Endogenous circadian regulation and phase resetting of clinical metabolic biomarkers

Abstract

Shiftwork and circadian disruption are associated with adverse metabolic effects. Therefore, we examined whether clinical biomarkers of metabolic health are under endogenous circadian regulation using a 40-h constant routine protocol (CR; constant environmental and behavioral conditions) and evaluated the impact of typical daily conditions with periodic sleep and meals (baseline; 8-h sleep at night, four meals during a 16-h wake episode) on the phase and amplitude of these rhythms. Additionally, we tested whether these circadian rhythms are reset during simulated shiftwork. Under CR (n=16 males, mean age ± SD = 23.4±2.3 years), we found endogenous circadian rhythms in cholesterol, HDL and LDL, albumin and total protein, and VLDL and triglyceride. The rhythms were masked under baseline conditions except for cholesterol, which had near-identical phases under both conditions. Resetting of the cholesterol rhythm and Dim Light Melatonin Onset (DLMO) was then tested in a study of simulated shiftwork (n=25, 14 females, 36.3±8.9 years) across four protocols; two with abrupt 8-h delay shifts and exposure to either blue-enriched or standard white light; and either an abrupt or gradual 8 h advance (1.6 h/day over 5 days) both with exposure to blue-enriched white light. In the delay protocols, the cholesterol rhythm shifted later by −3.7 h and −4.2 h, respectively, compared to −6.6 h and −4.7 h, for DLMO. There was a significant advance in cholesterol in the abrupt (5.1 h) but not the gradual (2.1 h) protocol, compared to 3.1 h and 2.8 h in DLMO, respectively. Exploratory group analysis comparing the phases of all metabolic biomarkers under both studies showed evidence of phase shifts due to simulated shiftwork. These results show that clinical biomarkers of metabolic health are under endogenous circadian regulation but that the expression of these rhythms is substantially influenced by environmental factors. These rhythms can also be reset, which has implications for understanding how both behavioral changes and circadian shifts due to shiftwork may disrupt metabolic function.

Introduction

The circadian system influences many aspects of behavior and physiology, including metabolism. Approximately ~10% of the hepatic transcriptome exhibits circadian rhythmicity suggesting circadian control of metabolic pathways regulated by the liver (1). Liver specific knockout of the circadian core clock gene BMAL1 alters glucose [hypoglycemia; (2)] and lipid [dyslipidemia; (3)] regulation, indicating a functional role of circadian gene expression in the liver in metabolism. In humans, under typical conditions with sleep at night and wake, activity and meals during the day, 24-h rhythms have been reported (4–8) in various biomarkers of metabolic function including hepatic protein production (e.g. albumin), circulating hepatic enzyme levels [e.g. alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate transaminase (AST) and gamma-glutamyl transpeptidase (GGT)] and in circulating levels of lipids and lipoproteins [e.g., cholesterol, high-density lipoprotein (HDL), low-density lipoprotein (LDL), and triglycerides]. Since periodic behavioral and environmental events such as sleep-wake, rest-activity, or feeding-fasting cycles were not removed in these studies, they cannot differentiate whether the observed 24-h rhythms were driven by an endogenous circadian pacemaker or evoked directly by the periodic changes in these factors, or a combination of both (9).

Recently, endogenous circadian regulation has been demonstrated in the human lipidome and metabolome by collecting samples over 24–40 hours (~1–1.5 circadian cycles) under constant routine (CR) conditions, which minimizes or uniformly distributes the timing of potential environmental and behavioral time cues throughout the 24-hour circadian cycle (10). These studies show that ~10–30% of the human metabolome exhibits circadian rhythmicity (11–14) and of these, the majority (~80%) were lipid metabolites (12). Lipidomics analysis under CR conditions showed ~13% of the human lipidome exhibits circadian rhythms (11). While these studies provide evidence of metabolism and lipid regulation being under endogenous circadian regulation in humans, these data were not compared to a standard basline day and so it is not clear the extent to which daily periodic events such as sleep/wake and meals impact the timing and amplitude of these rhythms. When behavior is abruptly misaligned from normal timing in simulated shiftwork studies, postprandial glucose and lipid metabolism are disrupted (15–19), as also observed in naturlistic studies with real-world conditions (20). Consistent with these observations, shift workers, who routinely experience circadian disruption, exhibit increased rates of metabolic syndrome, including obesity, diabetes, dyslipidemia, and impaired glucose tolerance compared to non-shift workers (21).

While the expression of these rhythms can be disrupted by changes in sleep, meal timing and other environmental conditions, it is unclear to what extent the endogenous circadian regulation of these rhythms is reset, or shifted, to a new phase (as opposed to masked directly) by circadian zeitgebers. A recent study showed that delaying meals by 5 hours resulted in delaying the circadian rhythms in plasma glucose and adipocyte gene expression in humans, without resetting the rhythms in plasma melatonin or triglyceride (22). In mice, time-restricted feeding can shift the circadian clock gene expression in the liver and lungs without shifting the clock in the central pacemaker located in the suprachiasmatic nucleus (23). These results suggest that 1) the rhythmic metabolic biomarkers may be reset by changes in light/dark and meal schedules; 2) the magnitude of the shifts may differ between different markers; 3) the endpoints likely have differential sensitivity to different stimuli such as light exposure and meal timing; and 4) that these metabolic outcomes are likely sensitive to circadian misalignment due to shiftwork (24, 25).

In the current study, we examined whether lipids [total cholesterol, triglycerides and free fatty acid (FFA)], lipoproteins (HDL, LDL, and VLDL), hepatic proteins (albumin, total protein) and enzymes (ALP, ATP, AST, GGT) are under endogenous circadian regulation. Furthermore, we evaluated whether these rhythms are reset in simulated shiftwork protocols where behavioral cycles (i.e., sleep, meal timing and activity) were phase advanced or delayed by 8 hours.

Materials and methods

Participants

To determine endogenous circadian regulation of metabolic biomarkers, we studied 16 males (mean age ± SD = 23.4 ± 2.3 years; range 20–28 years) in a 7-day inpatient protocol (Study 1). To determine circadian phase resetting of metabolic biomarkers, we studied another 15 males and 15 pre-menopausal females [total n = 30 (Table 1); n = 25 evaluable, mean age ± SD = 36.6 ± 9.0 years, range 25–55 years, 13 female] in four 8-day simulated shiftwork protocols (Study 2). Inpatient studies were completed at the Intensive Physiology Monitoring Unit in the Center for Clinical Investigation at Brigham and Women’s Hospital. All participants were healthy as determined by comprehensive physical, psychological and ophthalmologic examinations. Participants with extreme chronotypes [70<score<30, (26)] were excluded. Night- or shiftwork in the past 3 years or travel across more than two time zones in the previous 3 months were also exclusionary criteria. All participants maintained a self-selected, consistent 8 h:16 h sleep:wake schedule confirmed with time- and date-stamped calls at bedtime and wake-time for 3 weeks and actigraphy (Actiwatch-L, Minimitter, Inc., Bend, OR) for at least 1 week prior to entering the laboratory in order to stabilize circadian rhythmicity before starting the in-laboratory phase of the study. Scheduled sleep times that were maintained during the inpatient phase of the study were calculated based on the average bedtime, waketime and sleep duration in the week prior to laboratory admission. During the study, including the 3-week lead-in, participants were asked to refrain from use of prescription or non-prescription medications, supplements, melatonin, recreational drugs, caffeine, alcohol or nicotine. Compliance was verified by urine toxicology during admission to the inpatient phase of the study. The study was approved by the Partners Human Research Committee and all participants provided written informed consent.

Table 1.

Participant demographics for Study 1 and Study 2^†

Participant	Study	Condition		Age (years)	Sex	Waketime	DLMO
3057V	1	-		20	M	8.38	23.12
3090V	1	-		20	M	7.98	22.75
3174V	1	-		21	M	8.00	24.10
3187V	1	-		21	M	8.00	21.09
3139V	1	-		22	M	7.50	23.05
30H4V	1	-		23	M	7.08	21.23
3138V	1	-		23	M	8.18	20.25
3175V	1	-		23	M	8.15	23.94
30E7V	1	-		23	M	7.00	-
3126 V	1	-		24	M	8.62	22.91
3141V	1	-		24	M	7.98	23.89
30H6V	1	-		25	M	8.30	22.29
3109V	1	-		25	M	7.98	24.13
31A8V	1	-		25	M	6.92	22.47
3166V	1	-		27	M	8.02	21.41
3197V	1	-		28	M	5.87	19.99
3340V	2	Advance	Gradual	26	F	6.07	18.93
3214V	2	Advance	Gradual	30	F	6.18	19.59
3333V	2	Advance	Gradual	44	F	7.00	20.89
3164V	2	Advance	Gradual	46	F	7.83	22.85
3365V	2	Advance	Gradual	46	F	4.27	17.96
30H1V	2	Advance	Gradual	29	M	7.03	20.95
28E6Va	2	Advance	Gradual	30	M	8.55	-
3183V	2	Advance	Gradual	42	M	6.87	20.26
3351Va	2	Advance	Gradual	43	M	5.90	18.76
3323V	2	Advance	Slam	26	F	7.27	20.25
3314V	2	Advance	Slam	30	F	8.03	19.76
3357V	2	Advance	Slam	36	F	8.08	21.66
3341Va	2	Advance	Slam	49	F	7.65	21.82
3308V	2	Advance	Slam	25	M	7.85	21.89
3332V	2	Advance	Slam	30	M	7.32	19.86
3379V	2	Advance	Slam	39	M	7.03	19.75
3324V	2	Advance	Slam	44	M	7.87	19.67
3414Va	2	Advance	Slam	55	M	7.23	23.43
3467Vab	2	Delay	Standard	30	F	8.25	22.81
3506V	2	Delay	Standard	30	F	6.88	18.87
3545V	2	Delay	Standard	44	F	5.50	18.97
3431V	2	Delay	Standard	27	M	9.97	22.78
3447V	2	Delay	Standard	36	M	6.00	21.02
3443V	2	Delay	Standard	53	M	6.00	18.66
3437V	2	Delay	Optimized	28	F	7.42	20.45
3469V	2	Delay	Optimized	37	F	7.97	23.08
3448V	2	Delay	Optimized	55	F	6.17	22.22
3358V	2	Delay	Optimized	29	M	7.85	21.34
3439V	2	Delay	Optimized	35	M	7.98	20.60
3416V	2	Delay	Optimized	48	M	7.02	19.98
Study 1		Mean ± SEM / N (% male)		23.38 ± 0.57	16 (100)	7.75 ± 0.18	22.44 ± 0.35
Study 2: All participants		Mean ± SEM / N (% male)		37.40 ± 1.71	15 (50)	7.17 ± 0.20	20.66 ± 0.28
Advance Gradual		Mean ± SEM / N (% male)		37.33 ± 2.77	4 (44)	6.63 ± 0.41	20.02 ± 0.55
Advance Slam		Mean ± SEM / N (% male)		37.11 ± 3.51	5 (56)	7.59 ± 0.13	20.90 ± 0.45
Delay Standard Light		Mean ± SEM / N (% male)		36.67 ± 4.10	3 (50)	7.10 ± 0.70	20.52 ± 0.80
Delay Optimized Light		Mean ± SEM / N (% male)		38.67 ± 4.39	3 (50)	7.40 ± 0.29	21.28 ± 0.48

^†Wake and DLMO times are reported in decimal clock time. For Study 2, the average of all participants is reported along with the average for each phase resetting condition separately. DLMO in Study 1 represents DLMO calculated during CR, and in Study 2 calculated during pre-shift (baseline). DLMO = dim light melatonin onset; M= male, F= female.

^aParticipants removed from analysis due to IV failure, or unexpected phase-shifting response (3414 V).

^bDLMO calculated using saliva.

Study protocols

Study 1 – Endogenous circadian regulation of metabolic biomarkers.

Details of the protocol have been previously reported (27–29). The lipid data reported herein have been reported previously in a separate analysis to examine meal timing effects on absolute lipid levels (27). Participants were studied in an environment free of time cues (no access to windows, clocks, watches, live television, radio, internet, telephones and newspapers and continually monitored by study staff trained not to reveal time-of-day information). The study schedule consisted of i) 3 baseline days with 8 h:16 h sleep:wake schedule, ii) a 40-h constant routine (CR) followed by 8 h of recovery sleep, and iii) a 16-h light exposure day, followed by an 8-h sleep opportunity and then discharge (Figure 1). Results reported in the current study are from data collected until the end of the CR. Data on metabolic biomarkers were not collected beyond the CR. During baseline days, participants were minimally ambulatory around their ~23 sq m suite and not permitted to exercise. Participants received their daily nutritional intake in 3 meals (breakfast, lunch, dinner; each ~30% of total daily calories) and a snack (10% of daily calories) with macronutrient content 60% carbohydrate, 25% fat, 15% protein (150 mEq Na+/100 mEq K+ (± 20%); basal energy expenditure x 1.4; 2,500 mL fluids / 24h). During the CR, participants remained awake in constant semi-recumbent posture in dim light with daily nutrition intake divided into hourly portions (macronutrients: 60% carbohydrate, 25% fat, 15% protein; 150 mEq Na+/100 mEq K+ (± 20%) controlled nutrient, isocaloric [basal energy expenditure x 1.3] diet; 2,500 mL fluids/ 24h). Meals were ‘must finish’, and daily caloric requirements were adjusted for sex, weight, and age using the Mifflin-St Jeor equation.

Figure 1. Study protocols.

White bars represent wake episodes in ~25 μW/cm² (~90 lux) white fluorescent light, black bars represent scheduled sleep episodes with lights off (0 μW/cm², 0 lux), gray bars represent wake episodes in dim fluorescent light < 0.46 μW/cm² (<3 lux) not under CR conditions, and bars with a hatched diagonal pattern represent the same dim fluorescent light under CR. The red lines represent the timing of blood sampling in each protocol. Large black circles indicate main meals (breakfast, lunch, dinner) and smaller circles are snacks. Study days in the 7-day protocol (Study 1, A) not included in the current analyses are blank. Study 2 shiftwork simulations include two advance (slam shift: B; gradual shift: C) and two delay shifts (standard light: D; optimized light: E) schedules. Blue bars represent the 6- and 8-h blue-enriched LED light (6500K) exposures in the advance (B, C) and optimized light delay (E) schedules, respectively. The yellow bars represent the 8-h control LED light (4500K) exposure in the standard light delay schedule (D). Red bars represent the 2- and 0.5-h blue-depleted LED light (2700K) pre-sleep exposures in the advance (B, C) and delay (E) schedules, respectively. Pale yellow bars represent the 0.5-h dimmer ambient light exposure (4500 K) in the standard light delay schedule (D). In the slam advance schedule (B), the start of the light exposure was advanced by 1.6 hours each day. In the gradual advance schedule (C), the light exposure, meals and sleep were advanced by 1.6 hours each day. ADM= admit; D/C= discharge.

During the first 2.5 baseline days, maximum ambient fluorescent light during scheduled wake was 48 μW/cm² (~190 lux) when measured in the horizontal plane at a height of 187 cm and 25 μW/cm² (~90 lux) when measured in the vertical plane (137 cm). Halfway through day 3, maximum ambient light was decreased to ~0.5 μW/cm² (~1.5 lux) when measured in the vertical plane. This light level was maintained for the remainder of the study except during scheduled sleep where ambient light was switched off (0 lux).

Study 2 – Circadian phase resetting of metabolic biomarkers.

The protocols were designed to test the efficacy of different lighting interventions (see Supplemental Material for details) on phase-advance and phase-delay shifts of the central clock, as measured by shifts in the timing of dim light melatonin onset (DLMO). The two phase advance protocols compared the effect of two different shift schedules – either a gradual shift condition (n=9) where the sleep episode was advanced by 1.6 h per day over 5 days for a total shift of 8 h or an abrupt (or ‘slam’) shift condition (n=9) where the sleep episode was shifted by 8 hours in the first day and maintaining at that time for 5 days] but under the same experimental lighting intervention (Figure 1B–C). In contrast, the two phase delay protocols compared the effect of two different lighting interventions – either standard lighting (n=6) or optimized lighting (n=6) - but both having the same ‘slam’ shift of 8 h of the sleep schedule (Figure 1D–E). Participants were studied in the same facility as Study 1, in an environment free of time cues. The study schedule for all conditions consisted of i) a baseline day with 8 h:16 h sleep:wake schedule, followed by 5 days where the sleep episode was shifted relative to the self-selected sleep/wake scheduled maintained prior to the start of the inpatient phase of the study, then a 30-h CR and an 8-h recovery sleep opportunity and then discharge. The 30-h CR, and nutritional intake and activity on CR and non-CR days was conducted as described above for Study 1. During scheduled wake on days 1–6 of the study when the experimental light exposure did not occur, ambient light conditions were the same as described for the baseline days for Study 1.

Measurement of clinical metabolic biomarkers

Whole blood was collected from an indwelling IV cannula kept patent with a heparinized saline infusion (5 UI heparin/mL 0.45 NaCl infused at 40mL/h). Plasma samples were assayed for melatonin, cholesterol, triglyceride, FFA, HDL, albumin, total protein, ALT, AST, ALP and GGT. Four-hourly blood samples were assayed during baseline and CR in Study 1 and 2-hourly samples were assayed between 8–36 hours awake on admission day and during CR in Study 2. Clinical metabolic biomarker assays were conducted by a CLIA certified laboratory blind to the experimental conditions (LabCorp Raritan, Raritan, NJ).

Melatonin circadian phase assessment

Melatonin concentration was determined using RIA (Kennaway G280 antiserum; Specialty Assay Research Core Laboratory, Brigham and Women’s Hospital, Boston, MA). Intra- and inter-assay coefficients of variation (%CV) were 10% at 1.9 pg/mL and 7.2% at 21.9 pg/mL, and 12.65% at 3.06 pg/mL and 12.12% at 22.36 pg/mL, respectively. For Study 1 and 2, Dim Light Melatonin Onset (DLMO) was calculated using linear interpolation to determine the time at which melatonin levels crossed a threshold of 10 pg/mL.

Data analysis

Data are expressed as mean ± SEM unless otherwise specified. Data from seven participants (3057V, 30E7V, 30H4V, 30H6V, 3109V, 3126V, 3174V) were not available for the baseline portion in Study 1 because the IV was inserted the night before CR and only two timepoints were available per participant, which is insufficient for rhythmometric analyses. As sensitivity analyses, rhythmicity under CR conditions were tested excluding the data from these seven participants (Supplemental Table 1). In Study 2, data from one participant (3414V) in the slam-advance group was excluded from the analyses as they exhibited a 6.7-h phase delay (or 17.3-h phase advance) in DLMO and total cholesterol. Three participants (28E6V, 3341V, 3351V) had an IV failure during the baseline sleep episode, and another (3467V) during the CR, which precluded rhythmometric analyses to determine initial or final phases, respectively and were excluded from further analyses. Therefore, the final analysis included data from 25 out of 30 participants studied.

Cosinor analysis was performed to assess 24-h rhythms in the metabolic biomarkers at the group- and individual-level. All data were z-transformed for each individual prior to analysis. In Study 1 and 2, cosinor regression was applied to data from baseline and CR conditions separately. Group-level data were analysed using repeated-measures cosinor regression accounting for the correlated nature of the data from the same individual across the time series. The cosinor regression included a 24-h fundamental harmonic and a linear component: $y=Acos\left\{\left(\frac{2\pi }{\tau }\right)(x-\varphi )\right\}+(mx+D)$, where $\mathrm{A}$ = amplitude, $\mathrm{\tau }$ = period (24 hours), $\mathrm{\varphi }$ = acrophase, m= slope of linear term, and D = vertical intercept of linear term (29) [initial parameters: $\mathrm{A}=0.5;\mathrm{\varphi }=12;\mathrm{m}=0;\mathrm{D}=0$; convergence criteria (GCONV=1E-8) were satisfied for all fits]. The linear component was included in the model to estimate the influence of secular changes associated with the CR protocol (e.g., effects of accumulating time awake, repetitive meals, etc.). The regression was considered significant if the amplitude was significantly different from 0, with a two-tailed type I error threshold of 0.05 (one-tailed for determining significance of individual participant rhythms reported in Table 2). Where a significant nadir was detected by the regression model, the acrophase time was calculated as the peak 12-hours later. Regression analyses were conducted in SAS 9.4. (SAS Inc., Cary, NC, USA).

Table 2.

Acrophase, amplitude and proportion of individuals with significant rhythms under baseline and constant routine conditions in Study 1^†

Class	Parameter	Baseline			Constant routine
		Phase	Amp	% (n=9)	Phase	Amp	% (n=16)
Hepatic proteins	Albumina	16.43 ± 0.53	0.80 ± 0.11	89 %	14.47 ± 0.44	0.69 ± 0.09	81 %
	Proteina	16.54 ± 0.49	0.83 ± 0.10	89 %	14.95 ± 0.46	0.62 ± 0.08	81 %
Hepatic enzymes	ALP	16.37 ± 0.56	0.77 ± 0.11	89 %	15.08 ± 0.58	0.57 ± 0.10	75 %
	GGT	14.79 ± 0.70	0.67 ± 0.12	67 %	15.72 ± 0.46	0.74 ± 0.10	63 %
	ALTc	15.65 ± 3.88	0.14 ± 0.14	44 %	13.41 ± 1.39	0.28 ± 0.11	31 %
	AST	15.81 ± 1.54	0.36 ± 0.14	67 %	15.33 ± 1.33	0.29 ± 0.11	25 %
Lipids	Cholesterolb	15.49 ± 0.36	0.95 ± 0.08	89 %	15.43 ± 0.57	0.63 ± 0.10	94 %
	HDLb	15.28 ± 0.88	0.52 ± 0.11	67 %	15.48 ± 0.36	0.89 ± 0.09	94 %
	LDLb	12.57 ± 1.29	0.40 ± 0.14	44 %	15.39 ± 0.42	0.81 ± 0.10	94 %
	VLDLa	16.93 ± 0.66	0.75 ± 0.12	78 %	03.28 ± 0.46	0.72 ± 0.09	69 %
	Triglyceridea	16.93 ± 0.66	0.74 ± 0.12	78 %	03.37 ± 0.47	0.71 ± 0.09	69 %
	FFAc	02.37 ± 1.77	0.31 ± 0.14	22 %	07.68 ± 1.87	0.23 ± 0.10	56 %

^†Mean ± SE of estimated acrophase and amplitude derived from non-linear regression analyses of data under baseline and CR conditions. All acrophases are reported in relative decimal clock time referenced from the group-average scheduled wake time. All amplitude values are reported in standardised z-scored units. Phase= acrophase; Amp= amplitude; % = percentage of individuals with significant rhythms; ALP= alkaline phosphatase; GGT= gamma-glutamyl transferase; ALT= alanine aminotransferase; AST= aspartate aminotransferase; HDL= high-density lipoprotein; LDL= low-density lipoprotein; FFA= free fatty acid; VLDL= very low-density lipoprotein.

^aSignificant difference in phase between baseline and CR.

^bSignificant difference in amplitude between baseline and CR.

^cSignificant rhythms under CR only.

Statistical contrasts included population-level comparisons of 1) baseline versus CR amplitude and acrophase estimates in Study 1 for all metabolic biomarkers, and 2) pre-shift (baseline) versus post-shift (CR) acrophase in Study 2 for cholesterol and melatonin. Exploratory analysis included population-level comparisons of acrophase between CR conditions in Study 1 and post-shift CR conditions in Study 2. Acrophase and amplitude estimates of population-level rhythms in the metabolic biomarkers were considered significantly different (Type I error probability of 0.05) if the model-estimated 83.4% confidence intervals did not overlap (30). For melatonin, pre- and post-shift population mean DLMOs were considered significantly different if their respective 83.4% CI estimates did not overlap. Phase shifts were calculated as the difference between pre- and post-shift acrophases estimated at the population level for the metabolic biomarkers or the population mean DLMO for melatonin.

Results

Endogenous circadian rhythms in metabolic biomarkers

Endogenous circadian rhythms were observed under CR conditions for group analysis of all measures [albumin, total protein, ALP, GGT, total cholesterol, HDL, LDL, VLDL and triglyceride (p<0.001); ALT (p=0.019), AST (p=0.020) and FFA (p=0.043). All had acrophases during the afternoon and early evening (~15:00–17:00 h), except VLDL, triglyceride and FFA which peaked during the night (Figure 2, Table 2).

Figure 2. Circadian rhythms in metabolic parameters under baseline conditions with 8:16 sleep/wake and under constant routine conditions.

Group-mean ± SEM of z-scored data and cosinor regression fits ± SE (—) for all participants under baseline (●) and constant routine (CR; ○) conditions. Baseline data are re-plotted (grey line) during the CR for comparative purposes. The study protocol is shown at the top of Figures A and G, where white bars represent wake, black bars represent sleep, the grey bar represents dim light and the bar with the hatched diagonal pattern represents the constant routine (CR). The left panel shows the hepatic proteins and enzymes albumin (A), total protein (B), ALP (C), GGT (D), ALT (E), and AST (F), and the right panel shows the lipids cholesterol (G), HDL (H), LDL (I), VLDL (J) and triglyceride (K), FFA (L). Corresponding clock times are reported relative to scheduled wake. Time = 0 relative to scheduled wake was defined as 0730h based on the group mean wake time (mean ± SD: 0745 ± 0010h) for illustrative purposes. Solid bars represent group mean scheduled sleep times during baseline and dashed-bars represent where group mean scheduled sleep would have occurred during CR. Dashed lines (– – –) denote non-significant regression fits.

When the dataset was restricted to the 9 individuals who had both baseline and CR data, all metabolic parameters that demonstrated significant circadian rhythms in analyses of CR data in all 16 participants remained significant, except for AST and FFA (Supplementary Table 1). The phase and amplitude of rhythms also remained similar in the restricted dataset (Supplementary Table 1).

In analyses of individual-level data (Supplemental Figure 1), almost all individuals (15/16) exhibited significant rhythms in cholesterol, HDL, and LDL, and 13/16 participants had rhythms in albumin and total protein. VLDL and triglyceride exhibited significant rhythms in approximately two thirds of participants (11/15), whereas FFA (9/16) and GGT (10/16) were significant in fewer than two thirds but more than half of the participants. Less than half of the participants had rhythms in ALT and AST (Table 2).

Endogenous circadian rhythms in some metabolic biomarkers are masked by daily schedules with standard sleep/wake and meals

Under baseline conditions, there were significant 24-hour rhythms observed in albumin, total protein, ALP, AST, GGT, VLDL, triglyceride (p<0.001 for all), HDL (p=0.002) and LDL (p=0.023) (Figure 2, Table 2). ALT and FFA did not show significant rhythms under baseline. All the measures peaked between afternoon and early evening, except FFA which peaked in the morning (Table 2).

At the individual level (Supplemental Figure 1), significant rhythms were detected in at least three quarters of the participants in total cholesterol (8/9), VLDL (7/9), triglyceride (7/9), albumin (8/9), total protein (8/9), and ALP (8/9), and in two thirds of the participants (6/9) in AST, GGT and HDL. ALT, FFA and LDL were significant in less than half of the participants (Table 2).

In parameters that had significant 24-h rhythms under both baseline and CR conditions, the daytime peaks observed under CR were broadly replicated under baseline conditions, with cholesterol showing the least difference in peak timing between conditions (<3 minute difference; Table 2). There were small but significant differences in the daytime peak between conditions in albumin (1 h 57 min), total protein (1 h 35 min) and, LDL (2 h 49 min). There was a near-reversal of the peak time in triglyceride (10 h 26 min) and VLDL (10 h 21 min), from a night-time peak under CR to a daytime peak under baseline conditions. Furthermore, the amplitude of the LDL and HDL rhythms were significantly larger under CR compared to baseline, whereas the amplitude of the cholesterol rhythm was higher (+33%) under baseline compared to CR (Table 2). Amplitude did not differ significantly between CR and baseline for any of the other metabolic biomarkers.

Circadian rhythm in metabolic biomarkers are reset by simulated shiftwork

In Study 2, the cholesterol rhythm was used as the primary outcome to test whether circadian rhythms in biomarkers of metabolic function can be reset. The primary analysis was restricted to cholesterol because in Study 2, phase resetting was determined by a change in group-level acrophase between the pre-shift baseline (data were collected under ambulatory conditions) and post-shift (CR) conditions, similar to the group level analysis of the shift in DLMO. The other biomarkers, as shown in Study 1, exhibited altered phases between baseline and CR and were therefore not stable enough between conditions to use the group method. FFA and ALT were not significant under both baseline and CR; albumin, total protein, LDL, VLDL and triglyceride had significant differences in acrophase between baseline and CR; GGT was not assayed in Study 2; AST and HDL did not exhibit significant rhythms under all conditions in Study 2; ALP was excluded given the difference in phase between baseline and CR, although not significantly different, was >50% of the 2-hourly sampling frequency used in Study 2. As exploratory between-group analyses, however, we compared the post-shift CR acrophases of the metabolic biomarkers from Study 2 to the CR acrophases in Study 1 (no shift CR) (see Supplemental Material). Individual profiles for all outcomes are shown in Supplemental Figure 2.

The acrophase of cholesterol and the DLMO for melatonin pre- and post-shift are reported in Table 3 and illustrated in Figure 3. The acrophase of the cholesterol rhythm was significantly earlier following the 8-h advance slam shift as compared to baseline, indicating a phase advance of +5.15 h (p<0.05). In contrast, the acrophase of cholesterol was not significantly different following the 8-h advance gradual shift as compared to baseline. The acrophase of cholesterol was significantly delayed in both 8-hour slam delay protocols (p<0.05), indicating phase delays of −3.71 h and −4.24 h in the optimized and standard light conditions, respectively (Figure 3). By comparison, DLMO was significantly advanced in both advance-shift protocols and significantly delayed in both two delay-shift protocols (all, p<0.05). The magnitude of the phase advance in DLMO was the same in both advance-shift protocols (slam: 3.13 h; gradual: 2.84 h) but different in the delay-shift protocols (optimized light −6.62 h; standard light: −4.65 h).

Table 3.

Pre- and post-shift acrophase and phase shift magnitude of cholesterol and melatonin in the four simulated shiftwork protocols^†

Protocol		Total Cholesterol			Melatonin (DLMO)
		Pre-shift phase	Post-shift phase	Phase shift	Pre-shift phase	Post-shift phase	Phase shift
Advance	Gradual	15.85 ± 0.50	13.84 ± 1.04	+2.01	20.20 ± 0.60	17.36 ± 0.70*	+2.84
	Slam	16.15 ± 0.76	11.00 ± 0.91*	+5.15	20.40 ± 0.36	17.27 ± 0.37*	+3.13
Delay	Standard	14.23 ± 1.16	18.47 ± 1.24*	−4.24	20.06 ± 0.80	00.71 ± 1.08*	−4.65
	Optimized	16.29 ± 0.55	20.00 ± 1.04*	−3.71	21.28 ± 0.48	03.90 ± 0.79*	−6.62

^†Estimated acrophase ± SE derived from non-linear regression analysis of total cholesterol data and mean ± SE of individual participant DLMOs pre- and post-shift in the four simulated shiftwork protocols. Phase shifts were calculated as pre- minus post-shift phase. Positive and negative values indicate phase advances and delays, respectively. Phases and phase shifts are reported in decimal clock time.

^*denotes significant phase shifts (p<0.05).

Figure 3. Phase resetting responses of total cholesterol and melatonin.

The acrophase ± 83.4% CI of the rhythms in total cholesterol (○) and the phase of dim light melatonin onset (DLMO) (■) pre- and post-phase shift for each the advance and delay simulated shiftwork protocols. Non-overlapping confidence intervals (83.4%) indicate significant phase shifts. OPT = optimized lighting; STD = standard lighting.

Discussion

In the current study, we found evidence supporting endogenous circadian regulation of clinical biomarkers of metabolic health, including circulating lipids, lipoproteins, proteins and hepatic enzymes. Many of these outcomes were also rhythmic under baseline conditions with typical sleep/wake, activity and meal timing, but the endogenous rhythm was masked by these behaviors, changing the profile of the observed rhythm. In addition, in a series of simulated shiftwork protocols, we demonstrated that these endogenous rhythms, and particularly cholesterol, can be reset both in the advance and delay directions. Taken together, our results show that these metabolic biomarkers demonstrate two canonical properties of a circadian rhythm – they persist in the absence of external time cues and can be reset to a different phase. The specific external time cues that reset these rhythms (e.g., light, sleep-wake, meals) remain to be determined, however.

The timing of the 24-h rhythms in cholesterol, LDL, HDL, triglyceride, albumin, total protein, AST and ALP under baseline conditions were largely consistent with previous reports, which also assessed 24-h rhythms in the presence of 24-hour sleep/wake and meal schedules (4, 6–8). Furthermore, the endogenous timing of metabolic biomarker rhythms in our study was also consistent with those from prior reports that also employed constant routine conditions. For example, Morgan et al. showed that the rhythm in triglyceride peaked at night (~0400h) during a 26-hour CR (17) and a more recent lipidomics study (11), with a 40-hour CR, showed that many triglyceride species peaked predominantly during the night time/early morning hours (~0300 – 0700h). Additionally, Chua et al. (11) reported that two cholesterol species peaked at ~1500 and 1900h (11), similar to the peak time of total cholesterol found in our study under CR conditions.

The comparison of rhythmic profiles under both baseline and CR conditions in our study showed evidence of substantial masking effects on both timing and amplitude for some of the biomarkers (i.e., a change in the timing and/or amplitude of the endogenously regulated rhythm caused directly by an environmental or behavioral factor, such as light or sleep) (9). This is most dramatically illustrated by triglycerides and VLDL; both exhibited a significant nocturnal peak under CR conditions (11, 17) that was reversed to a signficant daytime peak under baseline conditions. These results underscore the significant impact that daily activities and periodic events such as sleep/wake, posture, light/dark schedules and meals can have on the expression of the endogenous circadian rhythms of triglyceride and VLDL and are in line with previous reports that external factors such as meals can impact the rhythms of these biomarkers (31, 32). Circulating proteins and LDL also showed smaller changes in phase between the two conditions, up to several hours, suggesting a more modest masking effect of environmental factors on their timing. The variability in the extent to which the timing and amplitude of the circadian rhythms of these biomarkers are masked by behavioral and environmental factors, coupled with their differential phase shifts, suggests that shiftwork will cause considerable internal circadian misalignment between multiple metabolic endpoints.

By first establishing the high stability of the circadian rhythm of cholesterol between baseline and CR conditions, we could examine whether the rhythm could be reset in a simulated shiftwork protocol, which measured initial phase under baseline conditions and final phase under CR. We assessed the phase resetting response to four simulated shiftwork protocols with either an 8-hour advance or 8-hour delay, and compared it to that for plasma melatonin, a gold-standard marker of the central circadian clock. We found significant phase resetting of the cholesterol rhythm in three of the four phase shifting protocols. In the two phase-delay protocols, where the non-photic time cues (i.e., sleep and meal timing) were identical, cholesterol shifted a similar amount in each, by about 4 hours (3.7 h and 4.2 h; Figure 3). Melatonin delayed to a greater degree (4.7 h and 6.6 h) with increased exposure to short-wavelength light in the experimental light exposures, as expected (33). In the two advance protocols, the timing of non-photic time cues differed while the blue-enriched light intervention remained the same. Under these circumstances, cholesterol advanced by 5 h when subjected to an abrupt ‘slam’ shift in the schedule but did not shift significantly (~ 2 h) when the schedule was gradually advanced. Melatonin advanced by approximately the same amount in both protocols (3.1 h and 2.8 h), most likely due to the similarity in the lighting interventions. These data suggest that the cholesterol rhythm is more sensitive to changes in the timing of non-photic time cues rather than light, opposite to melatonin (34).

The speculation that the total cholesterol and melatonin rhythms exhibit different phase resetting responses is consistent with previous work. For example, following 3 days on a simulated jet-lag schedule where sleep, meal and light timing were shifted by 12-hours, the diurnal rhythm in in vivo cholesterol fractional synthetic rate was delayed 10 hours, whereas the central clock marker cortisol was delayed by only 5.6 hours (35). A more recent study using a targeted metabolomics approach compared rhythms in plasma metabolites after a 12-hour shift in sleep-wake schedule with a non-shifted group in a between-subject design (25), similar to our exploratory anlayses of the other metabolic biomarkers examined in the current study. Rhythms were assessed after the phase shift using a 24-hour CR. After 3 days on the shifted schedule, which induced a ~2 h delay in DLMO, the acrophase of most metabolites (24 of 27) which were consistently rhythmic in individuals both on the night shift (12-hour shift) and the day shift (control) schedule, were significantly shifted, and in many cases were almost completely reversed (i.e., shifted by 12 hours). Similar results were also seen in a within-subject simulated shiftwork study where metabolites showed an average phase delay of 8.8 h in response to a 10-h delay in the sleep/wake schedule (24). While consistent with these previous studies, our findings extend these data and show evidence of phase resetting of the endogenous circadian rhythm of cholesterol and other commonly used clinical biomarkers of metabolic health. Specifically, having confirmed that total cholesterol exhibits an endogenously generated circadian rhythm in Study 1, and that the timing is similar under baseline conditions, and measuring cholesterol in a CR following the phase shift, our study shows that the change in the timing in cholesterol is not simply a masking effect but rather a phase shift of the endogenous rhythm. Furthermore, our study is the first to examine phase resetting of metabolic biomarkers in response to both phase advances and phase delays, which is important to establish that these endogenous rhythms are likely entrainable.

While the role of light as the principal phase-resetting stimulus for the central clock has been clearly established (36, 37), the role of light and non-photic stimuli on peripheral metabolic rhythms remains unclear. Based on our finding that total cholesterol and melatonin phase resetting responses differed by protocol, we hypothesize that this differential resetting may have been driven by meal timing. For example, in the two advance protocols, the experimental lighting was given at the same time but the meal timing was different and under these conditions we observed that the melatonin phase shifts were the same (i.e., consistent with the same light conditions), but the cholesterol shifts were different (i.e., consistent with differences in meal timing). Conversely, in the two delay conditions, the light stimulus was different, but meals were the same, and here we observed the phase shifts were different in melatonin but the same in cholesterol. Although our study precludes isolating the principal stimuli inducing the phase shifts in the measured metabolic rhythms, based on previous reports supporting the role of meal timing as an entrainment cue for peripheral clocks in rodents (23, 38) and studies in humans showing that delaying meal times by 5–6 hours phase delays the diurnal rhythm in cholestrol synthesis (39) and the circadian rhythm in glucose (22), we speculate that the phase shifting response in the metabolic rhythms may be associated with the shift in meal timing in our simulated shiftwork protocols. Future work with larger sample sizes, greater temporal separation of lighting and meals, and with varied meal/nutritional content (40) is needed to determine the principal time cue for entraining these metabolic rhythms and the relative contribution of light and meal timing to the phase shifts observed.

These studies have several limitations. First, the lack of a CR at the start of the simulated shiftwork protocol to assess pre-shift initial phase in Study 2 precluded formal analysis of the phase shifting response for other parameters; future studies with a CR both before and after the phase shift are needed to assess these in a within-subjects design. Second, it should be noted that we did not collect full pre-shift profiles for melatonin and therefore compared the phase shifts of melatonin based on the change in the timing of DLMO (calculated from a fixed threshold method) with 24-hour profile of cholesterol (with the peak time calculated from non-linear regression). DLMO is a well-established marker, however, and we would not expect the methodological difference to have a major effect on the ability to compare the magnitude and direction of the phase shifts (41). Additionally, although our findings suggest circadian regulation of some hepatic enzymes, these enzymes are not produced exclusively by the liver and therefore the source of rhythmicity in these and the other metabolic rhythms examined cannot be discerned from this study. Examination of changes in rhythmicity in disease states affecting specific organs will be an important step toward elucidating the source of rhythmicity in these parameters. Further research is also required to determine how pathological states (e.g., liver disease, dyslipidemia) and non-pathological factors such as diet, age and sex may also affect the circadian regulation of the biomarkers studied herein.

We identified robust, endogenous circadian rhythms in several lipid and hepatic markers commonly assessed in clinical practice to evaluate metabolic health. We further showed that the circadian rhythm in cholesterol could be reset by simulated shiftwork and may be regulated differently than the centrally-controlled rhythm in melatonin. There are two implications of this finding. The first is directly applicable to shift workers in that we have shown that shiftwork changes the timing of metabolic rhythms, which i) changes their timing relative to the central clock, as measured by melatonin; and ii) also changes the timing of these rhythms relative to food intake, as none of the rhythms measured shifted the full 8 hours scheduled. While the long-term health consequences of the multiple levels of circadian misalignment between the central clock, metabolic rhythms and meal timing reported herein are currently unknown, evidence of this type of desynchrony may contribute to future work to understand how circadian or environmental factors affect metabolic markers that increase the risk of cardiometabolic diseases in shift workers (21). Given the pervasive exposure to misalignment between meal schedules and light-dark exposure in the general population, however, the implications can also be extended beyond shift workers. The second implication is that these data provide additional support suggesting that central and peripheral clocks in humans can be reset at different rates in humans. Understanding the source of endogenous rhythmicity and how best to reset those rhythms is vital to understand how to address not only misalignment between centrally controlled rhythms and external time when treating circadian rhythm disorders, but also misalignment between internal phase relationships which might turn out to be as important, or even more so, in reducing disease burden.