ObjectiveTo investigate the independent associations of social isolation and loneliness with incident dementia and to explore the potential neurobiological mechanisms.MethodsWe utilized the UK Biobank cohort to establish Cox proportional hazard models with social isolation and loneliness as separate exposures. Demographic (sex, age and ethnicity), socioeconomic (education level, household income and Townsend deprivation index), biological (BMI, APOE genotype, diabetes, cancer, cardiovascular disease and other disabilities), cognitive (speed of processing and visual memory), behavioral (current smoker, alcohol intake and physical activity), and psychological (social isolation or loneliness, depressive symptoms and neuroticism) factors measured at baseline were adjusted. Then, voxel-wise brain-wide association analyses were used to identify gray matter volumes (GMV) associated with social isolation and with loneliness. Partial least squares regression was performed to test the spatial correlation of GMV differences and gene expression using the Allen Human Brain Atlas.ResultsWe included 462,619 participants (mean age at baseline 57.0 years [SD 8.1]). With a mean follow-up of 11.7 years (SD 1.7), 4,998 developed all-cause dementia. Social isolation was associated with a 1.26-fold increased risk of dementia (95% CI, 1.15-1.37) independently of various risk factors including loneliness and depression (i.e., full adjustment). However, the fully adjusted hazard ratio for dementia related to loneliness was 1.04 (95% CI, 0.94-1.16); and 75% of this relationship was attributable to depressive symptoms. Structural MRI data were obtained from 32,263 participants (mean age 63.5 years [SD 7.5]). Socially isolated individuals had lower GMVs in temporal, frontal and other (e.g., hippocampal) regions. Mediation analysis showed that the identified GMVs partly mediated the association between social isolation at baseline and cognitive function at follow-up. Social isolation-related lower GMVs were related to under-expression of genes that are down-regulated in Alzheimer’s disease and to genes that are involved in mitochondrial dysfunction and oxidative phosphorylation.ConclusionSocial isolation is a risk factor for dementia that is independent of loneliness and many other covariates. Social isolation-related brain structural differences coupled with different molecular functions also support the associations of social isolation with cognition and dementia. Social isolation may thus be an early indicator of an increased risk of dementia.