Recycling of autophagosomal components from autolysosomes by the recycler complex

Autolysosomes contain components from autophagosomes and lysosomes. The contents inside the autolysosomal lumen are degraded during autophagy, while the fate of autophagosomal components on the autolysosomal membrane remains unknown. Here we report that the autophagosomal membrane components are not degraded, but recycled from autolysosomes through a process coined in this study as autophagosomal components recycling (ACR). We further identified a multiprotein complex composed of SNX4, SNX5 and SNX17 essential for ACR, which we termed ‘recycler’. In this, SNX4 and SNX5 form a heterodimer that recognizes autophagosomal membrane proteins and is required for generating membrane curvature on autolysosomes, both via their BAR domains, to mediate the cargo sorting process. SNX17 interacts with both the dynein–dynactin complex and the SNX4–SNX5 dimer to facilitate the retrieval of autophagosomal membrane components. Our discovery of ACR and identification of the recycler reveal an important retrieval and recycling pathway on autolysosomes. Zhou and colleagues identify SNX4–SNX5–SNX17 as a multiunit complex that mediates the recycling of autophagosomal components from autolysosomes.


Results
STX17 is retrieved from autolysosomes. The autophagic SNARE protein STX17 is recruited to autophagosomes and translocated to autolysosome after fusion of autophagosomes with lysosomes [30][31][32] . We observed that the number of autophagic vacuoles with STX17 significantly increased after 2 h of starvation. However, the number of STX17 puncta significantly decreased during prolonged starvation of 5 h (Fig. 1a). Interestingly, no obvious degradation of STX17 was observed after prolonged starvation (Fig. 1b).
To investigate the fate of STX17 on the surface of autolysosomes, we examined the dynamics of STX17 by time-lapse imaging. Since GFP-STX17 blocks autophagosome-lysosome fusion and cannot be translocated to autolysosomes 33 , we used a GFP-tagged transmembrane domain of STX17 (GFP-STX17-TM) for this live imaging. Consistent with a previous report that STX17 fails to form puncta in completely autophagy-deficient cells 32 , the GFP-STX17-TM/ LAMP1 or Flag-STX17/LAMP1 double-positive vesicles are completely eliminated in autophagy-deficient FIP200 knockout and ATG9A knockout cells (Extended Data Figs. 1a,b,e,f and 2a,b,e,f), suggesting that these STX17/LAMP1 double-positive vesicles are autolysosomes.
The live images showed STX17 concentrated on one point of autolysosomal membrane, which then pinched off from the autolysosome ( Fig. 1c and Extended Data Fig. 3a). The signal intensity of STX17 on the bud was approximately fourfold of the signal on the opposite autolysosome membrane (Fig. 1d). The overall intensity of the STX17 signal on the autolysosomal membrane was reduced in a time-dependent manner (Fig. 1c,e). In contrast, the STX17 signal on existing autophagosomes remained relatively unchanged (Fig. 1e). These data suggest that STX17 on autolysosomes was not degraded to a large extent, but instead was retrieved from autolysosomes through a so far uncharacterized sorting process.
One autolysosome took approximately one to four rounds to clear the STX17 (Extended Data Fig. 3b and Supplementary Video 1), and about one event per autolysosome per minute on average (Extended Data Fig. 3c). This sorting process occurs in many other cell lines from different organisms, suggesting the generality and evolutionary conservation of this process (Fig. 1c,h and Extended Data Fig. 3d-f).
To distinguish ALR from this uncharacterized sorting process, we examined both processes simultaneously (Fig. 1f,g). STX17 translocated to autolysosomes through the fusion of autophagosome with lysosomes at 2 h after starvation. However, at 5 h after starvation, the STX17 signal on the autolysosomal membrane was completely undetectable, with almost no ALR occurrence. At 8 h of starvation, ALR was initiated on autolysosomes with no observable STX17 signal. Furthermore, depletion of ALR-essential genes, such as clathrin and AP2, had no effect on this sorting process (Extended Data Fig. 4a,b). These results suggest that this uncharacterized sorting process is completely different from ALR and occurs before ALR.
STX17 concentrated on the acidified autolysosome membrane (indicated by LysoTracker staining, an acid-dependent dye) and budded off (Fig. 1h). STX17 showed lower signal intensity on acidified autolysosomes than non-acidified autophagosomes (Fig. 1i,j). The vesicles budding off are not detectable by LysoTracker staining, despite strong positive staining occurring in the autolysosomal lumen (Extended Data Fig. 5a). Treatment with bafilomycin A1 (a lysosomal v-ATPase inhibitor) at 2 h after starvation to alkalize autolysosomes caused STX17 accumulation in autolysosomes (Extended Data Fig. 5b-d), and this STX17 accumulation is dependent on autophagy (Extended Data Fig. 1c,d,g,h and 2c,d,g,h).
SNX4 and SNX5 are needed for STX17 recycling from autolysosomes. Next, we investigated the molecular component(s) for STX17 recycling from autolysosomes. Since STX17-TM contains the essential components required for STX17 recycling from autolysosomes, we employed APEX2-STX17-TM to biotinylate its functionally related proteins on autolysosomes, followed by semi-quantitative mass spectrometry analysis. A total of 17 sorting nexins were frequently identified in our mass spectrometry list (Supplementary Tables 1-3). We characterized the localization and the STX17-recycling-related impacts for all 17 sorting nexin proteins. Ultimately, SNX4, SNX5 and SNX17 localize to STX17-positive autolysosomes, and their depletion leads to STX17 accumulation on autolysosomes (Extended Data Figs. 6 and 7).
In SNX4 or SNX5 knockdown cells, autophagosomes formed and fused with lysosomes at 2 h after starvation. However, the retrieval of STX17 was completely blocked by SNX4 or SNX5 depletion (Figs. 2a,b and 3a,b and Extended Data Figs. 8a and 9a), suggesting critical roles for SNX4 and SNX5 in STX17 recycling from autolysosomes. These results were further confirmed by the second short interfering RNA (siRNA) and rescue experiments (Extended Data Fig. 10 and Supplementary Figs. 1-3). is retrieved from autolysosomes. a, U2OS cells stably expressing Flag-STX17 were starved with EBSS for the indicated duration and stained with antibody against Flag. Scale bar, 5 μm. Quantification of STX17 puncta (bottom right). Data are mean ± s.d. (n = 3, 50 cells from three independent experiments were quantified), unpaired two-tailed t test. b, U2OS cells stably expressing Flag-STX17 were starved for the indicated duration and treated with or without 20 μM chloroquine (CQ) or 10 μM MG132. Cells were collected and lysed for immunoblotting with the indicated antibodies. c, U2OS cells stably expressing GFP-STX17-TM, LAMP1-mCherry were starved for 2 h with EBSS, and time-lapse images were taken. Scale bar, 1 μm. The arrows indicate the STX17 budding site. d, Fluorescence signal of STX17 on autolysosomes from the selected image in c. inset scale bar, 1 μm. e, Quantification of the fluorescence intensity of STX17 on the autolysosome in c. f, U2OS cells stably expressing GFP-STX17-TM, LAMP1-mCherry and CFP-LC3 were starved with EBSS for the indicated duration. Scale bar, 5 μm. inset scale bar, 1 μm. The arrow in the first line of images indicates isolation membrane. The arrows in the second line of images indicate autophagosomes. The arrows in the third line of images indicate STX17 + autolysosomes. The arrows in the fourth line of images indicate round STX17autolysosomes. The arrows in the fifth line of images indicate STX17tubular autolysosomes. g, Quantification of the puncta in the cells from f at the indicated timepoints. Data are shown as a percentage of the total puncta number (n = 3, 50 cells from three independent experiments were quantified). h, MEF cells stably expressing GFP-STX17-TM were starved with EBSS for 2 h and stained with LysoTracker. Time-lapse images were taken. Scale bar, 1 μm. The arrows indicate the STX17 budding site. i, U2OS cells stably expressing GFP-STX17-TM were starved with EBSS for 2 h and stained with LysoTracker. Scale bar, 5 μm. inset scale bar, 0.5 μm. j, Left: fluorescent intensity of STX17 and LysoTracker on autophagosomes and autolysosomes in i. Data are mean ± s.d. (n = 3, more than 50 cells from three independent experiments were quantified), unpaired two-tailed t test. Right: fluorescent intensity of STX17 on autophagosomes and autolysosomes. inset scale bar, 0.5 μm. Source numerical data and unprocessed blots are available in source data. Data are mean ± s.d. (n = 3, 50 cells from three independent experiments were quantified), unpaired two-tailed t test. c, U2OS cells stably expressing GFP-STX17-TM, mKATE2-SNX4 and LAMP1-CFP were starved with EBSS for 2 h, and live images were taken. Scale bar, 5 μm. inset scale bar, 1 μm. d, U2OS cells stably expressing Flag-STX17 with mKATE2-SNX4 were starved with EBSS for 2 h and stained with antibodies against Flag, LC3 and LAMP1. Scale bar, 5 μm. inset scale bar, 2 μm. e, U2OS cells stably expressing GFP-STX17-TM, mKATE2-SNX4 and LAMP1-CFP were starved with EBSS for 2 h, and videos were taken. Scale bar, 1 μm. The arrows indicate the STX17 budding site. f, HEK293T cells were transfected with empty vector or Flag-STX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with indicated antibodies. g, Co-immunoprecipitation (Co-iP) analysis of interactions between Flag-SNX4 with HA-STX17 in HEK293T cells. h, Glutathione sepharose beads bound with GST-STX17 were incubated with purified His-SNX4 for 16 h, and then eluted for immunoblotting. i, HEK293T cells were transfected with empty vector, HA-SNX4 with or without truncated variants of STX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with the indicated antibodies. j, HEK293T cells were transfected with empty vector or HA-STX17 with or without truncated variants of SNX4. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with the indicated antibodies. k, HEK293T cells were transfected with empty vector or HA-SNX4 BAR with or without truncated variants of STX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with the indicated antibodies. Source numerical data and unprocessed blots are available in source data.   k SNX4 recycles transferrin receptors through interaction with KIBRA on early endosomes in mammals 34 . Knockdown of KIBRA has no effect on STX17 recycling ( Fig. 2b and Extended Data Fig. 8b,e). SNX4 also forms heterodimers with SNX7 and SNX30 35,36 . Knockdown of SNX7 and SNX30 also caused no observable effect on STX17 recycling from autolysosomes ( Fig. 2b and Extended Data Fig. 8c-e). Together, these results show that SNX4 functions in STX17 recycling from autolysosomes independently of both the SNX4-KIBRA complex and the SNX4-SNX7/SNX30 complex.
SNX4 co-localized with STX17 on the rings and buds of autolysosomes (Fig. 2c, d). SNX5 localized only to the STX17-positive buds of autolysosomes (Fig. 3c,d). Time-lapse imaging revealed that SNX4 and SNX5 are recruited to a STX17-concentrated region on the autolysosome membrane gradually and co-localized with STX17 on the concentrated buds (Figs. 2e and 3e). At the conclusion of budding, STX17 disassociated from autolysosomes in small vesicles together with SNX4 and SNX5 (Figs. 2e and 3e). After the first budding event, some SNX4 signals remained on the budding sites of the autolysosomal membrane, and the second budding event occurred at the same site with the remaining SNX4 ( Supplementary  Fig. 4a). Sometimes, after the first budding, it appeared that SNX5 was again recruited to autolysosomes, where it began to sort STX17 for a second time ( Supplementary Fig. 4b).
SNX4 and SNX5 interacted with STX17 in vivo and in vitro (Figs. 2f-h and 3f-h). Further, SNX4 and SNX5 interacted with STX17 primarily through binding the C-terminal tail of the transmembrane domain (CT) and SNARE domain, while STX17 interacted with SNX4 and SNX5 through binding to the BAR domain (Figs. 2i-k and 3i-k and Supplementary Fig. 5a,b). These results suggest that SNX4 and SNX5 are essential for direct sorting of STX17 on autolysosomes.
Next, we identified three STX17 mutants that have disrupted binding with SNX4/SNX5, which inhibited the extent of STX17 recycling from autolysosomes (Supplementary Fig. 5a-c). The STX17 mutant variant (276-280 5A) significantly reduced the extent of localization of SNX4 and SNX5 to autolysosomes (Supplementary Fig. 5d-f). These results support that STX17's interactions with SNX4 and SNX5 contribute to the recruitment of SNX4 and SNX5 and represent a required step for STX17 recycling from autolysosomes.

SNX5 and SNX4 function cooperatively in STX17 retrieval.
On the basis of the results above, we hypothesized that SNX4 and SNX5 were functionally connected. In support of our hypothesis, (1) SNX5 exhibited significant co-localization with SNX4 on the buds of autolysosomes ( Fig. 4a and Supplementary Fig. 6a,b), (2) SNX5 interacted with SNX4 both in vivo and in vitro through their BAR domains ( Fig. 4b-e) and (3) SNX5 depletion blocked STX17 recycling from autolysosomes and concurrently led to a dramatic increase in the localization of SNX4 on STX17-positive autolysosomes ( Fig. 4f,g) and vice versa.
The BAR-domain-containing proteins have been characterized for their ability to sense or induce membrane structural curvature in the generation of carrier vesicles [43][44][45] . Here we examined the ability of SNX4 and SNX5 to remodel autolysosomal membrane. Cytosol from starved wild-type and SNX4, SNX5 and SNX17 knockout cells was incubated with trypsin-stripped autolysosomes, and were then observed via scanning electron microscopy. Bud/tubule formation on autolysosomes was significantly inhibited for cytosol from SNX4 Fig. 3 | SNX5 is required for STX17 recycling from autolysosomes. a, U2OS cells stably expressing Flag-STX17 were transfected with NC or siRNA against SNX5. Forty-eight hours after transfection, cells were starved with EBSS for the indicated duration. Scale bar, 5 μm. inset scale bar, 1 μm. b, Quantification of STX17-positive autolysosome number. images in a and Extended Data Fig. 9d were analysed. Data are mean ± s.d. (n = 3, 50 cells from three independent experiments were quantified), unpaired two-tailed t test. c, U2OS cells stably expressing GFP-STX17-TM, mKATE2-SNX5 and LAMP1-CFP were starved with EBSS for 2 h, and live images were taken. Scale bar, 5 μm. inset scale bar, 1 μm. d, U2OS cells stably expressing Flag-STX17 with mKATE2-SNX5 were starved with EBSS for 2 h and stained with antibodies against Flag, LC3 and LAMP1. Scale bar, 5 μm. inset scale bar, 2 μm. e, U2OS cells stably expressing GFP-STX17-TM, mKATE2-SNX5 and LAMP1-CFP were starved with EBSS for 2 h, and videos were taken. Scale bar, 1 μm. The arrows indicate the STX17 budding site. f, HEK293T cells were transfected with empty vector or Flag-STX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with indicated antibodies. g, Co-iP analysis of interactions between Flag-SNX5 with HA-STX17 in HEK293T cells. h, Glutathione sepharose beads bound with GST-STX17 were incubated with purified His-SNX5 for 16 h, and then eluted for immunoblotting. i, HEK293T cells were transfected with empty vector or HA-SNX5 with or without truncated variants of STX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with the indicated antibodies. j, HEK293T cells were transfected with empty vector or HA-STX17 with or without truncated variants of SNX5. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. k, HEK293T cells were transfected with empty vector or HA-SNX5 BAR with or without truncated variants of STX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with the indicated antibodies. Source numerical data and unprocessed blots are available in source data. Twenty-four hours after transfection, cells were starved with EBSS for 2 h, then stained with antibodies against LC3 and LAMP1. Scale bar, 5 μm. inset scale bar, 1 μm. Arrows indicate colocalization. b, Co-iP analysis of interactions between Flag-SNX5 with HA-SNX4 in HEK293T cells. c, Glutathione sepharose beads bound with GST-SNX4 were incubated with or without purified His-SNX5 for 16 h, and then eluted for immunoblotting. d, Co-iP analysis of interactions between Flag-SNX4, Flag-SNX4-BAR or Flag-SNX4-PX with HA-SNX5 in HEK293T cells. e, Co-iP analysis of interactions between Flag-SNX5, Flag-SNX5-BAR or Flag-SNX5-PX with HA-SNX4 in HEK293T cells. f, U2OS cells stably expressing GFP-STX17-TM, LAMP1-CFP and mKATE2-SNX5 or mKATE2-SNX4 were transfected with NC or siRNA against SNX4 or SNX5. Forty-eight hours after transfection, cells are starved with EBSS for 2 h. Arrows indicate the autolysosomes with SNX4 or SNX5. Scale bar, 5 μm. g, Quantification of SNX4 or SNX5 localization to autolysosomes in f. Data are mean ± s.d. (n = 3, 50 cells from three independent experiments were quantified), unpaired two-tailed t test. h, Purified autolysosomes from MEF cells starved for 2 h were treated with trypsin, incubated with the indicated cytosol from MEF cells starved for 2 h, washed and then observed under scanning electron microscopy. i, Samples from h were quantified to determine the percentage of autolysosome with buds or tubules. Data are mean ± s.d. (n = 50 autolysosomes from at least three independent experiments), unpaired two-tailed t test. Source numerical data and unprocessed blots are available in source data.    and SNX5 knockout cells, but not for cytosol from wild-type and SNX17 knockout cells (Fig. 4h, i). These data suggest that SNX4 and SNX5 functionally connect in STX17 recycling from autolysosomes, and they are not interchangeable in this process.
Further, we identified a mutant (SNX4 Y252A) that completely disrupted the interaction of SNX4 with SNX5 without effect on its self-interaction and its interaction with KIBRA ( Supplementary  Fig. 6c,d). STX17 retrieval was significantly blocked in SNX4 knockout cells stably expressing the Y252A mutant, but was rescued by wild-type SNX4 (Supplementary Figs. 6e,f and 7a,b). These data suggest that the cooperation between SNX4 and SNX5 is required for STX17 retrieval from autolysosomes.

SNX17 coordinates STX17-dynein-dynactin interactions.
The SNX-FERM family protein SNX17 was also implicated in STX17 recycling from autolysosomes by our initial experiment (Fig. 5a, Extended Data Figs. 6 and 7 and Supplementary Fig. 8a). Live cell imaging showed some SNX17 signals residing on lysosomes, and revealed translocation of SNX17 to autolysosomes via autophagosome-lysosome fusion and via autolysosome-lysosome fusion ( Supplementary Fig. 8b,c). Depletion of SNX17 also caused severe defects in STX17 recycling from autolysosomes (Fig. 5b,c and Supplementary Fig. 9a). These results were further confirmed using a second siRNA and in rescue experiments (Extended Data Fig. 10 and Supplementary Fig. 3).
SNX17 has also been shown to be involved in plasma membrane protein recycling together with the retriever complex [46][47][48] . However, knockdown of VPS26C and VPS35L, two retriever complex subunits, had no obvious effect on STX17 recycling from autolysosome ( Supplementary Fig. 9b-d and Fig. 5c). SNX17 interacted with STX17 directly (Fig. 5d-f). The F3 region in the FERM domain of SNX17 interacts with the CT region and SNARE domain of STX17 ( Supplementary Fig. 9e-h). These data suggest that SNX17 functions directly in STX17 recycling from autolysosomes independently of the retriever.
Given that SNX17 has been shown as a cargo adaptor and the dynein-dynactin complex recognizes cargoes through interactions between its subunits with adaptor proteins 48-57 , we reasoned that SNX17 might work as an adaptor cooperating with dynein-dynactin in STX17 recycling. SNX17 interacted with DCTN1, through its F3 region in the FERM domain ( Fig. 5g and Supplementary  Fig. 10a), and with dynein heavy chain (DHC1) (Fig. 5h). SNX4, SNX5 and STX17 also interacted with DCTN1 ( Supplementary Fig.  10b-d). Interactions were also observed between DHC1 and SNX4, SNX5 and STX17 (Fig. 5h). Silencing DCTN1, but not by any depletion of the endosomes or lysosome adaptor proteins HPS6, SKIP and KIBRA, blocked STX17 recycling from autolysosomes (Figs. 2b and 5i and Supplementary Figs. 10e-h and 11a,d).
Depletion of the heavy chain (DHC1H1) and dynein light chain LC8-type 1 (DYNLL1) of the dynein and the dynein inhibitor dynarrestin also blocked STX17 recycling ( Fig. 5i and Supplementary  Fig. 11b-f). Depletion of SNX17 reduced the interaction of DCTN1 with SNX4, SNX5 and STX17 ( Supplementary Fig. 12ac). Consistent with these observations, suppression of SNX17 and DCTN1 significantly increased the localization of SNX4 and SNX5 on autolysosomes ( Supplementary Fig. 12d-f). These results suggest that SNX17 plays a direct role in STX17 recycling by mediating the interactions between the STX17-SNX4-SNX5 and dyneindynactin complexes.
SNX4-SNX5-SNX17 forms the recycler complex. Since SNX4, SNX5 and SNX17 are essential for successful STX17 recycling from autolysosomes, all three co-localize on the same membrane (Fig. 6a), and all interact with STX17 (Figs. 2, 3 and 5). No additional effect on STX17 recycling from autolysosomes defects was observed when doubly or triply knocking down these three genes compared with single gene knockdown ( Supplementary  Fig. 13a,b), indicating their shared function as a complex in STX17 recycling from autolysosomes. Immunoprecipitation results show that each of these proteins consistently pulled down both of the other two (Fig. 6b). Moreover, tandem affinity immunoprecipitation (IP) using SNX5, then SNX4, as successive baits, clearly demonstrated that SNX4, SNX5 and SNX17 do indeed form a stable complex (Fig. 6c), and these interactions are confirmed by endogenous IP (Fig. 6d-f).
Overexpression of each individual gene resulted in substantially enhanced interaction between the other two proteins (for example, SNX4 overexpression promoted SNX5-SNX17 interaction, reciprocally true for overexpression of each gene) ( Supplementary  Fig. 13c,d), and specifically, disruption of the interaction between Flag-STX17 with LAMP1-mCherry were starved with EBSS for 2 h and stained with antibodies against Flag, SNX17 and LC3. Scale bar, 5 μm. inset scale bar, 1 μm. b, U2OS cells stably expressing Flag-STX17 were transfected with NC or siRNA against SNX17. Forty-eight hours after transfection, cells were starved with EBSS for the indicated duration. Scale bar, 5 μm. inset scale bar, 2 μm. c, Quantification of STX17-positive autolysosome number in b and Supplementary Fig. 9d. Data are mean ± s.d. (n = 3, 50 cells from three independent experiments were quantified), unpaired two-tailed t test. d, Co-iP analysis of interactions between Flag-STX17 with SNX17 in HEK293T cells. e, Co-iP analysis of interactions between Flag-SNX17 with HA-STX17 in HEK293T cells. f, Glutathione sepharose beads bound with GST or GST-STX17 were incubated with purified His-SNX17 for 16 h, and then eluted for immunoblotting. g, Co-iP analysis of interactions between Flag-DCTN1 with HA-SNX17 in HEK293T cells. h, Co-iP analysis of interactions between Flag-SNX17, Flag-SNX4, Flag-SNX5 and Flag-STX17 with DHC1 in HEK293T cells. i, Quantification of STX17-positive autolysosome number in Supplementary Figs. 10e,g and 11d. Data are mean ± s.d. (n = 3, 50 cells from three independent experiments were quantified), unpaired two-tailed t test. Source numerical data and unprocessed blots are available in source data. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. immunoblotting was performed with the indicated antibodies. c, HEK293T cells were transfected with HA-SNX4, Flag-SNX5 and Myc-SNX17. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. The elution from the first immunoprecipitation was subjected to a second round of immunoprecipitation with anti-HA antibody. immunoblotting was then performed with the indicated antibodies. d-f, interactions among endogenous SNX4, SNX5, SNX17 and STX17 in HEK293T cells; co-immunoprecipitation assay using SNX4, SNX5 and SNX17 as bait proteins and igG as the negative control. The 10% input shows results obtained from cell extracts without immunoprecipitation. g, HEK293T cells were transfected with the indicated plasmids. Twenty-four hours after transfection, cells were lysed and immunoprecipitated with anti-Flag antibody. h,i, Purified proteins were incubated together as indicated with GST beads. After 16 h of incubation, the elution was subjected to immunoblotting. Unprocessed blots are available in source data. SNX4 and SNX5 by SNX4 Y252A also dramatically decreased this complex formation (Fig. 6g). In vitro binding assays also demonstrated that each of these proteins interacts with the other two individually and in complex (Fig. 6h,i). SNX17 interacted with the BAR domain of SNX4 and SNX5 through its F1 and F3 regions, respectively ( Supplementary Fig. 13e-j). These results illustrate the mechanism by which these proteins form the SNX4/5/17 recycler complex to successfully conduct highly selective recycling of STX17 from autolysosomes.
ATG9A is retrieved from autolysosomes by recycler. ATG9A is a transmembrane autophagy-related gene (ATG) protein in mammals 19 . We found that ATG9A localization to autophagosomes and translocation to autolysosomes is autophagy dependent (Fig. 7a,b and Supplementary Figs. 14 and 15). Time-lapse imaging demonstrated the retrieval of ATG9A from autolysosomes (Fig. 7c), and revealed that this retrieval still occurred in STX17 knockout cells ( Supplementary Fig. 14e). These results suggest that ATG9A is recycled from autolysosomes independently of STX17.
More ATG9A accumulated on autolysosomes in SNX4, SNX5 and SNX17 depletion cells than in wild-type cells (Fig. 7d,e). SNX4, SNX5 and SNX17 co-localized with ATG9A on autolysosomes (Fig. 7f), but not with the isolation membrane markers WIPI2 and ATG16 ( Supplementary Fig. 16). ATG9A interacted with SNX4, SNX5 and SNX17 ( Supplementary Fig. 17a), and these interactions were further confirmed by endogenous IP (Supplementary Fig.  17b-d). ATG9A interacted with SNX4 and SNX5 through binding the BAR domain, and with SNX17 through binding the F3 domain ( Supplementary Fig. 17e-g). These results suggest that ATG9A is retrieved from autolysosomes by the recycler, but is not retrieved from isolation membranes.

Recycler is involved in the regulation of autophagy.
To investigate the function of the recycler in autophagy, autophagic flux was examined. The turnover of both LC3-II and p62 was inhibited in SNX4-, SNX5-and SNX17-deficient cells (Fig. 8a-c and Supplementary Fig. 18). The inhibition of autophagic flux in SNX4-deficient cells was rescued by wild-type SNX4, but not by SNX4 Y252A (Fig. 8d,e).
Autophagic flux was also examined using flow-cytometry-based RFP-GFP-LC3 and RFP-GFP-p62 reporter assays. Consistent with a previous report 58 , the GFP/RFP intensity ratio of LC3 significantly reduced after 4 h EBSS starvation. However, this reduction of GFP/RFP intensity ratio was suppressed in SNX4, SNX5 and SNX17 knockout cells ( Fig. 8f and Supplementary  Fig. 19a). Further, the suppressed reduction in SNX4 knockout cells was rescued by wild-type SNX4, but not SNX4 Y252A. Similar results were observed in RFP-GFP-p62 reporter assays ( Fig. 8g and Supplementary Fig. 19b). The pH and cathepsin D processing are almost unchanged in SNX4, SNX5 and SNX17 depleted cells and SNX4 depletion cells complemented with SNX4 Y252A (Supplementary Fig. 20). These data suggest that the recycler specifically regulates autophagy without apparent effects on lysosomal degradation activity.

Discussion
Autophagosomes subsume large amounts of membrane from varied origins within a cell and fuse with lysosomes, thus leading to the highly complex autolysosomes. The lysosomal membrane components on autolysosomes are recycled through ALR, but the fate of autophagosomal membranes and membrane proteins from different cellular origins on autolysosomal surface remains elusive. Here we describe the recycler complex, composed of SNX4, SNX5 and SNX17, which selectively sorts autophagosomal components (for example, STX17 and ATG9A) out of autolysosomes, and thereby prevents its accumulation on the autolysosomal surface, independently of the retromer and retriever complexes. We thus termed this process as autophagosomal components recycling (ACR). Further, loss of ACR function results in the inhibition of autophagy. The genetic basis of ACR differs from that of ALR, and ACR occurs immediately following autophagosome-lysosome fusion before ALR. So far, autophagy has been divided into four parts in time sequence: autophagosome formation, autophagosome-lysosome fusion, autophagosomal substrates degradation and ALR 3 . The discovery of ACR adds an essential part to the autophagy process.
Yeast ATG9 is incorporated into autophagosome outer membranes 59 . Mammalian ATG9A vesicles partially localize with LC3 and form dynamic contacts with autophagosomes 60 , and are not significantly incorporated into complete autophagosomes 61 . Here we detected endogenous ATG9A on autophagosomes. The discrepancy between our discovery and previous studies may relate to the dim ATG9A signal on autophagosomes. However, the dim ATG9A signal on autophagosomes was boosted upon blocking autophagosome-lysosome fusion. Live imaging results also revealed the localization of ATG9A on autophagosomes. ATG9A co-fractionated with STX17 and LC3 in autophagosomal fractions. All these results support the presence of ATG9A on autophagosomes.
Thus far, we found that STX17 and ATG9A were recycled from autolysosomes. It is possible, if not likely, that other cargoes on autolysosomes are also recycled, given that multiple sources contribute to the autolysosomal membrane. Ongoing research will necessarily focus on the identification of the suite of cargoes recycled from autolysosomal membranes to clearly determine the fate of autolysosomes. In addition, the final destination of recycling vesicles remains unclear due to technical difficulties of measurement, such as fast movement and quick running out of any focus plane, as has been observed with released STX17-carrying vesicles.
It is of particular interest that autolysosomal alkalization by bafilomycin A1 results in STX17 entry into autolysosomes here, suggesting that an ACR defect may occur. However, a previous study reported that STX17 disassociated normally after bafilomycin A1 treatment 32 . These different results obtained after bafilomycin A1 treatment may be caused by different treatment starting times and durations. It is known that bafilomycin A1 has complicated effects on autophagy due to the timepoints and time duration of bafilomycin A1 addition 62 . So far, we do not know how the STX17 enters lysosomes, but it is possible the rest of STX17 that is not recycled remains undegraded inside autolysosomal lumen. Further experiments are necessary to clarify the mechanisms that drive this phenomenon.  SNX proteins are involved in the sorting of diverse types and quantities of cargo, acting in various combinations and forming different complexes 35,38,63,64 . Here we identified an uncharacterized cooperation between SNX4 and SNX5 in which they form a heterodimer and recognize cargoes via their BAR domain in a similar manner to that of direct cargo recognition for ESCPE (endosomal SNX-BAR sorting complex for promoting exit-1)-related SNXs [65][66][67][68] . The region between SNX4 and SNX5 at the dimer interface is unlikely to serve in cargo recognition, since this domain is necessary for interaction between the SNX proteins and inaccessible for cargo recognition. We speculate that the interface between different SNX4-SNX5 dimers may instead be used for recognition since mapping analysis implied its involvement in the interaction between cargoes and SNX4 or SNX5. However, both SNX4 and SNX5 interact with both the CT region and the SNARE domain of STX17, and the transmembrane domain of STX17 is sufficient for its efficient retrieval from autolysosomes, suggesting that the SNARE domain is not essential. Further structural biology studies will provide more insight into the mechanisms underlying cargo recognition.
Previous studies have shown that SNX1/SNX2 and SNX5/SNX6 constitute the respective mammalian orthologues of yeast Vps5p and Vps17, therefore leading to redundancy and interchangeability between SNX1 and SNX2 40,63,69 . However, here SNX4 and SNX5 are both required for ACR and not interchangeable, suggesting a discrepancy between retromer and recycler in organization.
Previous studies of SNX4, SNX5 and SNX17 have mainly focused on their functions on endosomes 34,36,64,70 . Recently, two studies began to study the functions of sorting nexins on lysosomes/vacuoles. ATG24, the homologue of SNX4 in yeast, retrieves ATG27 on vacuoles together with SNX41 and SNX42 71 . SNX4 is also required for ATG9A recycling from endolysosomes to early endosomes in mammals 72 . Both of these studies revealed a sole ATG retrieval from vacuoles or lysosomes. Here we proposed that autophagosomal membrane components that are delivered to lysosomes are recycled from autolysosomes. These components not only include ATG proteins (for example, ATG9A), but also include the non-ATG proteins/components on autophagosomes (for example, STX17).
However, these studies reported some difference on the phenotype of autophagy 36,72 . Specifically, it has been reported that autophagic flux is abnormal with an increased ratio of LC3-II/LC3-I level in SNX4-deficient cells 72 . In another study, SNX4 depletion decreased the conversion of LC3-I to LC3-II 36 . Here, we found the LC3-II level increased without observable effects on autophagosome formation, but the autophagosome-lysosome fusion is inhibited. These differences may be caused by different cell lines and different treatments. Therefore, the specific regulation in different cells under different conditions may lead to different observations. It has been reported that STX17 functions in both autophagosome formation and autophagosome-lysosome fusion, while ATG9A functions in autophagosome formation 19,30,60,[73][74][75] . However, our results showed that blocking the recycling of STX17 and ATG9A leads to the inhibition of autophagic flux without autophagosome formation defect, probably because the STX17 involved in autophagosome formation does not participate in autophagosome-lysosome fusion and ACR. In addition, it is known that ATG9A traffics between the Golgi complex, endosomes and autophagosomes 76 . The recycling defect of ATG9A from autolysosomes caused by recycler depletion has no effect on autophagosome formation. This may reflect compensation by other ATG9A molecules in the cell, for example, those in the Golgi complex and/or endosomes.
Although STX17 contributes to SNX4 and SNX5 recruitment, phosphatidylinositols (PtdIns) may be also required for their recruitment. It is supported by previous studies that the class III phosphatidylinositol 3-kinase VPS34 inhibitor leads to the loss of SNX4 from endomembrane 72 , and phosphatidylinositol 3-kinase inhibitor wortmannin also results in exclusively cytosolic fluorescence of SNX5 77 . Autolysosome is a hybrid organelle, with multiple PtdIns that are received from late endosomes/lysosomes (with PI3P and PI(3,5)P2), from early endosomes (PI3P) and from autophagosomes (with PI3P, PI4P and PI5P), and on autolysosomes PI4P is transformed into PI(4,5)P2 3,28 . SNX4 binds PI3P 78 . SNX5 binds both PI (3,4)P2 and PI3P at a similar level 77 . SNX17 binds multiple PtdIns in the following ranked order (binding activity, strongest first): PI3P > PI5P > PI4P > PI(3,5)P2 > PI(4,5)P2 46 . Given the multiple PtdIns on autolysosomes, and considering the binding specificities of SNX4, SNX5 and SNX17 to the various PtdIns that are present on autolysosomes, we can ascertain a plausible explanation for the appearance of SNX4, SNX5 and SNX17 on autolysosomes. Specifically, once these three SNXs are recruited to autolysosomes, they have many chances to form a complex to carry out ACR.
Recycling and turnover of membranes and membrane proteins is crucial for maintaining organelle identity and normal function. In this study, we identified a multiprotein recycler complex (SNX4-SNX5-SNX17) that performs a recycling process for autophagosomal components on autolysosomes, independently of other retrieval processes. Further study is warranted to identify other machineries and regulators as part of the comprehensive understanding of the fate determination of autolysosomes, as well as maintenance of the identity and function of autophagosomes and lysosomes. In addition, given that SNX4 and SNX5 have been implicated in many human diseases, closer scrutiny of ACR and the recycler may reveal if their dysfunction is involved in SNX-related diseases.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41556-022-00861-8.

Fig. 8 | inhibited autophagic flux in recycler-deficient cells is rescued by SNX4, but not SNX4 Y252a, cells. a-c,
Autophagic flux is inhibited in recycler-deficient cells. Wild-type, SNX4 KD, SNX5 KD or SNX17 KD U2OS cells were starved with EBSS with or without 100 nM bafilomycin A1 (BFA) for indicated duration, and immunoblotting was performed with the indicated antibodies. The intensity of LC3-ii and p62 bands was normalized to actin. Data are mean ± s.e.m. of three independent experiments. d, Wild-type or SNX4-KO MEF cells stably expressing HA, HA-SNX4 or HA-SNX4 Y252A were starved with EBSS for indicated duration and immunoblotted with indicated antibodies. The intensity of LC3-ii and p62 bands was normalized to actin. Data are mean ± s.e.m. of three independent experiments. e, HEK293T wild-type or SNX4 KD cells were transfected with Flag, Flag-SNX4 or Flag-SNX4 Y252A. Twenty-four hours after transfection, cells were lysed and immunoblotted with indicated antibodies. The intensity of LC3-ii and p62 bands was normalized to actin. Data are mean ± s.e.m. of three independent experiments. f,g, Quantification values for the GFP/RFP intensity ratio by flow cytometry. Data are mean ± s.e.m. of three independent experiments. Unpaired two-tailed t test. Source numerical data and unprocessed blots are available in source data.
Cell culture and transfection. HEK293T, U2OS and HeLa cells were generous gifts from Dr. Qing Zhong (Shanghai Jiao Tong University, Shanghai, China). HEK293T, U2OS and HeLa cells were cultured in DMEM (Hyclone) supplemented with 10% FBS (Gibco) and 1% penicillin-streptomycin solution (Beyotime) at 37 °C with 5% CO 2 . For starvation treatment, cells were washed three times with PBS (Hyclone) and then incubated with EBSS for the indicated duration. HeLa cells were transiently transfected using Lipofectamine 3000 (Invitrogen) according to the manufacturer's instructions. Transient transfection of plasmid in HEK293T cells was performed using polyethylenimine according to the manufacturer's protocol. Cells were analysed 24 h after transfection. For RNA interference, siRNA duplexes were transfected into cells using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions. Forty-eight hours after transfection, cells were collected for analysis.
Immunostaining assays. Cells grown on coverslips were washed with PBS and fixed in 4% paraformaldehyde in PBS for 15 min at room temperature. After washing three times with PBS, cells were permeabilized with 0.1% saponin in PBS for 10 min, then blocked with 10% goat serum in PBS for 1 h. Then the cells were incubated with primary antibodies for 1 h or overnight at 4 °C. After washing three times with PBS, cells were incubated with appropriate secondary antibodies at room temperature for 1 h, and washed three times with PBS. The slides were mounted and images were acquired under a laser scanning confocal microscope (FV3000, Olympus).
Western blotting. Cells were lysed by SDS buffer. Samples extracted from cells were subjected to SDS-PAGE electrophoresis and immobilized on a PVDF membrane (BIO-RAD, 162-0177). After blocking non-specific proteins on PVDF membrane with 5% skimmed milk for 1 h at room temperature, target proteins were probed with specific primary antibodies for 12 h at 4 °C and then incubated with secondary antibody conjugated with horseradish peroxidase for 1 h at room temperature. Finally, visualization was carried out using enhanced chemiluminescence (Beyotime, P0018M-2) according to the manufacturer's protocol.
In vitro binding assay. Genes were cloned into pGEX-4T-1 or pET-28a vector for expression in E. coli BL21 (DE3). The recombinant proteins were purified by glutathione sepharose resin or Ni-affinity resin. In glutathione-S-transferase (GST) pull-down assays, GST and GST-tagged proteins were applied to GST resin, then incubated with 1 μg His-tagged proteins in binding buffer (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA and 0.5% NP-40) supplemented with protease inhibitor cocktail (Roche) for 2 h at 4 °C. After three washes, proteins were eluted and dissolved in sample buffer for SDS-PAGE and immunoblotting.
Live cell imaging. Cells were placed on a four-chambered cover glass (In Vitro Scientific, syD35-20-1-N) one day before observation. During live imaging, the culture dish was mounted on an inverted microscope (Olympus, FV3000) to maintain incubation conditions at 37 °C and 5% CO 2 using a Plan Apochromat N 100×/1.70 oil. Images or videos were recorded using confocal laser microscope system, then further processed and analysed using ImageJ1.52a.
Cytosol preparation. The cells were cultured to confluence and starved in EBSS for 2 h. Then the cells were collected by scraping and centrifuging at 600g for 5 min, washed with PBS followed by another 600g spin for 5 min and homogenized by passing through a 22 G needle in a 1× cell pellet volume of extraction buffer (50 mM HEPES-KOH pH 7.4, 500 mM sucrose, 2 mM PMSF, 4 mM EDTA and 4 mM EGTA) plus cocktail protease inhibitors (Roche), phosphatase inhibitors (Roche) and 0.3 mM DTT. The cell homogenates were centrifuged at 160,000g for 30 min, supernatant fractions were collected and the centrifugation was repeated three times to achieve a clarified fraction.
Field emission in-lens scanning electron microscopy. For field emission in-lens scanning electron microscopy, lysosome isolation was performed with a lysosome isolation kit (Sigma-Aldrich) according to the manufacturer's manual. Cytosol was incubated with trypsin-stripped autolysosomes on glass chips at 37    Extended Data Fig. 4 | aLR genes are not required for STX17 recycling from autolysosomes. a, ALR genes depletion has no effect on STX17 recycling from autolysosomes. U2OS cells stably expressing Flag-STX17 were transfected with indicated siRNAs. Forty-eight hours after transfection, cells were starved with EBSS for indicated hours and stained with antibodies against Flag and LAMP1. Scale bar, 5 μm. inset scale bar, 2 μm. b, Quantification of STX17 positive autolysosomes in a. Data are means ± s.e.m. (n = 3, 50 cells from 3 independent experiments were quantified). Unpaired two-tailed t-test. Source numerical data are available in source data.   Extended Data Fig. 8 | KiBRa, SNX7, and SNX30 are not required for STX17 recycling from autolysosomes. a-c, U2OS cells stably expressing Flag-STX17 were transfected with indicated siRNAs. Forty-eight hours after transfection, cells were subjected to immunoblot with antibodies against SNX4, KiBRA and SNX30. d, Representative mRNA level for the knockdown efficiency of SNX7. Data are presented as mean values ± s.d. e, The depletion of KiBRA, SNX7, and SNX30 has no effect on STX17 recycling from autolysosomes. U2OS cells stably expressing Flag-STX17 were transfected with non-targeting siRNA (NC) or siRNAs against KIBRA, SNX7, and SNX30. Forty-eight hours after transfection, cells were starved with EBSS for the indicated hours. Scale bar, 5 μm. inset scale bar, 2 μm. Source numerical data and unprocessed blots are available in source data. Extended Data Fig. 9 | Retromer is not required for STX17 recycling from autolysosomes. a-c, U2OS cells stably expressing Flag-STX17 were transfected with indicated siRNAs. Forty-eight hours after transfection, cells were subjected to immunoblot with antibodies against SNX5, VPS35 and SNX6. d, Depletion of VPS35 and SNX6 has no effect on STX17 recycling from autolysosomes. U2OS cells stably expressing Flag-STX17 were transfected with non-targeting siRNA (NC) or siRNAs against VPS35 and SNX6. Forty-eight hours after transfection, cells were starved with EBSS for the indicated hours. Scale bar, 5 μm. inset scale bar, 2 μm. Source unprocessed blots are available in source data.