https://www.nature.com/articles/s41586-026-10524-5
chatGPT:
Summary
The paper proposes “Sleep Chart”, an analysis linking self-reported sleep duration with 23 biological ageing clocks derived from MRI imaging, plasma proteomics and metabolomics. Its core finding is a U-shaped relationship: people reporting either short sleep or long sleep tended to have higher biological age gaps, while the lowest biological ageing signals clustered roughly between 6.4 and 7.8 hours, varying by organ system, omics layer and sex.
The study used UK Biobank sleep-duration data and ageing clocks across multiple organ systems. Out of 23 biological age gap measures, 9 showed statistically significant nonlinear associations with sleep duration. These included brain, pulmonary, hepatic, immune and skin proteomic clocks; endocrine metabolomic ageing; and MRI-based brain, adipose and pancreas ageing clocks. The paper also reports broader U-shaped patterns across 720 imaging-derived phenotypes, 342 organ-enriched plasma proteins and 107 organ-associated metabolites.
For disease outcomes, the authors divided sleep into short <6 h, normal 6–8 h, and long >8 h. Both short and long sleep were associated with higher all-cause mortality: short sleep had a hazard ratio of 1.50, and long sleep had a hazard ratio of 1.40, compared with 6–8 h sleep. Short sleep showed wider links with cardiovascular, metabolic, musculoskeletal, pulmonary, gastrointestinal and psychiatric disease outcomes, while long sleep showed a more focused relationship with brain-related and neuropsychiatric traits.
The paper also examines late-life depression. Its model suggests that short sleep may have a more direct association with late-life depression subtypes, whereas long sleep may be more indirectly associated through organ-specific biological ageing clocks, especially brain and adipose MRI-based ageing. The authors are careful that this mediation analysis is not definitive causal proof.
Novelty
The main novelty is the multi-organ, multi-omics scope. Previous work had linked sleep duration to ageing or brain-age measures, but this paper extends the analysis to biological ageing clocks across brain and body systems, using imaging, proteomics and metabolomics together. The paper explicitly frames itself as testing whether the known U-shaped sleep-ageing relationship generalizes beyond the brain and across molecular layers.
A second novel aspect is the distinction between short-sleep and long-sleep biology. The authors argue that short sleep appears more broadly and directly linked with systemic disease risk, while long sleep may often be a marker of, or mediated through, underlying ageing or disease processes. This is more nuanced than simply saying “7 hours is best”.
A third useful contribution is the integration of several evidence streams: generalized additive models for nonlinear sleep-ageing associations, GWAS/genetic correlations, Cox survival models, and structural equation models for depression pathways. That does not prove causality, but it gives a broader triangulated picture than a simple epidemiological association.
Critique
The biggest limitation is that sleep duration is self-reported. The authors acknowledge that questionnaire sleep is less objective than actigraphy or polysomnography and can capture a different aspect of sleep biology. This matters because reported sleep duration can be affected by mood, cognition, lifestyle, illness and recall bias.
The second major issue is causality. A U-shaped association does not show that too little or too much sleep accelerates ageing. Long sleep, in particular, may be a consequence or marker of subclinical disease, inflammation, depression, frailty, pain, medication use or low activity. The authors use Mendelian randomization and time-to-event models, but they still state that reverse causality and bidirectional effects cannot be fully excluded.
A third problem is that biological ageing clocks are not direct ageing mechanisms. A biological age gap can reflect inflammation, adiposity, illness, medication, organ pathology, acute stress or sampling variation, not necessarily the rate of fundamental ageing. This is especially relevant for proteomic and metabolomic clocks, which the authors note can fluctuate with illness, medication, diet and timing of sampling.
The paper’s “optimal” sleep windows should be treated cautiously. The reported minima differ by clock: for example, brain proteomic ageing had a minimum around 7.7–7.8 h, whereas brain MRI ageing had a minimum closer to 6.4–6.5 h. That variation may reflect real biology, but it may also reflect differences in cohorts, clock construction, measurement noise, statistical modelling or selection bias in the MRI subgroup.
There is also a UK Biobank generalizability problem. UK Biobank is healthier and more selected than the general population, and the authors note that analyses were mainly in people of European ancestry. This weakens claims about universal sleep optima across ethnic groups, cultures, socioeconomic conditions and age ranges.
Finally, the paper does not directly address key sleep dimensions such as sleep quality, fragmentation, circadian timing, insomnia phenotype, sleep apnoea severity, naps, shift work, chronotype or regularity. Duration alone is a crude measure. Someone sleeping 8.5 hours because of recovery, depression, sleep apnoea or chronic inflammation may be biologically very different from someone sleeping 8.5 hours with high sleep efficiency.
Bottom line
This is a large and technically sophisticated paper showing that both short and long self-reported sleep are associated with higher biological-age signals across multiple organs and molecular layers. Its strongest contribution is descriptive and integrative: sleep duration appears to track systemic ageing-related biology, not just brain health.
The main caution is that the study should not be read as proving that deliberately forcing sleep into a narrow range will slow ageing. The more defensible interpretation is: persistent short sleep and persistent long sleep are useful warning signals, and the long-sleep signal especially may reflect underlying illness or biological ageing rather than being a direct cause of it.