Exploratory Single-Cell Analysis of NK-Annotated Cells Across Marrow Compartments in Prostate Cancer Bone Metastasis

Prostate cancer preferentially colonizes bone, where marrow compartments differ in tissue structure and immune composition. We reanalyzed a 920,521-cell integrated prostate cancer atlas containing 38,842 NK-annotated cells from 126 sample identifiers, focusing on one bone-marrow cohort of 16 subjects and 32 samples. Distal marrow from patients with bone metastasis was compared with nonmalignant marrow from unrelated hip-replacement controls, while complete within-patient pairs were used for local comparisons. Sample- or subject-level composition, pseudobulk expression, pathway, and microenvironment analyses treated biological units, not cells, as replicates. The inherited CytokNKs_GZMH label was more frequent in distal marrow than in controls (25.4% versus 2.0%; centered log-ratio effect 2.82; false-discovery rate 0.0367). Paired tumor-versus-distal comparisons showed higher relative abundance of ImmSuppNKs_CREM and lower proportions of several other inherited labels at the 10% false-discovery threshold. A pooled comparison identified 62 differential genes within RegNKs_IL7R and generated a 24-gene exploratory program; distal-only models had no genes meeting the joint cutoff. Frequent T-cell- and myeloid-marker detection precluded assignment of this population as a validated NK subset. Program separation within the discovery cohort was subject to gene-selection bias and did not establish predictive performance. Full-atlas analyses described compartment-associated composition differences and candidate ligand-receptor expression pairs. Donor characteristics, treatment, sampling site, specimen processing, and annotation uncertainty remain unresolved sources of confounding. These findings are hypothesis-generating associations in recovered NK-annotated cells, not evidence of a metastasis-specific field effect, a regulatory NK population, or clinical utility.