Functional genomics platform for pooled screening and generation of mammalian genetic interaction maps

Abstract

Systematic genetic interaction maps in microorganisms are powerful tools for identifying functional relationships between genes and for defining the function of uncharacterized genes. We have recently implemented this strategy in mammalian cells as a two-stage approach. First, genes of interest are robustly identified in a pooled genome-wide screen using complex shRNA libraries. Second, phenotypes for all pairwise combinations of 'hit' genes are measured in a double-shRNA screen and used to construct a genetic interaction map. Our protocol allows for rapid pooled screening under various conditions without a requirement for robotics, in contrast to arrayed approaches. Each round of screening can be implemented in ∼2 weeks, with additional time for analysis and generation of reagents. We discuss considerations for screen design, and we present complete experimental procedures, as well as a full computational analysis suite for the identification of hits in pooled screens and generation of genetic interaction maps. Although the protocol outlined here was developed for our original shRNA-based approach, it can be applied more generally, including to CRISPR-based approaches.

Figure 1

Overview of the experimental procedure. (a) Primary genome-wide shRNA screen, in which hit genes are identified. (b) Cloning of individual shRNAs against hit genes from the primary screen, and confirmation of their phenotype in a batch retest. (c) Construction and screening of a double-shRNA library representing all pairwise combination of selected shRNAs from the batch retest, and calculation of a systematic genetic interaction map.

Figure 2

Lentiviral plasmids used in the experimental procedure (not drawn to scale). The EF1A promoter drives expression of an mRNA that encodes a puromycin resistance marker, an mCherry fluorescent marker, and either one or two shRNAs embedded in a minimal miR-30 context. (a) Vector pMK1047 contains the pooled, ultracomplex genome-wide shRNA library. The shRNA expression cassette is flanked by PvuII sites that are 1,122 bp apart; this restriction enzyme is used to size-fractionate genomic DNA from cell populations transduced with pMK1047. Primers oMCB800 and one of the primers oMK312 or oMK313 (shown in turquoise; sequences in Table 1) are used to amplify the shRNA cassette from genomic DNA, and primer oMK132 is used as primer for Illumina sequencing to identify the shRNA, starting with the guide strand. (b) Vector pMK1200 is used to clone individual shRNAs. This vector is a pool of plasmids with random N₁₀ barcodes upstream and downstream of the minimal miR-30 context. Annealed oligonucleotides encoding the shRNA of interest are ligated into BstXI-cut pMK1200. The combinations of barcodes and shRNA inserts are identified by sequencing with primers 5′ pSico-Eco or 3′ pSico-Pci (shown in purple; sequences in Table 1). The shRNA expression cassette is flanked by PvuII sites that are 1,177 bp apart. Primers oMCB922 and one of the primers oMK371 or oMK378 (shown in turquoise; sequences in Table 1) are used to amplify the shRNA cassette from genomic DNA, and the Illumina genomic sequencing primer (shown as SP) is used as primer for Illumina sequencing to identify the downstream barcode. (c) A double-shRNA library is created by digesting a pool of barcoded shRNA cloned in the pMK1200 backbone. This pool is digested with AvrII + KpnI to obtain an insert and with XbaI + KpnI to obtain a backbone, the ligation of which results in a double-shRNA library. The double-shRNA expression cassette is flanked by PvuII sites that are 1,369 bp apart. Primers oMCB847 and one of the primers oMK371 or oMK378 (shown in turquoise; sequences in Table 1) are used to amplify the barcode junction portion from genomic DNA. Note that although both primers will anneal to two places within the double-shRNA cassette, only one amplicon will be generated by PCR. The Illumina genomic sequencing primer (shown as SP) is used as primer for Illumina sequencing to identify the combinatorial barcode.

Figure 3

Gel purification steps during sample preparation. (a) Size fractionation of genomic DNA. Genomic DNA isolated from a screen cell population is digested with PvuII and the size range between 1,000 and 1,650 bp, which contains the shRNA expressing cassette for all constructs shown in Figure 2, is excised from an ethidium bromide-stained 0.8% agarose gel, as indicated by the dashed red box, and purified for further use as a PCR template. (b) PCR products are separated on a 20% polyacrylamide gel and visualized by ethidium bromide staining. In addition to the desired PCR product (in the example shown a 190 bp-long product amplified from genomic DNA of cells transduced with a double-shRNA library), shorter products result from long primer dimers (the unwanted band is present in the control reaction lacking template, but not the control reaction lacking reverse primer).

Figure 4

Graphical output of the script analyze_primary_screen.py (screenshot). (a) Each dot represents one shRNA. X-axis: log₂-transformed deep-sequencing counts of cells expressing this shRNA in the t0 population. Y-axis: log₂-transformed deep-sequencing counts of cells expressing this shRNA in the untreated population at the endpoint of the experiment. Black circles represent negative-control shRNAs. Colored circles represent shRNAs designed to target genes; they are colored according to a heatmap representing the growth phenotype (gamma). (b) Distribution of growth phenotypes (gammas) for negative control shRNAs (blue dots) and shRNAs designed to target genes (red dots). Grey dashed lines indicate the range of phenotypes containing the middle 95% of negative control shRNAs.

Figure 5

Graphical output of the script analyze_primary_screen.py (screenshot) (a) Each dot represents one gene. Log₁₀-transformed P values for an effect caused by gene knockdown were calculated by the Mann-Whitney test. Negative values indicate depletion of cells expressing shRNAs targeting the gene of interest from the population, positive values indicate enrichment. X axis: untreated cell population, Y axis: treated cell population. (b) This graph was generated interactively using the gene_rho() command after running the script (Step X). Each dot represents an shRNA. X-axis: log₂-transformed deep-sequencing counts of cells expressing this shRNA in the untreated population at the endpoint of the experiment. Y-axis: log₂-transformed deep-sequencing counts of cells expressing this shRNA in the treated population at the endpoint of the experiment. Grey semi-transparent dots represent negative control shRNAs. Solidly colored dots represent shRNAs targeting the gene VPS54; they are colored according to a heatmap representing the resistance phenotype (rho).

Figure 6

Graphical output of the script double_shRNA_phenotypes.py (screenshot). A two-dimensional histogram (implemented as hexagonally binned heatmap) of AB-BA differences in phenotype (rho) as a function of the average count numbers of the AB and BA double shRNAs in the t0 population.

Figure 7

Graphical output of the script calculate_GIs.py (screenshot). This graph was generated interactively using the ab_rho_sd_yellow_blue() command after running the script. Each dot represents an shRNA. X axis: average phenotype (rho) of double-shRNAs containing this shRNA and one of 12 negative control shRNAs. Y axis: phenotype (rho) of the double-shRNA containing the same shRNA in combination with shRNA SEC23B_4. Values for shRNAs with SEC23B_4 in the first and second position of the shRNA were averaged. Additionally, expected double-shRNA phenotypes according to three different definitions are indicated by solid lines: red, linear fit orange, sum definition; pink, product definition. These definitions are detailed in ref. .

Figure 8

High-density genetic interaction map created by hierarchical clustering of the genetic interaction matrix in Cluster 3.0 (ref. ), and displayed in Java Tree View. Genes encoding proteins that interact physically (e.g. COPI complex subunits) or functionally (e.g. ARF1 and its nucleotide exchange factor GBF1) form clusters and typically show buffering genetic interactions with each other.