Machine learning-assisted single-cell Raman imaging for rapid, sensitive detection and intracellular mapping of carotenoids in plant cell cultures

<h4>Key message</h4>CRaman imaging combined with a multi-layer perceptron neural network enables non-destructive, label-freeclassifi cation of tobacco BY-2 cells based on carotenoid composition. Carotenoids are natural tetraterpenoid pigments with important nutritional properties and broad industrial applications. Enhancing their production in plant-based biofactories offers a sustainable alternative to current manufacturing processes. In this work, we developed a label-free, single-cell analytical platform combining Raman imaging with a multi-layer perceptron neural network to classify tobacco BY-2 cells based on their carotenoid content. Carotenoid standards analysis, including astaxanthin, canthaxanthin, and β-carotene, was performed by surface-enhanced Raman scattering using hydrophobic gold nanostars due to the low concentration available. This analysis allowed the assignment of characteristic Raman peaks, specifically at 1160 cm^-1 and 1520 cm^-1, of key carotenoids and their identification inside of the cells by Raman imaging. The Raman fingerprints were correlated with carotenoid profiles obtained by HPLC, enabling accurate differentiation between wild-type and transgenic cell lines. In the analyzed transgenic lines, carotenoids accumulated in vesicle-like structures near the nucleus and along the cytoplasmic membrane. This method provides a non-destructive, label-free approach with high classification accuracy and sorting potential based on carotenoid composition, and may be a useful tool for plant synthetic biology and metabolic engineering.