Miraldi Lab

Research

Gene regulatory network modeling of disease-associated DNA methylation perturbations with Minji Byun.

Somatic mutations in DNMT3A and TET2 are common in the hematopoietic lineages of elderly individuals, estimated to affect more than 10% of adults over the age of 65. These mutations increase the risk for age-related comorbidities, including severe infection, atherosclerotic cardiovascular disease, osteoporosis, chronic kidney disease and hematologic malignancies, nearly doubling the mortality rate of affected individuals. DNMT3A and TET2 encode enzymes essential for remodeling DNA methylation during cellular differentiation. Animal studies suggest that mutations in these genes drive aberrant activation of immune cells, such as macrophages, which may underlie the disease associations. We recently developed a human pluripotent stem cell (hPSC)-derived macrophage model, where the differentiation-dependent effects of DNMT3A or TET2 perturbation can be precisely delineated. We discovered that DNMT3A- and TET2- perturbations impaired DNA methylation at thousands of regulatory loci, altering enhancer activities and expression of genes important for macrophage function. Our study highlighted the need for engineering approaches, and mathematical modeling in particular, to unravel the complex effects of DNMT3A and TET2 perturbations on cellular function and disease risk. Here, we pair novel computational modeling approaches with unique experimental resources to mechanistically connect site-specific changes in DNA methylation to aberrant immune responses and disease risk. Aim 1 builds deep neural network models (and requisite training data resources) to predict the effects of DNA methylation on chromatin binding of 100+ transcription factors (TFs), the “readers” of DNA methylation patterns that ultimately recruit RNA polymerase and co-activators to drive gene transcription. In Aim 2, we predict genome-scale TF-binding patterns from chromatin accessibility, transcriptional activity and DNA methylation data in our contexts of interest: DNMT3A- or TET2-perturbed human macrophages in response to viral and bacterial infection-induced immune activation. To discover links between existing and novel disease associations, we will intersect the TFBS predictions with curated sets of age-related disease risk variants, to nominate TFs and contexts where DNMT3A- or TET2-perturbation and downstream alterations in TF binding might mediate disease risk. In Aim 3, we will construct gene regulatory network (GRN) models of DNMT3A- and TET2-perturbed human macrophage to identify TFs driving differential gene expression responses to infection, hypotheses that (1) we will experimentally test and (2) could eventually lead to therapies that mitigate the negative, pathogenic consequences of common DNMT3A and TET2 mutations. Furthermore, we build significant generalizable resources (models, modeling methodologies and training data) that will enable future discoveries in new cell types and disease contexts where alterations in DNA methylation drive phenotypes.

Methods to maximize insight from (sc)ATAC-seq for gene regulatory networks and genetics with Matthew Weirauch PhD, Surya Prasath, PhD, Leah Kottyan, PhD, Artem Barski, PhD

Most human disease-associated genetic polymorphisms fall outside protein-coding sequences. They overlap significantly with enhancers, promoters and other locus-control regions. Causal variants are thought to contribute to disease phenotypes by altering gene expression in specific cell types. Gene regulatory networks (GRNs) describe the control of gene expression by transcription factors (TFs). GRN reconstruction for human cell types across diverse conditions will be crucial to identifying how noncoding genetic variants contribute to complex phenotypes through altered TF binding, chromatin looping and other mechanisms of transcriptional control. 

The Assay for Transposase Accessible Chromatin (ATAC-seq) opens new opportunities for GRN inference and genetics. This easy-to-use, popular technique provides high-resolution chromatin accessibility with low sample input requirements. Thanks to advances in single-cell (sc)ATAC-seq, it is now possible to computationally resolve the regulomes of individual cell types from heterogeneous tissues and limited clinical samples. Whether from single cells or homogenous bulk populations, our group has shown that integration of TF-binding predictions from ATAC-seq improves GRN inference from RNA-seq (Miraldi et al. (2019) Genome Research, Pokrovskii et al. (2019) Immunity). Although other experimental approaches more directly measure TF occupancy (e.g., ChIP-seq), they are prohibitively time-consuming and expensive. Profiling through ChIP-seq of >50 of the 100s of TFs expressed in a given cell-type condition has only been accomplished for a very few human cell types. Furthermore, for some rare cell types and physiological conditions, limited sample material precludes genomic measurement of TF occupancies. 

In response to these issues, the computational community pioneered methods to predict TF binding from chromatin accessibility data. In 2017, the “ENCODE-DREAM in vivo Transcription Factor Binding Site Prediction Challenge” established two top-performing TF-binding prediction algorithms that vastly improved performance over popular motif scanning (median AUPR .4. versus .1). Yet top-performing models are virtually never used in popular ATAC-seq analysis pipelines. Despite the promise of ATAC-seq for GRN inference and human genetics, ATAC-seq is greatly underutilized, even as ATAC-seq data generation grows exponentially. This collaboration is focused on construction of maxATAC, a user-friendly suite of top-performing deep neural network models for TF-binding prediction from ATAC-seq. 

We recently released a first-version of the maxATAC models and user-friendly codebase. 

An experimentally-refined dynamic gene regulatory network model of T-cell memory with Artem Barski, PhD.

Long-lived memory T cells are critical for protective immunity. Upon secondary exposure to pathogen, memory T cells mount an immediate defense, rapidly producing lineage-characteristic cytokines. In contrast, naïve T cells require days to produce cytokines after initial exposure. While this “rapid recall ability” of memory T cells is critical for pathogen defense and is the basis for vaccination, it also underlies allergy, asthma and anti-cancer immunity. Unfortunately, little is known about the mechanism(s) by which memory T cells are endowed with the ability to rapidly produce cytokines. Notably, such knowledge could be exploited to promote protective (e.g., vaccines, anti-cancer immunity) or to reduce pathologic (e.g., asthma, allergy, and autoimmunity) responses. 

The aim of this collaboration is to create an experimentally-validated, genome-scale model of memory immune response. Using single-cell genomics, we are characterizing the gene expression and chromatin dynamics of T cell activation in naïve and memory cells and building mathematical models that integrate these data (along with relevant existing genomics resources) into a dynamic gene regulatory network (GRN). Our GRN model will predict the molecular drivers (transcription factors) and regulatory elements that orchestrate rapid recall – hypotheses we will test with dynamic TF perturbation and occupancy measurements.

Although T-cell activation in naïve and memory cells similarly promotes nuclear translocation of inducible TFs, our data lead us to hypothesize that (1) chromatin remodeling upon initial pathogen exposure alters the occupancy of inducible TFs in memory T cells, and (2) this is the basis of rapid recall. 

We also aim to identify the mechanisms by which memory T cells maintain the epigenome conducive for rapid recall – over the human lifespan. We hypothesize that constitutive TFs maintain the epigenome poised for rapid recall. This study will help uncover basic mechanisms of T cell memory and identify potential targets for manipulating immunologic memory responses. Because rapid recall is the basis for vaccination and central to allergy, asthma, and cancer immunity, this study will have a broad impact on human health.

Dynamic regulatory network models of human response to influenza with Ivan Marazzi, PhD, Brad Rosenberg, MD, PhD, Chris Benner, PhD, Will Zacharias, MD, Matthew Weirauch PhD, Leah Kottyan, PhD, Surya Prasath, PhD

Human response to Influenza virus (IAV) infection varies dramatically between individuals, from mildly symptomatic to death. Here, we systematically measure and model the innate immune response to IAV at the molecular network level: from host-virus protein-protein interactions through cellular signal transduction to changes in gene expression in human cells. In addition, we aim to connect genetic variants to variability in early host transcriptional response to virus, through construction of deep neural networks that predict gene transcription and chromatin accessibility (the inputs to our gene regulatory network models) from DNA sequence. The resulting mathematical models will help identify network vulnerabilities to be exploited for IAV therapy and help predict differences in IAV response in the human population.  

Reverse-engineering therapeutic strategies for hepatic inflammation and injury resolution with Alex Miethke, MD.

Therapeutic options for autoimmune liver disease are limited, with many patients experiencing fibrotic end organ injury that requires liver transplantation. Our goal is to identify cell types and genes to serve as novel therapeutic targets in the resolution of liver inflammation. To achieve this goal, we are integrating cross-sectional patient liver and dynamic sc-genomic profiling of murine liver injury models to infer the cell-cell communication and transcriptional regulatory networks, identifying cell types and regulators that could be targeted to limit hepatic injury and improve healing.