Lysosomes with severe LMP have been reported to be selectively degraded by macroautophagy, known as lysophagy4,5, whereas milder LMP can be fixed through faster, direct membrane repair by the endosomal sorting complex required for transport6,7,8 (ESCRT). However, depletion of ESCRT subunits causes only partial defects in rapid lysosomal repair7, suggesting the existence of additional repair mechanisms. To search for such pathways, we designed an unbiased proteomic approach to identify proteins that are recruited rapidly to damaged lysosomes. Targeting TurboID, a promiscuous biotin ligase9,10, to the lysosomal surface (Extended Data Fig. 1a,b) enabled rapid biotinylation of lysosomal surface proteins that can be identified by streptavidin (Extended Data Fig. 1c). l-Leucyl-l-leucine methyl ester (LLOME) is the best characterized lysosomotropic reagent that polymerizes inside lysosomes to induce fast but reversible lysosomal membrane damage5,11. We purified all biotinylated, membrane-associated proteins from 293T cells expressing lysosome-targeted TurboID (Lyso-TurboID) with or without LLOME treatment for mass spectrometry analysis (Fig. 1a and Extended Data Fig. 1d,e). The autophagy adaptor protein p62 (also known as SQSTM1), previously reported to accumulate on damaged lysosomes5, was highly enriched in the purified samples from LLOME-treated cells (Fig. 1a), confirming the utility of our approach.

Fig. 1: An unbiased proteomic screen identifies PI4K2A-mediated PtdIns4P signalling in rapid lysosomal repair.
figure 1

a, Schematic illustration of the Lyso-TurboID screen. P20-SA↓, streptavidin (SA) pulldown from P20, the pellet fraction after centrifugation at 20,000 g. b, Scatter plot of the top proteins from more than 1,800 mass spectrometry hits. Three categories of proteins are labelled in colour. ORP9/11 are known PtdIns4P effectors. c, Left, OSBP-PH–GFP is recruited to damaged lysosomes in U2OS cells; the outlined region is magnified on the right. Right, a schematic illustration of the response. d, Pearson’s correlation coefficient (R) for OSBP-PH–GFP and IST1. Data are mean ± s.e.m.; n = 15 (−LLOME), n = 27 (+LLOME) cells from 3 trials. e, Endogenous PI4K2A accumulates rapidly on damaged lysosomes in U2OS cells; the outlined region is magnified on the right. f, Pearson’s correlation coefficient for PI4K2A and IST1. Data are mean ± s.e.m.; n = 27 cells from 3 trials for each condition. g, Quantification of OSBP-PH–GFP puncta in PI4K2A-KO U2OS cells expressing the indicated PI4K2A mutants. Approximately 50–100 random cells were counted per condition. Data are mean ± s.e.m.; n = 3 per condition. See fluorescence images in Extended Data Fig. 4c. WT, wild type. h, Lyso-pHluorin quantification in U2OS cells with indicated genetic modifications expressing the indicated PI4K2A mutants. More than 100 cells were counted for each condition. Data are mean ± s.e.m.; n = 3 per condition. Dotted line indicates 50% of repair. See fluorescence images in Extended Data Fig. 4d,e. i, Quantification of galectin-3 intensity above threshold in U2OS cells with indicated genetic modifications expressing the indicated PI4K2A mutants. Between 50 and 100 random cells were quantified for each condition. Data are mean ± s.e.m.; n = 3 per condition. See fluorescence images in Extended Data Fig. 4f. j, Percentage of cells exhibiting tau spreading after tau fibril exposure. Data are mean ± s.e.m.; n = 4 per condition. See fluorescence images in Extended Data Fig. 4h. All experiments were based on stable cell lines unless otherwise indicated. Statistical significance was determined by unpaired, two-tailed t-tests. Scale bars, 10 μm.

Source data

PtdIns4P accumulates on damaged lysosomes

Besides p62 and multiple ESCRT subunits previously reported as enriched on damaged lysosomes5,6,7,8, mass spectrometry identified three top hits likely within the same phosphoinositide signalling pathway (Fig. 1b, Supplementary Table 1). These included PI4K2A, an enzyme that generates the lipid messenger phosphatidylinositol-4-phosphate12 (PtdIns4P), as well as ORP9 (also known as OSBPL9) and ORP11 (also known as OSBPL11), both of which are PtdIns4P-binding proteins13,14,15. We confirmed by immunoblotting that these proteins were enriched on damaged lysosomes (Extended Data Fig. 1f). To further test whether LMP stimulates lysosomal PtdIns4P signalling, we established a stable cell line expressing the pleckstrin homology (PH) domain of OSBP fused to GFP (OSBP-PH–GFP), a well-established probe for PtdIns4P16. Consistent with previous observations16, under basal conditions, OSBP-PH–GFP puncta were found exclusively in the Golgi region (Fig. 1c, top). By contrast, LLOME treatment induced many new OSBP-PH–GFP puncta that colocalized with increased sodium tolerance 1 homologue (IST1), one of the ESCRT subunits used to mark damaged lysosomes (Fig. 1c, bottom and Extended Data Fig. 1g), and with multiple lysosomal markers including LAMP1, LAMP2 and CD63 (Extended Data Figs. 1h and 2a,b and Supplementary Video 1). This phenomenon was recapitulated across multiple cell lines, observed using different PtdIns4P probes, was specific for PtdIns4P but not for other phosphoinositide species, and was not dependent on the pool of PtdIns4P on the trans-Golgi network (TGN) (Extended Data Fig. 2c–i). Thus, specific accumulation of PtdIns4P is triggered on damaged lysosomes.

PI4K2A generates PtdIns4P on damaged lysosomes

PtdIns4P can be generated by direct phosphorylation of phosphatidylinositol by PI4K or dephosphorylation of PtdIns(4,5)P2. Since PtdIns(4,5)P2 was barely detectable on damaged lysosomes (Extended Data Fig. 2e), PI4Ks were probably responsible for lysosomal PtdIns4P production. Using verified PI4K antibodies, we observed selective enrichment of PI4K2A on damaged lysosomes by immunoblotting (Extended Data Fig. 3a,b). This was supported by immunofluorescence showing robust lysosomal accumulation of PI4K2A but not PI4K2B upon LLOME treatment (Fig. 1e,f and Extended Data Fig. 3c–f), with no change in total PI4K2A protein levels (Extended Data Fig. 3g). Of note, the recruitment of PI4K2A and the ESCRT complex were independent of each other (Extended Data Fig. 3h).

Other LMP-inducing strategies including expression of the SARS-CoV-2 protein ORF3A17 or knockout of CLN3, a gene frequently mutated in the lysosomal storage condition known as Batten’s disease18, also stimulated robust lysosomal recruitment of PI4K2A (Extended Data Fig. 3i–m and Supplementary Table 2). Lysosomes are well-known Ca2+ stores; LMP-induced lysosomal Ca2+ release probably contributes to PI4K2A recruitment, as agonists of TRPML1—the main lysosomal Ca2+ channel19—were sufficient to trigger PI4K2A accumulation on lysosomes (Extended Data Fig. 3n). Thus, PI4K2A is specifically recruited to damaged lysosomes.

To investigate whether PI4K2A is responsible for PtdIns4P production on damaged lysosomes, we generated PI4K2A-knockout (PI4K2A-KO) U2OS cells using CRISPR technology20. Without affecting the basal Golgi targeting of OSBP-PH–GFP, PI4K2A-KO completely abolished LMP-induced lysosomal recruitment of the PtdIns4P probe (Extended Data Fig. 4a), which was fully rescued by re-expressing PI4K2A in knockout cells (Extended Data Fig. 4a,b). Pretreating cells with brefeldin A, which disassembles the Golgi complex21, confirmed the robust de novo generation of lysosomal PtdIns4P only in cells with wild-type PI4K2A (Fig. 1g and Extended Data Fig. 4c). Two kinase-deficient point mutations of PI4K2A, K152A and D308A, have been previously characterized2223. Lysosomal OSBP-PH–GFP recruitment in PI4K2A-KO cells was partially rescued by re-expression of PI4K2A(K152A), but not by either PI4K2A(D308A) or the double mutant K152A/D308A (KDAA) (Fig. 1g and Extended Data Fig. 4c). Together, these results demonstrate that PI4K2A is the enzyme that generates PtdIns4P on damaged lysosomes.

PtdIns4P mediates rapid lysosomal repair

We established a lysosomal repair assay based on lyso-pHluorin, a lysosomal pH sensor that is activated in damaged lysosomes owing to pH neutralization24 (Extended Data Fig. 4d, left). Lyso-pHluorin puncta were not detected in either wild-type or PI4K2A-KO cells in resting conditions (Extended Data Fig. 4d, top), and a brief LLOME treatment induced extensive puncta in almost all cells, suggesting equal lysosomal damage (Extended Data Fig. 4d, middle). While lyso-pHluorin puncta disappeared quickly in wild-type cells with an average half-life of 1.1 h, puncta clearance in two independent PI4K2A-KO cell lines was significantly slower, with a calculated half-life of 5.4 to 11 h (Fig. 1h and Extended Data Fig. 4d,e).

Whereas lyso-pHluorin senses lysosomal pH, fluorescently-tagged galectin-3 has been used to detect large, unrepaired lysosomal membrane pores5,7. Stably expressed eGFP–galectin-3 in either wild-type or PI4K2A-KO cells showed no puncta under basal conditions (Extended Data Fig. 4f). Continuous LLOME treatment induced markedly more galectin-3 puncta in PI4K2A-KO cells than in wild-type cells, which was reversed by re-expressing wild-type but not kinase-dead PI4K2A (Fig. 1i and Extended Data Fig. 4f). Together, these data indicate that the kinase activity of PI4K2A is essential for rapid lysosomal repair. Of note, wild-type but not kinase-dead PI4K2B, when artificially anchored on the lysosomal surface, could also rescue the repair defects in PI4K2A-KO cells (Extended Data Fig. 4g).

In exploring the physiological significance of this repair pathway, we noted that loss of PI4K2A caused marked exacerbation of tau fibril spreading in a cell-based assay (Fig. 1j and Extended Data Fig. 4h), consistent with endolysosomal damage as a key step in tau fibril spreading25. Moreover, although CLN3-deleted skin fibroblasts normally accumulate lipofuscin, this lysosomal accumulation was further augmented upon loss of PI4K2A (Extended Data Fig. 4i).

PtdIns4P recruits ORPs to damaged lysosomes

The functions of phosphoinositide signalling pathways are often defined by lipid effectors26. Our mass spectrometry experiments using Lyso-TurboID identified four ORP family members—ORP9, ORP10, ORP11 and OSBP—as potential PtdIns4P effectors in this pathway (Supplementary Table 1). Using verified antibodies (Extended Data Fig. 5a), we found ORP9, ORP10 and ORP11 to be strongly enriched on damaged lysosomes, whereas OSBP was only weakly recruited to lysosomes (Fig. 2a and Extended Data Fig. 5b). Immunofluorescence revealed basal Golgi localization of ORP9 and ORP11, with robust lysosomal accumulation detected 10 min after LLOME treatment (Fig. 2b, Supplementary Videos 2 and 3 and Extended Data Fig. 5c–e). By contrast, ORP3 and ORP5 were not recruited to damaged lysosomes (Extended Data Fig. 5f,g). We confirmed that the LLOME-induced lysosomal recruitment of endogenous ORP9 and ORP11 and stably overexpressed eGFP–ORP9, eGFP–ORP10 and eGFP–ORP11 were all fully dependent on PI4K2A and its kinase activity (Fig. 2c and Extended Data Fig. 5h–m). Mutation of the PtdIns4P-binding residues in their N-terminal PH domains13,14 strongly suppressed the localization of ORP9, ORP10 and ORP11 to damaged lysosomes (Extended Data Fig. 5n–p). Thus, ORP9, ORP10 and ORP11 are recruited to damaged lysosomes in a PI4K2A- and PtdIns4P-dependent manner.

Fig. 2: PtdIns4P drives ORP-tethered ER–lysosomal contacts for rapid membrane repair.
figure 2

a, ORP family members are enriched on damaged lysosomes. Note that ORP10 is absent from S20 (supernatant after centrifugation at 20,000 g) (arrow head). Asterisk indicates a nonspecific band. b, Endogenous ORP9 is recruited to LAMP1-positive lysosomes upon brief LLOME treatment in U2OS cells. c, Punctate intensities of endogenous ORP9 and ORP11 before and after LLOME treatment. More than 30 random cells were quantified for each condition. Data are mean ± s.e.m.; n = 4 trials. See fluorescence images in Extended Data Fig. 5m. d, LLOME induces extensive wrapping of endogenous LAMP1 by eGFP–VAPA in U2OS cells. Right, schematic illustration of the process. e, Percentage of U2OS cells with more than five lysosomes extensively wrapped by eGFP–VAPA. More than 100 cells were counted for each condition. Data are mean ± s.e.m.; n = 3 per condition. See fluorescence images in Extended Data Fig. 6a. f, Representative microscopy and quantification of ER–lysosome wrapping in various ORP-knockout cells. More than 100 cells were counted for each condition. Data are mean ± s.e.m.; n = 3. See more details in Extended Data Fig. 6c–f. g, ORP-QKO cells stably expressing eGFP–VAPA and mCherry–ORP proteins were stimulated with LLOME for 10 min and the percentages of cells showing VAPA clusters on ORP proteins were quantified. Data are mean ± s.e.m.; n = 3 per condition. See more details in Extended Data Fig. 6j–m. ORP9/11 (FYAA/WT), co-expression of mCherry–ORP9(FYAA) and mCherry–ORP11(WT). h, Representative images and quantification of eGFP–galectin-3 puncta in wild-type and ORP-QKO cells. More than 100 cells were quantified per condition. Data are mean ± s.e.m.; n = 3 per condition. i, Quantification of eGFP–galectin-3 puncta in ORP-QKO cells stably expressing eGFP–galectin-3 and mCherry–ORP proteins. Approximately 50–100 cells were quantified for each condition. Data are mean ± s.e.m.; n = 5 per condition. Images are presented in Extended Data Fig. 6o. j, Schematic illustration of PtdIns4P-stimulated, ORP-mediated membrane tethering between the ER and damaged lysosomes. ORP9 carries a stronger FFAT motif than ORP10 and ORP11 and thus provides stronger ER tethering. Unpaired, two-tailed t-tests. NS, not significant. Scale bars, 10 μm.

Source data

ORPs establish ER–lysosome MCSs

The ORP family members often localize to membrane contact sites (MCSs) between the endoplasmic reticulum (ER) and other organelles27. Among the three lysosomal-recruited ORP family proteins, ORP9 carries a strong FFAT motif that binds to the ER transmembrane proteins, vesicle-associated membrane protein-associated proteins A and B13,14 (VAPA and VAPB) (Extended Data Fig. 5n,o). We observed extensive LMP-stimulated VAPA recruitment to lysosomes (Fig. 2d), which was completely dependent on PI4K2A and its kinase activity (Fig. 2e and Extended Data Fig. 6a) as well as the FFAT-binding residues in VAPA (Extended Data Fig. 6b). Extensive redundancy was detected in the recruitment of ORP proteins to damaged lysosomes (Extended Data Fig. 6c–e and Supplementary Table 3). Indeed, the deletion of all four recruited ORP family proteins (ORP9, ORP10, ORP11 and OSBP quadruple knockout (ORP-QKO)) (Supplementary Tables 1 and 4) was required to abolish VAPA clustering around damaged lysosomes (Fig. 2f and Extended Data Fig. 6f). ORP-QKO did not diminish upstream PtdIns4P signalling (Extended Data Fig. 6g,h) and in fact caused stronger and more prolonged lysosomal PtdIns4P production (Extended Data Fig. 6h,i). In agreement with the previously reported heterodimerization of ORP9 and ORP11 (ref. 15), re-expression of two ORPs (ORP9 and ORP10 or ORP9 and ORP11) was sufficient to fully rescue VAPA recruitment to damaged lysosomes in ORP-QKO cells, whereas individual expression of ORP9, ORP10 or ORP11 was insufficient (Fig. 2g and Extended Data Fig. 6j–l). Of note, although ORP9 could not homodimerize, it readily heterodimerized with either ORP10 or ORP11 (Extended Data Fig. 6m). By contrast, OSBP formed homodimers (Extended Data Fig. 6n) and was robustly recruited to damaged lysosomes in the absence of the other ORPs (Extended Data Fig. 6e,k). Similar to PI4K2A-KO cells (Fig. 1h,i), ORP-QKO cells also exhibited marked defects in rapid lysosomal repair (Fig. 2h), which were rescued by reconstitution with any two of ORP9, ORP10 and ORP11, with ORP9 and ORP11 reconstitution being the most robust (Fig. 2i and Extended Data Fig. 6o). Together, these data support the hypothesis that lysosomal PtdIns4P signalling recruits multiple ORP proteins, generating extensive ER–lysosomal MCSs, which are required for rapid lysosomal repair (Fig. 2j).

ORP9, 10 and 11 transport PS to damaged lysosomes

The ORP family are lipid transfer proteins that exchange PtdIns4P for sterols or phosphatidylserine (PS) at MCSs27,28,29. Among the twelve human ORP proteins, ORP5, ORP8, ORP9, ORP10 and ORP11 are evolutionarily closest to the yeast PS transporter Osh6, and previous reports have shown that ORP5, ORP8 and ORP10 specifically transfer PS but not sterols30,31,32. The main ORP proteins recruited to damaged lysosomes, ORP9, ORP10 and ORP11 (Supplementary Table 1), all carry highly conserved potential PS-binding sites (Extended Data Fig. 7a). Consistent with this structural homology, assessment of in vitro lipid transport using a fluorescence resonance energy transfer (FRET)-based assay demonstrated that ORP9 and ORP11 specifically transferred PS but not cholesterol (Fig. 3a,b and Extended Data Fig. 7b–d).

Fig. 3: ORP9, ORP10 and ORP11 mediate rapid lysosomal repair by ER-to-lysosome transfer of phosphatidylserine.
figure 3

a, Schematic of the in vitro lipid transport assay for ORP9 and ORP11. Note that nitrobenzoxadiazole (NBD) is attached to one fatty acid chain but not the head group of PS. PE, phosphatidylethanolamine; Rh, rhodamine; DGS–NTA(Ni), a nickel-chelating lipid that binds His-tagged proteins. b, Lipid transfer assay demonstrating the increase in NBD fluorescence stimulated by ORP9 and ORP11 in the presence of NBD–PS but not NBD–cholesterol (Chol) or TopFluor (TF)-cholesterol. Data are mean ± s.e.m.; n = 3 per condition. c, U2OS cells stably expressing the PS probe GFP–Lact-C2 were treated with LLOME for 20 min and washed with detergent before immunostaining of endogenous IST1 and LAMP1. Scale bars, 10 μm. d, Colocalization of GFP–Lact-C2 and LAMP1 in the indicated U2OS cell lines. Data are mean ± s.e.m. of Pearson’s correlation coefficient; n = 30 cells from 3 trials per condition. See fluorescence images in Extended Data Fig. 7f. e, Schematic illustration of different ORP mutants used in f,h. f, U2OS cells stably expressing GFP–Lact-C2 and indicated mCherry–ORP9 and ORP11 proteins (mutants are notated as ORP9 mutant/ORP11 mutant) were treated as in c, and the Pearson’s correlation coefficient for GFP–Lact-C2 and LAMP1 were quantified. Data are mean ± s.e.m.; n = 30 cells from 3 trials for each condition. See fluorescence images in Extended Data Fig. 7k. g, Damaged lysosomes purified from wild-type cells, but not those from ORP-QKO cells, are enriched in PS. Total lysosomal lipids were spotted onto PVDF membranes and PS was detected using purified eGFP–Lact-C2. Data are mean ± s.e.m.; n = 3 per condition. See more details in Extended Data Fig. 7m,n. h, eGFP–galectin-3 assays demonstrating restored rapid lysosomal repair in ORP-QKO cells reconstituted with wild-type ORP9 and ORP11, but not indicated mutants of ORP9 and ORP11. Data are mean ± s.e.m.; 50–100 cells, n = 3 per condition. See fluorescence images in Extended Data Fig. 8a. i, Schematic of chimeric proteins generated to replace ORPs. The substrate specificity of each transporter is indicated. Chol, cholesterol. j, eGFP–galectin-3 assays showing restored rapid lysosomal repair in ORP-QKO cells expressing OSBP–ORP chimeric proteins. Data are mean ± s.e.m.; 50–100 cells quantified, n = 3 per condition. See fluorescence images in Extended Data Fig. 8d. Unpaired, two-tailed t-tests.

Source data

In cells, the PS-specific probe GFP–lactadherin C2 domain33 (GFP–Lact-C2) showed robust lysosomal accumulation upon LMP (Fig. 3c,d and Extended Data Fig. 7e), which was dependent on PI4K2A and its kinase activity, and required the ORP proteins (Fig. 3d and Extended Data Fig. 7f). Consistent with the ORP proteins being PS transporters, the recruitment of ORP9 preceded the accumulation of lysosomal PS (Extended Data Fig. 7g,h). In ORP-QKO cells, lysosomal PS accumulation was successfully rescued by re-expression of any two of ORP9, ORP10 and ORP11, with the strongest effects again seen with expression of ORP9 and ORP11 (Extended Data Fig. 7i,j). ORP mutants deficient in lysosomal targeting (RE) or PtdIns4P–PS counter transport (AAA or HHAA) were completely incapable of generating PS accumulation on damaged lysosomes (Fig. 3e,f and Extended Data Figs. 5n–p and 7k), although some were still able to rescue ER–lysosome tethering (Extended Data Fig. 7l). ORP-dependent lysosomal PS transport was further supported by demonstrating increased PS levels in lysosomes purified from LLOME-treated wild-type but not ORP-QKO cells (Fig. 3g and Extended Data Fig. 7m,n). These data are consistent with ORP9, ORP10 and ORP11 orchestrating PtdIns4P-driven PS transport from the ER to damaged lysosomes, analogous to the previously reported ER-to-plasma membrane PS transfer catalysed by ORP5 and ORP8 (ref. 31).

PS mediates rapid lysosomal repair

Whereas reconstitution with any two of the three wild type ORP proteins robustly rescued lysosome repair in ORP-QKO cells (Fig. 2i), various OPR mutants defective in PS transport were incapable of restoring the lysosomal repair capacity (Fig. 3h and Extended Data Fig. 8a). Exploiting the capacity of OSBP to homodimerize (Extended Data Fig. 6n), we generated OSBP–ORP chimeras that bypassed the ORP-heterodimerization requirement for lysosomal recruitment (Fig. 3i and Extended Data Fig. 8c), which enabled each individual chimeric PS transporter to rescue lysosomal repair in ORP-QKO cells (Fig. 3j and Extended Data Fig. 8d). We further demonstrated that the mammalian ORP proteins could be replaced by yeast Osh6 (OSBP–Osh6) (Fig. 3i), which fully rescued rapid lysosomal repair through its PS transport activity (Fig. 3j and Extended Data Fig. 8c,d). Thus, lysosomal transport of PS mediates rapid lysosomal repair.

While the major membrane tethers ORP9/10/11 mediate PS transport for lysosomal repair, we identified OSBP as a parallel, complementary membrane tether (Extended Data Fig. 6e) that transported cholesterol to damaged lysosomes (Extended Data Fig. 8e, f), consistent with its known function29,34. Interestingly, reconstituting lysosomal cholesterol transport in ORP-QKO cells was also sufficient to restore rapid lysosomal repair (Extended Data Fig. 8g–h), in line with cholesterol’s known role in increasing membrane stability35. Thus, OSBP provides an auxiliary cholesterol transport pathway for rapid lysosomal repair.

ATG2 is a PS effector in lysosomal repair

ATG2A, a recently reported lipid transporter potentially activated by PS36,37,38,39, was rapidly recruited to damaged lysosomes (Fig. 4a,b and Supplementary Video 4). Unlike PI4K2A, lysosomal Ca2+ release was insufficient to recruit ATG2A (Supplementary Video 5). Cells lacking both ATG2A and ATG2B had marked defects in rapid lysosomal repair, which was fully rescued by ATG2A re-expression (Fig. 4c and Extended Data Fig. 9a,b). The lysosomal repair defects in ATG2A and ATG2B double-knockout (ATG2A/B-DKO) cells took longer to develop (2–4 h) than in PI4K2A-KO or ORP-QKO cells (0.5–1 h), probably owing to OSBP-mediated compensatory repair activity (Extended Data Fig. 9c–e). Thus, ATG2A and ATG2B act in parallel with OSBP and cholesterol but not PS, suggesting that ATG2A and ATG2B might be downstream effectors of PS in rapid lysosomal repair. Supporting this notion, the in vitro lipid transport activity of ATG2A was potently stimulated by the addition of PS in the acceptor membrane (Fig. 4d,e).

Fig. 4: PS-activated lipid transport by ATG2 mediates direct, rapid lysosomal repair.
figure 4

a, Live-cell imaging demonstrating rapid recruitment of eGFP–ATG2A to damaged lysosomes in U2OS cells stably expressing eGFP–ATG2A and LAMP1–mCherry. Scale bar, 10 μm. b, Quantification of eGFP–ATG2A puncta in individual U2OS cells upon LLOME treatment. c, eGFP–galectin-3 assay showing defects of rapid lysosomal repair in four independent clones of ATG2A/B-DKO cells. Data are mean ± s.e.m.; 50–100 cells, n = 3 for each condition. See more details in Extended Data Fig. 9a,b. d, Experimental design for the reconstitution of PS-stimulated lipid transport by ATG2A using a FRET-based assay. e, PS in acceptor liposomes potently stimulates ATG2A-dependent NBD fluorescence. Data are mean ± s.e.m.; n = 3 per condition. f, Illustration of lipid transport mutants of ATG2A. Flexible loops outside of the main structure are not shown. g, In vitro lipid transport assays testing the activity of various ATG2A mutants. Data are mean ± s.e.m.; n = 3 per condition. h, eGFP–galectin-3 assay demonstrating that ATG2A lipid transport mutants cannot rescue rapid lysosomal repair in ATG2A/B-DKO cells. Data are mean ± s.e.m.; 50–100 cells, n = 3 per condition. See fluorescence images in Extended Data Fig. 9f. i, In vitro ATG2A lipid transport assays in the presence of 1 μM maltose-binding protein (MBP) or MBP-tagged ATG2A-CT (MBP–CT). Data are mean ± s.e.m.; n = 3 per condition. j, Lysosomal repair defects in wild-type cells stably expressing LAMP1–CT but not in cells expressing LAMP1–mCherry or mCherry–CT. Data are mean ± s.e.m.; 50–100 cells were quantified, n = 3 per condition. See fluorescence images in Extended Data Fig. 9h. k, A cartwheel drawing of amino acids 1757–1789 from ATG2A-CT. Conserved residues mutated in the 5E mutant are indicated. l, The ATG2A(5E) mutant is not activated by PS-positive acceptor liposomes (A3) in lipid transport assays. Data are mean ± s.e.m.; n = 4. A2, charge-matched control acceptor liposomes. m, The ATG2A(5E) mutant cannot rescue rapid lysosomal repair in ATG2A/B-DKO cells. Data are mean ± s.e.m.; 50–100 cells were quantified, n = 3 per condition. See fluorescence images in Extended Data Fig. 9n. Unpaired, two-tailed t-tests.

Source data

ATG2 repairs lysosomes by lipid transport

ATG2 is proposed to transfer lipids between membranes through its hydrophobic tunnel36,37,38,39. Similar to previously reported mutants of another tunnel-like protein—VPS13 (ref. 40)—we designed three independent mutants (Mut1–3) in which the central hydrophobic tunnel of ATG2A is blocked (Fig. 4f), based on its predicted AlphaFold structure41,42. All three mutants were completely deficient in lipid transport in vitro (Fig. 4g) and were unable to rescue rapid lysosomal repair (Fig. 4h and Extended Data Fig. 9f) or macroautophagy (Extended Data Fig. 9g) when expressed in ATG2A/B-DKO cells. In addition, ATG2A without its carboxyl terminus (ΔCT) lost its in vitro lipid transport activity (Fig. 4g) and was unable to reconstitute lysosomal repair activity in ATG2A/B-DKO cells (Fig. 4h and Extended Data Fig. 9f). Therefore, the lipid transport activity of ATG2A is required for rapid lysosomal repair.

We next focused on the CT domain of ATG2, which is predicted to form amphipathic helices (Fig. 4f), a structure that is well known for membrane embedding43,44. Since ATG2A probably binds to the acceptor membrane (for example, damaged lysosomes) through its CT domain, we reasoned that the CT fragment could potentially compete with full-length ATG2A and ATG2B for membrane binding. Supporting this notion, a purified CT fragment (ATG2A amino acids 1723–1938) acted in a dominant-negative manner to inhibit in vitro lipid transport by full-length ATG2A (Fig. 4i). Similarly, overexpression of lysosomal-anchored ATG2A-CT (LAMP1–CT), but not soluble mCherry-tagged ATG2A-CT (mCh-CT), caused marked lysosomal repair defects in wild-type cells, which phenocopied ATG2A/B-DKO cells (Fig. 4j and Extended Data Fig. 9h) without however, affecting macroautophagy (Extended Data Fig. 9i). Indeed, rapid lysosomal repair, macroautophagy and lysosomal damage-induced LC3 lipidation appear to be three distinct pathways (Extended Data Fig. 9h,i,o–u and Supplementary Table 6). These data suggest that ATG2 uses its amphipathic CT to interact with lysosomal membranes for lipid delivery and that the role of ATG2A and ATG2B in lysosomal repair and macroautophagy are separable, although both processes require the lipid transport activity of ATG2.

We then tested whether PS regulates the membrane binding of ATG2A-CT. In vitro liposome pulldown assays revealed substantial PS-stimulated membrane binding of ATG2A-CT (Extended Data Fig. 9j). Positively charged residues on the polar face of amphipathic helices in other proteins have been reported to sense PS-positive membranes45,46. A similar group of highly conserved basic residues was identified on the polar face of ATG2A-CT, on the basis of which we generated the 5E mutant (Fig. 4k and Extended Data Fig. 9k,l). Purified ATG2A-CT(5E) retained normal basal membrane binding (Extended Data Fig. 9j), suggesting correct amphipathic folding, but did not demonstrate PS-stimulated membrane binding (Extended Data Fig. 9j). Notably, the 5E mutant of full-length ATG2A (ATG2A(5E)) acted similarly to ATG2A lacking the CT domain (ATG2A(ΔCT)). ATG2A(5E) was not activated by PS in lipid transport assays in vitro (Fig. 4l), exhibited weaker retention on damaged lysosomes (Extended Data Fig. 9m), and did not rescue lysosomal repair in ATG2A/B-DKO cells (Fig. 4m and Extended Data Fig. 9n). Thus, rapid lysosomal repair requires PS-stimulated lysosomal lipid transport by ATG2 (Supplementary Table 5).


Our study reveals a phosphoinositide-initiated membrane tethering and lipid transport (PITT) pathway that is essential for rapid lysosomal repair (Extended Data Fig. 10). PI4K2A generates PtdIns4P, the initiating signalling lipid that decorates damaged lysosomes. In response to PtdIns4P, the recruitment of multiple ORPs establish new and extensive ER MCSs, leading to ER-to-lysosome transport of PS and cholesterol, as complementary and parallel mechanisms to support rapid lysosomal repair. These observations are broadly consistent with other recent reports implicating OSBP in PtdIns4P-dependent endolysosomal cholesterol transport34,47. Finally, lysosomal PS accumulation activates ATG2-mediated lipid delivery for direct membrane repair, a function for ATG2 that is independent of its canonical role in autophagosome formation (Supplementary Table 6). The current work, as well as previous studies, are limited by the lack of evidence demonstrating high volume in vitro lipid transport by ATG2. Such in vitro activity might be achieved once the factors ensuring directional lipid transport by ATG2 are identified. Nevertheless, we demonstrated strict dependence of rapid lysosomal repair on ATG2 lipid transport activity, which suggests that cellular ATG2 activity is likely sufficiently robust to mediate membrane repair, although additional studies are required to solidify this conclusion.

Increasing evidence implicates a decline in lysosomal integrity and function in normal ageing and in age-related disease1,2,3. In cell-based models, we observed that deletion of PI4K2A caused exacerbation of a lysosomal storage disease (Extended Data Fig. 4i) and tau fibril spreading (Fig. 1j, Extended Data Fig. 4h). Coupled with the neurodegenerative phenotype previously reported in Pi4k2a-KO mice48, our observations suggest that the PITT pathway may have therapeutic implications for a wide range of age-dependent diseases characterized by impaired lysosomal function.


Cell culture, chemicals, and treatments

All cell lines were from ATCC. 293T, U2OS, PC3 and BJ cells were authenticated through short tandem repeat profiling and the profiling data are publicly available from ATCC. Cell lines in this study have different morphologies and growth rates and contamination were constantly monitored. All cell lines used in this study were free from mycoplasma contamination based on PCR detection and were regularly maintained with mycoplasma reagent. 293T, U2OS, COS7, and BJ cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM); PC3 cells were cultured in Roswell Park Memorial Institute (RPMI) 1640. Both media were supplemented with 8% fetal bovine serum (FBS) and penicillin/streptomycin. All cells were maintained at 37 °C with 5% CO2. LLOME (Sigma, L7393) was dissolved in ethanol and stored in aliquots at −20 °C. Gly-Phe-β-naphthylamide (GPN, Cayman Chemical, 14634), Brefeldin A (Sigma, B5936), ML-SA1 (Cayman Chemical, 29958), MEK6–83 (Cayman Chemical, 21944), O-methyl-serine dodecylamide hydrochloride (MSDH, Avanti Polar Lipids, 850546) were dissolved in dimethyl sulfoxide (DMSO) and stored at −20 °C. Nano-silica (Invivogen, tlrl-sio-2) was suspended in water and sonicated right before each use. Tau fibrils (SPR-329) were from StressMarq. Chemicals, silica, or tau fibrils were directly added to cell culture media to induce lysosomal membrane damage with equal volume of vehicle (ethanol or DMSO) added into control wells. For cell treatment that requires medium change, fresh media were pre-warmed overnight in an empty dish in the same incubator, which minimizes disturbance to cells caused by medium change.


The following antibodies were from Santa Cruz Biotechnology: ORP9 (A-7, sc-398961, immunofluorescence 1:1,000); PI4K2A (sc-390026, immunofluorescence 1:500); LAMP2 (sc-18822, immunofluorescence 1:200); CD2AP (sc-25272, western blot 1:1,000); CHMP3 (sc-166361, immunofluorescence 1:1,000); Golgin-97 (sc-59820, immunofluorescence 1: 500); LAMP1 (sc-20011, immunofluorescence 1:200); SQSTM1/p62 (sc-28359, immunofluorescence 1: 1,000, western blot 1: 2,000); GAPDH (sc-365062, western blot 1:5,000); tubulin (sc-5286, western blot 1:3,000); GFP (sc-9996, western blot 1:3,000, immunoprecipitation 2 μg per reaction). The PI4KB antibody (611816, western blot 1:1,000, immunofluorescence 1:200) and GM130 antibody (610822, immunofluorescence 1:1,000) were from BD Biosciences. Anti-rabbit IgG (H+L) CF350 (SAB4600412, immunofluorescence 1:200), Flag (M2, immunofluorescence 1:1,000; F7425, western blot 1:3,000), Flag M2 agarose (immunoprecipitation 10 μl beads per reaction) were from Sigma. Rabbit anti-LAMP1 monoclonal antibody (9091, immunofluorescence 1:100) was from Cell Signaling. Mouse anti-Alix antibody (634502, immunofluorescence 1:1,000) was from Biolegend. The following antibodies produced in rabbits were from Proteintech Group: STAM (12434-1-AP, western blot 1:1,000); IST1 (19842-1-AP, immunofluorescence 1:1,000, western blot 1:2,000); PI4KA (12411-1-AP, western blot 1:1,000); STAMBP (11346-1-AP, western blot 1:1,000); ORP9 (11879-1-AP, western blot 1:1,000); OSBP (11096-1-AP, western blot 1:1,000, immunofluorescence 1:1,000); PI4K2A (15318-1-AP, western blot 1:1,000); ATG2A (23226-1-AP, western blot 1:500); ATG2B (25155-1-AP, western blot 1:500), galectin-3 (14979-1-AP, immunofluorescence 1:200). The following antibodies were from Bethyl Laboratories: ORP11 (A304-580A, western blot 1:2,000, immunofluorescence 1:1,000) and ORP10 (A304-885A, western blot 1:1,000). Alexa-488/594- and Pacific Blue-conjugated secondary antibodies were obtained from ThermoFisher Scientific.

DNA cloning

For stable expression of proteins in this study, the relevant DNA sequences were cloned into pCDH-CMV-MCS between restriction sites BamHI and NotI. All cDNA sequences were of human origin unless otherwise specified. The cDNAs for ORP3, ORP5, ORP9, PI4K2A, ATG2A, Osh6 and Kes1 were from the DNASU plasmid repository at Arizona State University (HsCD00829348, HsCD00439845, HsCD00820675, HsCD00618068, HsCD00863023, ScCD00012609 and ScCD00009548, respectively). The ORP10 cDNA was a gift from V. Olkkonen. The ORP11 cDNA was purchased from Sino Biological (HG22085-U). The cDNA for PI4K2B was originally from Addgene (pDONR223-PI4K2B, 23507) which contained a 15-bp insert in the middle that introduced a premature stop codon, leading to the production of an inactive fragment majorly localized into the nucleus. This insert was removed when cloning PI4K2B into pCDH-CMV-MCS to express the full-length wild-type protein. For most fusion proteins, the peptide linker GSGSGS was used. To generate CRISPR guide RNA-resistant mutations or other point mutations, two fragments of the cDNA of interest were amplified with their joining ends containing the target mutations and then fused together into pCDH-CMV-MCS through infusion reactions. To generate the Lyso-TurboID construct (pCDH-LAMP1-mGFP-TurboID), the LAMP1–mGFP sequence was amplified from Addgene Plasmid 34831 and fused with TurboID sequence into pCDH-CMV-MCS. To constitutively target PI4K2B to lysosomes, the open reading frame (ORF) of PI4K2B was fused with the LAMP1–mCherry sequence into pCDH to achieve stable expression of LAMP1– mCherry–PI4K2B. Amino acids 1–382 of OSBP without its C-terminal cholesterol transport domain were used as an ER–lysosome tether when designing artificial fusion proteins to rescue ER-to-lysosome lipid transport in ORP-QKO cells. In mCherry–OSBP–ORP9, the sequences encoding mCherry, OSBP residues 1–382 and ORP9 residues 337–746 were fused together with GSGSGS linkers in between. Similarly, OSBP residues 1–382 was fused with ORP10 residues 348–764, ORP11 residues 322–747, and the full-length sequences of Kes1 or Osh6 to generate the relevant fusion proteins.

Stable cell line generation

Except for RNA interference and transient expression of GFP–P4M and SARS-CoV-2-ORF3A, all other experiments were based on stable cell lines with genetic knockout and/or stable ectopic expression of proteins. pCDH vectors carrying the ORF sequences of specific genes were used to generate lentiviruses to infect recipient cells, after which the cells were treated with puromycin to select against uninfected cells. In most situations, no selection was necessary, as the infection rates were close to 100%. A titration of viruses was usually performed to find out the lowest viral dosage that can lead to expression of the target protein in more than 90% of cells. To generate knockout cell lines, multiple CRISPR–Cas9 guide sequences for each gene were tested to identify the guide that most effectively diminished the target protein in the CRISPR pool. LentiCRISPR.v2 (Addgene, 52961) carrying the selected guide sequence was used for lentivirus packaging. The infection rate of U2OS cells by CRISPR viruses was remarkably high that no cell death was observed after puromycin treatment of infected pools when all uninfected control cells died. See Supplementary Table 4 for the CRISPR guide sequences.

RNA interference

Double knockdown of ALIX and TSG101 was performed to block the recruitment of ESCRT subunits to damaged lysosomes, as previously described7. For transient knockdown of PI4K2A and PI4KA, two different siRNA sequences were pooled and transfected at the same time using Lipofectamine RNAiMAX, following the manufacturer’s instructions. See Supplementary Table 4 for the siRNA sequences.

Biotinylation and purification of lysosomal surface proteins

293T cells stably expressing Lyso-TurboID were cultured until 80% confluent (Extended Data Fig. 1d). Cells were treated with 500 μM LLOME or ethanol (viechle control) for 30 min to induce lysosomal membrane damages and membrane repair events, followed by the addition of 50 μM biotin. After 30 min of incubation allowing for biotinylation of lysosomal surface proteins, cells were washed twice with 10 ml pre-chilled PBS, scraped into 1 ml of cold PBS per dish, and transferred into 1.5 ml tubes. Cells were spun down at 1,000g for 3 min and resuspended in 500 μl hypotonic buffer (10 mM HEPES, pH 7.5, 5 mM MgCl2, 10 mM NaCl). Resuspended cells were passed through a 25 G needle 10 times and spun at 1,000g for 3 min to remove intact cells and nuclei in the pellet (P1). The supernatant S1 was further spun at 20,000g for 20 min to precipitate most membranes including lysosomes with biotinylated proteins (P20). P20 was resuspended in lysis buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.5% Triton X-100, 2 mM NaF, 5 mM MgCl2, protease inhibitor cocktail) and briefly sonicated to completely dissolve the membrane. The samples were further spun at 15,000 g for 5 min to remove any undissolved aggregates. The supernatants were transferred to a new tube with 100 μl well-resuspended streptavidin magnetic beads and rotated at 4 °C for 2 h. The beads were washed twice sequentially with three different buffers: buffer 1 (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.8% SDS), buffer 2 (50 mM Tris-HCl, pH 8.0, 1 M NaCl, 0.5% Triton X-100), buffer 3 (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.5% Triton X-100). Proteins bound on the beads were eluted by adding 50 μl 2× SDS sample buffer (0.1 M Tris-HCl, pH 6.8, 4% SDS, 20% glycerol, 2% 2-mercaptoethanol, 0.01% bromphenol blue) and heated to 95 °C for 5 min.


Cells were seeded on glass coverslips (Warner Instruments, 64–0712) in 24-well plates. PC3 cells took 2 to 3 days to firmly attach to glass surfaces and were treated 2 to 3 days after seeding. All the other cell lines attached to glass coverslips quickly and were treated 24 h after seeding. LLOME and GPN were directly added to each well to induce lysosomal membrane damage. When medium change was needed after lysosomal damage was induced, fresh media were pre-warmed overnight in an empty dish in the same incubator to minimize any potential disturbance to the cells induced by medium change. Cells were fixed with 4% paraformaldehyde (PFA) in PBS for 30 min at room temperature, permeabilized with 0.1% Triton X-100 for 5 min at room temperature, and blocked for 1 h at room temperature with 1× fluorescent blocking buffer (ThermoFisher Scientific, 37565) which was also used to dilute all primary and secondary antibodies in immunofluorescence staining. Primary antibody incubation was performed at room temperature for 2 h or at 4 °C overnight. Cells on glass coverslips were washed with PBS three times and incubated with fluorescently labelled secondary antibodies at room temperature for 30 to 60 min, followed by PBS wash for three times. When necessary, DAPI staining of the nuclei was carried out after secondary antibody staining. The coverslips were mounted on slides using VECTASHIELD HardSet Antifade Mounting Medium (Vector Laboratories, H-1400). Images were taken using a Leica SP8 LIGHTNING confocal system using the built-in software Leica Application Suite X Multiple controls were included to avoid nonspecific staining and cross talks between channels. An Okolabstage-top incubator was used for live-cell imaging on the same confocal system. All images in the same figure panel were taken under the same software setting and equally processed in Adobe Photoshop. Colocalization was quantified by Coloc2 in ImageJ with Person’s coefficient used to determine the level of colocalization.


Immunoprecipitation was done as previously described49. 293T cells stably expressing Flag- and eGFP-tagged ORPs proteins were grown to confluent in six-well plates. Cells were washed briefly with cold PBS and lysed in 1 ml immunoprecipitation buffer per well (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl2, 2 mM NaF, 2 mM Na3VO4, 0.5% Triton X-100, protease inhibitor cocktail). Lysed cells were scraped into 1.5 ml tubes and briefly sonicated before centrifugation at 15,000 g for 10 min. The supernatants were collected as whole cell lysates, 100 μl of which were saved as input controls. The remaining 900 ul lysates were split into 3× 300 μl aliquots for immunoprecipitation with anti-Flag, anti-GFP, and normal IgG, respectively. Mouse anti-Flag M2 agarose gel (Sigma) was used to immunoprecipitate Flag-tagged proteins, whereas mouse anti-GFP (sc-9996, Santa Cruz) or normal IgG were used with protein A/G beads. The immunoprecipitation mixtures were rotated at 4 °C for 2 h and then washed three times with immunoprecipitation buffer without protease inhibitors. The immunoprecipitated proteins on beads were eluted by heating in 2× SDS loading buffer for 5 min, followed by immunoblotting analysis of the ORP proteins.

Lipofuscin detection

Cells on coverslips were fixed with 4% PFA in PBS for 30 min at room temperature and washed with PBS before mounted on slides using VECTASHIELD HardSet Antifade Mounting Medium (Vector Laboratories, H-1400). Lipofuscin was excited at 448 nm and emissions from 453–750 were collected as confocal images.


Cells were washed twice with cold PBS and scraped into lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% Triton X-100, 2 mM NaF, 5 mM MgCl2, protease inhibitor cocktail). The samples were briefly sonicated to disrupt membrane aggregates and then spun at 15,000 g for 10 min to remove insoluble components. The supernatants were collected as whole cell lysates which is mixed with equal volume of 2× SDS loading buffer (0.1 M Tris-HCl, pH 6.8, 4% SDS, 20% glycerol, 2% 2-mercaptoethanol, 0.01% bromphenol blue) and heated at 95 °C for 5 min. Equal amounts of total proteins were separated on a 4–20% precast polyacrylamide gel (Biorad, 4561096) and transferred onto 0.2 µm nitrocellulose membranes using the Trans-Blot Turbo system. The membrane was blocked with StartingBlock Blocking Buffer (ThermoFisher, 37542) for 30 min at room temperature and then incubated with primary antibody at 4 °C overnight. The membrane was then washed and incubated with horseradish peroxidase-conjugated secondary antibodies at room temperature for 1 h. The membrane was washed and exposed with SuperSignal West Pico PLUS Chemiluminescent Substrate (ThermoFisher, 34580) in a ChemiDoc MP Imaging System.

Lysosomal PtdIns4P production assay

The production of PtdIns4P on damaged lysosomes were tracked by a stably expressed PtdIns4P probe OSBP-PH–GFP17 or alternatively by transiently transfected GFP–P4M50 of which the stable cell line generation was not successful. Stable expression of OSBP-PH–GFP was achieved by a lentiviral approach. The DNA sequence of OSBP PH domain was cloned into pCDH, a lentiviral vector, with GFP fused on the C-terminal end. All cell types in which we have tried stable expression of OSBP-PH–GFP showed constitutive targeting of the probe to the Golgi complex with the other regions of the cell demonstrating only a diffuse signal. The PtdIns4P production on damaged lysosomes was assessed by tracking the lysosomal recruitment of OSBP-PH–GFP upon lysosomal damage by either 1 mM LLOME for 30 min or 100 μM GPN for 5 min. To rule out the interference of pre-existing Golgi-localized OSBP-PH-GFP puncta in assessing LMP-induced lysosomal puncta of the probe, cells were pre-treated with brefeldin A to disassemble the whole Golgi complex before LLOME-mediated lysosomal damage. Note that the lysosomal recruitment of PI4K2A and ORP9/10/11 happened much more quickly (within 10 min) than the recruitment of OSBP-PH–GFP (peaks at 30 min). This is likely because ORP9/10/11 have higher affinity to PtdIns4P than does OSBP-PH–GFP. Live-cell imaging and confocal analysis of fixed cells stained for endogenous markers of damaged lysosomes were performed. To ensure the lysosomal recruitment of OSBP-PH-GFP was due to lysosomal PtdIns4P production, we identified the enzyme PI4K2A as the source of PtdIns4P on damaged lysosomes. We confirmed that the lysosomal recruitment of OSBP-PH–GFP was abolished in cells lacking PI4K2A, even though their lysosomes could still be damaged.

Cell-based PS transport assay

Cells stably expressing the genetically encoded PS probe GFP–Lact-C2 (ref. 28) were treated with 1 mM LLOME for 20. This probe shows strong background signal from the cytosol and the plasma membrane, which was removed by a brief detergent wash before fixation. Specifically, cells with or without LLOME treatment were gently rinsed for 5–10 s with 0.1% Triton X-100, followed by PBS rinse for another 5–10 s. Cells were then immediately fixed in 4% PFA for 10 min. Fixed cells were blocked with 1× fluorescent blocking buffer (ThermoFisher Scientific, 37565) for 30 min at room temperature and stained with antibodies against the lysosome marker LAMP1 and/or the damaged lysosome marker IST1. The colocalization of GFP–Lact-C2 and LAMP1 was quantified by Person’s correlation coefficient.

Lysosomal repair assays

We aimed to analyse lysosomal repair in live cells with minimal interference from chemical staining or washing steps. Two approaches were established to monitor lysosomal repair in live cells, with one based on the stable expression of lyso-pHluorin and the other of eGFP–galectin-3.

Lyso-pHluorin is based on the lysosomal marker CD63 as the backbone with the GFP variant pHluorin that exhibits no fluorescence in the acidic lysosomal lumen but does emit when the lysosomal membrane is damaged with its lumenal pH neutralized20. Thus, the de-quenching and re-quenching of the lyso-pHluorin signal reflects lysosomal damage and repair after LLOME treatment. A U2OS lyso-pHluorin stable cell line was treated with 1 mM LLOME for 5 min and then washed and chased for up to 12 h. To minimize disturbance to cells caused by medium change especially after acute lysosomal damage, fresh media were pre-warmed overnight in an empty dish in the same incubator, which were used to gently rinse cells twice to wash out LLOME. Note, the washing step is critical to ensure efficient lysosomal repair, which can cause cell death and detachment if not properly handled. The lyso-pHluorin signals were checked every 1–2 h with live-cell imaging. Cells with more than 5 lyso-pHluorin puncta were defined as unrepaired. The number of repaired and unrepaired cells were manually counted. In each condition, 100–500 cells were counted. The whole process was independently repeated at least three times.

The other approach is based on stably expressed eGFP–galectin-3 which accumulates in the lysosomal lumen when the membrane pore is big enough to allow its entrance from the cytosol7,51. When stably expressed, eGFP–galectin-3 only shows diffuse cytosol signal similar to that of lyso-pHluorin. Upon lysosomal membrane damage, the number and sizes of eGFP–galectin-3 puncta correlate with the degree of membrane damage5,7. U2OS cells stably expressing eGFP–galectin-3 were treated continuously with 1 mM LLOME for 1 to 5 h and monitored for the formation of eGFP–galectin-3 puncta over time using live cells. The total punctate intensities above threshold were quantified in ImageJ. Note that images should be taken using live cells, as fixation causes puncta formation of eGFP–galectin-3.

Protein purification

ORP9, ORP11 and OSBP were purified by two-step purifications using 6×His tag and twin-strep-tag. All three proteins carry an N-terminal Flag-tag and a C-terminal twin-strep-tag. The FATT motif of ORP9/OSBP was replaced with a 6×His tag. A GGGS linker was added after the 6×His tag and a GSGSGS linker was added between the C-terminal end of each protein and the twin-strep-tag. As ORP11 does not carry a strong FATT motif, a 6×His tag followed with a GGGS linker was inserted into ORP11 between amino acids 321 and 322. The coding sequence for the three fusion proteins were cloned into pCDH for stable protein expression in HEK293T cells. The cells were washed with PBS, pelleted, and resuspended in lysis buffer containing 100 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10% glycerol, 1% Triton X-100, and protease inhibitor cocktail. Cells were sonicated and centrifuged at 15,000 g for 10 min. His-tagged proteins from the supernatant were purified using Ni-NTA agarose and eluted into elution buffer containing 100 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10% glycerol, 300 mM imidazole. The eluted protein was dialysed and further purified using Strep-Tactin column and eluted into 20 mM Tris-HCl, pH 8.0, 200 mM NaCl. The eluted protein was further dialysed with 20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM DTT, aliquoted, and stored at −80 °C.

Flag–MBP, Flag–MBP–CT, Flag–ATG2A and all related mutants were purified with anti-Flag M2 beads. Flag–MBP and Flag–MBP–CT were cloned into pCDH and then stably expressed in 293T cells by a lentiviral approach. The coding sequence for Flag–ATG2A with a 26-amino acid linker followed by a 6xHis tag at its C-terminal end was cloned into pCDNA3.0 and transiently transfected into 293T cells. Forty-eight hours after transfection, cells were washed with PBS and resuspended into lysis buffer containing 100 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10% glycerol, 1% Triton X-100, and protease inhibitor cocktail. After sonication and clearance centrifugation, the lysates were incubated with M2 beads for 1 h at 4 °C with rotations. The mixture was then transferred into an open column and sequentially, extensively washed with three buffers. Buffer 1: 100 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10% glycerol; buffer 2: 100 mM Tris-HCl, pH 8.0, 500 mM NaCl; buffer 3: 20 mM Tris-HCl, pH 8.0, 200 mM NaCl. Bound proteins were then eluted with 0.2 mg ml−1 3×Flag peptide in buffer 3. DTT was added to the eluted proteins at a final concentration of 1 mM. Proteins were aliquoted and stored at −80 °C.

Liposome preparation

The following lipids used for liposome preparation were from Avanti polar lipids: 18:1-12:0 NBD–PS (810195), 18:1 NBD–phosphatidylinositol (810145), NBD–cholesterol (810252), 18:1 lissamine–rhodamine–PE (810150), DOPC (850375), DOPE (850725), DOPS (840035), DGS–NTA (790404), brain PtdIns4P (840045), biotin–phosphatidylinositol (870282). Lipids dissolved in chloroform were mixed in glass tubes and dried to thin films under nitrogen flow and were desiccated overnight. Five hundred microlitres of rehydration buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM DTT; or 25 mM HEPES, pH 7.5 150 mM NaCl, 1 mM DTT) was added to dried lipids to make 1 mM final total lipid concentration. The tubes was shaken vigorously for 30 min at room temperature, subject to 10 freeze–thaw cycles using liquid nitrogen and 37 °C water bath, and then extruded through a 100-nm filter 21 times for ORP9/11-mediated lipid transport assays. The donor liposomes for ATG2A-mediated lipid transport tests were sonicated, and acceptor liposomes were extruded through 50 nm filter 31 times. ATG2A lipid transport activity was only observed when sonicated donors were used, whereas the sizes of extruded ATG2A acceptor liposomes did not affect the overall lipid transport activity.

In vitro reconstitution of lipid transport

Lipid transport in vitro was measured in 40 μl reaction buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM DTT) in a 384-well plate through a FRET-based assay29,36,37,38,39. The donor and acceptor liposomes (25 μM total lipids for each) were mixed and the NBD fluorescence was monitored (excitation at 460 nm, emission at 535 nm) upon the addition of 100 nM purified proteins.

For the ORP9/11-mediated lipid transport assays, the donor liposomes contained 66% DOPC, 25% DOPE, 5% DGS–NTA (Ni), 2% 18:1-12:0 NBD–PS (or 2% NBD–cholesterol or 2% TopFluor(TF)-cholesterol), and 2% Rh–phosphatidylethanolamine, the PtdIns4P acceptor liposomes contained 70% DOPC, 25% DOPE, 5% PtdIns4P, and the neutral acceptor liposomes contained 75% DOPC, 25% DOPE. Note that, to allow for efficient PS recognition by ORPs, 18:1-12:0 NBD–PS carries the NBD tag at the end of one fatty acid chain instead of the serine head group. We introduced a His tag in the middle of both ORP9/11 (replacing the FATT motif in ORP9) so that they can bind to the donor membrane containing lipids labelled with DGS–NTA (Ni), mimicking ORP interactions with VAPA/B on the ER through FATT motifs. The acceptor membrane carries 5% of PtdIns4P for ORP recruitment to mimic damaged lysosomes without fluorescent lipids. When testing the lipid transport activity of individual ORP9 or ORP11, 100 nM of each was used. When testing the activity of ORP9–ORP11 heterodimers, the two proteins were premixed at a molar ratio of 1:1 and 100 nM of total proteins (50 nM each) were added to the liposome mixtures to start the reaction.

For ATG2A assays, the donor liposomes contained 63% DOPC, 25% DOPE, 5% DGS–NTA (Ni), 2% 18:1 NBD–phosphatidylethanolamine, 2% Rh–phosphatidylethanolamine, and 3% DOPS, the PS-positive acceptor liposomes (A3) contained 50% DOPC, 25% DOPE, 5% PtdIns4P, and 20% DOPS, the PS-negative acceptor liposomes (A1) contained 70% DOPC, 25% DOPE, and 5% PtdIns4P, an additional set of charge-matched control acceptor liposomes (A2) contained 50% DOPC, 25% DOPE, and 25% PtdIns4P. ATG2A carried a 26 aa linker followed with His tag at its C-terminal end, as previously described37. OSBP, which does not transport PS or NBD–phosphatidylethanolamine, was used as a physiologically relevant membrane tether between the donor and acceptor liposomes. Note that OSBP appeared to exhibit nonspecific activity towards NBD–phosphatidylethanolamine which could be avoided after a pre-incubation of OSBP at room temperature for 1 h. Similar to ORP9/11, a His tag was added to the middle of OSBP (see protein purification for details) so that it can tether the donor and acceptor membrane through the His tag and its N-terminal PH domain. After mixing donor and acceptor liposomes with 100 nM OSBP, 100 nM ATG2A was added immediately before fluorescence reading. All ATG2A lipid transport mutants and the impact of MBP–CT on ATG2A activity were tested with A3 acceptor liposomes.

Lysosomal purification and lipid extraction

Lysosomes were purified through LAMP1-Twin-Strep and streptavidin beads as described52. U2OS cells stably expressing LAMP1-Twin-Strep grown to confluent in 15 cm dishes were washed three times with chilled PBS, scraped into 1 ml potassium buffer (KPBS, 136 mM KCl, 10 mM KH2PO4, pH 7.5). Cells were centrifuged for 5 min at 1,000 g, resuspend in 1 ml of potassium buffer, and then passed through a 25 G needle 10 times. Cell homogenates were centrifuged at 1,000 g for 2 min and 800 μl supernatant was transferred to a new tube with 100 μl of streptavidin magnetic beads. The mixture was incubated with rotation for 5 min, followed by bead collection and three washes with a magnetic stand. For each wash, the beads were resuspended in 1 ml potassium buffer and transferred into a new tube. After washing, half of the beads were directly heated in SDS loading buffer to extract lysosomal proteins as input controls. The other half of the beads were used for total lysosomal lipid extraction using the methyl-tert-butyl ether (MTBE) extraction protocol which allows fast and clean lipid recovery53. The washed beads were suspended in 200 ul water and transferred to a glass tube with a Teflon-lined cap followed by the addition of 1.5 ml methanol. The tube was vortexed and 5 ml of MTBE was added. The mixture was then incubated at room temperature on a shaker for 1 h. Phase separation was induced by the addition of 1.25 ml ultrapure water. After 10 min of incubation at room temperature, the tube was centrifuged at 1,000 g for 10 min. The upper organic phase was collected, and the lower phase was re-extracted with 2 ml of the MTBE:methanol:water (10:3:2.5, v/v/v) mixture. Combined organic phase solutions containing lysosomal lipids were dried under a nitrogen flow. Extracted lipids were dissolved in 200 μl of CHCl3:methanol:water (60:30:4.5, v/v/v) and 10 μl of each sample was spotted onto a PVDF membrane using a glass capillary. The membrane was completely dried for 1 h with air flow in a fume hood and then immediately used for PS detection by GFP–Lact-C2.

PS detection by lipid–protein overlay assay

PIP strips membrane or lipid-spotted PVDF membrane was incubated overnight at 4 °C with binding buffer (1× PBS, 3% BSA, 0.1% Tween 20) containing 1 μg ml−1 purified GFP–Lact-C2. The membrane was washed the next day with the binding buffer 3 times 5 min each, followed by imaging to capture GFP signal.

Liposome binding assay

Liposomes were made similarly to acceptors in the ATG2A lipid transport assay except the addition of 2% Biotin–phosphatidylethanolamine to all liposomes. Magnetic streptavidin beads (10 μl) were incubated with 10 μl 500 uM liposomes (total lipid concentration) in 500 μl binding buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM DTT) on a rotation mixer for 30 min at room temperature. Following this, 1 μg of purified MBP–CT or its 5E mutant was added together with 100 μg BSA. The samples were rotated for 10 min at room temperature to allow for liposome binding and then washed twice with the binding buffer above plus 0.04% BSA. For each wash, the beads were resuspended and transferred to new Eppendorf tubes. Proteins bound to immobilized liposomes were extracted by directly heating the beads in SDS loading buffer for 5 min, followed by immunoblotting analysis.

Tau fibril spreading assay

Active human recombinant tau441 (2N4R) P301S mutant protein pre-formed fibrils (SPR-329) were purchased from StressMarq and used for the cell-based tau spreading assay. U2OS cells stably expressing tau K18-P301L/V337M-mRuby2 without pre-formed mRuby2 puncta were incubated with 250 nM sonicated tau fibrils and the formation of K18–mRuby2 aggregates in live cells were examined by fluorescence microscopy on a daily basis.


Confocal images were taken using a Leica SP8 LIGHTNING confocal system with the built-in software Leica Application Suite X Images were processed and assembled in Adobe Photoshop 20.0.4. Schematic illustrations were made with Microsoft Office Power Point Professional Plus 2016 and Biorender. AlphaFold structure of ATG2A was visualized in the PyMOL Molecular Graphics System, Version 2.4.0 Schrödinger, LLC. Cartwheel drawing of amino acid sequence was performed with HeliQuest54. Unpaired, two-tailed t-tests Statistical analysis was performed in Microsoft Office Excel Professional Plus 2016. Graphs were generated in Excel and GraphPd Prism 9.0.0. For colocalization quantification, Pearson's correlation coefficient was quantified by Coloc 2 version 3.0.5 in ImageJ (Fiji 1.53f51).

Statistics and reproducibility

All experiments were independently reproduced at least three times, except the initial Lyso-TurboID screen which was performed once. Strict standards were applied to screen for robust and unbiased results. No statistical methods were used to predetermine sample size. The investigators were not blinded to allocation. Data were presented as mean ± s.e.m. Statistical significance was determined by unpaired, two-tailed t-tests.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.