Electrical readings track dying CHO cells in two stages, a bioprocessing review reports

The effect in the CHO data is large. In one study the review describes, cytoplasmic conductivity was about 0.45 siemens per meter in viable cells and about 0.05 S/m in apoptotic ones ^[8], a ninefold difference ^[13]. A follow-up measured 0.07 S/m in apoptotic cells by 52 hours ^[9]. The review calls CHO cells the "industry workhorse" ^[6].

The order of the changes interests me more than their size. Capacitance falls first, and gradually, as the membrane remodels. Conductivity collapses later, when the cell can no longer hold its ionic balance ^[7][9]. In the first study, apoptotic populations began to show up between 24 and 36 hours ^[8]. I'd expect any useful warning to come from the capacitance phase, because the conductivity plunge is measured in cells already classed as apoptotic ^[8].

The review is by Alaleh Vaghef-Koodehi of the University of Massachusetts Amherst and Blanca Lapizco-Encinas of the Rochester Institute of Technology ^[1]. It covers work from 2015 through 2026 on membrane capacitance, cytoplasmic conductivity, polarizability and surface charge as label-free readouts of cell state ^[2]. The main method is dielectrophoresis, which tracks how polarizable cells respond to a nonuniform electric field without labeling or destroying them ^[5]. Fluorescent staining and lactate dehydrogenase release assays run offline. They can destroy the sample and give limited time resolution ^[4].

Speed is the selling point. The commercial 3DEP platform has captured electrical signatures from roughly 20,000 cells within seconds ^[10]. The review says such measurements can reveal emerging apoptotic subpopulations "hours before traditional biochemical markers become detectable" ^[14].

The CHO studies described here all triggered death the same way, by starving the cells of nutrients ^[7]. That is a sensible first experiment, with a single controlled cause of death. It is still one cause, in one cell line, and a production batch going wrong in your plant is not a starved culture in a lab. GEN's account of the review does not say how many hours of lead time the electrical signal buys, what baseline the 90% accuracy figure beats, or whether the signatures have been checked against cultures failing for other reasons.

The machine-learning result needs the same care. The 90% figure comes from a 2026 analysis that trained a tuned random-forest model on electrical impedance data, proposed as a label-free "traffic-light" system for batch monitoring ^[11]. The seconds-scale throughput comes from 3DEP, a dielectrophoresis instrument ^[10]. Predictive accuracy on a research dataset is also a different quantity from a false-alarm rate on a manufacturing floor, where a wrong red light means an unnecessary intervention.

In the review, the authors call real-time monitoring of cell health a "critical bottleneck" in biologics manufacturing ^[3]. They see continued integration with computational modeling and machine learning as the route to real-time automated systems, and commercial dielectrophoresis platforms are already making the method more accessible ^[12]. By GEN's account, it is not yet routine bioprocess control ^[12].