Novel CSF tau biomarkers can be used for disease staging of sporadic Alzheimer’s

Background Staging Alzheimer’s disease (AD) allows for the identification of key milestones and inflection points in the disease course. However, most current in vivo disease staging methods rely on costly and low‐accessible measures such as PET. Our objective was to generate and evaluate a staging model based on multiple cerebrospinal fluid (CSF) biomarkers. Method We created a staging model using the Subtype and Stage Inference (SuStaIn) algorithm in 426 participants covering the whole AD spectrum (Table 1) by evaluating CSF Ab42/40 and 5 CSF tau markers: the proportion of tau phosphorylated at three threonine residues (pT205/T205, pT217/T217, and pT231/T231), and the concentrations of microtubule‐binding region of tau at residue 243 [MTBR‐tau243] and total‐tau [mid‐domain containing T212‐221]). We investigated associations between CSF stages and cross‐sectional and longitudinal changes in amyloid and tau pathologies, neurodegeneration and cognition using linear regression models. ROC curves were used to identify the optimal stage for predicting positivity in imaging biomarkers and to separate AD vs non‐AD cognitive symptoms. We also tested the relative risk of progression from cognitively unimpaired or MCI to AD‐type dementia using Kaplan‐Meier curves and Cox‐proportional Hazard models. Creation of the model and cross‐sectional analyses were replicated in an independent cohort (n = 222, Table 1). Result SuStaIn revealed that only one subtype was present and that five CSF biomarkers (ordered: Ab42/40, pT217/T217, pT205/T205, MTBR‐tau243, T212‐221) were sufficient to create an accurate disease staging model (Fig.1A‐B). Increasing stages (0‐5) were associated with increased abnormality in other AD‐related biomarkers (Fig.1C). In both cohorts, stage = 2 accurately predicted amyloid‐PET positivity (AUC = 0.89), stage = 4 predicted tau‐PET (AUC = 0.94) and neurodegeneration positivity (AUC = 0.81). In the main cohort, stage = 2 separated non‐AD from AD‐related cognitive impairment (AUC = 0.96). Longitudinal changes for several AD‐biomarkers by stages are presented in (Fig.2A). Higher stages (4‐5) at baseline were associated with higher hazard ratios (HR) of progressing to AD‐type dementia (CU&MCI: HR = 4.6[2.0,10.9], MCI: HR = 3.5[1.5,8.1], Fig.2B‐C). Conclusion Our results suggest that the AD continuum can be accurately staged with a single CSF sample, which may be useful in the clinical setting for patient management but also as a tool in clinical trials fo