Abstract
INTRODUCTION: Cancer development is intricately linked with metabolic dysregulation, including lactic acid metabolism (LM), which plays a pivotal role in tumor progression and immune evasion. However, its specific implications in lung squamous cell carcinoma (LUSC) remain unclear. METHODS: We used numerous datasets encompassing bulk and single-cell transcriptome, genome, intratumor microbiome, and digital pathome to systematically investigate the LM patterns in LUSC. Multiple machine learning algorithms were used to generate the LUSC classification. Histopathology image-based deep learning model was used to predict the classification. Casual mediation analysis was conducted to uncover the association among intratumor microbiota, LM, and immunity. RESULTS: Two LM-based subtypes were discovered endowed with distinct clinical outcomes and biological peculiarities, such as overall survival, somatic mutations, and intratumor microbiota structure. Moreover, the histopathology image-based deep learning model accurately predicted our LM-based LUSC taxonomy, significantly improving its clinical utility. Machine learning models based on seven LM-related genes (CHEK2, LIPT1, TUFM, NDUFA10, AGK, PNPLA2, and GFM1) accurately predicted immunotherapy outcomes for multiple cancer types, including LUSC, and outperformed other currently known biomarkers. Furthermore, mediation analysis identified potential association pathways involving tumor-resident microbes, LM-related gene signatures, and antitumor immune cells. DISCUSSION: Overall, this study advanced the understanding of the relationship between LM patterns and LUSC tumor biology, as well as its potential clinical implications, which might advance the tailored management of LUSC.