs. vase reports). Colors represent smoothed T-values obtained from cluster-based permutation testing of the contrast (face – vase; ns). (D) Time-frequency contrast in FFA (face vs. vase reports). Colors represent smoothed T-values obtained from cluster-based permutation testing of the contrast (face – vase; p_cluster = 0.029). Black lines surround the time-frequency gamma-range cluster that drove the significant statistical difference.

Fig. 2.

Prestimulus MEG connectivity is predictive of upcoming perceptual decisions. Shaded error regions represent SEM for within-subject designs (46). (A) No statistical differences in prestimulus spectral power between face and vase trials in either V1 (Left) or FFA (Right). (B) Compared with vase trials, face trials show increased prestimulus coherence between V1 and FFA in the alpha and beta frequency ranges. (C) Compared with vase trials, face trials show increased prestimulus feedback connectivity from FFA to V1 in the alpha range (Right) but no differences in prestimulus feedforward connectivity from V1 to FFA (Left) (49).

Fig. 3.

Prestimulus connectivity is correlated with poststimulus activity across participants; r values represent Pearson’s correlation coefficients, and shaded areas represent 95% confidence intervals. (A) Maximum prestimulus feedback Granger causality estimates are correlated with maximum poststimulus gamma differences (face – vase). (B) Maximum prestimulus feedback Granger causality estimates are correlated with maximum poststimulus decoding (AUC) scores. (C) Maximum poststimulus gamma differences (face – vase) are correlated with maximum poststimulus decoding (AUC) scores.

Fig. 4.

Trial structure.