The maternal blood lipidome is indicative of the pathogenesis of severe preeclampsia

Abstract

Preeclampsia is a pregnancy-specific syndrome characterized by hypertension and proteinuria after 20 weeks of gestation. However, it is not well understood what lipids are involved in the development of this condition, and even less is known how these lipids mediate its formation. To reveal the relationship between lipids and preeclampsia, we conducted lipidomic profiling of maternal sera of 44 severe preeclamptic and 20 healthy pregnant women from a multiethnic cohort in Hawaii. Correlation network analysis showed that oxidized phospholipids have increased intercorrelations and connections in preeclampsia, whereas other lipids, including triacylglycerols, have reduced network correlations and connections. A total of 10 lipid species demonstrate significant changes uniquely associated with preeclampsia but not any other clinical confounders. These species are from the lipid classes of lysophosphatidylcholines, phosphatidylcholines (PCs), cholesteryl esters, phosphatidylethanolamines, lysophosphatidylethanolamines, and ceramides. A random forest classifier built on these lipids shows highly accurate and specific prediction (F1 statistic = 0.94; balanced accuracy = 0.88) of severe preeclampsia, demonstrating their potential as biomarkers for this condition. These lipid species are enriched in dysregulated biological pathways, including insulin signaling, immune response, and phospholipid metabolism. Moreover, causality inference shows that various PCs and lysophosphatidylcholines mediate severe preeclampsia through PC 35:1e. Our results suggest that the lipidome may play a role in the pathogenesis and serve as biomarkers of severe preeclampsia.

Keywords: biomarker; classification; hypertension; lipidomics; machine learning; maternal blood; metabolomics; pathway; preeclampsia; pregnancy.

Fig. 1

Exploratory analysis of preeclampsia and control samples. A: Correlation matrix of the phenotypic variables on the 64 samples (20 control vs. 44 preeclampsia samples). Significant correlations (P < 0.05) are shown with ∗. B: Source of variation (SOV) results across the 64 samples using 729 lipid species. C: SOV results across the 64 samples using 280 lipid species, which all have F statistic values of at least 1 (F statistic of the error term). D: Heat map showing the correlations between the 280 lipid species and the confounding factors. The columns are lipids, and rows are clinical factors. Each entry of the heat map represents the Pearson correlation coefficient value between the lipid and the clinical factor.

Fig. 2

WGCNA network comparison between preeclampsia and control samples. A and B: WGCNA network in preeclampsia (A) and control (B), respectively. Each node represents a lipid species. C: Overlap between modules of networks in control and preeclampsia samples. D: Table showing more properties of modules in networks of control versus preeclampsia samples, including module density and enriched lipids in each module. E: Heat map of lipids with significant connectivity difference in WGCNA networks (A and B), between control versus preeclampsia. Connectivity value is defined as increased or decreased if the connectivity value in the preeclampsia network minus that in the control network is larger than 5 or less than −5, respectively.

Fig. 3

Lipids show significantly different levels in preeclampsia and control samples. A: Heat map of the 11 lipids with significant difference between preeclampsia and control samples. B: Box plots of the 11 lipids in our lipidomics data. C: Bipartite graph of lipids in (A) and their affiliated metabolic pathways. Elliptical nodes: lipid. Rectangular nodes: pathways from HMDB, PubChem, and KEGG databases. Blue color: downregulation in preeclampsia. Orange color: upregulation in preeclampsia. Note: lipids without any metabolomic pathway affiliations are omitted. The unseparated LPCs/PCs are shown in the same plot. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

Fig. 4

Biomarker classification model for preeclampsia. A: Performance of machine learning models on predicting severe preeclampsia using 11 potential lipid markers. Comparison of seven popular classification models on training data (left). From left to right: random forest (RF), gradient boosting (GBM), support vector machine (SVM), linear discriminant analysis (LDA), elastic net (LOG), decision tree (RPART), and nearest shrunken centroids (PAM). The performance on testing data from the winning method RF, based on training data, is shown on the right. All samples were randomly split into training data (80%) and testing data (20%) 10 times. The average value and standard error are shown for three performance metrics: area under the ROC curve (AUC), F1 statistic, and balanced accuracy. B: Heat map of correlation coefficients between 11 potential lipid markers and clinical variables. C: Precision-recall curve of RF model using 11 markers on: training data for severe preeclampsia and testing data for severe preeclampsia, gestational diabetes, and chronic hypertension, respectively. D: Normalized variable importance scores for the 11 lipid markers in the RF model.

Fig. 5

Predicted causality interactions among lipids and preeclampsia. Edges are from causes to results. Blue nodes are downregulated lipids, whereas red one is preeclampsia. Only significant (P < 0.05) causality interactions are shown. No significant causality interaction was found for upregulated lipid.

Fig. 6

A proposed model of lipidomic changes in the pathogenesis of severe preeclampsia.