Omegasome-proximal PtdIns(4,5)P2 couples F-actin mediated mitoaggregate disassembly with autophagosome formation during mitophagy

Cells govern their homeostasis through autophagy by sequestering substrates, ranging from proteins to aggregates and organelles, into autophagosomes for lysosomal degradation. In these processes cells need to coordinate between substrate remodeling and autophagosome formation for efficient engulfment. We found that in Parkin-mediated mitophagy, mitochondria to be turned over first become grape-like mitoaggregates, followed by their disassembly into smaller pieces via the actinomyosin system. At the disassembly step, we observed spatially-associated, synchronous formation of circular F-actin and BATS-labeled autophagy initiation sites near mitochondria, suggesting coordination between substrate downsizing and autophagosome formation during mitophagy. Interestingly, PtdIns(4,5)P2, instead of PtdIns(3)P, regulates this mitophagy-associated formation of circular F-actin and BATS-sites. Selective depletion of PtdIns(4,5)P2 near omegasomes, the endoplasmic reticulum (ER) subdomains involved in autophagosome formation, impaired mitoaggregate disassembly. Our findings demonstrate the presence of a pool of PtdIns(4,5)P2 adjacent to omegasomes, and that they coordinate mitoaggregate disassembly with autophagy initiation during Parkin-mediated mitophagy.

M itochondrial autophagy (mitophagy) is an organelle quality control pathway that sequesters damaged mitochondria into autophagosomes for lysosomal removal 1,2 . Defective mitochondria form clusters of mitochondrial aggregates (hereafter mitoaggregates) and can accumulate at the perinuclear region, a signature phenomenon reported in Parkinson's disease model systems 3 . Similar to the generation of protein aggresomes, mitoaggregate formation is considered a protective mechanism that clusters defective materials for the reduction of cellular toxicity. This process is driven by (1) Pink1/ Parkin-mediated ubiquitination of damaged mitochondria and subsequent accumulation of the ubiquitin (Ub) receptor p62 (also known as SQSTM1) that help cluster impaired mitochondria 4,5 , and (2) Trafficking of damaged mitochondria to the perinuclear region 6,7 . The apparent size mismatch between autophagosomes 8 (typically 0.5-1.5 μm in mammalian cells) and mitoaggregates suggests that mitoaggregates will eventually need to be dimensionally remodeled for efficient mitophagy.
A particularly interesting aspect related to mitoaggregate remodeling is its relationship with the autophagy initiation process. Synchronization between mitoaggregate remodeling and mitochondrial targeting by the autophagic machinery can reduce the exposure of free floating dysfunctional mitochondria to the cytoplasm, thereby minimizing cellular stress. It is known that autophagosomes originate from PtdIns(3)P-rich sites on the ER termed omegasomes: 9 LC3-PE containing autophagosomal membrane grows from these regions, which in turn sequesters substrates into complete autophagosome. Positioning mitoaggregate remodeling events directly at the omegasomes therefore represents one way to seamlessly integrate Parkin-labeled mitochondrial fragments into the autophagy pathway.
Here, we report that in Parkin-mediated mitophagy the mitoaggregates represent clusters of fragmented and swollen mitochondria. These mitoaggregates require disassembly into smaller pieces for productive Parkin-mediated mitophagy; during this process, circular F-actin and BATS-labeled autophagy initiation structures cooperatively form in proximity. The formation of both structures is controlled by the same upstream activator, a pool of PtdIns(4,5)P 2 that resides near the omegasome. Our results reveal the unrecognized role of phosphoinositides (PtdIns) in coordinating mitoaggregate remodeling with autophagy initiation during Parkin-mediated mitophagy.

Results
Mitoaggregate formation and disassembly during mitophagy. Using a photosensitizer 10 to selectively mediate the impairment of a fraction of cellular mitochondria (Fig. 1a), we observed that damaged mitochondria underwent mitoaggregate formation and disassembly within a time span of 3 h ( Fig. 1b; Supplementary Movie 1): first they clustered into mitoaggregates after Parkin recruitments (40-50 min after damage), followed by their disassembly into smaller fragments (90-100 min after damage). We then utilized FRAP to probe the status of Parkin-labeled mitoaggregates in HeLa cells expressing EBFP2-Parkin, EGFP-dMito (a mito-matrix-targeting EGFP), and KillerRed-dMito (another photosensitizer for mitophagy initiation). Compared to the control where EGFP-dMito can exchange between interconnected mitochondria ( Supplementary Fig. 1a), EGFP-dMito was unable to freely exchange within mitoaggregates (Fig. 1c, d; white arrow indicates the photobleached region). This demonstrated that the mitoaggregates were not composed of an entangled tubular mitochondrion. Correlative light and electron microscopy (CLEM) revealed that mitoaggregates represent clusters of fragmented and swollen mitochondria (Fig. 1e-g and Supplementary Fig. 1b, c). This indicated that damaged mitochondria underwent mitochondrial fission as well as clustering to form mitoaggregates, followed by their disassembly into smaller fragments for turnover. High electron densities found at the interfaces between individual mitochondrial fragments in the mitoaggregates (Fig. 1 h) support the presence of tethering factors (such as Ub and p62) that help mitoaggregate clustering. In support of this, p62-deficient HeLa cells exhibited defective formation of mitoaggregates ( Supplementary Fig. 1d). As shown in Fig. 1g, the dimensions of grape-like mitoaggregates are much larger than those of typical autophagosomes ( Fig. 1g; red arrow); thus, the disassembly of mitoaggregates into smaller pieces appears to be necessary for their efficient engulfment into autophagosomes. We therefore sought to determine how cells regulate mitoaggregate disassembly.
Circular F-actin structures during mitoaggregate disassembly. Time-lapse imaging of mitoaggregate disassembly (dynamics of Parkin-labeled mitochondria) revealed that the process included fragmentation as well as long-range dispersion, ( Supplementary  Fig. 1e). This observation led us to evaluate whether cytoskeleton is involved in mitoaggregate disassembly. Interestingly, we observed F-actin formation (using the yeast-derived F-actin marker LifeAct-EGFP 11 , Fig. 2a Fig. 2a, b), near Parkin-labeled mitochondria. Phalloidin staining also confirmed that F-actin structures formed in proximity to the mitoaggregates, both in cells expressing and not expressing LifeAct-EGFP ( Supplementary Fig. 2c-e). F-actin structures selectively appeared in the subcellular regions where Parkin-labeled mitoaggregates resided ( Fig. 2c and Supplementary Fig. 3a). The same phenomenon can also be observed in HEK293 cells, a cell line that possesses high levels of endogenous Parkin ( Supplementary Fig. 3b). The F-actin structures were not observed in HeLa cells lacking Parkin ( Supplementary Fig. 3c), suggesting that F-actin assemblies are spatially coupled to Parkinmediated mitophagy. The F-actin assemblies observed were transient and circular in dynamics (Fig. 2e, f), and occurred specifically during mitoaggregate disassembly (90-100 min following damage) (Fig. 2d). To better visualize their temporal overlap, we photoimpaired a small number of mitochondria within HeLa cells (Fig. 2g, h). As shown in Fig. 2i, multiple rounds of circular F-actin generation, which initially occurred 70 min following mitochondrial damage, accompanied mitoaggregate disassembly. The total LifeAct-EGFP intensities increased concurrently with Parkin-labeled mitochondrial particle count within light-illuminated areas (Fig. 2j). All of the above support a functional link between the observed circular F-actins and mitoaggregate disassembly.
Our observed circular F-actin differ in form and function from other LifeAct-EGFP imaged F-actins during mitochondrial fission. For example, LifeAct-EGFP has been used to reveal transient actin assembly along tubular mitochondria during mitochondrial fission 12 . In our assay, the dynamics and temporal occurrence of this fission-related actin cycling, when detected, were distinct from that of circular F-actin observed during mitoaggregate disassembly (Supplementary Fig. 5a-b). These revealed that mitoaggregate disassembly-associated circular Factins are different from those that drive mitochondrial fission.
Actinomyosin-dependent mitoaggregate disassembly. The functional relevance of actin polymerization on mitoaggregate disassembly was further evaluated using actin nucleation factor inhibitors. Mitoaggregate disassembly ( Fig. 3a and Supplementary  Fig. 6a) and circular F-actin formation (Fig. 3b, c and Supplementary Fig. 6b) were both significantly inhibited by treatment ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08924-5 with CK666 13 (an inhibitor of the actin nucleation factor Arp2/3) but not SMIFH2 14 (an inhibitor of the linear actin nucleation factor formin), indicating that Arp2/3-dependent actin mediates mitoaggregate disassembly. That mitoaggregate disassembly is formin-independent again confirms that the circular F-actin observed is distinct from fission-related actin (requires formin as reported) 12,15 . We also examined whether myosin II contributes to mitoaggregate disassembly, considering that interplay between myosin II and F-actin generates cellular force. Pre-treatment of cells with blebbistatin (a myosin II inhibitor) 16 before light-induced mitophagy resulted in the formation of abnormal mitoaggregates composed of tubular mitochondria rather than grape-like clusters ( Supplementary Fig. 5c-e). Given the involvement of myosin II in mitochondria fission 17 , these abnormal mitoaggregate structures likely arose from a block in mitochondria fission before mitoaggregate formation, further supporting the notion that grape-like mitoaggregates form after mitochondrial fission. To more specifically test whether myosin II partakes in mitoaggregate disassembly, we applied blebbistatin 40 min after photoillumination to allow the normal progression of   ) on the cellular region undergoing mitophagy in a after 635 nm illumination. Upon initial Parkin-labeling (left), mitochondria underwent biphasic morphology changes, including clustering (middle) and disassembly (right). c, d Analysis of mitoaggregate connectivity by FRAP. c MitoTracker DeepRed FM-stained HeLa cells co-expressing EBFP2-Parkin, EGFP-dMito, and KillerRed-dMito were illuminated with 559 nm to generate Parkin-labeled mitoaggregates (white-dotted region). EGFP-dMito within a small portion of clustered mitoaggregates were photobleached by 488-nm light (white circle). d Magnified view of clustered Parkin-labeled mitoaggregates. EGFP-dMito within the 488-nm illuminated regions (white arrow) were photobleached and monitored for recovery over a course of 30 min (none observed). A non-photobleached regions (red arrow) served as an internal controls. e-g CLEM analysis of Parkin-labeled mitoaggregates. e MitoTracker DeepRed FM-stained HeLa cells co-expressing EBFP2-Parkin and EGFP-dMito were illuminated with 635 nm to generate Parkin-labeled mitoaggregates. The cells were fixed 90 min after illumination and analyzed by electron microscopy. Scale bar: 10 μm. f TEM image correlating to the region of interest under fluorescence microscopy (white square region in e). g Magnified view of the white-dotted square region in f. Grape-like mitoaggregates were larger than mature autophagosomes (red arrow). Scale bar: 1 μm. h The expanded view shows that there are high electron densities at the interface of mitoaggregates NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08924-5 ARTICLE mitochondrial fission that occurs immediately upon mitochondrial damage as reported 18,19 (Supplementary Fig. 5f). Phenocopying the effect of CK666, this blebbistatin treatment led to defective disassembly of mitoaggregates (Fig. 3a). Another myosin II inhibitor, ML7, also blocked mitoaggregate disassembly and circular F-actin formation (Fig. 3b, c). All of the above point to a role of the actinomyosin system in driving mitoaggregate disassembly.
Mitoaggregate disassembly vs. autophagy initiation. Nondisassembled mitoaggregates (through Myosin II inhibition) were labeled by EGFP-LC3B ( Fig. 3d and Supplementary   These results demonstrated that EGFP-LC3B/NDP52 labeling seen on myosin II-blocked long-lived mitoaggregates did not represent complete autophagic engulfment of the mitoaggregates (LC3B may have become directly conjugated onto mitochondria owing to the observed colocalization between LC3B and Parkin-labeled mitoaggregates with blebbistatin treatment. However, under these conditions mRFP-EGFP-LC3B can still be used to track the delivery of damaged mitochondria into lysosome). Also, these observations let us hypothesize that actinomyosin-dependent mitoaggregate disassembly reduces mitochondrial dimensions for efficient sequestration by autophagic membranes. Driven by the above findings, as well as the fact that the actinomyosin system participates in multiple stages of the autophagy process 20 , we further investigated if mitoaggregate disassembly is functionally linked to autophagy initiation during Parkin-mediated mitophagy. We found that the BATS domain, the carboxyl terminus of Atg14L, displayed circular dynamics at where mitoaggregates disassembled ( Fig. 4a, b). BATS marked autophagy initiation sites during mitophagy, as they colocalized with Atg14L in fixed cells on Parkin-labeled mitochondria, and were spatially associated with autophagosome formation (Supplementary Fig. 9a-c). This is consistent with the fact that BATSlabeled structures are considered autophagy initiation sites due to their colocalization with Atg16 and Beclin 1 in starved cells 21,22 . Importantly, mitophagy-associated BATS structures appeared near circular F-actin at mitoaggregates ( Fig. 4c and Supplementary Fig. 9d), and displayed synchronous formation dynamics (Fig. 4d). Blocking mitoaggregate disassembly by CK666 inhibited the formation of mitoaggregate-associated BATS structures (Fig. 4e). Circular F-actin mediated mitoaggregate disassembly is therefore functionally linked to autophagosome generation against Parkin-labeled mitoaggregates.
Retarding disassembly blocks autophagy initiation. We further devised an optogenetic approach (Fig. 5a) to suppress disassembly through reversible crosslinking of mitoaggregate interfaces, to demonstrate that mitoaggregate disassembly is required for efficient autophagosome formation against damaged mitochondria. To achieve this, EGFP-Ub (or TagRFP-Ub) was fused to CRY2, a domain that undergoes oligomerization in response to blue light 23 . As shown in Fig. 5b, CRY2-EGFP-Ub colocalized with Parkin-labeled mitochondria 80 min after mitochondrial damage, indicating that CRY2-EGFP-Ub is able to be selectively conjugated onto mitoaggregates. The spatial enrichment of CRY2 at the mitoaggregate interfaces makes it possible to extend the duration of mitoaggregate clustering through blue light illumination. In our experiments, mitoaggregates were selectively illuminated by 488 nm light for 1 min twice, 80 and 140 min after mitochondrial damage ( Fig. 5b; dotted circles). Using this scheme we found that mitoaggregates were able to remain in the clustered form for up to 5 h (Fig. 5b, c; blue dotted region). As shown in Fig. 5c, these photoclustered, long-lived mitoaggregates lacked lysosomal markers and did not display EGFP quenching (CRY2-EGFP-Ub fluorescence would be lost in acidic lysosomes). This again demonstrated that Parkin-labeled mitoaggregates cannot be efficiently targeted into lysosomes for turnover in the absence of disassembly. Next we used the optogenetic photoclustering scheme to show that a direct block on mitoaggregate disassembly affects autophagy initiation against damaged mitochondrial fragments (Fig. 5d). We damaged two separate groups of mitochondria using 635-nm light within a cell (Fig. 5e); one mitoaggregate was subsequently photoclustered using 488-nm light, while the other was left as an internal control. 488-nm illumination caused local photobleaching in the green emission channel (Fig. 5f). However, because cytosolic hrGFP-BATS freely diffuses throughout the cytoplasm, the formation of BATS-labeled autophagy initiation sites can still be monitored and counted. Unlike at the control region, which showed increased BATSpositive structures during mitoaggregate disassembly ( Fig. 5f; top mitoaggregates), photoclustered mitoaggregates displayed reduced BATS counts ( Fig. 5f; bottom mitoaggregates and Fig. 5g). Apparent BATS structures on photoclustered mitoaggregates only occurred 5 h following mitochondrial damage ( Fig. 5f; red arrows). These BATS-sites grew only on the surfaces of photoclustered mitoaggregates, reminiscent of TEM images of blebbistatin-treated cells, where ISMs only showed up sparsely on the surface of mitoaggregates (Fig. 3e, f). Similarly, the circular Factin counts near photoclustered mitoaggregates were reduced ( Supplementary Fig. 10a, b). Collectively, these results again indicated that disassembly is not only necessary to dimensionally fit mitoaggregates into autophagic membranes but also to make them accessible to autophagy initiation. PIP 2 regulates disassembly and autophagy initiation. It is known that BATS domain confers binding ability to PtdIns(4,5) P 2 in vitro 22 , and this phospholipid association is essential for Atg14 complex formation and autophagy initiation. PtdIns(4,5)P 2 is also known to regulate F-actin formation 24 . Thus, we asked whether PtdIns(4,5)P 2 co-regulate mitoaggregate disassembly as well as the associated BATS-site formation. Mitoaggregateassociated BATS sites were not affected by the depletion of  Fig. 11a). Meanwhile, intracellular PtdIns(4,5)P 2 were detected near disassembled mitoaggregates through immunofluorescence ( Supplementary Fig. 11b). A tandem PH domain of PLCδ fused to EGFP was also utilized to monitor intracellular PtdIns(4,5)P 2 levels during mitophagy 25 . With the exception of the PtdIns(4,5)P 2 -enriched plasma membrane, we observed intracellular PtdIns(4,5)P 2 generation near Parkin-labeled mitochondria during mitoaggregate disassembly in HeLa cells expressing low levels of PH(PLCδ)x2-EGFP (Fig. 6a,  b), indicating a role for PtdIns(4,5)P 2 in the regulation of mitoaggregate disassembly and BATS-site generation.

Discussion
In this study, we discovered the presence of a pool of PtdIns(4,5) P 2 adjacent to omegasomes, and that they coordinate mitoaggregate disassembly with autophagy initiation during Parkinmediated mitophagy. Damaged mitochondria first become grape-like mitoaggregates. PtdIns(4,5)P 2 then coordinates mitoaggregate disassembly with autophagy initiation to allow for Step 1: 635 nm photoinduced mitophagy initiation Step 2: Mitoaggregates labeled by Parkin Step  The growing body of reports has indicated strong functional links between the actin cytoskeleton and autophagosome formation during starvation-induced autophagy [28][29][30][31][32][33] . Likewise, the mitophagy-associated F-actin could participate in the early stage of autophagosome formation owning to its synchronous formation with BATS-labeled autophagy initiation structures. Unlike CapZ-mediated PtdIns(3)P-dependent actin assembly in starvation-induced autophagy 33 , mitophagy-associated F-actin generation requires PtdIns(4,5)P 2 . We speculate that mitophagyassociated circular F-actin drive autophagosome generation by either serving as a scaffold for BATS (Atg14L) initiation structure formation or by disrupting mitoaggregates to increase the available surface accessible to ISMs growing from omegasomes. Threedimensional structured-illumination microscopy (3D-SIM) revealed that circular F-actins represented curved sheet rather than spherical cages surrounding Parkin-labeled mitochondria ( Supplementary Fig. 4a). Synchronous formation of BATS and circular F-actin near mitoaggregates indicates that it is possible to simultaneously observe circular F-actin, omegasomes, and BATS structures in TEM images. However, we failed to observe cradlelike structures 27 , which have previously been reported as omegasomal structures, and circular F-actin in our TEM images of glutaraldehyde-fixed samples. In this regard, it will be interesting to map the omegasomal structures (e.g., through the use of APEX-DFCP1) and circular F-actin in TEM images of cryogenically preserved samples to visualize transient and vulnerable structures that are not easily preserved by chemical fixation.
How may cells determine the onset of mitoaggregate disassembly? Given that Pink1-dependent phosphorylation on Parkin and Ub initiate the cascade to drive mitochondrial clustering, termination of Pink1/Parkin activity may therefore be involved in the disassembly step. Declining Pink1 activities could allow for subsequent dephosphorylation of poly-Ub and de-ubiquitination, resulting in a tendency for the mitoaggregates to disassemble. Indeed, we observed that phosphorylation levels of Ub at Ser65, a substrate of Pink1 kinase 34 , were increased on clustered mitoaggregates, and decreased upon their disassembly ( Supplementary  Fig. 12c, d). Therefore, it will be interesting to examine whether non-disassembled mitoaggregates [via photoclustering or depleting omegasome-proximal PtdIns(4,5)P 2 ] cause delayed dephosphorylation of Ub on Ser65, or vice versa, whether delayed dephosphorylation of Ub on Ser65 (by expressing constitutively active Pink1 35,36 ) affects PtdIns(4,5)P 2 -dependent mitoaggregate disassembly. FRB-DFCP1-PJ-Dead depleted PtdIns(4,5)P 2 near omegasomes but did not disturb the PtdIns(3)P-dependent DFCP1 localization on mitophagic substrates (Fig. 6d, e, g). On the other hand, wortmannin experiments showed that depleting PtdIns(3)P did not influence PtdIns(4,5)P 2 -dependent BATS formation (Supplementary Fig. 11a). These results suggest that PtdIns(4,5)P 2 and PtdIns(3)P signaling operate in parallel. This also suggests that PtdIns(3)P is not the sole signaling lipid for autophagy initiation; rather, PtdIns(4,5)P 2 also plays a role, indicating another level of autophagy regulation. Increasing evidence suggests that PtdIns(4,5)P 2 participates in autophagy, such as SNX18-dependent 37 and Arf6-dependent 38 autophagosome formation and Mss4-dependent mitophagy in yeast 39 . Whether PtdIns(4,5)P 2 is locally synthesized or derived from pre-existing membrane sources remains an interesting topic for future study. Indeed, it has been reported that Atg14L interacts with PIP5KIC 40 , indicative of PtdIns conversion from PtdIns(4)P to PtdIns(4,5)P 2 during autophagy initiation. Thus, it will also be interesting to examine whether PtdIns(4)P-enriched Golgi serves as a donor for PtdIns(4,5)P 2 generation near omegasomes.
Why do cells go through the trouble of clustering the to-beturned-over mitochondria, only to have them disassembled at a later stage for autophagic turnover? One reason is that this prevents the possibility of overwhelming the autophagy system, as well as reduces cellular stress. Dysfunctional mitochondria are first clustered for the reduction of cellular toxicity. Depending on the capacity and availability of the autophagy machinery, cells will then fine tune their rates of mitoaggregate disassembly so as to minimize the presence of free floating dysfunctional mitochondria in the cytoplasm. In accordance with this idea, we observed that mitoaggregate disassembly occurred at the autophagosome generating platform omegasomes, indicating that disassembly can only be as fast as the rate of autophagosome generation. In summary, our findings on the roles of omegasome-proximal PtdIns(4,5)P 2 in Parkin-mediated mitophagy, together with recent reports of early-onset Parkinsonism-associated synaptojanin 41,42 , may underlie the pathogenic roles of PtdIns metabolism. Further efforts to reveal the interplay between PtdIns (4,5)P 2 -and PtdIns(3)P-dependent signaling axes near the omegasome will provide new insights into mitochondrial autophagy and the pathogenesis of Parkinson's disease.
Fragmentation analysis and BATS/actin, DFCP1-sites, or PH(PLCδ)x2-EGFP intensity calculation. To count the number of EBFP2-Parkin labeled mitochondrial particles, images were thresholded, followed by particle identification using the Analyze Particles function in Image J. To count the number of BATS or DFCP1 structures near mitophagic substrates, regions containing EBFP2-Parkin labeled mitochondria were defined through the Color Threshold function in Image J. The total counts of BATS or DFCP1 within these defined regions were measured using the Analyze Particles function in Image J. The total counts of BATS were binned into 10 min intervals to calculate the total number of BATS as shown in Fig. 5g. The total number of circular actin was manually calculated. The raw intensities of PH(PLCδ)x2-EGFP within EBFP2-Parkin labeled mitochondria were calculated through the Measure function in Image J and were divided by area of EBFP2-Parkin labeled mitochondria to calculate mean PH(PLCδ)x2-EGFP intensities on mitoaggregates.
Quantifying mitophagosome maturation through mRFP-EGFP-LC3B. Quantification of mRFP-EGFP-LC3B fluorescence on EBFP2-Parkin labeled mitochondria was performed using Image J. In brief, regions containing EBFP2-Parkin labeled mitochondria were defined through the Color Threshold function in Image J. The total mRFP and EGFP signals within these identified regions were measured using the Measure function in Image J.
Structured-illumination microscopy. HeLa cells seeded on ibidi Grid-500 glass bottom (81168) dishes co-expressing EBFP2-Parkin, LifeAct-EGFP, and KR-dMito were 559 nm illuminated to activate for mitophagy. Cells were time-lapse imaged under a confocal microscope, and fixed with 4% paraformaldehyde when circular F-actin assemblies appear around Parkin-labeled mitochondria. The fixed samples were further imaged under a Zeiss ELYRA S1.
Statistical analysis. No statistical methods were used to predetermine sample size. Experiments were not randomized, and the investigators were not blinded to allocation during experiments and outcome assessment.

Data availability
The data sets generated from this manuscript are available from the corresponding author upon reasonable request.