Distinct roles of chronotype, daytime napping, and sleep duration in biological and functional aging: a univariable and multivariable Mendelian randomization study

Abstract

Background

Sleep traits have been linked to aging, but observational designs cannot separate correlated sleep behaviors or establish causality. We used Mendelian randomization (MR) to estimate both total and direct effects of chronotype, daytime napping, and sleep duration on multidimensional aging outcomes.

Methods

Two-sample MR used large GWAS instruments for the three sleep traits against leukocyte telomere length, facial ageing, epigenetic clocks (IEAA, HannumAge, PhenoAge, GrimAge), frailty index, and cognitive performance. Inverse-variance weighted (IVW) was primary, with MR-Egger, weighted median, and MR-PRESSO as sensitivity analyses. Univariable MR (UVMR) estimated total effects; multivariable MR (MVMR) estimated direct effects accounting for intercorrelations.

Results

In UVMR, napping was adversely associated with telomere length (β = − 0.11, 95% CI − 0.18 to − 0.05, PFDR = 0.002), facial ageing (β = 0.05, 0.01–0.09, PFDR = 0.036), GrimAge (β = 0.96, 0.09–1.83, PFDR = 0.048), frailty (β = 0.32, 0.20–0.45, PFDR < 0.001), and cognition (β = − 0.16, − 0.29 to − 0.03, PFDR = 0.036); HannumAge was nominal (β = 0.53, − 0.29–1.35, PFDR = 0.275). Chronotype related to HannumAge (β = 0.46, 0.15–0.78, PFDR = 0.032) with borderline links to IEAA (β = 0.36, 0.07–0.66, PFDR = 0.060) and poorer cognition (β = − 0.04, − 0.08 to − 0.01, PFDR = 0.060). Sleep duration was inversely associated with frailty (β = −0 .17, − 0.26 to − 0.08, PFDR = 0.001). In MVMR, napping retained associations with GrimAge (β = 1.08, 0.02–2.15, P = 0.046) and frailty (β = 0.29, 0.15–0.43, P < 0.001), while links to telomere length and cognition attenuated. Chronotype became protective for facial ageing (β = −0 .03, − 0.05 to − 0.01, P = 0.011) and cognition (β = 0.09, 0.01–0.17, P = 0.036) and showed no independent association with epigenetic clocks. The inverse association between sleep duration and frailty strengthened (β = − 0.36, − 0.49 to − 0.22, P < 0.001).

Conclusions

Sleep traits show distinct, partly independent causal links with aging: excessive napping adversely influences GrimAge and frailty; longer sleep duration independently protects against frailty; and chronotype benefits facial ageing and cognition after accounting for other sleep behaviors.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13148-026-02068-2.

Background

Aging is characterized not only by the gradual decline of physiological function but also by increased vulnerability across multiple systems, leading to higher risks of chronic diseases and mortality [1, 2]. According to the United Nations World Population Prospects report, by 2050, the global population aged 65 years and older is projected to reach 1.6 billion, accounting for nearly 16% of the total population [3]. With the accelerating process of population aging, delaying the aging process, preventing frailty, and maintaining cognitive function have become major global public health challenges. Identifying and intervening in modifiable risk factors are therefore of great importance for promoting healthy aging.

Aging is a complex, multidimensional, and irreversible biological process that can be evaluated from appearance, functional, and molecular perspectives. At the appearance level, facial aging reflects the visually perceived degree of senescence through facial image analysis [4]. At the functional level, the frailty index, based on the deficit accumulation model, quantifies overall functional decline by summing health deficits across multiple domains and serves as a widely used measure of biological aging [5]. Cognitive decline represents another core feature of aging, reflecting brain integrity and functional capacity [6]. At the molecular level, epigenetic clocks such as intrinsic epigenetic age acceleration (IEAA), GrimAge, HannumAge, and PhenoAge are DNA methylation–based biomarkers that estimate biological age and capture the pace of aging [7]. In addition, telomere length, which shortens progressively with each cell division, reflects cellular replicative senescence and serves as an established biomarker of biological aging [8].

Sleep, an essential physiological process for maintaining life and homeostasis, plays a critical role in aging and cognitive health [9]. Poor sleep habits can accelerate aging processes at biological [10], cognitive [11], and structural brain levels [12]. Sleep behaviors primarily include chronotype, daytime napping, and sleep duration. The chronotype reflects an individual’s preference for morning or evening activity, often categorized as morning, intermediate, or evening types [13]. Evening chronotypes are more prone to circadian rhythm disruption, chronic sleep debt, poor sleep quality, and physiological dysregulation [14]. Daytime napping, another common sleep behavior, may help restore alertness, improve cognitive performance [15], and benefit cardiovascular health [16]; however, prolonged or frequent napping can indicate underlying health issues and has been linked to metabolic abnormalities, depression, and dementia [17]. Sleep duration exhibits a U-shaped association with aging: both short and long sleep have been associated with accelerated biological aging, cognitive decline, and reduced health span [18]. Recent epigenetic studies also demonstrate significant associations between sleep disturbances and accelerated epigenetic aging [19].

Although prior studies have explored links between sleep traits and aging, most focused on single aging indicators and were largely cross-sectional, limiting causal inference. To address these gaps, we investigated the potential causal effects of multiple sleep traits—chronotype, daytime napping, and sleep duration—on a broad spectrum of aging-related markers using Mendelian randomization(MR) [20]. We implemented both univariable MR (UVMR) to estimate total causal effects and multivariable MR (MVMR) to disentangle the direct effects of each sleep trait while accounting for their intercorrelations [21, 22]. By integrating large-scale genetic data with multidimensional measures of aging, this study provides strengthened causal evidence and clarifies distinct, trait-specific pathways through which sleep behaviors may influence healthy aging.

Study design

This MR analysis was conducted using publicly available summary-level data from large genome-wide association studies (GWAS) [23]. Genetic association data for the exposures—chronotype, daytime napping, and sleep duration—were obtained from the UK Biobank (UKB), a large prospective cohort comprising more than 260,000 individuals of European ancestry. Aging-related outcome data were derived from multiple resources, including the UK Biobank and independent cohorts such as TwinGene and the Wisconsin Longitudinal Study (WLS), which enhanced the generalizability of the findings.

Single nucleotide polymorphisms (SNPs) were selected as instrumental variables (IVs) to estimate the causal effects of sleep traits on aging phenotypes. The selection process adhered to the three core assumptions of MR: (1) relevance—the instruments are robustly associated with the exposures; (2) independence—the instruments are not linked with potential confounders; and (3) exclusion restriction—the instruments affect the outcomes only through the exposures of interest [24, 25]. To ensure validity, SNPs were chosen at genome-wide significance (P < 5 × 10⁻⁸), pruned for linkage disequilibrium (r² < 0.001), and required to have F-statistics greater than 10. A schematic overview of the MR design is provided in Fig. 1.

Fig. 1.

Conceptual framework of the Mendelian randomization study design. Overview of the MR framework evaluating the causal effects of chronotype, daytime napping, and sleep duration on aging-related outcomes. Genetic variants were selected as instrumental variables under the three core MR assumptions, and analyses were performed using UVMR and MVMR with IVW as the primary method and complementary sensitivity models. Abbreviations: MR, Mendelian randomization; IVW, inverse-variance weighted; UVMR, univariable Mendelian randomization; MVMR, multivariable Mendelian randomization; SNP, single nucleotide polymorphism; IEAA, intrinsic epigenetic age acceleration

Genetic instrument selection

Genetic instruments for three sleep-related exposures—chronotype, daytime napping, and sleep duration—were derived from large-scale GWAS in populations of European ancestry.

For chronotype, we used summary statistics from the largest GWAS to date, which combined up to 697,828 individuals from the UK Biobank (n ≈ 449,734) and 23andMe (n ≈ 248,098) cohorts [26]. Chronotype was assessed through self-reported morning/evening preference, and 351 independent loci were identified at genome-wide significance, many of which were validated against accelerometer-derived sleep timing measures.

For daytime napping, genetic associations were obtained from a GWAS of 452,633 UK Biobank participants [27]. Napping frequency was assessed using a touchscreen questionnaire, identifying 123 loci at genome-wide significance. Replication in 23andMe (n ≈ 541,333) and meta-analysis across both cohorts confirmed 94 loci, with concordant associations observed in accelerometer-derived measures of daytime inactivity.

For sleep duration, we used data from a GWAS of 446,118 UK Biobank participants [28]. Sleep duration was reported in hours per 24-h period, with 78 loci reaching genome-wide significance. Replication in the CHARGE (n = 47,180) and EAGLE (n = 10,554) consortia confirmed many of these signals, and accelerometer-based analyses further supported the validity of the phenotype.

For the MR analyses, effect alleles and reference alleles were harmonized to ensure consistent alignment between exposure and outcome datasets. SNPs associated with each exposure at genome-wide significance (P < 5 × 10⁻⁸) were selected as candidate instrumental variables and clumped for independence using an r² threshold of 0.001 and a 1 Mb window [29]. To satisfy the MR assumption that instruments are not associated with confounders, SNPs linked to aging-related traits were screened and excluded using LDassoc [30]. Instrument strength was evaluated by calculating the F-statistic for each SNP, using the formula: F = [(N-k-1)/k] × [R²/(1-R²)], where N is the effective sample size of the exposure GWAS, k is the number of SNPs, and R² is the proportion of variance in the exposure explained by the instrument. R² was estimated as: R² = 2 × (1–MAF) × MAF × (β/SD)² [31, 32]. SNPs with F-statistics < 10 were excluded to reduce weak instrument bias.

Data source for epigenetic and frailty markers

Summary-level GWAS data for aging-related phenotypes were derived from large-scale studies conducted predominantly in populations of European ancestry. Details regarding sample sizes, age distributions, and study cohorts are summarized in Table 1.

Table 1.

Source GWASs and sample characteristics for work-related exposures and aging outcomes

Phenotype	Sample size	Gender Distribution	Data source	Population	Mean age (years)	Age range (years)	PMID/GWAS ID
Chronotype	697,828	Both males and females	UKB and 23andMe	European	56.8	40–69	30696823
Napping	993,966	Both males and females	UKB	European	56.5	40–69	33568662
Sleep duration	446,118	Both males and females	UKB	European	56.5	40–69	30846698
Telomere length	472,174	Both males and females	UKB	European	Not specified	40–95	34611362
Facial aging	423,999	Both males and females	MRC-IEU	European	Not specified	Not specified	ukb-b-2148
Epigenetic age markers	34,710	42.7% female, 57.3% male	Edinburgh DataShare	European	54.8	27.2–79.1	34187551
Frailty index (UKB)	164,610	51.3% female, 48.7% male	UKB	European	64.1 (2.8)	60–70	34431594
Frailty index (TwinGene)	10,616	52.5% female, 47.5% male	TwinGene	European	58.3 (7.9)	41–87	34431594
Cognitive performance	257,841	Both males and females	UKB	European	Not specified	12–20 (Add Health), Adults (WLS)	30038396

Summary of GWAS data sources and cohort characteristics for all exposures and aging-related outcomes included in this MR study. Work-related exposures (HMPW, SW, WAS) were derived from UKB questionnaires. Aging phenotypes (including TL, FA, epigenetic clocks, FI, and CP) were obtained from large-scale GWASs conducted in UKB, TwinGene, and WLS cohorts. Sample size, ancestry, age range, and reference IDs (PMID or GWAS ID) are provided for reproducibility

UKB, UK Biobank; TwinGene, TwinGene Study; WLS, Wisconsin longitudinal study; GWAS, genome-wide association study; PMID, PubMed Identifier; mean age for cognitive performance is not provided as the cohorts have distinct age groups (Add Health: 12–20 years; WLS: adults)

Leukocyte telomere length was measured using quantitative PCR in 472,174 participants from the UK Biobank, with summary statistics reported by Codd et al. (2021) [33].

Facial ageing data were also obtained from the UK Biobank and processed through the GWAS pipeline of the MRC-IEU [34]. Perceived facial ageing was assessed using the questionnaire item: “Do people say that you look younger than you are, older than you are, about your age, do not know, or prefer not to answer?” [35, 36].

Epigenetic aging traits, including PhenoAge, GrimAge, HannumAge, and IEAA, were derived from a GWAS of 34,710 individuals of European descent [7]. PhenoAge and GrimAge are second-generation DNA methylation clocks trained on mortality risk, with GrimAge further incorporating plasma protein markers and smoking history. HannumAge and IEAA are first-generation clocks that estimate biological age directly from methylation profiles.

For frailty, genetic variants were obtained from a meta-analysis of GWAS comprising 175,226 participants, including 164,610 individuals from the UK Biobank (aged 60–70 years) and 10,616 from the Swedish TwinGene study (aged 41–87 years). Frailty was defined using a deficit accumulation model encompassing 49 health-related variables [37], a widely validated approach in gerontology [38].

Cognitive performance data were drawn from a meta-GWAS combining the UK Biobank and the Cognitive Genomics Consortium (COGENT), with a total of 257,841 participants [39]. The phenotype was based on standardized tests evaluating memory, language, and processing speed.

All outcome GWAS were conducted in non-overlapping samples relative to the exposure datasets. As this study relied exclusively on publicly available summary-level statistics, no additional ethical approval was required.

Statistical analysis

We first conducted UVMR analyses within a two-sample MR framework to estimate the causal effects of chronotype, daytime napping, and sleep duration on eight aging-related outcomes: leukocyte telomere length, facial ageing, PhenoAge, GrimAge, HannumAge, intrinsic epigenetic age acceleration (IEAA), frailty index, and cognitive performance. The inverse-variance weighted (IVW) method with a random-effects model was used as the primary approach.

To assess the robustness of causal inferences and detect potential violations of MR assumptions, we performed four complementary sensitivity analyses. The weighted median method provides consistent estimates if ≥ 50% of the weight comes from valid instruments [40]. MR-Egger regression evaluates directional pleiotropy using the intercept term [41]. MR-PRESSO identifies and corrects for horizontal pleiotropic outliers [42]. The contamination mixture method accounts for the presence of invalid instruments [43]. Heterogeneity across SNP-specific estimates was tested using Cochran’s Q statistic, and instrument strength was evaluated using F-statistics.

For outcomes with nominal associations in UVMR (P < 0.05), we further applied MVMR using the multivariable IVW method to estimate the direct effect of each exposure conditional on the others [22]. Multivariable MR-Egger was additionally performed where appropriate.

To account for multiple testing, we applied the false discovery rate (FDR) correction using the Benjamini–Hochberg procedure [44]. Associations with PFDR < 0.05 were considered statistically significant, while those with P < 0.05 but PFDR ≥ 0.05 were regarded as suggestive. FDR correction was preferred over Bonferroni correction as it provides a better balance between type I error control and statistical power in multi-trait MR settings.

Causal effect estimates were reported as β coefficients with corresponding 95% confidence intervals for each exposure–outcome pair. All analyses were performed in R 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria) using the TwoSampleMR [45], MR-PRESSO [42], and MendelianRandomization [46] packages.

Results

UVMR

We first conducted UVMR analyses to estimate the causal effects of chronotype, daytime napping, and sleep duration on aging-related outcomes. The forest plot (Fig. 2) and scatter plots for individual instruments (Figs. 3, 4, 5) illustrate the associations across outcomes.

Fig. 2.

Forest plot of UVMR estimates based on IVW for sleep-related exposures on aging-related outcomes. Forest plot showing the total causal effects of three sleep traits—chronotype (green), daytime napping (orange), and sleep duration (blue)—on eight aging-related outcomes using the IVW method under a UVMR framework. Each point indicates the β estimate with 95% confidence intervals. P-values and FDR-adjusted P-values are provided in the accompanying table. Abbreviations: MR, Mendelian randomization; IVW, inverse-variance weighted; FDR, false discovery rate; IEAA, intrinsic epigenetic age acceleration

Fig. 3.

Scatter plots of genetic instruments for chronotype and aging-related outcomes in UVMR. Scatter plots illustrating the associations between SNP instruments for chronotype and aging-related outcomes under the UVMR framework. Lines represent causal estimates obtained from IVW, MR-Egger, and weighted median methods. Abbreviations: SNP, single nucleotide polymorphism; MR, Mendelian randomization; IVW, inverse-variance weighted

Fig. 4.

Scatter plots of genetic instruments for daytime napping and aging-related outcomes in UVMR. Scatter plots illustrating the associations between SNP instruments for daytime napping and aging-related outcomes under the UVMR framework. Lines represent causal estimates obtained from IVW, MR-Egger, and weighted median methods. Abbreviations: SNP, single nucleotide polymorphism; MR, Mendelian randomization; IVW, inverse-variance weighted

Fig. 5.

Scatter plots of genetic instruments for sleep duration and aging-related outcomes in UVMR. Scatter plots illustrating the associations between SNP instruments for sleep duration and aging-related outcomes under the UVMR framework. Lines represent causal estimates obtained from IVW, MR-Egger, and weighted median methods. Abbreviations: SNP, single nucleotide polymorphism; MR, Mendelian randomization; IVW, inverse-variance weighted

Genetically predicted daytime napping was consistently associated with unfavorable aging traits. Napping was significantly related to shorter leukocyte telomere length (β =  − 0.11, 95% CI − 0.18 to − 0.05, PFDR = 0.002), accelerated facial ageing (β = 0.05, 95% CI 0.01–0.09, PFDR = 0.036), higher GrimAge acceleration (β = 0.96, 95% CI 0.09–1.83, PFDR = 0.048), increased frailty index (β = 0.32, 95% CI 0.20–0.45, PFDR < 0.001), and poorer cognitive performance (β =  − 0.16, 95% CI − 0.29 to − 0.03, PFDR = 0.036). A nominal association was also observed with HannumAge (β = 0.53, 95% CI − 0.29 to 1.35, PFDR = 0.275).

For chronotype, UVMR indicated positive associations with HannumAge (β = 0.46, 95% CI 0.15–0.78, PFDR = 0.032) and IEAA (β = 0.36, 95% CI 0.07–0.66, PFDR = 0.060), suggesting accelerated epigenetic aging. Chronotype also showed a suggestive negative association with cognitive performance (β =  − 0.04, 95% CI − 0.08 to − 0.01, PFDR = 0.060).

In contrast, sleep duration was inversely associated with frailty index (β =  − 0.17, 95% CI − 0.26 to − 0.08, PFDR = 0.001), while no significant associations were observed for telomere length or epigenetic clocks. None of the exposures showed robust associations with PhenoAge.

MVMR

To disentangle independent effects, we performed MVMR analyses including chronotype, napping, and sleep duration simultaneously (Fig. 6). The detrimental effects of napping on GrimAge (β = 1.08, 95% CI 0.02–2.15, P = 0.046) and frailty index (β = 0.29, 95% CI 0.15–0.43, P < 0.001) remained significant, whereas its associations with telomere length and cognition attenuated to non-significance.

Fig. 6.

Forest plot of MVMR estimates based on IVW for sleep-related exposures on aging-related outcomes. Forest plot showing the independent causal effects of chronotype (green), daytime napping (orange), and sleep duration (blue) on eight aging-related outcomes using the IVW method under a MVMR framework. Each point indicates the β estimate with 95% confidence intervals. P-values are presented in the accompanying table. Abbreviations: MR, Mendelian randomization; MVMR, multivariable Mendelian randomization; IVW, inverse-variance weighted; IEAA, intrinsic epigenetic age acceleration

For chronotype, the previously observed positive associations with epigenetic aging were no longer significant. Instead, chronotype exhibited a protective effect against facial ageing (β =  − 0.03, 95% CI − 0.05 to − 0.01, P = 0.011) and a positive association with cognitive performance (β = 0.09, 95% CI 0.01–0.17, P = 0.036).

The protective role of sleep duration against frailty was further reinforced in MVMR (β =  − 0.36, 95% CI − 0.49 to − 0.22, P < 0.001), confirming its independent contribution. No significant associations were identified between sleep duration and molecular aging markers.

Collectively, these results indicate that daytime napping exerts independent detrimental effects on aging, particularly through GrimAge acceleration and frailty. Longer sleep duration consistently protects against frailty, while the role of chronotype appears heterogeneous, showing harmful associations with epigenetic aging in UVMR but protective effects on facial ageing and cognition in MVMR.

Heterogeneity and pleiotropy

Sensitivity analyses evaluating heterogeneity and horizontal pleiotropy for each exposure are summarized in Tables 2, 3, 4. Overall, the mean F-statistics across analyses ranged from 15.7 to 32.9, exceeding the threshold of 10 and confirming adequate instrument strength.

Table 2.

Sensitivity analyses for heterogeneity and pleiotropy in Mendelian randomization of chronotype

Outcome	MR-Egger regression		Cochran’s Q		MR-PRESSO
	Egger intercept	P value	Q value	P value	Global Test P	Outliers Detected	Distortion Test P
Telomere length	− 0.001	0.111	133.04	0.105	0.104	NA	NA	19.90
Facial aging	0.000	0.958	234.42	0.000	< 0.001	NO	NA	18.54
PhenoAge	0.007	0.611	162.62	0.011	0.004	NO	NA	18.22
GrimAge	− 0.006	0.577	154.40	0.025	0.205	NA	NA	17.63
HannumAge	0.107	0.325	143.05	0.058	0.169	NA	NA	18.22
IEAA	0.021	0.308	109.21	0.707	0.699	NA	NA	17.49
Frailty index	− 0.001	0.565	275.84	0.000	< 0.001	YES	0.896	18.38
Cognitive performance	− 0.001	0.277	194.79	0.000	< 0.001	YES	0.673	20.08

Table 3.

Sensitivity analyses for heterogeneity and pleiotropy in Mendelian randomization of daytime napping

Outcome	MR-Egger regression		Cochran’s Q		MR-PRESSO
	Egger intercept	P value	Q value	P value	Global Test P	Outliers Detected	Distortion Test P
Telomere length	0.000	0.972	116.73	0.003	0.002	NO	NA	21.66
Facial aging	− 0.001	0.084	134.83	0.000	0.001	NO	NA	29.71
PhenoAge	− 0.009	0.652	90.89	0.102	0.114	NA	NA	27.45
GrimAge	− 0.002	0.884	85.98	0.203	0.213	NA	NA	28.16
HannumAge	− 0.008	0.578	81.14	0.351	0.384	NA	NA	27.80
IEAA	0.021	0.193	102.73	0.032	0.032	NO	NA	28.53
Frailty index	− 0.001	0.925	138.98	0.000	0.103	NA	NA	28.91
Cognitive performance	0.000	0.933	165.87	0.000	< 0.001	YES	0.594	32.86

Table 4.

Sensitivity analyses for heterogeneity and pleiotropy in Mendelian randomization of sleep duration

Outcome	MR-Egger regression		Cochran’s Q		MR-PRESSO
	Egger intercept	P value	Q value	P value	Global Test P	Outliers Detected	Distortion Test P
Telomere length	0.001	0.801	79.37	0.005	0.007	NO	NA	17.87
Facial aging	0.002	0.055	89.19	0.002	< 0.001	NO	NA	16.57
PhenoAge	− 0.062	0.104	78.66	0.032	0.026	NO	NA	15.72
GrimAge	− 0.015	0.394	58.29	0.428	0.424	NA	NA	15.72
HannumAge	− 0.028	0.122	67.25	0.166	0.188	NA	NA	15.72
IEAA	0.003	0.883	65.01	0.218	0.226	NA	NA	15.72
Frailty index	− 0.003	0.199	108.81	0.011	< 0.001	NO	NA	16.88
Cognitive performance	− 0.002	0.501	157.03	0.015	< 0.001	YES	0.808	18.60

Results of MR-Egger regression, Cochran’s Q test, and MR-PRESSO global test for the associations between genetically predicted sleep duration and aging-related outcomes. Egger intercepts and P values test for directional pleiotropy, Cochran’s Q statistics indicate heterogeneity, and MR-PRESSO identifies horizontal pleiotropy and outlier SNPs. Distortion test P values assess whether causal estimates materially changed after removal of outliers. Mean F-statistics are reported to demonstrate instrument strength

MR, Mendelian randomization; MR-PRESSO, Mendelian randomization pleiotropy residual sum and outlier; SNP, single nucleotide polymorphism; IEAA, intrinsic epigenetic age acceleration

For chronotype (Table 2), the MR-Egger intercepts were close to zero and non-significant for all outcomes (all P > 0.05), suggesting no evidence of directional pleiotropy. Cochran’s Q tests indicated significant heterogeneity for facial ageing (Q = 234.42, P < 0.001), PhenoAge (Q = 162.62, P = 0.011), GrimAge (Q = 154.40, P = 0.025), frailty index (Q = 275.84, P < 0.001), and cognitive performance (Q = 194.79, P < 0.001). The MR-PRESSO global test further suggested pleiotropy for facial ageing, PhenoAge, frailty index, and cognitive performance (all P < 0.01). Outlier SNPs were identified for frailty index and cognitive performance, yet the distortion tests were non-significant (P = 0.896 and 0.673), indicating that the IVW estimates were not materially influenced.

For daytime napping (Table 3), MR-Egger intercepts again provided no evidence of directional pleiotropy (all P > 0.05). Significant heterogeneity was observed for telomere length (Q = 116.73, P = 0.003), facial ageing (Q = 134.83, P < 0.001), IEAA (Q = 102.73, P = 0.032), frailty index (Q = 138.98, P < 0.001), and cognitive performance (Q = 165.87, P < 0.001). The MR-PRESSO global test detected pleiotropy for telomere length, facial ageing, IEAA, and cognitive performance. Outliers were identified only for cognitive performance, but the distortion test was non-significant (P = 0.594).

For sleep duration (Table 4), Egger intercepts remained null across all outcomes (all P > 0.05), with the closest to borderline being facial ageing (P = 0.055). Significant heterogeneity was identified for telomere length (Q = 79.37, P = 0.005), facial ageing (Q = 89.19, P = 0.002), PhenoAge (Q = 78.66, P = 0.032), frailty index (Q = 108.81, P = 0.011), and cognitive performance (Q = 157.03, P = 0.015). The MR-PRESSO global test was positive for telomere length, facial ageing, PhenoAge, frailty index, and cognitive performance. Outliers were flagged for cognitive performance, yet the distortion test remained non-significant (P = 0.808).

Taken together, these results indicate that although heterogeneity and some pleiotropy signals were detected across chronotype, napping, and sleep duration, MR-Egger intercepts were consistently null and distortion tests were negative, supporting that the primary IVW estimates are robust and unlikely to be substantially biased by horizontal pleiotropy.

Discussion

This MR investigation evaluated the roles of three sleep traits—chronotype, daytime napping, and sleep duration—across molecular, functional, and appearance domains of aging. In UVMR (Fig. 2), napping was adversely associated with shorter telomere length, accelerated GrimAge, higher frailty, and poorer cognitive performance; sleep duration was inversely associated with frailty; and chronotype showed suggestive links with accelerated IEAA/HannumAge and lower cognition. In MVMR (Fig. 6), which accounts for intercorrelations among sleep traits, only a subset of signals persisted: napping remained positively associated with GrimAge and frailty; the inverse association of sleep duration with frailty strengthened; and chronotype exhibited protective associations with facial ageing and cognition, while its epigenetic-aging signals attenuated. These contrasts indicate that some UVMR findings likely reflect shared variance across sleep behaviors, which MVMR partitions to reveal direct effects.

UVMR–MVMR contrasts and interpretation

Several UVMR associations attenuated to the null in MVMR. First, napping–telomere length and napping–cognition lost significance after mutual adjustment, suggesting these links were partly carried by correlated behaviors (e.g., sleep timing or duration) rather than a direct napping effect. Second, chronotype’s UVMR associations with IEAA/HannumAge were not retained, whereas protective associations with facial ageing and cognition emerged in MVMR—consistent with confounding by co-occurring sleep behaviors in univariable settings. By contrast, two signals proved robust and independent: napping’s adverse associations with GrimAge and frailty, and the protective association of longer sleep duration with frailty. Methodologically, MVMR separates causal, mediated, and confounded paths; thus, persistence/strengthening in MVMR argues for direct effects, whereas attenuation points to shared or mediated pathways rather than independent causality.

Functional outcomes (frailty and cognition)

Functionally, more frequent napping (UVMR) and shorter sleep duration related to greater frailty, while evening chronotype and napping related to poorer cognition in UVMR. After multivariable adjustment, the beneficial association of chronotype with cognition emerged and reached significance, whereas the napping–cognition signal attenuated. These patterns align with observational evidence that longer naps are linked to worse cognition in men [47] and higher odds of cognitive impairment in Chinese older adults [48]; both cross-sectional and longitudinal analyses show excessive napping relates to declines in episodic memory and mental status [49]. A longitudinal cohort found long sleep predicted higher risk of mild cognitive impairment, whereas reduced sleep predicted greater frailty [50]. Together with our MR results, these findings suggest that timing preference and daytime sleep propensity map onto cognition and frailty through partly separable pathways—some independent of total sleep duration.

Molecular aging markers (epigenetic clocks and telomere)

At the molecular level, napping → GrimAge remained significant in both UVMR and MVMR, implicating a direct relationship with methylation-based aging. Chronotype’s associations with IEAA/HannumAge were confined to UVMR and disappeared in MVMR, consistent with confounding by correlated sleep features in univariable analyses. These results agree with reports that composite high-risk sleep profiles—excessive napping, short sleep, evening chronotype—associate with faster DNA-methylation aging (e.g., PhenoAge, GrimAge) [51], and with U-shaped relations between sleep duration and phenotypic aging acceleration, faster aging among evening chronotypes, and amplification with greater social jetlag [52]. UK Biobank analyses similarly connect long/short sleep, evening chronotype, and frequent napping with PhenoAge acceleration [53]. Notably, our UVMR signal linking greater napping to shorter telomere length did not persist in MVMR. Longitudinal evidence links short sleep, poor sleep quality, and altered architecture to telomere attrition over eight years [54]; given inverse correlations between epigenetic age acceleration and telomere length [55], one hypothesis is that excessive napping marks processes that hasten methylation-based aging and secondarily relate to telomere shortening—requiring future longitudinal and experimental confirmation.

Appearance domain (facial ageing)

We observed a positive UVMR association of napping with facial ageing, but an inverse (protective) MVMR association for chronotype, indicating that univariable signals for chronotype may be confounded by co-occurring sleep behaviors. Potential pathways include lifestyle patterns more prevalent among evening types—delayed meals, sedentary time, and lower activity [56, 57]—which can impair energy metabolism and slow epidermal turnover, manifesting as dullness and coarse texture [58]. Eveningness has also been linked to impaired glucose metabolism [59], favoring formation of advanced glycation end products that stiffen and yellow dermal collagen, reducing elasticity and accelerating facial ageing [60]. For excessive napping, prolonged side or prone positioning may contribute to “sleep wrinkles” via sustained shear, compression, and tensile forces on facial skin [61]. These mechanisms remain speculative and merit targeted study.

Robustness of inference

Sensitivity diagnostics support the credibility of the primary findings (Tables 2, 3, 4). Although heterogeneity (Q) was present for several outcomes, Egger intercepts were uniformly null, and MR-PRESSO distortion tests were non-significant wherever outliers were removed—arguing against material bias of IVW estimates by unbalanced horizontal pleiotropy. Because MVMR typically reduces power (due to collinearity and more parameters), the persistence/strengthening of napping → GrimAge/frailty and sleep duration → frailty is unlikely to be an artifact and more plausibly reflects genuine direct effects.

Taken together, MVMR reframes the sleep–aging landscape: napping exerts adverse, trait-specific direct effects on epigenetic aging and frailty; sleep duration confers an independent protective effect on frailty; and chronotype shows no independent molecular harm but benefits facial ageing and cognition once other sleep behaviors are accounted for. Practically, these results prioritize reducing excessive daytime napping and maintaining adequate nocturnal sleep as plausible levers for healthier aging, while interventions targeting timing preference should explicitly address co-occurring behaviors that may drive risk in unadjusted analyses.

Conclusion

Using UVMR and MVMR, we delineate trait-specific and independent pathways linking sleep to aging. Across frameworks, two signals were consistent: daytime napping showed a direct adverse relation to epigenetic aging (GrimAge) and frailty, whereas longer sleep duration was independently protective against frailty. Chronotype did not exhibit independent molecular harm but related favorably to facial ageing and cognition after accounting for other sleep behaviors. These findings refine causal understanding of sleep–aging relationships and motivate targeted, prospective work to test mechanistic pathways and evaluate sleep-focused interventions across diverse populations.

Strengths and limitations

A major strength of this study is the use of large-scale genome-wide association studies (GWAS) to derive genetic instruments for chronotype, daytime napping, and sleep duration, ensuring robust statistical power. By leveraging a Mendelian randomization framework, the study reduces confounding and reverse causation that typically limit observational sleep–aging research. The inclusion of multiple molecular and functional aging markers—including telomere length, epigenetic clocks, frailty index, and cognitive performance—allowed for a comprehensive evaluation of aging processes. Furthermore, both univariable and multivariable MR approaches were applied, enabling the disentanglement of independent effects among correlated sleep traits. Sensitivity analyses, including MR-Egger, weighted median, contamination mixture, and MR-PRESSO, further supported the robustness of the findings.

Several limitations should also be acknowledged. First, the analyses were restricted to individuals of predominantly European ancestry, which may limit generalizability to other populations. Second, sleep traits were primarily based on self-reported measures, which, although validated against accelerometer-derived data, remain susceptible to misclassification. Third, while MR reduces bias from confounding, pleiotropy cannot be fully excluded, as indicated by heterogeneity and MR-PRESSO global test signals in some outcomes, although distortion tests suggested minimal impact on causal estimates. Finally, the effect sizes were modest, and the biological mechanisms underlying the observed associations remain to be clarified.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Zheng Zhang and Xinglian Wang contributed equally to this work. Guang Yang and Chenggang Jiang conceived the study, supervised the project, and provided overall methodological guidance. Zheng Zhang and Xinglian Wang designed the analysis plan, harmonized exposure and outcome datasets, conducted the MR analyses (UVMR and MVMR) and sensitivity tests, and prepared the figures and tables. Haitang Qiu and Qinghua Luo curated GWAS sources, performed data quality control, and contributed to statistical interpretation and contextualization of findings. Zheng Zhang and Xinglian Wang drafted the manuscript. Guang Yang and Chenggang Jiang provided critical revisions of intellectual content and approved the final version for submission. All authors contributed to result interpretation, reviewed and edited the manuscript, approved the final manuscript, and agree to be accountable for all aspects of the work. Correspondence: Guang Yang; Chenggang Jiang.

Funding

This study was supported by the AI-based Fertility Assessment and Intelligent Decision System for Perinatal Depressive Disorder Development (CSTB2023TIAD-KPX0051), and the National Key Clinical Specialty Construction Project (Obstetrics and Gynecology), 2022. The funders had no role in the study design, data collection, analysis, interpretation, manuscript preparation, review, approval, or the decision to submit for publication.

Data availability

The summary-level data used in this study were obtained from previously published genome-wide association studies (GWAS) and publicly available repositories. GWAS summary statistics for the three sleep-related exposures—chronotype, daytime napping, and sleep duration—were sourced from the UK Biobank and collaborating cohorts via the IEU Open GWAS platform (https://gwas.mrcieu.ac.uk). Outcome data for telomere length, facial ageing, frailty index, and cognitive performance were also accessed through the same platform. Summary statistics for epigenetic age acceleration markers (IEAA, HannumAge, PhenoAge, and GrimAge) were retrieved from the University of Edinburgh’s DataShare repository (https://datashare.ed.ac.uk/handle/10283/3645).All datasets used are publicly accessible and appropriately referenced in the manuscript. No individual-level data were analyzed. Further information is available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted using publicly available summary-level data from previously published genome-wide association studies (GWAS). All original studies obtained ethical approval from the relevant institutional review boards and informed consent from participants. As only de-identified summary statistics were analyzed, no additional ethical approval was required for the present study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.