Intelligent programmable droplet with deterministic cell control for quantitative single-cell metabolic analysis

Single-cell metabolic heterogeneity influences cell state, disease progression, and therapeutic response, yet quantitative metabolic analysis remains constrained by stochastic cell loading and uncertain droplet occupancy. Here, we developed an intelligent programmable droplet microfluidic platform (ProgDrop-Met) that integrates dielectrophoresis (DEP)-assisted cell encapsulation, deep-learning (DL)- assisted label-free droplet classification and sorting, and chemiluminescence (CL) detection for quantitative single-cell glucose analysis. By synchronizing cell release with droplet formation, ProgDrop-Met programmed the generation of droplets containing one, two, three, or ≥4 cells. A Hough-transform-guided ResNet18 model classified five droplet-occupancy states with >97% accuracy and triggered closed-loop DEP sorting. The resulting target-cell encapsulation rates were 88.7%, 83.9%, 73.5%, and 72.6%, with recovery rates of 54.6%, 58.3%, 56.3%, and 50.6%, for single-cell, two-cell, three-cell, and ≥4-cell droplets, respectively. Coupling the sorted droplets with microsampling and CL detection enabled glucose quantification with a limit of detection of 0.85 μM. After 12 h of culture, glucose-consumption rates were 0.316 ± 0.049, 0.640 ± 0.062, and 0.833 ± 0.059 pmol⋅h^-1 for droplets containing one, two, and three K562 cells, respectively. The platform also resolved a marked difference between single K562 and GM12878 cells and supported time-resolved metabolic measurements over 24 h. ProgDrop-Met therefore provides an integrated, label-free framework for programmable droplet occupancy and quantitative single-cell metabolic phenotyping.