Abstract
White matter tracts form the structural basis of large-scale brain networks. We applied brain-wide tractography to diffusion images from 30,810 adults (U.K. Biobank) and found significant heritability for 90 node-level and 851 edge-level network connectivity measures. Multivariate genome-wide association analyses identified 325 genetic loci, of which 80% had not been previously associated with brain metrics. Enrichment analyses implicated neurodevelopmental processes including neurogenesis, neural differentiation, neural migration, neural projection guidance, and axon development, as well as prenatal brain expression especially in stem cells, astrocytes, microglia, and neurons. The multivariate association profiles implicated 31 loci in connectivity between core regions of the left-hemisphere language network. Polygenic scores for psychiatric, neurological, and behavioral traits also showed significant multivariate associations with structural connectivity, each implicating distinct sets of brain regions with trait-relevant functional profiles. This large-scale mapping study revealed common genetic contributions to variation in the structural connectome of the human brain.
INTRODUCTION
Cognitive functions and behaviors are supported by dynamic interactions of neural signals within large-scale brain networks (1). Neural signals propagate along white matter connections that link cortical, subcortical, and cerebellar regions to form the structural connectome (2). White matter connections also modulate neural signals and distribute trophic factors between brain regions (3), helping to establish and maintain functional specialization of subnetworks. Various heritable psychiatric and neurological disorders can involve altered white matter structural connectivity, relating, for example, to cognitive deficits, clinical presentation, or recovery (4, 5). It is therefore of great interest to understand which DNA variants, genes, and pathways affect white matter connections in the human brain, as they are likely to influence cognitive and behavioral variability in the population, as well as predisposition to brain disorders.
Diffusion tensor imaging (DTI) enables in vivo noninvasive study of white matter in the brain (6). This technique characterizes the diffusion of water molecules, which occurs preferentially in parallel to nerve fibers due to constraints imposed by axonal membranes and myelin sheaths (7). Metrics commonly derived from DTI, such as fractional anisotropy or mean diffusivity, reflect white matter microstructure and can index its integrity (7, 8). In contrast, tractography involves defining white matter connections at the macroanatomical scale, which permits the measurement of connectivity strengths by counting the streamlines that link each pair of regions. Streamlines are constructed to pass through multiple adjacent voxels in DTI data, when the principal diffusion tensor per voxel aligns well with some of its direct neighbors (9). Tractography therefore produces subject-specific measures of regional interconnectivity that are ideally suited for brain network-level analysis.
Recently, genome-wide association studies (GWAS) have reported that a substantial proportion of interindividual variability in white matter microstructural measures can be explained by common genetic variants, with single-nucleotide polymorphism (SNP)–based heritabilities ranging from 22 to 66% (10, 11). These studies also identified specific genomic loci associated with white matter microstructural measures (10, 11). However, microstructural measures do not necessarily capture topological properties of macroscale brain networks, such as the total amount of structural connectivity between distant pairs of brain regions. In principal, interindividual variability in topological features of the white matter connectome may be influenced by genetic variants that are partly distinct from those that influence white matter microstructure. For example, genetic influences on axon outgrowth and guidance during the development of long-distance connections may be most detectable in terms of connection strengths as measured through tractography, without necessarily affecting the microstructural integrity of those connections. However, to our knowledge, nerve fiber tractography has not previously been used for large-scale genome-wide association analysis of brain structural networks, likely because of heavy computational requirements for running tractography in tens of thousands of individuals.
Here, we aimed to characterize the genetic architecture of white matter structural network connectivity in the human brain, using fiber tractography. DTI data from 30,810 participants of the U.K. Biobank adult population dataset were used to construct the brain-wide structural connectivity network of each individual. In combination with genome-wide genotype data, we then carried out a set of genetic analyses of tractography-derived metrics, in terms of the sum of white matter connectivity linking to each of 90 brain regions as network nodes and 947 connectivity measures as network edges linking specific pairs of regions. The total connectivity of a node (brain region) likely relates to its global role in information transfer within multiple subnetworks, whereas individual connections between specific pairs of regions are more locally restricted measures. We anticipated that genetic influences on node-level and edge-level network measures might therefore be partly distinct, where some genetic effects are more relevant at larger scales whereas others could affect relatively specific circuit components.
Our genetic analyses included SNP-based heritability estimation, multivariate GWAS (mvGWAS), and biological annotation of associated loci. Then, to illustrate how multivariate gene-brain associations arose in the data and how the brain-wide mvGWAS results could be queried in relation to any specific brain network of interest, we used the results to identify genomic loci that are associated with structural connections between core language–related regions of the left hemisphere. Various aspects of language function—especially related to language production—show strong hemispheric lateralization, with roughly 85% of people having left-hemisphere dominance (12).
Last, we assessed how genetic disposition to brain disorders and other behavioral traits manifests in terms of white matter connectivity in the general population. To do so, we mapped multivariate associations of the brain-wide, white matter tractography metrics with polygenic scores for a variety of heritable brain disorders or behavioral traits: schizophrenia, bipolar disorder, autism, attention-deficit hyperactivity disorder, left-handedness, Alzheimer’s disease, amyotrophic lateral sclerosis, and epilepsy. We annotated the resulting brain maps with cognitive functions, using large-scale meta-analyzed functional neuroimaging data, to describe aspects of brain function that may be affected by polygenic dispositions to different forms of neurodivergence in the general population.
RESULTS
White matter connectomes of 30,810 adults
For each of 30,810 adult participants with diffusion magnetic resonance imaging (MRI) and genetic data after quality control, we performed deterministic fiber tractography (9) between each pair of regions defined in the Automated Anatomical Labeling atlas (13) (45 regions per hemisphere comprising cerebral cortical and subcortical structures) (Fig. 1 and Materials and Methods). In the structural connectivity matrix of each individual, each region was considered a node, and each connection between a pair of regions was considered an edge. We excluded edges when more than 20% of individuals had no streamlines connecting a given pair of regions, resulting in 947 network edges. To quantify a given edge in each individual, the streamline count for that edge was divided by the individual-specific gray matter volume of the two regions being connected (as larger regions tend to have more streamlines connecting to them). These volume-adjusted network edge measures were also used to calculate the node-level connectivity of each region, i.e., the sum of all volume-weighted edge measures connecting with a given region, for each participant. The resulting node and edge measures were adjusted for demographic and technical covariates and normalized across individuals (see the “Network construction and analysis” section), before being used for genetic analyses.
