Can you trust the datasheet? Validating IHC antibodies in your own lab

In May 2026, a researcher noticed that a western blot on an antibody product page looked edited. By late summer, data-integrity researchers had catalogued validation images showing signs of manipulation across 15 antibody suppliers.

It is a good moment to remember what a datasheet is: the supplier's claims, and the evidence behind them. Even when that evidence is genuine, your lab may use different fixation, a different retrieval set-up (microwave, pressure cooker or water bath) and a different detection system. Existing data help you choose a candidate. Only testing under your own conditions shows whether it works in your experiment.

My own numbers from one postdoc project: I ordered 60 antibodies. Twenty gave the expected staining at the first attempt, twenty needed substantial optimisation and validation, and twenty never worked at all, giving either no staining or background across the whole section. (Yes, I got my money back.) That is one project, not an industry statistic, but it is why I no longer treat a datasheet as a promise.

Trust, but always test.

This practical IHC antibody validation guide shows how to collect that evidence on a normal lab budget, with one illustrative example running through it.

What are the five pillars of antibody validation?

Antibody validation is not a new worry. In 2016, the International Working Group for Antibody Validation proposed five complementary "pillars" of evidence that an antibody binds its intended target in a given application (Uhlén et al., 2016):

1. Genetic strategies: knocking out or knocking down the target should remove or reduce the signal.
2. Orthogonal strategies: staining should agree with an antibody-independent measure of the target, such as RNA levels or mass spectrometry.
3. Independent antibodies: two antibodies against different epitopes should give the same pattern.
4. Expression of tagged proteins: staining should match a tagged version of the target.
5. Immunocapture followed by mass spectrometry: the protein the antibody pulls down should be the intended one.

Validation is application-specific: evidence for western blot does not cover IHC. And you rarely need all five. For an ordinary lab working with FFPE tissue, the first three are realistic, and this guide builds on them.

Can you trust the datasheet?

Yes, as a starting point. Many catalogue antibodies are sold under several brands, so the same images can appear on different vendors' pages. When you compare antibodies, look for:

• Tissue context and cellular detail. Whole-section or TMA views show distribution; a single selected field can hide unwanted staining.
• Data in your application. For FFPE IHC, you want FFPE IHC data; a clean western blot is a different test.
• Genetic evidence, such as knockout or knockdown material, in the same application.
• Stated conditions: retrieval buffer and pH, dilution, detection system.
• Independent data: YCharOS reports, the Human Protein Atlas, and published figures with the same catalogue number or RRID.

An antibody that passes this check is a good candidate, not a validated one.

Polyclonal, monoclonal, recombinant or multiclonal: what's the difference?

Clonality affects specificity, sensitivity and, above all, consistency from vial to vial. The types are steps along one production line:

1. Polyclonal. An immunised animal's serum contains many antibodies, made by many B-cell clones, against many epitopes. Antibody development still starts here, but as a finished product the polyclonal is becoming a thing of the past: nobody fully knows which antibodies are in the mix, or in what proportions.
2. Monoclonal. Single B-cell clones are isolated, immortalised as hybridomas and screened, and the best one is kept.
3. Recombinant monoclonal. The chosen clone's antibody genes are sequenced and expressed in cultured cells, so the antibody is defined by its sequence.
4. Recombinant multiclonal. Several recombinant monoclonals against different epitopes are mixed into one reagent, sometimes sold as "oligoclonal".
5. Chimeric. A recombinant antibody whose binding region comes from one species and its constant (Fc) region from another, for example a rabbit binder on a mouse Fc (multi-Fc-species production). Increasingly common on the market, they let one well-characterised clone be detected with a different secondary: handy in multiplex panels, or to keep an anti-mouse secondary off mouse tissue.

Consistency is not specificity, and every new lot of any antibody should be checked against reference material.

IHC antibody optimisation vs validation: what's the difference?

Optimisation asks which conditions give useful staining. Validation asks what evidence shows the staining is your target. They are easy to confuse because they happen on the same slides. A strong signal, a crisp image and a familiar-looking pattern are results of optimisation; none of them shows identity on its own. You can optimise an antibody to beautiful, reproducible and wrong staining.

What should you expect to see before you stain?

Write your expectations down before the first slide; once a crisp stain is on the microscope, it is tempting to accept whatever it shows. Good sources are the literature, the Human Protein Atlas (protein and RNA data) and UniProt: a compass, not an answer. HPA is a huge dataset and still the most useful starting reference for protein staining, but its staining is antibody-based too. For several targets I worked on, the HPA antibodies performed poorly, and HPA itself flagged the protein evidence as weak. So check each protein's reliability score (enhanced, supported, approved or uncertain), see which antibody was used, and compare the staining with HPA's RNA data.

Collect what you find in a short target profile:

Worked example (illustrative): Protein X. Protein X is a made-up cytoplasmic protein. Published data suggest it is strong in colon epithelium, low in breast epithelium and absent from liver and lymphoid tissue. Our candidate is a rabbit recombinant monoclonal with a clean colon image on its datasheet. We expect cytoplasmic staining in epithelium, and nothing in hepatocytes or lymphocytes.

Which tissues make good positive and negative controls for IHC?

A useful control is one that can fail. Abundant-target tissue alone will hide problems with sensitivity and with staining where there should be none.

A few practical rules:

• A low positive needs evidence of low expression, such as RNA data. Weak staining alone could be poor preservation or a masked epitope.
• Match fixation and processing to your study samples wherever you can control it, or at least know it. With archive or biobank material, choose controls processed the same way, or note the fixation times.
• A tissue name is not a control. Check that the relevant cells are actually present in the block you cut. A block labelled "breast" may contain only stroma and fat, with no epithelium at all.
• Use internal controls where the section has them: cell populations expected to be positive or negative, assessed on every slide.
• Consider FFPE cell blocks of lines with known expression: cheap and renewable, but without tissue architecture, so they complement tissue rather than replace it.

Worked example. We stain colon (strong), breast (low), and liver and tonsil (both expected negative). Colon epithelium is strongly positive, as expected. But hepatocytes show moderate granular cytoplasmic staining, and tonsil germinal centres are clearly positive. What a twist!

How do you titrate an IHC antibody without fooling yourself?

Start from the supplier's or a published protocol and test three or four dilutions around it, plus two retrieval conditions if relevant (for example pH 6 and 9), on the full control set.

Then judge discrimination, not brightness: the best condition keeps the low expresser visible while the negative tissue stays clean. Excess antibody can raise off-target binding along with the real signal, and wastes reagent. If you can, score the slides with the labels covered.

Go easy on the counterstain, and always blue it. During antibody testing, use a light hematoxylin. A strong counterstain can mask weak DAB signal, which is exactly the low-positive staining you are trying to judge. Unblued nuclei stay reddish-purple and are harder to tell apart from brown; bluing turns them crisp blue and restores the contrast. For the full workflow, see My Bulletproof IHC Protocol for FFPE Tissue.

Worked example. At the datasheet dilution (1:500), liver staining is moderate. At 1:2000, colon is still strong, breast is weak but clear, and liver is nearly negative. The tonsil germinal centres, however, stay clearly positive at every dilution. Titration solved one problem, most likely concentration-related background. The other needs an explanation.

What do IHC controls prove, and what don't they?

The most common mistake is treating a clean background control as evidence that the antibody binds the right protein.

Run a no-primary control whenever you set up an assay, and add a matched isotype control for monoclonals in immune-rich tissue. Neither replaces informative tissue controls (Torlakovic et al., 2014). And never present peptide blocking alone as proof of specificity.

Unexpected staining: how do you interpret it?

Knockout material and RNA in situ hybridisation give strong evidence but cost time and money most labs do not have. Start with the cheap steps:

1. Identify the stained cells. Get out your histology book: could the target plausibly be in those cell types?
2. Dig into the databases. Check HPA's RNA data, including single-cell data, for those cell types. It is free and independent of any antibody.
3. Test an independent antibody against a different, preferably non-overlapping epitope. Agreement supports the pattern. Disagreement needs investigating: cross-reactivity, sensitivity, epitope accessibility or isoform recognition could explain it. A different host alone does not make an antibody independent.
4. Use knockout or knockdown material if you have it, and check what the edit actually removes. Staining in a knockout can come from a truncated protein that keeps the epitope.

Let your conclusion match the evidence: "staining consistent with Protein X" and "Protein X localises to" are different claims. Stop and investigate if you see:

• Signal in verified negative tissue
• The wrong compartment, such as nuclear staining for a membrane protein
• Uniform staining, or signal tracking edges, necrosis or folds
• Staining in cell types where RNA data show little or no expression (though absent RNA does not prove absent protein)
• A pattern that changes with a new antibody lot
• Independent antibodies that disagree
• A mouse primary on mouse tissue: the anti-mouse secondary also detects endogenous IgG
• Brown pigment that is not DAB, such as melanin, haemosiderin or formalin pigment; check the no-primary slide

Worked example: the decision. RNA data show little Protein X transcript in lymphoid tissue. A second antibody, against a different epitope, stains colon and breast in the same pattern but leaves germinal centres negative. The epithelial pattern is convincing; the first antibody's germinal-centre signal is suspect, though a sensitivity difference could explain part of it.

Our choice: run the second antibody across the full control panel and, if it holds up, make it the study's primary choice. The first stays usable only for epithelial work at 1:2000, with the limitation recorded, since immune cells occur in most tissues and image analysis would count their signal.

Two last cautions. A singleplex-validated antibody may behave differently in a multiplex panel, after repeated retrieval and stripping. And sometimes every available antibody has some off-target binding. In situ proximity ligation assays (PLA) can help there: a signal forms only where two antibodies against different epitopes of the same protein bind close together, so off-target binding by one of them stays silent. This has been shown to improve specificity over single-antibody detection in tissue (Zieba et al., 2018). Conveniently, that may be the same independent antibody you already bought for validation.

Why keep your own antibody-validation TMA?

Everything above depends on staining the same informative tissues side by side, again and again. A small control TMA puts strong, low, negative and mixed tissue, plus cell-block cores, on one slide.

• Identical treatment: every control shares the same retrieval, incubation and detection.
• Fewer slides: comparing dilutions takes a handful of sections, not dozens.
• A fixed reference for every new lot and protocol change.
• Less tissue used: precious blocks are cored once, not sectioned for every test.

The limits: cores can miss heterogeneous expression, so confirm final conditions on whole sections, and cut slides close to use, as stored sections lose antigenicity.

In the next post: how to build your own validation TMA.

Validation costs a few extra slides and some patience, but far less than an experiment built on the wrong signal. Good luck at the bench, and may your negatives stay clean!

A one-page staining evaluation sheet

Proportion categories are adapted from Allred scoring and verdicts from NordiQC. It is a recording tool, not a validated score: define the scored cell population first, and remember that strong positives cannot outweigh staining in verified negative material.

Run details

Scores per control

Record no-primary and isotype controls separately: background (0–3) and where it sits.

Scales

• Proportion of the scored population stained: 0 none; 1 under 1%; 2 1–10%; 3 11–33%; 4 34–66%; 5 67–100%.
• Intensity: 0 none; 1 weak; 2 moderate; 3 strong.
• Background: 0 none; 1 faint; 2 visible but readable; 3 interferes with reading.
• Independent-antibody concordance (different epitope, same tissue): concordant / partly / discordant / not tested.

Overall verdict

• Optimal: correct pattern in all controls, low expresser clearly shown, negative clean.
• Good: acceptable in all controls; protocol could still improve.
• Borderline: a false-negative or false-positive result in one control; optimise before use.
• Poor: fails in several controls, or unreadable.

Decision: use / use with stated limits / investigate further / re-optimise / reject.

Open questions and limits of use: \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ (for example: "germinal-centre staining probably off-target; not for lymphoid tissue or immune panels")

End of evaluation sheet.

Sources and further reading

• Richardson R. At least 15 companies are selling antibodies using faked validation data (2026)
• The Scientist. Misleading antibody validation images controversy expands to 15 vendors (2026)
• Uhlén M et al. A proposal for validation of antibodies. Nature Methods (2016)
• Bordeaux J et al. Antibody validation. BioTechniques 48(3):197–209 (2010)
• Howat WJ et al. Antibody validation of immunohistochemistry for biomarker discovery. Methods (2014)
• Fowler CB et al. Antigen retrieval causes protein unfolding: evidence for a linear epitope model of recovered immunoreactivity. J Histochem Cytochem (2011)
• Holmseth S et al. Specificity controls for immunocytochemistry: the antigen preadsorption test can lead to inaccurate assessment of antibody specificity. J Histochem Cytochem (2012)
• Torlakovic EE et al. Standardization of negative controls in diagnostic immunohistochemistry. Appl Immunohistochem Mol Morphol (2014)
• Zieba A, Pontén F, Uhlén M, Landegren U. In situ protein detection with enhanced specificity using DNA-conjugated antibodies and proximity ligation. Modern Pathology (2018)
• Harvey JM et al. Estrogen receptor status by immunohistochemistry is superior to the ligand-binding assay for predicting response to adjuvant endocrine therapy in breast cancer. J Clin Oncol (1999)
• NordiQC assessment criteria
• Human Protein Atlas · YCharOS