Biomarkers.

Background: Differences in task-fMRI activation have recently been found to be related to neuropathological hallmarks of AD. However, the evolution of fMRI-based activation throughout AD disease progression and its relationship with other biomarkers remains elusive. Applying a disease progression model (DPM) to a multicentric cohort with up to four annual task-fMRI visits, we hope to provide a deeper insight into these relationships.

Method: We estimated AD disease stages using a multivariate Gaussian Process (GP) DPM including CSF-Aβ42/40 ratio, CSF-p-tau, hippocampal and entorhinal volume, ADAS13-Cog sum and PACC5 scores. Disease stages from 493 participants with longitudinal task-fMRI measurements from DELCODE (165 healthy controls (CN), 214 participants with SCD, 82 with MCI, 32 with suspected AD) were obtained. We derived subsequent memory and novelty contrasts from a visual memory encoding task using general linear modeling (GLM). Contrasts from all available follow-ups were then submitted to voxel-based group-level GLM analyses. Activations from resulting disease-stage-related clusters were (1) used to estimate cluster-level trajectory curves over disease stages using smoothing splines and (2) submitted to linear-mixed effects models to test longitudinal changes over follow-ups.

Result: Our DPM-derived disease stages were associated with clinical groups, fMRI performance and white matter lesions (Figure 1C-F). Generally, in both contrasts, activation increases were observed in task-negative clusters while activation decreases were observed in task-positive clusters (Figure 2C-F). We did not find indications for inverted u-shaped associations between disease stage and activation in whole brain voxel-wise cross-sectional analyses. However, smoothing splines revealed non-linear monotonically increasing biomarker abnormality for task-negative areas, showing earliest changes towards the beginning of disease progression. After a plateau, fMRI activation increases in abnormality conjointly with volume changes. For task-positive areas, we observed linear relationships with disease stages (Figure 3). Activation changes over follow-ups were not associated with disease stages.

Conclusion: Biomarker abnormality timing in our DPM reflected hypothetical AD progression. Changes in task-fMRI activation and deactivation were both associated with progression towards AD. Smoothing spline fits indicated abnormality changes in task-fMRI activation to begin in the earliest phases of the disease. Findings can be discussed as differential pathophysiological processes such as complex reorganization and neural noise.