Katherine S. Pollard, PhD - US grants

The identification of co-regulated genes is one of the great opportunities for biological research in the genomic era. Although the opportunity is great, the challenges are just as profound. Often existing clustering routines fail to find the main clusters or divide clear clusters. I have developed several new clustering methods, including the hierarchical ordered partitioning and collapsing hybrid (HOPACH), which combines the strengths of both partitioning and agglomerative clustering methods. The Conklin Lab is experienced with experimental biology, bioinformatics, and the development of public bioinformatics software tools (e.g.: GenMAPP, and MAPPFinder). I propose two specific aims for my postdoctoral fellowship: 1. To adapt the HOPACH algorithm for gene expression data analyses. This automated HOPACH algorithm will be written in the R language and will be a contributed package to the R open source statistical software projects. 2. To apply the HOPACH and other analytical tools to a large collection of cardiac and muscle-related gene expression datasets with the goal of developing new visualization methods to analyze gene expression data. The refined gene clusters will be integrated with GenMAPP and other analytical programs so that any biologist can compare expression profile with a larger reference dataset. These studies will begin to define the "topography" of gene expression to help identify new insights into complex biological systems.

DESCRIPTION (provided by applicant): Comparative genomics promises to shed light on those genetic changes that gave rise to the modern human species. Mounting evidence suggests that the vast majority of functional differences between the human and chimpanzee genomes are in regions that do not code for proteins. Focusing on these non-coding regions, we will investigate lineage-specific evolution in the human genome. Our approach includes developing likelihood ratio tests for identifying changes in either the rate or the pattern of nucleotide substitution in a single lineage. These novel methods will be implemented in open source software that can be used to scan an entire genome. We will apply this evolutionary analysis to multiple sequence alignments of human and other vertebrates, including several closely related species (macaque, chimpanzee, Neanderthal), allowing us to identify recent changes in the human genome. In order to concentrate on functionally relevant changes, evolutionary testing will be limited to sets of candidate regions with specific known or predicted functions (e.g. regulatory regions, RNA genes). Predicted functional regions will be identified using machine learning classification techniques. These classifiers will employ measures of sequence conservation as well as the rapidly expanding collection of experimental and bioinformatic annotations of the human genome, including results of the ENCODE Project and other functional genomic studies. After identifying those regions that were most significantly altered in the human lineage, we will use this functional information to develop testable hypotheses about the effects of the observed changes. Experimental investigations of these genomic regions will lead to new understanding of the evolution of human biology and health. PROJECT NARRATIVE: This project will vastly expand knowledge of biologically relevant features of the human genome that are unique to our species. Identification and characterization of the genetic changes leading to modern humans is of fundamental interest. These investigations also promise to contribute to our understanding of the causal mechanisms behind human diseases, leading to directed treatment and prevention strategies.

DESCRIPTION (provided by applicant): Congenital heart defects (CHDs) are among the most common and most devastating birth defects in humans. Networks of transcription factors regulate cardiac cell fate and morphogenesis, and dominant mutations in transcription factor genes lead to most instances of inherited CHD. The mechanisms underlying CHDs that result from disruption of these networks remain to be identified, but regulation of gene expression within a relatively narrow developmental window is clearly essential for normal cardiac morphogenesis. In addition to transcription factors, epigenetic regulation via histone modifications, chromatin remodeling, and non-coding RNAs have key roles in modulating gene expression programs. Elucidating on a genome scale the physical and functional interactions between transcription factors and epigenetic regulators will considerably enhance our understanding of the control of heart development and will have important implications for understanding the mechanistic basis of CHDs. We propose a project as part of the NHLBI Heart Development consortium to provide an integrated epigenetic landscape for heart development, with a focus on CHD-related genes. We propose three major aims. Aim 1: Define genome-wide occupancy maps of transcription factors with known roles in cardiac development and human disease, and epigenetic regulators of transcription, in differentiating cardiomyocytes. Aim 2: Define the global function of transcriptional and epigenetic regulation in heart development and congenital heart disease. We will examine the effect of loss of function of cardiac transcription factors on epigenetic regulation, and alterations in epigenetic regulation in disease-specific induced pluripotent cells from CHD patients. We will also evaluate the global role of histone modifications in mouse heart development. Aim 3;Integrate microRNA expression and function into the regulatory networks governing cardiac development. High-resolution occupancy maps from Aims 1 and 2 will be analyzed specifically for miRNA promoter occupancy and combined with quantitative sequencing of miRNAs in differentiating cardiomyocytes. We will study the function of highly altered miRNAs, specifically those that target disease-causing cardiac transcription factors. Our studies will yield an important and transformative epigenetic atlas of heart development, which will link for the first time transcriptional and epigenetic regulators in a comprehensive network that will illuminate mechanisms underlying CHDs. RELEVANCE (See instructions): The proposed project will for the first time allow a new understanding of the gene networks that underlie congenital heart disease. Congenital heart disease is the most serious childhood illness, affecting 1% of children, and leading to significant mortality and long-term illness. However the underlying causes of these diseases are not understood. Our project will link the so-called "epigenetic regulators" that control how genes are turned on or off, to congenital heart disease, bringing new important insights into these diseases.

The Advanced Bioinformatics Core will analyze and integrate data from multiple sources, e.g., affinity purification followed by mass spectrometry (AP-MS), chromatin immunoprecipitation followed by DNA sequencing (ChlP-Seq), gene expression by RNA-Seq and other methods to identify biologically relevant cellular pathways and processes in cardiac differentiation. It will make use of novel programs such as MiST for AP-MS data and other algorithms developed by the Gladstone Bioinformatics Core to accurately analyze ChlP-Seq and RNA-Seq data. This core will leverage the expertise in the existing Gladstone Bioinformatics Core but focus on protein-protein interaction data and its integration with other gene regulatory datasets to establish combinatorial interactions that control gene expression during cardiac differentiation.

DESCRIPTION (provided by applicant): Autoimmune diseases are debilitating and often life-threatening conditions that afflict a substantial minority of the human population. Aside from some known genetic variations, the causes of autoimmunity remain unclear, as does the reason for the rise in disease prevalence in Western countries. Recent studies of the human microbiome implicate various microbes and microbial proteins in the development of and response to autoimmune diseases. We propose to improve understanding of the relationship between the microbiome and autoimmune disease by investigating how the microbiome interacts with its host over the time course of disease onset and progression. These studies will use a mouse model of inflammatory bowel disease (IBD) where TFGb signaling is blocked on T cells by transgenic expression of a dominant negative form of TGFbRII (DNR). This system enables us to conduct carefully controlled experiments and to probe early, pre- symptomatic time points that are not easily accessible in human clinical studies. Our first aim is to characterize weight change and immunological markers from birth through severe IBD in DNR mice and use this data to identify disease checkpoints (initiation, pre-activation, post-activation severe disease). To explore the relationship between gut microbial communities and autoimmune disease, our second aim will use shotgun metagenomic sequencing and cutting-edge bioinformatics tools to profile the microbiome's protein repertoire at each disease checkpoint in DNR mice versus healthy wildtype (WT) littermates. We will map metagenomic sequencing reads into protein families and pathways and then use generalized linear models to test for significant differences in these physiological profiles between lines. These tests will identify candidate biomarkers that predict IBD onset or progression. Our third aim is to identify temporal biomarkers for IBD, microbial proteins and pathways that have different longitudinal trajectories. We will sequence metagenomes from additional time points?from birth through severe IBD?and use mixed effects models to identify microbial biomarkers that correlate with changes in host immunology and also distinguish DNR and WT mice. Our findings will then be related to metagenomic studies of IBD in humans to develop testable hypotheses about mechanisms of disease induction and to identify genes and pathways from our study that might also serve as inexpensive, early-onset and temporal diagnostics of IBD in humans. The overall goal of this study is to clarify the relationship between IBD and the mammalian gut microbiome. This study will establish the feasibility of using mouse models to study the role of the microbiome in human autoimmune disease and ultimately to develop microbiome-based therapeutics.

? DESCRIPTION (provided by applicant): Abnormal neuronal development can lead to a wide array of mental disorders. Genes important for neurodevelopment have been combed for coding mutations leading to psychiatric disease with limited success, suggesting that other regions in the genome could be causative. A variety of molecular and clinical data indicates that mutations associated with psychiatric disease can reside in gene regulatory sequences such as enhancers. However, only a few enhancers have been definitively linked with these disorders to date. This is primarily because regulatory mutations are challenging to functionally characterize and link to specific genes and phenotypes. To address this challenge, we will use functional genomics data, sequence motifs, and evolutionary signatures to train EnhancerFinder, software that we developed that predicts functional enhancers at high success rates, to now specifically identify active neurodevelopmental enhancers. Over 12,000 candidate neurodevelopmental enhancers will then be cloned and assayed en masse for their enhancer activity using massively parallel reporter assays (MPRAs) in three human embryonic stem cell (hESC) derived neuronal lines: early initiation, neural progenitor cell stage that produces only neurons upon further differentiation, and astrocytes. In addition, we will link enhancers to their target genes using a novel chromatin structure-based prediction approach, called TargetFinder, thereby establishing a network connecting regulatory regions to neurodevelopmental genes. By overlaying reproducible psychiatric disease associated loci with this network, we will identify and prioritize non-coding mutations that are likely to affect expression of neurodevelopmental genes with roles in psychiatric disease. These predictions will be validated using genome- editing techniques to knock out regulatory elements and then assay changes in chromatin interactions and gene expression in developing neurons. The key innovations of our approach are: (i) accurate, quantitative measurements of activity for thousands of psychiatric disease associated enhancer candidates in parallel, (ii) chromatin based inference of gene regulatory networks linking enhancer mutations to genes and pathways, and (iii) a well-characterized stem cell based system to apply these techniques in a high-throughput manner to developing human neurons. We will rapidly disseminate software, reagents, protocols, and datasets to enable follow-up functional studies in the labs of our mental health collaborators and many others. Our long-term aim is to pinpoint causative regulatory variants in the many genomic loci associated with psychiatric disease where an obvious coding mutation is lacking. This approach could easily be adapted to functionally characterize gene regulatory elements involved in other complex human diseases.

PROJECT SUMMARY/ABSTRACT The goal of this P01 project is to identify biomarkers that will enable us to predict the likely duration of the lag phase or ?remission? period prior to HIV rebound following discontinuation of antiretroviral therapy (ART) in HIV-infected individuals. In our study, a large number of virologic and immunologic parameters will be measured in a cohort of ~125 well-characterized HIV-infected individuals undergoing analytical treatment interruption (ATI), to determine if any of these measurements allow us to reliably predict the kinetics of viral rebound post-ART cessation. A large amount of high-dimensional data including next-generation sequencing data (transcriptomes and microRNA profiles) and CyTOF data will be generated in our proposed experiments. The Bioinformatics and Biostatistics Core will play a leading role in compiling, curating, analyzing and disseminating data generated in all three projects associated with this P01 application. To maximize our chances of identifying meaningful signatures predicting time until viral rebound, we will implement several statistical approaches and ensemble learning methods (e.g., gradient boosting, random forests) to develop theories, and we will rely on established classifier performance evaluation procedures (e.g. cross validation, recursive feature elimination, and feature importance measures) to rigorously determine the predictive potential of biomarkers under consideration. In Aim 1 of our Bioinformatics and Biostatistics Core project, we will evaluate the capacity of individual putative blood cell-associated biomarkers studied in Project 2 to predict time until viral rebound following ATI. Measurements include the frequency of replication-competent proviral genomes in CD4+ T cells and global characterization of the host cell transcriptome. In Aim 2, we will evaluate the capacity of individual putative cell- free plasma- and CSF-derived biomarkers studied in Project 3 to predict time until viral rebound following ATI. Measurements include circulating microRNA profile, extracellular vesicle phenotype, and multiplex cytokine and antibody characterization. Lastly, in Aim 3, we will perform a combined analysis of biomarkers across all 3 projects (including CyTOF immunophenotypic data generated in Project 1) to assess their relative performance and to identify potential synergies between predictors. Ensemble learning methods are ideal for discovering complex combinations of predictive features. They also provide a framework for evaluating the predictive importance of candidate biomarkers both individually and in combination with other biomarkers. The Bioinformatics and Biostatistics Core will play a central role in achieving our P01 objectives and in advancing the HIV cure agenda, transforming copious and diverse, high-dimensional data into robust predictors of HIV rebound following ART interruption.

Project Summary / Abstract Abnormal neuronal development can lead to a wide array of psychiatric disorders. Mutations disrupting protein coding genes have been found to cause some of these disorders but a large number of them still remain unsolved. A variety of molecular and clinical data suggests that mutations in gene regulatory sequences could be a major contributor to these disorders. However, only a few causal regulatory mutations have been found to date. This is primarily because functional regulatory elements are difficult to identify, particularly in mixed cell populations such as the developing brain. In addition, these elements are difficult to functionally characterize in a high-throughput manner in these cell types. To address these challenges, we propose to use novel single-cell genomic technologies along with massively parallel reporter assays (MPRAs) in human primary cells and organoids to characterize thousands of brain development associated genes, regulatory elements and pathways. First, using single cell RNA-seq (scRNA-seq) and ATAC-seq (sci-ATAC-seq) across multiple cortical areas and subcortical regions of developing human brain at three development stages, we will generate a comprehensive map of genes, regulatory elements and networks involved in human brain development (Aim 1). Next, we will use similar techniques (scRNA-seq and sci-ATAC-seq) on human cerebral organoid cultures derived from induced pluripotent stem cells (iPSCs). We will compare regulatory programs in organoid cells to cells present during normal human brain development. To assess the contribution of key transcription factors involved in psychiatric disorders to gene regulatory pathways in the developing brain, we will use genome editing on the same genetic background to create heterozygous loss-of-function mutations in key transcription factors involved in psychiatric disorders and assess their effects on gene expression (scRNA-seq) and gene regulation (sci- ATAC-seq) (Aim 2). Finally, we will functionally characterize over 37,500 candidate enhancers and nucleotide variants within them using a lentiviral-based MPRA (lentiMPRA) in disease-relevant cell types purified from human primary cells and organoids. Several of these sequences will also be assayed in organoids lacking key transcription factors deleted in Aim 2 to test the importance of these genes to regulatory activity and to identify interactions with regulatory variants (Aim 3). Data from all aims will be used to build predictive models of gene expression and enhancer activity as a function of regulatory sequences, which will be used to design lentiMPRA libraries and iteratively improve models using results from initial libraries. Combined our project will use cutting- edge techniques such as scRNA-seq, sci-ATAC-seq and MPRA coupled with advanced computational analyses to significantly increase the number of functionally characterized human brain developmental regulatory elements and how their activity changes in the presence of disease associated mutations to shed light on the genetic basis for psychiatric disorders.

PROJECT SUMMARY/ABSTRACT CORE B ? ADVANCED BIOINFORMATICS CORE The Advanced Bioinformatics Core will provide innovative and collaborative computational support for all three Projects, in close partnership with the Advanced Proteomics and Genome Editing Cores. We have several decades of experience in developing statistics and bioinformatics methods for genomics and proteomics, including open source software for analysis and visualization. For nearly 10 years, we have been working closely with the Srivastava, Bruneau, and Black labs to dissect cardiac regulatory mechanisms using the latest tools for data quantification and integrative analysis. These investigations identified several key cardiac transcription factors and chromatin-remodeling proteins that can alter cardiac development and cause CHDs when mutated. Yet we do not fully understand the mechanisms through which these regulatory proteins function. It is clear from our preliminary studies that mutations alter their subcellular localization, protein interactions, and genomic occupancy. New techniques in proteomics (e.g., APEX-MS) and genomics (single- cell RNA-seq) promise to shed light on how these effects translate into disease phenotypes through altered gene regulatory networks. But there are major challenges quantifying and jointly analyzing these diverse data types. The objective of the Advanced Bioinformatics Core is to rule out chance, bias and confounding in the data from Projects 1?3, using state-of-art analyses tools when available and developing new techniques where needed. Our rigorous statistical approach will enable the Core to integrate data within and across projects to decode novel mechanisms of CHD etiology that would otherwise be missed.

Gene regulation is an important determinant of the complex specialization of cells in the human brain, and nucleotide changes within regulatory elements contribute to risk for psychiatric disorders. We therefore hypothesize that these debilitating diseases are driven in part by genetic variants that alter gene expression and disturb the balance and function of cell types in brain tissue. Single-cell open chromatin assays are a promising approach to testing this hypothesis by mapping variants to regulatory elements specific to and shared across cell populations. There are two major barriers to this strategy, for which our project proposes modeling solutions. First, despite being the best assay currently, single-cell ATAC-sequencing (scATAC-seq) suffers from low resolution, meaning that an open chromatin region may be supported by zero or few reads in a given cell. This makes it hard to identify coherent cell populations. We propose a network model for semi-supervised clustering of cells in scATAC-seq that leverages information from higher-coverage bulk tissue experiments and single-cell RNA-sequencing (scRNA-seq), if available. The expected outcomes from applying this model to compendia of brain data from public repositories and our collaborators are (i) identification of open chromatin regions that differentiate cell types and states, and (ii) discovery of resolved cell populations whose open chromatin is enriched for psychiatric disorder associated genetic variants. These results alone may not be enough to develop a mechanistic understanding of how variants impact brain function. To address this second challenge, we will implement a computationally efficient, machine-learning framework for predicting the specific regulatory functions of single-cell open chromatin regions from our network model and other approaches. Gene regulatory enhancers are particularly amenable to this approach, because high-throughput mouse transgenics and massively parallel reporter assays have generated enough validated enhancers for supervised learning. Our framework will be easy to apply to other regulatory functions, such as insulating boundaries in chromatin capture data. By developing a compressed, yet flexible, featurization of massive bulk and single-cell data compendia, we will enable rapid iteration with computationally intensive prediction algorithms to be applied to single-cell open chromatin regions. Our approach will also incorporate transfer learning from data-rich (e.g., postmortem or mouse brains) to data-poor settings (e.g., human late-gestation brains). We expect predicted regulatory elements to be more enriched for psychiatric disorder genetic risk, to provide mechanistic insight regarding how variants cause disease, and to be useful molecular tools. Together our two proposed computational approaches will leverage the complementary strengths of bulk and single-cell data to resolve regulatory elements that drive the exquisite diversity of cells in developing and adult brains towards mapping the non-coding contribution of psychiatric disease.

CORE B ? ABSTRACT This program aims to discover the molecular drivers and consequences of network dysfunction in Alzheimer?s disease (AD) through rigorous characterization of cell-type specific gene regulation and multi-modal phenotypes. We will use human samples and a variety of mouse models. This breadth and depth of data across different organismal and cellular contexts present a unique opportunity for integrative modeling. To capitalize on this opportunity, however, the data must be quantitatively comparable across projects. To address this challenge, the Integrative Data-Science Core (Core B) will use the ?design for inference? approach, which means that the predictive modeling and hypothesis testing we plan to do will guide all stages of experimental design. To minimize and correct batch effects, we will standardize experimental protocols and establish a repeated-measures experimental design, which will boost the power for analyses. A second challenge is how to summarize and jointly model complex, high-dimensional phenotypes with single-cell and single-nucleus transcriptomic profiles. To solve this problem, we will develop innovative machine-learning and network models, with a focus on deep learning and sparse canonical correlation analysis to extract information from multivariate data and discover relationships between pairs of data types. To facilitate real-time sharing of results and exploration of data across projects, we will implement data tracking systems, Jupyter notebooks with pipelines and analytical code, and an interactive data portal with visualization and query capabilities. These collaborative tools will also help us share our data, code, and results rapidly with the AD research community through our Synapse website. Collectively, the activities of Core B will provide cutting-edge computational support to all four projects, enable cross-project discovery, and set new standards for the use of large-scale data integration to decipher molecular mechanisms in AD and other diseases.

PROJECT SUMMARY A major unanswered question is how chromatin topology coordinates human development and cellular differentiation, and how genome folding is differentially regulated in human disease. It is thought that three- dimensional (3D) chromatin organization is driven by transcriptional regulators, but fundamental mechanisms of this regulation as it relates to disease-relevant human cells have not been well explored. We propose to elucidate the temporally dynamic 3D nucleome (4DN) that underlies human cardiac differentiation, its molecular underpinnings, and the impact of mutations that underly defective 4DN organization in human congenital heart disease (CHD). CHDs are the most common birth defect and arise from abnormal heart development. The genetic basis of CHD is largely mutations in genes encoding chromatin modifiers (e.g. WDR5, KMT2D) and transcription factors (TFs, e.g. TBX5, GATA4), many of which also cause adult-onset arrhythmias. The impact of CHD mutations on the 4DN has not been explored. We hypothesize that 3D genome folding is highly regulated during cardiac differentiation and is impacted by disease-causing mutations in transcriptional regulators and non-coding elements. We will use iPS cell models and machine learning to elucidate dynamic 3D chromatin organization in human cardiomyocytes and endothelial cells during normal and diseased cardiac differentiation. We propose 3 specific aims: Aim 1: Establish a kilobase-scale 4D map of genome folding in human cardiomyocytes (CM) and endothelial cell (EC) differentiation. We will use directed differentiation of human iPS cells towards the two major cell types of the developing heart: CMs and ECs, and using microC across a fine time course of differentiation we will define at kilobase scale the 3D organization of the genome, capturing the states of developmental intermediates and the final differentiated cells. This aim will generate an essential integrated 4DN template for discovery in cardiac differentiation. In Aim 2: we will Determine the regulatory and disease-related basis for cardiac 3D chromatin organization. We will perform microC in iPS cell lines with CHD-associated mutations in transcriptional regulators, differentiated into CMs and ECs. These findings will establish the degree to which CHD is caused by abnormal genome folding and chromatin states, with important relevance to other human cardiovascular diseases. Finally, Aim 3 will address High-throughput screening of millions of CHD and synthetic non- coding mutations with a deep-learning model of dynamic genome folding. We will build a deep-learning model predicting 3D chromatin contact frequencies across cardiac differentiation at kilobase-resolution. By introducing thousands of CHD patient deletions and other non-coding mutations in silico, we will prioritize variants likely to interact with transcriptional regulators to cause disease through disrupted genome folding. Several candidates will be validated in engineered iPS cells differentiated into CMs and ECs. These results will provide a novel platform for computational discovery of disease variant impact across diverse human diseases