LYVAC/PDZD8 Is a Lysosomal Vacuolator

Abstract

Lysosomal vacuolation is commonly found in many pathophysiological conditions, but its molecular mechanisms and functions remain largely unknown. Here we show that the endoplasmic reticulum (ER)-anchored lipid transfer protein PDZ domain containing 8 (PDZD8), which we propose to be renamed as lysosomal vacuolator (LYVAC), is a general mediator of lysosomal vacuolation. Diverse lysosomal vacuolation inducers converged on lysosomal osmotic stress, triggering LYVAC recruitment through multivalent interactions. Stress-induced lysosomal lipid signaling contributed to both the recruitment and activation of LYVAC. Via directly sensing lysosomal phosphatidylserine (PS) and cholesterol, the lipid transfer domain of LYVAC mediated directional ER-to-lysosome lipid movement, leading to osmotic membrane expansion of lysosomes. These findings uncover an essential mechanism for lysosomal vacuolation with broad implications in pathophysiology.

Cytoplasmic vacuolation is a long-recognized phenomenon in mammalian cells, involving extensive dilation of intracellular membranes. (1, 2). It is observed in a broad spectrum of conditions including infection, chemotherapy, normal aging, lysosomal storage disease, cataracts, prion disease, and other neurodegenerative conditions (1–5). The most common cytoplasmic vacuoles are derived from late endosomes/lysosomes (hereafter referred to as lysosomes), but the exact roles of lysosomal vacuolation in pathophysiology remain unclear due to a lack of mechanistic studies.

Identification of LYVAC/PDZD8 as a lysosomal vacuolator

The loss-of-function of PIKfyve, a phosphatidylinositol 3-phosphate [PI(3)P] 5-kinase, is emerging as a driver of lysosomal vacuolation in multiple conditions, such as genetic mutations of PIKfyve or its interacting partners VAC14 and FIG4 (6), prion diseases (7), chronic ER stress (7), or cadmium toxicity (8). PIKfyve generates phosphatidylinositol 3,5-bisphosphate [PI(3,5)P2], a lipid messenger that regulates lysosomal ion homeostasis and osmotic balance (9–12). PIKfyve inhibitors, such as apilimod and YM201636, trigger robust lysosomal vacuolation, mimicking pathological conditions (13, 14). We used apilimod-induced vacuolation as a cellular model to investigate the underlying mechanisms. By stably expressing Lyso-TurboID (15), a promiscuous biotinylation enzyme (16) on the lysosomal surface, we specifically biotinylated, isolated, and characterized the lysosomal surface proteome before and after lysosomal vacuolation (Fig. 1A, fig. S1, A to C).

Figure 1. Unbiased proteomics reveal LYVAC as an essential mediator of lysosomal vacuolation.

(A) Schematic illustration of the Lyso-TurboID screen. P20-SA↓, streptavidin (SA) pulldown from P20, the pellet fraction after centrifugation at 20,000g. (B) Dot plot of proteins identified by Lyso-TurboID mass spectrometry. The top 959 proteins with ≥10 peptides were analyzed. Orange dots indicate proteins with 1.4-fold peptide enrichment following apilimod treatment. (C) Immunoblot showing the enrichment of LYVAC on lysosomal surface in 293T cells following apilimod treatment. The asterisk indicates the streptavidin band. (D) LYVAC is required for apilimod-induced lysosomal vacuolation. U2OS cells of the indicated genotypes were treated with apilimod, followed by bright-field microscopy to assess vacuole formation. (E) Quantification of average cellular vacuole area in (D). (F) In confocal DIC imaging, apilimod-induced vacuoles exhibited near-complete colocalization with LAMP1-mCherry, a marker of lysosomes. Note, fluorescent lysosomal vacuoles appear slightly smaller than DIC-detected vacuoles. (G) LYVAC is required for lysosomal vacuolation upon PIKfyve deletion. Indicated U2OS cells were infected with PIKfyve CRISPR knockout lentiviruses and imaged seven days post-infection. (H) Quantification of average cellular vacuole area in (G). (I) LYVAC is essential for lysosomal vacuolation induced by FIG4 deletion. Indicated U2OS cells were infected with FIG4 CRISPR knockout lentiviruses and imaged seven days post-infection. (J) Quantification of average cellular vacuole area in (I).

Bar, 10 μm. For all vacuole quantifications, cells were randomly and automatically selected using image analysis software; data are presented as mean ± s.e.m.; n = 300 cells per condition, pooled from three independent experiments. Two-way ANOVA with Tukey’s multiple comparison test for (E); ordinary one-way ANOVA with Tukey’s multiple comparison test for (H) and (J).

The top hit of this proteomic screen was a lipid transfer protein, PDZ-domain containing 8 (PDZD8), which we propose to be renamed as lysosomal vacuolator (LYVAC) (Fig. 1, B and C, Table S1). LYVAC is an ER-anchored protein carrying a synaptotagmin-like mitochondrial-lipid-binding protein (SMP) lipid transfer domain. It was initially proposed as an ER-mitochondria tethering protein (17), but was later found at ER-endolysosome membrane contact sites (MCSs) (18–22). While LYVAC has been linked to processes such as calcium dynamics, endolysosomal maturation, and lysosomal positioning (18–22), key biological questions remain: specifically, what signals trigger LYVAC recruitment, whether and how it transfers lipids at MCSs, and what function this lipid transfer serves.

Because lysosomal vacuolation requires lipids for membrane expansion, we hypothesized that LYVAC contributes to lysosomal vacuolation through bulk lipid transfer from the ER. To test this hypothesis, we generated LYVAC-knockout (LYVAC-KO) cells in two different cell lines (fig. S1, D and E). Apilimod-induced vacuolation was strongly suppressed in all LYVAC-KO cells and fully restored by LYVAC reconstitution (Fig. 1, D and E, fig. S1, F to H). High-resolution differential interference contrast (DIC) imaging showed almost complete overlap between apilimod-induced vacuoles and lysosome-associated membrane protein 1 (LAMP1) (Fig. 1F), confirming the lysosomal origin of the vacuoles as previously reported (6, 12). To mimic the loss of PIKfyve in prion disease (7) or cadmium toxicity (8), we genetically deleted PIKfyve. Loss of PIKfyve triggered lysosomal vacuolation only in wild-type and the rescue cells, but not in LYVAC-KO cells (Fig. 1, G and H). Similarly, deletion of FIG4, a PIKfyve-associated PI(3,5)P₂ phosphatase mutated in Charcot-Marie-Tooth disorder CMT4J (23), induced giant vacuoles that were also completely LYVAC-dependent (Fig. 1, I and J). Thus, LYVAC is essential for lysosomal vacuolation under PIKfyve deficiency.

LYVAC mediates lysosomal osmotic vacuolation

PIKfyve dysfunction causes lysosomal accumulation of chloride and ammonium, driving lysosomal water influx and osmotic vacuolation (12, 24, 25). Consistent with this notion, a water channel blocker phloretin abolished apilimod-induced lysosomal vacuolation (fig. S2A). Besides PIKfyve dysfunction, LYVAC was also required for vacuolation triggered by diverse lysosomal osmotic stress inducers, such as metoclopramide, a weak base compound that accumulates in acidic lysosomes (26) (Fig. 2, A and B, fig. S2, B and C), sucrose supplementation mimicking lysosomal storage disease (Fig. 2, C and D, fig. S2D), and cell exposure to hypotonic media (Fig. 2, E and F). The observed vacuoles were positive for multiple lysosome markers (fig. S2, E to G), but lysosomal vacuolation did not necessarily involve lysosomal pH neutralization or membrane damage (fig. S2, H to K). Although LYVAC is anchored on the ER, generation of ER-derived vacuoles during sorafenib-induced proteotoxic stress (27) was independent of LYVAC (fig. S2, L and M). Thus, LYVAC appears to provide a general mechanism for lysosomal osmotic vacuolation.

Figure 2. LYVAC is a general mediator of lysosomal osmotic vacuolation.

(A) Knockout of LYVAC blocks metoclopramide-induced lysosomal vacuolation. (B) Quantification of average cellular vacuole area in (A). Mean ± sem; n = 300 cells/condition pooled from three independent experiments. (C) Knockout of LYVAC blocks sucrose-induced lysosomal vacuolation. Note that sucrose cannot trigger vacuoles in U2OS cells. (D) Quantification of (C). Mean ± sem; n = 20 random fields pooled from three independent experiments for each condition. (E) Knockout of LYVAC blocks lysosomal vacuolation induced by hypotonic media. (F) Quantification of average cellular vacuole area in (E). Mean ± sem; n = 300 cells/condition pooled from three independent experiments. (G) LYVAC is recruited to LAMP1-positive lysosomes following apilimod treatment. (H) Quantification of LYVAC recruitment in (G). Mean ± sem; n = 62, 52, 55, and 65 random cells, respectively, pooled from three independent experiments. (I) LYVAC is recruited to lysosomes after sucrose treatment. (J) Quantification of sucrose-induced LYVAC recruitment to lysosomes in (I). Mean ± sem. n = 78 and 56 random cells, respectively, pooled from three independent experiments. (K) ATG2A is recruited to lysosomes damaged by LLOME, but not recruited to vacuolating lysosomes following apilimod treatment. (L) Quantification of EGFP-ATG2A puncta in (K). Mean ± sem; n = 35, 43, 42, and 36 random cells, respectively, pooled from three independent experiments. (M) Schematic illustration of ATG2A and LYVAC, two lipid transfer proteins, as specific sensors of different types of lysosomal membrane stress.

Bar, 10 μm. Ordinary one-way ANOVA with Tukey’s multiple comparison test for (B), (F), (H) and (L); two-way ANOVA with Tukey’s multiple comparison test for (D); student’s t-test for (J).

To explore the functional relevance of lysosomal vacuolation, we tested several models related to chemotherapy, antibiotics, or viral infection. Knockout of LYVAC sensitized U2OS osteosarcoma cells to multiple weak-base chemotherapeutics, including doxorubicin, topotecan, and sunitinib (fig. S3A), all known to trigger lysosomal osmotic stress due to their lysosomal sequestration as part of a chemoresistance mechanism (28, 29). LYVAC knockout also sensitized cells to cell death resulting from monensin (fig. S3B), an antibiotic and Na⁺/H⁺ ionophore that triggers lysosomal osmotic stress (30). In hepatitis A virus infection, the 3C viral protease triggers lysosomal vacuolation and cell death (31), both of which were reduced by LYVAC deletion (fig. S3, C to E), implicating 3C-dependent vacuolation as a pathogenic mechanism. Thus, LYVAC mediates lysosomal osmotic vacuolation in response to diverse stimuli of pathophysiological importance.

Lysosomes are prone to osmotic stress due to the high concentration of small solutes generated by macromolecule hydrolysis (32, 33). Consistent with LYVAC being a frequent responder to lysosomal osmotic stress, it was captured by five distinct lysosomal TurboID baits (fig. S3F) (34). Another lysosome-associated lipid transfer protein VPS13C (35) was also detected by lysosomal TurboID (fig. S3F), but it was inaccessible to most ER-anchored baits (fig. S3F), suggesting limited localization of VPS13C to ER-lysosome contacts. Consistent with this idea, knockout of VPS13C did not affect lysosomal vacuolation (fig. S3, G and H).

LYVAC is specifically recruited to osmotically stressed lysosomes

Both immunoblotting and immunofluorescence confirmed the recruitment of endogenous LYVAC to lysosomes in an apilimod-dependent manner (Fig. 1C, Fig. 2, G and H). LYVAC was recruited before vacuole formation (Fig. 2, G and H, fig. S4A), in agreement with its role in lysosomal expansion. Similar lysosomal recruitment of LYVAC was observed in different cell types (fig. S4, B and C) and in response to diverse lysosomal osmotic stress inducers, such as YM201636 (fig. S4D), metoclopramide (fig. S4E), monensin, nigericin (fig. S4, F to H), sucrose (Fig. 2, I and J, fig. S5, A and B), and hypotonic media (fig. S5, C to F). The water channel blocker phloretin abolished LYVAC recruitment (fig. S5, G and H). Thus, LYVAC is a general sensor of lysosomal osmotic stress.

The recruitment of ER-anchored LYVAC to lysosomes suggests inter-organelle membrane contacts. As expected, electron microscopy revealed significantly increased ER membrane contacts with apilimod-induced vacuoles (fig. S5, I and J). Notably, multi-lamellar membrane cores, a signature morphology of late endosome/lysosomes (36), were frequently found in the vacuoles (fig. S5I). Further examination ruled out obvious LYVAC recruitment to other organelles, such as lipid droplets, early endosomes, and mitochondria, in response to either lysosomal osmotic stress or specific stress of the other organelles (fig. S6, A to G). Thus, LYVAC is specifically recruited to osmotically stressed lysosomes.

LYVAC and ATG2 -- an autophagy-related lipid transfer protein -- responded to different types of lysosomal membrane stress. While ATG2 was quickly recruited to damaged lysosomes (15), it was not recruited to lysosomal vacuoles (Fig. 2, K and L). Conversely, damaged lysosomes did not recruit LYVAC (fig. S7, A and B). Blocking lysosomal repair by ATG2A/B double knockout (15) did not affect apilimod-induced LYVAC recruitment or vacuolation (fig. S7, C to G). Knockout of LYVAC, which blocked lysosomal vacuolation, did not impair lysosomal repair (fig. S7, H and I). Thus, while ATG2 detects lysosomal membrane damage, LYVAC senses lysosomal osmotic stress (Fig. 2M).

LYVAC is recruited via multivalent interactions

LYVAC carries six major domains: An N-terminal transmembrane domain (TM) that anchors to the ER membrane, an SMP lipid transfer domain, a PDZ domain, a split C2 domain, a C1 domain, and a coiled-coil (CC) domain (17, 37). AlphaFold Multimer (38–40) predicted LYVAC as a homodimer, with two SMP domains dimerizing in an anti-parallel way forming a tube-like structure (Fig. 3A, fig. S8A), similar to SMP domains in other proteins (41–43). Based on the predicted structure, a twisted, hydrophobic groove ran across the SMP dimer from one tip to the other (fig. S8A); the C2 domain from each protomer had direct, extensive contacts with both SMP domains (Fig. 3A); the PDZ domain was positioned proximally to the C2 domain through a brief linker (Fig. 3A); in contrast, the other domains including the TM, C1, and CC domains were connected to the SMP-C2-PDZ complex through extended, flexible linkers.

Figure 3. Multivalent weak interactions mediate LYVAC recruitment to osmotically stressed lysosomes.

(A) Structures of different LYVAC domains and dimerization predicted by AlphaFold Multimer. Disordered regions (flexible sequences) were removed. (B) Illustration of LYVAC mutants generated to investigate LYVAC recruitment and lysosomal vacuolation. Key conclusions were summarized on the right. (C) Apilimod-induced lysosomal recruitment of LYVAC is dependent on its TM, C1, and CC domains. U2OS LYVAC knockout cells stably expressing indicated LYVAC mutants with a C-terminal Flag-tag were treated with apilimod and then fixed for immunostaining of Flag and endogenous LAMP1. (D) Quantification of the colocalization between LYVAC and LAMP1 in (C). Mean ± sem; n = 89, 76, 74, 58, 52, 30, 81, 61, 71, 41, 39, 21, 58, 66, 69, 55, 50, 48, 56, and 58 random cells from left to right, pooled from three independent experiments. (E) Lysosomal vacuolation requires most LYVAC domains. U2OS LYVAC-KO cells stably expressing the indicated LYVAC mutants were treated with apilimod for 5 h and subjected to bright-field microscopy. (F) Quantification of average cellular vacuole area in (E). Mean ± sem; n = 300 random cells/condition pooled from three independent experiments. (G) TM-C1-CC serves as a probe for lysosomal osmotic stress. U2OS cells stably expressing TM-C1-CC-Flag were treated with 25 nM apilimod, hypotonic media, 37 μM monesin, or 10 μM nigericin for 1 h to induce lysosomal osmotic stress and then fixed for immunostaining of Flag and endogenous LAMP1. (H) Quantification of the recruitment of potential lysosomal osmotic stress probes. Mean ± sem; n = 48, 53, 52, 29, 58, 34, 73, and 38 random cells from left to right, pooled from three independent experiments. (I) Schematic illustration of lysosomal LYVAC recruitment through multivalent interactions involving TM, C1, and CC domains.

Bar, 10 μm. Two-way ANOVA with Tukey’s multiple comparison test.

To investigate how LYVAC senses lysosomal osmotic stress, we individually deleted each domain of LYVAC (Fig. 3B, fig. S8B). Deletion of either SMP or C2 strongly increased both basal and apilimod-stimulated LYVAC recruitment (Fig. 3, C and D, fig. S8C). Loss of the PDZ domain slightly reduced LYVAC recruitment, but deleting any remaining domain including TM, C1, or CC abolished LYVAC recruitment (Fig. 3, C and D). Deletion of TM, C1, or CC caused profound defects of LYVAC in apilimod-induced lysosomal vacuolation (Fig. 3, E and F, fig. S8D). Despite robust lysosomal recruitment, LYVAC-ΔSMP no longer mediated lysosomal vacuolation (Fig. 3, E and F), suggesting a role for lipid transfer in vacuole formation. LYVAC-ΔC2 also failed to mediate lysosomal vacuolation, phenocopying LYVAC-ΔSMP (Fig. 3, E and F). Thus, three loosely connected domains, TM, C1, and CC, are all required for the lysosomal recruitment of LYVAC, potentially allowing flexibility of the lipid transfer domain to support lysosomal vacuolation.

The requirement of three domains for LYVAC recruitment suggests that C1 or CC alone binds weakly to the lysosome. Indeed, when individually expressed, C1 or CC was not recruited to lysosomes (fig. S8E). A fragment containing all three required domains (TM-C1-CC) was sufficient to localize to osmotically stressed lysosomes (Fig. 3, G and H, fig. S8, F and G). Notably, TM-C1-CC and another fragment TM-C2-C1-CC exhibited stress-dependent recruitment (Fig. 3, G and H, fig. S8, F and G) without interfering with lysosomal osmotic vacuolation (fig. S8H, I). Thus, the shortest recruitable fragment of LYVAC, TM-C1-CC, can serve as a probe for lysosomal osmotic stress.

The LYVAC-CC domain directly binds to the small GTPase RAB7 (18, 20–22). RAB7, but not RAB5, was strongly enriched on vacuolating lysosomes (fig. S9, A to F). The LYVAC-C1 domain interacts with negatively charged lipids, particularly phosphatidylserine (PS) (22), but such interaction was likely too weak to drive C1 recruitment on its own (fig. S8E). Replacing C1 with the PH domain of OSBP (OSBP-PH), a PI(4)P-binding motif (44), caused constitutive LYVAC recruitment to RAB7-positive compartments that were profoundly dilated (fig. S9, G and H). Taken together, LYVAC is recruited to osmotically stressed lysosomes via a dual requirement of C1- and CC-mediated weak interactions, stabilized by the TM domain that retains LYVAC at local MCSs (Fig. 3I).

Lysosomal osmotic stress triggers lipid signaling

Inspired by the lipid-binding capacity of the LYVAC-C1 domain (22), we noticed multiple lipid transfer proteins from the oxysterol-binding protein (OSBP) related protein (ORP) family in our lysosomal proteomics (Fig. 4A). Immunoblot confirmed apilimod-induced lysosomal enrichment of five ORPs including ORP1, OSBP, ORP9, ORP10, and ORP11 (Fig. 4B). ORP1 has two splicing variants (45), and only the longer one ORP1L was recruited (fig. S9I) which carries an ANK domain for RAB7 binding (46). Like LYVAC, multiple ORPs accumulated on lysosomes following apilimod treatment before vacuole formation, as visualized by immunofluorescence (Fig. 4, C and D, fig. S9, J to O). ORPs are phosphoinositide effectors typically recruited to PI(4)P-enriched membranes (47–49). Phosphatidylinositol 4-kinase type 2 alpha (PI4K2A), a PI(4)P-producing enzyme, was enriched on lysosomes upon apilimod treatment (fig. S9, P and Q). PI4K2A and its kinase activity strongly promoted apilimod-induced lysosomal recruitment of ORP1L (Fig. 4E, fig. S10A), consistent with PI4K2A being upstream of the ORPs.

Figure 4. Lysosomal osmotic stress activates lipid signaling to promote LYVAC recruitment and vacuole formation.

(A) Mass spectrometry identified the enrichment of ORP family proteins on lysosomes following apilimod treatment. (B) Immunoblot showing apilimod-induced lysosomal enrichment of ORPs. (C) ORP1L is recruited to lysosomes upon apilimod treatment. (D) Quantification of ORP1L recruitment in (C); n = 38, 25, 26, and 24 random cells, respectively. (E) Quantification of lysosomal ORP1L recruitment in U2OS cells from fig. S10A; n = 34, 44, 31, 32, 19, 46, 32, and 29 random cells, respectively. (F) Quantification of apilimod-induced lysosomal recruitment of a PS probe in U2OS cells in fig. S10B; n = 35, 37, 46, 37, 40, 16, 33, 30, and 39 random cells, respectively. (G) Quantification of apilimod-induced lysosomal recruitment of a cholesterol probe in U2OS cells in fig. S10C; n = 32, 20, 14, 15, 13, 19, 23, 26, and 15 random cells respectively. (H) Schematic illustration of lysosomal osmotic stress-induced lipid signaling. (I) Lysosomal recruitment of LYVAC in genetically modified U2OS cells. (J) Quantification of LYVAC/LAMP1 colocalization in (I); n = 25, 25, 53, 59, 53, 38, 52, and 45 random cells, respectively. (K) Knockout of PI4K2A or ORPs suppresses apilimod-induced lysosome vacuolation. (L) Quantification of cellular vacuole area in (K); n = 230 cells/condition. (M-N) Quantification of cellular vacuole area in indicated U2OS cells in fig. S11C–D; n = 300 cells/condition. (O) Schematic illustration of lysosomal lipid signaling upstream of LYVAC. PI4K2A is recruited to osmotically stressed lysosomes where it produces PI(4)P to recruit five ORP family proteins. The ORPs exchange PI(4)P for cholesterol and PS, which promotes LYVAC recruitment and lysosomal vacuolation.

Bar, 10 μm. Mean ± s.e.m; total cell numbers were pooled from three independent experiments. Ordinary one-way ANOVA with Tukey’s multiple comparison test for (D); two-way ANOVA with Tukey’s multiple comparison test for (E), (F), (G), (J), (L), (M), and (N).

In line with established roles of ORPs in exchanging PI(4)P for PS and cholesterol at MCSs (47–49), lysosomal localization of PS (GFP-Lact-C2) and cholesterol (GFP-GRAM-W) probes was increased following apilimod treatment (Fig. 4, F and G, fig. S10, B to D). Among the five recruited ORP proteins, ORP9, ORP10, and ORP11 are PI(4)P/PS transporters (15, 50), whereas OSBP and ORP1L are specific PI(4)P/cholesterol exchangers (49) (Fig. 4H). Accordingly, apilimod-induced lysosomal recruitment of the PS probe was selectively lost in U2OS ORP9/10/11 triple knockout (ORP-TKO) cells (Fig. 4F, fig. S10B), while recruitment of the cholesterol probe was abolished only in OSBP-KO U2OS cells lacking ORP1L (Fig. 4G, fig. S10, C and D). Thus, lysosomal osmotic stress triggers PI4K2A-mediated PI(4)P signaling, thereby recruiting ORPs to ER-lysosome contacts, where they exchange PI(4)P for cholesterol and PS (Fig. 4H). This resembles PI(4)P-driven transfer of cholesterol and PS from the ER to the plasma membrane, Trans-Golgi network, and damaged lysosomes (15, 51, 52).

Lipid signaling mediates LYVAC recruitment and lysosomal vacuolation

We investigated whether lysosomal lipid signaling regulates LYVAC. Knockout of PI4K2A (PI4K2A-KO, fig. S10D) abolished apilimod- or hypotonic media-induced LYVAC recruitment, which was fully rescued by re-expressing wild-type but not kinase-dead PI4K2A in knockout cells (Fig. 4, I and J, fig. S10, E to G). We further tested the impact of ORPs downstream of PI(4)P signaling (Fig. 4H). Apilimod-induced LYVAC recruitment was also strongly reduced in ORP-TKO cells and partially suppressed in OSBP-KO cells (Fig. 4, I and J). Thus, PI(4)P-driven lysosomal PS transfer is important for LYVAC recruitment, which requires the PS-binding C1 domain of LYVAC (Fig. 3, C and D). Although protrudin, an interactor of LYVAC (18, 22), was enriched on osmotically stressed lysosomes (Fig. 4A), knockout of protrudin did not affect LYVAC recruitment or lysosomal vacuolation (fig. S10H). Likely due to competition between LYVAC and GFP-Lact-C2 in PS binding, a faster lysosomal accumulation of the PS probe was found in LYVAC-KO cells (fig. S11, A and B), further implicating PS as a key lipid in lysosomal osmotic stress response.

Consistent with PI4K2A being essential for LYVAC recruitment, PI4K2A-KO abolished apilimod-induced lysosomal vacuolation (Fig. 4, K and L). Reconstitution of wild-type but not kinase-dead PI4K2A rescued lysosomal vacuolation in knockout cells (Fig. 4M, fig. S11C). Downstream of PI4K2A, knockout of OSBP also abolished lysosomal vacuolation (Fig. 4, K and L), which was rescued by the re-constitution of wild-type OSBP but not a mutant lacking PI(4)P/cholesterol exchange activity (15, 49) (Fig. 4N, fig. S11D). Expression of ORP1L restored lysosomal vacuolation in OSBP-KO cells (fig. S11, E to G), suggesting redundant roles of OSBP and ORP1L in PI(4)P/cholesterol exchange. Furthermore, the lysosomal vacuolation activity of OSBP can be replaced by a yeast cholesterol transporter Kes1 (fig. S12, A to C). Thus, ER-to-lysosome cholesterol delivery is essential for lysosomal vacuolation. The ORP-TKO cells, which lack PI(4)P-driven, ER-to-lysosome PS transfer (15), showed a partial reduction in lysosomal vacuolation (Fig. 4, K and L), with the remaining vacuolation likely supported by the basal level of lysosomal PS. Reconstitution of the yeast PS transporter Osh6 in ORP-TKO cells largely rescued the vacuolar size (fig. S12, A to C). Thus, ORP-mediated transfer of PS and cholesterol from the ER to lysosomes is critical for LYVAC-driven vacuolation (Fig. 4O). Consistent with a key role for the PI4K2A-ORP-LYVAC pathway in lysosomal vacuolation, levels of PI4K2A, ORP9, ORP11, and LYVAC on individual lysosomes strongly correlated with vacuole size (fig. S12, D and E).

LYVAC-SMP binds PS/cholesterol-enriched membranes for lipid transfer

Although LYVAC extracts various lipids except cholesterol, no lipid unloading activity is detectable in vitro toward typical neutral acceptor membranes (22). Lysosomal lipid signaling motivated us to explore LYVAC regulation by PS and cholesterol (Fig. 4) on acceptor liposomal membranes. We started with light scattering assays (53) to measure liposomal membrane binding of purified LYVAC-SMP (Fig. 5A, fig. S13A). A relatively weak membrane tethering activity was detected when acceptor liposomes contained 20% PS and 20% cholesterol (Fig. 5B). Tethering was enhanced by increasing liposomal cholesterol levels, reaching a maximum at saturating levels of cholesterol in liposomal acceptors, without causing membrane fusion (Fig. 5B, fig. S13, B and C, Table S2).

Figure 5. Phosphatidylserine and cholesterol on acceptor membranes activate LYVAC-mediated, directional lipid transfer in vitro.

(A) Schematic illustration of the in vitro membrane tethering assay by LYVAC-SMP. (B) When acceptor liposomes contain 20% PS, increased level of cholesterol stimulates higher membrane tethering activity of LYVAC-SMP. (C) Schematic illustration of the in vitro reconstitution of lipid transfer by purified LYVAC-SMP. (D, E) Cholesterol and PS in acceptor liposomes each stimulate LYVAC-SMP–mediated lipid transfer in a dose-dependent manner; Mean ± sem; n = 7 in (D) and n = 6 in (E). (F) The lipid compositions of different acceptor liposomes and related LYVAC activity. PI4P, 5%; PS, 20%, cholesterol: saturating levels. (G, H) Both PS and cholesterol in the acceptor liposome are required for efficient membrane tethering (G) and lipid transfer (H) by LYVAC-SMP; Mean ± sem; n = 4 in (G); n = 3 in (H). (I) Schematic illustration of the in vitro assays to determine the direction of lipid transfer by LYVAC-SMP. (J) Similar membrane tethering by LYVAC-SMP in both the forward and the reverse lipid transfer assays in (I); Mean ± sem; n = 4. (K) LYVAC-SMP preferentially transfers lipids from neutral liposomes, where it is anchored, toward PS/cholesterol-enriched liposomes; Mean ± sem; n = 5 or 6. (L) Illustration of the setup of the molecular dynamics simulation. Note that the forward transfer is more relevant, as the reverse transfer was not favored than forward in vitro, regardless of the lipid substrates. (M) Lipid transfer from the neutral donor to PS/cholesterol-enriched acceptor results in reduced total membrane energy, while the reverse transfer increases it. Each dot represents one simulation. (N) The lipid transfer activity of LYVAC-SMP is increased when the acceptor liposomes have higher luminal osmolarity; Mean ± sem; n = 7. (O) Elevated luminal osmolarity in donor liposomes reduced LYVAC-SMP–mediated lipid transfer; Mean ± sem; n = 6 or 7.

See methods for details of all in vitro assays and MD simulations.

Using the same liposomes in fluorescence resonance energy transfer (FRET)-based lipid transfer assays (Fig. 5C), LYVAC-SMP showed robust lipid transfer activity (Fig. 5D), which positively correlated with membrane tethering activity and cholesterol levels in liposomal acceptors (Fig. 5B). Besides enhancing SMP membrane binding (Fig. 5B), cholesterol in liposomal acceptor membranes may also facilitate net lipid transfer by correcting interleaflet lipid imbalance (54). Additional free cholesterol particles also appeared to accelerate lipid transfer without affecting tethering (fig. S13C), likely by incorporating into the donor membrane. In addition to cholesterol, PS in acceptor liposomes also promoted LYVAC-SMP-mediated lipid transfer in a dose-dependent fashion (Fig. 5E). With acceptor liposomal membranes containing 10–20% of PS and cholesterol, LYVAC-SMP increased both the rate and total amount of lipid transfer in a dose-dependent manner (fig. S13D), demonstrating its sensitivity to physiological lipid levels. When total lipid transfer reached a maximum using liposomal acceptors carrying 20% PS and saturating cholesterol, additional LYVAC-SMP further accelerated the transfer rate without changing the total transfer amount (fig. S13E). Thus, total lipid transfer activity appears to be determined by the membrane composition rather than the LYVAC protein.

Because osmotically stressed lysosomes also upregulate PI(4)P, we incorporated various combinations of PI(4)P, PS and cholesterol into acceptor liposomes (Fig. 5F). Membrane tethering and lipid transfer by LYVAC-SMP were only stimulated when both cholesterol and PS were present in acceptor liposomes (Fig. 5, F to H, acceptor A6). Efficient membrane tethering by SMP required both the donor and the acceptor, with either alone showing much weaker tethering (fig. S13F). The removal of either cholesterol or PS strongly reduced tethering and fully abolished lipid transfer (Fig. 5, F to H, acceptors A2 and A3). Addition of PI(4)P to PS/cholesterol-enriched acceptor liposomes partially reduced membrane tethering and lipid transfer (Fig. 5, F to H, acceptor A7), whereas replacing PS with PI(4)P reduced tethering by ~50% and abrogated lipid transfer (Fig. 5 G and H, Acceptor A5). Thus, LYVAC-SMP directly recognizes PS and cholesterol but not PI(4)P on acceptor membranes, which activates membrane tethering and lipid transfer.

Full-length LYVAC is activated by PS and cholesterol on acceptor membranes

Like LYVAC-SMP, lipid transfer by full-length LYVAC was activated only when both cholesterol and PS were present in acceptor membranes (fig. S13, G to P). However, all mutants of LYVAC lacking individual domains showed normal lipid transfer activity in vitro except LYVAC-ΔSMP (fig. S13G and S14A, Table S2). Indeed, LYVAC-SMP and full-length LYVAC exhibited similar lipid transfer activity (fig. S14B). Compared with SMP alone, full-length LYVAC exhibited much lower membrane tethering activity (fig. S14C), which was not abolished by the deletion of SMP but strongly enhanced upon C2 deletion (fig. S14D). Thus, full-length LYVAC binds to acceptors through domains beyond SMP, where C2 may facilitate dynamic membrane interactions and lipid transfer.

LYVAC mediates directional lipid transfer toward PS/cholesterol-enriched membranes

The role of LYVAC-SMP in lysosomal expansion suggests directional lipid transfer toward PS/cholesterol-enriched lysosomes. To test lipid transfer direction in vitro, we compared the transfer activities using PS/cholesterol-enriched liposomes as either the acceptor (forward) or the donor (reverse). Because Liss Rhod-PE (Rh-PE) in PS/cholesterol-enriched membranes strongly suppressed SMP’s membrane tethering activity (fig. S14E–I), we removed Rh-PE from the FRET system (fig. S14E). Nitrobenzoxadiazole (NBD)-labeled lipids alone (Fig. 5I) are sufficient for lipid transfer assays (53), despite much weaker dequenching signals. The two pairs of liposomes measuring forward and reverse transfer, respectively, were equally tethered by LYVAC-SMP (Fig. 5J). Although reverse transfer of PS is expected to be gradient-driven, NBD-PS substrates with either intact or modified serine head groups (fig. S14J) exhibited consistently weaker reverse than forward transfer (Fig. 5K, fig. S14K), with similar forward rates observed for both substrates (fig. S14L). Thus, in vitro, LYVAC-SMP preferentially transfers lipids from neutral liposomes toward PS/cholesterol-enriched liposomes.

We next explored the mechanism underlying directional lipid transfer by LYVAC-SMP. Given the distinct lipid compositions between donor and acceptor membranes, we hypothesized that a chemical potential gradient could drive directional lipid movement. To assess thermodynamic favorability, we used molecular dynamics (MD) simulations to calculate total membrane energy before and after lipid transfer (Fig. 5L). Consistent with our in vitro assays, the total membrane energy was reduced after moving lipids from the neutral membrane to the PS/cholesterol-enriched membrane (Fig. 5M), supporting an energetically favorable forward lipid transfer. In contrast, the reverse transfer of the same lipids increased total membrane energy (Fig. 5M). Notably, higher levels of PS and cholesterol in the acceptor membrane enhanced the favorability of forward transfer (fig. S14M), consistent with the faster forward transfer observed in vitro (fig. S13D). Thus, PS and cholesterol on acceptor membranes promote both SMP binding and thermodynamically favorable forward lipid movement.

Membrane tension gradients are well-established drivers of lateral lipid movement (55–57), consistent with the fluid mosaic model of cellular membranes (58). When LYVAC-SMP bridges the ER and lysosomes, the higher membrane tension in osmotically stressed lysosomes is expected to pull lipids from the ER. To test this hypothesis in vitro, we induced membrane tension by applying a transmembrane osmotic gradient (59). Placing osmotic tension on acceptor liposomes consistently enhanced LYVAC-SMP-mediated lipid transfer (Fig. 5N), whereas applying the same tension to the donor substantially inhibited transfer (Fig. 5O). These results support osmotic membrane tension as another driving force promoting directional lipid movement to lysosomes. This directional transfer resembles fat-specific protein 27 (FSP27)-mediated lipid flux between lipid droplets, where higher internal pressure in smaller droplets drives the flow of neutral lipids into larger droplets through FSP27 conduits (60).

Directional lipid transfer by LYVAC-SMP suggests a bridge-like mechanism. To distinguish between a bridge or shuttle model, we engineered SMP heterodimers containing one wild-type and one lipid transfer-deficient mutant SMP domain (fig. S15, A to D). A shuttle would yield ~50% reduced transfer, but both heterodimers showed far greater impairment in lipid transfer (fig. S15, E and F), thus supporting a bridge mechanism. Taken together, these findings strongly support the idea that LYVAC drives directional lipid transfer from neutral donor membranes to PS/cholesterol-enriched acceptors.

LYVAC-SMP carries conserved PS- and cholesterol-sensing motifs

We aimed to further define how LYVAC-SMP directly senses lysosomal PS and cholesterol. Binding of the C2 domain to the middle of the SMP dimer precludes membrane access from this site (fig. S16A). Surface charge analysis revealed large positively charged patches extending from the SMP tip to C2 and PDZ domains (fig. S16, B and C). The concentrated charges at the SMP tip corresponded to conserved arginine (R) and lysine (K) residues (Fig. 6A, fig. S16, B to E), which may serve as binding sites for negatively charged PS. Mutations of this putative PS-sensing cluster fully abolished membrane tethering and lipid transfer by LYVAC-SMP in vitro (Fig. 6, A to C, fig. S16F) and abrogated the role of LYVAC in lysosomal vacuolation (Fig. 6D, fig. S16G).

Figure 6. Through PS- and cholesterol-sensing, LYVAC mediates ER-to-lysosome lipid movement and lysosomal vacuolation.

(A) Left: conserved motifs at the SMP tip. Middle: analogy between the FGK/R motifs and established cholesterol-binding CRAC/CARC motifs. Right: Mutations targeting predicted PS-sensing or cholesterol-sensing motifs. (B, C) All mutants of LYVAC-SMP lost activity in membrane tethering (B) and lipid transfer (C) toward PS/Cholesterol-positive acceptor liposomes (acceptor A6); Mean ± sem; n = 4 or 5. (D) Quantification of cellular vacuole area in fig. S16G; Mean ± sem; n = 300 cells/condition. (E) SRS ratiometric imaging of fixed U2OS wild-type (WT) cells reveals increased lysosomal lipid-to-protein ratio following apilimod (25 nM) treatment. Bar, 10 μm. (F) Quantification of the lysosomal lipid-to-protein ratios in WT and LYVAC-KO U2OS cells; Mean ± sem; n = 9 regions of interests (ROIs). (G) Lipid-to-protein ratios in lysosomes and adjacent ER within 1.5 μm of the lysosomal surface. Each dot represents one lysosome or its adjacent ER; lysosome–ER pairs are connected. (H, I) Quantification of the lipid concentration changes on isolated lysosomal (H) and ER (I) membranes in U2OS cells; Mean ± sem; n = 10 random ROIs. (J, K) Quantification of unsaturated lipid concentrations on isolated lysosomal (J) and ER (K) membranes in U2OS cells; Mean ± sem; n = 10 random ROIs. (L) Similarity scores of ER/lysosome lipid spectra before and after apilimod treatment in U2OS cells. Scores were calculated using the average spectra from 8–10 ROIs. (M) The average LAMP1 concentration on individual lysosomes decreases proportionally to the inverse square of lysosome diameter. Each data point represents one lysosome. See also S18C.

Each ROI for SRS ratiometric imaging contained 10–12 cells. Approximately 100 cells per condition, from ~9 ROIs pooled across three independent experiments, were used for quantification. Ordinary one-way ANOVA with Tukey’s multiple comparison test for (D); two-way ANOVA with Tukey’s multiple comparison test for (F) and (H) to (L).

The recognition of both PS and cholesterol by LYVAC-SMP (Fig. 5F) suggests the presence of additional cholesterol-sensing motifs. Cholesterol recognition is best characterized for transmembrane proteins with conserved cholesterol recognition/interaction amino acid consensus sequence (CRAC), L/V-X_1–5-F/Y/W-X_1–5-K/R, or its reverse version (CARC) (61) (Fig. 6A, middle). At the tip of the LYVAC-SMP dimer, we identified two conserved “FGK/R” motifs (Fig. 6A, fig. S16, D and E). They appeared to mimic the CARC motif in recognizing cholesterol through a partial membrane insertion (Fig. 6A). Mutations of the FGK/R motifs fully eliminated membrane tethering and lipid transfer activity of LYVAC-SMP in vitro (Fig. 6, A to C, fig. S16F). Introducing the same or similar mutations to full-length LYVAC blocked lysosomal vacuolation (Fig. 6D, fig. S16G). All the PS- or cholesterol-sensing mutants of LYVAC were normally expressed and recruited to lysosomes (fig. S17A). Thus, LYVAC-SMP directly senses lysosomal cholesterol and PS through conserved motifs at its tip (Fig. 6A), enabling specific lipid transfer to osmotically stressed lysosomes.

LYVAC-dependent ER-to-lysosome lipid movement in cells

To examine lipid movement in cells, we performed label-free stimulated Raman scattering (SRS) imaging (62) to visualize chemical changes of the ER and lysosomes in fixed cells. Lysosomal lipid-to-protein ratios were markedly increased in wild-type but not LYVAC-KO cells following apilimod treatment (Fig. 6, E and F, fig. S17B). Total ER lipid-to-protein ratios were similar in wild-type and LYVAC-KO cells (fig. S17C), likely due to the extended nature of the ER and signal contamination from cytosol. Paired ER-lysosome analysis showed that lysosomal vacuoles with elevated lipid-to-protein ratios were often surrounded by lipid-depleted ER (Fig. 6G). In contrast, such local ER lipid depletion was not found in LYVAC-KO cells (Fig. 6G), consistent with LYVAC-dependent net lipid movement from the adjacent ER to lysosomes.

To minimize interference from cytosol and other organelles, we isolated ER and lysosomal membranes for SRS-hyperspectral imaging (63) to obtain lipid-specific spectra with linear concentration dependence. After apilimod treatment, isolated membranes showed strongly increased total lipid concentration of lysosomes and a concurrent reduction in ER lipids, both of which were abolished by LYVAC-KO (Fig. 6, H and I, fig. S18A). A prominent Raman peak at 3015 cm⁻¹, marking unsaturated lipids, fell in the ER and rose in lysosomes after apilimod treatment, in a LYVAC-dependent manner (Fig. 6, J and K, fig S18A). Importantly, total lysosomal lipids adopted a more ER-like profile following apilimod treatment, in wild-type but not LYVAC-KO cells (Fig. 6L, fig. S18B). In line with lipid transfer as a major mechanism for lysosomal vacuolation, LAMP1 concentrations on individual lysosomes decreased as the lysosome radius increased, demonstrating an inverse square relationship (Fig. 6M, fig S18C). Taken together, these data are consistent with LYVAC-dependent directional, large-scale ER-to-lysosomal lipid transfer.

A model of LYVAC-mediated lysosomal vacuolation

While membrane fusion and fission also regulate lysosomal size (64), our findings reveal LYVAC as an essential contributor to lysosomal vacuolation through lipid transfer. Lysosomal osmotic stress, induced by diverse vacuolating conditions, stimulated PI(4)P-driven ER-to-lysosome transfer of PS and cholesterol, establishing a platform for strict control of LYVAC recruitment and activation (fig. S19). LYVAC recruitment required both the RAB7-binding CC domain and the PS-binding C1 domain, ensuring a specific response to osmotically stressed lysosomes. Lipid transfer by LYVAC depended on the presence of both PS and cholesterol on acceptor membranes, preventing nonspecific lipid delivery to other organelles. Thus, lysosomal osmotic vacuolation is a direct and highly regulated function of LYVAC. Accordingly, we propose renaming PDZD8 as lysosomal vacuolator or LYVAC to reflect its defining function.

We propose a three-step model by which LYVAC-SMP mediates directional lipid transfer. (i) Attachment: at ER-lysosome contacts, LYVAC-SMP binds to PS/cholesterol-enriched lysosomal membranes. (ii) Transfer: directional, large-scale ER-to-lysosome lipid transfer is driven by chemical potential gradients and lysosomal osmotic membrane tension, further facilitated by lipid scramblases. Membrane tension and lipid chemical potential gradients may serve as general mechanisms for large-scale lipid movement by bridge-like lipid transfer proteins. (iii) Detachment: incoming lipids locally dilute PS and cholesterol on the lysosome, weakening SMP’s lysosomal binding and triggering its dissociation, which prevents lipid backflow and enables dynamic cycles of lipid transfer.

DISCUSSION

The ER network, a known lipid source for autophagosome formation and lysosomal repair, is hereby identified as an integral component of the lysosomal osmotic stress response. A key effect of LYVAC-mediated ER-to-lysosome lipid transfer would be reduced lysosomal membrane tension, enhancing lysosomal osmoresilience and preventing rupture. Lipid signaling triggered by lysosomal osmotic stress may also prime cells for rapid lysosomal repair in anticipation of potential membrane damage. LYVAC-dependent lysosomal vacuolation emerges to shape cellular response to a wide range of conditions including chemotherapy and virus infection. LYVAC appears to be neuroprotective under normal conditions (65, 66), but its vacuolation activity in PIKfyve, FIG4 or VAC14 deficiency, or in prion disease, may contribute to neurodegeneration. Our findings thus suggest an essential mechanism for lysosomal vacuolation, which may have broad implications in pathophysiological conditions associated with lysosomal osmotic imbalance.

Materials and Methods

Cell culture, chemicals, and treatments

All cell lines were originally from ATCC. U2OS, 293T, HT1080, BEAS-2B, COS7, and BJ cells were authenticated through short tandem repeat profiling and their profiling data are publicly available from ATCC. The cell lines have different morphologies. The growth rates and morphologies of all cells were constantly monitored to avoid contamination. All cell lines used in this study were regularly tested by PCR to ensure that they were not contaminated with mycoplasma. All cell lines were maintained in Dulbecco’s modified Eagle’s medium (DMEM) with 10% FBS, 1% Penicillin Streptomycin, and 2.5 ug/ml plasmocin, at 37°C with 5% CO₂. Metoclopramide (Cayman Chemical, 23360) and sucrose (Sigma, S0389) were dissolved in H₂O and stored at −20°C. Apilimod (Cayman Chemical, 19094), YM201636 (Cayman Chemical, 13576), phloretin (Cayman Chemical, 14452), doxorubicin (Cayman Chemical, 15007), topotecan (Cayman Chemical, 14129), and sunitinib (Cayman Chemical, 13159) were dissolved in DMSO and stored at −20°C. Monensin (Cayman Chemical, 16488), nigericin (Cayman Chemical, 11437), and LLOME (Sigma, L7393) were dissolved in ethanol and stored at −20°C. The hypotonic media were made through 1:4 dilution of complete growth media in water.

Antibodies

Rabbit anti-LYVAC/PDZD8 (25512–1-AP; immunofluorescence 1:1,000, immunoblot 1:3,000; the LYVAC antibody had relatively high background in immunofluorescence, and multiple secondary antibodies were tested to achieve the best staining efficiency), rabbit anti-OSBP (11096–1-AP, immunoblot 1:1,000, immunofluorescence 1:1,000), rabbit anti-ORP9 (11879–1-AP, immunoblot 1:1,000), and rabbit anti-VPS13C (28676–1-AP, immunoblot 1:1,000) were from Proteintech. Mouse anti-ORP9 (A-7, sc-398961, immunofluorescence 1:1,000), mouse anti-PI4K2A (sc-390026, immunofluorescence 1:1,000), mouse anti-LAMP2 (sc-18822, immunofluorescence 1:1,000), mouse anti- LAMP1 (sc-20011, immunofluorescence 1:1,000) were from Santa Cruz Biotechnology. Rabbit anti-LAMP1 monoclonal antibody (9091, immunofluorescence 1:1,000), rabbit anti-Rab7 (9367, immunofluorescence 1:1000), mouse anti-Rab5 (46449, immunofluorescence 1:200) were from Cell Signaling. Flag (M2, immunofluorescence 1:10,000) was from Sigma. Mouse anti-CD63 (556019, immunofluorescence 1:1,000) was from BD Pharmingen. Rabbit anti-ORP1L (ab131165, immunoblot 1:1,000) was from Abcam. Rabbit anti-ORP10 (A304–885A, immunoblot 1:1,000) and Rabbit anti-ORP11 (A304–580, immunofluorescence 1:1,000) were from Bethyl Laboratories. Alexa-488/594- and Pacific Blue-conjugated secondary antibodies were from ThermoFisher Scientific.

DNA cloning

For stable protein expression through a lentiviral approach, the relevant DNA sequences were cloned into pCDH-CMV-MCS. The peptide linker GSGSGS was used when fusing two sequences together. To introduce point mutations, two DNA fragments were generated by PCR which carried the desired mutations at their overlapped ends. The two fragments were fused together into pCDH vector through infusion technologies. All cDNA sequences were of human origin unless otherwise specified. The cDNAs for ORPL1 (HsCD00877998), PDZD8/LYVAC (HsCD00438812), ORP9 (HsCD00820675), ORP11 (HsCD00438505), PI4K2A (HsCD00618068), and ATG2A (HsCD00863023) were from the DNASU plasmid repository at Arizona State University. Hepatitis A virus (HAV) 3C sequence was cloned from pAG416GALL-HAV_3C, a gift from Alejandro Chavez (Addgene plasmid #203481). EGFP-GRAM-W DNA was a gift from Yasunori Saheki (Addgene plasmid #211701). Lact-C2-GFP DNA was from Sergio Grinstein (Addgene plasmid #22852); OSBP sequence was cloned from a human OSBP ORF cDNA clone (Sino Biologcial, #HG20205-UT). Primer sequences for DNA cloning are summarized in Table S3.

Stable cell line generation

The study was conducted using stable cell lines with genetic knockouts and/or stable ectopic protein expression through lentiviral approach. Lentiviruses carrying specific ORF sequences were generated using pCDH vectors and used to infect recipient cells for stable expression. Puromycin treatment was applied, if necessary, to remove uninfected cells. However, given almost 100% infection rates, selection was often unnecessary. Viral titration was performed to determine the minimum virus titers required to induce target protein expression in over 90% of cells.

For generating knockout cell lines, various CRISPR-Cas9 guide sequences were evaluated for each gene to identify the most effective one that significantly reduced target expression in the CRISPR knockout pools. LentiCRISPR.v2 (Addgene #52961) carrying the selected guide sequence was used for lentivirus packaging (67). Primers containing guide sequences for CRISPR cloning are summarized in Table S3.

Biotinylation and purification of lysosomal surface proteins

293T cells stably expressing Lyso-TurboID were cultured until they reached 80% confluence. Subsequently, the cells were treated with apilimod (25 nM) or DMSO (vehicle control) for 1 hour to induce lysosomal osmotic stress. Following this treatment, 50 μM biotin was added and incubated for 30 minutes to allow for the biotinylation of lysosomal surface proteins. The cells were then washed twice with 10 ml pre-chilled PBS, scraped into 1 ml of cold PBS per dish, and transferred into 1.5-ml tubes.

The suspended cells were passed through a 25 G needle ten times and centrifuged at 1,000g for 3 minutes to remove intact cells and nuclei. The supernatant (S1) was further centrifuged at 20,000g for 20 minutes to precipitate most membranes, including lysosomes with biotinylated proteins (P20). P20 was resuspended in 500 ul 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 membranes. The samples were then centrifuged at 15,000g for 5 minutes to remove any undissolved aggregates. The resulting supernatants were transferred to a new tube containing 100 μl of well-resuspended streptavidin magnetic beads and rotated at 4°C for 2 hours.

The beads were washed twice with each of the following three buffers sequentially: 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), and buffer 3 (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.5% Triton X-100). Proteins bound to the beads were eluted by adding 50 μl of 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 at 95°C for 5 minutes.

Immunofluorescence

Cells were seeded onto glass coverslips (Warner Instruments, 64–0712) in 24-well plates and treated 24–48 hours later. Following treatment, cells were fixed with 4% paraformaldehyde (PFA) in PBS for 30 minutes at room temperature and then permeabilized with 0.1% Triton X-100 for 2 minutes. Subsequently, they were blocked with 1× blocking buffer (ThermoFisher Scientific, 37565) for 1 hour at room temperature. Both primary and secondary antibodies were diluted in the same blocking buffer. The cells were then incubated with the primary antibody overnight at 4°C. After three washes with PBS, the cells were exposed to secondary antibodies for 30 to 60 minutes at room temperature, followed by three additional washes with PBS. A brief 2-mininute DAPI staining was then performed as needed. Finally, coverslips were mounted onto slides using VECTASHIELD HardSet Antifade Mounting Medium (Vector Laboratories, H-1700) and left to settle for more than 1 hour at room temperature before imaging. The slides were stored at −20°C.

Image capture was conducted using a Leica SP8 LIGHTNING confocal system, with the built-in Leica Application Suite X 3.5.5.19976 software. To minimize nonspecific staining and crosstalk between channels, various negative controls were included. An Okolab stage-top incubator was employed for live-cell imaging on the same confocal system, ensuring consistent temperature and CO2 concentration. The same software settings and Adobe Photoshop processing were applied to all images within the same figure panel. Images were automatically analyzed by custom software codes in a blinded fashion.

Immunoblotting

Cells were washed with cold PBS and lysed in a buffer containing 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 0.5% Triton X-100, 2 mM NaF, 5 mM MgCl₂, and a protease inhibitor cocktail. Lysates were briefly sonicated and centrifuged at 15,000 × g for 10 minutes. Supernatants were collected and mixed with an equal volume of 2× SDS loading buffer (0.1 M Tris-HCl, pH 6.8, 4% SDS, 20% glycerol, 2% 2-mercaptoethanol, 0.01% bromophenol blue), then heated at 95 °C for 10 minutes. Equal amounts of total protein were separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) on Mini-PROTEAN TGX Precast Gels (4–20%, Bio-Rad, #4561091) and transferred to 0.22 μm TransBlot Turbo nitrocellulose membranes. Membranes were blocked with StartingBlock Blocking Buffer (ThermoFisher, #37542) for 1 hour at room temperature, then incubated overnight at 4 °C with primary antibodies diluted in the blocking buffer. After washing, membranes were incubated for 1 hour at room temperature with the appropriate HRP-conjugated secondary antibodies. Signals were developed using Immobilon Forte (Millipore, WBLUF0100) chemiluminescent substrates and visualized using the Invitrogen iBright FL1000 Imaging System (Thermo Scientific).

Protein purification

LYVAC and its mutants were purified by two-step purifications using 6×His tag and twin-strep-tag. An N-terminal Flag-tag and a C-terminal twin-strep tag were added to LYVAC and all its mutants. 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. The coding sequence for the protein was cloned into pCDNA3.0 and transiently transfected into 293T cells. Forty-eight hours after transfection, 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 dialyzed 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 dialyzed with 20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1 mM DTT, aliquoted, and stored at −80 °C until activity tests. Note that purified LYVAC proteins should be used within a few months, as the membrane tethering and lipid transfer activity of both LYVAC-SMP and full-length LYVAC gradually decrease over time and are mostly gone after 6 to 12 months of storage at −80 °C.

Liposome preparation

Liposomes were prepared using the following lipids from Avanti polar lipids: 18:1–12:0 NBD–PS (810195), 18:1 NBD-PS (810198), 18:1 NBD–phosphatidylethanolamine (NBD-PE, 810145), 18:1 lissamine–rhodamine–PE (Rh-PE, 810150), DOPC (850375), DOPE (850725), DOPS (840035), DGS–NTA (790404), brain PI(4)P (840045), cholesterol (700100). Lipids, dissolved in chloroform, were mixed in glass tubes and dried into thin films under a stream of nitrogen. The films were then left to desiccate overnight, ensuring complete removal of residual solvent. To rehydrate the dried lipids, 1 ml of rehydration buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, and 1 mM DTT) was added, resulting in a final total lipid concentration of 0.5 mM. The rehydrated lipid mixtures were vigorously shaken for 30 minutes at room temperature. Subsequently, they underwent10 freeze-thaw cycles, alternating between liquid nitrogen freezing and a 37 °C water bath thawing process. After these cycles, the lipid solutions were extruded through a 100-nm filter 21 times. All liposomes were used within the same day.

The donor liposomes contained 25% DOPE, 2% 18:1 NBD-PE or NB-PS, and 2% Rh-phosphatidylethanolamine (Rh-PE), and the remaining lipids were DOPC. DGS-NTA (Ni) was added as 5% when needed to anchor LYVAC to the liposome through an N-terminal Hisx6-tag. The acceptor liposomes contained various lipids as indicated, 25% DOPE, and the remaining various percentages were added as DOPC. PI(4)P was added as 5% when needed, whereas DOPS and cholesterol were added at different concentrations as indicated.

When comparing the impacts of PI(4)P, DOPS, and cholesterol on membrane tethering and lipid transfer in Fig. 5, 5% PI(4)P, 20% DOPS, and saturating cholesterol were used in different combinations as indicated. Saturating cholesterol was chosen to stimulate robust-yet-regulated membrane tethering and lipid transfer for easier comparison of the impacts from different lipids. Saturating cholesterol did not cause membrane fusion or non-specific binding as the observed tethering was still dependent on the co-presence of PS and cholesterol on acceptor liposomes and the PS-/cholesterol-sensing motifs at the tip of the LYVAC-SMP dimer. To generate cholesterol-saturated liposomes, the amounts of different lipids in the liposome were calculated based on a cholesterol concentration of 40% then extra cholesterol was added to make a final cholesterol concentration of 70%. After rehydration and freeze-thaw cycles, undissolved cholesterol particles larger than 100 nm were removed during liposome extrusion through the 100-nm filter. Note that cholesterol particles smaller than 100 nm may still be present in the solution of cholesterol-saturated liposomes, which may help alleviate lipid sparsity in the outer leaflet of the donor membrane caused by directional lipid transfer.

Free cholesterol nanoparticles were only used in fig. S13C. To generate free cholesterol particles, dried cholesterol was sonicated in the rehydration buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, and 1 mM DTT). A total of 1 mM cholesterol was fully dissolved by sonication, followed by extrusion through a 100 nm filter 21 times to generate nanoparticles. For lipid transfer assays, a final concentration of 50 μM free cholesterol was added during liposome mixing. For membrane tethering assays, as both donor and acceptor liposomes were used at a 7-times higher concentration, free cholesterol was also added 7-times more to a total concentration of 350 μM.

When comparing the lipid transfer activities in the forward and reverse directions, LYVAC-SMP domain instead of the full-length protein was used. This was because SMP not only resembles the full-length protein in PS/cholesterol-stimulated lipid transfer, but it also exhibits much stronger membrane-tethering, thereby facilitating mechanistic investigation. The presence of Rh-PE on the PS/cholesterol-positive membranes was found to block SMP binding to this membrane, likely due to the large headgroup of Rh-PE. When Rh-PE was on the neutral membrane with DGS-NTA(Ni), SMP binding to this membrane was not affected by Rh-PE as it can bind to DGS-NTA(Ni) through its His-tag. To enable fair comparison of the forward and reverse lipid transfer activities with similar membrane tethering in the two scenarios, we removed Rh-PE from the system, as NBD-lipids alone were sufficient to self-quench on the donor liposomes and can be de-quenched upon lipid transfer. When making these liposomes, 2% Rh-PE was replaced with DOPC. A saturating level of cholesterol was necessary for these assays because without Rh-PE, the dequenching signal was at least 10 times weaker. Such a weak increase in NBD signal can only be detected above the background signal when cholesterol-saturated acceptors were used to maximize lipid transfer.

When testing the impact of osmotic membrane tension on lipid transfer, isotonic liposomes were generated as described above. To generate hypertonic liposomes, dried lipids were rehydrated in a buffer containing 20 mM Tris-HCl, pH 8.0, 500 mM NaCl, and 1 mM DTT. The liposomes were finally diluted into the assay buffer containing 200 mM NaCl and gently mixed by tapping. LYVAC-SMP was added immediately, followed by NBD signal monitoring. The acceptors in these assays contained 20% PS and 20% cholesterol. When saturating levels of cholesterol were used in the acceptor, additional hyperosmolarity did not further increase lipid transfer.

In vitro reconstitution of lipid transport

A 6xHis-tag was added to the N-terminal end of LYVAC replacing the transmembrane domain. After mixing freshly made donor and acceptor liposomes (25 μM each, final total lipid concentration) in the assay buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, and 1 mM DTT), LYVAC or its mutant (40–100 nM as indicated) was added immediately before reading the NBD fluorescence (excitation: 460 nm, emission: 535 nm). The lipid transfer was recorded by fluorescence resonance energy transfer (FRET). The NBD fluorescence signal was initially quenched by rhodamine–PE (Rh-PE) due to their proximity in the donor liposome. Upon LYVAC-mediated lipid transfer, NBD-PS/PE and Rh-PE together with other lipids were transported to the acceptor liposomes where they were diluted and spaced away due to relatively low concentrations of them on the new membrane. Increased distance between NBD and Rh on the acceptor liposomes causes de-quenching of NBD.

In vitro membrane tethering

Membrane tethering tests were performed in 96-well transparent plates as previously described using light scattering assays (53). As indicated in specific figures, the tethering of neutral, ER-mimicking donor liposomes and PS/cholesterol-enriched, lysosome-mimicking acceptor liposomes were measured upon the addition of purified LYVAC variants which carry an N-terminal 6xHis-tag to anchor the protein to donor membranes. Briefly, 175 μM (total lipids) donor liposomes and 175 μM acceptor liposomes were mixed in a 90 ul reaction buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, and 1 mM DTT). Then, 500 nM purified proteins were added and the OD405 was monitored both before and immediately after protein addition. When indicated, 10 μl of a protease cocktail (1 M imidazole, 0.2 ug/ul proteinase K) was added to the reaction to remove membrane tethering by LYVAC-SMP, and followed by continuous OD405 monitoring for another 10 minutes.

Transmission electron microscopy sample preparation

Cells were treated with DMSO or apilimod as indicated, rinsed with PBS, and fixed in 2.5% glutaraldehyde in PBS for 1 hour at room temperature. Following fixation, cells were washed three times with PBS (10 minutes each), then post-fixed at 4 °C for 1 hour in 1% osmium tetroxide (OsO₄) containing 1% potassium ferricyanide. After post-fixation, cells were washed three times in PBS (10 minutes each), then dehydrated through a graded ethanol series: 30%, 50%, 70%, and 90% ethanol for 10 minutes each, followed by three changes of 100% ethanol for 15 minutes each. Samples were then infiltrated with Epon resin, replacing the resin with fresh Epon three times (1 hour per change) to ensure complete infiltration. For embedding, the final Epon resin was removed, and BEEM capsules filled with fresh Epon were inverted and placed directly over the monolayer regions of interest. The resin was polymerized overnight at 37 °C, followed by 48 hours at 60 °C to fully cure. After polymerization, the BEEM capsules were carefully popped off, lifting the embedded monolayer from the petri dish. The resulting resin blocks were trimmed and sectioned for transmission electron microscopy.

Cell death assays

Cell death assays for chemotherapy drugs and monensin were performed in 96 well plates. Cells were well suspended before seeded into 96-well plates to ensure even distribution, and their confluency was monitored before treatment. Chemicals were prediluted in fresh media to replace the original media in plates. To quantify cell viability, cells were fixed and briefly permeabilized, followed by DAPI staining and nuclear imaging under 4x objective.

Cell death assays for HAV-3C were performed in 6-well plates. Wild type and LYVAC-KO U2OS cells were infected with lentiviruses to express HAV-3C. Cells were analyzed 3 days after infection for vacuolation and cell death. Cell confluency was monitored to ensure a fair comparison. After bright-field images were taken for vacuolation analysis, cells were fixed and stained with DAPI for cell counting.

For cell number quantification in each condition, 20 to 40 total images from three experiments were pooled, with around 3000 cells quantified within each image. Dead cells were removed during fixation and cell washing. Dying or dead cells still attached to the bottom of the plates showed abnormally brighter and smaller nuclei, which were not counted by our software as live cells. Cell viability was calculated by dividing the number of cells in treated groups with the number of cells in control groups.

Molecular dynamics simulation and energy calculation

As simulating the whole process of LYVAC transferring the lipids to acquire the free energy from umbrella sampling or weighted histogram analysis method (WHAM) is a technical challenge in the field, we focused on the potential energy of the membrane system before and after lipid transfer and calculated whether such transfer led to a lower potential energy in the end state. Although net lipid movement between the two membranes should cause interleaflet lipid imbalance in both the donor and the acceptor (lipid sparsity in the outer leaflet of the donor due to lipid extraction and lipid crowding in the acceptor outer leaflet due to lipid influx), such imbalance should be effectively resolved in cells by the presence of cholesterol, flippases, and scramblases. Considering that in silico simulations are significantly limited by the time complexity of simulating a large membrane and the inherent resolution of the Martini model, we assumed that all lipids were redistributed evenly between the two leaflets of the membrane after the transfer.

Molecular dynamics (MD) inputs were generated using CHARMM-GUI Martini 3.0.0 models (68–70), and simulations were performed with GROMACS. In addition to the lipid bilayers, 22.5 Å of water was added above and below the bilayer. Sodium (Na⁺) and chloride (Cl⁻) ions were added to achieve a final salt concentration of 200 mM. The simulation temperature was maintained at 303.15 K. When generating donor and acceptor, the total number of lipids (donor + acceptor) was kept the same before and after lipid transfer. The total number of lipids was set to 2000 in each lipid layer, with the initial number of DOPE set to 500, DOPS and cholesterol set to the initial concentrations as indicated in the acceptors (0–20%), and the remaining lipids were filled with DOPC. 1% lipid transfer indicates the movement of 5 DOPE and 15 DOPC; the number of lipids transferred in each scenario was calculated accordingly for the other percentages. The x/y size of lipid bilayers was determined automatically by CHARMM-GUI. To minimize the potential effect of a slight rectangular shape differences before and after transfer on energy calculation and to offset potential randomness from different initializations, a series of lipid transfer percentages were simulated (1%, 3%, 5%, 10%, 20%, 30% from the donor to the acceptor and the reverse transfer) and a linear regression was performed in Python to track the lipid energy.

The equilibration process was performed as guided by the CHARMM-GUI. A 10-fold longer run was applied to confirm the equilibration state of the system. Then, a Van der Waals interactions cutoff of 2.0 nm and a Coulomb cutoff of 2.0 nm were used for the production run. The short-range Coulombic (electrostatic) interaction energy and the Lennard-Jones (van der Waals) interaction between the membrane molecules were recorded. Because no perturbation of the system was performed, we found that the Coulombic interaction potential among lipids, Lennard-Jones interaction potential among lipids, and the total potential energy of the whole system became stable within 2.5 ns of simulation during multiple 50 ns simulations. Therefore, a 5 ns simulation (250000 steps) was performed, and energy terms from 4 ns to 5 ns were used to calculate the average energy. The total energy of the donor and the acceptor was summed up to calculate the final energy terms.

As molecular-dynamics trajectories can vary with the random seeds used for CHARMM-GUI, equilibration/production runs, and parallel domain decomposition under different machines, we sought to exclude the possibility that our findings arise from such stochastic differences. Therefore, we intentionally ran one trajectory for each condition and tested the trend across multiple percentages of lipid transfer. The s.e.m. values were shown as bar errors. We limited our cholesterol concentration to a maximum of 20% for a known lipid simulation defect in Martini’s simulation of cholesterol, which causes random errors in equilibration (71). All input files and the scripts necessary to reproduce these simulations have been deposited in a public repository.

Sample preparation for SRS Imaging of fixed cells

U2OS cells stably expressing ER-GFP (lysozyme-GFP-KDEL) (72) and LAMP1-mCherry were seeded onto glass coverslips (Warner Instruments, Cat. No. 64–0712) in 24-well plates 48 hours prior to the indicated treatments. After treatment, cells were fixed with 4% paraformaldehyde (PFA) for 30 minutes at room temperature. Fixed cells were then mounted onto microscope slides using PBS and spacers (ThermoFisher Scientific, Cat. No. S24737) to preserve three-dimensional structure for imaging.

Sample preparation for SRS hyperspectral scanning of isolated membranes

U2OS cells expressing VAPA-GFP and LAMP1-mCherry were cultured in 10-cm dishes. After 48 hours, cells were subjected to specified treatments, then washed twice with 10 ml of ice-cold PBS. Cells were subsequently scraped into 1 ml cold PBS per dish and transferred to 1.5-ml microcentrifuge tubes. Pellets were obtained by centrifugation at 1,000 × g for 3 minutes at 4°C.

Pelleted cells were resuspended in 500 μl hypotonic buffer (10 mM HEPES, pH 7.5, 10 mM NaCl) and incubated on ice for 5 minutes. The suspension was then passed through a 22 G needle twice to physically break the cells, as confirmed by light microscopy. Teh suspension was then centrifuged at 1,000 × g for 3 minutes to remove intact cells and nuclei (P1). The resulting supernatant (S1) was centrifuged again at 20,000 × g for 20 minutes at 4°C to pellet membrane-enriched fractions (P20). The P20 fraction was resuspended in cold PBS and shipped overnight using cold chain packaging for further analysis. During imaging, the fluorescence signals of VAPA-GFP and LAMP1-mCherry were used to reliably identify the ER and lysosomal membranes.

SRS microscopy

SRS imaging is compatible with ER and lysosome fluorescence markers and enables quantitative analysis of endogenous lipids and proteins within individual lysosomes and the ER network. SRS imaging was used to collect protein signals at 2932 cm⁻¹ (corresponding to CH₃ vibrational modes) and lipid signals at 2850 cm⁻¹ (CH₂ vibrational modes). Lipid-to-protein ratiometric images were then generated by calculating the intensity ratio of 2850/2932 (CH₂/CH₃). Guided by ER-GFP and LAMP1-mCherry fluorescence markers, these ratiometric maps enabled precise localization and pixel-wise quantification of lipid-to-protein ratios within individual lysosomes and ER compartments.

The SRS images were collected from an upright laser-scanning microscope (DIY, multiphoton, Olympus), which was equipped with a 25x water objective (XLPLN, WMP2, 1.05 NA, Olympus). The synchronized pulsed pump beam (tunable 720–990 nm wavelength, 5–6 ps pulse width, and 80 MHz repetition rate) and stokes beam (wavelength at 1031nm, 6 ps pulse width, and 80 MHz repetition rate) from a picoEmerald system (Applied Physics & Electronics) were coupled and introduced into the microscope. Upon interacting with the samples, the transmitted signal from the pump and Stokes beams was collected by a high NA oil condenser (1.4 NA). A shortpass filter (950 nm, Thorlabs) was used to completely block the Stokes beam and transmit the pump beam only onto a Si photodiode array for detecting the stimulated Raman loss signal. The current output from the photodiode array was terminated, filtered, and demodulated by a lock-in amplifier at 20 MHz. The demodulated signal was fed into the FV3000 software module FV-OSR (Olympus) to form images using laser scanning. All SRS images of the fixed cells were obtained in a single frame of 512 x 512 pixels, at a dwell time 80 μs in a constant environmental condition. The raw images were output and subjected to minor contrast adjustment in Image J. All the ratiometric images were generated and ratio values were quantified in Image J manually or assisted by plugins including particleFinder and EzColocalization.

SRS hyperspectral imaging and analysis

SRS-hyperspectral imaging (63) was performed with the same setup described above. The hyperspectral imaging stacks of lysosome and ER were taken with 75 spectral steps at 40 μs pixel dwell time, covering the whole CH stretching region from 2700 to 3150 cm⁻¹. The intensity profiles of the hyperspectral stacks from the regions of interest were plotted in ImageJ and further processed by baseline correction, smoothing normalization in OriginPro 2017. The peak intensity of Raman spectra and the intensity ratios between peaks were quantified in OriginPro 2017.

Given that SRS-HSI provides high-fidelity Raman spectra with linear concentration dependence and a background-free signal, the lipid-specific component can be accurately obtained by subtracting the post-methanol wash (protein-only) spectrum from the pre-wash total signal. This subtraction yields a pure lipid spectrum, which was used for quantitative analysis. For the methanol wash experiment, the overall Raman spectra containing both protein and lipid signals, were first acquired from isolated ER and lysosome samples, respectively. This was followed by an in-situ methanol washing procedure consisting of five consecutive washes, each lasting 5 minutes. Throughout the entire process, the sample remained mounted on the microscope stage to maintain consistent imaging conditions and spatial registration. Pure protein Raman spectra were collected after the wash and then the lipid-specific components were obtained by subtracting the post-wash (protein-only) spectrum from the pre-wash total signal on the same organelle.

Raman spectra similarity was scored by a Penalized Reference Matching algorithm with Stimulated Raman Scattering (PRM-SRS) microscopy (73). All spectra were preprocessed and normalized in intensity to a range of 0 to 1. To ensure dimensional consistency for inner product calculations, spectra were interpolated at every integer wavenumber using the interp1 function. In cases where the reference spectrum exhibited a spectral shift beyond the range of the analyte spectrum, it was padded with zeros at the leading edge and trimmed at the trailing edge to maintain alignment. In this study, the ER spectra were considered as the reference.

Software

Confocal images were taken using a Leica SP8 LIGHTNING confocal system with the built-in software Leica Application Suite X 3.5.5.19976. Images were equally processed and assembled in Adobe Photoshop 21.0.2.

The colocalization images taken under Confocal Microscopy were quantified using a Python program. For the quantification of fluorescence images, the outlines of randomly selected cells were manually annotated in each image by an investigator blinded to the allocation. When quantifying the fraction of protein A positive for protein B recruitment (as puncta), a threshold was applied to minimize diffuse protein B signals. Thresholds for both proteins were determined by the target protein’s signal intensity percentile within a single cell, adjusted by a small constant. All cells from the same experiments were applied using the same threshold settings. Manual checking ensured accurate threshold applications.

The bright-field images were taken under an Echo Microscope. Then, the average cellular vacuole areas were quantified using a program written in Python and C++. First, cells in the images were detected by Cellpose (74), a deep-learning-based method. No data were excluded except for the following condition where exclusion was unavoidable: when detecting cells in average total vacuole area quantification, a cell size threshold is set to filter out detected cells that are too small in size due to Cellpose’s detection noise. To detect the vacuoles in each cell, the contrasts of the image were enhanced and equalized, the bright contours were detected, and vacuoles were filtered out from noisy background by the circularity of contours. The circularity was defined as the ratio of the contour area to the square of the contour perimeter. Then, the area of all vacuoles inside each cell was divided by the total area of this cell. For experiments with large amounts of data, a fixed number of cells were randomly sampled by software. The random seeds were fixed to enhance reproducibility. All intermediate results including the cell detection and the vacuole detection were confirmed manually for each experiment.

AlphaFold’s Multimer version (38–40) was used to predict the LYVAC dimer structure. The structure was obtained from an online platform supported with AlphaFold (ColabFold). The predicted top model #1 was visualized using PyMOL 2.5.7. Sequence alignment was performed using PROMALS3D multiple sequence and structure alignment (75).

After acquisition of the data, the comparisons of data in all groups were visualized in GraphPad Prism and the p values were calculated in Prism. The entire code and the environment setting to reproduce the code is released at GitHub and Zenodo. See Data and materials availability.

Statistics and reproducibility

All experiments were independently reproduced at least three times except the Lyso-TurboID mass spectrometry which was performed once. Strict standards were applied to screen for robust and unbiased results. No statistical methods were used to predetermine the sample size. The investigators were blinded to allocation during imaging and data analysis. Statistical significance was determined by unpaired, two-tailed t-test with the Analysis ToolPak in Microsoft Office Excel 2016 or one/two-way ANOVA with Tukey’s multiple comparison test in GraphPad Prism 8.0.2. Data were presented as Mean ± SEM, unless otherwise indicated; N numbers were defined in figure legends; whenever applicable, the total number of cells quantified were pooled from three independent experiments.

Supplementary Material

Figs. S1 to S19; MDAR Reproducibility Checklist; Table S1–S3.

Data and materials availability:

The mass spectrometry data have been deposited to the ProteomeX-change Consortium via the PRIDE partner repository under the data-set identifier PXD047404. All other data are provided within the paper and its Supplementary Information. The customized code for image quantification is deposited to an open-source repository at: https://github.com/jaytanlab/LYVAC and is archived in https://doi.org/10.5281/zenodo.15705618 (76). Source data and the AlphaFold predicted LYVAC/PDZD8 structures are deposited at Mendeley https://doi.org/10.17632/rhwnwm8s38.1 (77). Inquiries for materials in this manuscript can be directed to the corresponding author.