Disruption of mitochondrial quality control genes promotes caspase-resistant cell survival following apoptotic stimuli

In cells undergoing cell-intrinsic apoptosis, mitochondrial outer membrane permeabilization (MOMP) typically marks an irreversible step in the cell death process. However, in some cases, a subpopulation of treated cells can exhibit a sublethal response, termed “minority MOMP.” In this phenomenon, the affected cells survive, despite a low level of caspase activation and subsequent limited activation of the endonuclease caspase-activated DNase (DNA fragmentation factor subunit beta). Consequently, these cells can experience DNA damage, increasing the probability of oncogenesis. However, little is known about the minority MOMP response. To discover genes that affect the MOMP response in individual cells, we conducted an imaging-based phenotypic siRNA screen. We identified multiple candidate genes whose downregulation increased the heterogeneity of MOMP within single cells, among which were genes related to mitochondrial dynamics and mitophagy that participate in the mitochondrial quality control (MQC) system. Furthermore, to test the hypothesis that functional MQC is important for reducing the frequency of minority MOMP, we developed an assay to measure the clonogenic survival of caspase-engaged cells. We found that cells deficient in various MQC genes were indeed prone to aberrant post-MOMP survival. Our data highlight the important role of proteins involved in mitochondrial dynamics and mitophagy in preventing apoptotic dysregulation and oncogenesis.

In cells undergoing cell-intrinsic apoptosis, mitochondrial outer membrane permeabilization (MOMP) typically marks an irreversible step in the cell death process. However, in some cases, a subpopulation of treated cells can exhibit a sublethal response, termed "minority MOMP." In this phenomenon, the affected cells survive, despite a low level of caspase activation and subsequent limited activation of the endonuclease caspaseactivated DNase (DNA fragmentation factor subunit beta). Consequently, these cells can experience DNA damage, increasing the probability of oncogenesis. However, little is known about the minority MOMP response. To discover genes that affect the MOMP response in individual cells, we conducted an imaging-based phenotypic siRNA screen. We identified multiple candidate genes whose downregulation increased the heterogeneity of MOMP within single cells, among which were genes related to mitochondrial dynamics and mitophagy that participate in the mitochondrial quality control (MQC) system. Furthermore, to test the hypothesis that functional MQC is important for reducing the frequency of minority MOMP, we developed an assay to measure the clonogenic survival of caspase-engaged cells. We found that cells deficient in various MQC genes were indeed prone to aberrant post-MOMP survival. Our data highlight the important role of proteins involved in mitochondrial dynamics and mitophagy in preventing apoptotic dysregulation and oncogenesis. Apoptosis is a ubiquitous cellular self-elimination process that is critical for the homeostasis of various cell populations. Dysregulated apoptosis is integral to cancer progression and contributes to multiple diseases, including immune and neurodegenerative disorders. Many cancer therapies rely on the enhanced apoptotic death of tumor cells. Apoptosis frequently involves a "cell-intrinsic" pathway involving mitochondria. The central event in mitochondria-dependent apoptosis is mitochondrial outer membrane permeabilization (MOMP), which is induced by B-cell lymphoma 2 (BCL2)-associated X, apoptosis regulator (BAX) and Bcl-2 homologous antagonist/killer (BAK), the key proapoptotic BCL-2 family proteins (1,2). BAK is constitutively located on the mitochondrial outer membrane (MOM), whereas BAX is mostly soluble in the cytoplasm. When cells receive an apoptotic stress signal, molecules belonging to a subset of the BCL-2 family termed "Bcl-2 homology domain 3 (BH3)-only proteins" activate BAX and BAK. Consequently, BAX translocates to the MOM, and both BAX and BAK integrate into the outer membrane and trigger MOMP by inducing the formation of large membrane pores (3)(4)(5)(6)(7). In concert with these BCL-2 family members, additional MOM proteins facilitate BAX-induced pore formation (4). Antiapoptotic BCL-2 family members (including BCL-2, BCL-extra large, and myeloid cell leukemia-1) inhibit MOMP by sequestering the BH3-only proteins and by antagonizing the pore-forming activity of BAX and BAK.
The apoptotic pores in the MOM allow proteins normally residing in the mitochondrial intermembrane space to escape into the cytoplasm. Certain of these proteins are apoptogenic, including cytochrome c, second mitochondria-derived activator of caspase (SMAC)/direct inhibitor of apoptosis-binding protein with low pI, and OMI/high-temperature requirement factor A1. These proteins induce apoptotic protease-activating factor-1 (APAF-1)-dependent activation of caspase-9 and the effector caspases (most prominently caspase-3), which cleave certain protein substrates that actively promote cell destruction and engulfment (8)(9)(10). Even when caspase activity is blocked, MOMP per se usually leads to cell death, as it compromises energy metabolism in the permeabilized mitochondria (11)(12)(13). Although mitochondrial membrane potential and cellular ATP are maintained for some time after MOMP, cells treated with caspase inhibitors typically exhibit a gradual decline in mitochondrial respiration and a complete loss of clonogenic survival over the course of 2 or 3 days (12).
Recent reports describe a phenomenon, termed "minority MOMP," in which cells exposed to relatively weak cell stressors can evade MOMP-dependent death (14)(15)(16). In these atypical cells, only a fraction of the cell's mitochondria undergoes MOMP. As a result of minority MOMP, small amounts of cytochrome c and SMAC are released into the cytosol, and downstream caspases become activated at sublethal levels. This low-level caspase activation leads to a correspondingly limited activation of the apoptotic caspase-activated DNase (DNA fragmentation factor subunit beta) endonuclease, which enters the cell nucleus and produces a degree of DNA damage. The surviving cells exhibit genome instability and an increased propensity to become oncogenic (17).
The mechanisms controlling the uniformity of MOMP in cells, and hence the frequency of minority MOMP, are unclear. To discover genes that affect MOMP, we conducted a focused high-content siRNA screen that analyzed 1318 genes that had been annotated in databases as having some relationship to mitochondria. A novel aspect of our screen was the identification of genes whose silencing produced a heterogeneous MOMP phenotype in which some mitochondria within the cell were permeabilized, whereas others remained intact. Our screen yielded functionally diverse candidate genes, including a significant group of genes associated with mitophagy, autophagy, and mitochondrial dynamics, the processes constituting the system of mitochondrial quality control (MQC) (18,19). We hypothesized that an increased heterogeneity of MOMP observed in MQC-compromised cells could increase the frequency of minority MOMP (17). If so, downregulation of MQC-related genes would tend to allow cells to survive stresses that produce a degree of caspase activation, as a result of limited MOMP. To test this hypothesis, we developed a singlecell assay to measure the long-term clonogenic cell survival of cells that had exhibited a measurable degree of caspase activation, in response to treatment with a BH3-mimetic drug. Our results confirmed that deficiencies in various MQC proteins enhanced the survival of caspase-engaged cells. We conclude that mitochondrial dynamics and mitophagy play a critical role in limiting the frequency of minority MOMP. Thus, proper functioning of the MQC system can be important to prevent oncogenesis that results from an incomplete execution of the mitochondrial apoptosis program.

Screening strategy and assay development
To identify novel cell-intrinsic regulators of MOMP, we designed a high-content image-based siRNA screen. As we were primarily interested in mitochondrial function, we carried out a focused siRNA screen targeting genes with annotated relationships to mitochondria (Table S1). Our strategy took advantage of a HeLa cell line expressing two fluorescent reporter proteins (described previously by Llambi et al. (20)) that enabled us to simultaneously interrogate both BAX activation and MOMP. In living cells, Venus-BAX is cytoplasmic (diffuse), and OMI-mCherry is localized to mitochondria. When apoptosis is induced, Venus-BAX is translocated to mitochondria and OMI-mCherry is released into the cytoplasm and degraded. Normally, once apoptotic BAX translocation is initiated in a given cell, mitochondrial intermembrane space proteins (including cytochrome c or OMI) are released in a synchronous and rapid fashion (14,20,24). The release of proteins from the mitochondrial interior can only be restricted under special circumstances, for example, cristae junction remodeling by overexpression of a mutant form of the optic atrophy 1 (OPA1) protein (25) or MOMP inhibition by a recently described endolysosome-linked mechanism (26). However, in some cells, the release of intermembrane proteins is not an all-or-none event, and mitochondria within the same cell display heterogeneity in MOMP response (16,17). We hypothesized that downregulation of certain genes could increase MOMP heterogeneity and promote post-MOMP cell survival, even in caspase-proficient cells.
To begin to test our hypothesis, we analyzed punctate mitochondrial distribution of Venus-BAX and OMI-mCherry in cells treated with the apoptosis inducer, etoposide. The caspase inhibitor quinoline-Val-Asp-difluorophenoxy methylketone (Q-VD) was included in these experiments to decrease premature cell loss and detachment. Previous studies have demonstrated that caspase inhibition does not affect the kinetics of the release of intermembrane space proteins (12,24,27). We configured our automated image analysis to quantify three cellular phenotypes: (1) apoptotic cells with mitochondria that have undergone MOMP-containing translocated BAX and lacking intramitochondrial OMI (BAX puncta positive), (2) nonapoptotic cells with intact mitochondria lacking BAX and containing OMI (BAX puncta negative), and (3) atypical cells with mitochondria containing BAX but which have not released OMI (BAX/OMI puncta double-positive cells) (Fig. 1. A and B). Untreated (no etoposide) cells transfected with control siRNA were nonapoptotic and showed normal mitochondrial morphology (Fig. 1, A and E) indicating that our optimized transfection conditions produced minimal toxicity. As expected, most etoposide-treated cells fell into either of the first two categories, whereas BAX/OMI doublepositive cells were minimally present (Fig. 1, A and E). In assay validation experiments, we tested the effect of OPA1 siRNA. OPA1 mediates mitochondrial fusion in conjunction with its cleavage by the mitochondrial protease OMA1 (28)(29)(30) and has multiple roles in apoptotic signaling (22,25,(31)(32)(33). Knockdown of OPA1 protein increased the number of BAX/OMI double-positive cells compared with cells transfected with a nontargeting siRNA (Fig. 1F). Furthermore, highresolution confocal imaging of OMA1 siRNA-treated cells revealed mitochondrial heterogeneity in the MOMP response: in individual BAX/OMI double-positive cells, some mitochondria contained translocated BAX (green puncta) and lost OMI, whereas other mitochondria lacked BAX and retained OMI (red puncta) (Fig. 1D).
A BAX siRNA pool (which targets both endogenous BAX and ectopic Venus-BAX expression) potently inhibited etoposide-induced apoptosis and markedly reduced Venus-BAX fluorescence (Fig. 1, C and F). However, knockdown of BAK did not inhibit apoptosis, indicating that in these etoposide-treated cells, MOMP was predominantly BAX dependent. Transfection efficiency in high-throughput screening experiments was verified using a cell deathinducing transfection marker (siTOX) that consistently produced 90% loss of cells within 2 days post-transfection (not shown). Efficient knockdown of other proteins produced by siRNA pools was confirmed in supporting experiments (Fig. S1A).

The siRNA screen uncovers potential regulators of MOMP
The image-based assay was used to screen 1318 genespecific siRNA pools. We ranked gene scores in three groups corresponding to the cell phenotype categories noted previously: genes whose knockdown (I) increased the percentage of Venus-BAX puncta, (II) decreased the percentage of cells with Venus-BAX puncta, or (III) increased the percentage of BAX/ OMI double-positive cells. When we applied a 10% cutoff in the score distribution range, we identified 200 initial "hits" (with scores approximately threefold to fivefold above or below controls for the categories I and II). Hits in category III were from the top 10% tail of the score distribution (Fig. 1G). For a secondary screen, we chose to reassay 95 of the primary screen hits, based on their effect scores and our judgment concerning their biological interest (we decided not to pursue some "housekeeping" genes). In this secondary screen, we tested the four siRNAs from each siRNA pool individually. To reduce the likelihood of off-target effects, we required at least three of the four individual siRNAs to give concordant results (34). Applying this more stringent criterion yielded a final list of 63 candidate genes (Table S2). Several hits in categories I and II were consistent with previous reports. For example, the effect of siRNA targeting Bcl-2/adenovirus E1B 19-kDa interacting Figure 1. Imaging-based screening assay for regulators of apoptotic MOMP response. A, representative epifluorescence microscopy of untreated and etoposide-treated cells expressing Venus-BAX (green; YFP channel) and OMI-mCherry (red; Texas Red channel). Bottom panel illustrates enlarged cells with indicated phenotypes. Nuclei were stained with Hoechst-33342 (blue; DAPI channel). B, identification of Venus-BAX and OMI-mCherry puncta using Met-aXpress granularity application module. A representative enlarged image of etoposide-treated cells (left panel) and corresponding image segmentations. The granularity module identifies Venus-BAX (middle panel) and OMI-mCherry puncta. Note that cells with BAX "granules" do not contain OMI "granules" and vice versa. Nuclear segmentation settings correctly identify fragmented (apoptotic) and normal size nuclei. C, BAX siRNA inhibits Venus-BAX puncta formation and OMI-mCherry release in etoposide-treated cells. D, a confocal image of live Venus-BAX/OMI-mCherry cells transfected OMA1 siRNA. Arrows indicate cells with heterogeneous MOMP. The scale bars in (A-D) represent 20 μm. E and F, examples of phenotype quantification using granularity application module and high-throughput microscopy. E, quantification of indicated phenotypes in untreated and etoposide-treated cells transfected with a nontargeting (control) siRNA; "nondetermined" phenotype corresponds to a small fraction of nonfluorescent cells. F, effects of indicated siRNAs on the phenotype distribution. Data are mean ± SEM (n = 3 replicate wells). G, sorted assay scores for the primary screen siRNA set in triplicate plates. Numbers indicate the percentage of cells positive for both Venus-BAX and OMI-mCherry foci. Values shown are mean and SEM, n = 3. Hits were identified from the top 10% tail of the BAX/OMI score. BAX, BCL2-associated X, apoptosis regulator; DAPI, 4 0 ,6-diamidino-2-phenylindole; MOMP, mitochondrial outer membrane permeabilization. protein 3 (BNIP3) (category II) is consistent with the known cell death-promoting activity of this protein (35). Acting in the opposite manner, the siRNAs scoring in category I increased the percentage of cells undergoing MOMP. For example, several candidate genes in this group are required for metabolism of the mitochondrial lipid, cardiolipin (phospholipid scramblase 3 [PLSCR3], PRELI domain-containing protein 1 [PRELIDI], and TP53-regulated inhibitor of apoptosis 1 [TRIAP1]). Although cardiolipin is important for BAX pore formation (6,36), in certain paradigms, cardiolipin deficiency potentiates the release of apoptogenic proteins (37,38). In particular, the deficiency of p53-regulated protein TRIAP1 impaired cardiolipin level in mitochondria, compromised bioenergetics, and potentiated cytochrome c release (38). Potential novel regulators of MOMP include metabolic enzymes, long-chain acyl-coenzyme A dehydrogenase, which catalyzes one of the early steps in the circle of mitochondrial betaoxidation of fatty acids, and fumarylacetoacetate hydrolase domain-containing protein 1 with putative oxaloacetate decarboxylase activity in mitochondria. The roles of these proteins in mitochondrial metabolism and cell senescence are emerging (39,40), and their inhibitory effects on MOMP could be of interest for further investigations. In this study, we did not further pursue the hits in categories I and II but focused on the unique phenotype produced by the hits in category III. Of note, in nearly all cells containing both BAX and OMI puncta, BAX and OMI were not colocalized, implying that some mitochondria within a cell had undergone MOMP, whereas others had not (similar to the phenotype shown in Fig. 1D). Perhaps the most striking outcome of our screen is that a majority of hits in category III are related to the MQC system.

The interplay of MQC and MOMP
The concept of MQC is that mitochondrial dynamics (fission and fusion) work in tandem with mitophagy (the autophagic elimination of dysfunctional mitochondrial fragments) to maintain mitochondrial structural and functional integrity. If defective mitochondria cannot be repaired through fusion with functional organelles, they are prone to excessive fragmentation and elimination by mitophagy. It is postulated that asymmetrical fission segregates defective mitochondria, making the mitochondrial population inherently heterogeneous. These isolated small mitochondria are then redirected into a preautophagic pool (41,42). In cells dysfunctional for mitochondrial dynamics or mitophagy, damaged mitochondria would be predicted to accumulate, and the organelles would become abnormally heterogeneous with respect to bioenergetic function and the distribution of proteins involved in MOMP (43,44). Our screen tended to confirm this prediction.

Downregulation of mitophagy-related genes promotes MOMP heterogeneity
Among the hits in category III, both ubiquitin-dependent (e.g., autophagy-related 12 (45)(46)(47)(48) were represented (Table 1). ATG12 is ubiquitin-like protein that is integral for general autophagy and has additional roles in apoptosis and cell survival (49). For example, ATG12 deficiency compromises mitochondrial function and promotes cellular oncogenic transformation (50). One of the top-scoring hits was BLOC1S1/GCN5LI, which is reported to have multiple functions, including the coordinate regulation of mitochondrial biogenesis and mitophagy, through protein acetylation (51). BNIP3L/NIX (originally described as a BCL-2-related proapoptotic protein) and FUNDC1 directly interact with the light chain 3 protein in the autophagosome formation pathway and promote mitophagy facilitated by hypoxic conditions (48,50,52,53).

Other MQC-related hits in category III
Besides well-defined components of mitophagy pathways, our screen yielded multiple candidates that could affect MQC as part of their function. For example, protein tyrosine phosphatase-interacting protein 51 (encoded by the regulator of microtubule dynamics protein 3 [RMDN3] gene) was shown to be important for mitochondria-endoplasmic reticulum (ER) tethering, and therefore, its putative role in MQC and apoptosis could be linked to multiple functions regulated by ER-mitochondrial contacts, such as ER stress-induced Ca 2+ release (54), autophagosome formation, and mitochondrial fission (55, 56).
Another functional group represented by multiple hits in the screen (ATCAY, disrupted-in-schizophrenia 1 [DISC1], trafficking protein kinesin-binding 1, and spermatogenesisassociated 19 [SPATA19]) involves mitochondrial motility on microtubules, in which motor proteins such as kinesin 1 interact with certain mitochondrial membrane proteins, for example, mitochondrial Rho GTPase 1 (57,58). We suspect that these genes appeared as hits in our screen because microtubule-based motility is important for mitochondrial fusion, including the transient fusion event known as "kiss and run" (59,60). Since fusion is a key element of MQC, a loss of mitochondrial motility would be expected to increase the functional heterogeneity of the mitochondrial population within each cell.
Other interesting candidates include multifunctional E3 ubiquitin ligases ring finger protein 5 (RNF5) and MARCH5. Mitochondrial targets and pathways regulated by ring finger protein 5 remain to be identified. MARCH5 reportedly regulates activities of mitofusin-1 (MFN1) and dynamin-related protein 1 (DRP1) or its outer membrane receptor mitochondrial fission factor, the key mediators of mitochondrial fusion and fission, respectively (61-63). However, DRP1 and MFN themselves did not show significant effects in our screen, perhaps because of the redundancy. Among multifunctional proteins that may influence mitochondrial dynamics indirectly, leucine zipper-EF-hand containing transmembrane protein 1 (LETM1) is the mitochondrial Ca 2+ /H + antiporter with pleiotropic effects on ATP production, mitophagy, cristae structure, and mitochondria morphology (64). In particular, mitochondria lacking LETM1 are prone to undergo DRP1independent fission (65). In our screen, LETM1 knockdown also increased mitochondrial fragmentation (not shown) as well as MOMP heterogeneity. Other candidate genes in category III are highlighted in Table 1. Overall, our screen suggests that a deficiency in MQC disturbs the normal apoptotic response and, in particular, promotes heterogeneous MOMP.

An assay to measure clonogenic cell survival despite caspase activation
Based on the number of hits our screen being involved in MQC, we hypothesized that MQC-dependent mitochondrial heterogeneity could result in an increased frequency of "minority MOMP." In this scenario, when a minor fraction of mitochondria undergoes MOMP, the resulting sublethal caspase activation leads to DNA damage and promotes oncogenesis (17). To test our hypothesis, we developed a clonogenic survival assay based on that described by Ichim et al. (17) (Fig. 2). In this assay, we induced apoptosis with a BH3-mimetic compound, ABT-737 or ABT-199, incubated the cells with a fluorescent caspase-reporter compound, and used fluorescence-activated cell sorting to sort out cells that exhibited caspase activation. We then plated the cells and, after several days in culture, quantified the percentage of cells that survived and formed colonies.
We chose the BH mimetics for this assay, instead of etoposide used in the screen, for their well-defined molecular targets in MOMP (66,67) and no known toxic effects on other apoptotic pathways. Ichim et al. ((17) ; Fig. 1E) demonstrated that these mimetics induced minority MOMP in HeLa and U2OS cells. Therefore, we reasoned that these reagents would cause heterogenous MOMP as in the screen.
In order to validate the faithfulness of the caspase reporter dye, we compared the staining in APAF-1 CRISPR KO U2OS cells (obtained from Dr Stephen Tait, University of Glasgow) to that of their parental WT control cells. APAF-1 is required to activate caspase-9 and subsequently the effector caspase-3 and caspase-7, following the MOMP-dependent release of cytochrome c, SMAC, and OMI. As shown in Figure 3, only a trace amount of cells were stained by the dye in the APAF-1 KO cell population treated with a BH3 mimetic, ABT-737. In contrast, the WT cells showed a population of cells stained with the dye, indicating that the caspase reporter was faithful in detecting caspase activation in the cells. Based on the study by Ichim et al. (17), it is reasonable to assume that survival of cells from this caspase-activated population indicates the occurrence of minority MOMP. Given the challenging nature of the assay because of the sensitivity to temperature fluctuations, the cell density as well as the distribution in the wells (see also in Experimental procedures section), we only collected the data from the experiments flawlessly performed. Therefore, the data points for statistical analysis shown in Figures 4 and 5 were based only on two to three independent experiments. Further independent verification of these results will provide higher confidence in our findings.

MQC deficiency promotes minority MOMP
To test further whether MQC is important to mitigate minority MOMP, we used the clonogenic assay described previously to investigate the effects of deleting selected candidate genes from our screen as well as other known MQC factors. We generated CRISPR cells (U2OS) with perturbed expression of ATG12, RMDN3 (protein tyrosine phosphataseinteracting protein 51), and BNIP3L. Knockdown of each of these genes produced MOMP heterogeneity in our imaging assay (Table 1). To obtain stable KO cells, we generated CRISPR-edited cell pools and derived clonal populations from single cells (as described in the Experimental procedures section). Validation of the CRISPR-based gene-editing efficiency is shown in Fig. S2. As shown in Figure 4, we found that cells depleted of either RMDN3 or ATG12 showed significantly higher post-MOMP survival than WT cells. However, BNIP3L KO did not show an increase in minority MOMP. Reasons for this are unknown but could be related to other apoptosisrelated activities of BNIP3L.
To examine further the hypothesis that compromised MQC promotes minority MOMP, we performed experiments using previously derived cells in which genes known to be directly involved in mitochondrial dynamics or mitophagy had been deleted, namely MFN1 and MFN2 double KO, OPA1 KO, DRP1 KO mouse embryonic fibroblasts (MEFs), and "penta-KO" HeLa cells that had been shown to be deficient in mitophagy because of the deletion of five mitophagy receptor genes (21). As shown in Figure 5, all these MQC-deficient cells exhibited an enhancement of cell survival in our clonogenic assay, compared with their matched controls. The effect in DRP1 KO cells was not statistically significant; in this case, a weaker effect could reflect the slowing of cell proliferation that was reported with DRP1 deletion (68). Taken together, the loss of these genes, important for MQC, promoted the proliferative survival of caspase-engaged cells, again implying an increased frequency of minority MOMP.

Conclusion
Taken together, our data allow us to conclude that the MQC system is critical for mitigating the phenomenon of minority MOMP. This result has implications for oncogenesis; even a small increase in cell viability resulting from sublethal caspase activation could potentially raise the frequency of oncogenic cell transformation (69,70). Therefore, treatments developed to limit minority MOMP could improve the cytotoxic effects of anticancer treatments. For example, proteins that are known to inhibit mitophagy, such as seven in absentia homolog 3 (21), are potential therapeutic targets. Also, we predict, based on our observation that placing sorted caspase-engaged cells on ice eliminated their ability to survive clonogenicity (see the Experimental procedures section), that cryotherapy could prevent minority MOMP when used in combination with proapoptotic cancer therapeutics such as BH3-mimetic drugs. In conclusion, our study provides new insights into the mechanisms of aberrant apoptosis that are important for developing therapeutic strategies.

Cell lines
HeLa cells stably expressing Venus-BAX and OMI-mCherry cells (20) were constructed by the laboratory of Douglas Green (Department of Immunology, St Jude Children Hospital).  Youle (the National Institutes of Health) (21). Unless indicated otherwise, cells were maintained in Dulbecco's modified Eagle's medium (DMEM; Life Technologies) containing 10% fetal bovine serum (FBS) (GeminiBio) and 100 units/ml penicillin/ streptomycin at 37 C with 5% CO 2 .

siRNA screen: cell transfections and treatments
For the primary screen, gene-specific siRNA pools targeting 1318 mitochondria-annotated genes (Table S1) were cherrypicked from a Dharmacon genome-wide siRNA library (siGENOME). The list of genes was generated based on a "mitochondria/mitochondrial membranes" queries in the National Institutes of Health compiled databases (http://www. ncbi.nlm.nih.gov/gene). Pools of four individual gene-specific siRNAs and two nontargeting siRNAs were arrayed in 384well master plates. Each plate also contained Dharmacon siGENOME TOX and siGlo Red oligonucleotides as transfection indicators. In addition, an siRNA pool targeting BAX was used as a positive control for the inhibition of apoptosis and BAX puncta formation. For reverse transfection, 4.4 μl siRNA picked from each 1 μM stock siRNA solution in the master plate was mixed with 30.6 μl prediluted Lipofectamine RNAiMAX transfection reagent (Life Technology). RNAiMAX reagent was diluted 47 times in Gibco Opti-MEM reduced serum medium. The transfection mixtures (35 μl/well) were incubated for 20 min at room temperature in a 384-well mixing plate. During the incubation period, Venus-BAX/ OMI-mCherry HeLa cells were harvested by trypsinization and resuspended in antibiotic-free DMEM with 10% FBS at 30,000 cells per ml. After incubation, lipid-siRNA complexes were dispensed into triplicate tissue/culture Costar black-wall clear bottom 384-well plates (10 μl/well); siRNAs targeting the same gene were separated in the different plates. Plate handling, transfection reagent/siRNA mixing, and dispensing were performed with Hamilton Star Automated Liquid Handler contained within a Baker BioPROTECT class II biosafety cabinet. Cells were added at 40 μl (1200 cells) per well on top of the transfection mixture; final concentration of siRNA was 25 nM in the 50 μl per well volume. The plates were centrifuged at low speed (300g) for 1 min and transferred to a humidified CO 2 incubator (37 C). To minimize plate edge effects, the wells in two rows and columns at the plate edges were not used for transfections and contained only medium with cells. After 48 h, the medium was aspirated and replaced with fresh culture medium containing 400 μM etoposide (Sigma-Aldrich) and 20 μM of the caspase inhibitor Q-VD (Q-VD-OPH; SM Biochemicals LLC). Control wells were left untreated (no etoposide). By the time of etoposide treatment, most cells transfected with TOX siRNA had already shown morphological changes consistent with cell death, indicating efficient transfection. Following 24 h of etoposide treatment, the medium was removed, and cells were fixed with 0.5% glutaraldehyde as described previously (22). After two washes with PBS, cells were stained with Hoechst-33342 (Molecular Probes) diluted 1000 times in PBS. Plates were then washed twice with PBS, filled with 50 μl PBS per well, sealed, and stored at 4 C. For the secondary screen, four individual siGENOME siRNAs per gene were obtained separately from Dharmacon to target 95 gene candidates chosen from the primary screen. Individual siRNAs were arrayed in replicate 384-well plates, and the experiments were conducted as described previously.

High-throughput image acquisition and analysis
Image collection and processing were done using the Molecular Devices MetaXpress High Content Image Acquisition platform. Images of the cells were acquired in an Image Xpress Micro device at 20× magnification with the following filters: 4 0 ,6-diamidino-2-phenylindole (5060B), for Hoechst nuclear staining; YFP (2427A), for Venus-BAX fluorescence; and Texas Red (4040B), for OMI-mCherry fluorescence (numbers indicate the Semrock part number for the filters). Images were collected from 16 sites clustered in the center of the well. In our assay development, the punctate patterns of Venus-BAX translocation and OMI-mCherry retention in mitochondria were analyzed using the granularity module in MetaXpress (version 5.1) software (Molecular Devices). Individual cells were identified based on nuclear segmentation (4 0 ,6diamidino-2-phenylindole channel images). Z 0 factor for the BAX-positive or BAX-negative phenotype assay was calculated as 0.85 based on etoposide-treated and untreated conditions as positive and negative controls, respectively.
High-throughput analysis of primary and secondary screen data was done with a customized algorithm (created as a "journal macro" in MetaXpress) developed for automated phenotype quantification. We configured image analysis to count the percentage of cells that satisfied either of two different criteria: (1) a given cell contains Venus-BAX granules of diameters within the expected diameter range, with total intensity above a certain threshold or (2) a given cell contains both suprathreshold Venus-BAX and OMI-mCherry granules (not necessarily colocalized) within the appropriate diameter range. The thresholds were adjusted for stringency to limit false-positive scores from background fluorescence and debris. Plate-to-plate variability was minimal for BAX foci and acceptably low for cells double positive for BAX and OMI granules.

Confocal microscopy
Confocal images of Venus-BAX/OMI-mCherry cells were acquired with a 60× oil immersion objective on an Olympus FluoView FV10i automated confocal laser scanning microscope (Olympus Scientific Solutions America Corp).

CRISPR/Cas9-mediated gene depletion
CRISPR experiments were performed using modified synthetic single-guide RNAs (sgRNAs) from Synthego. Target sequences for guide RNAs were selected with the Synthego CRISPR design tool. RNA oligonucleotides were reconstituted in 10 mM Tris-HCl, 1 mM EDTA, pH 8.0 buffer according to recommendations from the manufacturer. Ribonucleoprotein complexes were formed from sgRNA and recombinant Cas9 two nuclear localization signal proteins (New England Biolabs) mixed at an sgRNA to Cas9 ratio of 4.5:1. After 20 to 30 min of incubation at room temperature, assembled ribonucleoproteins were delivered into cells by electroporation using a ThermoNeon device and 10 μM tips (Thermo Fisher Scientific). For one sample (2 × 10 5 cells), 3 μl of sgRNA (from 30 mM stock solution) were mixed with 1.5 μl Cas9 protein (from 20 μM stock solution). U2OS cells were electroporated at 1230 v/10 ms/4 (pulse voltage, width, and pulse number) settings. Immediately after electroporation, cells were transferred to 12-well plate with prewarmed antibiotic-free DMEM with 10% FBS. After 2 to 3 days of incubation, a portion of the control (unedited) and CRISPR cells were harvested for genomic DNA isolation using a QIAGEN genomic DNA purification kit. The remaining cells were left in culture for further propagation and clonal selection. Genomic DNA concentrations were measured using a NanoDrop instrument (Thermo Fisher Scientific). Primer design and PCR amplification of the edited region were done according to Synthego recommendations. Sanger sequencing of PCR amplicons was performed at Genewiz or Eton Bioscience; the resulting DNA sequencing chromatograms were analyzed using the Inference of CRISPR Edits algorithm (Synthego). Inference of CRISPR Edits analysis of CRISPR-edited genomic regions typically demonstrated at least 70% editing efficiency (i.e., 70% KO cells in the pool); sgRNA sequences used are shown in Fig. S2. For clone selection, CRISPR cells were expanded for 2 to 3 additional days, harvested, and subjected to sorting into 96-well plates using BD FACSAria-3 or FACSAria-4 Fusion instruments; typically, one to four single cells were dispensed into one well. After clonal expansion, cells were analyzed for KO efficiency as described previously. For each gene of interest, two clones with verified KO were combined for use in further experiments.
Assay for clonogenic survival after caspase activation via MOMP (outlined in Fig. 2)

Cells (CRISPR KO U2OS cells and MEFs [KO and matched
WT]) were plated out at 3 × 10 5 per well containing 1 ml media in 6-well plates the day before the experiment, two wells per condition. The cells were then treated with ABT-199 or ABT-737 at 2.5, 5, or 10 μM for 5 h. In the last 30 min of incubation, a caspase reporter dye, CellEvent Caspase3/7 Green (Life Technology; R37111), was added at 30 μl per well. Cells were harvested with trypsin/EDTA and washed with medium and PBS. Finally, cells were resuspended in 300 μl of sorting buffer consisting of 1% bovine serum albumin in PBS containing 20 μM of Q-VD, where Q-VD was included to stop the caspase reaction. We noticed that, when these caspaseactivated cells were left at room temperature or 4 C for more than 30 to 40 min, their ability to survive was compromised. Therefore, care was taken to sort the cells immediately after harvesting. To minimize artifacts from sample handling delay, we operated two cell sorters (FACSAria; BD) simultaneously for WT and gene-defective cells. Also, the sorting chamber and the plate holder were kept at 37 C during the run to prevent temperature shock to the cells. Six hundred cells within the population gated for green fluorescence were sorted in duplicate into a 48-well plate containing 300 μl of conditioned medium per well. The cells were grown for 8 to 12 days, during which an additional 500 μl of conditioned medium was added at days 4 to 5. Colonies were then stained with 6% glutaraldehyde containing 0.5% crystal violet. Nontreated cells distributed in the no-green gate were sorted as aforementioned and used as a control representing 100% growth. Because individual cell colonies tended to merge over time, we measured the total area occupied by colonies, using the ImageJ macros developed by Guzman et al. (23). The percentage of surviving caspase-engaged cells in the total population was normalized against the 100% growth control. The data are summarized from two to three independent experiments, as indicated. p Values were calculated using the mean and the standard deviation in each set of KO and WT cells with two-way ANOVA analysis using Prism 7 (GraphPad Software, Inc).

Data availability
All representative data are contained within the article.
Supporting information-This article contains supporting information.
Funding and additional information-This work was supported by R21CA216304 and a SPARK award from the La Jolla Institute for Immunology Board of Directors (to D. D. N.).
Conflict of interest-The authors declare that they have no conflicts of interest with the contents of this article.