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, which can introduce confounding batch effects. We present Design-Aware Integration of Single Cell ExpEriments (DAISEE), a new algorithm that models perturbation responses in single-cell datasets collected according to complex experimental designs. We demonstrate that DAISEE improves upon a previously available integrative nonnegative 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-h-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 toward other wild type states and induces novel states expressing early-acting endothelial genes. Overactive signaling is therefore capable of producing unexpected gene expression states in developing embryos.