ABSTRACT Background Prognostic biomarker panels identified through bulk sequencing approaches have shown utility in localized prostate cancer but are limited by underlying molecular heterogeneity. Spatial transcriptomics offers a complementary approach to investigate spatial gene expression patterns and the tissue‐ and cell‐type‐associated localization of their constituent biomarker genes. Methods Using publicly available data, we analyzed biomarker genes from four prognostic panels (Oncotype DX, Prolaris, Decipher, and ProClass—an in‐house candidate panel) across 37 tissue sections from two patients with localized high‐grade disease. Analyses were performed to quantify biomarker gene abundance across tissue sections using the Visium Spatial Platform, assess spatial variability using global Moran's I, and identify biomarker localization to specific biological niches through spatial co‐expression network analysis. Results Tissue type composition varied markedly between tissue sections. Several genes from Oncotype DX ( KLK2 , AZGP1 , TPM2 , GSN, FLNC, COL1A1 ) and Decipher ( ANO7 , MYBPC1 ) were consistently spatially variable across the samples (mean global Moran's I > 0.1). In contrast, cell cycle‐associated genes, predominantly in the Prolaris panel, exhibited weak expression limited to a small proportion of sample spots. ProClass genes also showed limited expression, impeding robust spatial analysis. Weighted gene co‑expression network analysis identified spatial modules closely aligned to histopathological annotations, linking most Oncotype DX genes to stromal networks, whereas Decipher genes spanned diverse networks. Conclusions Spatial transcriptomics revealed significant variability in biomarker gene expression across highly heterogeneous prostate cancer tissue sections from two patients, providing proof‐of‐concept for its potential use in prognostic biomarker panel research. Several Oncotype DX genes were notably spatially variable, predominantly localizing to stromal regions, whilst the majority of biomarker genes from other panels exhibited non‐spatial patterns. Although low‐abundance genes may be impacted by current technological limitations, integrating spatial data into biomarker research holds promise for developing “spatially robust” panels to provide reliable prognostic information amidst molecular heterogeneity in prostate cancer.
Taylor et al. (Mon,) studied this question.