Background
Alzheimer’s disease (AD) is a biologically heterogeneous disorder. Cerebrospinal fluid (CSF) contains thousands of proteins, with concentrations reflecting ongoing (patho)physiological processes. This provides the opportunity to study many biological processes at the same time in patients. Using a dual clustering technique, non‐negative matrix factorization, we studied whether biological subtypes of AD can be detected in CSF proteomics in two independent cohorts, and we will provide an overview how subtypes differed on biological clinical characteristics.
Method
We selected individuals with AD (defined as abnormal ab42 in CSF) and controls (normal cognition and normal AD biomarkers) from ADNI and the EMIF‐AD biomarker multimodality biomarker discovery study (ADNI: 32 preclinical AD, 103 prodromal AD, 62 AD‐dementia, 45 controls; EMIF‐AD: 56 preclinical AD, 90 prodromal AD, 77 AD‐dementia, and 82 controls). In ADNI we studied 149 proteins measured with mass reaction monitoring spectrometry and a multiplex rules based medicine assay, and in EMIF‐AD 556 proteins measured with tandem‐mass tag spectrometry. Proteins were selected when concentrations differed between AD and controls, scaled to include only positive values and then clustered with non‐negative matrix factorisation. Subtypes were compared on demographical, clinical and protein CSF markers not used for clustering.
Result
Both cohorts showed three protein profiles, based on which we allocated individuals to three subtypes: In EMIF‐AD, subtype 1 included 80 individuals (36%), subtype 2 included 71 individuals (32%) and subtype 3included 72 individuals (32%). In ADNI, subtype 1 included 97 individuals (45%), subtype 2 included 69 individuals (35%) and subtype 3 included 39 individuals (20%). Gene Ontology (GO) pathway analysis showed that proteins specifically increased in subtype 1 were enriched for synapse functioning, in subtype 2 for innate immune system activation and in subtype 3 for blood‐brain barrier dysfunction (all pFDR <0.05). Compared to controls, subtypes showed overlapping and distinct regional atrophy patterns: subtype 1 atrophy showed most pronounced atrophy in temporal areas; subtype 2 showed more involvement of parietal areas, and subtype 3 of frontal areas.
Conclusion
Biological heterogeneity amongst individuals with AD can be robustly detected with CSF proteomics, and this provides new opportunities for development of patient tailored treatment approaches.