Report from a Randomized Control Trial: Improved Alignment Between Circadian Biology and Sleep-Wake Behavior as a Mechanism of Depression Symptom Improvement in Evening-Type Adolescents with Depressive Symptoms

Abstract

Study Objectives:

An evening circadian preference is common among adolescents. It is characterized by a behavioral predilection for later sleep and wake timing and is associated with higher rates of Major Depressive Disorder (MDD). The present study aims to a) test the effectiveness of a cognitive behavioral sleep intervention (Transdiagnostic Sleep and Circadian Intervention; TranS-C) in a sample of adolescents with an evening circadian preference and clinically significant depressive symptoms and b) evaluate improved alignment between circadian biology and sleep-wake behavior as a potential mechanism in the relationship between sleep and depression symptom improvement.

Methods:

Adolescents with an evening circadian preference and clinically significant depressive symptoms were randomized to receive TranS-C (n=24) or a psychoeducation condition (PE; n=18). Alignment between circadian biology and sleep-wake behavior was measured using an objective biological measurement. Measures of sleep and circadian rhythm were taken at pre and post-treatment, and depression symptoms were measured at pre and post-treatment, and 6 and 12-month follow-up.

Results:

Mixed effects modeling revealed that when compared to an active control condition, TranS-C resulted in significant reduction in MDD severity at 12-months follow-up. A MacArthur mediation analysis conducted to explore alignment between circadian biology and sleep-wake behavior as a mediator of depression severity reduction through 12-month follow-up revealed a significant interaction between change in alignment between circadian biology and sleep-wake behavior and treatment arm, indicating that improved alignment between circadian biology and sleep-wake behavior at post treatment was associated with improvements in depression outcomes at 12-month follow up in the treatment condition.

Conclusions:

These results provide novel evidence for improved alignment between circadian biology and sleep-wake behavior as a specific mechanism of depression improvement, provide key clues into the complex relationship between sleep and depression and have significant clinical implications for adolescents with depression.

Introduction

Adolescence is characterized by a significant increase in the incidence of Major Depressive Disorder (MDD).¹ Epidemiological data indicate that the onset of MDD frequently occurs during adolescence.¹ In 2015, an estimated 3 million adolescents in the United States had at least one major depressive episode in the past year,² and approximately half of these adolescents experienced recurrent depressive episodes.³ Adolescent-onset MDD is associated with greater psychosocial impairment, physical health problems, psychiatric comorbidity, school failure, and suicidality.⁴ Even with the best evidence-based medication and psychosocial treatments, around 40% of depressed adolescents do not respond to treatment.⁵ Hence, there is an urgent need to identify novel, modifiable targets to improve MDD.

Evening circadian preference (ECP), defined as the behavioral predilection for later sleep and wake timing and is associated with having later biological circadian timing, is associated with increased depression severity, recurrent MDD episodes, increased suicidality and poor antidepressant response.^6–8 The human circadian rhythm is the autonomous, approximately 24-hour cycle for most human biological processes which has been implicated for decades in MDD.⁹ Numerous physiological and behavioral processes demonstrate circadian rhythms, including body temperature, hormones such as melatonin, concentration and mood.¹⁰ Circadian preference broadly is a measure of patterns of individual differences in preference of circadian sleep/wake behavior.¹¹ ECP specifically is the focus of the present manuscript. There is strong cross-sectional and longitudinal data that ECP is robustly associated with MDD diagnosis and increased depression severity and suicidal ideation.^7,12 Prevalence of ECP within populations with MDD are high, and higher levels of depressive symptoms are associated with an ECP in the general population.¹² Individuals with ECP are prone to long-term depression non-remission after treatment,⁸ suggesting vulnerability in this group which may not be adequately addressed by current treatments. During antidepressant treatment, ECP individuals also report higher levels of depressive symptoms and suicidality compared to non-ECP’s.⁶ Collectively, it appears that patients with ECP and MDD are an important, yet underserved, vulnerable sub-group to target for a personalized depression treatment intervention.

Cognitive-Behavioral sleep interventions improve MDD outcomes among patients with ECP and MDD.⁶ Consistent with the above-cited findings that ECP augurs poor treatment response, results from our trial of cognitive-behavioral sleep treatment (Cognitive Behavioral Therapy for Insomnia; CBT-I) augmentation of antidepressant management in adults with MDD and insomnia indicate that patients with ECP had diminished antidepressant response.⁴² Among ECP patients, we found that randomization to CBT-I (plus antidepressants) led to significant improvement in depression compared to the control intervention (antidepressant medication plus sleep education and desensitization).⁶ Research with adolescents lags behind adult research, however meta-analyses and systematic reviews have concluded that cognitive-behavioral sleep interventions significantly improve sleep in adolescents with and without MDD and that improvements are maintained for 6 to 12 months.¹³ Collectively, these findings suggest that 1) cognitive-behavioral sleep interventions are effective at improving sleep in adolescents with MDD, 2) patients with ECP and MDD are at increased risk for poor depression response, and 3) offering a cognitive-behavioral sleep intervention to patients with ECP and MDD may improve depression outcomes. However, a lack of clear understanding about why cognitive-behavioral sleep interventions may improve depression outcomes and the underlying pathophysiological mechanisms of depression improvement among youth with ECP is a major barrier to optimizing treatments.

Misalignment between biological circadian timing and sleep-wake behavior is a potential mechanism of depression among youth with ECP and MDD, which can be targeted for therapy¹⁴. The circadian system is regulated by the endogenous pacemaker in the suprachiasmatic nucleus of the hypothalamus.¹⁵ The suprachiasmatic nucleus influences many different aspects of sleep, but most importantly it influences the time of day when sleep occurs.¹⁶ One of the ways it does so is by controlling the timing of melatonin secretion via a multisynaptic pathway.¹⁷ As shown in Figure 1, the human body reliably produces its own (endogenous) melatonin starting approximately two hours before bedtime, provided the lighting is dim. Accordingly, the timing of melatonin onset under dim light conditions (dim light melatonin onset or DLMO) is the ‘gold standard’ marker of biological circadian timing.¹⁸ The production of melatonin at DLMO helps prepare the body for sleep.⁸³ Melatonin production peaks during the biological circadian night and decreases as biological circadian morning approaches.¹⁹ It has been repeatedly demonstrated that sleeping during this period is less disturbed and more restorative than sleeping at other times of the 24-hour cycle.^20,21 The relationship between DLMO, an indicator of biological circadian timing, and actual waketime, which is dependent on behavior, is referred to as the phase angle difference between DLMO and waketime (PAD_{DLMO WAKETIME} = time of DLMO - waketime; see Figure 1). Ideally, an individual would go to sleep around 2 hours after DLMO and would wake up at an hour that is coincident with their biological circadian wake time (e.g. they would wake up when their melatonin levels are starting to decline).¹⁹ In contrast, an individual who is going to bed or waking up at an hour that is out of alignment with their biological circadian timing would be prone to disturbed or insufficient sleep.²² For example, an individual who is waking up at an hour that is out of alignment with their biological circadian timing, that is, at a time too close to their DLMO, when their melatonin levels are still high, would have difficulty waking up in the morning. In this scenario, PAD_{DLMO WAKETIME} would be short (Figure 1b).

Figure 1.

Illustration of Healthy (a) and Unhealthy (b) Circadian Alignment in Relation to Endogenous Melatonin Profiles

Note. PAD is a difference score. For example, PAD_{DLMO WAKETIME} = time of DLMO- waketime. Figure 1B displays a PAD_{DLMO WAKETIME} that is shortened by a delayed DLMO and an advanced waketime. Note that total sleep time is shortened, and bedtime is advanced in this example.

Adolescents with ECP are particularly vulnerable to having misalignment between circadian biology and sleep/wake behavior.^23,24 An estimated 40% of adolescents endorse ECP.¹¹ DLMO findings indicate that ECP may be, at least in part, biologically driven and is associated with pubertal onset.¹¹ DLMO times therefore tend to be later in youth with ECP. At the same time, early school start schedules require ECP adolescents to wake-up at times that are often earlier than their biological circadian wake time, thus shortening the phase angle difference between their DLMO and waketime (PAD_{DLMO WAKETIME}; see Figure 1).²⁵ Indeed, a study examining PAD_{DLMO WAKETIME} before and after the initiation of remote learning due to COVID-19 among adolescents in Australia suggests that the PAD_{DLMO WAKETIME} was significantly shorter during in-person learning, compared to remote learning with large effect sizes (d=.9). While ECP was not considered in this study, others^12,14 find that ECP youth tend to have shorter PAD_{DLMO WAKETIME} during the school year than non-ECP youth.^12,14 Hence, adolescents are particularly likely to have ECP;¹¹ and due to a tendency towards late DLMO and early waketimes (due to school start), those with ECP are particularly likely to have short PAD_{DLMO WAKETIME}.²³ Both of these factors promote the general tendency for adolescents to have short PAD_{DLMO WAKETIME} and experience the well-established sleep consequences of sleeping out of alignment with the optimal biological sleep period. These consequences include daytime sleepiness and poor sleep quality which are highly prevalent in this age group.²⁶

ECP is modifiable with Transdiagnostic Sleep and Circadian Intervention (TranS-C), a cognitive-behavioral sleep intervention that improves the relationship between biological circadian timing and sleep/wake behavior among adolescents.²⁷ TranS-C is a comprehensive modular sleep health intervention; it consists of chronotherapy to address circadian misalignment common among youth with ECP.^27,28 It also includes many elements of CBT-I that address insomnia that frequently emerges in those with circadian misalignment,²⁹ as well as elements of Interpersonal and Social Rhythms Therapy to help stabilize social rhythms.³⁰ In 176 adolescents with ECP, our group found that TranS-C led to significant improvements in selected sleep and circadian outcomes at post-treatment compared to a psychoeducation control condition. These changes included greater reduction in self-reported ECP, earlier biological circadian timing (DLMO), less weeknight–weekend discrepancy in total sleep time, less daytime sleepiness, and better self-reported sleep quality compared to a control condition.³¹

The present study is a secondary data analysis of the aforementioned TranS-C study.^31,32 Taking a precision medicine approach, our goal was to better understand whether TranS-C was helpful for a specific sub-set of our sample. Moreover, the focus of the present study is on investigating a novel potential mechanism, misalignment between biological circadian timing and sleep-wake behavior (measured using PAD_{DLMO WAKETIME}, PAD_{DLMO BEDTIME}, and PAD_{DLMO MIDSLEEP}) as a driver of MDD severity in youth with ECP. This manuscript fills an important gap in scientific knowledge regarding the role of misalignment between biological circadian timing and sleep-wake behavior and MDD among youth. The first aim is to compare MDD severity outcomes up to one year following participation in TranS-C or an active control. We hypothesize that TranS-C, when compared to the active control condition, will result in significant improvements in MDD severity among adolescents with ECP and clinically significant MDD severity. The second aim is to explore PAD_{DLMO WAKETIME} and other ways of measuring misalignment between biological circadian timing and sleep-wake behavior (PAD_{DLMO BEDTIME}, and PAD_{DLMO MIDSLEEP} ) as mediators of MDD severity improvement at 12-month follow-up.

Methods and Materials

Participants

The data for the present study were collected as part of a larger clinical trial protocol (NCT01828320).³¹ A total of 398 participants were assessed for eligibility and 220 (55.6%) were excluded for not meeting inclusion criteria (n=154) or refusing to participate (n=66). A total of 102 female and 74 male participants (N=176) were randomly assigned to treatment. ³¹ The present study is a secondary analysis of the parent trial of 18 male and 24 female participants (N=42). Adolescents were eligible if they (a) were between 10 and 18 years old, living with a parent or guardian, and attending a class/job by 9am at least three days per week, (b) were fluent in English, (c) were able and willing to give informed assent, (d) reported an ECP as demonstrated by scoring in the lowest quartile on the Children’s Morningness-Eveningness Preference Scale (CMEP; 27 or lower),¹¹ had a 7-day sleep diary showing a sleep onset time of 10:40pm or later for 10–13 year olds, 11pm or later for 14–16 year olds, and 11:20pm or later for 17–18 year olds at least three nights per week, and this pattern had to be present for at least three months, and (e) fell in the “at-risk” range on measures in at least one of five health domains (behavioral, cognitive, emotional, social, physical) described in greater detail elsewhere.³¹ Adolescents were ineligible to participate, if (a) they could not communicate in English, (b) they had an active, progressive physical illness or neurological degenerative disease directly related to the onset and course of the sleep disturbance, (c) there was evidence from clinical diagnosis or report of sleep apnea, restless legs or periodic limb movements during sleep, (d) they had an intellectual disability, autism spectrum disorder, or other significantly impairing pervasive developmental disorder, (e) there was evidence from clinical diagnosis or report by youth or parent of Bipolar Disorder or Schizophrenia, or (e) they had a history of substance dependence in the past six months, and (f) current suicide risk to preclude treatment on an outpatient basis. Individuals ceased taking medications that were specifically prescribed for sleep (e.g., hypnotics) four weeks prior to the assessment (two weeks for melatonin) or were excluded.

The current study focuses on a subset of participants who were included if they met clinical criteria for clinically significant depressive symptoms at baseline using the Children’s Depression Rating Scale (CDRS; 40 or greater). ³³

Participants were recruited for the current study via clinicians and advertisements.

Measures of Demographic Characteristics.

Demographic characteristics assessed included parent reports of the adolescents’ age, biological sex, race/ethnicity and household income.

Medications.

A Medication Tracking Log was completed by the teen (corroborated by parents as needed) weekly during the acute treatment phase and at the 6 and 12 month follow-up sessions thereafter. Medication dosage, discontinuation, and any new medications added were tracked.

Depression Symptom Measure

Children’s Depression Rating Scale (CDRS).³³

The CDRS is an interviewer rating scale, which is widely used to assess the severity of depression and response to treatment. CDRS scores were collected at baseline, post-treatment, 6 and 12 month follow-up assessments. Teen CDRS interview scores were utilized in the present study and sleep items were removed.

Sleep and Circadian Measures

Circadian Preference Measure: Children’s Morningness-Eveningness Preferences Scale (CMEP).

The CMEP is a reliable and valid 10-item measure self-report measure, completed nby the teen, of circadian tendency in youth, to screen for ECP.¹¹ The total score ranges from 10 (Extreme evening preference) to 42 (Extreme morning preference). ¹¹ All participants in the study were required to have a score within the lowest quartile (27 or lower).

Sleep/Wake Behavior Measure: Sleep Diary³⁴.

The daily sleep diary is a valid and sensitive measure of sleep timing and sleep quality. Trained research assistants called the adolescents to collect their sleep diary each morning at an agreed-upon time, minimizing recall bias. Three variables of interest were extracted from the sleep diary: bedtime, midsleep and waketime. An a priori decision was made to focus on weeknight bedtime (Sunday to Thursday) and waketime (Monday to Friday) because these measures best address the sleep problems observed during adolescence. ³⁵

Sleep/Wake Behavior Measure: Actigraphy.³⁶

(Actiwatch^® Spectrum; Philips Respironics). Participants were asked to wear an actigraph for one week at each assessment (baseline, post-treatment and 6- and 12-month follow-ups). Actigraphy continuously measures movement (sampled in 60 second epochs), the data from which can be used to impute sleep/wake status using standard algorithms.³⁷ Sleep imputed via actigraphy is strongly correlated with sleep assessed via polysomnography for many sleep variables and is validated in adolescents.³⁶ Using a combination of the sleep diary and actigraphy, bed and wake times were set and from these, actigraphy-determined total sleep duration and sleep efficiency (duration/time in bed) were calculated.

Circadian Timing Measure: Dim Light Melatonin Onset (DLMO).

DLMO is the gold standard index of the endogenous circadian phase.³⁸ It is assessed with the serial saliva sampling method one night before and after treatment in an overnight stay in the Psychology Department at UC Berkeley based on an established protocol.³⁹ Thirteen saliva samples were collected for each participant before and after treatment, beginning 5.5 hours before average bedtime (computed from 7 nights of sleep diary) and ending 30 minutes after average bedtime. Saliva (~1 ml) samples were collected in 30-minute intervals in dim light (<50 lux) using untreated Salivettes (Sardtedt; Nümbrecht, Germany) and assayed for melatonin (SolidPhase, Inc.; Portland, Maine) using radioimmunoassay test kits (APLCO Diagnostics, Windham, NH). Assay sensitivities were 0.3 pg/ml and the minimum detectable dose was 0.05 pg/ml. Mean intra- and inter-assay coefficients of variation were 7.9% and 9.4%, respectively. DLMO was defined as the interpolated time at which melatonin exceeded 3.0 pg/ml. The selection of this threshold was based upon prior experience with melatonin as a marker of circadian phase and the visual inspection of each participant’s DLMO record.⁵⁵ Saliva samples were collected from all randomized participants. All participants provided saliva samples for melatonin analysis. Further detail is provided elsewhere.³¹

Misalignment between biological circadian timing and sleep-wake behavior was measured using phase angle difference (PAD) between DLMO and a recurring external cyclic event ⁴⁰ 3 different PAD’s were calculated; the difference (in hours and minutes) between DLMO and waketime (PAD_{DLMO WAKETIME}), DLMO and bedtime (PAD_{DLMO BEDTIME}), and DLMO and midsleep (PAD_{DLMO MIDSLEEP}; waketime, bedtime, and midsleep were all taken from sleep diary data).⁴⁰

Design

All procedures were approved by the University of California, Berkeley, Committee for the Protection of Human Subjects. All participants provided informed consent or assent. Participants were invited for an in-person assessment. If youth met criteria after the in-person assessment, they were enrolled in the study and scheduled for an overnight assessment. Seven to 10 days of sleep diary information was collected for each participant prior an overnight laboratory assessment. To determine DLMO collection times and wake-up times, an additional 7 nights of sleep diary were collected immediately prior to the overnight stay in the laboratory. The DLMO Protocol was conducted as part of the overnight assessment. Post-treatment, the procedures for the overnight assessment to collect DLMO were repeated, along with 7 days of sleep diary for the week prior to the overnight assessment.

Based in a university clinic, from March 2013 to March 2016, youth were randomly assigned, stratified by sex and age (10–14 years, 15–18 years), in a 1:1 parallel group design, to receive either TranS-C or PE. Sibling pairs (n = 3) were randomized to the same condition.

Assessors were blinded to treatment allocation. Assessments were conducted at baseline and at the end of treatment. Additional measures were collected at 6 and 12 month follow-up assessments as described elsewhere.³² Follow-up data at 6 and 12-months was missing for 8 participants. Participants were paid for completing assessments but not for participating in treatment.

Consolidated Standards Of Reporting Trials (CONSORT) Diagram, illustrating the flow of participants through the study is available elsewhere and in Appendix S1.³¹

Treatments

Therapists were graduate students in a doctoral program who had already completed their Master’s degree. Treatment involved 6 individual, weekly, 50-minute sessions delivered during the school year to minimize the impact of summer schedule variability.⁴¹

Transdiagnostic Sleep and Circadian Intervention for Youth (TranS-C).

Transdiagnostic Sleep and Circadian Intervention for Youth (TranS-C)⁴² is grounded in sleep and circadian basic science and combines elements from four evidence-based interventions: Cognitive Behavior Therapy for Insomnia,³⁵ Interpersonal and Social Rhythm Therapy,³⁰ Chronotherapy,⁴³ and Motivational Interviewing.⁴⁴ The goal is to reverse maintaining psychosocial, behavioral and cognitive processes impairing sleep health. The focus of treatment is on coaching youth through skills which are meant to simultaneously advance circadian rhythm and bedtime, delay waketime, improve sleep efficiency, extend total sleep time, and improve sleep quality. More details about TranS-C are available elsewhere and in Supporting Information.³¹

Psychoeducation (PE).

PE is an active comparison treatment associated with sleep improvement.⁴⁵ Sessions focus on the interrelationship among sleep, stress, diet, health, exercise, and mood (for more information see Supporting Information). There is no explicit focus on sleep except briefly mentioning sleep as a common contributor to each topic. Participants were also given the choice of sampling meditation, yoga, and/or outdoors appreciation. Within an Intrinsic Motivation framework,⁴⁶ therapists were trained to develop an environment in which the patient was self-directed regarding when materials were covered in order to promote youth autonomy. The emphasis was on providing information but not on specifically facilitating behavior change.

Analysis

All analyses were conducted in SPSS 26 (IBM Corporation, Armonk, NY, 2013). Statistical significance was evaluated using a 2-sided design with alpha set at 0.05.

Baseline characteristics were compared using t tests and χ². Analyses were carried out on the intent-to-treat sample.

Aim 1 (Model 1) was assessed using a mixed-effects linear model with autoregressive error structure to examine the differences in the rate of change (slope) in CDRS scores comparing the two treatment conditions. The model included a random intercept and slope for time, and fixed effects for treatment, time and their interaction.

Aim 2 (Model 2) was assessed using a MacArthur mediation model⁴⁷ to examine the possibility that misalignment between biological circadian timing and sleep-wake behavior (PAD_{DLMO WAKETIME,} PAD_{DLMO BEDTIME}, and PAD_{DLMO MIDSLEEP} ) mediated change in remission from MDD; that is, the part of the mechanism by which TranS-C affects depression lies in its ability to lengthen PAD_{DLMO WAKETIME}. The MacArthur Mediation model was designed specifically to clarify the distinction between moderation and mediation for clinical research (randomized clinical trials).⁴⁷ In the MacArthur Mediation Model, a moderator is defined as a variable that precedes the predictor variable, in this case treatment.⁴⁷ The MacArthur approach includes the interaction between treatment and mediator in the model and establishes mediation by the demonstrated presence of either a main effect of mediator or an interaction between time, treatment and mediator.⁴⁷ This analysis had 2 steps, progressing to the second step only if the model in the first step was statistically significant. In the first step, a regression analysis examined the relative differences between treatments on the rate of change in the PAD_{DLMO WAKETIME} from baseline to post-treatment. Change in the PAD_{DLMO WAKETIME} from baseline to post-treatment was estimated, based on all available data for each individual participant, as the slope of the regression line of the available PAD_{DLMO WAKETIME} scores from baseline to post-treatment. In the second step, a mixed effects linear model with autoregressive error structure (model 2) was tested to determine if the change in PAD_{DLMO WAKETIME} from baseline to post-treatment mediated depression symptom reduction from pre-treatment to 12 month follow-up. The model included treatment, change in PAD_{DLMO WAKETIME} from baseline to post-treatment (individual slopes), time and their interaction. The same models were run for the other measures of misalignment between biological circadian timing and sleep-wake behavior (PAD_{DLMO BEDTIME}, and PAD_{DLMO MIDSLEEP}).

To rule out the possibility that the mediation effect of PAD_{DLMO WAKETIME} could be better accounted for by weekday total sleep time and average weekly sleep efficiency (an index of insomnia severity), we conducted specificity and sensitivity analyses.

Results

Of the 176 participants in the larger study, 42 participants were determined to be depressed (CDRS>= 40); 18 were allocated to PE and 24 were allocated to TranS-C. This sub-sample is the focus of the present study. Demographic and clinical characteristics are presented in Table 1. There were no significant differences in demographic or clinical characteristics between treatment groups with elevated depressive symptoms, therefore no demographic or clinical characteristics were included as covariates in the models.

Table 1.

Demographic and Baseline Clinical Characteristics.

		Treatment		Analysis
Measure	Total (N=42)	TranS-C (n=24)	PE (n=18)	Test Statistic	P
Age, mean (SD), y	15.10 (1.62)	15.38 (1.91)	15.17 (1.04)	t=.42	.68
Sex, n (%)				χ2=1.17	.28
Male	24 (57.14)	12 (50)	12 (66.67)
Female	18 (42.86)	12 (50)	6 (33.33)
Race, n (%)				χ2= 4.19	.38
Caucasian	18 (58.07)	8 (53.33)	10 (62.5)
Black	2 (6.45)	0 (0)	2 (12.5)
Asian	4 (12.90)	3 (20)	1 (6.25)
Native Hawaiian or Other Pacific Islander	1 (3.23)	1 (6.67)	0 (0)
Mixed Race	6 (19.36)	3 (20)	3 (18.75)
Ethnicity, n (%)				χ2= .005	.95
Hispanic or Latino	4 (12.90)	2 (13.33)	2 (12.5)
Not Hispanic or Latino	27 (87.10)	13 (86.67)	14 (87.5)
Family Annual Income ($), n (%)				χ2=7.98	.24
≤ 20,000	4 (12.90)	1 (6.67)	3 (18.75)
20,001–50,000	3 (9.68)	1 (6.67)	2 (12.5)
50,001–100,000	6 (19.35)	4 (26.67)	2 (12.5)
>100,000	18 (58.07)	9 (60)	9 (56.25)
Antidepressant Use Baseline, n (%)	11 (26.19)	7 (29.17)	4 (22.22)	χ2=.26	.61
Current Grade, mean (SD), y	9.71 (1.53)	9.9 (1.73)	9.83 (1.20)	t=.31	.76
Sleep variables, mean (SD)
Bedtime, hh:mm	23:58 (1.03)	23:56 (1.01)	24:00 (.99)	t=−.169	.87
Waketime, hh:mm	7:35 (1.01)	7:36 (1.25)	7:59 (1.05)	t=−1.03	.31
TST, decimal hours	7.62 (1.01)	7.66 (1.24)	7.98 (1.01)	t=−.90	.38
DLMO, hh:mm	21:28 (1.27)	21:45 (1.47)	21:24 (1.16)	t=.78	.44
PADDLMO Waketime, decimal hours	10.35 (1.17)	10.16 (1.14)	10.54 (1.20)	t=−.91	.37
PADDLMO Bedtime, decimal hours	2.55 (.85)	2.49 (.95)	2.61 (.77)	t=−.54	.60
PADDLMO Midsleep, decimal hours	6.45 (.83)	6.32 (.80)	6.58 (.87)	t=−.29	.77
CDRS, mean (SD)	45.97 (5.90)	46.63 (7.02)	47.83 (7.35)	t=−.54	.59

^aPercent values are relative to non-missing values.

Abbreviations: SD = standard deviation, hh:mm= time displayed in hours and minutes (military time), TST = total sleep time, DLMO=Dim Light Melatonin Onset, PAD_{DLMO Waketime} = Phase Angle Difference between DLMO and waketime, PAD_{DLMO Bedtime}= Phase Angle Difference between DLMO and bedtime, PAD_{DLMO Midsleep}= Phase Angle Difference between DLMO and midsleep, CDRS= Children’s Depression Rating Scale.

Among the sub-sample included in the present study, medication use at study entry was as follows: antidepressants, n = 6 (19%); stimulants, n = 4 (13%); and benzodiazepines, n = 1 (1%). The doses of 93% of antidepressants, 100% of stimulants, and 100% of benzodiazepines remained stable across treatment. The doses of 93% of antidepressants, 75% of stimulants, and 100% of benzodiazepines remained stable during the follow-up period. There were no group differences in discontinuation or reduction during the treatment phase (TranS-C = 7%; PE = 6%) or the follow-up phase (TranS-C = 8%; PE = 6%). There were no group differences in antidepressant dose increase during the treatment phase (TranS-C = 7%; PE = 6%) or the follow-up phase (TranS-C = 0% PE = 6%). At study entry, 45% of participants were participating in therapy outside the study; there were no group differences in therapy participation (TranS-C = 53%; PE = 38%).

Depression Severity Trajectories

Figure 2 depicts the results of the mixed effects model to assess the Aim 1 hypothesis (Model 1) that TranS-C, when compared to an active control condition, will result in significant improvements in MDD severity among adolescents with ECP and clinically significant depressive symptoms. The rate of change in MDD severity from baseline to 12-month follow-up differed by treatment group (p=.002), a significant change of larger magnitude was observed in TranS-C (p= .001), whereas there was a smaller nonsignificant trend in PE (p=.09), as displayed in Table 2. At the 6-month follow-up the TranS-C group had a lower average CDRS score (without the sleep item; M=21.67, SD=7.35) than the Control Group (M=32.5, SD=6.9) and the CDRS scores were significantly different between treatment groups t(32) = 4.41, p<.001. At the 12-month follow-up the TranS-C group also had a lower average CDRS score (without the sleep item; M=24.97, SD=8.65) than the Control Group (M=32.75, SD=11.12) and the CDRS scores were significantly different between treatment groups t(32) = 2.29, p=.03. The difference between CDRS scores at post-treatment was not significant between groups (p>.05). Remission from MDD, defined as CDRS<=28 at 12 month follow-up, was not significantly different between groups (59% in TranS-C, 44% in PE).

Figure 2.

Change in Mean CDRS Scores From Baseline Through 12-Month Follow Up by Treatment Group^a

^a Vertical lines represent standard errors ± SEM.

Abbreviations: TranS-C= Transdiagnostic Sleep and Circadian Intervention, PE= Psychoeducation, CDRS= Children’s Depression Rating Scale after removing the sleep item.

Table 2.

Fixed Effects Estimates, Standard Errors and 95% Confidence Intervals for Mixed Effects Models.

Fixed Effect	Estimate	Standard Error	95% CI (Model Based)	P
Model 1: Aim 1 a
PE x Time	−2.62	1.51	(−5.67, .42)	.08
TranS-C x Time	−4.95	1.41	(−7.78, −2.12)	.0004**
Model 2: Aim 2 step 2 b
PADDLMO Waketime	.16	.88	(−1.65, 1.97)	.86
Time	−3.17	.86	(−4.96, −1.38)	.0002**
Time x Treatment Arm	−3.58	1.67	(−7.05, −.11)	.03*
Time x Treatment Arm x PADDLMO Waketime	3.86	1.44	(.86, 6.87)	.007**

^aModel 1 is a mixed-effects linear model with autoregressive error structure to examine relative differences between treatments with respect to the rate of change in CDRS. The model included a random slope and fixed effects for treatment.

^bModel 2 is a mixed effects linear model with autoregressive error structure to determine if the change in PAD_{DLMO Waketime} from baseline to post-treatment mediated depression symptom reduction through 12 month follow-up. The model included treatment, change in PAD_{DLMO Waketime} from baseline to post-treatment (individual slopes) and their interaction.

^*indicates a p-value of <.05;

^**indicates a p-value <.001

Relationship Between Improvements in PAD and Depression

The first step of the mediation analysis, with change in PAD_{DLMO WAKETIME} from pre to post-treatment as a dependent variable, found a differential effect comparing the two treatment conditions. The TranS-C group had significantly longer PAD_{DLMO WAKETIME} by post-treatment than the PE group (B= −.924, p=.035), indicating more alignment between the endogenous circadian phase and wakeup time (see Appendix S2). There was no treatment effect on PAD_{DLMO BEDTIME} or PAD_{DLMO MIDSLEEP}. Next, we found that elongating PAD_{DLMO WAKETIME} from pre- to post-treatment was associated with lower depression scores at 12-month follow-up (B= −.05, p=.03; see Figure 3).

Figure 3.

Relationship Between Change in Phase Angle Difference between DLMO and Waketime^a from Pre to Post-Treatment and Depression Severity at 12 Month Follow-Up in the TranS-C group^b

^a Change in PAD_{DLMO WAKETIME} from pre to post-treatment was defined as the difference in PAD_{DLMO WAKETIME} scores from pre to post-treatment (PAD Slope between pre and post treatment= baseline PAD_{DLMO WAKETIME}- post-treatment PAD_{DLMO WAKETIME}).

^b Vertical lines represent standard errors ± SEM.

Abbreviations: PAD= phase angle difference, TranS-C= Transdiagnostic Sleep and Circadian Intervention, CDRS= Children’s Depression Rating Scale after removing the sleep item.

Note. In the TranS-C group lengthened PAD_{DLMO WAKETIME} from baseline to post-treatment (a more negative PAD_{DLMO WAKETIME}) was associated with lower depression severity at 12 month follow-up.

Given the significant treatment effect on PAD_{DLMO WAKETIME} we proceeded to the second step of the mediation analysis, with depression severity as the outcome variable, analyses revealed a significant interaction between change in PAD_{DLMO WAKETIME} and treatment arm (p= .014), which according to the MacArthur Framework indicates a significant mediation.⁴⁷ The results of Model 2 are displayed in Table 2. The overall mediation model is displayed in Figure 4. Within each treatment arm, with depression severity as the outcome variable, there was a non-significant effect of change in PAD_{DLMO WAKETIME} in the PE and TranS-C arms (p>.05).

Figure 4.

Model displaying the fixed effects estimates and beta coefficients of the mediating effects of change in PAD_{DLMO WAKETIME} from baseline to post-treatment on the relationship between treatment group and depression at 12 months follow-up

* Indicates p-values <.05; ** indicates p-values <.001

Solid lines indicate fixed effects estimates from mixed effects analyses.

Dotted lines indicate beta coefficients from linear regression analyses.

Value in parentheses is the overall indirect effect of PAD_{DLMO WAKETIME} on the model.

Note. According to the MacArthur Mediation Framework a significant overall indirect effect (the interaction between time, treatment, and the mediator) establishes a mediation.

Sensitivity and Specificity Analyses

To assess the specificity of our analyses, we also examined total sleep time and sleep efficiency change from pre to post treatment as potential mediators. The models that included total sleep time (p=.69) and sleep efficiency (p= .12) as mediators were non-significant. To assess the sensitivity of our analyses, we ran Model 2 controlling for total sleep time and sleep efficiency. PAD_{DLMO WAKETIME} remained a significant mediator after controlling for both total sleep time (p=.03) and sleep efficiency (p= .02).

Discussion

We examined the effect of TranS-C, compared to PE, on depression symptoms in a sample of depressed adolescents with an evening circadian preference (ECP) through 12 month follow-up and examined potential mediators of antidepressant effects. Randomization to TranS-C, compared to PE, resulted in improved depression trajectories among adolescents with an ECP and clinically significant reduction in depressive symptoms (Aim 1). Depression symptom improvement occurred at a faster rate and with a greater magnitude in the TranS-C group, relative to the PE group. Lengthening the phase angle between DLMO and waketime (PAD_{DLMO WAKETIME}) during the acute treatment phase was a significant predictor of depression severity improvement at 12 month follow up and mediated the effect of treatment group on depression severity 12 months later (Aim 2). In other words, longer PAD_{DLMO WAKETIME} during the acute treatment phase mediated the effect of treatment on depression symptoms. This approach is consistent with recent calls for a proposed new clinical phenotype, ‘circadian depression,’ a phenotype whose onset is most likely to be in adolescence.⁴⁸

To our knowledge, this is the first report that has tested the impact of modifying the alignment between circadian biology and sleep/wake behavior on depression outcomes in adolescents with clinically significant depressive symptoms using a behavioral sleep intervention. Cross sectional data in multiple samples of adults with MDD show a strong association between depression and alignment between circadian biology and sleep/wake behavior.^49,50 Experimentally induced misalignment between circadian biology and sleep/wake behavior has been shown to increase depression symptom severity in healthy adults,⁵¹ shift workers⁵² and adults with MDD.⁵³ Moreover, a study using UK biobank data found that misalignment between circadian biology and sleep/wake behavior increased the odds of MDD even after controlling for circadian preference.¹² Only one experimental study of misalignment between circadian biology and sleep/wake behavior has been conducted among adolescents, which found that experimentally inducing misalignment among healthy adolescents had adverse consequences on neural networks of reward processing, a process which is impacted in MDD.⁵⁴ More research is needed to examine the relationship between misalignment between circadian biology and sleep/wake behavior and depression among adolescents, a group for whom misalignment between circadian biology and sleep/wake behavior is highly salient.⁵⁵

Moreover, there is a large body of work which has sought to advance circadian rhythms as a means of improving depression symptoms. This work, although largely conducted in adults, has found that phase advancement, through bright light chronotherapy, has a significant antidepressant effect among adults with seasonal depression⁵⁶; however, outcomes among patients with non-seasonal depression have been more mixed and have largely ignored circadian preference as a potential moderator.⁵⁷ Moreover, it is noteworthy that previous studies testing augmenting depression treatment with cognitive behavioral therapy for insomnia among patients with MDD and insomnia have failed to find an effect of a sleep intervention on depression symptoms among adolescent and adult patients with MDD.^58,59 This raises the question of whether the key sleep/circadian intervention target for adolescents with depression and ECP is correction of misalignment between circadian biology and sleep/wake behavior rather than insomnia or phase advancement. This also raises the question of whether alignment between circadian biology and sleep/wake behavior may be a mechanism from which other sleep problems may arise in adolescent depression (i.e., ECP adolescents who attempt to go to bed at an adverse circadian time then also tend to develop insomnia).

While the present study found a significant effect of treatment on MDD severity at the 6- and 12-month follow-up timepoints, we did not find a significant impact of treatment on MDD severity at post-treatment. Given that we saw an impact of treatment on the proposed mechanism, PAD_{DLMO WAKETIME}, at post-treatment it is likely that the lag in depression improvement until the follow-up period is in response to the changes in PAD_{DLMO WAKETIME} at post-treatment. Future studies should investigate how soon after lengthening PAD_{DLMO WAKETIME} depression improvements may occur.

While PAD_{DLMO WAKETIME} did emerge as a significant mediator of depression improvement, we were surprised to find no treatment effect on circadian alignment between PAD_{DLMO BEDTIME} or PAD_{DLMO MIDSLEEP}. It is possible that this lack of effect was due to a lack of power. However, given that our data spans a wide developmental period (12–18 years of age) data supports that PAD_{DLMO WAKETIME} may best control for developmental changes across adolescence due to the consistency in school start times across the period.²⁴ Moreover, TranS-C is a multidimensional treatment which improves circadian alignment by advancing DLMO and bedtime, delaying waketime (within the confines of school start times), increasing total sleep time and improving sleep efficiency. For example, youth may lengthen their PAD_{DLMO WAKETIME} during TranS-C by advancing their DLMO, delaying their waketime, or a combination of advancing their DLMO and delaying their waketime (this may result in increased total sleep time or not depending on when they went to bed). However, while TranS-C resulted in improvements in total sleep time, sleep efficiency, DLMO, PAD_{DLMO BEDTIME}, and PAD_{DLMO MIDSLEEP} analyses indicate that the change in in PAD_{DLMO WAKETIME} was the only variable that mediated MDD severity improvement. It is likely that PAD_{DLMO WAKETIME} is the measure of circadian alignment that best captures all of the changes that occur during TranS-C (i.e. delaying waketime, advancing DLMO, advancing bedtime, and extending total sleep time), whereas PAD_{DLMO BEDTIME} would only capture advances in DLMO and bedtime, and similarly PAD_{DLMO MIDSLEEP} would only capture advances in DLMO and bedtime and increase in total sleep time. Importantly, neither total sleep time nor sleep efficiency were associated with improvement in MDD severity. In summary, PAD_{DLMO WAKETIME} as the variable that best captures the complex sleep and circadian shifts that occur with TranS-C, may also best capture what about TranS-C is improving MDD severity. However, this finding needs to be replicated in a larger and more adequately powered sample.

Our findings should be interpreted in light of strengths and limitations. However, several important limitations should also be considered. First, this was a secondary analysis in a underpowered and relatively small sub-sample of adolescents with clinically-significant depressive symptoms enrolled in a parent clinical trial. In the parent trial, although participants were not selected on the basis of depression, risk in one of five health domains (including emotional health) was part of the inclusion criteria. These findings should be replicated in a larger and adequately powered sample. Second, depression treatment was not consistent between participants. Some participants took antidepressants, others were in psychotherapy, and others were untreated. To confirm that the results were not due to the heterogeneity of depression treatment we also conducted the analyses including medication use as a covariate and the results remained similarly significant. Third, generalizability to samples with fewer exclusion criteria and to lower-income families should be assessed. Fourth, while we did use statistical analyses that account for missing data, it is important to acknowledge that follow-up data was missing for 8 participants. Additionally, while in the current study we used a structured interview to screen for sleep disorders, objective sleep recordings are needed for the diagnosis of other sleep disorders (e.g., obstructive sleep apnea).

ECP affects over 80% of adolescents with depression.⁶⁰ Results from this secondary analysis of a larger behavioral sleep intervention trial for ECP adolescents provide novel evidence for improved alignment between circadian biology and sleep/wake behavior as specific mechanism of depression improvement. These preliminary data provide key clues into the complex relationship between sleep and depression severity and have significant clinical implications for adolescents with depression and ECP. Further research is needed in a larger sample to confirm these results.

Supplementary Material

SUP FIG1

Figure S1. CONSORT diagram illustrating the flow of participants through the study.

SUPINFO

Appendix S1. Description of Treatment Sessions.

Appendix S2. Additional Analyses.

SUP FIG2

Figure S2. Average Change in Phase Angle Difference between DLMO and Waketime Between Pre and Post Treatment by Treatment Group.

Key points.

• Evening Circadian Preference is common among adolescents and ) is associated with increased depression severity, recurrent MDD episodes, increased suicidality and poor antidepressant response.

• To our knowledge, this is the first report that has tested the impact of modifying the alignment between circadian biology and sleep/wake behavior on depression outcomes in adolescents with clinically significant depressive symptoms and evening circadian preference using a behavioral sleep intervention.

• Results provide novel evidence for improved alignment between circadian biology and sleep/wake behavior as specific mechanism of depression improvement among adolescents with an evening circadian preference.

• The results provide evidence that TranS-C, a low stigma, safe, sleep intervention can using a data driven precision medicine approach, improve depression for adolescents with clinically significant depression symptoms and evening circadian preference.