A researcher orders an antibody from a catalogue. The listing names a target protein, shows a clean western blot with a single band at the expected molecular weight, and cites a handful of papers. The vial arrives, the experiment works, a band appears where one was expected, and that band becomes a figure. The figure becomes a claim about biology.
The uncomfortable part is that in a substantial fraction of cases, nobody in that chain has established that the antibody binds the named protein and nothing else. Not the supplier, whose validation may have consisted of a single blot in a single cell line. Not the citing papers, which frequently inherited the assumption from each other. And not the researcher, who had no reason to doubt a reagent that produced the expected result.
Antibodies are not chemicals. They are biological products with batch-dependent behaviour, and their central property, specificity, is an experimental claim rather than a manufacturing specification. That gap between what a catalogue asserts and what has actually been demonstrated is one of the more consequential and least discussed sources of irreproducible findings in protein research.
This piece works through what specificity actually means, why the standard evidence is weaker than it looks, and which validation routes give a defensible answer.
Key takeaways
- Antibody specificity is an experimental claim about a particular sample type, not a fixed property of the vial.
- Catalogue validation data typically covers one context and rarely rules out cross-reactive binding.
- Genetic controls, where the target protein is removed or reduced, are the strongest single test available.
- Orthogonal methods that measure the protein without antibodies catch errors no antibody control can.
- Lot numbers and catalogue identifiers belong in the methods section, because reagent identity changes between batches.
What Specificity Means for an Antibody
An antibody binds through a small surface at the tip of its variable regions, which contacts a patch of the target protein a few amino acids across, or a shape formed where distant parts of a folded chain meet. That patch is the epitope. Everything about how the antibody behaves follows from the fact that it recognises the epitope, not the protein.
This immediately produces two failure modes. The first is cross-reactivity: another protein in the sample happens to present a similar surface, and the antibody binds it. Proteins within a family often share stretches of sequence, so an antibody raised against one member may be blind to the distinction between it and its close relatives. The second is context dependence: the epitope may be buried in the folded protein, masked by a binding partner, altered by a chemical modification, or destroyed by the fixation and heating steps an assay requires.
Because of the second point, specificity cannot be a single property. An antibody that gives a clean, single band on a denaturing western blot may perform badly in immunofluorescence, where the protein retains its fold and the epitope may be inaccessible. The reverse is equally common. Validation is therefore always validation for an application, and evidence gathered in one assay transfers to another only weakly.
Sensitivity matters too, but it is the less dangerous problem. An antibody that fails to detect anything gives an obviously null result. An antibody that detects the wrong thing gives a plausible, publishable, entirely wrong result.
Monoclonal and Polyclonal Trade-Offs
The two main classes of research antibody differ in ways that shape their failure modes directly.
A polyclonal antibody is purified from the serum of an immunised animal and is a mixture of many different antibodies, each recognising a different epitope on the target. Because several epitopes are targeted at once, a polyclonal preparation tolerates partial epitope loss well and usually gives a strong signal, which makes it forgiving in difficult applications. The cost is that the mixture is finite and irreproducible. When that animal’s serum runs out, the replacement comes from a different animal with a different immune history, and the new preparation is a genuinely different reagent that must be validated again.
A monoclonal antibody comes from a single immortalised antibody-producing cell line, so every molecule is identical and the supply can in principle continue indefinitely. That consistency is the main argument for monoclonals, and it is a strong one. The trade-off is that a single epitope carries all the risk. If that epitope is masked, mutated or modified in the sample being examined, the antibody sees nothing at all. If that one epitope happens to be shared with an unrelated protein, the cross-reactivity is absolute and no amount of purification removes it.
Recombinant antibodies, where the sequence encoding the binding region is cloned and expressed from a defined construct, are increasingly displacing both. They combine the consistency of a monoclonal with a permanent, sequence-defined identity, which means the reagent can be described exactly rather than by catalogue number alone.
| Property | Polyclonal | Monoclonal | Recombinant |
|---|---|---|---|
| Epitopes recognised | Many | One | One |
| Batch consistency | Poor, animal-dependent | Good within a cell line | High, sequence-defined |
| Tolerance of epitope masking | Good | Poor | Poor |
| Risk from a single cross-reaction | Diluted across the mixture | Total | Total |
| Long-term supply | Ends with the animal | Depends on the cell line | Renewable from sequence |
| Typical background signal | Higher | Lower | Lower |
Why Catalogue Validation Is Not Enough
The validation data on a supplier’s page is usually genuine and usually insufficient, and both halves of that sentence matter.
The commonest evidence offered is a western blot showing a band of the expected size. This is weaker than it appears. Molecular weight is a coarse property, and a great many proteins in any lysate run at a similar position. A single band tells you that one abundant species was detected under those transfer and blocking conditions in that lysate. It does not tell you the species is your protein, and blots are frequently shown cropped closely around the band of interest, removing exactly the information that would reveal additional binding elsewhere on the membrane.
The second common evidence is a set of literature citations. Citations record use, not validation. A reagent that produced usable-looking data in one laboratory accumulates citations, and those citations are then read as endorsement by the next buyer, which produces more citations. Nothing in that loop tests specificity, and a reagent can accumulate a long citation list while never having been checked against a sample lacking the target.
Third is the overexpression control, in which cells engineered to produce large amounts of the target are compared with normal cells. A signal that increases is reassuring about the direction of binding but says little about behaviour at native abundance, where the target may be a minor component competing with far more plentiful proteins for the antibody’s attention.
None of this means catalogue data should be ignored. It means it should be treated as a starting hypothesis about the reagent, to be tested in the sample type and application that the study actually uses.
Knockout and Knockdown Controls
The strongest available control is conceptually simple: run the assay on material where the target protein has been removed, and see whether the signal disappears.
Genetic knockout, most often achieved now with genome editing, produces cells or tissue lacking the gene entirely. If the antibody still gives a band or a stain in that material, whatever it is binding is not the target protein. This is close to a definitive negative control, and it is the reason knockout validation has become the benchmark that serious reagent initiatives push towards.
Knockdown, where the messenger RNA is degraded or its translation blocked, is easier to arrange but weaker as evidence. Reduction is partial, so the expected result is a diminished rather than absent signal, and judging whether a band has reduced enough is a subjective call. Knockdown is a reasonable substitute when knockout is impractical, provided the reduction achieved is independently confirmed at the RNA level.
Two cautions apply. Knockout cells adapt: removing a gene changes the expression of others, and a signal that shifts may reflect that adaptation rather than direct loss of the target. And some editing strategies produce truncated proteins rather than none at all, which can preserve or destroy the epitope depending on where the edit fell. The knockout should be characterised, not assumed.
Where the target is essential and cells cannot survive without it, conditional or inducible systems allow the protein to be depleted temporarily, and the same logic applies to the comparison.
Orthogonal Validation Strategies
Genetic controls test whether the antibody needs the target to give a signal. Orthogonal validation asks a different question: does the antibody’s readout agree with a measurement made by a method that does not involve antibodies at all?
Mass spectrometry is the most powerful of these. A band can be excised, digested and sequenced, identifying the protein present directly from its peptides. Targeted mass spectrometry can also quantify a protein across a set of samples, producing a pattern that the antibody signal should mirror. Where the two disagree, the antibody is the more likely culprit.
Comparing protein signal against messenger RNA measurement across a panel of cell lines or tissues is cheaper and surprisingly informative. Transcript abundance and protein abundance are not identical, so perfect correlation is not expected, but an antibody whose signal is high in samples where the transcript is absent is telling you something. This approach works particularly well when the panel includes tissues at both extremes of expression.
Using two independent antibodies raised against different, non-overlapping epitopes of the same protein is a further route. Agreement between them is meaningful because the two would be unlikely to cross-react with the same wrong protein through different epitopes. Disagreement is informative in itself and should be resolved rather than settled by picking the antibody that gives the expected answer.
Finally, tagging the endogenous protein with a short epitope tag through genome editing allows detection with a well-characterised anti-tag antibody, giving an independent readout of the same molecule in the same cells.
Lot-to-Lot Variation and Its Consequences
An antibody catalogue number identifies a product line. The lot number identifies the actual material in the vial, and for biological reagents that distinction carries weight.
For polyclonals, the lot corresponds to a bleed from a particular animal, and the epitope mixture differs between lots as a matter of course. Two lots of the same catalogue product can differ in optimal dilution, in background, and occasionally in whether they detect the target at all in a given application. Monoclonals are more stable, but cell lines drift, production conditions vary, purification and formulation change, and stabilising proteins added to the buffer can interfere with sensitive assays.
The practical consequence is that a mid-project lot change is an uncontrolled variable. A result that shifts when a new vial arrives may reflect biology or may reflect the reagent, and without a bridging experiment there is no way to tell. The remedy is unglamorous: when a lot is validated and performs well, buy enough of it to finish the project, record the lot number, and when a change is unavoidable run the old and new lots side by side on retained samples before committing.
Storage is the related, self-inflicted version of the same problem. Repeated freeze-thaw cycles degrade antibody preparations, and a reagent that worked in month one and fails in month nine has often simply been thawed too many times. Aliquoting on arrival costs an hour and removes an entire category of confusing results.
Recording Antibody Identity in Publications
A methods section that says an antibody against a named protein was used at a stated dilution does not identify the reagent. It does not permit replication, and it does not permit a later reader to check whether the antibody has since been shown to be non-specific.
The minimum useful record is supplier, catalogue number, lot number, host species and clonality, the application and dilution used, and the validation evidence relied on. Persistent identifiers assigned to research resources make this more robust, because catalogue numbers change and suppliers merge, whereas a stable identifier survives both. Where validation was performed in-house, the controls should be shown rather than asserted, ideally including the uncropped blot.
Full-length blot images matter more than the effort suggests. Cropping removes the evidence of additional bands, and the additional bands are precisely what a reader assessing specificity needs to see. Journals increasingly request the original images as supplementary material for this reason.
The broader point is that reagent identity is part of a result, not administrative detail attached to it. Protein research is unusually dependent on reagents whose behaviour is not fully specified by their description, and the field’s reproducibility problem is partly a documentation problem wearing a technical disguise.
Frequently asked questions
Does a single clean band on a western blot prove specificity?
No, and this is the most common misconception in the area. A single band shows that one detected species dominates at that molecular weight under those particular conditions, but many proteins share a similar apparent size, and blots are usually cropped to a narrow window around the expected position. Blocking conditions, exposure time and lysate preparation all affect what appears. A single band is consistent with specificity without demonstrating it. The evidence that carries weight is the disappearance of that band in material lacking the target protein.
Is an expensive antibody more likely to be well validated?
Price tracks production cost, purification method and market position far more closely than it tracks validation depth. Some inexpensive reagents have been characterised extensively because they became widely used in a well-organised field, while some costly ones carry little more than a supplier blot. The useful signals are the presence of genetic knockout data, uncropped images, application-specific evidence and clear lot documentation. None of those correlate reliably with price.
What should be done when two antibodies against the same protein disagree?
Treat the disagreement as the result rather than as noise to be resolved by preference. The first step is to check whether the two are being used in the same application and whether their epitopes lie in regions that might be differently affected by modification, cleavage or splicing, since a genuine biological difference can produce discordance between two specific antibodies. If that is ruled out, a genetic control usually settles it, because the antibody whose signal persists without the target is the non-specific one.
Do commercial validation initiatives solve the problem?
They help substantially without closing it. Independent characterisation programmes and structured validation frameworks have made it far easier to identify reagents with real evidence behind them, and the expectation of knockout data has become considerably more normal than it once was. What such initiatives cannot do is cover every protein, every application and every sample type, since the number of combinations is enormous. They shift the baseline rather than removing the need for context-specific checks.
How much validation is proportionate for a preliminary experiment?
The reasonable standard scales with the weight the result will carry. A quick screen that will be followed by orthogonal work needs less than a figure that anchors a paper’s central claim. A workable rule is that any antibody-based result which will not be independently confirmed by a non-antibody method should have a genetic control behind it. If the finding is exploratory and will be verified later by mass spectrometry, sequencing or a functional assay, lighter validation is defensible provided the limitation is stated rather than left implicit.
The habit that prevents most of this trouble is small: treat every new antibody as an untested hypothesis about a molecule, and design the first experiment to try to make it fail. A negative control sample lacking the target, an uncropped image, a recorded lot number and a second measurement by a different method together cost a few days at the start of a project. The alternative is discovering the problem after the figure has been published, when the cost is measured in retracted conclusions and in the years other groups spend chasing an artefact.
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.
