eoPred: Predicting the placental phenotype of early-onset preeclampsia using DNA methylation

Abstract

Abstract Background A growing body of literature has reported molecular and histological changes in the human placenta in association with preeclampsia (PE). Placental DNA methylation (DNAme) and transcriptomic patterns have revealed molecular subgroups of PE that are associated with placental histopathology and clinical phenotypes of the disease. However, the heterogeneity of PE both across and within subtypes, whether defined clinically or molecularly, complicates the study of this disease. PE is most strongly associated with placental pathology and adverse fetal and maternal outcomes when it develops early in pregnancy. We focused on placentae from pregnancies affected by preeclampsia that were delivered before 34 weeks of gestation to develop eoPred, a predictor of the DNAme signature associated with the placental phenotype of early-onset preeclampsia (EOPE). Results Public data from 83 placental samples (HM450K), consisting of 42 EOPE and 41 normotensive preterm birth (nPTB) cases, was used to develop eoPred - a supervised model that relies on a highly discriminative 45 CpG DNAme signature of EOPE in the placenta. The performance of eoPred was assessed using cross-validation (AUC = 0.95) and tested in an independent validation cohort (n = 49, AUC = 0.725). A subset of fetal growth restriction (FGR) and late-PE cases showed a similar DNAme profile at the 45 predictive CpGs, consistent with the overlap in placental pathology between these conditions. The relationship between the EOPE probability generated by eoPred and various phenotypic variables was also assessed, revealing that it is associated with gestational age, and it is not driven by cell composition differences. Conclusions eoPred relies on a 45 CpG DNAme signature to predict EOPE, and it can be used in a discrete or continuous manner. Using this classifier should 1) improve the consistency of future placental DNAme studies of PE and placental insufficiency, 2) facilitate identifying cases of EOPE in public data sets and 3) importantly, standardize the placental diagnosis to allow better cross-cohort comparisons. Lastly, classification of cases with eoPred should be useful for testing associations between placental pathology and genetic or environmental variables.