Increased REM activity (i.e., shortened REM latency, longer duration of the first REM period, higher percentage of REM over the course of the night, and greater REM density) has been widely reported in unipolar depression (e.g., Benca et al., 1992), and REM sleep has been proposed to have a mood-regulatory role that malfunctions in mood disorders (Cartwright et al., 1998). REM in BD has received less empirical attention than in unipolar depression, and the evidence is mixed. BD patients who are manic or depressed have been reported to exhibit decreased (e.g., Hudson et al., 1992), increased (e.g., Giles et al., 1986), or equivalent REM latencies (Jernajczyk, 1986) relative to healthy controls and to patients diagnosed with unipolar depression. In the inter-episode period, BD patients have shown sleep marked by increased REM activity (Knowles et al., 1986) and sleep that does not differ from that of controls (Sitaram et al., 1982). Studies of healthy relatives of individuals diagnosed with a mood disorder point to the possibility that disturbances in REM sleep may be a marker of affective disorder vulnerability. Such studies suggest that the sleep of relatives of individuals diagnosed with unipolar depression or bipolar disorder is marked by increased REM activity compared to controls (Giles, Kupfer, Rush, & Roffwarg, 1998; Modell, Huber, Holsboer, & Lauer, 2003). Additionally, greater REM density has been found to distinguish those relatives who go on to experience an affective episode from unaffected relatives and controls (Modell, Ising, Holsboer, & Lauer, 2007).

Slow wave sleep (SWS), which consists of Stages 3 and 4, is thought to have a restorative function in healthy individuals that includes memory consolidation and cell regeneration, and has been found to be decreased in unipolar depression (e.g., Benca et al., 1992). The two-process model of sleep regulation proposes that SWS is an index of the homeostatic process (Borbely, 1982), which may be weakened in depression (Borbely, 1987). Decreased SWS has also been reported in bipolar depression (Jovanovic, 1977) and in depressed adolescents who go on to have a BD illness course (Rao et al., 2002). However, most studies report SWS in mania and bipolar depression to be equivalent to that of healthy controls (Hudson et al., 1988; Hudson et al., 1992; Jernajczyk, 1986; Mendelson et al., 1987). Thus, the manner in which SWS is disturbed in BD is unclear, particularly during the inter-episode period.

Stage 2 sleep is also thought to have an important restorative and sleep-protective function (De Gennaro & Ferrara, 2003; Hayashi et al., 2005). Although it is understudied in mood disorders, decreased Stage 2 sleep has been found in both unipolar and bipolar depression (de Maertelaer et al., 1987). Additionally, in studies of clozapine, treatment response in BD was most dramatically associated with the enhancement of Stage 2 sleep (Hinze-Selch et al., 1997).

The present study was designed to clarify the association between BD and REM, SWS and Stage 2 sleep. We aimed to investigate whether there is a cross-sectional association between symptoms and sleep architecture, and whether sleep architecture predicts symptoms and mood-associated impairment at 3 month follow-up in patients with BD during the inter-episode period relative to a non-psychiatric control group. Our first hypothesis was that higher levels of depressive and manic symptoms at the baseline sleep assessment would be associated with increased REM activity (shorter REM latency, higher REM%, longer duration of first REM, greater REM density), and that increased REM activity would be associated with higher levels of depressive symptoms, manic symptoms, and impairment at 3 month follow-up. Second, we hypothesized that increased Stage 2 sleep would be associated with lower levels of manic and depressive symptoms and impairment at the baseline sleep assessment and at 3 month follow-up. Given the mixed findings for SWS reported in the literature, we included SWS on an exploratory basis.

To address medication effects, two of the authors (PE and JG) independently classified each medication reported by participants as having no effect, an enhancing effect, or a suppressing effect on REM, SWS, and/or Stage 2 sleep, based on a detailed literature search. In cases where participants reported taking medications with conflicting sleep effects (e.g., Bupropion is a REM sleep enhancer and Fluoxetine is a REM sleep suppressor; Ott et al., 2004; Rush et al., 1998) we reasoned that these effects canceled each other out. A list of potential medication effects on the sleep of each participant was generated by the two authors for 33% of participants with a reliability of 100% for those coded by both authors. As reported in Table 2, ANOVAs and Tukey post-hoc comparisons indicated that participants taking a medication that suppressed REM had a significantly longer latency to first REM than participants taking medications with no effect or an enhancing effect on REM. Participants taking a medication that suppressed REM also had greater REM density than participants taking a medication with no effect on REM. There were no other significant differences in sleep architecture based on medications (all p's > 0.10) (see Table 3).

Table 3 presents correlations of baseline sleep architecture with baseline and 3 month follow-up symptoms and impairment. Sleep architecture was not significantly correlated with concurrent manic and depressive symptoms at the baseline sleep assessment in either participant group. However, in the BD group there were 4 significant correlations (out of 12) between baseline sleep architecture and follow-up symptoms. Duration of first REM and SWS% were positively correlated with manic symptoms, while REM density was positively correlated with depressive symptoms. Stage 2% was negatively correlated with manic symptoms. Additionally, there were 3 significant correlations (out of 6) between sleep architecture and follow-up impairment in the BD group. REM density and SWS% were positively correlated with impairment, while Stage 2% was negatively correlated with impairment. In the control group, sleep architecture was not significantly correlated with follow-up symptoms. However, there was one significant correlation (out of 6) for impairment, indicating that REM density was negatively correlated with impairment in the control group. Thus, 7 significant correlations between baseline sleep architecture and follow-up symptoms and impairment emerged in the BD group, while only 1 significant correlation was found in the control group.

The first aim was to cross-sectionally examine the association between sleep architecture and symptoms and impairment in BD relative to non-psychiatric controls. We found no significant correlations between sleep architecture and concurrent symptoms at baseline in either group, although greater REM density was found in the BD group relative to the control group. These findings are in contrast to our hypotheses that REM and Stage 2 sleep would correlate with concurrent symptoms. One potential explanation may be our requirement that all participants be inter-episode. By definition, this restricts the range of depressive and manic symptoms in the sample, and decreases a likelihood of finding significant correlations (e.g., Guilford & Fruchter, 1978). A second potential explanation is that the neutral mood induction administered prior to sleep may have diffused pre-sleep mood and canceled out the effects inter-episode symptoms may have had on sleep architecture. However, the lack of significant group differences in mood prior to, and following, the neutral mood induction renders this account unlikely. Our finding greater REM density in the BD group relative to controls is consistent with reports of increased REM activity in inter-episode BD (e.g., Knowles et al., 1986) and in relatives of individuals with affective illness (e.g., Modell, Huber, Holsboer, & Lauer, 2003), raising the possibility that increased REM density is a persistent sleep disturbance in BD. Given that we also found that participants who reported taking REM suppressing medications demonstrated greater REM density, this group difference may also be attributable to medication effects. We will return to this possibility below.

The second study aim was to prospectively examine correlations between baseline sleep architecture and future symptoms and impairment in BD. A number of significant findings emerged. Taking REM first, several significant correlations supported the hypothesis that increased REM activity would be associated with increased symptoms and impairment at follow-up in the BD group. Specifically, in the BD group longer duration of first REM was associated with more manic symptoms at 3 month follow-up, while greater REM density was associated with more depressive symptoms and impairment at 3 month follow-up. Thus, duration of first REM and REM density may have predictive value in inter-episode BD. The negative correlation between REM density and impairment in the control group, along with the findings that REM density is positively correlated with impairment and depression in the BD group, is consistent with the suggestion that REM serves a mood-regulatory role that is adaptive in non-psychiatric individuals but malfunctions in mood disorders (e.g., Knowles et al., 1986).

Several findings also supported the hypothesis that greater amounts of Stage 2 sleep would be associated with less follow-up symptoms and impairment in the BD group. Specifically, greater Stage 2% was associated with lower levels of manic symptoms and impairment in the BD group at follow-up. This is consistent with theories that Stage 2 sleep fulfills an important restorative function (Hayashi et al., 2005). Although Stage 2 sleep has received comparatively little attention in mood disorder research to date, our findings raise the possibility that Stage 2 sleep may have some protective value. Studies demonstrating positive associations between treatment response and increased Stage 2 sleep in BD (e.g., Hinze-Selch et al., 1997) are also suggestive of the importance of Stage 2 sleep.

Given the mixed findings in the literature to date, we did not make specific hypotheses regarding SWS, including it in our analyses on an exploratory basis. Results indicated that greater SWS% predicted more manic symptoms and impairment in the BD group at the 3 month follow-up. This was somewhat surprising, given that past research has suggested that increased levels of SWS appear necessary to normalize depressed mood (e.g., Borbely, 1982; Jovanovic, 1977). There are several possible explanations for our findings. First, in consideration with findings that greater Stage 2% was associated with less manic symptoms and impairment at the follow-up, it is important to note that the durations of Stage 2 sleep and SWS may change in proportion to each other. Indeed, while it appears that disturbances in REM and NREM sleep are not reciprocal (Armitage, 2007), changes in the different stages of NREM sleep are likely to impact other NREM sleep stages. Second, perhaps even prior to the baseline assessment the participants who later developed manic symptoms were starting to increase their physical activity. Even a slight increase in exercise and physical activity is known to increase SWS (Naylor et al., 2000). Third, participants who later developed manic symptoms may have already been displaying reduced sleep need prior to the baseline sleep assessment and were, therefore, becoming sleep deprived. Based on the two-process model of sleep regulation and sleep deprivation studies, increased wakefulness is associated with greater SWS over the course of the night (e.g., Borbely, 1982; Brunner, Dijk, & Borbely, 1993). Fourth, perhaps there are bi-directional effects. Specifically, perhaps greater SWS may be detrimental in BD, overcompensating for the need to regularize depressed mood and inducing a shift to the opposite end of the mood spectrum toward manic symptoms.

Out third aim was to address the role of psychotropic medications. To address potentially confounding effects of medications in our analyses, we employed a multi-step approach based on a detailed patient-by-patient literature review of medication sleep architecture effects. By applying this approach, in lieu of requesting that all participants be medication-free prior to participating, we reasoned that we would be able to recruit a participant sample that is more representative of the larger population of adults with BD. In the present study, the main concern was the association of REM suppressing medications with greater REM density. It is possible that this medication effect confounded correlations between REM density and symptoms and impairment at 3 months in the BD group. If this is the case, this medication-related finding raises a number of possibilities. First, BD individuals who are prone to depression may also be more likely to respond to REM suppressing medications. Second, REM suppressing medications may render individuals more susceptible to depression and impairment, although this seems unlikely as neither follow-up depressive symptoms (t=-1.74, p>0.10) nor impairment (t=-1.70, p>0.10) were greater in participants taking REM suppressing medications than in participants taking medications with no REM effects. Third, REM suppressing medications may affect the mood-regulatory function of REM, thereby intensifying the association of REM density with future depression and impairment. Thus, further research on REM sleep in BD is needed to clarify the relationship between medication effects, REM density, and depressive symptoms.

There were several key lessons learned from our approach to medications that should be addressed in future studies of BD. First, we recognize that some of the medication subgroups were small (for example, only 1 participant was taking a medication that suppresses Stage 2 sleep). Thus, some statistical tests were under-powered to detect meaningful group differences in sleep architecture based on medication regimen. Future studies with larger samples could address medication confounds by utilizing medication subgroups as covariates in statistical analyses. Second, medication usage should be assessed at follow-up to check if a medication change or discontinuation may have occurred, potentially influencing the associations observed between sleep architecture and follow-up symptoms and impairment. Third, we reasoned that sleep architecture effects of medications that enhanced and suppressed the same sleep stage would cancel out. Although this decision makes sense at one level, medication interaction studies have not yet been conducted to assess the validity of this assumption. Fourth, investigators should be aware that medications may be prescribed to participants on the basis of underlying factors, such as a tendency to exhibit particular symptom clusters (e.g., Swann et al., 2002). These underlying factors may affect both, sleep architecture and the participant's medication regimen. In order to assess this possibility, it may be useful to obtain a more detailed illness history or obtain permission to contact the prescribing physician in order to discuss the physician's reasons for prescribing the medications taken by participants. Finally, the effects of medications on sleep architecture are undoubtedly quite complex and include effects other than increasing and decreasing sleep stages. For example, benzodiazepine receptor agonists may increase EEG beta activity while decreasing delta and theta activity (Bastien et al., 2003). While we recognize the limitations of our approach to medications, we aimed to begin a process of developing a method for dealing with pharmacological treatments that would allow for the study of a diverse and representative sample of individuals diagnosed with BD. We hope that future investigations will further elaborate on our approach, utilizing lessons learned about the issues discussed above.

Several limitations should be considered. First, the sample size was relatively small and largely composed of female participants. Replication in larger and more diverse samples is warranted. Second, we did not correct for multiple comparisons. While this increases the risk of Type I error, given that this is a relatively new research area, we were also concerned about decreasing the risk of Type II error (Nakagawa, 2004) so as to maximize information gleaned from this initial study. The small number of significant correlations in the control group is noteworthy. On the basis of chance, if the results reflected false positives, more significant findings should have emerged in the control group. Third, we acknowledge that the stages of NREM sleep assessed are not entirely independent and that we cannot ascertain causal relationships between sleep stages and subsequent symptoms and impairment. Fourth, we utilized one night of PSG as our assessment of sleep architecture. Future studies should assess sleep architecture over multiple nights. Finally, the impact of comorbid diagnoses should be considered. It may be useful to include a comparison group with a diagnostic profile similar to the comorbid diagnoses in the BD group in order to ascertain the impact of comorbid disorders on sleep architecture.