The 12 hallmarks of aging, and how each one becomes a drug target

The hallmarks of aging are the most useful map the field has produced. They turn a vague problem into twelve specific ones, and each of those twelve can be asked a concrete question: is there chemistry that touches it today? This is our working answer, target by target, from a programme that scores every candidate against all twelve.

López-Otín and colleagues set out nine hallmarks in 2013 and expanded them to twelve in 2023. The framework is now the common language of geroscience, and it is genuinely good for drug design because it forces you to name which part of ageing a molecule is supposed to touch.

The twelve, and where the chemistry is

Below is each hallmark with what it means in one line, how much usable small-molecule chemistry exists for it, and which of our targets sit against it. Chemistry ratings reflect how much published binder data is available to design from, not how important the hallmark is biologically.

Where the chemistry already exists

Five hallmarks have deep chemistry: nutrient sensing, epigenetic alterations, mitochondrial dysfunction, cellular senescence and chronic inflammation. These are where the well-known geroprotectors live. Rapamycin acts on nutrient sensing through mTOR. Metformin is associated with AMPK. The senolytic and senomorphic field targets senescence directly.

If you want to design a molecule with published binders to learn from, these are the hallmarks where that is possible. Our own target list is concentrated here for exactly that reason: a design engine needs reference chemistry to aim at.

Epigenetic alterations are the most active area right now

The strongest recent rejuvenation evidence sits on epigenetic enzymes rather than on the nutrient-sensing pathways that dominate the lifespan literature. METTL3 and KAT7 both have usable chemistry and both have perturbation data showing an effect on cellular age. KAT7 in particular came out of a 2021 rejuvenation study and has enough published binders to design against.

Where it does not, and why that is the opportunity

Three hallmarks are chemically sparse: telomere attrition, stem cell exhaustion and dysbiosis. Sparse does not mean unimportant. It means the tools available are mostly genetic or biological rather than small-molecule.

This is precisely where de novo design earns its place. A generative engine that assembles molecules through verified reactions does not need a library of known binders to produce a candidate. It needs a target and a set of building blocks.

How a candidate gets scored against twelve hallmarks

Every molecule we deliver carries a twelve-component hallmark profile. The value of that is not a single number, it is shape: two candidates with the same overall score can engage completely different parts of ageing, and a drug programme has to choose which one it wants.

The tissue axis matters as much as the hallmark axis. Senescence in adipose tissue is not senescence in the brain, and a molecule that cannot cross the blood-brain barrier cannot act on the second no matter what its hallmark profile says.

One measured caution about the framework

The twelve hallmarks are a taxonomy, not twelve independent dials. We measured how much independent variation they actually carry across our reference data, and the answer is smaller than the count suggests.

Two of the twelve were perfectly correlated in our data because they were constructed from the same compartment specification, so one was dropped before the analysis. Of the eleven that remained, the effective rank was about 2.9. That result survived a permutation control at roughly nineteen times the null floor, so the structure is real, but the framework compresses to fewer genuinely separate axes than its twelve labels imply.

Gems and de Magalhães made this argument from the biology in 2021: the hallmarks are a useful classification rather than a causal theory. Our number agrees with them from the data side. It does not make the map less useful, and it does mean a candidate that scores well on several hallmarks may be doing one thing rather than several.

Using this in practice

• ·Pick the hallmark before the molecule. A candidate is only as coherent as the mechanism it is aimed at, and "anti-ageing" is not a mechanism.
• ·Check the chemistry depth first. A hallmark with hundreds of published binders gives you a warm start and reliable predictions; one with almost none gives you novelty and wide error bars.
• ·Match the tissue to the disease. The hallmark tells you what goes wrong, the tissue tells you where, and only both together give you an indication.
• ·Treat multi-hallmark coverage carefully. Given the effective rank above, engaging four hallmarks may reflect one underlying axis rather than four wins.

Our target set is split into two programmes because lifespan evidence and rejuvenation evidence point at different genes, which is written up in lifespan versus rejuvenation targets. How a candidate is assembled with a working synthesis route is in synthesizability by construction, and the whole pipeline is documented in Methods.