Metabolite profiles and the risk of metabolic syndrome in early childhood: a case-control study

This study was conducted among children from the Family Atherosclerosis Monitoring In earLY life (FAMILY) study [24]. FAMILY is a prospective birth cohort study involving 857 predominantly white families that include 901 children recruited from Hamilton, Ontario, between 2002 and 2009 with a 10-year follow-up. Of the 676 children who completed a follow-up 5-year visit, 557 had complete anthropometric and clinical data and complete records of maternal gestational diabetes mellitus (GDM). Fasting serum specimens were available for 491 children who were considered for this study (consort diagram Additional file 1: Fig. S1).

Data collection occurred at the 5-year follow-up visit through questionnaires, physical measurements, and laboratory analysis of biospecimens as published previously [25]. Fasting blood samples were collected and serum was fractionated within 2 h from collection according to standard protocols, stored at − 80 °C and shipped on dry ice [25]. Cholesterol (LDL, HDL, and total cholesterol), triglycerides, and glucose from the fasting serum samples were analyzed at the McMaster Children’s Hospital and Hamilton Health Sciences Central Research Laboratories following standardized analysis protocols [24, 26].

Cohort-specific percentiles were calculated for waist circumference, systolic blood pressure, and serum fasting glucose using the most complete data sets for each variable from the total number of children who attended the 5-year visit. Standardized BMI-for-age z-scores were derived based on WHO child growth standards [27]. Sex-specific 90th percentile thresholds for waist circumference, blood pressure, and glucose were 57.0 cm, 112 mm Hg, and 5.0 mmol/L in females and 55.4 cm, 110.3 mm Hg, and 5.1 mmol/L in males, respectively. The 50th percentile threshold for waist circumference was 51.1 cm in girls and 51.2 cm in boys. Children were classified as having a higher risk of MetS and referred to as cases if they had one or more of the following abnormalities: (1) z-BMI ≥ 75th percentile and waist circumference ≥ median, (2) z-BMI ≥ 75th percentile and systolic blood pressure ≥ 90th percentile, (3) fasting serum glucose ≥ 90th percentile, and (4) maternal GDM diagnosis. Two hundred twenty-eight children who met these criteria were classified as “MetS cases” and were age- and sex-matched 1:1 to 228 controls (Additional file 1: Fig. S1).

Among these 456 children, we derived a continuous MetS risk score based on percentile curves created for 18,745 children aged 2.0–10.9 years of the European IDEFICS (Identification and prevention of Dietary-and lifestyle-induced health EFfects in children and infantS) cohort as reference [6, 28]. The MetS score was calculated summing sex- and age-specific z-scores according to the following formula by Ahrens et al. [6]: IDEFICS MetS score = z _{waist circumference} + (z _{systolic blood pressure} + z _{diastolic blood pressure})/2 + (z _{triglycerides} − z _HDL)/2 + z _{fasting glucose}. For more details on the IDEFICS MetS definition and score derivation, refer to Additional file 2: Supplemental Methods.

A validated, high-throughput method based on multisegment injection-capillary electrophoresis-mass spectrometry (MSI-CE-MS) [29, 30] was used for the analysis of polar ionic metabolites consistently measured in serum filtrate samples on an Agilent 6230 time-of-flight MS with a coaxial sheath liquid Jetstream electrospray ion source equipped to an Agilent G7100A CE (Agilent Technologies Inc., Mississauga, ON, Canada). This multiplexed separation platform takes advantage of a serial sample injection format of 13 samples [31, 32] within a single CE run including a pooled quality control (QC) sample prepared by combining equal aliquots of serum samples from all study participants (n = 456) for rigorous QC and data filtering procedures. Serum sample pre-treatment and data acquisition have been previously described in detail [30]. An iterative data workflow based on multiplexed injections pattern was used to reject spurious signals, redundant peaks, and background ions when performing nontargeted metabolomics based on analysis of a pooled serum sample that also served as QC for assessing technical precision [30, 31]. Fifty-eight serum metabolites were consistently measured in over 75% of the samples and satisfied QC criteria of a variance under 30% when using MSI-CE-MS under two configurations with positive and negative-ion mode detection. Fifty-two metabolites were unambiguously identified (level 1) and subsequently quantified using a calibration curve, where ion responses were normalized to a single internal standard. Six unknown serum metabolites were annotated based on their characteristic accurate mass and relative migration time under positive or negative ion mode. 2-hydroxybutyric and 3-hydroxybutyric acids were not baseline resolved and were reported as the sum of both isomers (which led to a higher variance of 40%). Metabolite combinations and ratios frequently used in the literature were derived for branched chain amino acids as the sum of leucine, isoleucine and valine, and glutamine/glutamate ratio. Any non-detects were replaced by a value that was 1/5 the detection limit, set to the smallest value in the data set. Finally, a robust QC-based batch correction algorithm was used to adjust for long-term signal drift of the mass spectrometer [33]. This stringent data pre-processing approach ensured that only fully authenticated serum metabolites were included to reduce false discoveries. Principal component analysis was used for data visualization (i.e., data trends/outlier detection) of the serum metabolome using MetaboAnalyst 4.0 [34]. Additional file 3: Table S1 lists the identification of all detected metabolites based on m/z:RMT:mode, molecular formula, the mean concentration measured, the % CV, and the % of data completeness for each metabolite.

Descriptive statistics are presented for the overall sample (n = 456) and by controls (n = 228) and cases (n = 228). Continuous variables including anthropometrics, blood pressure, and laboratory measurement are presented as mean and standard deviation, and categorical variables are presented as counts and percentages. Most metabolites were not normally distributed; thus, all metabolites were natural-log transformed and auto-scaled (mean-centered then standardized) for further analysis. Significant metabolite associations with MetS were tested using multivariable logistic regression as follows. To build this model, we identified a set of covariates known from the literature to be important predictors of MetS, which included diet quality, physical activity, time spent outside, screen time, night sleep, breastfeeding, socioeconomic status, and maternal education. Among the 456 study participants, all covariates were available for 433 children (95%). As a first step, we tested each of these in a series of univariate regression models against MetS; screen time was the only covariate which was associated with MetS with a p = 0.036, so this was retained as a covariate in all future models. Next, we entered all the previously identified predictors into a forward stepwise selection model to test if their addition on top of screen time explained more variance than screen time alone; from this procedure, screen time and diet quality were retained. Our final maximally adjusted analysis kept these two variables along with physical activity, night sleep, social disadvantage index, and child’s age and sex to improve direct comparability with previous work in the field. Additionally, we tested the MetS-metabolite associations in intermediate adjusted models to confirm the consistency of results (Additional file 4: Table S2). Next, to assess the association between the metabolites and the continuous MetS score, we used multivariable linear regression adjusting for the same covariates as in the maximally adjusted logistic model described above. Statistical significance for the multivariable logistic regression analyses was set at p < 0.05. Bonferroni correction for multiple hypothesis testing was considered for the multivariable linear regression model with a p value threshold < 0.0008 (0.05/60 tests). All analyses, tables and graphs were completed in R (v3.6.3; R Foundation: A Language and Environment for Statistical Computing). Finally, a pathway analysis was conducted by significant metabolites’ names to understand which pathways the identified metabolites may affect using MetaboAnalyst 5.0 pathway analysis tool [34] in conjunction with a literature review relevant to MetS and related conditions (Additional file 5: Table S3).

Characteristics of the study population included in this analysis (n = 456) are shown in Table 1. The mean age of the participants was 5.15 years; 50.4% were girls. Various measures for adiposity and values for glucose, systolic blood pressure, and diastolic blood pressure were higher in cases than in controls whereas serum lipids were not (Table 1). 57.4% of MetS cases had one risk factor, 27.6% had two risk factors, and 15% had more than two risk factors (Fig. 1A and Additional file 6: Fig. S2).

According to the IDEFICS criteria the proportion of children at 5 years with MetS was 2.9% (95% CI 2.0–3.7%) [6]. The IDEFICS definition is met when the value of three or more risk factors exceeds the 90th percentile for waist circumference, systolic OR diastolic blood pressure, HOMA-insulin resistance OR fasting glucose, and triglycerides OR is lower than the 10th percentile for HDL cholesterol. The prevalence of 2.9% would require a very large sample size to conduct association studies and therefore we employed a case-control design for the primary analysis.

The age- and sex-specific continuous MetS risk score and its individual components are presented in Table 2. The mean IDEFICS MetS score for cases was 1.00 and for controls was − 0.90, which further validated our selection criteria of cases and controls. Overall, the study participants at five years followed a normal distribution of the IDEFICS MetS score around a mean of zero (Fig. 1B).