Data-driven CSF biomarker profiling: imaging and clinical outcomes in a cohort at risk of Alzheimer's disease.

Background: Cerebrospinal fluid (CSF) biomarkers of synaptic dysfunction, neuroinflammation, and glial response, complementing Alzheimer's disease (AD) core biomarkers, have improved the pathophysiological characterization of the disease. Here, we tested the hypothesis that the co-expression of multiple CSF biomarkers will help the identification of AD-like phenotypes when biomarker positivity thresholds are not met yet.

Methods: Two hundred and seventy cognitively unimpaired adults with family history (FH) of sporadic AD (mean age = 60.6 ± 4.85 years, 64.8% women) underwent lumbar puncture, magnetic resonance imaging (n = 266) and positron emission tomography imaging (n = 239) protocols, and clinical evaluations. CSF Aβ₄₂, Aβ₄₀, p-tau₁₈₁, p-tau_217, p-tau_231, NfL, neurogranin, sTREM2, YKL40, GFAP, S100, α-Synuclein, SYT1, and SNAP25 were measured. Participants were clustered based on CSF biomarker co-expression with an agglomerative algorithm. The predictive value of the classification against brain and cognitive outcomes was evaluated.

Results: Three clusters (C) were identified. Higher Aβ burden and CSF p-tau was the hallmark of C1. The other two clusters showed lower Aβ burden but higher expression of glial (C2) or synaptic markers (C3). Participants in C1 showed an AD-like clinical phenotype, comprising participants with the overall highest percentage of two parent FH and APOE-ε4 carriers, in addition to comprising more females compared to C2. C3 displayed better vascular health compared to C1. C2 were older and comprised a lower percentage of females compared to C3. C1 showed an AD-like gray matter reduction in medial temporal (notably hippocampus) and frontal regions that were not observed in Aβ_42/40 + compared with Aβ_42/40 - . Furthermore, Aβ_42/40 - participants in C1 showed GM reduction in inferior temporal areas compared with Aβ_42/40 + participants overall. C1 membership also predicted cognitive decline in executive function, but not memory, beyond Aβ + status, overall suggesting a better prognosis in Aβ_42/40 + participants without C1 membership. Additionally, C1 displayed a higher rate of conversion to Aβ + (25%) over time.

Conclusions: Our results suggest that examining multiple CSF biomarkers reflecting diverse pathological pathways may complement and/or outperform AD core biomarkers and thresholding approaches to identify individuals showing a clinical and cognitive AD-like phenotype, including higher conversion to Aβ + , GM reductions and cognitive decline. The clinical utility of this approach warrants further investigation and replication in other cohorts.