Living breast cancer subtype classification by membrane-interfacing 3D surface-enhanced Raman spectroscopy substrates with multivariate analysis.

Surface-enhanced Raman spectroscopy (SERS) is a non-destructive and highly sensitive technique widely used for analyzing complex biological samples. However, conventional SERS approaches for living cell analysis face significant challenges. In particular, nanoparticle-based intracellular probes can induce cytotoxicity, and attempts to use 3D protruding nanostructures to interrogate cells externally have shown limited success so far. Furthermore, existing cell-interfacing SERS substrates typically capture only a small fraction of cellular biomolecular signals, producing sparse data that often limits classification tasks to simple binary outcomes. To address these limitations, we developed a 3D multilayer nanolaminate SERS substrate composed of gold and silica (Au/SiO₂) that interfaces directly with the cell membrane. Using this platform, we achieved high-speed, high-throughput, label-free SERS mapping of living breast cancer cells, obtaining large and information-rich spectral datasets. In contrast to prior methods, our approach enabled the classification of four breast cancer subtypes with 92.5 % accuracy using conventional machine learning algorithms. This label-free SERS platform demonstrates potential for more complex cellular analyses via advanced machine learning, including studies of cellular responses to external stimuli such as drug treatments.