Single-cell transcriptomics reveals prolactin-associated molecular stratification and remission risk in somatotroph PitNETs

Somatotroph pituitary neuroendocrine tumors (PitNETs) show heterogeneous postoperative remission outcome despite standardized surgical management, and the tumor-intrinsic basis of this variability remains poorly understood. We performed single-cell RNA sequencing on 55,648 cells from 12 somatotroph PitNETs to characterize tumor-intrinsic transcriptional states. Pseudobulk and trajectory analyses were used to characterize transcriptional programs. Patient-level stratification was achieved using non-negative matrix factorization. Tumor prolactin (PRL) expression was validated by immunohistochemistry and correlated with serum PRL levels. Clinical relevance was evaluated in an independent cohort of 188 unique patients with acromegaly using logistic regression and stratified analyses. Somatotroph tumors exhibited structured heterogeneity comprising classic, intermediate, and neuronal-associated states organized along a transcriptional continuum while preserving lineage identity. At the patient level, tumors segregated into PRL-high and PRL-low groups independent of cell-state composition. Tumor PRL transcript levels correlated with PRL immunoreactivity and preoperative serum PRL, supporting clinical applicability. In the independent clinical cohort, PRL-high status was associated with postoperative biochemical remission failure. Transcriptomic analyses revealed that PRL-high tumors lacked enrichment of canonical oncogenic pathways but instead showed distinct gene-level transcriptional differences, including receptor-associated candidate genes such as EPHA7. Single-cell transcriptomics identifies a PRL-associated transcriptional stratification in somatotroph tumors that is clinically detectable and associated with postoperative biochemical outcome. Preoperative PRL may serve as a potential adjunctive biomarker for risk stratification alongside anatomical and surgical predictors, although prospective external validation is required.