Intersubject consistent dynamic connectivity during natural vision revealed by functional MRI

Abstract

The functional communications between brain regions are thought to be dynamic. However, it is usually difficult to elucidate whether the observed dynamic connectivity is functionally meaningful or simply due to noise during unconstrained task conditions such as resting-state. During naturalistic conditions, such as watching a movie, it has been shown that local brain activities, e.g. in the visual cortex, are consistent across subjects. Following similar logic, we propose to study intersubject correlations of the time courses of dynamic connectivity during naturalistic conditions to extract functionally meaningful dynamic connectivity patterns. We analyzed a functional MRI (fMRI) dataset when the subjects watched a short animated movie. We calculated dynamic connectivity by using sliding window technique, and quantified the intersubject correlations of the time courses of dynamic connectivity. Although the time courses of dynamic connectivity are thought to be noisier than the original signals, we found similar level of intersubject correlations of dynamic connectivity to those of regional activity. Most importantly, highly consistent dynamic connectivity could occur between regions that did not show high intersubject correlations of regional activity, and between regions with little stable functional connectivity. The analysis highlighted higher order brain regions such as the default mode network that dynamically interacted with posterior visual regions during the movie watching, which may be associated with the understanding of the movie.

Keywords: Default mode network; Dynamic connectivity; Intersubject correlation; Movie connectome; Naturalistic condition; Supramarginal gyrus.

Figure 1

Illustration of the calculations of intersubject correlations of the time series of regional activity (A) and the time courses of dynamic connectivity between two regions (B).

Figure 2

Intersubject correlation maps of regional activity (A) and dynamic connectivity with different seeds (B through G). The seed regions were depicted in blue or green in respective rows. All voxels with positive correlations are shown. The numbers at the bottom represent z or x coordinates in Montreal Neurological Institute (MNI) space. LMV, left medial visual; RMV, right medial visual; LLV, left lateral visual; RLV, right lateral visual; LSMG, left supramarginal gyrus; RSMG, right supramarginal gyrus. All the maps are available at: https://identifiers.org/neurovault.collection:6245.

Figure 3

Differential intersubject correlations of dynamic connectivity among the medial visual, lateral visual, and supramarginal gyrus seeds (depicted on the left). All maps were thresholded at p < 0.001, and cluster thresholded at p < 0.0167 (0.05 / 3) after family-wise error (FWE) correction using nonparametric method. MV, medial visual; LV, lateral visual; and SM, supramarginal gyrus. Unthresholded statistical maps are available at: https://identifiers.org/neurovault.collection:6245.

Figure 4

Correlation matrices among the 9 regions of interest (ROI) using different methods. A) Mean functional connectivity across the 29 subjects. B) Correlations of the consistent component of each ROI (averaged time series across the 29 subjects). C) Intersubject correlations of dynamic connectivity. LMV, left medial visual; RMV, right medial visual; LLV, left lateral visual; RLV, right lateral visual; LSMG, left supramarginal gyrus; RSMG, right supramarginal gyrus; LPCG, left precentral gyrus; PCC, posterior cingulate cortex; and MPFC, medial prefrontal cortex.

Figure 5

Time courses of dynamic connectivity (Fisher’s z) for three pairs of regions of interest. Each thinner line represents the time course of one subject, and the thicker red lines represent the averaged time courses. LMV, left medial visual; RMV, right medial visual; LPCG, left precentral gyrus; RSMG, right supramarginal gyrus; PCC, posterior cingulate cortex.

Figure 6

A) and B) Intersubject correlations (ISC) of dynamic connectivity calculated from raw time series (A) and residual time series after regressing out the intersubject consistent components (B). C) Dynamic connectivity of the consistent component of regional activity between right supramarginal gyrus (RSMG) and posterior cingulate cortex (PCC). D) and E) Time courses of dynamic connectivity (Fisher’s z) between RSMG and PCC calculated from raw time series (D) and the residual time series after regressing out the intersubject consistent component. F) The time courses of intersubject correlations of regional activity in RSMG and PCC.

Figure 7

A) The effects of sliding-window length on the intersubject correlations of dynamic connectivity. B) Intersubject correlations of dynamic connectivity of three pairs of regions of interest: left medial visual (LMV) and right medial visual (RMV), LMV and left precentral gyrus (LPCG), and right supramarginal gyrus (RSMG) and posterior cingulate cortex (PCC). C) The time courses of dynamic connectivity between RSMG and PCC in three window lengths. TR, repetition time.