The current case-control study was nested in the Haguenau cohort. -

Description of the Haguenau cohort

Individuals are identified from the Haguenau community-based cohort derived from a population-based registry of the metropolitan area of the city of Haguenau in France. ²¹ Briefly, this registry included information about all pregnancies and deliveries occurring in the maternity unit of the hospital (1971-1985) with 80% degree of completeness. The SGA group (n=734) included singleton individuals born between 32 and 42 weeks of gestation with birth weight <10 ^th percentile for sex and gestational age, whereas the AGA group (n=886) was made of similar individuals born at the same gestational age, with birth weight between the 25 ^th and the 75 ^th percentiles and who were the first babies in the registry born immediately after a baby born SGA. As previously described, the mean ages at the two observations (Baseline and Follow-up) were 22 and 30 years respectively, with a participation rate of 80.7%. ^{18, 22} Individuals were all born full-term and sex distribution did not differ between the AGA and SGA groups. Clinical characteristics of the study population at birth were previously described in details ²³ : individuals born SGA were lighter (2.63±0.3 vs 3.37±0.27 kg), shorter (47.7±2 vs 50.3±1 cm) and thinner (11.6±1.1 vs 13.3±0.8 kg/m ²) at birth than individuals born AGA. The study protocol was reviewed and approved by the ethical committee of the University of Paris-St Louis, and all subjects gave written informed consent. -

Subject selection

For this study, a sub-population is selected among those with available serum samples, anthropometric and clinical data at baseline and follow-up (around 10, years later) for a case-control study design. Incident cases of MetS were selected at the end of the follow-up (at 30 years of age), from subjects free of MetS at inclusion (at 20 years). For defining MetS, the International Diabetes Federation (IDF) criteria is used: abdominal obesity and at least two of the four following parameters: hypertension, fasting hypertriglyceridemia, hyperglycaemia and low high-density level (HDL)-cholesterol concentrations. ^{24, 25} The control group, free of MetS at follow-up and inclusion, is first matched for sex. Because of a large age heterogeneity among subjects, it is not possible to match controls on exact age values. Therefore, two criteria are used for matching controls: age class at baseline (age≤20 y.o.; age ≥21y.o.) and the time difference between the two observations (5, 6, 7, 11, 12 and 13 years).

These cases are selected independently from SGA and AGA subjects, and the controls are drawn by random matching within strata. Of the 1,227 eligible individuals (participants present at follow-up, free of MetS at baseline), 93 developed MetS at 30 years, 49 being SGAs and 44 AGAs. From the 186 consequently selected subjects, 9 pairs (Case/Controls) are excluded because of the serum sample availability. Figure 1 summarizes the final subject selection and sample size of the subgroups.

Beyond clinical parameters related to MetS and SGA status, other phenotypic data are available on selected subjects such as insulin area under the curve following an oral glucose tolerance test, homeostasis model assessment of insulin resistance, waist to hip ratio, physical activity, and characteristics of parents at birth (body mass index, smoking during pregnancy).

Untargeted metabolomics will be performed following a slightly modified procedure described in Pereira et al. ²⁶ Briefly, serum samples (100 μL) will be deproteinized using 200 μL of cold methanol (ULC/MS grade, Biolsolve, catalogue number 136841); after centrifugation and evaporation under nitrogen, the dry residues were re-dissolved in 150 μL of 50/50 (v/v) acetonitrile (ULC/MS grade, Biolsolve, catalogue number 012041)/water containing 0.1% formic acid (ULC/MS grade, Biosolve, catalogue number 069141). Pooled quality control samples will be prepared by mixing 20 μL from each of the serum samples and prepared similarly. Metabolic profiles will be then determined using an ultra-high performance liquid chromatography (U3000 UHPLC system, Thermo Scientific) coupled with quadrupole-time of flight mass spectrometer (QToF, Impact HD2, Bruker), equipped with an electrospray source. Briefly, 5 μL of sample were injected into the UHPLC/QToF system and separations were carried out using an Acquity HSS T3 column (Waters) at 30 °C and a flow rate of 0.4 mL/min. Data will be acquired in positive and negative ion modes with a scan range from 50 to 1,000 mass-to-charge ratio (m/z). Samples will be randomized within the analytical sequence based on a Williams Latin Square strategy defined according to the main factors of the study. The stability of the analytical system will be monitored using pooled samples as quality control (QC), injected one time at the beginning of each sequence and then after each set of ten samples.

Data will be processed under the Galaxy web-based platform Worflow4Metabolomics, ²⁷ using first XCMS, ²⁸ followed by quality checks and signal drift correction according to the algorithm described by Van der Kloet et al. ²⁹ based on the use of pooled QC samples, to yield a data matrix containing retention times, masses and peak intensities that will be corrected for batch effects. These steps include noise filtering, automatic peak detection, and chromatographic alignment. In particular, all XCMS extractions will use a “minfrac” parameter of 0.2 to keep variables if present in at least 20% of the samples, since a large variability of profiles in the selected individuals is expected. After signal drift and batch effect correction, metabolite MS signals will be filtered using the following criteria: ratio of chromatographic peak areas of samples over blanks (above 3), repeatability of QC pool samples (coefficients of variation under 30%) and QC pool samples over biological samples (coefficients of variation ratio below 1.25). To reduce the analytical redundancy inside the obtained dataset, isotopes will finally be filtered.

Characteristics of the studied population will be analyzed using Mann-Whitney-Wilcoxon or Student’s t tests, depending on the variables’ distribution.

Serum metabolites modulated by MetS in young SGA adults

Extracted metabolomics data from follow-up will be first explored using principal component analysis (PCA) to assess main variations and show any trends or outlying data. Differential analysis will be performed using 2-way ANOVA tests to assess the effect of MetS with BMI as cofactor – obtained p-values will be corrected for multiple testing using the Benjamini-Hochberg correction. Calculation of a p-value for each feature will allow identifying significant differences (adjusted p-values ≤0.05). The log2 mean ratio of ion intensity of cases versus control subjects will be calculated to represent the fold change for selected features. Supervised multivariate analysis (Partial least-squares−discriminant analysis (PLS-DA, with prior Unit-Variance scaling) will also be used, to assess which metabolites among all detected features discriminate cases from controls. Variable importance on projection (VIP) values will be computed as indicators of importance of each metabolite in the PLS-DA model. Predictive discrimination ability by cumulative Q ² will be obtained according to a 7-fold cross validation method, with the validation of the reliability using permutation testing. ANOVA will be performed using the Galaxy web-service ²⁷ and PCA as well as PLS-DA will be performed using the SIMCA software (Umetrics, Umea, Sweden).

The same workflow will be applied to AGA metabolomics data at follow-up and modulated metabolites in both populations will be compared using Venn diagram. ³⁰

To explore the sub phenotypes, previously selected features will be analyzed by a hierarchical clustering analysis (HCA) using the R software. ³¹

Serum metabolites predicting evolution towards MetS in young SGA adults

Extracted metabolomics data from baseline will be explored to evaluate the possibility to predict the MetS status at follow-up, using Logistic Regression modelling. First, to reduce the number of candidate metabolites to include in the modeling procedure, ANOVA will be performed and p-values will be used to exclude the metabolites the most unlikely to carry individual predictive capabilities. Then, the remaining candidates will be evaluated through the building of logistic regression models, first in an individual way (one model per candidate), and then by combining the best candidates in a single multivariate logistic model with complementary selection using the optimisation of the Akaike Information Criterion. The prediction model performance will be evaluated using a confusion matrix, error rates, and area under receiver operating characteristic (ROC) curves (AUC) (R package “pROC” ³² ) with a confidence interval estimated with DeLong’s method. ³³ The increase in the AUC will be evaluated and tested for significance using the test also proposed by DeLong et al. ³³ All logistic regression models will be build using the R software.

A similar workflow will be performed on the available clinical variables, in order to be compared to the previously obtained metabolomics-based model. Complementary combined models will also be built by including metabolomic and clinical variables in common logistic regression models.

Metabolites contributing to the discrimination of the different phenotypes will be first identified using an in-house database containing the reference spectra of more than 1,000 authentic standard compounds analysed in the same analytical conditions. All standard compounds were purchased from Sigma-Aldrich. All were analytical standard grade (≥98%). Then, the remaining unknown compounds will be annotated on the basis of their exact masses which will be compared to those registered in the Human Metabolome Database (HMDB; www.hmdb.ca), ³⁴ and/or in Lipidmaps. ³⁵ Database queries will be performed with a mass error of 0.005 Da and a retention time difference of 0.1 min (for the in-house database). Database results will be confirmed using appropriate standards when available, isotopic patterns, and mass fragmentation analyses. For unidentified ions, the number of plausible elemental compositions will be restricted to a small number (or uniquely identified) with the support of additional chemical information i.e., the molecular formula of the parent and knowledge of possible metabolic pathways. Metabolites will be classified accordingly to Sumner et al. ³⁶ concerning the levels of confidence in the annotation process: identified (confirmed by standard), putatively annotated (based upon physicochemical properties and/or spectral similarity with public/commercial spectral libraries), putatively characterized compound classes.

Ethics and dissemination

The Haguenau study protocol was reviewed and approved by the ethical committee of the University of Paris-St Louis, and all subjects gave written informed consent. Research findings will be disseminated through peer reviewed publications and at national and international conferences with domain-specific societies, to researchers, clinicians, and policy markers sharing findings.

Patient and public involvement

No patients were involved in the study design.