Endoplasmic reticulum junctions serve as a platform for endosome-lysosome interactions through their stop-and-go motion switching

Abstract

Endosomes and lysosomes (collectively termed “endolysosomes”) traverse the cytoplasm in a stop-and-go manner, but the mechanisms underlying this motion remain poorly understood. Using deep learning–based image analyses, including particle tracking, spatial distribution, and endoplasmic reticulum (ER) morphology analysis, we found that ER junctions facilitate stop-and-go motion switching and serve as platforms for endolysosome interactions. Within the ER network, endolysosomes exhibit three dynamic states: fast movement, local slow movement, and pausing. Pauses occur mainly at ER junctions, where transient endosome-lysosome interactions often coincide with organelle fission and are followed by departure. Disruption of ER junctions impairs lysosomal motility and maturation. We further show that actin condensation around endolysosomes mediates motion switching, involving VAP-STARD3 interaction and the actin regulator YWHAH. Other organelles, such as lipid droplets and peroxisomes, also pause near ER junctions. These findings highlight ER junctions as regulatory hubs that orchestrate organelle dynamics, contributing to the spatial coordination of organelle distribution and interactions within the cytoplasm.

ER junctions coordinate endosome-lysosome interactions by regulating their stop-and-go motion for precise cargo exchange.

INTRODUCTION

Endolysosomes (referred to hereinafter as “endolysosomes” or “ELs”) undergo dynamic movements to explore the entire cytoplasmic space in an organized and regulated manner. Regulated movement is crucial for endolysosomes’ participation in various cellular processes, including cholesterol homeostasis, apoptosis, metabolic signaling, exosome release, and plasma membrane (PM) repair (1, 2). For precise delivery, receiving, degrading, and recycling of biological macromolecules, endolysosomes must traffic into close spatial proximity with cargo-containing vesicles (3), completing the signal or material exchange through subsequent membrane fusion, fission, or contact (4). Dysregulation of the mechanisms underlying endolysosomal dynamics and positioning may be responsible for their dysfunctions associated with various diseases, including lysosomal storage diseases (LSDs), cancer, and neurodegenerative diseases (5–7).

Various mechanisms regulate movement and distribution of the endolysosomal system. Classic bidirectional movement of endolysosomes is mediated by kinesin-driven anterograde and dynein-driven retrograde transport, respectively (8, 9). Additionally, actin polymerized by the Arp2/3 complex contributes to the movement and positioning of endolysosomes, influencing endocytic processes such as endosomal fusion, fission, and cargo sorting (10–12). In a third mechanism, the endoplasmic reticulum (ER) regulates the motility and distribution of endolysosomes through a complex system of tethering proteins at membrane contact sites (MCSs) (3, 6). For example, ER-anchored vesicle-associated membrane protein (VAMP)–associated protein A (VAPA) and B (VAPB) interact with the cholesterol-sensing protein oxysterol-binding protein-related protein 1L (ORP1L) and the cholesterol transport protein StAR-related lipid transfer domain protein 3 (STARD3) located on endolysosomes, thereby modulating microtubule-dependent endolysosomal movement in response to changes in cholesterol concentration (13–15). However, endolysosomal movement is not continuous but occurs in a stop-and-go manner (16). It is unknown how the multiple motility regulation mechanisms are spatiotemporally orchestrated to regulate the switching of endolysosomes between their static and motile states, or between their anterograde and retrograde movement.

The ER network is composed of interconnected tubules and sheets, which are linked by three-way junction (17). Structurally distinct ER domains show the preference for interacting with organelles. ER tubules form contacts with endolysosomes, facilitating their motility and the tubular split (18, 19). The tubules and three-way junctions, predominantly regulated by proteins such as Reticulon (RTN), Receptor Expression-Enhancing Protein (REEP) families, and Atlastin (18, 20), are the major sites for lipid synthesis and Ca²⁺ exchange (21–23). Disorganized ER network leads to enlarged and less active mature lysosomes in neurons (24), and mutations in genes encoding ER shaping proteins have been implicated in the pathology of neurodegenerative diseases, such as hereditary spastic paraplegia (HSP) (25–27), in which abnormal endolysosome structures have been observed (28). All these data highlight the critical role of ER morphology in organelle communication, while the specific roles of the ER’s local shape and the global integrity of the ER network remain unclear.

Here, we combine deep learning–based computational image analysis with live-cell imaging to quantitatively study how the ER network regulates the motility, interaction, and function of endolysosomes. We found that endolysosomes move along ER tubules and often pause and become confined near ER junctions, where the organelles start to move again, referred to as the stop-and-go motion switching. The stop-and-go motion of endolysosomes is mediated by local condensation of the ER network and the actin cytoskeleton, in which the VAP-STARD3-YWHAH pathway plays a key role. The motion attributes of endolysosomes modulated by ER morphology in turn regulate the frequency of their interactions, which are required for cargo sorting between endolysosomes. Together, these results reveal that ER mediates the spatial regulation of endosome-lysosome interactions via its unique network morphology. In addition to lysosomes and endosomes, organelles such as lipid droplets and peroxisomes were also found to pause at ER junctions. This suggests a general ER-mediated mechanism for spatial regulation of organelle stop-and-go movement and interaction.

RESULTS

Endolysosomes undergo stop-and-go motion switching at ER junctions

To visualize endolysosomal dynamics relative to the ER, we collected time-lapse images of COS-7 cells expressing markers for the ER [green fluorescent protein (GFP)–Sec61γ] with markers for early endosomes [EE, labeled with blue fluorescent protein (BFP)–Rab5], late endosomes (LE, labeled with mCherry-Rab7), or lysosomes (Ly, labeled with LAMP1-mCherry). We then analyzed endolysosomal motility patterns. Live-cell imaging revealed that endolysosomes underwent directional movement along ER tubules and often paused and became confined near ER junctions (Fig. 1, A to C, and movie S1). Maximum intensity projection analysis of endolysosomal trajectories revealed that ~90% of endolysosomes paused and became confined at ER junctions (Fig. 1D and fig. S1A). Live-cell imaging further demonstrated that ER junctions were consistently the sites for the reinitiation of endolysosomal directional movement (Fig. 1A and fig. S1B). Thus, the stop-and-go motion switching of endolysosomes takes place at ER junctions. Given the tendency of both endosomes and lysosomes to accumulate at ER junctions, subsequent experiments focused first on lysosomes. To quantitatively determine where lysosomes pause and resume movement, we extracted their trajectories using single-particle tracking and characterized the dynamic ER network using deep learning–based image segmentation (29). The position of each paused lysosome relative to neighboring ER junctions was normalized by the arc length of its resident ER tubule, termed its normalized distance. The average normalized distance of paused lysosomes was 0.36, compared to 0.5 for theoretically expected random pausing. It should be noted that lysosome size contributes, on average, ~0.3 to the normalized distance (Fig. 1E). We also assessed whether ER junctions are the major pausing sites for other organelles (30). The average normalized distances of paused peroxisomes and lipid droplets to their nearest ER junctions were 0.33 and 0.24, respectively (fig. S1, C and D), confirming that they also stop near ER junctions.

Fig. 1. Lysosomes and endosomes pause and become confined near ER junctions.

(A) Maximum intensity projection (MIP) image of lysosomes within the peripheral ER region of wild-type (WT) COS-7 cells (left) and Atlastin double knockout (ATL DKO) COS-7 cells (right). Images were computed from 15 frames over 30 s. Circles highlight lysosomes confined at ER junctions, and dashed lines show moving lysosomal trajectories. See movie S1. Scale bars, 10 μm. (B) Magnified view of the rectangular regions in WT and ATL DKO cells from (A), showing selected frames. (C) Confocal image of early endosomes (RAB4), late endosomes (RAB7), and lysosomes. In ATL DKO cells, lysosomal confinement appears random. The random pausing is rescued by ATL3 overexpression but not by RTN4a overexpression. Scale bar, 2 μm. (D) Percentage of paused lysosomes and endosomes localized near ER junctions, normalized to ER junction density under different conditions: WT, ATL DKO, RTN4a overexpression, and ATL DKO with RTN4a or ATL3 overexpression (89.7 ± 6.9%, 93.6 ± 8.4%, 93.7 ± 6.8%, 78.9 ± 22.8%, 75.4 ± 22.8%, 61.7 ± 20.8%, 92.8 ± 13.0%; n > 499 trajectories from 21 cells). (E) Distribution of normalized distances of lysosomes to their nearest ER junctions (0.30 ± 0.12, n = 2742 lysosomes). The normalized distance is calculated as the lysosome’s distance from the nearest ER junction divided by the total length of the ER tubule it resides on. pdf, probability density function. Data are represented as mean ± SD, with statistical significance denoted as ns (P > 0.05), ** (P < 0.01), *** (P < 0.001), and **** (P < 0.0001).

To further confirm the role of ER junctions in pausing endolysosomes, we analyzed the lysosome movement in cells with less branched ER networks that resulted from a double knockout (DKO) of ER tubule fusion factor Atlastin 2 (ATL2) and Atlastin 3 (ATL3) (31). The junction density of the peripheral ER was quantified by calculating the number of junctions per unit ER pixel area (fig. S1, E and F). The “frequency of confinement at junctions” was calculated as the proportion of trajectories that paused at ER junctions and was normalized by ER junction density to correct for differences in junction abundance across samples. Under DKO of ATL2 and ATL3, paused lysosomes were no longer restricted to ER junctions (Fig. 1, A to C, fig. S1A, and movie S1), and the fraction of paused lysosomes near ER junctions decreased to 75.3% (Fig. 1, C and D). Overexpression of ATL3 in Atlastin DKO cells restored the fraction of paused lysosomes confined near ER junctions to 92.8% (Fig. 1D and fig. S1A). The reductions in fractions of pausing lysosomes confined near ER junctions under Atlastin DKO may be caused directly by functional loss of Atlastins (32–34), indirectly by reductions of ER junctions, or both. To differentiate between these possibilities, we overexpressed reticulon4a (RTN4a), which reduces ER junctions by stabilizing ER tubules without disrupting Atlastin functions (18, 20). Under RTN4a overexpression, the fraction of paused lysosomes confined near ER junctions was reduced to 78.9% (Fig. 1, C and D, and fig. S1A). Under RTN4a overexpression with Atlastin DKO, the fraction of pausing lysosomes confined near ER junctions was further reduced to 61.7%, indicating that the topology of the ER network contributes to lysosome confinement (Fig. 1, C and D, and fig. S1A). These results indicate that ER junctions, whose formation and maintenance depend on the integrity of membrane proteins such as Atlastin and Reticulon, are preferentially involved in the pausing and resetting of endolysosomes, a process we refer to as motion switching in this study.

Stop-and-go motion switching is associated with elevated density and connectivity at ER junctions

Previous studies have suggested that endosomes and lysosomes move in a stop-and-go fashion with extensive and continuous contact with the ER (19, 31). We checked endolysosomes’ contact with the ER network by detecting overlaps in their fluorescence signals (fig. S2A and movie S2) and found that, at any given time, ~80% of endolysosomes maintain continuous contact with the ER (fig. S2B). We then designed a deep learning–based model Att-BiLSTM that combines hidden Markov model (HMM)–Bayes (35) with long short-term memory (LSTM) (36). This model captures both short-term and long-term dynamic changes in motion, enabling accurate detection of state transitions in the trajectories. As a result, periodic movement of endolysosomes can be classified into three modes: fast, slow, and paused (Fig. 2, A and B, fig. S2C, and movie S3). In the fast mode, which is characterized by a large diffusion coefficient, lysosomes explored different regions with high instantaneous velocities but no clear directionality. In the slow mode, which is characterized by a small diffusion coefficient, the particles largely remained in the same region. Paused mode was detected using three criteria: a diffusion coefficient smaller than 0.01 μm²/s, remaining in the slow diffusion state for at least 20 s, and movement confined within a circle of radius 1 μm. In our study, approximately 62.7% of lysosomes had two modes of movement. Around 16.8% of lysosomes stayed in a single mode, mostly in the slow mode, and about 20.5% of lysosomes exhibited three modes of movement (Fig. 2B). Similarly, endosomes exhibited different modes of movement and underwent motion switching, but with lower transition frequencies (fig. S2D and movie S4).

Fig. 2. Pausing and confinement of lysosomes in local ER regions with elevated density and connectivity.

(A) Representative lysosome trajectory illustrating three diffusion modes. (B) Proportion of lysosomes exhibiting different diffusive modes in COS-7 cells. (C) Time-lapse image sequence and network analysis of the peripheral ER within a 10-pixel-wide sliding window centered on a moving lysosome. Scale bar, 2 μm. (D) Ranking score of ER morphology parameters associated with pausing lysosomes (n = 135 lysosomes from three cells). (E) Representative images of ER organization illustrating both whole-cell and peripheral ER architecture in WT cells and in cells with reduced ER junctions following Atlastin 2/3 double knockdown (ATL DKO). Scale bar, 10 μm. (F to H) Quantitative analysis of lysosomal motion switch frequency [(F); 1.8 ± 0.6, 2.2 ± 0.6, 2.9 ± 1.1], transition frequencies between different motion states (H), and displacement [(G); 9.8 ± 2.4, 7.9 ± 0.5, 5.0 ± 4.2 μm] in WT, ATL DKO, and CLIMP63-overexpressing cells (>208 trajectories from >28 cells). (I) Lysosomal trajectories in COS-7 cells and ATL DKO cells imaged over 5 min. Lysosomal subpopulation analysis based on MSD revealed the following distributions (20.8 ± 3.0%, 24.1 ± 2.5%, 26.9 ± 5.3%, 21.5 ± 3.4%). Scale bar, 10 μm. Data are represented as mean ± SD, with statistical significance denoted as ns (P > 0.05), ** (P < 0.01), *** (P < 0.001), and **** (P < 0.0001).

To investigate specifically how ER morphology influences the movement of individual endolysosomes, we analyzed the relationship between their velocity changes and local ER morphology changes during their switch from movement to pause. Specifically, for each individual endolysosome, we centered a 10-pixel (1.1 μm × 1.1 μm) square sliding window on it. Local ER morphology, traced with the sliding window analysis, was quantified using ER density and three key metrics (29): junction degree, closeness centrality, and betweenness centrality, reflecting the connectivity, communication efficiency, and node influence in the ER network, respectively (fig. S3, A and B, and movie S5). Parameters were then ranked based on their correlation with lysosome pausing and confinement events (Fig. 2C). With a maximum ranking score of 100, the average ranking score was 68.9 for spatial density, 66.1 for junction degree, 66.1 for closeness centrality, and 58.1 for betweenness centrality (Fig. 2D). Consistently, the concentrated distribution area of motion switch sites coincides with the elevated spatial density on the ER density map (fig. S3C). These data suggest that increased network density and connectivity at ER junctions contribute to the stop-and-go motion switching of endolysosomes.

ER junctions are essential for patterned movement of lysosomes

To further examine the role of ER morphology in motion switching, we artificially altered the ER junction density by DKO of ATL2 and ATL3 (ATL DKO), the integral components of ER tubules (Fig. 2E and movie S6). Reduced junction density leads to random pausing along the ER tubule in ATL DKO cells (Fig. 1C) and increases the frequency of lysosomal motion switching by 19% (Fig. 2F), mainly because of twofold increase in the number of particles repetitively circling between the slow mode and the pause mode (7% to 13%) (Fig. 2H). As a result, the displacement of lysosomes is reduced by 19% (Fig. 2G). Overexpression of CLIMP63 resulted in a more pronounced reduction in branching (Fig. 2E), and it consistently induced approximately a 1.5-fold increase in motion switching frequency, primarily caused by a threefold increase (7% to 22%) in nondirected circling within the slow and pause modes (Fig. 2, F and H). As a result, the traveling displacement was reduced by 49% (Fig. 2G). A mean square displacement (MSD) assay was used to characterize the composition of particle populations at the whole-cell level (37), and revealed a 15.9% and 19.1% increase in confined particles, and a 20.0% and 49.2% decrease in directed-moving particles in ATL DKO and CLIMP63 O/E cells, respectively, compared to wild-type controls (Fig. 2I). To further dissect regional effects, we performed single-trajectory analysis in the peripheral regions of ATL DKO cells, which revealed more pronounced alterations compared to whole-cell scale. Specifically, the frequency of motion switching increased by 91% (fig. S3D), the proportion of confined particles increased by 34%, and the fraction of particles undergoing directed movement decreased by 33% (fig. S3E). Consequently, lysosomal displacement in the periphery was reduced by 47% (fig. S3F). These results highlight the importance of ER regions with elevated density and connectivity in regulating the dynamics of lysosomes with their stop-and-go pattern of movement. In the absence of junctions, lysosomes lose the ability to stop and go in a controlled manner, leading to less efficient motion.

Stop-and-go motion switching underlies endosome-lysosome interactions

Functioning of the endolysosomal system, including endosomal transport, directing material from or to the degradative pathway, and lysosomal acidification, requires dynamic and balanced interactions between particles (38, 39). We therefore aimed to monitor endolysosomal interactions and investigate whether motion switches are associated with these interaction events. We used live confocal fluorescence microscopy to image endosomes (BFP-Rab5) and lysosomes (LAMP1-mCherry) relative to ER (GFP-Sec61γ) every 2 s during a 5-min time-lapse video. For the interactions, the two particles initially approached each other through directed movement. Upon their meeting, endosome-lysosome pairs remained relatively stationary for 20 s on average and were often followed by fission, a process marked by the rapid movement of the component that detached (Fig. 3, A and C). The velocity of the interacting pair was 0.06 μm/s on average and increased 4.2 times to 0.24 μm/s after fission (Fig. 3B). Thus, the motion switch takes place throughout the entire interaction process at the junction.

Fig. 3. Motion switching is required for the interaction between lysosomes and endosomes at ER junctions.

(A and B) Time-lapse image (A) and velocity quantification (B) demonstrate that endosomes (white arrow) and lysosomes (yellow arrow) interact at ER junctions through motion switching. Notably, the fission process also involves motion switching (32 events from five cells). Scale bar, 2 μm. (C) A total of 54.1 ± 21.1% of fission endosomes are found to interact with lysosomes (623 events from 17 cells), and 10.2 ± 7.5% of these interacting pairs exhibit simultaneous fission of both endosomes and lysosomes (430 events from 12 cells). (D) A histogram illustrates the distance between the interaction center and the ER junction, alongside the diameters of endosome and lysosome particles in WT and ATL DKO cells (4406 events from five cells). (E and F) Trajectories and motility analysis of endosomes, lysosomes, and their interaction pairs show that most interactions are stationary (75.2 ± 11.5%, 87.2 ± 8.7%, 96.3 ± 2.4% stationary versus 24.8 ± 11.5%, 12.8 ± 8.7%, 3.7 ± 2.4% motile; >9258 interactions from >19 cells across >3 biological replicates, mean ± SD). Scale bar, 10 μm.

To quantitatively measure the interaction, we developed an interaction detection method. In this approach, the nearest neighbor track of the selected track was identified using the k-nearest neighbors (KNN) algorithm. Only the closest track pairs that persisted for more than four frames (8 s) were considered an interaction (fig. S4A). Detected interactions are associated with frequent endolysosomal fission events, with 54.0% of fission endosomes interacting with lysosomes, and 10.2% of these interactions involving simultaneous fission of both endosomes and lysosomes (Fig. 3C). By quantifying interactions, we found that ER junctions serve as key sites for the interaction. The distance between the center of interaction and the junction is nearly identical to the diameter of the endolysosomes (Fig. 3D). In ATL DKO cells, the distance between the interaction center and the junction is approximately 2.6 times the particle diameter (Fig. 3D). Another notable feature observed for endosome-lysosome interaction pairs was that they were relatively stationary. MSD analysis showed that, compared to the tracks of individual endosomes and lysosomes, the interacting particles exhibited the shortest traveling distances (Fig. 3E) and the highest percentage (>90%) remaining in the immobilized state (Fig. 3F). These results underscore the motion switching process integral to endosome-lysosome interactions.

ER junctions coordinate endosome-lysosome interactions and are spatially coupled to the microtubule and actin cytoskeletons

Our data suggest that ER enrichment, particularly at junction sites, is involved in endolysosomal motion switching and interaction. Therefore, we investigated whether the extensive ER network, which spans a substantial portion of the cell, globally regulates these motion switching events and synchronizes endosome-lysosome interactions at the whole-cell level. Clusters, defined by a higher density compared to neighboring areas in our previous study, represent a distribution pattern of endolysosomes that increases local density through the recruitment of lysosomes undergoing directed movement and the confinement of preexisting lysosomes (37, 40). To identify these clusters computationally, we performed mean shift–based spatial clustering analysis and observed that endolysosomes were clustered in both the perinuclear and peripheral regions of cells (Fig. 4A). We first focused on the peripheral regions. The interaction rate was defined as the ratio of the interaction events to the total number of particles. The relative interaction rate, comparing the interaction ratio within the cluster to that in the adjacent nonclustered region, revealed 1.5- and 1.4-fold increases in endosomes and lysosome clusters, respectively. Notably, the intersection region, defined as the region where endosome and lysosome clusters overlap, exhibited the highest relative interaction rates (2.5-fold) (Fig. 4B). Similar results were observed at the whole-cell level. Compared to the outside nonclustered region, the interaction frequencies of both endosome and lysosome clusters were elevated by ~1.2-fold, and the most notable increase was observed in intersecting cluster regions, which showed a 1.6-fold elevation (Fig. 4C). Thus, the interaction between endosomes and lysosomes correlates with their spatial density (Fig. 4C). We also found that increased ER density correlates with the cluster formation. At the periphery, ER density increased by 1.1-, 1.2-, and 1.2-fold in endosome clusters, lysosome clusters, and the intersection region, respectively. Whole-cell level analysis revealed consistent increases in ER density, with approximately 1.2-, 1.6-, and 1.7-fold increases in endosome clusters, lysosome clusters, and intersection regions, respectively (Fig. 4, B and C).

Fig. 4. ER coordinates the endosome-lysosome interaction at whole-cell level.

(A) Clusters of endosomes and lysosomes were identified computationally. Scale bars, 5 μm (peripheral region) and 10 μm (whole-cell view). (B) Left: Interaction ratio for endosomes or lysosomes within the cluster region relative to those outside the cluster in the peripheral region (1.5 ± 1.0, 1.4 ± 0.7, 2.5 ± 1.4). Right: ER density inside the cluster region relative to that outside the cluster (1.1 ± 0.1, 1.2 ± 0.2, 1.2 ± 0.2), 57 regions selected from 22 cells. (C) Left: Interaction ratio for endosomes or lysosomes within the cluster region relative to those outside the cluster in the whole cell (1.2 ± 0.7, 1.2 ± 0.7, 1.6 ± 1.0). Right: ER density inside the cluster region relative to that outside the cluster (1.5 ± 0.4, 1.6 ± 0.6, 1.7 ± 0.8), five cells. (D) Using an interaction detector, endosome-lysosome pairs (distance <0.11 μm, interaction duration >10 s) are marked by boxes and circles. Scale bar, 10 μm. (E) The interaction frequency exponentially increases with lysosome number in WT cells, but not in VAPA KO cells or ATL DKO cells. (N > 19,913 events from >18 cells for each type). (F) Western blot analysis using anti–cathepsin D antibodies reveals the positions of procathepsin D (53 kDa) and the mature chain of cathepsin D (31 kDa), as indicated. (G and H) Confocal fluorescence images (G) of WT, ATL DKO, and VAPA KD cells stained with Magic Red. Quantification is shown in (H) (n = 30 cells from 12 biological replicates). Scale bar, 10 μm. Data are represented as mean ± SD, with statistical significance denoted as ns (P > 0.05), ** (P < 0.01), *** (P < 0.001), **** (P < 0.0001).

The interaction rate is also closely correlated with the abundance of both lysosomes and endosomes, and the second-order polynomial model fits the observed increase well (fig. S4B). In contrast, this interdependency was lost in ATL DKO cells, where the ER network was disrupted, as well as in VAPA knockdown (KD) cells, which exhibit impaired ER-lysosome MCSs (Fig. 4, D and E). This suggests that the ER-enriched regions, where endolysosomes cluster, represent the area with more frequent endosome-lysosome interactions.

The lack of ER junctions could impede the maturation process of endolysosomes. Analysis of lysosome acidity and proteolytic activity, both important indicators of endosome and lysosome maturation (39), revealed a decrease in the mature form of cathepsin D in ATL DKO cells and VAPA KD cells, indicating impaired maturation (Fig. 4F). Additionally, Magic Red, a membrane-permeable cathepsin substrate (41), showed substantially reduced intensity in ATL DKO cells and VAPA KD cells (Fig. 4, G and H), suggesting that loss of ER junction affects the acidity and proteolytic activity of lysosomes.

How does the ER contribute to stop-and-go behavior of endolysosomes? While microtubules and motor proteins are well-established determinants of endolysosomal motility (3, 42), increasing evidence suggests that they are not the sole regulators of this process. Actin filaments are another candidate regulator of motion switching, as they have been reported to modulate endolysosomal dynamics by promoting tubule formation and cargo sorting, facilitating early-to-late endosome maturation, and driving vesicle transport following fission (43, 44). We therefore investigated whether these three components (microtubules, actin filaments, and ER) are spatially and functionally interconnected for endolysosomal motion switching. Coimaging of lysosomes and microtubules revealed that while most lysosomes undergo bidirectional movement along microtubules, a subset also moves and remains in regions lacking detectable microtubules (fig. S4C). We also investigated the spatial colocalization between ER junctions and the two cytoskeletal systems. By extracting the coordinates of ER junctions, we defined the enrichment ratio as the fluorescence intensity of the cytoskeletal signal at the junction relative to that in the surrounding region. An enrichment ratio greater than 1 was considered indicative of local cytoskeletal enrichment at the ER junction. Using this criterion, we found that 57.5% of ER junctions exhibited microtubule enrichment and 45.8% showed actin enrichment (Fig. 5A). Furthermore, we conducted a detailed examination of the roles of the actin and microtubule cytoskeleton in maintaining ER morphology, particularly in the formation and stability of ER junctions. Actin depolymerization using cytochalasin D or latrunculin rapidly disrupted the peripheral ER network. In contrast, microtubule depolymerization using nocodazole, despite being highly effective (Fig. 5B and fig. S4D), led to a slower and less pronounced reduction in ER junctions (Fig. 5C). Conversely, manipulating ER morphology, specifically modifying ER shape, did not produce detectable changes in overall microtubule organization, including average microtubule length and directional distribution (fig. S4, E to H). This observation is consistent with our motility analysis and previous findings from our laboratory (37), highlighting the role of actin-dependent ER shape maintenance and endolysosomal motility in addition to the contribution of the microtubule cytoskeleton.

Fig. 5. Spatial and structural coupling among the ER, microtubules, and the actin cytoskeleton.

(A) Microtubule and actin enrichment at ER junctions. The ER, microtubules, and actin filaments were labeled with GFP-Sec61β, mCherry-tubulin, and mCherry-utrophin, respectively. Microtubules and actin are enriched at ER junctions, as indicated by yellow circles (>1249 events from >24 cells were analyzed). (B and C) Time-lapse imaging of peripheral ER labeled with GFP-Sec61β was conducted to monitor ER morphology during cytoskeletal depolymerization. Cells were treated with 2 μM nocodazole to depolymerize microtubules, or with 1 μM latrunculin A or 2 μM cytochalasin D to depolymerize actin filaments. ER junction density (number per unit area) was quantified over time using network analysis and is shown in (C). Data are presented as mean ± SD. Scale bar, 10 μm.

YWHAH, recruited by VAPA-STARD3 to ER-endolysosome contact sites, regulates their motility

Beyond their spatial alignment with ER junctions, microtubules and actin filaments also functionally contribute to lysosomal motility through molecular mechanisms such as kinesin activation and actin nucleation (42, 45). A prominent example is the MCSs between ER tubules and lysosomes, which are preserved during lysosomal trafficking and rely on vesicle-associated membrane protein-associated proteins (VAPs) as tethers to recruit motor proteins (30, 46, 47). VAPA KD in COS-7 cells reduced lysosomal mobility by increasing the proportion of particles exhibiting constrained diffusion and decreasing the proportion showing directed motion (Fig. 6, A and B, and fig. S5A). Similarly, KD of ORP1L (30, 47), a VAP-binding protein crucial for dynein-mediated lysosomal transport along microtubules, also impaired lysosomal mobility (fig. S5, A and B). Depletion of STARD3, a lipid exchange mediator at contact sites, also impaired lysosomal motility, increasing constrained diffusion by 10% and reducing directed movement by 16% (Fig. 6, A and B). Only reexpression of STARD3, but not VAPA or other VAPA-interacting proteins such as Protrudin, rescued the increased confinement and restored directed movement caused by STARD3 KD (fig. S5C). Re-expression of VAPA or VAPB, but not STARD3, restored the reduced lysosomal motility caused by VAPA KD (fig. S5D). Additionally, GFP-tagged VAPA localized throughout the ER and showed enrichment around lysosomes, whereas overexpression of the dominant-negative mutant VAPA KDMD reduced accumulation on lysosomes (Fig. 6C and movie S7). VAPA mediates membrane contacts via its major sperm protein (MSP) domain, which interacts with the FFAT motif of its binding partners (48). Consistently, overexpression of a contact-deficient mutant of STARD3 (STARD3ΔFFAT), fused with GFP at its N terminus, disrupted its localization around lysosomes (Fig. 6C). Protrudin exhibits a subcellular localization pattern similar to STARD3, being predominantly ER-localized and dynamically accumulating at ER-lysosome contact sites (fig. S5E). The interaction between STARD3 and VAPA was further validated by coimmunoprecipitation of GFP-STARD3 with endogenous VAPA, and the MSP domain of VAPA successfully pulled down GFP-STARD3 from cell lysates (Fig. 6, D and E).

Fig. 6. VAPA-STARD3–mediated membrane tethering recruits YWHAH to regulate endolysosomal dynamics.

(A and B) Characterization of lysosome movement under perturbation of the VAP-STARD3-YWHAH pathway, based on MSD assay (19.5 ± 2.2%, 21.5 ± 3.3%, 21.2 ± 2.2%, 22.2 ± 2.2%, 33.22 ± 3.7%, 26.0 ± 6.4%, 27.9 ± 3.4%, 24.8 ± 6.9%, >10,521 trajectories from >46 cells, in more than three independent replicates) and displacement quantification (11.2 ± 2.5, 9.0 ± 1.6, 7.8 ± 1.9, 5.5 ± 1.4, >59 cells). (C) Perturbation of the VAP-STARD3 interaction alters ER morphology around lysosomes. (D and E) IP and pulldown assays reveal that VAPA interacts with STARD3, and STARD3 interacts with YWHAH. Scale bar, 10 μm. (F) Confocal images of cells expressing BFP-YWHAH, mCherry-UtrCH, and lysosomes labeled with dextran 647, showing actin cytoskeleton accumulation on YWHAH-positive lysosomes. BFP-YWHAH is diffusely distributed in the cytoplasm but shows local enrichment on a subset of lysosomes in 76% of wide-type cells. In contrast, no such enrichment is observed in STARD3 KD cells (0%, n = 30). (G and H) YWHAH is dynamically recruited to STARD3-enriched puncta in a movement-associated manner (G). The enrichment of YWHAH on lysosomes occurs concurrently with local actin accumulation (indicated by dashed circles) and is accompanied by lysosomal movement (H). Right panels show the complete trajectory of a representative lysosome, including velocity and fluorescence intensity changes. Dashed lines indicate acceleration (Acc.) and deceleration (Dec.) phases. Scale bar, 10 μm. Data are represented as mean ± SD, with statistical significance denoted as ns (P > 0.05), ** (P < 0.01), *** (P < 0.001), **** (P < 0.0001).

How does VAP-STARD3–mediated membrane contact modulate lysosomal movement, and why does the disruption of STARD3 lead to the confinement phenotype? We hypothesized that VAP-STARD3 engages with additional downstream factors. An analysis of the STARD3 interactome identified tyrosine 3 monooxygenase/tryptophan 5-monooxygenase activation protein eta (YWHAH), a 14-3-3 protein family member known to bind actin regulatory proteins such as cofilin and its phosphatase slingshot. This interaction can reorganize the actin cytoskeleton (32, 49) and forms a functional partnership with phosphatidylinositol-4 (PI4) lipid phosphatase SAC1 and PI4-kinase-IIIβ (PI4KIIIβ) to regulate phosphoinositide homeostasis (50, 51). To test the role of YWHAH, we conducted colocalization, functional, and interaction analyses. RNA interference (RNAi)–mediated depletion of YWHAH (fig. S5A) phenocopies the lysosomal dynamic defects observed in VAPA or STARD3 KD. MSD analysis revealed that a higher fraction of lysosomes (22.2 ± 2.2%) exhibited constrained diffusion, while the proportion of lysosomes undergoing directed movement decreased to 24.8 ± 6.9% (Fig. 6, A and B). Additionally, Flag-YWHAH and GFP-STARD3 coimmunoprecipitated with each other from cell extracts, as confirmed by Western blotting (Fig. 6, D and E). YWHAH was diffusely distributed throughout the cytoplasm but was enriched in prominent puncta colocalized with utrophin calponin homology domain (UthCH)–labeled F-actin structures and lysosomes in 76% of wild-type cells. In contrast, no such enrichment was observed in STARD3 KO cells, where actin accumulation around lysosomes was also absent (Fig. 6F). We further examined the dynamic colocalization of STARD3, YWHAH, and actin on lysosomes. Time-lapse imaging revealed that YWHAH progressively accumulated on STARD3-positive puncta, and this accumulation coincided with the directed movement of these puncta (Fig. 6G). Moreover, during these movements, the enrichment of YWHAH on lysosomes occurred in parallel with actin polymerization at the same sites (Fig. 6H). These spatiotemporal colocalization patterns are consistent with the biochemical interaction between STARD3 and YWHAH.

Targeting YWHAH to lysosomes promotes actin assembly and partially rescues lysosomal dynamics and acidification in ER junction– or membrane tethering–deficient cells

Actin comets have previously been described as the propelling force for endogenous organelles and are nucleated by N-WASP (a Wiskott-Aldrich syndrome family member) in response to the abundance of phosphoinositide (45) or lysosomal trapping (37). In our observation, actin comets were frequently enriched near ER junctions, and in some cases, F-actin formed ring structures encircling ER polygons (fig. S6A). We also observed that, as indicated by fluorescence signals, lysosomes were colocalized with the dynamic actin accumulation during their movement (Fig. 7A). To avoid potential artifacts caused by fluorescent protein overexpression, we stained actin with phalloidin and lysosomes with anti-LAMP1 antibodies, and imaged by confocal microscopy. Although actin fibers are present near lysosomes in general, we observed the enrichment of actin around lysosomes in wild-type cells but not in ATL DKO cells (fig. S6B).

Fig. 7. YWHAH-mediated actin dynamics regulate the motion switching of endolysosomes.

(A) Time-lapse confocal images of Cos-7 cells transfected with the F-actin probe mCherry-UtrCH and dextran 647–labeled lysosomes. The accumulation and shape of actin comets dynamically change with lysosomal movement. Scale bar, 2 μm. (B) TLCC analysis between actin and velocity changes. (C and D) Examples and statistical analysis of the correlation between lysosomal velocity and actin dynamics. (E) Representative confocal image showing STARD3 localization and the associated actin enrichment on lysosomes. (F) Quantification of STARD3/actin enrichment on lysosomes. The enrichment index was defined as the ratio of STARD3/actin fluorescence intensity within lysosomes to the average intensity in the surrounding 10 μm × 10 μm region. Actin enrichment associated with STARD3 was compared between control RNAi and YWHAH RNAi conditions, as well as in cells expressing STARD3 mutants: ΔFFAT (deficient in VAP binding), phospho-deficient (KAAANP), and phospho-mimetic (KDADNP) variants. (G) Correlation analysis of STARD3 and actin dynamics during lysosomal movement. Colocalization between STARD3 and actin was maintained during periods of sustained lysosomal acceleration and deceleration lasting more than 10 s. Scale bar, 10 μm. Data are represented as mean ± SD, with statistical significance denoted as ns (P > 0.05), ** (P < 0.01), *** (P < 0.001), **** (P < 0.0001).

Destabilizing actin filaments with latrunculin led to ER network collapse and impaired lysosomal dynamics. ER network complexity, quantified as junction frequency (number of junctions per unit ER tubule length), showed a 24% reduction in latrunculin-treated cells (fig. S6C). To investigate the role of actin dynamics during lysosome motion switching, we analyzed periods of acceleration and deceleration lasting over 10 s in individual trajectories and quantified real-time actin assembly on lysosomes. Time-lag cross-correlation (TLCC) analysis was performed to examine the leader-follower relationship between velocity and actin. Distinct patterns emerged: During sustained deceleration, slowing down preceded changes in actin, whereas during acceleration, actin change occurred before increases in velocity (Fig. 7B). Besides the chronological order, we specifically analyzed the relationship between actin dynamics and acceleration/deceleration events. In our analysis, if an increase in actin signal coincided with a velocity increase and the correlation coefficient was greater than 0.8, we considered actin to have a positive correlation with the acceleration phase. A correlation coefficient between −0.8 and 0.8 was considered indicative of a weak correlation, while a value below −0.8 suggested a negative correlation (Fig. 7C). In wild-type cells, 50% of lysosomal acceleration events were positively correlated with actin assembly. In contrast, in the ATL DKO cells, the effect of actin was less pronounced, with the proportion of positive correlations reduced to 10%. More than 60% of acceleration events showed a weak correlation with actin dynamics (Fig. 7D). The proportion of events with weak correlation to deceleration increased from 30% in wild-type cells to 60% in ATL DKO cells (Fig. 7D), underscoring the importance of ER junctions in mediating actin assembly during endolysosomal motion switching.

We also performed a quantitative analysis of STARD3’s lysosomal localization and its role in actin recruitment. We defined a lysosomal enrichment index as the ratio between STARD3 fluorescence intensity within lysosomes and the average intensity within a surrounding 10 μm × 10 μm region. An index greater than 1 was considered indicative of STARD3 accumulation on lysosomes. Using this criterion, we found that 89.8% of lysosomes exhibited STARD3 enrichment. This accumulation was highly dependent on interactions with VAP proteins, as deletion of the FFAT motif produced a secondary population of lysosomes with enrichment index near 0. YWHAH RNAi moderately reduced lysosomal enrichment to 68.0%, but did not show accumulation near the index of 0 (Fig. 7, E and F). To probe the STARD3–YWHAH interaction and its link to actin, we further quantified local actin enrichment around lysosomes. Actin localization displayed a similar pattern to STARD3’s lysosomal accumulation: In cells overexpressing GFP-STARD3, 75.7% of lysosomes showed local actin enrichment. This percentage decreased to 43% in cells expressing the FFAT motif–deleted mutant. Notably, in YWHAH RNAi cells, actin enrichment was most markedly reduced to 15% (Fig. 7, E and F). To examine the dynamic interplay between STARD3, YWHAH, and actin during lysosomal movement, we tracked lysosome trajectories with a focus on continuous 10-s acceleration and deceleration phases, analyzing the correlation between dynamic STARD3 and actin signals. In wild-type cells, STARD3 signal positively correlated with actin during both acceleration (61% positive, 6% negative, 33% no correlation) and deceleration phases (52% positive, 14% negative, 33% no correlation). In contrast, both the GFP-STARD3 ΔFFAT mutant and GFP-STARD3 expressed under YWHAH RNAi conditions exhibited markedly reduced correlation with actin during the acceleration phases (acceleration: 17% and 14% positive, 19% and 11% negative, 64% and 75% no correlation; deceleration: 21% and 26% positive, 18% and 13% negative, 53% and 54% no correlation) (Fig. 7G). These results demonstrate that the spatial organization of the VAPA-STARD3-YWHAH complex is essential for actin recruitment, highlighting the critical role of ER-lysosome interactions in this process.

STARD3 can be phosphorylated (fig. S6D), with serine-209 and serine-213, which lie within the FFAT motif critical for binding to VAP proteins (52), identified as potential phosphorylation sites. It has also been suggested that STARD3 utilizes the 392-KSASNP-397 sequence to bind 14-3-3 proteins in a phosphorylation-independent manner (53). We found that neither the phospho-deficient nor the phosphomimetic mutant affected STARD3 accumulation on lysosomes, actin recruitment (Fig. 7, E and F), or its interaction with YWHAH, as assessed by glutathione S-transferase (GST)–YWHAH pull-down assays (fig. S6E).

On the basis of this observation, we hypothesized that YWHAH at MCSs may recruit actin to regulate endolysosomal dynamics. To test this, we artificially fused YWHAH to LAMP1 and evaluated the colocalization frequency of actin foci with lysosomes in COS-7 cells overexpressing either LAMP1-YWHAH or LAMP1. Colocalization frequency was quantified by measuring the overlap between actin comet trajectories and lysosomal tracks. Cells expressing LAMP1-YWHAH showed a higher colocalization rate of actin comets with lysosomes compared to those expressing LAMP1 (80% versus 70%) (Fig. 8A). MSD assay revealed that overexpression of LAMP1-YWHAH, but not LAMP1 in wild-type cells, enhanced lysosomal motility by increasing the number of particles undergoing directed movement and extending their traveling distances in wild-type cells, and partially rescued impaired lysosomal dynamics in ATL DKO cells (Fig. 8, B to D). This incomplete rescue may be attributed to the lack of contact with ER junctions. In wild-type cells, lysosomes labeled with either construct frequently paused near ER junctions. In contrast, in ATL DKO cells, lysosomes frequently paused at locations distant from ER junctions or even away from ER tubules, and this pattern was not rescued by the expression of either LAMP1 or LAMP1-YWHAH, highlighting the importance of ER junctions in regulating lysosomal motility (fig. S6F).

Fig. 8. VAPA-STARD3-YWHAHA is required for lysosomal acidity and proteolytic activity.

(A) Abundant actin accumulation around lysosomes was observed in Cos-7 cells expressing the chimera protein LAMP1-YWHAH-mCherry, compared to cells expressing LAMP1-mCherry. The ratio of actin comet-positive lysosomes to the total lysosome population in the entire cell was measured (70 ± 18%, 83 ± 18%, >20 cells). (B) Lysosomal trajectories in a COS-7 cell imaged over 5 min. (C) Percentage of lysosomal subpopulations in the MSD assay (23.3 ± 4.1%, 19.7 ± 2.7%, 28.8 ± 7.7%, 36.7 ± 9.1%, >39,257 tracks from >26 cells, in more than three independent replicates). (D) Traveling distance of lysosomes (9.6 ± 1.7, 10.7 ± 1.7, 7.4 ± 1.4, 8.4 ± 1.4, >36 cells). (E) The intensity of Magic Red was reduced in ATL DKO, VAPA KD, STARD3 KD, and YWHAH KD cells. This reduction was partially rescued by overexpression of LAMP1-YWHAH-GFP (purple dashed line) compared to nonexpressing cells (blue dashed line). (F) Western blot analysis using anti–cathepsin D antibodies revealed the positions of procathepsin D (53 kDa) and the mature chain (31 kDa) of cathepsin D, as indicated on the margin. The overexpression of LAMP1-YWHAH-mCherry restored the level of mature cathepsin D. (G) A schematic diagram illustrating endolysosome movement regulated by the coordinated interactions among the ER, microtubules, and the actin cytoskeleton. Endosomes and lysosomes converge at ER junctions via a motion switch mechanism mediated by the interaction between VAPA and STARD3. Beyond microtubule-based transport, further interactions between STARD3 and YWHAH, together with subsequent actin assembly, are required to facilitate the interplay among the ER, endosomes, and lysosomes. Scale bar, 10 μm. Data are represented as mean ± SD, with statistical significance denoted as ns (P > 0.05), ** (P < 0.01), *** (P < 0.001), and **** (P < 0.0001).

We next examined whether VAPA, STARD3, and YWHAH are functionally associated with lysosomal acidity and proteolytic activity. Overexpressing LAMP1-YWHAH-GFP partially recovered Magic Red fluorescence intensity in ATL DKO cells, as well as in VAPA-, STARD3-, and YWHAH-depleted cells (Fig. 8E). Consistently, Western blot analysis revealed an increase in the mature form of cathepsin D in these cells (Fig. 8F). These findings suggest that YWHAH, recruited by STARD3, facilitates endolysosomal dynamics and maturation by promoting actin assembly.

DISCUSSION

Dynamic movement of endolysosomes on the ER network has been studied extensively. It is now widely believed that the localization and movement of endolysosomes rely on intricate interplay between microtubule-based motility, actin-based motility, and organelle membrane contacts. The localization and movement of endolysosomes mediate intracellular responses to changing nutrient levels and lipid distributions in the membrane. Although various mechanisms for recruiting motor proteins to endolysosomes have been identified, the regulatory mechanisms underlying the switching of endolysosomes between static and motile states as well as between anterograde and retrograde transport are still not well understood. The biological functions of the dynamic and complex motion switching of endolysosomes remain largely unknown.

Our study shows that the switching of endolysosomal movement states is jointly regulated by the morphology of the ER network, organelle interactions, the microtubule cytoskeleton, and the actin cytoskeleton. In particular, ER junctions serve as critical sites for switching the movement states of endolysosomes. The ER network structure provides a global “map” that coordinates the stop-and-go pattern of endolysosome movement within the cell, increasing the likelihood of their encounters at junctions and facilitating their interactions. At the molecular mechanism level, VAP and STARD3 are localized on the ER and lysosomes, respectively, tethering the organelles and regulating actin assembly on lysosomes through the recruitment of YWHAH, thereby completing the “stop-and-go” motion switching (Fig. 8G). Therefore, the dynamics of lysosomes should no longer be studied in isolation but within the context of an integrated system. Coordination among multiple mechanisms enables the precise regulation of lysosomal movement and distribution.

Extensive studies have also demonstrated that membrane curvature is a critical link between surface topography sensing and the intracellular organization of the actin cytoskeleton. Changes in membrane curvature are mediated by the activation of curvature-sensing proteins such as FBP17, which initiates branched actin polymerization through N-WASP, cortactin, and the Arp2/3 complex (54, 55). These findings support our observations: ER junctions inherently exhibit high membrane curvature and function as hubs that integrate lysosomal stop-and-go motility, VAP-mediated membrane contacts, and local actin enrichment. Furthermore, our findings provide mechanistic insights that complement previous studies, suggesting a possible molecular basis whereby VAP-mediated interactions and ER membrane morphology jointly contribute to contact formation. Obara et al. reported that VAPB-enriched contact sites contain dynamic subdomains associated with ER membrane curvature (56). VAPB dynamically accumulates at these sites, forming a central-to-peripheral gradient within ER-mitochondria contact regions. This local enrichment is proposed to enhance membrane curvature by increasing membrane adhesion (56). Similarly, membrane curvature has been shown to influence the spatial distribution of organelle interactions at ER-PM contact sites (57). These contact sites likely facilitate the recruitment of tethering proteins as well as the exchange of signaling molecules, lipids, and metabolites (58, 59). These potential links highlight the possibility that membrane curvature acts as a general coordinating platform, coupling cytoskeletal dynamics with both intracellular and extracellular membrane systems. Such coordination may play a pivotal role in a wide range of dynamic cellular processes.

Recent studies have also substantially expanded our understanding of the functional organization of endosomes and lysosomes. Once viewed as static organelles limited to processing and recycling roles, they are now recognized as forming highly dynamic and spatially regulated networks. While lysosomal distribution, size, and the accumulation of intraluminal contents have been used to infer potential pathological changes, quantitative characterization of collective organelle motion at the subcellular level remains underexplored (60). To advance endolysosome dynamics analysis, computational analysis tools are needed to ensure efficiency and accuracy, providing valuable insights into the role of endolysosomes in disease progression. Here, we have developed a computational analysis pipeline that integrates lysosomal motion analysis, spatial distribution analysis, and ER network analysis, applying it to mechanistic studies of individual and collective behaviors of endolysosomes. Nevertheless, our study has its limitations. Leveraging the flat profile of COS-7 cells, we conducted ER morphology and single-particle motion analyses in two-dimensional imaging. Expanding these analyses tothree-dimensional analysis will enable broader applications across different cell types. Additionally, our data show that endolysosomes have a 30% probability of pausing and becoming confined at ER junctions, indicating that beyond ER junction regulation, lysosomal dynamic distribution is subject to other regulatory mechanisms. A more comprehensive assay capable of analyzing diverse organelles and cytoskeletal components is essential for understanding mechanisms underlying lysosomal dynamic positioning.

Our data demonstrate the prominent regulatory role of microtubules in lysosomal motility while also revealing an essential role of actin in orchestrating the stop-and-go behavior of lysosomes. This observation aligns with the well-recognized functions of actin in various endosomal processes, such as biogenesis, maturation, morphology, movement, positioning, and cargo sorting (19, 61, 62). Additionally, endolysosomes are frequently observed to bind to actin-based comet tails, indicating an actin-dependent mechanism in their dynamic positioning and transport (12). The Wiskott-Aldrich Syndrome Protein and SCAR Homolog (WASH) complex may directly induce potent actin nucleation on endolysosomes in response to the abundance of phosphoinositide (45) or facilitate endolysosome transport by linking actin and microtubule cytoskeletons (63, 64). Dong et al. have shown that VAP-mediated ER-endosome contacts affect WASH functions (45). Altered interaction probabilities observed in VAP KD cells may also reflect the role of VAPs in regulating the HOPS complex, which is essential for late endosomal tethering and fusion (13). Our findings further provide an example of inter-organelle communication involving actin regulation, showing that YWHAH is recruited to ER-endolysosome contact sites, where it influences actin nucleation. This insight suggests mechanisms by which VAP mutations contribute to neuronal disease pathology. Mutations in the FFAT-binding VAP-MSP domain as well as in the WASH complex have been implicated in neurodegenerative conditions such as amyotrophic lateral sclerosis (ALS), Alzheimer’s disease, and Parkinson’s disease (65–69). Notably, in ALS-affected spinal cord tissue, YWHAH expression is up-regulated, triggering a cascade of events that ultimately leads to motor neuron degeneration (70). In Alzheimer’s disease, YWHAH also serves as a critical hub gene implicated in disruptions of the autophagosome-lysosome pathway, neurotransmitter synthesis, and amyloid-β (Aβ) clearance (71). Further exploration of the protein network discussed here could provide insights into the pathogenic mechanisms underlying these diseases.

MATERIALS AND METHODS

Plasmids and primers

Plasmids encoding fluorescent fusion proteins were either purchased from Addgene or constructed in house using ligation (NEB M2200S) or In-Fusion (Clontech 638947). A detailed list of the plasmids is provided in table S1. For RNAi KD of YWHAH, two short hairpin RNA (shRNA) sequences were used as in table S2. The shRNA plasmids were constructed based on the lentiviral backbone PLKO.1 (Addgene 8453) following the protocol provided by Addgene (https://www.addgene.org/protocols/). For CRISPR-Cas9 KD of VAPA, STARD3, and ORP1L, two sites were designed for each gene, and the sequences were provided in table S2. The single guide RNA (sgRNA) plasmids were constructed based on the lentiviral backbone LentiCRISPRv2 (Addgene 52961) following previously described protocol (72). Briefly, forward and reverse primers of each shRNA and sgRNA were annealed by preincubation at 37°C for 30 min in T4 ligase buffer followed by incubation at 95°C for 5 min and then ramped down to 25°C at 5°C/min. The annealed inserts were digested with EcoRI/AgeI for PLKO.1 and Bsm BI for LentiCRISPRv2 and ligated using the Quick Ligation Kit (NEB M2200S).

Cell culture, transfection, and generation of stable KD cell lines

Cos-7 (Cercopithecus aethiops, ATCC CRL-1651, RRID:CVCL_0224) and human embryonic kidney (HEK) 293T cells (Homo sapiens, ATCC CRL-11268, RRID:CVCL_0063) were obtained from the American Type Culture Collection (ATCC) and cultured in Dulbecco’s modified Eagle’s medium (CellMAX CGM101.05) supplemented with 10% fetal bovine serum (CellMAX SA101.02) and 1% penicillin-streptomycin (CellMAX CPS101.02). Cells were seeded in a six-well plate at a density of 1 × 10⁵ cells per well before transfection. Transfection of plasmid DNA was performed using Lipofectamine 3000 (Invitrogen L3000015) according to the manufacturer’s instructions. Briefly, cells were incubated in 1000 μl of Opti-MEM media containing 2 μg of plasmid and 5 μl of Lipofectamine for 4 hours. After transfection, cells were reseeded at a density of 1.6 × 10⁵ cells per dish in glass-bottom dishes (MatTek P35G-0-7-C) for subsequent imaging or selection of stable KO lines. For gene KD, plasmids were cotransfected with the packing plasmids pVSVG (Addgene 8454) and psPAX2 (Addgene 12260) into HEK293T. Stable KD cell lines were generated after infection using the produced virus for 24 h, and subsequently selected and validated using polymerase chain reaction and Western blot.

Live-cell staining, immunofluorescence staining, and imaging

Lysosome acidity was evaluated using Magic Red Cathepsin B Assay Kit (ImmunoChemistry 937) according to the instructions. For immunofluorescence staining, cells were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 15 min at room temperature, followed by permeabilization with 0.1% Triton X-100 in PBS for 10 min. After blocking with 5% bovine serum albumin in PBS for 1 hour at room temperature, cells were incubated with a primary antibody against LAMP1 (1:200 dilution) overnight at 4°C. After washing, cells were incubated with an Alexa Fluor 488–conjugated secondary antibody (1:500) for 1 hour at room temperature. F-actin was stained by incubating the cells with Alexa Fluor 568–conjugated phalloidin (1:200) for 30 min. For live-cell imaging, fluorescently labeled cells were grown on glass-bottom MatTek dishes at 60% confluency. Cells were imaged using a 100× 1.45 numerical aperture (NA) oil immersion objective on an Eclipse Ti2-E inverted microscope (Nikon) with a CSUW1 Spinning Disk scanning head (Yokogawa) and a Prime 95B sCMOS camera (Photometrics), all controlled by Nikon Elements software. Cells were maintained at 37°C with 5% CO₂ in an on-stage incubator (Tokai Hit). Cells expressing endosome, lysosome, and/or ER markers were imaged in a single focal plane for 2 min with images taken every 2 s. For single-particle tracking, images were taken at 2-s intervals for maximal temporal resolution.

Western blot

For sample preparation, cells were lysed in radioimmunoprecipitation assay lysis buffer strong (Beyotime, P0013B) supplemented with protease inhibitor cocktail (Roche, 4693116001). Cells were lysed at 4°C for 20 min and centrifuged at 14,000 rpm for 10 min to remove insoluble debris. Protein concentrations were quantified using the Bradford assay (Beyotime, P0012). SDS samples containing 4 to 8 μg of total proteins were separated with 12% Hepes-tris polyacrylamide gel electrophoresis (PAGE) gel (MEILUNBIO, MA0243) and transferred to polyvinylidene difluoride membranes (Millipore, 70584-3). The membrane was blocked with 5% skim milk in TBST buffer (20 mM tris, 150 mM NaCl, 0.1% Tween 20, pH 7.4) and then incubated with the indicated primary antibodies at room temperature for 1 h. After washing three times with TBST, the horseradish peroxidase–conjugated secondary antibodies were incubated at room temperature for 1 hour. The antibodies are listed in table S3.

Pull-down assay and immunoprecipitation

VAPA-MSP domain was amplified from a human cDNA library and cloned into pGEX4T vectors. The resultant plasmids were transformed to Escherichia coli strain BL21 (DE3) and grown at 37°C in LB media. Protein expression was induced at OD₆₀₀ (optical density at 600 nm) = 0.6 by isopropyl-β-d-thiogalactopyranoside (IPTG, 0.3 mM final concentration) at 16°C for 16 hours. E. coli were harvested and disrupted by sonication in the lysis buffer [25 mM tris-HCl (pH 7.5), 150 mM NaCl, 1% NP-40] supplemented with protease inhibitor cocktail (cOmplete, Roche). After centrifugation at 40,000g for 30 min, cell lysate was incubated with glutathione Sepharose beads (GE 17-0756) for 60 min at 4°C. Beads were washed with PBS containing 250 mM NaCl, reequilibrated in PBS supplemented with 20% glycerol, and flash-frozen in liquid nitrogen for pulldown assay. A total of ~1 × 10⁷ HEK293T cells expressing GFP-STARD3 were lysed with 1 ml of lysis buffer supplemented with a protease inhibitor cocktail (cOmplete, Roche). Lysates were clarified by centrifugation at 12,000g for 60 min and mixed with the beads coated with 2 mg of GST or GST-VAPA-MSP for 60 min at 4°C. The beads were washed with PBS supplemented with 250 mM NaCl and resuspended with an SDS sample buffer followed by SDS-PAGE. For immunoprecipitation (IP), a total of ~1 × 10⁷ HEK293T cells stably expressing the GFP-STARD3 or Flag-tagged YWHAH were harvested and lysed using the lysis buffer supplemented with protease inhibitor (Roche). Clarified lysate was mixed with prewashed 10 μl of GFP-Trap A beads (Chromotec gta-20) or Anti-DYKDDDDK Affinity Gel (Yeasen 20585ES03) and incubated at 4°C for 2 hours. The beads were then washed twice using high-salt IP lysis buffer (IP lysis buffer supplemented with 500 mM NaCl), with 5-min incubation on a rotor. The final wash was performed using regular IP lysis buffer supplemented with SDS sample buffer.

Data analysis pipeline

Single-particle tracking

Trajectories of particles were extracted using TrackMate plugin (73) of FIJI software. First, the LoG detector was selected, which applied a Laplacian of Gaussian filter to the image. In this step, spurious spots were filtered out depending on the chosen threshold. Second, the simple LAP tracker was selected, which was based on the Linear Assignment Problem mathematical framework. In this step, particle linking relied mainly on the settings of gap-closing max distance. Finally, particle positions at each time point were obtained as csv files, from which particle velocity was estimated for further analysis.

Particle tracking and classification based on MSD

Using TrackMate plugin (73) of FIJI software (74), trajectories of endosomes or lysosomes in time-lapse videos were obtained as csv files. Then, it was analyzed using our customized MATLAB program (The MathWorks Inc., 2017) to obtain the MSD of each trajectory and to further classify particles into confined or directed. Among particles lasting for over 5 frames, MSD was calculated using MSD analyzer (75) with a maximum lag of 20 frames. The following model is used to classify different modes of movement

$$\text{MSD}\left(t\right)=A{t}^{\mathrm{\alpha }}+B$$

in which $\mathrm{\alpha }$ defines the mode as described below (76). After taking logarithm, the positive linear relationship was characterized by least-squares fit. $r^2$ refers to variance. When the anomalous diffusion exponent (α) is less than 0.9 and the coefficient of determination (r²) is greater than 0.8, the mode of movement is classified as confined. When α is greater than 1.2 and r² is greater than 0.8, the movement is classified as directed.

Motion switch analysis

We model the behavior of lysosomes that pause and become confined as undergoing confined diffusion (77). For a specific lysosome, we classify it as in the state of pausing and confinement if they meet the following three criteria. First, its current diffusion coefficient is smaller than 0.01 μm²/s. Second, it stays in this state for at least 20 s. Third, its movement is confined within a radius of 1 μm. We then searched for the nearest time interval in which the lysosome undergoes fast movement, defined by a diffusion coefficient that is three times or higher than the diffusion coefficient in confinement. Now that we have determined the time and location of the lysosome before and after its state switch, we examined the density as well as connection properties of the ER network. We calculated the density of the ER network within a square window of 11 pixels. We then calculated the ratio between the densities of the two regions.

Only particles tracked for more than 15 frames were selected for sequence analysis to reduce random error. The preprocessed coordinates are fed into a deep learning model named Att-BiLSTM (78) to get the motion state sequences of all particles. The state sequences illustrate the motion switching process between different diffusion modes of every particle.

A deep learning model named Att-BiLSTM was used to quantitatively describe the switching of different diffusion modes during particle dynamic motion. The model is a time-series data processing method based on LSTM and HMM-Bayes. We input the motion feature sequence to the model and get the state sequence from the model, which belongs to a many-to-many task with the same length of input and output. There are three layers in our model: BiLSTM layer: generate higher feature representation from the input sequence; local attention layer: produce local attention weights within a window and multiply corresponding hidden layer output to predict the state at time $t$ in the next step; fully connected layer: get the final state sequence using the context vectors.

To analyze movement attributes of endosomes and lysosomes, the movement attributes of particles, which include displacement $\mathrm{\Delta x}$ and $\mathrm{\Delta y}$ in x and y direction, velocity $v_x$ and $v_y$ in x and y direction, movement angle ${\mathrm{\alpha }}_x$ and ${\mathrm{\alpha }}_y$ , etc., were calculated. Statistical analysis of motion states: calculating the start frame ID $f_{\mathrm{s}}$ , end frame ID $f_{\mathrm{e}}$ , and the number of continuous frames of each diffusion mode. Here, we considered Brownian $F_{\mathrm{B}}$ , directed $F_{\mathrm{D}}$ , and confined $F_{\mathrm{C}}$ three diffusion modes. Visualization of particle movement: In the original video, we dynamically use different colors to mark the different diffusion modes in the trajectories of particles for the researcher to conduct an intuitive analysis.

ER morphology analysis

ER image segmentation

The fluorescent image of ER was segmented using the PE-Net deep learning model to extract its morphology. PE-Net has reduced the down-sampling path and expanded stages of convolution compared to the traditional U-Net architecture (79).

Convert ER image to network graph

The graph representation of segmented ER is constructed using a network topological toolkit (80). ER skeleton is extracted from the segmentation result, and a graph construction algorithm is developed to detect the ER junctions and tubules. A junction was defined as a pixel associated with more than two tubules. To further calculate the ER topological complexity, four common junction properties, namely, degree centrality, degree, closeness centrality, and effective size (78), and two customized properties, namely, mesh density and junction density, are calculated from the graph constructed. The degree represents the number of edges connected to a junction. The degree centrality measures the degree of junction normalized by the maximum possible degree of the network, and the closeness centrality is the mean shortest path to all the other junctions. The junction density of a network is defined as the total number of junctions divided by the total pixel length of edges.

Statistical analysis of attribute of ER

We perform further statistical calculations on the morphological properties of ER, including the average intensity $\overline{{\text{intensity}}_{\mathrm{o}}}$ of ER original image, the average intensity $\overline{{\text{intensity}}_{\mathrm{s}}}$ of ER segmented image, percentage $p_{{\text{deg}}_{\mathrm{\omega }}}$ of the node number whose degree ≥ 3, the maximum closeness $\text{max}{\text{closs}}_{\mathrm{\omega }}$ of all nodes in the local window, maximum betweenness $\text{max}{\text{between}}_{\mathrm{\omega }}$ of all nodes in the local window, percentage $p_{{\text{deg}}_{\mathrm{k}}}$ of the node number whose degree ≥ 3 within the KNN, and the average distance $\overline{{\text{dis}}_{\mathrm{k}}}$ between the KNN nodes whose degree >3 and the moving particle.

Local ER complexity analysis

We extract N frames in which particle speed dropped rapidly with the range (0.055 to 0.44 μm/s). Then, we calculate the ER attribute interval values of the corresponding frames. Pearson’s linear correlation coefficient is generated using speed and ER attribute statistical parameters. The pairwise linear correlation coefficient was set to ensure statistically reliable results. Finally, we count the number of particles that meet the correlation requirements under different biological experimental conditions. The Pearson coefficient of the corresponding ER attributes during the change of speed from high to bottom of particles was analyzed.

ER-endolysosome contact analysis

The morphology of ER was segmented from the time-lapse video by a customized convolutional neural network (U-Net). The morphologies and trajectories of endosomes/lysosomes were obtained by a combination of background subtraction and object detection via TrackMate plugin (FIJI software) (73) as TIF files and csv files separately. Contacts were defined as the overlapping signals between the ER and endosomes/lysosomes and marked blue (shown in fig. S2 and movie S3). The contact frequency was defined as the number of lysosomes/endosomes contacting with ER divided by the total number of lysosomes/endosomes in systemically picked frames in the time-lapse video. The formulation is as follows

$$\begin{array}{c}\text{Contact frequency}=\hfill \\ \frac{\text{number of lysosomes or endosomes contacting}\text{ER}}{\text{total number of lysosome or endosomes}}\hfill \end{array}$$

Five frames from each time-lapse video were used for the analysis. Endosomes were counted as always ER-associated if they remained in contact with the ER during every frame of the movie, partly ER-associated if the endosome contacted the ER in some frames of the movie, and not ER-associated if no ER contact was visible.

Stop and fission at ER junction

Particle trajectories were obtained from compressed 30-s images using the Fiji Z stack maximum intensity function. Dot-shaped trajectories indicate stationary particles, noted as “stop,” and line-shaped trajectories are noted as “move.” The nearest distance of a single trajectory to the ER junction was measured. Distance less than 0.2 μm was defined as a stop at ER junction. Similarly, fission events were determined by manual selection, and the position on ER junction was annotated as “yes” if either the docking endosomes or fission site is at the junction. The criteria used to define fission events are as follows: (i) absence of apparent surrounding trajectories that could potentially interfere with the identification of fission events during the process; (ii) morphological transformation of the vesicle, e.g., the appearance of a distinct budding structure; and (iii) following the budding event, the vesicle displays clear motility and separates from the parent structure.

Interaction between particles and ER

The nearest neighbor endosome of a given lysosome was calculated using the KNN algorithm, implemented via the MATLAB function KNN search function. The trajectories of the pairwise candidates were used to further characterize the interaction stability of endolysosome. Pairs that persisted for over 10 s were considered interacting. The motion behavior of the resulting interactions was analyzed by using the mean standard distance (MSD) analyzer and the Att-BiLSTM. The network complexity of the ER around these interactive endolysosome pairs was then analyzed to characterize the ER morphological attributes.

Cluster assay

The distribution of lysosomes and endosome in peripheral region or in whole-cell scale was characterized using the mean shift clustering algorithm, which was implemented by using the python library sklearn.cluster.MeanShift (81). The locations of the particles were estimated using the Trackmate plugin in Fiji. The argument “bandwidth” of the function sklearn.cluster.MeanShift was determined by the function sklearn.cluster.estimate_bandwidth (https://scikit-learn.org/0.16/modules/generated/sklearn.cluster.estimate_bandwidth.html), in which the argument “quantile” ranges from 0.05 to 0.3. We set a relatively large value of the argument quantile for sparse particles. Next, the motion behavior of the particles within and out of the clusters was separately analyzed using the algorithm above.

Lysosome position relative to ER junctions

To assess the spatial relationship between paused lysosomes and the surrounding ER architecture, we calculated the normalized distance of each paused lysosome relative to the nearest ER junction. Specifically, for each paused lysosome, the linear distance between the lysosome and its nearest ER junction was measured along the arc of the ER tubule on which the lysosome resided. This value was then normalized by the total arc length of the corresponding ER tubule segment, yielding a dimensionless value ranging from 0 to 1. A value of 0 indicates that the lysosome is located at an ER junction, whereas a value of 1 indicates that it is positioned at the maximal possible distance from any junction along its tubule.

Microtubule analysis

Microtubule structures were segmented using a threshold-based method. Individual microtubule lengths were measured with the Ridge Detection plugin in Fiji. Microtubule lengths were normalized to the length of the cell’s major axis to account for cell size differences. For orientation analysis, the segmented microtubule images were processed using the Directionality plugin in Fiji to quantify the angular distribution of microtubule alignment.

Supplementary Materials

The PDF file includes:

Figs. S1 to S6

Tables S1 to S3

Legends for movies S1 to S7

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S7