Single-cell mapping of the endocannabinoid system in breast cancer reveals distributed expression and ECS-NK correlations that track immune composition

Abstract Background Bulk RNA studies link the endocannabinoid system (ECS) to natural killer (NK) cells but cannot show which cell type produces the signal. We asked whether ECS expressions are concentrated in NK cells and whether bulk ECS-NK correlations survive immune adjustment. Methods We attributed 19 ECS genes to ten cell types in a 314,343-cell breast single-cell atlas using pseudobulk centroids, validated NK identification in an independent 91,432-cell atlas, and in three bulk cohorts (4,875 patients, 1,114 deaths) re-estimated ECS–NK correlations as partial Spearman coefficients adjusted for immune, CD8, interferon-gamma and TIL scores. Results None of the 12 ECS genes with sufficient read support was NK-dominant; seven others were too sparsely detected to classify. The highest NK expression share was 20.5% (TRPV2, macrophage-dominated) and the abundance-weighted NK share never exceeded 10.4%. CB2 (CNR2) was plasma-cell dominated, CB1 (CNR1) myofibroblast-dominated and undetected in NK cells, and eicosanoid genes myeloid and stromal. Bulk CNR2–NK correlation fell from ρ = 0.74 to 0.06 after adjustment, and no ECS gene retained |partial ρ| ≥ 0.20 in two cohorts. An 8-gene NK-enriched signature was protective (pooled HR 0.859, 95% CI 0.800-0.924) and survived CD8 and cytolytic adjustment (HR 0.831, p = 0.018), but was collinear with CD8 scores and null after TIL adjustment (HR 0.876, p = 0.111). Conclusions Breast tumor ECS expression is distributed across cell lineages, and bulk ECS-NK correlations are not independent of measured immune composition. Bulk ECS expression is not evidence of NK-specific signaling; cell-type-resolved validation should precede such claims.