Gene Expression and DNA Methylation in Human Placental Cell Types: Applications to Deconvolution, Prenatal Metals Exposure, and Preeclampsia Molecular Epidemiology

Abstract

The human placenta mediates adverse perinatal outcomes and current research examines molecular changes in bulk placental tissue. However, bulk molecular measures represent a convolution of signals across cell types, obscuring biological mechanisms and biasing study results. Placental cell composition in preeclampsia is not well-understood and limits interpretation of bulk gene expression measures. In Chapter 2, I integrated a single-cell RNA-sequencing atlas of 19 fetal and 8 maternal cell types from placental villous tissue (n=9 biological replicates) at term (n=40,494 cells) and deconvoluted eight published microarray case-control studies of preeclampsia (n=173 controls, 157 cases). Preeclampsia was associated with excess extravillous trophoblasts (POR = 1.94, 95% CI [1.61, 2.34]) and fewer mesenchymal (POR = 0.79, 95% CI [0.73, 0.85]) and Hofbauer cells (POR = 0.67, 95% CI [0.59, 0.77]). Cellular composition mediated the association between preeclampsia and FLT1 (37.8%, 95% CI [27.5%, 48.8%]) overexpression. My findings demonstrated placental cellular composition heterogeneity in preeclampsia drives previously observed bulk gene expression differences. This novel deconvolution reference allows for cellular composition-aware investigation into adverse perinatal outcomes. To enable robust estimation of placental cell composition from bulk DNA methylation, in Chapter 3, I integrated a DNA methylation atlas (n=81) of 5 placental and 7 non-placenta cell types (p=192 fractions). Methylation was quantified via the Illumina DNA methylation microarray (450k or EPIC), and common probes were quality filtered (n=407,628 DNA methylation sites). To identify cell type-discriminating DNA methylation sites, I ranked the top 50 hyper- and hypomethylated sites per cell type by F-test. To deconvolute bulk placental tissues, I applied robust partial correlation. Consistent with placental biology, bulk placental tissue (n=35) cell type proportion estimates (mean ± standard deviation) were predominately syncytiotrophoblast (57.8% ± 8.3%), stromal (20.6 ± 5.9%), cytotrophoblast (11.0% ± 4.1%), endothelial (7.5% ± 2.2%), and Hofbauer cells (1.5% ± 1.2%). This atlas can robustly estimate cell composition from placental DNA methylation data to detect unexpected non-placental cell types and improve casual inference. Malapropos perinatal exposure to essential and toxic metals is widespread and linked to adverse outcomes. Healthy placental morphology is essential to pregnancy, but the relationship between metals exposure and placental cell composition is poorly understood. In Chapter 4, I analyzed data from two prospective pregnancy cohorts: MARBLES (n=83) and EARLI (n=94). Urinary metals (p=18) concentrations were measured during early or late gestation. Placental DNA methylation was measured with EPIC or 450k microarrays. I estimated cellular composition via the Chapter 3 reference. Demographics-adjusted beta regression models tested associations between metals concentrations (p=18) during early or late gestation and each placental cell type proportion (p=5). Results were meta-analyzed. A doubling in late gestation barium concentration was associated with a 0.2% increment in mean Hofbauer cell proportion to 2.5% (POR = 1.08, 95% CI [1.02, 0.14], q=0.25). Cadmium exposure was associated with a 3.0% decrement in mean syncytiotrophoblast proportion to 59.5% (POR = 0.88, 95% CI [0.78, 0.98], q=0.31). Divalent metals exposure may disrupt placental structure, particularly among Hofbauer cells and syncytiotrophoblasts. In conclusion, I applied state-of-the-art laboratory and bioinformatic approaches to develop cost-effective and scalable reference-based deconvolution methods for researchers to estimate cell composition in bulk placental tissue and I demonstrated the power of these novel approaches through applications to human populations. To provide valid, more causally interpretable results, future perinatal investigations should carefully consider placental cell composition using approaches developed in this dissertation.