Poor Sleep Patterns Accelerate Aging via Shared Inflammatory Protein Signatures
UK Biobank data links fragmented, irregular sleep to multiple hallmarks of aging through key proteins like IL1RN, GDF15, and LEP.
Summary
Researchers used wrist accelerometers worn for a week by more than 93,000 UK adults to estimate total sleep, REM sleep, deep sleep, light sleep, nighttime waking, and sleep regularity. They then tracked who later developed diseases linked to nine biological hallmarks of aging. More total sleep, deep sleep, and especially REM sleep were associated with lower risk across most hallmarks. More time awake after falling asleep and more irregular sleep were linked to modestly higher risk. Light sleep showed no clear link. Blood protein analysis in about 9,300 people pointed to immune and inflammatory pathways as a shared thread, with proteins such as leptin, GDF15, FABP4, and IL1RN recurring. The study is observational and cannot prove cause and effect.
Detailed Summary
Sleep is widely viewed as a pillar of healthy aging, yet most evidence relies on self-reported sleep and examines single diseases. This study asked a broader question: how do objectively measured sleep duration, sleep stages, and sleep regularity relate to the nine recognized hallmarks of aging, and do they share blood protein signatures? Answering that could point toward the biology connecting nightly rest to the pace of aging, and toward markers clinicians might eventually track.
Researchers analyzed 93,249 UK Biobank adults (mean age 56.2 years; 56.3% women) free of age-related disease who wore a wrist accelerometer for seven days. A deep-learning model called SleepNet, trained against laboratory polysomnography, estimated six sleep measures: total sleep, REM sleep, deep sleep (N3), light sleep (N1/N2), wakefulness after sleep onset (WASO), and sleep irregularity, defined as the seven-day standard deviation of sleep duration. Hallmarks were defined by mapping 83 age-related diagnoses onto nine hallmarks. Cox models adjusted for demographics, socioeconomic status, BMI, physical activity, smoking, and alcohol, with follow-up through October 2022 and false discovery rate correction.
By the end of follow-up, 62.4% of participants had developed at least one hallmark. Each additional hour of total sleep was linked to roughly 5% lower risk of stem cell exhaustion (HR 0.95, 95% CI 0.93-0.96) and deregulated nutrient sensing (HR 0.95, 0.94-0.96). REM sleep showed the strongest association: each extra hour corresponded to a 15% lower risk of stem cell exhaustion (HR 0.85, 0.83-0.88). Deep sleep was tied to lower risk of epigenetic alterations (HR 0.94, 0.91-0.97). Higher WASO and irregularity raised risk modestly (HRs about 1.04), while light sleep showed no significant associations.
Spline analyses suggested U-shaped relationships for light sleep, deep sleep, REM sleep, and irregularity (all p for nonlinearity <0.001), so more is not always better. Across 83 individual diseases, REM sleep was linked to 22 after correction, total sleep to 16 (including a 24% lower Parkinson's disease risk per hour, HR 0.76, 0.69-0.84), WASO to 7, irregularity to 5, deep sleep to 4, and light sleep to 3. Associations of total sleep and irregularity with epigenetic alterations were stronger in women (FDR-adjusted p for interaction = 0.03).
In 9,275 participants with Olink proteomics (2,920 proteins), 1,307 proteins were associated with total sleep and 1,262 with REM sleep, versus only 32 for light sleep. Overlapping proteins, associated with both a sleep pattern and a hallmark, ranged from 1 to 558 per pair and were enriched in inflammatory response, complement activation, leukocyte migration, and chemotaxis. Leptin (LEP), GDF15, FABP4, and IL1RN recurred; GDF15 and LEP ranked among the top five proteins for eight and seven hallmarks respectively.
These are observational associations with small per-hour hazard ratios, and hallmarks were defined through disease diagnoses rather than direct molecular measures. SleepNet agreed only moderately with polysomnography (F1 0.49 for three-class staging; total sleep off by about 48 minutes), blood was drawn about five years before accelerometry, and the cohort was 96.5% White. Results held in sensitivity analyses excluding sleep disorders and early events, but the authors frame them as hypothesis-generating.
Key Findings
- Each additional hour of REM sleep was associated with a 15% lower risk of stem cell exhaustion (HR 0.85, 95% CI 0.83-0.88) and lower risk across all nine hallmarks
- Each additional hour of total sleep was associated with 5% lower risk of stem cell exhaustion and of deregulated nutrient sensing (both HR 0.95)
- Deep sleep was linked to lower risk of epigenetic alterations (HR 0.94, 95% CI 0.91-0.97); light sleep showed no significant associations with any hallmark
- Greater WASO and sleep irregularity raised risk modestly, with HR 1.04 for deregulated nutrient sensing (WASO) and loss of proteostasis (irregularity)
- Each additional hour of total sleep was associated with 24% lower Parkinson's disease risk (HR 0.76, 95% CI 0.69-0.84); REM sleep was linked to 22 of 83 age-related diseases after FDR correction
- Overlapping sleep-hallmark proteins ranged from 1 to 558 and were enriched in inflammatory response, complement activation, and leukocyte migration; GDF15 and LEP ranked top five for 8 and 7 hallmarks
- Light sleep, deep sleep, REM sleep, and irregularity showed U-shaped dose-response curves with most hallmarks (all p for nonlinearity <0.001)
Methodology
Prospective analysis of 93,249 UK Biobank participants free of 83 age-related diseases at baseline, who wore an Axivity AX3 wrist accelerometer for 7 days. Sleep patterns were derived with the SleepNet deep-learning model, validated against polysomnography. Cox proportional hazards models (follow-up to October 31, 2022) were fully adjusted for demographic, socioeconomic, BMI, activity, smoking, and alcohol covariates, with Benjamini-Hochberg FDR correction. Proteomic overlap used 2,920 Olink proteins in 9,275 participants (linear regression for sleep, Cox for hallmarks), followed by Gene Ontology enrichment, subgroup analyses by sex and age, and five sensitivity analyses.
Study Limitations
Sleep stages were estimated by accelerometer-based deep learning with only moderate agreement with polysomnography, and proteomic blood samples were collected about five years before sleep measurement. Hallmarks were inferred from disease diagnoses rather than direct molecular measures, the cohort is predominantly White, and residual confounding and reverse causality cannot be excluded in an observational design. The authors describe the protein findings as hypothesis-generating; no conflict-of-interest information was included in the supplied text.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.
Enter your email to subscribe: