Biomarkers.

Background: Growth/differentiation factor-15 (GDF15) has been associated with dementia risk, yet its predictive value across cohorts and sub-population, as well as its relationship with endophenotypes relevant to dementia, remains unknown.

Methods: Using the Atherosclerosis Risk in Communities (ARIC) study as the discovery cohort, we examined the relationship between plasma GDF15 levels (SomaScan) and risk for incident all-cause dementia (ACD) in late-life (N=4,287, 7-year follow-up, M=75±5) and in midlife (N=11,595, 20-year follow-up, M=57±6). Utilizing the UK Biobank (UKB; replication cohort), we related plasma GDF15 (Olink) to incident ACD (N=35,673, 14-year follow-up, M=61±5), vascular dementia (VaD) and Alzheimer's disease dementia (AD). Finally, we examined the cross-sectional association of plasma GDF15 (SomaScan) with brain volume (N=994), white matter lesions (N=911), and plasma biomarker levels (Aβ, GFAP, NfL, and pTau-181) in cognitively normal Baltimore Longitudinal Study of Aging (BLSA) participants. Analyses were stratified by APOEε4 status, cardiometabolic diseases, education, sex, race, and obesity.

Results: Late-life GDF15 abundance was associated with ACD risk in the full ARIC sample (HR=1.61 per log increase; [95% CI: 1.36-1.90]) and in all but one subgroup - obese individuals (Figure 1A-B). GDF15 measured during late-life predicted 7-year ACD risk with an AUC of 0.63 (AUC for GDF15+Age: 0.71). Midlife GDF15 was also associated with ACD risk in the full ARIC sample (HR=1.55 per log increase; [1.32-1.82]) and in each subgroup (Figure 1C-D). GDF15 measured during midlife predicted 20-year ACD risk with an AUC of 0.65 (AUC for GDF15+Age: 0.75). We replicated GDF15's relationship with ACD risk in the UKB sample and found that GDF15 has a much stronger association with VaD (HR=1.76 per log increase; [1.48-2.10]) compared to AD (HR=1.11; [1.01-1.23]; Figure 1E). In the BLSA, higher GDF15 was significantly associated with lower total and regional brain volume, a pattern of accelerated structural brain aging (SPARE-BA), and elevated plasma NfL and pTau181 (Figure 2A-G).

Conclusions: Plasma GDF15 is associated with incident dementia risk across multiple cohorts, particularly vascular dementia, independent of cardiometabolic diseases and obesity. While our analyses indicate that GDF15 may play a role in dementia risk as early as midlife, GDF15, alone, provided only modest accuracy for dementia prediction.