Circadian Protein Expression Patterns in Healthy Young Adults

Abstract

Objectives:

To explore how the blood plasma proteome fluctuates across the 24-hour day and identify a subset of proteins that show endogenous circadian rhythmicity.

Methods:

Plasma samples from 17 healthy adults were collected hourly under controlled conditions designed to unmask endogenous circadian rhythmicity; in a subset of 8 participants, we also collected samples across a day on a typical sleep-wake schedule. A total of 6,916 proteins were analyzed from each sample using the SomaScan aptamer-based multiplexed platform. We used differential rhythmicity analysis based on a cosinor model with mixed effects to identify a subset of proteins that showed circadian rhythmicity in their abundance.

Results:

1,063 (15%) proteins exhibited significant daily rhythmicity. Of those, 431(6.2%) proteins displayed consistent endogenous circadian rhythms on both a sleep-wake schedule and under controlled conditions: it included both known and novel proteins. When models were fitted with two harmonics, an additional 259 (3.7%) proteins exhibited significant endogenous circadian rhythmicity, indicating that some rhythmic proteins cannot be solely captured by a simple sinusoidal model. Overall, we found that the largest number of proteins had their peak levels in the late afternoon/evening, with another smaller group peaking in the early morning.

Conclusions:

This study reveals that hundreds of plasma proteins exhibit endogenous circadian rhythmicity in humans. Future analyses will likely reveal novel physiological pathways regulated by circadian clocks and pave the way for improved diagnosis and treatment for patients with circadian disorders and other pathologies. It will also advance efforts to include knowledge about time-of-day, thereby incorporating circadian medicine into personalized medicine.

1. Introduction

Circadian rhythms are powerful regulators of physiology.¹ In mammals, these are driven by a central pacemaker located in the suprachiasmatic nucleus (SCN), which receives direct input from the visual system,² allowing synchronization to the external 24-hour day. The central clock, in turn, is connected through neuronal and endocrine networks to other tissues which have their own endogenous circadian clocks, allowing each system to coordinate temporal programs optimal to the organism.^3–5 Although much is known regarding how each cell generates its own circadian rhythm through a process involving transcriptional/translational feedback loops,⁶ the synchronization between peripheral and central clocks is poorly understood, and there is evidence these systems can become dissociated. As an example, in rodents and humans, restricting food availability can shift the phase of metabolic processes to times when food is available, with deleterious metabolic consequences.^7,8 For this reason, there is interest in understanding and describing how these processes are organized.

Traditionally, timing (or phase) of the human central circadian clock is measured using the Dim Light Melatonin Onset (DLMO),^9,10 because secretion of melatonin by the pineal gland is controlled via a multi-synaptic pathway from the SCN through the superior cervical ganglion. Melatonin is at very low or non-detectable levels in blood and saliva during the daytime and begins secretion in the evening, reaching peak levels in the middle of the subjective night. While the rhythm of melatonin has served a useful purpose for assessing the phase of the central clock for research purposes, it does have several drawbacks. First, melatonin is acutely suppressed by light, and assessment of melatonin in saliva or blood requires the participant to remain in dim light throughout the sampling segment. Second, measurement from saliva requires an awake participant, while measurement from blood requires either multiple needle sticks (again waking the participant) or a special blood sampling system that allows sampling without waking the participant. These issues, together with the duration of melatonin sampling required to obtain a phase estimate^11,12 have severely limited the assessment of melatonin timing in sleep disorders patients. Furthermore, while melatonin provides useful information about central circadian phase, it does not provide any indication of the phase of peripheral oscillators, nor whether peripheral rhythms are synchronized to the central clock or to each other.

For these reasons, researchers have sought alternatives for assessing endogenous circadian phase in humans.¹⁷ In animals, tissues can be collected and analyzed for metabolites, gene expression, or protein content to assess the phase of both the central and peripheral oscillators.⁴ Such studies have led to a wealth of information on how organ clocks are organized, with the limitations that most animal models are nocturnal, inbred, and live in artificial lighting and environmental conditions. Animal models have shown that by measuring multiple parameters at a given time point and considering a stable relationship between phases of various processes, it is possible to derive the central circadian phase with a single time point, a concept called a “time stamp”.¹⁸ Subsequently, attempts have been made at establishing time stamp measures that could replace melatonin phase in humans, but in most cases to achieve accuracy within 1–2 hours of DLMO phase at least two samples collected hours apart have been needed.¹⁹ One exception is BodyTime, a measure of circadian phase from gene expression in isolated monocytes derived from a single blood sample, which has shown good performance in studies against DLMO.²¹ Although powerful, BodyTime measures the circadian clock in a specific immune tissue that could be perturbed in conditions such as infections. Nonetheless, it has the advantage of measuring core circadian genes themselves and thus has heuristic value.

In the present work, we hypothesized that measuring the circulating plasma proteome might lead to future development of a single sample proteomic circadian phase measure, and, as a first step are describing proteins that have circadian influences. Indeed, it is now possible to measure an increasing number of proteins in smaller and smaller amounts of blood, either using Olink, a DNA bar-coded antibody-based technology, or Somalogic, an aptamer-based technology.²² Although debates are ongoing regarding the merit of Olink versus Somalogic, these are comparable in relation to mass spectrometry.²² The reliability of these technologies is revealed by many genetic effects that have been shown to regulate levels of individual proteins and that are mostly located close to the gene encoding each analyte (Cis-QTL). These signals are often themselves also known as expression QTLs.^23,24 These proteins are secreted or the result of normal tissue shedding and damage and can be both of intra and extracellular origin. Proteomics offers several advantages over metabolomics²⁰ because many metabolites are unstable and of gastrointestinal origin, as well as advantages over transcriptomics,¹⁹ which utilizes whole blood where different cell types show rhythmic fluctuations in number and internal gene expression. Furthermore, proteins are closer to physiology than gene transcripts, as they are the structural and acting component of the human body. Although transcript and protein amounts correlate within tissues,²⁵ there are many exceptions. Another advantage of measuring the plasma proteome is that it is likely to contain many secreted proteins and hormones that are used in inter-organ communications.

As a first step in the process of determining whether proteomics could be used to develop a single timepoint circadian phase measure, we carried out a study to explore whether the blood proteome shows 24-hour rhythmic activity and whether any such rhythmic proteins show endogenous circadian variation. Somalogic currently offers a larger number of analytes (~7,000 proteins) from single small blood samples (200 μL plasma) and is therefore the platform we elected to use.

2. Participants and Methods

2.1. Participants and Protocol

Samples were collected from two studies carried out at the Intensive Physiological Monitoring Unit of the Center for Clinical Investigation (CCI) at Brigham and Women’s Hospital (BWH), part of the Harvard Catalyst Clinical and Translational Science Center.

In the first study, 8 adults (2 men, 6 women) between ages 21 and 35 were recruited for a 5-day study. To be included, participants had to have a self-reported habitual sleep duration between 7 and 9 hours, a body mass index (BMI) between 18 and 29.9, normal sleep quality (Pittsburgh Sleep Quality Index score <5), no daytime sleepiness (Epworth Sleepiness Scale score <10), no depression symptoms (Beck Depression Inventory II score <10), no medications (except for hormonal contraceptives), and no sleep disorders. For at least a week prior, participants followed a regular 8-hour sleep schedule of their own choosing at home, wore a wrist activity monitor, and maintained a daily sleep diary. Participants had a blood sampling catheter inserted on the afternoon of admission. After a 9-hour sleep episode centered at the participant’s sleep average times from the week prior, the participant was awakened for a 15-hour Baseline Day in the laboratory. They were served three regular meals, were allowed to shower in the morning, and spent the day in their study room pursuing sedentary activities such as reading, listening to music, etc. Hourly blood sampling began just after wake time. At the end of the Baseline Day a second 9-hour scheduled sleep episode occurred during which hourly sampling continued. Upon waking the next morning, participants remained in bed for a 39-hour constant routine (CR, see below for description). At the end of the CR the participant had a 12-hour recovery sleep episode during which hourly blood sampling continued. The study ended the following early afternoon. See Figure 1, panel A for a schematic of the study protocol.

Figure 1. Overview of study protocols (upper panel) and example of melatonin profiles from each study condition (lower panel).

A: schematic of study protocols from Study 1 and Study 2. Relative clock times of study events, lighting levels (white=room light, grey=dim light, black=lights out/dark), sleep and wake times, study sampling segments (in Study 1 blue=sampling from the 24-hour sleep-wake day; orange=sampling from the first 24 hours of Constant Routine (CR); purple=sampling from the final 15 hours of CR and recovery night; in Study 2 green=first 24 hours of sampling from CR), and approximate timing of proteomic samples and melatonin samples from each segment. Lower panel: Data from one exemplar participant in Study 1 is shown in B and data from one participant in Study 2 is shown in C; background colors correspond to study sampling segments shown in A. Black dots indicate samples for which both melatonin and proteomics analysis was carried out, grey dots indicate sample times when only melatonin was measured. The 3-harmonic fit to the melatonin data is shown in the grey curve, and the vertical dashed line indicates the time at which the rising melatonin reached 25% of the fundamental harmonic level, denoted as Dim Light Melatonin Onset (DLMO). The arrows with clock time indicate the clock time of DLMO.

In the second study, nine adults (6 men, 3 women) between ages 18 and 25 were recruited for a 4-day study of sleep and brain structure. Inclusion/exclusion criteria were the same as in Study 1, with the exception that they had to have a usual bedtime between 22:00 and 24:00 and a usual wake time between 06:00 and 09:00. After an 8-hour sleep episode, the participant was awakened and began a CR. A blood sampling intravenous catheter was inserted shortly after waketime, and hourly blood sampling began and continued for the next 36 hours. After 36 hours, the CR was ended, the catheter removed, and the participant escorted out of the CCI to the BWH Magnetic Resonance Imaging Research Center where the study continued. The first 24 hours of samples from the CR were used in the proteomics analysis (see below and schematic in Figure 1, panel A).

Overall, the average age of the 17 participants was 23.9 ± 3.9 years (mean ± standard deviation), their average BMI was 22.9 ± 2.0, and they were on average “neither” types based on their Morningness-eveningness scores (55.3 ± 6.5). Sleep times during the week prior to study averaged from 22:31 ± 0:58 to 07:00 ± 0:52.

2.2. Constant Routine

The CR consists of a regimen of enforced semi-recumbent wakefulness (sitting at a ~45° angle in bed) in dim light, with nutritional intake divided into identical hourly snacks.^26,27 Trained staff remain with the participant to ensure compliance and maintain wakefulness; waking electroencephalographic (EEG) data is collected to verify wakefulness. The CR ensures that specimens are collected under conditions of controlled posture, activity level, feeding, and sleep-wake state so that any observed rhythmicity in the data is due to endogenous sources rather than due to periodic behaviors (such as postural changes, sleep-wake state, or eating) or changes in the environment.

2.3. Light Conditions

Lighting was provided by ceiling-mounted fluorescent lamps (T8 or T12 lamps with CCT of 4100K, Philips Lighting Eindhoven, The Netherlands) transmitted through ultraviolet (UV)-shielding ceiling filters (Lexan, GE Plastics, Pittsfield, MA). All lighting was always controlled by the experimenters and participants had no access to any other lighting.

For Study 1, lighting on the admission day was ~0.23 W/m² (~89 lux) at 137 cm from the floor facing the walls and had a maximum of 0.48 W/m² (150 lx) at 187 cm from the floor facing the ceiling anywhere in the room. Throughout Baseline Day and CR, lighting was set to ~0.0087 W/m² (~3.3 lux) at 137 cm from the floor facing the walls and had maximum of 0.048 W/m² (15 lux) at 187 cm from the floor facing the ceiling anywhere in the room. Throughout each of the scheduled sleep episodes, all lighting was turned off (complete darkness).

For Study 2, lighting at admission was ~0.23 W/m² (~89 lux) at 137 cm from the floor facing the walls and had a maximum of 0.48 W/m² (150 lx) at 187 cm from the floor facing the ceiling anywhere in the room. Beginning ~6h before bedtime on the admission day, ambient lighting was reduced to ~0.0087 W/m² (~3.3 lux) at 137 cm from the floor facing the walls and had maximum of 0.048 W/m² (15 lux) at 187 cm from the floor facing the ceiling anywhere in the room. The same light level was used throughout the CR. Throughout the 8-hour scheduled sleep episode, all lighting was turned off.

2.4. Blood Sampling

A CCI research nurse inserted a 20-gauge intravenous catheter into a forearm vein for blood collection from each participant. The catheter was connected to a triple-stopcock manifold (Cobe Laboratories Inc., Lakewood, CO) via an intravenous loop with a 12-foot small-lumen extension cable (Liberty Medical, Inc.) so that blood could be sampled from outside the room while the participant was sleeping. Between samples, a solution of 0.45% saline with 5,000 IU/liter of heparin was infused at a rate of 40 mL/hour to maintain patency.

Blood was collected hourly, and a 2 mL aliquot placed into a 3 mL EDTA vacutainer tube. The tube was inverted 8–10 times to ensure the EDTA was mixed with the blood, and then the tube was centrifuged at 3,000 rpm for 10 minutes at room temperature. A 200 μL aliquot of plasma was placed into a 2 mL microtube and frozen at −80° C. Every other sample was subjected to proteomics analysis in the present study to reduce overall costs.

An additional 1 mL of blood from each sample was put into a separate 3 mL EDTA vacutainer tube. Plasma from that tube was pipetted into a different microtube and frozen at −20°C for melatonin assay. Melatonin assays were performed by Solidphase, Inc. (Portland, ME) using the Bühlmann radioimmunoassay (NovoLytiX GmbH, Witterswil, Switzerland). This assay, based on the Kennaway G280 anti-melatonin antibody,²⁸ has a sensitivity of 0.84 pg/ml, a range of 1–81 pg/ml, an intra-assay precision of 6.7%, and an inter-assay precision of 10.4%.

2.5. Ethical Approval

Protocols were reviewed and approved by the Partners Health Care (now Mass General Brigham) Human Subjects Committee and conducted in accordance with the principles outlined in the Declaration of Helsinki. Each participant gave written informed consent prior to study.

2.6. Protein Quantification

Plasma was assayed using the SomaScan aptamer-based multiplexed platform (SomaLogic Inc., Boulder, CO), which utilizes aptamers and hybridization to quantify proteins from plasma.²⁹ The platform is designed so that protein levels can be measured over a large range of concentrations. It includes both extracellular and intracellular proteins with soluble domains of membrane proteins. SomaScan assays have shown validity and reproducibility as well as stability.³⁰ SomaLogic also conducts data quality controls at the sample and protein level to adjust for variability between and within samples and provides population-based normalized outputs of relative protein expression levels. Detailed information on SomaLogic’s quality control (QC) technique can be found on the manufacturer’s website (https://somalogic.com/technology/). Briefly, QC procedures use pooled matrix-matched samples (e.g., plasma, serum) run in the SomaScan Assay alongside clinical samples to quantify the quality of each assay run by determining the accuracy of the median replicate signal for each SOMAmer reagent compared to the reference. QC check is performed after hybridization normalization, intraplate median signaling normalization, plate scaling, calibration, and adaptive normalization to a reference have been applied. SomaLogic provided two output files, each with different levels of population-based normalization, whereby the present study utilized the most normalized output based on our sample distributions. This study used the SomaScan platform of ~7,000 proteins. After removal of non-human proteins (e.g., mouse), a total of 6,916 proteins were quantified and analyzed. We further excluded samples that were flagged by SomaLogic as potential outliers.

2.7. Statistical Analysis

We utilized the concept of internal time by calculating DLMO for each participant in each study condition. This involved fitting a 3-harmonic curve to the melatonin data, using a periodicity ranging from 24 to 24.3 hours with a step size of 1 minute. Subsequently, we determined the time at which the rising melatonin level reached 25% of the level of the fundamental harmonic. This DLMO marker (DLMO 25%) served as an indicator of the participant’s internal circadian time in each condition (see Figure 1 panels B and C for examples of DLMO timing).^5,10

To identify rhythmic diurnal and circadian proteins, we used a differential rhythmicity analysis³² based on a cosinor model³¹ with mixed effects³³ that accounted for baseline changes between conditions and between participants. We divided the data into four “conditions”. In Study 1 we had a 24-hour Baseline sleep-wake day; the first 24 hours of the CR; and the final 15 hours of the CR plus the first 9h of the recovery sleep. In Study 2 we used the first 24 hours of data from the CR (Figure 1).

Next, we fit cosinor models to estimate acrophases (peak times) and amplitudes of the protein profiles. The equation for a protein is:

$$\begin{gathered} \mathit{\text{log}}(\mathit{\text{protein}})~(1\mid \mathit{\text{subject}})+(1+{\sum }_{i=1}^n\mathit{\text{cos}}\left(\frac{2\pi \times 24}{i}{t}_{\mathit{\text{DLMO}}+}\right)+ \\ \mathit{\text{sin}}\left(\frac{2\pi \times 24}{i}{t}_{\mathit{\text{DLMO}}+}\right)\mid \mathit{\text{condition}}), \end{gathered}$$

where t_DLMO+ represents number of hours after DLMO which we use as internal time, and n is the number of harmonics. We tested the null hypothesis that the sine and cosine terms were equal to zero, allowing us to identify proteins exhibiting daily rhythmicity. Proteins that surpassed a false discovery rate (FDR) rhythm threshold (rℎy_fdr) of 0.01 and met a minimum amplitude requirement (amp_cutoff) of 0.1 in at least one condition were classified as rhythmic (but not necessarily circadian).

Next, we compared variations across the four conditions for proteins identified as rhythmic. We tested the null hypothesis of equal sine and cosine terms for each protein across conditions. If the null hypothesis were rejected under an FDR threshold (com_fdr) of 0.05, it indicated a significant difference in rhythms between conditions for that protein, suggesting a behavioral or environmental source such as posture, sleep-wake state, eating, or light-dark changes causing rhythmicity. Conversely, if the null hypothesis could not be rejected, the protein was considered to have a consistent rhythm between conditions, suggesting an endogenous circadian origin. We adjusted the FDR thresholds (rhy_fdr and com_fdr) and changed the amplitude requirement (amp_cutoff) to strike for a balance between sensitivity and specificity, ensuring identification of circadian proteins while rejecting false positives (Figure 2).

Figure 2. Tuning hyperparameters to balance specificity and sensibility.

We identified rhythmic proteins based on two criteria: the ability to satisfy the false discovery rate (FDR) rhythm threshold (displayed in the left panel) and the amplitude requirement (shown in the middle panel) in at least one condition. The amplitude was determined using the 2-norm of its components. Each hyperparameter was investigated separately while keeping other parameters at default values. Additionally, proteins that showed a rhythmic pattern and did not have a significant difference in the pattern across the different conditions (based on the FDR comparison threshold), were categorized as circadian (shown in the right panel). We merged 1-harmonic (dot curve) and 2-harmonic (dot-dash curve) results by taking the union (full curve). The default parameter values used in our study were rhy_fdr = 0.01, amp_cutoff = 0.1, and com_fdr = 0.05.

To accommodate variations in baseline expression between conditions and participants, we employed linear mixed models. These models were used to estimate and account for individual-specific baseline levels of protein expression, capturing any inherent differences in baseline between conditions and between participants. This adjustment enabled us to discern genuine circadian rhythmicity by focusing on variations attributed to the endogenous circadian system rather than overall differences in protein level between individuals or conditions.

We performed cosinor analysis with both one and two harmonic components to accommodate for complex circadian behaviors that could manifest higher harmonics. This approach enabled us to capture additional variations in the protein expression patterns, thereby enhancing sensitivity.

We used a soft-dtw barycenter technique³⁴ to compute the average circadian expression profiles (Figure 3). A regulation value of 10 was utilized, except for the final 15 hours of CR data and the initial 9 hours of recovery sleep data in Study 1. We used a regulation five times stronger for this last condition to compensate for fewer data points. To assess stability of the barycenter, we implemented a leave-one-participant group-out strategy, providing insights into robustness and reliability in capturing overall patterns across participants.

Figure 3. Expression patterns of selected proteins.

Panel A shows examples of four rhythmic proteins detected using the 2-harmonic approach, with the rightmost protein showing circadian rhythmicity. Panel B showcases the eight most robust circadian proteins. The individual curves are derived through a series of steps, including log-normalization, mean-filtering, and soft-dtw barycenter computation.³⁴ X-axis is the time relative to Dim Light Melatonin Onset (DLMO). Blue: protein pattern from the 24-hour sleep-wake day in Study 1; orange: protein pattern from the first 24 hours of Constant Routine (CR) in Study 1; purple: protein pattern from the final 15 hours of CR and recovery night in Study 1; green: protein pattern from the first 24 hours of CR in Study 2. Transparent curves represent variations computed using the same process but deviating from the main barycenter by a single subject. Note that not all proteins were assessed in Study 2, which was tested with a different version of the Somalogic panel that did not include them. Additionally, SomaLogic sometimes uses two types of aptamers for the same protein, which is indicated here by a sequence id at the end of the protein’s name for Anterior gradient protein 3 in Panel A.

We performed a biological pathway enrichment analysis on identified circadian proteins with peak expression occurring within ±1 hour intervals, utilizing the GO_Biological_Process_2021, Reactome_2022, and KEGG_2021_Human databases using the Gene Set Enrichment Analysis (GSEA) computational method.^35,36 To map pathway activity over the 24 hours, we normalized combined scores along the time axis.^37,38

3. Results

Overall, we assayed 432 samples from the 17 subjects: Study 1, 24-hour sleep-wake day (122 samples from 8 participants); first 24 hours of CR (112 samples); the final 15 hours of CR and first 9h of recovery sleep (91 samples); Study 2, first 24 hours of data from the CR (107 protein samples from 9 participants).

Utilizing the default hyperparameters presented in Figure 2, our one harmonic analysis identified 804 proteins exhibiting rhythmic patterns. Among these, 293 displayed consistent 24-hour rhythms between conditions and participants, indicating a circadian origin. In our two harmonics model, we discerned 895 rhythmic proteins, with 328 showing circadian characteristics. Notably, 259 proteins identified using the two harmonics approach were not detected in the single harmonic method (Figure 3.A). Combining insights from both models, among the 6,916 proteins we identified a total of 1,063 (15%) proteins with rhythmic tendencies. Among these, 431 (6.2%) exhibited clear circadian rhythms across varied conditions and participants. Further characterization of these circadian proteins by amplitude and phase revealed deeper insights into their oscillation strength and timing (Supplemental Table A.1). Figure 3.B displays a selection of the most robust circadian proteins, emphasizing their inherent oscillatory patterns along with their respective amplitudes and phases.

Among the most robust circadian proteins were pro-opiomelanocortin (POMC) and parathyroid hormone (PTH), two well established circadian hormones (Table 1 and Figure 3). POMC is a precursor of adrenocorticotropic hormone (ACTH), and peaked 11.3 hours after DLMO, a time consistent with the well-known morning peak of cortisol. PTH peaked 2.1 hours after DLMO, although a second, smaller peak was present under entrained sleep-wake conditions as has been described for circulating levels of the hormone^{3 39} (see blue curve in Figure 3).

Table 1:

Characteristics of the proteins in Figure 3.

* C for Circadian, R for Rhythmic					1-harmonic model				2-harmonic model
Seq-id	Name	Gene	UniProt	Class*	p-rhy	p-com	A	φ	p-rhy	p-com	A1	A2	φ1	φ2
23326–10	Glutathione S-transferase A2	GSTA2	P09210	C	3.76E-05	1.22E-01	0.223	7.7	2.24E-08	3.20E-04	0.230	0.070	7.5	18.7
9204–33	Pro-opiomelanocortin	POMC	P01189	C	9.12E-42	2.06E-01	0.175	11.3	5.44E-44	2.19E-01	0.176	0.060	11.2	16.2
7921–65	Four-jointed box protein 1	FJX1	Q86VR8	C	6.24E-47	4.51E-01	0.164	3.5	8.28E-49	3.27E-01	0.166	0.047	3.5	12.0
3509–1	C-C motif chemokine 15	CCL15	Q16663	C	2.20E-17	5.84E-01	0.132	5.0	1.79E-15	5.85E-01	0.133	0.020	5.0	15.9
8620–56	Kallikrein-14	KLK14	Q9P0G3	C	5.75E-09	5.54E-02	0.100	7.6	2.84E-09	1.09E-02	0.100	0.023	7.7	15.1
5954–62	Parathyroid hormone	PTH	P01270	C	5.75E-05	1.13E-01	0.075	2.1	2.44E-06	4.16E-02	0.078	0.045	2.0	10.0
24490–16	THAP domain containing protein 4	THAP4	Q8WY91	C	7.34E-08	1.97E-02	0.096	8.8	2.39E-06	5.48E-02	0.098	0.022	8.9	19.2
2212–69	Tissue-type plasminogen activator	PLAT	P00750	C	4.63E-11	1.88E-01	0.165	10.7	6.65E-11	2.63E-02	0.160	0.005	10.7	13.8
10574–10	Beta-2-microglobulin	B2M	P61769	C	5.96E-05	7.92E-01	0.088	7.0	5.25E-04	7.50E-01	0.089	0.023	6.9	13.2
24957–6	Espin	None	B1AK53	C	4.39E-08	2.47E-01	0.086	6.5	9.92E-09	1.28E-01	0.086	0.033	6.3	11.8
18289–16	C-C motif chemokine 15	CCL15	Q16663	C	3.38E-21	2.88E-01	0.101	5.2	7.58E-19	2.03E-01	0.101	0.007	5.3	16.8
25048–30	Protein kinase C and casein kinase substrate in neurons protein 2	PACSIN2	Q9UNF0	C	1.45E-04	1.29E-01	0.194	21.1	4.05E-05	2.83E-02	0.197	0.075	21.2	4.8
21495–134	Prosaposin receptor GPR37	GPR37	O15354	C	1.51E-15	6.85E-02	0.086	4.8	7.46E-16	6.52E-03	0.087	0.014	4.7	13.8
9264–11	Cathepsin O	CTSO	P43234	C	9.87E-10	6.31E-02	0.075	13.9	5.13E-08	8.64E-02	0.078	0.003	13.9	0.6
15686–49	Inhibin beta C chain	INHBC	P55103	C	5.70E-10	4.21E-01	0.087	18.5	8.82E-09	2.35E-01	0.087	0.011	18.4	8.8
15375–49	Carboxypeptidase B	CPB1	P15086	C	3.80E-05	1.13E-01	0.073	21.8	2.15E-04	1.20E-01	0.079	0.022	21.7	4.8
5668–49	Anterior gradient protein 3	AGR3	Q8TD06	R	4.22E-02	1.45E-01	0.032	23.4	1.34E-05	6.52E-03	0.035	0.044	23.3	10.1
17342–13	Anterior gradient protein 3	AGR3	Q8TD06	R	2.97E-02	2.38E-01	0.086	23.5	1.08E-05	9.84E-03	0.090	0.098	23.5	11.3
13118–5	SPARC-related modular calcium-binding protein 1	SMOC1	Q9H4F8	R	2.43E-14	4.27E-12	0.026	19.9	1.82E-17	1.92E-12	0.026	0.031	19.9	10.2
9384–17	Cathelicidin antimicrobial peptide	CAMP	P49913	C	4.90E-01	3.45E-01	0.008	14.9	4.57E-04	1.46E-01	0.010	0.032	14.3	21.9

Notes: The provided values for amplitudes and acrophases represent the average results across the four conditions employed in the analysis. Seq-Id is a unique id used by SomaLogic to measure proteins. UniProt provides a unique reference to a protein in UniProt database. In column class, “C” stands for Circadian while “R” stands for for Rhythmic. P-rhy is the adjusted p-value that tests if there is a rhythm. P-com is the adjusted p-value that tests if rhythms are different. The first block of proteins corresponds to the proteins in Figure 3 B while the last block corresponds to Figure 3 A. The middle block provides additional circadian proteins.

Previously unknown circadian proteins that were identified include glutathione S-transferase A2 (GSTA2); four-jointed box protein 1, a protocadherin involved in cancer progression; chemokine (C-C motif) ligand 15; kallikrein-14, a protein involved in skin desquamation and prostate cancer; tissue-type plasminogen activator; espin, a microfilament binding protein; and GPR37.

When we examined the time at which the circadian proteins peaked, we found that their acrophases were not distributed randomly across the 24-hour day. Rather, more proteins peaked in the late afternoon/evening, during the wake maintenance zone when the circadian rhythm of temperature is at its peak and circadian drive for alertness is highest^40–42 (Figure 4). A second, smaller peak of circadian protein acrophases was observed during the latter half of the habitual sleep episode, in the early morning when the endogenous circadian rhythm of temperature is at its nadir and the circadian rhythm of sleep propensity is at its peak (Figure 4).

Figure 4. Biological pathway enrichment analysis for endogenous circadian proteins with respect to peak times.

Pathway analysis was carried out on circadian proteins peaking within ±1 hour with GO_Biological_Process_2021, Reactome_2022, KEGG_2021_Human database using the Gene Set Enrichment Analysis (GSEA) computational method.^35,36 The combined score was normalized along the time axis to visualize when the pathway is the most active (see also Supplementary Tables 2 and 3). The solid black line illustrates the number of endogenous circadian proteins peaking at each time (right axis) used to conduct the pathway analysis. The vertical dashed lines delineate the two times of day when most circadian protein peaks occurred. Relative clock hour is presented along the top x-axis, while time relative to Dim Light Melatonin Onset (DLMO) is presented along the bottom x-axis.

Finally, we carried out pathway analysis for proteins peaking within ±1 hour, adding additional insight into many of the novel endogenous circadian proteins identified in our study. The consolidated pathways having the most significant loadings are presented in Supplemental Table A.3, which includes the peak activity time for each pathway. We next grouped the pathways into four segments by the time within the 24-hour day at which they peaked. Supplemental Table A.2 lists all the pathways and the time segment at which each peaked. This is presented visually in Figure 4. Table 2 presents the five most significant pathways peaking within each of these segments. As Table 2 and Supplemental Table A.4 show, pathways peaking in the evening are diverse and include immune (T cell development), prostaglandin metabolism, chondrocyte maturation, and protein localization to axons. Among the most significant pathways, positive regulation of nucleotide-binding oligomerization domain containing 2 (NOD2) signaling,⁴³ an inflammatory pathway, and two heat acclimatation pathways peaked in the morning, a time when few pathways were active (Figure 4). Pathways active in the late night/early morning included innate antibacterial humoral responses, anion transport, negative regulation of metallopeptidase activity, mucosal immune responses, glial and endothelial cell regulation.

Table 2:

Pathways analysis per periods

Periods	Term	Combined Score
0am – 8am	regulation of glial cell apoptotic process (GO:0034350)	1092.6
	negative regulation of glial cell apoptotic process (GO:0034351)	837.4
	positive regulation of mast cell chemotaxis (GO:0060754)	648.5
	regulation of mast cell chemotaxis (GO:0060753)	460.2
	lymphocyte mediated immunity (GO:0002449)	460.2
8am – 2pm	positive regulation of nucleotide-binding oligomerization domain containing signaling pathway (GO:0070426)	2185.5
	cellular heat acclimation (GO:0070370)	1568.4
	regulation of nucleotide-binding oligomerization domain containing 2 signaling pathway (GO:0070432)	1568.4
	heat acclimation (GO:0010286)	1568.4
	positive regulation of microtubule nucleation (GO:0090063)	1207.9
2pm–8pm	actin filament depolymerization (GO:0030042)	802.3
	regulation of glomerular mesangial cell proliferation (GO:0072124)	504.4
	regulation of plasma membrane organization (GO:1903729)	504.4
	Cytosolic tRNA Aminoacylation R-HSA-379716	369.2
	regulation of immature T cell proliferation in thymus (GO:0033084)	356.9
8pm–0am	regulation of growth hormone receptor signaling pathway (GO:0060398)	293.8
	positive regulation of smooth muscle cell differentiation (GO:0051152)	293.8
	membrane raft distribution (GO:0031580)	293.8
	determination of left/right asymmetry in lateral mesoderm (GO:0003140)	293.8
	positive thymic T cell selection (GO:0045059)	293.8

Notes: Using the peaking circadian proteins and GO_Biological_Process_2021, Reactome_2022, KEGG_2021_Human databases, we present the most significant pathways in the given periods segments of the 24-hour day.

4. Discussion

We described the physiology of the blood circadian proteome in humans by measuring nearly 7,000 proteins in samples collected every two hours from 17 healthy participants using the gold-standard constant routine (CR) protocol. Applying both a single and dual harmonic method of circadian analysis to the data, we identified 1,063 diurnal proteins, 431 of which exhibit endogenous circadian rhythmicity. Thus, at least ~10–15% of circulating proteins are regulated by the circadian system, with proteins peaking at different times across the 24-hour cycle, although most commonly in the late afternoon or early evening hours as illustrated in Figure 4. Our approach to studying the blood circadian proteome in humans allowed us to identify and characterize proteins with circadian behaviors beyond simple sinusoidal rhythms. The distinction between diurnal and endogenous circadian proteins, coupled with assessment of amplitude and phase, provides insights into the complexity, stability, and temporal characteristics of the multi-oscillatory circadian timing system in humans.

Using the CR protocol, we were able to distinguish endogenous circadian proteins from those with diurnal fluctuations, many of which could be driven by the timing of sleep, activity, posture, or food intake. This distinction, often overlooked in studies of human circadian rhythmicity, is critical for understanding clock dysfunction.²⁶ Rhythms driven by behaviors rather than by endogenous circadian clocks would be expected to be altered in shift workers, patients with circadian rhythm sleep-wake disorders, and others (such as patients in Intensive Care Units); understanding the distinction between endogenous circadian rhythms versus diurnal rhythms is critical to be able to use proteins to identify endogenous circadian phase as well as to identify the presence of circadian disruption. Reassuringly, POMC, the precursor of ACTH and a driver of cortisol, had one of the most robust patterns, peaking almost antiphase to DLMO, as expected.

Other circadian proteins identified included tissue-type Plasminogen Activator (tPA), which is involved in fibrinolysis and is associated with sleep apnea⁴⁴. Interestingly, GSTA2, an enzyme involved in hematopoiesis and detoxification, notably for chemotherapeutic agents such as paclitaxel⁴⁵ and in the hepatotoxicity of selected medications,⁴⁶ peaked late at night, of possible significance for chrono-pharmacology. Chemokine (C-C motif) ligand 15, an important chemotaxis agent for neutrophils, monocytes, and lymphocytes, also peaked late at night, and may, in addition to the modulatory effects of cortisol, explain well-known circadian variations in white blood cell subtypes across the 24-hour day.^47,48 Carboxypeptidase B, a protease involved in the biosynthesis of neuropeptides and peptide hormones⁴⁹ is primarily secreted by the pancreas, is involved in complement activation and other endocrine functions, and peaks in the afternoon. Prosaposin/prosaptide receptor GPR37, a brain receptor for saposins, important neurotrophic and glioprotective factors for Parkinson disease,⁵⁰ and SPARC-related modular calcium-binding protein 1 (SMOC-1) a biomarker of tau involvement in Alzheimer’s disease, were also strongly circadian.^51,52 Future studies in which analysis of relative phases of protein subsets is elucidated will allow better understanding of the coordination of rhythmicity across organs and cell types. The importance of this is clear from animal models, in which desynchronization between central and peripheral oscillators can induce metabolic disease and immune dysfunction, and may account for the numerous health complications reported among shift workers.⁵³ These examples exemplify the far-reaching implications of circadian regulation in health and disease, and hint at the power of circadian proteomics for use in precision medicine.

Our study of nearly 7,000 plasma proteins allowed us to probe the output of many organs and physiologic functions, showing the feasibility of large-scale proteomic studies of circadian rhythmicity. Grouping the proteins by pathways revealed additional insight. Pathways regulating T cell differentiation and activation in the thymus occurred in the late afternoon. Cell numbers in blood⁵⁴ as well as T cell egress from the thymus⁵⁵ are influenced by sleep and circadian timing, peaking in the evening and early night, a few hours after this pathway is activated. Prostaglandin metabolism and chondrocyte maturation, strongly circadian processes,⁵⁶ also peaked at this time. In contrast, pathways relating to endothelial cell proliferation and repair peaked in the early morning, reflecting circadian and organ specific modulation.⁵⁷ Understanding the timing of such pathways may provide new insights into the best time of day to administer certain medications or treatments.⁵⁸A notable finding was the strong endogenous circadian regulation of the heat acclimatation pathway, which peaked in the late morning when most other pathways are silent. This may reflect anticipation of the warmest time of day, a timely finding considering the threat of global warming. Similarly, the NOD2 pathway, important for autoinflammatory syndromes such as Chron’s disease, Blau syndrome, and NOD2-associated autoinflammatory diseases,⁴³ is also active in the late morning.

Proteomics is increasing as a field of inquiry, yet studies rarely if ever control for time of day or circadian effects, effects that the present findings demonstrate could be confounding, or if considered could increase power for discovering new associations. Only one study has explored circadian effects using a smaller panel of 1,300 proteins in a simulated shiftwork protocol.⁵⁹ It is becoming more evident that circadian abnormalities are present (and even at the core of) in many disorders from psychiatric to neurodegenerative, thus understanding how rhythmic proteins may be altered in these conditions has its own value. Two proteomics platforms are available and offer comparable performance,^60–62 Olink, an antibody-based platform, and Somalogic, an aptamer-based technology. However, correlations with mass spectrometry or enzyme-linked immunosorbent assay (ELISA) vary, so that any individual result must be interpreted cautiously. Nonetheless, integration of these results with genomic data has generally validated many targets by revealing cis-pQTLs close to each gene of interest, and these findings are now being integrated with genomic data through Mendelian randomization studies.^23,24,63,64 Similar experiments may, in the future, allow convergence of natural variation associated with morningness/eveningness or other diseases with specific pathways or organ specific clocks.⁶⁵

The discovery of proteins peaking at various times also offers the possibility of building an algorithm that could predict central circadian clock phase with a single blood sample, a circadian “time stamp”.¹⁷ Although this has been attempted using gene expression and metabolomics in human blood,^20,66 precision has been moderate except when two blood samples are collected. Further, in some cases validation has been done in entrained conditions that cannot distinguish diurnal from circadian and thus may not accurately reflect the status of the circadian system. One exception has used gene expression within a single cell population, monocytes, and has shown good correlation with DLMO.²¹ It is likely that noninvasive measures of physiology through wearables, together with cell specific measures of circadian rhythmicity and global variation of the human proteome and metabolome will lay the foundation for improved circadian time stamp methods.⁶⁷

5. Conclusions

A major strength of our study was the use of the CR protocol which allowed us to distinguish between diurnal rhythms (likely driven by rhythmic behaviors) and endogenous circadian rhythms. Our approach revealed hundreds of plasma proteins that show endogenous circadian rhythmicity, from a wide range of organ systems and physiologic functions. Future analyses will likely reveal novel pathways affected by circadian timing and pave the way for improved diagnosis and treatment for patients with circadian rhythm disorders and other sleep pathologies. It will also advance efforts to include knowledge about time-of-day, thereby incorporating circadian medicine into the broader field of personalized medicine.

6. Limitations

Despite its many strengths, there are several limitations to this work. While we assayed samples every 2 hours from the participants, the overall number of participants was small. Because this was an initial study of circadian regulation of plasma proteomics we selected only young healthy adults as participants. The amplitude and/or timing of the circadian rhythm of proteins may differ in older adults (as has been recently demonstrated for metabolomic rhythms⁶⁸), in individuals with specific medical disorders, or in patients with circadian rhythm disorders. Additional studies in such individuals should be carried out to understand whether the circadian proteins we have identified are reliably rhythmic in those populations.

In addition, analysis of the phase relationships between protein subsets in healthy individuals of a wider range of ages will allow better understanding of the coordination of rhythmicity across organs and cell types under normal conditions so that we can harness that knowledge in patients. While it was beyond the scope of this study, understanding the phase relationships between protein subsets in patients who may show desynchronization between central and peripheral oscillators should be carried out, as it will likely elucidate how such desynchrony leads to metabolic disease and immune dysfunction (for example, in shift workers). Analysis of the relative timing of protein rhythms in different physiological pathways also provide important insight for the development of circadian medicine, for example the best time of day to administer certain medications or treatments.

Public Health Relevance.

Our ability to incorporate circadian timing information into clinical decision-making for personalized medicine is impaired by our inability to measure circadian phase quickly and easily. The most common method for assessing circadian timing is measuring melatonin in saliva or plasme, an expensive and time-consuming procedure that necessitates multiple samples over several hours in strictly controlled lighting conditions. Furthermore, melatonin informs about the status of the central circadian clock but does not provide information about peripheral clocks, which have been demonstrated to become internally desynchronized from the brain’s central clock in some conditions. Assessment of rhythmic proteins can be further developed to fill this technological gap, allowing for development and refinement of circadian biomarkers that can be used in both sleep medicine and more broadly in precision medicine to inform about central clock timing as well as synchronization between the central clock and peripheral clocks.

Data Availability Statement

Information about each of the identified proteins are available in Supplementary Table 1.