Metabolomic Profiles During and After a Hypertensive Disorder of Pregnancy: The EPOCH Study

Abstract

Hypertensive disorders of pregnancy are associated with a higher risk of later cardiovascular disease, but the mechanistic links are unknown. We recruited two groups of women, one during pregnancy and another at least two years after delivery, including both cases (with a hypertensive disorder of pregnancy) and controls (with a normotensive pregnancy). We measured metabolites using liquid chromatography-mass spectroscopy and applied machine learning to identify metabolomic signatures at three time points: antepartum, postpartum, and mid-life. The mean ages of the pregnancy cohort (58 cases, 46 controls) and the mid-life group (71 cases, 74 controls) were 33.8 and 40.8 years, respectively. The levels of 157 metabolites differed significantly between the cases and the controls antepartum, including 19 acylcarnitines, 12 gonadal steroids, 11 glycerophospholipids, nine fatty acids, six vitamin D metabolites, and four corticosteroids. The machine learning model developed using all antepartum metabolite levels discriminated well between the cases and the controls antepartum (c-index = 0.96), postpartum (c-index = 0.63), and in mid-life (c-index = 0.60). Levels of 10,20-dihydroxyeicosanoic acid best distinguished the cases from the controls both antepartum and postpartum. These data suggest that the pattern of differences in metabolites found antepartum continues to distinguish women who had a hypertensive disorder of pregnancy from women with a normotensive pregnancy for years after delivery.

Keywords: case–control study; gestational hypertension; machine learning; metabolomics; preeclampsia.

Figure 1

Study overview: Study participants were enrolled in two groups. The pregnancy cohort (left panel) was invited to make two study visits, one antepartum (AP) and one postpartum (PP). The mid-life (ML) group was invited to make a single study visit.

Figure 2

Metabolite levels of cases vs. controls at three time points during and after pregnancy. Data are presented at three time points: antepartum (AP) on the top row, postpartum (PP) in the middle row, and mid-life (ML) in the bottom row. Within each row, the left panel displays the natural log of the fold-change in levels between cases and controls, plotted on the horizontal axis, and the adjusted p-value, plotted on the vertical axis (log scale); the dashed horizontal line indicates an adjusted p-value of 0.05. Red dots indicate that the metabolite level is significantly higher in cases and blue dots indicate that the metabolite level is significantly lower in cases. The center panel of each row displays the receiver operating characteristic (ROC) curves for the multivariate model of the metabolite signature that distinguished cases from controls. The horizontal axis shows the false-positive rate (1-specificity) of the multivariate model, while the vertical axis indicates its sensitivity. The right panel(s) of each row display box-and-whisker plots of the most discriminating metabolites between cases and controls. The top line of the box is the 75th percentile, the bottom line is the 25th percentile, and the line in the box is the median. The point indicates the mean value in each group **** = < 0.0001.

Figure 3

Antepartum metabolites that differ between cases and controls. Each line displays a single metabolite, labeled on the left, with blue dots indicating lower levels in cases than in controls, while red dots indicate higher levels in cases than controls. The size of the dots is proportional to the −log10 p-value, and the color indicates the log2 fold change between cases and controls. The metabolites are grouped by metabolic pathway, which is labeled on the right. The numbers of metabolites in various classes are summarized in the lower right corner.

Figure 4

Multivariate models applied at subsequent time points, and adjusted for baseline covariates. Left panel: Discrimination of models applied at different time points, displayed as AUCs. The blue line indicates the discrimination of the model developed with the antepartum samples when applied to the postpartum samples (AUC = 0.63), and to the mid-life samples (AUC = 0.60). The orange line indicates the discrimination of the model developed in the postpartum samples when applied to the mid-life samples (AUC = 0.65). Right panel: Effect of adjustment for covariates on the performance of the postpartum model. The vertical axis indicates the partial correlation coefficient, and the horizontal axis indicates the results at three study time points: antepartum (AP), postpartum (PP), and mid-life (ML). The black points indicate the performance of the unadjusted model, and the error bars indicate its 95% confidence limits. Model performance adjusted for body mass index (BMI) is shown with blue symbols, model performance adjusted for gestational diabetes (GDM) is shown with orange symbols, and model performance is adjusted for study visit timing (weeks of gestational age (GA) antepartum, and weeks after delivery postpartum and mid-life) is shown in green symbols. Statistical significance levels indicated as: 0.01< * = <0.05; 0.001< ** = < 0.01; 0.0001< *** = < 0.001; **** = < 0.0001.