Systematic perturbation of 1983 transcription factors during human cardiomyocyte differentiation reveals that congenital heart disease genes are organized in a scale-free regulatory network where CHD pathogenesis commonly involves altered cardiac differentiation programs, and this experimentally-derived regulatory architecture—extended to include enhancer-TF-gene relationships—enables deorphanization of novel CHD transcription factors, interpretation of non-coding variants, and machine learning-based prediction of dysregulated transcription factors from patient-derived cardiomyocyte transcriptomes.

Over 100 genes have been implicated in congenital heart disease (CHD), yet the genetic basis for >50% of CHD remains unexplained. A key challenge is to define the regulatory architecture of CHD genes. Here, we systematically perturb 1983 transcription factors (TFs) during cardiomyocyte differentiation of human stem cells. Our analysis links TFs to gene expression phenotypes across scales and nominates TFs in cardiac cell fate commitment, dosage sensitivity, and transposable element regulation. By deriving TF-gene regulatory networks from experimental perturbation, we gain insights into how developmental networks are structured, use this map to interpret the regulatory architecture of CHD, and deorphanize TFs with roles in CHD that have been under-sampled by genetic studies. To extend these networks to include enhancer-TF-gene linkages, we also perturb 981 putative enhancers of TFs. Finally, we develop a deep learning transformer model to accurately predict perturbed TFs from altered transcriptomes, which we apply to interpret CHD patient transcriptomes. This reference map represents a foundational platform to model the functions of TFs and aids in interpreting CHD variants and mechanisms.

Highlights

Systematic perturbation of 1983 human TFs in cardiac differentiation.

Defining the regulatory architecture of CHD genes and cardiac cell fate commitment.

Nominating TFs in dosage sensitivity and regulation of transposable elements.

Extended enhancer-TF-gene networks improve fine mapping of patient-derived variants.

Perturbation models nominate TFs driving altered regulatory networks in patients.