Optical Pooled Screens in Human Cells

Abstract

Genetic screens are critical for the systematic identification of genes underlying cellular phenotypes. Pooling gene perturbations greatly improves scalability but is not compatible with imaging of complex and dynamic cellular phenotypes. Here, we introduce a pooled approach for optical genetic screens in mammalian cells. We use targeted in situ sequencing to demultiplex a library of genetic perturbations following image-based phenotyping. We screened a set of 952 genes across millions of cells for involvement in nuclear factor κB (NF-κB) signaling by imaging the translocation of RelA (p65) to the nucleus. Screening at a single time point across 3 cell lines recovered 15 known pathway components, while repeating the screen with live-cell imaging revealed a role for Mediator complex subunits in regulating the duration of p65 nuclear retention. These results establish a highly multiplexed approach to image-based screens of spatially and temporally defined phenotypes with pooled libraries.

Keywords: CRISPR; functional genomics; high-content screening; in situ sequencing; optical pooled screen; pooled screen.

Figure 1.. Optical pooled genetic screens

(A) In pooled screens, a library of genetic perturbations is introduced, typically at a single copy per target cell. In existing approaches, cellular phenotypes are evaluated by bulk NGS of enriched cell populations or single-cell molecular profiling (e.g. single-cell RNA-seq). In optical pooled screens, high-content imaging assays are used to extract rich spatiotemporal information from the sample prior to enzymatic amplification and in situ detection of RNA barcodes, enabling linkage between the phenotype and perturbation genotype of each cell. (B) Targeted in situ sequencing is used to read out RNA barcodes expressed from a single genomic integration. Barcode transcripts are fixed in place, reverse transcribed, and hybridized with single-stranded DNA padlock probes, which bind to common sequences flanking the barcode. The 3’ arm of the padlock is extended and ligated, copying the barcode into a circularized ssDNA molecule, which then undergoes rolling circle amplification. The barcode sequence is then read out by multiple rounds of in situ sequencing-by-synthesis. See also Figure S1A.

Figure 2.. Identification of perturbation barcodes by in situ sequencing

(A) Schematic of perturbation detection by in situ sequencing. Barcodes representing perturbations are expressed on a Pol II transcript and enzymatically converted into cDNA and amplified by RCA. RCA products serve as templates for sequencing-by-synthesis, in which barcodes are read out by multiple cycles of fluorescent nucleotide incorporation, imaging and dye cleavage. (B) A 125-nt oligo pool encoding perturbations (sgRNAs) and associated 12-nt barcodes was cloned into a lentiviral vector (lentiGuide-BC) and delivered into HeLa cells. Expressed barcode sequences were read out by padlock detection, rolling circle amplification, and 12 cycles of sequencing-by-synthesis. A linear filter (Laplacian-of-Gaussian, kernel width σ = 1 pixel) was applied to sequencing channels to enhance spot-like features (scale bar 10 μm; composite image of DAPI and four sequencing channels). See also Movie S1. (C) >80% of barcodes map to 40 designed sequences out of 16.7 million possible 12-nt sequences. See also Figure S1B-D. (D) Most cells contain multiple barcode reads that map to the designed library. (E) Cellular read distribution further categorized by read identity for 77% of cells containing at least one read. (F) The number of possible barcodes scales geometrically with barcode length. Sufficient 12-nt barcodes can be designed to cover a genome-scale perturbation library while maintaining the ability to detect and reject single or double sequencing errors (minimum pairwise Levenshtein distance d = 2 or 3, respectively).

Figure 3.. Accuracy of phenotype-to-genotype mapping assessed with a fluorescent reporter

(A) Workflow for CRISPR-Cas9 knockout-based screening of a genetically-encoded frameshift reporter. A library of targeting and non-targeting guides was cloned into either LentiGuide-BC or the CROP-seq vector and transduced into cells at low MOI. Cas9 expression generates indels at the frameshift reporter target locus in cells with a targeting guide and leads to expression of a nuclear-localized HA epitope. HA expression was assayed by immunofluorescence and correlated with sgRNAs detected by in situ sequencing. Frameshift reporter accuracy was estimated using the relative abundances of HA⁺ cells mapped to targeting and non-targeting guides (X and Y, respectively). Scale bar is 30 μm. (B) Targeting and control barcodes expressed from LentiGuide-BC in HeLa-TetR-Cas9 cells were well separated by fraction of HA⁺ cells. (C) The same cell library was screened by flow sorting cells into HA⁺ and HA^- bins and performing next-generation sequencing of the genomically integrated barcode. See also Figure S3D. (D) The experiment was repeated across a panel of cell lines using the CROP-seq library and an optimized padlock detection protocol, yielding a similar distribution of mapped reads (top) and frameshift reporter accuracies (bottom). Error bars indicate the range between two replicate sequencing experiments. Cell types are indicated by the same colors in both plots.

Figure 4.. A screen for regulators of NF-κB signaling

(A) Workflow for CRISPR-Cas9 knockout-based screening using a fluorescently tagged reporter cell line. Screen hits were identified by the failure of p65-mNeonGreen to translocate to the nucleus following stimulation with IL-1β or TNFα cytokines. (B) Known NF-κB regulators were identified as high-ranking screen hits. Cells were assigned translocation scores based on the pixelwise correlation between mNeonGreen fluorescence and a DAPI nuclear stain; thus a score 1 indicates maximum translocation while a score of −1 indicates maximum cytoplasmic localization. The translocation defect for a gene was defined based on the integrated difference in the distribution of translocation scores relative to non-targeting control sgRNAs across three replicate screens. See also Tables S1 and S3. (C) Cumulative distributions of translocation scores (second-ranked guide) of known NF-κB regulators in response to both cytokines. The shaded areas depict the difference between the translocation score distributions for targeting sgRNAs and non-targeting control sgRNAs (gray). (D) NF-κB pathway map (KEGG HSA04064) color-coded as in (B). KEGG pathway members colored gray did not show a translocation defect when individually knocked out in HeLa cells. (E) Top-ranked genes were validated with arrayed CRISPR-Cas9 knockouts (scale bar 10 μm). Histograms show the cumulative distributions of IL-1β and TNFα-induced translocation scores (averaged over two guides) for each gene knockout compared to wild type cells (gray). See also Figure S4B.

Figure 5.. A live-cell screen identifies a role for Mediator components in regulating p65 kinetics

(A) The initial 952 gene translocation screen was repeated using live-cell imaging to monitor p65 translocation kinetics (n = 361,587 analyzed cells). Hierarchical clustering of the time-dependent translocation difference between each gene and the non-targeting controls grouped together KEGG-annotated regulators, as well as other positive and negative regulators with distinct cytokine-specific kinetic signatures. See also Table S4. (B) and (C) A validation live-cell screen was performed to improve sampling of regulators identified by the primary screen and helped group regulators with distinct p65 translocation kinetics. Individual traces represent different sgRNAs for each gene. See also Figure S5 and Table S4. (D) Clonal knockouts of MED12 and MED24 recapitulated the increased retention time phenotype seen in the primary screen. Each trace represents a different knockout clone (scale bar 10 μm). (E) Cytokine stimulation of Mediator clonal knockouts led to differential activation of NF-κB target genes, including increased expression of chemokines after stimulation with TNFα (1 ng/mL) or IL-1β (30 ng/mL). Error bars show the range among knockout clones or wild type biological replicates, clones same as in (D). (F) IL1B expression was induced in the Mediator clonal knockouts, but not in wild type cells, by TNFα stimulation.