Cell counters and viability systems

A cell counter reports two numbers, concentration and viability, and only one of them is a straightforward measurement. Concentration is a counting problem. Viability is an inference from a proxy, and which proxy is chosen determines what the number means.

Trypan blue answers a narrower question than it appears to

Trypan blue is a large anionic azo dye. An intact plasma membrane excludes it; once inside, it binds cytosolic protein and the cell appears blue. The assay is therefore a test of membrane integrity, nothing more. A cell in early apoptosis has an intact membrane while its caspase cascade is already irreversible, and it is scored as viable. In a culture that has begun to die, trypan blue systematically overstates viability, and it does so most in exactly the situation where an accurate answer matters. The dye is also cytotoxic and the staining window is short — leave the mixture too long and viability falls because of the measurement — and it stains free protein, so debris-rich late-stage broth produces spurious blue objects.

Image-based instruments automate the same chemistry with brightfield optics and segmentation. Their real difficulty is not detection but clump resolution: a doublet counted as one object and a clump counted as one large object both bias concentration low, and the deconvolution algorithms that split them are opinionated and vendor-specific. Two instruments disagreeing on the same suspension usually disagree about segmentation, not about staining.

Impedance and fluorescence answer different questions

The Coulter principle measures something else entirely. A cell drawn through a small aperture between two electrodes displaces conductive electrolyte, and the resulting resistance pulse has an amplitude proportional to the cell’s volume. This gives excellent size distributions and accurate counts, and it is completely blind to whether the cell is alive. Its own limit is coincidence: at high concentration two cells occupy the aperture together and register as one larger particle, so the sample must be diluted into a regime the instrument can resolve.

Fluorescence counting separates the two questions cleanly. A membrane-permeant nuclear stain such as acridine orange labels every nucleated cell; propidium iodide, excluded by an intact membrane, labels only those that have lost it. Counting nuclei rather than objects removes debris and non-nucleated particles from the result, which is why fluorescence methods track better than brightfield in harvest-stage material. The trade is that anucleate cells are invisible to it. To see the population trypan blue miscounts, a different marker is needed: annexin V binds phosphatidylserine externalised early in apoptosis, before membrane permeability changes, and is read by flow cytometry.

Metabolic assays — tetrazolium reduction, resazurin — are sometimes treated as cell counts. They measure reductase activity. Signal is proportional to cell number only if activity per cell is constant, and it is not: it shifts with growth phase, confluence and stress. These assays report metabolic capacity, and reading them as viability is a category error.

The statistical floor

Counting is a Poisson process, so the relative standard deviation of a count of N events is 1/√N. Counting one hundred cells carries about 10 % uncertainty; a hemocytometer field is in that range. Distinguishing 99 % from 97 % viability requires observing thousands of cells, because the quantity being estimated is the small dead fraction. Automated instruments earn their place mainly by making N large enough for the answer to be stable, not by staining better than a person can.