Loading evaluation data…
TREM2 plays multiple functional roles in microglia and variants are associated with increased risks of Alzheimer's disease (AD). Genetic polymorphisms reducing expression of the functionally related protein CD33 are protective. Here we have contrasted cellular pathology in human post-mortem brain with and without AD to test mechanisms associated with the differential genetic risks conferred by R47H and R62H TREM2 variants (TREM2var) with and without heterozygosity for the protective rs3865444 CD33 polymorphism. Epistasis between CD33 and TREM2 was demonstrated by relative normalisation of differences in β-amyloid load in TREM2var carriers of the protective CD33 allele. These functional differences were mirrored by differential microglial transcriptomic responses to β-amyloid. Controlling for CD33 genotype, microglial transcriptional responses to increasing β-amyloid were lower for TREM2var, particularly for R47H compared to CV, and there was a reduction in expression of neuroplasticity pathways in TREM2var. R62H microglial signatures were distinguished from those of R47H by upregulation of genes associated with phagocytosis and from CV by differences in inflammatory gene expression including those involved in NF-kappaB signalling. Differential gene expression with increasing β-amyloid also suggested upregulation of β-amyloid production and binding pathways in excitatory neurons in TREM2var heterozygotes. There was lower enrichment for pathways positively adaptive to pathology and expressed in inhibitory neurons from CV samples for both TREM2var. Exploratory bulk tissue proteomics support these observations with evidence for adaptive plasticity in response to β-amyloid pathology in CV tissue not found for the TREM2var, which showed evidence of increased β-amyloid formation and neuroplasticity changes. Together, these results highlight differences in molecular pathology between CV and TREM2var and between the TREM2var risk variants. They highlight mechanisms of AD risk mediated by secondary effects on astroglial and neuronal functions. Demonstration of strong epistasis between TREM2 and CD33 with AD supports the therapeutic potential of modulators of CD33 inhibition or expression.
Epistasis between CD33 and TREM2 was demonstrated by relative normalisation of differences in β-amyloid load in TREM2var carriers of the protective CD33 allele.
small stratified groups and no reported formal interaction test limit the observational epistasis inference
Controlling for CD33 genotype, microglial transcriptional responses to increasing β-amyloid were lower for TREM2var, particularly for R47H compared to CV.
unequal sample sizes, low R47H power, and unclear CV accounting confound DEG-count comparisons
Differential gene expression with increasing β-amyloid also suggested upregulation of β-amyloid production and binding pathways in excitatory neurons in TREM2var heterozygotes.
pathway associations are appropriately hedged but derive from small, unequal observational genotype groups
They highlight mechanisms of AD risk mediated by secondary effects on astroglial and neuronal functions.
cross-sectional post-mortem expression associations cannot identify causal mediation of disease risk
Derived from the full evaluation — not a separate score.
Strengths
The study integrates IHC, imaging mass cytometry, snRNA-seq, and proteomics across R47H, R62H, and common TREM2 genotypes. This provides a useful human-tissue resource for comparing rare risk variants.
↳ Methods; Results, Figures 1–5
Nuclei are aggregated into donor-specific pseudocells, and donor identity is included as a random effect alongside pathology, genotype, region, sex, APOE4, CD33, and technical covariates.
↳ Methods, Differential gene expression analysis
The Discussion addresses rare-variant scarcity, subgroup imbalance, snRNA-seq sparsity, post-mortem artefacts, contralateral pathology measurement, and absent spatial or molecular plaque characterization.
↳ Discussion, limitations paragraph
Limitations
The Methods call the CV snRNA-seq set 48 blocks but list 18 NDC and 21 AD blocks, while the regional AD counts sum to 30. This uncertainty bears directly on central genotype-specific transcriptomic comparisons.
↳ Methods, Single nucleus isolation for single nuclear RNA sequencing
The paper calls epistasis demonstrated and strong, although the key R47H result comes from small CD33-stratified groups, Figure 1e reports trends, and no formal TREM2-by-CD33 interaction test is reported.
↳ Abstract; Results, Figure 1c and 1e; Discussion, final paragraph
Only 34.8% of equivalent CV-associated effects were estimated detectable at the R47H sample size, yet the Discussion treats the reduced DEG count as genuine biological dampening. Low detectability cannot distinguish dampening from underpowering.
↳ Discussion, paragraph beginning “Functional changes with TREM2 KO”
The principal value lies in combining pathology, cell-type transcriptomics, and proteomics across two rare TREM2 variants in human cortex. The donor-aware pseudocell model is appropriate for repeated nuclei, but the rare genotype-by-CD33 groups are very small and the reported CV block count is internally inconsistent. Figure 1 supports an R47H-associated CD33-stratified amyloid difference, not a definitive interaction across variants, while the transcriptomic comparisons are strongly affected by unequal power. The final mechanistic and therapeutic framing therefore exceeds what cross-sectional post-mortem associations can establish.
Nabu’s assessment, alongside the field’s view.
Are you an author of this paper?
Limited2.9
Confidence mediumThe study meaningfully extends TREM2 research into multimodal human post-mortem tissue, comparing R47H, R62H, and common variants across pathology and cell-type expression. Its advance is constrained because the proposed human CD33–TREM2 interaction remains exploratory rather than demonstrated.
“Epistasis between CD33 and TREM2 was demonstrated”
The pseudocell analysis models donor dependence and includes relevant genotype and technical covariates, but rare-variant strata are small and important age, PMI, and RIN imbalances remain. Cross-sectional associations are interpreted mechanistically, and the central CV snRNA-seq sample accounting is inconsistent.
“Donor identity was included as a random effect”
The progression from histology through cell-type transcriptomics and proteomics is readily followable. However, headline terms such as “demonstrated,” “strong epistasis,” and “genuine biological dampening” communicate greater certainty than the interaction and power evidence warrants.
“The attenuated response in R47H therefore reflects a genuine biological dampening”
The paper engages relevant mouse, in-vitro, human-tissue, and therapeutic literature and gives a substantive limitations discussion. Contextual honesty is reduced because acknowledged scarcity, imbalance, and low detectability are followed by strong biological conclusions that those limitations undercut.
“34.8% of CV-associated genes would be detectable at 0.8 power”
Concerns4 of 4 checks
The reported sample accounting for central transcriptomic analyses is internally inconsistent, and several conclusions are stronger than the stated power and subgroup sizes support. Other cohort descriptors also conflict across the Methods, Table 1, and Figure 1.
Human-tissue ethics and named data repositories are declared, with no text-visible conduct contradiction. Repository contents and supplementary files were not independently verified and are treated as assessability gaps.
Flags: 2 declared / 5 total
75 references in manuscript 74 of 74 checkable references found in an index 1 are books, websites or datasets — counted, but not index-checkable
No retraction notice found in Retraction Watch.
Sources: Retraction Watch ✓
Where this paper’s evidence sits on the path from initial observation to real-world use.
The work is at the human-tissue mechanism-characterization stage rather than intervention testing, biomarker validation, or clinical evaluation. Several functional and prospective validation steps separate the findings from therapeutic use.
“may also support the therapeutic potential”
AI-generated, human-governed. Something look off? Contact us to request a review.