Non-invasive determination of gene expression in placental tissue using maternal plasma cell-free DNA fragmentation characters
Kun Li,Zhiwei Guo,Fenxia Li,Shijing Lu,Min Zhang,Yuyan Gong,Jiayu Tan,Chao Sheng,Wenbo Hao,Xuexi Yang
DOI: https://doi.org/10.1016/j.gene.2024.148789
IF: 3.913
2024-11-30
Gene
Abstract:Background: The expression profiles of placental genes are crucial for understanding the pathogenesis of fetal development and placental-origin pregnancy syndromes. However, owing to ethical limitations and the risks of puncture sampling, it is difficult to obtain placental tissue samples repeatedly, continuously, multiple times, or in real time. Establishing a non-invasive method for predicting placental gene expression profiles through maternal plasma cell-free DNA (cfDNA) sequencing, which carries information about the source tissue and gene expression, can potentially solve this problem. Methods: Peripheral blood and placental samples were collected simultaneously from pregnant women who underwent cesarean section. Deep sequencing was performed on the separated plasma cfDNA and single-cell sequencing was performed on peripheral blood mononuclear cells (PBMC), chorioamniotic membranes (CAM), placental villi (PV), and decidua basalis (DB). The aggregation of corresponding information for each gene was combined with the transcriptome of PBMCs and a differential resolution transcriptome of the placenta. This combined information was then utilized for the construction of gene expression prediction models. After training, all models evaluated the correlation between the predicted and actual gene expression levels using external test set data. Results: From five women, more than 20 million reads were obtained using deep sequencing for plasma cfDNA; PBMCs obtained 32,401 single-cell expression profiles; and placental tissue obtained 156,546 single-cell expression profiles (59,069, 44,921, and 52,556 for CAM, PV, and DB, respectively). The cells in the PBMC and placenta were clustered and annotated into five and eight cell types, respectively. A "DEPICT" gene expression prediction model was successfully constructed using deep neural networks. The predicted correlation coefficients were 0.75 in PBMCs, 0.84 in the placenta, and 0.78, 0.80, and 0.77 in CAM, BP, and PV respectively, and greater than 0.68 in different cell lines in the placenta. Conclusion: The DEPICT model, which can noninvasively predict placental gene expression profiles based on maternal plasma cfDNA fragmentation characteristics, was constructed to overcome the limitation of the inability to obtain real-time placental gene expression profiles and to improve research on noninvasive prediction of placental origin pregnancy syndrome.