Preview

PDF, English Download (75MB) | Terms of use

Abstract

Cells, despite having identical genetic information, show various functions and structures, which are determined by their unique gene expression profiles. Key elements such as chromatin, tran- scription factors (TF), and genes play an important role in regulating these profiles, together forming complex gene regulatory networks (GRNs). Understanding GRNs is crucial for inter- preting how cellular identity is established and maintained, and how it can be disrupted in diseases. However, developing computational methods to reconstruct GRNs presents significant challenges, particularly in evaluation due to the absence of a definitive gold standard or ground truth. Relying only on experimentally validated connections for evaluation could lead to a bias toward well-established TFs. To address this, I have developed GRaNPA, a novel method that evaluates networks unbiasedly based on their ability to predict gene expression perturbations. GRaNPA has a dual purpose: as a benchmarking tool, it compares different GRNs based on the hypothesis that true connections between TFs and genes can, to some extent, predict gene expression perturbations. In addition, it identifies key TFs essential for predicting these variations in gene expression. This functionality of GRaNPA is particularly beneficial for unraveling the underlying biological mechanisms. I applied GRaNPA to assess an enhancer-based GRN (eGRN) that I constructed using GRaNIE for an iPSC-derived macrophage dataset. GRaNIE is a method that reconstructs GRNs based on co-variation across individuals, establishing connections both from TFs to enhancers and from enhancers to genes. For the macrophage eGRNs, I initially demonstrated their ability to accurately predict differential expression values between naive macrophages and those infected with Salmonella. Subsequently, by comparing their predictive accuracy with networks derived from AML and CD4+ T cells and demonstrating that these networks are predictive only for differential expression values specific to their cell type, I confirmed the cell type specificity of these eGRNs. Additionally, I showed that eGRNs from different cell types contain TFs with almost entirely distinct regulons, highlighting the various roles of TFs in different cell types. I also utilized GRaNPA’s second function to identify important TFs under various conditions, such as infection with Salmonella, breast cancer, and tuberculosis disease. Beyond identifying well-known TFs like NFKB, which is influential in M1 macrophages exposed to INF-γ, I discovered lesser-known TFs like PURA, potentially playing a proinflammatory role in macrophages. Overall,this demonstrates GRaNPA’s utility in evaluating and identifying key TFs in various conditions, helping our understanding of the underlying biology. Understanding and identifying important TFs could be beneficial for unraveling gene regula- tion in the context of cell fate determination. A few TFs, such as terminal selectors or master regulators, can drive specific cell lineages. Additionally, safeguard repressors actively repress alternative cell fates to induce and maintain cell identity. Following a computational screening for potential safeguard candidates, my collaborators and I identified Prox1 as a possible safeguard repressor for hepatocytes. I then explored Prox1’s role as a potential tumor suppressor in a human hepatocyte carcinoma cell line and found that it restrains proliferation by targeting the key TF MYC. Further, I confirmed Prox1’s role in hepatocyte fate induction at the single-cell level and utilized GRaNPA to identify its important targets, Pparg and Prrx1, which are key regulators in adipocyte and fibroblast fates. We also demonstrated that Prox1 can prevent transdifferentiation, suggesting that its absence might lead to an identity shift from hepatocyte carcinoma to cholan- giocarcinoma. In conclusion, Prox1 promotes hepatocyte fate by targeting alternative fates and acts as a tumor suppressor in cancer. In summary, understanding GRNs is key to interpreting the gene expression profiles of distinct cells. I developed GRaNPA, a method for assessing GRNs and identifying important TFs that explain specific variations, thereby enhancing our understanding of complex scenarios, including diseases.