Machine learning-based lactate-related genes signature predicts clinical outcomes and unveils novel therapeutic targets in esophageal squamous cell carcinoma

Chuting Yu; Yan Bian; Ye Gao; Yunfei Jiao; Yusi Xu; Wei Wang; Lei Xin; Han Lin; Luowei Wang

doi:10.1016/j.canlet.2025.217458

Machine learning-based lactate-related genes signature predicts clinical outcomes and unveils novel therapeutic targets in esophageal squamous cell carcinoma

Cancer Lett. 2025 Jan 27:613:217458. doi: 10.1016/j.canlet.2025.217458. Online ahead of print.

Authors

Chuting Yu¹, Yan Bian¹, Ye Gao¹, Yunfei Jiao¹, Yusi Xu¹, Wei Wang¹, Lei Xin², Han Lin³, Luowei Wang⁴

Affiliations

¹ Department of Gastroenterology, Changhai Hospital, Naval Medical University, Shanghai, 200433, China; National Clinical Research Center for Digestive Diseases (Shanghai), Shanghai, 200433, China.
² Department of Gastroenterology, Changhai Hospital, Naval Medical University, Shanghai, 200433, China; National Clinical Research Center for Digestive Diseases (Shanghai), Shanghai, 200433, China. Electronic address: aip_xin@163.com.
³ Department of Gastroenterology, Changhai Hospital, Naval Medical University, Shanghai, 200433, China; National Clinical Research Center for Digestive Diseases (Shanghai), Shanghai, 200433, China. Electronic address: babyhan831@aliyun.com.
⁴ Department of Gastroenterology, Changhai Hospital, Naval Medical University, Shanghai, 200433, China; National Clinical Research Center for Digestive Diseases (Shanghai), Shanghai, 200433, China. Electronic address: wangluoweimd@126.com.

PMID: 39864540
DOI: 10.1016/j.canlet.2025.217458

Abstract

Esophageal squamous cell carcinoma (ESCC), a predominant subtype of esophageal cancer, typically presents with poor prognosis. Lactate is a crucial metabolite in cancer and significantly impacts tumor biology. Here, we aimed to construct a lactate-related prognostic signature (LPS) for predicting prognosis in ESCC and uncovering potential therapeutic targets. We designed a computational framework to identify lactate-related genes (LRGs) and applied machine-learning to generate an optimal LPS model from 103 combinations. The LPS was evaluated for its predictive accuracy regarding patient prognosis, chemotherapy, radiotherapy, and immunotherapy. Analysis also covered genomic and proteomic traits linked to LPS-defined subtypes. The LPS model demonstrated robust and reliable accuracy in predicting survival outcomes in patients with ESCC. Patients with low LPS scores exhibited a more favorable prognosis and an enhanced response to both chemotherapy and radiotherapy. Conversely, patients with high LPS scores exhibited increased sensitivity to BI-2536 and panobinostat. Furthermore, a low LPS score was associated with better prognosis in multiple immunotherapy datasets across cancer types. Genetic amplifications and deletions were detected more frequently in the high-LPS than in the low-LPS group; however, no significant correlation was observed with the tumor mutation burden. Knockdown of GATM, a key LRG, significantly inhibited cell viability, proliferative capacity, and migration and invasion abilities in ESCC cell lines. In conclusion, the LPS score can be used to predict the prognosis of patients with ESCC and facilitate a more precise approach for selecting patients likely to respond to treatment.

Keywords: Esophageal squamous cell carcinoma; Lactate-related gene; Machine learning; Prognosis.