Multi‐cohort cerebrospinal fluid proteomics identifies robust molecular signatures for asymptomatic and symptomatic Alzheimer’s disease

Abstract Background Changes in Amyloid‐β (A) and hyperphosphorylated Tau (T) in the brain and cerebrospinal fluid (CSF) precedes AD symptoms, making the CSF proteome a potential avenue to understand disease pathophysiology and facilitate reliable diagnostics and therapies. Method We used the Somascan assay for measuring the protein levels of 7,029 analytes in CSF of 2,286 participants from four different cohorts. We employed a three‐stage analytical approach (discovery, replication, and meta‐analysis). Firstly, discovery was performed in 1,170 samples from the Knight‐ADRC and FACE cohorts (stage1) using the ATN framework (A‐T‐ =680 and A+T+ =490). Secondly, proteins that passed FDR correction in the stage1 were further tested in 593 individuals (A‐T‐ =235 and A+T+ =358) from the ADNI and Barcelona‐1 replication cohorts (stage2). Lastly, the proteins that passed Bonferroni corrections on the meta‐analysis of stage 1 and 2 were further utilized for creating AD prediction models, conducting pathway enrichment, disease progression, and protein‐protein interaction (PPI) network analyses to gain mechanistic insights into AD pathophysiology. Result We identified 2,173 proteins dysregulated in AD, that were further validated in a third totally independent cohort (Stanford‐ADRC). Using machine learning, we identified a distinctive signature of 11 proteins with robust and high predictive capability for AD (AUC = 0.97–0.99) in three independent cohorts (stage1, stage2, and Stanford‐ADRC). The identified proteomic signature was unique to AD and showed no predictive power for other dementias such as frontotemporal dementia (AUC=0.61), dementia Lewy body (AUC=0.73), or Parkinson’s disease (AUC=0.57). The developed proteomic signature displayed significant association with disease progression (β=0.35, p=2.1×10 ‐04 ) and individual’s probability of not developing AD (p=2.2×10 ‐58 ). The associated proteins cluster in four different protein pseudo‐trajectory groups spanning the AD continuum and were enriched in specific pathways including neuronal death, apoptosis and tau phosphorylation (early‐stages), microglia dysregulation and endolysosomal dysfunction (mid‐stages), and late microglia‐neuron crosstalk (late‐stages). Conclusion Our findings show the promising potential of AD CSF proteomics in developing reliable and robust AD prediction model. We identified proteins and biological pathways that are compromised in AD, thereby, increasing our understanding of AD biology and facilitating the development of effective therapies targeted at earliest molecular triggers of AD.