Granger causal time-dependent source connectivity in the somatosensory network

Abstract

Exploration of transient Granger causal interactions in neural sources of electrophysiological activities provides deeper insights into brain information processing mechanisms. However, the underlying neural patterns are confounded by time-dependent dynamics, non-stationarity and observational noise contamination. Here we investigate transient Granger causal interactions using source time-series of somatosensory evoked magnetoencephalographic (MEG) elicited by air puff stimulation of right index finger and recorded using 306-channel MEG from 21 healthy subjects. A new time-varying connectivity approach, combining renormalised partial directed coherence with state space modelling, is employed to estimate fast changing information flow among the sources. Source analysis confirmed that somatosensory evoked MEG was mainly generated from the contralateral primary somatosensory cortex (SI) and bilateral secondary somatosensory cortices (SII). Transient Granger causality shows a serial processing of somatosensory information, 1) from contralateral SI to contralateral SII, 2) from contralateral SI to ipsilateral SII, 3) from contralateral SII to contralateral SI, and 4) from contralateral SII to ipsilateral SII. These results are consistent with established anatomical connectivity between somatosensory regions and previous source modeling results, thereby providing empirical validation of the time-varying connectivity analysis. We argue that the suggested approach provides novel information regarding transient cortical dynamic connectivity, which previous approaches could not assess.

Figure 1

(a) Spatial distribution of MEG averaged responses. At each sensor location, the traces illustrate signals recorded by two orthogonal gradiometers. (b) Group averaged waveforms from each channel are plotted in different colors and superimposed and scalp topographies at 71 ms and 201 ms peak latencies are displayed.

Figure 2

Representative source locations and the corresponding grand average source time courses of somatosensory MEG responses from all participants are presented.

Figure 3

Time-frequency representations of rPDC as a measure of time-dependent Granger causal influences within the neural network between SI-l, SII-l and SII-r averaged across all the subjects. The regions circled by blue lines had significantly larger rPDC values than those in the reference interval from -100 ms to 0 ms (P < 0.01).

Figure 4

Graph obtained from rPDC analysis applied to source time series located at SI-l, SII-l and SII-r for somatosensory evoked MEG.