Placental and immune cell DNA methylation reference panel for bulk tissue cell composition estimation in epidemiological studies.

To distinguish DNA methylation (DNAm) from cell proportion changes in whole placental villous tissue research, we developed a robust cell type-specific DNAm reference to estimate cell composition. We collated new and existing cell type DNAm profiles quantified via Illumina EPIC or 450k microarrays. To estimate cell composition, we deconvoluted whole placental samples ( = 36) with robust partial correlation based on the top 30 hyper- and hypomethylated sites identified per cell type. To test deconvolution performance, we evaluated root mean square error in predicting principal components of DNAm variation in 204 external placental samples. We analyzed DNAm profiles ( = 368,435 sites) from 12 cell types: cytotrophoblasts ( = 18), endothelial cells ( = 19), Hofbauer cells ( = 26), stromal cells ( = 21), syncytiotrophoblasts ( = 4), six lymphocyte types ( = 36), and nucleated red blood cells ( = 11). Median cell composition was consistent with placental biology: 60.9% syncytiotrophoblast, 17.3% stromal, 8.8% endothelial, 3.7% cytotrophoblast, 3.7% Hofbauer, 1.7% nucleated red blood cells, and 1.2% neutrophils. Our expanded reference outperformed an existing reference in predicting DNAm variation (PC1, 15.4% variance explained, IQR = 21.61) with cell composition estimates (mean square error of prediction: 8.62 vs. 10.79, -value < 0.001). This cell type reference can robustly estimate cell composition from whole placental DNAm data to detect important cell types, reveal biological mechanisms, and improve causal inference.