Ghost mitochondria drive metastasis through adaptive GCN2/Akt therapeutic vulnerability

Significance Exploitation of mitochondrial functions promotes tumor traits, including metastasis, which is responsible for >90% of all cancer deaths. In this study, we investigated how mitochondrial fitness impacts tumor behavior. We found that acutely damaged, de-energized, and reactive oxygen species-producing mitochondria not only persist in cancer but are also key enablers of metastasis. These “ghost” mitochondria originate from the heterogeneous and often reduced expression of Mic60, an essential scaffold of organelle structure, in certain human cancers. The compensatory activation of gene expression programs as well as GCN2/Akt kinase signaling enables the survival of Mic60-low tumors but also provides a new therapeutic target in advanced and hard-to-treat malignancies.

T he rewiring of metabolic pathways is a ubiquitous cancer trait that confers cellular plasticity, expands clonal heterogeneity, and enables disease progression (1). There is now a consensus that mitochondria are important for this process, titrating energy output, buffering oxidative stress, and controlling a host of cell death programs (2). In particular, exploitation of mitochondrial functions has been linked to metastatic competence (3,4). This involves oxidative bioenergetics (5) and redox balance (6) but also deregulated mitochondrial dynamics (7), a process that controls the size, shape, and distribution of mitochondria and their trafficking to the cortical cytoskeleton, where they fuel pivotal steps of cell motility, such as membrane lamellipodia dynamics, turnover of focal adhesion (FA) and phosphorylation of signaling kinases (8).
However, the environment of tumor growth is highly unfavorable to mitochondrial fitness. Erratic oxygen concentrations, high levels of oxidative radicals (9), constantly changing metabolic needs (10), and vulnerabilities of the mitochondrial proteome (11) are all potent stimuli to disrupt mitochondrial integrity, shut off organelle functions, and activate cell death (12). Quality-control measures activated in these settings, in particular mitophagy (13), are designed to remove such subpar, "ghost" mitochondria and restore homeostasis. However, the role of these pathways in cancer is far from clear, and activation of mitophagy has been paradoxically linked to tumor progression (14) as well as treatment resistance (15).
An important regulator of mitochondrial integrity is Mic60, also called Mitofilin or inner membrane mitochondrial protein.
Mic60 is an essential constituent of a MICOS complex (16) that maintains cristae architecture (17), organizes respiratory complexes (18), and ensures outer membrane biogenesis (19). Whether this pathway is important in cancer has not been determined, but there is evidence that Mic60 participates in mitochondrial fitness, including PINK1/Parkin-directed mitophagy (20) and mitochondrial dynamics (21).
In this study, we investigated how mitochondrial fitness may impact cancer traits and potentially expose therapeutic vulnerabilities in advanced disease.

Significance
Exploitation of mitochondrial functions promotes tumor traits, including metastasis, which is responsible for >90% of all cancer deaths. In this study, we investigated how mitochondrial fitness impacts tumor behavior. We found that acutely damaged, de-energized, and reactive oxygen species-producing mitochondria not only persist in cancer but are also key enablers of metastasis. These "ghost" mitochondria originate from the heterogeneous and often reduced expression of Mic60, an essential scaffold of organelle structure, in certain human cancers. The compensatory activation of gene expression programs as well as GCN2/Akt kinase signaling enables the survival of Mic60-low tumors but also provides a new therapeutic target in advanced and hard-to-treat malignancies.

Results
Mic60 Expression in Cancer. To study the role of mitochondrial fitness in cancer, we focused on Mic60 as an essential scaffold of organelle integrity and function (18). Inspection of the Human Protein Atlas database showed that Mic60 expression was highly heterogeneous in cancer, as several tumor types had reduced, increased, or unchanged levels of Mic60 compared to normal tissues (SI Appendix, Fig. S1A). Immunohistochemical (IHC) staining of a universal tumor microarray (TMA; n = 5 to 8 cases per tumor type) gave similar results (Fig. 1A), where Mic60 expression was reduced in colorectal adenocarcinoma (COREAD) and glioblastoma (GBM), unchanged in breast (BRCA) and prostate adenocarcinoma (PRAD), or increased in lung adenocarcinoma (LUAD) compared to adjacent normal tissue (SI Appendix, Fig. S1B). Consistent with these results, Mic60 mRNA levels in The Cancer Genome Atlas (TCGA) database were reduced in GBM and COREAD but prominently upregulated in LUAD (SI Appendix, Fig. S1C). Other potential Mic60low tumors in this analysis included malignancies of kidney, thyroid, head and neck, and soft tissue, whereas uterine and cervix cancer had higher Mic60 mRNA levels compared to normal tissues (SI Appendix, Fig. S1C). Although breast cancer showed increased Mic60 protein (SI Appendix, Fig. S1A) and mRNA (SI Appendix, Fig. S1C) in public databases, our TMA analysis did not reach statistical significance (SI Appendix, Fig. S1B). Heterogeneous Mic60 expression was also observed intratumorally. When analyzed in patient samples of pancreatic ductal adenocarcinoma (PDAC), Mic60 expression ranged from focal perinuclear distribution in normal pancreatic acinar cells to disordered, submembranous or linear staining in in situ and invasive neoplastic epithelium to absence in poorly differentiated (basaloid) carcinomas by IHC (Fig. 1C). Mechanistically, differentiation of patient-derived GBM neurospheres, a process associated with the modulation of stemness and proliferative potential (22), lowered Mic60 as well as HIF1α mRNA levels (SI Appendix, Fig. S1D).
Mic60-Dependent Mitochondrial Integrity in Cancer. Next, we examined the function of Mic60 in cancer. Using a proteomics screen in PRAD PC3 cells, we identified 119 high-confidence mitochondrial proteins that associate with Mic60 (Fig. 1D). Bioinformatics analysis of this dataset identified multiple regulators of mitochondrial membrane transport and organization, protein sorting, Ca 2+ homeostasis, and oxidative phosphorylation (SI Appendix, Fig. S1E). Therefore, we sought to reproduce the phenotype of Mic60-low tumors (  (shRNA) or CRISPR-Cas9 (SI Appendix, Fig. S2A, Top). Small interfering RNA (siRNA) sequences targeting Mic60 were also characterized in PC3 cells, normal diploid fibroblasts, MRC5, breast adenocarcinoma MDA231, and osteosarcoma HT1080 cells (SI Appendix, Fig. S2A, Bottom). Using these approaches, silencing of Mic60 caused a catastrophic collapse of mitochondrial integrity in tumor cells, with disassembly of tubular network and cristae organization (Fig.  1E). This was accompanied by acute mitochondrial damage, characterized by increased outer membrane permeability ( Fig.  1F) and depolarization of the inner membrane ( Fig. 1G and SI Appendix, Fig. S2B). As a result, Mic60-depleted tumor cells exhibited decreased oxygen consumption rates (SI Appendix,  Fig. S4A), as well as an inhibition of colony formation (Fig. 2D). An analysis of the DepMap Portal (23) revealed dependency scores < 1 for all tumor types examined after Mic60 silencing by RNA interference (RNAi) or CRISPR-Cas9 ( Fig. 2E and SI Appendix, Fig. S4B), consistent with a general requirement of Mic60 for tumor cell proliferation. Accordingly, PC3 clones with reduced Mic60 levels by shRNA or CRISPR-Cas9 formed slow-growing superficial tumors in immunocompromised mice (Fig. 2F) with decreased K i -67 reactivity (SI Appendix, Fig. S4C), a marker of cell proliferation. Conversely, control PC3 cells formed exponentially growing flank tumors ( Fig.  2F) with high K i -67 staining (SI Appendix, Fig. S4C).
Despite cellular damage and activation of mitophagy, tumor cells with reduced Mic60 did not upregulate ferroptosisassociated genes (SI Appendix, Fig. S4D), a type of cell death induced by mitochondrial stress, and oxidized lipid content, a marker of ferroptosis, was unchanged compared to control cultures (Fig. 2G). Similarly, Mic60 depletion only modestly affected mitochondrial apoptosis, as quantified by caspase (DEVDase) activity ( Fig. 2H) or hypodiploid DNA content and flow cytometry (SI Appendix, Fig. S4E). Finally, no significant differences in necroptotic cell death were observed in control or Mic60-silenced cells over a 3-d ( Fig. 2I) or 5-d culture (SI Appendix, Fig. S4F), by analysis of plasma membrane integrity and light microscopy.
Mechanistically, siRNA silencing of STING (SI Appendix, Fig. S5E), a key regulator of mitochondrion-directed innate immunity, abolished the increase in cytokine mRNA levels after Mic60 loss (Fig. 3D). In addition, Mic60 depletion was accompanied by an increased phosphorylation of STAT1 (Ser727) and extracellular release of HMGB1 (SI Appendix, Fig. S5F), which are two effectors of IFN signaling. In contrast, senescenceassociated β-galactosidase staining was unchanged in control or Mic60-depleted cultures (Fig. 3E).
Mic60 Regulation of Tumor Cell Invasion and Metastasis. SASP signaling has been associated with increased tumor cell invasion and metastasis (25). Accordingly, Mic60 depletion changed the morphology of PC3 cells to a flattened, elongated, and spindle-shaped appearance, characterized by rearrangement of the actin cytoskeleton and redistribution of mitochondria to the cortical cytoskeleton (Fig. 4A). This was associated with reduced cellular roundness, increased surface area (Fig. 4B), and the appearance of epithelial-mesenchymal transition (EMT) markers, including E-and N-cadherin switch, and upregulation of vimentin, SLUG, and SNAIL (Fig. 4C). Functionally, Mic60 depletion enhanced FA turnover (SI Appendix, Fig. S6A), increasing the fraction of new and decayed FA, while reducing stable FA (SI Appendix, Fig.  S6B). This resulted in a greater speed of single-cell movements, longer distance traveled by individual cells (Fig. 4D), and accelerated directional cell motility in a "wound" closure assay, shortening the half-time (t1/2) of wound closure from 16. Although impaired in primary tumor growth (Fig. 2F), superficial flank tumors of Mic60-depleted PC3 cells gave rise to increased metastatic dissemination to the lungs of immunocompromised mice (Fig. 4G). As a second, independent model of metastasis, we injected control or Mic60-depleted PC3 cells in the spleen of immunocompromised animals and looked at liver metastasis after 11 d. Here, Mic60-silenced PC3 cells generated more numerous and larger liver metastases compared to controls ( Fig. 4 H and I). Based on these data, we next looked at matched patient samples of primary and metastatic LUAD. In this comparison, Mic60 mRNA levels increased in metastases compared to the primary tumor (Fig. 4J, Top) and the nonneoplastic lung tissue (Fig. 4J, Bottom), suggesting that Mic60 becomes re-expressed at established metastatic sites to support tumor cell proliferation. Mechanistically, siRNA silencing of mitochondrial GTPase RHOT1 or RHOT2 (SI Appendix, Fig. S8A), which mediate mitochondrial trafficking in tumors (26), normalized the speed of mitochondrial movements and the distance traveled by individual mitochondria after Mic60 depletion (SI Appendix, Fig.  S8B). Buffering oxidative stress gave similar results, as the reconstitution of Mic60-silenced cells with antioxidant Prx3 (27) corrected the increase in single-cell motility (SI Appendix, Fig. S8C), lowered the speed of cell movements and the total distance traveled by individual cells to levels of control cultures (SI Appendix, Fig. S8D), and reversed the increase in Matrigel invasion in these settings (SI Appendix, Fig. S8E). Reconstitution of Mic60-silenced cells with a mitochondrial-targeted superoxide scavenger, MitoTempo, also normalized tumor cell invasion to control levels (SI Appendix, Fig. S8F).

Discussion
In this study, we have shown that Mic60, an essential scaffold of mitochondrial structure, is heterogeneously expressed and often reduced in human cancer compared to normal tissues. As modeled in tumor cell lines, even a partial reduction in Mic60 levels was sufficient to induce an acute loss of mitochondrial fitness, leading to bioenergetics defects, cellular starvation, and oxidative stress. Despite the activation of quality-control measures of autophagy and mitophagy, tumor cells harboring such extensively damaged, ghost mitochondria managed to evade cell death, slowed down cell proliferation, and activated mitochondrial dynamics to fuel increased cell invasion and metastasis. This response was accompanied by the expression of an IFN/SASP-like transcriptome, as well as adaptive activation of GCN2/Akt survival signaling, which provided an actionable therapeutic target in these metastasis-prone tumors.
The basis for the heterogeneous expression of Mic60 in cancer remains to be elucidated. This may result from mitochondrial   and/or environmental stress conditions associated with tumor growth, including defective oxidative phosphorylation (28), or alternatively, mechanisms of tumor evolution, as suggested here by Mic60 downregulation during differentiation of patientderived GBM neurospheres.
Irrespectively, decreased Mic60 levels generate subpar, ghost mitochondria that bear striking similarities to the mitochondrial defects seen in aging (29). Consistent with this parallel, in vivo models of aging have been associated with reduced Mic60 levels (30), heightened SASP signaling (31), mitochondria-tonuclei retrograde gene expression (32), and a general, proinflammatory environment (33). However, key differences between aged and Mic60-low mitochondria were also noted (31). Aside from distinct profiles of cytokine induction, Mic60depleted tumor cells were negative for senescence-associated β-galactosidase and did not undergo permanent G1 cell cycle arrest, and gene expression changes in these settings were unrelated to p53 status.
A unique adaptive response of Mic60-low tumors was the upregulation of a nuclear transcriptome combining a type I IFN response characteristic of innate immunity (34) and cytokines/chemokines reminiscent of SASP signaling (25). Mechanistically, mitochondrial stressors induced by Mic60 loss, such as ROS (35), loss of membrane integrity (36), and energy starvation (37), have all been associated with mitochondria-tonuclei retrograde gene expression (38). Activation of a type I IFN response in these settings fits well with a key role of mitochondria in innate immunity (39), in line with the activation of STAT1 and the requirement of STING for cytokine production observed here. While there is evidence that SASP-associated cytokine/chemokine signaling promotes tumor growth, favors an immunosuppressive microenvironment, and enhances metastasis (25), the role of a type I IFN response in cancer is likely timeand context-specific. Whereas acute IFN signaling has been associated with antitumor immunity and improved treatment responses (34), sustained inflammatory conditions are protumorigenic (40) and chronic IFN stimulation enables myeloiddirected immunosuppression (41) and propagation of cancer stemness (42).
Consistent with this scenario, Mic60-low tumors switched from a proliferative to a highly motile and prometastatic phenotype, contributed by EMT, sustained FAK phosphorylation, and heightened mitochondrial dynamics (8). Described as phenotype switching (43), this reversible transition between proliferative and migratory states has been proposed as a potential escape mechanism for tumor cells to leave a stress-laden, unfavorable microenvironment and colonize distant sites, i.e., metastasis (44). Regulators of mitochondrial dynamics, such as syntaphilin (27), FUNDC1 (45), and now Mic60 depletion (this study), are important mediators of phenotype switching, reprogramming oxidative bioenergetics, and redox balance to promote heightened cell migration and invasion at the expense of cell proliferation. Consistent with this model, oxidative stress generated in Mic60-low tumors was a key mediator of increased mitochondrial trafficking and tumor cell movements, in keeping with a central role of ROS in tumor cell motility (46), EMT, and metastasis (47).
Despite an extensive loss of mitochondrial fitness, activation of autophagy/mitophagy, and high ROS production, Mic60-low tumor cells managed to persist likely through the activation of compensatory cell survival mechanisms. The induction of GCN2/ISR as well as Akt signaling (48,49) observed in these settings appears ideally poised to adjust metabolism under stress (50,51), preserve mitochondrial integrity (22,52) and oppose cell death (53,54). Although correlating with shortened patient survival, a potential dependence of Mic60-low tumors to GCN2/ISR/Akt adaptive signaling was therapeutically exploitable, and proof-of-concept studies shown here demonstrated that pharmacologic or genetic targeting of this pathway can restore mitochondrial cell death and inhibit proliferation selectively of Mic60-low tumor cells.
In summary, we have shown that persistent, acutely degraded ghost mitochondria are major signaling hubs in cancer, driving multiple, adaptive responses of nuclear gene expression, ISR activation, and suppression of mitochondrial cell death to enable metastatic competence. This reinforces the role of mitochondrial reprogramming as an important therapeutic target in cancer (55), especially in hard-to-treat and metastasis-prone malignancies with currently limited therapeutic options.

Materials and Methods
Patient Samples. Primary patient samples arranged in a universal TMA were examined for differential Mic60 expression by IHC, as described (56). Archival tissues and clinical records were obtained from Fondazione Istituto di Ricovero e Cura a Carattere Scientifico (IRCCS) Ca' Granda Hospital in Milan, Italy, under a protocol approved by the Institutional Review Boards (IRBs) of Fondazione IRCCS Ca' Granda-Ospedale Maggiore Policlinico (code 179/2013). Because of the retrospective nature of this study and the use of data anonymization practices, the need for written informed consent was waived. For the glioma series, fresh-frozen material was available from Fondazione IRCCS Ca' Granda Hospital under IRB protocol 275/2013, and written informed consent from all patients was obtained before surgery. Clinically annotated patient samples with a confirmed histologic diagnosis of PDAC (n = 5) were obtained from the archival database of the Department of Pathology at Yale New Haven Hospital upon approval from the Yale University IRB and examined for intratumoral heterogeneity of Mic60 expression by IHC.
Proteomics. Immune complexes of Mic60 or nonbinding immunoglobulin G (IgG) were precipitated from PC3 cells and separated by sodium dodecyl sulphate gel electrophoresis for ∼5 mm followed by fixing and staining with colloidal Coomassie. The region of the gel-containing proteins was excised and digested with trypsin. Tryptic peptides were analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) on a Q Exactive highfield (HF) mass spectrometer (Thermo Scientific) coupled with a Nano-ACQUITY ultra performance liquid chromatography (UPLC) system (Waters  diameter [i.d.] × 2 cm packed with 5-μm C18 resin; Waters), and tryptic peptides were separated by RP-HPLC on a BEH C18 nanocapillary analytical column (75-μm i.d. × 25 cm, 1.7-μm particle size; Waters) using a 90-min gradient. Eluted peptides were analyzed in data-dependent mode where the mass spectrometer obtained full MS scans from 400 to 2,000 m/z at 60,000 resolution. Full scans were followed by MS/MS scans at 15,000 resolution on the 20 most abundant ions. Peptide match was set as preferred, and the exclude isotopes option and charge-state screening were enabled to reject singly and unassigned charged ions. MS/MS spectra were searched using MaxQuant 1.6.5.0 (57) against the Uni-Prot human protein database (October 2017). MS/MS spectra were searched using full tryptic specificity with up to two missed cleavages, static carbamidomethylation of Cys, variable oxidation of Met, and variable protein N-terminal acetylation. Consensus identification lists were generated with false discovery rates of 1% at protein and peptide levels. Undetected protein intensity values of 0 were floored to the value of 10 6 (minimum nonzero detected intensity was 1,233,300), and a total of 1,534 detected proteins were taken for further annotation analysis. Proteins were then annotated as mitochondrial related using the MitoCarta 2.0 database, and 119 mitochondrial proteins detected with at least 5 peptides at an intensity over 10-fold versus IgG control were considered as Mic60-associated proteins.
Mitochondrial Time-Lapse Videomicroscopy. Cells (2 × 10 4 ) growing on high optical quality glass-bottom 35-mm plates (MatTek Corporation) were incubated with 100 nM Mitotracker Deep Red FM dye for 1 h and imaged on a Leica TCS SP8 × inverted laser scanning confocal microscope using a 63× 1.40NA oil objective as described (58).
Single-Cell Motility. Cells (2 × 10 4 ) were seeded in 4-well Ph+ Chambers (Ibidi) in complete growth medium and allowed to attach for 16 h at 37°C. Timelapse videomicroscopy was performed over 10 h, with a time-lapse interval of 10 min. Stacks were imported into Image J Fiji software for analysis, and at least 10 to 20 cells per condition were tracked using the Manual Tracking plugin for Image J Fiji. Tracking data were exported into the Chemotaxis and Migration Tool v. 2.0 (Ibidi) for graphing and calculation of the mean and SD of speed and accumulated distance of movement. For directional cell migration, wounds were made in monolayers of PC3 cells using a 10-μL pipette tip. Cell debris were washed off, and cultures were maintained in complete growth medium containing 10% fetal bovine serum at 37°C and 5% CO 2 . Time-lapse imaging of migrating cells was performed using a TE300 inverted microscope (Nikon) equipped with an incubator set at 37°C, 5% CO 2 , and 95% relative humidity. Each image was acquired using a 10× objective of the same fields at each 10-min interval for a total of 24 h.
Small-Molecule Drug Screening. Cell viability screening against the MedChem Express anti-cancer library (1,820 compounds) was performed using CellTiter-Glo (Promega). PC3 cells stably transduced with pLKO or Mic60-directed shRNA (shMic60) were maintained in complete media, trypsinized, and plated (500 cells/well) in 40 μL of complete medium the day before the experiment in white, clear-bottom 384-well plates. A total of 50 nL of test compound was added to each well using the Janus MDT Nanohead (Perkin-Elmer). Each compound was screened at a final concentration of 10, 1, 0.1, and 0.01 μM. After a 72-h incubation at 37°C in the presence of 5% CO 2 , 20 μL of the CellTiterGlo reagent was added to each well. After 15 min, luminescence was measured using the Envision Multimode plate reader (PerkinElmer). The raw data were normalized to % inhibition, where 0% is the relative light units (RLU) in the presence of dimethylsulfoxide, and 100% is the RLU in the presence of 1 μM bortezomib. Estimated half maximal inhibitory concentration (IC 50 ) values for each compound were determined using nonlinear regression fits on the data to a one-site binding model in XlFit (ID Business Solutions Ltd. [IDBS]). Because only 4 data points were used in this calculation, the top and bottom of the curve was fixed to 100% and 0%, respectively, with a constant slope value of 1.
Statistical Analysis. Data are expressed as mean ± SEM or mean ± SD of multiple independent experiments or replicates of representative experiments out of a minimum of two or three independent determinations. Two-tailed Student's t test or Wilcoxon rank sum test was used for twogroup comparative analyses. For multiple-group comparisons, ANOVA or Kruskal-Wallis test with post hoc Bonferroni's procedure were applied. All statistical analyses and graphing were performed using GraphPad software package (Prism 9.0) for Windows. A P value of <0.05 was considered statistically significant.
Data Availability. All study data are included in the article and/or SI Appendix.