How many gamma delta T cells are there? Calling strategy, quality control and receptor-locus specificity in 3-prime single-nucleus RNA sequencing of human small intestine.

Public single-nucleus transcriptomes are increasingly reused to count gamma delta T cells in the human small intestine, but how far that count depends on analytical choices has not been quantified. Using a public 3-prime single-nucleus dataset of 33,600 nuclei from six small-intestinal biopsies, we benchmarked three calling strategies across three quality-control levels, tested six module-score cutoff rules, and applied two specificity checks that need no additional assay. The recovered fraction spanned 3.38% to 12.49% on the deposited matrix and 5.46% to 21.62% under strict filtering, a 3.70-fold to 4.08-fold range depending on the filter. Tightening the module-score cutoff corrected the count but not the membership: at a cutoff matched to the gated call, agreement with it reached only 0.408 by Jaccard index. Within one quality-control stratum the call rate rose from 0.51% to 12.17% across deciles of genes per nucleus, so detection sensitivity rather than abundance dominates. Constant-region genes were near-absent from non-T epithelial nuclei (TRGC2 0.25% versus 36.19% of CD3-positive nuclei), indicating that broad TRGC2 detection reflects expression that is not lineage restricted rather than ambient RNA. However, 96.3% of gated calls also carried TRAC or TRBC counts, which is hard to reconcile with excision of the TRD locus during TRA rearrangement and points to unquantified alpha-beta admixture. Variable-region genes were near-undetectable, so subtype assignment was infeasible. We conclude that absolute gamma delta counts from 3-prime nuclear data are not interpretable on their own, and we supply a reporting checklist and open code.