# Automated ELISA analyzers

The binding and enzyme chemistry underneath automated microplate immunoassay, and the two failure modes — the high-dose hook effect and timing drift — that decide whether an automated result can be trusted.

Why a sandwich immunoassay turns colour in proportion to analyte — and why, above a certain concentration, it stops.

Source: https://en.bioecon.ru/docs/bioproduction-equipment/analytics-pat/automated-elisa-analyzers/
Updated: 2026-09-04



An ELISA is an amplifier built out of two antibodies and an enzyme. A capture antibody is adsorbed onto polystyrene, which binds protein passively through hydrophobic contact rather than any covalent chemistry — the reason plate surface treatment, and not the reagent, often explains why one lot of plates behaves unlike another. Unoccupied surface is then blocked with irrelevant protein, sample is added, and a second antibody carrying horseradish peroxidase binds a different epitope on the same analyte. Each captured molecule therefore ends up tethered to one enzyme, and that enzyme runs continuously: with tetramethylbenzidine and hydrogen peroxide it generates a blue radical cation, and acid stops the reaction and shifts the product to a yellow diimine read at 450 nm. One binding event becomes millions of coloured molecules. That amplification is the whole point, and it is also where the assay's fragility comes from.

## Why the readout is a clock, not a state

Absorbance at the moment of stopping is the integral of enzyme turnover over the substrate incubation. Nothing about it is an equilibrium property. If one row of a plate sits thirty seconds longer than another, that row reads higher, and the difference is indistinguishable from real analyte. The same applies upstream: antibody–antigen association has not reached equilibrium at typical incubation times, so binding is partly kinetic and partly thermodynamic, and both depend on temperature. Plate edge wells warm and cool faster than centre wells, producing the classic edge effect — a systematic gradient, not noise.

This is the honest argument for automation, and it is a metrological one rather than a labour one. A machine that dispenses, incubates and stops every well on the same schedule removes a variance term that no amount of careful pipetting eliminates. Washing carries the same weight: residual unbound conjugate is the background floor, and how completely the wash head aspirates each well sets the lower limit of quantitation more directly than antibody affinity does.

## The hook effect

Sandwich formats have a defect that a monotonic calibration curve hides. At very high analyte concentration, free analyte saturates the capture antibody and, separately, saturates the detection antibody in solution before it reaches the surface. No bridge forms. Signal falls, and the dose–response curve turns back down. A grossly positive sample can read as low, or as negative, and nothing in the result flags it. Because the function is non-monotonic, a single dilution cannot resolve it; two dilutions that do not agree proportionally are the only routine detector. Automated analyzers can be programmed to re-run out-of-pattern samples, but the chemistry is not removed by the instrument.

## The rest of the error budget

Calibration is a four- or five-parameter logistic fit, and precision is worst at both asymptotes, so an assay's usable range is narrower than its measurable one. Sample matrix contributes interference of its own: heterophilic and anti-animal antibodies in human serum can bridge capture and detection antibodies with no analyte present, producing false positives that dilution does not linearise. Method validation under ICH Q2(R2) exists to bound these effects — specificity, range, accuracy and precision — rather than to certify that the chemistry is well behaved. It is not.