Mapping the Genetic Landscape of Human Cells

Abstract

Seminal yeast studies have established the value of comprehensively mapping genetic interactions (GIs) for inferring gene function. Efforts in human cells using focused gene sets underscore the utility of this approach, but the feasibility of generating large-scale, diverse human GI maps remains unresolved. We developed a CRISPR interference platform for large-scale quantitative mapping of human GIs. We systematically perturbed 222,784 gene pairs in two cancer cell lines. The resultant maps cluster functionally related genes, assigning function to poorly characterized genes, including TMEM261, a new electron transport chain component. Individual GIs pinpoint unexpected relationships between pathways, exemplified by a specific cholesterol biosynthesis intermediate whose accumulation induces deoxynucleotide depletion, causing replicative DNA damage and a synthetic-lethal interaction with the ATR/9-1-1 DNA repair pathway. Our map provides a broad resource, establishes GI maps as a high-resolution tool for dissecting gene function, and serves as a blueprint for mapping the genetic landscape of human cells.

Keywords: CRISPR; CRISPRi; epistasis; functional genomics; genetic interactions.

Figure 1.. A large-scale quantitative GI mapping platform in human cells.

(A) Schematic of the overall GI mapping approach. (B) Histogram of gene growth phenotypes (γ) from a CRISPRi v1 growth screen (Gilbert et al., 2014). A subset of these genes were selected for inclusion in the GI map based on exhibiting a moderate growth phenotype and a high-confidence p-value. (C) Cellular processes represented in GI map, with number of genes in parentheses (see also Table S2). (E) Approach for quantifying epistasis between sgRNAs, based on the relationship between single sgRNA phenotypes and the corresponding pair phenotypes with a given “query” sgRNA. (F) Example of sgRNA epistasis with query sgRNA sgANAPC13-1. Negative control sgRNAs are circled in red, and red line corresponds to quadratic fit of all sgRNA single and pair phenotypes (see also Figure S3A).

Figure 2.. A large-scale CRISPRi-based GI map.

(A-B) sgRNA GI scores (A) and GI correlations (B) from two independent replicates performed in K562. Contours correspond to 99th, 95th, 90th, 75th, 50th, and 25th percentiles of data density. Pearson correlation (R) is of all sgRNA pair correlations. Due to the size of the dataset, Pearson P-values here and throughout the manuscript are < 10⁻³⁰⁰ unless otherwise stated. (C) Histogram of sgRNA GI correlations calculated from replicate-averaged sgRNA pair phenotypes. Smoothed histograms of all pairs of sgRNAs or only sgRNA pairs targeting the same gene or complex were generated with Gaussian kernel density estimation. (D-E) Gene-level GI scores (D) and GI correlations (E), displayed as in A-B. (F) Histogram of gene GI correlations from replicate-averaged screens, displayed as in C. (G) Full gene-level GI map in K562. Dendrogram indicates average linkage hierarchical clustering based on uncentered Pearson correlations between genes. Clusters were annotated by assigning GO annotations if the GO term was significantly enriched in that cluster (hypergeometric P ≤ 10⁻⁹) and not more enriched in another cluster.

Figure 3.. GI correlations identify members of protein complexes and functionally related pathways.

(A) Histogram of all correlations between K562 gene GI profiles (green) or between non-targeting (NT) control and gene GI profiles (black). (B) Cumulative distribution of GI correlations for all genes (as in A), for gene pairs within mitochondria or early trafficking, or for pairs with one gene in each compartment. (C) Fraction of gene pairs with a given GI correlation annotated by the STRING experimentally validated interaction set. GI correlations were binned to the next-lowest tenth. (D) Histogram of the correlations between GI score profiles in K562 and Jurkat maps. Only genes present in both K562 and Jurkat are included. (E) Comparison of GI correlations within each GI map. (F) Gene networks of the most highly correlated genes in K562. Edges represent correlations greater than 0.6. GI correlations that correspond both to STRING-annotated interactions and to MitoCarta gene pairs were labeled according to their STRING interaction confidence. Edge lengths were determined by force-directed layout. Asterisks indicate gene pairs that have closely neighboring TSSs.

Figure 4.. Oxidative metabolism is highly correlated with poorly characterized gene TMEM261 and anti-correlated with glycolytic metabolism.

(A) Selected GIs with mitochondrial complex I and glycolytic genes from the K562 GI map. (B) ATP levels relative to baseline ATP following one-hour incubation in either respiratory or glycolytic conditions. Data show mean ± SEM, and N=16 experimental replicates per group from two independent experiments. *** indicates P<0.001 versus NT sgRNA in each condition by one-way ANOVA with Dunnett’s multiple comparisons test. (C) GI scores for genes paired with ATP5A1 and PGK1. (D) GI correlation with ATP5A1 for genes involved in carbon metabolism.

Figure 5.. Structure of genetic interactions in the GI map.

(A) Histogram of all GI scores between unique gene pairs in K562. Same-gene pairs were not included. (B) Relationship between GI correlation and GI score. GI correlations were binned to the next-lowest tenth. (Left) Boxplot of scores within each bin. (Middle) Percent strong buffering interactions within each bin. (Right) Percent strong synergistic interactions within each bin. (C) Enrichment of correlations and strong interactions for gene pairs between the indicated cellular compartments. Values indicate the percent of all gene pairs between the compartments that are correlated or have a GI score above the stated thresholds. (D) Average GI score between GO-annotated clusters in the K562 GI map. Clusters correspond to those displayed in Figure 2G. Numbers in parentheses indicate number of member genes.

Figure 6.. Repression of FDPS is synthetic lethal with HUS1 and results in accumulation of the cholesterol intermediate IPP.

(A) Selected interactions with HUS1 and with FDPS in the K562 GI map. (B) Individual validation experiments sgRNAs targeting HUS1 and FDPS, performed as in Figure S4A. Lines represent mean of two experimental replicates (open circles). (C) Schematic of the cholesterol biosynthesis pathway. Gene names are colored by mean validation GI (see also Figure 6D) of all sgRNA pairs targeting HUS1 and the indicated gene. (D) sgRNA pair epistasis for sgRNAs targeting HUS1 and cholesterol biosynthesis genes. Epistasis was calculated as the measured double-sgRNA phenotype subtracted by the sum of the individual phenotypes and by epistasis with non-targeting (NT) sgRNA. Bars represent mean of duplicate experiments and error bars represent the maximum and minimum data points. (E) Epistasis between sgRNAs targeting HUS1 and FDPS in the presence of DMSO control or 4 μM lovastatin. (F) IPP concentration in cells containing NT or FDPS-targeting sgRNAs grown in the presence or absence of 4 μM lovastatin for 48 hours. N=6 replicates each (4 for lovastatin-treated samples).

Figure 7.. Chemical and genetic perturbation of FDPS causes replicative DNA damage via deoxynucleotide depletion.

(A) Western blot measuring CHEK1 and CHEK1 p-S345 abundance in K562 cells expressing sgRNAs targeting HUS1 or FPDS. (B) Western blots measuring CHEK1 and CHEK1 p-S345 abundance in K562, HEK293T, and iPSC cells treated with the indicated concentrations of alendronate. (C) Western blots measuring CHEK1 and CHEK1 p-S345 abundance in K562 treated with 4 μM lovastatin, 200 μM alendronate, or both drugs. (D) Sensitivity or resistance to alendronate induced by CRISPRi repression of genes involved major DNA repair pathways, excerpted from an unbiased alendronate screen in K562 cells (see methods and Table S7). (E) Cell cycle analysis of K562 cells before and after treatment with 250 μM alendronate. Cells undergoing DNA synthesis incorporate EdU and propidium iodide labels overall DNA content. (F) Sensitivity or resistance to alendronate induced by CRISPRi repression of genes that modify RNR activity, as in Figure 7D. P-values were calculated by Mann-Whitney test of all 10 sgRNAs targeting a given gene compared to negative controls; * indicates P<0.05, *** indicates P<0.001. (G) dATP and ATP concentration in K562 cells expressing NT or FDPS-targeting sgRNAs grown in the presence or absence of 4 μM lovastatin, measured by LC-MS/MS as in Figure 6F. N=6 replicates each (4 for lovastatin-treated samples). *** indicates P<0.001. (H) Schematic of proposed mechanism of FDPS/RRM1/HUS1 synthetic interactions. Red lines indicate the observed consequences of chemical and/or genetic perturbations.