Systematic Identification of Regulators of Oxidative Stress Reveals Non-canonical Roles for Peroxisomal Import and the Pentose Phosphate Pathway

SUMMARY

Reactive oxygen species (ROS) play critical roles in metabolism and disease, yet a comprehensive analysis of the cellular response to oxidative stress is lacking. To systematically identify regulators of oxidative stress, we conducted genome-wide Cas9/CRISPR and shRNA screens. This revealed a detailed picture of diverse pathways that control oxidative stress response, ranging from the TCA cycle and DNA repair machineries to iron transport, trafficking, and metabolism. Paradoxically, disrupting the pentose phosphate pathway (PPP) at the level of phosphogluconate dehydrogenase (PGD) protects cells against ROS. This dramatically alters metabolites in the PPP, consistent with rewiring of upper glycolysis to promote antioxidant production. In addition, disruption of peroxisomal import unexpectedly increases resistance to oxidative stress by altering the localization of catalase. Together, these studies provide insights into the roles of peroxisomal matrix import and the PPP in redox biology and represent a rich resource for understanding the cellular response to oxidative stress.

Graphical Abstract

In Brief

Despite its importance in metabolism and disease, a comprehensive analysis of the cellular response to oxidative stress is lacking. Here, Dubreuil et al. use genome-wide screens to identify cellular regulators of oxidative stress. They investigate paradoxical mechanisms by which disruption of the pentose phosphate and peroxisomal import pathways protect cells.

INTRODUCTION

Oxidative stress has diverse deleterious effects and can lead to tumorigenesis, cell death, neurological disease, and aging (Busciglio and Yankner, 1995; Conger and Fairchild, 1952; Cunningham et al., 1987; Finkel and Holbrook, 2000; Guo et al., 2011; Ishii et al., 2005; Liochev, 2013; Nagai et al., 2009; Sakurai et al., 2008; Totter, 1980; Wu et al., 2003). Conversely, reactive oxygen species (ROS) also have normal physiological roles and can promote autophagy (Chen et al., 2009; Scherz-Shouval et al., 2007) as well as signal proliferation and survival by activating various MAPK proteins (Ichijo et al., 1997; Matsuzawa et al., 2005; Meng et al., 2002; Ray et al., 2012). Diverse antioxidant systems help the cell maintain a redox environment permissive to normal metabolism and ROS signaling while preventing toxic ROS accumulation (Go and Jones, 2008). These systems include antioxidants such as vitamin C, reducing molecules such as NADPH and glutathione and antioxidant enzymes such as superoxide dismutase (SOD) and catalase. However, under conditions of metabolic or environmental stress, these mechanisms can be insufficient, and ROS levels can increase and cause DNA damage, protein dysfunction, and lipid oxidation (Kong and Chandel, 2018; Nathan and Cunningham-Bussel, 2013; Schieber and Chandel, 2014). Though a number of studies have begun to uncover the genetic effectors of ROS toxicity using model organisms and targeted screens in mammalian cells (Ayer et al., 2012; Kimura et al., 2008; Reczek et al., 2017; Ueno et al., 2012), much remains to be discovered, and a comprehensive screen in mammalian cells has not been performed.

Hydrogen peroxide (H₂O₂) is a ubiquitous ROS in biological systems. Endogenously, H₂O₂ is produced as a by-product of oxidative metabolism in peroxisomes and mitochondria and is converted from superoxide anion by SOD. Less reactive and longer lived than superoxide anion, H₂O₂ often acts as a membrane-permeable signaling molecule, promoting autophagy, growth, and survival in various contexts, including cancer (Moloney and Cotter, 2018). However, at higher concentrations, H₂O₂ can induce apoptosis and senescence as well as oxidative damage to proteins, lipids, and DNA (Kuehne et al., 2015; Nathan and Cunningham-Bussel, 2013; Nagai et al., 2009; de Oliveira et al., 2014; Pillai et al., 2005; Schuster and Feldstein, 2017; Sekine et al., 2012; Varani and Ward, 1994). H₂O₂ concentrations vary greatly in the human body. Though there is some disagreement regarding the level of H₂O₂ in blood and plasma, H₂O₂ levels have been found in the low micromolar range (Forman et al., 2016; Go and Jones, 2008; Roberts et al., 2005). H₂O₂ concentrations of 5–15 μM have been measured at sites of inflammation, which can induce oxidative stress in proximal cells (Buchmeier et al., 1995; Forman and Torres, 2002; Liu and Zweier, 2001; Test and Weiss, 1984; Varani and Ward, 1994; Weiss, 1980). Furthermore, UV radiation induces production of superoxide anion and H₂O₂ in melanocytes, creating localized H₂O₂ concentrations up to 1 mM in individuals with pigment deficiencies (Denat et al., 2014; Maresca et al., 1997; Schallreuter et al., 1999, 2012; Song et al., 2009). In addition, H₂O₂ levels have been shown to exceed 100 μM in human urine and are thought to fluctuate along the digestive tract (Go and Jones, 2008; Long and Halliwell, 2000; Long et al., 1999; Varma and Devamanoharan, 1990). Tumor cells are also known to produce high levels of ROS, although they typically upregulate antioxidant activity to counter increased ROS levels (Cairns et al., 2011; Szatrowski and Nathan, 1991). H₂O₂ thus represents an archetypical ROS that requires delicate control to maintain essential redox signaling without incurring cellular oxidative damage.

H₂O₂ toxicity is mediated by free (labile) iron or other transition metals, which decompose H₂O₂ into the highly reactive and damaging hydroxyl radical via the Fenton reaction (Halliwell and Gutteridge, 1990; Halliwell et al., 2000; Ueda et al., 1996). Iron is transported into cells via clathrin-mediated endocytosis of the transferrin receptor (TFRC), which binds to iron-bound transferrin. Labile iron is released in the early endosome (85%–95% of iron uptake) or when TFRC is degraded in the lysosome (5%–15% of iron uptake) (Tortorella and Karagiannis, 2014). The majority of labile iron is found in the lysosome, which is also responsible for the breakdown of the iron storage protein ferritin. Likely because of this high local free iron concentration, lysosomal membranes are sensitive to oxidative damage-induced permeabilization, which can initiate apoptosis (Turk and Turk, 2009). Thus, the transport and metabolism of iron are a key point of regulation for oxidative stress.

One of the first responses for a cell undergoing oxidative stress is to upregulate the pentose phosphate pathway (PPP) to increase production of the reducing agent NADPH (Kuehne et al., 2015). The PPP branches off of glycolysis, metabolizing glucose-6-phosphate to reduce NADP⁺ into NADPH and create purine nucleotide precursors (Patra and Hay, 2014). NADPH is a reducing agent that maintains the antioxidant capacities of the cell, including the glutathione antioxidant system and catalase (Kirkman et al., 1999; Lu, 2009, 2013). In some cells, the PPP acts as a first line of defense against ROS by increasing glucose flux through the oxidative branch of the PPP (Kuehne et al., 2015). Specifically, the metabolite 6-phosphogluconate (6PG), the substrate of phosphogluconate dehydrogenase (PGD), inhibits carbon flux through glycolysis, increasing carbon flux through the PPP (Parr, 1956). This inhibition of glycolysis quickly redirects the glucose flux of the cell into the oxidative PPP, producing NADPH, until the oxidative threat subsides. Thus, the PPP represents a first line of defense for the cell when undergoing oxidative stress.

Integral to the regulation of cellular redox balance, peroxisomes are single-membrane organelles that both detoxify and produce intracellular H₂O₂. Peroxisomes are centers for plasmalogen biosynthesis and the catabolism of diverse substrates, including fatty acids, select amino acids, xenobiotics, and glyoxylate (Schrader and Fahimi, 2006). These reactions, in particular the β-oxidation of fatty acids, produce H₂O₂. Peroxisomes also act as centers of antioxidant activity to detoxify these ROS. The majority of peroxisomal matrix proteins are imported into peroxisomes through the peroxisome targeting signal type-1 (PTS1) localization domain, which is recognized by PEX5, a cytosolic receptor that coordinates functions with eight other PEX proteins to import cargo and cycle back to the cytoplasm (Bottger et al., 2000; Fujiki et al., 2014; Harper et al., 2003; Miyazawa et al., 1989; Stanley et al., 2006). Catalase is key substrate of this pathway localized inside of peroxisomes, where it reduces H₂O₂ into water at high efficiency, potentially converting millions of H₂O₂ molecules per second under the proper conditions, thereby preventing ROS from escaping the peroxisome (Otera and Fujiki, 2012; Young and Woodside, 2001). Thus, while peroxisomes are a significant source of intracellular ROS, they also have substantial antioxidant roles in cells.

Despite the critical importance of oxidative stress, a comprehensive study of the genetic factors regulating the response in mammalian cells is lacking. Here, we use CRISPR/Cas9 knockout (KO) and short hairpin RNA (shRNA) knockdown screens (Barbie et al., 2009; Bassik et al., 2013; Koike-Yusa et al., 2014; Shalem et al., 2015; Silva et al., 2008; Wang et al., 2014; Zhou et al., 2014) to systematically identify the genetic factors that modulate cellular sensitivity to exogenous H₂O₂. We then validate these hits using targeted CRISPR-KO and CRISPRi libraries in HeLa and K562 cells, demonstrating conserved function across cell lines for a broad set of regulators of oxidative stress sensitivity. These screens identified several pathways with paradoxical ROS-sensitivity phenotypes that were inconsistent with their expected roles. In particular, we demonstrate that perturbation of the peroxisomal matrix protein import pathway protects cells from oxidative stress by causing the accumulation of catalase in the cytoplasm. We also investigate the function of PGD, a key member of the PPP; surprisingly, KO of this gene is highly protective against ROS and results in a dramatic increase of upstream PPP metabolites. Together, these results provide unexpected insights into the roles of peroxisomal matrix import pathway and the PPP in redox biology and a systematic and rich resource for studying the cellular response to oxidative stress.

RESULTS

Parallel Genome-wide CRISPR and shRNA Screens Identify Genetic Effectors of Cellular Sensitivity to Oxidative Stress

To systematically identify the genetic factors that modulate cellular sensitivity to oxidative stress, we performed genome-wide knockdown shRNA and KO single guide RNA (sgRNA)/Cas9 screens in K562 chronic myelogenous leukemia cells in the presence of H₂O₂. We infected K562 cells with genome-wide shRNA (Deans et al., 2016; Kampmann et al., 2015) or CRISPR/Cas9 (Morgens et al., 2017) libraries, then passaged the cells untreated or treated them with a 50% lethal concentration (LC₅₀) dose of H₂O₂ repeatedly (allowing for recovery) (Figure 1A). In this manner, protective genes were enriched in the treated conditions relative to negative control cells, while sensitizing genes were depleted compared with negative control cells. The frequencies of the shRNA and sgRNA elements in each screening population were measured using high-throughput sequencing (Figure 1A) and analyzed using our previously published Cas9 high-throughput maximum likelihood estimator (casTLE) (Morgens et al., 2016). Using a 5% false discovery rate (FDR), we identified 511 significant gene hits with the CRISPR/Cas9 screen and 1,250 significant gene hits with the shRNA library (Figures 1C and 1D; Figures S1A and S1B; Table S1).

Figure 1. Parallel Genome-wide CRISPR and shRNA Screens Identify Genetic Effectors of Cellular Sensitivity to Oxidative Stress.

(A) Screening schematic for sgRNA and shRNA screens.

(B) GO term analysis for hits passing 5% FDR from genome-wide Cas9 knockout and genome-wide shRNA knockdown screens.

(D) Volcano plot showing casTLE analysis of independent K562 genome-wide Cas9 knockout screens (n = 2). Protective genes at 5% FDR cutoff are colored in blue, sensitizing genes at 5% FDR cutoff in yellow, and genes chosen from key pathways in red.

(E) Cell diagram showing the 50 most significant hits from shRNA and sgRNA genome-wide screens. Large circles represent hits in either screen’s top 50 most significant hits, and smaller circles represent relevant genes outside of the top 50 in either screen. Protective hits are shown in blue and sensitizing hits in yellow. Shades of colors are based on casTLE scores.

We and others have shown that KO and knockdown screen technologies often reveal distinct gene signatures because of differences in strength of on-target activity, off-target effects, efficacies of gene-targeting elements between libraries, compensatory mechanisms, and the ability to target essential genes (Barrangou et al., 2015; El-Brolosy et al., 2019; Frock et al., 2015; Grimm et al., 2006; Jackson and Linsley, 2010; Kaelin, 2012; Ma et al., 2019; Morgens et al., 2017; Pruett-Miller et al., 2009; Tsai et al., 2015; Weiss, 1980). By using both approaches, the genetic modifiers of a cellular phenomenon can be more comprehensively identified than if either approach was used alone (Morgens et al., 2016; Rosenbluh et al., 2017). Here we found that shRNA or CRISPR screens identified genes enriched in a range of pathways as measured by Gene Ontology (GO) term analysis, some of which were unique to a particular perturbation technology and some of which were shared by both (Figure 1B). Interestingly, the CRISPR/Cas9 screen had stronger signatures for cell cycle regulators, histone modifiers, members of the Wnt signaling pathway, and the peroxisomal matrix import pathway. In contrast, the shRNA screen identified regulators of iron-sulfur cluster assembly and mitochondrial respiratory chain complex I assembly genes.

These screens identified both known and unexpected genetic modifiers of cellular sensitivity to oxidative stress (Figure 1E). We found that perturbation of many of the known antioxidant systems sensitized cells to H₂O₂, including catalase (CAT), peroxiredoxin 1 (PRDX1), and the glutamate-cysteine ligase complex (GCLC and GCLM), which provides glutathione for the glutathione peroxidases. Disruption of genes involved in the cellular response to DNA damage, including ATM, TP53PB1, and MMS19, among others, was sensitizing to H₂O₂ (Figure 1B), consistent with previous observations that H₂O₂ can cause DNA damage and activate the ATM pathway (Driessens et al., 2009; Guo et al., 2010; Moiseeva et al., 2016). Furthermore, we also found that deletion of KEAP1 was protective, consistent with its role in negatively regulating NRF2, an oncogenic transcription factor that drives expression of antioxidant and prosurvival genes (Blake et al., 2010; DeNicola et al., 2011; Kobayashi et al., 2004).

Interestingly, both screening methods detected strong effects for genes involved in all aspects of cellular iron homeostasis (Figure 1E; Figure S1C). Iron can exacerbate the damaging effects of ROS by acting as a catalyst for the Fenton reaction, and thus, cells with less iron are more resistant to oxidative stress (Mantzaris et al., 2016; Mello Filho et al., 1984; Meneghini, 1997; Nishida, 2012). Indeed, one of the strongest protective hits was IREB2, a central iron homeostasis gene that regulates the translation of mRNAs that promote iron uptake and downregulates the translation of mRNAs regulating iron export and storage. Disrupting the expression of IREB2 would recapitulate high-iron conditions and decrease iron levels in the cell. In contrast, FBXL5 and ZFP36L2, which under high-iron conditions degrade IREB2 and TFRC mRNA, respectively, were both sensitizing hits in the screens (Patial et al., 2016; Vashisht et al., 2009). We also detected TFRC itself as a strongly protective hit, as well as the genes involved in clathrin-mediated endocytosis, the mechanism by which TFRC brings iron into the cell. These genes include members of the AP2 clathrin adapter complex, CLTC, CLTH, DNM2, PICALM, and RAB5C (Figure S1C; Gulbranson et al., 2019; Loerke et al., 2009; Tortorella and Karagiannis, 2014). Once TFRC is trafficked to the early endosome and releases transferrin and Fe³⁺, STEAP3 reduces Fe³⁺ to Fe²⁺, which is then transported out of the early endosomes by SLC11A2 (DMT2) as labile iron to iron chaperone PCBP2 (Bogdan et al., 2016). Disruption of both STEAP3 and SLC11A2 were protective against H₂O₂ toxicity.

Consistent with the toxic potential of intracellular labile iron, loss of the labile iron chaperones PCBP1 and PCBP2 had sensitizing effects. Both chaperones mediate direct metallation of non-heme-binding proteins and storage of labile iron in its non-toxic form within ferritin complexes (Bogdan et al., 2016; Leidgens et al., 2013; Shi et al., 2008; Yanatori et al., 2016). Finally, perturbation of genes involved in the assembly and formation of Fe/S clusters in the mitochondria (FXN, NFS1, ISCU, LYRM4, and ABCB7) were protective. Once Fe/S clusters are formed in the mitochondria, they are chaperoned by either FAM96A (CIA2A) or FAM96B (CIA2B) to distinct sets of proteins with differing effects on H₂O₂ toxicity. FAM96A, along with CISD1 and NARFL, had protective effects when perturbed and has been shown to directly bind to IREB2, potentially stabilizing its expression (Alvarez et al., 2017; Stehling et al., 2013). On the other hand, FAM96B and MMS19 negatively regulate IREB2 activity and chaperone Fe/S clusters to DNA damage repair proteins and enhanced H₂O₂ toxicity when perturbed in the genome-wide screens. Together, these results reveal a comprehensive picture of the diverse mechanisms of iron transport and metabolism that control cellular sensitivity to oxidative stress.

We also detected a number of seemingly paradoxical signatures. For example, although mitochondrial metabolism has been shown to be involved in the generation of ROS, it was somewhat surprising that our screens identify both sensitizing (tricarboxylic acid [TCA] cycle) and protective hits (mitochondrial complex I). Particularly striking among these pathways were protective effects observed upon deletion of genes involved in the peroxisomal import pathway and the PPP. Despite the strong antioxidant activity present in peroxisomes and their central role in redox regulation, our screen revealed that deletion of every member of the peroxisomal import pathway was highly protective against oxidative stress. Additionally, although the PPP is a significant source of reducing power, we found that perturbing the PPP at the level of PGD resulted in strong protection against oxidative stress. Given their prominent and puzzling effects, we further investigate the mechanisms of oxidative stress protection from perturbation of the peroxisomal import pathway and PGD below.

Hits from Genome-wide Screens Validate in Competitive Growth Assays and in Multiple Cell Lines

To further investigate the genome-wide screen hits in high throughput, we first performed batch retest screens. We and others have shown that smaller “batch” screens performed at higher coverage (cells/sgRNA) can reduce false positives and false negatives (Bassik et al., 2013; Han et al., 2017; Haney et al., 2018; Parnas et al., 2015). In this case, we sought to conduct batch retest screens not only with higher coverage but in additional cell types and using orthogonal perturbations in order to highlight robust hits for further investigation.

To do this, we designed a batch retest library designed to target the 511 hits passing 5% FDR from our genome-wide CRISPR screen, 282 manually identified genes from the literature that have been previously shown to be involved in redox biology or are related to pathways identified in the screens, and additional high-confidence genes from the genome-wide CRISPR screen, for a total of 1,000 genes (Figure S2A; Table S2). This set of 1,000 genes was used to generate both CRISPR KO libraries (Table S2) as well as knockdown libraries using CRISPRi (dCas9-KRAB; Table S2) both with ten sgRNAs per gene (Horlbeck et al., 2016; Morgens et al., 2017). CRISPR/Cas9 KO and CRISPRi/dCas9-KRAB knockdown K562 and HeLa cell lines were then lentivirally infected with the two batch retest libraries. After selection, the four libraries were pulsed four times with LC₅₀ doses of H₂O₂, 30 μM and 300 μM for K562 and HeLa, respectively. After selection, libraries were prepared, sequenced, and analyzed as described above for the genome-wide screens.

The K562 batch KO screen correlated well with the genome-wide screen (r² = 0.65), providing further validation of our initial genome-wide screen results (Figure 2B). Comparing results across cell lines, H₂O₂ doses, and CRISPR technologies, we found that genes representing major pathways showed consistent effects in both K562 and HeLa cells. Genes involved in iron homeostasis, vesicle trafficking, and the peroxisomal matrix import pathway all had protective effects across screens, while genes involved in the DNA damage response, the Krebs cycle, and antioxidants had sensitizing phenotypes across cell lines (Figure 2A; Figures S2B–S2E; Table S3). As expected for two different cell types, certain genetic pathways showed differences between HeLa and K562 cells. For example, although perturbation of genes involved in flavin metabolism and cytochrome P450 activity were protective in all cases, their magnitude of significance and effect was much greater in HeLa cells (Table S3). Nonetheless, our focused validation screens suggest that the genetic pathways implicated in oxidative stress from the genome-wide screens, including the seemingly paradoxical roles identified for the peroxisomal import and PPPs, are not specific to K562 cells and could be observed with multiple perturbation technologies.

Figure 2. Hits from Genome-wide Screens Validate in Competitive Growth Assays and in Multiple Cell Lines.

(A) CasTLE effect sizes for secondary Cas9 knockout and CRISPRi screens in K562 and HeLa cells ± 95% credible interval. Blue circle, K562 Cas9-KO; green circle, K562 CRISPRi; red square, HeLa Cas9-KO; yellow square, HeLa CRISPRi. Genes shown represent key hit pathways.

(B) Comparison of secondary screen and genome-wide Cas9 knockout screens in K562 (r² = 0.65). Genes chosen from key pathways are in red.

(C) Results from sgRNA competitive growth assays. Bars represent mean across experiments (n = 3 per sgRNA) ± SD across sgRNAs. Dark blue dots show fold enrichment for individual experiments of one sgRNA, while light blue dots show fold enrichment for individual experiments for another sgRNA. **p < 0.01 and ***p < 0.001, by Student’s t test.

(D) Results from shRNA competitive growth assays. Bars represent mean across experiments (n = 3 per shRNA) ± SD across shRNAs. Dark orange dots show fold enrichment for individual experiments of one shRNA, while light orange dots show fold enrichment for individual experiments for another shRNA. *p < 0.05, **p <0.01, and ***p < 0.001, by Student’s t test.

To further confirm our screen results, we validated select hits from the genome-wide screen using competitive growth assays with individual sgRNAs. We mixed unlabeled control cells with fluorescently labeled cells expressing shRNAs or CRISPR sgRNAs targeting hits identified in the screen (Figure S2F). Using two sgRNAs or shRNAs per gene, we observed statistically significant protective and sensitizing effects for hits involved in iron homeostasis, iron trafficking, antioxidant activity, peroxisomal matrix import, and the PPP (Figures 2C and 2D), all of which corresponded to their phenotypes in the primary genome-wide and batch retest screens.

Disruption of the Peroxisomal Import Pathway Leads to Cytosolic Localization of Catalase and Protection against ROS

Strikingly, our results revealed that disruption of every peroxisomal matrix protein import pathway gene was protective against oxidative stress (PEX5, PEX13, PEX14, PEX2, PEX12, PEX10, PEX1, PEX6, and PEX26) (Figure 3A), despite the role of the peroxisome as a center for reducing activity. At the same time, we identified no peroxisomally localized proteins (other than the PEX genes themselves) whose loss of function could explain the observed effect, with the notable exception of catalase. Catalase, deletion of which was highly sensitizing, is imported into the peroxisome by the PEX5 pathway, where it detoxifies H₂O₂. Thus, we hypothesized that perturbation of the peroxisomal matrix protein import pathway causes localization of catalase to the cytoplasm, where it may more effectively detoxify cellular ROS.

Figure 3. Disruption of the Peroxisomal Import Pathway Leads to Cytosolic Localization of Catalase and Protection against ROS.

(A) Diagram showing peroxisomal matrix protein import pathway hits.

(B) Fold enrichment of knockout cells over control cells. Bars represent mean across experiments (n = 3) ± SE. **p < 0.01 and ***p < 0.001, by Student’s t test. Results from K562 cells are shown in blue and results from HeLa cells in red.

(C) Digitonin-mediated subcellular fractionation assay using negative control and PEX5-KO HeLa cells showing that catalase is enriched in the cytoplasm in PEX5-KO cells. Cells were fractionated into whole-cell extract (WCE), cytoplasm (C), and peroxisome-enriched pellet (P). LDH was immunoblotted as a cytoplasm control, and peroxisomal membrane proteins PEX13 and PEX14 were stained as peroxisomal controls. Peroxisomal matrix proteins with PTS1, such as ACOX1 and LONP2, were also immunoblotted. Only the B-chain of ACOX1 was shown in immunoblot with anti-ACOX1 antibody.

(D) Digitonin enzymatic release assay experimental diagram. Intact membranes are depicted with solid lines, and digitonin-permeabilized membranes are shown with dotted lines.

(E) Digitonin enzymatic release assay showing mean catalase enzymatic activity released (n = 3 for control and PEX5-KO, n = 2 for PEX12-KO) ± SE in HeLa cells. Curve represents hill equation regression modeling of data with r² values for fit to data as well as the half maximal effective concentration (EC₅₀) of digitonin for each cell line on the basis of fitted line.

(F) Representative immunofluorescence microscopy images for two negative control HeLa cell lines and HeLa cells expressing PEX5 and PEX12 sgRNAs. Cells were stained with DAPI to stain nuclei, PMP70 antibodies to stain peroxisomal membranes, and catalase antibodies. Scale bar represents 10 μm.

(G) Normalized area under the curve (A.U.C.) lethal fraction scores showing cell death of WT (gray) and CAT-KO (blue) cells expressing control and PEX sgRNAs after 48 h treated with 300 μM H₂O₂. Bars represent mean across experiments (n = 8) ± SE. NS, nonsignificant (p > 0.05); ***p < 0.001 (comparisons with negative control sgRNAs by Student’s t test).

To test whether perturbation of the peroxisomal matrix protein import pathway causes localization of catalase to the cytoplasm, we first created clonal KO lines of PEX5 and PEX12 and confirmed their protective effects in competitive growth assays (Figure 3B; Figure S3A). We then performed digitonin-permeabilization and subcellular fractionation to measure the localization of catalase in HeLa PEX5-KO or control cells (Figure 3C). We observed a complete shift of peroxisomal catalase to the cytoplasm in the PEX5-KO cells compared with the cells expressing a negative control sgRNA, while peroxisomal membrane proteins (PEX14 and PEX13) and a cytoplasmic marker (LDH) did not change localization. Impaired import of peroxisomal matrix proteins in PEX5-KO cells was verified by immunoblot indicating that B-chain of a PTS1 enzyme, ACOX1, derived from intraper-oxisomal processing of the full-length A-chain (Miyazawa et al., 1989) was not discernible and that another PTS1 protein, LONP2, was detectable only in the cytoplasm with lower stability (Figure 3C). Peroxisome function in PEX5-KO HeLa and K562 cells was assessed by analyzing plasmalogen levels by liquid chromatography-tandem mass spectrometry (LC-MS/MS). We found reduced levels of plasmalogens in both cell lines, suggesting that PEX5-KO results in non-functional peroxisomes (Figure S3B).

To further quantify catalase localization in the cytoplasm, we measured its enzymatic activity by selectively permeabilizing cellular membranes with digitonin and observing H₂O₂ degradation by the catalase released from the cells in negative control, PEX5-KO, and PEX12-KO HeLa cells (Tsukamoto et al., 1990). Because digitonin is a cholesterol-dependent membrane permeabilizing agent, lower concentrations of digitonin will selectively permeabilize the plasma membrane (which has higher cholesterol content), while only higher concentrations will permeabilize the peroxisomal membrane (Figure 3D). The assay is sensitive to catalase released into the assay buffer; thus the titration point of the catalase activity is dependent on the localization of catalase within the cell. We observed a much lower digitonin titration point for PEX5-KO and PEX12-KO HeLa cells than HeLa cells expressing a negative control sgRNA, strongly supporting that catalase is enriched in the cytoplasm in PEX5-KO and PEX12-KO cells (Figure 3E). Finally, we used immunofluorescence microscopy in PEX5 and PEX12 KO HeLa cells, which also demonstrated the localization of catalase to the cytoplasm, as well as the canonical “peroxisomal ghost” phenotype of these two KOs revealed by PMP70 staining (Figure 3F).

We next tested whether catalase is necessary for the protective effect of PEX5 and PEX12 sgRNAs. We created a clonal catalase-null (CAT-KO) HeLa cell line and expressed safe-targeting (control), PEX5, or PEX12 sgRNAs with Cas9 in either these cells or in HeLa-WT (wild-type) cells (Figure S3C). Cells were treated with 300 μM H₂O₂ and monitored using automated microscopy for 48 h post-H₂O₂ treatment to quantify cell death by uptake of SYTOX Green dye. We observed an overall greater sensitivity to the 300 μM H₂O₂ dose in the HeLa CAT-KO cells compared to the HeLa-WT cells, consistent with the screening results (Figure 3G). While sgRNAs targeting PEX5 and PEX12 were significantly protective in HeLa-WT cells treated with H₂O₂, they did not have a protective effect in HeLa-CAT-KO cells. Notably, we observed no major defects in peroxisomal number or morphology in CAT-KO cells using immunofluorescence staining for PMP70, which suggests that catalase KO cells do not have an aberrant peroxisomal phenotype (Figure S3D). Thus, only disruption of catalase-positive peroxisomes protects cells from ROS. Together, these data demonstrate that catalase is localized to the cytoplasm in peroxin gene KOs and is necessary for the protective effect observed in the context of oxidative stress.

Localization of Catalase to the CytoplasmIs Sufficient to Protect the Cell from ROS

To test whether non-peroxisomal catalase was sufficient to protect cells against oxidative stress, we examined if targeting CAT to different locations in the cell could protect against ROS (Figure 4A). We hypothesized that because PEX-KO cells have enriched cytoplasmic catalase and catalase is necessary for the PEX-KO protective effect, non-localized catalase should have an increased protective effect (Figure S4A). Native catalase has a non-canonical peroxisomal localization motif at its C terminus and localizes incompletely to the peroxisome (Purdue and Lazarow, 1996; Sheikh et al., 1998; Williams et al., 2012). We created expression constructs for catalase with its WT localization domain (K-A-N-L, WT-CAT), catalase with no localization domain (CAT-DKANL), and catalase with the canonical PTS1 peroxisomal localization domain (S-K-L, CAT-SKL), all in addition to an N-terminal FLAG epitope. Expression constructs were stably infected into HeLa cells using lentivirus (Figure S4B). Analysis of the expression of CAT variants by western blot showed levels similar to those of WT CAT (Figure 4B). Immunofluorescence microscopy confirmed the predicted localization of constructs in HeLa cells: WT-localized and PTS1 catalase localized to the peroxisomes, while the non-localized CAT-ΔDKANL localized diffusely across the whole cell (Figure 4C). We tested the sensitivity of HeLa cells expressing the differentially localized catalases to ROS by treating cells with 350 μM H₂O₂ (higher dose needed to achieve LD₅₀ in catalase-expressing cells). Cells were monitored using automated microscopy for 48 h post-H₂O₂ treatment to quantify live cells by nuclear-localized mCherry and cell death by uptake of SYTOX Green dye (Figure 4D). Basal levels of cell death and growth were similar among untreated cell lines. However, HeLa cells expressing WT-CAT and CAT-ΔDKANL had significant reductions in cell death after H₂O₂ treatment compared with HeLa cells with no plasmid, by lethal fraction score (Forcina et al., 2017). Importantly, non-localized catalase had the lowest amount of cell death across any of the tested overexpression constructs. Interestingly, overexpressing CAT-SKL did not significantly decrease cell death compared with cells with no construct after H₂O₂ treatment. These results indicated that catalase localization determines its efficacy in protecting cells from oxidative damage and that cytoplasmic catalase is more protective against ROS than peroxisomal catalase.

Figure 4. Localization of Catalase to the Cytoplasm Is Sufficient to Protect the Cell from ROS.

(A) Diagram showing constructs used in differentially localized catalase panel.

(B) Western blot showing overexpression of catalase constructs in wild-type HeLa cells.

(C) Representative immunofluorescence microscopy images showing subcellular localization of catalase constructs. Cells were stained with DAPI to stain nuclei, PMP70 antibodies to stain peroxisomal membranes, and FLAG antibodies to stain FLAG-tagged catalase. Scale bar represents 10 μm.

(D) Normalized area under the curve (A.U.C.) lethal fraction scores showing cell death of HeLa cells expressing differentially localized catalase constructs or noplasmid. Bars represent mean across experiments (n = 9) ± SE. **p < 0.01 and ***p < 0.001, by Student’s t test.

PGD KO Decreases Glycolysis, Increases Upper PPP Glucose Flux, and Protects Cells against Oxidative Stress

Surprisingly, the genome-wide CRISPR/Cas9 and shRNA screens showed that deletion of specific members of the PPP protect cells against oxidative stress, including 6-phosphogluconolactonase (PGLS), PGD, and transketolase (TKT). This is counterintuitive given that the PPP metabolizes glucose-6-phosphate to reduce NADP⁺ into NADPH, which maintains the antioxidant capacity of the cell, and so removal of this pathway might be expected to be sensitizing (Kirkman et al., 1999; Lu, 2009; Patra and Hay, 2014). We generated KO lines for PGLS, PGD, and TKT using two sgRNAs per gene and validated their protective effects against H₂O₂ treatment by competitive growth assay (Figure 2C). PGD KO cells exhibited profound protection against H₂O₂ compared with other PPP KOs. Although PGD KO alone had a large negative growth effect, this was unlikely to be the cause of the protective effect, as there were many other genes with equivalently strong growth effects in the screen but no protection against ROS (Figure S5A). We also found no increase in the key antioxidant transcription factor, NRF2, in PGD-KO cells (Figures S5C and S5D). We hypothesized that this dramatic ROS-protective effect could be explained by the accumulation of the substrate of PGD, 6PG, which inhibits glycolysis and promotes glucose flux through the oxidative PPP, generating more NADPH during oxidative stress conditions compared with WT cells (Figure 5A; Kuehne et al., 2015; Parr, 1956). Indeed, we observed a very significant accumulation of 6PG in PGD KO cells using mass spectrometry: compared with negative control cells, 6-phosphogluconolactone (6PGL) and 6PG were present at 100and 20-fold higher levels, respectively (Figures 5B and 5C).

Figure 5. PGD Knockout Decreases Glycolysis, Increases Upper PPP Glucose Flux, and Protects Cells against Oxidative Stress.

(A) Diagram showing pentose phosphate pathway hits from genome-wide screens.

(B) Abundance of 6-phosphogluconolactone by mass spectrometry in K562 cells either untreated or treated with 30 μM H₂O₂ 0, 5, 15, and 30 min before sample preparation. Bars represent mean across experiments (n = 3) ± SE.

(C) Abundance of 6-phosphogluconate by mass spectrometry in K562 cells either untreated or treated with 30 μM H₂O₂ 0, 5, 15, and 30 min before sample preparation. Bars represent mean across experiments (n = 3) ± SE.

(D) Rate of anaerobic glycolysis and anaerobic glycolytic capacity measured by Seahorse glycolytic stress test assay of K562 cells expressing negative control orPGD sgRNAs. Bars represent mean extracellular acidification rate (ECAR) across experiments (n = 7) ± SE. **p < 0.01 and ***p < 0.001, by Student’s t test.

(E) CM-H₂DCFDA fluorescence in PGD knockouts and negative control K562 cells. Background of non-dye cells was subtracted for each experiment. Bars represent mean across experiments (n = 5) ± SE. *p < 0.05 by Student’s t test.

(F) Western blot showing K562 polyclonal population knockouts of PGD and G6PD.

(G) Representative percentages for K562 competitive growth assay cultures showing the growth (left) and protective (right) effects of PGD knockout cells and the abrogation of the phenotype in the presence of a G6PD knockout. Bars represent mean across experiments (n = 3) ± SE. Experiment was performed twice with two sgRNAs per gene tested; full experimental results are shown in Figure S5B.

To test whether the accumulation of 6PG was responsible for the protective effect of PGD deletion against H₂O₂ treatment, we knocked out the upstream pathway member G6PD (which produces 6PGL, leading to 6PG production) in the background of PGD-KO cells (Figure 5F). Our hypothesis suggests that this should prevent both the accumulation of 6PG and the consequent upregulation of PPP flux, thus removing the protective phenotype of the PGD KO. Strikingly, whereas G6PD KO alone had little to no effect on growth rate or H₂O₂ sensitivity, the double G6PD/PGD KO almost completely reversed the phenotypes of the PGD KO, restoring both growth rate and sensitivity to H₂O₂ to WT levels (Figure 5G; Figure S5B). Interestingly, knocking out another upstream gene, PGLS, might also be expected to prevent accumulation of 6PG and cause protection against H₂O₂. However, we observed little effect on H₂O₂ sensitivity, possibly because of spontaneous hydrolysis of PGLS substrate, 6PGL, to 6PG, which had been previously observed (Miclet et al., 2001; Figure S5B). Overall, our data show that the growth phenotype and protective effect of PGD KO depend on G6PD activity.

To determine if the accumulation of 6PG in a PGD KO would decrease glycolysis (Figure 5A), we measured anaerobic glycolysis rates in K562 PGD-KO and negative control cells. We found decreased basal glycolysis rates and glycolytic capacity (Figure 5D; Figure S5E). We also found that G6PD KO decreased glycolysis rates, but to a lesser extent than PGD-KO cells. The double KO of PGD/G6PD had decreased glycolysis levels but was closer to the glycolytic rates of G6PD-KO compared with PGD-KO (Figure S5F). Thus, the decreased anaerobic glycolytic rate of PGD-KO cells and the partial rescue by KO of G6PD is consistent with inhibition of glycolysis rates by accumulation of 6PG in the oxidative branch of the PPP.

We next investigated the effect of PGD KO on cellular redox levels. We used a chloromethyl derivative of 2’,7’-dichlorodihydrofluorescein diacetate (CM-H₂DCFDA), a ROS-sensitive fluorescent dye, to monitor the intracellular redox environment of K562 cells after H₂O₂ treatment. After H₂O₂ treatment, PGD-KO K562 cells had only minor changes to H₂DCFDA fluorescence compared with untreated cells, while negative control cells had steadily increasing levels of ROS signal after H₂O₂ treatment (Figure 5E). This observation suggests that PGD-KO cells have stronger antioxidant responses to ROS compared with non-PGD-KO cells. Similarly, in HeLa cells basal levels of ROS were comparable between negative control and PGD-KO HeLa cells, but after 5 min of H₂O₂ treatment, PGD-KO cells had significantly lower ROS signal than negative control cells (Figure S5I). Interestingly, despite protection against H₂O₂ in PGD-KO HeLa cells (Figure 2A), time points at 15, 30, and 60 min showed no statistically significant difference in ROS levels between HeLa PGD-KOs and controls. The reduction in ROS was also reversed when G6PD and PGD were both deleted (Figures S5G–S5J), consistent with the reversal of PGD-KO phenotypes on growth, protection against H₂O₂, and glycolysis. Interestingly, there was no difference in steady-state NADPH/NADP⁺ ratio in PGD-KO cells in response to H₂O₂ treatment (Figures S5K and S5L).

Taken together, our data support a model in which PGD-KO protects cells against ROS toxicity through an accumulation of 6PG, which inhibits glucose-6-phosphate isomerase (GPI) and promotes glucose flux through the oxidative PPP at the expense of glucose flux through glycolysis. We observed increased PPP metabolites in PGD KO as well as decreased glycolysis, indicating an upregulation of PPP activity. Decreased ROS burden after H₂O₂ treatment, as shown by H₂DCFDA fluorescence, suggests that PGD-KO cells have a stronger antioxidant response to acute increases in ROS compared with controls. Finally, we observed a complete buffering of the growth inhibition, ROS protection, and other phenotypes of PGD KO when G6PD was also deleted. These observations suggest that PGD is an important control point in the metabolic cellular response to acute oxidative stress.

DISCUSSION

In this study we conducted genome-wide shRNA and CRISPR screens to comprehensively identify diverse genetic factors that modulate cellular sensitivity to oxidative stress. We show through focused secondary sgRNA library screens that these major pathways are critical regulators of ROS response in both K562 and HeLa cells. Among these, we demonstrate roles for genes involved in DNA damage response, key antioxidants, iron homeostasis, peroxisomal protein import, and the PPP. We provide detailed insight into two pathways whose deletion showed unexpectedly protective phenotypes, contradictory to their annotated roles in redox biology: the peroxisomal matrix import pathway and the PPP. Although our results identify certain genes that were found in previous screens in model organisms (Ayer et al., 2012; Ueno et al., 2012), early RNAi screens (Kimura et al., 2008), and targeted screens in mammalian cells (Reczek et al., 2017), the majority of genes we found as significant hits were not identified in these screens.

One of the strongest protective signatures from our screens was one of the most puzzling: the peroxisomal matrix import pathway protects the cell from ROS when disrupted. Our screens identified all nine members of the PTS1 pathway involved in the import of catalase as protective hits. With three orthogonal methods, we show that perturbation of the peroxisomal matrix import pathway causes enrichment of non-peroxisomal catalase (Figures 3C–3F). Our data demonstrate that catalase-null cells have no protective phenotype against H₂O₂ when expressing sgRNAs targeting PEX genes, suggesting that catalase is necessary for the protective effect observed in PEX KOs (Figure 3G). Using a panel of differentially localized catalase constructs, we show that non-localized catalase is most protective in cells treated with H₂O₂ (Figure 4D). This suggests that the localization of catalase in the cell determines its efficacy in ameliorating oxidative stress.

Our results are consistent with an emerging model in the field of peroxisomal biology that catalase is actively transported outside of peroxisomes during periods of oxidative stress, where it confers a greater protective effect for the cell (Walton et al., 2017). A recent study identified redox-sensitive sites on PEX5 that inactivate peroxisomal protein import under oxidative conditions, which would prevent newly translated catalase from entering the peroxisome (Apanasets et al., 2014). Additional studies have suggested that BAK can permeabilize peroxisomal membranes during oxidative stress and may be involved in actively transporting catalase from the peroxisomes to the cytoplasm (Hosoi et al., 2017). Thus, although localization of catalase to the peroxisome is needed for detoxifying basal levels of ROS produced there, catalase export to the cytoplasm is likely a functional mechanism for ameliorating acute oxidative stress. Our results suggest that peroxisomal catalase export is a key control point in the cellular ROS response and that peroxisomes may act as antioxidant reservoirs for cells. Furthermore, delivery or manipulation of catalase localization could be a promising therapeutic strategy for diseases related to ROS (Barnham et al., 2004; Ebadi et al., 1996; Haney et al., 2015; Wu et al., 2003).

Another paradoxical finding was the protective effect of deleting the core PPP gene PGD. The role of the PPP in cellular survival and stress is complex. PGD has been shown to be upregulated in cancer metastases that are resistant to ROS (Bechard et al., 2018; McDonald et al., 2017; Shan et al., 2014), but inhibition of the PPP has also been shown to prevent erastin-induced ferroptosis, a form of cell death mediated by iron and ROS (Dixon et al., 2012). We show that deletion of PGD is strongly protective against H₂O₂ in HeLa and K562 cells in screens and individual competitive growth assays. Our data suggest that this is due to accumulation of the upstream metabolite 6PG (Figure 5C), which can inhibit GPI and thus glycolysis, increasing glucose flux through the PPP (Figure 5A; Kuehne et al., 2015). The increase of glucose flux through the oxidative PPP, which would result in increased NADPH production, could explain our observation that PGD-KO cells have significantly reduced ROS burden after H₂O₂ treatment (Figure 5E; Figures S5H and S5I). Finally, knocking out G6PD in PGD-KO cells buffered all of the phenotypes we observed (H₂O₂ protection, slow growth, reduced ROS), suggesting that the PGD-KO mechanism relies on the activity of G6PD and accumulation of PPP metabolites (Figure 5G). Although the precise mechanisms by which PGD-KO confers such dramatic cellular oxidative stress protection are unclear, it is interesting to note that we found no increase in the key antioxidant transcription factor, NRF2, in PGD-KO cells. This suggests that protection is not conferred by upregulation of this classic mechanism of antioxidant activity (Figures S5C and S5D). In addition, we find that the steady-state NADPH/NADP⁺ ratio in PGD-KO cells is unchanged, consistent with previous observations that perturbation of PGD protein levels does not affect the tightly controlled cytoplasmic NADPH/NADP⁺ ratio in cells (Figures S5K and S5L; Au et al., 2000; Christodoulou et al., 2019; Tao et al., 2017). Further investigation will be required to fully understand how PGD contributes to the control of glycolysis and cell survival, but our results suggest that it is a critical regulator of response to oxidative stress. Taken together, our results represent a rich resource for the study of oxidative stress and highlight a number of potential targets that might be exploited for therapeutic intervention.

STAR★METHODS

LEAD CONTACT AND MATERIALS AVAILABILITY

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Michael C. Bassik (bassik@stanford.edu). Plasmids generated during this study are deposited at Addgene, numbers available in key resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit FLAG antibody	Cell Signaling Technology	#14793; RRID:AB_2572291
Mouse CAT antibody	Biomol International	YIF-LF-MA0003; RRID:AB_1611839
Mouse GAPDH antibody	Invitrogen	AM4300; RRID:AB_437392
Rabbit PGD antibody	Cell Signaling Technology	#13389; RRID:AB_2798202
Rabbit NRF2 antibody	Cell Signaling Technology	#12721; RRID:AB_2715528
Rabbit TKT antibody	Cell Signaling Technology	#8616; RRID:AB_10950823
Rabbit CAT antibody	Abcam	ab16731; RRID:AB_302482
Rabbit PMP70 antibody	Abcam	ab3421; RRID:AB_2219901
Rabbit PEX12 antibody	Abcam	ab103456; RRID:AB_10711955
Rabbit PGLS antibody	Abcam	ab127560; RRID:AB_11130290
Rabbit b-actin antibody	Abcam	ab8227; RRID:AB_2305186
Rabbit IREB2 antibody	Abcam	ab181153
Mouse PMP70 antibodies	Sigma	SAB4200181; RRID:AB_10639362
Rabbit PEX5 antibody	Otera et al., 2000	N/A
Rabbit PEX13 antibody	Mukai and Fujiki, 2006	N/A
Rabbit PEX14 antibody	Shimizu et al., 1999	N/A
Rabbit ACOX1 antibody	Tsukamoto et al. 1990	N/A
Rabbit LONP2 antibody	Okumoto et al., 2011	N/A
Chemicals, Peptides, and Recombinant Proteins
Hydrogen Peroxide	Millipore-Sigma	#216763
Herculase II Fusion DNA Polymerase	Agilent	Cat# 600675
Digitonin	Wako Chemicals USA	#NC0141730
Triton X-100	Aqua Solutions	SKU# T9010
SYTOX Green	Thermo Fisher Scientific	S7020
CM-H2DCFDA	Thermo Fisher Scientific	C6827
Critical Commercial Assays
Qiagen DNA Blood Maxi kit	Qiagen	Cat# 51194
Promega NADP/NADPH-Glo Assay	Promega	G9081
Seahorse glycolysis stress test	Agilent	#103017–100
Deposited Data
Raw Screening Data	This paper	SRA: SRP235195
List of CRISPR/Cas9 K562 genome-wide screen hits	This paper	N/A
List of shRNA genome-wide K562 hits	This paper	N/A
List of CRISPR/Cas9 K562 secondary batch screen hits	This paper	N/A
List of CRISPRi K562 secondary batch screen hits	This paper	N/A
List of CRISPR/Cas9 HeLa secondary batch screen hits	This paper	N/A
List of CRISPRi HeLa secondary batch screen hits	This paper	N/A
Experimental Models: Cell Lines
Human: K562 Cells	ATCC	CLL-243
Human: HeLa Kyoto Cells	Hein et al., 2015	CVCL_1922
Oligonucleotides
Oligos for sgRNAs and shRNAs, see Table S4	This paper	N/A
Primers for TIDE, see Table S4	This paper	N/A
Primers for library sequencing, see Table S4	This paper	N/A
Recombinant DNA
pMCB320 (pSico-pU6-sgRNA EF1A-Puro-T2A-mCherry)	Morgens et al., 2017	Addgene Plasmid #89359
pMCB306 (pSico-pU6-sgRNA EF1A-Puro-T2A-GFP)	Han et al., 2017	Addgene: #89360
pMCB280 (SFFV-Puro-T2A-mCherry-shRNA)	Deans et al., 2016	N/A
pMCB309_pMK1221 (pSicoR-CMV-shRNA SFFV-puro-mCherry)	Kampmann et al., 2015	Addgene: #84220
pMMD650_UbC-G418-T2A-WT-CAT	This paper	Addgene: #137990
pMMD651_UbC-G418-T2A-CAT-ΔKANL	This paper	Addgene: #137991
pMMD652_UbC-G418-T2A-CAT-SKL	This paper	Addgene: #137992
Software and Algorithms
casTLE screening analysis	Morgens et al., 2016	https://bitbucket.org/dmorgens/castle
IncuCyt®e S3 Base Software	Essen Biosciences	V2016A/B
Excel	Microsoft	N/A
Progenesis QI 2.3 software	Progenesis	N/A

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell Culture and antibiotics

K562 cells (ATCC) were grown in Roswell Park Memorial Institute (RPMI) 1640 Medium (Life Technologies, 11875093) supplemented with 10% Fetal Bovine Serum (Fisher, Cat# SH30910), 2 mM L-glutamine (Fisher, Cat# SH3003401) and 1% penicillin-streptomycin (Fisher, Cat#SV30010), and cultured at 37°C with 5% CO₂. HeLa Kyoto cells (CVCL_1922) were grown in Dulbecco’s Modified Eagle’s Medium (Life Technologies, Cat# 11995073) supplemented with 10% FBS, 2 mM L-glutamine, and 1% penicillinstreptomycin, and cultured at 37°C with 5% CO₂. Cell cultures were routinely tested and found negative for mycoplasma infection (MycoAlert, Lonza).

METHOD DETAILS

Genome Wide sgRNA Screen in K562 Cells

Lentiviral screening library preparation, lentiviral infection, library selection, genomic DNA extraction, PCR amplification, and casTLE analysis were performed as previously described (Deans et al., 2016; Jeng et al., 2019; Morgens et al., 2017). K562 cells, with a BFP-Cas9 lentiviral transgene, were lentivirally infected with a genome-wide sgRNA library as described (Morgens et al., 2017), containing ~200,000 sgRNAs targeting 20,549 protein-coding genes and 13,500 negative control sgRNAs at an MOI of 0.3–0.4 (as measured by the mCherry fluorescence from the lentiviral vector) with approximately 1,000x coverage per library element, and selected for lentiviral integration using puromycin (1 μg/ml) for 3 days as the cultures were expanded for the screens. In one flask, cells were grown with four pulses of 30 μM H₂O₂ (Millipore-Sigma #216763), the LC₅₀ dose for K562-Cas9 cells, with a four day recovery in-between pulses or until cells began to double nearly daily again. The other flask was grown in the absence of H₂O₂. All conditions and replicates were maintained at a minimum of 1000x coverage per element for the duration of the screen. Untreated cells were diluted to a density of 500,000 cells/mL each day. H₂O₂-treated cells were diluted to a density of 500,000 cells/mL as needed. The cell culture screening lasted for 21 days total, including a final recovery period, with approximately 9.5 and 24 doublings for the treated and untreated populations, respectively. After the cell culture period, untreated and H₂O₂-treated cells were pelleted by centrifugation for genomic DNA extraction using the Qiagen DNA Blood Maxi kit (Cat# 51194), as described by manufacturer’s instructions. Approximately 2.5×10⁸ cells were used for each genomic DNA extraction to maintain 1000x coverage per element. The sgRNA-encoding constructs were PCR-amplified using Agilent Herculase II Fusion DNA Polymerase (Cat# 600675) (See Table S4 for the primer sequences used). These libraries were then sequenced across an Illumina NextSeq flow cell (40 M reads per condition; ~200× coverage per library element). Computational analysis of genome-wide screen was performed as previously described using casTLE, which is a maximum likelihood estimator that uses a background of negative control sgRNAs as a null model to estimate gene effect sizes (Morgens et al., 2016). See Table S1 for the K562 genome-wide sgRNA screen results.

Genome Wide shRNA Screen for K562 Cells

Lentiviral screening library preparation, lentiviral infection, library selection, genomic DNA extraction, PCR amplification, and casTLE analysis were performed as previously described (Deans et al., 2016; Jeng et al., 2019; Morgens et al., 2017). K562 cells were lentirvirally infected with the genome-wide shRNA library as described previously (Deans et al., 2016), containing 506,995 sgRNAs targeting 19,128 protein-coding genes and 7,689 negative control shRNAs at an MOI of 0.3–0.4 (as measured by the mCherry fluorescence from the lentiviral vector) with approximately 1,000x coverage per library element, and selected for lentiviral integration using puromycin (1 mg/ml) for 3 days as the cultures were expanded for the screens. Infected cells were expanded and split into two flasks. Treated cells were grown with either 5 or 6 increasing doses of H₂O₂ (Millipore-Sigma #216763) (45 μM, 100 μM, 125 μM, 125 μM, and 125 μM or 30 μM, 35 μM, 40 μM, 50 μM, 100 μM, and 100 μM, respectively) until there was a roughly two-fold difference in doublings between the untreated and treated libraries (31.7 and 15.8 doublings or 29.9 and 17.7 doublings, respectively). The treated library was allowed to recover so cells were roughly doubling each day before the next treatment. All conditions and replicates were maintained at a minimum 1000x coverage per element for the duration of the screen. Untreated cells were diluted to a density of 500,000 cells/mL each day. H₂O₂-treated cells were diluted to a density of 500,000 cells/mL as needed. Cell culture screening lasted for 24 or 25 days total, including a final recovery period. After the cell culture period, untreated and H₂O₂-treated cells were pelleted by centrifugation for genomic DNA extraction using Qiagen DNA Blood Maxi kit (Cat# 51194), as described by manufacturer’s instructions. Approximately 5.0×10⁸ cells were used for each genomic DNA extraction to maintain 1000x coverage per element. The shRNA-encoding constructs were PCR-amplified using Agilent Herculase II Fusion DNA Polymerase (Cat# 600675). These libraries were then sequenced across an Illumina NextSeq flow cell (~80 M reads per condition; ~200× coverage per library element). Computational analysis of genome-wide screen was performed as previously described using casTLE, which is a maximum likelihood estimator that uses a background of negative control shRNAs as a null model to estimate gene effect sizes (Morgens et al., 2016). See Table S1 for the K562 genome-wide shRNA screen results.

Batch Screen for K562 Cells

The batch screen library included the following, non-comprehensive sets of genes (1,000 genes in total, ~10 sgRNAs per gene, plus 5,000 negative control sgRNAs [2,000 non-targeting scrambled controls, 3,000 safe-targeting guides]): 511 hits passing 5% FDR from our genome-wide CRISPR screen, 282 manually identified genes from the literature that have been previously shown to be involved in redox biology or are related to pathways identified in the genome-wide screens, and finally 207 genes just outside the 5% FDR threshold in the genome-wide CRISPR screen for a total of 1,000 genes (see Table S2 for a complete list of genes and complete 14,917 element list). The same library of genes was used to generate a CRISPRi sgRNA library as well, with the same number of control sgRNAs (see Table S2 for complete 15,460 element list). The library oligos were synthesized by Agilent Technologies and cloned into pMCB320 using BstXI/BlpI overhangs after PCR amplification. Lentiviral screening library preparation, lentiviral infection, library selection, genomic DNA extraction, PCR amplification, and casTLE analysis were performed as previously described (Deans et al., 2016; Jeng et al., 2019; Morgens et al., 2017). The Cas9 and dCAS9-KRAB K562 cell lines were lentivirally infected with the secondary library at an MOI of 0.3–0.4 as described for the genome-wide screens. After puromycin selection (1 0μg/ml for 3 days) and expansion, 30 M (~2,000 coverage per library element) cells were treated with four pulses of 30 μM H₂O₂ (Millipore-Sigma, #216763), with a four day recovery in-between pulses or until cells began to double nearly daily again with a logarithmic growth (500,000 cells/ml) maintained as in the K562 genome-wide screen. Thirty million cells under each condition were used for genomic extractions, sequenced (~6–10 million reads per condition; 1000–20003 coverage per library element) and analyzed using casTLE as described for the genome-wide screens. See Table S3 for the K562 secondary screen results

Batch Screen for HeLa Cells

Lentiviral screening library preparation, lentiviral infection, library selection, genomic DNA extraction, PCR amplification, and casTLE analysis were performed as previously described (Deans et al., 2016; Jeng et al., 2019; Morgens et al., 2017). Cas9 and dCAS9-KRAB HeLa cell lines were lentivirally infected with the secondary library (14,917 and 15,460 elements, Table S2) at an MOI of 0.3–0.4 as described for the genome-wide screens. After puromycin selection (1 μg/mL for 3 days) and expansion, ×15 million Cas9 or dCas9-KRAB HeLa cells (~1,000× coverage of sgRNA library) were treated with three pulses of 300 μM H₂O₂ (Millipore-Sigma, #216763), with recovery in-between pulses until cells began to double nearly daily again with logarithmic growth. Cells were split every day to maintain logarithmic growth and prevent confluence. 15 million (~1,000× coverage) cells per condition were used for genomic extractions and sequencing as described for the genome-wide screens. See Table S2 for the HeLa secondary screen results.

Generation of Individual sgRNA-Expressing Cells/Stable Cell Lines

Lentivirus production and infection were performed as previously described (Morgens et al., 2016). Briefly HEK293T cells were transfected with third-generation packaging plasmids and the sgRNA-expressing vector. Lentivirus was harvested after 48 h and 72 h and filtered through a 0.45μm Polyvinylidene Fluoride (PVDF) filter (Millipore). K562 cells expressing lentiCas9-BFP were infected by centrifugation at 1000 x g for 2 h. HeLa cells expressing lentiCas9-BFP were infected by incubating the cells in lentivirus-containing media for 24 h. 3 days after infection, cells were selected with puromycin (1.2 μg/mL for 3 d for K562 and HeLa cells). Clonal knockout lines were generated for K562 PEX5-KO, K562 PEX12-KO, HeLa PEX5-KO, and HeLa PEX12-KO cell lines by single-cell sorting puromycin-selected cells into 96-well plates and expanding them for 2–3 weeks. Gene editing efficiency was determined by Sanger sequencing and analyzing the resulting chromatograms using TIDE software (Brinkman et al., 2014).

Western Blots

Live cells were lysed for 20 min at 4°C in protein extraction buffer (1% Triton X-100, 150 mM NaCl, 50 mM Tris pH 7.5, 1 mM EDTA) with protease inhibitors and centrifuged to at 21,000 x g to collect the supernatant lysate. For protein extraction buffer of nuclear lysates, cells were instead lysed with RIPA buffer (150 mM NaCl, 1% nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, 25mM Tris pH 7.4 in ddH2O) and sheared with a needle to extract nuclear protein before centrifugation. The cell lysate was measured with Bradford reagent (Biorad). For each western blot, 4x LDS Sample Buffer (Invitrogen, NP0007) and DTT (to final added concentration of 125mM) were mixed with each cell lysate sample, which were then denatured at 70°C for 10 minutes. Protein lysates were then separated on 4–12% acrylamide SDS-PAGE gels (Invitrogen, NP0322BOX) and transferred to nitrocellulose membranes (BioRad, #162–0146). After blocking nitrocellulose membranes for 30 minutes with blocking buffer (LI-COR, #927–70001) and washing with TBST (137mM NaCL, 2.7mM KCl, and 19mM Tris base), membranes were stained with the primary antibody diluted in 5% milk in TBST, according to antibody manufacturer’s instructions, for 1 hour at room temperature. After washing and secondary antibody staining (45 minutes, room temperature), membranes were visualized on an Odyssey CLx and analyzed with Image Studio 5.x.

Immunofluorescence

Live HeLa cells were grown on glass coverslips overnight in normal cell culture conditions. Cells were fixed with 4% paraformaldehyde (EMS, #15710) for 15 minutes at room temperature. Cells were permeabilized with 0.1% Triton X-100 (Aqua Solutions, SKU# T9010–500ML) at room temperature for 30 minutes. Coverslips were then blocked with 1% BSA for 30 minutes at room temperature before adding primary antibody stain to coverslips for 1 hour at room temperature. After primary staining, cells were stained with secondary antibodies for 1 hour at room temperature. Coverslips were incubated with Hoechst stain (Thermo Fisher Scientific, H3570) at 10 mg/mL for 5 minutes. Coverslips were adhered to a microscopy slide (Corning #2947–75×25) using permafluor (Thermo Fisher Scientific, #TA-030-FM) according to manufacturer’s instructions. Coverslips were imaged on a Nikon Eclipse Ti.

Antibodies

Rabbit FLAG antibody (#14793, Cell Signaling Technology), mouse CAT antibody (YIF-LF-MA0003, Biomol International), mouse GAPDH antibody (AM4300, Invitrogen), rabbit PGD antibody (#13389, Cell Signaling Technology), rabbit TKT antibody (#8616, Cell Signaling Technology), rabbit NRF2 antibody (#12721, Cell Signaling Technology), rabbit CAT antibody (ab16731, Abcam), rabbit PMP70 antibody (ab3421, Abcam), rabbit PEX12 antibody (ab103456, Abcam), rabbit PGLS antibody (ab127560, Abcam), rabbit β-actin antibody (ab8227, Abcam), rabbit IREB2 antibody (ab181153, Abcam), mouse PMP70 antibodies (SAB4200181, Sigma). Rabbit antibodies to PEX5 (Otera et al., 2000), PEX13 (Mukai and Fujiki, 2006), PEX14 (Shimizu et al., 1999), ACOX1 (Tsukamoto et al., 1990), LONP2 (Okumoto et al., 2011) were used as described in corresponding references.

Gene editing measurements by Sanger sequencing

Both HeLa and K562 PEX5-KO and PEX12-KO clonal cells were harvested and total genomic DNA was isolated using QuickExtract DNA Extraction Solution (VWR, QE09050). PCR was prepared using 5x Phusion HF reaction buffer (Thermo Fisher Scientific, F-518) and Phusion DNA polymerase (Thermo Fisher Scientific, F-530L), 10cmM dNTPs and primers designed about 400–550cbp upstream (MMD371_TIDE_PEX5_fwd, 5′-CTAGGTATGGTCGGGCTGTT; MMD373_TIDE_PEX12_fwd, 5′-AGTCACTGTTGGACCCTGAA) and 400–550cbp downstream (MMD372_TIDE_PEX5_rev, 5′-CTCAGACCACCACCACCTG; MMD374_TIDE_PEX12_rev, 5′-TGACGTG CAGCATTTGACAA) of the predicted cut site. PCR reactions were run on a C1000 Touch Thermo Cycler (Bio-Rad). PCR products were then purified over an Econospin DNA column (Epoch, 1910–250) using buffers PB and PE (Qiagen, 19066 and 19065). Sanger sequencing was performed and applied biosystems sequence trace files (.ab1 files) were obtained from Quintara Biosciences. Editing efficiencies of knockout cell lines was analyzed using the Netherlands Cancer Institute’s online TIDE analysis software (Brinkman et al., 2014).

Digitonin Enzymatic Release Assay

Assay was conducted in similar fashion to protocol outlined in (Tsukamoto et al., 1990) and in Figure 3C. HeLa cells were plated at 2 × 10⁶ cells in 6cm dish the day before conducting assay. HeLa cells were then trypsinized, resuspended, and count the cell number. Cells were kept on ice and washed cells in 1.5 mL tubes (1 × 10⁶ cells per tube) twice with cold Hepes-sucrose buffer (HSB) (250 mM sucrose, 20 mM Hepes-KOH (pH 7.4)). Cells were then diluted to 500,000 cells/mL in HSB. A dilution series of digitonin (Wako Chemicals USA, #NC0141730) was made in HSB in 1.5 mL tubes and 50 μL of each digitonin concentration, including a 2% Triton X-100 control (Aqua Solutions, SKU# T9010–500ML), was pipetted into the bottom of glass test tube. 100 μL of HSB wad added to the blank/no cell control glass tubes. All tubes were placed in a crushed-ice water bath (at 0C). At the very start of the assay, 8.75 μL of 30% H₂O₂ to 25 mL of substrate solution (0.02 M imidazole, 0.1% BSA, and 250 mM sucrose) to create the active substrate solution for the assay. Next, 50 μL of the cell dilutions (500,000 cells/mL) was added to digitonin dilutions, mixed briefly with vortex (setting #1), and incubated in crushed-ice water (0°C) for 6 minutes. After 6 minutes, 1.0 mL of the active substrate solution was added to samples, mixed briefly by vortexing, and incubated in crushed-ice water (0°C) for 15 minutes. After 15 minutes, 1.0 mL of Ti(SO₄)₂ solution was added and solutions were briefly mixed by vortexing to stop catalase reaction. After vortexing, glass test tubes were incubated at RT for 10 minutes. Finally solutions were pipetted into a 96-well plate and the OD 410 was measured. EC₅₀ curves were drawn and calculated by using the previously described Hill algorithm for each cell line (Gadagkar and Call, 2015).

Digitonin Subcellular Fractionation

HeLa cells, negative control and PEX5-KO, were plated the day before and grown to 80% confluence. Cells were washed twice with ice-cold PBS (4°C) and once with 4°C ice cold HSB, 1 mM EDTA, and protease inhibitors). Next, 400 μL of HSB containing 100 μg/mL digitonin (Wako Chemicals USA, #NC0141730) was added to cells. Cells and buffer were scraped into a 1.5 mL tube and incubated on ice for 10 minutes. Cells were divided into two different 1.5 mL tubes (180 μL/tube) for total lysate and cytosolic/membrane fraction lysates. The cytosolic/membrane fraction was centrifuged at 20,000 x g for 15 minutes at 4°C. After centrifugation, 150 μL of the supernatant was pipetted into a new 1.5 mL tube for the cytosolic fraction and 30 mL of 6x SDS-PAGE sample buffer (375 mM Tris-HCl, pH 6.8), 6% SDS, 6% b-mercaptoethanol, 30% glycerol) was added to sample. The supernatant left on the pellet was carefully removed and 50 μL of HSB was added to the pellet, or membrane fraction, which was further centrifuged at 20,000 x g for 15 minutes at 4°C to rinse the pellet. After carefully removing the HSB, 216 mL of 1x SDS-PAGE sample buffer was added to the membrane pellet, which was then sonicated in a sonic bath. Thirty-six μL of the 6x SDS-PAGE sample buffer was added to the 180 μL of total lysate fraction. Finally, all samples were boiled for 5 minutes before undergoing the standard western blot protocol as described above.

Incucyte Lethal Fraction Score Experiments

HeLa cell lines measured using the Incucyte were lentivirally infected with plasmid pMMD212-puro-3xNLS-mCherry. After puro selection (1 μg/ml for 3 days) and expansion, respective HeLa cell lines were trypsinized while in logarithmic growth and measured using a BD Accuri. After calculating live cells/mL, cell lines were diluted to 80,000 cells/mL and 0.5 mL was then plated in a 24 well plate (Corning #393047) and incubated under normal cell culture conditions for 2 hours. After two hours, 0.5 mL of DMEM medium with 40 ng/mL of SYTOX Green (Thermo Fisher Scientific, S7020) with or without 2x concentration of H₂O₂ (Millipore-Sigma, #216763) was added to treated or untreated wells. Cells were then placed in an Incucyte Zoom for 48 hours and were measured every 4 hours using exposure times of 400 ms for 488nm and 800 ms for 547nm.

Calculating Lethal Fraction Scores

To calculate lethal fraction score, we followed calculations outlined in (Forcina et al., 2017) combining live and dead cell counts every 4 hours across the experiment. Automated image analysis routines were optimized for HeLa cells using the Zoom software package (V2016A/B) and training data from untreated and H₂O₂ (Millipore-Sigma, #216763) treated samples.

HeLa CM-H2DCFDA

HeLa cells were plated at 40,000 cells/well 24 hours before initiating the experiment. Medium was replaced before beginning experiment and cells were treated with 300 μM H₂O₂ (Millipore-Sigma, #216763) for 5, 15, 30, and 60 minutes before washing 3 times with DMEM medium. After wash, cells were incubated in the dark at 37°C for 30 minutes with 5 μM CM-H₂DCFDA (Thermo Fisher Scientific, C6827) DMEM medium. After incubation, cells were washed 3 times with warm DMEM medium and then imaged using an Incucyte S3 live-cell analysis system (Sartorius).

K562 CM-H2DCFDA

K562 cells were plated at 400,000 cells/well in a 24-well plate and adhered to plate using poly-L-lysine and a 5 minute 300 x g spin. Medium was replaced before beginning experiment and cells were treated with 30 μM H₂O₂ (Millipore-Sigma, #216763) for 5, 15, 30, and 60 minutes before washing 3 times with RPMI medium. After wash, cells were incubated in the dark at 37°C for 30 minutes with 5 μM CM-H₂DCFDA (Thermo Fisher Scientific, C6827) RPMI medium. After incubation, cells were washed 3 times with warm RPMI medium and then imaged using an Incucyte S3 live-cell analysis system (Sartorius) or quantified using a BD Accuri C6 and BD Csampler software.

Seahorse Glycolysis Assays

K562 cells were plated at 30,000 cells/well in a Seahorse cell culture miniplate after washing twice with DPBS and adhered to plate using poly-L-lysine and a 5 minute 300 x g spin. Cells were incubated in Seahorse assay medium supplemented with 1% L-Glut at 37°C in a non-CO₂ incubator for 45–60 minutes. Cells were then assayed using the Seahorse glycolysis stress test (Agilent, #103017100) according to kit instructions. Each assay was conducted in triplicate for each cell line.

Metabolite Extraction for Untargeted Metabolomics by LC-MS

K562 cells were grown in RPMI medium and split into untreated and 4 different lengths of incubation with 30 μM H₂O₂ (MilliporeSigma, #216763) per genetic background: 0 minutes (H₂O₂ added immediately prior to washing), 5 minutes, 15 minutes, and 30 minutes. After H₂O₂ incubation, cells were washed three times with 25 mL of ice-cold (4°C) PBS buffer to remove any remaining medium, and aliquoted into 5×10⁶ cells and flash frozen in cell pellets. A previously described comprehensive untargeted metabolomics platform (Contrepois et al., 2015) was used to profile metabolites extracted from cell pellets. Cell pellets were re-suspended in 500 μl of 4:1 methanol:water including 17 internal standards to control for extraction efficiency and evaluate LC-MS performance. Re-suspended cells were vortexed for 30 s, sonicated using a bath sonicator for 30 s and incubated on ice for 30 s (repeated 3 times). Proteins were then precipitated by incubating samples at −20°C for two hours. Samples were centrifuged at 10,000 rpm for 10 min at 4°C, metabolite extracts were evaporated to dryness under nitrogen and reconstituted in 100 μl 1:1 methanol:water before analysis.

Data Acquisition for Untargeted Metabolomics by LC-MS

Each sample was analyzed four times using HILIC and RPLC separation in both positive and negative ionization modes. Data were acquired on a Thermo Q Exactive HF mass spectrometer for HILIC and a Thermo Q Exactive mass spectrometer (Thermo Fisher Scientific) for RPLC. Both mass spectrometers were equipped with a HESI-II probe and operated in full MS scan mode. For HILIC experiments, a ZIC-HILIC column (2.1 × 100 mm, 3.5 μm, 200 Å; EMD Millipore) was used with mobile phase solvents (A) 10 mM ammonium acetate in 50:50 acetonitrile:water and (B) 10 mM ammonium acetate in 95:5 acetonitrile:water. For RPLC experiments, a Zorbax-SB-aq column (2.1 × 50 mm, 1.7 μm, 100 Å; Agilent) used with mobile phase solvents (A) 0.06% acetic acid in water and (B) 0.06% acetic acid in methanol. Before running the sequences, LC-MS systems were equilibrated by injecting 12 and 6 pooled quality control samples (QCs) for HILIC and RPLC, respectively. MS/MS data were acquired on pooled QCs consisting of an equimolar mixture of all the samples in the study at normalized collision energies (NCE) of 25 and 35 for HILIC and 25 and 50 for RPLC. Multiple quality control measures were performed to ensure data quality. All samples were randomized prior to protein extraction and data acquisition. Further, mass accuracy, retention time and peak shape of internal standards were reviewed in each sample.

Data Processing for Untargeted Metabolomics by LC-MS

Raw data were imported into Progenesis QI 2.3 software (Water, Milford, MA, USA) to align and quantify chromatographic peaks. Data from each acquisition mode were processed independently and then merged. MS drift with time was corrected using the LOESS normalization method on pooled QCs injected every 10 injections in the sequence. A targeted list of metabolites containing 6-phosphogluconolactone and 6-phosphogluconate were formally identified by matching retention time and fragmentation spectra to authentic standards when possible.

Measurement of plasmalogens by LC-MS/MS

Total lipids were extracted by the Bligh and Dyer method (Bligh and Dyer, 1959). Briefly, 50 μg of total cellular proteins were suspended in methanol/chloroform/water at 2:1:0.8 (vol/vol/vol) and then 50 pmol of 1-heptadecanoyl-sn-glycero-3-phosphocholine (LPC; Avanti Polar Lipids #855676P), 1, 2-didodecanoyl-sn-glycero-3-phosphocholine (DDPC; Avanti Polar Lipids #850335P), and 1, 2-didodecanoyl-sn-glycero-3-phosphoethanolamine (DDPE; Avanti Polar Lipids #850702P) were added as internal standards, followed by incubating for 5 min at room temperature, adding 1 ml each of water and chloroform, centrifuging at 720 g for 5 min. The lower organic phase was collected. To re-extract lipids from the remaining water phase, 1 ml chloroform was added. The combined organic phase was evaporated under a nitrogen stream and the extracted lipids were dissolved in methanol. Total cellular plasmalogens were analyzed by LC-ESI-MS/MS (Abe et al., 2014) using a 4000 Q-TRAP quadrupole linear ion trap hybrid mass spectrometer (AB Sciex) with an ACQUITY UPLC system (Waters). The data were analyzed and quantified using Analyst software (AB Sciex).

Measuring NADPH/NADP+ Levels

NADP/NADPH-Glo™ Assay (G9081, Promega) was used to measure NADPH and NADP+ levels in HeLa cells according to instructions. HeLa cells were seeded into a 24-well plate at approximately 40,000 cells per well at the time the assay was performed. DMEM medium was replaced before beginning experiment and cells were treated with 300 μM H2O2 (Millipore-Sigma, #216763) for 5, 15, 30, and 60 minutes before quickly removing H2O2-treated media. 160 μL of PBS was added to each well before adding an additional 160 μL of PBS with 1% DTAB (Dodecyltrimethylammonium bromide) and 0.2N NaOH to lyse the cells for a final concentration of 0.5% DTAB and 0.1N NaOH in PBS. NADPH standards were diluted in the same concentration of DTAB and NaOH from freshly resuspended NADPH (#N8035, Sigma Aldrich). After lysis, each sample was divided into NADP+ and NADPH fractions. NADP+ samples were used directly, NADPH fractions were diluted by half using 0.5% DTAB, 0.1N NaOH PBS. 50 μL of each fraction for each sample was then added to a 96-well plate. 25 μL of 0.4 HCl was added to the NADP+ samples and all samples were then incubated in the covered plate for 15 minutes at 60°C, and then 10 minutes at room temperature. After incubation, 25 mL of 0.5M Trizma base was added to NADP+ samples and 50 μL of HCl/Trizma solution was added to NADPH samples. 24.5 μL of each sample was then transferred into two wells in a new white 96-well luminometer plate (#3917, Corning Costar). 24.5 μL of the kit NADP/NADPH-Glo detection reagent was then added to each well. After 30 minutes incubation at room temperature, luminescence from the plate was read on a TECAN Spark, using an integration time of 1,000 ms. Data was analyzed according to assay instructions.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were performed using Microsoft Excel and python using the indicated tests and custom scripts where referenced. Lethal fraction scores calculated from live/dead cell counts using previously published formula (Forcina et al., 2017) and Incucyte Zoom and S3 software (Essen). Immunofluorescence image analysis was performed using NIS Elements. Western blots were imaged using an Odyssey CLx and analyzed with Image Studio 5.x. Statistical details of experiments can be found in the figure legends. N represents the number of replicates for each experiment. Randomization, blinding, and sample-size estimation were not applicable in this study.

DATA AND CODE AVAILABILITY

Sequencing data are available at Sequence Read Archive (SRA) accession number SRP235195 under BioProject accession number PRJNA594164.

Highlights.

• Genome-wide shRNA and CRISPR screens identify modifiers of oxidative stress sensitivity

• PEX gene perturbation protects cells via catalase redistribution

• Disruption of the pentose phosphate pathway protects cells by metabolic rewiring