Manifold lateralisation and variability in the language connectome at 7T.

Language lateralisation has long been viewed as a hallmark of left hemisphere specialisation. Yet the structural foundations of this asymmetry remain unclear. Using ultra-high-field 7T diffusion MRI from 172 Human Connectome Project participants, we reconstructed seven language-recruited white matter tracts and characterised the architecture of hemispheric asymmetry by combining a new Variability Index, Bayesian inference, and manifold learning. Joint analysis across tracts reveals that individuals vary along a continuous manifold of structural lateralisation rather than forming discrete left, right, or bilateral lateralisation phenotypes. Additionally, lateralisation is robust but tract-specific: the arcuate long segment, inferior longitudinal fasciculus, and frontal aslant tract show leftward asymmetry, while the anterior arcuate segment and uncinate fasciculus show rightward asymmetry. Despite these population-level results, microstructural variability is substantial, with certain tracts exhibiting greater dispersion in one hemisphere than the other. Bayesian evidence indicated that structural lateralisation is unrelated to handedness or language performance. Together, these results redefine hemispheric specialisation as a graded rather than distinctly distributed language connectome, balancing conserved structural scaffolds with individual flexibility.