AUTHOR=Zhong Fangmin , Yao Fangyi , Liu Jing , Fang Qun , Yu Xiajing , Huang Bo , Wang Xiaozhong 

TITLE=Autophagy crosstalk with the immune microenvironment in chronic myeloid leukemia and serves as a biomarker for diagnosis and progression

JOURNAL=Frontiers in Immunology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2025.1570903

DOI=10.3389/fimmu.2025.1570903

ISSN=1664-3224

ABSTRACT=BackgroundPrevious studies have shown that autophagy is closely related to the occurrence, development, and treatment resistance of chronic myeloid leukemia (CML) and has dual roles in promoting cell survival and inducing cell death.MethodsWe analyzed autophagy levels in CML samples via transcriptome data and evaluated the relationships between autophagy and the immune microenvironment, treatment response, and disease progression. A consensus clustering algorithm was used to identify autophagy-related molecular subtypes. The value of autophagy-related genes (ARGs) in diagnosis and treatment evaluation was analyzed and verified by a variety of machine learning algorithms.ResultsCompared with normal samples, CML samples had significantly lower autophagy scores and more downregulated ARGs. The autophagy score was positively correlated with the activity of immune and signal transduction-related pathways and negatively correlated with proliferation-related pathways. Patients with high autophagy scores had a greater proportion of regulatory T-cell infiltration and greater cytokine–cytokine receptor interaction signaling pathway activity, while patients with low autophagy scores had greater γδT cell infiltration and PD-1 expression. Low autophagy scores are also associated with malignant progression and nonresponse to treatment. The immune landscape and chemotherapy sensitivity significantly differed between the two autophagy-related molecular subtypes. Three diagnostic ARGs (FOXO1, TUSC1, and ATG4A) were identified by support vector machine recursive feature elimination, least absolute shrinkage selection operator, and random forest algorithms, and the combined diagnostic efficiency of the three was further improved. The diagnostic value of the three ARGs was verified by an additional validation cohort and our clinical real-world clinical cohort, and they can also be used for the differential diagnosis of CML from other hematological malignancies.ConclusionOur study revealed that CML samples exhibit decreased autophagy, and autophagy may induce Tregs to undergo immunosuppression through cytokines. Autophagy-related molecular subtypes are helpful for guiding the clinical treatment of CML. The identification of ARGs by a variety of machine learning algorithms has potential clinical application value.