Thesis defense Sourena Soheili-Nezhad (Donders series 585)

28 November 2023

Promotor: prof. dr. C. Beckmann
Co-promotors: dr. E. Sprooten, dr. G. Poelmans

Multivariate neuroimaging genetics: from brain networks to genetic factors

All human brains share a common anatomy and functional specialisation, but there are many fine-grained variations in the structure and function of our brains. Many of these differences are moderately heritable and to some extent driven by our DNA. Neuroimaging genetics aims to understand the way that DNA shapes the brain, and by extension, to understand the three-way associations between genetic factors, brain, and brain–related disorders. Both genomic and neuroimaging data encompass multiple and sometimes up to millions of variables. This burden of dimensionality has been traditionally eased by conducting hypothesis-driven studies, for example by focusing on specific brain regions, using anatomical atlases or limiting studies to risk gene(s) of a particular disease condition.

Genome-wide association studies (GWAS) test associations of (brain) traits with variants across the entire DNA in a hypothesis-free manner. GWAS is a univariate method since every SNP’s effect on every brain trait is tested independently. However, the influence of the genome on brain may better be captured by models that consider the concerted ways through which the genetic factors influence multiple phenotypes. Due to the underlying biology, these associations are usually not univariate but higher order structured. For example, a genetic factor that modifies cortical thickness in a single region may also affect other brain regions to some extent and at the same time modulate brain connectivity. In addition, this gene probably doesn’t act alone, but in concert with multiple genes in a complex biological pathway.

The global aim of this thesis was to identify multivariate associations between genetic, neuroimaging and clinical features. By applying independent component analysis (ICA), an exploratory source separation method for decomposing a complex signal into its constituents, we report changes in the functional connectivity of the occipital and the default-mode networks in patients with migraine. Using ICA, we also identified structural deficits of the frontostriatal network in attention-deficit/hyperactivity disorder (ADHD) and validated this finding in an independent sample of subjects. Extending our data driven investigations to Alzheimer’s disease (AD), we discovered a new genetic polymorphism in a synaptic gene, SHARPIN, in association with the volume of a structural brain component mapping to the limbic system. By focusing on the biological function of SHARPIN, a new model for the molecular pathogenesis of brain aging and dementia was developed and tested, which implicates synaptic adhesion molecules, DNA damage and somatic mutations in long genes as the etiology of neurodegeneration. In the final chapter, we developed and tested a new method for data-driven neuroimaging research, genomic ICA, in the large community-based UK Biobank database. This method separates the influence of genome-wide effects on thousands of brain MRI traits into independent genomic sources. The results indicate that univariate imaging and genetic associations can be approximated by a number of multivariate genomic sources. A number of these latent components are highly polygenic and sensitive to molecular pathways and cell-specific gene expression profiles. Overall, this work shows that, as we transfer to biobank-scale data and larger sample sizes, transitioning from univariate to multivariate imaging genetic data fusion methods can aid the discovery of new gene-by-phenotype associations and help biological interpretations.