Melatonin circadian phase timing and stability: Associations with mood symptoms among a clinically-enriched sample of young adults with mood disorders

Abstract

Circadian rhythms regulate many aspects of human health, and disruptions in these rhythms have been implicated in mood disorders. Dim light melatonin onset (DLMO) provides a gold standard measure of circadian timing, yet few studies have examined its stability or relation to mood symptoms in psychiatric populations. This study examined DLMO across three time points over 20 days in a clinically-enriched sample of 124 young adults with and without mood disorders. Participants also reported mood symptom levels thrice daily via ecological momentary assessments (EMA). Circadian timing and stability were quantified and their associations with mood symptoms and diurnal mood patterns were assessed. Overall, DLMO demonstrated substantial stability, and did not differ by diagnostic group. The mood disorder group had earlier circadian timing than healthy controls. Hypomanic, but not depressive, symptoms were consistently associated with later DLMO in both cross-sectional and prospective analyses. Moreover, hypomanic symptoms exhibited a diurnal pattern that differed for participants with earlier and later circadian timing. Future work should examine endogenous circadian rhythms in relation to mood disorder subtypes and mood episodes to determine whether circadian timing changes can be used to predict or intervene on psychiatric symptoms.

Introduction

Circadian rhythms are processes that repeat approximately every 24 hours and regulate many aspects of human health.¹ Disruptions or irregularities in circadian rhythms have been implicated in several psychiatric conditions, most notably mood disorders (i.e., major depression and bipolar spectrum disorders [BSD]).² Longitudinal research indicates circadian disruption increases risk for major depression,^3,4 while empirical and theoretical work support circadian disruption in the etiology of BSD as well.^5–7 For example, abnormal melatonin secretion, rest-activity rhythm instability, and altered core body temperature rhythms are associated with BSD,^8–10 with some evidence suggesting indices of variability, stability, and circadian amplitude can predict mood episode recurrence.¹¹ Prominent biopsychosocial models of BSD assert that circadian dysregulation can lead to hypo/manic and depressive episodes,^5,12 and changes in seasons and time zones have been linked to the onset of bipolar mood episodes.¹³ However, the mechanisms linking mood disorders and the circadian system are not fully understood.

Circadian timing and stability may be especially relevant aspects of circadian physiology for advancing our understanding of mood disorders.¹⁴ Most research in this area has focused on chronotype, the self-reported preference for the timing of sleep and activity. Eveningness, the preference for wakefulness and activity later in the day, consistently has been associated with mood disorders.^15–20 Eveningness prospectively predicts depressive episodes and greater symptom severity in bipolar disorder and similarly is linked to higher depressive symptoms in the general population.^16,17,21,22 However, chronotype is only a proxy for endogenous circadian timing, and is vulnerable to “masking” or obscuring by lifestyle and environmental factors. Dim light melatonin onset (DLMO) is the gold standard for circadian timing, but the few studies on DLMO and mood symptoms are limited by small samples, likely due to burdensome protocols. Salivary melatonin levels, which are about one-third lower than plasma levels, can offer a non-invasive window into endogenous circadian rhythms.²³

A clearer understanding of the extent to which the timing of the internal biological clock is related to mood disorders and their symptoms is necessary to pinpoint novel treatment targets. Extant research is mixed. One small study did not find differences in DLMO among depressed and non-depressed individuals²⁴, whereas another found that DLMO was later in adults with seasonal affective disorder (i.e., major depression with seasonal pattern) relative to controls²⁵. In a larger study of 50 depressed and non-depressed young adults, a data-driven approach was used to identify a “delayed circadian timing” cluster, which was predominantly (82%) composed of depressed individuals with significantly more severe depressive and hypomanic symptoms relative to the conventional timing cluster²⁶. This study also reported that individuals diagnosed with depression had DLMOs ~1.5 hours later than the non-depressed control group²⁶. Conversely, some evidence suggests that individuals with bipolar disorder exhibit earlier circadian timing -- ~1.5 hours earlier than healthy controls²⁷.

Beyond understanding circadian timing in mood disorders, the stability of circadian timing across time may be a related construct of importance.²⁸ Emerging research into sleep irregularity (i.e., the timing of sleep and wake behaviors) suggests that sleep irregularity is associated with depressive symptoms, and a hypothesized mechanism driving these findings may be disrupted circadian physiology or irregular circadian timing.^29,30 However, there are very few studies examining the stability of endogenous circadian timing. In one small study, 15 college students completed a DLMO protocol in naturalistic settings on three occasions over a 100-day period, showing high circadian phase stability.³¹ Another study of eight individuals with delayed sleep phase syndrome found that although their DLMO timing was later than that of controls, it remained consistent across three sampling sessions spaced five days apart,³² despite the tendency for individuals with delayed sleep phase to have more irregular sleep patterns.³³ However, circadian stability has not yet been explored in clinical populations or in association with clinical symptoms.

Circadian timing remains underexplored regarding when mood symptoms are most prominent. Historically, the melancholic subtype of depression—marked by anhedonia and reduced mood reactivity—has been associated with more severe depressive symptoms in the morning;³⁴ whether circadian physiology underlies this observation is unknown. Compared to morning chronotype, evening chronotype has been linked to more pronounced negative emotional symptoms in the evening, although this study did not collect a biological indicator of circadian timing.³⁵ Among adolescents with evening chronotype, later DLMO was associated with higher negative affect in the evening, but there was no significant relationship between DLMO and morning affect.³⁶

A separate literature has investigated diurnal variation in positive and negative affect, finding that positive affect demonstrates a reliable circadian rhythm, starting low in the morning, peaking in the early afternoon, and declining throughout the evening.^37,38 In contrast, negative affect does not appear to have a regular circadian pattern. Prior work suggests the diurnal variation in positive affect is blunted among individuals with depressive symptoms,³⁹ but this work has not yet determined whether diurnal variation differs among those with delayed circadian timing, and whether these findings extend to mood disorder symptomatology. Addressing this question is a step toward the goal of not only identifying who is most at risk for negative mental health outcomes but also determining when they are most at risk.

The aims of this study were as follows: (1) Quantify circadian timing and its stability and determine whether these differ by diagnostic group (i.e., those with and without a mood disorder), (2) Test whether mood symptom severity is concurrently or prospectively associated with circadian timing and/or its stability, and (3) Investigate diurnal variation in mood symptoms and test whether circadian timing moderates potential time-of-day effects. We hypothesized that mood disorders would be characterized by later, more variable, and less stable DLMO. We also expected that later DLMO would be associated with greater mood symptom severity and that symptom timing would be positively correlated with DLMO timing such that those with later DLMO would experience higher mood symptoms later in the day.

Methods

Participants.

Participants were drawn from an ongoing research study entitled Project TEAM (Teen Emotion and Motivation), a prospective longitudinal study examining risk factors for the onset and course of bipolar spectrum disorder (BSD).⁴⁰ Additional information about Project TEAM selection criteria is in the supplement. Participants were divided into two groups: those with and without a mood disorder diagnosis. Participants in the mood disorder group were individuals who met DSM-IV-TR criteria for a major depressive episode, bipolar I, bipolar II, bipolar NOS, or cyclothymia at any Project TEAM follow-up prior to starting this EMA study.^a

This study included 124 Project TEAM participants recruited ~3 years after project onset (between 2014–2017). Participants underwent a 20-day ecological momentary assessment (EMA) protocol with thrice-daily smartphone surveys of depressive and hypo/manic symptoms and saliva collection for DLMO estimation on days 1, 10, and 20. EMA prompts were delivered randomly within three 4-hour windows (0800h–1200h, 1200h–1600h, 1600h–2000h). Self-reported mood measures from the most recent Project TEAM biannual follow-up also were included (mean duration prior to DLMO assessment = 262 days, SD=387 days). The study was approved by the Temple University IRB, and all participants provided written informed consent.

Measures.

Dim light melatonin onset (DLMO).

Participants completed the DLMO protocol at home on the evenings of Day 1, 10, and 20 of the EMA study. Nonsteroidal anti-inflammatory drugs (NSAID) were prohibited during the protocol, and alcohol and caffeine were prohibited for > 24 h before each DLMO procedure. On the day of sampling, participants were instructed to refrain from consuming beverages containing artificial colorants, bananas, and chocolate for the entire day of the sampling protocol to minimize potential dietary influences on salivary melatonin levels. Participants were instructed to complete their evening meal at least 30 minutes before the sampling began and to avoid any food intake during the procedure. Water was allowed only within five minutes following each saliva sample collection. Five hours prior to habitual bedtime (and 30 minutes before the first sample), participants donned light-attenuating goggles with side shields with 4% transmission (Model U23, Noir Medical Technologies, South Lyon, MI) to minimize light exposure. This method has been used in several other in-lab and at-home DLMO studies.^41–44 Habitual bedtime was determined via self-report by asking participants to recall their bedtime the week preceding the DLMO protocol. The same sampling schedule was used for the three timepoints. Participants were told not to engage in physical activity and to limit themselves to passive activities throughout the 4.5 hour sampling window (e.g., listening to music, reading). Participants provided nine saliva samples at 30-minute intervals. Absorbent polyester swabs were placed in the mouth for 5 min and then deposited into Salivette tubes. Participants kept the salivettes in the freezer until they were brought to the lab where they were stored at −20°C pending assay.

Salivary melatonin concentrations were assessed using enzyme-linked immunosorbent assay (ELISA). The minimum detectable limit of the assay was 0.5 pg/ml with intra- and inter-assay coefficients of variation of 6.8% and 7.3%, respectively. For quality control, 10% of the samples were run in duplicate. DLMO, the gold standard marker of circadian phase, was defined as the first interpolated point at 3.0 pg/ml on the rising curve of melatonin concentration, consistent with the approach used in other in-lab and at-home DLMO procedures.^41–46

Diagnostic Interview.

The expanded Schedule for Affective Disorders and Schizophrenia – Lifetime (exp-SADS-L) is a structured clinical diagnostic interview that was administered by trained clinical interviewers at Project TEAM baseline to ascertain lifetime psychiatric disorders. The expanded Schedule for Affective Disorders and Schizophrenia – Change (exp-SADS-C) was administered at biannual follow-ups. The exp-SADS has demonstrated high inter-rater reliability (κ>.96).^40,47

Mood Symptoms.

Depressive Symptoms.

Prior to the EMA protocol, participants completed the full Beck Depression Inventory-II⁴⁸ (BDI-II) to assess depressive symptom severity. Four items from the BDI-II (items #1, 2, 7, 15) were selected for use in three daily EMA prompts, which assessed sadness, hopelessness, low self-esteem, and low energy. On EMA surveys, participants rated each item on a scale of 1–4 and responses from the four items were summed to create a composite score.

Hypomanic symptoms.

Participants also completed the Altman Self-Rating Mania Scale (ASRM) prior to starting the EMA protocol to assess hypo/manic symptoms. Four items from the ASRM were selected for use in the three daily EMA surveys (inflated self-confidence, talkativeness, reduced need for sleep, and elevated mood). Participants rated how they felt at that moment on a five-point Likert scale from 0–4 and responses were summed to create a composite score.

Data analysis.

Mood symptom variables and DLMO were z-standardized. Linear regression tested diagnostic group differences in DLMO (controlling for sex, race, and age) and examined cross-sectional and prospective associations with mood symptoms. For participants with at least two DLMO estimates, standard deviations were computed and then averaged by diagnostic status. Intraclass correlation coefficients (ICC) were calculated based on 5,000 bootstrapped samples to examine individual consistency in DLMO.

To evaluate whether mood symptoms changed as a function of time of day, models were estimated with and without linear, quadratic, and cubic effects of time of day, as well as with and without interactions with DLMO. Bayesian information criterion (BIC) statistics were compared to determine the best fitting specification. We tested an autoregressive parameter to account for autocorrelation based on hourly time, but results were very similar to models without such a specification, so we reported the latter. Models did not converge when random slopes were introduced, so all models were specified with three random intercepts (day, ID, and day nested within ID), and zero random slopes.

Results

Table 1 displays sample characteristics. A total of 150 participants were recruited for the overarching EMA study, but only 124 participants (mean age = 21.87, SD=2.11, range = 18–27) had data on the study variables relevant to this investigation. Attrition analyses suggest there were no major differences in participants included and excluded (see Supplement for details). Approximately 44% were healthy controls (n=54) and 56% were in the mood disorder group (n=44 major depression, n=2 bipolar I disorder, n=16 bipolar II disorder, n=3 bipolar NOS, n=5 cyclothymia). On average, participants contributed interpretable data for approximately two DLMO estimates. Reasons for lack of interpretability (e.g., DLMO threshold not reached, non-compliance, etc.) are detailed in the supplement. In brief, the reasons for lack of interpretability did not differ by diagnostic or clinical variables, although the mood disorder group provided more interpretable estimates on average relative to the control group (2.07 vs. 1.76). Compliance with the EMA surveys was high overall (84% completion rate); refer to the supplement for additional details. Depressive symptoms on the BDI were in the “minimal” range (M=5.40, SD=6.18, range 0–27) and mean hypomanic scores on the ASRM were 4.45 (SD=3.71, range = 0–15). Four participants were in a depressive episode at the time of the EMA study.

Table 1.

Sample characteristics

	Mean (SD) / N (%)
Age	21.87 (2.11)
Sex (Female)	74 (60%)
Race
White	79 (64%)
Black	26 (21%)
Asian	14 (11%)
Multiracial	3 (2%)
Other	2 (2%)
Ethnicity (Hispanic)	11 (9%)
Diagnostic Category
Mood Disorder	70 (56%)
Healthy Control	54 (44%)
Psychotropic Medication Use	7 (6%)

Notes: Full sample = 124 participants.

CIRCADIAN TIMING AND STABILITY

The mean DLMO for the sample was 22.01h (SD=1.44h). Mean DLMO per timepoint was: T1 M=21.95h (SD=1.58; n=96), T2 M=21.98 (SD=1.56, n=79), and T3 M=21.80 (SD=1.25; n=65). When stratified by diagnostic group, mean DLMO was 22.28h (SD=1.28h) for healthy controls (n=54) and 21.82h (SD=1.52h) for the mood disorder group (n=70) (see Figure 1).^b Mood disorder diagnosis was significantly associated with an earlier DLMO than healthy controls (b=−.51, SE=.24, p=.04).

Figure 1.

Average dim light melatonin onset by diagnostic group (n=54 healthy controls; n=70 mood disorders)

The majority (44/58) of the 58 participants with T1 and T2 samples had DLMOs within one hour of the other. Of the remaining 14, six had T2 DLMO ≥ one hour earlier than T1, and eight had T2 DLMO ≥ one hour later than T1. Approximately 95% (55/58) had DLMOs within 2 hours of one another. The mean absolute difference between T1 and T2 was 42m (SD=45.6). There were 44 participants with T2 and T3 samples. A total of 59% (n=26) had T2 and T3 DLMO within one hour of the other. Of the remaining 18 participants, 12 had T3 DLMO later than T2 DLMO by ≥ one hour, whereas 6 had T3 DLMO earlier than T2 DLMO ≥ one hour. All but one participant had T2 and T3 DLMOs within 2 hours of one another. The mean absolute difference between T2 and T3 was 52.8m (SD=37.2m). The percentage of participants with consecutive DLMOs more than an hour apart was 36.74% for the mood disorder group and 22.58% for healthy controls; this difference was not statistically significant (X²=1.17, df=1, p=.28). Of the 50 participants with T1 and T3 samples, the mean intra-individual (absolute value) difference was 54.6m (SD=38.4).

Among the 80 unique participants with two or more DLMOs, the mean standard deviation across DLMO was 36.6m (SD=28.2m). Variability did not differ by diagnostic group (t(66.68)=.08, p=.94). Figure 2 shows DLMO across all three timepoints for the 36 participants with complete data (mood disorders: n=26; controls: n=10).

Figure 2.

Dim light melatonin onset at T1, T2, and T3 (n=36)

CIRCADIAN TIMING AND MOOD SYMPTOMS

The strength of the intraclass correlation coefficient (ICC) for the entire sample was in the “substantial” range (0.75 [95% CI=.60–.83]). When stratified by group, the ICC remained in the “substantial” range for healthy controls (0.67 [95% CI=.46–.87]) and the mood disorder group (0.78 [95% CI=.65–.87]). Overlapping confidence intervals suggest the ICCs did not significantly differ by group.

Depressive symptoms (BDI-II) were not prospectively predictive of average DLMO (B=−.08, SE=.10, p=.37), but hypomanic symptoms (ASRM) were prospectively associated with average DLMO (B=.24, SE=.09, p=.01), such that greater hypomanic symptoms were associated with later DLMO. Neither BDI score nor ASRM score were predictive of DLMO variability (all ps>.56). Mean depressive symptoms throughout the 20-day EMA period were not concurrently associated with average DLMO (B=−.03, SE=.09, p=.78), but mean EMA-assessed hypomanic symptoms were positively associated with average DLMO (B=.20, SE=.09, p=.03), such that greater hypomanic symptoms were associated with later DLMO. The overall pattern of results remained unchanged after including psychotropic medication use as a covariate in the models, and in sensitivity analyses excluding the four participants who were in a major depressive episode.

CIRCADIAN TIMING, MOOD SYMPTOMS, AND TIME OF DAY

Results are presented in Table 2. For depressive symptoms, the best specification was a model without any fixed effects for hourly time.^c The main effect of DLMO was not significant (B=−0.01, SE=.04, p=.76). For hypomanic symptoms, a cubic model of time of day interacting with DLMO (hour, hour², and hour³ each interacting with DLMO) produced the best fitting model. The interaction was probed and predictions based on this model (for individuals with DLMO +/− 1SD) are presented in Figure 3 and Table 3. The results show a sharp increase in hypomanic symptoms in the morning and early afternoon. People with earlier DLMOs show hypomanic symptoms peaking at around 1400 hours, and then the symptoms level off. People with later DLMOs have gradual increases in symptoms such that they peak in the evening hours. Controlling for psychotropic medication yielded comparable results. Actual data were plotted by hour in Supplemental Figure 1, with data plotted separately for individuals with DLMO earlier and later than the sample mean. Overall, the data are consistent with the modelled pattern.

Table 2.

Time-of-day effects on mood, moderated by DLMO

	Depressive Symptoms			Hypomanic Symptoms
	B	SE	p	B	SE	p
Hour				−76.43	10.28	<.001
Hour2				7.08	10.48	.50
Hour3				−3.96	10.95	.72
DLMO	−.01	.04	.76	.10	.04	.02
DLMO × Hour				3.78	.47	<.001
DLMO × Hour2				−.60	.47	.20
DLMO × Hour3				.34	.50	.50

Notes: DLMO = Dim light melatonin onset. DLMO and mood symptoms were z-standardized. Models controlled for age, sex, and race. Time (hour) was centered on its mean of 15.43. Full maximum likelihood was used to produce the models. Models included 6,227 observations nested in 20 days nested in 124 participants.

Figure 3.

Modelled hypomanic symptom level data throughout the day among individuals with earlier and later DLMO

Note: Hypomanic symptoms are in original units.

Table 3.

Effect of hour on hypomanic symptoms by time of day and by DLMO levels

Time	−1 SD DLMO			+1 SD DLMO
	B	SE	p	B	SE	p
0800	.16	.03	<.001	.24	.03	<.001
1200	.04	.01	<.001	.08	.01	<.001
1600	−.02	.01	<.01	.01	<.01	.39
2000	−.02	.01	.06	.01	.01	.36

Notes: DLMO = Dim light melatonin onset; SD = standard deviation

Discussion

This study examined circadian timing across three time points in a clinically-enriched sample of young adults, providing insight into melatonin phase and stability in a real-world setting and their relevance to mood disorders. Four main findings emerged: (1) circadian phase was relatively stable over 20 days, with no differences in DLMO stability between diagnostic groups; (2) the mood disorder group had earlier circadian timing than the control group; (3) hypomanic -- but not depressive -- symptoms were consistently associated with later DLMO in both prospective and cross-sectional analyses; and (4) hypomanic symptoms showed a diurnal pattern, with earlier versus later circadian timing linked to distinct symptom profiles.

The sample’s DLMO ICC was in the “substantial” range,⁴⁹ slightly lower than that reported for another real-world study (ICC=.85 [in the “almost perfect” range]), but generally consistent with previous reports. Nearly all participants had consecutive DLMOs within 2 hours of each other, which is similar to a laboratory-based study in which researchers found that 16 of 18 participants maintained DLMO within 2 hours of one another, even when the follow-up interval was substantially longer (9 months – 3 years).⁵⁰ Another metric of stability, the mean absolute difference between timepoints, ranged from 42–53 minutes, which aligns with that reported by Benloucif and colleagues⁵¹ (43-minute change across 3 weeks).

We did not find evidence supporting our hypothesis that individuals with mood disorders would have more DLMO instability or variability. This is somewhat in contrast to prior findings indicating that individuals with mood disorders experience notable social rhythm disruptions and are sensitive to social rhythm disrupting events.^52,53 However, although conceptual and theoretical models assert that social and circadian rhythms are correlated, if not causally related,^54,55 there is a dearth of empirical evidence quantifying their association. Further, the number of uninterpretable samples—whether due to starting above the sampling threshold or never reaching it—artificially restricted the observable range of DLMO estimates (thereby precluding the detection of extreme variability), and may have contributed to the null findings. A longer or more flexible sampling window may be necessary to capture DLMO more reliably in clinical or at-risk samples.

It has been shown that day-to-day changes in sleep are associated with daily fluctuations in mood disorder symptoms.⁴⁵ If night-to-night sleep variability or within-person changes in sleep metrics are an indicator for mood pathology, a related question is whether changes in endogenous circadian timing metrics also emerge at this granular of a level, or whether chronic sleep changes are necessary to be able to detect changes in circadian physiology. There is limited research on the temporal dynamics of endogenous circadian timing, and it is possible that three measurements across a 20-day period is not the appropriate sampling frequency to detect meaningful changes or variability in DLMO, and a shorter follow-up interval would be necessary to detect these changes. Interestingly, prior research demonstrated that DLMO stability did not differ when follow-up intervals were <1 month versus >3 months⁵¹. Given the observed stability of endogenous circadian timing over weeks/months, detecting variability may require more frequent measurement than our protocol provided.

Alternatively, circadian instability may be more closely associated with mood episodes, reflecting a state, rather than trait difference. If that is the case, circadian instability would be detectable only during a manic or depressive episode. Nearly all participants (all but four) in the mood disorder group were euthymic during the protocol, which also may have contributed to the lack of between-group stability differences. This possibility fits nicely with literature demonstrating that sleep irregularity, which is hypothesized to be linked to circadian instability, is correlated with depressive symptom severity.^29,30 Extending this line of reasoning, circadian instability may be more pronounced in individuals with later circadian timing, consistent with evidence linking delayed circadian phase to irregular sleep patterns and mood symptoms.³³

Contrary to expectations, we observed that DLMO was earlier for the mood disorder group than the control group by approximately 30 minutes. This was most aligned with the previous finding that individuals with bipolar disorder have advanced circadian timing.²⁷ Given that the umbrella study from which this sample was drawn, Project TEAM, oversampled for individuals hypothesized to be at-risk for BSD due to elevated reward sensitivity, our sample may reflect a stronger bipolar diathesis. Although the clinical significance of a 30-minute phase advance is not fully understood, it is interesting to note that some evidence suggests that a phase advance may serve as a state-sensitive marker of mania. One study found that during acute episodes, the cortisol circadian rhythm acrophase was advanced by approximately seven hours compared to recovered states⁵⁶ (though cortisol rhythms do not always align with melatonin rhythms).⁵⁷ Another study investigated relationships between chronotype and several psychiatric traits, finding that the majority of traits were associated with eveningness, whereas mania was the only characteristic associated with morning chronotype.³⁵ Future research would benefit from additional investigation into whether a curvilinear relationship exists between DLMO and health status, where both early and late DLMOs reflect an abnormal state. Indeed, prior work has found that long and short phase angles (ostensibly reflecting DLMO earlier or later than expected relative to sleep timing) are associated with depressive symptoms.⁵⁸ Extreme DLMO timing (either advanced or delayed) may not only serve as a risk marker on its own but also reflect misalignment between the circadian clock and sleep-wake cycles. Given empirical evidence that circadian alignment is an important correlate of depressive symptoms,^46,59 it will be important for future work to consider circadian timing both as a standalone phenomenon and in the context of alignment.

Regarding the symptom-related results, hypomanic symptoms, but not depressive symptoms, were tightly linked to circadian timing. Results indicated a main effect of hypomanic symptoms such that both symptoms reported prior to the start of the EMA protocol and EMA-assessed hypomanic symptoms were associated with later circadian timing. Greater symptom severity on the ASRM was predictive of later circadian timing, and brief EMA-assessed hypomanic symptoms were concurrently associated with later circadian timing. Although at first this result may seem inconsistent with the previous finding that the mood disorder group had earlier DLMOs, it lends weight to the possibility that state and trait differences are involved. The consistency in these findings across two measures of hypomanic symptoms (and null findings across both depressive measures) increases the robustness of this observation. Future research is needed to replicate these findings.

Investigation into diurnal patterns in hypomanic symptoms suggest the main effect of hypomanic symptoms on delayed circadian phase is driven by elevated hypomanic symptoms in the afternoon and evening particularly. As shown in Figure 3, hypomanic symptoms sharply increase between ~0700h and ~1000h for the whole sample, but symptoms continue to increase in the early afternoon and gradually increase in the evening for those with later DLMO. Those with earlier circadian timing experience peak hypomanic symptoms earlier in the day (1300h) before declining slightly. The diurnal patterns of depressive and hypomanic symptoms largely paralleled those observed for negative and positive affect. Depressive symptoms showed no diurnal variation (consistent with prior findings assessing negative affect during typical waking hours). One study suggests negative affect may peak during the circadian night (~0200h-0300h)⁵⁹, but this protocol’s lack of 24-hour monitoring prevented us from testing this. Hypomanic symptoms, like positive affect, exhibited distinct diurnal patterns.^32,33 The time-of-day effects in mood disorder symptomatology may share a common underlying diathesis with affective processes. Participants with later timing (ostensibly self-identified evening types) endorsed higher levels of symptoms later in the afternoon and evening. To our knowledge, no other study has directly assessed circadian phase as a moderator of time-of-day effects, but Hasler and colleagues⁶⁰ tested a related question, whether diurnal patterns of positive affect varied by chronotype. Their pattern of findings differed from ours in that in their study, investigations into diurnal rhythms of positive affect showed that evening types had a later phase and lower amplitude,⁶⁰ which could reflect the measurement of chronotype rather than endogenous rhythms, or positive affect instead of hypomanic symptom profiles.

Several limitations are important to consider for interpretation and future research. First, the modest retention for the full DLMO protocol precluded a more nuanced investigation of variability and stability metrics. This also impacted our ability to determine between-group differences within the mood disorder group (i.e., bipolar disorder versus major depression). In light of preliminary research suggesting depressed individuals with bipolar disorder may have later circadian timing compared to those with major depression, this limitation should be explored further in future research.⁶¹ Second, this study focused on young adults, an important population given their increased vulnerability to psychiatric problems,⁶² but the findings may not generalize to older adults who tend to have a different (advanced) sleep profile.⁶³ Third, although the at-home DLMO sampling procedure enhanced ecological validity, it limited the research team’s ability to verify adherence to all collection guidelines and control lighting conditions. Specifically, we did not have measures to monitor ambient light levels or to objectively verify that participants adhered to activity, dietary, and other protocol requirements (e.g., wearing the light-attenuating goggles). The most common reason for uninterpretable DLMO estimates was failure to reach the melatonin threshold. This may reflect light exposure during the protocol, the use of self-reported rather than actigraphy-derived bedtimes to set the sampling window, or characteristics inherent to the sample that may require a longer or more expansive sampling window. If clinical samples indeed exhibit circadian irregularities or misalignment as hypothesized, habitual bedtime may be a less reliable anchor for determining the appropriate sampling window. Fourth, this study did not assess melatonin offset, which precluded investigation into this aspect of circadian timing in relation to mood symptoms. Prior work suggests that melatonin offset may be delayed in depressed individuals.⁶⁴ Because melatonin offset cannot be measured with saliva samples without disrupting sleep, we were unable to assess the duration of melatonin secretion—a circadian marker that has been shown to vary seasonally in individuals with seasonal affective disorder.⁶⁵ Fifth, future research should consider circadian alignment, measured by the phase angle of entrainment between sleep onset and DLMO, as another possible indicator of circadian health that may be relevant to advancing our understanding of circadian disturbance in mood disorders.

This study demonstrates that circadian phase is stable over short timescales in a relatively large sample of young adults, and that instability in melatonin onset timing does not appear to be an endophenotype of mood disorder pathology. Differences in circadian timing distinguished those with and without mood disorders, with the mood disorder group having an earlier circadian phase than the healthy control group. At the same time, hypomanic symptoms were consistently linked to later circadian timing and exhibited a distinct diurnal pattern. Future research would benefit from incorporating measures assessing endogenous rhythms (not just self-reported chronotype), as we did in this study, in relation to psychopathology and investigate how circadian timing pertains to mood disorder episodes, not just diagnostic group and symptom profile, to better predict and intervene on negative mental health outcomes.