Immune Cell-Specific Genetic Signals Reshape the Alzheimer’s Question
A single-cell Mendelian randomization analysis has tied Alzheimer’s disease risk to genetically predicted gene expression across peripheral immune cells, sharpening the view that the disorder is not only a brain disease but also a disease with measurable immune-genetic circuitry outside the central nervous system. Using cis-eQTLs from 14 peripheral blood immune cell types in the OneK1K single-cell dataset and Alzheimer’s GWAS summary statistics, investigators identified 187 significant gene-cell type associations involving 161 candidate causal genes.
The scale of the signal matters because the associations were not smeared evenly across immune biology. One hundred forty-one genes showed significance in only a single immune cell type, a pattern that points to highly localized immune regulation rather than a broad, generic inflammatory effect. In practical terms, the findings support a model in which specific immune subsets carry distinct genetic programs that may influence Alzheimer’s susceptibility through separate molecular routes.
TNF in CD4+SOX4+ T Cells Emerges as the Strongest Shared Signal
Among the colocalized signals, TNF expression in CD4+SOX4+ T cells stood out. Bayesian colocalization showed a shared genetic signal between TNF expression and Alzheimer’s risk, with PP.H4 = 0.833, a value that supports a plausible common causal variant underlying both traits.
That result is biologically credible. TNF sits at the center of immune activation, cytokine signaling, and downstream inflammatory amplification. In T-cell biology, altered TNF expression can reshape activation thresholds, cell communication, and the broader balance between immune surveillance and pathological inflammation. In Alzheimer’s disease, where immune dysfunction and chronic inflammation have long been implicated in neurodegenerative progression, a peripheral T-cell TNF signal offers a plausible bridge between systemic immunity and disease risk.
This does not prove causality in a clinical sense. It does, however, provide a tightly defined genetic clue that can be pursued in functional work and in independent human cohorts.
Cell-Type Specificity Was the Dominant Theme Across the Dataset
The strongest message from the analysis is not a single gene but the degree of compartmentalization. The fact that most significant genes appeared in only one immune cell type suggests that peripheral immunity may influence Alzheimer’s risk through cell-specific mechanisms rather than through a single shared immune axis.
Functional enrichment analysis reinforced this interpretation. Significant genes across immune cell types clustered in biological processes such as MHC class II antigen presentation, phagocytosis, membrane invagination, and positive regulation of T cell activation. These pathways map directly onto core immune functions: antigen presentation governs how immune cells recognize molecular threats, phagocytosis controls clearance of debris and cellular material, and T-cell activation governs adaptive immune tone. Each of those processes could plausibly shape the inflammatory environment that influences neurodegeneration over time.
A biological signal this dispersed is rarely accidental. It reflects layered immune regulation, where different cell populations contribute different pieces of the same disease mechanism.
Replication Added Support, but Not Uniform Confirmation
To test robustness, the analysis moved into the 1M-scBloodNL replication dataset. Several candidate genes showed the same direction of effect and nominal significance, suggesting that at least part of the discovery signal persisted in an independent single-cell eQTL resource. The replication results are useful because they reduce the chance that the main findings are a byproduct of one dataset’s structure or noise profile.
Still, the validation was partial rather than definitive. Nominal significance is not the same as strong independent confirmation, and the replication resource did not erase the need for further testing in additional populations. The work does, however, strengthen the claim that the discovery set captured biologically meaningful, cell-specific signals rather than statistical artifacts alone.
PheWAS Kept TNF in the Immune-Hematology Corridor
Phenome-wide association study results for the TNF instrumental variable were concentrated mainly in immune- and hematology-related traits. No genome-wide significant neurological phenotypes emerged. That pattern is informative because it suggests the TNF signal is not simply tagging a broad neurological liability profile in the available datasets. Instead, its genetic footprint appears to remain anchored in immune biology.
For bedside translation, that distinction is important. It keeps the focus on immune mechanisms rather than implying that peripheral TNF expression is already a ready-made neurological biomarker. It also argues against overreading the result as a direct clinical predictor of dementia risk. The signal belongs to a mechanistic layer, not to a validated diagnostic panel.
What This Means for Alzheimer’s Biology Right Now
This study adds weight to the idea that peripheral immune cells are not passive bystanders in Alzheimer’s disease genetics. The integration of single-cell eQTL data with Mendelian randomization and colocalization makes it possible to isolate candidate genes in specific immune compartments, rather than treating blood-based immunity as a uniform compartment. That approach is especially useful in a disease as heterogeneous as Alzheimer’s, where multiple biological systems probably converge on the same clinical endpoint.
The most consequential implication is methodological as much as biological. Single-cell transcriptome-wide Mendelian randomization can narrow the search from broad association to cell-specific hypothesis generation, which is exactly where mechanistic neuroscience often stalls. Here, it produced a shortlist that includes TNF in CD4+SOX4+ T cells, a finding that is mechanistically coherent and statistically supported by colocalization.
Why the Findings Still Need Functional Proof
The analysis relied on publicly available summary statistics, not new patient-level biological sampling. It also used genetic proxies rather than direct measurements of protein activity, cell behavior, or disease progression. Those features make the work powerful for causal inference but limited for clinical deployment.
Independent population validation remains necessary, especially across ancestries and data platforms that may differ in immune-cell annotation, expression quantification, and genetic architecture. Functional experiments are also needed to establish whether TNF expression in CD4+SOX4+ T cells actively contributes to disease biology or simply marks a correlated immune state.
For now, the study leaves Alzheimer’s genetics with a more precise immune map and one especially durable clue: peripheral immune-cell expression, not just bulk inflammation, may carry disease-relevant information that is visible only when the transcriptome is resolved at the single-cell level.