Systematic reconstruction of molecular pathway signatures using scalable single-cell perturbation screens

In this study, we introduce a highly scalable Perturb-seq workflow and apply it to perturb signaling regulators across 30 separate biological contexts. We demonstrate how Perturb-seq is compatible with recent and commercially available advances in combinatorial indexing as well as a new sequencing-by-synthesis platform, sequencing 2.6 million cells in total. We performed Perturb-seq experiments in six different cancer cell lines from different tissues of origin: A549 (lung), MCF7 (breast), HT29 (colon), HAP1 (bone marrow), BxPC3 (pancreas), and K562 (bone marrow). To facilitate multiplexed gene knockdown screens, we modified each of these lines to express a CRISPR interference (CRISPRi) dCas9-KRAB-MeCP2 cassette. To explore different environmental signaling contexts, we exposed each cell line to five distinct stimuli representing well-established pathway regulators, each of which has been broadly implicated in cellular responses and disease pathogenesis. Our selected pathways included interferon-beta (IFNB), interferon-gamma (IFNG), transforming growth factor beta (TGFB), tumor necrosis factor-alpha (TNFA), and insulin (INS). Together, our combination of six cell lines and five stimuli created a diverse matrix of biological samples, which we used as input to Perturb-seq.Our goal is to characterize the molecular responses after the perturbation of known regulators. Therefore, for each pathway, we selected 44 to 61 genes based on a literature review of known regulators. For each gene, we selected three independent single guide RNAs (sgRNA) from the Dolcetto genome-wide CRISPRi library, as well as 14 non-targeting (NT) controls. We next created five pooled sgRNA libraries, one independently created for each pathway. For each pathway, we separately infected all six cell lines with the pathway-specific sgRNA library, and then stimulated infected cells with the corresponding cytokine for 24 hours to activate signaling. This process resulted in a total of 30 distinct Perturb-seq samples.