The autonomic brain: Multi-dimensional generative hierarchical modelling of the autonomic connectome

The autonomic nervous system governs the body's multifaceted internal adaptation to diverse changes in the external environment, a role more complex than is accessible to the methods—and data scales—hitherto used to illuminate its operation. Here we apply generative graphical modelling to large-scale multimodal neuroimaging data encompassing normal and abnormal states to derive a comprehensive hierarchical representation of the autonomic brain. We demonstrate that whereas conventional structural and functional maps identify regions jointly modulated by parasympathetic and sympathetic systems, only graphical analysis discriminates between them, revealing the cardinal roles of the autonomic system to be mediated by high-level distributed interactions. We provide a novel representation of the autonomic system—a multidimensional, generative network—that renders its richness tractable within future models of its function in health and disease.

based data, including curvature and sulcal depth maps. This method employed both intensity and continuity information from the entire 3D MR volume in segmentation and deformation procedures to produce representations of cortical thickness, calculated as the closest distance from the gray/white boundary to the gray/CSF boundary at each vertex on the tessellated surface. The maps are created using spatial intensity gradients across tissue classes and are therefore not fully reliant on absolute signal intensity. Maps produced are not restricted to voxel resolution of the original data, thus can detect submillimetre differences between groups.
Procedures for measurement of cortical thickness have been validated against histological analysis and manual measurement. Statistical analysis was to investigate the left and right cortical hemispheres with regards to cortical thickness and cortical volumes.

Subcortical-specific measures | FSL FIRST
Vertex and volumetric analyses were performed using FSL-FIRST 5.0, a Bayesian modelling toolkit developed for segmentation, shape and volumetric analysis of the subcortex (Patenaude, Smith, Kennedy, & Jenkinson, 2011). This package firstly skull-strips with the FSL brain extraction tool (BET), following which performs segmentation of each patient structural scan into 15 subcortical structures: bilateral nucleus accumbens, amygdala, caudate, hippocampus, pallidum, putamen and thalamus, as well as the brainstem. Segmented structures are then registered to the Montreal Neurological Institute (MNI) 152 1mm template. The morphometry (shape deformations) of the subcortex were statistically tested. Subcortical volumes were extracted using FSL-STATS.

Network-based statistics
Network-based statistic (NBS) is a non-parametric statistical method which corrects for multiple comparisons and controls for family-wise error rate (FWE). The NBS is a graph analogue of cluster-based statistical methods used in mass-univariate testing on all voxels in an image and produces clusters in topological space, as opposed to physical space. The NBS relies on permutation testing (Freedman & Lane method (Freedman & Lane, 1983)) to determine significance within the GLM, which includes regression of nuisance predictors, permuting resulting residuals and subsequently adding permuted residuals back to nuisance signal to give a realisation of data under the null hypothesis. This approach recognizes that permuting raw data is not desirable as it may engender some variability explained by nuisance predictors. Rather, it is the error terms that can be permuted and estimated under the null hypothesis as a part of the data not explained by the nuisance regressors; that is, the residuals (Anderson & Robinson, 2008). The method permits derivation of FWER-corrected p values using permutation testing when investigating brain networks (Sporns, Tononi, & Kötter, 2005).

Sympathovagal balance
We identified a significant positive correlation of cortical thickness to the RMSSD/CSI ratio involving the anterior cingulate gyrus, paracingulate and orbitofrontal cortex, insula, and precentral gyrus (all p=0.0002), inferior frontal gyrus and superior frontal gyrus (all p=0.007), angular gyrus, supramarginal gyrus, lateral occipital cortex and middle temporal gyrus (all p=0.02). On the left hemisphere, we identified a negative correlation of cortical thickness to RMSSD/CSI ratio involving the lateral occipital cortex (p=0.02) (Supplementary Figure 3). No significant relationships to cortical volume were identified. Subcortical analysis identified shape changes at the left accumbens (p=0.04), right caudate (p=0.03) and left thalamus (p=0.03), contingent on the RMSSD/CSI ratio (Supplementary Figure 1). Network based statistics identified a gray-matter morphometric network negatively related to RMSSD/CSI, consisting of 87 nodes and 100 edges (p=0.04), including the frontal pole/orbitofrontal cortex (degree 45), insula (degree 6), cingulate (degree 6) and caudate nucleus (degree 5).

Sympathovagal balance
With tract based spatial statistics, there were no significant differences in white matter skeleton fractional anisotropy associated to the RMSSD/CSI ratio. Similarly, network-based statistics did not identify a network specific to sympathovagal balance after correction for multiple comparisons. proportional to summed z-statistics from functional, gray-matter morphometry and tract based spatial statistics. Edge width is proportional to summed network-based statistics. Hierarchical node colour is proportional to its coherence with the alternate autonomic contrast.

Sympathovagal balance
Representative regions implicated in the parcellation are colour-coded according to the colour of the lowest level community of the hierarchical model.

Supplementary Movie 1:
Dynamic representation of generating the multi-modal, highdimensional autonomic connectome. Node colour and size is proportionate to the effect size of its involvement in autonomic nervous system function. Edge colour and width is proportionate to the sum of its effect size across both gray matter, tractography and functional brain networks. Individual frames (60fps) correspond to single iterations of the model. Note how regions which are 'more connected', i.e., those with wider and brighter yellow edges, will begin to move closer together yet those which are 'less connected', i.e., those with thinner more purple edges, will be pushed away from the centre of the graph as the procedure iterates.