Variation of DNA methylation on the IRX1/2 genes is responsible for the neural differentiation propensity in human induced pluripotent stem cells

• Published in: Regenerative therapy Vol. 21; pp. 620 - 630
• Main Authors: Sekiya, Asato, Takasawa, Ken, Arai, Yoshikazu, Horike, Shin-ichi, Akutsu, Hidenori, Umezawa, Akihiro, Nishino, Koichiro
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01.12.2022; Japanese Society for Regenerative Medicine; Elsevier
• Subjects: Differentiation propensity; DNA methylation; Human iPSCs; Machine learning; Neural stem cells; Original; PSC; Lasso; Human iPSCs; PAX6; EGFP; IRX; NSCs; C5orf38; dpc; Machine learning; DNA methylation; Differentiation propensity; Neural stem cells; EGFP, enhanced green fluorescent protein; PAX6, Paired Box 6; NSCs, neural stem cells; dpc, days post coitum; PSC, pluripotent stem cell; IRX, Iroquois Homeobox; Lasso, least absolute shrinkage and selection operator; C5orf38, Chromosome 5 open reading frame 38
• ISSN: 2352-3204; 2352-3204
• DOI: 10.1016/j.reth.2022.11.007

Human induced pluripotent stem cells (hiPSCs) are useful tools for reproducing neural development in vitro. However, each hiPSC line has a different ability to differentiate into specific lineages, known as differentiation propensity, resulting in reduced reproducibility and increased time and funding requirements for research. To overcome this issue, we searched for predictive signatures of neural differentiation propensity of hiPSCs focusing on DNA methylation, which is the main modulator of cellular properties. We obtained 32 hiPSC lines and their comprehensive DNA methylation data using the Infinium MethylationEPIC BeadChip. To assess the neural differentiation efficiency of these hiPSCs, we measured the percentage of neural stem cells on day 7 of induction. Using the DNA methylation data of undifferentiated hiPSCs and their measured differentiation efficiency into neural stem cells as the set of data, and HSIC Lasso, a machine learning-based nonlinear feature selection method, we attempted to identify neural differentiation-associated differentially methylated sites. Epigenome-wide unsupervised clustering cannot distinguish hiPSCs with varying differentiation efficiencies. In contrast, HSIC Lasso identified 62 CpG sites that could explain the neural differentiation efficiency of hiPSCs. Features selected by HSIC Lasso were particularly enriched within 3 Mbp of chromosome 5, harboring IRX1, IRX2, and C5orf38 genes. Within this region, DNA methylation rates were correlated with neural differentiation efficiency and were negatively correlated with gene expression of the IRX1/2 genes, particularly in female hiPSCs. In addition, forced expression of the IRX1/2 impaired the neural differentiation ability of hiPSCs in both sexes. We for the first time showed that the DNA methylation state of the IRX1/2 genes of hiPSCs is a predictive biomarker of their potential for neural differentiation. The predictive markers for neural differentiation efficiency identified in this study may be useful for the selection of suitable undifferentiated hiPSCs prior to differentiation induction. •Ability of hiPSCs to differentiate into neural stem cells differs from line to line.•Epigenetic markers to predict NSC differentiation ability of hiPSCs were identified.•DNA methylation of IRX1/2 related to neural differentiation ability in female hiPSCs.•Overexpression of IRX1/2 in undifferentiated hiPSCs impaired neural differentiation.