A silibinin-standardised extract was dosed at 0.05% and 0.2% onto human skin explants under methylglyoxal and UV stress. CML fell by 73.89% and 82.60%. This article sets out what those two numbers support, what they do not, and how to formulate so that they still apply in a finished emulsion.
At a glance
INCI | Silymarin |
Active | Silibinin ≥80% by HPLC; measured 83.3% on the reference lot (silibinin A 32.6%, silibinin B 50.7%) |
Product ID | PMKZ3101 (PMAnti-AGEs® Silibinin, 80 grade) |
Model | Ex vivo human skin explants; methylglyoxal 2 mM plus UVA 30 J/cm² and UVB 50 mJ/cm², four cycles over four days |
CML inhibition | 73.89% at 0.05%; 82.60% at 0.2% (P < 0.01 versus the glycated control, n = 3) |
Assay control | Aminoguanidine sulfate 3 mM: 80.31% |
Matrix proteins | 11 of 11 increased at both doses (P < 0.01); collagen XVII +355.56% at 0.2% |
Use level | 0.05–0.2% carries the data; the product reference range is 0.1–0.5% |
Solubility | 1.0–1.1 g/100 g in propanediol or butanediol; insoluble in water and in oils |
Table 1. Summary of the ingredient and the study behind the figures in this article.
What glycation does to dermal protein — and why CML is not “AGEs”
Short answer
Glycation is a non-enzymatic reaction in which reducing sugars and reactive dicarbonyls bind to lysine and arginine residues on dermal proteins, forming crosslinks that stiffen the extracellular matrix. Its end products are a family, not a single molecule. CML — Nε-carboxymethyl-lysine — is one member, and the one most often quantified.
Glucose reacts slowly. Dicarbonyls such as methylglyoxal and glyoxal react orders of magnitude faster, which is why almost every controlled glycation experiment uses one of them as the challenge rather than sugar alone. In dermal tissue the consequence is mechanical before it is visible: collagen fibres crosslink, the elastic network loses recoil, and the dermis stiffens from the inside.
The end products of that chemistry are not one substance. CML, pentosidine, pyrraline, argpyrimidine and the hydroimidazolones are all advanced glycation end products, formed by different routes from different precursors, and they behave differently in tissue. CML is the one that gets measured most, for a practical reason: a validated antibody exists, so it can be stained and quantified in situ. Pentosidine can be read by its fluorescence instead, which is cheaper but less specific.
A product that lowers CML by 82.60% has shown something specific. It has not shown an 82.60% reduction in “AGEs”.
That distinction is not pedantry. It decides whether a claim survives review. “Reduces CML, one of the advanced glycation end products” is a statement about a measured endpoint. “Reduces AGEs” is a statement about a family of compounds, most of which were never assayed. The first is defensible with one report; the second needs a panel.
Glycation is also not oxidation, although the two are often sold together. An antioxidant assay — ORAC, DPPH, a FRAP value — measures radical scavenging and says nothing about whether a carbonyl has been prevented from attaching to a lysine. Plamed keeps the two as separate evidence streams, because a formulator who needs an anti-glycation claim cannot substitute an antioxidant number for it.
The model: glycated human skin explants, two doses, two controls
Short answer
The study used human skin explants rather than cultured cells. Tissue was challenged with 2 mM methylglyoxal together with UVA at 30 J/cm² and UVB at 50 mJ/cm², four cycles over four days, and dosed topically at 0.05% and 0.2%, with three replicates per group.
Plamed commissioned the work on PMAnti-AGEs® Silibinin from Shaanxi Biocell General Testing Technology, and the report — G4814-2604274 — is the single source for every figure in this article. Explants were maintained in six-well plates with 3.7 mL of medium at 37 °C and 5% CO₂, medium changed daily, and equilibrated for two days before treatment began.
Group | Treatment | What it is for |
|---|---|---|
Blank (BC) | Medium only, no challenge | Untreated baseline for CML and for the matrix proteins |
Negative control (NC) | Methylglyoxal 2 mM + UVA + UVB | The glycation challenge, and the denominator for every inhibition figure |
PC1 | Vitamin C 100 µg/mL + vitamin E 7 µg/mL + challenge | Positive control for the eleven matrix proteins |
PC2 | Aminoguanidine sulfate 3 mM + challenge | Positive control for CML |
Sample 0.05% | Topical, plus challenge | Low dose |
Sample 0.2% | Topical, plus challenge | High dose |
Table 2. Test groups. Two positive controls, one per endpoint family, are the reason the results can be interpreted at all.
Each cycle ran the same way: the sample was applied to the surface, methylglyoxal was added to the medium, the tissue incubated for 24 hours, and then the explants were irradiated. Four cycles over four days, after which residual material was washed off, the tissue fixed in 4% paraformaldehyde for 24 hours, and sections stained. CML and the collagens were read by immunofluorescence at 200×; hyaluronic acid binding protein by biotin chromogenic staining. Intensity was expressed as integrated optical density per area and compared with a two-tailed t-test in GraphPad Prism.
Dose needs translating before it can be compared with anything. 0.05% and 0.2% are concentrations of the ingredient, not of silibinin: at ≥80% silibinin, they correspond to roughly 0.04% and 0.16% silibinin equivalents. Shin and colleagues, in the study that characterised silibinin against methylglyoxal-induced glycation in explants, used 1% of a milk thistle flower extract containing 0.1% silibinin. The high dose here therefore sits in the same range as the published explant work. Plamed publishes the conversion because a dose quoted without it cannot be compared with anything else.
What “n = 3” and a two-tailed t-test actually buy you
Three replicates per group is the minimum for this design, and it is what the published protocol used as well, so the two studies are comparable in structure. What it buys is a statement of the form “the difference from the glycated control is larger than the within-group variation in this tissue”. What it does not buy is a population inference, a rank order against an active tested in another laboratory, or a prediction about a different donor.
Explant response varies between donors, and a single study is a signal rather than a distribution. This is why Plamed reports the whole design — challenge, controls, replicates, statistics — next to the number, and why the number alone should not travel without them.
The result: 73.9% and 82.6% lower CML
Short answer
In the glycated control, CML rose to 6.55× the untreated baseline. At 0.05% the ingredient lowered it by 73.89%, and at 0.2% by 82.60%, both P < 0.01 against that control. Residual CML at 0.2% was 1.14× baseline, and the two doses differed significantly (P = 0.006).
Group | Relative IOD / area | Inhibition | P vs NC |
|---|---|---|---|
Blank (BC) | 1.00 ± 0.08 | — | — |
Negative control | 6.55 ± 0.31 | — | 0.000 vs BC |
PC2 · aminoguanidine 3 mM | 1.29 ± 0.18 | 80.31% | < 0.01 |
Sample 0.05% | 1.71 ± 0.15 | 73.89% | < 0.01 |
Sample 0.2% | 1.14 ± 0.12 | 82.60% | < 0.01 |
Table 3. CML by immunofluorescence, integrated optical density per area. Inhibition is calculated against the negative control, not against blank. Source: Biocell report G4814-2604274, table 26.
PMAnti-AGEs® Silibinin · glycated human skin explants, 4 days
CML by immunofluorescence, integrated optical density per area, as fold of the untreated control (dashed line = 1.0). Inhibition percentages are versus the glycated control. P < 0.01, n = 3.
Source: Shaanxi Biocell General Testing Technology, report G4814-2604274. Values are means; inhibition calculated against the glycated control. Figure prepared by Plamed.
Figure 1. CML in glycated human skin explants. The dashed line marks the untreated control at 1.0, so a bar below it would mean less CML than unchallenged tissue; the 0.2% dose returns CML to 1.14× that level. Source: Biocell report G4814-2604274, table 26.
Read the first two rows before the last three. The challenge multiplied CML by 6.55, which is what makes the inhibition percentages meaningful; without that rise the same percentages would be arithmetic on noise. Then read the two sample rows against the control row, not against each other: 73.89% and 82.60% are both reductions from a six-fold elevated baseline.
The gap between the doses is 8.71 percentage points, and the report gives a P-value of 0.006 for the comparison between them. That supports “dose-dependent within 0.05–0.2%”, and nothing beyond that range. There is no data above 0.2%, so a higher use level is not an extrapolation this report authorises.
Why 1.14× baseline matters more than 82.60%
A percentage moves with its denominator. 82.60% of a six-fold elevation still leaves a residue, and the useful question for a formulator is where that residue sits relative to untreated tissue. Here it sits at 1.14× — close to the blank, versus 6.55× with no treatment. That is the sentence Plamed puts in front of an R&D reviewer, because it describes a position rather than a ratio.
It is not zero, and the design does not allow “removal”. The tissue was dosed while glycation was being induced, not after it had accumulated. Four days of simultaneous protection is a different claim from clearing what is already crosslinked, and only the first one is on the table.
Why the control group decides whether the number means anything
Short answer
A percentage is only informative if the assay could have detected an effect of that size. Aminoguanidine sulfate at 3 mM, run as the positive control, inhibited CML by 80.31% — which is what makes 73.89% and 82.60% interpretable, rather than impressive in isolation.
Assay validity has two halves, and reports often show only one. The challenge has to work: the negative control must differ from blank, and here it does, at P < 0.01. The method has to be able to see an effect of the size being claimed: a known inhibitor, run in the same plate on the same tissue, must move the endpoint. Without the second half, a null result and an insensitive assay look identical.
Aminoguanidine is a reference glycation inhibitor from the pharmacology literature. It is not a cosmetic ingredient and Plamed does not position it as one. It appears in the design for one reason: a method that cannot detect a known inhibitor cannot be trusted to detect a new one. The 80.31% is evidence about the assay, not about the market.
A number without a control is a number without a denominator.
That is also why the 80.31% is not a benchmark to beat. Two figures produced in the same assay, on the same tissue, at different concentrations, are not a competition; a superiority claim would need a head-to-head dose-response for both, and this study is not that design. Plamed reports the control because withholding it would make the sample figures unreadable, and makes no comparative claim from it.
The same samples: eleven matrix proteins up at both doses
Short answer
All eleven extracellular matrix proteins measured increased at both doses (P < 0.01). At 0.2% and 0.05% respectively: collagen XVII 355.56% and 266.67%, collagen III 340.00% and 235.00%, fibrillin-1 307.69% and 200.00%, and hyaluronic acid binding protein 317.39% and 221.74%.
Marker | 0.05% vs NC | 0.2% vs NC | Compartment |
|---|---|---|---|
Collagen XVII | +266.67% | +355.56% | Hemidesmosome / basement membrane |
Collagen III | +235.00% | +340.00% | Interstitial dermis |
HABP | +221.74% | +317.39% | Hyaluronic acid binding protein |
Fibrillin-1 | +200.00% | +307.69% | Microfibrils at the DEJ |
Collagen I | +168.18% | +290.91% | Interstitial dermis |
Collagen VI | +200.00% | +268.42% | Microfibrillar network |
Collagen IV | +131.58% | +257.89% | Basement membrane |
Elastin | +154.55% | +230.30% | Elastic fibre |
Collagen XV | +130.00% | +196.67% | Multiplexin, basement membrane zone |
Collagen V | +63.89% | +141.67% | Interstitial dermis |
Collagen VII | +47.37% | +134.21% | Anchoring fibrils |
Table 4. Eleven matrix proteins, immunofluorescence except HABP (biotin chromogenic). All comparisons P < 0.01 versus the negative control. Source: Biocell report G4814-2604274.
Two features of the PMAnti-AGEs® Silibinin panel matter more than the ranking. The first is coverage: basement membrane (IV, XVII), interstitial collagens (I, III, V), anchoring fibrils (VII), the microfibrillar collagens (VI, XV), elastin, fibrillin-1 and hyaluronic acid binding protein were all read in the same sections. A single-marker panel can be moved by a single artefact; eleven moving together is harder to explain away. Plamed reports all eleven rather than the three that moved most, because a panel that shows only its best markers is not a panel.
The second is the pairing with CML. Fibrillin-1 is unusually sensitive to glycation — Shin and colleagues showed that methylglyoxal reduces fibrillin-1 at the dermal–epidermal junction while CML rises, and that the two run in opposite directions. This study reproduced that opposition: CML down 82.60%, fibrillin-1 up 307.69%, both at 0.2%, both in the same tissue. That is a mechanistic consistency check, not a marketing point, and it is the strongest single reason to trust the panel.
What the panel is not: immunofluorescence measures protein present in the section, not the rate at which new protein is being made, and not the mechanical property of the tissue. A 355.56% increase in collagen XVII staining is not a 355.56% increase in firmness, and it should never be written as one.
What the data does not prove
Short answer
It does not show that the ingredient removes glycation that already exists, that it lowers AGEs other than CML, or that a finished emulsion behaves the way the test system did. Dosing began with induction, ran four days, and stayed ex vivo; human in-vivo confirmation is not part of this set.
Plamed limits its claim for this ingredient to the four boundaries below.
Not removal. Protection was concurrent with the challenge. A study that glycates first and applies the ingredient afterwards answers a different question, and this one does not attempt it.
Not every AGE. One marker was measured. Pentosidine, argpyrimidine, the hydroimidazolones and total AGE fluorescence were not. Any sentence that generalises from CML to the family is unsupported by this report.
Not a finished formula. The test article was the raw material applied to the surface. Whether 0.2% of it survives emulsification, reaches the viable epidermis, and holds its concentration over a shelf life is a formulation question, addressed in the next section but not answered by this study.
Not human. Explanted tissue is closer to skin than a monolayer and further from a person than a patch test. At the time of writing there is no in-vivo data set for this ingredient, and Plamed does not imply one.
Formulating so the data still applies
Short answer
Dose at 0.05–0.2% of the ingredient, pre-disperse it in propanediol or butanediol at 1:20 or leaner, and add it after emulsification. It will precipitate in any aqueous base regardless of pH, and colour is stable only up to pH 8.
Solvent | Solubility (g/100 g) | Practical note |
|---|---|---|
Water | Insoluble | Do not build an aqueous system around it |
1,3-Propanediol | 1.0 – 1.1 | Preferred; sonicate over 8 h or warm to dissolve |
1,3-Butanediol | 1.0 – 1.1 | Preferred; interchangeable with propanediol in practice |
1,2-Pentanediol | 0.3 – 0.4 | Partial; not a primary solvent |
1,2-Hexanediol | 0.1 – 0.2 | Slow and limited; not recommended as primary |
GTCC (caprylic/capric triglyceride) | < 0.01 | Not for oil-only systems |
Table 5. Solubility in cosmetic solvents at room temperature with sonication. Data from Plamed’s internal solubility screen.
The number that changes practice for PMAnti-AGEs® Silibinin is not the 1.0 g/100 g figure but the dispersion ratio. In Plamed’s application work, a 1:20 product-to-glycol ratio dispersed evenly with no short-term precipitate, while 1:10 precipitated within minutes; 1,2-pentanediol precipitated even at 1:40. So Plamed’s handling instruction is a ratio, not a percentage: pre-disperse at 1:20 or leaner, do not aim for a clear solution, and add the slurry after the emulsion has formed.
Disperse the powder in propanediol or butanediol at 1:20 (product : glycol). Warm to 40–45 °C and stir for about 30 minutes if a smoother slurry is needed; complete dissolution is not required.
Complete the emulsification first. Add the dispersion at the cool-down stage with stirring, then homogenise briefly to distribute the particles.
Check the final pH. Keep it at or below 8, and keep alkaline raw materials out of direct contact with the pre-dispersion.
Where it will fail: water, oil, and pH 9
Clear aqueous systems. A gel, a toner or a transparent serum will show precipitate regardless of pH, because the event is a solubility limit and not a pH reaction. This is the single most common formulation error with silymarin, and it is not fixable with a chelator or a solubiliser swap at the same use level.
Pure oil systems. Silicone, synthetic esters, vegetable oils and GTCC all gave cloudy dispersions in Plamed’s tests, and no photographs were retained because the result was uniformly negative. An anhydrous balm is not a candidate.
pH 9 and above. Colour is stable from pH 3 to 8 and darkens visibly above that, so the ingredient should not be put in direct contact with alkaline raw materials. For emulsions that drift upward over shelf life, buffer rather than rely on the initial reading.
For process temperature, Plamed’s guidance is to add at cool-down; finished-formula stability data above 45 °C is not yet available, and the ingredient should not be described as heat-stable on the strength of a raw-material assay. Antioxidant co-use — vitamin E or a sulfite — is the usual way to slow colour development in a system that also contains easily oxidised oils.
How to word the claim in your own file
Short answer
Put the concentration, the model, the duration, the control and the report number in one sentence. Under Regulation (EU) No 655/2013 a claim has to be supported by evidence that matches its wording, so the narrowest sentence is also the most defensible one.
Tempting wording | What the data supports | Why |
|---|---|---|
“Reduces AGEs” | “Reduced CML, one of the advanced glycation end products, by 82.60% at 0.2% in glycated human skin explants” | CML is one marker, not the family |
“Proven anti-glycation effect” | “Inhibited CML formation by 73.89% at 0.05% in ex vivo human skin over four days” | Ex vivo is not clinical; no human study exists |
“Boosts collagen 3.5×” | “Increased collagen XVII by 355.56% versus the glycated control at 0.2%” | Denominator is the challenged control, and the readout is staining intensity |
“Removes existing glycation” | “Inhibited CML formation during a four-day glycation challenge” | Dosing was concurrent with induction |
“Reduces wrinkles” | Not supported by this data set | No wrinkle endpoint was measured |
Table 6. Substitutions that keep the claim inside the evidence. The left column quotes wording that Plamed does not use, shown so that the replacement is explicit.
Plamed keeps the sentence and its evidence together, and supplies the report itself rather than a summary of it: request the full report from Plamed. In a product information file, the claim, the report number, the concentration and the model should sit in one block; if the sentence changes, the summary changes with it. The version of this that survives review is narrower than the version that reads best, and that is the normal trade.
The figures here come from a single production lot submitted as a representative sample. They are not an industry benchmark and not a specification minimum; treat them as this lot, this design, this laboratory.
Frequently asked
Is CML the same as AGEs?
No. CML — Nε-carboxymethyl-lysine — is one advanced glycation end product among several, and the one most often measured because a validated antibody makes it quantifiable in tissue. A CML result is a statement about CML.
Can I claim “reduces AGEs” on pack?
Not as written. “Reduces CML, one of the advanced glycation end products” is supportable by report G4814-2604274 at the stated concentrations. The unqualified version asserts an effect on compounds that were never assayed.
What use level should I start with?
0.05% to 0.2% is the range that carries data, and 0.2% is the level with the largest measured effect. The product reference range is 0.1–0.5%; above 0.2% the anti-glycation numbers are extrapolated, not measured.
Is Silymarin water soluble?
No. Solubility is 1.0–1.1 g/100 g in propanediol or butanediol and effectively zero in water. It is used as a dispersion in a glycol, added to an emulsified base.
Can it go into a clear serum or a gel?
Not at a use level that carries the data. Any aqueous system will show precipitate, and pH adjustment does not prevent it. Emulsion systems with a thickener are the workable route.
Is aminoguanidine a competing ingredient?
No. It is a reference inhibitor used to prove the assay can detect an effect. It is not a cosmetic raw material, and the 80.31% figure is evidence about the method, not a product comparison.
References
Shaanxi Biocell General Testing Technology Co., Ltd. Test report G4814-2604274: collagen and CML endpoints in a glycated human skin explant model. 2026.
Shin S, Lee J-A, Kim M, Kum H, Jung E, Park D. Anti-glycation activities of phenolic constituents from Silybum marianum (milk thistle) flower in vitro and on human explants. Molecules. 2015;20(3):3549–3564. doi:10.3390/molecules20033549
Commission Regulation (EU) No 655/2013 of 10 July 2013 laying down common criteria for the justification of claims used in relation to cosmetic products. EUR-Lex
Plamed Green Science Group. Technical data sheet: PMAnti-AGEs® Silibinin, 80 grade (PMKZ3101). 2026.