A global genetic interaction map of a human cell reveals conserved principles of genetic networks

Abstract

We generated a genome-scale, genetic interaction network from the analysis of more than 4 million double mutants in the haploid human cell line, HAP1. The network maps ~90,000 genetic interactions, including thousands of extreme synthetic lethal and genetic suppression interactions. Genetic interaction profiles enabled assembly of a hierarchical model of cell function, including modules corresponding to protein complexes, pathways, biological processes, and cellular compartments. Comparative analyses showed that general principles of genetic networks are conserved from yeast to human cells. A genetic interaction network mapped in a single genetic background complements the DepMap gene co-essentiality network, recapitulating many of the same biological connections and also capturing unique functional information to reveal roles of uncharacterized genes and molecular determinants of specific cancer cell line genetic dependencies.

Fig. 1.. Genome-scale genetic interaction analysis in HAP1 cells.

(A) Diagram of genetic interaction analysis pipeline in co-isogenic cell lines. The quantitative genetic interaction (qGI) score is based on the difference between log fold change measurements for a given library gene in the query mutant (i.e. double mutant) versus WT (i.e. single mutant) cell populations. (B) Scatterplots depicting genetic interactions for the indicated query genes. FANCG, PDCD5, VPS52 screen identifiers correspond to GIN192, GIN189, and GIN 241, respectively. Negative (blue) and positive (yellow) genetic interactions that satisfied a standard genetic interaction threshold (|qGI| > 0.3, FDR < 0.1) are shown. Specific negative and positive interactions identified in each screen are indicated. (C) Heatmap of qGI values for selected reproducible genetic interactions (columns) from biological replicate screens (n=5) for the indicated query genes (rows). Negative qGI scores are shown in blue and positive qGI scores in yellow. Genes labeled in panel B are indicated in bold face. Functions enriched among specific groups of library genes are indicated. (D) Examples of functionally enriched gene modules derived from clustering of the entire genetic interaction dataset, as described (39). Node color represents shared general function and the poorly characterized HEATR6 gene is shown in red.

Fig. 2.. A genetic interaction profile similarity network for a human cell.

(A) HAP1 genetic interaction network comprising 3784 human genes (nodes). Gene pairs were connected by profile similarity (PCC > 0.41) and graphed using a spring-embedded layout algorithm (edges)(161). Genes sharing similar genetic interaction profiles are positioned near each other, whereas genes with less similar genetic interaction profiles are farther apart. (B) HAP1 genetic interaction network annotated using SAFE (66) for bioprocess terms. (C) HAP1 genetic interaction network highlighting network regions that are enriched for proteins in the same cellular compartment. Node opacity reflects genelevel enrichment significance, with more enriched genes displayed more opaquely. Dashed lines represent network regions enriched for bioprocesses indicated in panel B. (D) HAP1 genetic interaction network annotated by highlighting CORUM protein complexes. Nonredundant protein complexes were identified by assigning each gene to the largest complex it belongs to with >2 unique members. The centroid of the network positions of the genes annotated to a given protein complex was used to create the protein complex node. Nodes are colored according to the biological process-enriched region of the network to which they localized. Dashed lines indicate the network regions enriched for bioprocesses indicated in panel B. (E-F) Genes belonging to the bioprocess–enriched network region highlighted in the inset were extracted from the HAP1 network and genes (nodes) in the subnetworks were colored according to their Corum protein complex annotation.

Fig. 3.. Genetic interaction density analysis.

(A) Bar chart showing genetic interaction density (observed interactions/total gene pairs screened) for library genes by category (all genes, nonessential (noness), nonessential with fitness phenotypes (noness fitness), essential) at |qGI| >0.3, FDR < 0.1. Negative (blue), positive (yellow) and total (grey) interaction densities, along with the number of genes in each category, are indicated. (B) Density distribution of negative (blue) and positive (yellow) interactions, highlighting the top 5% of genes with the highest interaction density. (C) Genetic interaction density heatmap visualized as a function of a gene’s single mutant standard deviation in the DepMap dataset (x-axis, CERES score std. deviation) and the single mutant mean phenotype (y-axis, CERES score mean). Darker purple represents increased total genetic interaction density (positive and negative interactions) in the HAP1 GI network. Right bar plots show positive and negative density and CERES score mean for the genes TSR1 and DERL2. The dotted line indicates the boundary between high and low genetic interaction density. (D) The average negative (blue) and positive (yellow) interaction density for library genes as a function of expression in HAP1 cells, with dotted lines indicating background interaction densities across all tested library genes. (E) The distribution of genes belonging to each gene set is plotted as a function of a gene’s single mutant standard deviation (x-axis, CERES score std. deviation) and mean phenotype (y-axis, CERES score mean) in the DepMap dataset. Plots show (i) HAP1 essential genes with lowest 50% GI density, (ii) HAP1 non-essential genes with significant fitness effects, and (iii) HAP1 essential genes with the top 20% total interaction density (right). The contour lines reflect the density of the corresponding gene sets in this two-dimensional space. The dotted line indicates the boundary between high and low genetic interaction density as defined in C. Grey nodes represent library genes with at least 1 genetic interaction (|qGI| > 0.3, FDR < 0.1). The purple nodes indicate HAP1 essential hub genes. (F) Negative genetic interaction density among pairs of duplicated genes with increasing sequence identity (i.e. paralogs).

Fig. 4.. Relating genetic and physical interactions.

(A) Bar charts indicating significant foldenrichment (P < 0.05, hypergeometric test) for gene pairs encoding physically interacting proteins (PPI), proteins within the same protein complex (Co-complex), proteins within the same pathway (Co-pathway) or co-expressed gene pairs among negative (blue) and positive (yellow) genetic interactions. Enrichment was measured for all gene pairs, essential gene pairs, pairs of nonessential genes associated with a fitness phenotype, and nonessential genes lacking a fitness phenotype. Grey bars indicate non-significant enrichment. Genes with roles in mitochondrial-related functions were excluded from this analysis. (B) Network of coherent negative (blue) or positive (yellow) genetic interactions among genes of the RETROMER, GARP, HOPS and COG protein complexes. Node color indicates members of the same protein complex. (C) Bar charts depicting the percentage of CORUM protein complexes whose members were enriched for any type of genetic interaction (grey), negative (blue) or positive (yellow) interactions with each other within (top) or between (bottom) protein complexes. Only single complexes and complex-complex pairs with at least 5 tested gene pairs were included in the analysis. Genes with mitochondrial-related functions were excluded from this analysis. (D) Distribution of complex-complex pairs with respect to between-complex genetic interaction purity scores (51). A score of −1 indicates that genetic interactions occurring between a pair of protein complexes are exclusively comprised of negative interactions whereas a purity score of 1 indicates pairs of complexes connected strictly by positive interactions. The dotted grey line indicates the random expectation based on purity scores generated by sampling negative/positive interaction signs randomly according to a binomial distribution. Genes with mitochondrial-related functions were excluded from the analysis.

Fig. 5.. Functional distribution of genetic interactions.

(A) (i) Schematic of genetic interactions within the functional hierarchy of the HAP1 genetic interaction profile similarity network, showing genetic interactions that occur within the same complex/pathway, biological process, or cellular compartment, and distant interactions between compartments (see Fig. 2). (ii) Line graph showing the observed genetic interaction density for genes within the same hierarchy level for negative (blue) and positive (yellow) genetic interactions (|qGI| > 0.3, FDR < 0.1). Pie charts indicate the total number of gene pairs examined at each level of the functional hierarchy. Horizontal dashed line show background density of negative and positive interactions. Analysis includes ~1600 genes with high-confidence profiles but excludes mitochondrial-related genes (right). (B) Functional distribution of all negative (blue) and all positive (yellow) interactions (|qGI| > 0.3, FDR< 0.1) in the genetic network hierarchy. Genes with mitochondrial-related function are excluded from this analysis. (C) Fraction of negative (blue) and positive (yellow) interactions within specified qGI score ranges connecting genes within different functional levels. Different shades of blue and yellow correspond to levels of functional relatedness shown in B. Analysis includes ~1600 genes with high-confidence profiles, excluding mitochondrial-related genes. (D) Network density of genetic interactions (|qGI| >0.3, FDR < 0.1) within and across biological processes for 14 enriched gene sets, as defined in Fig. 1B. Diagonal nodes represent interactions within bioprocesses, off-diagonal nodes represent interactions between bioprocesses. Node size reflects the fraction of interacting gene pairs. The average density of negative and positive interactions observed within and between bioprocesses is shown in the box plots. (E) (i) Network map showing regions of the HAP1 profile similarity network enriched for genes with negative (blue) or positive (yellow) consensus genetic interactions with a VPS52 query gene (n=5 biological replicates). (ii) Genes encoding members of vesicle tethering complexes showing coherent genetic interactions with VPS52. (iii) Most genes with roles in the cholesterol biosynthesis pathway show positive interactions with the VPS52 query gene.

Fig. 6.. Genetic suppression interactions.

(A) Specific examples of genetic suppression. Arrows indicate direction of suppression. Grey nodes indicate genes whose mutant fitness phenotype is suppressed and colored nodes represent suppressor genes. Dotted arrow indicates weak suppression interactions that did not satisfy a suppression score threshold (score >0.5). VPS52 suppressors involved in cholesterol biosynthesis (blue) and sphingolipid biosynthesis (pink) are highlighted (B) Box plot showing the mean percentage of DepMap cell lines that depend on the indicated groups of HAP1 essential genes for viability. Numbers of essential genes tested in each group are indicated. Genes with mitochondrial-related functions were excluded from this analysis.

Fig. 7.. Genetic interaction conservation.

(A) Scatter plot comparing human (qGI < −0.3, FDR <0.1) and yeast (SGA score < −0.08, P <0.05) negative interaction densities within bioprocesses (dark blue) and between pairs of bioprocesses (light blue). (B) Pie chart shows the fraction of conserved gene pairs tested in HAP1, while the donut plot summarizes negative and positive interactions in HAP1 among conserved gene pairs. Dark blue and dark yellow represent conserved negative and positive interactions, while light blue and light yellow indicate interactions found only in HAP1. (C) Bar graph illustrating enrichment for negative (blue) and positive (yellow) interactions in yeast among conserved gene pairs that showed a negative or positive genetic interaction in HAP1 cells (left). Bar graph illustrating enrichment for negative (blue) and positive (yellow) interactions in HAP1 cells among conserved gene pairs that showed a negative or positive genetic interaction in yeast (right). * indicates level of statistical significance (*** P < 10⁻⁶, Fisher’s exact test). (D) Consensus genetic interactions for PTAR1. Mean negative (blue) and positive (yellow) qGI scores (|qGI| > 0.3 and FDR < 0.1) based on genetic interactions from 4 independent PTAR1 genome-wide screens are shown. Conserved negative (dark blue) and positive (orange) genetic interactions identified in HAP1 and yeast screens using human PTAR1 and yeast ECM9 orthologous query genes are shown and specific examples of conserved interactions are indicated. (E) Bar graphs illustrating enrichment for PTAR1 negative (blue) and positive (yellow) genetic interactions in HAP1 cells among conserved gene pairs that showed a negative or positive genetic interaction with yeast ECM9, and vice versa. * indicates level of statistical significance (*** P < 10⁻⁵, Fisher’s exact test). (F) Yeast-two hybrid analysis illustrating the physical interaction between α and β subunits of the indicated prenyltransferases. (G) Tetrad analysis showing that co-expression of the human PTAR1-RABGGTB GGTaseIII complements essentiality of yeast ECM9. Yeast ECM9/ecm9Δ heterozygous deletion strains carrying a vector control or a plasmid expressing human PTAR1 and RABGGTB expressed from a bidirectional galactose-inducible promoter were sporulated. The meiotic progeny derived from four tetrads were dissected and tested for spore germination (denoted a-d) on either glucose (Glu.) medium, where the promoter is repressed or galactose medium (Gal.) where the promoter is induced. Black circles indicate spore progeny that are predcited to carry the ecm9Δ deletion. Blue circles indicate ecm9Δ deletion mutants where the ECM9 essential phenotype is rescued by galactose-inducible expression of the human PTAR1-RABGGTB GGTaseIII (bold). (H) Schematic model for dual lipid modification-dependent activation of Ykt6. (I) Immunoblot for Ykt6 palmitoylation assessed by mPEG replacement chemistry using protein extracts from the three indicated yeast strains (51). (J) Chemical-genetic interaction profile mapped for the depalmitoylase inhibitor ABD957. Negative (blue) and positive (yellow) chemical-genetic interactions that satisfied a standard confidence threshold are shown, with select genes highlighted. The chemical structure of ABD957 is shown. Bar graph shows PTAR1 mutant fitness in ABD957 and DMSO conditions.

Fig. 8.. Relationship between DepMap cancer cell line expression dependency and HAP1 genetic interactions.

(A) Scatter plots illustrating the relationship between PTAR1 single mutant fitness and expression of either (i) ABHD16A or (ii) YKT6 across DepMap cancer cell lines. (iii) The PTAR1-ABHD18 gene pair shows a negative Expression-Dependency (ED) score and a positive genetic interaction score (qGI). The PTAR1-YKT6 gene pair shows a positive ED score and a negative qGI score. (B) Schematic illustrating the distribution of ED scores for all gene pairs tested in this study and the overlap between ED and qGI scores. (C) Selected examples of specific gene pairs that exhibited significant ED-qGI combinations.

Fig. 9.. An integrated functional network based on genetic interaction and co-essentiality profiles.

(A) Comparison of the overlap between correlated gene pairs in the complete HAP1 GI profile similarity and DepMap co-essentiality sampled networks. Co-essentiality networks were constructed by selecting non-overlapping random samples of 298 screens from the DepMap dataset (20Q2). This was repeated to generate 10 co-essentiality networks. Network overlap was assessed by computing Jaccard indices at increasing network similarity thresholds (Pearson’s correlation coefficient thresholds). The same procedure was used to measure similarity of each DepMap co-essentiality network to the HAP1 genetic interaction profile similarity network. Continuous lines represent the mean Jaccard index of the DepMap-DepMap network comparisons (blue) and the DepMap-GI network comparisons (purple). The dotted lines represent the quartiles of Jaccard indices. (B) Scatter plot of Z-scores for modules or gene clusters identified from the genetic interaction profile similarity network. Modules with significant similarity in the DepMap co-essentiality network (blue) and modules without significant similarity (purple) are plotted. The grey dashed line indicates Z-score threshold DepMap co-essentiality network similarity. (C) Examples of modules derived from the genetic interaction profile similarity network. (D) Box plots of mean fitness in HAP1 cells and fitness in DepMap cancer cell lines for genes in significant modules that share similar co-essentiality profiles (blue bars) or do not have strongly correlated co-essentiality profiles (purple bars). (E) Bar plot illustrating the fraction of genetic interaction profile similarity network modules (z-score > 2, File S22) enriched for the same GO-BP terms (hypergeometric test, Benjamini-Hochberg-corrected FDR < 0.2) or PPIs (hypergeometric test, Benjamini-Hochberg-corrected FDR < 0.05), including modules uniquely identified in the genetic interaction network profiles (GI module z-score > 2 and DepMap module z-score < 2, purple bars), or genetic network-derived modules that share highly similar DepMap co-essentiality profiles (GI module z-score > 2 and DepMap module zscore > 2, blue bars). (F) Precision-recall plots for genes exhibiting similar DepMap co-essentiality profiles (blue), genetic interaction profiles (light purple) or profiles from the integrated network. True positives (TP) involve gene pairs co-annotated to a gold standard set of GO-BP terms (left panel) or gene pairs encoding members of the same CORUM protein complex (right panel). Grey dashed line represents background co-annotation rates. Genes annotated with a mitochondrial-related function were excluded because profile similarity profiles tend to be dominated by mitochondrial genes (19). The same precision-recall analysis based on all genes, including mitochondrial genes, is shown in fig. S20F. (G) Comparison of individual GO-BPs or CORUM protein complexes captured by the DepMAP co-essentiality network and the integrated network. Nodes above (purple) or below (blue) the diagonal indicate better performance by the integrated or DepMap networks, respectively, based on AUPRC (Area Under a Precision Recall Curve) values, generated per GO-BP process or CORUM complex.