Short sleep and insomnia are associated with accelerated epigenetic age

Abstract

Objective:

Short sleep and insomnia are each associated with greater risk for age-related disease, which suggests that insufficient sleep may accelerate biological aging. We examine whether short sleep and insomnia alone or together relate to epigenetic age among older adults.

Methods:

A total of 3,795 men (46.3%) and women aged 56–100 years from the Health and Retirement Study were included. Insomnia was defined as reporting at least one insomnia symptom (difficulty falling asleep, waking up at night, or waking up too early in the morning) and feeling unrested when waking up most of the time. Those reporting <6 hours of bedtime were categorized as short sleepers. Three second- or third-generation epigenetic age acceleration clocks were derived from the 2016 HRS Venous Blood Study. The linear regression analysis was adjusted for age, sex, race/ethnicity, education, and obesity status.

Results:

Insomnia and short sleep were associated with an 0.49 (95%CI:0.03–0.94; P:0.04) and 1.29 (95%CI:0.52–2.07; P:0.002) years acceleration of GrimAge, respectively, as well as a faster pace of aging (DunedinPACE; 0.018 (95%CI:0.004–0.033; P:0.02); 0.022(95%CI:−0.004–0.048; P:0.11)). Compared to healthy sleepers, individuals with the combination of short sleep and insomnia had an accelerated GrimAge (0.97 years; 95%CI:0.07–1.87; P:0.04) and a greater DunedinPACE (0.032; 95%CI:0.003–0.060; P:0.04).

Conclusion:

Our findings indicate short sleep, insomnia, and the combination of the two, are linked to epigenetic age acceleration, suggesting that these individuals have an older biological age that may contribute to risk for comorbidity and mortality.

Introduction

Insomnia is a disorder characterized by an inability to initiate or sustain sleep at night, accompanied by daytime dysfunction. Estimated prevalence increases with age with roughly half of adults over age 65 reporting sleep disturbances, and 10–25% experiencing more severe clinical insomnia.(1–4) Optimal sleep duration of 7 to 8.5 hours per night is recommended by the American Academy of Sleep Medicine and Sleep Research Society,(5) whereas short sleep duration of less than 6 hours, or long sleep duration of 9 or more hours, are associated with worse health. Given that the U.S. adult population prevalence of short sleep duration is estimated to be around 30–35%, this is a public health concern.(6,7)

Insomnia and short sleep duration independently are associated with a myriad of age-related outcomes, including an increased risk for cardiovascular disease and mortality.(8–10) Despite both insomnia and short sleep having effects on health, the combination is thought to be the most detrimental for age-related disease and death.(11–24) Therefore, it is important to identify the effects of short sleep and insomnia separately, as well as synergistically, on their health outcomes. The process through which short sleep or insomnia has its effect on health is unknown, although one hypothesis is that inadequate sleep modifies the biological aging processes.(25) The extent of biological aging, as opposed to chronological aging, has recently been quantified with DNA methylation patterns used to calculate epigenetic age estimates. These epigenetic age estimates infer biological age, as compared to chronological age. They are strongly associated with higher comorbidity and mortality risk, and can serve as surrogate endpoints for these risks.(26–29) As such, studying the association between sleep duration and insomnia symptoms, independently and interactively, with epigenetic age acceleration can provide further understanding of the role that insomnia symptoms and sleep duration have on increased risks for morbidity and mortality, as well as identify the potential role biologic aging has as a mechanism of this risk.

The first-generation epigenetic age clocks were trained to predict chronologic age; second-generation clocks have been created to determine the biologic age based on manifestation of phenotypic signs of aging, the development of morbidities, and risk for mortality. Two second-generation (e.g., GrimAge, PhenoAge) and one third-generation clock (e.g., DunedinPACE) were developed for this purpose.(26,28,29) GrimAge is based on a mortality risk estimated through a two-step approach that combines seven DNA methylation (DNAm) estimated proteins levels, DNAm estimated smoking pack years, chronological age and sex.(28) PhenoAge was developed to estimate a multi-system estimate of “Phenotypic Age”, using both mortality and clinical data.(26) An increased epigenetic PhenoAge, relative to chronological age, was found to be associated with among other pathways, an increase in inflammatory pathways and a decrease in DNA damage response.(26) DunedinPACE was developed on a longitudinal cohort in Dunedin, New Zealand to estimates the rate or “pace” of aging by modeling methylation patterns that reflect a change in 19 indicators of organ-system function over two decades of life, where the reference value is a rate of 1 year of aging per chronological year.(29)

Prior analyses have tested the relationship of insomnia symptoms and sleep duration with epigenetic aging among postmenopausal women, finding links of insomnia symptoms with older epigenetic age using first generation clocks.(30) Recent findings have also linked short sleep duration among postpartum mothers (the majority of whom had sleep disturbances) with an older PhenoAge.(31) In addition, increased epigenetic age has been observed among young female adults who reported short sleep duration during college.(32) Following current epidemiological guidelines, we will also consider potential interaction of sex and race on the association between sleep and epigenetic age acceleration, especially given that previous studies on the association between insomnia, short sleep duration and epigenetic age acceleration have been restricted to women only.

Several studies have examined the relationship of sleep duration and insomnia symptoms with telomere length, another biomarker of aging.(25) In addition, previous studies have indicated an association between sleep apnea and epigenetic age acceleration.(33,34) However, to our knowledge, no previous studies have examined the association of short sleep duration with and without insomnia symptoms with epigenetic age acceleration using the second- and third-generation clocks in an older adult sample, including the new DunedinPACE.

We hypothesize that short sleep and insomnia symptoms would be associated with an increased epigenetic age. This could provide evidence that biologic aging may be one of the mechanisms through which sleep duration and insomnia symptoms influence the increased mortality and morbidity risks that have been observed in previous studies.(8–10)

Methods

For this project, we used data from participants of a subset of the Health and Retirement Study (HRS). The HRS is an ongoing, longitudinal study of men and women aged 51 years or older at baseline, which has been described in more detail previously.(35) The Health and Retirement Study data is sponsored by the National Institute on Aging (grant number U01AG009740).(36) We restricted the population to those with DNA methylation assays performed from samples collected during the HRS Venous Blood Study.(37) This sample consists of 4,018 individuals age 56 and older who were chosen from the almost 10,000 participants who consented to a venous blood draw. Those included in the blood draw participated in the 2016 interview, which was not their first interview, answered for themselves, and were not in a nursing home. When using a weighted analysis approach, this sample was designed to be representative of the U.S. aging population. All study protocols regarding human participants have been approved by their local Institutional Review Board and informed consent was given by all participants. All data are publicly available through the HRS or the NIAGADS website.(38,39)

Sleep characteristics

The sleep characteristics were established by questionnaire during the 2014 or 2016 visit. A random sample of the population sample received this questionnaire during the 2014 survey, and the other half of the population received this questionnaire during the 2016 survey. Sleep duration was estimated as the number of hours in bed, calculated based on the time the individual reported he or she went to bed and woke up the day before the interview. Given that previous research has shown a non-linear effect of sleep duration on health risk with health risk at both extremes,(5) we categorized sleep duration into four categories known to be correlated with increased health risks. Based on prior work linking both short and long sleep to health risks,(5) sleep duration was categorized as: short sleep (less than 6 hours); less than optimal sleep (6–7 hours); healthy sleep (7–8.5 hours); and long sleep (more than 8.5 hours).Based on prior work linking short and long sleep to health risk,(5) sleep duration was categorized as: short sleep (less than 6 hours); less than optimal sleep (6–7 hours); healthy sleep (7–8.5 hours); and long sleep (more than 8.5 hours).

Insomnia symptoms was determined using the following questions: “How often do you have trouble falling asleep?”, “How often do you have trouble because of waking up during the night?”,” How often do you have trouble with waking up too early and not being able to fall asleep again?”, and ”How often do you feel really rested when you wake up in the morning?”. To approximate a clinical diagnosis of insomnia, (40) we combined the evidence of sleep disturbance (reporting “most of the time” trouble falling asleep, or sleeping throughout the night) with non-restorative sleep defined as not feeling rested when waking up. In secondary analyses, reported in the Supplemental Digital Content, we calculated an insomnia symptom score where each question was counted as a one if they answered that this happened rarely/never, a two if they answered this happened sometimes, and a three if this happened almost all the time. The four questions were then summed into the insomnia symptoms score with a range between 4 and 12, where a higher score indicates more or more frequent insomnia symptoms.

Insomnia symptoms with short sleep duration.

As it has been hypothesized that the combination of short sleep duration (less than 6 hours) and insomnia symptoms have even more detrimental health effects, (11–24) we compared individuals with short sleep and insomnia against those who reported healthy sleep, defined as a sleep duration between 7 and 8.5 hours without insomnia.

Epigenetic age

Blood samples from the 2016 visit were utilized to derive DNA methylation using the MethylationEPIC BeadChip (Infinium) microarray (Illumina Inc.).(37) Quality control of the DNA methylation data has been previously reported.(41) The epigenetic clocks were calculated by HRS using the DNAmAge website for the first generation and the PhenoAge and GrimAge clocks.(42) The DunedinPACE package was used to calculate the DunedinPACE.(29) These clocks were previously created and are available through the HRS website.(43)

Final study population, covariates, and analyses

We used sample weights provided by HRS that are designed to make the sample population representative of community-dwelling older adults in the US. We excluded participants who did not have a survey weight for the Venous Blood Study (N = 142). These participants were too young to be part of the HRS cohort or were not community-dwelling, and were therefore not representative of the target population.

We included individuals with information about sleep duration, insomnia symptoms and DNA methylation (12 individuals were missing information regarding their sleep). We excluded individuals who did not have information about covariates that were included in the analyses. We considered two groups of covariates in our model based on extant literature. The first model included: chronological age; sex; race/ethnicity (divided into White, not Hispanic; Black, not Hispanic; Hispanic; and other, not Hispanic; N missing: 11); and education (divided into four categories: <12 years; 12 years; 13 to 15 years; and 16 or more years of education; N missing: 18). Age, sex, education, and race/ethnicity are associated with both sleep duration and insomnia symptoms,(44–46) as well as epigenetic age.(47,48)

The second model included the covariates from the first model plus obesity status given the links between obesity and epigenetic aging,(49–52) and obesity status and sleep disorders.(53) Obesity status was categorized in four categories based on an individual’s Body Mass Index (BMI). The groups were under/normal weight (BMI<25); overweight (BMI between 25 and 30); obese (BMI between 30 and 35); and morbidly obese 2 (BMI>35). Obesity status was missing for 40 individuals leading us to a total study population of 3795 individuals who were included in the analysis.

We analyzed the association between the main sleep characteristics of interest (sleep duration; insomnia symptoms; and the combination of short sleep duration and insomnia symptoms) with the three second/third generation epigenetic clocks (PhenoAge, GrimAge, and DunedinPACE) by performing linear regression analyses.

Sensitivity analysis

After performing analysis on our main exposures of interest, we also analyzed insomnia symptoms and sleep duration using a continuous scale. Given that previous studies have shown a non-linear effect with the strongest effects of health outcomes among those reporting duration at the ends of the spectrum, we performed a spline model with a knot at 7.5 hours.

We performed additional sensitivity analyses by stratifying by the visit where sleep information was gathered; by sex; by race; and by retirement status. In addition, we performed a sensitivity analysis excluding those who reported sleep apnea or other sleep disorders (N=553; 502 individuals reported sleep apnea). Finally, to explore potential pathways of epigenetic age differences, we analyzed the association between short sleep, insomnia symptoms and other (first-generation) epigenetic clocks, namely the Horvath and Hannum clocks; the seven GrimAge components, and DNAm-based estimates of white blood cell counts (reported in Supplemental Digital Content, see supplemental note 1).

Results

Descriptive Statistics

The study populations had an average chronological age of 68.6 years and consisted of 53.8% women. Most of the individuals in this population identified as White (78.0%), followed by: Black (10.1%), Hispanic (8.5%), and other race/ethnicity (3.4%). Many of the individuals were either overweight (37.1%) or obese (35.8%). See Table 1 for more characteristics of this population, stratified by insomnia symptoms.

Table 1.

Descriptive characteristics of the study population. Weighted proportions of the demographic characteristic are provided stratified by insomnia symptoms (yes/no), except for age and average duration in bed which are provided in years and hours, respectively.

Descriptive Statistics
		No Insomnia symptoms N=2996	Insomnia symptoms N=799	Total N=3795
	Age (years)	68.8 (SD: 9.3)	67.7 (SD: 9.0)	68.6 (SD: 9.2)
	Average duration in bed (hours)	7.84 (SD: 1.51)	7.68 (SD: 1.96)	7.81 (SD: 1.59)
Duration in bed (categorial)
	Less than 6 hours	0.08	0.16	0.10
	Between 6 and 6.99 hours	0.15	0.17	0.15
	Between 7 and 8.49 hours	0.43	0.30	0.40
	More than 8.5 hours	0.34	0.36	0.35
	Missing info on duration	757 (27.5%)	220 (25.3%)	977 (25.7%)
Sex
	Male	0.48	0.40	0.46
	Female	0.52	0.60	0.54
Race/Ethnicity
	Black, not Hispanic	0.10	0.12	0.10
	Hispanic	0.08	0.09	0.09
	White, not Hispanic	0.79	0.76	0.78
	Other, not Hispanic	0.04	0.03	0.03
BMI
	Under/Normal Weight (BMI<25)	0.29	0.21	0.27
	Overweight (BMI:25–30)	0.38	0.33	0.37
	Obese (BMI:30–35)	0.21	0.27	0.22
	Morbidly Obese (BMI>35)	0.12	0.19	0.13
Smoker Status
	Never Smoked	0.45	0.41	0.44
	Current Smoker	0.10	0.15	0.11
	Past Smoker	0.45	0.45	0.45
Educational Attainment
	0–11 Years	0.12	0.19	0.14
	12 Years	0.30	0.32	0.30
	13–15 Years	0.26	0.24	0.26
	16+ Years	0.32	0.26	0.31

Abbreviations: N – Number; SD – Standard Deviation; BMI: Body Mass Index

Sleep characteristics

Among this population, 3773 individuals provided information on insomnia symptoms, 799 individuals (21.1%) had suspected insomnia defined as most of the time not feeling rested upon wakening and at least one of the sleep disturbance questions most of the time. The insomnia symptoms score ranged from 4 to 12; with the majority of individuals reporting a low score (65.33% less than a score of 8; and only 11.95% with a score of 10 and more). We reported the frequency of the insomnia symptoms score in Table S1, Supplemental Digital Content.

The average duration in bed was 7.8 (SD: 1.6) hours. There was a low correlation between duration in bed and insomnia symptoms score (correlation: −0.04, p-value 0.01) or dichotomized insomnia symptoms (−0.04, p-value 0.01). Especially, among individuals with an insomnia symptoms score of 10 or above, the average duration in bed appears to slightly decrease (from 7.85 with a score less than 10 to around 7.5 hours with a score of 10 or more). In addition, the standard deviation of duration increases from approximately 1.5 to 2.0 hours (p-value of F-test 1e-4, Supplemental Table S1), suggesting that those with more insomnia symptoms are more likely to either have short or long sleep duration.

A total of 290 individuals reported a short sleep duration (see also Supplemental Table S2, less than 6 hours in bed, 10.3%); 417 individuals between 6 and 7 hours (14.8%); 1100 individuals between 7 and 8.5 hours (39.5%); and 1011 individuals with a long sleep duration defined as more than 8.5 hours (35.9%). A total of 98 individuals reported both short sleep duration and insomnia symptoms (3.5%), 192 individuals reported short sleep duration without insomnia symptoms (6.8%), and 183 individuals insomnia symptoms with a normal sleep duration (6.5%), compared to 917 (32.5%) individuals who reported healthy sleep (between 7 and 8.5 hours in bed and no insomnia symptoms). One-third of those who reported short sleep also reported insomnia symptoms (N=98/290; 33.8%), while only 16.9% (N=98/579) of those who reported insomnia symptoms also reported short sleep.

Sleep and epigenetic age

In chronological age, sex, race/ethnicity, and education adjusted models, individuals with insomnia symptoms were on average epigenetically older, both in terms of GrimAge and the DunedinPACE, compared to those who reported no insomnia symptoms (Table 2). On average, they were 0.52 years older in GrimAge (95% Confidence Interval [CI]: 0.08 – 0.97, p-value: 0.025), and their pace of aging was faster by 0.026 years/year (95% CI: 0.012 – 0.041, p-value: 0.001). Further adjustment by obesity slightly attenuated these estimates but remained statistically significant (GrimAge: 0.49 years; 95%CI: 0.03 – 0.94, p-value: 0.042, and DunedinPACE: 0.018 years/year; 95% CI: 0.004 – 0.033, p-value: 0.016). No associations were seen with PhenoAge and insomnia symptoms.

Table 2.

Linear regression analysis for sleep characteristics and three epigenetic ages (PhenoAge, GrimAge, and DunedinPACE).

		PhenoAge				GrimAge				DunedinPACE
	Cov. Adj.	Eff	95% CI		P	Eff	95% CI		P	Eff	95% CI		P
Insomnia symptoms (N=3795)
Insomnia symptoms (N=799)	Model 1	0.08	−0.67	0.82	0.84	0.52	0.08	0.97	0.025	0.026	0.012	0.041	0.001
	Model 2	−0.04	−0.79	0.71	0.92	0.49	0.03	0.94	0.042	0.018	0.004	0.033	0.016
Sleep duration (N=2818) divided in 4 categories; reference category are individuals who reported time in bed between 7 and 8.49 hours (N=1100)
In Bed: Less than 6 hours (N=290)	Model 1	0.45	−0.56	1.47	0.39	1.36	0.62	2.11	0.001	0.029	0.003	0.055	0.037
	Model 2	0.36	−0.69	1.41	0.54	1.29	0.52	2.07	0.002	0.022	−0.004	0.048	0.11
In Bed: Between 6 and 6.99 hours (N=418)	Model 1	0.27	−0.61	1.14	0.56	0.19	−0.36	0.74	0.50	0.012	−0.005	0.029	0.17
	Model 2	0.20	−0.70	1.09	0.67	0.15	−0.39	0.70	0.58	0.009	−0.008	0.026	0.31
In Bed: 8.5 hours or more (N=1011)	Model 1	0.34	−0.27	0.94	0.28	0.72	0.32	1.12	0.001	0.020	0.005	0.035	0.011
	Model 2	0.27	−0.36	0.89	0.41	0.67	0.26	1.08	0.003	0.017	0.002	0.032	0.033
Short sleep and insomnia symptoms combined versus individuals who reported healthy sleep (N=917)
Short Sleep and Insomnia symptoms (N=98)	Model 1	0.51	−0.78	1.80	0.44	1.11	0.26	1.95	0.013	0.045	0.017	0.072	0.003
	Model 2	0.27	−1.08	1.62	0.70	0.97	0.07	1.87	0.040	0.032	0.003	0.060	0.037
Insomnia symptoms with normal sleep duration (N=183)	Model 1	−0.52	−1.96	0.93	0.49	0.04	−0.98	1.06	0.94	0.017	−0.012	0.046	0.26
	Model 2	−0.57	−2.01	0.87	0.44	0.01	−1.01	1.02	0.99	0.013	−0.013	0.04	0.32
Short Sleep without insomnia symptoms (N=192)	Model 1	0.29	−0.93	1.50	0.65	1.11	0.18	2.04	0.024	0.019	−0.014	0.052	0.27
	Model 2	0.23	−1.02	1.48	0.72	1.08	0.13	2.02	0.031	0.017	−0.017	0.050	0.33

Abbreviations: Cov. – Covariate; Adj. – Adjustments; 95% CI – 95% Confidence Interval; P – P value

Model 1 was adjusted for age, sex, race/ethnicity and education; Model 2 was model 1 plus obesity status.

For the analysis of sleep duration in clinically relevant categories, short sleep duration (less than 6 hours per night) was associated with an older epigenetic age for GrimAge of 1.36 years (95% CI: 0.62 – 2.11, p-value: 0.001) as compared to healthy sleep duration (7 to 8.5 hours per night). Short sleep was also associated with a faster DunedinPACE of aging by 0.029 years/year (95% CI: 0.003 – 0.055, p-value: 0.037) when compared to those with healthy sleep duration of 7–8.5 hours. Further adjustment for obesity status had a modest effect for both GrimAge and DunedinPACE. While the DunedinPACE showed a suggestion for an increase in the pace of aging, the confidence interval included the null (0.022 years/year; 95% CI: −0.004 – 0.048, p-value: 0.11). The GrimAge effect estimates were attenuated but remained statistically significant (1.29 years; 95%CI: 0.52 – 2.07, p-value: 0.002).

Those who reported a longer sleep duration (>8.5 hours a night) were on average 0.72 years older in their GrimAge (95%CI: 0.32 – 1.12, p-value: 0.001) and after adjusting for obesity status were 0.67 years older in GrimAge (95%CI: 0.26 – 1.08, p-value: 0.003). The DunedinPACE of aging was 0.020 years/year (95% CI: 0.005 – 0.035, p-value: 0.011) faster for long sleep duration compared to health sleep duration. This effect remained after adjustment for obesity status (0.017 years/year (95% CI: 0.002 – 0.032, p-value: 0.033)).

Insomnia symptoms in combination with a short sleep duration were associated with epigenetic age acceleration compared to those reporting insomnia symptoms without the short sleep duration. When reviewing an interaction between short sleep duration and insomnia symptoms, the subgroups become relatively small with a subsequent loss of power in analysis. The interaction model did not show a statistically significant interaction between short sleep and insomnia symptoms. However, when reviewing the effect estimates within the subgroups, those reporting short sleep duration and insomnia symptoms compared to healthy sleepers had an increase in GrimAge of 0.97 years (95%CI: 0.07 – 1.87, p-value: 0.040) older than healthy sleepers. Similarly, short sleepers with insomnia symptoms had a DunedinPACE rate of 0.032 years/year (95%CI: 0.003 – 0.060, p-value 0.037) faster than healthy sleepers. Among individuals who reported insomnia symptoms with a normal sleep duration, there was no statistically significant effect on the epigenetic age acceleration. Finally, among those reporting a short sleep duration without insomnia symptoms, we observed an increase in GrimAge of 1.08 years (95%CI: 0.13 – 2.02, p-value: 0.031), while no statistically significant effect was seen for DunedinPACE.

Sensitivity analyses

Effect of insomnia symptoms and sleep duration using continuous variables

Similar effects were seen when using the insomnia symptom score instead of the dichotomized insomnia symptoms. Specifically, we observed that an increase in insomnia symptom score is associated with an increased epigenetic age, both in terms of GrimAge and the DunedinPACE, compared to those who reported no insomnia symptoms (Supplemental Table S3). On average, each point increase in insomnia symptoms score is associated with 0.10 years older in GrimAge (95% Confidence Interval [CI]: 0.02– 0.19 p-value: 0.017), and their pace of aging was faster by 0.005 years/year (95% CI: 0.002 – 0.008, p-value: 0.001). Further adjustment by obesity slightly attenuated these estimates but remained statistically significant (GrimAge: 0.09 years; 95%CI: 0.01 – 0.18, p-value: 0.034, and DunedinPACE: 0.004 years/year; 95% CI: 0.001 – 0.006, p-value: 0.010). No statistically significant associations were seen between PhenoAge and insomnia symptoms though results were suggestive for an increase in epigenetic age.

For the analysis of continuous sleep duration on epigenetic age acceleration, we used a spline model with a knot at 7.5 hours of bedtime duration to account for the non-linear effects of sleep duration. For individuals who reported a bedtime duration less than 7.5 hours, every hour of increase in sleep duration was associated with a decrease in epigenetic age, specifically for GrimAge (−2.91 years; 95%CI: −5.07 to −0.75, p-value: 0.011) and the DunedinPACE (−0.067 years/year; 95% CI: −0.128 to −0.006, p-value: 0.038). Adjusting for obesity status, attenuated the effects. The effects of GrimAge remained statistically significant (GrimAge: −2.71 years; 95%CI: −4.96 – −0.46, p-value: 0.023). The effects of sleep duration on DunedinPACE were attenuated and had a confidence interval that crossed the null, with every hour of increase in sleep duration the DunedinPACE decreased with −0.048 years/year (95% CI: −0.112 – 0.015, p-value: 0.14). There were no effects of sleep duration on epigenetic age among individuals who reported a time in bed longer than 7.5 hours.

For the sensitivity analysis of the first-generation epigenetics clocks, GrimAge components, and epigenetic cell type estimates, we would like to refer to Supplemental Note 1 and Supplemental Table S4. Briefly, no effects were seen between the categorical sleep characteristics and first-generation epigenetic clocks, DNAm-based estimated white blood cells, or the DNA-m based GrimAge components: leptin, tissue inhibitor metalloproteinases 1 (TIMP1), and adrenomedullin (ADM). There were associations between insomnia symptoms and the following GrimAge components: DNAm-based packyears and Plasminogen Activator Inhibitor-1 (PAI-1); between short sleep duration and DNAm-based packyears, growth differentiation factor 15 (GDF15), beta-2-microglobulin (B2M), and possibly Cystatin C.

Effect of sleep characteristics on epigenetic age stratified by race/ethnicity and sex

When examining the effects of insomnia symptoms and short sleep on epigenetic age stratified by race/ethnicity, results suggest that the relationship is stronger among individuals in the Black and Hispanic populations compared to the White non-Hispanic population (Supplemental Table S5). We observed an interaction with racial subgroups and insomnia symptoms on GrimAge for those identifying as Non-Hispanic Black (p-value for interaction: 0.046) and possibly for Hispanics (p-value interaction: 0.086) compared to those identifying as non-Hispanic White. For non-Hispanic Blacks, insomnia symptoms increased the average GrimAge by 1.74 years (95%CI: 0.52 – 2.95, p-value: 0.008), for Hispanics with 1.36 years (95%CI: 0.20 – 2.53, p-value 0.034), while for non-Hispanic Whites the average GrimAge did not increase (0.14 years, 95%CI: −0.39 – 0.67, p-value: 0.61). Though the effects of insomnia symptoms among non-Hispanic Blacks and Hispanics was larger for DunedinPACE compared to non-Hispanic Whites, the confidence intervals were wide and the p-values for interaction were not significant (p-value: 0.13 and 0.27). Similarly, though the effect estimates are larger among Hispanics compared to Whites for short sleep duration and GrimAge, there is no statistically significant interaction when testing (p-value 0.16).

Finally, we observed an interaction between the combined short sleep and insomnia symptoms and non-Hispanic Blacks and non-Hispanic Whites (p-values interaction PhenoAge: 0.010, GrimAge: 0.019, DunedinPACE: 0.016), and between Hispanics and non-Hispanic Whites (p-value interaction GrimAge: 0.034). For non-Hispanic Blacks, the combination of short sleep and insomnia symptoms, increased the PhenoAge by 2.97 years (95%CI: 0.75 – 5.18, p-value: 0.018), GrimAge by 2.81 years (95%CI: 1.06 – 4.56, p-value: 0.006), and DunedinPACE by 0.12 years/year (95%CI: 0.05 – 0.19, p-value: 0.004). For Hispanics, the combination of short sleep and insomnia symptoms are associated with an increase in GrimAge by 3.53 years (95%CI: 1.14 – 5.92, p-value: 0.028). See Figure 1 for comparison of effect estimates within each race/ethnic group.

Figure 1.

Visual representation of the effects of insomnia symptoms, short sleep duration, and the combination of short sleep duration and insomnia symptoms on the epigenetic clocks, stratified by race categories. The error bars represent the range between the upper and lower 95% confidence interval. Figure 1A–C are the effects of insomnia symptoms on PhenoAge, GrimAge, and DunedinPACE respectively; figures 1D–F the effects of short sleep duration (less than 6 hours in bed) on PhenoAge, GrimAge, and DunedinPACE respectively; and finally figures 1G–I the effects of the combination of insomnia symptoms and short sleep duration on PhenoAge, GrimAge and DunedinPACE. Abbreviations: NH-White – Non-Hispanic White; yrs – years; yr – year

When stratifying the population by sex, the overall effect estimates of sleep on epigenetic age are mainly driven by the strong effects of short sleep and insomnia symptoms on epigenetic age among men. There was statistically significant interaction between sex and short sleep on GrimAge (p-value 0.040). Among women, the effect estimates are attenuated though in the same direction as among men (see supplemental Table S6). A further sensitivity analysis included stratifying by: visit, retirement status, and excluding those reported sleep disorders. The effects were similar among each subgroup (see Supplemental Tables S7 to S9).

Discussion

Short sleep duration and insomnia symptoms, separately and in combination, were associated with an increased GrimAge and a faster pace of aging using the DunedinPACE in this older adult cohort, which was representative of the U.S. population. These results indicate that individuals with these sleep characteristics are biologically aging faster, which is linked to an increased mortality and morbidity risks.(26–29)

Our findings indicated an association with both the GrimAge and DunedinPACE. Though there was a trend towards an increased epigenetic age in PhenoAge, the results had wide confidence intervals including the null. Each of these epigenetic clocks capture unique components of biological aging. Even though all three epigenetic clocks are highly correlated with mortality and morbidity risks, (26–29) they differ in terms of how they were designed, the training population, and primary outcomes. GrimAge was developed to track with biomarkers that are predictive of time to death in the Framingham Heart Study, while the DunedinPACE was developed using a longitudinal birth cohort from New Zealand starting in 1970’s to track with the pace at which the cohort ages from childhood into adulthood using changes in biomarkers over the first five decades of life.

In our sensitivity analysis we tried to delve further into potential pathways for an increased GrimAge. We observed that insomnia symptoms and the combination of insomnia symptoms and short sleep duration are associated with several biomarkers of inflammatory and senescence pathways. Specifically, our sensitivity analysis indicated an association between insomnia symptoms and the combination of short sleep and insomnia symptoms with DNAm-based PAI-1, and between short sleep and DNAm-based GDF-15, B2M, and Cystatin C. DNAm-based PAI-1 estimate is associated with both cardiovascular and metabolic disorders,(54) inflammation and cellular senescence.(55) GDF-15 is associated with age-related mitochondrial dysfunction and inflammation, and a novel biological aging marker,(56,57) while B2M and cystatin C are linked to inflammation and kidney function, and have also been linked with cognitive decline and cardiovascular disease.(57–62) These analyses support a body of literature suggesting insomnia symptoms and short sleep influence age-related disease risk through inflammatory and cell aging pathways,(25,63) though further analysis of the specific pathways and how they are associated with both sleep and health outcomes or epigenetic aging are needed.

In our sensitivity analysis stratifying by race, the results indicated that the effects were more pronounced among individuals identifying as Black or Hispanic. This is consistent with data showing the prevalence of short sleep duration and insomnia symptoms are higher among the Black American and Hispanic population,(64–70) and suggests that this has a greater toll on long term health. Health disparities among these groups may be driven by a number of factors. Previous studies have identified various psychosocial stressors,(71) perceived racisms and adverse environments,(71–78) are associated with poor sleep among the Hispanic and Black American populations, suggesting our measures of sleep could function synergistically with these other biobehavioral drivers of aging. Given the social-economic and health disparities among these racial/ethnic populations, and prior findings in HRS and elsewhere of an accelerated biological age,(79–81) it is plausible that these biobehavioral factors play a role in the higher prevalence of both short sleep duration, insomnia symptoms, as well as health-related outcomes that are associated with epigenetic age. As our findings were limited to ad-hoc analysis, it is important to replicate our findings in other ethnically diverse population, and identifying the underlying mechanisms that cause poor sleep is necessary to prevent aging and aging-related outcomes.

In addition to more pronounced effects among individuals identifying as Hispanic and Black, we also observed stronger effects among men compared to women. The confidence intervals are overlapping and studying larger populations with short sleep duration by sex would be necessary to confirm if this is indeed a difference in effect size. These findings are in line with a previous study which identified an association between long sleep duration and increased mortality, partially due to an increased cardiovascular risk among men, whereas no effects among women were observed.(82)

One major strength of this study is its large sample size powered to identify potential associations with accelerated epigenetic age. In addition, HRS is a representative aging population in the United States which increases generalizability. Further research will be needed to determine if effects are observed among younger individuals. One noted limitation in our analyses is that we were unable to exclude sleep disorders, as this data was not available in the entire cohort. However, our sensitivity analyses suggest results are similar among the subset who provided sleep disorders data, the observed effects are likely not driven solely by underlying sleep disorders. In addition, we were unable to determine night shift work status, a known risk factor for cancer and other age-related disease.(83–85) However, given the average age of the HRS population and that they are predominantly retired (75%), prevalence of shiftwork is likely low. In addition, sensitivity analyses among the retired population yielded similar findings to the overall sample, making it unlikely that this contributed to our results.

We observed an association between long sleep duration and epigenetic age acceleration. Other studies have shown that individuals who report long sleep duration are at increased risk for mortality, cardiovascular disease, and cognitive decline.(86–91) Several hypotheses have been described in the literature. First, this association may be due to reverse causation, those who report longer sleep duration already have health problems leading to higher sleep demands.(91) Second, the higher sleep demand is to counteract poor sleep quality, and this poor sleep quality in and of itself is associated with worse long-term outcomes.(91,92) Various other hypotheses have been proposed, including, but not limited to physiological changes in immunological processes, changes in photoperiod, psychological, and underlying medical conditions, among others.(91)

As this study was derived from secondary data analysis, we were constrained to relatively limited information about their sleep. Insomnia was measured using a questionnaire, and although insomnia is diagnosed through self-report,(93,94) there remains the potential that some individuals were misclassified. Similarly, sleep duration was estimated by self-report of time in bed. Self-reported sleep duration is less reliable than objectively assessed sleep, although it is routinely related to health outcomes in epidemiological samples.(96) Given these limitations, especially when asking about hours in bed, we may have underestimated actual hours of sleep and don’t have information about sleep efficiency. These limitations may have resulted in measurement bias, potentially skewing results away from null.

The prevalence for the sleep characteristics were within the range of what has been previously reported.(1–4,6,7) However, the prevalence of healthy sleep, defined as a sleep duration between 7 and 8.5 hours and no insomnia symptoms, was relatively low in this population (24%). This emphasizes that insomnia symptoms or poor sleep duration are highly prevalent. Further evaluation of the potential reason for this is warranted. Furthermore, the prevalence of insomnia symptoms was very high among those who reported a short sleep duration (33.8%). Though we did not find any statistically significant interaction between short sleep and insomnia symptoms, we did observe an increased epigenetic age among those who reported a combination of short sleep and insomnia symptoms, whereas no effect was observed among individuals with insomnia symptoms with a normal sleep duration suggestive of an interaction. Also, we observed an increased GrimAge, but no increased DunedinPACE, among individuals who reported short sleep duration without insomnia symptoms, thereby suggesting that the combination of short sleep duration and insomnia symptoms is indeed associated with an increased pace of aging, whereas short sleep duration may in and off itself be associated with GrimAge

Another limitation is the nature of the timing of the sleep data with the blood collection. Sleep questionnaires are derived from two separate interview periods, where one half of this study sleep data occurred two years prior to the blood draw, with the remaining occurring at the same time interval. In addition, the possibility of reverse causation cannot be completely excluded, and those with an older biological age may develop poorer sleep, for example secondary insomnia, due to underlying health issues.(91)

In this study, we used two different types of covariate adjustments. The full adjustment model included obesity status. Obesity could both be considered a confounder, where obesity status is associated with sleep characteristics, as well as a mediator given that short sleep duration and insomnia symptoms are associated with metabolic and dietary changes.(97,98) Adjusting for obesity status did attenuate our results, though most results were still statistically significant. Hence, though obesity could potentially be a mediator, confounder or a potential proxy for apnea-related sleep disorders, most of the effect is not dependent on obesity status. More advanced longitudinal analysis using advanced methods could further elucidate the role of obesity in the pathway between short sleep and insomnia symptoms and epigenetic age acceleration.

Given the high prevalence of sleep disorders in the aging population and the identified effect of short sleep duration and insomnia symptoms on accelerated epigenetic age, this warrants further attention to diagnosis and treatment of these sleep disorders. This study only examined the effect of sleep on epigenetic age acceleration. Cognitive behavior therapy has been shown to be effective in treating clinical insomnia (99,100) and decreasing various inflammatory and other biomarkers associated with biological aging.(101–103) In addition, treatment for obstructive sleep apnea has been shown to partially reverse epigenetic aging.(34) This would suggest that treatment may be able to reverse the effects of epigenetic age acceleration and thereby potentially decreasing mortality and morbidity risk.

Conclusion:

Our study found that short sleep duration and insomnia symptoms in an older population is associated with an accelerated GrimAge and a faster pace of epigenetic aging via DunedinPACE, which indicates that these individuals may be at an increased risk for age related disease and early mortality. Notably, these effects were stronger amongst individuals identifying as Black or Hispanic, pointing to an enhanced impact of poor sleep on health in these groups.

Abbreviations:

DNAm

DNA methylation

HRS

Health and Retirement Study

N

Number

BMI

Body Mass Index

%

Percentage

SD

Standard deviation

TIMP1

Tissue inhibitor metalloproteinases 1

ADM

adrenomedullin

PAI-1

Plasminogen Activator Inhibitor-1

GDF15

growth differentiation factor 15

B2M

beta-2-microglobulin

CI

Confidence Interval