rs3865444 — CD33 CD33 microglial Alzheimer's variant

CD33 — The Microglial Brake on Amyloid Clearance

The brain's immune cells — microglia¹¹ microglia
Resident macrophage-like cells of the central nervous system; account for roughly 10–15% of all brain cells and are the primary defence against pathogens and cellular debris — are responsible for patrolling the brain and clearing misfolded protein aggregates, including amyloid-beta (Aβ) plaques. How well they do this job depends in part on the proteins on their surface. CD33 (also known as Siglec-3) is one such surface receptor: it acts as an inhibitory brake on phagocytic activity. When CD33 levels are high, microglia engulf less amyloid-beta. When CD33 is reduced or absent, clearance improves dramatically. rs3865444 sits 372 base pairs upstream of CD33's first exon and, through tight linkage with a functional exon 2 splice variant, is one of the most replicated genetic controls of this brake.

The Mechanism

rs3865444 is in near-perfect linkage disequilibrium with rs12459419, a variant at the fourth base of CD33 exon 2. The rs12459419 T allele (coinherited with rs3865444 A) promotes skipping of exon 2 during mRNA splicing²² The rs12459419 T allele (coinherited with rs3865444 A) promotes skipping of exon 2 during mRNA splicing
Minigene experiments in BV2 microglial cells showed the T minigene produced 10.3% exon-2-skipped transcripts versus only 3.4% for the C minigene (p = 0.034). Exon 2 encodes the immunoglobulin variable domain — the sialic acid-binding site that enables CD33 to suppress phagocytosis. A CD33 protein lacking exon 2 (D2-CD33) is functionally inert as an inhibitory receptor.

Each copy of the protective A allele at rs3865444 increases the proportion of D2-CD33 by roughly 11% and reduces total full-length CD33 protein by approximately 23% per allele, with AA homozygotes showing ~46% lower full-length CD33 than CC homozygotes³³ AA homozygotes showing ~46% lower full-length CD33 than CC homozygotes
Adjusted R² = 0.76 for the expression model; rs3865444 genotype p = 0.012. The functional consequence is unambiguous: human microglia lacking CD33 phagocytize significantly greater amounts of Aβ42 than CD33-expressing microglia⁴⁴ human microglia lacking CD33 phagocytize significantly greater amounts of Aβ42 than CD33-expressing microglia
Microglia were transfected with CD33-targeting siRNA; knockdown cells showed ~50% increase in Aβ42 uptake versus controls. In transgenic APP/PS1 mice, complete CD33 knockout reduced cortical plaque burden by 37.2% (p < 0.01) and hippocampal plaque by 33.5% (p < 0.05).

The Evidence

Two parallel GWAS consortia published genome-wide significant findings for rs3865444 in 2011. Hollingworth et al.⁵⁵ Hollingworth et al.
Combined four GWASs plus replication across stages 2–3; meta P = 1.6×10⁻⁹ for CD33 rs3865444 and Naj et al.⁶⁶ Naj et al.
Three-stage design with Alzheimer Disease Genetics Consortium data; P = 1.6×10⁻⁹ across all stages combined both independently identified CD33 as a late-onset AD susceptibility locus — a striking concordance. These remain among the most robustly replicated non-APOE loci for AD.

A meta-analysis of 18 case-control studies⁷⁷ meta-analysis of 18 case-control studies
50,030 cases and 77,405 controls; largest synthesis of rs3865444 data to date confirmed the protective A allele confers OR 0.93 (95% CI 0.90–0.97) overall. Genotype-specific analyses from independent studies show a dose-dependent effect: each copy of the A allele confers approximately 13–18% relative risk reduction (CA OR ≈ 0.87, AA OR ≈ 0.82 vs CC). The effect is most robust in European and North American populations; evidence in East Asian cohorts is weaker and heterogeneous, possibly reflecting lower A allele frequency (~19% in Korean populations versus ~34% in Europeans) or population-specific LD patterns.

Neuropathological confirmation⁸⁸ Neuropathological confirmation
Arizona Study of Aging; 193 brain-bank donors stratified by rs3865444 genotype shows AA carriers have significantly lower CD33 protein in cortex, fewer CD33-positive microglia, and lower insoluble Aβ42 levels than CC carriers, providing direct biological validation in human brain tissue.

Practical Actions

CD33 rs3865444 is a modest but mechanistically compelling AD risk modifier. Its effect is substantially smaller than APOE ε4 (which multiplies AD risk 3-fold per allele) but among the strongest non-APOE GWAS signals. CC homozygotes carry the highest CD33 expression and weakest microglial amyloid clearance from this locus. Protective strategies that support microglial function and reduce amyloid burden — particularly those with independent clinical evidence — are especially relevant for CC carriers.

Emerging therapeutic interest focuses directly on this pathway: CD33-blocking antibodies and CD33 gene-editing approaches are in preclinical development for AD, following the same logic that reducing CD33 activity improves microglial phagocytosis.

Interactions

The strongest documented interaction is with APOE ε4 (rs429358 / rs7412). APOE ε4 impairs microglial lipid metabolism and amyloid clearance through a partially distinct mechanism from CD33 inhibition; individuals carrying both CC at rs3865444 and one or more ε4 alleles face compounding microglial dysfunction. The CD33 effect appears largely additive on top of APOE ε4 risk rather than synergistic, but CC + ε4 carriers represent a high-priority group for AD risk management.

A second layer of interaction involves TREM2 (rs75932628) — another microglial receptor whose R47H variant impairs phagocytosis through a different surface receptor pathway. Individuals homozygous CC at rs3865444 and heterozygous R47H at TREM2 would have parallel deficits in two independent microglial clearance pathways.