Dynamic Causal Modelling and physiological confounds: A functional MRI study of vagus nerve stimulation

Abstract

Dynamic Causal Modelling (DCM) has been proposed to estimate neuronal connectivity from functional magnetic resonance imaging (fMRI) using a biophysical model that links synaptic activity to hemodynamic processes. However, it is well known that fMRI is sensitive not only to neuronal activity, but also to many other psychophysiological responses which may be task-related, such as changes in cardio-respiratory activity. They are not explicitly taken into account in the generative models of DCM and their effects on estimated neuronal connectivity are not known. The main goal of this study was to report the face validity of DCM in the presence of strong physiological confounds that presumably cannot be corrected for, using an fMRI experiment of vagus nerve stimulation (VNS) performed in rats. First, a simple simulation was used to evaluate the principled ability of DCM to recover directed connectivity in the presence of a confounding factor. Second, we tested the experimental validity using measures of the BOLD correlates of left 5Hz VNS. Because VNS mostly activates the central autonomic regulation system, fMRI signals were likely to represent both direct and indirect vascular responses to such activation. In addition to the inference of standard statistical parametric maps, DCM was thus used to estimate directed neural connectivity in a small brain network including the nucleus tractus solitarius (NTS) known to receive vagal afferents. Though blood pressure changes may constitute a major physiological confound in this dataset, model comparison of DCMs still allowed the identification of the NTS as the input station of the VNS pathway to the brain. Our study indicates that current developments of DCM are robust to psychophysiological responses to some extent, but does not exclude the need to develop specific models of brain – body interactions within the DCM framework to better estimate neuronal connectivity from fMRI time series.