Normative modelling of cortical networks identifies subtypes of type 2 diabetes associated with distinct clinical and transcriptomic profiles

Type 2 diabetes (T2D) is characterized by substantial clinical heterogeneity, yet its neurobiological substrates remain unclear. Here, we leverage normative modelling of cortical morphometric networks to quantify individual-level brain deviations using structural MRI data from 3,723 participants (1,941 T2D patients and 1,782 healthy controls) across three independent cohorts. Data-driven clustering reveals two robust T2D biotypes characterized by widespread positive (biotype 1) and negative (biotype 2) brain deviations. Despite comparable overall metabolic burdens, these biotypes feature distinct brain-metabolic coupling patterns: biotype 1 is primarily associated with lipid metabolism, whereas biotype 2 reflects a multifactorial metabolic burden spanning glycaemic control, adiposity, lipid, vascular risks, and disease duration. These divergent brain vulnerability profiles have distinct cognitive consequences; notably, biotype 2 shows poorer performance in complex cognitive tasks, aligning with negatively deviated connectivity gradients along the sensorimotor-to-association axis. Spatial transcriptomic analyses further link biotype 2 deviations to gene expression patterns enriched in insulin signalling, mitochondrial functions, tight junctions, and neurodegenerative disease pathways, alongside specific involvement of inhibitory neurons and oligodendrocyte lineage cells. Our findings provide a neurobiological understanding of T2D heterogeneity and establish a data-driven framework for characterizing personalized brain vulnerability, with implications for advancing precision diabetes medicine.