Effective Connectivity and Inter-Subject Variability: using a Multi-Subject Network to test Differences and Commonalities

Andrea Mechelli, Will Penny, Cathy Price, Darren Gitelman, Karl Friston, Karl Friston, Karl Friston

Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, U.K.
Cognitive Neurology and Alzheimer’s Disease Center, Northwestern University, Chicago, USA

Subject: Modeling & Analysis

Abstract
Introduction
Previous studies of effective connectivity have dealt with inter-subject variability by analysing data from different subjects separately or pretending they came from the same subject. The aim of this work is to illustrate how differences in connectivity among subjects can be addressed explicitly using Structural Equation Modelling (SEM). This involves constructing a multi-subject network which comprises m regions of interest for each of the n subjects studied, resulting in a total of m x n nodes. Constructing a network of regions from different subjects (i) allows differences among subjects to be tested directly, and (ii) provides additional degrees of freedom to estimate the model’s free parameters. We present an fMRI study which used a multi-subject network to investigate inter-subject variability in effective connectivity in the context of single word and pseudoword reading.

Methods
Thirteen subjects were scanned while silently reading words and pseudowords presented with different stimulus duration (i.e. 200, 600 and 1000 msec) and while resting. Data were analysed using SPM99.
Based on the neuroanatomical and neuroimaging literature, we selected 4 left-lateralized regions (fusiform, posterior inferior temporal, posterior superior temporal and insula) that activated when reading both words and pseudowords in 10 subject (p<0.001 uncorrected). A multi-subject network was then constructed, which contained 10 x 4 = 40 nodes and all possible reciprocal connections within each subject (see Figure 1). The common influence of exogenous stimulation was modelled by connecting a virtual node, whose time course corresponded to stimulus onsets, with the fusiform gyrus in all subjects. This analytical device allowed us to model both endogenous (or intrinsic) variance and exogenous variance induced by the experimental design. In order to ensure that each parameter of the model could be estimated uniquely, reciprocal connections were set to have the same value within each subject. The effects of word type on the reading-induced coupling were modelled using first order interaction terms1. Inter-subject variability of this modulatory effect was assessed independently for each connection by comparing a free model (in which the effects of word type on connectivity were free to vary across subjects) with a restricted model (in which the effects of word type on connectivity were set to be the same in all subjects). SEM analysis was performed using the SEM Toolbox of SPM99. Inferences were made at p<0.05.

Results
A number of functional connections were stronger when reading pseudowords relative to words. These included both forward (Fusiform→Inferior Temporal, Fusiform→Insula, and Superior Temporal→Insula) and backward (Insula→Superior Temporal) connections. The backward connection between the Insula and the Superior Temporal region showed further significant inter-subject variability.

Conclusions
Our findings illustrate that (i) differences between processing words and pseudowords can be characterised in terms of context-sensitive interactions among brain areas, and (ii) a pattern of connectivity estimated over subjects may not be a good approximation to the underlying patterns of connectivity in all of the subjects studied. Rather, in some cases a model which allows for inter-subject variability would be preferred.

References
1 Büchel C. and Friston K.J. 1997. Cereb. Cortex 7:768-78.