Prediction of late-onset preeclampsia using plasma proteomics: a longitudinal multi-cohort study

Preeclampsia is a pregnancy disorder with substantial perinatal and maternal morbidity and mortality. Pregnant women at risk of preeclampsia would benefit from early detection for follow-up, timely interventions and delivery. Several attempts have been made to identify protein biomarkers of preeclampsia, but findings vary with demographics, clinical characteristics, and time of sampling. In the current study, we combined three independent longitudinal pregnancy cohorts (Detroit, Stanford and Oslo) resulting in 124 late-onset preeclampsia (LOPE) cases and 178 gestational age matched controls, and analyzed > 1000 proteins in maternal plasma sampled between 12 and 34 weeks of gestation. Differential abundance analysis of combined protein data revealed increased deviation in protein abundance trajectories throughout gestation in women destined to develop LOPE compared to controls. There were no differentially abundant proteins at time interval T1 (12-19 weeks), yet 31 differentially abundant proteins were found at time interval T2 (19-27 weeks), and 48 proteins at time interval T3 (27- 34 weeks). Multi-protein random forest models assessed via cross-validation predicted LOPE with an area under the ROC curve of 0.72 (0.65-0.78), 0.76 (0.71-0.81) and 0.80 (0.75-0.85) at time interval T1, T2 and T3, respectively. The results at T3 were confirmed using a leave-one-cohort-out analysis suggesting cross-cohort consistency, and at T1 and T2 when the largest two cohorts were used as training sets.