THE CATEGORY
“Undoing glycation” is not one problem. Depending on which chemical species is targeted, it can mean four different things — and most writing in this field collapses at least two of them. This reference page separates the four kinds and attaches the primary literature to each, so that any claim about reversing glycation can be read precisely.
Kind one · Precursor detoxification
Reactive dicarbonyls such as methylglyoxal are neutralised before they ever modify a protein. The adduct never forms. The human glyoxalase system — glyoxalase-1 working with glutathione — does this continuously, and its impairment under inflammation is documented in age-related osteoarthritis (Trellu et al., 2019). Detoxifying a precursor is not removing an adduct: nothing already bound to a protein is touched.
Kind two · Early-adduct deglycation
The first stable products of glycation — fructosamines (Amadori products) on lysine residues — can still be enzymatically removed. Human fructosamine-3-kinase (FN3K) does this endogenously, and its structural mechanism is published (Lokhandwala et al., 2024). A combined FN3K and fructosyl-amino-acid-oxidase treatment reduced autofluorescence in glycated ex vivo human skin (De Decker et al., 2023), and a fructosyl amino oxidase has been proposed as a therapeutic enzyme in age-related macular degeneration (Delanghe et al., 2024). Early adducts are reversible; the chemistry further downstream is a different problem.
Kind three · Stable-adduct removal
Further downstream sit stable adducts such as Nε-carboxymethyl-lysine (CML), historically considered irreversible. This is the kind that had no published enzymatic answer until 2026: CMLase (CrGO-897), engineered from a glycine-oxidase scaffold, oxidises the CML side chain and restores the native lysine residue. The demonstration is in vitro and in prepared human tissue samples — lens proteins, arterial tissue and skin — with no living-organism study performed (Trabosh et al., 2026). The CMLase result belongs to this kind, and stating that precisely is what separates the finding from the headlines around it.
Kind four · Crosslink breaking
The stiffest glycation damage is crosslinks — glucosepane and related structures that tie protein molecules together and drive the mechanical decline of collagen tissues (Eekhoff et al., 2018; Giannetti et al., 2026). No enzyme published to date breaks these crosslinks, and CMLase is not reported to break glucosepane cross-links. Kind three and kind four are different chemistry; a result in one says nothing about the other.
| Kind | Target species | Endogenous route | Published enzymatic route | Primary sources |
|---|---|---|---|---|
| 1 · Precursor detoxification | Reactive dicarbonyls (methylglyoxal) | Glyoxalase-1 system | Not applicable — prevention, not removal | PMID 30635030 |
| 2 · Early-adduct deglycation | Fructosamines (Amadori products) | Fructosamine-3-kinase (FN3K) | FN3K + FAOD combination, ex vivo human skin | PMIDs 39173621, 37240327, 38732004 |
| 3 · Stable-adduct removal | Stable adducts (CML) | None known | CMLase (CrGO-897), in vitro + ex vivo human tissue | PMID 42448719 |
| 4 · Crosslink breaking | Crosslinks (glucosepane) | None known | None published | PMIDs 42334941, 29873266 |
The table is a taxonomy of the field, not a claim about any single programme. Prevention (kind one) is not removal (kinds two and three), and neither touches crosslinks (kind four).