3D convolutional neural networks predict cellular metabolic pathway use from fluorescence lifetime decay data

Area of Science:

• Biophysics
• Cell Biology
• Machine Learning

Background:

• Fluorescence lifetime imaging of reduced nicotinamide adenine dinucleotide (NADH) provides label-free detection of cellular metabolic changes.
• Interpreting NADH lifetime variations to understand specific metabolic pathway usage remains challenging due to complexity.

Purpose of the Study:

• To develop a 3D convolutional neural network (CNN) capable of identifying metabolic pathway usage in cancer cells using NADH fluorescence lifetime imaging.
• To enable robust, label-free, single-cell resolution metabolic measurements.

Main Methods:

• A 3D CNN was trained using 3D NAD(P)H lifetime decay images of MCF7 breast cancer cells cultured under isolated metabolic conditions (glycolysis, oxidative phosphorylation, glutaminolysis).
• Multiphoton fluorescence lifetime microscopy was used to acquire NADH lifetime images, which were then segmented into individual cells for model input.

Main Results:

• The 3D CNN models achieved over 90% accuracy in classifying cancer cells based on their reliance on glycolysis, oxidative phosphorylation, or glutaminolysis.
• The model successfully predicted metabolic phenotype differences in macrophages from control and POLG-mutated mice, demonstrating cross-species applicability.

Conclusions:

• Integrating autofluorescence lifetime imaging with 3D CNNs allows discrimination of multiple metabolic phenotypes by analyzing NADH intensity patterns and lifetime decay dynamics.
• This approach eliminates the need for complex decay fitting, paving the way for live-cell and in vivo metabolic measurements with single-cell resolution.