Simulated Night Shift Disrupts Circadian Rhythms of Immune Functions in Humans

About 15-30% of the workforce in industrialized countries operates outside the standard daytime hours (1, 2). This often requires a drastic change in the sleep schedule every few days or weeks. As a result, a conflict between external signals (e.g., light, food intake) and the organism’s internal timing system can be observed, leading to a state of internal desynchrony between the different clocks of the circadian system and the shifted sleep–wake cycle. The circadian system allows a coordination of different molecular, physiological, and behavioral functions with each other. In the case of circadian disruption, the equilibrium among the different functions is lost, which can have important consequences. Indeed, studies in rodents and in humans using environmental circadian disruption (e.g., simulated shift work, jetlag, light–dark cycles different from 24 h) associated circadian misalignment with various health consequences (3). For instance, rodents subjected to repeated jetlag exhibit faster tumor growth (4, 5) and more severe septic shock in response to endotoxin administration (6). In humans under a 28-h day protocol (forced desynchrony), altered leptin levels, as well as glucose and insulin response after meals (up to a prediabetic level), were observed when the circadian and behavioral cycles were misaligned compared with the days when they were well aligned (7). Thus, these circadian disturbances are thought to be an important contributor to shift work–associated medical disorders, such as increased risk for infection, autoimmune diseases, obesity, diabetes, cardiovascular disorders, and cancer (2, 8–13).

The circadian system is known to regulate many biological functions in a rhythmic fashion. Among them, the innate and adaptive immune systems were shown to display daily rhythms in rodents and humans (14). This includes rhythms in cell counts in the blood (15–18) and peripheral lymphoid organs (19), lymphocyte proliferation (20), cytokine levels in the blood (21–24), and following ex vivo stimulation with various stimuli (16, 19, 25–28).

The relative contribution of the different parts of the circadian system to the rhythmic regulation of immune functions has only started to be deciphered. In mammals, the circadian system comprises a central clock located in the suprachiasmatic nucleus (SCN) of the hypothalamus and peripheral clocks located in various cell types and organs (29, 30). Although the peripheral clocks can function in an autonomous fashion, within the organism they are entrained by the SCN central clock via neuronal and endocrine pathways (31). The SCN clock can be reset by environmental light–dark cycles, whereas the peripheral clocks can also respond to other synchronizers, such as meal times, temperature, and humoral factors. In central and peripheral clocks, rhythms with a period of ∼ 24 h are generated by autoregulatory feedback loops involving clock genes (e.g., CLOCK, BMAL1, Period or PER1-3, Cryptochrome or CRY1-2) (32). Hence, immune cells (e.g., monocytes/macrophages, lymphocytes) have the molecular clock machinery and display circadian gene expression (19, 20, 25, 33–35). The rhythmic functions, such as cytokine production, seem to be under the control of the immune clocks (19, 25, 28), whereas the number of circulating WBCs appears to be controlled by the central clock via humoral and neuronal signaling (18, 36, 37). It is hypothesized that immune clocks are able to gate the integration of central clock’s signals to limit or reinforce their effects at certain times of day (14).

In conjunction with the circadian system, sleep is known to regulate immune functions, although it is difficult to distinguish the respective influence of these two regulatory systems (38). From studies in humans, sleep, sleep deprivation, sleep restriction, and mistimed sleep appear to affect immune functions, but discrepancies were observed (38). Given the deleterious effects of circadian disruption and insufficient sleep on many biological processes (e.g., endocrine, metabolic, cardiovascular, cognitive), as observed in some shift workers (39), more studies are needed to characterize the effects of such disruption on the functions of the innate and adaptive immune systems. The objective of the current study was to subject individuals to a simulated night shift protocol and to investigate the effects on innate and adaptive immune responses. More specifically, we tested the hypothesis that the rhythms of human T lymphocyte and monocyte counts are mainly controlled by the central clock, whereas the rhythms of cytokine production are controlled by several factors, including local immune clocks and the sleep/wake cycle. To our knowledge, we tested for the first time whether night shifts differentially affect the circadian rhythms of immune functions, resulting in a temporal misalignment between these rhythms, as a result of the entrainment (cytokine release) or absence of entrainment (WBC counts) to a night-oriented schedule.

Ten healthy subjects (22.6 ± 3.4 y old; one woman) were enrolled, and eight subjects completed the study (S20 only participated to the baseline condition; immune data for S14 were not available). They were studied individually in time isolation for 6 d at the Centre for Study and Treatment of Circadian Rhythms of the Douglas Mental Health University Institute in May to August in 2013 and 2014. Participants were recruited as previously reported (34), and they gave their written informed consent, as approved by the Douglas Institute Ethic Board and within the ethical standard of the Declaration of Helsinki. For ≥7 d prior to laboratory admission, subjects maintained a stable sleep/wake schedule, with an average sleep duration of 07h59 ± 00h02, to stabilize circadian rhythmicity (34). Subjects were admitted to a time-free laboratory environment for 6 d (Fig. 1) (40). Following the first day (experimental day 1), they underwent a procedure called constant posture (CP) for 24 h (experimental day 2), involving minimal levels of activity (subject had to stay in a semirecumbent position), dim light exposure, and hourly isocaloric snacks. This first CP sampling period, under a day-oriented schedule (referred to as the baseline condition), served to determine baseline rhythms of immune functions and plasma melatonin. At the start of experimental day 3, a 4-d simulated night shift work schedule began. The sleep schedule was delayed by 10 h relative to habitual sleep times and was maintained for the remainder of the study. During the waking period of the first simulated night shift (experimental day 3), subjects were given hourly isocaloric snacks, whereas on the second and third simulated night shifts (experimental days 4 and 5), subjects received three meals per day (+45 min, +4 h, and +10 h after lights on), as well as a snack (2 h before lights off). Following the 8-h sleep episode of experimental day 5, subjects underwent a second 24-h CP (experimental days 5 and 6) in simulated night shift conditions. Thereafter, this second CP sampling period (referred to as the night shift condition) served to determine the effects of simulated night shifts on the rhythms of immune functions and plasma melatonin. Following the 8-h sleep episode of experimental day 6, subjects were allowed to leave the laboratory.

Throughout the six experimental days, ambient light intensity was tightly controlled and continuously monitored via an Actiwatch with a light-mounted sensor (Actiwatch-L, MiniMitter, Bender, OR). During waking periods, it was also monitored in the average angle of gaze every 2 h using a research photometer (IL1400A; International Light Technologies, Peabody, MA). During the waking period of experimental day 1, average ambient light levels were 295.5 ± 13.2 W/m² (100.9 ± 4.5 lx). During experimental days 2–6, average ambient light levels were decreased to reach an average of 7.6 ± 0.6 W/m² (2.6 ± 0.2 lx) during waking periods. Levels of light were always <0.1 W/m² (0.03 lx) during sleep episodes.

An indwelling catheter was inserted in a forearm vein (33–35) a few hours prior to the beginning of each CP sampling period, allowing the sampling of blood without disturbing the subject’s sleep. Every 4 h, 5 ml whole blood was collected in a heparin-coated tube and centrifuged for 30 min at 370 × g on a Ficoll density gradient (Histopaque-1077; Sigma-Aldrich, Oakville, ON, Canada) to isolate PBMCs. Following three washes with 1 × PBS, PBMCs were resuspended in a solution for cryopreservation (50% RPMI 1640, 40% FBS, 10% DMSO kept at room temperature) (41). PBMCs were placed in a Styrofoam box for 24 h at −80°C before being stored in liquid nitrogen until further processing.

For a given subject, PBMC samples for both conditions (baseline and night shift) were thawed (41–43) and placed in culture at the same time. Cryopreserved PBMCs were transferred from liquid nitrogen to a 37°C water bath for 10 min, transferred to a 15-ml tube, and washed twice with RPMI 1640 at 200 × g. PBMCs were resuspended in 1 ml RPMI 1640 (supplemented with 5% FBS, 2 mM l-glutamine, 100 U/ml penicillin/streptavidin, 10 mM HEPES, 1 mM sodium pyruvate, 1× nonessential amino acids) and set at 1 × 10⁶ cells/ml following their count in trypan blue using a hematocytometer. For stimulation, 0.4–0.5 × 10⁶ cells were incubated at 37°C with 5% CO₂ in humidified air in 24-well plates and stimulated with 10 μg/ml LPS (Sigma-Aldrich; 4-h incubation to stimulate monocytes to release IL-1β, IL-6, and TNF-α) or 13.3 μg/ml PHA (Sigma-Aldrich; 24-h incubation to stimulate T lymphocytes to release IL-2 and IFN-γ). Unstimulated cells (no LPS or PHA added) were used as controls. Plates were centrifuged at 200 × g for 10 min, and supernatants stored at −80°C until further processing.

Cytokines were measured by ELISA, according to the manufacturer’s instructions (Life Technologies, Burlington, ON, Canada, for IL-1β, IL-6, TNF-α, and IFN-γ; Affymetrix eBioscience, San Diego, CA for IL-2). The sensitivity of the assays was 1 pg/ml for IL-1β, 2 pg/ml for IL-6 and IL-2, 1.7 pg/ml for TNF-α, and 4 pg/ml for IFN-γ. The intra- and interassay coefficients of variation were ≤8.6 and ≤8.4%, respectively, for all assays. Each sample was tested in duplicate in the same assay.

In addition to the ex vivo stimulation, we used 0.4 × 10⁶ cells/sample to quantify the relative proportion of monocytes (CD14⁺/CD11b⁺), T lymphocytes (CD3⁺), and platelets (CD42b⁺) using the following Abs (BioLegend, San Diego, CA): CD3 PerCP (SK7), CD11b allophycocyanin (M1/70), CD14 Alexa Fluor (M5E2), and CD42b PE (HIP1). Following the blockade of FcRs with Human TruStain FcX (BioLegend), cells were stained for 30 min at 4°C with a mix of anti-CD3 (6 mg/ml), anti-CD11b (1.25 mg/ml), anti-CD14 (8 mg/ml), and anti-CD42b (1 mg/ml). Cells were analyzed on a FACSCalibur flow cytometer (BD Biosciences, Mississauga, ON, Canada). Raw data were analyzed with FlowJo 8.7.3 (Tree Star, Ashland, OR). For technical reasons, cells from only five subjects (S18, S19, S21, S22, and S23) were used for this quantification. Within a gate covering all PBMCs, monocytes and T lymphocytes were identified as CD11b⁺/CD14⁺ and CD3⁺ cells, respectively. Platelets (CD42b⁺) were either free (i.e., not bound to other cells) in the platelet gate or bound to other cells (e.g., monocytes) in the PBMC gate, as previously described (44). Although we could not eliminate the PBMC-bound platelets, the relative proportion of free platelets was always low (≤3.9% of total cell number included in the platelet gate and the PBMC gate) and stable over time. This showed that our thorough washing procedure minimized the presence of platelets in the samples, making it unlikely that cytokine secretion was affected by their presence (45, 46).

Every ∼60 min, 2 ml whole blood was obtained in K₂EDTA-coated tubes to measure plasma levels of melatonin. Samples were centrifuged at 1494 × g for 15 min at 4°C, and the plasma was stored at −80°C until further analysis. According to the manufacturer’s instructions, plasma melatonin was quantified in duplicates by RIA (Labor Diagnostika Nord, Nordhorn, Germany; range, 3–1000 pg/ml; sensitivity, 2.3 pg/ml; intra- and interassay coefficients of variation of 8.1 and 4.2%, respectively).

All data are presented as mean ± SEM, with the exception of data for age, body mass index, and chronotype, which are reported as mean ± SD. The p values < 0.05 were considered significant. Bedtimes and wake times were assigned a relative clock time of 00:00 and 08:00, respectively, to align subjects’ data with each other relative to their wake time. All phases reported in the text and figures are based on values adjusted to this relative clock time.

Melatonin data are given in pg/ml, whereas the other data were transformed into z-score for each subject [z = (x − μ)/σ, where x is the raw value, μ is the mean, and σ is the SD for the subject for 24 h], to account for interindividual variability in the response to the ex vivo stimulation and were averaged per group.

Harmonic regressions were applied to group data (nonlinear mixed-effect models using nlmixed SAS procedure; SAS Institute, Cary, NC) (35). We set the period parameter of the regression to 24 h, facilitating the assessment of phase shifts between and within subjects. A rhythm was considered significant when the 95% confidence interval of the calculated 24-h and/or 12-h amplitude did not include the zero value, as confirmed by p values given by the nlmixed procedure (Tables I–III, Supplemental Table I) (35).

For plasma melatonin, time of fitted maximum (phase) and amplitude were assessed using a three-harmonic regression (35, 47). We calculated the composite phase via a manual process using fitted amplitude and phase values from the three harmonics (6-, 12-, and 24-h periods), followed by a bootstrap strategy (1000 repetitions) to calculate the SEM (35, 48). The amplitude was defined as the mean-to-trough difference of the first harmonic of the regression (49).

For WBC counts and cytokine levels, we applied dual-harmonic regressions (35) rather than single-harmonic regressions, because a visual inspection revealed bimodal patterns for most of the data series. We calculated the composite phases via a manual process using fitted amplitude and phase values from the two harmonics (12- and 24-h periods), followed by a bootstrap strategy (35, 48). For most of the dual-harmonic regressions, two composite phases could be calculated (Tables I–III, Supplemental Table I). To estimate the relative size of the second phase compared with the main one, we first calculated the peak-to-trough variation for each of the composite phases as follows: z-score value observed at the time of the first (or second) peak − z-score value observed at the time of the main trough (provided by the regression). Then, we expressed the peak-to-trough variation observed for the second phase as a percentage of the peak-to-trough variation calculated for the first phase. We considered the second phase substantially smaller when the percentage obtained was <50%. All of these values are reported in Table II and Supplemental Table II.

We used F statistic provided by the nlmixed procedure to assess amplitude changes in the melatonin rhythm induced by night shift. All of the phases reported in the 8Results section were composite phases (as described above) that were not generated by the nlmixed procedure; thus, F statistic (taking into account repeated measures) could not be used. Instead, we applied unpaired t tests to analyze changes in phases between baseline and night shift conditions. We assessed phase shifts based on the nighttime peak (ϕ₁ in Fig. 2) observed at baseline because this was the peak present for all of the cytokines studied and often the most prominent. Following night shifts, we used the timing of the first peak, which was the prominent peak for all the cytokines, as ϕ₂ (Fig. 2). Phase shifts were calculated as ϕ₁ − ϕ₂.

In subjects living under a day-oriented schedule (baseline in CP, involving minimal levels of activity, dim light exposure, and hourly isocaloric snacks) (Fig. 1), we found significant rhythms of absolute levels of cytokine secretion in response to a short stimulation of monocytes and T lymphocytes with LPS and PHA, respectively (Fig. 2, Table I). Throughout this article, results are presented following these stimulations unless stated otherwise. The cytokines released by monocytes (IL-1β, IL-6, and TNF-α) followed a bimodal profile with two periods of high secretion (similar peak-to-trough variation; Supplemental Table II; see 2Materials and Methods), at night and in the evening, as seen in individual curves (Supplemental Fig. 1A–C) and in averaged group data (Fig. 2A–C, Table I). The cytokines released by T lymphocytes (IFN-γ and IL-2) displayed a rhythm with only one major peak at night (Fig. 2D, 2E, Table I, Supplemental Fig. 1D). A second peak of secretion resulting from the dual-harmonic regression was seen for IFN-γ and IL-2, but its peak-to-trough variation represented only 19 and 44% of that measured for the main peak, respectively (Supplemental Table II). Nonetheless, we reported the time of both peaks for all cytokines in Fig. 2 and Table I. As expected, control assays without stimulation showed much lower mean cytokine levels (Table I).

In subjects living under a night-oriented schedule for four consecutive 24-h cycles (night shift CP; Fig. 1), absolute levels of cytokine secretion after stimulation were still significantly rhythmic, with the exception of IL-2 (Fig. 2, Table I). The average cytokine levels were similar between night shift and baseline conditions (Table I). The cytokines released by monocytes still displayed a bimodal profile (Fig. 2A–C, Supplemental Fig. 1A–C) with a main peak of secretion occurring at ∼21:40 (Table I), ∼3–4 h after waketime under the night-oriented schedule. The regression displayed a second peak at ∼10:30 for IL-1β, IL-6, and TNF-α (Table I, i.e., around bedtime under the night-oriented schedule). In contrast to baseline, in which the two peaks were of similar peak-to-trough variation, the second peak was reduced compared to the first one (Supplemental Table II). Only one peak of secretion was observed for IFN-γ, at 22:22 ± 03:52 (Fig. 2D, Table I, Supplemental Fig. 1D, Supplemental Table II), whereas no significant peak was seen for IL-2 (Fig. 2E, Table I, Supplemental Fig. 1E, Supplemental Table II). As expected, control assays without stimulation showed much lower mean cytokine levels (Table I).

We then statistically assessed phase shifts by comparing phases obtained at baseline with those observed following night shifts. We observed a significant phase advance (p < 0.001) for IL-1β (+ 5 h 17 min), IL-6 (+ 6 h 14 min), TNF-α (+ 4 h 29 min), and IFN-γ (+ 4 h 30 min) in response to night shift (Fig. 2). We then decided to verify whether the numbers of circulating monocytes and T lymphocytes at each time point influenced the shape of the cytokine-secretion curve. We first analyzed the rhythm of circulating cell counts and then calculated the secretion/cell, for each cytokine and at each time point.

We first quantified the monocytes (CD14⁺/CD11b⁺ cells) present in each PBMC sample (see Fig. 3A, 3D for examples). In both baseline and night shift conditions, we found an average, over 24 h, of ∼14% monocytes among PBMCs (Table II), which is within the habitual range in humans (50). As the result of interindividual variability, the variations observed for the averaged z-scores did not display a significant rhythm (Fig. 3B, 3E, Table II). Nevertheless, by looking at the individual profiles we could separate subjects into two subgroups. At baseline, although all subjects had a trough in monocyte numbers during the night (sometime between 00:00 and 08:00), a peak was observed during the morning (between 08:00 and 12:00) in one subgroup (n = 2; m in Fig. 3C) and during the evening (between 16:00 and 20:00) in the other subgroup (n = 3; e in Fig. 3C). Thus, we concluded that monocyte numbers were lower at night than during the day, which is in line with results from previous studies (15, 16).

In the night shift condition, no significant rhythm of monocyte counts was observed for averaged z-scores, but individual profiles could also be divided in two subgroups. We observed a trough in monocyte counts between 02:00 and 06:00 (i.e., middle of the wake period under the night-oriented schedule) for all subjects and a peak between 18:00 and 22:00 for two subjects (“e” in Fig. 3F) and between 10:00 and 14:00 for three subjects (“m” in Fig. 3F). Overall, no phase shift was observed for the relative number of monocytes in response to night shift, with the exception of the rhythm of one subject (S22) that shifted from the e subgroup to the m subgroup.

Of the nine subjects who participated in the study, we could only use PBMCs from five subjects to quantify the relative proportion of different cell types. Therefore, before calculating the cytokine secretion/cell, we verified that the rhythmic profiles of absolute levels of cytokine secretion were similar in this subgroup of five subjects compared with the entire group of nine subjects. As expected, the averaged data curves for IL-1β, IL-6, and TNF-α were similar (Supplemental Fig. 2), and most of the dual-harmonic regressions revealed significant rhythms with similar phases (Supplemental Fig. 2, Supplemental Tables I, II). This allowed us to calculate the secretion of IL-1β, IL-6, and TNF-α/monocyte for each subject and at each time point.

The cytokine secretion/cell was significantly rhythmic at baseline (Fig. 4A–C, Table III) with some differences compared with the rhythms of absolute levels of cytokine secretion (Fig. 2A–C). On one hand, the night peak of secretion for IL-1β, IL-6, and TNF-α was still present and was the most prominent peak, as indicated by the regression. On the other hand, when looking at the averaged group data, the second peak of secretion was barely present in the evening. This was confirmed by the regression displaying a second peak but with reduced peak-to-trough variation compared with that of the night peak (Supplemental Table II). Altogether, these results indicate a higher responsiveness of monocytes to LPS primarily at night, whereas a higher number of monocytes during the daytime contributed, in part, to the greater release of cytokines during that time.

Following night shifts, the secretion/monocyte of IL-1β, IL-6, and TNF-α was significantly rhythmic (Fig. 4A–C, Table III). Although the main peak of secretion/cell occurred in the late evening (∼22:10, Table III) as for absolute levels of cytokine secretion (Supplemental Table I), differences were observed for the second peak. When looking at the averaged group data, there was no second peak of secretion (Fig. 4A–C) for IL-1β, IL-6, or TNF-α. This was confirmed by the regression, which did not detect a second peak for IL-1β and found a very low peak-to-trough variation for the second peak of IL-6 and TNF-α (Fig. 4A–C, Supplemental Table II). The strong reduction or lack of the second peak observed under baseline and night shift conditions indicates that the phase shift assessment that we applied for the rhythms of absolute levels of cytokine secretion using ϕ₁ and ϕ₂ (Fig. 2) was appropriate; therefore, we calculated phase shifts for the cytokine secretion/cell accordingly. We found significant phase advances for IL-1β (+ 5 h, p < 0.01), IL-6 (+ 5 h 30 min, p < 0.001), and TNF-α (+ 5 h 12 min, p < 0.001) in response to the night shift (Fig. 4). This confirms that the night shift had an impact on the rhythm of cytokine secretion (for both absolute levels and levels/cell), whereas it did not significantly affect the rhythm of circulating monocyte numbers.

We quantified T lymphocytes (CD3⁺ cells) in each PBMC sample (see Fig. 3G, 3J for examples). We found an average, over 24 h, of ∼50 T lymphocytes among PBMCs before and after night shifts (Table II), which is within the habitual range in humans (50). In addition, T lymphocyte counts displayed a significant rhythm in both conditions (Fig. 3H, 3K). At baseline, in both averaged (Fig. 3H) and individual data (Fig. 3I), a main peak occurred at night (Table II). The regression indicated the presence of a second peak during the afternoon that was negligible based on the averaged z-scores (Fig. 3H, Table II). Following night shifts, we observed a main peak in T lymphocyte number at 01:52 (middle of the wake period, Fig. 3K, Table II) and a second negligible peak (Fig. 3K, 3L, Table II). An analysis of the group phase shift revealed that the small apparent phase advance (+ 59 min) in response to night shift was not significant (p = 0.79).

We confirmed that the rhythms of absolute IFN-γ and IL-2 secretion levels were similar for the five subjects for whom cell counts were assessed and for the nine subjects enrolled in the study (compare Supplemental Fig. 2D, 2E and Supplemental Tables I, II with Fig. 2D, 2E and Table I and Supplemental Table II). We then calculated the level of IFN-γ and IL-2 secreted per T lymphocyte and applied dual-harmonic regressions. We observed a trend toward significance (p = 0.07) for the rhythm of IFN-γ in the baseline and night shift conditions (Fig. 4D, Table III), whereas the rhythm of IL-2 was only significant at baseline (Fig. 4E, Table III). Yet, the averaged data displayed peaks of secretion/cell similar to those observed for the absolute levels of cytokine secretion; therefore, we calculated phase shifts and found that, in response to night shift, the rhythms of IFN-γ and IL-2 secretion/cell were significantly phase advanced (p < 0.001) by + 5 h 28 min and + 5 h 8 min, respectively (Fig. 4). In summary, the rhythm of circulating T lymphocyte number was not shifted in response to night shift, whereas the rhythm of cytokine secretion (absolute levels and levels/cell) was phase advanced by ∼5 h. Thus, at baseline, the rhythm of IFN-γ and IL-2 secretion is influenced by both a higher responsiveness of T lymphocytes and higher cell counts at night, whereas following a simulated night shift, shifted peak levels are primarily due to a shift in the rhythm of T lymphocyte responsiveness.

We measured the plasma melatonin rhythm in all subjects at baseline (n = 10) and under the night-oriented schedule (n = 9) to ensure that the central clock was not shifted in response to four cycles of night shift. We detected the presence of significant plasma melatonin rhythms at baseline and following night shifts (Fig. 5, p < 0.001 for both), with acrophases (i.e., time of maximum) occurring at 4:24 ± 0:40 and 4:11 ± 1:12, respectively. There was no significant phase shift (+ 13 min, p = 0.87) or change in amplitude (p = 0.11) in response to night shift. Thus, similar to the cell count rhythms (Fig. 3), the melatonin rhythm does not shift under night shift conditions (Fig. 5), supporting a role for SCN-derived signals in controlling cell trafficking (Fig. 6).

The present study demonstrates that cytokines secreted in response to ex vivo stimulation follow a 24-h rhythm that is primarily controlled by the endogenous circadian system. To our knowledge, this is the first study performed in humans kept in constant conditions that analyzed cytokine release by cells of the innate and adaptive immune systems. Importantly, we also showed that four 24-h cycles of simulated night shift were able to partially shift the phase of the cytokine-secretion rhythm, whereas the rhythm of circulating monocytes and T lymphocytes remained unaffected.

By using a CP procedure (40), we were able to reduce the masking effects that might have affected the expression of the circadian rhythms of immune functions during waking periods. This procedure allowed us to observe rhythms while minimizing the effects of confounding factors, such as changes of posture, levels of activity, rhythms of feeding, and light–dark exposure. The CP procedure includes sleep episodes scheduled at night or during the day for the baseline and night shift conditions, respectively. This allowed us to study each subject under a day-oriented and a night-oriented schedule. To perform this highly controlled protocol, we had to study the effects of night shift in simulated conditions rather than in the field. This led to the recruitment of a limited number of participants following strict screening procedures. Future field studies performed in shift workers will be required to confirm these results obtained under laboratory conditions. The protocol that we used in this study has the advantage of closely mimicking night shift work schedules compared with other studies in which the subjects were assessed under other protocols (e.g., forced desynchrony) that are useful for determining circadian parameters and the effect of misalignment, but are not as close to shift work.

The effects of sleep on human immune functions (compared with continuous wakefulness) are controversial. Some studies showed that sleep is associated with a reduction in the numbers of immune cell types (16, 51), including monocytes and T lymphocytes, whereas other studies reported an enhancement or no change (51–53). The situation is more complex for cytokines, with some affected (increase and decrease) and others unchanged, with discrepancies observed among studies (38). These differences are likely the result of different methods of assessment (e.g., whole blood versus isolated cells; stimulated versus unstimulated cytokine production) and the degree of control of confounding factors (e.g., activity levels, light exposure, food intake). In this study, we used isolated PBMCs to prevent an influence of RBCs and platelets on the levels of cytokines released (45, 46, 54, 55), and we studied cytokine secretion in stimulated cells (using both absolute levels and levels/cell). Under these conditions, the change in the timing of sleep from a day-oriented to a night-oriented schedule did not result in overall changes in the rhythm of immune cell counts or in the amounts of cytokines released with or without stimulation. However, we observed a robust shift in the acrophase (i.e., time of fitted maximum) of cytokine release.

Using a cell preparation (the PBMCs) that contains different cell types (T and B lymphocytes, monocytes/macrophages, NK cells) (50) raises the question of which cells respond to the stimuli used in the experiments and secrete the cytokines that are measured. Among PBMCs, T cells are the main PHA-responsive cells, and only NK cells can also be stimulated by PHA. However, given that only T cells secrete IL-2 and that the profiles we observed for IFN-γ and IL-2 were very similar (Figs. 2, 4, right panels), it is very likely that the response to PHA that we observed is primarily due to T cells. As for LPS, monocytes/macrophages are the only cells among PBMCs that express LPS receptor TLR4 and are able to secrete the cytokines that we studied (IL-6, IL-1β, and TNF-α). Moreover, IL-6, IL-1β, and TNF-α show identical profiles (Figs. 2, 4, left panels), so the secretion of these cytokines can be ascribed to monocytes/macrophages. Also, using PBMCs for the stimulation experiments makes it possible that other cells in the population influence the response to the stimuli. In any case, we observed very robust rhythms of cytokine release, and these rhythms show clear and distinct responses to the night shift procedure compared with baseline conditions, which clearly supports our hypothesis that night shift conditions affect the immune response.

In line with previous studies (15–18), we demonstrated that subjects living under a day-oriented schedule expressed rhythms of circulating T lymphocyte and monocyte counts that peaked during the night and day, respectively. With four 24-h cycles of night shift, the phases of these rhythms remained similar to those observed at baseline and, therefore, were not aligned to the shifted sleep–wake/feeding cycle. Hence, changing the timing of sleep and feeding for four consecutive days does not modify the rhythm of circulating T lymphocytes and monocytes. Studies in humans indicated that the number of circulating T cells is regulated by the central clock via glucocorticoid signaling (18, 36). The present study, as well as previous studies, shows that, for most night shift workers, central clock–derived signals, such as melatonin and cortisol rhythms, do not spontaneously shift in response to night shift (39, 56). Therefore, we infer that the rhythms of circulating T lymphocytes and monocytes, which were not shifted in response to night shift, might be controlled, at least in part, by the central clock.

Cytokine-secretion rhythms might be due to variations intrinsic to the immune cells (i.e., sensitivity to stimuli and capacity to produce cytokines), or they might be a direct consequence of blood cell count rhythms. To distinguish between these possibilities, we compared absolute levels of cytokine secretion with levels/cell. In subjects living under a day-oriented schedule, IL-1β, IL-6, TNF-α, IFN-γ, and IL-2 exhibited a peak of secretion occurring at night. Because this peak is seen, even when normalizing for cell numbers, we conclude that it is due to a higher responsiveness of monocytes and T lymphocytes. For IL-1β, IL-6, and TNF-α, this interpretation is further supported by the fact that monocyte counts are low at night. Accordingly, the magnitude of the nocturnal peak is even greater when values are normalized according to cell numbers, which compensates for the decrease in monocyte count at night. For IFN-γ and IL-2 secretion at baseline, the magnitude of the nocturnal peak is similar when we compare absolute values with values expressed per cell. Although cortisol and its anti-inflammatory properties were proposed to be associated with rhythmic cytokine secretion (cortisol levels are low when cytokine release is high and vice versa) (27, 38), studies in rodents (25) and humans (28) indicate a control of the cytokine secretion by the peripheral clocks located in immune cells.

In our study, switching from a day-oriented to a four-cycle night-oriented schedule led to an ∼5-h phase advance in cytokine-release rhythms. The presence and absence of a phase shift for cytokine release and immune cell numbers, respectively, support the hypothesis that the two rhythms are regulated by different mechanisms. We propose a model for the differential circadian regulation of immune cell numbers and their capacity to secrete cytokines (Fig. 6). On one hand, as the result of the absence of a phase shift in response to night shift for immune cell counts rhythms and melatonin (the current study), as well as for cortisol rhythms (56), we propose that immune cell distribution between blood and tissues is primarily regulated by the central clock, via neuronal and/or humoral regulators (e.g., cortisol) acting on adhesion molecules and chemokines (18, 37, 57). On the other hand, due to the presence of a phase shift in cytokine-secretion rhythms in response to night shift, we infer that the sensitivity of immune cells to stimuli (e.g., LPS or PHA in our study) or their capacity to produce cytokines in response to these stimuli is dependent on the local clocks intrinsic to these immune cells. These immune cell clocks could act in two ways: by regulating the expression of molecules important for the synthesis and secretion of cytokines by these cells and by gating the response to the regulatory effects of cues from the central clock. A local direct regulation of immune cell functions (25, 28) and a gating of the response to systemic cues (14, 58, 59) were shown to occur in rodents. However, we cannot exclude that the shift in the sleep–wake/feeding cycle might have contributed, to some extent, to the changes observed in the cytokine-release rhythm.

Importantly, we believe that this differential regulation of distinct aspects of the immune system can lead to a state of desynchronization between immune parameters under conditions of shift work. One of the main roles of immune rhythms is the coordination of the different immune functions with each other and with other physiological functions. We demonstrate that, following night shifts, the relative phase of the rhythms of cytokine secretion and immune cell counts is disturbed, and these two rhythms no longer demonstrate the conventional phase relationship with the sleep–wake cycle and with the external environment. Thus, night shift leads to a disruption of the rhythmicity of the immune system, probably as a consequence of the desynchronization between the central and peripheral clocks, as we described previously when comparing central markers (e.g., cortisol and melatonin rhythms) with peripheral markers (e.g., clock gene expression in PBMCs) (35, 39, 40). Such a disruption might play a role in the increased incidence of various medical conditions in shift workers, such as the increased risk for infections, autoimmune diseases, cancer, and cardiovascular and metabolic disorders (2, 8–13). Interestingly, an increased prevalence of autoimmune conditions, such as multiple sclerosis and autoimmune thyroid disorders, was associated with shift work (11, 12, 60–63). In autoimmune thyroid disorders, there is an increase in the recruitment of Th1 lymphocytes in thyroid tissue (based on changes in cell trafficking between blood and tissues), leading to an increase in the production of IFN-γ and TNF-α, further stimulating the recruitment of Th1 lymphocytes and initiating an amplification feedback loop (63). Because a cascade of events is implicated in this process, the timing of each of these events might be important; thus, it is possible that the lack of temporal coordination among the different immune functions seen during night shifts creates a state in which this amplification phenomenon could be favored. The same could be true in multiple sclerosis, in which cascades of events were described involving Th1 lymphocytes and IFN-γ (61), as well as IL-2 and IL-7 (60).

In conclusion, this is the first study, to our knowledge, demonstrating that night shift leads to the disruption of rhythmic circadian immune functions in humans. We already know that some medical conditions reported in shift workers (e.g., autoimmune diseases, obesity, atherosclerosis) are related to alterations in immune functions (64, 65). The present study reveals the possibility that some of these alterations are associated with or result from a dysregulation of immune clocks and other clocks that are part of the circadian system.

We thank Drs. S. Rhéaume and A. Solignac, as well as A. Azzoug, M.R. Guertin, and A. Tchomgang (research nurses) for medical supervision; Dr. A.S. Zandi, Dr. E.A. Begum, and W.H. Yeh for help during blood-drawing sessions, all of the research assistants who helped during the experiment; all of the subjects for their participation; and Dr. N. Labrecque for discussions and critical review of the manuscript.

D.B.B. and P.B. consult for AlphaLogik Consultants Inc., a D.B.B.-founded/owned company. The other authors have no financial conflicts of interest.