Disrupted developmental signaling induces novel transcriptional states

Signaling pathways induce stereotyped transcriptional changes as stem cells progress into mature cell types during embryogenesis. Signaling perturbations are necessary to discover which genes are responsive or insensitive to pathway activity. However, gene regulation is additionally dependent on cell state-specific factors like chromatin modifications or transcription factor binding. Thus, transcriptional profiles need to be assayed in single cells to identify potentially multiple, distinct perturbation responses among heterogeneous cell states in an embryo. In perturbation studies, comparing heterogeneous transcriptional states among experimental conditions often requires samples to be collected over multiple independent experiments. Datasets produced in such complex experimental designs can be confounded by batch effects. We present Design-Aware Integration of Single Cell ExpEriments (DAISEE), a new algorithm that models perturbation responses in single-cell datasets with a complex experimental design. We demonstrate that DAISEE improves upon a previously available integrative non-negative matrix factorization framework, more efficiently separating perturbation responses from confounding variation. We use DAISEE to integrate newly collected single-cell RNA-sequencing datasets from 5-hour old zebrafish embryos expressing optimized photoswitchable MEK (psMEK), which globally activates the extracellular signal-regulated kinase (ERK), a signaling molecule involved in many cell specification events. psMEK drives some cells that are normally not exposed to ERK signals towards other wild type states and induces novel states expressing a mixture of transcriptional programs, including precociously activated endothelial genes. ERK signaling is therefore capable of introducing profoundly new gene expression states in developing embryos. Significance StatementSignaling perturbations produce heterogeneous transcriptional responses that must be measured at the single-cell level. Data integration techniques allow us to model these responses which, however, can be confounded by batch effects. We present a computational tool (DAISEE) for extracting the common and perturbation-specific features of single-cell datasets representing multiple experimental conditions while achieving efficient batch effect correction. DAISEE outperforms its predecessor and will enable accurate analysis of a broad range of single-cell datasets. DAISEE applied to new single-cell RNA sequencing data from zebrafish embryos shows that gain-of-function signaling perturbations can induce novel states. Our analysis suggests that a wild type endothelial cell-specification program can be activated in abnormal developmental contexts when the extracellular signal-regulated kinase (ERK) pathway is deregulated.