Sleep and Parasympathetic Activity during Rest and Stress in Healthy Adolescents and Adolescents with Bipolar Disorder

Abstract

Objective:

Sleep disruption contributes to the pathophysiology of mental disorders, particularly bipolar illness, but the biobehavioral mechanisms of this relationship are insufficiently understood. This study evaluated sleep duration, timing, and variability as prospective predictors of parasympathetic nervous system activity during rest and social stress in adolescents with bipolar disorder, reflecting sleep-related interference in stress regulatory systems that may confer vulnerability to mood episodes.

Method:

Participants were adolescents with bipolar disorder (n=22) and healthy adolescents (n=27). Sleep duration and timing were measured by actigraphy for one week prior to a laboratory social stress task, during which high frequency heart rate variability (HF-HRV) was indexed using electrocardiography. Multilevel models were used to evaluate group, sleep characteristics, and their interactions as predictors of initial HF-HRV, and change in HF-HRV, during rest and stress.

Results:

Associations between group and changes in HF-HRV during stress were moderated by sleep duration mean (z=2.24, p=.025) and variability (z=−2.78, p=.006). There were also main effects of mean sleep duration on initial HF-HRV during rest (z=−5.37, p<.001) and stress (z=−2.69, p=.007). Follow-up analyses indicated that, in bipolar adolescents during stress, shorter and longer sleep durations were associated with lower initial HF-HRV (z=−5.44, p<.001), and greater variability in sleep duration was associated with less change in HF-HRV (z=−2.18, p=.029).

Conclusions:

Sleep durations that are relatively short or long, which are characteristic of mood episodes, are associated with parasympathetic vulnerability to social stress in adolescents with bipolar disorder. Obtaining regular sleep of moderate duration may favorably affect responses to stress in bipolar youth.

INTRODUCTION

Disruptions in sleep duration and circadian timing contribute to the onset, maintenance, and recurrence of psychiatric disorders, including depressive and manic episodes in bipolar disorder (1–3). For example, the onset of mood episodes in bipolar disorder is commonly preceded by insomnia and other sleep problems (4, 5), and is also associated with delayed circadian phase and evening chronotype (6). In laboratory studies, sleep deprivation can precipitate mania or hypomania in depressed adults with bipolar disorder (7). In clinical applications, there is emerging evidence that treatment for sleep and circadian disruption can prevent the recurrence of mood episodes in bipolar disorder (8).

Adolescence may be a particularly vulnerable period for sleep- and circadian-related mood disruption. A substantial body of research shows that bipolar disorder commonly onsets during adolescence (9), and is a valid diagnosis that can be reliably detected in youth (10, 11). Sleep is often shorter and later during normative adolescence due to delays in the endogenous circadian clock and reductions in sleep drive, leading to later bedtimes coupled with early wake-times for school and work that often results in insufficient sleep on weekdays (12). Consequent “make-up” sleep on weekends, coupled with adolescents’ greater responsibility for setting their own sleep schedules, also generates sleep schedule inconsistency (12). The combination of short and long sleep duration, later sleep timing, and sleep schedule inconsistency may all contribute to risk for mood episodes.

One mechanism by which sleep and circadian disruption may confer risk for mood episodes is by interfering with neuroautonomic stress response systems, including the parasympathetic nervous system (13). Parasympathetic activity can be indexed non-invasively by measuring high frequency heart rate variability (HF-HRV), which reflects the capacity of the central autonomic network to regulate cardiac stress response via the vagus nerve (14, 15). Low tonic parasympathetic activity, indicated by resting HF-HRV, is an early risk factor for cardiac-related morbidity and mortality (16), and it is consistently blunted in adults with bipolar disorder (17, 18). We also recently reported that adolescents with bipolar disorder have blunted HF-HRV (19). One intriguing possibility is that sleep disruptions, which are characteristic of mood episodes in bipolar disorder, contribute to HF-HRV by reducing parasympathetic tone. In support of this hypothesis, sleep-related impairment in HF-HRV is produced by experimental partial sleep restriction in healthy adults (20), and correlational studies indicate that low HF-HRV is associated with longer sleep latencies in children (21).

Disturbed sleep may be particularly impactful when it occurs in the context of stressful life events. For example, short sleep duration has been associated with larger decreases in HF-HRV during psychosocial stress in young adult men (22), and healthy sleep can protect adolescents from the detrimental effects of social stress on cognitive function (23). Notably, adolescents report more stressful life events than children (24, 25) and are more reactive to stressors than adults (26). If healthy sleep can protect adolescents from the detrimental effects of social stress, then sleep could be leveraged to promote healthy stress response in vulnerable populations. For example, management of sleep and social rhythms is one of the theoretical mechanisms of interpersonal social rhythm therapy for bipolar disorder (27, 28).

The aim of the present study was to evaluate whether sleep characteristics that have been associated with risk for mood episodes – namely, shorter and longer sleep duration, later sleep timing, and greater variability in sleep duration and timing – are associated with parasympathetic activity during rest and social stress in adolescents, and whether the effects of these sleep characteristics differ between healthy adolescents and adolescents with bipolar disorder. Based on evidence that short sleep is associated with low resting HF-HRV and greater HF-HRV reactivity in healthy young adults (20, 22), and that both short and long sleep durations are associated with negative health outcomes (29), we predicted that shorter and longer sleep would be associated with low resting HF-HRV, and more HF-HRV reactivity to social stress. In addition, we expected sleep duration, timing, and variability in sleep duration and timing to differ between adolescents with bipolar disorder and healthy controls, and these sleep characteristics were expected to moderate associations between bipolar disorder and HF-HRV during rest and social stress.

Methods

Participants

Participants were adolescents with bipolar disorder (n=27; type I n=9, type II n=9, not otherwise specified n=9) and psychologically healthy adolescents (n=28) who participated in a study of sleep, stress response, and brain function. Data were collected between 2/2011 and 2/2013 using procedures approved by the University of Pittsburgh IRB (#10060014). Details about the study participants and procedure for assessing stress reactivity are described in a previous report (19). Briefly, adolescents with bipolar disorder were recruited from a specialty bipolar disorders clinic at the University of Pittsburgh Medical Center, and healthy adolescents were recruited from the community. Exclusion criteria for both groups included: evidence of intellectual disability, pervasive developmental disorder, or organic central nervous system disorder (e.g., epilepsy) during medical record review or diagnostic assessment; or a life-threatening medical condition requiring immediate treatment. In addition, individuals with bipolar disorder were excluded if they had current psychiatric symptoms severe enough to require hospitalization, or an apnea-hypopnea index of ≥15 events/hr of sleep assessed via ApneaLink (ResMed, San Diego, CA). Healthy participants were excluded if they had a past mood or anxiety disorder or any current psychiatric diagnosis; a first-degree relative with history of bipolar disorder; a sleep disorder or other major medical problem, including apnea-hypopnea index of ≥15 events/hr of sleep; or were taking medications known to affect sleep/wake function.

Clinical Assessments

The Schedule for Affective Disorders and Schizophrenia for School-Age Children – Present and Lifetime (30) was used to assess current and past psychiatric disorders defined by the Diagnostic and Statistical Manual for Mental Disorders (DSM-IV-TR; 31). Detailed procedures for diagnosis of bipolar type I, II, and not otherwise specified are described in previous reports (32, 33). Differences between the DSM-IV and DSM-5, in which the major change was reclassification of mixed episode subtype to mixed-episode specifier, do not have significant implications for the generalizability of our results to samples diagnosed with more recent DSM-5 criteria. Out of 13 recordings reviewed for inter-rater reliability (23.6%), Kappa statistics were high for diagnosis of bipolar disorder (0.90), differential diagnosis of bipolar disorder subtypes (0.79), and diagnosis of disorders other than mood disorders (0.80). Current sleep disorders were assessed using a locally-developed Structured Clinical Interview for Sleep Disorders and ApneaLink assessment.

Depressive symptom severity and body mass index (BMI) were assessed using the 33-item Mood and Feelings Questionnaire–Child Version (MFQ; 34), and measures of height and weight, respectively. Scores above 29 on the MFQ indicate a probable depressive episode (35). The internal consistency of the MFQ was excellent (Cronbach’s alpha=0.965, N=48).

Medical record review and participant report were used to identify psychiatric medications that were current at the initial study visit. Medications were given one of six classifications – antidepressant, antipsychotic, anxiolytic, mood stabilizer, sedative/hypnotic, or stimulant – and the dose of each medication was coded as absent (0), low (1), or high (2) using a 3-point rating scale (36). Low and high dosages were compared to the mean effective daily dose based on chlorpromazine equivalents for antipsychotics (37), and referenced to the midpoint of the recommended daily range in the Physician’s Desk Reference for other medications (38). Ratings were summed to represent the overall medication load within medication class.

Actigraphy

Sleep characteristics were recorded for up to 14 days preceding the social stress task using the Actiwatch 2 (Philips Respironics, Murraysville, PA) and a modified electronic version of the Pittsburgh Sleep Diary (39). Actigraphy is a valid measure of sleep-wake patterns when combined with behavioral and/or self-report measures of bedtime and wake time (40, 41). Actiwatches were worn on the non-dominant wrist, and participants were instructed to press an event-marker button at bedtime (i.e., in bed and trying to sleep) and wake time (i.e., out of bed and starting the day) for major sleep intervals and naps. Philips Actiware software was used to prepare the data for analysis. Rest intervals were identified by a trained research assistant based on the following rule: event markers if present and coinciding with a decrease in activity consistent with quiet rest; sleep diary times if event markers were not present; rest interval set by automated algorithm if neither button presses nor diary data were available. Automated scoring parameters were set to: “medium” for wake threshold selection, 10 minutes for sleep onset and sleep offset interval detection, and a white light threshold of 1000.0. Unusual cases were presented for consensus review.

Actigraphy-derived sleep midpoint and total sleep time (TST) were calculated from sleep onset and offset as identified by the Philips Actiware sleep/wake scoring algorithm. Data presented here are for the major rest period only (excluding naps). Actigraphy TST is the total minutes of sleep occurring within the major rest period scored as “sleep” by the software. Actigraphy midpoint – a measure of sleep timing – is the midpoint of the sleep period [(start time – end time)/2]. The mean TST, mean sleep midpoint, standard deviation of TST, and standard deviation of sleep midpoint were calculated across all days of actigraphy for each participant. The number of days of actigraphy did not differ by group; t(47) = 0.77, p = 0.45; Bipolar M = 6.91, SD = 0.29, Control M = 6.96, SD = 0.19.

Social Stress Task

Participants completed a modified version of the Trier Social Stress Task (TSST; 42) between 4:30 pm and 8:00 pm. Task timing was restricted to control for diurnal patterns in stress physiology (43). The TSST is a laboratory-based social-evaluative threat task that reliably elicits subjective and physiological stress (44). The standard TSST includes a resting baseline, followed by preparation to give a speech (10 min), delivery of the speech (5 min), and a serial subtraction task (5 min) in front of a video recorder and a panel of two impassive judges who are introduced as experts in public speaking. The task was modified in the current study to include 6 min each for speech preparation, speech delivery, and mental arithmetic; the speech and serial subtraction tasks were performed in front of a large video camera on a tripod and two observers -- the research assistant and a judge who was described as an expert in non-verbal behavior; and participants were asked to make a speech as if defending themselves in court regarding shoplifting charges while being as persuasive as possible (45, 46). The social stress task (SST) was preceded by a 20 min pre-stress resting baseline period, and followed by a 60 min recovery period, both of which included viewing a nature video.

Heart Rate Variability Data Collection and Reduction

Electrocardiography was performed continuously during the pre-stress baseline, SST, and recovery periods at 1024 Hz using Grass amplifiers and Harmonie collection software (Natus Medical Incorporated, Pleasanton, CA). Data files were imported into MindWare HRV 3.0 (Gahanna, OH) and examined for artifact, as well as arrhythmic and ectopic beats. Abnormal beats were manually interpolated using the immediately preceding and subsequent beats. Segments were excluded if beats could not be reasonably estimated or had more than 8 arrhythmic beats within any 2-minute segment. Autoregressive spectral analyses of the interbeat interval were used to derive 2 min averages of heart rate (49 segments total; 10 segments during baseline, 9 segments during the SST, 30 samples during recovery). Heart rate variability within the high frequency range (0.15–0.40 Hz; HF-HRV) was calculated using absolute power from spectral analyses.

Analytic Approach

Descriptive analyses.

First, we evaluated the extent of missing data, the distribution of the data (normality, equality of variance, outliers), and bivariate correlations between sleep variables and initial HF-HRV during the resting baseline and stress. Participants with fewer than 5 days of actigraphy (n=3) or fewer than 50% usable 2-min HR segments (n=3; i.e., due to poor signal quality) were removed from analyses; as a result, analyses are based on data from 49 participants (22 bipolar adolescents, 27 healthy adolescents). Natural log transformation was applied to the HF-HRV data so it would approximate a normal distribution.

Hypothesis tests.

Two multilevel models were used to evaluate whether sleep characteristics measured during the week prior to laboratory stress were associated with HF-HRV at baseline and in response to social stress, and whether these associations differed between healthy adolescents and adolescents with bipolar disorder. Multilevel models were performed in MPlus-7 (47) using robust maximum likelihood (MLR) estimation. Based on mean plots and model fit statistics for a previous report (19), first-level models for the current manuscript estimated the intercept (initial value) and linear change in HF-HRV over time for the baseline period (min 2–20) and SST (min 22–38). Time was scaled to equal minutes divided by the sampling frequency (2 min) and centered to the initial value for each period of analysis (min 2 for baseline, min 22 for SST). Analyses for baseline included an average of 9.4 heart rate segments (based on 2 min averages) for each of the 49 participants (461 segments total), and analyses for stress included an average of 8.7 heart rate segments (427 segments total). This number of observations is well above minimum levels for unbiased effect size estimates, which are based on simulation studies with as few as 5 level-1 observations (heart rate segments in this study) and 30 level-2 observations (participants in this study) (48).

The individual intercepts and slopes derived from first-level models were then regressed on the following nine second-level predictors: the quadratic function of mean total sleep time (TST²), standard deviation of TST, mean sleep midpoint, standard deviation of sleep midpoint, group, TST² × group, standard deviation of TST × group, mean sleep midpoint × group, and standard deviation of sleep midpoint × group. The inclusion of multiple sleep variables in a single model for baseline, and a single model for stress, provided a more conservative test of our hypotheses than would be provided by independent tests with each sleep characteristic. We estimated the effects of quadratic total sleep time (TST²), rather than linear TST or linear and quadratic TST, after comparing fit statistics and the size of the TST and TST² effects in competing second-level models (Supplemental Digital Content, Text S1). The quadratic function of mean TST was calculated by subtracting the grand average of sleep duration (6.764 h) from each individual’s TST, and squaring the result. All sleep variables and group (0=control, 1=bipolar) were grand mean centered for analysis. Analyses were considered significant if the probability of obtaining the effect size under the null hypothesis (p) was ≤.05. When sleep characteristics moderated the effects of group on HF-HRV, subsequent multilevel models were run separately in each group to evaluate within-group effects of sleep characteristics. Higher order effects were required to be significant at each stage of post-hoc testing to provide a conservative test of the hypotheses.

Sensitivity analyses.

Sensitivity analyses were performed to evaluate the influence of demographic and health characteristics on the primary hypothesis tests. First, we performed sensitivity analyses to evaluate whether age, sex, or BMI moderated the effects of sleep characteristics and group on HF-HRV during rest and stress. Age, sex (0=male, 1=female), and BMI were added as three additional second level predictors to the 9-predictor multilevel models described above (12 second-level predictors total). All variables were grand mean centered. Missing BMI values for three participants were imputed using the multiple imputation function in MPlus, with five iterations.

We also performed exploratory analyses to evaluate whether medication load and depression severity moderated the effects of sleep characteristics on HF-HRV during social stress in participants with bipolar disorder. Medication load for the three most common medication classes – antipsychotics, mood stabilizers, and stimulants, each prescribed in at least 40% of cases – were included as additional second-level predictors of HF-HRV. A separate model examined depression severity in adolescents with bipolar disorder by including MFQ score as a second level predictor of HF-HRV. The effects of medication load and MFQ score were assessed in participants in bipolar disorder, rather than both groups, because medication load and MFQ score were both dependent on diagnostic status (i.e., none of the healthy controls were taking medications, and controls by design were not depressed and thus had markedly lower depressive symptoms than the bipolar participants).

Results

Descriptive Characteristics

Sample characteristics are presented in Table 1 and Supplemental Digital Content Table S1. The distributions of age, sex, race and BMI were balanced between the participant groups. Participants with bipolar disorder had higher depression symptom scores on the MFQ relative to healthy controls; t(22.64) = −3.14, p = .005 with correction for unequal variance. The bipolar group also had longer mean sleep durations than controls; t(32.35) = −4.10, p≤ .001 with correction for unequal variance. Mean sleep midpoint, and the standard deviations of sleep duration and sleep midpoint, did not differ by group. Frequency distributions for sleep characteristics in each group are presented in Supplemental Digital Content, Figure S1 and Figure S2. Bivariate correlations between sleep variables and initial HF-HRV during baseline and stress are presented in Supplemental Digital Content, Table S2 for both groups, and Table S3 for the bipolar group. The correlation coefficients for sleep variables with initial HF-HRV during baseline ranged from r=−0.09 (association with sleep midpoint SD) to r=−0.60 (association with mean TST²). The correlation coefficients for sleep variables with initial HF-HRV during stress ranged from r=−0.01 (association with sleep midpoint SD) to r=−0.67 (association with mean TST²).

Table 1.

Descriptive characteristics of the sample

Characteristic	Control (n=27)	Bipolar (n=22)	Χ2/t(df)	p
Sex, n	18 female	15 female	.013(1)	.910
Race, n	4 Asian, 6 Black, 17 White	3 Black, 19 White	4.65(2)	.098
Age, M(SD)	19.13(2.93)	18.53(2.74)	0.74(47)	.460
BMI, M(SD)	24.33(5.27)	27.32(6.79)	−1.68(44)	.100
MFQ, M(SD)	4.11(4.83)	15.86(16.62)	−3.50(46)	.001
Mean TST, M(SD)	6.22(0.72)	7.43(1.23)	−4.11(32.39)	<.001
Mean TST2, M(SD)	0.79(1.01)	1.88(2.61)	−1.84(26.09)	.078
SD TST, M(SD)	1.48(0.59)	1.50(0.50)	−0.15(47)	.879
Mean Midpoint, M(SD)	4.54(1.15)	4.56(1.49)	−0.07(47)	.945
SD Midpoint, M(SD)	1.08(0.43)	1.13(0.51)	−0.39(47)	.700

Note. One participant with bipolar disorder did not complete the MFQ (n=21) and three participants with bipolar disorder did not complete BMI measurements (n=19). Time units are in hours. Significant group differences appear in bold font.

Associations of Sleep Characteristics and Group with HF-HRV

Resting state.

Analyses of HF-HRV during the pre-stress resting baseline indicated that there was a significant association of the quadratic function of mean total sleep time with the initial HF-HRV and linear change in HF-HRV (Table 2, Figure 1). Longer and shorter sleep durations were associated with lower initial HF-HRV, and greater increases in HF-HRV across the pre-stress resting baseline, than moderate sleep durations. In addition, adolescents with bipolar disorder had lower initial HF-HRV than healthy controls, consistent with our previous analyses with this sample (19). Main effects and interactions for variability in sleep duration, mean sleep midpoint, and variability in sleep midpoint were not significant (ps > .05).

Table 2.

Associations of group and sleep characteristics with HF-HRV during pre-stress baseline and social stress.

Contrast	Z	b(SE)	p	Z	b(SE)	p
	Baseline			SST
Intercept
Intercept	45.99	6.09(0.13)	<.001	37.45	5.59(0.15)	<.001
Linear slope	0.43	0.00(0.01)	.667	4.29	0.07(0.02)	<.001
Group
Intercept	−2.58	−0.72(0.28)	.010	−2.37	−0.71(0.30)	.018
Slope	−0.77	−0.02(0.02)	.443	1.99	0.06(0.03)	.046
Mean TST2
Intercept	−5.37	−0.39(0.07)	<.001	−2.69	−0.27(0.10)	.007
Slope	2.50	0.02(0.01)	.013	−0.29	−0.00(0.01)	.774
SD TST
Intercept	−0.97	−0.32(0.34)	.335	0.74	0.33(0.45)	.461
Slope	0.67	0.02(0.02)	.506	−1.29	−0.05(0.04)	.199
Mean Midpoint
Intercept	0.71	0.08(0.12)	.480	1.29	0.14(0.11)	.197
Slope	−0.48	−0.00(0.01)	.630	−0.50	−0.01(0.01)	.616
SD Midpoint
Intercept	−0.82	−0.27(0.33)	.413	−1.08	−0.43(0.40)	.280
Slope	0.78	0.02(0.03)	.435	1.63	0.05(0.03)	.103
Group × Mean TST2
Intercept	−0.64	−0.09(0.14)	.524	−1.74	−0.32(0.19)	.083
Slope	−0.61	−0.01(0.02)	.540	2.24	0.04(0.02)	.025
Group × SD TST
Intercept	1.71	1.26(0.74)	.087	1.24	1.19(0.96)	.215
Slope	−0.05	−0.00(0.05)	.963	−2.78	−0.21(0.08)	.006
Group × Mean Midpoint
Intercept	0.84	0.21(0.25)	.403	0.48	0.11(0.22)	.633
Slope	−1.50	−0.02(0.01)	.135	−0.64	−0.02(0.02)	.523
Group × SD Midpoint
Intercept	−0.36	−0.25(0.71)	.721	−0.47	−0.40(0.85)	.637
Slope	0.29	0.02(0.06)	.769	1.44	0.09(0.07)	.151

Note. Intercepts represent starting values, and slopes represents the linear change in values over time. Significant z-scores appear in bold font. Group and sleep variables were grand mean centered

Figure 1.

Mean (SE) HF-HRV by group and mean total sleep time during pre-stress baseline (2–20 min) and social stress (22–38 min; grey background).

Social stress.

Analyses of HF-HRV during the SST indicated that the quadratic function of sleep duration, and variability in sleep duration, both moderated the effect of group on parasympathetic reactivity to stress; Table 2, Figures 1 and 2. There was also a significant main effect of the quadratic function of sleep duration on initial HF-HRV, where longer and shorter sleep durations were associated with lower HF-HRV than moderate sleep durations. Furthermore, consistent with our previous analyses, adolescents with bipolar disorder had lower initial HF-HRV and greater change in HF-HRV over time (more stress reactivity) than healthy adolescents (19). Main effects and interactions for mean sleep midpoint and variability in sleep midpoint were not significant (ps > .05).

Figure 2.

Mean (SE) HF-HRV by group and standard deviation of total sleep time during pre-stress baseline (2–20 min) and social stress (22–38 min; grey background).

Subsequent multilevel models in each group indicated that longer and shorter sleep durations were associated with lower initial HF-HRV in adolescents with bipolar disorder, while the quadratic function of TST was not significantly associated with initial HF-HRV in healthy adolescents; Table 3; Figure 3. These models also indicated that longer and shorter sleep durations predicted a marginally significant increase in slope in adolescents with bipolar disorder, and a nonsignificant decrease in slope in healthy adolescents. In addition, greater variability in TST was associated with a significant decrease in slope in adolescents with bipolar disorder, while variability in TST was associated with a nonsignificant increase in slope in healthy adolescents. Mean sleep midpoint and standard deviation in sleep midpoint were excluded from within-group analyses because their main effects and interactions were not significant in the full model.

Table 3.

Associations of mean total sleep time and standard deviation of total sleep time with HF-HRV during social stress for each group.

Contrast	Z	b(SE)	p	Z	b(SE)	p
	Control			Bipolar
Intercept
Intercept	33.66	5.98(0.18)	.000	20.99	4.92(0.23)	<.001
Slope	2.56	0.05 (0.02)	.011	5.64	0.12(0.02)	<.001
Mean TST2
Intercept	−0.94	−0.16(0.17)	.346	−5.44	−0.42(0.08)	<.001
Slope	−1.02	−0.02(0.02)	.307	1.84	0.02(0.01)	.065
SD TST
Intercept	−0.68	−0.23(0.33)	.498	1.22	0.57(0.47)	.223
Slope	1.36	0.05(0.04)	.174	−2.18	−0.09(0.04)	.029

Note. Intercepts represent starting values, and slopes represents the change in values over time. Sleep variables were grand mean centered. Significant z-scores appear in bold font.

Figure 3.

Scatter plot of mean total sleep time by initial HF-HRV power during the SST. Lines represent SPSS-generated quadratic functions for each group.

Sensitivity Analyses

The results of sensitivity analyses were broadly consistent with the results of the primary analyses. Analyses that adjusted for age, sex, and BMI indicated that these variables did not account for a significant portion of the variance in initial HF-HRV or change in HF-HRV during the resting baseline or stress; Supplemental Digital Content, Table S4. Consistent with primary analyses of resting parasympathetic tone, there was a significant association of the quadratic function of mean total sleep time with the initial HF-HRV and linear change in HF-HRV, and group status was significantly related to initial HF-HRV. Consistent with primary analyses of HF-HRV during the SST, the quadratic function of mean TST, and the SD of TST, both moderated the association of group status with the linear change in HF-HRV, though the association of mean TST was reduced to marginal significance.

Analyses within the bipolar group alone indicated that longer and shorter sleep durations were associated with lower initial HF-HRV during the SST after adjusting for the load of antipsychotics, mood stabilizers, and stimulants; Supplemental Digital Content, Table S5. Analyses that adjusted for depressive symptom severity within the bipolar group indicated that the effects of quadratic TST on initial HF-HRV and HF-HRV reactivity were reduced, but still significant, in participants with higher MFQ scores (higher intercepts, lower slopes); Supplemental Digital Content, Table S6. In addition, consistent with primary analyses, greater variability in TST was associated with a significant decrease in slope that was especially pronounced in participants with higher MFQ scores (significant interaction of MFQ × TST SD on change in HF-HRV during stress).

Discussion

The aim of the present research was to evaluate whether behaviorally-assessed sleep characteristics predict resting and stress-related parasympathetic activity in adolescents, and whether they moderate associations between bipolar disorder and parasympathetic response. Consistent with our hypotheses, we found that mean sleep duration during the previous week was associated with resting parasympathetic activity in healthy adolescents and adolescents with bipolar disorder, and moderated associations between bipolar disorder and parasympathetic response to stress. Specifically, shorter and longer sleep durations were associated with lower vagal tone at the start of the resting baseline and social stress across both groups (although post-hoc analyses indicated that the quadratic function of mean sleep duration predicted HF-HRV in youth with bipolar disorder but not healthy controls). We also found that greater variability in total sleep time during the previous week was associated with a slower rate of change (slope) in HF-HRV during the course of the stressor in adolescents with bipolar disorder, indicating that greater sleep variability is associated with a more persistent stress response.

The association between immoderate sleep duration and lower HF-HRV is consistent with evidence that shorter sleep is associated with low resting HF-HRV in healthy young adults (20, 22), and short and long sleep durations predict negative health outcomes (29). Short and long sleep durations are also associated with higher depressive symptoms and poorer quality of life and functioning in patients with bipolar illness (49). Notably, adolescents with bipolar disorder obtained more sleep on average than the control group (respectively, 7hr 26min versus 6hr 13min). The association between the quadratic function of sleep duration and initial HF-HRV in the full sample, with longer and shorter sleep durations associated with lower initial parasympathetic tone during the resting baseline and stress, may be driven by the relatively larger number of healthy adolescents who were in the lower tertile for mean sleep duration (n=14/27), combined with the larger number of bipolar adolescents who were in the upper tertile for mean sleep duration (n=13/22). This interpretation is supported by post-hoc analyses in each group, in which the quadratic function of mean sleep duration predicted HF-HRV in youth with bipolar disorder but not healthy controls. Notably, the average sleep duration in our sample is consistent with other samples of adolescents and young adults (50, 51), and is considered insufficient to adequately meet sleep need (52). In addition, only one bipolar adolescent had an average sleep duration of less than 5 hours. This suggests that associations between sleep duration and parasympathetic tone can be observed in adolescents with bipolar disorder even when their typical sleep duration is within the “normal” range. However, shorter sleep durations (< 5 hours) may have a different relationship with parasympathetic function in bipolar illness and could be evaluated in future research.

Depression symptom severity did not account for the relationship between sleep duration and parasympathetic tone in bipolar adolescents. This suggests that the relationship between sleep duration and HF-HRV may be more trait-based than state-based. However, this interpretation is tentative because we did not examine whether mania symptom severity moderates the relationship between sleep duration and HF-HRV, as symptoms of mania were less severe and less variable than symptoms of depression at the time of the study.

The associations of sleep duration and variability with parasympathetic function run parallel to the results of a related study that examined associations between sleep characteristics and brain response during a stressful cognitive control task (53). In this fMRI study, which included a sub-sample of the participants reported here, shorter and longer sleep durations predicted rostroventral anterior cingulate cortex (ACC) function in adolescents with bipolar disorder, shorter sleep durations predicted greater rostroventral ACC response in healthy controls, and greater variability in sleep duration predicted less activation in the dorsal ACC in bipolar adolescents, but not controls. Rostroventral and dorsal ACC are both involved in regulation of autonomic function (14, 15). Collectively, these results demonstrate the impact of immoderate sleep duration, and high sleep variability, across stress response systems (brain and heart).

Mean sleep duration was the only sleep characteristic that differed significantly between the bipolar and healthy samples in this study, potentially reflecting longer sleep durations in more depressed bipolar youth. Studies that evaluate more acutely symptomatic adolescents with bipolar disorder (particularly during mania), or high-risk samples of adolescents (e.g., those with a parental history of bipolar illness), may be better positioned to evaluate whether sleep timing, and variability in sleep duration and timing, moderate the relationship between bipolar disorder and parasympathetic activity. Replication of these results with larger and unmedicated samples is also warranted (though the latter may be unrealistic in a population diagnosed with bipolar disorder), and would help address the disproportionate data retention of the current study (5 participants with fewer than 5 days of actigraphy or usable HF-HRV data in the bipolar group, versus 1 participant in the control group). Prospective longitudinal studies are necessary to determine whether sleep duration or other sleep characteristics prospectively predict the onset of mood episodes, bipolar onset, and episode recurrence.

Sleep-related variation in HF-HRV in adolescents with bipolar disorder may be due to one or more unexamined variables. Sleep characteristics such as sleep onset latency and sleep efficiency, while not the focus of the present study, could be evaluated in future work. In addition, our multilevel models included four different sleep indicators and their interaction with group as a conservative test of the study hypotheses, but the partial distribution of variance between predictors also increases the potential for false negatives. Future studies could begin to address these limitations using experimental study designs that systematically restrict or extend sleep in bipolar youth. However, experimental sleep manipulation in adolescents with bipolar disorder is difficult because sleep disruption can precipitate mood episodes (7).

Changes in sleep, circadian timing, and social stress are typical during adolescent development and may have synergistic effects on risk for mood episodes in bipolar disorder. These risk factors may also have their greatest impact when they occur in the context of continued development in affective neural networks, contributing to long-term disruption of autonomic response systems that might mediate stress-related mood disruption (54, 55). HF-HRV reflects activity in the central autonomic network (56, 57), an affective system that modulates autonomic stress response (58). Disruptions in affect-related brain function are a putative mechanism of bipolar disorder (59–63), have been associated with both sleep duration and sleep variability in youth with bipolar disorder (53), and are predicted during adolescence generally by insufficient sleep (64–66), delayed and inconsistent sleep timing (67, 68), and social stressors (69–71). An exciting potential direction for future work is to evaluate the degree to which HF-HRV is correlated with the function of affective neural networks, and whether improvement in sleep can buffer neuroaffective and autonomic systems from the effects of psychosocial stress.

Acronyms:

BMI

body mass index

BP

bipolar

HC

healthy control

HF-HRV

high frequency heart rate variability

MFQ

Mood and Feelings Questionnaire

SST

social stress task

TSST

Trier Social Stress Task

TST

total sleep time

TST²

quadratic function of total sleep time