Plasma proteomics reveals continuous molecular heterogeneity rather than discrete subtypes in Alzheimer’s disease
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
INTRODUCTION
Alzheimer’s disease is clinically and biologically heterogeneous. We asked whether plasma proteomics separates patients into discrete molecular subtypes or instead reflects continuous biological variation.
METHODS
We studied 5,895 Global Neurodegeneration Proteomics Consortium (GNPC) participants with Alzheimer’s disease or mild cognitive impairment using protein coexpression networks, clustering, and continuous molecular-axis analysis. External analyses used Stanford Alzheimer’s Disease Research Center (ADRC) biomarker/imaging data and UK Biobank proteomics.
RESULTS
Four continuous axes captured 81.5% of module-level proteomic variation. Although a two-cluster solution was reproducible, separation was weak and added little clinical information beyond the continuous axes. Stanford ADRC analyses showed selected fluid biomarker associations, but imaging and PET results did not provide consistent support. In UK Biobank, projected axes were more strongly related to APOE genotype and systemic hematologic, renal, lipid, inflammatory, and hepatic traits than to clear dementia-risk replication.
DISCUSSION
Plasma proteomics did not support robust Alzheimer’s disease subtypes. Continuous molecular coordinates better describe plasma proteomic heterogeneity and may guide future biological stratification.
Research in Context
Systematic Review
Prior work has established that Alzheimer’s disease is biologically heterogeneous. Neuropathological and imaging studies have described hippocampal-sparing, limbic-predominant, typical, cortical, or subcortical patterns, while transcriptomic and CSF proteomic studies have reported molecular subtypes with distinct cellular, immune, vascular, synaptic, or metabolic signatures. Large-scale plasma proteomic studies and biobank resources show that circulating proteins capture dementia-related and systemic biology, but plasma-based subtype claims require caution because circulating proteins may reflect both central nervous system-linked and peripheral processes. We therefore considered a robust plasma proteomic subtype claim to require reproducible clustering, clear separation, incremental clinical information beyond continuous dimensions, and directionally coherent support from external biomarkers, imaging, or independent cohorts.
Interpretation
In this study, a brain-region- and cell-system-inspired plasma proteomic subtyping hypothesis was not supported as a set of discrete molecular classes. In GNPC, the most reproducible clustering solution showed weak separation and provided little additional clinical information after accounting for four continuous module-derived axes. External analyses further supported a cautious interpretation: Stanford biomarker analyses showed partial but partly discordant associations, whereas UK Biobank Olink projections most strongly contextualized the axes through APOE genotype and systemic hematologic, renal, lipid, inflammatory, and hepatic traits rather than through unambiguous dementia-risk replication. Together, these results suggest that plasma proteomic heterogeneity in Alzheimer’s disease is better represented by continuous molecular coordinates than by hard subtype labels.
Future Directions
The present findings argue for a shift from categorical plasma-proteomic subtyping toward continuous molecular stratification in Alzheimer’s disease. Replication in independent Alzheimer’s disease and mild cognitive impairment cohorts, ideally with longitudinal follow-up and harmonized biomarker and imaging measures, will be necessary to determine whether these axes are stable and clinically useful. Integrating plasma proteomics with phosphorylated tau, neurodegeneration markers, amyloid and tau imaging, and structural MRI may help distinguish Alzheimer’s disease–linked variation from systemic physiology. Finally, brain-region and cell-type annotations should be interpreted as biological context rather than evidence of tissue origin unless supported by paired tissue, imaging, or mechanistic data.
