Caloric Restriction, the Menstrual Cycle, and Sleep in Women Without Obesity

Abstract

Introduction

Short-term caloric restriction is a common practice even in lean and underweight women. We studied the impact of dietary restriction on sleep and its interplay with reproductive hormones across the menstrual cycle in women without obesity.

Methods

Seventeen healthy women without obesity, aged 23.6 ± 2.3 years (mean ± SD) underwent a neutral (± 0%) and deficient energy availability diet (−55%) in the early follicular phase of 2 menstrual cycles. Actigraphic data and urinary LH, estrone-3-glucuronide (E1G), and pregnanediol-3-glucuronide (PDG) were collected daily. Blood orexin and leptin were collected on the fifth day of each diet. Sleep was analyzed in relation to menstrual cycle phase, diet, and hormones.

Results

Decreased energy availability and menstrual cycle phase independently affected wake after sleep onset (WASO; P = .004, P = .007 for diet and cycle phase, respectively) and number of awakenings (NOA; P = .03, P = .0006, respectively) with the greatest sleep disruption in the late luteal phase. Sleep efficiency (SE) was lower and duration of awakenings was longer in association with dietary restriction. Orexin was positively associated with WASO (P = .02), the sleep fragmentation index (P = .001), and NOA (P = .009) and inversely related to SE (P = .02). Increasing PDG was associated with WASO (P < .05) and duration of awakenings (P < .05) and inversely associated with SE (P < .01). Increasing E1G was positively associated with WASO (P < .05) and NOA (P < .01).

Conclusion

Short-term modest caloric restriction independently disrupts sleep and exacerbates changes in sleep that occur across the menstrual cycle in healthy, young women without obesity.

Key Points.

• Objective sleep parameters changed across the menstrual cycle in women without obesity.

• Five days of moderate caloric restriction disrupted sleep and worsened the impact of some menstrual cycle phases on sleep.

• Sleep disruption was associated with plasma orexin-A and urinary metabolites of progesterone and estradiol.

Short- or longer term calorie reduction is widely prevalent in our society and has become increasingly popular due to suggestions of beneficial effects on longevity, cognition, some cancers, cardiovascular risks, and insulin sensitivity derived from a combination of animal and cross-sectional human studies (1-5), in addition to weight loss and maintenance. An analysis of data from the National Health and Nutrition Examination Survey collected between 1999 and 2016 revealed a steady increase in the proportion of individuals who attempted to lose weight and do so primarily through dieting (6). This behavior is not limited to overweight and obese individuals, as the same survey indicated that over a third of normal or underweight women attempted weight loss in the previous year. Popular weight loss strategies included reduced food intake, frequent water intake, and, more recently, an increase in the consumption of plant-based foods (6), although the efficacy of such dieting regimens on a population basis is uncertain, given the concomitant sustained rise in body weight and body mass index (BMI) among US adults (6).

Energy balance and sleep act in concert to modulate central and peripheral homeostasis, ensuring the maintenance of cellular function in accordance with available resources. During wakefulness, hunger dominates as a result of heightened energy demand, whereas satiety promotes restorative sleep (7). In men and women, restricted and disrupted sleep result in increased insulin resistance, reduced energy expenditure, and increased caloric intake (8, 9). There is also evidence that dramatic energy restriction may have a detrimental effect on sleep. Disrupted sleep is seen during extremes of basal metabolic rates related to starvation (10, 11) and severe thyroid hormone dysfunction (12), and sleep patterns are suppressed in starved Drosophila melanogaster mediated by neuropeptidergic noncircadian neuronal pathways (13-15). Previous studies have shown an adaptive response to dietary restriction, marked by increases in orexigenic hormones such as orexin-A and a concomitant decrease in anorexigenic peptides like leptin (16, 17), both of which act on hypothalamic neuronal centers. While assessing centrally active orexin-A and leptin in the cerebrospinal fluid is challenging, these difficulties can be overcome by measuring their plasma levels, which correlate well with cerebrospinal fluid levels in human studies (18, 19). Specifically, orexin-A relays signals to both energy-regulating and sleep/wake-regulating hypothalamic centers (20) and has been linked to disrupted sleep (16, 21), while leptin is linked to sleep homeostasis (22). The time course of recovery of neurohormonal changes that affect sleep after caloric restriction is unclear given these adaptive responses, and it is possible they persist well beyond the period of the dietary intervention. Teleologically, the disruptive impact of energy restriction on sleep may be linked to the heightened wakefulness necessary for locomotor and foraging behavior in nonhuman species in response to reduced energy availability (20). Numerous studies in individuals with obesity have noted a positive effect of energy restriction on sleep; however, relief of obstructive sleep apnea appears to be the mechanism underlying this observation (23). Short-term “yo-yo” dieting is associated with poor sleep quality (24), but controlled interventional studies of short-term dietary restriction are lacking. An important unresolved question is whether short-term moderate energy restriction has deleterious effects on sleep in individuals without obesity.

Sleep is closely linked to the reproductive hormonal axis in girls and women. During puberty, gonadotropin secretion initially occurs during nighttime sleep and a majority of LH pulses follow slow-wave sleep (25, 26). In contrast, once ovulatory cycles have been established, a decrease in LH pulse frequency is apparent during sleep, particularly during the early follicular phase (EFP) (27). Sleep disruption has been associated with the hypoestrogenism of menopause, an observation that is associated both with and without the occurrence of hot flashes (28, 29). The interrelationship between sleep and gonadal steroids is less clear during the reproductive years and has been challenging to study (28) due to the dynamic changes in steroid hormone secretion that occur across the menstrual cycle (30). There is conflicting evidence of changes in sleep across the menstrual cycle, with sleep disruption in the luteal phase and during menses reported in some (31, 32) but not all (33-36) studies, possibly due to the use of surrogate markers and/or infrequent sampling of gonadal hormones to assess the timing of ovulation. Importantly, the role that ovarian hormonal secretion across the menstrual cycle plays in sleep disruption is not well understood (37).

We hypothesized that moderate short-term energy restriction and periods of hormonal transition across the menstrual cycle are associated with sleep disruption in women without obesity. To delineate the relationship between metabolism, sleep regulation, and ovarian hormones, we studied the effect of a short-term, moderate decrease in energy availability in the EFP on actigraphy-defined objective sleep across the menstrual cycle and further examined the changes in metabolic and reproductive hormones that may mediate these effects (38).

Materials and Methods

Subjects

Thirty-eight subjects met criteria for enrollment between 2016 and 2020, of whom 17 completed both neutral energy availability (NEA) and decreased energy availability (DEA) interventions and complete actigraph and urine sample collection for reproductive hormone analyses (Fig. 1). Actigraphic sleep parameters were preplanned outcomes of interest in this study. Subjects were healthy, nonsmoking, sedentary, nulliparous, eumenorrheic (menstrual cycle lengths between 25 and 35 days) women with proven ovulatory cycles between the ages of 18 and 28 years with a BMI between 18.5 and < 27 kg/m² at the time of screening and a postmenarche duration of ≤ 14 years. Physiological characteristics varied slightly between indirect calorimetry and BodPod® assessed at least 1 month apart without differences in body weight and BMI (Supplement 1, Table 1) (39). Those with serious medical, gynecological, or endocrine conditions were excluded. We also excluded women who had a habitual energy intake of < 35 or > 55 kcal/kg lean body mass (LBM)/day determined by 4-day food diaries, as well as those who were dieting (defined as habitual energy intake of < 35 kcal/kg LBM/day and/or dietary weight loss of ≥ 2 kg in the preceding 3 months); had a V̇O­₂ max > 40 mL/kg/min or were engaging in rigorous daily exercise (defined as > 4 hours of aerobic exercise per week); or were taking prescribed or over-the-counter hormone or sleep medications, stimulants, antidepressants, homeopathic substances, herbal remedies, or supplements. All subjects had normal hemoglobin, TSH, testosterone, and prolactin levels. None of the subjects reported a history of diagnosed sleep disorder, premenstrual dysphoric disorder, endometriosis, or leiomyomas. There was no reported recent night shift work. Seven subjects reported travel over 2 or more time zones at some time within the 3 months prior to the start of their study participation until their final visit. Actigraphic analysis verified that all participants who reported travel across time zones returned to the same time zone for at least 10 days prior to starting the NEA and DEA interventions and for the entire duration of the corresponding menstrual cycles. Five subjects did not provide an answer regarding recent travel.

Figure 1.

Enrollment and participation of study subjects.

Table 1.

Participant demographic and physiologic characteristics (mean ± SD unless otherwise indicated)

Demographic characteristics
Age (y)a	23.6 ± 2.3
Age of menarche (y)a	12.6 ± 0.9
Gynecological age (y)a	11.0 ± 2.3
Race/ethnicitya
White, n (%)	6 (35.3)
Black, n (%)	6 (35.3)
Asian, n (%)	2 (11.8)
Hispanic/Latino, n (%)	3 (17.6)
Physiologic characteristics	NEA (n = 17)	DEA (n = 17)
Body weight (kg)b	60.8 ± 7.5	59.9 ± 7.2c
Body mass index (kg/m2)b	23.2 ± 2.5	22.7 ± 2.2c
Body fat (%)b	29.1 ± 6.9	28.1 ± 7.3
Lean body mass (kg)b	42.9 ± 6.1	42.8 ± 5.6
Resting energy expenditure (kcal/day)b	1180.6 ± 157.2	1173.2 ± 143.1
Menstrual cycle length (day)	29.5 ± 3.4	28.5 ± 3.2
Follicular phase length	17.6 ± 3.4	16.3 ± 2.7d
Luteal phase length	11.8 ± 2.0	12.2 ± 2.0
E1G (μg/g Cr) (mean ± SEM)
Follicular phase	50.1 ± 1.9	48.4 ± 2.3
Luteal phase	70.1 ± 2.2	67.0 ± 2.2
PDG (μg/g Cr) (mean ± SEM)
Follicular phase	983.0 ± 35.0	963.5 ± 40.0
Luteal phase	4798.8 ± 262.6	5159.6 ± 355.5
Peak LH (IU/g Cr) (mean ± SEM)	152.4 ± 18.2	99.7 ± 12.8e
Fasting leptin (ng/mL) (mean ± SEM)b (n = 16)	12.12 ± 1.34	9.25 ± 1.11c
Fasting orexin-A (ng/mL) (mean ± SEM)b (n = 16)	0.58 ± 0.04	0.63 ± 0.06

Mean ± SD values (unless otherwise indicated) are reported for NEA and DEA cycles for 17 participants.

Abbreviations: BMI, body mass index; Cr, creatinine; DEA, decreased energy availability; E1G, urinary estrone-3-glucuronide; IU, international units; NEA, neutral energy availability; PDG, urinary pregnanediol glucuronide.

^aAge, gynecological age, and race/ethnicity were determined at enrollment. Gynecological age is defined as the number of years postmenarche.

^bFasting hormones, body composition measures, and resting energy expenditure were assessed on day 5 of the corresponding dietary intervention.

^c P < .01 by paired sample t-test.

^d P < .05 by paired sample P-test.

^e P < .05 by signed rank test.

This study was approved by the National Institutes of Health Institutional Review Board (initial approval: July 15, 2016), and all participants provided informed consent prior to study procedures. All study procedures had consistent IRB oversight and adhered to the tenets of the Declaration of Helsinki.

Protocol

Participants were provided with 2 5-day dietary interventions in the EFP of separate menstrual cycles—1 of NEA and the other of DEA. Subjects completed a baseline visit during the EFP to the mid-follicular phase (MFP) of the pre-NEA cycle. Resting energy expenditure (REE) and maximum oxygen consumption (V̇O₂­ max) were measured by monitoring respiratory gases using open-circuit spirometry and a calibrated metabolic cart (TrueOne 2400 Metabolic Measurement System, Parvo Medics Inc., Sandy, UT, USA). LBM was measured using dual-energy x-ray absorptiometry (Lunar iDXA, GE Medical Systems, Madison, WI, USA). Baseline diet was logged in a 4-day food diary by each individual to determine whether they were habitually under- or overeating and, therefore, ineligible. Diets were based on individual LBM and energy expenditure measurements collected at baseline. Individualized diet and exercise regimens were established to result in an NEA of 45 kcal/kg LBM/day and a DEA of 20 kcal/kg LBM/day, keeping macronutrient composition as constant as possible. Standard meal replacements were provided as cookies (Lenny & Larry®), bars (OhYeah®, Quest®, and One Bar®), and/or shakes (GNC Total Lean Shakes®). For each individual, exercise (eg, treadmill, exercise bicycle, run/walk) was kept constant between NEA and DEA interventions and designed to result in a caloric expenditure of 150 to 300 kcal/day lasting up to 60 minutes/day. Thus, the reduction in energy availability in the DEA intervention was due to a reduction in dietary intake alone.

Subjects recorded menstrual diaries and underwent pelvic ultrasounds (transabdominal or transvaginal; Samsung H60 system, Seoul, Republic of Korea) in the cycle before each intervention. Subjects recorded events related to spotting/bleeding, ovulation, and premenstrual symptoms. Day of ovulation was ascertained by a combination of ultrasound measurement of the preovulatory dominant follicle and home ovulation kits (First Response^TM Ovulation Plus Pregnancy Test, Church & Dwight Co. Inc., Ewing, NJ) to predict the timing of the next menses and the institution of the diet intervention. Subjects consumed the prescribed diets for 5 days in the EFP and were evaluated on the fifth day. Energy restriction is known to cause changes in reproductive hormones and menstrual cycle dynamics (40, 41), and even short-term dieting may have deleterious impacts on menstrual cycle characteristics that extend for several months beyond the duration of the dietary intervention (42). For this reason, the order of diets was not randomized to avoid potential confounding of results. This approach offers the advantage of avoiding the carryover effects of an energy-negative dietary intervention but carries the potential to introduce bias in our results. Subjects returned to the Clinical Research Unit at the National Institute of Environmental Health Sciences on day 5 of each diet intervention, at which time compliance with the prescribed diet and exercise intervention was determined by a review of daily logs, empty food wrappers, and urine ketone strip testing (Multistix® 10 SG Urine Test Strips, Siemens Corporation, USA). LBM, body fat percentage, and REE were measured by air displacement plethysmography using a BodPod (COSMED, Rome, Italy), and fasting blood samples were drawn for the measurement of orexin-A and leptin. BodPod estimates REE based on measured fat mass and LBM using the Nelson equation, which has been validated as a reliable predictor of REE measurements obtained via indirect calorimetry (R² = .99, P < .001) (43). Daily early morning urine strips were collected throughout the study period and stored as per manufacturers’ instructions (ZRT Laboratory, Beaverton, OR, USA). Urine samples were collected from the pre-NEA cycle ultrasound visit until menses following DEA.

Actigraphy

ActiGraph GT3X-BT wrist band monitor devices (ActiGraph, LLC, Pensacola, FL, USA) were worn continuously by subjects starting before ovulation in the pre-NEA cycle and ending at the end of the DEA cycle. Wear-time validation of the raw actigraphy data was performed using the Choi (2011) algorithm to identify nonwear periods. We estimated energy expenditure using the Freedson VM3 Combination algorithm (2011) and METs (the ratio of working metabolic rate relative to resting metabolic rate) using the Freedson Adult (1998) algorithm (44). We analyzed sleep using 60-second epochs using the Cole-Kripke Sleep Scoring algorithm made available by ActiLife (ActiGraph data analysis software). The Cole-Kripke Sleep Scoring algorithm has high performance in detecting sleep-wake compared to polysomnography (PSG) (sensitivity 0.92, specificity 0.65, accuracy 0.89) (45). Sleep periods, defined as the period from time in bed to time out of bed for each night, were detected using the ActiGraph Auto Sleep Period Detection tool. This tool utilized the Troiano Wear Time Validation algorithm, which distinguished sleep periods from nonwear periods. Daytime naps were excluded.

Six objective predefined measures were used: (1) total sleep time (TST, minutes); (2) sleep efficiency (SE; percent); (3) wake after sleep onset (WASO; minutes); (4) number of awakenings (NOA) per night; (5) average duration of awakenings (DOA; minutes); and (6) sleep fragmentation index (SFI). TST refers to the total duration of a person's sleep based on their movement patterns detected by the actigraph computed by the time difference in sleep onset (reduced movements) and sleep offset (increased movements) (46). SE indicates the ratio of the duration of sleep and time spent in bed (47). WASO represents the number of minutes of wakefulness between the onset of sleep and final offset of sleep (46). NOA is the number of awakenings that last more than 60 seconds during sleep (46). Average DOA is the ratio of total time spent awake and the number of awakening episodes during sleep (46). SFI is computed through a combination of a movement index (period of activity) and fragmentation index (ratio of 60-second periods of sleep and count of all periods of sleep) during the sleeping period, with a higher SFI signifying more disrupted sleep (46, 48).

Assays

Urinary metabolites of estradiol (E2) and progesterone (P4) [estrone glucuronide (E1G), RRID: AB_3665778 and pregnanediol glucuronide (PDG), RRID: AB_3665779, respectively] and LH (Cal Biotech® Catalog # LH231F, RRID not available) were measured by ELISAs that were validated for analysis of dried urine strips at ZRT Laboratory (Beaverton, OR, USA) and used in prior studies (49, 50). Coefficients of variation were acceptable if they fell below 15% to 20% for each assay (51). For the E1G assay, the intra-assay coefficients of variation (CV) were 7.2% for low controls (10.1 ± 0.7 ng/mL) and 4.9% for high controls (91.3 ± 4.5 ng/mL). The E1G interassay CVs were 16.2% for low controls (18.4± 3.0 ng/mL) and 12.0% for high controls (124.5 ± 14.9 ng/mL). The PDG intra-assay CVs were 10.7% for low controls (544.5 ± 58.2 ng/mL) and 6.4% for high controls (3726.5 ± 239.1 ng/mL). The PDG interassay CVs were 12.3% for low controls (1678 ± 207 ng/mL) and 9.5% for high controls (8701 ± 825 ng/mL). The LH intra-assay CVs were 11.6% for low controls (3.0 ± 0.3 mIU/mL) and 3.1% for high controls (180.7 ± 3.1 mIU/mL). The LH interassay CVs were 18.7% for low controls (8.1 ± 1.5 mIU/mL) and 16.3% for high controls (136.3 ± 22.2 mIU/mL). All dried urine specimens were analyzed for creatinine (Cr) to adjust for hydration and variation in filter paper strip saturation using the Jaffe reaction (52) with results of E1G, PDG, and LH reported per gram of Cr. The Cr intra-assay CVs were 1.3% for low controls (0.47 ± 0.01 mg/mL) and 0.6% for high controls (2.12 ± 0.01 mg/mL). The Cr interassay CVs were 14% for low controls (0.41 ± 0.06 mg/mL) and 6.4% for high controls (2.2 ± 0.1 mg/mL). All samples for both cycles in a given individual were sent for analysis at the same time and coefficient values referenced were obtained for samples kept at room temperature for the month-long testing period.

Plasma orexin-A was measured using a competitive immunoassay (ELISA EK-003-30 developed by Phoenix Pharmaceuticals). The sensitivity of the assay was 0.22 ng/mL with an intra-assay CV <10% and interassay CV <15% within the linear range of 0.22 to 2.28 ng/mL. All samples were measured in the same assay. Plasma leptin levels were measured using the multiplex immunoassay method through a Clinical Laboratory Improvement Amendments-certified commercial laboratory (RRID: not available, Myriad Rules Based Medicine, Austin, TX, USA). The sensitivity of the assay was 0.15 ng/mL, the intra-assay CV was 4.40%, and the interassay CV was 6.67% within the linear range of 0.15 to 215 ng/mL.

Data Analyses

Analysis was conducted on 17 subjects who met enrollment criteria and completed all study procedures with the exception of orexin-A and leptin for which 1 subject had missing data. There were no demographic differences between the 17 “completers” and the 21 subjects meeting enrollment criteria who did not complete study procedures (“noncompleters”) (Supplement 1, Table 2) (39).

Table 2.

Sleep outcomes in relation to dietary conditions and menstrual cycle phases

Sleep outcomes	EFP	MFP	LFP	ELP	MLP	LLP	LMM results
							Effect: P-valuea
Total sleep time (minutes)
NEA	400.2	403.8	419.2	409.3	403	406.8	Phase: NS
	385.4-414.9	389.3-418.3	404.1-434.3	393.7-424.9	387.1-418.8	390.8-422.8	Diet: NS
DEA	409.1	408.3	416.6	418.1	413.6	397.1	Phase × diet: NS
	394.3-423.9	393.6-423.1	401.3-431.9	402.4-433.7	393.6-429.3	380.6-413.6
Wake after sleep onset (minutes)
NEA	32	28.3	28.7	34.3	29.4	40.4	Phase: .008
	28.9-35.4	25.6-31.2	25.8-31.9	30.8-38.2	26.3-32.9	36.1-45.2	Diet: .005
DEA	30.2	34.5	39	35.1	37.8	41.4	Phase × diet: NS
	27.3-33.4	31.2-38.2	35.1-43.3	31.5-39.1	33.9-42.1	36.9-46.5
Number of awakenings
NEA	14.3	14.9	14.9	14.8	14.1	16.9	Phase: .0006
	12.9-15.7	13.5-16.3	13.5-16.3	13.4-16.2	12.6-15.5	15.4-18.3	Diet: .02
DEA	14.2	15.2	16.9	15	16.2	18.2	Phase × diet: NS
	12.9-15.6	13.8-16.6	15.4-18.3	13.5-16.4	14.7-17.6	16.7-19.7
Sleep fragmentation index
NEA	18.8	17.7	17.7	17.4	18	19.8	Phase: NS
	17.1-20.6	16.2-19.4	16.1-19.5	15.8-19.2	16.3-19.8	17.9-21.8	Diet: NS
DEA	17.7	19.3	20.4	18.4	18.3	21.1	Phase × diet: NS
	16.2-19.5	17.6-21.1	18.5-22.4	16.8-20.3	16.6-20.1	19.1-23.2
Duration of awakenings (minutes)
NEA	2.8	2.3	2.2	2.7	2.6	2.9	Phase: .04
	2.6-3.0	2.1-2.4	2.0-2.4	2.5-2.9	2.4-2.8	2.7-3.1	Diet: NS
DEA	2.5	2.6	2.7	2.8	2.7	2.7	Phase × diet: .046
	2.3-2.7	2.4-2.8	2.5-2.9	2.6-3.0	2.5-2.9	2.5-2.9
Sleep efficiency (%)
NEA	90.8	92	92.2	90.8	91.4	89.1	Phase: .002
	90.0-91.6	91.3-92.8	91.4-92.9	90.0-91.7	90.5-92.2	88.2-90.0	Diet: .01
DEA	91.5	90.5	90.1	90.8	89.8	89.2	Phase × diet: .02
	90.7-92.2	89.7-91.3	89.2-90.9	90.0-91.6	88.9-90.6	88.3-90.1

Least square means ± SEM and P-values for sleep outcomes in relation to diet, menstrual cycle phase, and diet × menstrual phase. In each cycle, there was a 5-day diet intervention (NEA vs DEA) in the early follicular phase of the cycle. Wake after sleep onset, duration of awakenings, and sleep fragmentation index were log-transformed to approach normality, while sleep efficiency was cube-transformed for analysis. The back-transformed values are reported here.

Abbreviations: DEA, decreased energy availability; EFP, early follicular phase; ELP, early luteal phase; LFP, late follicular phase; LLP, late luteal phase; LMM, linear mixed model; MFP, mid-follicular phase; MLP, mid-luteal phase; NEA, neutral energy availability; NS, not significant.

^a P-value for type 3 tests of fixed effects from interaction models (statistically significant values are bolded).

The day of the luteal transition (DLT), used as a proxy for the day of ovulation, was determined for each menstrual cycle. In the pre-NEA cycle, DLT was determined by a home ovulation kit. In the NEA and DEA intervention cycles, DLT was determined retrospectively from the collected urinary hormone concentrations (LH, E1G, PDG), utilizing a combination hierarchical algorithm developed by O’Connor et al (53). This algorithm employed a multistep approach combining in order of priority, a rise in urinary LH, a sustained rise in PDG, and a decline in E1G (all Cr normalized) to estimate the DLT (53), which was then confirmed by visual inspection of the plotted urine hormone concentrations. Follicular phase length was calculated as the number of days from menses onset up to and including the DLT. Luteal phase length was calculated as the number of days after the DLT until the day before next menses.

To compare sleep data across cycles between and within subjects, cycles were standardized to a 28-day cycle centered on the DLT split equally into a 14-day follicular phase (which included the day of presumed ovulation) and a 14-day luteal phase before analysis. The standardized follicular phase (days 1-14) was divided into EFP (days 1-5), MFP (days 6-10), and late follicular phase (LFP; days 11-14). The standardized luteal phase (days 15-28) was divided into early luteal phase (days 15-19), mid-luteal phase (MLP; days 20-24), and late luteal phase (LLP; days 24-28) as previously described (30, 54).

The E1G and PDG values for each subject were fitted with smooth curves using the LOESS procedure in SAS/STAT® (SAS Institute Inc., Cary, NC, USA) for both NEA and DEA cycles. Differences in mean hormone levels were then assessed using 2-way ANOVA with repeated measures of the natural-log transformed E1G and PDG concentrations with cycle phase (EFP, MFP, etc.) and diet intervention (NEA, DEA) as the 2 factors.

Statistical Analysis

BMI, body weight, body fat percentage, LBM, REE, and fasting day-5 orexin-A and leptin were compared using paired sample t-tests or Wilcoxon signed rank tests. Nonstandardized total cycle, follicular phase, and luteal phase lengths between the NEA and DEA diet cycles were compared using paired sample t-tests or Wilcoxon signed rank tests. Two-way repeated measures ANOVA was used to evaluate changes in energy expenditure determined by ActiGraph across the menstrual cycle between NEA and DEA.

We used linear mixed models (SAS Proc Mixed) to analyze sleep measures with post hoc Tukey-Kramer tests to account for multiple comparisons of least square means. We analyzed sleep measures as a function of (1) cycle phase and diet intervention (NEA vs DEA cycle) and their interaction (phase * diet);( 2) reproductive hormones (E1G and PDG) over time, diet intervention, and their interaction (diet * hormone); (3) fasting appetite-regulating hormone levels (orexin-A and leptin) on day 5 of NEA vs DEA, menstrual cycle phase, and their interaction (phase * hormone). Interaction terms in all 3 types of models were deleted when nonsignificant to produce main effects models. Subject identity was included in all models to identify the repeated measurements for a given subject. All models incorporated random intercepts for subjects and used an unstructured covariance matrix. Models also controlled for weekday vs weekend when marginally or significantly related to sleep outcome, but otherwise, this term was dropped. We compared luteal phase sleep characteristics of NEA to corresponding periods during an ad libitum diet prior to NEA (pre-NEA).

Some sleep parameters, as well as E1G and PDG, were transformed to make their distributions more normal. Thus, log10 values were used for analysis for WASO, DOA, and SFI, and SE was cubed while TST and NOA did not require transformation. Log10 transformations were also used for EIG and PDG. P-values ≤ .05 were considered significant using 2-sided comparisons throughout.

Results

There was a small but significant decrease in BMI and body weight, measured on day 5 of each dietary intervention, from NEA to DEA. Body fat percentage, LBM, and REE calculated using the BodPod did not differ between diet conditions (Table 1). REE measured by indirect calorimetry (metabolic cart) at least 1 month before NEA was highly related to REE estimated by the BodPod (adjusted R² = 0.78, P < .001) during NEA. Actigraphy data confirmed that total daily energy expenditure remained unchanged between habitual activity (pre-NEA) and NEA and DEA interventions [daily energy expenditure (mean ± SEM); 1655 ± 192, 1619 ± 176, and 1686 ± 211 kcal/day, respectively; P = .97]. The macronutrient composition of the 2 diets was similar (carbohydrates: NEA 37%, DEA 31%; fat: NEA 29%, DEA 31%; and protein: NEA 35%, DEA 38%), and dietary compliance was above 90% in all participants (NEA 92% and DEA 97%), with an average dietary reduction of 960 kcal/day. The change in body weight is consistent with the expected effect of this degree of caloric deficit in the sedentary subjects included in this study who had a median V̇O₂ max of 34.4 (interquartile range 28.5-37.9) mL/kg/min at baseline. Energy expenditure and METs (the ratio of working metabolic rate relative to resting metabolic rate) determined by actigraphy did not differ between the 2 diet intervention periods or between the cycles that immediately followed the diet interventions.

Hormonal data, including confirmation of ovulation, were available across both menstrual cycles in all 17 participants. Follicular phase length was shorter in the DEA compared with the NEA diet cycle, while there was no difference in the total cycle or luteal phase length (Table 1). Peak LH was lower in the DEA compared with the NEA cycle, but E1G and PDG normalized to standard follicular and luteal phase lengths did not differ between the 2 diet conditions (Table 1 and Fig. 2).

Figure 2.

Fitted curves of mean ± SEM for E1G and PDG concentrations in relation to cycle phase. Hormonal data were combined for NEA and DEA as these were not different between NEA and DEA cycles.

Abbreviations: Cr, creatinine; DEA, decreased energy availability; E1G, estrone-3-glucuronide; EFP, early follicular phase; ELP, early luteal phase; LFP, late follicular phase; LLP, late luteal phase; MFP, mid-follicular phase; MLP, mid-luteal phase; NEA, neutral energy availability; PDG, pregnanediol-3-glucuronide.

Effect of Diet, Menstrual Cycle Phase, and Their Interactions on Sleep

Among the participants, a total of 918 nights were recorded using actigraphy across the 2 diet cycles. Controlling for weekday vs weekend and for multiple comparisons, we demonstrated changes in sleep measures that indicate energy deficiency and/or specific menstrual cycle phases are associated with poorer sleep.

Menstrual cycle phase

TST was greater on weekend nights compared to weekday nights (adjusted P < .0003) but did not change as a function of menstrual cycle phase. However, controlling for weekend vs weekday, sleep was more disrupted in the LLP compared with the EFP, as evidenced by increased WASO (adjusted P = .008) and NOA (adjusted P = .0006). The overall difference in WASO between the EFP and LLP was 9.9 minutes (range 3.3-16.7) with 3.2 more awakenings (range .6-6.1) in the LLP than EFP (Fig. 3 and Table 2), while there was no main effect of cycle phase on TST or SE. Sleep characteristics were not different between pre-NEA ad libitum cycles and NEA cycles: WASO (mean ± SEM 33.2 ± 3.9 vs 32.4 ± 3.8, P = .91), TST (400.63 ± 13.38 vs 405.93 ± 13.42, P = .52), SE (90.35 ± 0.87 vs 90.47 ± 0.87, P = .79), NOA (16.05 ± 1.48 vs 14.96 ± 1.48, P = .1), DOA (2.53 ± 0.15 vs 2.72 ± 0.15, P = .14), and SFI (19.10 ± 1.59 vs 18.49 ± 1.55, P = .48).

Figure 3.

Effect of menstrual cycle phase on sleep parameters (least square means + SEM). Menstrual cycle phase directly affected (A) WASO and (B) number of awakenings. All analyses were performed with LLP as a reference. WASO was log10-transformed to approach normality. *P < .05 vs LLP; **P < .01 vs LLP; ***P < .001 vs LLP. P-values for the cycle phase differ slightly from Table 2 because nonsignificant interactions were dropped from the model.

Abbreviations: LLP, late luteal phase; WASO, wake after sleep onset.

Dietary restriction

Short-term caloric restriction disrupted sleep independent of changes related to menstrual cycle phase (Fig. 4, Table 2). WASO, for example, increased by an average of 4.3 minutes across the menstrual cycle (P = .005) from NEA to DEA. Similarly, there was an increase in NOA (P = .03) and a trend for SFI to be higher in DEA than NEA (P = .08).

Figure 4.

Effect of diet intervention on sleep parameters. WASO and NOA increased between the NEA and DEA diet interventions, and there was a trend to increased SFI. For analysis, WASO and SFI were log-transformed to approach normality. The back-transformed values are shown.

Abbreviations: DEA, decreased energy availability; NEA, neutral energy availability; NOA, number of awakenings per night; SFI, sleep fragmentation index; WASO, wake after sleep onset.

Interaction between diet and menstrual cycle phase

In addition to the main effects of diet and menstrual cycle phase on sleep, there was an interaction between the 2 (Table 2) that was significant for both DOA (P = .046) and SE (P = .02) with DEA resulting in increased DOA in the LFP and reduced SE in the MFP, LFP, and MLP (Fig. 5).

Figure 5.

Interaction of diet and menstrual cycle phase on sleep parameters. Diet interacted with cycle phase for (A) DOA and (B) SE. SE was cube-transformed to approach normality, and back-transformed values are shown. Differences in DOA were significant in the late follicular phase, and SE were significant in the mid-follicular phase, late follicular phase, and mid-luteal phase in post hoc testing but not in any other phase.

Abbreviations: DEA, decreased energy availability; DOA, duration of awakenings; NEA, neutral energy availability; SE, sleep efficiency.

Relationship of Reproductive Hormones to Sleep

PDG and E1G levels correlated positively with greater sleep disruption (Table 3). Increasing PDG was associated with reduced SE and increased WASO and DOA. Increasing E1G was associated with increased WASO and NOA (Table 3). To provide context, an increase in PDG from 1000 μg/g Cr in the follicular phase to 5000 μg/g Cr in the luteal phase would be expected to increase WASO by a mean (SEM) of 3.3 ± 0.4 minutes. An increase in E1G from 40 μg/g Cr in the EFP and MFP to 80 μg/g Cr in the LFP and MLP would be expected to increase WASO by a mean (SEM) of 2.3 ± 0.25 minutes. TST, WASO, and NOA were lower on weekday nights compared to weekend nights, and thus linear mixed models controlled for weekday/weekend nights.

Table 3.

Regression coefficients ± SEM from linear mixed models evaluating the relationship between objective sleep, diet intervention (NEA and DEA), and reproductive hormones

Sleep outcomes	Factors	log(PDG)	log(E1G)
TST	Hormone	2.50 ± 6.99	19.53 ± 12.69
	DEA vs NEA	5.04 ± 5.79	5.59 ± 5.79
SE (cubed)	Hormone	−25751 ± 9310a	−23032 ± 16972
	DEA vs NEA	−19734 ± 7706a	−19863 ± 7747a
WASO (log)	Hormone	0.06 ± 0.02b	0.09 ± 0.04b
	DEA vs NEA	0.05 ± 0.02a	0.05 ± 0.02a
NOA	Hormone	0.95 ± 0.52	2.93 ± 0.95a
	DEA vs NEA	0.53 ± 0.43	0.60 ± 0.43
DOA (log)	Hormone	0.03 ± 0.01b	−0.005 ± 0.022
	DEA vs NEA	0.02 ± 0.01b	0.02 ± 0.01b
SFI (log)	Hormone	0.01 ± 0.01	0.03 ± 0.03
	DEA vs NEA	0.01 ± 0.01	0.01 ± 0.01

Each sleep outcome was analyzed using a linear mixed model with hormone (separate models for PDG and E1G), diet intervention, and weekday/weekend as independent variables. PDG, E1G, WASO, DOA, and SFI were log-transformed to approach normality and SE was cube-transformed. Weekday/weekend was significant in both PDG and E1G models for TST, WASO, and NOA (data not shown) and was deleted when nonsignificant.

Abbreviations: DEA, decreased energy availability; DOA, average duration of each awakening; E1G, estrone-3-glucuronide; NEA, neutral energy availability; NOA, number of awakenings per night; PDG, pregnanediol glucuronide; SE, sleep efficiency; SFI, sleep fragmentation index; TST, total sleep time; WASO, wake after sleep onset.

^a P < .01.

^b P < .05.

Relationship of Metabolic Hormones to Sleep

Orexin-A was positively associated with WASO (β ± SEM 0.27 ± 0.11, P = .02), SFI (0.26 ± 0.08, P = .001) and NOA (7.62 ± 2.93, P = .009) and was inversely associated with SE (−109371 ± 48181, P = .02) in models adjusted for weekday vs weekend. There was no interaction between fasting EFP orexin-A and cycle phase on sleep parameters. To provide context, an increase in orexin-A from 0.58 to 0.63 ng/mL is predicted to result in an increase in WASO, SFI, and NOA of 3.21%, 3.40%, and 2.49%, respectively with a small decrease in overall SE of 0.24%.

Fasting leptin measured in the EFP on the fifth day of each dietary intervention decreased with DEA (12.12 ± 1.34 vs 9.25 ± 1.11 ng/mL, NEA vs DEA, respectively; P = .011). There was an interaction between the effect of EFP fasting leptin and cycle phase on WASO (P = .03) with an inverse relationship between leptin and WASO in the MFP, LFP, MLP, and LLP but a positive association in the EFP and early luteal phase when gonadal hormones are lowest. Leptin was not significantly related to TST, SE, NOA, DOA, or SFI either alone or in relation to the menstrual cycle phase.

Discussion

In the current study, we used actigraphy in a habitual setting to provide objective measures of sleep over a duration of approximately 2.5 months to determine the impact of a controlled dietary intervention on sleep measures across the menstrual cycle. This study adds important information on changes in sleep characteristics across the menstrual cycle using daily hormone measures that confirmed ovulation and precisely delineated cycle phases, demonstrating that poorer sleep is most pronounced in the LLP and associated with higher levels of P4 and E2 in young, normally cycling women without obesity. Of additional importance, we used a controlled intervention that demonstrated that short-term, moderate energy restriction has a deleterious effect on sleep and showed that the effect of energy restriction at the beginning of a cycle on sleep persists across the menstrual cycle. We showed that the increase in orexin that occurs with energy restriction is positively associated with sleep disturbance, while the effect of the concomitant decrease in leptin depends on menstrual cycle phase. Together, these associations provide a potential mechanistic link between the effect of short-term dietary restriction and sleep disruption.

While the impact of sleep disruption on feeding behavior and metabolic responses is now well established (55), there is far less information on the effect of decreased dietary intake on sleep, and studies have almost exclusively included obese participants in whom resolution of obstructive sleep apnea was likely the cause of sleep improvement (23). In women without obesity, we found that short-term, moderate dietary restriction resulted in worse sleep quality, as evidenced by increases in WASO and NOA. In the current study, there was a 25% increase in wakefulness after the onset of sleep from the beginning to the end of the menstrual cycle. While the effect of short-term, modest energy restriction was less pronounced, it added an additional 12.5% to sleep disruption across the cycle. These results provide preliminary evidence that short-term dietary restriction negatively impacts sleep across the menstrual cycle and suggest that diet studies should consider the inclusion of sleep assessment in their design. For some outcomes like DOA and SE, dietary restriction-related sleep disruption was evident during certain cycle phases, pointing to a combined effect of appetite-related hormones and changes in ovarian steroids on sleep. In the only previous human study that addressed the effect of an energy deficit on sleep in individuals without obesity, 12 men were studied before and after 2 days of 90% calorie restriction (56). In this study, there was an increase in slow wave sleep (SWS) without changes in rapid eye movement (REM) sleep, WASO, TST, or SE. Contrary to results from the current study and what would have been predicted from preclinical studies, in this study of near-complete energy deficiency, orexin was positively rather than negatively associated with sleep (56).

Our findings suggest that in young women without obesity, orexin and leptin may be involved in both the neurometabolic responses to a decrease in caloric intake and the homeostatic control of sleep/wake. Orexin was positively related to sleep disruption, as predicted from animal studies. Although named for its function as an appetite stimulator, orexin is central to the homeostatic control of wakefulness and sleep. In mice, orexin stabilizes wake-promoting neurons, stimulating arousal and the downstream activities of grooming and locomotion, increasing energy expenditure (16). The relationship of leptin to sleep was more complex, with the expected positive association between leptin and poorer sleep in cycle phases in which estrogen is low, while the inverse was true in cycle phases associated with higher estrogen levels. Leptin, which originates from and is proportional to adipose cell mass and BMI, suppresses appetite such that decreased leptin is a marker of caloric insufficiency and would be expected to increase appetite (17). Studies in rodents indicate that reduced leptin levels are associated with increased activity of orexin neurons, suggesting that leptin may relay information on energy availability to sleep centers through its effect on orexin (57) and that diet-induced decreases in leptin may contribute to sleep disturbance, albeit indirectly. Leptin has been shown to vary across the menstrual cycle in some (58-61) but not all (62, 63) studies and has been positively associated with E2 levels in an interventional study (61). This relationship may account for the interaction between leptin and the menstrual cycle on sleep, but future studies in which leptin and sex steroids are measured across the cycle in conjunction with sleep will be required.

In the current study, we demonstrated the occurrence of poorer sleep in the LLP in reproductive-aged women with proven ovulation, as evidenced by increased WASO and NOA, as well as an increase in DOA and a decrease in SE interacting with energy deficiency. Sleep patterns were similar in the luteal phase of the NEA intervention and the pre-NEA cycle, suggesting that sleep findings from our standardized NEA dietary intervention are applicable to the ad libitum diet in a real-world setting. Our findings extend those from a previous actigraphy study in late reproductive or early perimenopausal women. In this study, NOA was higher in the apparent LLP, although the ovulatory status of the women was unknown (31). Our finding of changes in sleep in relation to the menstrual cycle, as determined by actigraphy, support subjective accounts of poorer sleep quality in the week before menses, which corresponds to the LLP (37, 47, 64, 65), but did not support the finding of greater subjective sleep disturbance during menses that was reported in 1 of these earlier studies (65). For some sleep parameters (SE and DOA), diet-related sleep differences were accentuated in certain menstrual cycle phases and may be explained by the differential impact of gonadal steroids on sleep characteristics (66); however, underlying mechanisms will need to be elucidated in future studies.

Studies in which sleep is assessed using PSG allow for the assessment of sleep architecture, but cannot provide the continuous assessment of sleep that is possible in actigraphy studies or the advantage of measurement in the individual's habitual setting. Actigraphically measured values of WASO (∼30-40 minutes), SE (∼85-90%), NOA (∼14-17), SFI (∼17-21), and TST (∼400 minutes) in our NEA cycle were comparable to PSG reference values for healthy, young adults (67, 68), providing confidence in our findings. Consistent with findings in the current study, 3 PSG-based studies reported greater sleep disruption in the luteal phase (31, 32, 66). Increased arousals and reduced SWS were reported in the LLP in a PSG study of perimenopausal women (31). Similarly, in a study evaluating sleep in women with premenstrual syndrome and unaffected control women (mean age ∼30yrs), reduced non-REM and stage 1 sleep and a higher number of micro-arousals were reported in the LLP (32). Other PSG studies have demonstrated luteal phase changes in sleep architecture, including higher sleep spindle frequency, reduced REM sleep, and increased non-REM sleep, but did not find significant effects on the homeostatic processes identified by changes in sleep/wake such as WASO or SE (34, 69).

We found that increasing levels of urinary metabolites of both E2 and P4 were associated with disrupted sleep, providing a potential mechanism for changes in sleep as a function of the menstrual cycle. Results of the current study provide more precise hormonal data that extends the findings from several previous studies. In 1 study that included a single follicular and a single luteal phase hormone measurement, E2 and P4 were positively associated with WASO and reduced REM sleep (32). In a second study in premenopausal women, there was a negative relationship between SE and PDG, with the lowest SE in the time period consistent with the LLP, although cycle phases were not determined in this study (70). In contrast to our results, this study reported a weakly positive association between E1G and SE (70). However, in rodents, a reduction in both NREM and REM sleep has been reported during proestrus when estrogen is elevated (71-73), and E2 administration decreased NREM (74) and REM sleep (74, 75), consistent with our findings. While these results contrast with studies in postmenopausal women in whom estrogen replacement improved subjective sleep quality (76) and reduced NOA and DOA (77, 78), studies in postmenopausal women are confounded by vasomotor symptoms, which are associated with awakenings and are improved with estrogen administration.

Both sleep-promoting and sleep-disrupting effects of P4 have been reported in animal studies (79, 80). In men, P4 administration increased lighter stages of sleep at the expense of deep sleep (8). P4 administration increased SWS and reduced wake time in postmenopausal women (81, 82) although, as with the effect of E2 on sleep it is difficult to disentangle the impact of P4 on vasomotor symptoms in postmenopausal women.

There are several important strengths of the current study. The study subjects were rigorously screened for regular, ovulatory cycles as well as habitual eating, exercise, and absence of sleep problems or shift work. We collected sleep data in a habitual setting using actigraphy, and diet and exercise were strictly controlled in both interventions. We documented > 90% compliance with the diet interventions and found no difference in activity either before or during the dietary interventions using daytime actigraphy. Finally, we measured orexin and leptin, both of which impact sleep/wake centers, as well as daily reproductive hormones over 2 complete menstrual cycles. Measurement of daily reproductive hormone levels allowed us to confirm not only ovulation but also the day of ovulation, which permitted us to standardize data across 6 menstrual cycle phases in each cycle to account for individual variation in cycle and follicular and luteal phase lengths and to permit assessment of the effect of E2 and P4 on sleep parameters. Only 1 previous study has evaluated changes in sleep during the menstrual cycle using such complete hormonal data, but it was limited to a single cycle and did not present data longitudinally in relation to the menstrual cycle to assess the effect of cycle phase on sleep (70). Finally, we were specifically interested in the effect of the common practice of short-term dietary restriction, as more prolonged energy restriction has the potential for greater adaptation over time. Despite the short duration and moderate degree of caloric restriction in the current study, there were modest yet significant changes in sleep parameters, plausibly mediated by metabolic intermediaries such as orexin. Importantly, impaired sleep quality and disrupted sleep architecture may diminish restorative sleep (83), ultimately reducing overall well-being and potentially contributing to the failure of short-term energy restriction approaches for weight loss. Despite these strengths, our study is limited by a relatively small sample size, and our results cannot be generalized to women with obesity. Furthermore, likely in part due to the rigorous and intensive nature of this study, 17 subjects (out of 38 initially enrolled subjects) contributed to the data reported. Moreover, although we implemented several measures to enhance dietary compliance, such as collecting returned wrappers, reviewing dietary logs, and conducting urine ketone testing, and observed the anticipated weight reduction following DEA, we cannot exclude the potential for behavioral changes due to awareness of being observed (the Hawthorne effect). We elected to not randomize the order of the diets as it is known that even modest energy restriction may disrupt normal menstrual cycles and the duration of this carryover effect is unknown (42). We attempted to minimize the order effect of the intervention by beginning daily monitoring (actigraph and daily urine collection) several weeks before any intervention such that participants were well acquainted with procedures before both interventions. In addition, study staff (with the exception of the principal investigator and associate principal investigator) believed that the actigraph data would be used to assess activity levels, and the participants were also unaware that we were interested in collecting sleep data. The lack of randomization may result in a type II error, which would decrease the likelihood of finding a true effect of the intervention. Although their effects are not well characterized, the use of processed meal replacements may have contributed to changes in the neurohormonal axis, which is a limitation of the present study but would not have contributed to differences between the 2 interventions. We did not collect subjective sleep data (sleep diaries), although subjective sleep does not always change in parallel to objective sleep measures. Furthermore, despite the advantages of actigraphy including its validation and use in prior studies, PSG studies would have provided additional information on sleep architecture that we were unable to capture. Finally, it is unfortunate that we were unable to measure ghrelin in our study for technical reasons. Ghrelin also influences feeding and sleep/wake behaviors in animal studies (84, 85). Howevver, in humans, the response to ghrelin appears to be sex-specific as ghrelin administration promotes sleep in men but does not have the same effect in young or older women (86-89).

Taken together, the results of the current study in normal young women without obesity demonstrate that sleep is disrupted for at least 3 weeks following 5 days of modest dietary restriction and that these changes add to the sleep disruption that occurs in the luteal phase of the menstrual cycle. Neurometabolic hormones including orexin and leptin may mediate the effect of dietary restriction both independently and in conjunction with ovarian steroids, although interventional studies will be required to confirm causality. Beyond its contribution to our understanding of the biology of sleep, this study has implications for understanding and managing sleep complaints in young women in whom short-term caloric restriction is common. Further studies will be required to determine whether our studies are generalizable to women with overweight or obesity but without underlying obstructive sleep apnea and if more profound acute or chronic energy restriction will have a greater effect on sleep. The answers to such questions will help define the need for the inclusion of sleep hygiene counseling or other interventions in dieting strategies.

Funding

This research was supported by the Intramural Research Program (ZIDES102465 and ZID ES103323) of the National Institute of Environmental Health Sciences, National Institutes of Health, USA.

Disclosures

The authors have no conflicts of interest to disclose

Data Availability

Data will be made available upon reasonable request to the corresponding author through written agreements with the authors and the data partner.

Clinical Trial Information

Trial registration number:  Clinicaltrials.gov Identifier: NCT02858336