Autophagy adaptors mediate Parkin-dependent mitophagy by forming sheet-like liquid condensates

To better understand the role of each autophagy adaptor, we analyzed their localisation during Parkin-mediated mitophagy. When mitophagy was induced by treatment with the mitochondrial uncoupler carbonyl cyanide m-chlorophenylhydrazone (CCCP) in HeLa cells expressing exogenous Parkin, the autophagy adaptors p62, OPTN, NBR1, NDP52 and TAX1BP1 translocated to mitochondria as previously observed (Fig. 1A,B) (Geisler et al, 2010; Wong and Holzbaur, 2014; Heo et al, 2015; Moore and Holzbaur, 2016; Gallagher and Holzbaur, 2023). Ubiquitin and all of these autophagy adaptors were distributed evenly on the surface of separate mitochondria (Mt; Fig. 1A). In contrast, these adaptors exhibited distinct localisation patterns during the formation of isolation membranes on mitochondria (Mt–IM; Fig. 1B,C). The signals of OPTN and NDP52 were enriched in areas where LC3B signals colocalized, while they were mostly absent from the LC3B-negative side of the same mitochondria (Fig. 1B,C). Endogenous OPTN and NDP52 also showed clear enrichment on the LC3B-positive areas in comparison to the LC3B-negative side, confirming that this localisation is not caused by the overexpression of adaptors (Fig. 1D). However, this inhomogeneous distribution pattern was not observed for ubiquitin, p62, NBR1, and TAX1BP1 (Fig. 1B,C). Unlike OPTN and NDP52, p62 showed enrichment between clustered mitochondria (Mt–Mt), which is consistent with previous reports (Fig. EV1A,B) (Wong and Holzbaur, 2014), which appears to be consistent with the role of p62 in promoting mitochondrial clustering (Narendra et al, 2010; Okatsu et al, 2010). These accumulation patterns cannot be explained by the stoichiometric interaction of the autophagy adaptors with ubiquitinated proteins that are evenly distributed on mitochondria.

The localisation of autophagy adaptors may be affected by their interaction with other adaptors (Turco et al, 2021; Gubas and Dikic, 2022). To examine the localisation of each adaptor on its own, we used HeLa cells lacking all five autophagy adaptors (penta KO cells) to exclude the effect of other adaptors (Lazarou et al, 2015). In penta KO cells, OPTN and NDP52 exhibited significant enrichment in LC3B-positive areas, similar to that observed in wild-type cells, suggesting that this accumulation occurred independently of heterologous interaction with other adaptors (Fig. 1E,F). Mitophagy was not restored by exogenous expression of p62 or NBR1 in penta KO cells, and thus colocalization with LC3 could not be tested for these two adaptors (Lazarou et al, 2015). The accumulation of p62 between clustered mitochondria was also observed in penta KO cells (Fig. EV1C,D). Given the uniform distribution of ubiquitinated proteins on mitochondrial surfaces, these data suggest that autophagy adaptors exhibit non-stoichiometric enrichment between membranes. Hereafter, we used penta KO cells to study each autophagy adaptor individually.

We hypothesised that the non-stoichiometric distribution of the autophagy adaptors on ubiquitinated mitochondria could be explained by LLPS. Typical condensates generated by LLPS exhibit rapid exchange of their components, which is often demonstrated by fluorescence recovery after photobleaching (FRAP) analysis (Taylor et al, 2019; McSwiggen et al, 2019). We first conducted FRAP experiments for separate (i.e. unclustered) mitochondria in CCCP-treated cells. As expected, ubiquitin showed virtually no recovery because it was conjugated to mitochondrial membrane proteins (Fig. 2A,B). Additionally, p62 recovered only slightly, suggesting a minute exchange between the mitochondrial surface and cytosol (Fig. 2A,B). In contrast, OPTN and NDP52 showed a rapid recovery, indicating a dynamic exchange of these adaptors between the mitochondrial surface and the cytosol (Fig. 2A,B).

Next, we examined the dynamics of adaptors between mitochondria and isolation membranes. OPTN and NDP52 showed partial recovery (Fig. 2C,D). Although this result indicates some exchange of these adaptors between membranes, the incomplete recovery suggests the existence of a gel-like or immobile fraction. Treatment with 1,6-hexanediol, which can be used to dissolve condensates, dispersed OPTN and NDP52, but not p62, from mitochondria without isolation membranes (Figs. 2E and EV2A). In contrast, 2,5-hexanediol, which is considered by some studies as a negative control for 1,6-hexanediol, did not disperse the OPTN or NDP52 condensates (Fig. EV2B). These data suggest that OPTN and NDP52 accumulation is supported by weak interactions. In contrast, OPTN and NDP52 at the mitochondria-isolation membrane contact sites remained unaffected by 1,6-hexanediol (Fig. EV2C). This may be due to the additional interaction with ATG8 proteins, not only with ubiquitin. These results indicate that OPTN and NDP52 on ubiquitinated mitochondria are mobile and can quickly exchange between the mitochondrial surface and the cytosol, supporting the hypothesis that they form phase-separated condensates on damaged mitochondria.

To simulate the behaviour of autophagy adaptors in the context of phase-separated condensates, we developed mathematical models (see Materials and Methods for details). First, we considered the case of two mitochondria approaching each other (Fig. 3A). In Parkin-mediated mitophagy, various proteins in the mitochondrial outer membrane are ubiquitinated. We modelled the mitochondrial outer membrane as a circle with a diameter of $D_{\mathrm{Mt}}=600$ nm. On these mitochondria, we assumed a ring-shaped ubiquitin layer with a thickness $w_{\mathrm{Ub}}=20$ nm and area-fraction (hereafter referred to as concentration) ${\varphi }_{\mathrm{Ub}}=0.1$ (Milo and Phillips, 2015; Berry et al, 2018) (Fig. 3A). The closest distance between the two mitochondrial outer membranes was set to $d_{\mathrm{Mt}}$ (Fig. 3A). Here, we considered the dynamics of concentration changes of the adaptor proteins and solvent components. We assumed that mitochondria were very large compared with proteins and did not move on a similarly short time scale, and that ubiquitin was tightly bound to mitochondrial membranes and immobile, as suggested by the FRAP experiment (Fig. 2B). The adaptor proteins were assumed to diffuse freely in solution and to bind to ubiquitin and themselves (self-interactions) with strengths ${\chi }_{\mathrm{Ub}}$ and ${\chi }_{\mathrm{self}}$, respectively. The adaptors showed accumulation and distribution on mitochondria similarly to what was observed in cells (Fig. EV1) when ${\chi }_{\mathrm{Ub}}\ge 12k_{B}T$ and ${\chi }_{\mathrm{self}}=3k_{B}T$ (Fig. EV3A), with the Boltzmann constant $k_B$ and temperature T. Therefore, we used ${\chi }_{\mathrm{Ub}}=12k_{B}T$ and ${\chi }_{\mathrm{self}}=3k_{B}T$ for further analyses.

Using the mathematical model, we assessed how the condensates deformed when the distance between two mitochondria $d_{\mathrm{Mt}}$ was changed. When two mitochondria were far apart ($d_{\mathrm{Mt}}=900\mathrm{nm}$), the model predicted that adaptors cover the entire mitochondrial surface (Fig. 3B). When the mitochondria were close enough to come into contact ($d_{\mathrm{Mt}}=60\mathrm{nm}$), the adaptors moved to the area between the mitochondria and filled the cleft between them (arrowheads, Fig. 3B).

Next, we considered the case in which a growing isolation membrane surrounded and enclosed a single mitochondrion (Fig. 3C,D). A mitochondrion with ubiquitin was modelled as mentioned above, and the isolation membrane was modeled as a region of thickness $w_{\mathrm{IM}}=50\mathrm{nm}$ bound by two semicircles with the same centre as the mitochondrion. The closest distance between the isolation membrane and the mitochondrial outer membrane was $d_{\mathrm{IM}}=100\mathrm{nm}$, and the isolation membrane bent at an angle α to surround the mitochondrion. The surface of the isolation membrane was assumed to be covered by an LC3 region with a thickness $w_{\mathrm{LC}\mathrm{3}}=20$ nm and concentration ${\varphi }_{\mathrm{LC}\mathrm{3}}=0.1$. The adaptors interacted with both LC3 and ubiquitin with strengths ${\chi }_{\mathrm{LC}\mathrm{3}}$ and ${\chi }_{\mathrm{Ub}}$, respectively. A clear accumulation of the autophagy adaptor in the region in contact with the isolation membrane, along with depletion from the non-contacting region, was observed at ${\chi }_{\mathrm{LC}3}\ge 12k_{B}T$ (Fig. EV3B), similar to the OPTN distribution observed in cells (Fig. 1B). Therefore, we used ${\chi }_{\mathrm{LC}\mathrm{3}}=12k_{B}T$ for further analyses.

Using this model, we determined the distribution of the adaptors when the isolation membrane appeared and elongated (changing the angle α surrounding the mitochondrion). In the absence of the isolation membrane, the adaptors wet the entire mitochondrial surface uniformly (Fig. 3D). However, with the appearance of the isolation membrane, the adaptors remobilized to the area between the isolation membrane and mitochondrion, and this adaptor-enriched region elongated together with the isolation membrane. These modelling results suggest that the distribution of the adaptors on a mitochondrion can change upon contact with another mitochondrion or an isolation membrane.

Our mathematical models predict that the OPTN distribution changes when mitochondria cluster or are engulfed by isolation membranes. We then validated these predictions by using live-cell imaging. When two small mitochondria with uniform OPTN distributions on their surface approached each other, OPTN redistributed and formed a smooth surface that covered the two adjacent mitochondria (Fig. 4A–D; Movie EV1). Notably, strong OPTN enrichment was observed in the cleft between the two mitochondria (arrowheads, Fig. 4A,C), showing a pattern distinct from that of the outer membrane protein Omp25 (OPTN signal peaks localised outside of Omp25 signal peaks) (Fig. 4B,D). This phenomenon was apparent when the size of both (Fig. 4A) or one (Fig. 4C; Movie EV1) of the two mitochondria was less than 1 µm. Although OPTN formed sheet-like rather than spherical condensates, this phenomenon appeared to be similar to condensate coalescence, which is one of the hallmarks of liquid-like condensates (Hyman et al, 2014).

Furthermore, when an isolation membrane started to engulf a mitochondrion, OPTN, which was initially distributed uniformly on the mitochondrial surface, underwent redistribution to the area contacting the isolation membrane (Fig. 4E; Movie EV2). OPTN signals on the isolation membrane-negative side of the mitochondrion diminished during this process (Fig. 4E,F). The OPTN-enriched region expanded together with the isolation membrane thereafter (Fig. 4E). These live-cell observations are consistent with the predictions of our mathematical models (Fig. 3). Taken together, our experimental data support the hypothesis that OPTN forms sheet-like phase-separated condensates on the surface of ubiquitinated mitochondria, exhibiting liquid-like properties.

To investigate the importance of the liquid-like nature of OPTN condensates on mitochondria, we sought to reduce their fluidity by strengthening the interaction between ubiquitin and GFP–OPTN constructs. To achieve this, we utilised the anti-green fluorescent protein (GFP) nanobody, which exhibits a high binding affinity to GFP (Rothbauer et al, 2008) (Fig. 5A). Consequently, the interaction between anti-GFP nanobody-fused ubiquitin and GFP–OPTN was stronger than that between ubiquitin and GFP–OPTN (Fig. 5B). Expression of anti-GFP nanobody-fused ubiquitin (nanobody–mRuby–Ub) did not affect the recruitment of GFP–OPTN to mitochondria (Fig. 5A, (i) and (ii)) but partially reduced its FRAP recovery (Fig. 5C), suggesting a reduction in mobility. We then evaluated the effect of reduced OPTN fluidity on mitophagy using the recently developed HaloTag cleavage assay (Yim et al, 2022). This assay utilises the ligand-dependent conformational change of HaloTag (Halo); ligand-free Halo is efficiently degraded in lysosomes, whereas ligand-bound Halo becomes resistant to lysosomal degradation. When Halo is expressed in the mitochondrial matrix by fusing to the mitochondrial presequence of Fo-ATPase subunit 9 and SNAP-tag (mtHalo–SNAP), we can measure the extent of mitochondrial degradation by quantifying the amount of processed free Halo out of the total amount of Halo (mtHalo–SNAP + processed Halo). Mitophagy activity, which was impaired in penta KO cells, was restored by the expression of GFP–OPTN (Fig. 5D,E). Mitophagy activity was partially reduced in cells expressing GFP–OPTN and nanobody–mRuby–Ub (Fig. 5D,E).

The partial reduction in both FRAP recovery and mitophagy activity may be attributed to the remaining interaction between OPTN and endogenous ubiquitin. Therefore, we deleted the ubiquitin-binding domain in ABINs and NEMO (UBAN domain) from OPTN (GFP–OPTNΔUBAN), abolishing its interaction with ubiquitinated proteins (Fig. 5B). GFP–OPTNΔUBAN failed to accumulate on mitochondria in the presence of mRuby–Ub (Fig. 5A, (iii)). In contrast, the expression of nanobody–mRuby–Ub rescued the recruitment of GFP–OPTNΔUBAN to mitochondria (Fig. 5A, (iv)), albeit to a reduced extent, likely reflecting the loss of condensate formation. FRAP experiments revealed that, compared with wild-type GFP–OPTN (Fig. 5A, (i)), GFP–OPTNΔUBAN (Fig. 5A, (iv)) showed a substantial decrease in recovery on mitochondria when co-expressed with nanobody–mRuby–Ub, indicating a reduction in its dynamic properties (Fig. 5C). GFP–OPTNΔUBAN failed to restore mitophagy, even when it was co-expressed with nanobody–mRuby–Ub to rescue ubiquitin binding and mitochondrial localisation (Fig. 5A,B,D,E). This impairment was indeed due to the loss of binding to ubiquitin because a point mutation that abolished ubiquitin binding (E478G) (Li et al, 2018) also showed normal mitochondrial localisation, low FRAP recovery, and mitophagy defect (Figs. 5D,E and EV4A,B). Furthermore, the impaired mitophagy was not simply due to the altered binding orientation of OPTN. Replacement of its UBAN domain with GFP (OPTNΔUBAN–GFP), mimicking its binding to ubiquitin at the C-terminus, did not restore FRAP recovery or mitophagy activity, although it also showed clear recruitment to mitochondria in the presence of nanobody–mRuby–Ub (Fig. EV4A–D). The restricted orientation of OPTN due to the stoichiometric interaction between GFP and nanobody, as opposed to the more flexible orientation in liquid-like condensates, may also contribute to the impairment of autophagy.

Consistent with the results of mitophagy activity, the accumulation of the ULK1 complex component FIP200, likely representing isolation membranes, was observed in the presence of mRuby–Ub and GFP–OPTN (Fig. EV5A). In contrast, no clear FIP200 accumulation was observed in cells expressing nanobody–mRuby–Ub and GFP–OPTNΔUBAN, suggesting that mitophagy failed at the initiation step in these cells (Fig. EV5A). Together, these data suggest that the localisation of OPTN on mitochondria and its binding to ubiquitin alone is insufficient for mitophagy, and the fluidity of OPTN is important for effective mitophagy induction.

ATG9 vesicles are considered to be one of the sources of autophagosomal membranes (Yamamoto et al, 2012). Upon mitophagy induction, ATG9 vesicles accumulate at the sites of autophagosome formation (Itakura et al, 2012). OPTN plays a crucial role in this process by recruiting ATG9 vesicles to induce mitophagy (Yamano et al, 2020). In general, liquid-like condensates can be involved in clustering small vesicles such as synaptic vesicles, which exemplify complete wetting (Milovanovic and De Camilli, 2017; Sansevrino et al, 2023). Indeed, ATG9 vesicles are incorporated into condensates containing a glaucoma-associated OPTN mutant (O’Loughlin et al, 2020) and synapsin (Park et al, 2023). We, therefore, hypothesised that the liquid-like properties of OPTN condensates on mitochondria facilitate the recruitment of ATG9 vesicles. Consistent with previous reports (Itakura et al, 2012; Yamano et al, 2020), ATG9A was recruited to depolarised mitochondria in cells expressing GFP–OPTN (Fig. 6A,B). However, this ATG9A recruitment was almost completely abolished in cells expressing nanobody–mRuby–Ub and GFP–OPTNΔUBAN (Fig. 6A,B), even though GFP–OPTNΔUBAN still retains the ATG9A-interacting domain and the ability to bind with the ATG9 vesicles (Fig. 6C,D). These data suggest that the fluidity of sheet-like OPTN condensates on the mitochondrial surface is crucial for recruiting ATG9 vesicles and executing mitophagy.

Furthermore, TBK1 is recruited by OPTN to the mitophagy initiation site to be phosphorylated and activated, driving the feed-forward mechanism of OPTN accumulation and TBK1 activation, which in turn phosphorylates the PI3K complex to initiate mitophagy (Nguyen et al, 2023; Yamano et al, 2024). Consistently, a clear accumulation of phosphorylated TBK1 on mitochondria was observed in the presence of GFP–OPTN and mRuby–Ub (Fig. EV5B). In contrast, only very faint signals of phosphorylated TBK1 were observed in the presence of nanobody–mRuby–Ub and GFP–OPTNΔUBAN (Fig. EV5B). These data suggest that the fluidity of OPTN is also important for the feed-forward mechanism between OPTN and TBK1 for the initiation of mitophagy. Together, we propose that autophagy adaptors phase separate on ubiquitinated mitochondria to amplify the initiation signals and wet the ATG9 vesicles to facilitate efficient mitophagy.

In the present study, we discovered that autophagy adaptors accumulate on ubiquitinated mitochondria with non-stoichiometric distribution compared with ubiquitin signals. They displayed characteristics consistent with phase-separated structures, including dynamic exchange with cytosol and redistribution upon coalescence. Notably, we observed the enrichment of OPTN at the isolation membrane-positive region, which expanded along with the elongation of the isolation membrane. These findings suggest that phase-separated adaptors facilitate the effective engulfment of mitochondria by the isolation membrane through the wetting effect and capillary forces (Fig. 6E, right panel). Moreover, the fluidity of OPTN condensates is critical for the localisation of ATG9 vesicles onto ubiquitinated mitochondria, indicating that partial or complete wetting of phase-separated adaptors to ATG9 vesicles may be crucial for their recruitment to and/or retention on mitochondria (Fig. 6E, left panel). Therefore, we propose that LLPS of autophagy adaptors plays a critical role in two distinct steps of mitophagy by producing capillary forces, first, at the initiation step by recruiting and retaining the ATG9 vesicles in the condensates and, second, at the autophagosome elongation step by facilitating the attachment between ubiquitinated mitochondria and the isolation membrane. Additionally, our experimental data suggest that NDP52 also exhibits liquid-like properties. Further studies are needed to determine whether condensate formation is necessary for mitophagy induction by NDP52.

Recent reports have shed light on the importance of LLPS in selective autophagy (Noda et al, 2020). In lysophagy, p62 undergoes LLPS on lysosomes to facilitate efficient lysophagy together with HSP27, which maintains the fluidity of p62 and prevents its gelation (Gallagher and Holzbaur, 2023). This may also be mediated by the wetting between damaged lysosomes and the isolation membrane by phase-separated p62. Indeed, p62 itself forms phase-separated condensates along with ubiquitinated proteins, producing capillary forces by the wetting effect to facilitate the engulfment by the isolation membrane (Agudo-Canalejo et al, 2021). In our study, p62 exhibited low mobility, as determined by FRAP and 1,6-hexanediol experiments (Figs. 2B and EV2A). While these data may suggest that p62 is immobile, they appear to contradict the observed redistribution of p62 to the area between mitochondria upon clustering (Fig. EV1A,B). Thus, we speculate that p62 may exhibit some degree of fluidity despite its low exchange between the condensate and cytosol. Similarly to p62, Ape1, the cargo of the cytosol-to-vacuole targeting (Cvt) pathway in yeast, forms gel-like condensates with Atg19 on the surface of Ape1 condensates to be sequestered by the isolation membrane (Yamasaki et al, 2020). This association could also be mediated by partial wetting. These examples, together with our observations, highlight the importance of LLPS and the wetting phenomena in various selective autophagy processes.

LLPS may also be important for non-selective bulk autophagy. In yeast, the Atg1 complex (the ULK1 complex in mammals) forms pre-autophagosomal structures (PASs), which have been shown to be condensates driven by LLPS (Fujioka et al, 2020). Phase separation in this context provides a high local concentration of the Atg1 complex to support the autoactivation of Atg1 kinase (Yamamoto et al, 2016). Mammalian autophagy factors also accumulate at the autophagosome formation site; FIP200, a subunit of the autophagy-initiation complex, undergoes LLPS triggered by calcium transients (Zheng et al, 2022). The liquid-like nature of these autophagy-initiating structures may also be important for the recruitment of Atg9/ATG9 vesicles through the wetting effect.

In contrast to Parkin-mediated mitophagy, hypoxia-induced mitophagy does not involve autophagy adaptors, and instead, the outer mitochondrial membrane proteins NIX (also known as BNIP3L) and BNIP3 directly interact with ATG8 homologues on the isolation membrane (Novak et al, 2010; Onishi et al, 2021). In this context, ATG8 proteins interact with NIX or BNIP3, likely in a one-to-one manner. This raises the question of why phase separation of autophagy adaptors is essential for Parkin-mediated mitophagy. One possibility is that the quick induction and execution of Parkin-mediated mitophagy requires amplification of the reactions through increased local concentration of adaptors and autophagy-related proteins facilitated by LLPS. This is consistent with the fact that mitophagy-inducing signals are also amplified by a mechanism involving phosphorylated ubiquitin (Kane et al, 2014; Kazlauskaite et al, 2014; Koyano et al, 2014) and OPTN-TBK1 axis (Nguyen et al, 2023; Yamano et al, 2024). Indeed, we observed that the amount of OPTN recruited to mitochondria was greater with wild-type OPTN than with OPTNΔUBAN (in the presence of nanobody-Ub), likely reflecting the enhanced recruitment by condensate formation (the reduced recruitment of OPTNΔUBAN may also be due to a lower expression level of nanobody–mRuby–Ub compared to abundant endogenous ubiquitin, which may further contribute to reduced mitophagy activity) (Figs. 5A and EV4A). Alternatively, Parkin-mediated mitophagy requires a high level of precision and selectivity; mitophagy should be induced only for damaged mitochondria, which is ensured by the recruitment and retention of ATG9 vesicles only on phase-separated adaptor-positive mitochondria. This strict specificity would enable proper quality control of the mitochondria.

HeLa cells, HeLa penta-knockout (penta KO) cells and human embryonic kidney (HEK) 293T cells authenticated by RIKEN were cultured in Dulbecco’s modified Eagle’s medium (DMEM; D6546; Sigma-Aldrich) supplemented with 10% foetal bovine serum (FBS; 173012; Sigma-Aldrich) and 2 mM l-glutamine (25030-081; Gibco) in a 5% CO₂ incubator at 37 °C. For stable expression, retrovirus was produced with HEK293T cells transfected with pMRX-IP-based, pMRX-IB-based, pMRX-IU-based, or pMRX-IN-based retroviral plasmids, pCG-VSV-G, and pCG-gag-pol by using Lipofectamine 2000 (11668019; Thermo Fisher Scientific). After transfection, cells were further incubated at 37 °C for 24 h. The viral supernatant was collected by filtration through a 0.45-μm filter unit (Ultrafree-MC; Millipore) and then used for infection. Cells were plated onto 6-cm dishes 18 h before infection, and the medium was replaced with viral supernatant diluted 1.5-fold with 8 μg/mL polybrene (H9268; Sigma-Aldrich). Two days later, cells were selected in a medium containing 2 μg/mL puromycin (P8833; Sigma-Aldrich), 4 μg/mL blasticidin S hydrochloride (022-18713; FUJIFILM Wako Pure Chemical Corporation), 1.5 mg/mL neomycin or 250 μg/mL zeocin (R25005; Thermo Fisher Scientific).

The pMRX-IPU, pMRX-IBU and pMRX-INU plasmids were generated by modifying the multi-cloning site of pMRX-IP (Saitoh et al, 2002; Kitamura et al, 2003), pMRX-IB, and pMRX-IN, respectively. DNA fragments encoding ubiquitin, p62 (NM_003900.5), OPTN (NM_001008211.1), NBR1 (NM_001291571.2), NDP52 (NM_001261391.2), TAX1BP1 (NM_001206901.1), ATG9A (NM_001077198.3), LC3A (NM_032514.4) and LC3B (NM_022818.5) were inserted into pMRX-IPU, pMRX-INU, or pMRX-IBU. DNAs encoding the HA epitope, monomeric enhanced GFP with the A206K mutation (mGFP), codon-optimised ultra-stable GFP (muGFP) (Scott et al, 2018), codon-optimised mRuby3 (Bajar et al, 2016), HaloTag7 (G1891; Promega), 3×FLAG, and SNAP-tag (New England BioLabs, N9181S) were used for tagging. The mitochondrial presequence of Neurospora crassa Fo-ATPase subunit 9 (residues 1–69) was added to HaloTag7-SNAP to deliver the reporter into the mitochondrial matrix (mtHalo–SNAP) (Eura, 2003). Truncated OPTN (OPTNΔUBAN, aa 445–502) was prepared by PCR-mediated site-directed mutagenesis. The resulting plasmids were sequenced.

The primary antibodies used in this study were rabbit polyclonal anti-OPTN (Proteintech, 10837-AP), rabbit polyclonal anti-ATG9A (MBL, PD042), mouse monoclonal anti-Halo (Promega, G9211), mouse monoclonal anti-HSP90 (BD Transduction Laboratories, 610419), rabbit polyclonal anti-GFP (Thermo Fisher Scientific, A6455), rabbit polyclonal anti-NDP52 (Proteintech, 12229-1-AP), rabbit polyclonal anti-RB1CC1 (Proteintech, 17250-1-AP) and rabbit monoclonal anti-phospho-TBK1 (Ser172) (Cell Signaling Technology, 5483) antibodies. The secondary antibodies used were HRP-conjugated goat polyclonal anti-rabbit IgG (Jackson ImmunoResearch Laboratories, 111-035-144) and HRP-conjugated goat polyclonal anti-mouse IgG (Jackson ImmunoResearch Laboratories, 115-035-003) antibodies. To induce mitophagy, HeLa cells were treated with 20 μM carbonyl cyanide m-chlorophenyl hydrazine (CCCP; Sigma-Aldrich) for 45 min, or 10 μM oligomycin (Cabiochem, 495455-10MGCN) and 4 μM antimycin A (Sigma-Aldrich, A8674) for 18 h. After cells had been treated with oligomycin and antimycin A for >6 h, 10 μM Q-VD-OPH (SM Biochemicals, SMPH001) was added to block apoptotic cell death. To inhibit the formation of isolation membranes in the 1,6-hexanediol (Sigma-Aldrich, 240117-50 G) treatment experiment, wortmannin (Sigma-Aldrich, W1628-1MG) was added into the medium.

Cell lysates were prepared in a lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, EDTA-free protease inhibitor cocktail [19543200; Roche]). After centrifugation at 17,700×g for 10 min, the supernatants were subjected to immunoprecipitation using anti-FLAG M2 magnetic beads (M8823-1ML; Sigma-Aldrich). Precipitated immunocomplexes were washed three times with lysis buffer and boiled in sample buffer (46.7 mM Tris-HCl, pH 6.8, 5% glycerol, 1.67% sodium dodecyl sulfate, 1.55% dithiothreitol, and 0.02% bromophenol blue). For immunoprecipitation of ATG9 vesicles, cells were disrupted by Dounce homogenisation with hypotonic lysis buffer (10 mM HEPES, pH 7.9, 1.5 mM MgCl₂ and 10 mM KCl, EDTA-free protease inhibitor cocktail [19543200; Roche]). Disruption was carried out by applying 35 strokes while the cell suspension was cooled on ice. The supernatants were subjected to immunoprecipitation using anti-FLAG M2 magnetic beads (M8823-1ML; Sigma-Aldrich). For immunoblotting, the samples were separated by SDS-PAGE and transferred to Immobilon-P polyvinylidene difluoride membranes (Millipore, WBKLS0500) with the Trans-Blot Turbo Transfer System (Bio-Rad). After incubation with the relevant antibody in 5% skim milk in 20 mM Tris-HCl, 150 mM NaCl, and 0.1% Tween 20 (02194841-CF; MP Biomedicals), the signals from incubation with SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific, 34579) were detected with the FUSION SOLO.7S.EDGE imaging system (Vilber-Lourmat). Contrast and brightness adjustment and quantification were performed using the image processing software Fiji (Schindelin et al, 2012).

In-cell fluorescence recovery after photobleaching (FRAP) analyses were performed with an Olympus Fluoview FV3000 confocal microscope equipped with a 60× oil-immersion objective lens (1.40 NA, Olympus). The chamber was maintained at 37 °C and continuously supplied with humidified 5% CO₂. Bleaching was performed using 80% laser power (488 or 561 nm laser), and images were captured every 5 s for 30 frames. Recovery curves and fitting were analyzed using OriginPro 2022. The fluorescence intensity of the bleached spot, an unbleached control spot, and the background was measured using the software package Fiji. Background intensity was subtracted, and the intensity values of the region of interest are reported relative to the pre-bleached images during image acquisition. Each data point represents the mean and standard error of fluorescence intensities in more than three unbleached (control) or bleached (experimental) spots.

Living imaging was conducted with the Olympus SpinSR10 spinning-disk confocal super-resolution microscope equipped with a Hamamatsu ORCA-Flash 4.0 camera, a UPLAPO OHR 100 × (NA 1.50) lens, and the SORA disk in place. The microscope was operated with Olympus cellSens Dimension 2.3 software. Cells were passaged onto a four-chamber glass-bottom dish (Greiner Bio-One) more than 24 h before imaging. To induce mitophagy, cells were incubated with 10 μM CCCP in the presence of 200 nM SF650-conjugated Halo ligand (GoryoChemical, A308-02). Images were processed using the image processing software package Fiji (Schindelin et al, 2012). The fluorescence intensity of indicated fluorophores was measured by Fiji, and spline-connected graphs were created using OriginPro 2022 software.

Cells were fixed with 4% PFA in 0.1 M phosphate buffer (pH 7.3) for 15 min at room temperature (RT) and washed with PBS. Fixed cells were incubated with 10 µg/mL digitonin for 5 min at RT and washed with PBS. Then, cells were incubated with primary antibodies diluted (1:1000) in blocking buffer (3% BSA, in PBS) for 1 h at RT, washed with PBS, incubated with Alexa Fluor-conjugated secondary antibodies (1:1000) in blocking buffer for 60 min at RT, washed again, and mounted on coverslips. Colocalization was calculated by Fiji plugins BIOP JACoP, with the threshold of the ATG9A channel fixed to 500. ATG9A-positive areas out of GFP-positive areas were calculated to obtain the overlapping areas.

We formulated a mathematical model of autophagy adaptor condensate formation. For the case of two mitochondria approaching each other (Fig. 3A), the free energy, F, of the system can be written aswith the interaction energy, $f_{\mathrm{int}}$, and the entropic energy, $f_{\mathrm{ent}}$, expressed aswhere ${\varphi }_{\mathrm{ad}}$, ${\varphi }_{\mathrm{Ub}}$, and ${\varphi }_{\mathrm{sol}}=1-{\varphi }_{\mathrm{ad}}-{\varphi }_{\mathrm{Ub}}-{\varphi }_{\mathrm{Mt}}$ are the adaptor, ubiquitin, and solvent concentrations, respectively, ${\chi }_{\mathrm{self}}$ and ${\chi }_{\mathrm{Ub}}$ represent the strength of the adaptor self-interaction and adaptor–ubiquitin interaction, respectively, σ represents the surface tension of the condensates, and ${\chi }_{\mathrm{Mt}}$ represents the mitochondrial area exclusion effect, introduced to prevent proteins from entering the mitochondrial interior (${\varphi }_{\mathrm{Mt}}$).

For the case in which an isolation membrane surrounded a mitochondrion (Fig. 3C), the interaction energy, $f_{\mathrm{int}}$, is instead expressed aswhere ${\varphi }_{\mathrm{LC}\mathrm{3}}$ and ${\chi }_{\mathrm{LC}\mathrm{3}}$ represent the LC3 concentrations and the strength of the adaptor–LC3 interaction, respectively, and ${\chi }_{\mathrm{IM}}$ represents the mitochondrial area exclusion effect, introduced to prevent proteins from entering the lumen of the isolation membrane (${\varphi }_{\mathrm{IM}}$). The solvent concentration that appears in the entropic energy (3) was also modified to ${\varphi }_{\mathrm{sol}}=1-{\varphi }_{\mathrm{ad}}-{\varphi }_{\mathrm{Ub}}-{\varphi }_{\mathrm{LC}\mathrm{3}}-{\varphi }_{\mathrm{Mt}}-{\varphi }_{\mathrm{IM}}$.

For the numerical simulation, the space was discretized by a lattice with 10 nm on each side, and a concentration field was assigned to each location. The time evolution of the system was assumed to follow the Cahn–Hilliard equation,which is commonly used in analyses of phase separation dynamics (Berry et al, 2018). Here, $M=D/k_{B}T$ is the mobility of the protein with the diffusion constant D = 10 μm²/sec (Milo and Phillips, 2015) expressed in terms of the Boltzmann constant $k_B$ and temperature T. We considered a 3000 nm × 3000 nm square region with periodic boundary conditions containing two mitochondria and surrounding cytosolic components as a system and assumed that adaptor proteins are uniformly distributed in the cytosol around mitochondria at a concentration of $\varphi =0.1$ in the initial state.

Statistical analysis was performed using OriginPro 2022 software. The statistical methods used for each analysis are specified in the figure legends. No sample size calculation was performed, the experiments were not randomised, and investigators were not blinded during experiments.