For decades, aging research has operated on an assumed iron law: once advanced glycation end-products (AGEs) form in the body, they are irreversible — like rust permanently corroding elastic tissue. Revel Pharmaceuticals (San Francisco) and collaborators have rewritten that premise with an AI-engineered enzyme. Their enzyme, CMLase, cut CML levels in aortic tissue from a 75-year-old donor by about 70%, down to concentrations typical of a 30-year-old. Similar reductions were observed in skin and eye lens proteins from a 64-year-old donor. The paper was published in *Nature Communications*.
Why AGEs are called the body's "rust"
The golden crust of baked bread comes from sugars reacting with proteins under heat. The same reaction happens slowly inside the body at ~37°C, producing AGEs. Over decades, they stiffen elastic tissues and fuel chronic inflammation — classic hallmarks of cardiovascular disease, diabetes, and eye and kidney damage.
The extracellular matrix makes up roughly 70% of the body and turns over extremely slowly (whole-body collagen half-replacement takes about 15 years). The longer it persists, the more AGE-damaged it becomes, leading to sagging skin, weakened tendons, stiff joints, and organ decline. Existing drugs can only block *new* AGE formation; they cannot remove AGEs already embedded in tissue, let alone reverse underlying protein damage.
How AI found this "molecular lawnmower"
The team targeted N-carboxymethyllysine (CML), the most common AGE type — hard to clear, pro-inflammatory, and known to trigger cells to secrete factors that stiffen tissue and harm brain microglia. The challenge: no human enzyme exists as a template. So they turned to microbes — organisms that decompose AGE-modified proteins in decay may carry engineerable enzymes.
Approach: 1. Used AI to screen DNA from 50,000+ microbial species, modeling the structures of their encoded enzymes 2. Filtered for candidates able to penetrate large molecules like collagen and reach hidden CML residues 3. The best hit came from a thermophilic hydrothermal vent bacterium 4. The initial enzyme's activity was mediocre, so the team applied directed evolution (the Nobel-winning method that accelerates natural selection) over 5 rounds spanning 500+ million variants, yielding CMLase — over 10× faster than its precursor
On CML-modified collagen, retinal proteins, and hemoglobin, CMLase cleaved the chemical adducts and restored native conformations — effectively "repainting" damaged molecules.
Why this matters (chemistry × biology × AI)
1. From prevention to repair. It advances anti-aging from "blocking new damage" to "reversing old damage," directly challenging the foundational assumption that AGE damage is immutable — a rare pro-repair demonstration in aging biology.
2. A textbook AI4S methodology. AI sequence/structure screening (50,000 microbial genomes) → physics-informed modeling → a directed-evolution feedback loop, circumventing the dead end of "no template found in nature." This belongs to the same lineage as AlphaFold/RFdiffusion and the Baker lab's de novo enzyme design — but tested on real human tissue samples rather than mice, a harder and more valuable leap.
3. Part of the broader AI biomolecule-design wave. An April 2026 *Nature* review (Wei Yang et al.) judges that protein design's structural problems are nearly solved; the live question is "what to design." A *Food Research International* review the same year notes AI platforms predicting anti-aging peptide sequences from food protein libraries (targeting mTOR/Nrf2/IIS), with engineered microbes providing scalable synthesis. CMLase offers rare preclinical, human-tissue evidence along the chain: AI discovery/design → synthetic biomanufacturing → human validation.
Caveats
Chemical reversal is not the same as functional rejuvenation, and ex vivo tissue evidence is still far from repair inside living humans. Revel's CEO acknowledged "more work is needed." But overturning the premise of "irreversibility" has already reset the starting point for R&D in this field.
In one sentence: when an AI-designed enzyme can clip the rust out of the body, anti-aging finally has a candidate for *repair* — not just maintenance.
Sources: *Nature Communications* (Revel Pharmaceuticals et al., CMLase study); BIOSCIENCE/The Scientist coverage; April 2026 *Nature* protein design review; *Food Research International*, DOI 10.1016/j.foodres.2026.119189 (review on AI-discovered anti-aging peptides); SpaceDaily (overview of AI-designed novel proteins).