Blood proteomics to identify AD CSF molecular subtypes

Background We identified three distinct molecular subtypes in Alzheimer’s disease (AD) using cerebrospinal fluid (CSF) proteomics: one subtype was characterised by neuronal hyperplasticity with increased amyloid metabolism, the second subtype was characterised by innate immune activation and the third subtype by a disrupted blood‐brain barrier. Each subtype will likely need specific treatment, but for patient selection CSF proteomics might be impractical. Here we investigated if AD molecular subtypes can also be detected in blood. Method We selected from EMIF‐AD MBD 149 individuals with AD across the clinical spectrum who had plasma proteomics as well as a CSF subtype definition available. We compared CSF subtypes on plasma levels using general linear models. We selected a subset of proteins that best predicted AD subtype by trying a random forest classifier on 70% of the patient sample, and computed prediction accuracy statistics on patients that were left out. Result Plasma levels of 407 (13%) proteins differed between AD subtypes (figure). The plasma profiles specific to AD subtype showed altered levels for corresponding molecular processes: The CSF hyperplasticity subtype (n = 53) showed low plasma levels of 76 proteins associated with neuron development and axon guidance. The CSF innate immune activation subtype (n = 48) showed low plasma levels of 142 proteins related to regulation of immune response. The CSF blood‐brain barrier subtype (n = 48) showed the low plasma levels of 189 proteins associated with blood vessel morphogenesis. The random forest could classify subtypes with an overall accuracy of 77% (range 70%‐85%), and the top 5 contributing proteins were FGF10, CTCF, TCN2, PLXNA1, and EMILIN3. Conclusion It is possible to detect distinct AD CSF molecular subtypes in blood. The next step is to validate this in independent cohorts, and to test if subtypes are related to different treatment responses.