To visualize endosomal vesicles in yolk sac VE cells, an embryonic day 8.5 (E8.5) whole embryo was labeled with a fluorescent membrane probe, FM1-43, and subsequently stained with organelle markers. The embryo was put on a glass-bottom dish, and VE cells were observed by use of confocal microscopy (Figure 1A). These observations revealed that different types of endosomal vesicles were aligned from the apical to the basal area in the supranuclear portion of VE cells, reflecting the repeated fusion and enlargement of vesicles as they migrated deeper into the cell (Figure 1B and C). In the most superficial portion (within 0.5–1 μm of the apical surface; designated as zone 1, Figure 1—figure supplement 1A), fuzzy FM1-43 staining was observed (Figure 1B). The small vesicles in this zone were positive for EEA1 (Figure 1B), indicating that they were early endosomes. Immediately beneath zone 1, many medium-sized round vesicles were observed (zone 2; at a depth of 1–6 μm from the apical surface, Figure 1—figure supplement 1A). Electroporation of EGFP-Rab7 revealed that these vesicles were positive for Rab7, suggesting that they were late endosomes. In the deep portion (zone 3; at a depth of 6–16 μm from the apical surface, Figure 1—figure supplement 1A), large amorphous lysosomes, positive for LysoTracker Red and LAMP1, were observed (Figure 1B; Koike et al., 2009).

Figure 1 with 1 supplement see all

Download asset Open asset

To examine whether these vesicles constitute endocytic pathways, we traced vesicle trafficking in VE cells. After E8.5 whole embryos were pulse-labeled with Alexa 488-transferrin for 5 min, fluorescently labeled endocytic vesicles were tracked over time. The fluorescent signals were initially detected in small vesicles (area <2.0 μm²) in zone 1; 10–20 min after labeling, the signals were mainly detected in medium-sized vesicles (2.0–25 μm²) in zone 2; 40 min after labeling, the signals were strongly observed in large lysosomes (>25 μm²) in zone 3 (Figure 1D and E). The time course of the numbers of labeled vesicles in each zone showed that endocytosed transferrin was rapidly transported from early endosomes to lysosomes via late endosomes in VE cells, as reported previously (Richardson et al., 2000). In addition, rhodamine-labeled dextran, which was transported by pinocytosis pathways, was also transported to lysosomes in a similar fashion (Figure 1—figure supplement 1B).

To examine the dynamic changes of endocytic vesicles, we performed time-lapse imaging of Alexa 488-transferrin- or FM 1–43-labeled vesicles. We focused on late endosomes in zone 2 because these vesicles were large enough and suitable for microscopic observation of membrane fusion processes, as described below.

First, from 5 min after pulse labeling, fusion of two neighboring late endosomes was observed frequently (Figure 2A and B; Video 1). The fusion showed an explosive mode, in which a membrane pore quickly expanded and two vesicles fused to become a single vesicle (Figure 2B). The size distribution of the late endosomes that underwent fusion (7.3±6.6 μm², n=34) was similar to that of the total late endosomes in the observed area at the start of recording (5.7±6.4 μm², n=265; Figure 2C), indicating that late endosomes of any size underwent homotypic fusion. The fusion was completed within 20–50 s (the time required from membrane fusion to single spherical vesicle formation was 31.9±0.2 s, n=37), and the time was weakly correlated with the vesicle size (r=0.60, Figure 2D).

Figure 2

Download asset Open asset

Video 1

Download asset

Second, from 15 min after pulse labeling, some late endosomes shrank gradually and disappeared from the focal plane (Figure 3A and Video 2). Because a previous paper reported that late endosomes were engulfed by lysosomes in E6.5 mouse VE cells (Kawamura et al., 2012), we examined the mechanism of the disappearance of late endosomes in more detail using different approaches. First, we used differential labeling of late endosomes and lysosomes. When whole embryos were incubated with rhodamine-dextran for 20 min and subsequently pulse-labeled with Alexa 488-transferrin for 5 min, lysosomes and late endosomes were labeled with dextran and transferrin, respectively, at the time of observation (15 min after transferrin labeling). In a diminishing late endosome, the fluorescence intensity of transferrin in the vesicle increased initially and then decreased quickly (Figure 3B–D). At the same time, the rhodamine-dextran in the adjacent lysosome entered the late endosome (Figure 3B–D), indicating that the vesicular contents of the late endosomes and lysosomes were mixed. Second, using electron microscopy, we observed pore formation between a late endosome and a lysosome (Figure 3E). These data suggest that the disappearance of late endosomes is a consequence of fusion with the underlying lysosomes, but not of engulfment in E8.5 embryos. This type of fusion, so-called bridge fusion, described in a previous study (Nicot et al., 2006; Roberts et al., 1999), occurred as a result of the slow mixing of vesicular contents through a narrow fusion pore. Actually, heterotypic fusion was slower than homotypic fusion and was completed within 90–250 s (154.8±1.4 s, n=25), and the fusion time (the time required from membrane fusion to disappearance of the vesicle) did not correlate with the vesicle size (Figure 3F). The size of the labeled late endosomes at this time point (10.1±8.1 μm², n=256) was larger than that at 5 min after labeling (6.7±6.2 μm², n=242), indicating that larger late endosomes tended to undergo heterotypic fusion (Figure 3G, see also below). However, the size of the late endosomes that underwent heterotypic fusion (10.2±9.2 μm², n=34) did not differ from that of the total late endosomes at 15 min after labeling (Figure 3G), indicating that the size of late endosomes is not the sole trigger for heterotypic fusion.

Figure 3

Download asset Open asset

Video 2

Download asset

To examine whether the differences in membrane deformation of late endosomes depend on their fusion targets, we adopted a mathematical modeling approach that describes the membrane deformation processes after pore formation. According to thermodynamics, the deformation of a physical object proceeds as its free energy decreases. The free energy of a vesicle consists of two terms: bending energy and osmotic energy. The bending energy, $b$, is the integral of the square of the mean curvature all over the surface (Helfrich, 1973). The osmotic energy is the product of the osmotic pressure difference between the inside and the outside of the vesicle, $\Delta p$, and the vesicle volume, $V$. Thus, the free energy is given by

Note that the bending energy depends only on its shape; in other words, two membranes with the same shape but with different sizes have the same bending energy. On the other hand, the volume depends on both the size and the shape. Therefore, the vesicle size intrinsically affects the balance of both energy sources and may be a critical factor for membrane deformation during fusion.

We focused on a closed membrane having a constricted neck region and introduced two shape parameters, the neck width, $w$, and the area ratio between both sides of the neck, $a$ (Figure 4A), so that the basic processes of the two fusion types are described by the changes in $w$ and $a$ just after pore formation, $w=0$. The relaxation of the neck in explosive fusion is expressed by the increase in $w$. On the other hand, the absorption of smaller vesicles by the larger one in bridge fusion is the decrease in $a$ while $w$ remains small.

Figure 4 with 1 supplement see all

Download asset Open asset

When calculating the free energy as a function of $w$ and $a$, the deformation is expressed as the process of descending the free-energy landscape in the space. Because the shape giving the minimum free energy is most likely to be realized, we first specified the minimum-free-energy shape and then calculated the free energy value for each parameter set.

We considered two extreme conditions: the membrane size is sufficiently small or large. In the former case, the osmotic energy is negligible and the free energy is approximated to

The free-energy landscape on the shape parameter space $\left(w,a\right)$ is shown in Figure 4B. In the latter case, although the osmotic energy is dominant, the bending energy is also not negligible because the bending energy at the neck region becomes considerably large. Thus, the free energy for the large membrane case is

where $B^{neck}$ is the bending energy at the neck. The free-energy landscape for a vesicle with the projected area of $10\left[\mu m^2\right]$ is shown in Figure 4C.

These free-energy landscapes (Figure 4B and C) clearly differ from each other, implying that they drive different deformation processes. In particular, the contour of the $F^{small}$ landscape around w 0 is perpendicular to the $w$ axis, whereas the contour of the $F^{large}$ landscape is parallel to the $w$ axis. Typical orbits of the shape deformations from the initial condition $\left(w,a\right)=\left(\mathrm{0.0,0.3}\right)$ generated by the gradients of the landscape are shown as gray arrows in Figure 4B and C. The orbit in the $F^{small}$ landscape is characterized by a smooth increase in $w$; that is the neck expands quickly (Figure 4D), resulting in explosive fusion. On the other hand, in the $F^{large}$ landscape, $a$ increases through the orbit, whilst $w$ remains small, which evokes bridge fusion (Figure 4E).

The deformation processes observed in our modeling seem to be qualitatively consistent with our experimental results showing that the size of the fusion target changes the fusion modes (Figures 2 and 3). Next, we made a rough estimate of how much of the vesicle size the mathematical model expects to switch the types of fusion. The typical bending energy of a (spherical) vesicle is $b\simeq 2\times {10}^{-18}\left[J\right]$ (Evans and Rawicz, 1990). Referring to the density differences in the dominant ions of endosomes and cytoplasm (Scott and Gruenberg, 2011), we assume that the total ion difference is in the range of ${\displaystyle 1\sim 10mM}$, and that the osmotic pressure difference is ${\displaystyle \mathrm{\Delta }p\simeq 2.5\sim 25\left[Pa\right]}$. From these assumptions, ${\displaystyle 0.22\sim 1.0\left[\mu m^2\right]}$ would be the vesicle size to switch fusion modes. However, the size of the observed late endosomes (5–20 μm² in Figure 3F) is much larger than the predicted size, suggesting that homotypic fusion between late endosomes should show the bridge fusion mode.

To resolve these inconsistencies, we considered the fluctuation effects. Intracellular membranes receive various fluctuations including thermal noise and active fluctuation from the cytoskeletons and motor proteins (Biswas et al., 2017; Chen et al., 2009). Although orbits descending the $F^{large}$ landscape are energetically restricted to keep $w$ small, these fluctuations may force the orbits to deviate from the restricted path (the dashed arrow in Figure 4C); that is they are expected to induce explosive fusion. Then, we performed Monte-Carlo simulations to generate orbits descending the free-energy landscape with various magnitudes of fluctuations and determined the probabilities of orbits that deviate from the path (Figure 4F). When we consider only the thermal noise with room temperature with the estimated osmotic pressure difference (${\displaystyle \mathrm{\Delta }p\simeq 2.5\sim 25\left[Pa\right]}$), as for late-endosome-sized vesicles (${\displaystyle 2.0\sim 25\left[\mu m^2\right]}$), the simulation predicts that vesicles will show bridge fusion (the green-shaded region in Figure 4F). However, if we take into account the fact that a larger fluctuation allows explosive fusion (the blue-shaded region in Figure 4F), fluctuations generated by the cytoskeleton and motor proteins, much larger than thermal noise, may induce explosive fusion.

Previously we found that lysosome biogenesis was regulated by actin dynamics in VE cells (Koike et al., 2009). To examine the relationship between actin filaments and endocytic vesicles, actin filaments were visualized by use of Alexa 488-phalloidin. Endocytic vesicles were visualized by immunostaining with anti-mouse IgG because IgG is endocytosed from the maternal tissue in utero and can be stained without incubation of embryos in vitro for the pulse labeling of endosomes. In VE cells, strong signals were detected to be associated with the cell membrane (Figure 5A–C). In addition, in zone 1, fuzzy cytoplasmic staining was observed (Figure 5A), which appeared to correspond to the actin meshwork in the terminal web. In zone 2, small punctates and short filaments were detected in the cytoplasm (Figure 5B) and were bound to the surface of late endosomes (Figure 5B, B’ and B’’) and infrequently to lysosomes in zone 3 (Figure 5C, C’ and C’’). Optical cross-sectional images and 3D reconstruction showed that several rod-like actin filaments, extending from the apical side of VE cells into the cytoplasm, surrounded and held the late endosomes (Figure 5D–F). These data indicate that most actin punctates observed in the x-y plane correspond to the cross-sectional images of actin filaments extending in the apical-to-basal direction. In addition, when a fusion protein of an actin-bundling protein, fascin-1, with a red fluorescent protein was expressed in VE cells by electroporation, fascin-1 was highly colocalized with the actin punctates, suggesting that most of the actin punctates surrounding the late endosomes were bundled (Figure 5G).

Figure 5

Download asset Open asset

To investigate the endosome-associated actin filaments in more detail, we performed high-resolution imaging analysis using a deconvolution technique, which had resolution comparable to super-resolution microscopy. It demonstrated that fine actin filaments extended radially from the endosomal membranes in the x-y plane, in addition to the actin filaments extending in the apical-basal axis (Figure 5H).

To examine the dynamics of actin filaments associated with late endosomes, E7.5 whole embryos were electroporated with a pEGFP-actin vector, and the EGFP signals in VE cells were observed by use of confocal microscopy 24 hr after electroporation. Because high levels of actin overexpression led to the deterioration of cell functions (see white arrows in Figure 6—figure supplement 1A), cells expressing low levels of EGFP-actin were selected for analysis: those selected cells were normal in morphology and endocytosis (see white arrowheads in Figure 6—figure supplement 1A). Punctate signals of EGFP-actin were seen on the late endosomal membrane similarly to those observed in Alexa 488-phalloidin labeling. Time-lapse imaging showed that some new spots had appeared, others had disappeared, and two actin spots were unified on the membrane surface of endocytic vesicles (Figure 6A). To examine the turnover rate of actin spots in VE cells, we performed fluorescence recovery after a photobleaching (FRAP) assay. After EGFP-actin on a late endosome was photobleached, recovery of its fluorescence was measured (Figure 6B). Recovery of actin spots on late endosomes occurred more rapidly than the control recovery measured at the cell surface (t_1/2 = 29.0 s; Figure 6C–E). Especially, the recovery of actin spots in the apical area of late endosomes was faster than in the basal area (t_1/2 = 11.5 s in the apical area vs. t_1/2 = 18.9 s in the basal area; Figure 6C and E). These results indicated that actin turnover on the surface of late endosomes was dynamically regulated, with a tendency to be more active in the apical area than in the basal area. The time-lapse images showed that when late endosomes moved in an x-y plane in zone 2, spots of EGFP-actin were always at the rear-end of the endosomes along the moving direction (Figure 6F, n=13). Because it is known that Arp2/3-dependent actin polymerization on the membrane acts as the driving force of the motility of intracellular pathogens (Welch and Way, 2013) and endosomes/lysosomes in Xenopus egg extracts (Taunton et al., 2000), we examined the localization of Arp2/3 in VE cells to test the possible role of Arp2/3 in endosome migration. When a fusion protein of Arp3-EGFP was expressed in VE cells by electroporation, the EGFP signals were distributed in dots on the late endosomes and partially colocalized with phalloidin spots in the x-y plane in zone 2 (Figure 6—figure supplement 1B). In addition, Arp3-EGFP signals were always seen at the rear-end of the endosomes along the moving direction (Figure 6—figure supplement 1C). Taken together, these data suggest that the radial branched filaments are temporally polymerized to provide the propulsive force on the membrane for intracellular movement.

Figure 6 with 1 supplement see all

Download asset Open asset

To examine the roles of actin in late endosomal fusion, we tested whether perturbation of actin dynamics caused defects in vesicle fusion. Treatment with cytochalasin D (a cell-permeable inhibitor of actin polymerization) for 5 min significantly changed the distribution of actin filaments. High-resolution imaging analysis using a deconvolution technique showed that phalloidin-positive filamentous structures decreased and instead actin aggregation was observed in the cytosol (Figure 7—figure supplement 1A). The treatment reduced the frequency of homotypic fusion in a dose-dependent manner (Figure 7A and B; Video 3). In addition, it altered the speed and mode of the fusion. Treatment with 0.1 or 1.0 μM cytochalasin D slowed down homotypic fusion (Figure 7C and D; Figure 7—figure supplement 1B). After membrane fusion, two endosomes remained in a peanut shape for a long time, indicating slower neck expansion than that in the control cells (Figure 7—figure supplement 1B). FM 1–43 labeling revealed that the membranes were rapidly fused, whereas it took a long time for the gourd-shaped vesicles to form a large round vesicle (Figure 7—figure supplement 1B). Furthermore, upon 1.0 μM cytochalasin D treatment, about one-third of homotypic fusion showed bridge fusion (Figure 7C and E; Video 4). Therefore, cytochalasin D slowed down the fusion processes and changed the fusion mode to bridge fusion at higher concentrations. These data are consistent with our mathematical model showing that larger fluctuation is necessary for explosive fusion in homotypic fusion of late endosomes (Figure 4F). Treatment with jasplakinolide (a cell-permeable stabilizer of actin filaments) also changed the actin dynamics. After the treatment, the actin meshwork beneath the plasma membrane (zone 1) was denser (Figure 7—figure supplement 2). Furthermore, actin filaments extending radially from the late endosomes were thicker and the fluorescence intensity was increased, although the actin filaments extending from the apical-basal axis were not changed (Figure 7—figure supplement 2). The treatment reduced the frequency of homotypic fusion in a dose-dependent manner (Figure 7A and B; Video 5), indicating that dynamic actin turnover is required for homotypic fusion.

Figure 7 with 4 supplements see all

Download asset Open asset

Video 3

Download asset

Video 4

Download asset

Video 5

Download asset

We next observed the distribution of actin filaments during homotypic fusion. When late endosomes underwent fusion, EGFP-actin punctates were excluded from the membrane docking sites and highly polymerized transiently at both ends of the fusing vesicles (opposite to the docking region) during the rounding period after membrane fusion (Figure 7F). The fluorescence intensity of EGFP-actin on endosomes was quantitated in the proximal, lateral, and distal regions at the predocking, docking, rounding, and postfusion stages. This analysis indicated that actin filament distribution was spatially regulated during fusion (Figure 7G, dark gray bar), whereas FM 1–43 distribution, used as an unchanged control, was uniform on the membrane throughout the rounding stages (Figure 7G, light gray bar). Furthermore, the total fluorescence intensity of EGFP-actin on late endosomes significantly increased transiently after pore formation (Figure 7H), indicating that actin filaments were polymerized temporarily, not simply distributed differently. However, cytochalasin D treatment stabilized EGFP-actin spots and thereby reduced movement of EGFP-actin spots and temporal polymerization during homotypic fusion (Figure 7—figure supplement 3). Moreover, accumulation of Arp3-EGFP in the distal region during the rounding stage was observed similarly to EGFP-actin accumulation (Figure 7—figure supplement 4), and treatment with CK-666, an Arp2/3 inhibitor, resulted in a decrease in the frequency of homotypic fusion (Figure 7B) and slowed down homotypic fusion (Figure 7C and D). These results suggest that branched actin filaments nucleated by the Arp2/3 complex on the distal surface of late endosomes are necessary to transform the vesicle into a large round sphere quickly after membrane fusion of late endosomes.

We next examined the role of actin filaments in heterotypic fusion. Cytochalasin D treatment led to a significant reduction in the frequency of heterotypic fusion in a dose-dependent manner (Figure 8A–C and Video 6). In contrast, jasplakinolide treatment did not inhibit heterotypic fusion and somehow increased the frequency of heterotypic fusion with lysosomes at as early as 5 min after pulse labeling (Figure 8B and Video 7). In addition, CK-666 treatment had no effects, suggesting that Arp2/3 was not required for heterotypic fusion (Figure 8C). These data suggest that actin played distinct roles in homotypic fusion and heterotypic fusion and that stable actin filaments are implicated in driving heterotypic fusion.

Figure 8 with 1 supplement see all

Download asset Open asset

Video 6

Download asset

Video 7

Download asset

Myosins are a superfamily of actin-based motor proteins and their members are grouped into many classes according to their structures and functions (Hartman and Spudich, 2012). To determine which class of myosins is important for vesicle fusion and motility, we examined the effects of inhibitors for several myosin classes that are known to be involved in vesicle transport. We tested inhibitors of myosin I (pentachloropseudilin, PClP), nonmuscle myosin II (NM II) (blebbistatin), myosin V (MyoVin-1), and myosin VI (2,4,6-triiodophenol, TIP; Bond et al., 2013). Heterotypic fusion frequency was reduced to 20% by MyoVin-1 and to a lesser extent, but significantly, by blebbistatin, but was not affected by PClP or TIP (Figure 8D). In contrast, homotypic fusion frequency was significantly reduced by PClP, MyoVin-1, and TIP (Figure 8—figure supplement 1). EGFP-myosin IIA signals overlapped with phalloidin-positive spots on late endosomes (Figure 8E), whereas EGFP-myosin Va signals were juxtaposed to phalloidin-positive spots on late endosomes (Figure 8E).

Late endosomes undergo homotypic fusion almost exclusively at the beginning and then switch to heterotypic fusion over time after endocytosis (Figure 9A). Given that actin plays a different role in homotypic fusion and heterotypic fusion and that pharmacologic changes in actin dynamics shift the fusion mode as shown above, it is suggested that actin dynamics are a critical factor for the transition of fusion modes. Thus, we next decided to look into the regulation of actin dynamics in two different fusion modes. Previously, we found that lysosome biogenesis was regulated by actin dynamics through cofilin downstream of the autotaxin–LPA receptor–Rho–ROCK–LIM kinase pathway in VE cells (Koike et al., 2009). When VE cells were incubated with the S3 peptide, which contains the N-terminal 16 amino acid sequence of cofilin fused to a cell permeation sequence and is used as a competitive inhibitor for LIM kinase and serves as a cofilin activator (Maekawa et al., 1999; Aizawa et al., 2001), lysosomes were fragmented in VE cells and the fluorescence intensity of phalloidin in VE cells was reduced (Koike et al., 2009).

Figure 9 with 1 supplement see all

Download asset Open asset

We thus focused on cofilin in endosomal fusion in VE cells. YFP-cofilin showed small punctate staining that colocalized with phalloidin-positive spots in the x-y plane and extended along the phalloidin signal (Figure 9B and C). When the fluorescence intensity of Alexa 546-phalloidin- and YFP-cofilin-positive filaments was quantitated along the apical-basal axis of late endosomes in zone 2, cofilin was slightly abundant in the apical region, whereas the phalloidin signal was stronger in the basal region (Figure 9D). Note that the possibility that overexpression of YFP-cofilin changed the distribution of actin filaments cannot be excluded. EGFP-actin on a late endosome showed that treatment of E7.5 whole embryos with the S3 peptide (15 μg/ml) for 24 h resulted in a decrease in radially extended actin filaments in the apical area of zone 2, whereas the actin filaments extending along the apical-basal axis appeared unchanged (Figure 9—figure supplement 1). In addition, a FRAP assay showed that the treatment of S3 peptide induced faster turnover of actin filaments of late endosomes when compared with treatment with the RV peptide (Aizawa et al., 2001), which contains the reverse sequence of cofilin and is used as a negative control (t_1/2 = 10.9 s with the S3 peptide, n=9 vs t_1/2 = 18.3 s with the RV peptide, n=9; Figure 9E). These results indicate that cofilin distributes with actin filaments on late endosomes and regulates actin dynamics on late endosomes. Finally, the roles of cofilin in late endosomal fusion were investigated. After treatment with the S3 peptide for 24 hr, Alexa 488-transferrin-labeled vesicles were observed by use of time-lapse imaging. When compared with the RV peptide, the S3 peptide strongly inhibited homotypic fusion (Figure 9F) but not heterotypic fusion (Figure 9G). These data suggest that spatial differentiation of actin dynamics along the apical-basal direction, partially contributed to by cofilin, may be a critical factor for the transition of fusion modes (Figure 10).

Figure 10

Download asset Open asset

We found two types of actin filaments associated with late endosomes in VE cells: those that extended along the apical-basal axis and those that extended radially from the endosomes (Figure 5). Our findings resemble the previous observation that early endosomes were distributed along with actin cables and that Arp2/3-dependent small cloud-like patches protruded from the membrane (Morel et al., 2009). Actin is reportedly distributed around vesicles in different patterns. In addition to the well-known interaction of vesicles with actin rails in directional vesicle transport (Titus, 2018), actin was localized at the vertex ring in vacuole fusion in yeast (Wang et al., 2002) and entirely surrounded phagosomes in macrophages (Liebl and Griffiths, 2009). Advanced microscopy techniques have clarified the fine structure of actin, which is assembled into polyhedron-like lattices around vesicles in exocrine glands (Ebrahim et al., 2019).

We observed strong actin signals as patches on late endosomes in the focal planes on confocal microscopy. Three-dimensional reconstruction showed that most of these patches represent the cross-section of rod-shaped actin filaments extending from the apical side into the cytoplasm in VE cells. In addition to these patches, some signals seem to represent actin filaments extending radially from membranes. Furthermore, we showed that Arp3-positive actin spots appeared in the site opposite to that of endosome movement (Figure 6F, Figure 6—figure supplement 1C) like a comet tail (Goley and Welch, 2006), that Arp3-positive actin spots were asymmetrically distributed and temporarily polymerized in the distal region during endosomal fusion (Figure 7F and G; Figure 7—figure supplement 4), and that Arp2/3 inhibition reduced homotypic fusion (Figure 7B). These results suggest that radial actin filaments polymerized by Arp2/3 generate force for homotypic fusion in VE cells. This notion is compatible with the findings of previous reports that the radial filaments were nucleated and extended with Arp2/3-dependent branches on the lateral membrane of endosomes to push vesicles (Goley and Welch, 2006; Svitkina, 2018). Given that nucleation and Arp2/3 activities are regulated by nucleation-promoting factors (NPFs; Goley and Welch, 2006), the activity of NPFs and membrane fusion processes are likely tightly linked to accurate membrane fusion. Differential regulation of NPF activity may give rise to different patterns of actin surrounding vesicles and membrane dynamics. Future investigation is necessary to elucidate how the NPF activity is spatially and temporally regulated on membranes.

Actin filaments distribute on endosomes in various patterns depending on the cell types. However, why cells show different patterns of actin filament distribution on vesicles and whether different distribution patterns of actin filaments have different functions remain unclear. In this study, by taking advantage of the large size of endosomes, we found that there are two different types of actin filament distribution on the surface of late endosomes in VE cells and that these actin filaments regulate different aspects of membrane fusion, movement, and endosomal distribution. Taken together, these data suggest that different patterns of actin filament distribution on vesicles have various effects on specific functions of vesicles.

In this study, we found that the membranes of late endosomes behave differently depending on the fusion mode. In homotypic fusion, the fusion pore formed between docked endosomal vesicles expanded quickly and the fusion was completed within 20-30 s (explosive fusion, Figure 2). In contrast, in heterotypic fusion with lysosomes, the pore opened very slowly and late endosomes gradually shrank in diameter and disappeared over 1-2 min (bridge fusion, Figure 3). Using mathematical modeling, we showed that vesicle size is an important factor determining the fusion mode. When the size of the vesicle becomes larger, the free energy for neck relaxation decreases and the pore is kept small. Our modeling predicted that the late endosomes are large enough to allow bridge fusion to occur, although they exhibit explosive fusion in vivo. Similarly, homotypic fusion in an explosive mode was observed in very large vesicles, such as vacuoles in yeast, endosomal vesicles in coelomocytes of C. elegans, and secretory granules in the salivary glands (Liu et al., 2016; Masedunskas et al., 2011; Wang et al., 2002). These observations indicate that vesicle size is not the sole factor determining the fusion mode. Lipid composition, a well-known parameter affecting its bending modulus, is also involved, but the effect on fusion mode may be less dominant when compared with the vesicle size effects (Simson et al., 1998; Faizi et al., 2019; Ridolfi et al., 2021). One of the critical factors other than vesicle size that determine the fusion mode is membrane fluctuation. We found that homotypic fusion was converted to bridge-like fusion after cytochalasin D treatment (Figure 7C), which suggests that actin dynamics around vesicles cause fluctuation and affect the fusion modes. This finding can be successfully explained by introducing minor modifications in our mathematical modeling so that a larger fluctuation can shift the fusion mode from bridge to explosive fusion.

VE cells vigorously endocytose maternal proteins and supply the digested products to the developing embryo to support embryonic development. This seems to be why VE cells have such huge endosomal and lysosomal vesicles. However, larger vesicle size leads to the reduction in the vesicle fusion speed and subsequent nutrient trafficking efficiency. If vesicle size were the only factor determining the fusion mode, this disadvantage would occur, but if fluctuation is taken into account, it can be overcome, as shown in our mathematical modeling, and the forces generated by actin are thought to be responsible for increasing the rate of vesicular trafficking. This idea is supported by previous reports that knockout mice of actin-related genes (e.g. Arc and Myh9) showed abnormal VE cells and embryonic lethality (Conti et al., 2004; Liu et al., 2000).

In heterotypic fusion, late endosomes fused with lysosomes in a bridge-like manner. This may be because lysosomes are too large in comparison with late endosomes and actin accumulation on the lysosomal membrane is scarce, resulting in insufficient force of actin for explosive fusion. Bridge fusion was previously observed in live cells only when giant vesicles were artificially generated by gene modification, for example, in endosomes enlarged by expression of constitutively active Rab5 (Roberts et al., 1999) and lysosomes enlarged in the intestinal cells of ppk-3-deleted C. elegans (Nicot et al., 2006). These findings suggest that the enlargement of these vesicles exceeds the intrinsic forces of actin to induce explosive fusion.

Like homotypic fusion, heterotypic fusion is also dependent on actin, as shown by reduction in fusion frequency upon cytochalasin D treatment. Intriguingly, in heterotypic fusion, NM II, myosin V, and stabilized actin filaments, but not Arp2/3, were required, indicating a distinct contribution of actin to heterotypic fusion when compared with homotypic fusion. We showed that, consistent with the well-known roles of myosin V in trafficking of lysosomes, ER, melanosomes, and synaptic vesicles (Hammer and Sellers, 2011), myosin V facilitates the directional transport of late endosomes towards lysosomes along stabilized actin filaments that serve as a rail (Figure 10). Moreover, our results indicate that NM II, which is primarily responsible for generating contractile force (Svitkina, 2018), partially contributes to heterotypic fusion. It is well established that NM II facilitates the constriction of the membrane into the plasma membrane after membrane pore formation in exocytosis (Ebrahim et al., 2019) in cooperation with actin depolymerization by cofilin-1 (Miklavc et al., 2015). In contrast, our data showed that actomyosin appeared not to form a lattice on the surface of endosomes for the contraction and that cofilin did not show a robust influence on heterotypic fusion. Rather, actin stabilization tended to promote heterotypic fusion (Figure 8A and B). These data suggest that NM II may contribute to heterotypic fusion through a different mechanism from that of exocytosis. NM II is also implicated in the movement of the nucleus along with cell migration, which is akin to a ‘conveyor-belt’ by the sliding of the nuclear membrane along the actin filament (Gomes et al., 2005). Further investigation is necessary to elucidate the mechanisms by which NM II transports late endosomes along actin filaments.

Late endosomes, which fuse frequently with neighboring late endosomes, shift the target of fusion towards lysosomes. What triggers the transition of the fusion targets? We found that larger endosomes tend to fuse with lysosomes (Figure 3G). Thus, one possibility is that the larger endosomes located in the basal position have a higher opportunity to fuse with lysosomes than do the smaller endosomes in the more apical portion. However, upon treatment with jasplakinolide, even smaller endosomes distributing in the more apical area underwent heterotypic fusion with lysosomes (Figure 8A and B), thus suggesting rather that the size and the position of vesicles are not the sole factors and that the different dynamics of actin shift the fusion modes. Furthermore, we showed weaker actin polymerization/depolymerization activity in the basal area as compared with that in the apical area (Figure 6C), a decrease in the frequency of homotypic fusion but not of heterotypic fusion induced by S3 peptide treatment (Figure 9F and G), a decrease in radially extending actin filaments by S3 peptide treatment (Figure 9—figure supplement 1), and an increase in actin filaments on endosomes in the basal area as compared with in the apical area (Figure 9D). These findings support the notion that cofilin-dependent regulation of actin polymerization/depolymerization may play a critical role in determination of the fusion target. In parallel with VE cells, it was shown that cofilin was specifically distributed in the apical area of epithelial cells and was highly active under the plasma membrane (Morris et al., 2011; Vitriol et al., 2013). Cofilin regulates actin remodeling for lamellipodia formation and endosomal trafficking (Harrington et al., 2011; Vitriol et al., 2013). Furthermore, early endosomes located close to the plasma membrane fuse frequently when compared with late endosomes located at the perinuclear region (Gruenberg et al., 1989). Taken together, higher dynamics of actin filaments regulated by cofilin at the cell peripheral area may increase the fusion frequency in many cells.

In this study, we developed a method of observing and manipulating late endosomes in VE cells. This method enables, for the first time, the observation of fusion of single vesicles with two different targets and is very useful to examine the molecular mechanisms that determine the fusion targets of vesicles and regulate different fusion modes.

All the experiments using animals were approved by the Animal Care and Use Committee of the University of Tsukuba (#24–248) and of the University of Toyama (A2021ENG-1) and performed under their guidelines. Noon of the day on which a vaginal plug was observed was taken as embryonic day 0.5 (E0.5). Time-pregnant ICR mice were purchased from CLEA Japan and Japan SLC.

For whole-mount immunohistochemistry, E8.5 embryos were fixed with 4% paraformaldehyde (PFA) in PBS at 4 °C for 3 hr and incubated with FITC-conjugated anti-EEA1 (1:100, BD Bioscience) or Cy3-labeled anti-mouse IgG (1:500, Jackson ImmunoResearch). For phalloidin staining, the embryos were fixed with 4% PFA in PBS at 37 °C for 15 min and incubated with 0.2 U/ml Alexa Fluor 488- or Alexa Fluor 546-conjugated phalloidin (Molecular Probes) for 60 min.

The whole embryo culture was performed as described previously (Koike et al., 2009). In brief, embryos were dissected at E7.5 and cultured in 100% rat serum (Charles River) supplemented with 2 mg/ml glucose in a culture bottle placed in a rotation drum culture system (Ikemoto Rika) at 37 °C under 5% O₂, 5% CO₂, and 90% N₂ for 24 hr. In the pharmacologic experiments, embryos were incubated with 15 μg/ml S3 and RV peptides (a kind gift of Dr Kensaku Mizuno, Tohoku University).

Whole embryos with intact yolk sacs dissected out at E8.5 or embryos cultured for 1 d from E7.5 were transferred into DMEM (Sigma-Aldrich) supplemented with 0.1% BSA. Embryos were incubated in 100 nM LysoTracker Red (Molecular Probes) or 12 μM FM 1–43 (Molecular Probes) in 0.1% BSA/DMEM at 37 °C for 5 min. For tracing endocytic vesicles, embryos were incubated with 20 μg/ml Alexa Fluor 488-labeled transferrin (Molecular Probes) or 100 μg/ml dextran rhodamine B 70,000 MW (Molecular Probes) at 37 °C for 5 min. For differential labeling of late endosomes and lysosomes, whole embryos were initially cultured in 100% rat serum supplemented with 2 mg/ml glucose and 75 μg/ml dextran rhodamine B 70,000 MW at 37 °C for 20 min and then incubated with 20 μg/ml Alexa Fluor 488-labeled transferrin for 5 min. To test the effects of actin inhibitors, embryos were incubated with 0.01–1 μM cytochalasin D (Sigma-Aldrich), 0.01–1 μM jasplakinolide (Calbiochem), 20 μM CK666 (Calbiochem), 10 μM pentachloropseudilin (AOBIOUS), 10 μM blebbistatin (Sigma-Aldrich), 10 μM MyoVin-1 (Calbiochem), and 10 μM 2,4,6-triiodophenol (Tokyo Chemical Industry) for 5 min, and the same drugs were added to the medium during observation.

For high-resolution imaging, the samples on the glass-bottom dish were mounted with ProLong Glass (Thermo Fisher Scientific). Images were captured by use of a TCS SP8 laser confocal microscope (Leica Microsystems) equipped with an HC PL APO CS2 100×/1.40 OIL objective (Leica). Alexa Fluor 488 and Alexa Fluor 594 dyes were excited at 488 and 594 nm of a supercontinuum white light laser, respectively. The emitted fluorescence was sequentially detected by use of HyD-SMD detectors through a spectral separation module (505–550 nm for Alexa Fluor 488 and 605–700 nm for Alexa Fluor 594). The pinhole size was set at 0.40 Airy units (60.7 μm). The pixel size and z-step were set at 45 nm and 182 nm, respectively. The acquired images were processed using the Lightning deconvolution algorithm on LAS-X software (Leica). Three-dimensional reconstructions of the images were carried out on NIS-Elements software (Nikon) or Imaris software (Bitplane).

An embryo was placed on a glass-bottom dish (No.1S thickness, Matsunami Glass), and confocal fluorescent images were obtained by use of an LSM510 laser scanning microscope (Carl Zeiss) equipped with a C-APOCHROMAT 63 x/1.2 W Korr objective lens (Carl Zeiss) and a pinhole set to 1.0 AU. Images of zone 2 and zone 3 were taken by focusing on the z-axis where late endosomes and lysosomes occupied the largest area in the observation field, respectively. To compare endosomal sizes, the area of all intact endosomes in each image was measured by use of Image J. Time-lapse images were obtained by collecting 40 images with time intervals of 5 or 10 s at room temperature. The series of images of the confocal z-stacks were obtained with a lateral resolution of 1024x1024 pixels and a scanning depth of 6–8 μm with 13–17 layers and 0.5 μm intervals. To quantify fusion frequency, the number of membrane fusion events in the 400 s time-lapse images was counted, and the number of fusions per minute a single late endosome underwent was calculated. To compare the time required for single fusion processes, fusion time was defined as follows. For homotypic fusion, fusion time was defined as the time from the start of membrane fusion of two late endosomes until the formation of a single, round vesicle. For heterotypic fusion, fusion time was defined as the time from when the late endosomes began to shrink until they disappeared as the result of absorption by a lysosome.

The electroporation of whole embryos was described previously (Koike et al., 2009). An E7.5 embryo was inserted into a small hole of an agarose mold (2% agarose/PBS, 7x7 mm) placed in Hanks buffered saline solution (HBSS). The solution bathing the embryos was then replaced with HBSS containing 2 mg/ml DNA. Three square electric pulses (30 V, 50ms duration, 1 pulse/s) were delivered to the agarose mold by use of an electroporator (CUY21, Nepa Gene) and a 10 mm electrode (CUY650P10, Nepa Gene). After electroporation, the embryo was rinsed with HBSS and cultured as described in the previous section. Rab7, nonmuscle myosin IIA, and myosin Va subcloned into a pEGFP-C2 expression vector (Clontech), Arp3 subcloned into a pEGFP-N1 expression vector (Clontech), fascin-1 subcloned into a pmRFP-C1 expression vector (Clontech), pEGFP-Actin (BD Biosciences), and YFP-cofilin (kind gifts of Dr Kensaku Mizuno) subcloned into the pEGFP-C1 vector (Clontech) were used. For cloning, following primers were used: myosin IIA Fwd, 5’-ttcgaattctgcatggctcagcaggctgcag-3’; myosin IIA Rev, 5’-cagaattcctattcagctgccttggcatc-3’; myosin Va Fwd, 5’-ttcgaattctgcatggccgcgtccgagctctac-3’; myosin Va Rev, 5’- cagaattctcagacccgtgcgatgaag-3’; fascin-1 Fwd, 5’-ttcgaattctgcatgaccgccaacggcacg-3’; fascin-1 Rev, 5’-cagaattcctagtactcccagagtgagg-3’; Arp3 Fwd, 5’-tcgaattcatggcgggacggctgcc-3’; and Arp3 Rev, 5’-gtggatcccgggacatgactccaaacactg-3’.

Photobleaching analysis was performed with the LSM510 laser scanning microscope. The EGFP-actin-electroporated embryo was observed by use of a 63 x objective lens. For the bleaching, an ROI was drawn around a late endosome (4 μm x 4 μm) or cell boundary and was bleached by use of a 488 nm laser beam for 8 bleach iterations at 100% intensity, and fluorescence recovery into the bleached area was monitored. To calculate the percentage of recovery after photobleaching, the intensity profiles for the actin filaments in the bleached area, a background region, and the whole cells were collected. The background was subtracted from all the data, and the intensity of the whole cells was used to correct for fluorescence loss during image acquisition.

E8.5 mouse embryos were fixed with 2% glutaraldehyde and 2% PFA in 0.1 M phosphate buffer (pH 7.4) overnight. After three washes with 0.1 M phosphate buffer, pH 7.4, the embryos were postfixed with 1.0% osmium tetroxide in 0.1 M phosphate buffer for 2 hr, dehydrated, and embedded in Epon 812 according to a standard procedure. Ultrathin sections were stained with uranyl acetate and lead citrate and observed by use of an electron microscope (H-7100, Hitachi).

In general, the free energy governing the shape relaxation process of a vesicle (closed lipid membrane) consists of 2 terms: the bending energy and the osmotic energy,

The bending energy, $b$, is given by

where c is the mean curvature defined at each point of the surface of the vesicle and the integration is performed all over the vesicle. $\kappa$ is the bending modulus (Helfrich, 1973). The mean curvature has the dimension of the inverse of the length. Suppose a similarity transformation of a membrane with the length scaling ratio $\lambda$, the mean curvature of the transformed membrane is rescaled by $1/\lambda$ and the surface area is rescaled by ${\lambda }^2$. Thus, the bending energy is invariant under a scale change in size.

The osmotic energy is the product of the osmotic pressure difference between the inside and the outside of the membrane, $\Delta p$, and the volume enclosed by the membrane, $V$. The volume depends on both the size and the shape of the membrane. Here we represent the size of a vesicle with its area $A$, since it is invariant under shape changes, and we introduce a rescaled volume $v$ ($v=0\sim 1/6\sqrt{\pi }$) with

Thus,

In the above form, a similarity transformation only changes the scale factor term $A\sqrt{A}$. $b$ and $v$ depend only on the shape and are invariant under a scale change in size.

In this study, we consider a process in which 2 spherical vesicles with different areas connect and the integrated closed membrane proceeds shape relaxation. At the onset of the connection, the membrane shape has rotational symmetry for the axis through the centers of the mass of 2 vesicles and the connection point. We assume that the succeeding shape relaxation also holds the rotational symmetry, because some sort of asymmetric force is necessary to violate the rotational symmetry, which is a more complicated problem than can be discussed here.

For further simplification, we only consider membrane shapes that have a constricted neck region, and introduce 2 shape parameters: the neck width, $w$, and the area ratio between both sides of the neck, $a$. The constricted neck region is the trace of the connection event and the time evolution of the 2 parameters expresses the approximate deformation processes of 2 vesicle fusions as follows. Just after fusion, $w=0$ and $a$ is the area ratio between the 2 vesicles. The relaxation of the neck (“explosive” fusion) is expressed by the increase in $w$. The absorption of smaller vesicle by the larger one (“bridge” fusion) is expressed by the decrease in $a$, while $w$ remains small.

Membrane shapes confined by the shape parameters still display a variety. We specify the typical shape of the membrane for a given set of shape parameters $\left(w,a\right)$. The typical shape must be more energetically stable than atypical shapes having the same shape parameter values. We choose the shape having the minimum free energy as the typical shape. Then, we obtain the free energy of the typical shape for each set of shape parameters and display the free-energy landscape on the two-dimensional shape parameter space.

Suppose a considerably small membrane, which is described by small $A\sqrt{A}$, the free energy is approximated by

To generate rotationally symmetric shapes and to evaluate the bending energy, we consider a curve in the radial and axial space and parametrize it by the function of angle, $\theta$, against the arc-length along the curve, $s$ (Figure 4—figure supplement 1A). Rotating the curve around the axis gives a membrane shape. The mean curvature is $\frac{d\theta }{ds}+\frac{sin\left(\theta \right)}{r}$ (Seifert et al., 1991) and the bending energy for the shape is $b=2\pi \kappa {\int }_0^{s_1}\frac{r}{2}{\left\{\frac{d\theta }{ds}+\frac{sin\left(\theta \right)}{r}\right\}}^{2}ds.$

We independently determine the shapes giving the minimum bending energy for the right and left parts of the membrane with an additional boundary condition: a smooth connection between the angles at the boundary ($\theta =\frac{\pi }{2}$ at the boundary).

Here we suppose a membrane with the radius of opening $\overline{w}$, area 1, and give the free energy

With the Lagrange function

where $\gamma$ is a Lagrange multiplier. Then the Euler-Lagrange equations for the system are

Solving these equations with boundary conditions ${\left(\theta ,r\right)}_{s=0}=\left(\mathrm{0,0}\right)$, ${\left(\theta ,r\right)}_{s=s_1}=\left(\frac{\pi }{2},\overline{w}\right)$ and $2\pi {\int }_0^{s_1}rds=1$, we get the rotationally symmetric shape having a minimum bending energy $\theta ={\theta }_{min}\left(s;\overline{w}\right)$.

We calculated the minimum energy shape for $\overline{w}\in \left[0,1/\sqrt{2\pi }\right]$, and plotted the bending energy against the radius of the opening. The function is well fitted with a hyperbolic curve

with $\lambda \simeq 7.467$ (Figure 4—figure supplement 1C). Thus, we generate the minimum-bending-energy shape for a given set of shape parameters by combining two rescaled partials spheres with scaling ratios $a$ and $1-a$, and we get the free energy as a function of $\left(w,a\right)$

Contrary to the small membrane, a considerably large membrane is described by a large $\Delta P$. We tentatively approximate the free energy of the large membrane with

The equation indicates that the minimum-free-energy shape corresponds to the shape with the largest volume for the given shape parameters $\left(w,a\right)$. Moreover, we independently determine the shapes giving the maximum volume for the right and left parts of the membrane across the constricted neck region, since the area ratio and the size of the boundary (the constricted neck region) are determined by shape parameters. The shape of each part of the membrane corresponds to a partial sphere (Figure 4—figure supplement 1B).

A general partial sphere with the radius of opening $\tilde{w}$, area 1, and the volume $\tilde{v}$, is given by

where, $\alpha$ and $\psi$ are the radius and angle of the partial sphere, respectively (Figure 4—figure supplement 1B). Solving the volume as a function of ${\displaystyle \tilde{w}}$, we get

We generate the maximum volume shape for a given set of shape parameters by combining two rescaled partials spheres with scaling ratios ${\sqrt{a}}^3$ and ${\sqrt{1-a}}^3$, and the volume of the shape is calculated by

Therefore, we get the free energy as a function of $\left(w,a\right)$

The resulting shape of the membrane has a sharply bent region at the neck, which stores large bending energy. Moreover, at the onset of fusion, the osmotic energy has no gradient in the $w$ direction

Thus, the above approximation of neglecting the bending energy for large membranes is not always valid. To draw the time evolution of $w$, we have to incorporate the bending energy as a modification term.

Suppose that the neck region is curved with radius $R_0$ (Figure 4—figure supplement 1D and E) which is considerably smaller than the actual neck width $W=\sqrt{A}w$, the bending energy of the neck region is

with $s=r_0\theta ,\xi =\pi -\theta ,\partial \xi /\partial s=\partial \xi /\partial \theta \cdot \partial \theta /\partial s=-1/r_0$, and the volume change by introducing the curved neck region is negligible. The coordinate system of the calculation is shown in Figure 4—figure supplement 1D and E. Using the relation $W\gg R_0$, the bending energy becomes

Since $R_0$ is assumed to be the minimum curvature radius determined by the physical properties of the membrane, it should be independent from the scale change in size (in other words, the thickness of the neck is fixed under the scale change). Therefore, the bending energy of the neck region depends on the size. The right-hand side of the above equation is decomposed into the size-dependent parameter and the size-independent function of $\left(w,a\right)$,

Finally, we get the modified free energy for the large membrane

Since $\Gamma \propto \sqrt{A}$, the contribution of the osmotic energy is dominant for the large membrane except the region where the slope of the osmotic energy vanishes.

Monte-Carlo simulations on $\left(w,a\right)$ space were performed with free energy $F^{large}\left(w,a\right)$. We assumed that active fluctuations caused by nonequilibrium processes or activities of motor proteins are modeled by thermal fluctuations with shifted temperature (effective temperature) (Ben-Isaac et al., 2011). Although this system has three parameters—the size of the vesicle $\sqrt{A}$, osmotic pressure difference $\Delta p$, and energy of the effective temperature $\Xi$—it is invariant under the transformation $\left(\sqrt{A},\Delta p,\Xi \right)\to \left(\lambda \sqrt{A},\Delta p/{\lambda }^2,\lambda \Xi \right).$ It means that the essential dimension of the parameters is two. Thus, we chose two rescaled parameters $\Delta pV/\Gamma$ and $\Xi /\Gamma$ with $V=4A\sqrt{A}/3\sqrt{\pi }$, $\Gamma =4\pi \kappa \sqrt{A}/R_0$ to be changed. With the initial condition $\left(w,a\right)=\left(\mathrm{0.0,0.3}\right)$, Monte-Carlo simulations are performed 1000 times for each set of rescaled parameters using the Metropolis method and the frequency of neck expansions in the courses of shape relaxations was calculated. Neck expansion was judged when $w=0.2$.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

This work was supported in part by KAKENHI (17024006) and the 21st Century COE program from the MEXT, Japan, KAKENHI (JP22H04926), Grant-in-Aid for Transformative Research Areas – Platforms for Advanced Technologies and Research Resources “Advanced Bioimaging Support” from JSPS, Japan, and Takeda Science Foundation. We thank Drs K Mizuno, K Ohashi, and N Ohsumi, for reagents, protocols, and advice; Y Sato (Carl Zeiss, Japan) for help in 3D reconstruction; and H Bito and F Miyamasu for useful comments on the manuscript.

All the experiments using animals were approved by the Animal Care and Use Committee of the University of Tsukuba (#24-248) and of the University of Toyama (A2021ENG-1) and performed under their guidelines.

1. Preprint posted: December 18, 2023
2. Sent for peer review: February 2, 2024
3. Reviewed Preprint version 1: March 20, 2024
4. Reviewed Preprint version 2: September 11, 2024
5. Version of Record published: October 23, 2024

You can cite all versions using the DOI https://doi.org/10.7554/eLife.95999. This DOI represents all versions, and will always resolve to the latest one.

© 2024, Koike et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.