Background Lactate metabolism is a hallmark of cancer metabolic reprogramming, shaping tumor immunity and therapeutic resistance, yet clinically accessible and low-cost methods to assess intratumoral lactate activity remain limited. Methods We curated a lactate-related 59-gene signature and characterized its biological and clinical relevance across TCGA, GEO, and single-cell RNA-seq datasets. By integrating multi-omic, spatial, and computational analyses, we linked lactate metabolism to the tumor microenvironment and developed a deep learning framework to infer lactate metabolic states directly from routine HE whole-slide images. Results High lactate activity (LACH) was associated with enhanced tumor proliferation, suppressed immune infiltration, and poor response to both immunotherapy and radiotherapy in HNSCC. The pathology-based model achieved robust performance in distinguishing LACH from LACL tumors (AUC = 0. 73–0. 82 in HNSCC) and demonstrated strong generalizability across 12 TCGA cancer types (AUC = 0. 78–0. 89). Importantly, external validation in an independent real-world SAZHU-HNSCC cohort confirmed that model-predicted LACH tumors exhibited significantly increased protein expression of LDHA and MCT1 by immunohistochemistry, supporting the biological validity of the digital lactate biomarker. Conclusions This study integrates multi-omics and digital pathology to infer tumor lactate metabolism from routine histology, providing a scalable and clinically practical digital biomarker for metabolism-informed precision oncology.
Feng et al. (Thu,) studied this question.