An integrative framework to identify gene regulatory programs in mouse notochord and its derived nucleus pulposus cells

Abstract

The notochord is a signature of chordate as a flexible rod. It is significant for embryo patterning. In mice, segmentation occurs at E12.5-E15.5 as condensed cells form the vertebrae and non-condensed cells form the intervertebral discs (IVDs). Notochord cells are progenitors and differentiate into chondrocyte-like cells in nucleus pulposus (NP) in IVDs. Although it has been reported that the decrease of notochordal cells in NP is associated with the onset of disc degeneration, the mechanisms remain unclear. Characterizing the gene regulatory programs in mouse notochord and its derived cells, using multiple OMICS data, would help to determine the cellular and molecular basis of notochord and IVD development, and facilitate the induction of stem cells into notochordal cells (NCCs) or NP cells (NPCs) as potential therapy for IVD degeneration. In this thesis, I first characterized active enhancer profiles, via the analysis of H3K27ac ChIP-seq data, in mice NCCs and NPCs at four developmental stages: embryonic day 8.5-10.5 (E8.5-10.5), embryonic day 12.5 (E12.5), postnatal day 2-3 (P2-3), and adult 8-10 weeks (8-10wks). The profiles for E8.5-10.5 and E12.5 have many differences, while the profiles for E12.5, P2-3 and 8-10wks are much more similar. With integrative analyses of ChIP-seq and RNA-seq data, potential target genes of the enhancers and active transcription factors (TFs) were identified in E12.5 and P2-3 NPCs. Jun-AP1 was shown to be active at P2-3. I then constructed gene regulatory networks with secondary structure (GRsN) to describe the regulation of groups of genes with groups of TFs. The strategy is based on a linear model via l1-l2,1 and l0-l2,0 norm based Multivariate Sparse Group Lasso (Msgl), estimated using Proximal Gradient Algorithm (PgaMsgl). Thorough simulation studies showed good performance of l0-l2,0 based PgaMsgl for group structured sparse linear regression problems, outperforming Lasso and Group Lasso based methods on simulated problems. We also demonstrated with mouse embryonic stem cell (mESC) data that PgaMsgl can be applied to GRsN construction with two grouping methods. The first considered motif spacing of TFs and promoter-promoter interactions, while the second was purely based on expression profiles. The regulatory relationships of key transcription factors in mESCs were always involved in the results no matter of the model and grouping methods. The application of l0-l2,0 based PgaMsgl on NCC/NPC expression profiles identified regulatory relationships, which are significantly cross-validated with the appearance of the TF’s motif in the enhancer(s) assigned to the target gene. Pathway enrichment of these genes indicates the importance of the AP-1 network, pathways related to extracellular matrix components formation and organization, TGF-β signaling pathway, as well as Smad2/3 signaling in NCCs/NPCs. This work provides mechanistic insights into the gene regulatory profiles for IVD development, which will be a useful resource for the study of IVD degeneration. The enhancers and regulatory relationships identified may also be useful for improving the induction of ESCs to NCCs/NPCs for the cell therapy of degenerative disc disease.