A connectome-based mechanistic model of focal cortical dysplasia

Abstract

Neuroimaging studies have consistently shown distributed brain anomalies in epilepsy syndromes associated with a focal structural lesion, particularly mesiotemporal sclerosis. Conversely, a system-level approach to focal cortical dysplasia has been rarely considered, likely due to methodological difficulties in addressing variable location and topography. Given the known heterogeneity in focal cortical dysplasia histopathology, we hypothesized that lesional connectivity consists of subtypes with distinct structural signatures. Furthermore, in light of mounting evidence for focal anomalies impacting whole-brain systems, we postulated that patterns of focal cortical dysplasia connectivity may exert differential downstream effects on global network topology. We studied a cohort of patients with histologically verified focal cortical dysplasia type II (n = 27), and age- and sex-matched healthy controls (n = 34). We subdivided each lesion into similarly sized parcels and computed their connectivity to large-scale canonical functional networks (or communities). We then dichotomized connectivity profiles of lesional parcels into those belonging to the same functional community as the focal cortical dysplasia (intra-community) and those adhering to other communities (inter-community). Applying hierarchical clustering to community-reconfigured connectome profiles identified three lesional classes with distinct patterns of functional connectivity: decreased intra- and inter-community connectivity, a selective decrease in intra-community connectivity, and increased intra- as well as inter-community connectivity. Hypo-connectivity classes were mainly composed of focal cortical dysplasia type IIB, while the hyperconnected lesions were type IIA. With respect to whole-brain networks, patients with hypoconnected focal cortical dysplasia and marked structural damage showed only mild imbalances, while those with hyperconnected subtle lesions had more pronounced topological alterations. Correcting for interictal epileptic discharges did not impact connectivity patterns. Multivariate structural equation analysis provided a mechanistic model of such complex, diverging interactions, whereby the focal cortical dysplasia structural makeup shapes its functional connectivity, which in turn modulates whole-brain network topology.

Keywords: MRI; brain networks; epilepsy; focal cortical dysplasia; resting state fMRI.

Figure 1

Functional analysis framework. (A) Using whole-brain connectivity mapped on the AAL-derived 1000 regions of interest parcellation, we identified canonical functional communities based on control and patient cohorts. (B) Applying the same parcellation scheme, we subdivided every FCD into lesion parcels and extracted seed-based functional connectivity. After, normalizing the lesion parcels connectivity matrix L with respect to controls, connection profiles were split into same functional community (intra-community) and different communities (inter-community). (C) Connectivity profiles were reordered into intra- and inter-communities, and sorted by their z-scores. (D) After estimating similarity in intra- and inter-community connectivity between parcels, we applied agglomerative hierarchical clustering to objectively discover biotypes (refer to the ‘Materials and methods’ section for details).

Figure 2

Connectome-based clustering and profiling. Hierarchical clustering identified three classes based on the goodness of fit index (i.e. the ratio of intra-cluster over inter-cluster distance shown in the left panel). In the reordered connectivity matrix, the flame scale indicates connectivity strength normalized with respect to healthy controls. (A) Average z-scores of intra- and inter-community connectivity (asterisks indicate difference at **FDR < 0.05/*tendency compared to controls). (B) Proportion of functional communities across classes. (C) Average z-scores of cortical thickness, grey and white matter (GM, WM) interface blurring, and intensity. (D) Direct correlation between averaged functional connectivity and structural anomalies (z-scored) without clustering stratification.

Figure 3

Within-lesion heterogeneity in structure and function. Lesional classes are shown in two patients (Case 1: FCD type IIA; Case 2: FCD type IIB). Classes are colour-coded and projected onto the cortical surface model (left). For each class, the averaged inter- and intra-community functional connectivity (FC) strength and the load of averaged structural anomalies (SA) normalized with respect to healthy controls (z-score) is indicated.

Figure 4

Global functional network topology. Bar and line plots display topological parameters (clustering, C^w; path length, L^w; modularity, Q^w; rich-club coefficient, ${\phi }_{norm}^w$) in controls and patients, as well as separately for each lesional class. Asterisks indicate ^**significant difference (FDR < 0.05) and *tendency compared to controls. Right: In the graph, the grey area shows the ‘rich club’ regime in controls; k represents degree levels (*group differences at FDR < 0.05).

Figure 5

Mechanistic model of structure-function relationship in FCD. Structural equation modelling defining the relation between latent and observed variables. The numbers on each arrow represent path coefficient and its standardized z-score, indicating statistical causality between variables. ^†Fixed coefficient (=1) providing a scaling factor relative to other variables. The fitness of the model was evaluated by chi-square, root mean square error approximation (RMSEA) and comparative fit index (CFI).