LC3/GABARAPs drive ubiquitin-independent recruitment of Optineurin and NDP52 to amplify mitophagy

Current models of selective autophagy dictate that autophagy receptors, including Optineurin and NDP52, link cargo to autophagosomal membranes. This is thought to occur via autophagy receptor binding to Atg8 homologs (LC3/GABARAPs) through an LC3 interacting region (LIR). The LIR motif within autophagy receptors is therefore widely recognised as being essential for selective sequestration of cargo. Here we show that the LIR motif within OPTN and NDP52 is dispensable for Atg8 recruitment and selectivity during PINK1/Parkin mitophagy. Instead, Atg8s play a critical role in mediating ubiquitin-independent recruitment of OPTN and NDP52 to growing phagophore membranes via the LIR motif. The additional recruitment of OPTN and NDP52 amplifies mitophagy through an Atg8-dependent positive feedback loop. Rather than functioning in selectivity, our discovery of a role for the LIR motif in mitophagy amplification points toward a general mechanism by which Atg8s can recruit autophagy factors to drive autophagosome growth and amplify selective autophagy. Selective autophagy receptors are thought to selectively recruit Atg8 positive membranes to cargo via their LIR motif. Here, the authors show the LIR motifs in OPTN and NDP52 are dispensable for selectivity, functioning instead to recruit additional receptors and amplify mitophagy.

Atg8 family members are required for efficient autophagosome formation and can regulate autophagosome size [35][36][37][38] . Despite their importance during PINK1/Parkin mitophagy, recent evidence indicates that the Atg8 family does not play an essential role in selective sequestration of mitochondria. Cells lacking Atg8s can successfully sequester mitochondria within autophagosomes 38 , and Atg8 lipidation deficient cells also contain autophagic membranes surrounding mitochondria 30 . Given that Atg8s do not mediate selective recognition, the role of LIRmediated interactions between autophagy receptors and Atg8 family members during PINK1/Parkin mitophagy is unclear.
In this study, we discover that the LIR motif within OPTN or NDP52 is not essential for either LC3 or GABARAP subfamily recruitment. However, the LIR motif plays an important role in driving ubiquitin-independent recruitment of OPTN and NDP52 to autophagic membranes following autophagosome initiation. The post-initiation recruitment of OPTN and NDP52 promotes additional ULK1 complex recruitment resulting in an amplification of autophagosome biogenesis and mitophagy. All five of the major ubiquitin binding autophagy receptors are capable of LIRmediated recruitment via Atg8s, but only OPTN and NDP52 robustly amplify PINK1/Parkin mitophagy. In silico modelling supports the existence of an Atg8-dependent positive feedback loop of autophagy receptor recruitment and autophagosome formation that amplifies mitophagy. This may represent a general mode of signal amplification in other selective autophagy pathways.

Results
Analysis of the LIR motif during PINK1/Parkin mitophagy. To determine the role of the LIR motif during PINK1/Parkin mitophagy, we first assessed the Atg8 recruitment profile of OPTN and NDP52 following PINK1/Parkin activation. LIR amino acid sequences vary between autophagy receptors, with different Atg8 binding specificities reported for each protein. For example, the LIR motif in OPTN has a high affinity for GABARAP 39 , but the affinity can switch toward LC3B when OPTN is phosphorylated at S177 by TANK binding kinase (TBK1) 40 . NDP52 contains a non-canonical LIR motif (also termed CLIR) that is highly selective for LC3C binding 23 . To assess the Atg8 recruitment profile of OPTN and NDP52 during PINK1/Parkin mitophagy, we utilised penta knockout (KO) cells which lack the five major autophagy receptors (OPTN, NDP52, TAX1BP1, NBR1 and p62) 14 .
Atg8 localization was assessed using HA-tagged variants of all six mammalian Atg8s (LC3A, LC3B, LC3C, GBRP, GBRPL1 and GBRPL2), to avoid the functional impairment associated with GFP-tagged LC3/GABARAPs 14,38 . Penta KO cell lines stably expressing untagged Parkin and individual HA-Atg8s were rescued with either GFP-OPTN or GFP-NDP52, before treatment with oligomycin and antimycin A (OA) for 3 h to activate PINK1/ Parkin mitophagy. HA-tagged Atg8s were not recruited to mitochondria in the absence of autophagy receptor expression (Fig. 1a), whereas all HA-tagged Atg8s were recruited in penta KOs rescued with either GFP-OPTN or GFP-NDP52 (Fig. 1b, c). Neither OPTN nor NDP52 displayed a preference for a particular Atg8 family member during mitophagy (Fig. 1b, c).
The LIR motif promotes efficient mitophagy. Our results reveal that the LIR motif is not essential for Atg8 family recruitment nor selectivity, however, several studies have demonstrated the importance of the LIR motif for selective autophagy 17,23,24 . We therefore assessed the mitophagy activity of LIR-mutant GFP-OPTN(F178A) and GFP-NDP52(V136S). Penta KO cells rescued with GFP-tagged WT and LIR-mutant autophagy receptors, or p62 as a negative control 14 , were treated with OA for 21 h before immunoblotting for the mitochondrial DNA encoded protein CoxII to assess mitophagy (Fig. 2a, b). LIR-mutant GFP-OPTN (F178A) robustly degraded CoxII similarly to WT GFP-OPTN, whereas GFP-NDP52(V136S) had a significant mitophagy defect relative to WT GFP-NDP52. Despite the mitophagy deficiency observed for LIR-mutant GFP-NDP52(V136S), it could still drive mitophagy to a significant degree when compared to the penta KO control which lacks autophagy receptor expression ( Fig. 2a; compare lanes 2 and 7, 2b). The ability of OPTN(F178A) and NDP52(V136S) to drive CoxII degradation indicates that the LIR motif is not essential for mitochondrial clearance. However, the decrease in mitophagy levels observed for NDP52(V136S) pointed toward a defect in mitophagy efficiency. Given that  mitophagy efficiency defects may only be apparent at earlier time points 24 , we utilised mtKeima to measure mitochondrial delivery to lysosomes during the first 3 h of OA incubation ( Fig. 2c-h; Supplementary Fig. 1j). Indeed, LIR-mutant GFP-OPTN(F178A) was observed to have a significant defect in the rate of mitophagy ( Fig. 2c-e), while GFP-NDP52(V136S) had a much greater mitophagy defect (Fig. 2f-h) consistent with the CoxII degradation data (Fig. 2a, b). Expression of GFP-p62 failed to rescue mitophagy in any of the conducted assays ( Fig. 2b; Supplementary Fig. 1f, i). Thus, the LIR motif within OPTN and NDP52 functions to drive efficient mitochondrial clearance during PINK1/Parkin mitophagy.
To determine how the LIR motif might contribute to mitophagy efficiency, we analysed the stages of mitophagy following autophagy receptor recruitment. OPTN and NDP52 play a key role during autophagosome initiation by promoting the recruitment of the ULK1 complex 14 . We therefore quantified the formation of autophagosome initiation foci containing the ULK1 complex subunit Atg13 29 . Penta KO cells expressing WT or LIRmutant autophagy receptors were immunolabelled for endogenous Atg13 after an OA treatment time-course. The analysis revealed that Atg13 foci in penta KO cells expressing GFP-OPTN (F178A) were significantly smaller than those in cells expressing WT GFP-OPTN (Fig. 3a-c), although the overall number of Atg13 foci was not significantly different (Fig. 3b). In contrast, cells expressing GFP-NDP52(V136S) contained both fewer and smaller Atg13 foci relative to cells expressing WT GFP-NDP52 (Fig. 3a, e, f). Recruitment of the initiation machinery to mitochondria by WT and LIR-mutant autophagy receptors was also analysed using subcellular fractionation ( Supplementary  Fig. 3f,g and i,j). Mitochondrial fractions from penta KO cells expressing OPTN(F178A) or NDP52(V136S) had reduced levels of the ULK1 complex subunit FIP200 relative to WT controls, corroborating the microscopy analyses. Consistent with the defects in Atg13 foci formation (  Fig. 2d-f) were also defective in cells expressing LIR-mutant autophagy receptors. Penta KO cells either lacking autophagy receptor expression or expressing GFP-p62 failed to initiate mitophagy as demonstrated by a lack of Atg13 foci ( Fig. 3a; Supplementary Fig. 2a, d). Importantly, our analyses of Atg13 foci also revealed that the intensity of GFP-OPTN(F178A) and GFP-NDP52(V136S) at phagophores was significantly reduced relative to WT GFP-OPTN (Fig. 3d) and GFP-NDP52 (Fig. 3g). Taken together, our results reveal that mutation of the LIR motif in OPTN and NDP52 causes an autophagosome formation defect at early stages of mitophagy including phagophore formation. In addition, the reduction of autophagy receptor intensity at phagophore formation sites indicates that mutation of the LIR motif affects events upstream and/or during autophagosome initiation.
We next asked whether the reduced intensity of LIR-mutant OPTN and NDP52 at autophagosome formation sites was caused by reduced autophagy receptor translocation. GFP-OPTN (F178A) (Fig. 3h, j) and NDP52(V136S) (Fig. 3i, l) translocated significantly slower than their wild-type counterparts. Mitochondrial translocation of NDP52(V136S) was particularly low, consistent with the greater mitophagy defect observed in penta KO cells expressing this mutant (Fig. 2). Defects in mitochondrial translocation of OPTN(F178A) and NDP52(V136S) were also detected via analysis of isolated mitochondrial fractions (Supplementary Fig. 3f-k). To confirm that reduced translocation of LIRmutant OPTN and NDP52 was due to their inability to bind Atg8s, and to exclude the potential effect of protein alteration from mutated residues, we then assessed WT GFP-OPTN and GFP-NDP52 translocation rates in Atg8 hexa KO cells that lack all six LC3/GABARAPs (  Supplementary Fig. 2r), which displayed very similar translocation rates to the LIR mutants expressed in penta KO cells (Fig. 3j, l). Autophagy receptor translocation during PINK1/Parkin mitophagy is thought to be entirely dependent on binding to ubiquitin chains, with both OPTN and NDP52 previously shown to preferentially bind to ubiquitin chains phosphorylated at S65 by PINK1 14,17 . However, analysis of pS65-Ub levels on mitochondria in WT and hexa KO cells confirmed that hexa KO cells did not have reduced levels of pS65-Ub chains ( Supplementary Fig. 3a), thus eliminating the possibility that altered pS65-Ub levels had affected autophagy receptor translocation. Collectively, these results demonstrate that OPTN and NDP52 recruitment efficiency during PINK1/Parkin mitophagy is directly influenced by LIR-motif-mediated interactions with LC3s and/or GABARAPs.
LIR-mediated recruitment of autophagy receptors in mitophagy. OPTN and NDP52 recruitment to mitochondria during mitophagy is dependent on ubiquitin binding 14,17,40 . It is therefore unclear how Atg8s can influence OPTN and NDP52 recruitment. We hypothesised that following the initiation of autophagosome biogenesis, the LIR motif within OPTN and NDP52 may enable ubiquitin-independent recruitment of the autophagy receptors via binding to lipidated Atg8s ( Fig. 4a; Supplementary Fig. 4a). Thus, we sought to determine whether the LIR motif was sufficient to mediate the translocation of autophagy receptors independently of ubiquitin binding. First, we analysed the localization of OPTN(D474N) and NDP52(C443K) UBD mutants in penta KO cells following PINK1/Parkin mitophagy activation (Fig. 4b). As expected, both mCherry(mCh)-NDP52(C443K) (Fig. 4c) and mCh-OPTN(D474N) (Fig. 4d) failed to translocate to mitochondria after 3 h OA incubation corroborating previous studies 14,41 .
Next, we determined whether UBD mutant autophagy receptors can translocate via their LIR motif during phagophore formation by co-expressing a WT autophagy receptor (Fig. 4e).
A positive feedback model of selective autophagy. Our results point toward a model in which OPTN and NDP52 are initially recruited to mitochondria via their UBD. After their recruitment, OPTN and NDP52 promote the recruitment and activation of the ULK1 complex 14 . Subsequent activation of the phosphoinositide 3-kinase complex by ULK1 stimulates local production of PtdIns (3)P at the phagophore nucleation site 29 , which recruits WIPI2b 34 . Recruitment of the Atg8 conjugation machinery (Atg12-Atg5-Atg16L1 complex) by WIPI2b defines the site of LC3/GABARAP lipidation 33 . The Atg8-positive phagophores then recruit additional OPTN and NDP52 via LIR-mediated interactions, which drives further rounds of ULK1 translocation and Atg8 lipidation. This forms an Atg8-dependent positive feedback loop that amplifies the rate of autophagosome biogenesis. The positive feedback loop is lost in the absence of LIRmediated interactions between autophagy receptors and LC3/ GABARAPs (Fig. 5i). This model also provides a mechanistic explanation for the reduced autophagosome size and reduced autophagosome formation efficiency observed during PINK1/ Parkin mitophagy in cells lacking Atg8s 37,38 .
To test the positive feedback model, we generated a computational model of PINK1/Parkin activation and autophagosome formation. The in silico model combines the PINK1/Parkin pS65-Ub-dependent positive feedback loop 8,9 , and the LIR-dependent positive feedback loop identified in this study. A mathematical kinetic model of autophagy initiation by NDP52 was formulated using a system of 21 ordinary differential equations (ODEs) with starting conditions and rate constants derived from empirical time-course data (Figs 2, 3, 4). The model scheme is shown in Supplementary Fig. 4 and a detailed model description is presented in Methods. The system of ODEs (Supplementary  Tables 1-2) was then numerically solved using Mathematica (Wolfram Research) to quantify total lipidated Atg8 (Atg8PE), which was the surrogate model variable chosen to represent  Supplementary Fig. 5b). b-g Automated image analysis of the average Atg13 foci count per cell (b, e), Atg13 foci volume (c, f) and intensity of foci in the GFP channel (d, g), in penta KO HeLa cells stably expressing untagged Parkin, rescued by stable expression of either GFP-OPTN/GFP-OPTN(F178A) (b-d) or GFP-NDP52/GFP-NDP52(V136S) (e-g), immunostained for Atg13, HSP60 and GFP after time-course incubation with OA. (Data from other time points shown in Supplementary Fig. 2g-j). h, i, j, l Representative immunofluorescence images (h, i) and automated image analysis of GFP-tagged receptor translocation (j, l) in penta KO HeLa cells stably expressing untagged Parkin after rescue expression of either GFP-OPTN/GFP-OPTN(F178A) (h, j) or GFP-NDP52/GFP-NDP52(V136S) (i, l), immunostained for HSP60 and GFP after time-course incubation with OA (times indicated). k, m Automated image analysis of GFP-OPTN (k) or GFP-NDP52 (m) translocation in WT and hexa KO HeLa cells stably expressing untagged Parkin, after time-course incubation with OA and immunostaining for HSP60 & GFP. (Representative images for (k, m) provided in Supplementary Fig. 2q, r). Data in (b-g) and (j-m) are mean ± s.d. from three independent experiments. *P < 0.05, **P < 0.005, ***P < 0.001, ****P < 0.0001 (two-way ANOVA). ns: not significant. Scale bars: overviews, 10 µm; insets, 2 µm NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08335-6 ARTICLE NATURE COMMUNICATIONS | (2019) 10:408 | https://doi.org/10.1038/s41467-019-08335-6 | www.nature.com/naturecommunications mitophagy output of the system. Upon simulated depolarization of mitochondria (reaction 4), the modelled numerical outputs recapitulated the switch-like response of mitophagy induced by WT NDP52 over time (compare Fig. 5h with 2 h). In contrast, the modelled outputs during mitophagy induction by NDP52(V136S) lacked the positive feedback loop and had significantly diminished mitophagy (Fig. 5h). The PINK1/Parkin mitophagy simulations demonstrated a high degree of robustness, as random perturbation of reaction rate constants had a limited effect on the outcomes (Supplementary Fig. 4d). Taken together, the high level of agreement between in silico model simulations and the biological data strongly support the presence of a positive feedback loop mediated by Atg8s and autophagy receptor LIR binding (Fig. 5i). According to the positive feedback model ( Fig. 5i), phosphoinositide 3-kinase (PI3K) inhibitors should block LIR-mediated recruitment of selective autophagy receptors by preventing Atg8 lipidation at initiation sites 28,29,31 . To test the model, the translocation of mCh-NDP52(C443K) (Fig. 5j) and mCh-OPTN(D474N) (Supplementary Fig. 5a) was assessed in the presence of the PI3K inhibitor Wortmannin (Fig. 5j; Supplementary Fig. 5a). Indeed, wortmannin treatment blocked LIRmediated OPTN and NDP52 recruitment ( Fig. 5j; Supplementary  Fig. 5a), without affecting upstream ubiquitin-dependent recruitment of LIR-mutant autophagy receptors (Fig. 5i).
In summary, we have discovered a ubiquitin-independent mechanism of autophagy receptor recruitment which amplifies PINK1/Parkin mitophagy signalling by OPTN and NDP52. We argue that cargo selectivity during PINK1/Parkin mitophagy is achieved via de novo autophagosome biogenesis on the surface of mitochondria 14 , but not through LIR-mediated bridging with Atg8s on autophagic membranes. This is supported by reports showing that cells lacking Atg8s successfully sequester mitochondria within autophagosomes 38 , and LC3/GABARAP-lipidationdeficient cells contain autophagic membranes surrounding mitochondria 44 . By clarifying the role of the LIR motif in mitophagy, we have identified an autophagy associated positive feedback loop that drives efficient mitophagy. PINK1/Parkin mitophagy is therefore regulated by two independent but complementary ubiquitin(-like) positive feedback loops. The first is dependent on phospho-ubiquitin generated by PINK1/ Parkin 8,9 . The second is dependent on the family of ubiquitinlike Atg8 proteins that are lipidated by the E3 ligase-like Atg12-Atg5-Atg16L1 complex 33,34 . Together, the two positive feedback loops enable OPTN and NDP52 to promote rapid clearance of damaged mitochondria, which would help prevent the release of mitochondrial factors that trigger inflammation and cell death 45,46 .

Methods
Cell culture, antibodies and reagents. The penta KO HeLa and hexa KO HeLa cell lines were generated previously 14,38 . All cell lines used in the present study tested negative for mycoplasma contamination. Cell lines were cultured in DMEM supplemented with 10% (v/v) FBS (Cell Sera Australia), 1% penicillin-streptomycin, 25 mM HEPES and GlutaMAX (Life Technologies). Lipofectamine LTX (Life Technologies) transfection reagent was used as per manufacturers specifications.  14 . Point mutants were generated using overlap-extension PCR, and truncation mutants were generated using Gibson assembly. p62(ΔUBD) was generated by truncating p62 at residue 385; 52 TAX1BP1(ΔUBD) was generated by truncating TAX1BP1 at residue 755; 53 NBR1(ΔUBD) was generated by truncating NBR1 at residue 906 54 . The primers used for all mutageneses are provided in Supplementary  Table 3. Plasmids generated during the study are available from AddGene. Stably transfected cell lines were generated using retroviral systems as described previously 14 , and protein expression levels were normalized between all cell lines by fluorescence activated cell sorting (FACS). Expression level data for the autophagy receptors used in this study are provided in Fig. 5e; Supplementary Fig. 1j Immunoblotting. HeLa cells were cultured in 6-well plates for 24 h prior to incubation with either fresh culture medium, EBSS starvation medium, or fresh medium containing 10 μM oligomycin (Calbiochem), 4 μM antimycin A (Sigma) and 10 μΜ QVD (ApexBio) for the times indicated. For immunoblot analysis of cell lysates (Figs 1a, 5e; Supplementary Fig. 5n), cells were lysed in 1 × LDS sample buffer (Life Technologies) supplemented with 100 mM dithiothreitol (DTT; Sigma) and heated to 99°C with shaking for 7-10 min. For isolation of crude mitochondrial fractions (Supplementary Fig. 3a, f, i), cells were homogenized in 20 mM HEPES (pH 7.6), 220 mM mannitol, 70 mM sucrose, 1 mM EDTA, and 0.5 mM phenylmethylsulfonyl fluoride. After centrifugation of the cell homogenates at 800× g for 10 min (4°C), mitochondria in the post-nuclear supernatant were pelleted by centrifugation at 10,000×g for 20 min (4°C). The mitochondrial pellets were lysed in 1× LDS sample buffer (Life Technologies) supplemented with 100 mM dithiothreitol (DTT; Sigma) and heated to 99°C with shaking for 7-10 min. Approximately 25-50 μg of protein per sample was separated on 4-12% Bis-Tris gels (Life Technologies) according to manufacturer's instructions, electrotransferred to polyvinyl difluoride membranes, then immunoblotted using antibodies as indicated. Uncropped data for immunoblots are provided (Supplementary Fig. 7).
Co-immunoprecipitation. Cells were lysed in lysis buffer (1% TritonX-100, 50 mM Tris-HCl pH 7.4, 150 mM NaCl) supplemented with cOmplete protease inhibitor Fig. 4 The LIR motif mediates autophagy receptor recruitment after initiation of autophagosome biogenesis. a Schematic of the proposed mechanism by which Atg8s influence receptor translocation, depicting wild-type receptor recruitment via ubiquitin and Atg8 proteins and LIR-mutant receptor recruitment exclusively via ubiquitin (the upstream stages of this model are in depicted in Supplementary Fig. 4a). b, e, h Model schematics depicting the predicted translocation outcomes in cells expressing a UBD mutant receptor alone (b), co-expressing a UBD mutant receptor and a wild-type receptor (e), or co-expressing a UBD/LIR double mutant receptor and a wild-type receptor (h). c, d, f, g, i, j Representative images of penta KO HeLa cells stably expressing untagged Parkin, with co-expression of either mCh-NDP52(C443K) (c), mCh-OPTN(D474N) (d), GFP-OPTN & mCh-NDP52(C443K) (f), GFP-NDP52 & mCh-OPTN(D474N) (g), GFP-OPTN and mCh-NDP52(V136S/C443K) (i), or GFP-NDP52 and mCh-OPTN(F178A/D474N) (j), after 3 h OA incubation and immunostaining for HSP60, GFP and mCherry. k, l Automated image analysis of mCherry translocation in the cell lines shown in (f and i) (k) or in (g and j) (l), after time-course incubation with OA and immunostaining for HSP60, mCherry & GFP. m, n, o Representative images (m) and automated image analysis of mCherry translocation (n) and GFP translocation (o) in the cell lines shown in (g and j), immunostained for HSP60, mCherry and GFP after 3 h incubation with OA in the presence or absence of 1 µM BX795. Data in (k, l, n, o) are mean ± s.d. from three independent experiments. *P < 0.05, **P < 0.005, ***P < 0.001, ****P < 0.0001 (two-way ANOVA). ns: not significant. Scale bars: overviews, 10 µm; insets, 2 µm cocktail (Roche Applied Science). Lysates were cleared by centrifugation (20,000× g for 10 min at 4°C), and subjected to immunoprecipitation using anti-HA magnetic beads (ThermoFisher). The beads were washed three times with wash buffer (0.1% TritonX-100, 50 mM Tris-HCl pH 7.4, 150 mM NaCl), before elution with 1× LDS sample buffer (Life Technologies) and heating to 99°C with shaking for 7-10 min followed by immunoblot analysis of input and eluted fractions. . The coverslips were rinsed three times with PBS, before mounting on slides using a TRIS buffered DABCO-glycerol mounting medium. All samples were imaged in 3D by optical sectioning using an inverted Leica SP8 confocal laser scanning microscope equipped with an ×63/1.40 objective (Oil immersion, HC PLAPO, CS2; Leica Microsystems). Imaging was conducted with a minimum z-stack range of 2.8 µm and a maximum voxel size of 180 nm laterally (x,y) and 320 nm axially (z). All figure images are displayed as z-stack maximum projections. Images for automated analysis were acquired after blindly adjusting the stage position to image three random fields-of-view per coverslip. The randomization procedure was repeated if the field-of-view contained fewer than 30 cells.
Automated image analysis. All 3D image data were processed and analysed by automated 3D image segmentation using the 3D ROI Manager (v3.93) 56 and FeatureJ (v3.0.0) 57 plugins for FIJI (v1.52 h) 58 . The image analysis workflow was divided into three discrete stages; object detection, object measurement and analysis. Detection and segmentation of mitochondria employed a local thresholding strategy. Each image was pre-processed by a 2D sliding paraboloid background subtraction (50 pixel radius), contrast enhancement (0.2% saturation) and a 2D median filter (1.25 px kernel). Seeds for object based segmentation were retrieved by maximum local filter (x,y,z;8,8,4px) before 3D noise reduction (3D median filter; 2 voxels) and automatic 2D local thresholding using the Bernsen method (4 pixel radius). The 3D ROIs were then extracted using a 3D seed-based watershed (5 pixel radius). Detection and segmentation of Atg13 foci, WIPI2b foci, or Atg16L1 foci employed a global thresholding strategy, to account for the lack of foci in the untreated conditions. Linear histogram normalization between experimental repeats was performed by assembling a montage of maximum intensity projections for all images in a given experiment. The montage was used to calculate new minimum and maximum intensity values, which were, respectively, assigned to a given intensity if the number of pixels with that intensity exceeded 0.01% of all pixels in the montage. The histogram of each individual image was then rescaled between these two values before 3D noise filtering (3D median filter; 1.5 voxels) and extraction of the 3D ROIs using a global threshold (96, arbitrary) via simple 3D thresholding (300 voxel maximum size). Each individual 3D ROI was then applied to the original corresponding image to measure the volume and fluorescence intensity of each segmented object in all available channels. All measurements and ROIs were stored for subsequent analyses using Mathematica (version 11). The mitochondrial volume positive for GFP-tagged autophagy receptors and HA-tagged Atg8s was quantified by inspecting the intensity mean and standard deviation of each segmented object in the corresponding channel. Segmented objects were deemed to be positive for the protein of interest if the mean intensity of the ROI in the corresponding channel was greater than 5 (arbitrary), and the intensity standard deviation exceeded an experiment specific threshold (arbitrary). The following standard deviation thresholds were used for quantification of GFP- Mito-Keima autophagosome-lysosome fusion mitophagy assay. Treated cells were trypsinized then pelleted for resuspension in sorting buffer (1 mM EDTA, 10% FCS in 1x PBS) before analysis using a LSRFortessa cell sorter (BD Biosciences). Measurements of lysosomal mito-Keima were made using dual-excitation ratiometric pH measurements at 488 (pH 7) and 561 (pH 4) nm lasers with 695 nm and 670 nm emission filters, respectively. Additional channels were analysed for samples containing GFP (Ex/Em; 488 nm/530 nm) or iRFP670 (Ex/Em; 628 nm/ 670 nm) for fluorescence compensation. For each sample, 20,000 events were collected and subsequently gated for GFP/Keima double-positive cells, or GFP/ Keima/iRFP670 triple-positive cells where relevant. Data were analysed using FlowJo (version 10). The gate used for mtKeima measurements was defined using the untreated sample from each cell line, by aligning the edge of a triangular gate with the untreated bulk population, then moving the gate to encapsulate 99.8% of the population (Supplementary Fig. 1j, k).
In silico modelling implementation and assumptions. The PINK1/Parkin/ NDP52 mathematical model was formulated using ODEs. The model's schematic diagram containing all model reactions is given in Supplementary Fig. 4. Model ODEs, rate equations and the reference parameter values used for simulations are given in Supplementary Tables 1 and 2. The model was implemented and numerically simulated using Wolfram Mathematica 59 . The model consists of two modules ( Supplementary Fig. 4): a PINK1/Parkin module (Supplementary Fig. 4b) and a NDP52 module (Supplementary Fig. 4c). In response to mitophagy stimulation such as OA, PINK1 is assumed to undergo dimerization and becomes activated. It has been reported that dimerized PINK1 is more stable than PINK1 monomers, thus the degradation rate of PINK1 dimer (reaction 3) is assumed to be smaller than that of the monomer (reaction 2). Activated PINK1 subsequently phosphorylates ubiquitin on Serine 65 (reaction 7), which is dynamically opposed by a reverse dephosphorylation reaction by an unspecified enzyme(s). All phosphorylations and similar conversion reactions were modelled with opposing reactions to enable equilibration of the system, as is common practice in computational modelling of biological systems [60][61][62][63][64] . We assume a positive feedback where Parkin once being recruited by phosphorylated ubiquitin is phosphorylated by PINK1 (reaction 11) and phosphorylated Parkin subsequently promote the availability of ubiquitin for phosphorylation by PINK1 (reaction 5) 1,65,66 .  Supplementary Fig. 3b). b, d automated image analysis of mCherry translocation in the cell lines shown in (a) (b) and in (c) (d) after time-course incubation with OA and immunostaining for HSP60, mCherry and GFP. e, f Penta KO HeLa cells stably expressing untagged Parkin rescued by stable expression the indicated receptor combinations were analysed by immunoblotting after 8 h incubation with OA (e), for quantification of the remaining CoxII levels (f). g FACS analysis of the percentage of 561 nm mtKeima-positive cells in penta KO HeLa cells stably expressing mtKeima, untagged Parkin and GFP-OPTN(F178A) alone, or with co-expression of either iRFP670-NDP52(C443K) or iRFP670-NDP52(V136S/C443K). (representative FACS plots for mtKeima and iRFP670 expression provided in Supplementary Fig. 1k). h In silico modelling of mitophagy using lipidated Atg8 as the surrogate output variable, under wild-type NDP52 or the LIR-mutant NDP52(V136S). (See Methods for details; Supplementary Tables 1-2 for modelled reactions; Supplementary Fig. 4 for overview of the model). i The proposed model of LIR-mediated autophagy receptor coalescence in which autophagy receptors are initially recruited by S65 phospho-ubiquitin. The post-initiation lipidation of Atg8s to the expanding phagophore subsequently recruits more autophagy receptors via their LIR motif. The additional autophagy receptors accelerate autophagosome biogenesis to generate a positive feedback loop. j Representative images of methanol fixed penta KO cells stably expressing untagged Parkin with co-expression of GFP-OPTN(F178A) & mCh-NDP52(C443K), immunostained for LC3A/B, HSP60, GFP and mCherry after 1 h OA treatment in the presence or absence of 1 µM wortmannin. Data in (b, d, f, g) are mean ± s.d. from three independent experiments. *P < 0.05, **P < 0.005, ***P < 0.001, ****P < 0.0001 (two-way ANOVA). ns: not significant. Scale bars: overviews, 10 µm; insets, 2 µm The receptor NDP52 is assumed to bind phosphorylated ubiquitin through its UBD domain (reaction 12). Ubiquitin-bound NDP52 could activate ULK1, which in turn stimulates the conversion of PtdIns to PtdIns(3)P (reactions [14][15][16][17]. PtdIns (3)P controls Atg8 lipidation by indirectly promoting recruitment of the Atg12-5-16L complex, and Atg8 is deconjugated by Atg4 (reactions [18][19]. Importantly, lipidated Atg8 (Atg8PE) can recruit wild-type receptors independently of ubiquitin (reaction 20), because Atg8 can recruit UBD mutant receptors (Figs 4 and 5) and wild-type NDP52 recruitment is defective in hexa KO (Fig. 3). Once NDP52 has been recruited to the phagophore, it is likely it can also interact with ubiquitin (reaction 21) and no evidence exists to suggest otherwise. Similarly, we assume that ubiquitin-bound NDP52 can also bind Atg8 (reaction 13). Atg8-bound NDP52, like ubiquitin-bound NDP52, can also activate ULK1 and thus further promote the production of Atg8PE, generating a positive feedback mechanism. Numerical output of the model used Atg8PE as the surrogate variable for mitophagy.
The parameter values used as the reference set for model analysis and simulations were guided by typical ranges of physiological values and constrained by biologically plausible values 67,68 . Specifically, the rates of protein-protein interactions are given by mass-action (MA) law, and that of (de)phosphorylation reactions are given by Michaelis-Menten (MM) law [60][61][62] . The k on (association) rates are limited by the rate of collisions, which is limited by the rate of diffusion approximately ranging from 0.1 to 10 nM −1 s −1,69 . Michaelis-Menten constants (K m ) typically vary over a broad range and to explore a wide parameter space, they range from 1 to 1000 (nM). Catalytic constants (k c ) are set between 0.0001 to 1 (s −1 ) and the maximal velocities (V m ) from 0.001 to 10 nMs −1 . We further constrained the model parameters by fitting the model to the dynamics of PINK1, phosphorylated ubiquitin (pS65-Ub) and mitophagy obtained in Hela cells ( Supplementary Fig. 3a). The constrained parameter values are displayed in Supplementary Table 1. To simulate the effect of the LIR-mutant NDP52(V136S), the association constants for reactions 13 and 20 were assigned to zero.
Statistical calculations. All statistical comparisons were conducted on data originating from three or more biologically independent experimental replicates (as indicated in figure legends), with similar data variances observed between groups. Comparisons between groups were planned prior to statistical testing, and target effect sizes were not pre-determined. No statistical methods were used to determine sample size. All statistical data were calculated and graphed using GraphPad Prism 7. For multi-variable and time-course data, statistical comparisons between groups were performed by two-way ANOVA with post hoc testing by unprotected Fisher's LSD Test. For data representing a single time-point and condition, statistical comparisons between groups were performed by one-way ANOVA with post hoc testing by unprotected Fisher's LSD Test. P values exceeding 0.05 were considered non-significant.

Data availability
Plasmids generated during the study are available from AddGene. Additional data that support the findings of this study are available from the corresponding authors upon reasonable request.