Abstract Body: With a prevalence of 4:10,000, dilated cardiomyopathy (DCM) is the most common hereditary heart disease, characterized by dilatation of one or both ventricles and impaired systolic and diastolic function. Numerous mutations have been identified to cause DCM. TITIN-truncating variants (TTNtv) are commonly observed in 20% of the genetic DCM cases. Progress in drug discovery for DCM has been hampered by the lack of robust high-throughput/high-content analysis pipelines of disease-specific cell and tissue platforms for deep structural and functional phenotyping.
To address this challenge, we have developed a two-tiered human induced pluripotent stem cell (iPSC)-based screening approach, comprised of high-content cardiomyocyte imaging with artificial intelligence (AI)-driven image and data analyses as well as deep tissue phenotyping in engineered heart muscle (EHM) models. As a DCM use case, we have developed several isogenic human iPSC-lines with and without a previously reported and pathologically relevant A-band-TTNtv mutation (c.70692_70693insAT; p.T23565SfsX5; Gerull et al. 11788824, Gramlich et al. 25759365, Fomin et al. 34731013)
Using fixed weights from a convolutional deep neural network trained on ImageNet, we generated unbiased deep embeddings from each 2D cell model and applied these to train machine learning (ML) models to detect morphological disease phenotypes. Our ML model is able to confidently separate healthy controls from TTNtv cells with high accuracy (>93%), specificity (WT >79%) across different batches/plate layouts and generate concentration-response curves, demonstrating platform robustness and sensitivity. Building upon this work, we will present our latest efforts applying the platform technology for a chemogenomic screening campaign to identify novel cardioprotective drugs using a library of 7,100 bioactive compounds and validation assays, including measuring functional changes in EHMs.
The phenotypic profiling platform, as presented here, can be readily adapted to other disease-relevant cell types and pathologies. Our results demonstrate the power of combining iPSC-CMs with cytological profiling and deep learning as well as tissue-level phenotyping to accelerate drug discovery for patients with DCM.