SUMMARY
Human heart development represents an ideal model system for understanding 1) how normal development can
produce all the cell types necessary for cardiac function and 2) how genetic variation can perturb this process
and lead to disease. We will generate large-scale single cell data sets that will enable the development of
accurate computational models capable of predicting the effects of both genetic changes to regulatory elements
(REs) and perturbations to trans-acting regulatory factors on gene expression during the complex developmental
process of human heart development. We will study a medically relevant, human, in vitro, temporally dynamic,
3D cardiac organoid differentiation system that faithfully recapitulates fetal differentiation patterns for
differentiation towards various cell types, including cardiomyocyte, neural crest, and cardiac endothelium. For
each of these differentiation trajectories, we will work in distinct aims toward mapping, perturbing, modeling,
and model validation: Mapping: we will generate systematic, single cell multi-omic (RNA-seq, ATAC-seq, and
protein quantification) and bulk data to map REs, chromatin contacts, RNA polymerase, and gene expression
through differentiation of human induced pluripotent stem cells to heart tissue. Perturbing: We will use CRISPR-
based methods to comprehensively perturb transcription factors (TFs) required for the different differentiation
trajectories, and map the single-cell gene regulatory and expression impact of perturbing these factors at multiple
time points across these differentiation. Modeling: We will develop multi-input, nucleotide-resolved neural
networks to learn dynamic gene regulatory networks using these mapping and perturbation data sets. These
models will aim to understand the changing landscape of regulation and grammars of TF motifs over
differentiation time and will predict both chromatin and gene expression effects expected from genetic
perturbations. Model validation: We will apply our network models to identify, investigate, and experimentally
test perturbations relevant to understanding disease variation, by knocking down TFs, perturbing REs, and
editing disease-associated noncoding variants. Finally, we will extract and test molecular properties of TF
function from validated models. Successful completion of our project will provide mechanistic interpretations for
how genetic variants may impact development (by disrupting REs that in turn disrupt gene expression) in human
heart development. Our Stanford team comprises a diverse collection of investigators with a history of
collaboration and work in consortia, with expertise in genomics methods development (Greenleaf, Engreitz,
Bassik), single cell methods and analysis (Greenleaf), heart development and disease genetics (Gifford,
Quertermous), and deep learning for genomic data sets (Kundaje). This project will produce gold-standard data
defining the trans-acting factor network driving heart development, and a model capturing these complex
dynamics capable of quantitatively linking changes in genotype to effects phenotype relevant to human disease.
