Paradoxical Effects of Prolonged Insufficient Sleep on Lipid Profile: A Pooled Analysis of 2 Randomized Trials

Abstract

Background

Insufficient sleep is associated with increased cardiovascular disease risk, but causality is unclear. We investigated the impact of prolonged mild sleep restriction (SR) on lipid and inflammatory profiles.

Methods and Results

Seventy‐eight participants (56 women [12 postmenopausal]; age, 34.3±12.5 years; body mass index, 25.8±3.5 kg/m²) with habitual sleep duration 7 to 9 h/night (adequate sleep [AS]) underwent two 6‐week conditions in a randomized crossover design: AS versus SR (AS–1.5 h/night). Total cholesterol, low‐density lipoprotein cholesterol (LDL‐C), high‐density lipoprotein cholesterol, triglycerides, and inflammatory markers (CRP [C‐reactive protein], interleukin 6, and tumor necrosis factor‐α) were assessed. Linear models tested effects of SR on outcomes in the full sample and by sex+menopausal status (premenopausal versus postmenopausal women+men). In the full sample, SR increased high‐density lipoprotein cholesterol compared with AS (β=1.2±0.5 mg/dL; P=0.03). Sex+menopausal status influenced the effects of SR on change in total cholesterol (P‐interaction=0.04), LDL‐C (P‐interaction=0.03), and interleukin 6 (P‐interaction=0.07). Total cholesterol and LDL‐C decreased in SR versus AS in premenopausal women (total cholesterol: β=−4.2±1.9 mg/dL; P=0.03; LDL‐C: β=−6.3±2.0 mg/dL; P=0.002). Given paradoxical effects of SR on cholesterol concentrations, we explored associations between changes in inflammation and end point lipids under each condition. Increases in interleukin 6 and tumor necrosis factor‐α during SR tended to relate to lower LDL‐C in premenopausal women (interleukin 6: β=−5.3±2.6 mg/dL; P=0.051; tumor necrosis factor‐α: β=−32.8±14.2 mg/dL; P=0.027).

Conclusions

Among healthy adults, prolonged insufficient sleep does not increase atherogenic lipids. However, increased inflammation in SR tends to predict lower LDL‐C in premenopausal women, resembling the “lipid paradox” in which low cholesterol associates with increased cardiovascular disease risk in proinflammatory conditions.

Registration

URL: https://www.clinicaltrials.gov; Unique identifiers: NCT02835261, NCT02960776.

Nonstandard Abbreviations and Acronyms

AS

adequate sleep

SR

sleep restriction

TST

total sleep time

Research Perspective.

What Is New?

• Six weeks of 1.5 h/night sleep restriction reduced low‐density lipoprotein cholesterol and increased high‐density lipoprotein cholesterol and differentially affected circulating levels of the inflammatory cytokine interleukin 6 in premenopausal women versus postmenopausal women+men.

• Under conditions of short sleep, increases in interleukin 6 and tumor necrosis factor‐α tended to predict lower low‐density lipoprotein cholesterol in premenopausal women, suggesting a maladaptive lipid response to sleep restriction.

What Question Should Be Addressed Next?

• Future research should elucidate the mechanisms underlying the effects of long‐term insufficient sleep on lipid profile and inflammation and determine sex and life stage differences.

Sleep is a restorative process that plays an important role in human physiology and behavior and can influence a broad range of health outcomes. ¹ Short sleep is strongly associated with increased risk for hypertension, coronary heart disease, and stroke, ² leading the American Heart Association to include sleep duration as the eighth metric of cardiovascular health in Life's Essential 8. ³ Elucidating the pathways linking insufficient sleep with cardiovascular diseases (CVDs) is key to developing effective strategies to promote cardiovascular health. Although hyperlipidemia and inflammation are key factors in the pathogenesis of CVD, ⁴ , ⁵ it is unclear whether long‐term insufficient sleep, a lifestyle pattern that affects over one‐third of US adults, ⁶ adversely affects lipid and inflammatory profiles.

Studies investigating the impact of sleep duration on lipid profile have used an extreme, short‐term sleep curtailment model (1–14 days) and report inconsistent findings. An early study observed increased total cholesterol (TC) and low‐density lipoprotein cholesterol (LDL‐C) in response to severe sleep restriction (SR) of ≤4 h/night for 3 nights in middle‐aged women. ⁷ Similarly, 40 hours of total sleep deprivation altered plasma lipid species, indicative of dysregulated lipid metabolism. ⁸ Although these studies suggest that short sleep could lead to hyperlipidemia, we observed no change in lipid profile after 5 nights of restricting sleep to ≤4 h/night under controlled feeding conditions in young, normal weight adults. ⁹ Conversely, short‐term episodes of SR have been found to reduce triglycerides ¹⁰ , ¹¹ , ¹² and LDL‐C levels. ¹³ The heterogeneity across studies in participant characteristics and duration and intensity of experimental SR precludes firm conclusions about the effects of sleep curtailment on lipid profile. In addition, the severe SR paradigm used in these studies may not represent “real‐life” patterns of sleep curtailment. Instead, the predominant pattern of insufficient sleep appears to be long‐term, mild sleep curtailment attributable to maintaining work/life balance in modern societies. ¹⁴

In addition to lipid alterations, chronic inflammation is an important contributor to the pathogenesis of cardiometabolic diseases ¹⁵ , ¹⁶ and may underlie a possible link between insufficient sleep, dyslipidemia, and CVD. Experimental short‐term SR increases circulating levels of inflammatory cytokines, ¹⁷ and a network analysis of genetic pathways affected by short sleep implicates inflammation as a mediator of altered lipid profile in response to sleep loss. ¹⁸ Furthermore, we recently found that mild SR over 6 weeks promotes endothelial cell inflammation in healthy women. ¹⁹ However, the effects of a common sleep pattern of prolonged, mild sleep curtailment on lipid profile and systemic inflammatory markers have not been investigated.

Considering that >1 in 3 US adults reports sleeping less than the recommended minimum of 7 h/night, ⁶ and the strong association of insufficient sleep with CVD risk, elucidating a causal impact of prolonged insufficient sleep on established CVD risk factors is urgently needed. We evaluated changes in lipid and inflammatory profiles in response to 6 weeks of mild SR in healthy adults with adequate habitual sleep duration (adequate sleep [AS]) and explored potential sex and menopause‐based differences.

Methods

Data or full trial protocols will be provided upon reasonable request to the corresponding author (M.‐P.S.‐O.).

Study Overview

This investigation consisted of 2 clinical trials designed to evaluate the impact of long‐term, mildly insufficient sleep on cardiometabolic risk and performance. Both trials used randomized crossover designs with identical sleep interventions and enrolled healthy adults with adequate habitual sleep duration ≥7 h/night. The trials had similar eligibility criteria; however, 1 enrolled only women as part of an American Heart Association Go Red for Women Strategically Focused Research Network (NCT02835261), whereas the other enrolled both men and women (NCT02960776). Aside from this and corresponding sex‐specific criteria, eligibility criteria were identical between the studies. Both trials were preregistered on ClinicalTrials.gov.

Enrolled participants underwent 2 outpatient 6‐week study phases representing conditions of AS or mild SR. Study phases were separated by a 2‐ to 6‐week washout period. During each phase, sleep was objectively monitored daily using wrist actigraphy, and compliance with the sleep protocol was verified weekly. At baseline, midpoint, and end point of both study phases, blood samples were collected, from which circulating levels of lipids and inflammatory markers were quantified. All study procedures were conducted at Columbia University Irving Medical Center (New York, NY) and approved by its institutional review board. Participants provided written informed consent before taking part in any study procedures, and prespecified financial remuneration was provided throughout the study.

Participants and Procedures

Recruitment and Screening

Adults from the New York City area, aged 20 to 75 years, with overweight/obesity (body mass index [BMI], 25–35 kg/m²), or without overweight/obesity (BMI, 20–24.9 kg/m²), and ≥1 parent with cardiometabolic disease, but otherwise good cardiometabolic health were recruited through print and online advertisements. Screening and recruitment procedures did not differ between the 2 trials comprising this pooled analysis. Notably, the full sample of participants was recruited from the same geographic location and, therefore, is equally representative of the broader population of the area.

Individuals interested in the study contacted our staff to review initial eligibility criteria and learn more about the study. Those who met initial eligibility criteria were scheduled for an in‐person screening visit. At this screening visit, the consent form was reviewed with a trained member of the research staff, and participants had the opportunity to ask questions. After providing written consent, height and weight were measured, with shoes and outer garments removed, using a calibrated research‐grade scale with stadiometer (Tanita WB‐3000; Tanita Corporation of America, Inc, Arlington Heights, IL). Measures were taken in duplicate, and BMI was calculated from average values. Individuals falling within the acceptable BMI range were asked to provide detailed information about their demographics, health history, medication use, and menopausal status (defined as 12 consecutive months of amenorrhea) and to complete a variety of questionnaires to address additional screening criteria. Questionnaires included the Berlin Questionnaire (measure of sleep apnea risk), ²⁰ Morningness‐Eveningness Questionnaire (measure of chronotype), ²¹ Pittsburgh Sleep Quality Index, ²² Beck Depression Inventory‐II, ²³ and a modified version of the Caffeine Consumption Questionnaire. ²⁴ Exclusion criteria, assessed via these forms and questionnaires, included the following: diagnosis of cardiovascular or metabolic diseases; diagnosis of a sleep or psychiatric disease or disorder; contraindications for any of the procedures; depression; pregnancy; active lactation or <1 year postpartum; use of hormonal contraceptives or hormone replacement therapy; use of medications known to affect sleep; recent weight loss; smoking; history of drug or alcohol abuse; excessive caffeine consumption (>300 mg/d); high risk for sleep apnea; engaging in shift work; definite evening or morning chronotype (Morningness‐Eveningness Questionnaire scores of 16–30 or >69); poor sleep quality (Pittsburgh Sleep Quality Index score of >5); and recent or planned travel across time zones. In addition, individuals signed an agreement to not operate a vehicle during the SR phase of the study. Men and women meeting all eligibility criteria were invited to participate in a 2‐week sleep screening.

The final stage of screening was a 2‐week home‐based assessment of habitual sleep patterns. Individuals were provided an Actigraph GT3X+ triaxial accelerometer (Actigraph LLC, Pensacola, FL), which they were instructed to wear on their nondominant wrist at all times aside from bathing or swimming. This wrist‐worn device is validated for sleep measurement, and application of the Cole‐Kripke scoring algorithm (used in this study) to the raw actigraphy data provides comparable estimates of total sleep time (TST) to polysomnography, the gold standard measure of sleep. ²⁵ To enhance the accuracy of sleep measurement, participants completed sleep diaries concurrently. Individuals with average TST of 7 to 9 h/night or sleeping ≥7 h/night on at least 70% of nights were enrolled in the study (Figure 1).

Figure 1. The CONSORT flow diagram for this randomized crossover study with 2 sleep conditions. CONSORT indicates Consolidated Standards for Reporting of Trials.

Study Procedures

For enrolled participants, the 2‐week sleep screening period provided a representative assessment of their habitual sleep patterns, which were used to create personalized sleep schedules for each condition. For AS, target bed and wake times were designed to approximate average times recorded during the screening period. For SR, a nightly reduction in TST of ≈1.5 hours was induced by delaying bedtimes relative to the average bedtimes recorded during screening while maintaining constant wake times. Study phases were separated by a multiweek washout period to: (1) minimize any carryover effects on sleep before beginning the second phase and (2) ensure that women were in the same phase of their menstrual cycle at the start of each study phase.

Following the sleep screening, enrolled participants were randomized to a study condition (coded as “D” or “V”) order using a computer‐based random‐sequence generator. Participants were informed of their order of sleep conditions on completion of the phase 1 baseline visit, at which time target bed and wake times were determined. Participants were required to always wear the Actigraph device throughout each phase and to complete nightly sleep diaries. Study staff met with participants on a weekly basis to review sleep data and determine compliance. If necessary, adjustments were made to the sleep schedule to maintain the target TST for each condition.

Throughout the study, metabolic assessments were scheduled to occur in the morning with participants in the fasted state. At baseline, week 3, week 4 (n=47), and end point visits of each study phase, participants were escorted to the Irving Institute for Clinical and Translational Research at Columbia University Irving Medical Center for blood collection. Whole blood was collected into EDTA‐treated tubes by a trained phlebotomist and spun for 20 minutes at 4 °C and 16.1g following collection. The resulting plasma was stored at −80 °C until biomarker analyses were performed. Although lipid profile was a primary outcome for 1 of the trial registrations, both trials included assessment of immune function as an outcome of interest. Given the shared overarching goal of the trials to study the effects of mild SR on cardiometabolic health outcomes and to increase power to detect overall and sex+menopause‐specific effects, lipid and inflammatory profiles were assessed in both trials.

Outcome Measures

Lipid and inflammatory profiles were determined from fasting plasma samples. Baseline samples represent the participants' usual state, before any change in sleep, whereas all other time points represent the influence of the sleep intervention with increasingly prolonged time spent in SR and in AS. Circulating levels of TC, LDL‐C, high‐density lipoprotein cholesterol (HDL‐C), and triglycerides were measured via liquid chromatography–mass spectrometry (Agilent 6490 Triple Quadrupole MS integrated with an Agilent 1260 Infinity LC system). CRP (C‐reactive protein), interleukin 6 (IL‐6), and tumor necrosis factor‐α (TNF‐α) were analyzed via analyzer (CRP) and ELISA (IL‐6, TNF‐α). All assays and quantification of lipids were performed by the Biomarkers Core Laboratory of the Irving Institute for Clinical and Translational Research.

Statistical Analysis

Participant characteristics are presented as mean±SD for continuous variables and number (percentage) for categorical variables. Primary outcomes of interest for this pooled analysis were plasma levels of lipids (TC, LDL‐C, HDL‐C, and triglycerides) and cytokine mediators of inflammation (CRP, IL‐6, and TNF‐α). All analyses were performed by the study statistician, who was blinded to the study conditions. To test the effect of sleep condition on change in outcomes, repeated‐measures linear‐mixed models (PROC MIXED in SAS v9.4), adjusted for the baseline value of the outcome variable, were used. Study condition (SR versus AS), week, age, and sex were treated as fixed effects in models, whereas subject was a random effect. Effect sizes are presented as β±SE. To confirm that the pattern of results for main effects did not differ significantly by study, we also included an interaction of study condition with trial (NCT02835261 versus NCT02960776). The term was dropped when nonsignificant. Because an aim of the overall project was to evaluate sex differences in response to short sleep, we also evaluated whether sex+menopausal status influenced effects of SR on outcomes. Where interactions were significant, analyses were stratified by premenopausal women only and postmenopausal women+men. We chose to group participants in this way given established differences in sex hormones between premenopausal and postmenopausal women and the corresponding increased CVD risk profiles of postmenopausal women and men relative to premenopausal women. ²⁶ Furthermore, this method of grouping allowed for more balanced numbers in stratified analyses, as premenopausal women were enrolled in both trials, whereas postmenopausal women and men were enrolled in 1 of the 2 trials (postmenopausal women in NCT02835261 and men in NCT02960776). Statistical models for stratified analyses were the same as those described for the full sample, but sex was not included as a fixed effect in the model.

Post hoc analyses using linear models were conducted to explore the relationship between change in inflammatory markers with lipid outcomes under conditions of AS and SR. The goal of this post hoc analysis was to gain mechanistic insight into paradoxical effects of SR on lipid profile. Certain proinflammatory conditions are characterized by low cholesterol levels, which represent a maladaptive response to inflammation. ²⁷ Given our previous work demonstrating that mild SR increases activation of nuclear factor‐κB, ¹⁹ which upregulates transcription of proinflammatory genes, ²⁸ we explored whether changes in circulating levels of inflammatory markers under different sleep conditions predicted lipid outcomes. Exposure variables were the change in CRP, IL‐6, and TNF‐α from baseline to end point, and outcomes were end point measures of TC, LDL‐C, and HDL‐C. Associations of each inflammatory marker with the different outcomes were evaluated in separate models. Given consistent effects of sex+menopausal status on the impact of SR on outcomes, these exploratory analyses were conducted separately among premenopausal women and postmenopausal women+men. In initial models, we included sleep condition (SR versus AS) as a fixed effect along with an interaction term between sleep condition and change in inflammatory marker to formally evaluate whether exploratory associations differed between AS and SR. However, given limited power to detect interactions in a stratified sample, we also evaluated associations of change in inflammatory markers with end point lipids separately in each of the 2 sleep conditions. All participants with outcome data were included in relevant analyses. All analyses were conducted using SAS 9.4 statistical software (SAS, Cary, NC). For primary analyses (ie, main effects of sleep condition on primary outcomes) for which we had a priori hypotheses, results were considered statistically significant at P<0.05. For post hoc analyses, a Bonferroni correction for potential type I error attributable to multiple testing was applied. Given 3 separate exposures, results were considered statistically significant at P<0.02.

Results

Participant Characteristics

Data were available from 78 participants (59% minority race and 29% Hispanic ethnicity) who completed the study (ie, completed both phases or withdrew following completion of a full phase [n=5]) between August 2016 and May 2021. Of those 78 participants, 70 (90%) completed the study before a pause in research activities from March to July of 2020 because of the COVID‐19 pandemic. Descriptive characteristics of participants are provided in Table 1. Twelve women (21%) were postmenopausal. Per protocol, average BMI at baseline was in the overweight range, whereas TC, LDL‐C, HDL‐C, and triglycerides were all within normal ranges (Table 1).

Table 1.

Descriptive Characteristics of the Study Participants at Baseline

Characteristic	Full sample (N=78)	Premenopausal women (N=44)	Postmenopausal women+men (N=34)
Age, y	34.3±12.5	29.3±5.8	40.9±15.6
Race, N (%)
White	32 (41)	18 (41)	14 (41)
Black	20 (26)	13 (29)	7 (21)
Asian	17 (22)	11 (25)	6 (17)
All other races/unknown	9 (11)	2 (5)	7 (21)
Ethnicity, N (%)
Non‐Hispanic	55 (71)	33 (75)	22 (65)
Hispanic	23 (29)	11 (25)	12 (35)
Body mass index, kg/m2	25.8±3.5	25.1±3.0	26.8±3.8
Total cholesterol, mg/dL	169.0±29.0	159.1±21.6	182.6±32.6
LDL cholesterol, mg/dL	89.8±26.7	60.6±14.8	100.9±29.4
HDL cholesterol, mg/dL	56.2±14.8	82.0±21.8	50.1±12.6
Triglycerides, mg/dL	85.5±54.3	69.7±24.9	107.3±73.6
CRP, mg/dL	1.86±2.12	1.66±2.16	2.16±2.07
Interleukin 6, pg/mL	1.23±0.92	1.08±0.72	1.42±1.12
Tumor necrosis factor‐α, pg/mL	0.73±0.26	0.66±0.29	0.83±0.15
Sleep duration, min	457.3±22.5	459.5±22.2	454.4±22.8
Midpoint of sleep*	03:49±1:21	03:44±1:24	03:57±1:16
Sleep efficiency, %	91.1±2.8	91.7±2.8	90.3±3.1
Sleep fragmentation index	23.5±6.2	23.6±6.5	23.5±6.0

Values are presented as mean±SD for continuous variables and count (percentage) for categorical variables. CRP indicates C‐reactive protein; HDL, high‐density lipoprotein; and LDL, low‐density lipoprotein.

^*Mean values are clock times, and SD values are duration in hours and minutes.

Sleep Data

Participants had adequate nightly sleep duration during the 2‐week actigraphy screening (457.3±22.5 minutes) and Pittsburgh Sleep Quality Index scores indicative of good sleep quality (3.0±1.8). As was expected given the stringent sleep screening criteria, habitual sleep duration was nearly identical between those participating before versus after the halt in research activities attributable to COVID‐19 pandemic (449.3±11.9 versus 449.2±12.7 min/night). Fidelity to the sleep protocol in this outpatient study was excellent. Over the 6 weeks of SR, mean±SE of TST was 368.6±2.0 min/night compared with 449.2±1.9 min/night in SR, representing a mean reduction of 80.6±1.6 min/d in SR relative to AS (P<0.001; Figure S1). During SR, 49% of participants achieved an average reduction in TST from screening ≥90 min/night, whereas 74% achieved a reduction in TST ≥80 min/night.

Lipid Profile

In the full sample, HDL‐C increased (β=1.2±0.5 mg/dL; P=0.03) and LDL‐C decreased (β=−3.3±1.6 mg/dL; P=0.03) in SR relative to AS, whereas triglycerides and TC were unaffected by sleep condition (both P≥0.50; Figure 2). Results did not differ by trial. Notably, sex+menopausal status influenced the relationship between sleep condition and TC (P‐interaction=0.04; Figure 2E) and LDL‐C (P‐interaction=0.03; Figure 2F). In analyses stratified by sex+menopausal status, SR reduced TC (β=−4.2±1.9 mg/dL; P=0.03) and LDL‐C (β=−6.3±2.0 mg/dL; P=0.002) relative to AS in premenopausal women but not in postmenopausal women+men (TC: β=3.1±2.8 mg/dL; P=0.26; LDL‐C: β=0.4±2.4 mg/dL; P=0.88) (Figure 2).

Figure 2. Mean±SEM levels of circulating TC (A), LDL‐C (B), HDL‐C (C), and TG (D) across weeks under conditions of AS and mild SR in the full sample.

Across women of both menopausal statuses and men, HDL‐C was increased in SR relative to AS (P=0.03). Sex+menopausal status influenced the relationship between sleep condition and change in TC (P‐interaction=0.04; E) and LDL‐C (P‐interaction=0.03; F). Stratified analyses showed that TC (E) and LDL‐C (F) decreased in SR relative to AS among premenopausal women (Pre; TC: P=0.03; LDL‐C: P=0.002) but not postmenopausal women+men (Post+Men; both P>0.20). AS indicates adequate sleep; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; SR, sleep restriction; TC, total cholesterol; and TG, triglycerides.

Markers of Inflammation

Levels of inflammatory markers CRP, IL‐6, and TNF‐α did not differ between SR and AS in the full sample (Figure 3), and results did not differ by trial. However, an interaction of sex+menopausal status with condition approached significance for IL‐6 (P‐interaction=0.069), which increased in SR relative to AS in premenopausal women compared with postmenopausal women+men. In analyses stratified by sex+menopausal status, effects of SR on IL‐6 did not reach statistical significance (premenopausal women: β=0.2±0.2 pg/mL; P=0.29; postmenopausal women+men: β=−0.2±0.1 pg/mL; P=0.20).

Figure 3. Mean±SEM levels of circulating proinflammatory cytokines CRP (A), TNF‐α (B), and IL‐6 (C) across weeks under conditions of AS and mild SR in the full sample.

Sleep condition did not significantly affect any markers of inflammation in the full sample. There was a trend toward a significant influence of sex+menopausal on the relationship between sleep condition and IL‐6, which increased in SR relative to AS in premenopausal women (Pre) compared with postmenopausal women+men (Post+Men) (P‐interaction=0.069; D). AS indicates adequate sleep; CRP, C‐reactive protein; IL‐6, interleukin 6; SR, sleep restriction; and TNF‐α, tumor necrosis factor‐α.

Associations of Markers of Inflammation With Lipids

Given the observed effects of SR on cholesterol levels and variations in inflammatory markers, we explored possible associations between change in inflammation and end point lipid levels (Table 2). First, to formally evaluate associations of change in inflammatory markers with end point lipids in the different sex+menopausal status groups that differed by sleep condition, we tested the interaction of sleep condition with change in inflammatory markers on the primary outcomes. In premenopausal women, increases in TNF‐α were associated with lower TC after SR relative to AS (β‐interaction=−88.7±30.5 mg/dL; P‐interaction<0.005), whereas in postmenopausal women+men, increases in TNF‐α were associated with increased TC in SR relative to AS (β‐interaction=207.6±88.9 mg/dL; P‐interaction=0.024). No other interactions between condition and change in inflammatory marker on lipid outcomes were observed for either premenopausal or postmenopausal women+men.

Table 2.

Associations of Change in Inflammatory Markers With Cholesterol Levels at End Point During Each Study Condition in the Full Sample and Stratified by Sex+Menopausal Status

Exposure	Full sample		Premenopausal women		Postmenopausal women+men
	SR	AS	SR	AS	SR	AS
Outcome: total cholesterol (mg/dL) at end point
ΔTNF‐α, pg/mL	−13.4±18.3, 0.46	−12.8±36.0, 0.72	−19.4±13.6, 0.16	69.3±27.3, 0.016	98.8±54.0, 0.08	−108.7±71.0, 0.14
ΔIL‐6, pg/mL	−3.5±3.3, 0.29	−2.5±5.2, 0.62	−1.7±2.5, 0.50	−0.8±5.8, 0.89	0.3±9.6, 0.97	−3.3±8.3, 0.70
ΔCRP, mg/dL	1.7±2.9, 0.56	0.7±2.5, 0.78	2.2±2.1, 0.30	−2.1±1.7, 0.23	−6.3±9.5, 0.52	9.6±7.3, 0.20
Outcome: LDL cholesterol (mg/dL) at end point
ΔTNF‐α, pg/mL	−22.7±17.6, 0.20	−7.0±27.1, 0.80	−32.8±14.2, 0.027*	3.0±27.3, 0.91*	93.8±51.5, 0.08*	−7.6±52.3, 0.88*
ΔIL‐6, pg/mL	−4.4±3.2, 0.18	0.8±4.0, 0.84	−5.3±2.6, 0.051	−1.8±5.3, 0.73	10.3±9.0, 0.26	2.2±6.0, 0.72
ΔCRP, mg/dL	3.4±2.9, 0.25	1.6±1.9, 0.40	3.8±2.3, 0.11	0.1±1.6, 0.94	−2.7±9.9, 0.79	6.0±5.2, 0.26

Outcome: HDL cholesterol (mg/dL) at end point
ΔTNF‐α, pg/mL	−4.2±7.8, 0.59	15.2±13.9, 0.28	−7.9±8.8, 0.37	21.6±18.7, 0.26	−15.0±17.8, 0.41	−2.9±18.4, 0.87
ΔIL‐6, pg/mL	−0.5±1.4, 0.71	0.6±2.0, 0.78	−0.3±1.6, 0.87	1.9±3.7, 0.62	−6.8±2.8, 0.020	−0.1±2.1, 0.97
ΔCRP, mg/dL	−0.5±1.3, 0.70	−2.8±0.9, 0.003	−0.3±1.4, 0.83	−2.5±1.1, 0.026	−0.3±3.2, 0.93	−3.1±1.7, 0.09

Values are presented as β±SE (per 1‐unit increase in exposure), P value. AS indicates adequate sleep; CRP, C‐reactive protein; HDL, high‐density lipoprotein; IL‐6, interleukin 6; LDL, low‐density lipoprotein; SR, sleep restriction; and TNF‐α, tumor necrosis factor‐α.

^*Interaction of sleep condition with change in inflammatory marker (premenopausal women: P=0.005; postmenopausal women+men: P=0.02).

Among premenopausal women, exploratory analyses stratified by sleep condition showed that change in CRP tended to be inversely related to end point HDL‐C during AS (β=−2.5±1.1 mg/dL; P=0.026) but not SR (β=−0.3±1.4 mg/dL; P=0.83). Inverse associations of IL‐6 with LDL‐C failed to reach statistical significance but were stronger in SR than AS (SR: β=−5.3±2.6 mg/dL; P=0.051 versus AS: β=−1.8±5.3 mg/dL; P=0.73). Changes in TNF‐α were marginally associated with end point LDL‐C in SR (β=−32.8±14.2 mg/dL; P=0.027) but not AS (β=3.0±27.4 mg/dL; P=0.91). Changes in TNF‐α were positively associated with end point TC during AS (AS: β=69.3±27.3 mg/dL; P=0.016) but not SR (SR: β=−19.3±13.6 mg/dL; P=0.16).

Among postmenopausal women+men, exploratory analyses stratified by sleep condition demonstrated that increasing IL‐6 during SR tended to predict lower end point HDL‐C (β=−6.8±2.8 mg/dL; P=0.020), with no association noted in AS (β=−0.1±2.1 mg/dL; P=0.97). Associations of change in TNF‐α during SR with end point TC and LDL‐C in postmenopausal women+men failed to reach statistical significance (TC: β=98.8±54.0 mg/dL; P=0.080; LDL‐C: β=93.8±51.5 mg/dL; P=0.081).

Discussion

The major findings of this study are that 6 weeks of mild SR, mimicking “life‐like” suboptimal sleep habits, reduced levels of TC and LDL‐C, changes that would be considered favorable for cardiovascular health. ³ However, decreases in LDL‐C in response to SR were concomitant with increases in markers of systemic inflammation in premenopausal women relative to postmenopausal women+men. These findings, paired with trending inverse associations between changes in inflammatory markers and LDL‐C after SR, suggest a potential maladaptive lipid response, akin to those reported in proinflammatory conditions, such as rheumatoid arthritis and celiac disease. ²⁷ , ²⁹

Previous studies have reported inconsistent effects of a short‐term, severe SR on cholesterol levels. ⁷ , ⁹ , ³⁰ Using prolonged, mild sleep curtailment, which aligns more closely with sleep habits observed in the general population, we find that LDL‐C is reduced during prolonged SR, whereas HDL‐C is increased, concordant with findings of 2 previous studies of 5 and 14 days of sleep curtailment to 4 ¹⁸ and 5.5 h/night, ¹³ respectively. Results of a multi‐omics analysis of the effects of sleep loss may help to elucidate mechanisms underlying these findings. Specifically, short sleep decreased the expression of genes encoding cholesterol transporters and increased expression in pathways involved in inflammatory responses. ¹⁸ In conditions of short sleep, genes encoding toll‐like‐receptor 4 and TNF‐α are upregulated, which suppresses activity of liver X receptor, ³¹ a regulator of reverse cholesterol transport. ³² Network analyses indicate that SR reduces LDL‐C via inflammation‐induced suppression of liver X receptor. ¹⁸ Increases in HDL‐C in response to prolonged SR in this study may also represent a similar maladaptive response to inflammation; the composition and functional capacity of the HDL complex is altered in inflammatory states, and proinflammatory HDL is elevated and correlates directly with disease activity. ³³ , ³⁴ The present study cannot confirm a direct pathway linking SR‐induced inflammation with changes in composition and function of cholesterol fractions; however, initial findings from this investigation and others ¹⁹ , ³⁵ suggest a role of inflammation in the observed effects of SR on lipid profile, which could heighten CVD risk. Thus, elucidating the mechanistic underpinnings of these effects of SR should be a priority of future studies.

We recently reported that prolonged SR increased endothelial inflammation, a key step in the development and progression of CVD, in healthy women. ¹⁹ Here, we find that this model of long‐term, mildly insufficient sleep also lowers circulating levels of LDL‐C while increasing markers of systemic inflammation in premenopausal women relative to postmenopausal women+men. Reductions in LDL‐C in response to positive lifestyle changes can improve cardiometabolic risk profile and are recommended for promotion of cardiovascular health. ³ , ³⁶ However, reductions in circulating cholesterol in response to inflammation are considered maladaptive because this phenotype is associated with worse health outcomes in chronic inflammatory diseases. ²⁷ , ³⁷ , ³⁸ This is particularly true of autoimmune diseases, which are characterized by dysfunctional hyperactivity of the immune system. ³⁹ In rheumatoid arthritis, an increase in inflammatory markers is accompanied by low levels of TC and its fractions, whereas reductions in inflammation are associated with increases in cholesterol levels. ⁴⁰ , ⁴¹ , ⁴² Similar inverse associations of inflammatory markers with cholesterol levels have been reported in other inflammatory conditions, including lupus, ⁴³ cancer, ⁴⁴ sepsis, ⁴⁵ , ⁴⁶ heart failure, ⁴⁷ and postmyocardial infarction, ⁴⁸ with greater mortality in those with lower cholesterol levels. ⁴⁷ , ⁴⁹ , ⁵⁰ Aligning with this literature, we recently provided evidence that prolonged exposure to suboptimal sleep impacts the epigenome of hematopoietic stem and progenitor cells, increasing hematopoiesis and shifting differentiation of immune cells toward a myeloid fate, thereby increasing the likelihood of hyperinflammation. ³⁵ Indeed, we reported recently causal evidence that prolonged, mild SR leads to endothelial dysfunction and inflammation in premenopausal women. ¹⁹ Taken together, these observations suggest findings of reduced cholesterol in the current study may be akin to the lipid paradox observed in other chronic inflammatory conditions. Thus, the lipid phenotype observed in conditions of long‐term short sleep may reflect heightened CVD risk attributable to a maladaptive response to increased inflammation. ¹⁹ , ³⁵ , ⁵¹

We did not observe an impact of prolonged SR on triglycerides under free‐living conditions. Assessment of changes in triglycerides in response to SR was of interest considering that elevated triglycerides increase CVD risk. ⁵² However, lower triglycerides levels are also associated with adverse metabolic conditions, such as hyperinsulinemia. ⁵³ Indeed, short‐term SR consistently leads to reduced insulin sensitivity and glucose tolerance, ⁵⁴ an effect that has been observed with concomitant decreases in triglycerides levels ¹⁰ , ¹¹ , ¹² and that is attributed to metabolic shifts toward fatty acid oxidation and gluconeogenesis. One explanation for the discordance in findings between studies could be differences in dietary intake conditions. In our study, participants self‐selected their diet, and we expected that lipid profile would be adversely affected in part via worsening of dietary intakes. Prior research from our group and others suggests that SR leads to increased intakes of fat and refined carbohydrates. ⁵⁵ , ⁵⁶ However, the null result observed in the current study is concordant with our prior findings of no effect of SR on triglycerides levels in healthy individuals fed a controlled, well‐balanced diet. ⁹ It is thus unlikely that our findings are attributable to alterations resulting from dietary modifications. Furthermore, a meta‐analysis reported no significant association between short sleep and elevated triglycerides. ⁵⁷ Thus, hypertriglyceridemia is unlikely to contribute to CVD risk associated with insufficient sleep.

We also explored potential sex‐based differences in the effects of prolonged SR on CVD risk factors. Interestingly, sex+menopausal status was found to influence the effect of SR on TC and LDL‐C. In premenopausal women, TC and LDL‐C were reduced in SR relative to AS, with no effect on HDL‐C levels. In contrast, SR increased HDL‐C with no effect on LDL‐C in postmenopausal women+men. Analogous to our findings in humans, sleep loss in a murine model increased HDL‐C levels in males and ovariectomized females, without affecting HDL‐C in intact male/female. ⁵⁸ In addition, LDL‐C levels decreased marginally in intact female rats following SR, aligning with our observations in premenopausal women. Furthermore, short‐term experimental SR produces different effects on the lipidome based on sex, with most lipid metabolites affected by SR being elevated in men and reduced in women. ⁵⁹ Given the role of sex hormones in CVD risk ⁶⁰ , ⁶¹ and differential effects of insufficient sleep on lipid and inflammatory profiles between premenopausal women and postmenopausal women+men in this study, future research is needed to better characterize sex and menopausal status as determinants of CVD risk in insufficient sleep.

The current study is both the longest SR trial to date and 1 of the first to induce mild SR in a free‐living context. A key strength of this study is that the SR intervention brought nightly sleep duration over the 6 weeks to ≈6 hours on average, resembling what is reported by most short sleepers. ⁶ In addition, we ensured a high level of fidelity to the sleep conditions through objective continuous tracking of sleep. In addition, recruitment of men and of women of varying menopausal status allowed for exploration of differences in the response to SR based on sex+menopausal status, corresponding with differences in CVD risk profiles. ²⁶ A key goal of future research should be to elucidate the pathway linking prolonged SR with lipid and inflammatory profiles. For example, changes in physical activity or sedentary time under different conditions of sleep could contribute to effects of SR on lipid profile. Another key area of investigation will be to formally evaluate the influence of sex hormones on the relationship between long‐term short sleep and cardiometabolic outcomes.

In conclusion, prolonged, mild SR induced reductions in TC and LDL‐C in premenopausal women and increases in HDL‐C in postmenopausal women+men. Coupled with inverse associations of changes in inflammatory markers with TC and LDL‐C and recently reported adverse effects of prolonged SR on immune ³⁵ and endothelial function, ¹⁹ these data suggest a maladaptive response to inflammation in insufficient sleep, akin to the lipid paradox in other chronic inflammatory conditions. ²⁷ Elucidating the mechanisms underlying the effects of long‐term short sleep on lipid profile and inflammation is of great importance, as it may help to identify short sleepers at the greatest risk of developing CVD. Consequently, this information could be leveraged to augment, and personalize, lifestyle interventions for CVD prevention by targeting sleep health as a component of lifestyle modification.

Sources of Funding

This work was supported by American Heart Association grants 16SFRN27950012 (M.‐P. St‐Onge), 16SFRN29050000 (S. Jelic), and 16SFRN27960011 and AHA811531 (B. Aggarwal); National Institutes of Health (NIH)/National Heart, Lung, and Blood Institute grants R01HL128226 and R35HL155670 (M.‐P. St‐Onge) and R01HL106041 and R01HL137234 (S. Jelic); and a Clinical and Translational Science Award UL1 TR001873 awarded to Columbia University. R. Barragán is supported by Generalitat Valencia and Fondo Social Europeo fellowship (APOSTD/2019/136). F.M. Zuraikat is a Berrie Fellow in Diabetes and Obesity Research and is supported by NIH grants T32HL007343 and R01DK128154.

Disclosures

None.

Supporting information

Figure S1