https://www.cell.com/cell-reports/fulltext/S2211-1247(26)00975-7
chatGPT5.6paid:
Overall assessment
This is a valuable methods-and-resource paper, but a less convincing drug-discovery paper. Its real novelty is the integrated screening infrastructure across five organisms, not definitive proof that it has discovered conserved anti-ageing drugs.
The individual drug results should be treated as candidates for replication. They do not yet justify human translation or demonstrate that the compounds slow ageing rather than alter mortality through antimicrobial, nutritional, toxicological or behavioural effects.
Paper: Perez et al., Cell Reports 45, 117897 (2026).
Summary
The authors construct a pipeline for testing pharmacological effects on lifespan at progressively greater biological complexity:
| Model | Platform | Main results |
|---|---|---|
| S. cerevisiae | Miniaturised 96-well chronological-lifespan assay using propidium iodide and flow cytometry | Screens hundreds of compounds and a separate 950-compound FDA-approved library. Finds candidate hits including baicalein, epiberberine, furazolidone and geldanamycin, but also shows that high-dose pH buffering and early toxicity followed by regrowth can create false lifespan extension. |
| C. elegans | 24-well, no-transfer imaging with FUdR and YOLOv8 detection of immobile/dead worms | Approximately 315 compounds tested with live and UV-killed OP50. Effects depend strongly on bacterial diet. Forskolin and taselisib work in both dietary conditions; several PI3K inhibitors are prominent hits. |
| D. melanogaster | Multi-fly vials imaged twice daily on flatbed scanners; 12 mL food plus ampicillin removes routine transfers | About 90 compounds in females and 101 in males. Forskolin, rifampicin, astaxanthin, piperine and taurine extend lifespan in both sexes; 4-PBA is reproduced. |
| Killifish | Drug-loaded pellets, with pellet content and tissue exposure checked by LC-MS/MS | Resveratrol extends female median lifespan by 18%. Baicalein extends male lifespan by 30.5%, while baicalein plus forskolin produces a 39.9% increase in males. |
| Mice | Late-life drug-containing chow in male C57BL/6J or HET3 mice, plus home-cage activity monitoring | Several nominal survival differences are reported, including rifampicin, 17α-estradiol, enalapril, benazepril and some combinations. Several cohorts were still substantially alive, so many results concern early or interim survival rather than completed lifespan. |
Within the main three high-throughput models:
- 423 distinct compounds were included in the integrated dataset.
- 130 were positive in at least one model.
- 32 were positive in at least two.
- Only metformin, rifampicin and furazolidone were classified as positive in all three.
- No compound was shown to significantly extend completed lifespan across all five species.
Target annotation suggested convergence on PI3K–AKT–mTOR, stress-response and proteostasis pathways, although this analysis is descriptive rather than mechanistic.
What is genuinely novel?
1. The integrated operational pipeline
The strongest innovation is bringing yeast, worms, flies, killifish and mice into one coordinated programme with increasing cost and complexity. This could be a useful “funnel” in which inexpensive models narrow the candidates entering vertebrate testing.
It is probably the first lifespan-screening campaign to combine all five models at this scale within one programme.
2. Useful technical improvements
The component technologies are mostly developments of existing methods:
- Flow-cytometric yeast chronological-lifespan measurement already existed; the advance is its miniaturisation, automation and explicit handling of evaporation, pH and regrowth artifacts.
- Automated worm survival was previously demonstrated by the Lifespan Machine and WormBot. The new contribution is the compact 24-well microscope format and YOLO-based analysis.
- Flatbed-scanner fly monitoring was previously developed in DIAMonDS. The important improvement here is transfer-free whole-lifespan measurement in multi-fly vials rather than individual microplate wells.
- The standardised killifish pellet formulation, coupled to analytical verification, is particularly useful because chronic oral dosing has been a major practical obstacle.
3. Systematic identification of artifacts
Arguably the most scientifically useful result is not a drug hit but the demonstration that apparent lifespan extension can result from:
- extracellular pH buffering in yeast;
- early toxicity and delayed culture outgrowth;
- drug–bacteria interactions in worms;
- sex-specific responses in flies and fish;
- altered food intake, weight or activity in mice;
- inaccurate or unstable exposure in formulated diets.
4. The pharmacological results are less novel
A substantial part of the worm dataset, and some fly findings, extend the team’s earlier study, which had already screened more than 200 compounds and examined bacterial-diet effects and cross-species validation. See Phelps et al., 2025. The present paper adds automation, more compounds and vertebrate integration, rather than providing an entirely new drug dataset.
Strengths
- It measures actual survival rather than relying solely on ageing biomarkers.
- Negative results are reported, including failure of rapamycin and metformin in killifish and metformin and resveratrol in mice.
- Controls appear reasonably stable over time.
- Machine-learning train, validation and test sets were separated by experiment, reducing leakage.
- Both live and killed bacterial diets were used in worms.
- Both fly sexes and both killifish sexes were considered, even though compounds were not always tested in both.
- Vertebrate pellet content and at least limited systemic exposure were verified analytically.
- The authors explicitly investigate several artifacts rather than treating every positive survival curve as geroprotection.
Major weaknesses
1. Multiple testing is not properly addressed
Hundreds of compounds, multiple concentrations, two worm diets and two fly sexes generate thousands of opportunities for nominal significance. The statistical section uses (p<0.05), but reports no systematic false-discovery correction or pre-specified hit-selection procedure.
Across the 22 displayed mouse intervention arms, most positive results have (p)-values between 0.038 and 0.05; rifampicin is strongest at (p=0.005). None of the positive mouse findings would survive a simple experiment-wide Bonferroni correction. Cross-species replication reduces this problem somewhat, but testing was selective and uneven.
2. Experimental-unit and clustering problems
Animals share wells, vials, tanks or cages:
- worms: roughly 15 per well, only three wells per condition;
- flies: 15 per vial, three vials per condition;
- killifish: groups of five related fish per tank;
- mice: approximately five per cage with dietary treatment delivered by cage.
The analysis uses ordinary individual-level log-rank tests, with no reported frailty model, cluster-robust variance or treatment of vial/tank/cage as the experimental unit. This can overstate precision.
The yeast analysis is more concerning: the authors arbitrarily set (N=50) for survival analysis “to match other models,” even though the measurement is a repeated population viability proportion from a culture well. Statistical significance should be based on independent cultures or plates, not a synthetic sample size.
3. “Cross-species” is not a uniform comparative experiment
The same library was not screened at matched exposures in all five models:
- 347 compounds were assessed in yeast, 315 in worms and only 101 in flies.
- Only around 20 sex-specific intervention conditions were examined in killifish.
- Approximately 22 mouse intervention arms were distributed across seven batches, two genotypes and different starting ages.
- Vertebrate treatments were usually tested at one dose.
- Compounds entering fish and mouse experiments were selectively chosen from prior results and literature.
Consequently, the overlap diagram is useful for prioritisation but cannot estimate the general probability that an invertebrate hit will translate to vertebrates.
4. Several mouse “lifespan” results are interim-survival results
For multiple mouse cohorts, more than half of the animals remain alive at the end of the displayed curve. Supplementary Table 2 therefore reports survival at 75% of the control lifespan rather than median or complete lifespan.
Thus enalapril, benazepril, rifampicin, 17α-estradiol and several combinations may have reduced early post-treatment mortality, but full lifespan extension, late-life hazard and maximum lifespan remain unresolved.
This is much weaker than the multisite, both-sex, completed-survival standard established by the NIA Interventions Testing Program.
5. An internal inconsistency weakens the three-species headline
Metformin is counted as one of only three compounds positive in yeast, worms and flies. Yet the authors themselves show that its yeast effect appears only at 50 mM and is associated with extracellular pH buffering—precisely the artifact they warn against.
Likewise, rifampicin’s worm benefit is strongly dependent on live bacteria in the team’s earlier work, suggesting that at least part of its apparent conservation reflects antimicrobial effects rather than conserved host ageing biology.
6. Machine-learning accuracy is adequate for screening, not definitive validation
Reported performance is modest:
- worms: about 80% precision and 70% recall;
- flies: about 70% precision and 61% recall.
Movement is used as the primary indication of life. Drugs that cause paralysis, sedation or stimulation could therefore produce treatment-dependent classification errors. Agreement with manual scoring is encouraging, but there is no independent-laboratory validation or validation specifically across behavioural drug classes.
The authors share code and weights, but the annotated training dataset is available only on request, preventing complete independent retraining from the public materials. See the project repository.
7. Pathway convergence is suggestive, not demonstrated
The drug–target network is based on supplier, DrugBank and literature annotations. There are no genetic epistasis experiments for most new hits, target-engagement measurements or formal enrichment analysis against the composition of the screened library.
Because the library intentionally contains many PI3K and mTOR inhibitors, recovery of PI3K–AKT–mTOR targets is partly expected. It does not establish that this pathway caused the lifespan effects.
8. Lifespan is not necessarily slowed ageing
Antibiotics, pesticides and oncology drugs could extend survival by suppressing infection, altering food organisms, reducing reproduction, slowing growth or selecting resistant subpopulations. Mouse activity changes could represent stimulant or sedative pharmacology rather than rejuvenation.
Healthspan assessment is limited: body weight, food/water intake, cage-level activity and selected blood counts do not establish reduced frailty, preserved organ function or compressed morbidity.
Most interesting candidates
- Baicalein plus forskolin: The 39.9% male-killifish median extension, (p=0.0010), is probably the strongest new vertebrate signal. However, it comes from one strain, sex, dose and batch, without a formal statistical interaction comparison against both single agents.
- Forskolin: The most coherent prioritisation signal across worms and both fly sexes, with favourable but non-significant trends in killifish and mice. It deserves replicated vertebrate dose-response studies.
- Baicalein: Strong in yeast and male killifish, but the fly result is marginal and there are no mouse data.
- Rifampicin: Interesting mouse early-survival signal, but incomplete mouse follow-up and antimicrobial confounding make it difficult to interpret as an anti-ageing effect.
- Taselisib/ZSTK474 and related PI3K inhibitors: Strong discovery hits in worms, but oncology-class toxicity and absence of vertebrate lifespan validation substantially limit translational relevance.
Bottom line
The paper establishes a credible and potentially important screening infrastructure. It demonstrates that context and assay artifacts dominate many reported longevity effects, which may be its most durable contribution.
It does not yet establish a new conserved geroprotector. The next convincing step would be prospectively pre-specified replication of a small number of candidates—especially forskolin and baicalein plus forskolin—using multiple doses, both sexes, independent laboratories, cluster-aware statistics, completed survival and substantive healthspan/pathology endpoints.