Subjects

Healthy children (7–12 years; n = 432) from three schools in a Mediterranean area of Spain were recruited between October 2014 and June 2016 (ClinicalTrials.gov ID: NCT02895282), as previously described [11]. Approval for this study was obtained from the University of Murcia Ethics Committee (ID: 1868/2018). All procedures followed the ethical standards of the institutional and national research committees and complied with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The present study is a secondary analysis of the same cohort of children studied in our previous works [10, 11, 16,17,18].

Obesity-related trait measurements

Body weight, body fat percentage, and waist circumference (WC) were measured in the whole population (n = 432) on the first day of the week of the study and at the same time in the morning, as we previously published [11]. Body weight was measured in barefooted children wearing light clothes using a digital scale accurate to the nearest 0.1 kg. Height was determined using a portable stadiometer (rank: 0.14–2.10 m). The children were positioned upright, relaxed, and with their heads in the Frankfort plane. These data were used to calculate body mass index (BMI) according to the formula weight (kg)/height2 (m2). BMI was further transformed to the BMI Z-score according to the World Health Organization growth reference [19]. Total body fat was determined by bioelectrical impedance using TANITA TBF-300 equipment (Tanita Corporation of America, Arlington Heights, IL) [11]. Several cardiometabolic traits, such as glucose, triglyceride, and cholesterol levels, and inflammatory markers such as C-reactive protein, immunoglobulin A, interleukin (IL)-8, IL-1b, tumor necrosis factor-alpha and monocyte chemotactic protein 1 were determined from serum samples obtained from a subset of 79 children in the morning under fasting conditions as previously described [11].

Global circadian score determination

An already developed formula was used to determine the GCS for each child [11] (Supplementary Table 1). The score for each factor was calculated by multiplying each variable by its eigenvalue. Then, each factor score was multiplied by the percentage of the variance in the total information explained by that factor. Factors 1 and 2 explained 28% of the variance and were loaded mainly by TAP characteristics, factor 3 explained about 10% (loaded by cortisol measures), and factors 4 and 5 together explained about 13% of the variance and loaded on the timing of food intake (breakfast and dinner). The five factors together explained 50% of the variance in total information.

Factors and variables included in the GCS

A description of each variable included in the formula is presented in Table 1. Factors 1 and 2 from the GCS included variables derived from TAP. To determine TAP, children wore a wristwatch in their non-dominant hands during the seven days of the study. This wristwatch integrates two sensors: a temperature sensor [20] (Thermochron iButton DS1921H, Dallas, Maxim, Dallas, TX, USA), which is programmed to collect information every 5 minutes; and an accelerometer sensor (G Acceleration Data Logger UA-004–64; Onset Computer, Bourne, MA, USA), which measures physical activity and body position rhythms to record data every 30 seconds [14]. The position was expressed in degrees (°), and activity was expressed as the change in degrees per minute (Δ°/minute). TAP was calculated as previously described [14]. From TAP, different parameters were derived (Table 1) using an integrated package for temporal series analysis (Kronowizard; https://kronowizard.um.es/kronowizard; Chronobiology Laboratory, University of Murcia, Murcia, Spain, 2015). Factor 3 from the GCS comprised salivary cortisol measurements. Saliva samples were collected on the same day at three different times using salivates (Sarstedt, Barcelona, Spain): before breakfast (9:00), before lunch (14:00), and after dinner (23:00). The saliva concentration of cortisol was determined by radioimmunoassay (IZASA, Barcelona, Spain). Factors 4 and 5 from the GCS included dinner start time and finish time and breakfast start time and finish time determined by 7-day dietary records completed by children and parents. Melatonin and lunch timing were not included in the GCS because, in the factor analysis performed for developing the GCS, melatonin and lunch timing did not appear within the first five factors that accounted for about 50% of the variance.

Table 1 Description of each variable included in the global circadian score formulaLifestyle variablesSleep characteristics, individual chronotype, and related variables

We determined sleep timing, sleep duration, and daytime napping timing and duration using 7-day sleep diaries; children and parents recorded the wake/sleep cycle, which has been demonstrated to be a convenient tool for assessing sleep quality and duration [15, 21]. Individual chronotypes were subjectively determined using an age-appropriate Spanish version of the Munich ChronoType Questionnaire (MCTQ) [22], while melatonin was used as an objective proxy of individual chronotypes. Melatonin concentrations were determined via radioimmunoassay (IBL, Germany) in saliva samples collected at night (01:00) and before lunch (14:00). Social jet lag, which represents different sleep schedule behaviors between free days and schooldays, was measured as the difference between mid-sleep on free days and mid-sleep on schooldays. We considered that children had social jet lag when there was more than a two-hour difference in the midpoint of sleep between free days and school days [23].

Food intake

Energy and macronutrient distributions across meals and the percentage of energy provided by macronutrients were determined by 7-day food diaries adapted for this age group [24], which specify the timing of food intake, type of food and amount/weight/quantity of food eaten [15]. Total energy intake and macronutrient composition were analyzed with a nutritional evaluation software program (Grunumur 2.0 8) [25] based on Spanish food composition tables [26]. Total morning intake was defined as the sum of every intake in the morning, including lunch. Every intake after lunch, including dinner, was considered total evening intake. The timing of meals was determined by averaging every meal timing during the seven days of the dietary records completed by the children and parents.

Physical activity

The average physical activity was determined by an accelerometer (G Acceleration Data Logger UA-004–64; Onset Computer, Bourne, Massachusetts, USA) and was calculated as the average accumulative degree change in three-axis tilt per minute during the most active ten hours during the wake period (i.e., nonsleep time), which correlates with motor activity measured by the wrist-worn Actiwatch [11, 27]. We classified physical activity levels by dividing the average physical activity into tertiles (low, medium and high). The day–night physical activity contrast was determined as the relative amplitude of the activity rhythm.

Light exposure

We evaluated light exposure near the face (eyes) via a pendant luxmeter in the neck (HOBO Pendant Light Data Logger, UA-002–64; Onset Computer, Bourne, Massachusetts, USA). The instrument was programmed to collect light information continuously every 30 seconds for 7 days. We calculated the lux logarithm collected every 30 seconds to assess the light average. We further calculated the amplitude, percentage of rhythmicity (PR), interdaily stability (IS), and circadian function index (CFI) of the light data rhythm. The description and calculation of these variables are included in Table 1. Children were instructed to wear the pendant over their clothing. When sleeping, they had to leave it on the bedside table. Cortisol and melatonin salivary determinations were performed on Sundays after the 7-day recording of temperature, activity, position, and light was completed to avoid these determinations being affected by the melatonin sampling.

Statistical methods

Differences in GCS were analyzed among children stratified into groups according to three categories of obesity (children with normal weight, overweight and obesity) according to the sex- and age-specific BMI cutoff points proposed by the International Obesity Task Force [28] via ANCOVAs. When significant, we explored differences among groups using post hoc analysis (Bonferroni). The median GCS was used as a cutoff point for classifying children according to good or poor circadian health. We fitted multinomial logistic regression models to estimate the odds ratios and 95% confidence intervals of having poorer circadian health (as an outcome) in the presence of obesity/overweight (combination of overweight and obesity categories [28]) or high WC (divided according to the median, considering as "low WC" a WC lower or equal to 64 cm, and "high WC" a WC greater than 64 cm).

We tested for linear regression models of the GCS and its components with BMI, BMI Z-score, and body fat percentage. To understand the impact of lifestyle on the association between GCS and obesity, we explored the significant interactions between the GCS and lifestyle factors (already identified) and BMI. To identify those lifestyle factors, we previously tested linear regression models to determine whether there were significant associations between GCS and lifestyle factors or between the GCS and obesity traits (BMI, BMI Z-score, and body fat percentage). Because no significant interaction effect was found between sex and GCS for BMI, we performed statistical analyses of the total number of children studied without separating by sex. All analyses were performed using SPSS version 25.0 (SPSS, Chicago, Illinois, USA). P < 0.05 was considered to indicate statistical significance and is presented in the tables and figures in bold. The results of borderline statistical significance (considering those with P < 0.1) are presented in italics in all tables and figures.

We adjusted all analyses, including ANCOVA, multinomial logistic regression, linear regression, and interactions according to sex, age, race, and academic year. We chose those variables as a way to control for sociodemographic status.