Do nuclear magnetic resonance (NMR)-based metabolomics improve the prediction of pregnancy-related disorders? Findings from a UK birth cohort with independent validation

Abstract

Background Prediction of pregnancy-related disorders is usually done based on established and easily measured risk factors. Recent advances in metabolomics may provide earlier and more accurate prediction of women at risk of pregnancy-related disorders. Methods We used data collected from women in the Born in Bradford (BiB; n = 8212) and UK Pregnancies Better Eating and Activity Trial (UPBEAT; n = 859) studies to create and validate prediction models for pregnancy-related disorders. These were gestational diabetes mellitus (GDM), hypertensive disorders of pregnancy (HDP), small for gestational age (SGA), large for gestational age (LGA) and preterm birth (PTB). We used ten-fold cross-validation and penalised regression to create prediction models. We compared the predictive performance of (1) risk factors (maternal age, pregnancy smoking, body mass index (BMI), ethnicity and parity) to (2) nuclear magnetic resonance-derived metabolites (N = 156 quantified metabolites, collected at 24–28 weeks gestation) and (3) combined risk factors and metabolites. The multi-ethnic BiB cohort was used for training and testing the models, with independent validation conducted in UPBEAT, a multi-ethnic study of obese pregnant women. Results Maternal age, pregnancy smoking, BMI, ethnicity and parity were retained in the combined risk factor and metabolite models for all outcomes apart from PTB, which did not include maternal age. In addition, 147, 33, 96, 51 and 14 of the 156 metabolite traits were retained in the combined risk factor and metabolite model for GDM, HDP, SGA, LGA and PTB, respectively. These include cholesterol and triglycerides in very low-density lipoproteins (VLDL) in the models predicting GDM, HDP, SGA and LGA, and monounsaturated fatty acids (MUFA), ratios of MUFA to omega 3 fatty acids and total fatty acids, and a ratio of apolipoprotein B to apolipoprotein A-1 (APOA:APOB1) were retained predictors for GDM and LGA. In BiB, discrimination for GDM, HDP, LGA and SGA was improved in the combined risk factors and metabolites models. Risk factor area under the curve (AUC 95% confidence interval (CI)): GDM (0.69 (0.64, 0.73)), HDP (0.74 (0.70, 0.78)) and LGA (0.71 (0.66, 0.75)), and SGA (0.59 (0.56, 0.63)). Combined risk factor and metabolite models AUC 95% (CI): GDM (0.78 (0.74, 0.81)), HDP (0.76 (0.73, 0.79)) and LGA (0.75 (0.70, 0.79)), and SGA (0.66 (0.63, 0.70)). For GDM, HDP and LGA, but not SGA, calibration was good for a combined risk factor and metabolite model. Prediction of PTB was poor for all models. Independent validation in UPBEAT at 24–28 weeks and 15–18 weeks gestation confirmed similar patterns of results, but AUCs were attenuated. Conclusions Our results suggest a combined risk factor and metabolite model improves prediction of GDM, HDP and LGA, and SGA, when compared to risk factors alone. They also highlight the difficulty of predicting PTB, with all models performing poorly.

Maternal Body Mass Index, Early-Pregnancy Metabolite Profile, and Birthweight

“…These metabolites improved the prediction of birthweight by increasing the explained variance for birthweight from 4.4% to 6.5% after addition of maternal fasting metabolites to a model with maternal prepregnancy BMI and from 7.1% to 9.2% after addition of 1-h metabolites. A study among 8212 women obtained nuclear magnetic resonance-derived metabolite concentrations in mid-pregnancy ( 36 ). They observed that addition of 66 metabolites, including AA, fatty acids, phospholipids, apolipoproteins, cholesterol, and very low density lipoprotein, to a risk factor model with maternal age, pregnancy BMI, ethnicity, and parity improved prediction of LGA newborns.…”

Do Mass Spectrometry-Derived Metabolomics Improve the Prediction of Pregnancy-Related Disorders? Findings from a UK Birth Cohort with Independent Validation

“…In our previous work, we found that NMR-derived metabolomics improves upon risk factors for the prediction of GDM, LGA, SGA and combined PE and GHT (hypertensive disorders of pregnancy—HDP). We reported the best discrimination for GDM and LGA, and our findings here suggest that metabolites from different platforms are valuable for the prediction of GDM and LGA [ 25 ]. In previous work in POPs and BiB 1000, we found that the amino acid 4-hydroxyglutamate was a novel predictor of PE, and the metabolite ratio described above was a better predictor of fetal growth restriction/SGA than a biomarker ratio used in the diagnosis of PE (sFlt1:PlGF) [ 27 , 28 ].…”

“…We have recently shown that a targeted nuclear magnetic resonance (NMR)-derived metabolomics panel of 156 (mostly lipid) traits can improve the prediction of GDM, LGA and hypertensive disorders of pregnancy (HDP) in Born in Bradford (BiB), a large general population pregnancy cohort. This work was externally validated in the UK Pregnancies Better Eating and Activity trial (UPBEAT), a cohort of obese pregnant women [ 25 ]. We have also identified novel metabolite predictors for specific pregnancy outcomes using a mass spectrometry (MS)-derived metabolites platform in the Pregnancy Outcome Prediction study (POPs), which were externally validated in the BiB cohorts.…”

Putrescine as a Novel Biomarker of Maternal Serum in First Trimester for the Prediction of Gestational Diabetes Mellitus: A Nested Case-Control Study

“…The amount of information obtained by this approach is still not fully understood. The GDM prediction model from the UK Pregnancies Better Eating and Activity Trial (UPBEAT), which included multiple serum metabolites, reached an AUC of 0.78 ( 30 ). Other metabolite models that included adiponectin reached an AUC of 0.79-0.85 ( 24 , 33 ).…”

“…Therefore, their clinical benefit when applied to the most recent GDM definition has not yet been well investigated. Recently, a number of new markers have been evaluated for use in predicting GDM with variable success, such as protein biomarkers (23), adiponectin (24) and leptin (25), pentraxin 3 (26), trace elements (27), RNA (28), singlenucleotide polymorphisms (29), and other combined metabolite models (30)(31)(32)(33). The AUC values of some of these models can reach above 0.8.…”