Social Jetlag Linked to Sleep Quality and Heart Regulation During Sleep in Young Healthy Men

Participants with regular weekly schedule were recruited via advertisements in social media (Facebook groups) and mailing list of university students. Initially 35 students participated in the study, but two of them were excluded either due to experiencing high emotional stress or for lack of appropriate cooperation. Therefore, data of 33 subjects were involved in the final analysis. The sample was homogenous with respect to age (mean ± SD of 23.2 ± 1.5 years), BMI (mean ± SD of 22.8 ± 2.5), and social situation (university students in Budapest). All subjects were healthy and none of them were taking medication. None of the subjects had a history of sleep disorder or cardiovascular problem and neither one of them experienced jetlag in the last month prior to the study. In a short interview, participants were asked about smoking habits, caffeine consumption, physical activity, blood pressure, and cholesterol levels. Answers did not indicate notable cardiovascular risk factors in any participants. Participants were asked not to undergo extreme physical activity, and to avoid drinking alcohol during the study week. Following the measurements, participants were interviewed about any unexpected events (high psychical and physical stress or any health problems). The study protocol was approved by the Semmelweis University Regional and Institutional Committee of Science and Research Ethics (Ethical approval 170/2016). Participants gave written consent prior to entering the study and were paid for their participation. The study took place in winter (2016/2017) and in the first half of spring (2017) and was paused for 2 weeks after changing to daylight saving time. The number of participants with lower and higher SJL was similar in every month. The study for each subject lasted for a week, during which the participants performed their daily activities according to their usual habit, including waking to an alarm clock on workdays but waking on their own on free days. On a workday (Wednesday) and a free day (Saturday) participants wore an ambulatory heart rate recorder (Actiheart, CamNtech Ltd., United Kingdom) during the night and filled out a sleep quality questionnaire upon waking on the next day. On the days of HRV detection, subjects collected urine samples in the evening before going to bed and in the morning directly after getting up.

To recruit participants with different SJL and regular weekly schedule, students were asked to fill out the Hungarian version of the Munich Chronotype Questionnaire (MCTQ) (Haraszti et al., 2014). Chronotype was assessed as the midpoint of sleep on free days (MSF) corrected for oversleep on free days (MSFsc), and SJL was calculated by subtracting the midpoint of sleep on workdays (MSW) from MSF, based on Wittmann et al. (2006). During the week of the study each participant kept a sleep diary, based on which the SJL characteristic for the study period was calculated. The mean SJL ± SD was 92.1 ± 52.9 min, which fits well with previously reported SJL values for this age group (Wittmann et al., 2006; Haraszti et al., 2014; Pilz et al., 2018). The median SJL (93 min) was used as a cut-off value to divide the participants into two groups with either lower (SJL ≤ 93 min) or higher SJL (SL > 93 min). Using this cut-off value resulted in a 10-minutes gap between the two groups. There was no significant difference between the two groups regarding age (23.3 ± 1.9 and 23.1 ± 0.9 years mean ± SD for the lower and the higher SJL group, respectively) and BMI (22.1 ± 2.7 and 23.6 ± 2.1 years mean ± SD for the lower and the higher SJL group, respectively). Sleep onset and waking time on the days of HRV detection were estimated by the Actiheart software. These time points were controlled by comparing them with data from the sleep diary. The estimated sleep onset and waking times were used to calculate exact SJL for the days the measurements were performed. This SJL value was used to investigate correlation between SJL and HRV parameters.

Urine samples collected by the subjects in the evening and morning were stored at −20°C before analysis. Urinary 6-sulfatoxymelatonin, a stable melatonin metabolite, was measured by competitive ELISA according to the instructions of the manufacturer (IBL International). The creatinine concentration of the samples was determined by enzymatic assay (Diagnosticum Zrt., Budapest) to obtain a normalized 6-sulfatoxymelatonin/creatinine ratio (nmol/mmol).

The Actiheart monitor was used for the recording of heart rate. Prior to the measurements, participants came into the laboratory where a signal test was performed along with the programing of the device to start recording in the evening. On the next day participants brought the device back for data collection and recharge, before taking it again for the weekend. The Actiheart monitor collects inter-beat-interval (IBI) data, which can be further processed and used for the calculation of HRV parameters. Studies comparing Actiheart and standard ECG recordings did not find any significant difference in IBI detection (Brage et al., 2005; Barreira et al., 2009; Kristiansen et al., 2011). All HRV analyses were performed according to the standards of measurements (Malik, 1996) and the recent recommendations by Laborde et al. (2017). First, reliability of raw data was checked by visual inspection. To ensure a consistent artifact detection and correction, the ARTiiFACT software was used (Kaufmann et al., 2011). According to the international guidelines the data were decomposed into 5 min long segments for analysis of HRV parameters in both the time and frequency domains (Malik, 1996). Segments with more than 10% artifact ratio were excluded. Excluded segments and loss of electrode contact resulted in slightly varying sample sizes in the different analyses. Parameters calculated for the 5 min long segments were averaged for either 1.5- or 2-h blocks, depending on the analysis performed, as indicated in the text and legends.

In the time domain the standard deviation of the normal to normal interval (SDNN), the root mean square of successive differences (RMSSD), and the percentage of successive normal to normal intervals differing by more than 50 ms (pNN50) were calculated using the ARTiiFACT software. For the analysis of HRV in the frequency domain the DADiSP 6.7 software (DSP Development Corp., United States) was used. IBI data were first resampled to 4 Hz using cubic spline interpolation, detrended and tapered with a Hanning window. Power spectral density (PSD) was then calculated with mixed-radix Fast Fourier Transformation, yielding 600 frequency bins in the range of 0–2 Hz. From the average PSD of the successive 5 min segments very low frequency (VLF, 0.003–0.04 Hz), low frequency (LF, 0.04–0.15 Hz), and high frequency (HF, 0.15–0.4 Hz) absolute powers were calculated by numerical integration of the respective frequency ranges. Normalized units of LF (LFnu = LF/LF + HF) were also calculated. Bin-wise and band-wise PSD values were log normalized (natural base logarithm) before the statistical analyses. In our subjects respiratory rate, determined on the basis of the peak in the high frequency region of the FFT plot, was between 0.2 and 0.3 Hz (12–18/min). At this respiratory rate (resting state) HF corresponds to vagal activity, and correction of the HRV parameters for respiration is not required (Larsen et al., 2010; Bertsch et al., 2012; Laborde et al., 2017).

For the assessment of subjective sleep quality, we used the Groningen Sleep Quality Scale (GSQS) (Meijman et al., 1988). The Hungarian version of the questionnaire translated from English was validated by Simor et al. (2009). The participants filled out the GSQS questionnaire directly upon waking following the nights with HRV measurements. The questionnaire contains 15 short, true or false questions about the previous night’s sleep. The first question is not evaluated, therefore the GSQS is scored between 0 and 14, a higher score indicating lower quality of sleep.

Normality of data (sleep times, HRV measures, questionnaire scores, and 6-sulfatoxymelatonin levels) was assessed by the Kolmogorov-Smirnov test. 6-sulfatoxymelatonin and bin-wise spectral HRV data showed non-normal distribution. For the analysis of the former the non-parametric Mann-Whitney U test was used. Bin-wise spectral HRV data was log normalized to achieve normality. If normal distribution was present, we used parametric statistical procedures as follows. Workday-free day differences were assessed by paired sample t-tests. Group differences were tested by two sample t-tests. For the analysis of HRV data in consecutive time blocks, repeated measures analysis of variance (ANOVA) was performed. Fisher’s LSD test was used as post hoc test. In addition, the association of SJL with sleep quality and HRV was examined by Pearson’s correlation. Log-normalized bin-wise spectral data was expressed as a difference (free day – workday) over the frequency axis (0–0.4 Hz). One-sample t-tests were run in order to test if the differences were significantly deviating from the null hypothesis (difference = 0). Inflation of Type-I error in this case was handled by the procedure of descriptive data analysis (Abt, 1987). Statistical significance threshold was set at p < 0.05. If not otherwise indicated, data are presented as mean ± standard error of the mean (SEM). Statistical analyses were performed using Statistica software version 13.2 (StatSoft, Tulsa, OK, United States).

To control that the participants’ actual sleep schedule during the study week was similar to their usual sleep timing, data from the sleep diary recorded during the study week were compared with data from the MCTQ filled in prior to the study. Reliable correlation for both chronotype (indicated by MSFsc) and SJL were obtained (Supplementary Table S1). Participants were divided into two groups based on the median SJL of the whole sample. Table 1 shows sleep data (timing and duration) for both the workday and free day when subjective sleep quality and HRV were assessed. On workdays the two groups displayed no differences in sleep timing and duration, whereas on free days the group with higher SJL had a significantly later bedtime, sleep onset, and waking time than the other group. Both groups had longer sleep duration on free day compared to workday. As a factor of homeostatic sleep regulation we calculated the time spent awake on the days before the HRV measurements. However, neither a group nor a group x day effect was obtained (Supplementary Table S2).

As a marker of circadian rhythm, urinary levels of 6-sulfatoxymelatonin, the stable metabolite of melatonin, were determined in evening, and morning urinary samples. The marked difference in the levels of 6-sulfatoxymelatonin between the morning and evening samples (mean individual morning/evening ratios ± SEM, 6.75 ± 0.75) confirmed the rhythmic physiology of our participants. However, neither the evening nor the morning excretion of 6-sulfatoxymelatonin was associated with SJL (). Supplementary Table S3

To assess how sleep timing affects autonomic cardiac control during sleep, individual HRV parameters were compared between workdays and free days. Average heart rate (HR) for the whole sleep period did not differ between the two days in either group of participants. However, in the group with higher SJL, SDNN, a basic measure of HRV, was significantly higher on the free day than on the workday (Table 2). To analyze HRV in the course of sleep, we calculated mean HRV parameters for consecutive 1.5-h sleep blocks from the time point of falling asleep. As the minimum sleep length on both days was 6 h, four sleep blocks were analyzed for each night and the data sets of workdays and free days were compared. Figure 1 shows data for RMSSD and pNN50, the time-domain parameters most commonly used for the characterization of vagal control of heart function (Malik, 1996; Laborde et al., 2017). Repeated measures of ANOVA with three factors (group, day, and sleep block) revealed no significant effect of group or day, only sleep block showed significance among the main factors (Supplementary Table S4). Furthermore, a day × sleep block × group interaction was obtained (p = 0.015; F(3,72) = 3.74) for pNN50. While the HRV values nearly overlapped in the group with lower SJL, in the first two time blocks data of subjects with higher SJL showed intra-individual differences as their values were significantly lower on the workday than on the free day (Figure 1A) (higher SJL, workday, first block: 31.3 ± 4.1; second block: 38.9 ± 4.2; free day, first block: 47.3 ± 4.7, second block: 46.8 ± 4.4). Post hoc test did not reveal difference between the two groups in any sleep block. RMSSD showed a similar tendency, as values diverged in the first two blocks in the group with higher SJL, however, the interaction did not reach the level of significance [p = 0.059; F(3,72) = 2.59] (Figure 1B). To exclude that the HRV difference was counteracted by an opposite HRV difference after the sleep period previously analyzed (first 6 h), we performed a further analysis comparing the average HRV in the last 1.5 h of sleep on the workday and in the corresponding sleep segment (same time form sleep onset) on the free day (Figures 1A,B). However, the end-of-sleep HRV data did not display differences between workday and free day.

Next, we addressed the relationship between SJL and the HRV difference between the workday and the free day. Analyzing HRV parameters for the first 3 h, corresponding to the first two sleep blocks in Figure 1, a positive correlation between SJL and the free day-workday differences in pNN50 was obtained (Pearson’s r = 0.39, p = 0.033, n = 30). When the analysis was restricted to the first 2 h, both RMSSD and pNN50 differences were positively correlated with SJL (r = 0.365, p = 0.047, n = 30 for RMSSD and r = 0.4135, p = 0.023, n = 30 for pNN50) (Figure 2A).

In the next analysis, we compared the beginning of sleep for a duration equal to the difference of sleep onset between the 2 days, referred as shift of sleep onset (Figure 2B). Data of all participants with more than 25 min shift were involved in this comparison. Both RMSSD and pNN50 parameters were significantly lower on the workday than on the free day (54.7 ± 5.3 and 67.5 ± 6.0, p = 0.016 for RMSSD, and 29.6 ± 4.1 and 39.8 ± 5.0, p = 0.009 for pNN50).

To further characterize the HRV differences in the two groups of participants, frequency domain analyses were also performed. When the PSD of HRV was plotted for the first 2 h of sleep for the work and free day, the curves overlapped in the group with lower SJL, whereas they differed in the group with higher SJL (Supplementary Figure S1). We calculated the individual bin-wise free day – workday differences in the PSD. The difference was significant in several frequency ranges in subjects with higher SJL, but not in those with lower SJL (Figure 3A). The frequency band-based analysis of PSD showed that both LF [0.82 ± 0.02 and 0.85 ± 0.02 (lnms²) on workday and free day, respectively, p = 0.027] and HF [1.45 ± 0.06 and 1.58 ± 0.06 (lnms²) on workday and free day, respectively, p = 0.031], but not VLF [0.29 ± 0.01 and 0.30 ± 0.01 (lnms²) on workday and free day, p = 0.103] bands were involved in the free day elevation of HRV of subjects with higher SJL (Figure 3B). In contrast, no differences between work and free day were found in the group with lower SJL [Figure 3B; VLF, 0.28 ± 0.01 and 0.29 ± 0.01; LF, 0.79 ± 0.02 and 0.81 ± 0.02; HF, 1.40 ± 0.07 and 1.43 ± 0.06 (lnms²) on workday and free day, respectively]. To approach sympathetic modulation, normalized units of LF were also calculated. LFnu powers were relatively low as expected during sleep in healthy young people (Brandenberger et al., 2003; Stein and Pu, 2012). Workday and free day LFnu powers did not differ significantly in either group (Figure 3B).

To assess sleep quality, we used the GSQS where a higher score indicates worse quality of sleep. We found that SJL was positively associated with the GSQS score on workday, whereas no association was obtained on the free day (Supplementary Figure S2). While participants in the group with lower SJL reported similar sleep quality for both days, participants with higher SJL had a significantly worse sleep quality on the workday than on the free day (p = 0.027) (Figure 4), and their workday scores were also significantly higher than those of the other group (two-sample t-test, p = 0.043).