Cross-Cohort Characterization of an LHB-Associated Transcriptomic State in Prostate Cancer

We sought to characterize an LHB-associated tumor transcriptomic state across public prostate cancer datasets. Tumor LHB expression and the correlation-derived signature were treated as molecular measurements only; serum luteinizing hormone and testosterone were not available or analyzed.

In 501 TCGA-PRAD tumors, the 25 genes most positively and 25 most negatively correlated with LHB defined a direction-weighted signature. The frozen score was projected without refitting into GSE54460, GSE116918, and two GSE21034 platform views. Cohort-specific grade, biochemical recurrence, and compact expression programs were evaluated. Program genes did not overlap the signature, and LHB was excluded from steroidogenesis. In GSE176031, upper and lower signature-score quartiles were compared within each of 53 samples.

The score was higher in high-grade TCGA-PRAD tumors (mean difference, 0.327; P = 7.12 × 10^-9) and had positive but heterogeneous grade differences in the external datasets. Across all five bulk views, the score correlated negatively with androgen receptor signaling (r = -0.365 to -0.627) and positively with cell cycle (r = 0.283-0.692) and epithelial-mesenchymal transition (r = 0.217-0.734). Direct LHB transcripts were detected in 30 of 53,765 cells. Sample-level single-cell contrasts showed lower androgen receptor signaling and higher cell-cycle and epithelial-mesenchymal-transition scores in signature-high epithelial states; the LHB-excluded steroidogenesis contrast was near zero.

The LHB-correlation-derived signature identifies an AR-low, proliferative, mesenchymal-associated expression state across datasets. It does not measure circulating hormones, establish LHB-dependent biology, or support clinical use. Paired molecular and functional studies are required.