Computational identification of FOXP3-associated spatial prognostic markers in HCC via digital pathology

Traditional differential gene identification relies on bulk analysis, which lacks spatial resolution and limits the detection of spatially variable genes due to intratumoral heterogeneity. Spatial transcriptomics addresses this, but high costs reduce sample sizes and statistical power.

Building upon our previous work, SciSt, on spatial gene inference, this study constructs a computational pipeline to identify prognostically relevant spatial markers using digital pathology slides from the TCGA-LIHC cohort. We identified prognostically significant genes by analyzing their spatial distribution patterns and mapped their expression onto segmented tumor and stroma regions to derive biologically meaningful spatial features. Tissue types were segmented using established marker genes and a pathologist-annotated slide.

FOXP3 emerged as the sole gene showing a significant association with prognosis (P < 0.01). Spatial kurtosis was the main driver of this association. Based on FOXP3 expression across tissue compartments, six quantitative spatial features were derived, enabling stratification of patients into three groups: FOXP3_Sp1, FOXP3_Sp2 and FOXP3_Sp3. This classification served as an independent prognostic (HR = 1.57, 95 % CI 1.03-2.41). Accounting for observed racial disparities, subgroup survival analysis remained significant in White patients (P = 0.035). Patients in FOXP3_Sp1 exhibited the poorest prognosis, characterized by abnormally high FOXP3 expression in tumor regions and low expression in stroma. Conversely, FOXP3_Sp2 displayed the opposite pattern and intermediate prognosis, while FOXP3_Sp3 showed balanced expression across both regions and the most favorable outcome.

We validated six prognostic spatial markers in HCC and elucidated the dual role of FOXP3 within tumor and stromal compartments, demonstrating the potential of this approach as a practical tool for personalized clinical decision-making in HCC. This study pioneers the computational identification of spatially differential genes at spot-level resolution, offering a cost-effective and scalable approach for spatial biomarker discovery.