PitNET tissue deconvolution: rastreo de residuos de tejido normal y dinámica inmunitaria

Bulk RNA sequencing (RNA-seq) has substantially advanced the understanding of pituitary neuroendocrine tumors (PitNETs). However, its limited ability to resolve cellular heterogeneity - particularly in samples containing residual non-tumor pituitary cells - remains a significant challenge.

We developed and validated a tissue deconvolution framework using a reference dataset derived from single-nucleus RNA sequencing (snRNA-seq) of normal pituitary tissue, aimed at estimating cellular composition in PitNETs from bulk RNA-seq data and characterizing the tumor microenvironment (TME).

Marker-based (CIBERSORT, MuSiC) and single-cell-based (CIBERSORTx, MuSiC) deconvolution approaches were benchmarked across simulated, pseudobulk, and bulk RNA-seq datasets to identify the most reliable tools.

CIBERSORTx demonstrated the highest sensitivity (r > 0.85) for detecting pituitary cell types, although accuracy decreased for TME components. Application to ten GH-secreting PitNETs with known histological contamination and to public datasets consistently revealed residual normal tissue across hormone-secreting subtypes, excluding silent tumors. Contaminated samples - averaging 43% ± 19% with CIBERSORTx and 37% ± 22% with CIBERSORT - displayed distinct transcriptomic profiles compared to uncontaminated, lineage-matched tumors, based on clustering analyses.

This study establishes snRNA-seq-based deconvolution as a robust strategy for reconstructing cellular composition in PitNETs, mitigating the impact of histological contamination and improving the reliability of downstream transcriptomic analyses.