Machine Learning\nClassification of Integrin-Expression-Based\nMagnetic Sorted SW 620 Cells by Simultaneous O‑PTIR and SERS

Immortalized cell lines are commonly used for in vitro\nstudies\nsuch as drug efficacy, toxicology, and life cycle due to their cost\neffectiveness and accessibility; however, subpopulations within a\ncell line can arise from random mutations or asynchronous cell cycles\nwhich may lead to results that make interpretation difficult. A method\nthat could classify these differences and separate unique subpopulations\nwould increase our understanding of heterogeneous cellular responses.\nIn the present work, we explore spectroscopic signals associated with\nsubpopulations of cells magnetically sorted on the basis of α<sub>5</sub>β<sub>1</sub> integrin binding to cyclic-RGDfC which\nmimics fibronectin in the extracellular matrix. SW620 colon cancer\ncells were incubated with cyclic-RGDfC functionalized gold-coated,\niron core nanoparticles and magnetically sorted. The subpopulations\nfrom the sort were imaged (<i>N</i> = 10 positive and <i>N</i> = 10 negative, number of cells) via simultaneous surface-enhanced\nRaman scattering (SERS) and optical–photothermal infrared spectroscopy\n(O-PTIR). Pearson correlations of the standard peptide–protein\ninteraction in the SERS channel allowed for visualization of the cyclic\nRGDfC–integrin α<sub>5</sub>β<sub>1</sub> interaction.\nPartial least-squares discriminant analysis of the O-PTIR spectra\ncollected from cell maps successfully classified the positively or\nnegatively sorted cells. These results demonstrate that biochemical\nchanges within a single cell line can be sorted via an integrin-activity-based\nassay using simultaneous SERS and O-PTIR.