The human cortex is folded. These folds develop during the fetal stage and are unique to each individual, forming a true "brain fingerprint" that partially depends on the ability of cells to proliferate, migrate, and differentiate during development—and thus on genetic factors. Studying these patterns could reveal hidden events occurring during brain development, as the resulting shapes remain relatively stable throughout life. However, the high interindividual variability of these patterns makes them very difficult to associate with specific pathologies.
Le plissement cortical sous influence génétique
A team from the GAIA Laboratory (UMR BAOBAB/NeuroSpin), in collaboration with LaPsyDé (Université Paris Cité) and the University of Cambridge, has developed a method to identify patterns influenced by genetic factors. They focused on a key brain region, the anterior cingulate cortex, where the paracingulate sulcus is of particular interest in psychiatry. Their method relies on a comprehensive regional representation of fold variability, estimated using a self-supervised deep learning algorithm applied to MRI scans of 41,000 individuals from the UK Biobank database, the world's largest population-based imaging-genetics cohort.
This representation can be used to linearly distinguish folding patterns and generalize manual labeling across large datasets. In this case, manual labeling of folding patterns in the anterior cingulate cortex (considering the presence or absence of the paracingulate sulcus as a phenotype) did not reveal clear genetic associations.
Vers des biomarqueurs du neurodéveloppement
However, the researchers demonstrated that it is possible to discover new genetic loci associated with cortical folding patterns directly from this deep learning-derived representation, without manual labeling. In a discovery cohort of 36,000 subjects, they identified loci linked to folding patterns in the anterior cingulate cortex: four in the right hemisphere and ten in the left. Although only one locus was replicated in a smaller cohort (5,000 subjects), several of these loci are already known to be associated with brain anatomy or psychiatric conditions.
These preliminary findings confirm that cortical foldings are indeed biomarkers of neurodevelopment, but they also suggest that human labeling captures genetic information less effectively than deep learning representations.
European fundings This work received European funding through the
R-Link and
EBRAIN2 projects.
FRANCE 2030The research was supported by a structuring project coordinated by
Jean-François Mangin, selected under the high-risk research program
"Audace!", part of the
FRANCE 2030 initiative.
Contact Frédéric-Joliot Institute for Life sciences:
This text was translated with the assistance of Mistral AI.