Deep learning reveals FLAD1-mediated mitochondrial metabolic reprogramming in hypoxic tumors

Area of Science:

• Oncology
• Metabolic Engineering
• Computational Biology

Background:

• Hypoxia is a critical hallmark of solid tumors, driving cancer progression and therapeutic resistance.
• Tumor metabolic reprogramming under hypoxia presents unique vulnerabilities exploitable for cancer therapy.

Purpose of the Study:

• To systematically compare metabolic network differences between hypoxic and normoxic tumor cells.
• To develop a deep learning model (DepFormer) for identifying hypoxia-dependent metabolic genes.
• To identify FLAD1 as a key metabolic vulnerability and therapeutic target in hypoxic tumors.

Main Methods:

• Systematic comparison of metabolic networks in hypoxic versus normoxic cells.
• Development and application of DepFormer, a transformer-based deep learning model.
• Functional validation of FLAD1's role in tumor cell adaptation to hypoxia.
• Identification of a FLAD1 inhibitor.

Main Results:

• Oxidative phosphorylation identified as a significantly hypoxia-dependent pathway.
• FLAD1 predicted as a key hypoxia-dependent metabolic gene, with amplified locus and upregulated expression in tumors.
• FLAD1 depletion disrupts mitochondrial complex II, causing succinate/fumarate imbalance and hindering hypoxia adaptation.
• A FLAD1 inhibitor selectively inhibits hypoxic tumor cell growth.

Conclusions:

• DepFormer is an effective framework for predicting state-specific metabolic dependencies.
• FLAD1 represents a significant metabolic vulnerability in hypoxic tumors.
• FLAD1 inhibition offers a novel therapeutic strategy for targeting hypoxic tumors.