A Validated Test Method Was Withdrawn Because the Tissue Left the Market

On 25 June 2025 the OECD published 56 updates to its Test Guideline Programme. Most were the sort of technical housekeeping that only a validation scientist reads. Two of them mattered more than the rest. Test No. 431 for in vitro skin corrosion and Test No. 439 for in vitro skin irritation were both amended to remove a reconstructed human epidermis model — because that model is no longer commercially available.

Read that again as a market event rather than a scientific one. A test method that had passed validation, that regulators in 38 economies accept as a substitute for a rabbit study, and that sits inside a formal OECD guideline, stopped being usable because a supplier stopped making the tissue it runs on. The science did not fail. The supply chain did.

That is the actual structure of the in vitro toxicology market, and it is why the growth numbers in this sector are so widely misread.

A validated method is three products, not one

What a contract research organisation sells a regulated customer is not an assay. It is a dossier that a regulator will accept, and an accepted dossier rests on three separate things being simultaneously true: a biological model exists and is reproducible, a protocol exists that has been run against reference chemicals and produced the expected answers, and a regulatory text exists that names the method as acceptable.

The third part is where the money is. OECD Mutual Acceptance of Data means a study conducted to an adopted Test Guideline in one of the 38 member countries is accepted by the others. That reciprocity is the whole commercial engine of the sector: it converts a validated method into a product that can be sold once and used in every filing. A method that is not in a guideline, or inside an adopted defined approach, is not worthless — the US Food and Drug Administration will accept non-guideline data as supporting evidence — but it cannot carry a dossier on its own. It is evidence, not compliance.

Which is why the update to Test No. 497 on defined approaches for skin sensitisation, in the same June 2025 package, is quietly more consequential than any instrument launch this year. It allows the in vitro and in chemico methods in Test Nos. 442C, 442D and 442E to be used as alternate sources of information, and adds a defined approach for determining a point of departure. That is the regulatory text moving: a defined approach lets a fixed combination of non-animal methods produce a classification without animal data at all. Every time one is adopted, a slice of future demand is created and a slice of animal-testing demand is deleted.

The demand signal came from the FDA, not from the market

On 10 April 2025 the FDA published its Roadmap to Reducing Animal Testing in Preclinical Safety Studies, with a stated three-year implementation window. It is unusually specific about mechanism, and three of its provisions matter to anyone selling into this space.

The first is that the agency will accept organoid or in silico results inside investigational and licensing packages as supportive data, and will offer regulatory relief — fewer animal study replicates — to sponsors who submit them. That is a statement about volume: the incentive on offer is the deletion of animal work, not the purchase of an equal quantity of something else.

The second is a pilot waiver. Where a monoclonal antibody targets a human-specific receptor and the only available animal model is a transgenic mouse, FDA suggests a sponsor could substitute a battery of human in vitro tests, or microphysiological systems plus a physiologically based pharmacokinetic model, in place of that study. Again: a multi-year, multi-million-dollar animal study is being replaced by a battery of cell-based assays and a simulation.

The third is the admission that the evidence base itself is missing. The roadmap states plainly that no comprehensive database of animal and human toxicity data currently exists, and proposes to build an open-access repository by expanding the Tox21 programme and combining international datasets. That is a public-good investment in the exact asset — reference data — on which private validation businesses charge a premium.

What the substitution is actually worth

The mistake is to model this as a market that grows with drug discovery budgets. It grows with endpoint substitution, and the arithmetic of substitution runs against revenue per decision.

Pheonix Research’s own forecast for the platform layer of this shift, organ-on-chips, runs from USD 0.78 billion in 2025 to USD 3.25 billion in 2033 at a stated 19.50 per cent — the endpoints compound at 19.53 per cent, so the headline holds. Liver-on-chip is the largest product segment and multi-organ platforms are the fastest growing. That 19.5 per cent is best read as a substitution rate rather than a category growth rate: it is what a displacement technology grows at while it takes share from an incumbent method.

The displacement is deflationary in the service layer and inflationary in the platform layer. High-content screening reads several endpoints off one plate; multiplexed panels put more answers into a single run; automated imaging removes technician hours. Price per data point falls. What rises is price per accepted answer, because the scarce item is no longer measurement — it is validation and documentation.

That is why the value in this sector is migrating away from the bench. Two businesses are being confused with each other: the laboratory that runs assays, which grows with sample volume and prices per test, and the supplier of a validated platform plus its reference dataset, which prices per method and per licence. Only the second one gets to compound at twenty per cent.

The lesson in a deleted model

The removal of a tissue model from Test No. 439 is not a curiosity. Reconstructed tissue is a living biological product with a short shelf life, a manufacturing site and a commercial rationale that can change. A guideline that names one model has a single point of failure, and a toxicology programme built on one model has the same.

The practical implication for anyone buying or building in this market is unglamorous and worth stating. Validation status is an asset with a counterparty, not a permanent property of a method. Ask what happens to the dossier if a model is withdrawn. Ask who holds the reference data. Ask whether the method is in a guideline, inside an adopted defined approach, or merely published.

We do not publish a market size for “in vitro toxicology testing” here, and that is deliberate. The perimeter collapses three unrelated purchases into one number: reagents and consumables bought per well, instruments bought per laboratory, and studies bought per dossier. They have different buyers, different gross margins and different growth rates, and adding them produces a total that no supplier can act on.

The number that matters is not how large the market is. It is how many endpoints have a validated non-animal method behind them, and how many of those methods have a second supplier.