Principles of regional covariance in brain structure

Abstract: The human brain varies between individuals in both shape and size. These variations are not unique for each brain region. This causes grey matter density between regions to covary, a phenomenon known as “structural brain covariance”. The reasons for this structural covariance, and its possible relations to functional connectivity, remain poorly understood. In order to study this morphological variation, a standard brain atlas – also called a brain template – is commonly used to enable consolidation of information across individuals. In the early days of computerised brain research, this was done by creating an average of a large number of brain images from young adults resampled into a common stereotaxic space known as the “MNI space”, which is still widely used today. However, this space is less ideal for ageing studies since the brain changes structurally with age. The aim of this thesis was to provide a deeper understanding of the principles governing structural covariance across the age span. In Study I, the need of a computerised brain atlas in ageing was addressed by constructing a standard non-linear brain template from 314 older individuals (average age 75 years) together with a regional atlas and corresponding tissue probability maps. This template was constructed to be linearly mapped to MNI space while forming its own non-linear ageing space. The tissue probability maps also allowed us to construct a non-linear transformation to other spaces and warp the regional atlas to other study cohorts. This approach was used in Study II-IV. In Study II, the nature of structural covariance in ageing was investigated by calculating for each grey matter voxel (data point) its number of significant correlations with all the other grey matter voxels in the brain, in a large sample of 960 healthy individuals (age range 68-83 years). Voxels with many significant correlations (known as “hubs”) were found in the basal ganglia, the thalamus, the brainstem, and the cerebellum. No significant difference in the covariance structure could be found between relatively younger (68-75 years) and older (76-83 years) individuals or between men and women, suggesting that the hubs represent a fundamental property of structural brain variation that is relatively unaffected by the ageing process. Study III investigated if the subcortical hub regions from Study II would also be present as hubs with a high level of covariation in a study cohort of 138 young adults between 18-35 years. Secondly, we explored if the observed patterns of structural covariation were related to patterns of functional connectivity during resting state. We replicated the finding from Study II that the basal ganglia, the thalamus, and the brainstem were structural hub regions, further strengthening the support that these hubs are not caused by old age. Comparisons of structural covariance patterns and patterns of functional connectivity during rest demonstrated only limited overlap, suggesting that functional connectivity does not cause structural covariance as a general principle. In the final study (Study IV), a dimensionality reduced latent space representation of the cohort from Study III was examined using a convolutional variational autoencoder. The results revealed that only four dimensions – or latent factors – were required to reconstruct most of the structural covariance, including the hubs. Regions with low overall structural covariability typically showed an inconsistent pattern of intercorrelations with other regions in their scores on different factors (e.g. significantly correlated on one factor, but not on other factors). In contrast, hub regions tended to covary across the whole latent space. The factors that correlated positively with the subcortical hubs were also positively related to an increase in functional connectivity during resting state in wide-spread cortical regions. In summary, these results show that subcortical hubs in human brains are robust across the age span and that structural covariance only shows weak relations to patterns of functional connectivity. Further studies in genetically informative samples would be required to investigate the genetic basis of structural covariation in the human brain.