PINK1 drives production of mtDNA-containing extracellular vesicles to promote invasiveness

Tumor cells up-regulate glutaminolysis to support anaplerosis and growth while establishing cytoprotective programs, such as glutathione synthesis, to counteract metabolically stressful tumor microenvironments (Zhu and Thompson, 2019). Key to glutathione synthesis is the xCT cystine-glutamate antiporter (SLC7A11), which imports cystine at the expense of equimolar glutamate efflux (Combs and DeNicola, 2019), and this leads to increased glutamate in the extracellular milieu. Increased circulating glutamate has been reported in patients with cancer, and in the MMTV-PyMT autochthonous mouse model of mammary carcinoma using mice expressing the polyoma-middle-T (PyMT) oncogene under the control of the mammary epithelial mouse mammary tumor virus (MMTV) promoter, serum glutamate levels correlate closely with tumor burden (Budczies et al., 2015; Dornier et al., 2017). Extracellular glutamate activates metabotropic glutamate receptor 3 (mGluR3), and this drives endosomal trafficking of the membrane type 1 matrix metalloproteinase (MT1-MMP) to promote disruption of basement membranes, which contributes to invasiveness, and lung colonization by mammary carcinoma cells is opposed by i.p. administration of a small-molecule inhibitor of mGluR3 (Dornier et al., 2017).

Mobilization of endosomal MT1-MMP requires Rab27-dependent delivery of late endosomes to the plasma membrane (Dornier et al., 2017; Macpherson et al., 2014), and fusion of late endosomes with the plasma membrane also leads to release of extracellular vesicles (EVs; Ostrowski et al., 2010). EVs are important mediators of intercellular communication during cancer progression and can render certain organs, such as the lung and liver, more susceptible to metastatic seeding, a process that has been termed “metastatic niche priming,” due to recruitment of myeloid cells to, and on ECM deposition within, metastatic target organs (Wortzel et al., 2019). Thus, it is key to understand the mechanisms underlying how EVs mediate communication between cells and organs. A range of potential EV cargoes include oncoproteins, mRNAs, miRNAs, and DNA, and many of these may mediate transfer of various phenotypes between cancerous and other cells. However, before we can assemble a full picture of EV-mediated intercellular communication, it is necessary to understand the mechanistic details underlying the processes that enable this cascade, including identification of relevant transferrable EV cargoes and their sorting within donor cells, and determination of the cellular signaling systems that allow recipient cells to respond to tumor EVs. Mechanisms underlying some of these processes have been elucidated. Mutant p53-expressing pancreatic tumors influence the ECM microenvironment of the lung by altering sorting of podocalyxin into EVs (Novo et al., 2018). In turn, podocalyxin-containing EVs influence ECM deposition by altering integrin recycling in fibroblasts, highlighting the endosomal system of recipient cells as a key element of the cellular response to EVs.

In the present study, we report that the combination of increased extracellular glutamate and altered mitochondrial function leads to increased production of EVs, which promote endosomal trafficking and invasive behavior in recipient cells. We show that the mitochondrial chromosome is packaged into these EVs via a mechanism that involves phosphatase and tensin homolog–induced putative kinase 1 (PINK1)-dependent interaction of late endosomes with mitochondria. Finally, by reducing mtDNA in exosomes through the production of Rho0 cells, and by using synthetic unilamellar liposomes to mimic mitochondrial DNA (mtDNA)-containing EVs, we demonstrate that this cargo is necessary and sufficient to evoke trafficking and invasive responses in recipient cells via activation of Toll-like receptor 9 (TLR9).

Glutaminolysis drives invasive behavior by increasing extracellular glutamate, which activates mGluR3 to promote Rab27-dependent trafficking of MT1-MMP (Dornier et al., 2017), and Rab27s also control EV release (Ostrowski et al., 2010). Therefore, we used differential centrifugation followed by nanoparticle tracking analysis (NTA) and Western blotting to determine whether mGluR3 drives EV release. Treatment of MDA-MB-231 breast cancer cells with the mGluR3 antagonist LY341495 (LY95) reduced EV release (Fig. 1 a). Consistently, following glutamine starvation (to deplete extracellular glutamate), addition of glutamate to the extracellular medium increased release of EVs (Fig. 1 b). To further establish a role for mGluR3 in EV release, we used immunogold labeling for the EV marker CD63 in combination with transmission EM. This indicated that CD63-positive EVs released by MDA-MB-231 cells were ∼100 nm in diameter and that mGluR3 inhibition decreased release of these (Fig. 1 c). EV release was also opposed by combined siRNA of Rab27a and Rab27b (Fig. S1 a). In addition to being a marker for EVs, CD63 plays an active role in intraluminal budding from multivesicular bodies (MVBs; van Niel et al., 2011). Consistently, siRNA of CD63 reduced EV release from MDA-MB-231 cells (Fig. S1 b).

We considered whether mGluR3 might influence oxidative respiration. A 5-d exposure to LY95 significantly reduced basal and maximum oxygen consumption rate (OCR; Figs. 2 a and Fig. S2 a) without altering the mitochondrial mass or membrane potential (Fig. S2 b). A shorter exposure (18 h) to LY95 was ineffective in reducing OCR, indicating that defects in mitochondrial respiration need several days to accrue following inhibition of mGluR3 (Fig. S2 c). To determine whether mGluR3’s influence on oxidative respiration was mediated via control of endosomal function and/or EV production, we measured OCR following Rab27 or CD63 knockdown (Fig. S2 d). OCR, but not mitochondrial mass, was reduced following siRNA of Rab27 or CD63 for 5 d, with CD63 knockdown being most effective (Fig. 2 a; and Fig. S2, a and b). These data indicate that proper function of the late endosomal/MVB network and EV production is required to maintain mitochondrial function.

Mitochondrial proteins are commonly present in EV proteomes (Choi et al., 2015). Western blotting indicated that mitochondrial components such as glutamate dehydrogenase 1 (GLUD1; matrix), voltage-dependent anion channel 1 (VDAC; outer membrane), and cyclophilin D (matrix) are abundant in EV preparations. Consistently, addition of LY95 reduced the quantity of mitochondrial proteins associated with EV preparations (Fig. 2 b). Mitochondria possess 16-kbp circular genomes containing genes encoding respiratory chain subunits, mitochondria-specific tRNAs, and mitochondrial ribosomal RNAs. Quantitative PCR (qPCR) indicated that a range of mitochondrial genes were detectable in EV preparations (Fig. S3 a), and long-range PCR confirmed the presence of the intact mitochondrial chromosome (Fig. 2 c). mtDNA associated with EV preparations (including genes encoding tRNA-Leu, cytochrome c oxidase I [COX1], NADH dehydrogenase 1 [ND1], ND6, cytochrome b [CYB], and mtDNA nt positions 8339–9334) was more resistant to DNase than DNA encoding a nuclear gene (β-globin), indicating that the mitochondrial genome is preferentially packaged into the EV lumen (Fig. 2 d; Fig. S3 a). Consistently, inhibition of mGluR3 and knockdown of Rab27 or CD63 reduced release of mtDNA associated with EV preparations (Fig. 2 e and Fig. S3 b). Finally, we used sucrose density gradient centrifugation to confirm that mtDNA was associated with CD63-positive vesicles with a flotation density of 1.1–1.15 g/ml, consistent with that of EVs, and treatment with LY95 reduced this (Fig. 2 f).

PINK1 becomes stabilized at the surface of damaged/uncoupled mitochondria, where it phosphorylates and activates the E3-ubiquitin (E3-Ub) ligase Parkin and Ub moieties attached to mitochondrial proteins. This initiates phosphorylation/ubiquitylation cascades that promote mitophagy (Koyano et al., 2014; Lazarou et al., 2015; McWilliams and Muqit, 2017). Consistent with previous observations (Narendra et al., 2010), treatment with the mitochondrial uncoupler carbonyl cyanide 3-chlorophenylhydrazone (CCCP) elevated PINK1 levels and PINK1-dependent phospho-Ub levels (Fig. S4, a and b). We confirmed that prolonged suppression of PINK1 reduced basal and maximal OCR (Fig. S4 c). We then considered whether PINK1 might contribute to production of mtDNA-containing EVs. siRNA of PINK1 reduced release of mtDNA-containing EVs, as determined by differential centrifugation and sucrose density gradient centrifugation followed by NTA, Western blotting, and qPCR for the mitochondrial gene ND1 (Fig. 3, a and b). Furthermore, we confirmed that CRISPR depletion of PINK1 opposed release of CD63-postive EVs from cells derived from a mouse mammary carcinoma driven by the PyMT oncogene (Fig. S4 d). Finally, we transfected PINK1 knockdown cells with a “rescue” vector encoding an siRNA-resistant YFP-tagged PINK1 (PINK1-YFP; Lazarou et al., 2015). Indeed, expression of PINK1-YFP rescued CCCP-driven levels of phospho-Ub and EV release following PINK1 knockdown (Fig. 3, c and d).

We investigated whether increased PINK1 levels might be sufficient to drive EV release. Addition of CCCP (to increase PINK1 levels) increased EV release (Fig. 4 a), and this was opposed by siRNA of PINK1 or CD63 (Fig. 4 b). Overexpression of PINK1-YFP (Fig. 4 c) led to an increase in the release of EVs associated with mtDNA (Fig. 4, d and e). Moreover, a proportion of this EV-associated mtDNA is resistant to DNase and therefore is likely to be contained within these PINK1-driven EVs (Fig. 4 e).

We hypothesized that PINK1-dependent interaction between mitochondria and late endosomes/MVBs might underpin transfer of mitochondrial material to EVs. We used time-lapse fluorescence microscopy to determine the relationship between mitochondria (DsRed-mito) and late endosomes/MVBs (CD63-GFP). Mitochondria and late endosomes formed regular contacts that persisted for an average of ~70 s (Fig. 5 a; and Videos 1, 2, 3, and 4). The frequency of contacts was opposed by siRNA of PINK1 (Fig. 5 b). Moreover, the duration of contacts formed between DsRed-mito and CD63-GFP–positive structures was significantly increased by addition of CCCP and opposed by siRNA of PINK1 (Fig. 5 b). These data establish a role for PINK1 in promoting association between mitochondria and late endosomes/MVBs.

The autophagy machinery has recently been shown to regulate a novel secretory pathway, termed “microtubule-associated protein-1 light chain 3 (LC3)-dependent EV loading and secretion,” which leads to release of EVs containing RNA-binding proteins (Leidal et al., 2020). Given PINK1’s role in mitophagy, an autophagic process, we determined whether the autophagy regulators autophagy-related 5 (Atg5) and Atg7 and the mitophagy gatekeeper FAK family interacting protein of 200 kD (FIP200; Vargas et al., 2019) might contribute to EV production. However, neither siRNA of Atg5/7 (Fig. S5, a and b) nor suppression of FIP200 (Fig. S5, c and d) opposed release of EVs.

PINK1’s contribution to mitophagy is via phosphorylation of Parkin and the Ubs conjugated to mitochondrial proteins, which then engage the autophagy machinery (Koyano et al., 2014; Lazarou et al., 2015; McWilliams and Muqit, 2017). Given our finding that release of EVs associated with mtDNA is not ATG5/7 dependent, we wished to determine whether there is a role for PINK’s kinase activity in EV release. We transfected PINK1 knockdown cells with a PINK1-YFP rescue vector containing three point/substitution mutations (K219A, D362A, and D384A) to render the protein kinase dead (Beilina et al., 2005). Surprisingly, expression of this kinase-dead PINK1 completely restored EV release in PINK1 knockdown cells, despite the fact that this construct was unable to rescue CCCP-driven phospho-Ub levels following siRNA of PINK1 (Fig. 3, c and d). Taken together, these data indicate that PINK1 promotes physical interaction of late endosomes with mitochondria and contributes to release of mtDNA-containing EVs via a mechanism that is distinct from PINK’s canonical, kinase-dependent role in mitophagy.

EVs can alter cell migration in recipient cells by promoting endosomal trafficking (Novo et al., 2018). We determined whether EVs released from “donor” cells might influence trafficking of MT1-MMP, a receptor linked to invasiveness, in “recipient” cells. We collected conditioned media from donor cells in the presence and absence of LY95, and we purified EVs from these. EVs were then incubated with mCherry-MT1-MMP–expressing recipient cells that had previously been starved of glutamine to deplete extracellular glutamate and suppress MT1-MMP trafficking (Fig. 6 a). Trafficking of mCherry-MT1-MMP vesicles to the ventral plasma membrane of recipient cells was then determined using total internal reflection fluorescence (TIRF) microscopy (Koyano et al., 2014; Lazarou et al., 2015; McWilliams and Muqit, 2017). EVs from glutamine-replete donor cells increased delivery of mCherry-MT1-MMP–positive vesicles to the plasma membrane of recipient cells, whereas EVs from LY95-treated donors were less effective in this regard (Fig. 6 b). Consistently, donor cell EVs enabled glutamine-starved recipient cells to execute MT1-MMP–dependent invasive processes, such as collagen degradation (Fig. 6 c) and invasion, into 3D fibroblast-derived ECM (Fig. 6 d). By contrast, EVs from donor cells in which mGluR3 was inhibited or PINK1 was knocked down were unable to drive collagen degradation or invasive behavior (Fig. 6, c and d). Consistently, suppression of Rab27 or CD63 in donor cells also rendered EV preparations to be ineffective in driving invasiveness in recipient cells (Fig. 6, c and d).

Although DNase protection approaches indicated that a higher proportion of EV-associated mtDNA than nuclear DNA is likely to be encapsulated within the EVs, we wished to determine the specificity of mtDNA in the ability of EVs to evoke invasive responses from recipient cells. We generated mtDNA-deficient “Rho0” cells and found that these release quantities of exosomes similar to those of control cells, but with reduced mtDNA (but not nuclear DNA) content (Fig. 7, a and b). Interestingly, exosomes from Rho0 cells are unable to promote collagen degradation (Fig. 7 c) or to drive extension of invasive protrusions (Fig. 7 d) when they are incubated with glutamine-starved recipient cells. These data indicate that it is specifically the mitochondrial (as opposed the nuclear) DNA associated with tumor EVs that is required for these EVs to evoke invasive responses.

The TLR family mediates responses to moieties associated with infective agents and cell damage, and certain TLRs are localized to endosomes (Luo et al., 2020). Because TLR9 is an endosomally localized receptor that is activated by double-stranded DNA, this is a good candidate for mediating cellular responses to the mtDNA associated with tumor cell EVs. Direct activation of TLR9 using a specific agonist (oligodeoxynucleotide [ODN] 2006) drove collagen degradation and extension of invasive protrusions in glutamine-starved cells (Fig. 8, a and b). Consistently, pharmacological blockade (using ODN-TTAGGG-A151) of TLR9 opposed the ability of mtDNA-containing EVs to drive extension of invasive protrusions (Fig. 8 c). Moreover, siRNA of TLR9 EVs inhibited invasive behaviors, including extension of protrusions (Fig. 8 c) and collagen degradation (Fig. 8 d), evoked by tumor cell EVs in glutamine-starved recipient cells. These data indicate that cancer cells (in which glutaminolysis is active) release mtDNA-containing EVs that transfer endosomal trafficking and invasive characteristics to glutamine-starved cells via activation of TLR9.

EVs contain several cargoes, including mtDNA, with potential to activate TLRs. We adopted a reconstitution approach to determine whether the mtDNA was itself sufficient to drive endosomal trafficking and invasive behavior in recipient cells. We constructed unilamellar liposomes with dimensions similar to those of EVs (90–120-nm diameter), loaded them with mtDNA, and treated them with immobilized DNase to remove DNA that was not lipid encapsulated (Fig. 9, a and b). Using capture ELISA recycling assays and TIRF microscopy, we determined that mtDNA-loaded liposomes promoted recycling of pro-invasive receptors, including MT1-MMP and α₅β₁ integrin, from endosomes to the plasma membrane, whereas control liposomes were ineffective in this regard (Fig. 9 c). Moreover, EV-induced delivery of MT1-MMP–containing vesicles to the ventral plasma membrane was opposed by pharmacological inhibition of TLR9 (Fig. 9 d). Consistently, mtDNA-loaded liposomes restored invasive behaviors to glutamine-starved cells, and this was opposed by inhibition or siRNA of TLR9 (Fig. 9 e).

These data indicate that mtDNA is packaged via a PINK1-dependent mechanism into EVs released by breast cancer cells and that mtDNA is a key EV cargo both necessary and sufficient to transfer glutaminolysis-driven invasive phenotypes between cells.

We have identified a means of intercellular communication in which processes evoked to mitigate cytotoxicity in metabolically stressed cells can evoke invasive behavior in other cells. This proceeds via three sequential mechanistic events. First, PINK1, activated in response to mitochondrial damage, promotes physical interaction of late endosomes with mitochondria. Second, increased secretion of glutamate activates mGluR3 and drives Rab27-dependent exocytosis of EVs loaded with mtDNA. Third, mtDNA transported within EVs activates a TLR9-dependent mechanism to promote pro-invasive endosomal trafficking in other cells (Fig. 10).

Communication of metabolic information to nearby cells may be conveyed in several ways. Metabolites can be exchanged between cells in proximity. Glycolytic and glutaminolytic tumor cells generate lactate and glutamate, which are exported by monocarboxylate transporters and the xCT antiporter, and these can be assimilated by neighboring cells (Doherty and Cleveland, 2013; Sonnewald and Schousboe, 2016; Whitaker-Menezes et al., 2011). Extracellular metabolites may influence other cells by interacting with surface receptors, such as GPR81 lactate receptor, GPR91 for succinate, and mGluRs for glutamate (Ristic et al., 2017). Mice bearing invasive MMTV-PyMT–driven carcinoma display elevated serum glutamate (Dornier et al., 2017), indicating that tumor metabolites are not locally restricted, introducing the possibility that tumor metabolites can evoke systemic responses. Glutamine metabolism can impact the release of EVs with pro-tumorigenic properties (Fan et al., 2020), and EVs have a well-established role in mediating communication between tumors and metastatic target organs (Costa-Silva et al., 2015; Novo et al., 2018; Peinado et al., 2012). Our findings highlight the likelihood that tumor cells experiencing metabolic stress use EVs to communicate their metabolic state to other cells within the tumor and farther afield.

Donor cells influence recipient cell signaling and sorting of protein and nucleic acid cargoes into EVs. Proto-oncogenes and oncogenic receptors, such as EGF receptor and c-MET (Fan et al., 2020; Peinado et al., 2012; Zhang et al., 2017), can evoke transformed phenotypes in recipient cells, but how these receptors attain the appropriate membrane localization and topology in recipient cells is unclear. Annexin-6 and podocalyxin are EV cargoes known to promote metastatic niche priming. Annexin-6 promotes CCL2 induction in neutrophils (Keklikoglou et al., 2019), but details of how this cargo is sorted into EVs and then interfaces with recipient cell signaling machinery are not known. Mechanistically better defined is podocalyxin’s sorting into cancer cell EVs, which is controlled by Rab35, and this influences ECM deposition in metastatic target organs by influencing Rab-coupling protein–dependent integrin trafficking machinery in fibroblasts (Novo et al., 2018). miRNAs may alter expression of their cognate targets in recipient cells (Cooks et al., 2018), although to do this, they must access the mRNA processing and/or translational machinery in the cytosol by as yet undefined mechanisms. Extracellular nucleic acids evoke cellular responses by engaging with pattern recognition receptors, and miRNAs associated with tumor EVs have been shown to activate pattern recognition receptors. EVs from pancreatic cancer cells regulate TLR4 in dendritic cells via miR-203 (Zhou et al., 2014). However, although a recent report has identified the LC3-conjugation machinery as key to sorting a cohort of RNA-binding proteins and small noncoding RNAs into EVs (Leidal et al., 2020), metabolic landscapes that might influence packaging of nucleic acid cargoes into EVs are yet to be determined. Macrophages and dendritic cells express high levels of TLRs, indicating the likelihood that PINK1-dependent EV release offers a mechanism by which tumor cells undergoing mitochondrial stress may communicate with the innate immune system, and this will be an interesting line of investigation in the future.

When DNA is present in the cytosol, resulting from viral infection or following its (aberrant) release from the nucleus or mitochondria, this leads to immune and inflammatory responses by activating cytosolic DNA-sensing pathways such as cyclic GMP-AMP synthase (cGAS)/stimulator of interferon genes (STING; Riley et al., 2018) and AIM2 (absent in melanoma 2; Dang et al., 2017). Recently, activated T cells were shown to release mtDNA-containing EVs to stimulate cGAS/STING in dendritic cells (Torralba et al., 2018), but how EV-associated mtDNA can access the cytoplasmically located cGAS/STING machinery still presents a topological puzzle. Interestingly, these investigators reported that EV-associated mtDNA was more oxidized than endogenous mtDNA, suggesting that EV release from T cells may be linked to the altered redox metabolism. Cancer cells have also been found to release EVs in response to stresses associated with altered redox metabolism. Thermal and oxidative stress induces EV release from cancer cells (Hedlund et al., 2011; Wang et al., 2014). Moreover, EVs are released from breast tumors in response to treatment with taxanes and anthracyclines, forming a mechanism through which chemotherapeutics promote metastatic niche priming (Keklikoglou et al., 2019). However, although these investigators demonstrated a requirement for Rab27 in this process, mechanisms linking chemotherapy-induced metabolic stresses with EV release have not been addressed. Our data provide a mechanism for how a nucleic acid cargo (mtDNA) is packaged into EVs, and we provide insight into the metabolic landscapes that promote this. In the future, it will be interesting to investigate the role played by xCT-driven glutamate fluxes and PINK1 in EV release following chemotherapy and to determine whether the packaging of mtDNA into EVs influences metastatic niche priming following chemotherapy.

Mitochondria have been shown to be released from cells. Larger vesicles containing damaged mitochondria are released by Caenorhabditis elegans neurons and murine cardiomyocytes (Melentijevic et al., 2017; Nicolás-Ávila et al., 2020). More recently, neutrophils were reported to leave damaged mitochondria behind them as they migrate, and this has been suggested to constitute an alternative to mitophagy for maintenance of mitochondrial homeostasis and quality control (Jiao et al., 2021). However, it is likely that cells export mitochondria from cells for reasons other than to maintain mitochondrial homeostasis. Intercellular transfer of functional mitochondria has been reported to increase aerobic respiration in recipient cells (reviewed in Liu et al., 2021). EVs produced by stromal cells may transfer functionally intact mtDNA to cancer cells with impaired metabolism, allowing restoration of metabolic activity and release from metabolic quiescence (Sansone et al., 2017). Although we agree with these investigators that the intact mitochondrial genome is packaged into EVs, we have been unable to demonstrate any EV-mediated intercellular transfer of mitochondrial function. Indeed, we have purified EVs from cells with WT mitochondria and incubated these with recipient cells harboring a loss-of-function mutation in the mitochondrial ATP synthase gene. In these experiments, EVs from cells with WT mitochondria did not rescue metabolic defects of cells with mutant mitochondria. Thus, we conclude that a primary function of PINK1-dependent sorting of mtDNA into EVs is to influence pro-invasive endosomal trafficking in recipient cells (via activation of TLR9), thus engaging in intercellular communication without direct transfer of functional mitochondrial components or genetic information.

PINK1 has an established role in mitochondrial homeostasis, and this is achieved via its ability to phosphorylate Parkin and Ub to increase phospho-Ub levels at the mitochondrial surface (Koyano et al., 2014; Lazarou et al., 2015; McWilliams and Muqit, 2017). This triggers recruitment of the ULK1 complex via the Ub-binding adaptor, FIP200 (Vargas et al., 2019). ATG proteins then mediate assembly of the autophagosome to encapsulate the damaged mitochondrion (Koyano et al., 2014; Lazarou et al., 2015; McWilliams and Muqit, 2017). Involvement of PINK1 in packaging of mtDNA into EVs suggests that mitophagy and mtDNA-containing EV production may share features. However, PINK1’s kinase activity and the mitophagy machinery (ATG5/7 and FIP200) are not required for release of mtDNA-containing EVs. Consistently, a recent study has shown that mitochondrial material and mtDNA is routinely packaged into released EVs only when Parkin expression is absent (Todkar et al., 2021). Taken together, these observations indicate that PINK1 contributes to production of mtDNA-containing EVs in a way that is mechanistically distinct from its role in mitophagy. Also, although the two processes both involve congress between mitochondria and late endosomes/lysosomes, they do not share the same membrane topology. More topologically aligned to packaging of mtDNA into EVs is production of mitochondria-derived vesicles (MDVs). Vesicles are established to bud from mitochondria in a PINK1- and Parkin-dependent, but ATG5/7-independent, manner and these are trafficked to various intracellular locations, including the peroxisomes and late endosomes/lysosomes (Neuspiel et al., 2008; Sugiura et al., 2014; Yang and Yang, 2013). These connections to oxidative and degradative organelles have prompted proposals that MDV production may be an early response to oxidative stress that assists with reconditioning of damaged mitochondria in a manner more selective than mitophagy (McLelland et al., 2014; Soubannier et al., 2012). Our data clearly indicate that knockdown of Rab27 and CD63, factors with established roles in the production, plasma membrane delivery, and release of EVs, leads to attrition of mitochondrial function. Therefore, the likelihood that PINK1, acting in a kinase-independent manner, in conjunction with the endosomal regulators, CD63 and Rab27, to promote packaging of mitochondrial components into EVs may function in piecemeal or bit-by-bit mitophagy (as has been proposed for MDV production) must be considered, and it will be interesting to determine how PINK1 achieves this by acting as an adaptor rather than as a kinase.

Therapeutic responses and relapse of tumors with known oncogenic mutations, such as those in KRAS and p53, may be monitored using PCR detection of circulating DNA (Kilgour et al., 2020). Furthermore, circulating DNA increases following metabolic insults, such as paracetamol-induced hepatic injury, and this may be used to monitor these conditions (Laurent et al., 2020). However, the use of circulating DNA as an early detection marker in cancer is limited because many of the mutations occurring in tumors are present in normal epithelia (Martincorena et al., 2018) and are not necessarily good markers of malignant or dysplastic disease. Our data indicate that adaptations enlisted by tumors to counter redox stress may be reflected in altered circulating factors. For instance, the glutamate release that results from up-regulation of the xCT transporter leads to detectably increased circulating glutamate in the serum in tumor-bearing subjects (Dornier et al., 2017). The present study provides a mechanistic rationale connecting increased glutamate secretion to release of mtDNA-containing EVs and demonstrates how these events are linked to invasive characteristics. Moreover, a recent study indicating that a number of tumor types harbor particular mutations in mtDNA may assist in identification of mtDNA-containing EVs in the circulation that have emanated from invasive cancer cells (Gorelick et al., 2021). Therefore, it will be interesting to determine the potential for combining measurement of circulating glutamate levels and EV-encapsulated mtDNA into a biomarker strategy for detection of breast cancer and for distinguishing indolent from more aggressive disease.

MDA-MB-231 cells were maintained in DMEM (Life Technologies) supplemented with 10% FBS, 2 mM glutamine (unless otherwise stated), and antibiotics. A primary mouse breast cancer cell line was isolated from mammary tumors of mice expressing the PyMT oncogene under the control of the mammary epithelial MMTV promoter. Tumors from mice expressing the MMTV-PyMT transgene were surgically excised and minced using a McIlwain tissue chopper. Minced tumor specimens were transferred to DMEM supplemented with 10% FBS, 20 ng/ml EGF, 10 µg/ml insulin, 2 mM glutamine, and antibiotics and kept in culture for up to 20 passages.

The following antibodies were used for Western blotting and diluted 1:1,000 unless otherwise stated: hCD63 (BD Biosciences; 556019), mCD63 (Abcam; ab217345), CD9 (Cell Signaling Technology; 13174S), flotillin (Cell Signaling Technology; 18634), β-actin (Sigma-Aldrich; 1:5,000, 1978), Rab27A (Abcam; 55667), VDAC (Abcam; ab14734), cyclophilin D (Abcam; ab110324), GLUD1 (Abcam; ab153973), PINK1 (Novus Biologicals; BC100-494), LC3B (Cell Signaling Technology; 2775), and FIP200 (Proteintech; 1:2,000, 17250-1-AP). The following antibodies were used for ELISA-based recycling assays at a concentration of 5 µg/ml in 0.05 M Na₂CO₃, pH 9.6, at 4°C: MT1-MMP (MilliporeSigma; mab3328), α₅-integrin (BD Pharmingen; 555651), and active β1 integrin (BD Biosciences; 09351D).

Plasmid expression vectors used included the following: mCherry-MT1-MMP, kindly gifted by Philippe Chavrier (Institut Curie, Paris, France), and CD63-GFP as previously described (Diaz-Vera et al., 2017). Expression vectors encoding PINK1-YFP (WT and kinase-dead; Lazarou et al., 2015) were kindly donated by Richard Youle (National Institutes of Health, Bethesda, MD). Rescue vectors for PINK1-YFP were generated by introducing silent mutations in the region where PINK1 siRNA oligo 3 (Dharmacon; J004030-08-0005) binds. Mutations were introduced using the QuikChange Multi Site-Directed Mutagenesis Kit (Agilent; 210515) using the following primers: forward: 5′-GTA​CAG​GGA​TAG​TTC​TTC​ATG​ACT​AGA​AAT​AGT​GTC​CGG​CCA​TGG​CCC​AGG​C-3′; and reverse: 5′-GCC​TGG​GCC​ATG​GCC​GGA​CAC​TAT​TTC​TAG​TCA​TGA​AGA​ACT​ATC​CCT​GTA​C-3′.

TLR9 reagents (InvivoGen) were all used at 1 µM: TLR9 agonist ODN 2006 (ODN 7909) and ODN 2006 control (ODN 2137), as well as TLR9 antagonist ODN TTAGGG (A151) and ODN TTAGGG control. Drug treatments were as follows with the working concentration in parentheses: group II glutamate metabotropic receptor agonist LY95 (100 nM; Tocris) and CCCP (10 µM; Sigma-Aldrich).

siRNA oligonucleotides for Rab27, CD63, PINK1, FIP200, TLR9, ATG5, and ATG7 were all derived from SMARTpool (Dharmacon). For certain experiments (see Fig. 8 d), “stealth” siRNAs (Thermo Fisher Scientific/Life Technologies) were used to knock down TLR9. CRISPR/Cas9 knockout of PINK1 was performed as previously described (Novo et al., 2018), and the guide sequences were as follows: PINK1 1 CRISPR guide forward: 5′-CAC​CGC​AAG​CGC​GTG​TCT​GAC​CCA​C-3′; reverse: 5′-AAA​CGT​GGG​TCA​GAC​ACG​CGC​TTG​C-3′; PINK1 2 CRISPR guide forward: 5′-CAC​CGT​GTC​GGA​CTT​GAG​ATC​CCG​A-3′; reverse: 5′-AAA​CTC​GGG​ATC​TCA​AGT​CCG​ACA​C-3′. Transient transfections of were performed using the AMAXA transfection system (Lonza) according to the manufacturer’s instructions. Approximately 5 × 10⁶ cells were used for each transfection with 100 µl AMAXA solution V plus 5 µl siRNA reagent at 20 µM, or with 0.5 µg plasmid construct, and they were electroplated using the X-013 program. For overexpression of PINK1, cells were transfected with 2 µg expression vector. For rescue experiments, 2 µg of the appropriate expression vector was combined with 5 µl of 20 µM PINK1 3 siRNA (Dharmacon).

qPCR and long-range PCR analysis was performed on DNA extracted from cells/EVs/liposomes using the GeneJET Genomic DNA Purification Kit (Thermo Fisher Scientific) with primers obtained from Invitrogen. Human gene primers used were as follows: ND1 forward: 5′-CGA​AAG​GAC​AAG​AGA​AAT​AAG​C-3′; ND1 reverse: 5′-CTG​TAA​AGT​TTT​AAG​TTT​TAT​GCG-3′; β-globin forward: 5′-CCA​CTT​CAT​CCA​CGT​TCA​CC-3′; β-globin reverse: 5′-GAA​GAG​CCA​AGG​ACA​GGT​AC-3′; tRNA-leu forward: 5′-CAC​CCA​AGA​ACA​GGG​TTT​GT-3′; tRNA-leu reverse: 5′-TGG​CCA​TGG​GTA​TGT​TGT​TA-3′; COX1 forward: 5′-CCA​ATA​CCA​AAC​GCC​CCT​CT-3′; COX1 reverse: 5′-AGA​ATG​GGG​TCT​CCT​CCT​CC-3′; ND6 forward: 5′-AAT​CAT​ACA​AAG​CCC​CCG​CA-3′; ND6 reverse: 5′-TGG​GGT​TAG​CGA​TGG​AGG​TA-3′; CYB forward: 5′-ATC​ACT​CGA​GAC​GTA​AAT​TAT​GGC​T-3′; CYB reverse: 5′-TGA​ACT​AGG​TCT​GTC​CCA​ATG​TAT​G-3′; and mtDNA nt positions 8339–9334 forward: 5′-CCA​CCC​AAC​AAT​GAC​TAA​TC-3′; mtDNA nt positions 8339–9334 reverse: 5′-GTA​TGA​GGA​GCG​TTA​TGG​AG-3′. Mouse gene primers used were as follows: ND1 forward: 5′-CCT​ATC​ACC​CTT​GCC​ATC​AT-3′; ND1 reverse: 5′-GAG​GCT​GTT​GCT​TGT​GTG​AC-3′; and genomic DNA forward: 5′-CAG​CGC​TAG​AGT​GAG​CGG​ATT​A-3′; genomic DNA reverse: 5′-GAA​TTG​ATC​AGT​CTG​CAG​ACA​GCC-3′.

qPCRs were prepared using the SYBR Green qRT-PCR kit (Qiagen), and amplified products were analyzed by a CFX96 Touch Real-Time PCR Detection System (Bio-Rad Laboratories) and CFX manager software from which the comparative cycle threshold was determined and presented as the fold change relative to control.

For long-range PCR, 100 ng of each sample was amplified in a 25-µl reaction containing 0.5 µl Takara PrimerSTAR GXL DNA polymerase and 5 µl of the corresponding buffer, 2 µl deoxynucleoside triphosphate, 0.5 µl mtDNA forward sequence 5′-TTA​AAA​CTC​AAA​GGA​CCT​GGC-3′ (10 µM), and 0.5 µl reverse primer sequence 5′-AGG​GTG​ATA​GAG​CTG​TGA​TC-3′ (primer stock concentration 10 µM). DNA was amplified under the following conditions: 94°C for 1 min, then 30 cycles of 98°C for 10 s and 68°C for 15 min, followed by 72°C for 10 min. The product was run at 45 V for 3 h in 0.8% agarose gel alongside lambda HindIII digest ladder as per the manufacturer’s instructions (New England Biolabs).

MDA-MB-231 cells or primary mouse mammary tumor cells (after transfection with siRNA if required) were plated into 15-cm tissue culture dishes at 2 × 10⁶ cells/dish. Where indicated, cells were fed with glutamine-free media, and 4 × 10⁶ cells/dish were plated. After 48 h, cells were washed with PBS, and 15 ml of EV-free DMEM was added. EV-free media was obtained by overnight centrifugation at 100,000 g. After 48 h, EVs were collected via differential centrifugation from the media as described previously (Novo et al., 2018). The volume of 0.02-µm filtered PBS used to resuspend final EV pellets depended on the original volume of conditioned media as follows: For instance, 22 µl of filtered PBS was used per 15-cm dish from which the EVs were collected. All EV isolations were stored at 4°C and used within 48 h.

NTA was performed using the NanoSight LM10 instrument (Malvern Panalytical) according to the manufacturer’s instructions. Resuspended EVs were diluted 1:50 in filtered PBS before being introduced into the instrument for measurement. Sucrose density gradient centrifugation was performed as described previously (Novo et al., 2018).

OCRs were determined using the Seahorse Extracellular Flux Analyzer (Agilent) according to the manufacturer’s instructions. Briefly MDA-MB-231 cells were plated at 3 × 10⁴ cells/well in a 96-well plate, and assay medium was prepared with phenol-free DMEM. OCR was measured during stepwise injection of 1 µM oligomycin, 1 µM carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP), and 1 µM rotenone/antimycin A (Sigma-Aldrich). After completion, cells were lysed with radioimmunoprecipitation assay buffer, and OCR was normalized to protein content determined using a Pierce bicinchoninic acid protein assay kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. Basal and maximal OCRs were calculated by subtracting the average post-rotenone/antimycin A OCR value from the average initial OCR value (for basal respiration) and the post-FCCP value (for maximal respiration). To calculate basal OCR, the mean post-rotenone/antimycin A values were subtracted from the mean preoligomycin values. For maximum respiration rate, the mean post-rotenone/antimycin A values were subtracted from the mean post-FCCP values.

To calculate mitochondrial mass, cells were seeded at 3 × 10⁵ cells/well in six-well tissue culture dishes, and, after 16 h, they were pretreated with the fluorescent mitochondrial probe MitoTracker Green (100 nM; Thermo Fisher Scientific) for 30 min, and the mean fluorescence intensity per cell was determined using FACS. Cells were detached by trypsin and washed in PBS before resuspension in 1 ml PBS and transferred to FACS tubes (STEMCELL Technologies). Samples were imaged using the FACSCalibur machine (BD Biosciences) and analyzed using FlowJo single-cell analysis software to determine the average per-cell fluorescence intensity.

MDA-MB-231 cells were treated with LY95 (for 3 or 5 d) and plated into six-well dishes (3 × 10⁵ cells/well). Cells were then incubated overnight in the presence or absence of CCCP (10 µM). Media were then supplemented with tetramethylrhodamine ethyl ester (TMRE; 50 µM, ab113852; Abcam) for 30 min at 37°C. Cells were then detached by trypsinization, resuspended in 1 ml PBS containing DAPI and analyzed using flow cytometry. Data were analyzed using FlowJo software with gates positioned to remove duplicate cells and dead cells (via positive DAPI signal). The mean fluorescence of TMRE was then plotted as the fold change compared with untreated control.

MDA-MB-231 cells pretreated with EVs or liposomes were transfected with mCherry-MT1-MMP plasmid plated directly onto 3.5-cm glass-bottomed plates, which were precoated with fibronectin (Sigma-Aldrich), diluted in PBS to 50 µg/ml. After 18 h, the cell media were changed to DMEM lacking glutamine plus treatments including EVs, liposomes, and/or drug treatments for at least 1 h before imaging in a 5% CO₂ atmosphere at 37°C using either a 100× or 60× plan apochromat TIRF objective on a Nikon Eclipse Ti microscope. Images were collected every 1 s for 5 min as time-lapse videos and analyzed in ImageJ using the TrackMate plugin (Tinevez et al., 2017) to determine the appearance of fluorescent puncta per minute of each video. Kymographs were obtained via the orthogonal view of each video adjusted to 40 µm.

2 d before imaging, MDA-MB-231 cells were transfected with DSRed-mito (Clontech) and GFP-CD63 expression plasmids with either nontargeting control or PINK1 targeting SMARTpool siRNAs (Dharmacon). The following day, cells were detached and plated onto fibronectin-coated 3.5-cm glass-bottomed dishes. At least 3 h before imaging, media were changed to full DMEM. CCCP (10 μM) or DMSO was added 1 h before image collection.

Images were acquired on a Zeiss LSM 880 Airyscan confocal microscope in Airyscan Fast mode. To minimize bleed-through, channels were imaged separately. Cells were imaged in an environmental chamber maintained at 37°C and 5% CO₂. For each time-lapse series, 500 frames were collected with an interval of 0.59 s. Images were then processed using the Airyscan processing function of the ZEN software.

For each distinct mitochondrion, the mitochondrial area encompassed at the beginning of the video (t = 0 s), together with the number and duration of contacts made with CD63-GFP–positive endosomes, was recorded. For each cell, all mitochondria whose envelope could be distinguished at t = 0 were analyzed, and values were normalized to the area of each mitochondrion. ImageJ Fiji software was used for measurements. Perinuclear mitochondria and CD63-GFP–containing endosomes formed large static clumps that could not easily be distinguished. These were unaltered by transfection with PINK1-targeting siRNA oligonucleotides and were omitted from analysis.

For pseudopod invasion assays, MDA-MB-231 cells were seeded at 8 × 10⁴ cells/well into fibroblast-derived matrices made as described previously (Cukierman et al., 2001) using human immortalized fibroblasts prepared as described elsewhere (Timpson et al., 2011). 4 h later, cells were imaged using time-lapse phase-contrast microscopy (Axiovert S100; Carl Zeiss Microscopy; 10× objective) in an atmosphere of 5% CO₂ at 37°C as described previously (Rainero et al., 2012). Pseudopod lengths were measured using ImageJ software. Fluorescent gelatin degradation assays were performed as described previously (Dornier et al., 2017).

MDA-MB-231 cells were treated with ∼900,000 liposomes/10 ml media (without glutamine) for at least 24 h. Then cells were incubated in serum-free DMEM in the absence of glutamine for 1.5 h, transferred to ice, washed twice in cold PBS, and surface labeled at 4°C with 0.2 mg/ml NHS-SS-biotin [succinimidyl 2-(biotinamido)-ethyl-1,3′-dithiopropionate; Pierce] in PBS for 30 min. Cells were transferred to serum-free and glutamine-free DMEM for 30 min at 37°C to allow internalization of the tracer. Cells were returned to ice, washed twice with ice-cold PBS, and biotin was removed from proteins remaining at the cell surface by reduction with MesNa (2-mercaptoethanesulfonate). The internalized fraction was then chased from the cells by returning them to 37°C in serum-free DMEM. At the indicated times, cells were returned to ice, and biotin was removed from recycled proteins by a second reduction with MesNa. Biotinylated MT1-MMP, active β1, and α5 integrin were then determined by capture ELISA using MaxiSorp (Nunc) plates coated with antibodies recognizing each receptor as described previously (Dozynkiewicz et al., 2012).

MDA-MB-231 cells with reduced mtDNA content (Rho0) were generated as previously described (Giampazolias et al., 2017). Briefly, cells were treated with ethidium bromide (500 ng/ml) plus uridine (50 µg/ml) for 5 d. Following this, cells were incubated in EV-free media for EV collection over a 48-h period.

DNase beads were made using Affi-Gel 10 beads (Bio-Rad Laboratories) and DNase I lyophilized powder (Qiagen). 10 ml of Affi-Gel 10 bead slurry was washed via a vacuum pump with >400 ml ice-cold water followed by >300 ml coupling buffer (0.1 M Hepes, 80 mM CaCl₂, pH 7.2), then the volume was reduced to 8 ml. DNase I, reconstituted in 4 ml coupling buffer, was added to 8 ml resin and incubated on a roller at 4°C overnight. Then the beads were washed with >400 ml coupling buffer with 1 M NaCl followed by >400 ml coupling buffer. Following this, the slurry was reduced to ∼4 ml, then supplemented with 4 ml coupling buffer and 4 ml glycerol and kept at −20°C in 500-µl aliquots.

45-µl fractions of 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC) lipids (25 mg/ml; Avanti Polar Lipids) were dried in glass tubes for 1–2 h in a Genevac centrifugal evaporator to remove all traces of chloroform. Lipids were then stored at −20°C or used directly. Dried lipids were resuspended in 500 µl PBS to achieve a concentration of 3 mM. For encapsulated mtDNA, lipids were resuspended in PBS containing 1,000 ng mtDNA (previously extracted from MDA-MB-231 cells using an mtDNA isolation kit; Abcam). Resuspended lipids were then sonicated in a water bath for 10 min at 45°C and then subjected to five rounds of snap freezing/thawing using liquid nitrogen. Lipids were then extruded through a 100-nm filter at RT using a Mini Extruder (Avanti Polar Lipids), and this was repeated 21 times. Extruded lipids were stored for up to 2 wk at 4°C, and their size and number were assessed using NTA.

To remove mtDNA not encapsulated by liposomes, 25 µl prepared liposomes were added to 50 µl DNase resin, 20 µl MgCl₂ buffer (50 mM Tris, pH 7.5, 12.5 mM MgCl₂, 2.5 mM CaCl) and PBS, then incubated at 37°C at 450 rpm in a thermomixer (Eppendorf) for 1 h. Following this, the DNase-conjugated beads were precipitated by centrifugation, and 100 µl supernatant (containing the liposome-encapsulated mtDNA) was stored for up to 2 wk at 4°C.

For immunolabeling, EVs isolated by differential centrifugation were fixed using PFA (4%). An aliquot (10 µl) of fixed EV suspension was placed onto a 100-mesh formvar-carbon–coated grid for 20 min to allow the EVs to adhere. Grids were washed in PBS, and the nonspecific binding site was blocked using BSA (3% in PBS) for 1 h. Adherent EVs were then incubated sequentially with an antibody recognizing CD63 and a secondary antibody conjugated to 10-nm gold particles (Aurion) for 1 h. For contrast, grids were then incubated in a mixture of oxalate/uranyl acetate for 10 min in the dark, then in methylcellulose/uranyl acetate for 10 min on ice in the dark, and allowed to dry in metal loops overnight. The grids were imaged using a JEOL 1200 EX transmission electron microscope operating at 80 kV. All acquired images were processed (brightness and contrast), and all EV diameters were determined, using Fiji software (Schindelin et al., 2012).

Fig. S1 describes the consequences of siRNA of Rab27 on release of EVs from MDA-MB-231 cells, and these data pertain to Fig. 1. Fig. S2 displays data outlining the consequences of siRNA of Rab27 and CD63 on oxygen consumption by MDA-MB-231 cells, and this pertains to Fig. 2 a. Fig. S3 displays an analysis of mitochondrial genes within EVs that pertains to Fig. 2, b–f. Fig. S4 displays Western blot and oxygen consumption data that pertain to Fig. 3. Fig. S4 also displays data illustrating the consequences of CRISPR knockout of PINK1 in cells from the MMTV-PyMT mouse model of mammary carcinoma. Fig. S5 contains data indicating that disruption of autophagy regulators (ATG5/7 or FIP200) does not impact EV release from MDA-MB-231 cells. Video 1 shows a clip of the dynamics of late endosomes and mitochondria in control cells. Video 2 shows a clip of the dynamicsof late endosomes and mitochondria in PINK1 knockdown cells. Video 3 shows a clip of the dynamics of late endosomes and mitochondria in CCCP-treated cells. Video 4 shows a clip of the dynamics of late endosomes and mitochondria in CCCP-treated PINK1 knockdown cells. Table S1 shows the catalog numbers of the antibodies used in the study.

The data supporting the findings of this study are available within the article and its supplementary information files and from the corresponding author upon request.

We thank Richard Youle (National Institutes of Health, Bethesda, MD) and Chunxin Wang (Institut Curie, Paris, France) for the PINK1-YFP constructs, Anh Hoang Le (Machesky laboratory, Cancer Research UK Beatson Institute) for help with the gelatin degradation protocol, and Jaclyn Long (Ryan laboratory, Cancer Research UK Beatson Institute) for advice with LC3 blots.

This work was funded by Cancer Research UK and Breast Cancer Now. We acknowledge the Cancer Research UK Glasgow Centre (C596/A18076) and the Biostatistics Unit facilities at the Cancer Research UK Beatson Institute (C596/A17196).

The authors declare no competing financial interests.

Author contributions: N. Rabas, S. Palmer, L. Mitchell, L. Lemgruber Soares, and J.C. Norman performed the experiments and analyzed experimental data. J.C. Norman supervised the experimental work. S. Ismail, A. Gohlke, J.S. Riley, S. Tait, P. Gammage, and I.R. Macpherson contributed to the conceptual design and planning of experiments and provision of reagents. N. Rabas, S. Palmer, and J.C. Norman prepared the figures and wrote the manuscript.