An enormous share of the numbers produced by laboratories every day come from a plastic tray with ninety-six small wells in it. Hormone levels, antibody titres, allergen screening in food, cytokine measurements in research, infectious disease serology: all of them frequently reduce to how much yellow appeared in a well and how that compares with a row of standards.
The technique is called the enzyme-linked immunosorbent assay, and its durability comes from a simple trick. Antibodies are exquisitely good at recognising specific molecules but produce no signal when they bind. Attach an enzyme to the antibody and the binding event becomes visible, because a single enzyme molecule converts a great many substrate molecules into a coloured product. Specificity comes from the antibody; sensitivity comes from the enzyme’s catalytic amplification.
What confuses people is that four quite different arrangements of the same components are all called ELISA. They differ in what is stuck to the plate, what is being detected and, in one case, whether more signal means more or less of the thing you are measuring.
Key takeaways
- Specificity comes from the antibody and sensitivity from enzymatic amplification of the signal.
- Direct, indirect, sandwich and competitive formats answer different questions with the same hardware.
- In a competitive ELISA the signal runs backwards: more colour means less analyte.
- The standard curve is not a straight line, and results outside its usable range should not be extrapolated.
- Washing removes unbound reagent, and most background problems trace back to washing or blocking.
The Microplate Well as a Reaction Vessel
The plate looks trivial and is not. Each well is a small vessel whose inner surface has been treated to bind protein, and that surface is where the assay physically happens.
Binding to untreated polystyrene occurs through hydrophobic interaction between the plastic and non-polar regions of the protein. It is passive, requires only incubation in a suitable buffer, and works well for many proteins. It is also uncontrolled: molecules land in random orientations, some with their binding sites facing the plastic and therefore useless, and the amount bound varies with protein concentration, buffer pH and ionic strength. Manufacturers offer plates with modified surfaces, some more hydrophilic for proteins that adsorb poorly, some carrying reactive chemical groups that form covalent bonds for molecules that will not stay put otherwise.
Whatever is coated, the remaining bare surface must then be filled with irrelevant protein, exactly as in other solid-phase immunoassays, or every subsequent reagent will stick to it directly and generate signal unrelated to the analyte.
Two physical properties of the plate matter for the reading step. Wells intended for absorbance measurement are clear and flat-bottomed so light passes cleanly through a uniform path length; wells for fluorescence or luminescence are opaque, black or white, to stop light leaking between neighbours. And plate position has real effects: wells at the edges of a plate lose more liquid to evaporation and equilibrate to temperature differently from those in the middle, producing the well-known edge effect in which the outer ring reads systematically differently. Careful assays either avoid the outer wells or distribute samples so that the effect does not align with the comparison being made.
Direct, Indirect and Sandwich Formats
The three non-competitive formats differ in how many layers sit between the plate surface and the enzyme.
In a direct ELISA, the sample is coated onto the well and a single enzyme-labelled antibody binds the target. It is the shortest protocol with the fewest cross-reactivity opportunities, but it requires the detection antibody to be individually conjugated, and it offers no amplification beyond the enzyme itself. It is also limited by the fact that coating a crude sample puts everything in the sample onto the plate, so the target competes for surface with everything else.
In an indirect ELISA, the antigen is coated, an unlabelled primary antibody binds it, and a labelled secondary antibody against the primary’s species provides the signal. Several secondaries can decorate each primary, so sensitivity improves, and one labelled secondary reagent serves any number of unlabelled primaries. This is the standard arrangement for serology, where the coated antigen is from a pathogen and the primary antibody is the patient’s own, with the secondary directed against human immunoglobulin.
In a sandwich ELISA, an unlabelled capture antibody is coated first, the sample is added, and the target is caught from solution before a labelled detection antibody binds a second site on the same molecule. This is the workhorse for quantifying a protein in a complex fluid, because the capture step provides purification: only the target is retained on the plate and everything else washes away.
| Format | What is coated on the well | What is measured | Typical use |
|---|---|---|---|
| Direct | The sample itself | Analyte bound by one labelled antibody | Quick screening, purified samples |
| Indirect | Known antigen | Antibody present in the sample | Serology, antibody titres |
| Sandwich | Capture antibody | Analyte captured from solution | Quantifying proteins in serum or culture medium |
| Competitive | Antigen or capture antibody | Analyte, by how much signal it suppresses | Small molecules with a single binding site |
The sandwich format has one structural requirement worth naming: the target must be large enough to be bound by two antibodies at once, at sites that do not overlap. Small molecules cannot satisfy this, which is precisely why the fourth format exists.
Competitive ELISA and Inverted Signals
A hapten, a drug metabolite, a hormone of small molecular size, a mycotoxin: none of these can host two antibodies simultaneously. Competition solves the problem by measuring the analyte indirectly, through its ability to occupy binding sites that would otherwise be filled by a labelled competitor.
A common arrangement coats the well with antigen. The sample is mixed with a limited amount of antibody and added. Analyte in the sample binds that antibody in solution, and any antibody that is already occupied cannot bind the antigen on the plate. After washing, the remaining plate-bound antibody is detected. A sample containing a lot of analyte leaves little antibody available, so little signal develops. A sample containing none leaves the antibody free to bind the plate, and the signal is maximal.
The consequence is a curve that slopes downward: absorbance is inversely related to concentration. This is straightforward once expected and a serious hazard when not. Reading a competitive result with the mental model of a sandwich assay inverts the answer entirely, and it has produced published errors.
Competitive designs bring other characteristics. They are generally more tolerant of complex sample matrices than sandwich assays because the critical binding happens in solution. They tend to be more sensitive to small variations in incubation time and temperature, since the outcome depends on a competition reaching a particular point. Their dynamic range is often narrower, and precision is poorest at the extremes of the curve, where it flattens.
Enzyme Labels and Colour Development
Two enzymes dominate. Horseradish peroxidase is small, fast and cheap, and it oxidises substrates that change colour on oxidation. Alkaline phosphatase is larger and slower per molecule but more stable, and it removes phosphate groups from substrates to yield coloured or fluorescent products.
The most common peroxidase substrate produces a blue product that is converted to yellow by adding acid, which also stops the reaction. That stop step is not a formality. The enzyme reaction proceeds for as long as substrate and enzyme are in contact, so colour keeps deepening, and the relationship between analyte and absorbance holds only if every well is stopped after the same interval. A plate developed by hand where the first column has been reacting appreciably longer than the last carries a gradient that has nothing to do with the samples.
Substrate consumption sets the upper limit of the assay. Once local substrate is depleted or the enzyme is saturated, extra analyte produces no extra colour and the curve flattens. Wells reading at the top of the instrument’s range are in that flat region and cannot be quantified; they must be diluted and repeated.
Choosing between chromogenic, fluorescent and chemiluminescent detection is a trade between convenience and sensitivity. Colorimetric readings are simple and need only an absorbance reader. Fluorescent and luminescent substrates extend sensitivity by one or more orders of magnitude and widen the usable range, at the cost of a more capable reader and, for luminescence, a signal that changes with time and must be read on a schedule.
Standard Curves and Reading Absorbance
An ELISA does not measure concentration. It measures optical density, and concentration is inferred by comparing that reading against standards of known concentration run on the same plate.
Those standards must be on the same plate, not on a previous one, because plate-to-plate variation in coating, incubation and development is large enough to make a stored curve unreliable. Running standards in duplicate or triplicate is normal, and the agreement between replicates is the first indication of whether the plate worked.
The curve is not linear. It is characteristically sigmoid, with a flat region at low concentrations where signal cannot be distinguished from background, a usefully steep central region, and a flat region at high concentrations where the system is saturated. Fitting is usually done with a four-parameter logistic model, which describes that shape properly. Fitting a straight line through the middle points is a common shortcut that introduces error towards the ends of the range.
Only the steep central region gives reliable quantification. Samples reading below the lowest standard should be reported as less than that standard rather than assigned a number, and samples reading above the highest should be diluted and rerun rather than extrapolated. Extrapolation beyond the curve is the single most frequent misuse of ELISA data, and it produces numbers that look precise and are essentially invented.
Blanks and controls complete the picture. A blank containing everything except the analyte defines the background to be subtracted. Quality control samples of known concentration, ideally at low and high points on the curve, run on every plate, are what allow results from different plates and different days to be compared at all.
Washing Steps and Background Noise
Between every stage, unbound reagent must be removed, and washing is where more assays fail than any other single step.
The purpose is straightforward: any labelled reagent left in the well when substrate is added contributes signal regardless of whether it is bound to anything. Washing exploits the difference in binding strength between the specific antibody-antigen interaction, which is strong and slow to dissociate, and non-specific adsorption, which is weak. Repeated cycles of filling, soaking briefly and aspirating strip away the weakly held material while leaving the specific complexes intact.
Wash buffer normally contains a low concentration of a non-ionic detergent, which greatly improves removal of loosely adsorbed protein. Too much detergent starts to strip specific complexes as well and reduces signal, so the concentration is not a free parameter.
Under-washing produces high background across the whole plate, compressing the difference between low and high standards and destroying sensitivity. Incomplete aspiration leaves residual volume that dilutes the next reagent inconsistently. Manual washing tends to be uneven between wells, and an automated washer with a blocked dispense or aspirate needle produces a striking pattern of aberrant wells in a single column or row, which is diagnostic once recognised.
One further habit matters: wells should not be allowed to dry out between steps. A dried surface denatures the bound proteins and behaves unpredictably afterwards, which is why the next reagent is added promptly after the final aspiration.
Sources of Plate-to-Plate Variation
The reason every plate carries its own standards is that a great many things differ from one plate to the next, and most of them are invisible.
Temperature is a leading contributor. Antibody binding and enzyme activity are both temperature-dependent, so a plate incubated on a bench in a warm room develops differently from one in a controlled incubator. Within a single plate, temperature gradients produce the edge effect described earlier, and stacking plates during incubation makes it considerably worse because the ones in the middle equilibrate more slowly.
Timing is next. Every incubation and the development step have specified durations, and the assay assumes they were equal across wells. With a multichannel pipette and ninety-six wells, several minutes can separate the first and last well of a plate unless the operator works deliberately in a consistent direction and stops the reaction in the same order.
Reagents drift. Coated plates degrade in storage, conjugate loses activity, substrate is light-sensitive and develops background as it ages, and lot changes in any antibody can shift the curve. A plate that produces a standard curve noticeably flatter or lower than previous runs usually points to a reagent rather than to the samples.
Pipetting accounts for the rest. Small volumes into small wells amplify technique differences, and residual liquid on tip exteriors, inconsistent depth and a poorly calibrated multichannel all leave patterns in the data. Replicates are what expose this: consistently poor agreement between duplicate wells is a technique signal, not noise to be averaged away.
The practical response is not to eliminate variation, which is impossible, but to make it visible and to design around it. Standards on every plate, controls at known concentrations, samples that must be compared placed on the same plate rather than across plates, replicates that are genuinely independent, and a record of reagent lots. An ELISA run this way produces numbers that can be trusted and compared. One run without those controls produces numbers that look identical on the page and cannot be defended when someone asks how they were obtained.
Frequently asked questions
Why are my duplicate wells so different from each other?
Consistently poor replicate agreement almost always indicates a physical handling problem rather than genuine biological variation. Pipetting technique is the first suspect, particularly with small volumes, worn tips or an uncalibrated multichannel device. Uneven washing is the second, since a well that retained residual buffer will dilute the next reagent differently. Bubbles in a well interfere with the optical reading, and a plate that was not properly mixed after adding a reagent will have concentration gradients. If the poor replicates cluster in a particular row or column, the cause is almost certainly the washer or the pipette rather than the samples.
What is the hook effect and how do I avoid it?
In a sandwich assay, an extremely high concentration of analyte can saturate both the capture and detection antibodies separately, so that analyte molecules occupy capture sites without being linked to labelled detection antibody. The result is a paradoxically low signal from a very high sample, which can be misread as a low concentration. It is dangerous precisely because the result looks plausible. Protection comes from running suspect samples at more than one dilution: a genuine result scales predictably with dilution, whereas a sample affected by this phenomenon produces a higher apparent concentration when diluted further.
Can I use a stored standard curve instead of running standards every time?
It is not advisable, because the sources of plate-to-plate variation described above are large enough to shift the curve meaningfully between runs. Coating density, reagent age, incubation temperature and development timing all differ, and a stored curve carries none of that information. Some highly automated platforms with tightly controlled conditions and calibrator adjustment do work this way, but a manually run research ELISA does not have those controls. Running standards on every plate costs wells and is the price of comparable numbers.
Does a positive ELISA result confirm a diagnosis?
Not on its own, and screening assays are generally designed with this in mind. Immunoassays are usually tuned to favour sensitivity, accepting some false positives in order to miss as few true positives as possible, which means a reactive result in a population where the condition is uncommon has a meaningful chance of being incorrect. Cross-reactivity with related molecules, interfering substances in the sample and non-specific binding all contribute. This is why screening protocols specify confirmation by a different method that relies on a different principle, rather than simply repeating the same assay.
What causes an entire plate to read high or low?
A uniformly high plate usually points to background: inadequate blocking, insufficient washing, conjugate at too high a concentration, or substrate that has aged and begun developing colour on its own. A uniformly low plate points to a missing or inactive component, with expired conjugate, an omitted reagent, incorrect buffer pH or a substrate that was exposed to light being the usual candidates. Checking the blank wells separates the two possibilities quickly: high blanks indicate background, while low blanks combined with weak standards indicate a detection failure.
This is education, not medical advice. Laboratory results only carry meaning alongside your symptoms, history and examination. Talk to a qualified clinician about your own results before changing anything about your care or supplements.
