Novel machine vision-based cytometry approach for high-throughput phenotypic pooled CRISPR screening

Functional genomics screening of subcellular phenotypes is currently restricted to detection by imaging using conventional high resolution microscopy, therefore prohibiting scaling up to large pooled screening libraries. Here we present a novel phenotypic screening approach using high-throughput machine vision-based cell sorting (ViCS) and sequencing analysis to study morphological subcellular phenotypes by flow in a highly scalable and comprehensive manner. ViCS combines the use of high-speed imaging in flow and machine learning analysis without image reconstruction to identify and sort cells exhibiting phenotypes of interest. As a proof of concept, we have demonstrated the application of this method in a CRISPR pooled screening using the human monocytic leukemia cell line THP-1 to identify genes regulating the nuclear translocation of nuclear factor kappa B (NF-κB) upon lipopolysaccharide (LPS) stimulation.  We screened a pathway specific library of 7,290 sgRNAs targeting 729 kinase genes in a pooled format using fluorescent staining of NF-kB nuclear localization as a readout for its activation. First, we created a machine learning training model using cells with and without LPS stimulation, which resulted in a classification score of 0.981 that allows for the accurate separation of the phenotypes. Then, a pool of Cas9 expressing THP-1 cells transduced with the kinase CRISPR library was stimulated with LPS, stained for NF-kB and cells without nuclear translocation of NF-κB were sorted. NGS analysis of the sorted cells identified significant enrichment of sgRNAs targeting genes downstream of the TLR4 signaling including MAP3K7, IRAK4, IKBKB, and IKBKG, which upon deletion prevented the translocation of NF-kB.  Overall, we demonstrated that our ViCS technology is well-suited for high-throughput pooled phenotypic CRISPR screening, and can provide significant advantages over traditional arrayed screens such as, (1) scalability to large screening libraries without any further modification, (2) applicability to a variety of morphological phenotypes in both adherent and suspension cells, (3) and compatibility with commercial sequencing platforms to identify target gene perturbations.