Altered sleep architecture following consecutive nights of presleep alcohol

Abstract

Study Objectives

Alcohol consumption before sleep decreases sleep latency, explaining the common use of alcohol as a sleep aid. The full impact of alcohol on sleep architecture is not well understood, particularly the potential cumulative effects of presleep alcohol consumption across consecutive nights. Here, we describe the effects of presleep alcohol on sleep architecture across three consecutive nights.

Methods

Thirty adult participants took part in a crossover, within-participants study consisting of two sets of three consecutive nights of in-lab polysomnography. For each series of nights, participants drank one of the two beverages: a mixer only or a mixer plus alcohol (targeting a BrAC of 0.08 mg/L), ending 1 hour before lights out. Polysomnography (PSG) was used to stage sleep, and standard sleep variables were extracted. Linear mixed-effect analysis and generalized additive modeling were used to examine the effect of alcohol on sleep architecture.

Results

Alcohol before sleep increased the rate of slow wave sleep (SWS) accumulation across all three nights and decreased the rate of rapid eye movement (REM) sleep accumulation at the start of each night. Alcohol also decreased the total amount of REM sleep but did not affect the total amount of SWS each night.

Conclusions

These data indicate that drinking alcohol before sleep substantially affects sleep architecture, including changes to the rate of accumulation of SWS and REM sleep. We show that alcohol disrupts normal sleep architecture, leading to a significant decrease in REM sleep; thus, the use of alcohol as a sleep aid remains a public health concern.

Graphical Abstract

Statement of Significance.

Alcohol is the most used and abused psychoactive substance globally, yet our understanding of how it alters brain function and behavior is incomplete. Alcohol use before sleep may profoundly impact individuals who regularly use alcohol; thus, a better understanding of the relationships between alcohol and sleep is required. Previous studies have failed to characterize alcohol-induced sleep changes at high resolution, with prior studies being limited to either whole-night or coarse divisions thereof (e.g. thirds). Moreover, the effects of consecutive nights of alcohol consumption on sleep remain largely unknown. Thus, the present study combines experimental alcohol administration with overnight physiological sleep studies to explore how alcohol use across consecutive days alters sleep architecture.

Introduction

Alcohol is often used as a somnogen due to its ability to accelerate sleep onset, with nearly 20% of adults in the United States reporting using alcohol as a sleep aid [1, 2]. This rate increases to 30% in individuals with insomnia and over 60% in those with alcohol use disorder [1, 3–5]. Nevertheless, the effects of alcohol on sleep have only been partially characterized. A series of initial studies were limited in their participant samples (primarily young Caucasian men) and by their inability to control for sleep or alcohol use patterns before entering the study. Alcohol may have pronounced effects on the sleep of individuals who engage in regular drinking; therefore, controlled studies of this population are required. Finally, the potential cumulative effects of consecutive nights of alcohol have seldom been examined in a research setting, despite the typical use of alcohol commonly manifesting across multiple nights of drinking [6, 7]. Thus, this study seeks to contribute to this gap in the research by providing highly controlled sleep data following consecutive nights of alcohol consumption.

Prior PSG studies have shown that presleep alcohol is associated with next-day fatigue and decreased working memory and cognitive performance [8, 9]. Alcohol use prior to sleep has also been associated with increased sleep fragmentation and more wakefulness during the second half of the sleep episode [10]. This fragmentation is thought to be a result of several variables, including (1) alcohol-induced night sweats due to alcohol-induced increase in body temperature, (2) increased need to use the restroom due to alcohol’s diuretic effects, (3) increased apnea episodes due to alcohol-induced muscle relaxation, and (4) a potential rebound effect leading to increased wakefulness after alcohol is metabolized [11–14]. This “wakeful rebound” refers to a phenomenon where alcohol consumption prior to sleep leads to increased wakefulness in the second half due to various factors, including blood sugar fluctuations, dehydration, and digestive system discomfort [11, 15]. Specifically, blood sugar levels can increase as alcohol is metabolized, resulting in wakefulness [16, 17]. Alcohol is also a diuretic, which leads to increased urination, resulting in increased bathroom trips during the sleep episode, as well as dehydration, which can cause discomfort and also lead to increased waking during the second half of the night [13, 18]. Furthermore, alcohol consumption can increase the production of stomach acid, leading to various discomforts that can disrupt sleep, including heartburn, acid reflux, and general stomach pain [19, 20]. Overall, the wakefulness experienced in the early morning hours after a drinking episode reflects complex processes triggered by alcohol consumption in the hours prior. Taken together, this data demonstrates that alcohol use before sleep results in an overall increase in sleep disruptions and a decrease in an individual’s sleep quality. Considering that disrupted sleep has been directly associated with deficits in memory and cognitive performance and next-day fatigue, alcohol-induced sleep fragmentation may underlie some of the memory deficits and fatigue associated with nocturnal alcohol use [9, 21].

Alcohol use before sleep has also been shown to alter the architecture of sleep, including a decrease in sleep latency, an increase in rapid eye movement (REM) sleep latency, an increase in slow wave sleep (SWS) [22–27], and a decrease in REM sleep during the first portion of the sleep episode [24, 27–33]. However, these findings are not consistent across the study, possibly owing to variable alcohol use and sleep schedules prior to entering the study, as well as the low sample sizes [22–33]. For example, several PSG studies reported an alcohol-induced effect on SWS but failed to detect an alcohol-induced effect on REM sleep [22, 23, 25, 26]. In contrast, several additional studies report an alcohol-induced effect on REM sleep but fail to detect an alcohol-induced effect on SWS [28–33]. These conflicting results point to limitations in experimental designs, including a lack of control for prior sleep and narrow participant pools. Furthermore, previous PSG studies varied the timing of alcohol administration in relation to sleep time and timing of day of alcohol use, and these experimental differences may be underlying much of the discrepancies in results [22–33]. Thus, the conflicting data currently available from experiments investigating the effects of alcohol on sleep necessitates further, well-controlled studies.

In addition to altering the timing and distribution of sleep stages across the sleep period, regular alcohol use can also exacerbate or induce sleep disorders, including insomnia (sleep maintenance and sleep onset insomnia), sleep apnea, restless leg syndrome, and REM behavioral disorder [14, 34–39]. A potential underlying cause of worsening sleep disorders with alcohol use is the circadian rhythm disturbances experienced by individuals who regularly drink [40–42]. This data underscores the importance of clearly understanding and communicating the effects and potential risks of alcohol use on sleep and circadian rhythms.

Most prior alcohol and sleep studies have focused on either acute intoxication (i.e. a single night of alcohol use) or the long-term effects of alcohol (i.e. individuals with alcohol use disorder). Not studied, however—yet critically relevant for ecological validity—is examining multiple consecutive nights of drinking in individuals from the general population without Alcohol Use Disoder (AUD) [7]. The few existing studies on consecutive nights of alcohol and sleep generally show a decrease in REM sleep (as in the acute studies referenced above [23, 24]) that partially recovers over subsequent nights and has no effect on wake-after-sleep onset or NREM sleep [29]. Each study outlined above—acute or repeated administration—experiences the same sample limitations [29, 30]. Thus, understanding the effects of presleep alcohol on sleep across consecutive nights requires a larger sample size.

Finally, prior studies have been limited by focusing on the whole night’s sleep or coarse divisions (e.g. half or thirds) of the night. While informative, such approaches cannot examine ultradian dynamics within sleep as alcohol metabolism progresses. By expanding our perspective to the dynamics of sleep across the night, advances in science-based recommendations regarding drinking and sleep, as well as potential therapeutic interventions for buffering the harmful effects alcohol has on sleep, may become possible. The current study introduces a novel method capable of modeling moment-by-moment sleep architecture in a nonlinear manner [43, 44] across multiple nights of drinking, using within participants’ design to fill this critical knowledge gap.

The present study seeks to expand our current knowledge of the effects of presleep alcohol on sleep architecture across consecutive nights using both conventional analysis and analysis with high temporal resolution. The aim of this study is to test the hypothesis that sleep architecture is altered following presleep alcohol administration compared to no-alcohol nights and that these effects will dampen across consecutive nights of administration as the body habituates to the dose of alcohol. Specifically, we hypothesize that first, alcohol before sleep will increase the amount of time spent in SWS at the start of the night compared to nights with no alcohol. Second, we hypothesize that the amount of REM sleep will be low at the start of the night due to enhanced SWS and increase toward the end of the night as a rebound from the initial deficit compared to nights with no alcohol. Third, because alcohol is known to produce tolerance to a number of effects [45, 46], we hypothesize that these sleep effects will dampen across consecutive nights. Our analyses test these hypotheses using linear-effects models to investigate whole-night trends and thirds-of-night trends across each sleep stage and each night and generalized additive models to explore the moment-by-moment dynamics of sleep architecture within nights.

Materials and Methods

All experimental procedures were approved by the Lifespan Office of Human Research Participant Protection

Participants.

Thirty (15F; aged = 22–57, mean = 33 years) healthy adult participants, free of medical, neurological, or psychiatric illness were recruited into the current study. Twenty-four participants identified as white (80%), three as black (10%), one as Asian (3%), and two (7%) as more than one racial identity. Two individuals identified as Hispanic (7%) (Table 1). All individuals reported moderate levels of habitual drinking behavior, assessed using a standard timeline follow-back interview [47, 48]. Males must have reported consuming at least 12–15 drinks, and females consuming at least 9–12 drinks in 1 week over the last 12 months. At least one of these drinking episodes must have met the criteria for heavy episodic drinking (three to four drinks for women or —four to five drinks for men). Additional inclusion criteria included being at least 21 years old, BMI between 18.5 and 39.9, and having proficiency in English. Exclusion criteria for enrollment in the study included: currently seeking treatment for alcohol problems, self-reported regular sleep time <7 hours a night or >9.5 hours a night, diagnosed sleep disorder, extreme chronotypes, history of shiftwork within the last 2 years, positive pregnancy test, medical diagnosis that requires ongoing medical intervention (e.g. cancer, diabetes, and kidney disease), fever or illness at the start of the in-lab assessments, current use of medications known to impact sleep or circadian rhythms, use of illicit substances (tested each day of the in-lab procedure), history of brain injury, history of alcohol withdrawal, nicotine addiction that would prevent them from spending 15 continuous hours in a nonsmoking environment, clinically significant depression, physical handicaps that could interfere with the ability to participate (e.g. hearing loss, impaired vision, or motor function), inability or aversion to beverages administered in the study, inability to limit caffeine use throughout the study, and current evidence of major psychopathology in the participant. After enrolling in the study, all participants completed a series of at-home and laboratory assessments as indicated below.

Table 1.

Demographics of Participants in Study

	Female	Male	Total
N	15	15	30
Age	31 +/− 11.6	35 +/− 12.0	33.0 +/− 12.0
Race	11 white, 1 black, 1 Asian, 2 more than 1 race	13 white, 2 black	24 white, 3 black, 1 Asian, 2 more than 1 race
Ethnicity	2 Hispanics, 13 non-Hispanics	15 non-Hispanics	2 Hispanic, 28 non-Hispanic

Protocol overview

The overall study design is illustrated in Figure 1 and consists of a pre-laboratory at-home sleep stabilization period followed by two 3-night laboratory stays separated by ~4 days. Beverage administration was counterbalanced (counterbalance was randomized) across the two three-night sessions and involved presleep beverage consumption (alcohol + mixer vs. mixer only) followed by a polysomnography-monitored sleep opportunity (Figure 1).

Figure 1.

Serial alcohol study protocol overview. Blue represents in-lab time, and green represents at-home time that was monitored by the lab using actigraphy and sleep diaries. Light green represents time spent at home between overnight sessions in the lab. During these times, participants took a battery of cognitive tests; otherwise, they continued their daily routines. Light gray represents sleep time out of the lab, and dark gray represents sleep time in the lab. Time is presented above and below the protocol. Adapt represents adaptation night and D1–D6 denote days 1–6 of the in-lab protocol.

Pre-study sleep.

Prior to sleeping in the lab, participants were asked to sleep at home on a fixed sleep schedule of 8 hours in bed (±30 minutes) set to habitual rise times while refraining from recreational drugs, alcohol, medications that may impact sleep, and excessive caffeine use (> 360 mg/day or any after 02:00 pm). Sleep was monitored with wrist-actigraphy (Micro Motionlogger Watch, AMI, Ardsley, NY, USA) together with daily diaries and call-ins to the laboratory.

In-laboratory polysomnographic procedures.

Each night in the lab, participant sleep was monitored by routine polysomnography (PSG) using a Grael HD amplifier (Compumedics, Inc., Charlotte, NC) digitized and stored with a sampling rate of 400 samples per second, consisting of six EEG derivations (C3, C4, O1, O2, F3, and F4) referenced to contralateral mastoid together with measures of respiratory effort, airflow, and oxygen saturation. Additionally, electro-oculograms (EOG) from left and right canthi, chin electromyogram (EMG) from mentalis/submentalis, and electrocardiogram (EKG) using modified lead II. An adaptation evening was provided prior to starting the study. At this visit, PSG equipment was applied, and the participant remained in the lab for a few hours to become familiar with the setup before beginning the in-lab portion of the study.

Serial presleep alcohol or mixer-only beverage administration.

For the 3 nights of each counterbalanced session (alcohol + mixer vs. mixer only) participants were asked to consume their experimental beverage, split evenly into three cups, over a 45-minute drinking period (15 min/cup). Our protocol for beverage consumption was as follows: First, participants were screened for any recent alcohol use through a AlcoSensor IV breath analyzer upon arrival to the lab (Intoximeters, Inc, St. Louis, MO). The AlcoSensor IV’s accuracy and performance have been demonstrated in a prior study [49]. If the participant’s breath alcohol content (BrAC) was over 0.0 when they arrived at the lab, the participant could not complete the study. Following the BrAC measurements upon arrival at the lab, BrACs were taken every 15 minutes after the drinking episode began and continued until 15 minutes before lights out. BrAC measurements began again in the morning ~45 minutes after waking. BrAC measurements stopped if the participant had a BrAC of 0.0. If not, BrAC measurements were taken ~15 minutes until the participant reached 0.0 and was safe to leave the lab. Participants were required to have someone transport them to and from the lab. If this was not possible, the lab paid for transportation to and from the lab. For each BrAC measurement, we used the following approach to confirm the accuracy of the measure: two BrAC recordings were taken back-to-back, and the average was reported for each BrAC measurement. Second, all meals in the lab were standardized and provided 3 hours prior to beverage administration. Third, beverage administration started 1.75 hours before bed and ended 1 hour before scheduled bedtime (Figure 1).

Each participant drank alone in a separate room monitored by research staff. On alcohol + mixer nights, the experimental beverage consisted of Everclear (180 proof) or vodka (90 proof) mixed with a common sugar-free powdered drink mixed in water. On mixer nights, the beverage excluded the alcohol yet matched for volume. The quantity of beverage given was determined for each participant using height, weight, and sex [50] to target a resulting breath alcohol concentration of 0.08 mg/L. The final BrACs before sleep recorded in this study ranged from 0.038 to 0.087 mg/L (mean = 0.066 mg/L). For each participant, the total serving of the assigned beverage was split equally between three cups. Each participant was given 15 minutes to finish drinking each cup. As outlined above BrAC was measured every 15 minutes after the drinking period began, with the final BrAC being recorded 15 minutes before lights out.

Data analysis

Sleep staging.

PSG records were scored and processed in the open-source MATLAB-based Hume toolbox [51]. Sleep was scored in 30-second epochs using Rechtschaffen & Kales (1968) criteria [52]. A total of four trained sleep scorers scored the PSG records. Scoring reliability was reevaluated every 10 records, with trained technicians comparing their scores to validated PSG records to a criterion of 85%. All six nights for each participant were staged by a single scorer who was blind to beverage condition. Sleep continuity and architecture variables were calculated for each night. SWS, REM sleep, and wake were also examined for each third of the night (T1–T3). Thirds of the night were calculated by separating each participant’s sleep period into thirds, beginning with the first epoch of scored sleep and ending at lights on. Each participant’s alcohol nights were compared to their mixer-only nights. Restroom trips throughout the night were scored as wake, and thus, these epochs were not included in sleep-related analyses.

Data analytics and visualization.

We used two approaches to examine the effect of alcohol versus nonalcohol mixer and study night on sleep variables. The first approach used linear mixed-effect (LME) models to test for differences in sleep variables summarized across the full sleep period (sleep onset to lights on) and across each night utilizing a repeated measured design [53]. The second approach used generalized additive mixed-effect modeling (GAMM) to explore the temporal dynamics of each sleep stage throughout the night while accounting for the nesting of epochs within participants [43, 54–56]. GAMMs apply nonparametric smoothing splines to the data to achieve nonlinear relationships between variables, such as the accumulation of time spent in each sleep stage across the night. In this study, we use GAMMs to explore the cumulative amount of each sleep stage across the night, from epoch-scored PSG recordings.

LMEs modeling of sleep architecture.

LMEs models investigated the main effects and interaction of beverage type (alcohol + mixer vs. mixer only), and serial night of presleep drinking (nights 1, 2, and 3) on sleep structure while accounting for nesting of nights within participants using a compound symmetry variance/covariance assumption, and accounting for missing data [43]. The following sleep microarchitectural variables were considered in independent models: slow wave sleep across the full night (SWS; NREM stages 3 + 4), SWS broken into thirds of the night (SWS T1, T2, and T3), REM sleep, REM sleep broken into thirds of the night (REM T1, T2, and T3), and wake time; each was examined as a percentage of total sleep time (sleep onset to lights on) and as a percentage of each third of the sleep period and run in separate models. Additionally, we looked at REM latency (REML) and examined sleep onset latency and wake-after-sleep onset (WASO) to investigate the continuity of sleep. Models were fit using the lme4 version 1.1-32 R package. A Bonferroni correction was used when evaluating results from the third-of-night models to address inflated family-wise errors from multiple tests.

General additive mixed-effect modeling of sleep stages across time.

To explore alcohol-induced changes to sleep architecture, we used generalized additive modeling of epoch-by-epoch sleep stages. Generalized additive mixed effect models (GAMMs) [43, 44] estimate smooth, functional relationships between predictor variables and responses [57] while accounting for the nesting of epochs within participants, enabling a more detailed examination of sleep architecture across each night.

To obtain a closer look at the dynamics of sleep across the night, the additive sum of epochs for SWS, REM sleep and wake across the sleep period for each participant was calculated. Separate GAMMs were applied to the accumulation of each sleep stage or wake, reported percentage of the total amount of a given sleep stage or wake or in total minutes smoothed using b-spline with the smoothing parameters determined using restricted maximum likelihood. Smoothing was allowed to differ by study night, which was coded into four categories: alcohol night 1, alcohol night 2, alcohol night 3, and mixer. Mixer nights (mixer night 1, mixer night 2, and mixer night 3) were averaged to provide a stabilized “baseline” against which the alcohol + mixer nights can be compared. Each model thus compared alcohol + mixer nights 1, 2, and 3 to mixer. Separate models were generated for SWS, REM sleep, and wake. Models were fit using the mgcv version 1.8-42 package. GAMMs produce a smooth trajectory across the night with a 95% confidence interval ribbon. Results are best interpreted as a lack of overlap of the confidence ribbons of the time-varying estimates, indicating a difference in the accumulation of a given sleep stage at a given time.

Results

Sleep continuity and architecture

For all third of the night results, three separate models were run for each outcome, and we used a Bonferroni correction to adjust the significance level to 0.016. A main effect of alcohol revealed the percentage of time spent in SWS was increased relative to mixer for the first third (T1) of the night across all 3 nights (% SWS T1, F(1, 145) = 20.46, p < 0.001) whereas REM was decreased during the first third of the night (% REM T1, F(1, 145) = 18.12, p < 0.001) (Table 2). A main effect of alcohol also revealed the percentage of time spent in SWS was decreased relative to mixer during the subsequent two-thirds of the night (T2 and T3, respectively) across all three nights (% SWS T2, F(1, 145) = 12.95, p < 0.001; % SWS T3, F(1, 145) = 5.93, p = 0.016) (Table 2). There was no main effect of alcohol on time spent in REM sleep on the second two-thirds of the night (Table 2).

Table 2.

Mixed-Effect Analysis Examined Beverage Type, Study Night, and the Interaction of Beverage and Night for Sleep Variables

Variable	df	F	Sig.
Main effect of alcohol on sleep variables
Beverage X % REM	(1, 145)	5.8	0.017*
Beverage X % REM T1	(1, 145)	18.2	0.000*
Beverage X % REM T2	(1, 145)	0.001	0.98
Beverage X % REM T3	(1, 145)	0.05	0.82
Beverage X REML	(1, 145)	11.0	0.001*
Beverage X SleepL	(1, 145)	0.2	0.64
Beverage X % SWS	(1, 145)	0.002	0.97
Beverage X % SWS T1	(1, 145)	20.5	0.000*
Beverage X % SWS T2	(1, 145)	12.9	0.000*
Beverage X % SWS T3	(1, 145)	5.9	0.015*
Beverage X WASO	(1, 145)	0.15	0.70
Beverage X % wake T1	(1, 145)	0.04	0.79
Beverage X % wake T2	(1, 145)	1.1	0.15
Beverage X % wake T3	(1, 145)	1.7	0.18
Beverage X % TST	(1, 145)	0.002	0.97
Main effect of Night on sleep variables
Night X % REM	(1, 145)	1.6	0.21
Night X % REM T1	(1, 145)	0.6	0.55
Night X % REM T2	(1, 145)	4.6	0.01*
Night X % REM T3	(1, 145)	0.14	0.87
Night X REML	(1, 145)	9.9	0.000*
Night X SleepL	(1, 145)	6.2	0.003*
Night X % SWS	(1, 145)	1.3	0.28
Night X % SWS T1	(1, 145)	1.70	0.18
Night X % SWS T2	(1, 145)	0.78	0.46
Night X % SWS T3	(1, 145)	0.05	0.95
Night X WASO	(1, 145)	5.65	0.004*
Night X % wake T1	(1, 145)	1.9	0.15
Night X % wake T2	(1, 145)	0.15	0.70
Night X % wake T3	(1, 145)	0.02	0.82
Night X % TST	(1, 145)	0.21	0.62

No significant interactions were discovered. * represents p < 0.05.

Interestingly, a main effect of alcohol was only noted when SWS was broken into thirds of the night and not when examined across the full night (% SWS). A main effect was noted for REM sleep across the full night (% REM) and revealed a decrease in the percentage of time spent in REM sleep relative to the mixer nights (% REM, F(1, 145) = 5.80, p = 0.017). Furthermore, a main effect of alcohol indicated that REM latency (REML) increased compared to the mixer nights (REML, F(1, 145) = 11.00, p = 0.001) (Table 2). There was no main effect of alcohol on the % wake T1, T2, or T3, sleep latency, or WASO (Table 2; Figures 2 and 3).

Figure 2.

Sleep variables following presleep alcohol or mixer consumption across three consecutive nights. (A) % SWS across total sleep time (TST). Alcohol + mixer N1–N3 is shown in blue, and mixer-only N1–N3 is shown in gray. No significant main effect of beverage or night. (B–D) % SWS T1–T3. All variables are reported in % of total sleep time in the specified third of night. A Bonferroni correction was used, and significance is represented by * and signifies p < 0.016. No main effect of night was noted. (E) % REM sleep across total sleep time (TST). (F–H) % REM T1–T3. All variables are reported in % of total sleep time in the specified third of night. A significant effect of night is indicated by an * below the x-axis. (I) Percent wake-after-sleep onset reported in minutes. (J–L) % wake T1–T3 reported in % total wake.

Figure 3.

Latency variables following presleep alcohol or mixer consumption across three consecutive nights. (A) Sleep latency is reported in minutes from lights out to sleep onset. Alcohol + mixer nights are shown in blue (to the left), and mixer-only nights are shown in gray (to the right). (B) REM sleep latency is reported in minutes from sleep onset to the first epoch of REM sleep.

A main effect of night, indicating differences occurring across the three nights, was identified for sleep latency, F(2, 145) = 6.21, p = 0.003; total sleep time F(2, 145) = 6.28, p = 0.002; % REM T2, F(2, 145) = 4.64, p = 0.011 and REML, F(1, 145) = 9.934, p < 0.001 and WASO, F(2, 145) = 5.65, p = 0.004. No significant interactions were identified (Table 2; Figures 2 and 3).

Generalized additive mixed-effect modeling

The GAMM models provided timeframes when the amount of time spent in each sleep stage differed between the alcohol nights and the mixer baseline. The variable % SWS showed a marked increase from 1.8 hours into the sleep period to 6.5 hours on alcohol nights (Figure 4A). This trend is found across all three alcohol nights when compared to the mixer baseline variable.

Figure 4.

Generalized additive models (GAMs) of sleep variables following presleep alcohol or mixer consumption across three consecutive nights. (A) GAM model of percent slow wave sleep accumulation across the night. A 95% confidence interval is shown in a lighter color around each line. Chart below indicates where the GAM model lines for each condition show no overlap (including no overlap of 95% confidence interval). Mixer nights were collapsed to create the baseline “Mixer” variable (shown in purple). A raster plot that represents the end of each participant’s night can be found at the bottom right of each GAM model. (B) GAM model of percent REM sleep accumulation across the night. (C) GAM model of percent wake-after-sleep onset accumulation across the night.

There was a decrease in time spent in REM sleep on alcohol night 1 compared to the mixer baseline starting about 1.4 hours into the sleep period to 7.8 hours into the sleep episode (Figure 4B). This alcohol-associated decrease in REM sleep was also apparent on alcohol nights 2 and 3 and occurred during a narrower window, from 2.1 hours into the sleep period to 5.3 hours (Figure 4B). There was no difference in percent wake across the three alcohol nights compared to the mixer baseline (Figure 4C). The alcohol nights show differences from each other in the cumulative amount of time awake from 1.0 to 2.7 hours for alcohol night 1 compared to alcohol night 2 (Figure 4C). Alcohol night 2 and night 3 showed a difference in wake from 5.1 to 6.7 hours from sleep onset (Figure 4C).

In contrast to proportions or relative amounts of SWS, absolute minutes had a different pattern across the night. The amount of SWS for alcohol nights increased compared to the mixer baseline between 1.8 and 6.1 hours only on alcohol night 2 (Figure 5A).

Figure 5.

Generalized additive models (GAMs) of sleep variables following presleep alcohol or mixer consumption across three consecutive nights. (A) GAM model of total minutes of slow wave sleep accumulation across the night. A 95% confidence interval is shown in a lighter color around each line. Chart below indicates where the GAM model lines for each condition show no overlap (including no overlap of 95% confidence interval). Mixer nights were collapsed to create the baseline “Mixer” variable (shown in purple). (B) GAM model of total minutes of REM sleep accumulation across the night. (C) GAM model of total minutes of wake-after-sleep onset accumulation across the night.

There was a decrease in the amount of REM sleep on alcohol nights compared to the mixer baseline variable from 1.3 to 7.8 hours for alcohol night 1, from 2.1 to 5.3 hours for alcohol night 2, and from 2.3 to 5.2 hours for alcohol night 3 (Figure 5B). There was also a difference between alcohol nights. Alcohol nights 1 and 2 diverged from 6.9 to 8.0 hours, and alcohol nights 2 and 3 diverged from 6.3 to 7.9 hours (Figure 5B).

The GAMMs produced for total amount of WASO showed divergence from the mixer baseline variable only for alcohol night 1 from 5.6 hours to the end of the night. Alcohol nights 2 and 3 show no divergence (Figure 5C). Wake also showed differences across alcohol nights. Alcohol nights 1 and 2 diverged from hour 5.2 to the end of the night. Alcohol nights 1 and 3 diverged from hour 5.4 to the end of the night (Figure 5C).

Discussion

Our findings indicate that alcohol alters the temporal dynamics of SWS and REM sleep and that these changes persist over multiple nights of drinking. SWS predominates at the start of the night, when blood alcohol levels are high, and likely contributes to a delay of REM sleep onset and a reduction of total REM sleep. These effects are largely present across all three alcohol nights, though most notable on the first night of alcohol administration. Considering time spent in REM sleep increases across sleep cycles, peaking in the second half of the sleep period, the decrease in total REM sleep on alcohol night 1 may also be a result of more fragmented sleep and an increase in wakefulness at the end of the sleep period. This effect is lessened on the second and third presleep alcohol night, indicating that the human nervous system likely develops a tolerance to repeated doses of alcohol, perhaps to counteract the alcohol-induced reduction in REM sleep.

Our research investigated the impact of drinking alcohol for multiple nights on sleep with a robust sample size, well-controlled pre-study screening of sleep and alcohol use habits [47, 48], and novel temporal analyses. Specifically, the present study addressed concerns with experimental designs from prior studies by monitoring sleep before the study using actigraphy, sleep diaries, and phone calls and by ensuring each participant passed urine drug screenings and had no alcohol in their system (tested with a breathalyzer) [49] each day of the in-lab study. The current study controlled for circadian factors by maintaining consistent timing of meals and alcohol administration prior to each participant’s bedtime. Finally, the participants in this study include more diversity in gender and age to produce a richer dataset that better reflects a diverse population. The present study also provides a high-resolution look at the dynamics of alcohol-induced changes to sleep architecture using GAMMs. For example, the GAMMs indicated a significant difference in the proportion of time spent in SWS across much of the night on alcohol nights compared to mixer-only nights; however, when looking at the data from the full-night LME, we see no detectable difference in SWS between alcohol night and mixer-only nights. This difference would not be detected using coarser analyses (i.e. full-night summary values). Given that many prior studies investigated the impact of alcohol on sleep across the whole night or large sections of the night, this averaging of the data may dampen the effects of alcohol on sleep architecture, and thus, this may underlie some of the inconclusive or conflicting findings in prior studies.

Furthermore, the present study bridges a gap in the current literature by examining the effects of consecutive nights of presleep alcohol on sleep architecture using GAMMs. By analyzing the differences between our absolute and proportional accumulation GAMMs, we describe sleep architecture changes that occur across consecutive nights of alcohol use with higher resolution than previously shown.

High-resolution imaging of sleep architecture

This study documented the utility of viewing sleep through a temporal lens that allows the identification of specific windows most affected by alcohol. More broadly, using high-resolution analyses provides additional insights into the effects of other sleep interventions on a moment-by-moment basis. These methods can be expanded to provide detailed pictures of how a given challenge (including drugs of abuse, medications, and sleep timing manipulations) impacts sleep across the night. Furthermore, GAMMs and similar modeling may be helpful when examining the effects of various sleep disorders, as well as biological variability in age and sex, that fall outside the scope of the current investigation. Using such fine-grain visual analyses may also aid in detecting sleep changes that are missed using summary-based statistics. In this study, for example, individuals did not spend a significantly different amount of time in SWS but demonstrated significant and lasting changes in the time course of SWS accumulation. This significant impact was overlooked in analyses of SWS for the whole night.

Generalized additive mixed-effect modeling of sleep architecture

Our temporal analysis identified a window beginning about 1.8 hours after sleep onset where SWS appeared most affected by alcohol administration as the cumulative amount of SWS increased more rapidly than the mixer baseline variable (Figure 4A). This pattern was found on each of the three alcohol nights compared to the mixer baseline variable, with a slight delay occurring on alcohol nights 2 and 3. Given that 1.5 hours is approximately the length of a sleep cycle, these data suggest that the effects of alcohol on SWS are more pronounced following the first sleep cycle. Considering SWS often predominates in the first sleep cycle during normal sleep, this trend may also be more notable in the second sleep cycle, when SWS levels begin to decrease as alcohol is metabolized.

We noted that differences in REM sleep also do not become apparent until after the first 1.5 hours and continue to differ from the baseline until 7.8 hours into the sleep episode on alcohol night 1 and until about 5.4 hours into the sleep episode on alcohol nights 2 and 3. Interestingly, % REM approached mixer levels more quickly on alcohol nights 2 and 3 than on alcohol night 1. Furthermore, our GAMM of absolute REM sleep time indicated that participants spent less total time in REM sleep on nights with alcohol compared to without. The first night’s administration of alcohol resulted in a large, statistically significant decrease in total REM sleep (approximately 11-minute decrease in REM sleep), and these effects were less pronounced on subsequent alcohol nights (approximately 4-minute decrease). Taken together, these results indicate that the REM deficit in alcohol night 1 is not as pronounced on the following two alcohol nights despite alcohol consumption remaining consistent, suggesting that REM sleep may be less impacted by the same dose of alcohol across consecutive nights. Furthermore, these results suggest that the nervous system adapts to alcohol levels to ensure time is spent in REM sleep following a deficit, perhaps highlighting REM sleep’s important evolutionary role [46, 58, 59].

The effects of presleep alcohol on sleep architecture are examined with linear mixed-effect modeling

Linear mixed-effect modeling allowed for the general trends in the data to be examined using statistics that are familiar within the scientific community. Obtaining reportable statistics remains an important aspect of scientific communication, and the utility of relaying the results of our LME models facilitates this and provides a complement to our GAMMs. According to the LME models, alcohol consumption significantly increased % SWS in the first third of the night; however, the effects of SWS varied across the night (increased T1 and decreased for T2 and T3), resulting in no detectable difference when averaging across the full night. Furthermore, there was a main effect of alcohol indicating a decrease in percent time spent in REM sleep on alcohol night compared to the mixer nights of the full night; however, this effect was only statistically significant for the first third of sleep (% REM T1) and not the latter two-thirds (% REM T2–T3) (Figure 2, E–H). These findings indicate that the main effect of alcohol on % REM was primarily impacted by differences that occurred within the first third of the night. Furthermore, these findings may suggest that the neural mechanisms underlying REM sleep may be sensitive to acute intoxication and that these effects decrease as alcohol levels decrease.

The gradual reduction of the REM sleep deficit across each alcohol night, as shown by our GAMMs, may indicate that LME modeling is insufficiently sensitive to detect temporal changes in sleep. Taken together, these data emphasize the need for improved resolution time course of sleep stages across the night to ensure that effects are not shrouded by averaging across the night.

Consecutive nights of alcohol use differentially impact SWS and REM sleep variables

Our data indicate that over multiple nights of drinking there may be an attenuation of the impact of alcohol on REM sleep, whereas the effects on SWS appear to remain constant across consecutive nights of presleep alcohol. The persistent differences in relative REM sleep GAMMs across all three alcohol nights, emphasizing the lasting time-course difference of REM sleep accumulation, complimented by the diminishing differences in the absolute REM sleep GAMM across alcohol nights, further suggests that some compensatory mechanisms rescue the alcohol-induced decrease in total REM following presleep alcohol.

Underlying mechanisms of alcohol-induced sleep architecture changes

The reported increase in SWS at the start of that night, which is noted in the present study, is in congruence with several previously published studies [22–24, 26, 27, 32]. The increased time spent in SWS is speculated to be attributed to alcohol’s depressant effects on the nervous system, Alcohol is known to increase the activity of the inhibitory GABAergic neurons and decrease the activity of the excitatory glutamatergic system, resulting in hyperpolarization of much of the cortex, which is associated with an increase in synchronous activity across the brain [26, 27]. This synchronous activity of cortical neurons is reflected in the EEG as high amplitude, low-frequency waves in the delta range (<4 Hz); that is, the “slow wave” frequency after which SWS is named. It follows from its depressant effects that alcohol would alter sleep by increasing early SWS. However, it is important to emphasize that several studies have failed to identify an effect of presleep alcohol on SWS [28–33]. Considering that alcohol induces a neural environment conducive to SWS, other sleep stages (like REM sleep) may not be actively inhibited rather, less time is spent in all other stages as a function of more time spent in SWS.

Integration of current study into contemporary literature

The data collected in this study provides a high-resolution complement to the contemporary PSG literature. Overall, our trends of increased SWS and decreased REM at the start of the night have been demonstrated in prior work; however, this study adds to this literature by demonstrating both effects in the same population. Many prior studies only detected an effect on either SWS or REM, with few detecting both [24, 27]. We may have detected differences in sleep architecture that were missed by averaging across larger portions of the night in prior work. Furthermore, our study showed that alcohol did not significantly affect wake during the second half of the night. We see the most variation in time spent in wake during alcohol night 1 (Figure 4C) and this variation decreases on subsequent nights. This may suggest that some of the wakefulness recorded in prior studies may be partially due to the first-night effect that has been documented in previous sleep studies. This emphasizes the need to allow participants to adapt to sleeping in the lab before applying an experimental variable. Finally, our work demonstrates that brain regions underlying sleep can compensate for changes early in the sleep episode to strive to maintain comparable amounts (compared to nights with no alcohol) of each sleep stage. However, it is important to emphasize that our work shows a rebound during the second half of the sleep episode and that many individuals do not sleep over 8 hours outside of the lab. Thus, our data underscores the importance of communicating the need for sufficient sleep to combat the negative side effects of presleep alcohol use.

Clinical implications

Gaining a better understanding of the dynamic way in which alcohol modifies sleep architecture can inform future treatment strategies. By characterizing the effects of presleep alcohol at higher resolutions, we can begin to unpack intervals of time that may be especially sensitive to the effects of alcohol. For example, the data presented in this study demonstrates a decrease in time spent in REM sleep across much of each alcohol night compared to mixer only. However, this disparity decreases (as seen on alcohol night 1) or disappears (as seen on alcohol nights 2 and 3) by the end of the sleep period. This suggests that the body compensates for the initial decrease in REM sleep to ensure comparable amounts of REM occur on alcohol nights, but only during the latter part of the night. Given the time in bed in our laboratory allows for an adequate night’s sleep (~8.5 hours) and that many adults in the United States report sleeping less than 7 hours a night, they may miss this opportunity to recoup lost REM sleep [60]. Considering the important role that REM plays in memory and emotional regulation [61–63], it may have great benefits to communicate this effect with individuals who are trying to decrease their drinking.

Although our study expanded the number of participants included in the study and maintained an equal number of men and women, we still were not able to greatly increase the racial diversity included in the study. This is due, in part, to the population surrounding our immediate area consisting primarily of individuals from Caucasian backgrounds. In the future, it is incredibly important to explore these topics with individuals from more diverse backgrounds. By continuing to expand participant pools to gather richer data about the widespread effects of presleep alcohol and the potential individual differences in sensitivity.

In summary, consecutive nights of presleep alcohol increased the rate of accumulation of SWS and decreased the rate of accumulation of REM sleep. The recovery of REM sleep following an initial deficit on alcohol night 1 may indicate that tolerance to the dose of alcohol occurs for that stage. Interestingly, such tolerance to alcohol was not seen for SWS. This suggests that the sedating effects of alcohol may interfere with the nervous system’s normal sleep regulation when alcohol levels are high at the start of the night. Importantly, these trends are demonstrated across a short period of alcohol use. Taken together, these findings underscore the detrimental impact of alcohol on sleep; thus, frequent use of alcohol before sleep should be further investigated to better understand the scope of these concerns. The potential impacts on cognition and memory, as well as the risks to cardiovascular health, need further exploration to inform the populations of potential risks associated with habitual use.

Funding

This work was supported by NIH grants: R01AA025593.

Disclosure Statements

Financial disclosure: None.

Nonfinancial disclosure: None of these organizations had any role in the study conception, design, interpretation, or the decision to publish these data. The findings and conclusions in this publication are those of the authors and do not represent the views of the U.S. Department of Veterans Affairs or the National Institutes of Health, and do not represent any US Government determination, position, or policy.