Computer-predicted anti-aging drugs hit 30% in worms, led by retinoic acid

Finding drugs that slow aging is hard because the number of candidate compounds is effectively unlimited. Computational models that predict anti-aging potential from drug-induced gene expression, protein interactions and chemical structure could narrow the field, but they depend on reliable validation data. This study used the Caenorhabditis Intervention Testing Program (CITP), which standardizes methods and uses hierarchical statistics across labs, to test how well such predictions hold up.

The team started from the top 10% of a published prediction set (Fuentealba et al., 2019). They cross-referenced it with two other prediction efforts and the DrugAge database, and removed compounds already known to extend C. elegans lifespan. That left 16 compounds, each tested at 4–5 concentrations in C. elegans N2. Eleven (aldosterone, arecoline, bortezomib, dasatinib, decitabine, dexamethasone, erlotinib, everolimus, gefitinib, temsirolimus and thalidomide) did not change median lifespan or were toxic. Bortezomib was strongly toxic, with lifespan falling 10.5% to 47.3%. Five compounds extended lifespan: ritonavir, fisetin and berberine modestly (roughly 6–12%), and propranolol and atRA substantially. The hit rate was 31% for this set and about 30% (7/23) when previously tested compounds were included.

Propranolol extended median lifespan by 44% at 1 mM but was toxic at 5 mM. It also reduced growth of the bacterial food source. The beta-1 blocker metoprolol had no effect, and propranolol shortened lifespan on metabolically inert, paraformaldehyde-treated bacteria. These results suggest its benefit may be indirect and food-dependent, so the authors set it aside. They focused on atRA, which extended N2 median lifespan by 23.5% at 150 µM without affecting bacterial growth. Across three C. elegans strains (N2, JU775, MY16) and three labs, atRA extended lifespan by 18.8–44.4%. Variance from site (3.9%) and experimenter (7.1%) was small, so the effect was reproducible. The effect was species-specific. C. briggsae showed no benefit, and C. tropicalis showed at most a small, inconsistent one. On the Automated Lifespan Machine, C. elegans benefits held, while some non-elegans strains showed slight toxicity.

Per the abstract, genetic analysis showed that atRA's lifespan extension requires both AKT-1 and AKT-2. The canonical Akt target FOXO/DAF-16 was largely dispensable, but the Akt targets SKN-1 (Nrf2) and HSF-1, and the AMPK catalytic subunit AAK-2, were all necessary. This implicates a conserved metabolic and stress-response circuit rather than classic insulin/FOXO signaling. The authors argue that co-opting endogenous signaling molecules is a promising route to aging interventions.

Caveats: all results are in nematodes, and the mammalian translation is a prediction. Because the text provided was truncated, the healthspan (swimming) and detailed mutant data were not available to this summary, so the genetic results rely on the abstract. Effect sizes were also inconsistent in places, such as atRA at 23.5% in the screen versus 29% in the table. Retinoic acid effects in worms have been contradictory across earlier studies, and nematodes lack identified canonical retinoid receptors.