Combinatorial CRISPR-Cas9 Metabolic Screens Reveal Critical Redox Control Points Dependent on the KEAP1-NRF2 Regulatory Axis

Abstract

The metabolic pathways fueling tumor growth have been well characterized, but the specific impact of transforming events on network topology and enzyme essentiality remains poorly understood. To this end, we performed combinatorial CRISPR-Cas9 screens on a set of 51 carbohydrate metabolism genes that represent glycolysis and the pentose phosphate pathway (PPP). This high-throughput methodology enabled systems-level interrogation of metabolic gene dispensability, interactions, and compensation across multiple cell types. The metabolic impact of specific combinatorial knockouts was validated using ¹³C and ²H isotope tracing, and these assays together revealed key nodes controlling redox homeostasis along the KEAP-NRF2 signaling axis. Specifically, targeting KEAP1 in combination with oxidative PPP genes mitigated the deleterious effects of these knockouts on growth rates. These results demonstrate how our integrated framework, combining genetic, transcriptomic, and flux measurements, can improve elucidation of metabolic network alterations and guide precision targeting of metabolic vulnerabilities based on tumor genetics.

Keywords: KEAP1; NADPH; NRF2; combinatorial CRISPR screening; genetic interactions; glycolysis; metabolic enzyme compensation; metabolic enzyme essentiality; metabolic flux analysis; redox metabolism.

Figure 1. Experimental design

(A) Schematic pathway diagram of carbohydrate metabolism, and list of 51 targeted enzymes. (B) Schematic overview of the combinatorial CRISPR-Cas9 screening approach. A dual-gRNA library in which each element targets either gene-gene pairs or gene-scramble pairs, to assay dual and single gene perturbations, was constructed from array-based oligonucleotide pools. Competitive growth based screens were performed, and the relative abundance of dual-gRNAs were sampled over multiple time points. The fitness and genetic interactions were computed via a numerical Bayes model and key hits were validated using both competitive cell growth assays and measurement of metabolic fluxes. See also Figure S1 and Table S1.

Figure 2. Combinatorial CRISPR screens reveal metabolic network dependencies

(A) SKO fitness scores for HeLa cells, plotted as f_g (day⁻¹), with a more negative score representing a loss in fitness with SKO. Plotted as mean ± SD. (B) Multi-isoform family member fitness scores and gene expression for HeLa (top) and A549 (bottom) cells. (C) Relative comparison of SKO fitness scores (f_g) across both cells. (D) Relative comparison of genetic interaction scores (π_gg) across both cell lines. (E) Combined genetic interaction map of both cell lines. Green solid line represents interactions observed in both cell lines. Red and blue lines represent significant genetic interactions in A549 and HeLa cells respectively. See also Figure S2 and Table S2–S4.

Figure 3. Screening results validated through targeted fitness and metabolic flux measurements

(A) Schematic of cell competition assay used to validate growth. A Cas9-expressing cell is transduced with a sgRNA lentivirus of interest (tdTomato−) and mixed with a control Cas9-expressing cell transduced with a tdTomato lentivirus (tdTomato+). The cells are grown together and the percentage of control (tdTomato+) cells is used to assess relative fitness of SKO. (B) Non-targeting control (top) is stable for duration of experiment and shows no fitness changes. SKO of ALDOA (bottom) shows decreased fitness over time as control cells take over population. (C) SKO competition assay of ALDO isozyme family. ALDOA shows greatest loss of fitness. (D) Growth validation of PFKM/PGD genetic interaction. DKO (green) shows significantly greater decrease in fitness over additive SKO effect (black). (E) Growth validation of ALDOA/GAPDH interaction. (F) Atom transition map depicting glycolysis. Fully labeled ([U-¹³C₆]glucose) leads to fully labeled pyruvate, lactate, and alanine. (G) Metabolic validation of DKO interaction in ENO1/ENO3. DKO significantly lowered flux through glycolysis over control or SKOs. † indicates statistical significance (p<0.05) for all conditions as compared to DKO (H) Growth validation of ENO1/ENO3 interaction. (I) Atom transition map depicting oxPPP tracing. [3-²H]glucose labels cytosolic NADPH through oxPPP. Labeling on glycolytic intermediates from [1,2-¹³C]glucose is changed by shunting of glucose through oxPPP. (J) Metabolic validation of PGD SKO at day 4. oxPPP contributes less NADPH with PGD knockout. Plotted as mean ± 95% CI. * indicates statistical significance by non-overlapping confidence intervals. (K) Metabolic validation of G6PD SKO at day 7. Less glucose is shunted through oxPPP with G6PD knockout. (L) SKO competition assay of oxPPP enzymes. All experiments were performed in HeLa cells. (C–E, G–H, K–L) Data plotted as mean ± SEM. See also Figure S3.

Figure 4. KEAP1 mutational status alters redox metabolism and impact of oxPPP gene knockouts on cellular fitness

(A) Plot of cell-specific fitness scores for expressed genes. Positive scores are SKOs that are essential in A549s and negative scores are SKOs more essential in HeLa cells. The cell-specific essentiality scores respond to the z-score transformed residuals of linear regression of HeLa and A549 SKO fitness, shown in Figure S4A. (B) Immunoblot of KEAP1 SKO in HeLa cells. (C) Contribution of oxPPP to cytosolic NADPH with KEAP1 SKO in HeLa cells. Plotted as mean ± 95% CI. * indicates statistical significance by non-overlapping confidence intervals. (D) Relative PGD SKO effect in A549s with KEAP1 mutant panel. (E) Contribution of oxPPP to cytosolic NADPH in A549s with KEAP1 mutant panel. Plotted as mean ± 95% CI. * indicates statistical significance by non-overlapping confidence intervals. (F) Immunoblot of A549s with KEAP1 mutant panel. (G) Normalized relative gene expression of A549s with KEAP1 mutant panel. (H) Glutathione measurement in A549 with KEAP1 mutant panel (n=5). (I) Schematic of how KEAP1 mutational status alters relative metabolism and oxPPP dispensability. (D, G, H) Data plotted as mean ± SEM. See also Figure S4.