A molecule is pipetted straight onto cultured human fibroblasts. A collagen-related signal rises. Solid result. Give it a short journey through a product cycle and it may return wearing a larger claim: the ingredient has been shown to support collagen in human skin cells.
The experiment is legitimate. The claim is questionable.
In the dish, the cells received a known dose for a known length of time. The ingredient didn’t have to remain stable in a finished formula, leave that formula after application, cross the outer barrier or arrive at the intended tissue in sufficient quantity. It was escorted past the difficult part and placed beside its target.
This doesn’t make cell culture weak evidence. It makes it evidence for a narrower and often very important question: can this molecule alter this cell under these conditions? Problems begin when biological capability is presented as if formulation, delivery and an outcome in your skin came free with it.
This is the translation gap behind many peptide, growth-factor and exosome claims, and one reason the route through the barrier deserves as much attention as the pathway at the end of it. A cell result is neither empty nor complete. It needs to remain the size of the experiment that produced it.
The Target Has Already Been Reached
Cell culture is powerful partly because it removes context. Researchers can control the concentration, exposure time and surrounding medium, then compare treated and untreated cells without an entire tissue changing ten things at once. If a signalling pathway changes, the relationship can be isolated with a precision that would be difficult to recover from a person using a serum at home.
The context removed from the experiment is also part of the underlying biology. A fibroblast no longer sits beneath an epidermis, within its usual extracellular matrix, alongside immune cells, blood vessels and the mechanical forces of living tissue. A keratinocyte may be missing the neighbouring melanocytes, fibroblasts or immune signals that alter its response. Calling the cells human doesn’t restore the human skin around them.
For a dermal target, the missing journey is considerable. A molecule applied at the surface has to leave its vehicle, pass through or around the stratum corneum, cross the living epidermis and reach the dermis without being degraded, trapped or diluted below an effective concentration. Even an epidermal target is constrained by the formulation-release problem and at least part of the barrier. In a dish, the pipette has handled the commute.
The strongest conclusion from an isolated-cell experiment is therefore conditional: if this molecule reaches this cell, at this concentration and for this long, it can produce the measured response. This is evidence of biological capability. A topical product must also establish biological access.
In vitro shouldn’t be used as a polite synonym for poor evidence. It means that the experiment took place outside a living organism, and covers models with very different levels of structure. A well-designed cell study may be the cleanest way to establish a mechanism. It becomes poor support when it is asked to prove penetration, tissue exposure or a clinical change it never measured.
In Vitro Is Not One Model
Researchers can add context back in stages. Two cell types can be grown together. Keratinocytes can be organised into reconstructed human epidermis. Fibroblasts can sit beneath them in a fuller skin equivalent. Excised human skin can retain its native barrier and tissue arrangement for a limited experiment. Each model restores something. None restores everything, and they don’t form one tidy league table.
A model is validated for a defined method and endpoint, not awarded a general certificate for being sufficiently skin-like. The OECD’s 2025 Test Guideline 439 uses reconstructed human epidermis to identify irritant chemicals through loss of tissue viability. Cells convert the dye MTT into a coloured product, and a fall below the specified viability threshold supports an irritant classification. The method is built for that regulatory question.
The OECD’s skin-absorption guideline asks another question altogether. A skin sample separates donor and receptor chambers while researchers measure how much test substance moves into and through it over time. The same principle applies when interpreting an ingredient whose effect changes with concentration and tissue context: the model has to suit the claim. A system validated for irritation doesn’t automatically become a penetration model because both experiments contain something resembling skin.
More tissue can make an experiment more representative in one respect and less controllable in another. Excised human skin brings back the original permeability barrier, but not circulation, immune recruitment or long-term repair. A human trial restores ordinary use, yet the mechanism can become harder to isolate. The question isn’t which model sounds most advanced. It is whether the model can carry the conclusion placed on it.
The Model Was Skin-Like Until the Molecule Changed
Fritz Schmook, Josef Meingassner and Andreas Billich made that dependence unusually visible in 2001. They mounted human, pig and rat skin, along with two reconstructed human models, in diffusion cells and compared the movement of four topical drugs with different chemical properties: salicylic acid, hydrocortisone, clotrimazole and terbinafine.
One reconstructed model, Graftskin LSE, provided a reasonably similar barrier to salicylic acid. Change the molecule to the more hydrophobic clotrimazole and the resemblance collapsed. Its measured flux through Graftskin was about 900 times higher than through human skin, while the concentration retained in the model was about 50 times higher. In the other reconstructed epidermis, penetration of hydrophobic compounds could also exceed human skin by hundreds of times.
The model hadn’t become less human between assays. The function being tested had changed with the compound. A result that was passable for one permeability problem was badly misleading for another. Skin-like, without an endpoint attached, was doing too much.
The paper is twenty-five years old and reconstructed tissues have continued to develop, so its numerical errors shouldn’t be assigned to every model now in use. The principle has aged better than the models: performance must be shown for the particular method, material and question.
Sometimes the Missing Context Hides the Effect
If simplified models only exaggerated what ingredients could do, this would be an easier article. Sometimes the missing context contains the mechanism, and the effect disappears when the biology is taken apart.
Tomohiro Hakozaki and colleagues followed niacinamide through several levels of context in 2002. Niacinamide didn’t inhibit purified mushroom tyrosinase, an enzyme often used as a proxy for pigment production. It also didn’t reduce melanogenesis when melanocytes were cultured alone. On those experiments, a direct effect on pigment synthesis was absent.
When melanocytes and keratinocytes were grown together, the answer changed. Niacinamide inhibited transfer of pigment-containing melanosomes between the two cell types by 35 to 68 per cent. Pigmentation also fell in a reconstructed epidermis containing the relevant cellular relationship. The melanocyte hadn’t failed the ingredient. It had been asked the wrong question in an empty room.
The paper then moved into people. Eighteen participants with hyperpigmentation used a 5 per cent niacinamide moisturiser and its vehicle in a paired design. A second trial included 120 participants with facial tanning and compared vehicle, sunscreen and 2 per cent niacinamide with sunscreen. After four weeks, the niacinamide formulations improved measured pigmentation outcomes relative to vehicle. The clinical result could not, by itself, prove melanosome transfer was the route. The preceding models supplied that mechanistic account.
Several authors, including Hakozaki and researchers involved in the formulations, worked for Procter & Gamble. Commercial involvement is therefore central rather than incidental. The vehicle-controlled comparisons and progression across purified enzyme, isolated cells, co-culture, reconstructed tissue and human use give the paper real substance. They don’t turn it into independent replication or allow every modern niacinamide claim to borrow its authority.
The example runs opposite to the usual warning. Direct exposure can create false confidence by bypassing delivery. Isolation can also create false absence by removing a neighbouring cell, structural component or metabolic step that the response requires. Context can enlarge an opportunity. It can hide one too.
Sometimes the Missing Context Hides the Effect
When a product says an ingredient worked in human skin cells, start with the unglamorous questions. Which cells? Was the raw molecule tested or the finished formulation? At what concentration, for how long, and in what medium? Was the endpoint a change in gene expression, an amount of secreted protein, cell survival or something that resembles the promised result?
Then put the missing journey back. Can the formula preserve the ingredient and release it after application? Can the molecule cross the relevant part of the barrier? Could the target cells encounter anything close to the experimental concentration, and can the response survive the competing signals of living tissue? The longer the claim, the more handovers it contains.
A rise in collagen-related expression in fibroblasts isn’t more collagen in the dermis, and more collagen in the dermis isn’t automatically a visible change in a face. These may be connected steps. They remain separate steps until the evidence joins them.
So, does an ingredient that worked in cells work on skin? It may, and it’s a positive start to the investigation. The cell experiment establishes what can happen once the molecule reaches the target under the arranged conditions. The product still has to prove its worth in tissue.
Keep the cell result. Just don’t ask it to smuggle the rest of the evidence in behind it.
The experiment establishes the possibility. The formulation and the skin decide whether that possibility is made available.
References
Schmook, F.P., Meingassner, J.G. and Billich, A. (2001). Comparison of human skin or epidermis models with human and animal skin in in-vitro percutaneous absorption. International Journal of Pharmaceutics, 215(1-2), pp. 51-56. https://doi.org/10.1016/S0378-5173(00)00665-7
Hakozaki, T., Minwalla, L., Zhuang, J., et al. (2002). The effect of niacinamide on reducing cutaneous pigmentation and suppression of melanosome transfer. British Journal of Dermatology, 147(1), pp. 20-31. https://doi.org/10.1046/j.1365-2133.2002.04834.x
OECD (2025). Test No. 439: In Vitro Skin Irritation: Reconstructed Human Epidermis Test Method. OECD Guidelines for the Testing of Chemicals, Section 4. Paris: OECD Publishing. https://doi.org/10.1787/9789264242845-en
OECD (2004). Test No. 428: Skin Absorption: In Vitro Method. OECD Guidelines for the Testing of Chemicals, Section 4. Paris: OECD Publishing. https://doi.org/10.1787/9789264071087-en
This article contains no affiliate links, sponsored content or product recommendations. Scientific claims are grounded in the peer-reviewed literature and methodological standards cited above. The Schmook model comparison came from a commercial drug-development setting linked to Novartis; its conclusion was critical of the reconstructed models then available, and it tested defined drug solutions rather than finished cosmetics. The Hakozaki niacinamide paper included researchers from Procter & Gamble and the University of Cincinnati. Its industry origin and product-development aim deserve attention alongside its vehicle-controlled comparisons and progression from mechanistic models to human studies. Neither paper becomes worthless because industry was involved, and neither should be stretched beyond the compounds, models and endpoints it tested.
