Measuring Metabolic Changes in Cancer Cells Using Two‐Photon Fluorescence Lifetime Imaging Microscopy and Machine‐Learning Analysis

ABSTRACT Two‐photon (2P) fluorescence lifetime imaging microscopy (FLIM) was used to track cellular metabolism with drug treatment of auto‐fluorescent coenzymes NAD(P)H and FAD in living cancer cells. Simultaneous excitation at 800 nm of both coenzymes was compared with traditional sequential 740/890 nm plus another alternative of 740/800 nm, before and after adding doxorubicin in an imaging time course. Changes of redox states at single cell resolution were compared by three analysis methods: our recently introduced fluorescence lifetime redox ratio (FLIRR: NAD(P)H‐a 2%/FAD‐a 1%), machine‐learning (ML) algorithms using principal component (PCA) and non‐linear multi‐Feature autoencoder (AE) analysis. While all three led to similar biological conclusions (early drug response), the ML models provided statistically the most robust significant results. The advantage of the single 800 nm excitation of both coenzymes for metabolic imaging in above mentioned analysis is demonstrated. Three different ways for data analysis—machine learning, principal component analysis (PCA) and fluorescence lifetime redox ratio—are compared, measuring the metabolic/redox response of cancer cells after treatment. Fluorescence lifetime imaging microscopy (FLIM) images of the two coenzymes NAD(P) and FAD are acquired at three different two‐photon excitation protocols. The single‐wavelength (800 nm) of simultaneous excitation of both coenzymes performs well versus the traditional sequential 740‐ and 890‐nm excitations, offering image acquisition efficiency, spatial and temporal coherence of the coenzymes with less laser power exposure.