A systematic mammalian genetic interaction map reveals pathways underlying ricin susceptibility

Abstract

Genetic interaction (GI) maps, comprising pairwise measures of how strongly the function of one gene depends on the presence of a second, have enabled the systematic exploration of gene function in microorganisms. Here, we present a two-stage strategy to construct high-density GI maps in mammalian cells. First, we use ultracomplex pooled shRNA libraries (25 shRNAs/gene) to identify high-confidence hit genes for a given phenotype and effective shRNAs. We then construct double-shRNA libraries from these to systematically measure GIs between hits. A GI map focused on ricin susceptibility broadly recapitulates known pathways and provides many unexpected insights. These include a noncanonical role for COPI, a previously uncharacterized protein complex affecting toxin clearance, a specialized role for the ribosomal protein RPS25, and functionally distinct mammalian TRAPP complexes. The ability to rapidly generate mammalian GI maps provides a potentially transformative tool for defining gene function and designing combination therapies based on synergistic pairs.

Figure 1. Pooled High-Coverage RNAi Screen for Ricin Resistance and Sensitization

(A) Experimental strategy: A population of K562 cells was infected with a pooled high-coverage shRNA library and split into two subpopulations, one of which was treated with ricin. The frequency of shRNA-encoding constructs in each subpopulation was determined by deep sequencing. (B) Based on the frequency in the treated and untreated subpopulations, a quantitative resistance phenotype ρ was calculated for each shRNA. Comparing the distribution of ρs for shRNAs targeting a gene of interest to the ρ distribution for negative control shRNAs using the Mann-Whitney U test yielded a P value for the gene. RAB1B knockdown protects cells from ricin (P = 6.9·10⁻⁸) whereas knockdown of COPA sensitizes cells to ricin (P = 2.4·10⁻⁸). (C, D) Increasing the coverage of the shRNA library improves the detection of hit genes above background. P values for each gene in a test library were calculated on the basis of random subsets of the data; the number of shRNAs included per gene was varied. Random subsampling was repeated 100 times; means of −log₁₀ P values are shown. Gray dotted lines indicated a coverage of 25 shRNAs per gene, which we chose for our genome-wide library. (C) Means of −log₁₀ P values +/− SD for three example genes: a strong hit (RAB1A), a moderate hit (STX16), and a non-hit (CRYAB). (D) Means of −log₁₀ P values for all 1,079 genes targeted by the library (left panel) and for the top 50 hits based on the P value calculated from 45 shRNAs (right panel).

Figure 2. Hits from a Genome-Wide Screen Recapitulate Known Ricin Biology

(A) Overview of ricin intoxication of mammalian cells. Ricin is taken up by endocytosis and traffics retrogradely to the ER, where ricin A and B chains dissociate. The A chain retrotranslocates to the cytoplasm and cleaves ribosomal RNA, thereby inhibiting protein synthesis and ultimately triggering apoptosis. (B) GO-term enrichment analysis for top hits. Top hits were defined as the set of 73 protective genes with an FDR < 0.05 and 83 sensitizing genes with an FDR < 0.02. Non-redundant GO-terms with an FDR < 0.05 are shown; biological process (black bars), cellular component (gray bars). (C) Visualization of top hits in cellular pathways as blue circles (protective hits) and red circles (sensitizing hits); circle area is proportional to −log₁₀ P value. Selected hits below the top hit cutoff were included (pink and light blue circles) if they were part of a known physical complex containing a top hit, or if they were part of the GI map presented in Figure 5. Gray ovals indicate known physical complexes, the asterisk identifies the SRI complex identified in this study.

Figure 3. Characterization of Hit Genes from the Primary Screen

(A) K562 cells were treated with ricin in the presence or absence of atorvastatin for 24 h, and then allowed to recover in the continued presence of atorvastatin. The percentage of viable cells was quantified using flow cytometry. (B) Cells expressing ER-localized SNAP were intoxicated with benzylguanine-labeled ricin and covalent ricin-SNAP complexes were detected by anti-SNAP Western blot. (C) Quantification of ricin modified fraction of ER-SNAP. (D) Raji B cells were infected with shRNAs targeting the indicated genes, and a competitive growth assay was performed in the presence of either ricin or shiga toxin. (E) COPZ1 knockdown increases levels of ER-localized ricin as measured by the SNAP assay

Figure 4. Effects of Combinatorial Gene Knockdowns by Double-shRNAs

(A) Experimental strategy: Active shRNAs targeting hit genes from the primary screen were individually cloned and barcodes are added upstream and downstream of the mir30 context. Pooled ligation yielded a library of all pairwise combinations of shRNAs. Ricin resistance phenotypes of double-shRNAs were determined as for the primary screen; double-shRNA were identified by sequencing the combinatorial barcode. (B) Reproducibility between phenotypes of individual shRNAs in a batch retest (mean of two experiments +/− spread) and the same shRNAs paired with negative control shRNAs in a double-shRNA screen (mean +/− SD for combinations with 12 different negative control shRNAs). (C) Reproducibility between two permutations of double shRNA constructs representing (negative control + targeted) or (targeted + negative control), mean +/− SD for combinations with 12 different negative control shRNAs. (D) Genetic interactions are calculated as deviations from the typical double-mutant phenotype. The relationship between single shRNA phenotypes and double-shRNA phenotypes in combination with an shRNA of interest (in this example SEC23B_i) is typically linear (red line). Deviations from this line are defined as genetic interactions. Buffering interactions (yellow) are closer to WT phenotype than expected, as in this case found for double-shRNAs targeting SEC23B twice. Synergistic interactions (blue) are further away from WT than expected, as in this case found for double-shRNAs targeting both isoforms of SEC23, SEC23A and SEC23B. (E) Phenotypes for individual and combinatorial SEC23A, SEC23B knockdown measured in competitive growth assay (mean of triplicate experiments +/− SD). (F) Quantification of ER localization of ricin measured by the SNAP assay in different knockdown strains (mean of triplicate experiments +/− SD).

Figure 5. A GI Map Reveals Functionally and Physically Interacting Genes

(A, B) Correlations of GI patterns between shRNA pairs: shRNAs targeting the same gene in orange, shRNAs targeting different genes in previously known physical complexes in purple, other pairs of shRNAs in gray. (A) Reproducibility of GI correlations between shRNA pairs in two experimental replicates. (B) High inter-gene and inter-complex correlation of GIs. Distribution of correlation coefficients between shRNA pairs are shown for the three classes of shRNA pairs. The anti-correlated part of the bimodal distribution of intra-complex shRNA pairs is fully accounted for by pairs including shRNAs targeting TRAPPC9, SEC23A, and RPS25. (C) GIs for all gene pairs were calculated (shown as a yellow-cyan heatmap) and genes were clustered hierarchically based on the correlation of their GIs. Individual phenotypes are indicated by sidebars using a red-blue heatmap. Genes marked with asterisks were imported from a separate double-shRNA screen we conducted with a partially overlapping gene set. Known physically or functionally interacting groups of genes are labeled by vertical lines; diamonds mark interactions defined in this study.

Figure 6. Novel Interactions Predicted From the GI Map: RPS25/ILF2/3 and SRIC

(A) Buffering genetic interactions between shRNAs targeting ILF3, the ribosomal subunit RPS25, and ILF2/ILF3. (B) Correlation and buffering genetic interactions between shRNAs targeting ILF2, ILF3 and RPS25 in an shRNA-based genetic interaction map. (C, D) The poorly characterized, genetically correlated proteins SRI1 and SRI2 interact physically, as shown by reciprocal co-immunoprecipitation and MS. (E) GFP-SRI1 partially colocalizes with the autophagosome/lysosome marker mCherry-LC3 in HeLa cells. (F) Total cellular ricin levels after intoxication, as quantified by western blotting, are increased upon knockdown of degradation-related genes and SRI1 (which sensitizes to ricin). The asterisk indicates statistically significant differences (P < 0.05, Student’s t test). (G) SRI1 and COPZ1 knockdown increase levels of ER-localized ricin as measured by the SNAP assay, whereas TRAPPC8 knockdown decreases levels of ER-localized ricin. The asterisks indicate statistically significant differences (**, P < 0.01; ***, P < 0.001; Student’s t test). (H) Model: Ricin partitions between degradation and productive intoxication pathways; inhibition of degradation increases productive intoxication.

Figure 7. Functional Dissection of the TRAPP Complex

(A–B) All TRAPP complex members (other than TRAPPC9/10) specifically co-immunoprecipitate with TRAPPC11 (A) and TRAPPC8 (B), as quantified by mass spectrometry. (C) Correlation of genetic interactions with TRAPPC11 and buffering genetic interaction with TRAPPC11 is shown for each gene included in the genetic interaction map. TRAPP complex members are shown in green, the functionally related SEC22B in green. TRAPPC9 shows a strongly anti-correlated genetic interaction pattern when compared to other TRAPP complex members. (D) Abundance (quantified as LFQ) of each TRAPP subunit in the immunoprecipitation is indicated by color scale. (E) Extracts from K562 cells were fractionated by size exclusion chromatography on a superose 6 column. Western blot could detect co-migration of TRAPPC8 and TRAPPC11, which were larger in size than TRAPPC10. The core component TRAPPC3 migrated with both components. EXT=unfractionated extract. (F) Immunoprecipitation of TRAPPC8 or TRAPPC10 tagged with GFP showed specific association of TRAPPC8 with SEC31A. (G) Association of GFP-TRAPPC8 with SEC31A was assessed by immunoprecipitation in extracts from cells stably expressing shRNAs targeting the indicated TRAPP components. (H) Hypothetical model for mammalian TRAPP complexes. We propose that at least two complexes exist, which contain a core set of proteins (yellow) and unique subunits, either TRAPPC9/10 (mTRAPPII) or TRAPPC8/11/12/13 (mTRAPPIII), which associate with COPI or COPII vesicles, respectively.