Abstract

   Degradation of defective mitochondria is an essential process to
   maintain cellular homeostasis and it is strictly regulated by the
   ubiquitin-proteasome system (UPS) and lysosomal activities. Here, using
   genome-wide CRISPR and small interference RNA screens, we identified a
   critical contribution of the lysosomal system in controlling aberrant
   induction of apoptosis following mitochondrial damage. After treatment
   with mitochondrial toxins, activation of the PINK1-Parkin axis
   triggered a BAX- and BAK-independent process of cytochrome c release
   from mitochondria followed by APAF1 and caspase 9–dependent apoptosis.
   This phenomenon was mediated by UPS-dependent outer mitochondrial
   membrane (OMM) degradation and was reversed using proteasome
   inhibitors. We found that the subsequent recruitment of the autophagy
   machinery to the OMM protected cells from apoptosis, mediating the
   lysosomal degradation of dysfunctional mitochondria. Our results
   underscore a major role of the autophagy machinery in counteracting
   aberrant noncanonical apoptosis and identified autophagy receptors as
   key elements in the regulation of this process.
     __________________________________________________________________

   Mitophagy counteracts Parkin-mediated apoptosis.

INTRODUCTION

   Parkin is an E3 ubiquitin ligase that participates in the processes of
   proteasomal and lysosomal degradation ([42]1). In its autoinhibited
   inactive state, Parkin is located in the cytosol and, upon
   mitochondrial impairment, translocates onto mitochondria ([43]2), where
   it shows exquisite specificity for proteins on the outer mitochondrial
   membrane (OMM) ([44]3); ubiquitination of which is considered to be the
   initial step of the complex process termed mitophagy ([45]3). This
   process mediates the degradation of dysfunctional mitochondria to
   maintain cellular homeostasis. During mitophagy, the protein
   PTEN-Induced Kinase 1 (PINK1) is required for Parkin activation. Unlike
   Parkin, PINK1 localizes to the OMM, where it is imported and degraded
   in healthy mitochondria by mitochondrial proteases ([46]4) and released
   into the cytosol for degradation by the ubiquitin-proteasome system
   (UPS) ([47]5). Mitochondrial membrane depolarization stabilizes PINK1
   on the OMM ([48]4, [49]6) in association with the translocase of the
   OMM complex ([50]7). PINK1 phosphorylates ubiquitin ([51]8) and the
   Parkin ubiquitin-like domain ([52]9). This phosphorylation event
   results in the full activation state of Parkin and its translocation to
   the OMM ([53]3). Here, the E3 ubiquitin ligase function of Parkin
   ubiquitinates substrates on the OMM, amplifying the signal that
   recruits both the UPS ([54]10–[55]13) and a small class of cytosolic
   proteins, the autophagy receptors, to the damaged mitochondrion. These
   receptors contain a microtubule-associated protein 1A/1B light chain 3
   (LC3)–interacting region (LIR) domain that recruits members of the
   autophagy related 8 (ATG8) family ([56]14), finally giving rise to the
   formation of autophagosomes around mitochondrial remnants and their
   fusion with lysosomes ([57]11–[58]13).

   Mutations in both PRKN and PINK1, encoding Parkin and PINK1,
   respectively, are associated with genetic forms of early-onset
   Parkinson’s disease (PD) ([59]15, [60]16). Because both PINK1 and
   Parkin are required for mitochondrial homeostasis ([61]17, [62]18) and
   regulate mitophagy ([63]19, [64]20), it is thought that dysfunction in
   mitophagy can cause PD, suggesting that the accumulation of
   dysfunctional mitochondria may induce loss of neuronal function and
   eventually cell death ([65]20). PRKN mutations and loss of
   heterozygosity have also been observed in different types of cancer
   ([66]21). In a landscape of somatic copy number alterations across
   human cancer, Parkin is one of the most deleted genes, therefore
   considered a tumor suppressor.

   Other possible mechanisms, such as escape from cell death, can drive
   the process of tumorigenesis ([67]22). A specific link between PINK1
   and Parkin activity to promote cell death has been suggested by several
   studies ([68]23–[69]26). A type of autophagy-dependent cell death that
   could explain the onset of cell death in the context of aberrant
   intracellular catabolic processes such as mitophagy has been previously
   described ([70]27, [71]28), but the mechanism of action remains
   elusive, and the interpretation of this phenomenon is controversial
   ([72]29).

   We therefore sought to reassess the role of Parkin in cell death. Using
   in vitro and in vivo models as well as two different whole-genome
   screening approaches, we report that Parkin expression sensitized cells
   to die and that this process required the kinase activity of PINK1 and
   the degradation activity of the UPS. This process of cell death was
   independent of the B cell lymphoma 2 (Bcl2) family members but involved
   caspase activation mediated by assembly of the apoptosome. We found
   that Parkin-dependent cytochrome c (cyt c) release from the
   intermembrane space was mediated by UPS activity on the OMM and that
   lysosomal degradation of mitochondrial remnants was protected from cell
   death. Last, we identified the recognition of ubiquitinated OMM
   proteins by autophagy receptors as a cell fate decision point
   downstream the activation of the PINK1-Parkin axis.

RESULTS

Parkin sensitizes cells to apoptosis in a PINK1-dependent manner

   To ask whether Parkin plays a role in regulating cell death,
   immortalized mouse embryonic fibroblasts (MEFs) stably expressing
   yellow fluorescent protein–Parkin (YFP-Parkin MEFs) (fig. S1A) were
   treated with chemical compounds that induce mitochondrial
   depolarization, e.g., valinomycin or the combination of antimycin A and
   oligomycin. As expected, both treatments rapidly dissipated the
   mitochondrial transmembrane potential, as detected by the
   potentiometric dye tetramethylrhodamine ethyl ester perchlorate (TMRE)
   (fig. S1B), and increased mitochondrial reactive oxygen species
   production, as assessed using the Mitochondrial Superoxide Indicator
   (MitoSOX) (fig. S1C). Consistent with previous reports ([73]2, [74]6,
   [75]30), chemically induced mitochondrial uncoupling triggered
   mitophagy in YFP-Parkin MEFs, as depicted by the pH-sensitive
   mitochondrial probe mitochondrial mKeima (mt-mKeima) that visualizes
   the delivery of impaired mitochondria to lysosomes (fig. S1, D and E).
   Treatment of YFP-Parkin MEFs with these mitochondrial agents resulted
   in cell death as assessed by propidium iodide (PI) positivity
   ([76]Fig. 1, A and B). Cotreatment with the pan-caspase inhibitor
   q-VD-OPh (qVD) delayed cell death, suggesting that cell death occurred
   by apoptosis, a conclusion supported by the typical apoptotic
   morphology of YFP-Parkin MEFs treated with valinomycin ([77]Fig. 1, A
   and B, and fig. S1F). As controls, compounds that induce intrinsic
   apoptosis, such as actinomycin D, thapsigargin, and tunicamycin were
   able to induce cell death independently of the expression of Parkin
   ([78]Fig. 1A). Similar observations were made using the human colon
   cancer cell line HCT116 (fig. S1, A and G).

Fig. 1. Parkin expression sensitizes cells to apoptosis and requires PINK1.

   [79]Fig. 1.
   [80]Open in a new tab

   (A) Cell death quantification by IncuCyte of wild-type (WT) or
   YFP-Parkin MEFs after 24 hours of treatment with 5 μM valinomycin,
   additional mitochondrial toxins (5 μM antimycin A plus 1 μM
   oligomycin), or inducers of intrinsic apoptosis (1 μM actinomycin D, 1
   μM thapsigargin, and 1 μM tunicamycin) in combination or not with 40 μM
   qVD. (B) Representative IncuCyte images showing PI positivity (red) and
   cell death morphology (black arrows) in WT or YFP-Parkin MEFs, related
   to (A). Scale bar, 70 μm. (C) Clonogenic assay of WT or YFP-Parkin MEFs
   after 2 hours of treatment with 600 nM valinomycin. (D) Results of
   CRISPR screen indicating single guide RNA (sgRNA) enrichment. (E)
   Clonogenic assay of YFP-Parkin MEFs expressing individual or the
   combination of three sgRNA, targeting Pink1 after 15 min of treatment
   with 600 nM valinomycin. (F) Cell death quantification by IncuCyte of
   YFP-Parkin HCT116 with a doxycycline (Dox)–inducible active form of
   PINK1 (acPINK1) after 24 hours of treatment with 2.5 μM valinomycin,
   BH3 mimetics (5 μM ABT737 plus 5 μM [81]S63845), or Dox (1 μg/ml), with
   or without 40 μM qVD. Error bars in (A) and (F) represent the SD of the
   means of triplicate samples. Data are representative of n = 4 (A), (B),
   and (F) or n = 3 (C) and (E) independent experiments. *P < 0.05; ***P <
   0.001; n.s., not significant.

   To gain additional insights into the effects exerted by valinomycin
   treatment, we performed a clonogenic survival assay with MEFs
   expressing YFP-Parkin. Notably, increasing concentrations of
   valinomycin caused progressive loss of clonogenic survival only in
   YFP-Parkin MEFs ([82]Fig. 1C). We then took advantage of this
   observation to perform a genome-wide CRISPR-Cas9 knockout (KO) screen
   to identify genes that, when ablated, could restore clonogenic survival
   under our experimental conditions (fig. S1H). Using two different
   concentrations of valinomycin and three rounds of treatment, we
   enriched for pools of cells that significantly expressed multiple
   single guide RNAs (sgRNAs) targeting the same gene. We found and
   validated that Pink1 was the only gene whose absence could restore the
   clonogenic survival of the YFP-Parkin MEFs ([83]Fig. 1, D and E; fig.
   S1, I and J; and table S1).

   To investigate how PINK1 participates in this Parkin-mediated form of
   cell death, we developed an inducible in vitro system to express a
   stable, mitochondrial-targeted, and active form of PINK1 (acPINK1) by
   which Parkin translocation to the OMM is achieved without the use of
   any mitochondrial uncoupler ([84]6). Doxycycline (Dox) induction of
   acPINK1 induced apoptosis of YFP-Parkin HCT116 cells, as depicted by PI
   positivity and poly(adenosine diphosphate–ribose) polymerase 1 (PARP)
   cleavage, which was inhibited by qVD ([85]Fig. 1F and fig. S1K).
   Moreover, we confirmed a previous report ([86]6) showing that
   expression of acPINK1 triggered mitophagy, as depicted here by the
   degradation of the mitochondrial protein cyt c oxidase subunit 5B (COX
   Vb), and this process was not inhibited by qVD. These results indicate
   that Parkin expression sensitizes cells to apoptosis upon mitochondrial
   depolarization, and PINK1 is both necessary and sufficient to trigger
   the cascade of events leading to cell death.

Parkin-mediated apoptosis is a BAX- and BAK-independent mechanism

   To further characterize the observed Parkin-mediated form of apoptosis,
   we ablated in YFP-Parkin MEFs the apoptotic protease activating factor
   1 (APAF1), which is the core component of the apoptosome required for
   the activation of mitochondrial apoptosis (fig. S2A) ([87]31). As
   expected, cells lacking APAF1 were protected from cell death, induced
   upon treatment with mitochondrial uncouplers, a combination of BH3
   mimetics (ABT737 and [88]S63845), or actinomycin D. In contrast, we
   observed that PINK1 KO cells were only protected from death induced by
   mitochondrial uncouplers ([89]Fig. 2A). However, while the absence of
   PINK1 restored clonogenic survival in YFP-Parkin MEFs treated with
   valinomycin, the absence of APAF1 did not ([90]Fig. 2B). Moreover, in
   the same cells treated with BH3 mimetics, the absence of PINK1 or APAF1
   could not restore clonogenic survival. This suggests that activation of
   the PINK1-Parkin pathway might trigger a mitochondrial catastrophe that
   ultimately leads to a caspase-independent form of cell death ([91]32).

Fig. 2. Parkin-mediated apoptosis is BAX/BAK-independent.

   [92]Fig. 2.
   [93]Open in a new tab

   (A) Cell death quantification by IncuCyte of YFP-Parkin MEFs parental,
   sgRNA PINK1 KO (sgPINK1 KO), and sgAPAF1 KO after 24 hours of treatment
   with 5 μM valinomycin, 5 μM antimycin A plus 1 μM oligomycin
   (Antim./Oligom.), or inducers of intrinsic apoptosis (1 μM actinomycin
   D and 5 μM ABT737 plus 5 μM [94]S63845). (B) Clonogenic survival assay
   of YFP-Parkin MEFs parental, sgPINK1 KO, and sgAPAF1 KO after 15 min of
   treatment with 600 nM valinomycin for 2 hours with BH3 mimetics (5 μM
   ABT737 plus 5 μM [95]S63845). (C) Cell death quantification by IncuCyte
   of Bax^−/− Bak^−/− MEFs with or without YFP-Parkin after 24 hours of
   treatment as in (A) and (B) ± 40 μM qVD. (D) Clonogenic survival assay
   of WT or Bax^−/− Bak^−/− MEFs expressing YFP-Parkin, after 20 min of
   treatment as in (B). (E) MOMP quantification (see Materials and
   Methods) in Bax^−/− Bak^−/− YFP-Parkin MEFs after 6 hours of treatment
   as in (B). (F) Cell death quantification by IncuCyte of the indicated
   Bax^−/− Bak^−/− YFP-Parkin MEFs after 24 hours of treatment as in (A)
   and (B). (G) Cell death quantification by IncuCyte of Bax^−/− Bak^−/−
   YFP-Parkin MEFs expressing the indicated proteins after 24 hours of
   treatment with 2 μM valinomycin or BH3 mimetics (5 μM). (H) Flank tumor
   growth (left) and volume after 16 weeks (right) following the injection
   of nonobese diabetic scid gamma mice with Bax^−/− Bak^−/− MEFs with or
   without YFP-Parkin and treated peritumorally with vehicle or
   valinomycin (0.4 mg/kg) at days 4 and 8 after cells injection. (I)
   Survival curves of mice related to (H). Data in (A), (C), (F), and (G)
   represent means ± SD of triplicate samples. Data are representative of
   n = 4 (A) and (C), n = 2 (F) and (G), or n = 3 (B), (D), (H), and (I)
   independent experiments. DMSO, dimethyl sulfoxide. ***P < 0.001; ****P
   < 0.0001; n.s., not significant.

   To further explore this phenomenon, we generated cells lacking the two
   major proapoptotic Bcl2 effector proteins, the BCL2 associated X
   apoptosis regulator (BAX) and the BCL2 antagonist/killer 1 (BAK), which
   are known to be required for mitochondrial outer membrane
   permeabilization (MOMP), leading to caspase activation and apoptosis
   ([96]33). As expected, YFP-Parkin MEFs or YFP-Parkin HCT116 lacking BAX
   and BAK (Bax^−/− Bak^−/−) were completely resistant to cell death
   induced by BH3 mimetics and other conventional inducers of apoptosis
   ([97]Fig. 2C and figs. S1A and S2B). They remained, however, fully
   susceptible to cell death induced by raptinal, which is known to bypass
   BAX and BAK activation ([98]34), and by valinomycin or by the
   combination of antimycin A and oligomycin, suggesting that BAX and BAK
   are dispensable for this Parkin-dependent cell death ([99]Fig. 2C and
   figs. S1A and S2B). In line with this observation, Bax^−/− Bak^−/−
   YFP-Parkin MEFs exhibited clonogenic survival when treated with a
   combination of BH3 mimetics but not after treatment with valinomycin
   ([100]Fig. 2D). Furthermore, the absence of BAX and BAK did not
   influence the capacity of the cell to undergo mitophagy (fig. S2C).

   To further characterize this Parkin-dependent cell death, we assessed
   MOMP by monitoring mitochondrial cyt c release by flow cytometry
   analysis of digitonin-permeabilized cells ([101]35). Bax^−/− Bak^−/−
   YFP-Parkin MEFs treated with valinomycin showed cyt c release from the
   mitochondria, which, as expected, was not detectable after BH3 mimetics
   treatment ([102]Fig. 2E). In addition, in the same cells, while this
   form of cell death was delayed by the CRISPR deletion of either Apaf1
   or caspase 9 (Casp9), it was not affected by the enforced expression of
   the antiapoptotic Bcl2 proteins—BCL2, the BCL2-like 1 (BCL-xL), or the
   myeloid cell leukemia sequence 1 (MCL1) ([103]Fig. 2, F and G, and fig.
   S2, D to F). Together, these results imply that the apoptosis-inducing
   activity of valinomycin in cells expressing Parkin is mediated
   independently of the activity of the proapoptotic Bcl2 family members
   BAX and BAK but in a manner that required cyt c release into the
   cytosol and subsequent APAF1 activation.

   Last, because the induction of cell death by valinomycin has been
   previously associated with antitumor activity both in vitro and in vivo
   ([104]36–[105]38), we sought to determine whether the observed
   Parkin-mediated apoptosis, in the absence of BAX and BAK, could engage
   tumor control in vivo. To this end, we implanted Bax^−/− Bak^−/− MEFs
   expressing or not YFP-Parkin into the flanks of nonobese diabetic scid
   gamma (NSG) immunodeficient mice and monitored tumor growth and
   survival. Notably, only the Parkin-expressing tumors were sensitive to
   valinomycin treatment that led to reduced tumor volume and improved
   survival ([106]Fig. 2, H and I), consistent with our in vitro
   observations.

A genome-wide small interference RNA screen identifies genes regulating
Parkin-dependent apoptosis

   To investigate the molecular events that lead to the activation of
   Parkin-dependent apoptosis, we conducted a genome-wide small
   interference RNA (siRNA) screen in valinomycin-treated Bax^−/− Bak^−/−
   YFP-Parkin MEFs and measured Casp3/7 activity coupled to a high-content
   optical imaging system (fig. S3A). We identified genes whose silencing
   either inhibited or increased Parkin-dependent caspase activation
   (named “Class 1” and “Class 2,” respectively) ([107]Fig. 3A; fig. S3, B
   and C; and table S2). Among the Class 1 members, we identified several
   genes known to regulate apoptosis, such as Apaf1, cyt c somatic (Cycs;
   encoding cyt c), Holocytochrome C Synthase (Hccs) ([108]39), and Casp3
   ([109]31, [110]40), as well as genes previously characterized as
   regulators of PINK1 stability, such as Tomm7 and translocase of inner
   mitochondrial membrane 23 (Timm23) ([111]7). We also identified RAB10,
   member RAt Sarcoma virus (RAS) oncogene family (Rab10), a small
   guanosine triphosphatase (GTPase) that accumulates on mitochondria
   after depolarization and regulates mitophagy ([112]41); ryanodine
   receptor 3 (Ryr3), which has been previously associated with the
   process of autophagic cell death ([113]42); dynein light chain
   roadblock-type 1 (Dynlrb1), which is part of the cytoplasmic dynein
   complex and regulates traffic-induced degradation response for
   secretion ([114]43, [115]44); and ubiquitin specific peptidase 5
   (Usp5), a member of the deubiquitinating enzymes ([116]45). Class 2
   members that increased caspase activation are discussed below.

Fig. 3. Genes regulating Parkin-dependent apoptosis and the role of
proteasomal activity in mediating OMM breaks and cyt c release in the
cytosol.

   [117]Fig. 3.
   [118]Open in a new tab

   (A) Relative inhibition of Casp3/7 activation (RICA) in Bax^−/− Bak^−/−
   YFP-Parkin MEFs subjected to a genome-wide siRNA screen after 6 hours
   of treatment with 500 nM valinomycin (blue dashed line, RICA of
   siPINK1; red dashed line, RICA NT siRNA). Blue dots (Class 1) indicate
   genes that behave similarly to Pink1 (Z > −2.0). Red dots (Class 2)
   indicate genes that increase Casp3/7 activity when compared to NT
   siRNA. (B) GSEA analysis showing significant enrichment of cellular
   processes containing candidate genes from Class 1. (C) Percentage of
   Casp3/7 activity in Bax^−/− Bak^−/− YFP-Parkin MEFs as relative light
   units (RLU) normalized to NT siRNA (top) and percentage of mitophagy by
   flow cytometry analysis (bottom) after 48 hours of siRNA silencing of
   the indicated proteins and 9 hours of treatment with 1 μM valinomycin.
   (D) Cell death quantification by IncuCyte of Bax^−/− Bak^−/− YFP-Parkin
   MEFs after 24 hours of treatment with 2.5 μM valinomycin plus 0.1 μM
   bortezomib, 0.1 μM epoxomicin, or 1 μM ONX-0914. (E) Transmission
   electron microscopy (TEM) analysis of Bax^−/− Bak^−/− YFP-Parkin MEFs
   after 6 and 12 hours of treatment with 2.5 μM valinomycin, 20 μM qVD, ±
   0.1 μM bortezomib. Normal (blue) and abnormal (red) mitochondrial
   morphology are indicated by asterisks and reported per cell and as pie
   chart. n, nucleus. Scale bars, 0.98 μm. One representative image from
   each condition, n ≥ 11. (F) Absolute numbers of normal mitochondria
   scored on per cell basis in TEM. Each condition, n ≥ 11. (G) MOMP
   assessed by fluorescence-activated cell sorting (FACS) of Bax^−/−
   Bak^−/− YFP-Parkin MEFs treated with 1 μM valinomycin, qVD, ± 10 μM
   bortezomib. Means ± SD of triplicate samples (C) and (D). Data are
   representative of n = 3 (C), (D), and (G) independent experiments. *P <
   0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; n.s., not significant.

   To identify potential regulatory pathways of Parkin-mediated cell
   death, we subjected the identified candidate Class 1 genes to gene set
   enrichment analysis (GSEA) ([119]Fig. 3B and fig. S3D). The most
   significant pathways included the following: “proteasome,” “parkinson
   disease,” “PINK1/Parkin-mediated mitophagy,” and “cyt c-mediated
   apoptosis response.” Notably, in accordance with the observation that
   Parkin-dependent cell death is independent of BAX and BAK function and
   regulation, GSEA did not identify pathways related to intrinsic
   apoptosis signaling, which is regulated by Bcl2 family members.
   Furthermore, pathways such as “regulation of autophagy” and “Rab
   regulation of trafficking” were not significantly enriched, suggesting
   that after valinomycin treatment, although YFP-Parkin MEFs undergo
   mitophagy (fig. S1E), genes required for the final stage of lysosomal
   degradation during autophagy might not be necessary for this
   Parkin-dependent process of apoptosis. Together, our genome-wide siRNA
   screen defined two major classes of genes that regulate the process of
   Parkin-dependent cell death in either a positive or negative manner.

Proteasomal activity mediates OMM breaks and cyt c release into the cytosol

   We then focused on the “proteasome” gene signature, as it was the most
   significantly enriched pathway for Class 1 members. To validate this
   result, we selected the proteasome 26S subunit adenosine triphosphatase
   4 (Psmc4) as representative of the canonical proteasome as well as the
   proteasome 20S subunit beta 8 (Psmb8), Psmb9, and Psmb10 as components
   for the immune proteasome (fig. S3, D to F). siRNA silencing of these
   genes protected the Bax^−/− Bak^−/− YFP-Parkin MEFs from apoptosis
   induced by valinomycin, and consistent with previous reports ([120]46),
   silencing of these genes also delayed mitophagy ([121]Fig. 3C and fig.
   S3G). Moreover, either inhibitors of the proteasome, such as bortezomib
   and epoxomicin, or the inhibitor of the immune proteasome ONX-0914
   protected Bax^−/− Bak^−/− YFP-Parkin MEFs from valinomycin-induced
   apoptosis ([122]Fig. 3D), as well as from mitophagy (fig. S3H)
   ([123]10–[124]13).

   Because one of the known functions of the proteasome is to degrade
   ubiquitinated proteins on the OMM ([125]13), we hypothesized that this
   activity could create breaks in the OMM, leading to the observed
   release of cyt c into the cytosol upon valinomycin treatment. We
   therefore conducted an electron microscopy analysis on Bax^−/− Bak^−/−
   YFP-Parkin MEFs treated at different time points with valinomycin in
   combination or not with bortezomib and assessed mitochondrial integrity
   ([126]Fig. 3E). When compared to control samples, cells treated with
   valinomycin showed mitochondria with abnormal morphology, typified by
   an enlarged shape and a reduced number of cristae, and breaks in or a
   total lack of OMM. This pattern became more evident at the latest time
   point, where the presence of normal mitochondria on a per cell basis
   decreased by 80% in the valinomycin-treated samples. Consistent with
   previous observations ([127]13), the presence of bortezomib largely
   rescued the mitochondrial morphology even after 12 hours of cotreatment
   with valinomycin, as well as preserved the total number of normal
   mitochondria per cell ([128]Fig. 3, E and F).

   We then asked whether the rescued mitochondrial integrity would
   consequently lead to a decreased release of cyt c from the mitochondria
   upon proteasome inhibition. Bortezomib treatment reduced the amount of
   cyt c released to the cytosol during valinomycin-induced MOMP
   ([129]Fig. 3G). These results, together with our previous observations,
   pointed to a mechanism in which Parkin-mediated ubiquitination triggers
   proteasome-dependent degradation of the OMM, consequent release of cyt
   c into the cytosol, activation of caspases, and ultimately apoptosis
   (fig. S3I).

A phenotypic classification model identifies genes mediating apoptosis

   Next, we designed a secondary follow-up siRNA screen with the intention
   of validating the 313 candidate hits from the Class 1 members that
   showed a high relative inhibition of Casp3/7 activation (RICA). First,
   we narrowed the list of hits by excluding genes known to be
   indispensable for cell viability or siRNAs that induced either
   cytotoxicity or impaired expression of YFP-Parkin (fig. S4A). Then, we
   used this list of siRNAs to perform a siRNA screen in
   valinomycin-treated Bax^−/− Bak^−/− YFP-Parkin MEFs and assessed
   Casp3/7 activity. We identified a validated population of 170 genes
   with a RICA value higher than 0.5, recapitulating the phenotype of
   protection from caspase activation after treatment with valinomycin
   (table S3).

   Following mitochondrial damage, Parkin translocates to the impaired
   mitochondria ([130]2), and this event precedes cell death. We therefore
   reasoned that our observed process of cell death might be controlled at
   two different stages. To investigate this, we asked whether our
   validated list of hits could include genes that regulate the initial
   process of Parkin translocation to mitochondria and, therefore, caspase
   activation (“Class 1a”), in addition to genes that regulate only the
   later process of apoptosis but not Parkin translocation (“Class 1b”)
   (fig. S4B).

   For this purpose, we took advantage of the intracellular fluorescence
   signal of YFP-Parkin being cytosolic and diffuse in physiological
   conditions but clustered and unevenly distributed after treatment with
   mitochondrial uncouplers. While Bax^−/− Bak^−/− YFP-Parkin MEFs treated
   with a non-targeting siRNA (NT siRNA) showed a clustered, perinuclear
   YFP-Parkin localization ([131]Fig. 4A, left), PINK1 siRNA inhibited
   Parkin translocation, resulting in a homogenous distribution of
   YFP-Parkin fluorescence in the cytosol, even if cells were treated with
   valinomycin ([132]Fig. 4A, right).

Fig. 4. A phenotypic classification model identifies genes mediating
apoptosis.

   [133]Fig. 4.
   [134]Open in a new tab

   (A) Representative microscopy images showing YFP-Parkin localization
   and corresponding quantification of Edge and Saddle scores using a
   phenotypic classification model for per cell–based YFP-Parkin
   translocation. On the XY plot, the red line delineates the limit
   separating type A and B cells. After 48 hours of siRNA silencing,
   Bax^−/− Bak^−/− YFP-Parkin MEFs were treated with 500 nM valinomycin
   for 5 hours. Top, ×60 magnification; the bottom shows digital
   magnification of the corresponding white dotted square. Scale bars, 20
   μm. (B) Bidimensional distribution of a subset of genes subjected to a
   secondary validation siRNA screen; data are plotted as RICA (Y axis)
   and relative inhibition of YFP-Parkin translocation to mitochondria
   (RIPT) (X axis). Samples treated with siPINK1 and valinomycin were
   centered at X,Y = 1,1; samples treated with NT siRNA and valinomycin
   were centered at X,Y = 0,0. (C) Percentage of Casp3/7 activity, as RLU
   normalized to NT siRNA, of Bax^−/− Bak^−/− YFP-Parkin MEFs after
   silencing of the indicated genes and 9 hours of treatment with 1 μM
   valinomycin. (D) Flow cytometry analysis showing percentage of
   mitophagy, quantified as lysosomal-positive mt-mKeima, after silencing
   of the indicated genes and 9 hours of treatment with 1 μM valinomycin.
   Error bars in (C) and (D) represent the SD of the means of triplicate
   samples. Data are representative of n = 3 independent experiments. *P <
   0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; n.s., not significant.

   We then developed a machine learning algorithm, based on YFP-Parkin
   clustering after valinomycin treatment (fig. S4, C to E), which defined
   type “A” and “B” cellular subtypes with reduced or increased
   translocation of YFP-Parkin onto the mitochondria, respectively
   ([135]Fig. 4A, bottom). We then calculated a relative inhibition of
   YFP-Parkin translocation (RIPT) value that we used to differentiate
   Class 1a from Class 1b genes ([136]Fig. 4B and table S3).

   Tomm7, which is important for PINK1 stabilization and Parkin
   translocation ([137]7), was identified as a member of Class 1a genes,
   showing a high RIPT value (fig. S4C). In contrast, classic apoptotic
   genes such as Apaf1, CytC, Hccs, and Casp3 as well as the proteasome
   subunit Psmb8 were characterized as Class 1b members, supporting the
   notion that none of those genes regulate the initial step of Parkin
   translocation onto mitochondria.

   To further test our classification, we used individual siRNAs to
   validate the Autophagy Related 4C Cysteine Peptidase (Atg4c) and the
   Dynlrb1 as members of the Class 1a and the Pellino E3 Ubiquitin Protein
   Ligase 1 (Peli1), Leucine Rich Repeat and Sterile Alpha Motif
   Containing 1 (Lrsam1), and Procollagen-Lysine,2-Oxoglutarate
   5-Dioxygenase 3 (Plod3) as members of the Class 1b. As predicted, the
   silencing of each of the tested genes was protected from caspase
   activation following valinomycin treatment, supporting the validity of
   our phenotypic classification model algorithm ([138]Fig. 4C and fig.
   S4F). In addition, the validated genes in Class 1a, restraining Parkin
   translocation, not only inhibited caspase activation but also mitophagy
   ([139]Fig. 4D). This was not the case for all the validated members
   belonging to Class 1b. Among them, Peli1 and Lrsam1, which are both E3
   ubiquitin protein ligases capable of mediating proteasomal degradation
   and autophagy ([140]47, [141]48), also inhibited Parkin-dependent
   mitophagy. However, Plod3 specifically inhibited apoptosis but failed
   to influence the process of autophagy-dependent mitochondrial
   degradation. Therefore, using our phenotypic classification model
   algorithm, we generated a useful framework for better understanding and
   characterizing the different steps of the observed Parkin-dependent
   apoptosis.

Failure in autophagy accelerates Parkin-dependent apoptosis

   We next focused our attention on the Class 2 genes, as identified by
   our genome-wide siRNA screen, whose silencing exacerbated caspase
   activation when compared to NT siRNA samples ([142]Fig. 3A). Pathway
   enrichment analysis on this set of genes showed significant enrichment
   for pathways such as “lysosome,” “phagosome,” “vesicle organization,”
   “regulation of macroautophagy,” “vesicular acidification,” and
   “autophagosome organization” ([143]Fig. 5A).

Fig. 5. Failure in autophagy accelerates Parkin-dependent apoptosis.

   [144]Fig. 5.
   [145]Open in a new tab

   (A) Overrepresentation enrichment analysis showing significant
   enrichment of pathways containing candidate genes from the Class 2
   members with a RICA < 0 as related to [146]Fig. 3A. (B) (Top)
   Percentage of Casp3/7 activity as RLU normalized to NT siRNA; (bottom)
   percentage of mitophagy by flow cytometry analysis after 48 hours of
   siRNA silencing of the indicated Rab family members and 9 hours of
   treatment with 1 μM valinomycin. (C) Cell death quantification by
   IncuCyte of Atg3^+/+ and Atg3^−/− YFP-Parkin MEFs after 24 hours of
   treatment with 2.5 μM valinomycin or 5 μM antimycin A plus 1 μM
   oligomycin. (D) Percentage of Casp3/7 activity as RLU normalized to
   control in HeLa cells WT, LC3 TKO, GABARAP TKO, ATG8 6KO, and ATG-R
   5KO, all expressing YFP-Parkin, after 9 hours of treatment with 1 μM
   valinomycin. (E) Casp3/7 activity normalized to time (t) = 0 in HeLa
   YFP-Parkin WT and YFP-Parkin ATG-R 5KO after 6 and 9 hours of treatment
   with 1 μM valinomycin in combination or not with 1 μM BX-795. (F)
   Immunoblot analysis showing expression of ATF4, total eIF2α, and
   phospho-eIF2α in Bax^−/− Bak^−/− YFP-Parkin MEFs treated for 24 hours
   with 40 μM qVD or qVD plus the indicated doses of valinomycin. (G)
   Quantitative polymerase chain reaction–relative mRNA expression of
   Irf7, Oasl1, and Ifnb1 genes in Bax^−/− Bak^−/− YFP-Parkin MEFs treated
   with 40 μM qVD or qVD plus 2 μM valinomycin for the indicated hours.
   Error bars in (B), (C), (E), and (G) represent the SD of the means of
   triplicate samples. Data are representative of (B), (C), and (E) n = 3
   or (G) n = 2 independent experiments. Error bars in (D) represent the
   SD of the mean of n = 3 independent experiments.

   Considering these results, we hypothesized that the activity of the
   autophagy machinery and lysosomal degradation are crucial for
   segregating and degrading mitochondria that otherwise promote
   apoptosis. To test this, we first interrogated some of the hits that
   belonged to the Rab family of small GTPases, such as Rab5 and Rab7,
   which play a central role in controlling autophagosome and lysosomal
   trafficking ([147]49) and regulating mitophagy ([148]50), and Rab9,
   which has been shown to mediate an alternative degradative pathway,
   driving small portions of mitochondria directly to lysosomal
   degradation (fig. S5A) ([149]51).

   In line with our screening results, we validated that Bax^−/− Bak^−/−
   YFP-Parkin MEFs treated with valinomycin and lacking Rab7, as well as
   the three isoforms of Rab5 or Rab9, showed reduced mitophagy when
   compared to an NT siRNA sample ([150]Fig. 5B, bottom). This delay in
   mitophagy was accompanied by a marked increase in caspase activation
   ([151]Fig. 5B, top), suggesting that the inhibition of lysosomal
   maturation leads to a pronounced acceleration of cell death. Similarly,
   Atg3^−/− YFP-Parkin MEFs, which lack ATG8 lipidation, were mitophagy
   deficient when treated with valinomycin or the combination of antimycin
   A and oligomycin (fig. S5, B and C) but were fully capable of
   undergoing apoptosis ([152]Fig. 5C).

   Parkin-dependent ubiquitination of OMM proteins leads to UPS- ([153]3)
   or autophagy-mediated degradation (fig. S5D) ([154]52). We observed
   that, although these two processes were both triggered by Parkin
   activity and mediated mitochondrial degradation, they exhibited
   opposite effects on cell death ([155]Fig. 3 and [156]Fig. 5, B and C).
   The discriminating factor between these two pathways is the recruitment
   of the ubiquitin binding autophagy receptors, five of which were
   previously characterized in the context of mitophagy ([157]14). Among
   them, optineurin (Optn) was identified as a Class 2 member
   ([158]Fig. 3A). We therefore postulated that the specific recognition
   of OMM-ubiquitinated proteins by autophagy receptors is a fate decision
   point in the context of Parkin-dependent apoptosis.

   To test this hypothesis, we used HeLa ATG-R 5KO cells, which lack the
   five autophagy receptors: Tax1 Binding Protein 1 (TAX1BP1), calcium
   binding and coiled-coil domain 2 (NDP52), NBR1 autophagy cargo receptor
   (NBR1), sequestosome 1 (SQSTM1), and OPTN ([159]14). These YFP-Parkin
   HeLa ATG-R 5KO cells showed a strong increase in caspase activation
   upon valinomycin treatment when compared to their wild-type (WT)
   counterpart ([160]Fig. 5D) but were defective in mitophagy (fig. S5, E
   and F).

   After cargo recognition through their ubiquitin binding domain, the
   autophagy receptors bind to ATG8 family members using their respective
   LIR domain ([161]53). HeLa cells express six members of the highly
   conserved Atg8 family: LC3A, LC3B, LC3C, γ-aminobutyric acid Type A
   Receptor-Associated Protein (GABARAP), GABARAP-like 1 (GABARAPL1), and
   GABARAPL2. Two of them, Gabarapl1 and Lc3a, were identified as Class 2
   genes ([162]Fig. 3A). Therefore, we evaluated the effect of valinomycin
   treatment in YFP-Parkin HeLa cells deficient for the LC3 (LC3 TKO),
   GABARAP (GABARAP TKO), or all members of the ATG8 family (6KO)
   ([163]54). Consistent with the results observed in cells lacking the
   autophagy receptors, we observed enhanced caspase activity (compared to
   the WT line) after valinomycin treatment in the two TKO cell lines,
   which appeared additive in the 6KO line ([164]Fig. 5D).

   Following mitochondrial damage, TANK Binding Kinase 1 (TBK1) mediates
   recruitment of the autophagy receptors OPTN, NDP52, and SQSTM1 to
   depolarized mitochondria ([165]55). Both Tbk1 and its binding partner
   TRAF (tumor necrosis factor receptor–associated factor) family
   member–associated NF-κB (nuclear factor κB) activator (Tank) were also
   found among our Class 2 members. We therefore addressed their role in
   our model system, and, in line with the previously analyzed Class 2
   members, Bax^−/− Bak^−/− YFP-Parkin MEFs treated with the TBK1
   inhibitor BX-795 showed a substantial reduction in valinomycin-induced
   mitophagy ([166]56) and increased caspase activation and apoptosis
   ([167]Fig. 5E and fig. S5, G to I).

   To test whether TBK1 regulates cell death through its action on
   autophagy receptors, we treated HeLa ATG-R 5KO with valinomycin in
   combination with BX-795. While BX-795 treatment increased caspase
   activation in YFP-Parkin HeLa WT cells, it did not have any effect on
   YFP-Parkin HeLa ATG-R 5KO cells, supporting a model in which TBK1
   negatively regulates Parkin-dependent cell death by facilitating the
   recruitment of autophagy receptors to damaged mitochondria
   ([168]Fig. 5E).

   Previous studies have shown that the engagement of the mitochondrial
   pathway of apoptosis can induce the integrated stress response, as
   evidenced by eukaryotic translation initiation factor 2A (eIF2α)
   phosphorylation and expression of activating transcription factor 4
   (ATF4) ([169]57). Similarly, such engagement can induce type I
   interferon (IFN) responses when caspases are inhibited, because of the
   activation of the cyclic guanosine monophosphate-adenosine
   monophosphate synthase (cGAS)-stimulator of interferon genes (STING)
   pathway by released mitochondrial DNA ([170]58, [171]59). Consistent
   with these findings, we observed that treatment of Parkin-expressing
   HeLa cells with valinomycin in the presence of the caspase inhibitor,
   qVD, robustly induced eIF2α phosphorylation and ATF4 expression
   ([172]Fig. 5F) as well as expression of several IFN response genes
   ([173]Fig. 5G). It remains unclear how mitophagy may regulate these
   processes.

   In conclusion, our results suggest that ubiquitinated OMM proteins can
   be degraded by UPS, thereby increasing OMM rupture, cyt c release, and
   cell death. Alternatively, they can be recognized by autophagy
   receptors, triggering the engulfment of mitochondria into
   autophagosomes and driving autophagic degradation of damaged
   mitochondria. These contrasting effects dictate cell fate in
   Parkin-expressing cells in response to mitochondrial depolarization.

DISCUSSION

   We found that Parkin expression sensitizes cells to die after
   treatments that depolarize mitochondria. PINK1 was necessary and
   sufficient to drive this Parkin-dependent cell death. It is therefore
   conceivable that, in tissues or cell types expressing sufficiently high
   levels of Parkin, this form of cell death might take place
   physiologically in response to mitochondrial stressors.

   Parkin-dependent cell death is associated with caspase activation and
   therefore is a bona fide apoptosis that requires APAF1 activation and
   cyt c release from the intermembrane space ([174]26). In the absence of
   caspase activity, cells nevertheless died, likely because of
   catastrophic mitochondrial damage, similar to that of
   caspase-independent cell death following MOMP ([175]32). Our genetic
   approach showed that Parkin-dependent cell death did not require the
   effectors BAX and BAK. Previous reports have suggested a central role
   for MCL1 in a similar BAX- and BAK-independent cell death ([176]23,
   [177]24); however, in our experimental setting, neither BCL2, BCL-xL,
   nor MCL1 regulated cyt c release and the subsequent extent of
   apoptosis. Thus, Parkin-mediated apoptosis appears to overlap with
   known mechanisms of classical apoptosis downstream of cyt c release but
   remains mechanistically distinct at the level of outer membrane
   permeabilization.

   Our observations are consistent with the literature describing the role
   of Parkin activity at the OMM in triggering mitophagy. During
   mitophagy, the UPS-mediated degradation of OMM proteins is followed by
   the complete degradation of mitochondrial remnants through autophagy.
   Using pharmacological or genetic approaches, coupled to the
   mitochondrial targeted probe for lysosomal degradation, mt-mKeima, we
   observed that the inhibition of either UPS or autophagy block lysosomal
   degradation of mitochondrial proteins. However, while inhibition of the
   UPS inhibited cell death, inhibition of autophagy did not, suggesting
   that the processes of mitochondrial degradation and cell death are
   differentially regulated downstream of the activation of the UPS.
   Therefore, our observations on Parkin-dependent cell death cannot be
   explained in light of the previously described process of autophagic
   cell death ([178]29).

   While our genome-wide CRISPR screen identified Pink1 as the only gene
   regulating cell death upstream of Parkin, our siRNA screen and the
   following validation with a machine learning algorithm for the analysis
   of Parkin translocation onto mitochondria gave us a more nuanced
   picture defined by a complex set of genes involved in the regulation of
   this form of cell death. With this approach, we identified major
   intracellular pathways that positively or negatively regulate this
   process. While, the UPS was confirmed as necessary for the induction of
   cell death, the inhibition of pathways regulating autophagy and
   lysosomal function accelerated cell death. Among them, and in line with
   a previous report ([179]60), we found that several members of the Rab
   small GTPase family, which play a major role in the
   autophagosome-lysosomal maturation process during mitophagy ([180]41,
   [181]50, [182]51), increased caspase activation.

   Our work also defines autophagy receptors as important in the cell fate
   decision between survival and cell death. Previous studies in
   Drosophila suggested that the ubiquitination status of the Voltage
   Dependent Anion Channel 1 (VDAC1) is an important discriminant between
   mitophagy and cell death in the context of Parkin activation ([183]25).
   Considering our results, it is conceivable that the VDAC1
   ubiquitination status determines its ability to recruit autophagy
   receptors. Moreover, our model could explain, at least in part,
   previous observations involving autophagy receptors. For instance,
   mutations in Optn, which reduced OPTN levels and/or activity, were
   associated with neurodegeneration and cell death ([184]61) and have
   been reported to induce apoptosis, which was antagonized by autophagy
   ([185]62). Furthermore, the activity of the autophagy receptors OPTN,
   NDP52, and SQSTM1 is regulated by TBK1-mediated phosphorylation
   ([186]55), and similarly, RAB7A is regulated by TBK1 during mitophagy
   ([187]63).

   In our study, we observed that in the context of Parkin activation,
   pharmacological regulation of TBK1 had a major effect on the ability of
   cells to undergo programmed cell death. TBK1 therefore might represent
   a target for the development of future therapies in the context of
   cancer and neurodegenerative diseases.

MATERIALS AND METHODS

Cell lines

   HeLa [American Type Culture Collection (ATCC), CCL-2], 293T (ATCC,
   CRL-3216), Phoenix-AMPHO (ATCC, CRL-3213), and primary MEFs were
   maintained at 37°C/5% (v/v) CO[2] in a humidified incubator in
   Dulbecco’s modified Eagle’s medium (Gibco, 11971–025) supplemented with
   10% fetal bovine serum, 2 mM l-glutamine (Gibco, 25030-081),
   penicillin-streptomycin (100 U/ml; Gibco, 15140-122), 1 mM sodium
   pyruvate (Gibco, 11360-070), nonessential amino acids (Gibco,
   11140-050), and 55 μM β-mercaptoethanol (Gibco, 21985-023). HCT116
   (ATCC, CCL-247) were maintained in McCoy’s 5 A medium (ATCC, 30-2207)
   and supplemented as above. MEFs were immortalized with the SV40 large T
   antigen (pBabe-Neo-SV40-LTA). WT HCT116, Bax^−/− Bak^−/− HCT116, and
   Bax^−/− Bak^−/− MEFs cell lines were a gift from R. Youle ([188]64).
   For the generation of Atg3^−/− YFP-Parkin MEFs, Atg3^f/f mice (provided
   by H. W. Virgin) were bred to CMV-Cre mice (the Jackson Laboratory,
   006054) for deletion of loxP-flanked genes in all tissues. The obtained
   Atg3 heterozygous mice were subsequently mated to finally generate
   Atg3^−/− KO MEFs. The following HeLa cells were provided by M. Lazarou:
   WT, LC3 TKO (lacking LC3A, LC3B, and LC3C), GABARAP TKO (lacking
   GABARAP, GABARAPL1, and GABARAPL2), ATG8 6KO (lacking the six ATG8
   members), and ATG-R 5KO (lacking the five autophagy receptors TAX1BP1,
   NDP52, NBR1, SQSTM1, and OPTN) ([189]14, [190]54). Cell lines were
   routinely tested for mycoplasma using the MycoAlert Plus mycoplasma
   detection kit (Lonza).

Plasmids and retroviral transduction

   For generating YFP-Parkin cells, the pLZRS-YFP-Parkin construct was
   cloned as previously described ([191]65). To create the
   acPINK1-inducible system, OPA3-PINK1 Δ1–110, provided by R. Youle
   ([192]6), was subcloned into the Dox-inducible vector pRetroX-TRE3G
   (Clontech, 631188) as previously described ([193]66). To stably express
   the pH-sensitive mitochondrial mt-mKeima probe, cells were retrovirally
   transduced using the pCHAC-mt-mKeima plasmid, a gift from R. Youle
   (Addgene, 72342). All BCL2 family member cDNAs used in this study were
   of human origin; FLAG-BCL2, FLAG-BCL-xL, and FLAG-MCL1 were subcloned
   into the retroviral vector pMX-IRES-Cerulean as previously described
   ([194]66). All stable cell lines were generated by retroviral
   transduction. Briefly, Phoenix-AMPHO (amphotropic) cells (0.5 × 10^6
   per 10-cm dish) were transfected with retroviral constructs using
   Lipofectamine 2000 (Life Technologies, 11668027). Two days later, the
   virus-containing supernatant was harvested, filtered, and used to
   infect 1 × 10^5 target cells in the presence of hexadimethrine bromide
   (polybrene) (5 μg/ml; Sigma-Aldrich, H9268). Forty-eight hours
   postinfection, stable transductants were selected after adding
   puromycin (2 to 0.5 μg/ml; Sigma-Aldrich, P8833) or were sorted by flow
   cytometry for YFP positivity.

Reagents, cell death induction, and chemical inhibitors

   The following reagents and concentrations were used, unless indicated
   otherwise: 0.3 to 5 μM valinomycin (Sigma-Aldrich, V0627), 5 μM
   antimycin A (Sigma-Aldrich, A8674), 1 μM oligomycin (Sigma-Aldrich,
   O4876), 1 μM actinomycin D (Sigma-Aldrich, A1410), 1 to 2 μM
   thapsigargin (ENZO, BML-PE180-0001), 1 to 2 μM tunicamycin
   [MedChemExpress (MCE), HY-A0098], 20 to 40 μM qVD (MCE, HY-12305), 5 μM
   ABT-737 (MCE, HY-50907), 5 μM [195]S63845 (MCE, HY-100741), 5 μM
   raptinal (MCE, HY-121320), 100 μM etoposide (MCE, HY-13629), 1 μM
   rotenone (MCE, HY-B1756), 0.1 to 10 μM bortezomib (Cayman, 10008822),
   0.1 μM epoxomycin (MCE, HY-13821), 1 μM ONX-0914 (Cayman, 16271), Dox
   (1 μg/ml; Clontech, 631311), and 1 μM BX-795 (MCE, HY-10514).

Mitochondrial potential measurement

   Mitochondrial transmembrane potential was assessed using the
   potentiometric dye TMRE (Thermo Fisher Scientific, T669). A total of 50
   nM TMRE dye was added to the cells for 5 min, and fluorescence was
   acquired by flow cytometry before (time 0) or after addition of 5 μM
   valinomycin or 5 μM antimycin A plus 1 μM oligomycin for the indicated
   time. The fluorescence emission of the dye was acquired with a 5-lasers
   Aurora (Cytek) spectral flow cytometer, and median fluorescence
   intensity (MFI) was analyzed using FlowJo (v10; Tree Star).

Mitochondrial reactive oxygen species measurement

   Mitochondrial reactive oxygen species production was assessed using the
   Mitochondrial Superoxide Indicator MitoSOX-RED (Thermo Fisher
   Scientific, [196]M36008). Cells were treated for the indicated times
   with 40 μM qVD or qVD plus 5 μM valinomycin. At the end of the
   treatment, cells were harvested using trypsin-EDTA (Corning, 25053CI)
   and washed, and 5 μM MitoSOX-RED was added to the collected cells for
   30 min at 37°C and 5% CO[2]. Cells were washed and the fluorescence
   emission of the dye was measured with a 5-lasers Aurora (Cytek)
   spectral flow cytometer, and MFI was analyzed using FlowJo (v10, Tree
   Star).

Cell death measurement

   After the indicated treatment, cell death was quantified using a
   two-color IncuCyte Zoom in-incubator Imaging System (Sartorius). The
   percentage of cell death was obtained by normalizing the dead cell
   count with the total cell count detected by the uptake of 1 μg/ml of
   the cell-impermeable dye propidium iodine (PI) (Sigma-Aldrich, P4170)
   and 100 nM of the cell-permeant dye SYTO 16 (Invitrogen, S7578),
   respectively. Casp3/7 activity was assessed by luminescence using the
   Caspase-Glo 3/7 Assay System following manufacturing instructions
   (Promega, G8091).

Clonogenic assay

   Cells were seeded in 24-well plates (5 × 10^4 per well) 24 hours before
   the assay. After treatment, wells were washed three times with
   phosphate-buffered saline (PBS), and fresh complete medium was added.
   After 5 to 7 days in culture, the plates were gently washed with PBS
   and then stained with methylene blue solution [1% (w/v) methylene blue
   in a 50:50 methanol/water solution] for 30 min at room temperature. The
   plates were then washes with distilled H[2]O, air dried, and scanned
   using a document scanner.

Assessment of mitophagy by mt-mKeima flow cytometry assay

   Mitophagy was assessed by flow cytometry with the mt-mKeima fluorophore
   that has bimodal excitation under neutral (440 nm) and acidic
   conditions (586 nm). Cells were seeded in 24-well plates (5 × 10^4 per
   well). The next day, cells were treated with compounds as specified in
   the figures, in the presence of 20 μM qVD. After treatment, cells were
   harvested with trypsin-EDTA (Corning, 25053CI), washed, and resuspended
   in annexin binding buffer (Invitrogen, V13246) in the presence of
   annexin-V–allophycocyanin (APC) (Biolegend, 640941) to exclude death
   cells. Mitophagy was quantified by detecting lysosomal mt-mKeima using
   dual-excitation ratiometric pH measurements at 488- or 405-nm (pH 7)
   and 561-nm (pH 4) lasers, as previously described ([197]14, [198]46).
   Data were analyzed using FlowJo (v10; Tree Star).

CRISPR-Cas9 genome editing of cell lines

   Highly edited KO MEFs cells were generated using CRISPR-Cas9 technology
   using CRISPR All-in-One (pLenti-U6-sgRNA-SFFV-Cas9-2A-Puro) lentiviral
   vectors (ABMgood, C442). A set of three oligonucleotide pairs of sgRNAs
   per each gene was used: sgPINK1-1 (GCTGATCGAGGAGAAGCAGG), sgPINK1-2
   (GATCTCCTGACAGGCCGAGG), sgPINK1-3 (TGCCAGCATCGAGTGTCCAG), sgAPAF1-1
   (TGGCGTCTTGTCAGTGATG), sqAPAF1-2 (GGCTGCTCGTTGATATGAG), sgAPAF1-3
   (TACAACGCTCTGCTACACGA), sgCASP9-1 (CTTCACGCGCGACATGATCG), sgCASP9-2
   (TGTTAAACCCCTAGACCACC), and sgCASP9-3 (GGGGTGAGGTTTCCCACGGC). Single KO
   clones were generated by selecting the KO pools with puromycin (0.5
   μg/ml) and sorting them for single cells by flow cytometry (St. Jude
   Flow Cytometry and Cell Sorting Immunology Facility). Deletions were
   confirmed by Western blot or by targeted next-generation sequencing
   (NGS).

CRISPR KO screen

   The mouse CRISPR-Cas9 KO (GeCKOv2) pooled library was a gift from
   F. Zhang (Addgene, 1000000052), and it was used following a previously
   described protocol ([199]67). Briefly, the gRNA pooled library in the
   vector lentiCRISPRv2 was amplified in Endura competent cells, and NGS
   was performed to determine sgRNA distribution. Lentivirus was then
   generated in human embryonic kidney 293T cells and virus titer was
   determined. Target YFP-Parkin MEFs were lentivirally transduced with
   the library at 0.3 multiplicity of infection, treated or not with two
   concentrations of valinomycin (300 or 600 nM) for three rounds of
   treatment and enriched by clonogenic survival for pools of cells
   expressing multiple gRNAs targeting the same gene. Genomic DNA was
   harvested and prepared for NGS. For NGS analysis, reads were aligned
   using PinAPL-Py ([200]http://pinapl-py.ucsd.edu/). Counts per million
   reads of samples treated with valinomycin were normalized to untreated
   samples, and two biological replicates were averaged to calculate the
   respective fold enrichment for each gRNA. The significance of each
   sgRNA was determined using one-sided permutation test. P value was
   calculated as the fraction of events when the enrichment was greater
   than 10,000 randomly chosen sgRNAs. Significance at the gene level was
   tested by performing a two-sided Welch’s t test of the enrichment of
   the sgRNAs recovered for each gene against six randomly chosen sgRNAs.
   Data were displayed as negative logarithm of the gene-level P values
   against the mean enrichment and number of significant sgRNAs. Python
   scripts are available from the authors upon request. Validation of the
   candidate gene for the screening phenotype was performed by cloning
   three different sgRNAs into the plasmid backbone of the sgRNA library
   and by performing the screening workflow YFP-Parkin MEFs PINK1 KO
   single-cell clones, or the poll of the three different clones, as
   described above. NGS data have been deposited with links to BioProject
   accession number PRJNA956539 in the NCBI BioProject database
   ([201]www.ncbi.nlm.nih.gov/bioproject/).

Cyt c release assay measured by flow cytometry

   A total of 1 × 10^5 cells were seeded 1 day before the experiment. The
   next day, cells were treated with drugs in combination with 40 μM qVD
   to prevent caspase activation. After treatment, cells were harvested by
   trypsinization, washed twice with cold PBS, and permeabilized on ice
   with 500 μl of digitonin (25 μg/ml) in PBS for 10 min. After two washes
   with PBS/1 mM EDTA/1% bovine serum albumin [fluorescence-activated cell
   sorting (FACS) buffer], cells were resuspended in
   Fixation/Permeabilization buffer (BD, 554714) and incubated for 20 min
   on ice. Cells were then washed with BD Perm/Wash buffer and stained on
   ice in 50 μl of the same buffer with Alexa Fluor 647–conjugated
   anti–cyt c antibody (1:200; BioLegend, 612310) or mouse IgG1k Isotype
   Ctrl (Invitrogen, 51-4714-81) for 30 min. Last, cells were washed two
   more times with BD Perm/Wash buffer and resuspended in FACS buffer
   before flow cytometric analysis.

Assessment of tumor growth in vivo

   A total of 1 × 10^6 Bax^−/− Bak^−/− MEFs expressing or not YFP-Parkin
   were resuspended in 100 μl of PBS and injected subcutaneously into the
   flanks of males NSG immunodeficient mice (the Jackson Laboratory,
   005557). At days 4 and 8 after cell injection, mice were treated by
   peritumoral injection with vehicle or valinomycin (0.4 mg/kg). Tumor
   growth was measured every 7 days, and mouse survival was monitored. All
   mice were housed in pathogen-free facilities, in a 12-hour light/dark
   cycle in ventilated cages, with chow and water supply ad libitum. All
   animal procedures were performed in accordance with protocols approved
   by the Institutional Animal Care and Usage Committee of St. Jude
   Children’s Research Hospital.

Genome-wide siRNA screen

   Briefly, 20,518 mouse genes were screened using RNA interference (RNAi)
   for relief from PINK1/Parkin-dependent cell death (fig. S3A). Bax^−/−
   Bak^−/− YFP-Parkin MEFs were reverse-transfected using Dharmafect II
   (Dharmacon, T-2002-04) with 25 nM of a pool of four siRNAs (Mouse
   siGENOME siRNA Library, SMARTpool; Dharmacon, G-015005-E2-05),
   targeting each gene per well arrayed in a 384-well format. After 48
   hours of RNAi knockdown, cells were challenged by addition of
   valinomycin (Sigma-Aldrich, V0627) to a final concentration of 500 nM.
   Six hours after valinomycin challenge, cell death activity was assayed
   using the Caspase-Glo 3/7 Assay System (Promega) and the luminescence
   for each well was quantified using the Envision Multilabel plate reader
   (PerkinElmer) (fig. S3A). For the primary screen,
   PINK1/Parkin-dependent cell death was evaluated relative to the
   negative control NT siRNAs (siGENOME NT siRNA #4; Dharmacon,
   D-001210-04) and the positive control PINK1 siRNA (siGENOME PINK1 siRNA
   SMARTpool; Dharmacon, M-044666-01). The luminescence for each well in
   the primary screen (table S2) was processed such that values were log
   transformed, adjusted for plate position bias using median polish, and
   two-point normalized where NT siRNA and PINK1 siRNA centered to the
   values 0 and 1, respectively, using Genedata Screener software. During
   the primary screen, a composite Z′ factor of 0.42 was achieved,
   attesting to the robustness of the screening approach (table S2)
   ([202]68, [203]69). Candidate hits required for PINK1/Parkin-dependent
   cell death were selected after z score transformation using Pink1
   knockdown population mean and SD. A total of 313 gene knockdowns that
   compare favorably with Pink1 knockdown (z score greater than −2.0) were
   selected and shown (fig. S3, B and C) as a RICA value close to 1. Those
   hits were defined as Class 1 and were identified as genes behaving as
   we observed for Pink1, where their silencing inhibited Parkin-dependent
   caspase activation. We also identified a second class (Class 2) of
   genes whose silencing increased caspase activation when compared to a
   NT siRNA treatment, therefore receiving, after normalization, negative
   RICA values.

Secondary follow-up screen

   In total, 419 genes of interest (top hits and selected genes) were
   screened in a validation experiment. Fresh pools of four siRNAs were
   obtained from Dharmacon, and the RNAi knockdown was induced following a
   reverse transfection protocol similar to the primary screen. The 419
   genes of interest were screened for relief from PINK1/Parkin-dependent
   cell death after valinomycin challenge following an experimental
   procedure similar to the primary screen (table S3). To extend the
   findings, RNAi-associated cell toxicity was directly interrogated 48
   hours after siRNA transfection (in the absence of valinomycin
   challenge) using the CellTiter-Glo luminescent viability assay
   (Promega, G7572) and read using an Envision plate reader. In a separate
   experiment, YFP-Parkin expression was characterized after RNAi-induced
   gene knockdowns using automated imaging and image analysis. Hoechst
   staining and YFP-Parkin expression were captured using 405- and 488-nm
   laser excitations, respectively, of four fields per well using a GE
   InCell 6000 analyzer (GE Healthcare) at ×20 magnification. All data
   were then uploaded to the Columbus Image Data Storage and Analysis
   System (PerkinElmer) for image analysis. Nuclear and cytoplasmic
   boundaries were first segmented using staining from the 405 channel,
   and any objects not completely within the field of view were removed
   from further analysis. Cells were additionally selected for analysis by
   passing minimum nuclear roundness and size thresholds. The mean
   YFP-Parkin intensity was measured for every cell, and the median value
   of that population was reported for each well. Thirty-five gene
   knockdowns negatively affected cell health (fig. S4A, left), and
   additional nine gene knockdowns were associated with reduced YFP-Parkin
   expression (fig. S4A, right). The validation screen identified 170
   genes as required for PINK1/Parkin-dependent cell death.

Phenotypic classification model

   A machine learning algorithm classified two cellular subtypes on the
   basis of the intracellular YFP-Parkin distribution after 5 hours of
   valinomycin treatment. Hoechst staining and Parkin-YFP expression were
   captured using 405- and 488-nm laser excitations, respectively, of nine
   fields per well using a GE InCell 6000 analyzer (GE Healthcare). All
   data were then uploaded to the Columbus Image Data Storage and Analysis
   System (PerkinElmer) (fig. S4, C and D) for image analysis. Nuclear and
   cytoplasmic boundaries were first segmented using staining from the 405
   channel, and any objects not completely within the field of view were
   removed from further analysis. Cells were additionally selected for
   analysis by passing minimum nuclear roundness and size thresholds. The
   Columbus PhenoLOGIC algorithm (PerkinElmer) was then used to calculate
   the “Edge” and “Saddle” textures for each cell using a 2-pixel seed
   size. Last, once these textures were captured by the software,
   empirical classification was used in combination with machine learning
   to classify the cells into two subpopulations (fig. S4, D and E): type
   A cells, defined as events with a lower Saddle and Edge score and with
   a reduced translocation of YFP-Parkin onto the mitochondria after
   valinomycin treatment (YFP evenly distributed throughout the cells),
   and type B cells, with higher Saddle and Edge scores and with increased
   YFP-Parkin translocation onto the mitochondria (YFP accumulated in
   puncta). The number of cells in class “B” were then divided by the
   total cells in classes “A” and “B” to generate a single value per well
   to describe the fraction of cells with YFP-Parkin in distinct puncta.
   We then extrapolated a unique value describing YFP-Parkin translocation
   onto mitochondria (RIPT) and, after normalization, calculated a RIPT
   value that we used to differentiate Class 1a from Class 1b genes
   ([204]Fig. 4B and table S3). As a negative and positive control, NT and
   PINK1 siRNAs were assigned a RIPT value of 0 and 1, respectively.

siRNA transient transfection

   siRNA transient transfection was performed over 48 hours using
   Dharmafect II (Dharmacon, T-2002-04) with 25 nM of a mixture of a pool
   of four siRNAs (Dharmacon, Mouse siGENOME siRNA, SMARTpool), selected
   from the Mouse siGENOME siRNA Library, SMARTpool (Dharmacon,
   G-015005-E2-05). siRNA oligo sequences can be retrieved from the siRNA
   ID listed in table S2.

Pathway enrichment analysis

   GSEA was performed against MSigDB database with R package
   clusterProfiler (version 3.18.1) ([205]Fig. 3B and fig. S3D). RICA was
   used for gene ranking. Overrepresentation enrichment of the Class 2
   members with a RICA < 0 was performed against MSigDB with R package
   clusterProfiler (version 3.18.1) ([206]Fig. 5A). Moreover, to examine
   whether the enrichment was due to Pink1 and Prkn (encoding for Parkin)
   in the gene sets, the two genes were excluded from the gene sets and
   similar results were obtained.

Electron microscopy

   Samples for electron microscopy were fixed in 0.1 M cacodylate buffer
   containing 2.5% glutaraldehyde and 2% paraformaldehyde. Samples were
   postfixed in reduced osmium tetroxide and contrasted with aqueous
   uranyl acetate. Dehydration was by an ascending series of ethanol to
   100% followed by 100% propylene oxide. Samples were infiltrated with
   EmBed-812 and polymerized at 60°C. Embedded samples were sectioned at
   ~70 nm on a Leica UC-7 ultramicrotome and examined in a Thermo Fisher
   Scientific TF20 transmission electron microscope at 80 kV. Digital
   micrographs were captured with an Advanced Microscopy Techniques
   imaging system controlled by SerialEM for acquiring montages of whole
   cells for analysis. Montages were stitched and exported as tiff files
   for analysis using the IMOD software
   ([207]https://bio3d.colorado.edu/imod/). Unless otherwise indicated,
   all reagents were provided by Electron Microscopy Sciences.

Western blotting and antibodies

   Cells were lysed in radioimmunoprecipitation assay buffer [50 mM
   tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40, 0.5% deoxycholic acid, and
   0.1% SDS], complete protease inhibitors (Roche, 11836153001), or
   phosphatase inhibitors (Roche, 04906837001). Protein concentration was
   assessed by bicinchoninic acid (BCA) assay (Pierce, 23225) and
   normalized before Western blotting. Samples were run on 4–12% Criterion
   XT Bis-Tris Precast Gels (Bio-Rad) and transferred to Hybond-C Extra
   membranes (GE Healthcare). After antibody incubation and
   chemiluminescent substrate exposure (homemade ECL), membranes were
   acquired by a ChemiDoc Touch Imaging System (Bio-Rad) and analyzed by
   Image Lab software (Bio-Rad). The following antibodies were used for
   immunoblotting: anti-BAK (Santa Cruz Biotechnology, sc832), anti-BAX
   clone N-20 (Santa Cruz Biotechnology, sc493), anti-MCL1 (Rockland,
   600–401–394), anti–β-actin clone C4 (Santa Cruz Biotechnology,
   sc47778), anti-Parkin (Santa Cruz Biotechnology, sc32282), anti–green
   fluorescent protein (Santa Cruz Biotechnology, sc8334), anti-PINK1
   (Novus, BC100-494), anti-COX Vb (Abcam, ab110263), anti-PARP (Cell
   Signaling Technology, 9542), anti-APAF1 (Bio-Rad, AHP-487), anti-FLAG
   (Sigma-Aldrich, F7425), anti-Casp9 (Cell Signaling Technology, 9504),
   anti-ATG3 (Cell Signaling Technology, 3415), anti-ATF4 (Cell Signaling
   Technology, 11815), total eIF2α (Cell Signaling Technology, 9722), and
   phospho-eIF2α (Cell Signaling Technology, 9721).

Real-time quantitative polymerase chain reaction

   Total RNA was extracted using the RNeasy Mini Kit (QIAGEN, 74104).
   Reverse transcription reaction was performed with M-MLV reverse
   transcriptase (Invitrogen, 28025013) and random hexamers (IDT).
   Real-time quantitative polymerase chain reaction (qPCR) was performed
   using SYBR Green (Invitrogen, 4309155) in a 7500 Fast Real-Time PCR
   System (Applied Biosystems) using the following PCR conditions: 50°C in
   2 min, 95°C in 10 min, 45 cycles of 95°C in 15 s, and 60°C in 1 min.
   β-actin mRNA expression was used for normalization. The following
   primers (IDT) were used in this study: mouse_β-actin (forward:
   AGAGGGAAATCGTGCGTGAC; reverse: CAATAGTGATGACCTGGCCGT), mouse_Pink1
   (forward: GGCTTCCGTCTGGAGGATTAT; reverse:AACCTGCCGAGATATTCCACA),
   mouse_Psmb8 (forward: ATGGCGTTACTGGATCTGTGC; reverse:
   CGCGGAGAAACTGTAGTGTCC), mouse_Psmb9 (forward: CATGAACCGAGATGGCTCTAGT;
   reverse: TCATCGTAGAATTTTGGCAGCTC), mouse_Psmc4 (forward:
   CAGCACTGTCCGTGTCTCG; reverse: CTGCTCGTCCTTGATATACTCCT), mouse_Psmb10
   (forward: GAGGAATGCGTCCTTGGAACA; reverse: CAC AACCGAATCGTTAGTGGC),
   mouse_Atg4c (forward: AGATGAAAGCAAGATGTTGCCT; reverse:
   CCCTGTAGGTCAGCCATATTCTA), mouse_Dynlrb1 (forward: AGCCAGAAAGGAGTGCAGG;
   reverse: GGGATTGTCCATTGTGCTCTTG), mouse_Peli1 (forward:
   CAAGGACCAGCATAGCATATCAT; reverse: GTTGACCGACCAATCTGAAACA), mouse_Lrsam1
   (forward: CTCGAAAACGCCTGGAGTACC; reverse: AGGTGAGGTGGTTTGTATGGA),
   mouse_Plod3 (forward: AAGAAGTTTGTTCAGAGTGGCA; reverse:
   TGAATCCACCAGAGTTGAGGA), mouse_Rab7 (forward: AAGCCACAATAGGAGCGGAC;
   reverse: AGACTGGAACCGTTCTTGACC), mouse_Rab5c (forward:
   TGGTCCTCCGCTTTGTCAAG; reverse: TGACCGTTGTATCGTCTAAGCA), mouse_Rab5b
   (forward: GGGAAGTCTAGCCTGGTGTTA; reverse: GCTTTCCTGGTATTCATGGAACT),
   mouse_Rab5a (forward: GCTAATCGAGGAGCAACAAGAC; reverse:
   CCAGGCTTGATTTGCCAACAG), mouse_Rab9 (forward: ATGGCAGGAAAATCGTCTCTTT;
   reverse: GCATGGTAACAAAATGTCCGTCC), mouse_Irf7 (forward:
   TTTGGAGAGTGGCTATTGGGG; reverse: CCTACGACCGAAATGCTTCCA), mouse_Oasl1
   (forward: CATGCTCCCAAGCTTCTCTCT; reverse: CTGCCATGGCTCCTCCTTTTT), and
   mouse_Ifnb1 (forward: AACTCCACCAGCAGACAGTG; reverse:
   AGTGGAGAGCAGTTGAGGAC).

Statistical analysis

   Data were shown as means ± SD; please refer to figure legends for
   detailed description. Data were plotted and analyzed with GraphPad
   Prism 9.0 software (GraphPad Software). Statistical significance was
   determined using one- or two-way analysis of variance (ANOVA) followed
   by the Sidak’s multiple comparisons test. The thresholds for
   significance were denoted as asterisks: *P < 0.05, **P < 0.01, and ***P
   < 0.001; n.s., not significant.

Acknowledgments