Eight adult male and twelve adult female Wistar rats were obtained from Charles River Laboratories (Raleigh, NC, United States) and housed in pairs under a reversed light/dark (12:12 h) cycle with ad libitum access to food and water within a temperature-controlled vivarium (22–23°C). At the start of the experiment, male and female rats weighed approximately 250–350 g, with approximate age-matching between sexes (9–10 weeks old). Their body weights were monitored throughout the course of the experiment (Supplementary Figure 1A). All procedures conducted during this experiment were approved by the Animal Care and Use Committee of the National Institute on Drug Abuse Intramural Research Program.

Animals were anesthetized using isoflurane and implanted with F40-EET telemetry devices (Data Sciences International; DSI; St. Paul, MN, United States) in the lower right abdomen and secured into the muscle wall. Two electromyography (EMG) and electroencephalography (EEG) leads each were secured into the nuchal muscle and skull (A/P: +1.0 mm, M/L: +1.0 mm; A/P: −2.0 mm, M/L: −1.0 mm), respectively. During the 3-day post-operative period, animals were treated with meloxicam (1 mg/kg) and gentamicin (1 mg/kg). Animals recovered for 4–6 weeks before the experimental procedures continued.

Following recovery from surgery and baseline recordings, all animals were pair-housed in home cages that were placed inside the chronic ethanol vapor chambers to induce alcohol dependence as described previously (Gilpin et al., 2008; Vendruscolo and Roberts, 2014). The chronic, intermittent ethanol vapor (CIEV) model of alcohol dependence elicits rapid tolerance, without detrimental health effects, typically within 2–4 weeks. Six to eight hours after the end of exposure to CIEV, rats exhibit both somatic and motivational signs of withdrawal, and the motivational signs can still be observed 3–5 weeks into alcohol withdrawal (Gilpin et al., 2009; Koob et al., 2014; Vendruscolo and Roberts, 2014). Mean body weights at the start of CIEV exposure were 277.4 ± 6.6 and 460.9 ± 6.2 g for females and males, respectively (Supplementary Figure 1A; baseline). Here, rats underwent cycles of CIEV for 12 h ethanol vapor OFF [zeitgeber time (ZT) 0–12, where ZT 0 indicates beginning of light phase] and 12 h ethanol vapor ON (ZT 13–24, where ZT 13 indicates beginning of dark phase). The amount of alcohol vaporized was independently controlled per sex to maintain the desired BAL in the range of 150–225 mg/dL, regardless of body weight. We targeted BALs in this range as it is known to cause both somatic (e.g., rigidity, irritability, hyperalgesia) and motivational (e.g., anxiety-like behavior, increased alcohol self-administration) signs of withdrawal (Gilpin et al., 2008, 2009; Vendruscolo and Roberts, 2014; Glover et al., 2021). Animals were tested weekly for BALs to determine intoxication induced by CIEV exposure. Mean BALs of 195.2 ± 13.4.9 and 226.0 ± 17 mg/dL were achieved for the females and males, respectively, after 4 weeks of CIEV exposure (Supplementary Figure 1B; acute withdrawal). Mean body weight after 4 weeks of CIEV exposure were 303.2 ± 6.0 and 550.1 ± 15.9 g for females and males, respectively (Supplementary Figure 1A).

All sleep/wakefulness recordings were collected from ZT 0–12. Rats were first habituated to their home cages within the ethanol vapor chambers (vapor OFF), and baseline data were collected before any exposure to alcohol. Following baseline, rats were exposed to CIEV for 4 weeks. For acute alcohol withdrawal, recordings started at ZT 0 immediately following the last vapor exposure (i.e., the preceding ZT 13–24). BALs are expected to decrease continually from the targeted 150–225 mg/dL at time ZT 0 and reach negligible BALs in about 4.5 h as shown previously in dependent male rats (Gilpin et al., 2009). Protracted abstinence data were collected 4 weeks following acute withdrawal (see Figure 1 for experiment timeline). Sleep macroarchitecture (i.e., time spent in each sleep/wakefulness state and sleep onset latency) and microarchitecture (bout rate, bout duration, and sleep spindle characteristics) were analyzed for each timepoint (baseline, acute withdrawal, and protracted abstinence). For telemetry recordings, the EEG and EMG signals were amplified using a Data Exchange Matrix and acquired using Ponemah software (DSI, version 5.2). Each 12-h recording was digitized at a rate of 500 Hz and then decomposed using a Fast Fourier Transformation to parse out each wavelength into 0.25 Hz bins. A bandpass filter (0.5–50 Hz) was used to exclude any incoming noise at high and low frequencies. EEG/EMG signals were manually scored in 2-s epochs into three general vigilance/sleep states: rapid eye-movement (REM) sleep, non-rapid eye-movement (NREM) sleep, or wakefulness (see Figures 2A,C,E for representative traces) using Neuroscore software (DSI, version 3.3). Transitioning into a different sleep state occurred only if the new state lasted 10 s or more. If the new sleep/wakefulness state was shorter than 10 s, then the animal did not undergo a transition, and the most recent scored state persisted. The time the rat spent (minutes) in a specific sleep/wakefulness state was used to calculate sleep onset latency (minutes from ZT 0 h to maintain REM or NREM sleep for at least 1 min), average bout rate (number of bouts/minute spent in sleep state) and average bout duration (number of minutes/bout). Sleep spindles occurring in NREM sleep were detected and analyzed using the Detect Spindles EEGLAB plugin (Ray et al., 2015), which is a validated, automated method of detecting sleep spindles. Complex Demodulation was used to extract power (in μV) within the sigma frequency band (11–16 Hz), and the signal was transformed into z-scores and utilized a 60-s sliding window to account for intra- and inter-individual differences in spindle characteristics. Parameters for spindle detection consisted of a central frequency of 13.5 Hz, a bandwidth of 5 Hz, a minimum of 0.25 s between each spindle event, and a minimum duration of 0.49 s for each spindle event. Spindle duration, frequency, and peak amplitude (μV) were extracted across all NREM sleep epochs over the 12-h recording. The average duration and intra-spindle frequency of sleep spindles for each hour were calculated in MATLAB. Additional number (counts) and amplitude measures for sleep spindles can be found in Supplementary Figure 5. Spectral power was examined for REM and NREM sleep over the 12-h recording using the MATLAB plugin EEGLAB (Delorme and Makeig, 2004; see Supplementary Figure 6).

All data are expressed as mean and standard error of the mean (+SEM). Statistical analyses were performed using GraphPad Prism software (version 8.02). The time spent in each sleep/wakefulness state (REM, NREM, and wakefulness) for each hour within the 12-h period (ZT 1–12) was collapsed across sex (no sex differences were observed in an analysis of the whole 12-h period; see Supplementary Section 1.2). The time spent in each sleep/wakefulness state within each hour was then analyzed using a repeated-measures two-way analysis of variance (ANOVA) with alcohol withdrawal state (baseline, acute withdrawal, and protracted abstinence) as the within-subjects factor and time as a between-subjects factor. Bout rates, bout duration, and sleep onset latency for REM and NREM sleep states over the whole 12-h period (ZT 0–12) were each analyzed using a repeated-measures two–way ANOVA, with alcohol withdrawal state (baseline, acute withdrawal, and protracted abstinence) as the within-subjects factor and sex as a between-subjects factor. Due to missing data points where values equaled 0 (i.e., the rat had no sleep spindles during that hour), spindle duration and frequency were analyzed using a mixed-effects model, applying a restricted maximum likelihood approach to analyze repeated-measures. When appropriate, post hoc comparisons were performed using Tukey’s multiple-comparison test. For all analyses, α < 0.05 was considered statistically significant.

Each stage of the addiction cycle (binge/intoxication, withdrawal/negative affect, and preoccupation/anticipation) is associated with unique neuroadaptations and to various sleep disturbances, alterations of sleep architecture, and the development of insomnia in individuals with AUD (Koob and Colrain, 2020). A primary focus on changes in sleep macroarchitecture during the binge/intoxication stage in the literature leaves a significant gap in our understanding of acute alcohol withdrawal and protracted abstinence on sleep among individuals with AUD. To date, there has been inconsistent clinical and preclinical evidence regarding the influence of alcohol abstinence on the time spent in REM or NREM sleep, a discrepancy largely relative to the number of days of abstinence. The time spent in REM sleep in recently or long-term abstinent individuals with AUD compared to controls has been reported to either not change or increase, whereas the time in NREM sleep increases or decreases, but largely has been shown to decrease (for review, see Koob and Colrain, 2020). Similar variabilities in the time spent in REM and NREM sleep have been reported in preclinical rodent studies, depending on the type and length of alcohol exposure, and the number of days of abstinence (Ehlers and Slawecki, 2000; Veatch, 2006; Thakkar et al., 2010, 2015; Sanchez-Alavez et al., 2019; Sharma et al., 2021). The current studies used a 4-week exposure to ethanol vapor during the dark cycle, with sleep recorded during the light cycle prior to and during acute withdrawal, and also following protracted abstinence. Somatic and motivational signs of withdrawal in rats have been observed in as few as 2 h after CIEV exposure (for review, see Vendruscolo and Roberts, 2014), with brain alcohol levels returning to baseline by 4.5 h post-CIEV (Gilpin et al., 2009). The current studies found significant changes in sleep macroarchitecture largely in the second half of the sleep/light cycle while the rats were in acute withdrawal.

The time spent in REM sleep decreased in both females and males during acute alcohol withdrawal with the changes occurring during the ZT 6, 7, 8, and 10 h time points. In contrast, decreased REM sleep has been observed in humans in the first half of the sleep cycle following acute alcohol intoxication, likely due to initially higher BALs and increased time in SWS (Rundell et al., 1972; Prinz et al., 1980). The decrease in REM sleep during acute withdrawal in the current studies returned to baseline levels following protracted abstinence. These findings are consistent with a previous study showing a decrease in REM sleep in male rats 1 day into withdrawal following chronic alcohol exposure, followed by a REM sleep rebound by day 3 of alcohol withdrawal (Mendelson et al., 1978). This contrasts with a study conducted in male mice chronically exposed to ethanol vapor, showing no change in the percent time spent in REM sleep 1 day into alcohol withdrawal, and a subsequent increase in the percentage of REM sleep by day 3 of alcohol withdrawal (Veatch, 2006). It is notable that both of these prior studies used a significantly shorter, albeit repeated, exposure to alcohol than the current studies, making direct comparisons difficult. The current studies also showed a return of REM sleep to baseline levels during protracted abstinence compared to acute withdrawal in both females and males. The decrease in REM sleep during acute withdrawal may be a product of repeated stress associated with alcohol withdrawal experienced during CIEV. Likewise, other studies have shown that repeated restraint or footshock stress can cause a decrease in both NREM and REM sleep in female and male rats (Jean Kant et al., 1995; Papale et al., 2005; Hegde et al., 2011; Gargiulo et al., 2021). Furthermore, it has been hypothesized that REM sleep is associated with the processing and stabilization of memory transformed into long-term memory during SWS (for review, Rasch and Born, 2013). Thus, whereas an increase in REM sleep may facilitate the consolidation of aversive alcohol withdrawal memory acquired during wakefulness, it is possible that the inverse, such as the reduction in REM sleep during acute withdrawal observed herein, may be protective to some degree. It is also possible that rats experienced rebound sleep during the dark phase (not recorded here), which would mimic the REM sleep rebound observed in humans that is used as a predictor for the propensity for relapse (Gillin et al., 1994; Brower et al., 1998; Clark et al., 1998). Indeed, a suppression of REM sleep during the light phase and a subsequent increase in REM sleep during the dark phase 3 weeks following chronic alcohol cessation in male rats have been reported (Kubota et al., 2002).

In regards to NREM sleep macroarchitecture, we found no changes in overall time spent in NREM sleep compared to baseline, although there was an increase in NREM sleep during protracted abstinence compared to acute alcohol withdrawal. There are inconsistent reports in the literature of preclinical chronic alcohol models addressing the impact of acute alcohol withdrawal or protracted abstinence on NREM sleep and SWS, ranging from a decrease in NREM sleep during acute withdrawal (Veatch, 2006; Sharma et al., 2021) to having no impact on the time spent in SWS in either acute withdrawal or protracted abstinence (Ehlers and Slawecki, 2000) in males. The differences in the route of chronic alcohol administration and the time of withdrawal between each of these studies may be contributing to opposing findings. Additionally, considering that rats spend 50–70% of their time asleep during the light cycle, subtle changes in the total time spent in NREM sleep may be difficult to capture, suggesting that other measures of NREM sleep microarchitecture may also be useful.

The current studies found a sex-specific effect on REM sleep onset latency, as demonstrated by shortened REM sleep onset latency following protracted abstinence in females. There were no effects of alcohol withdrawal on NREM sleep onset latency in either females or males. This contrasts with previous reports of an increase in NREM sleep onset latency during acute withdrawal (Sharma et al., 2021) or protracted abstinence (Ehlers and Slawecki, 2000) in males, or no change in REM or NREM sleep onset latency during early abstinence in females (Amodeo et al., 2020). Typically, a shortened sleep onset latency indicates increased tiredness and sleep pressure, potentially due to a prior sleep deficit or some underlying chronic sleep disorder. Clinically, it has been reported that shortened REM sleep onset latency is associated with higher relapse rates in individuals with AUD (Gillin et al., 1994; Brower et al., 1998). Investigating whether a shortened REM sleep onset latency during protracted abstinence is associated with the reinstatement of alcohol seeking or another rodent alcohol-relapse model is an important topic for future studies.

The current studies demonstrate an effect of alcohol withdrawal after chronic alcohol exposure on sleep macroarchitecture in both female and male rats, as shown by a reduced time spent in REM sleep in females and males during acute withdrawal, and decreased REM sleep onset latency during protracted abstinence in females. These results highlight the importance of REM sleep and including both sexes into future studies investigating alcohol withdrawal effects on sleep.

Although standard sleep measures (i.e., macroarchitecture), such as the time spent asleep or awake and sleep onset latency, are important, they do not completely characterize sleep dysfunction. Thus, the current studies considered further analyses of sleep microarchitectural components, including sleep bout and spindle characterization, during acute alcohol withdrawal and following protracted abstinence from alcohol. The current studies revealed an overall higher REM sleep bout rate in females compared to males, although there was no effect of withdrawal state on REM sleep bout measures. However, the examination of hourly data for females and males combined showed an overall decrease in REM sleep bout duration during acute withdrawal that returned to baseline levels during protracted abstinence (see Supplementary Figure 3). This decrease in REM sleep bout duration may reflect the reduced time spent in REM sleep, or could be indicative of reduced quality of REM sleep, particularly in the latter half of the light/sleep cycle. Further REM sleep bout analysis is warranted to determine whether REM sleep is asynchronous, or fragmented, during acute withdrawal. An inverse outcome on wakefulness bout measures was also noted, with a decrease in wakefulness bout rate and increase in wakefulness bout duration during acute withdrawal. Finally, although there was a significant decrease in NREM sleep bout rate during protracted abstinence in males, suggesting a longer time spent in each bout of NREM sleep, there was no significant effect of withdrawal state on NREM sleep bout duration detected in either males or females. Our findings are in contrast to another study in which they observed decreased NREM sleep bout duration during acute withdrawal following CEIV (Sanchez-Alavez et al., 2019), although this could be due to differences in length of alcohol exposure.

Although there was no significant effect of alcohol withdrawal on the overall time spent in NREM sleep, the examination of sleep spindles, a hallmark of NREM sleep, revealed changes in NREM sleep microarchitecture in both female and male rats. During acute alcohol withdrawal, despite no overall change in spindle duration, there was an increase in intra-spindle frequency (i.e., the rate of spindle oscillation) that returned to baseline levels during protracted abstinence. Sleep spindles are considered to be sleep-protective elements, shielding NREM sleep from spontaneous sensory input or disruptions, possibly at the expense of REM sleep (Fernandez and Lüthi, 2020; Sippel et al., 2020). Thus, the lack of change in the time spent in NREM sleep and the decreased time in REM sleep found in the current study may reflect the influence of NREM sleep spindles during acute withdrawal from alcohol. Stress has consistently been shown to be a profound contributing factor to AUD, particularly during alcohol withdrawal in both humans and rodents (Becker, 2012; Tunstall et al., 2017; Vendruscolo and Koob, 2020). Both clinical and preclinical studies have demonstrated a consistent interaction between physiological systems of stress/overactive arousal and sleep (Rodenbeck and Hajak, 2001; Riemann et al., 2015; Kalmbach et al., 2018; Lo Martire et al., 2020) including increased spindle frequency in individuals with post-traumatic stress disorders (Van Reeth et al., 2000; Wang et al., 2020). Therefore, the increased stress response associated with acute alcohol withdrawal may contribute to the increase in intra-spindle frequency observed in the current studies. However, the role of stress systems in sleep dysfunction associated with alcohol withdrawal has yet to be elucidated.

Together, these results indicate a role for acute alcohol withdrawal in the changes in sleep microarchitecture. Although standard sleep measures, such as the time spent asleep or awake, are important tools, restricting sleep measures to standard macroarchitecture components may be limiting our full understanding of alcohol-related sleep dysfunction. Here, we showed that microarchitectural changes not only in the continuity of sleep state, but also in sleep-spindle elements. These indices of sleep microarchitectural change during alcohol withdrawal and may be potential sleep biomarker candidates for the prediction of AUD treatment outcomes in future studies.

A limitation of the present study is that we did not track estrous cycle in female rats throughout the experiment. The estrous cycle has been reported to influence sleep patterns, with proestrus being associated with reduced REM sleep, enhanced waking, and increased sleep spindle density at the end of the light cycle when progesterone and estradiol are highest (Schwierin et al., 1998; Swift et al., 2019). Regarding alcohol drinking and CEIV-induced alcohol drinking, we have reported that the estrous cycle had minimal influence on these behaviors (Priddy et al., 2017). Hormonal effects during estrous cycle also seem to be most influential during acquisition of drug-taking behavior and less so once compulsive drug-taking has been established (Becker and Koob, 2016). However, whether there is an interaction effect between passive alcohol exposure and estrous cycle on sleep architecture remains to be determined.

An additional possible limitation of the current studies lies in the sleep spindle characterization. It has been reported that alcohol decreases sigma band power (10–15 Hz), typically associated with sleep spindles, during NREM sleep in humans (Dijk et al., 1992; Chan et al., 2015). In analyzing the sigma band during NREM sleep for spindles in the current studies, we did not differentiate between slow (<12 Hz) and fast (>12 Hz) oscillating spindles, which largely vary based on topographical location (Terrier and Gottesmann, 1978; Gandolfo et al., 1985; Sitnikova et al., 2014). It remains to be seen whether the distinction between slow and fast spindles has any functional significance, although evidence suggests that fast oscillating sleep spindles are associated with sleep-dependent memory processing (Mölle et al., 2011), which may have implications for processes of extinction and thus, addiction recovery (Feld and Born, 2020; Valentino and Volkow, 2020). This is an exciting future direction for sleep microarchitecture research in the field of addiction.