Defining the dynamin-based ring organizing center on the peroxisome-dividing machinery isolated from Cyanidioschyzon merolae

Organelle membranes constantly reshape for distinct types of biological events, including organelle division and the transport of materials by endocytosis or budding organelles. For example, organelles such as mitochondria, plastids and peroxisomes contract a part of their membrane for division. For homeostatic maintenance of plasma membrane proteins, the plasma membrane invaginates and forms endocytotic vesicles. Likewise, transport vesicles pinch off from the organelle membrane. As the membrane reshapes during organelle division and vesicle generation, the membrane must close at the last step through membrane fission. This process leads to elastic stress and requires external force to accomplish it. Numerous molecular genetic studies have thus determined fission factors involved in the fission process of many membranes – these include a member of the GTPase dynamin family in the division of mitochondrion, plastid and peroxisome (Bleazard et al., 1999; Miyagishima et al., 2003; Koch et al., 2003) and in synaptic vesicle endocytosis (Obar et al., 1999; Chen et al., 1991; van der Bliek and Meyerowrtz, 1991). Despite their biological importance, the ultrastructure and molecular mechanism of supramolecular division machinery remain enigmatic because isolation and direct observation of the machinery in vivo is difficult as most of the model organism contains numerous organelles and vesicles that are shaped irregularly and divide randomly. To conquer these problems, light–dark-stimulated organelle division system in simple model organisms, such as the unicellular red alga Cyanidioschyzon merolae, has recently been used. C. merolae contains a minimum set of organelles – i.e. only one plastid, a mitochondrion and a peroxisome (Miyagishima et al., 1999a; Matsuzaki et al., 2004; Imoto et al., 2010). Division of the organelles takes place once per cell cycle and is synchronized by light–dark circadian rhythm stimulation (Suzuki et al., 1994; Miyagishima et al., 2014). Thus, organelle divisions can be observed in a strictly sequential order through the cell cycle (Miyagishima et al., 1999b; Imoto et al., 2011). These unique biological features make C. merolae an ideal model system in which to study organelle division, as illustrated by the first identification and recent successful isolation of the electron-dense organelle division apparatus – the mitochondrion-dividing machinery, the plastid-dividing machinery (Kuroiwa et al., 1993; Yoshida et al., 2006, 2010, 2013) and the peroxisome-dividing (POD) machinery (Imoto et al., 2012, 2013). POD machinery that sustains in vivo structures can be purified to a high degree from cells, and ultrastructural analysis shows that the machinery consists of a closed-ring structure comprising an outer string (dynamin-based ring) and an inner skeletal filament (filamentous ring). Hence, POD machinery is a typical and ideal model system to further understanding of the ultrastructure and molecular mechanisms underlying membrane fission machinery. The motive force for contraction is generated by the dynamin-based ring, which includes the GTPase dynamin-like protein Dnm1 (Imoto et al., 2013). Dnm1 is responsible for the division of the peroxisome and mitochondrion that is mediated by GTPase activity and is highly conserved in plant, yeast and mammalian cells (Bleazard et al., 1999; Smirnova et al., 2001; Sesaki and Jensen, 1999; Koch et al., 2003; Li and Gould, 2003; Mano et al., 2004; Tanaka et al., 2006; Motley and Hettema, 2007). Importantly, Dnm1-associated proteins such as fission1 (Fis1) (Mozdy et al., 2000), mitochondrial division (Mdv1) (Tieu and Nunnari, 2000) and mitochondrial fission factor (Mff) (Gandre-Babbe and van der Bliek, 2008) are also used in both divisions of peroxisomes and mitochondria (Motley and Hettema, 2007; Motley et al., 2008). Moreover, Dnm1 is phylogenetically linked to Dnm2/DRP5B, which is involved in plastid division (Miyagishima et al., 2003; 2008), and dynamin-1, which is involved in synaptic vesicle endocytosis (Hinsaw and Schmid, 1995). Thus, the dynamin-based ring is likely to be a core structure of the membrane fission machineries. However, the mechanism by which the dynamin-based ring is organized and formed around the membrane fission plane is unknown. POD machinery, including the dynamin-based ring, is specifically present during the late M phase (Imoto et al., 2013), although Dnm1 is continuously expressed during the cell cycle (Nishida et al., 2003). Dnm1 resides in the cytosol and relocates to the cytosolic side of the division plane of mitochondrion during the early M phase (Nishida et al., 2003) and to that of the peroxisome during late M phases (Imoto et al., 2013). Hence, the molecular dynamics and assembly of Dnm1 surrounding the division plane at the cytosolic side is apparently important. Understanding how Dnm1 molecules assemble into a ring-shaped supramolecular nanomachinery should provide new insights into the function and molecular mechanics of the membrane fission machinery.

Herein, we successfully isolated Dnm1 in the cytosol and determined the GTP–GDP state of Dnm1 in comparison to that of isolated POD machineries. We provide evidence that turnover between the GTP-bound Dnm1 in the cytosol and GTP-hydrolyzed Dnm1 in the dynamin-based ring is important for maintaining dynamin-based ring homeostasis under ‘semi-in-vivo’ conditions. Furthermore, a semi-in-vivo assay was developed for dynamin-based ring formation, enabling us to directly observe that the dynamin-based ring is formed through unidirectional assembly of Dnm1 from the dynamin-based ring organizing center (DOC) in a manner that is dependent on GTP concentration. Moreover, single or double DOCs are formed on the POD machinery and function as the site at which dynamin-based ring formation begins, and this is likely to be critically important to dynamin-based peroxisomal division. The DOC-mediated dynamin-based ring formation in POD machinery provides the basis for dynamin-based membrane division – including mitochondrial division, plastid division and vesicle endocytosis – among Eukaryota.

To explore the mechanism of dynamin-based ring formation, sequential events of Dnm1 dynamics were observed in detail by using light–dark-stimulated synchronization of the cell cycle. During the peroxisome division period that follows mitochondrial division (Fig. S1A), Dnm1 was observed as cytosolic Dnm1 patches and at the peroxisomal division plane (Fig. 1A). After pharmacological perturbation using oryzalin, an inhibitor of microtubule polymerization that increases the number of dividing peroxisomes (Imoto et al., 2010, 2013) (Fig. S1B), cytosolic Dnm1 patches were barely visible during the peroxisomal division period (Fig. 1B). Moreover, the number of cytosolic Dnm1 patches and the fluorescence intensity of these Dnm1 patches decreased during the peroxisomal division period (Fig. 1C and D), while the fluorescence intensity of Dnm1 at the peroxisomal division plane increased (Fig. 1D). When cells were treated with oryzalin, a decreased number of cytosolic Dnm1 patches and increased fluorescence intensity of Dnm1 at the peroxisomal division plane were observed (Fig. 1C and D). Thus, Dnm1 in C. merolae has two localization sites: (1) the organelle division plane, and (2) cytosolic patches, and these two pools have a quantitative relationship.

To compare the characteristics of Dnm1 localized at the peroxisome division plane with those of Dnm1 in the cytosolic patches, we attempted to determine the nucleotide state of Dnm1 in these two pools. First, an isolation method for cytosolic Dnm1 patches was developed. Upon cell homogenization and Percoll density gradient ultracentrifugation, Dnm1 concentrated in the organelle fractions and not in the cytosolic fraction – even in synchronized cells collected during the G1 phase, at which stage the majority of Dnm1 is localized to the cytosolic Dnm1 patches (Fig. 2A). This indicated that the cytosolic Dnm1 patches are not freely soluble in the cytosol. After ultracentrifugation on an Opti-prep density gradient, a couple of Dnm1-containing fractions (3.0-4.0 ml from the bottom) were shown to be completely free of mitochondrion, peroxisome, Golgi body, ER and lysosome contamination (Fig. 2B, square boxed). Immunofluorescence microscopy showed that Dnm1 in the cytosolic patches isolated from this fraction was localized to a vesicle-like dense structure, which was distinct from the pattern in the POD machinery fraction, and the Dnm1-positive isolated vesicle-like structures were almost the same size as Dnm1 patches observed in whole cells (Fig. 2C). Thus, we termed these fractions as cytosolic Dnm1 patch fractions.

To investigate the biochemical features of the isolated cytosolic Dnm1 patches, we determined the nucleotide state of the cytosolic Dnm1 patch fraction (Fig. 2B) and the isolated POD machinery fraction (Fig. S2A) from the synchronized cells. Quantitative liquid chromatography-coupled tandem mass-spectrometry (LC-MS/MS) analysis based on the genomic database of C. merolae (Matsuzaki et al., 2004) showed that Dnm1 is a major constituent of the GTPase or GTP-binding protein fractions within the isolated POD machinery portions (Fig. S2B). Liquid chromatography and electrospray ionization mass spectrometry (LC-ESI-MS/MS) showed that the molar ratio of GTP to Dnm1 was about six-fold higher in the POD machinery fraction than in the cytosolic Dnm1 patch fraction (91.5±1.2% in the POD machinery fraction; 15.9±1.5% in the cytosolic Dnm1 fraction; mean±range) (Fig. 2D; Table S1, Fig. S2C-E), which suggests that the GTP concentration is important for dynamin-based ring formation. To further support this conclusion, we induced the expression of Dnm1–mCherry or Dnm1-K39A–mCherry, corresponding to K44A of the human dynamin1 mutation (Fig. 2E; Fig. S2G and H), through transient gene expression. The K44A mutation in human dynamin 1 abolishes GTP-binding activity of dynamin and exhibits dominant-negative effects (van der Bliek et al., 1993). Expression of Dnm1–mCherry rescued impaired peroxisomal division through the suppression of endogenous Dnm1 expression (Fig. S2I and J). Immunofluorescence microscopy showed that the morphology of dividing peroxisomes was altered to that of a ‘dumbbell’, and division of peroxisomes in Dnm1-K39A–mCherry-expressing cells was impaired (Fig. 2E and F). The density of the mCherry fluorescence intensity in the cytosol of Dnm1-K39A–mCherry-expressing cells was higher than that of Dnm1–mCherry-expressing cells and was lower at the division plane (Fig. 2G). These results strongly suggest that the GTP-bound form of Dnm1 is important transition of the protein from the cytosolic patches to the division plane.

Nucleotide analysis of Dnm1 showed the involvement of GTP in Dnm1 transition from cytosolic patches to the division plane, which is encircled by a dynamin-based ring. Then, we investigated dynamin-based ring formation around the division plane using the mCherry-fused Dnm1 recombinant protein and isolated dividing peroxisomes from C. merolae. Recombinant Dnm1–mCherry has the same GTPase activity as non-tagged recombinant Dnm1, while Dnm1-K39A–mCherry has significantly reduced GTPase activity (Fig. S3A and B). Dnm1–mCherry or Dnm1-K39A–mCherry were incubated with non-dividing peroxisomes that had been isolated during G1 phase or with dividing peroxisomes that had been isolated during peroxisomal division under GTP or GTP-free conditions (semi-in-vivo assay for dynamin-based ring formation) (Fig. S3C). Immunofluorescence microscopy and immunoblotting analysis revealed that Dnm1–mCherry did not bind to isolated non-dividing peroxisomes, regardless of GTP concentration (Fig. 3A and B). Dnm1–mCherry did bind to isolated dividing peroxisomes, forming a punctate structure on the division plane under GTP-free conditions, and was localized around the division plane upon addition of GTP (Fig. 3C and D). The level of Dnm1–mCherry bound to isolated dividing peroxisomes increased in a GTP concentration-dependent manner (Fig. 3D). When Dnm1–mCherry and isolated dividing peroxisomes were incubated with guanosine 5′-(β,γ-methylene) triphosphate (GMP-PCP), a non-hydrolysable analog of GTP, Dnm1–mCherry bound to isolated dividing peroxisomes around the division plane, as was the case with GTP (Fig. 3E and F; GMP-PCP addition). When Dnm1-K39A–mCherry and isolated dividing peroxisomes were incubated in the presence of GTP, Dnm1-K39A–mCherry apparently bound to isolated dividing peroxisomes in a similar pattern on the division plane to that observed under GTP-free conditions (Fig. 3E and F; K39A+GTP addition). Dnm1-K39A–mCherry binding did not increase under increasing GTP concentrations (Fig. 3F), and essentially the same results were obtained using GDP (Fig. 3E and F; GDP addition). This indicates that GTP-binding is required for Dnm1 assembly around the division plane, but is independent of GTP hydrolysis.

To confirm dynamin-based ring formation, we investigated Dnm1–mCherry assembly on the POD machinery. The POD machinery was isolated from dividing peroxisomes that were bound to recombinant Dnm1–mCherry under various concentrations of GTP. Isolated POD machineries were analyzed by immunofluorescence microscopy and stained with mCherry- and Dnm1-specific antibodies (Fig. 4A). The protein level of Dnm1–mCherry (molecular mass range 100–150 kDa) in the supernatant was found to decrease, whereas in the POD machineries, the level increased with increased GTP concentration (Fig. 4B). The protein level of endogenous Dnm1 (molecular mass range 75–100 kDa) in the supernatant increased, while in POD machineries, the level decreased with increased GTP concentration (Fig. 4B). The total fluorescence intensity of Dnm1 in each ring of the POD machinery (detection of both Dnm1–mCherry and endogenous Dnm1) was constant, regardless of the diameter of the ring or GTP concentration (Fig. 4C). The total fluorescence intensity of mCherry in each ring of the POD machinery (detecting Dnm1–mCherry) increased in a GTP concentration-dependent manner (Fig. 4D). Thus, endogenous Dnm1 molecules are likely to be replaced by Dnm1–mCherry molecules in the POD machinery. Dnm1–mCherry was observed in three types of pattern: (1) one or two puncta where Dnm1–mCherry was localized on the POD machinery; (2) a semi-ring structure where Dnm1–mCherry was localized along part of the POD machinery; and (3) an intact ring structure (Fig. 4E). One or two puncta of Dnm1–mCherry were also observed on the semi-ring or intact ring (Fig. 4E; semi-ring and ring), and endogenous Dnm1 also formed punctate structures (Fig. S3D). Large rings with diameters of 501–700 nm, which corresponded to the diameter of peroxisomes during the contraction phase in microscopy analyses, contained one punctate structure; however, smaller rings with diameters of 300–500 nm, corresponding to the diameter of the peroxisome during the fission phase, contained two punctate structures (Fig. 4F). To understand the relationship between the punctate structures and dynamin-based ring formation, rings were analyzed under various concentrations of GTP. For rings during contraction phase (Fig. 5A), one punctate structure was observed on the tip of the Dnm1–mCherry semi-ring, and the assembly ratio (for explanation, see Fig. 5A) increased in a GTP concentration-dependent manner. Under GTP-free conditions, ∼90% of the Dnm1–mCherry was observed in punctate structures. The frequency of Dnm1–mCherry rings peaked at the assembly ratio group 0.11–0.2 in 0.1 µM GTP and at 0.31–0.5 in 10 µM GTP. With 100 µM GTP, >60% of the Dnm1–mCherry rings exhibited assembly ratios 0.81–1.0. These results indicate that the dynamin-based ring is formed via a punctate structure through assembly of Dnm1 in a GTP-dependent manner. Hereafter, punctate structures are termed DOCs. In rings during the fission phase (Fig. 5B), >90% of Dnm1–mCherry formed one or two DOCs under GTP-free conditions, and two DOCs were detectable on the tip of each of the two Dnm1–mCherry semi-rings in the presence of GTP. The peak of the assembly ratio of the Dnm1–mCherry semi-ring (calculated using the total length of the two Dnm1–mCherry semi-rings, see Fig. 5B) increased with increasing GTP concentration, and >70% of the Dnm1–mCherry rings showed a peak circumference ratio of 0.81–1.0 in 100 µM GTP. These results suggest that the dynamin-based ring is also formed via both DOCs through the assembly of Dnm1 in a GTP concentration-dependent manner in smaller rings with diameters of 300–500 nm. We also examined the effects of GMP-PCP, Dnm1-K39A–mCherry and GDP on dynamin-based ring formation (Fig. 5C). With 100 µM GMP-PCP, >60% (contraction phase) and >70% (fission phase) of the Dnm1–mCherry rings had assembly ratios 0.81–1.0, as observed when incubated with 100 µM GTP. In the presence of 100 µM GTP with Dnm1-K39A–mCherry or 100 µM GDP, DOCs were observed but dynamin-based ring formation was not in ∼90% of POD machinery. Thus, dynamin-based ring formation is not required for GTP hydrolysis and the GTP-bound form of Dnm1 is important.

We next studied the molecular mechanism of the formation of the second DOC in the POD machinery during the fission phase. Both Dnm1–mCherry semi-rings were elongated in the same direction and had almost the same length regardless of the diameter of the rings (Fig. 6A and B). The second DOC forms essentially on the opposite side to the first DOC on the POD machinery (Fig. 6C). These results suggest that the dynamin-based ring is formed from both DOCs with rotational symmetry. We also found that formation of the second DOC depends on the concentration of Dnm1 (Fig. 6D and E). In the presence of 500 nM Dnm1–mCherry, >80% of the POD machinery structures during fission phase were found to contain two DOCs; the rest contained a single DOC. In the presence of 100 nM Dnm1–mCherry, >90% of POD machinery during fission phase were found to contain a single DOC. POD machinery during the contraction phase contained a single DOC regardless of the concentration of Dnm1–mCherry. In the POD machinery structures that had a single DOC during fission phase, when 100 nM Dnm1–mCherry and GTP were added, Dnm1–mCherry semi-ring formation could be completed without the second DOC (Fig. 6F). Moreover, section and cross-section analysis of the POD machinery showed that the first DOC and second DOC form at different layers (Fig. 7A–C). In the presence of 500 nM Dnm1–mCherry, cross-sectional images of the POD machinery showed that the fluorescence intensity of Dnm1–mCherry at the position of the first DOC and that at the second DOC peaked at different positions, which correspond to the upper and lower layers of the POD machinery, respectively (Fig. 7B and C; 500 nM Dnm1–mCherry). Serial cross-sectional images showed that both Dnm1–mCherry rings also formed in different layers (Fig. 7B and D; 500 nM Dnm1–mCherry). In the presence of 100 nM Dnm1, cross-sectional and serial cross-sectional images showed that the fluorescence intensity of Dnm1–mCherry peaked on one side of dynamin-based rings that had been stained with an antibody against Dnm1 (Fig. 7B–D; 100 nM Dnm1–mCherry). Serial cross-sectional images also showed that the thickness of each Dnm1–mCherry semi-ring in the POD machinery during the fission phase in the presence of 500 nM Dnm1–mCherry and of 100 nM Dnm1–mCherry, and in the POD machinery during the contraction phase in the presence of 500 nM Dnm1–mCherry were very similar (Fig. S4A and B). Conversely, the thickness of the dynamin-based ring in POD machinery during the fission phase was observed to be about twice that observed in POD machinery during the contraction phase (Fig. S4C). These results strongly suggest that a second DOC is required for double dynamin-based ring formation in POD machinery during the fission phase.

Here, we investigate dynamin-based ring formation using two methods. One involved isolation of cytosolic Dnm1 patches, and enabled the analysis of the nucleotide state of Dnm1 by ESI-LC-MS/MS. The other method was a semi-in-vivo assay for dynamin-based ring formation, which enabled the observation of sequential events of dynamin-based ring formation on isolated POD machinery. We identified the DOC, which served as an initiation point for dynamin-based ring formation. So far, studies of C. merolae (Nishida et al., 2003; Miyagishima et al., 2003; Yoshida et al., 2006; Imoto et al., 2013) and structural analysis in vitro (Ingerman et al., 2005; Mears et al., 2011) have previously been used to propose that dynamin-based machinery is likely to generate constriction force for membrane fission of the mitochondrion, plastid and peroxisome. Thus, understanding the formation, function and components of dynamin-based machinery is likely to be the key to understanding the mechanism underlying organelle division. Our data, including the identification of the DOC, illustrate how dynamin-like protein Dnm1 molecules assemble into division machinery and how the dynamin-based rings elongate from the DOC on the division machinery during organelle division.

C. merolae contains two dynamin-like protein homologues, Dnm1 and Dnm2, involved in both POD and mitochondrion-dividing machinery (Nishida et al., 2003; Imoto et al., 2013), and in plastid-dividing machinery (Miyagishima et al., 2003), respectively. Dnm2 localizes on small vesicle-like structures in the cytosol (Miyagishima et al., 2003), and these small vesicles localize along the plastid-dividing machinery (Kuroiwa et al., 1998). Thus, the proposed plastid-dividing machinery formation model suggests that Dnm2 is recruited to the plastid-dividing machinery from the cytosol via small vesicles (Kuroiwa et al., 1998). Likewise, Dnm1 localizes on small puncta termed cytosolic Dnm1 patches in the cytosol, and Dnm1 molecules are hypothesized to be recruited to the division plane from the cytosolic Dnm1 patches (Nishida et al., 2003; Imoto et al., 2013). In this study, we showed that the two pools of Dnm1 between the cytosolic Dnm1 patch and peroxisomal division plane are in a quantitative relationship. Dnm1 in the POD machinery has a sixfold higher GTP-binding affinity than that of the cytosolic Dnm1 patches. Dnm1 K39A was mostly localized in the cytosol. Additionally, DOC-mediated dynamin-based ring formation was dependent on GTP concentration but not hydrolysis. Thus, we hypothesize that GTP-binding is crucial for the assembly of Dnm1 from cytosolic patches to the dynamin-based ring. The GTP concentration in the cytosol at the S–M phases does change (Fig. S2F), therefore, the assembly of Dnm1 is likely to be regulated by local concentrations of GTP or, specific adaptor proteins that enhance the GTP-binding affinity of Dnm1, such as Mdv1 (Lackner et al., 2009).

Punctate cytosolic localization of the Dnm1 ortholog, Drp1 (or DLP1), is also reported in mammalian cells (Yoon et al., 1998), and DLP1 is suggested to be recruited to the mitochondrial division plane (Yoon et al., 2003). Thus, the dynamics of Dnm1 between cytoplasmic vesicle-like structures and the organelle division plane is considered to be conserved in eukaryotes. As Dnm1-family members execute both mitochondrial and peroxisomal division (Schrader and Yoon, 2007), it is most likely that cytosolic vesicle-like structures – the cytosolic Dnm1 patches – serve as a ‘stockyard’ for the distribution of Dnm1 to the organelle division planes.

The semi-in-vivo assay for dynamin-based ring formation shows that dynamin-based ring assembly depends on GTP concentration but is independent of GTP hydrolysis. The Dnm1–mCherry ring contains one or two DOCs, discernible at the tip of one side of the semi-rings. The circumference determined from the DOC to the tip of the semi-ring increases with increasing GTP concentration (Fig. 5A and B). Under GTP-free conditions or in the presence of Dnm1 K39A, >80% of the isolated POD machinery structures contained a DOC, but semi-ring structures were not detectable (Fig. 5A–C). This observation indicates that the dynamin-based ring elongates unidirectionally from the DOC. Additionally, under GTP-free conditions, we rarely observed the DOC on the endogenous Dnm1 ring-free POD machinery with diameters of 501–700 nm (Fig. S3E), suggesting that the DOC forms before dynamin-based ring assembly. Taken together, Dnm1 does not require GTP-binding for DOC formation, and the DOC more likely serves as a nucleation site for dynamin-based ring formation. How the DOC is formed is still mysterious, except for the finding of nucleated Dnm1, but there are two hypotheses. One is that Dnm1 adapter proteins are present. For example, mitochondrial division factors Fis1p and Mdv1 (Mozdy et al., 2000; Tieu and Nunnari, 2000) are required for Dnm1-dependent peroxisomal division (Motley et al., 2008), and human Fis1 serves as the Dnm1 recruiter to peroxisomes (Koch et al., 2005). Fis1p and Mdv1 are conserved in C. merolae (Matsuzaki et al., 2004), and there are >20 candidates that could be components of POD machinery (Imoto et al., 2013). Thus, these proteins are strong candidate components of DOC. The other possibility is that endoplasmic reticulum (ER) contact sites play a role. Recently, it has been reported that ER tubules determine the mitochondrial division site (Friedman et al., 2011), and ER–mitochondria contact sites are likely to be important for Drp1 recruitment via the ER-associated formin INF2 and actin polymer complex (Korobova et al., 2013). In C. merolae, a single ER tubule tip stretches to the division plane of the mitochondrion and peroxisome during M phase (Yagisawa et al., 2012). As Dnm1 is used in divisions of both mitochondrion and peroxisome, ER involvement in DOC formation – including in the nucleation of Dnm1 – is likely to be important. Future studies should clarify the mechanism of DOC formation.

The DOC was discernible regardless of the diameter of the POD machinery. Additionally, endogenous Dnm1 disassembles from the POD machinery, and Dnm1–mCherry was incorporated into the POD machinery in the semi-in-vivo assay for dynamin-based ring formation (Fig. 4B–D). In C. merolae, Dnm1 mRNA is constitutively expressed during the cell cycle, regardless of the light–dark time scale (Fujiwara et al., 2009; Miyagishima et al., 2014), and Dnm1 levels remain constant during the cell cycle (Nishida et al., 2005). Thus, for dynamin-based ring formation and homeostasis, the following two aspects are important: (1) supply of the GTP-bound form of Dnm1 from the cytosol to the dynamin-based ring, while internal Dnm1 in dynamin-based ring disassembles; and (2) DOC serves as nucleation site for Dnm1 assembly.

Interestingly, over 95% of POD machinery with diameters of 501–700 nm, corresponding to the contraction phase, contained only one DOC, while POD machineries with diameters of 300–500 nm, corresponding to the fission phase, contained two DOCs. We propose that formation of the second DOC is important for double dynamin-based ring formation during the fission phase. This postulate is consistent with a previous study showing that the dynamin-based ring, stained with Dnm1, becomes thicker during the fission phase (Imoto et al., 2013). Because the inner skeletal filament disassembles during the contraction of the POD machinery, the dynamin-based ring is considered to be closely attached to the membrane and performs membrane fission (Imoto et al., 2013). Thus, we hypothesise that the double dynamin-based ring during the fission phase might promote constriction or affinity to the lipid, and facilitate severing of the membrane of the peroxisome. Interestingly, an in vitro cryoelectron microscopy study showed that yeast Dnm1 forms a ‘two-start’ double helix around lipid tubules, while endocytotic dynamin forms a ‘one-start’ single helix (Mears et al., 2011). Constriction of lipid tubules by Dnm1 is stronger than that by endocytotic dynamin (Mears et al., 2011). Whether the dynamin-based ring contains a double-helix structure is unknown, but serial cross-sectional analysis showed that the fluorescence intensities of double Dnm1–mCherry rings are not completely separated but overlap at the equator region (Fig. 7B and D). Hence, it is possible that the membrane-constricting activity is more effective at the equator of the dynamin-based ring. Interestingly, the mammalian homologue of Dnm1, Drp1, has most recently been reported to split into two during a later stage of mitochondrial constriction, and then endocytotic dynamin-2 mediates mitochondrial membrane fission (Lee et al., 2016). Therefore, the formation of the second dynamin-based ring during the later stage might be conserved in mitochondrial division. On the other hand, genes encoding endocytotic machinery, including dynamin-2, are not encoded by the genome of C. merolae (Matsuzaki et al., 2004). As dynamin-2 is a member of a family different from that of Dnm1 and Dnm2, which are involved in the division of the mitochondrion and peroxisome, and in plastid division, respectively (Miyagishima et al., 2003, 2008; Nishida et al., 2003; Imoto et al., 2013), the other unknown mechanism is likely to be involved in a final step of membrane fission, or the double-dynamin-based ring alters mammalian dynamin-2 function in single-celled eukaryotes.

Results from the semi-in-vivo assay for dynamin-based ring formation led to the proposal of a model for dynamin-based ring formation that enables the correct function of the POD machinery (Fig. 8). Initially, Dnm1 forms a single DOC on POD machinery, independent of GTP-binding. Second, a dynamin-based ring is formed through unidirectional elongation from the DOC by recruiting Dnm1 from the cytosol to the tip of the growing dynamin-based ring. Finally, the second DOC is formed at the side opposite to that of the first DOC, and a double dynamin-based ring assembly originates from DOCs during the later period of contraction, the membrane fission step. Our novel findings uncover the mechanism of formation of the dynamin-based ring containing conserved GTPase dynamin-like protein Dnm1. This study sheds light on the unique mechanisms of organelle division and the dynamin-based membrane fission phenomenon, including endocytosis and vesicle formation in eukaryotic cells.

For synchronization, C. merolae 10D (Matsuzaki et al., 2004) cell cultures were subcultured to 1.0×10⁷ cells/ml as described previously (Suzuki et al., 1994). For the fractionation, cells were treated with oryzalin, and membrane fractions were isolated using a French Pressure Cell (SLM Aminco) and Percoll gradient centrifugation, as described previously (Imoto et al., 2013). After centrifugation, membrane fractions were collected, and the top 10 ml fraction was collected as the cytosolic fraction. The fraction containing cytosolic Dnm1 patches was separated from the membrane fraction using 42.3% Opti-prep dissolved in isotonic buffer and centrifugation at 200,000 rpm for 3 h in a NVT65.2 rotor (Beckman Instruments). Dividing peroxisomes and POD machinery were isolated from the membrane fraction as previously described (Imoto et al., 2013).

To generate an anti-Pex14 antiserum in rabbit, the open reading frame (ORF) of the CMN043C protein, C. merolae Pex14, was amplified by PCR using the following primers: 5′-cgcGGATCCatgcgtgaggactgggtcca-3′ and 5′-cgcCTCGAGcgtcggcagatccgcc-3′ (BamHI and XhoI sites, respectively, are capitalized). The amplified DNA fragment was cloned into pGex6p-1 using restriction digestion at the BamHI and XhoI sites. The recombinant protein was purified on a GSTrap HP column (GE Healthcare) and subcutaneously injected into rabbits for immunization (T.K. Craft corp.). Preparation of the other antibodies used in this study have been described previously, as follows: rabbit anti-Dnm1 antibody (Nishida et al., 2003), guinea pig anti-EF1α antibody (Nishida et al., 2004), guinea pig anti-porin antibody (Fujiwara et al., 2009), rat anti-Sed5 antibody, rat anti-calnexin antibody and rabbit anti-V-ATPase antibody (Yagisawa et al., 2009), and mouse monoclonal anti-mCherry antibody (Clontech). For immunoblotting analysis, the secondary antibodies were alkaline phosphatase (AP)-conjugated IgG (BioRad). For immunofluorescence microscopy, the secondary antibodies were Alexa Fluor 488- and Alexa Fluor 555-conjugated IgG (ThermoFisher).

Cells of C. merolae were fixed and blocked as described previously (Nishida et al. 2003). Phase-contrast and immunofluoresence images were captured using a fluorescence microscope (BX51; Olympus). Section and cross-section images of POD machinery were acquired with a confocal scanning microscope (LSM 710; Carl Zeiss) and analyzed using the ZEN 2012 acquisition software (Carl Zeiss). Immunofluorescence profiles were acquired in ImageJ.

A trypsinized protein mixture of isolated POD machinery was analyzed in an ion trap mass spectrometer (LCQ-Deca, Finnigan) equipped with a nano-LC electrospray ionization source at the Human Proteome Research Center Kyushu University, Japan, as described previously (Matsumoto et al., 2005). Protein identification and calculation of exponentially modified protein abundance index (emPAI) values were performed using the MASCOT (Matrix Science) running on the local server using the C. merolae genome database (http://merolae.biol.s.u-tokyo.ac.jp/).

To extract nucleotides from the fractions of the isolated cytosolic Dnm1 patches and POD machinery, 1.5×10¹⁰ cells of synchronized C. merolae were cultured synchronously in 2× Allen's medium and harvested for fractionation, as described above. To extract nucleotides in the cytosol, 1.0×10⁷/ml C. merolae cells in 500 ml of 2× Allen's medium were cultured synchronously, and 50 ml was sampled at each time point. Cells were harvested, and the pellet was quenched in liquid nitrogen. Fractions were taken as quickly as possible from the isolation tubes and immediately quenched, to halt cellular metabolism, at −30°C in 2.5 times the volume of the fraction with methanol. Intracellular metabolites were extracted by the addition of 1.25× volumes of chloroform. Subsequently, 2 mM raffinose, as a quality control standard used for normalization between the different fractions, was added, and all samples were lyophilized and dissolved in methanol:water (1:1 v/v). After centrifugation at 10,000 g for 5 min, the supernatant was evaporated at room temperature for 6 h and subjected to LC-ESI-MS/MS analysis. LC-ESI-MS/MS was performed using a 4000 Q-TRAP quadrupole linear ion trap hybrid mass spectrometer (AB Sciex) with an ACQUITY UPLC System (Waters). Samples were injected into an ACQUITY UPLC BEH C18 column (1.0×150 mm; Waters) and then directly subjected to ESI-MS/MS analysis. A 10 µl sample volume was separated by step gradient elution with mobile phase A (water, 0.1% formic acid and 0.028% ammonia) and mobile phase B (20% acetonitrile, 0.1% formic acid and 0.028% ammonia) at the following ratios: 95:5 (for 0–5 min), 0:100 (5–25 min), 0:100 (25–30 min) and 95:5 (30–40 min). The flow rate was 50 µl/min at 30°C. The source temperature was 400°C, the declustering potential was −85 and the collision cell exit potential was −11. The ion transitions at m/z 506>79, 426>79, 522>79, 442>79 and 503>179 for ATP, ADP, GTP, GDP and raffinose, respectively, were used in the multiple reaction monitoring mode. Data were analyzed and quantified using Analyst software (AB Sciex). To calculate the molar ratios of GTP to Dnm1, the concentration of Dnm1 in the fractions was determined as shown in Fig. S2D and E and summarized in Table S1.

To construct the plasmid for the expression of the recombinant Dnm1–mCherry fusion protein (pGex6p-1Dnm1mCherry), the Dnm1 (CME019) sequence was amplified from C. merolae genomic DNA with primers 5′-cgcGGATCCatggagcgcctaatacctatcg-3′ (BamHI site is capitalized) and 5′-cgcGAATTCaatctcctctttgacgtgaacg-3′ (EcoRI site is capitalized). The amplified fragment was digested with BamHI and EcoRI (Nippon Gene) and inserted into pGex6p-1 (pGex6p-1Dnm1). The mCherry sequence was amplified from pmCherry C1 with primers 5′-cgcGAATTC atggtgagcaagggc-3′ (EcoRI site is capitalized) and 5′-cgcCTCGAGcttgtacagctcgtccatgcc-3′ (XhoI site is capitalized). The amplified fragment was digested with EcoRI and XhoI (Nippon Gene) and inserted downstream of the Dnm1 sequence in pGex6p-1. To construct the plasmid for the expression of the recombinant Dnm1-K39A–mCherry fusion protein (pGex6p-1Dnm1k39AmCherry), pGex6p-1Dnm1mCherry was amplified with primers 5′-tcgggaGCGtcgagcgtattggagaacg-3′ (alanine codon is capitalized) and 5′-gctcgacgctcccgaggactgggcg-3′. The amplified linear construct with Lys-39 replaced by Ala was recovered using the In-Fusion HD cloning kit (Clontech).

All the PCR products were combined using the In-Fusion HD cloning kit. Dnm1–mCherry and Dnm1-K39A–mCherry sequences were amplified with primers 5′-TCTACCCatggagcgcctaatacctatcg-3′ and 5′-ACGATCTGcttgtacagctcgtccatgcc-3′ (overlap region for In-Fusion cloning is capitalized) from pGex6p-1Dnm1mCherry and pGex6p-1Dnm1k39AmCherry, respectively. Linear construct pTH-2PL (Ohnuma et al., 2009) was amplified with primers 5′-GTACAAGcagatcgttcaaacatttggc-3′ and 5′-GCTCCATgggtagaaacagcacgaatcc-3′ (the overlap region for In-Fusion cloning is capitalized) and combined with amplified Dnm1–mCherry or Dnm1-K39A–mCherry. The combined plasmids were linearized with primers 5′-AGGCTTCatggagcgcctaatacctatcg-3′ and 5′-TTGTTGTGctcgagagcttggcactgg-3′ (the overlap region for In-Fusion cloning is capitalized). The Dnm1 promoter with a 1000-bp upstream region of the Dnm1 ORF was amplified with primers 5′-CTCTCGAGcacaacaaccatgcggatacc-3′ and 5′-GCTCCATgaagcctgtcgcgttcg-3′ (the overlap region for In-Fusion cloning is capitalized). Amplified Dnm1 promoter was combined with linear products of Dnm1–mCherry or Dnm1-K39A–mCherry cloned into a pTH-2PL based vector. To construct the plasmid for the mCherry, pDnm1-Dnm1ORF-mCherry-NOS described above was amplified with primers 5′-CAGGCTTCatggtgagcaagggcgag-3′ and 5′-TCACCATgaagcctgtcgcgttcg-3′ (the overlap region for In-Fusion cloning is capitalized). The PCR product lacks the Dnm1 ORF region, and it was re-ligated using the In-Fusion HD cloning kit. To construct the plasmid for the antisense Dnm1 and Dnm1–mCherry constructs, antisense Dnm1 was amplified from C. merolae genomic DNA with primers 5′-CAGGCTTCctcctctttgacgtgaacgtctc-3′ and 5′-ACGATCTGccaacgcgattatattggctg-3′ (the overlap region for In-Fusion cloning is capitalized). The linear construct contains a 1000-bp upstream region of the Dnm1 ORF, and the NOS terminal region was amplified from Dnm1–mCherry for transient expression, as described above, with primers 5′-GCGTTGGcagatcgttcaaacatttggc-3′ and 5′-AGAGGAGgaagcctgtcgcgttcg-3′ (the overlap region for In-Fusion cloning is capitalized). Amplified antisense Dnm1 was combined with the amplified linear construct (resulting in the plasmid ‘AS Dnm1’). Dnm1–mCherry, including a 1000-bp upstream region of the Dnm1 ORF and the NOS terminal region was amplified from Dnm1–mCherry for transient expression with primers 5′-AAACATGagggttttcccagtcacgac-3′ and 5′-TGTTGTGgctggcgtaatagcgaagag-3′. The linear construct was amplified from pDnm1-AS Dnm1-NOS with primers 5′-ACGCCAGCatgcttggatgcttggatgc-3′ and 5′-CCAAGCATcatgtttgacagcttatcatcgg-3′ (the overlap region for In-Fusion cloning is capitalized). Amplified products were combined with the amplified linear construct (resulting in the plasmid ‘AS Dnm1+Dnm1-mCherry’). To construct the plasmid for the antisense Dnm1+mCherry construct, the pDnm1-mCherry-NOS sequence was amplified from the mCherry plasmid described above with primers 5′-CTAGATCgcggcatcagagcagattg-3′ and 5′-CAAGCATgacgggcttgtctgctcc-3′ (the overlap region for In-Fusion cloning is capitalized). Linear antisense Dnm1 was amplified with primers 5′-AGCCCGTCatgcttggatgcttggatgc-3′ and 5′-CAAGCATgacgggcttgtctgctcc-3′ (overlap region for In-Fusion cloning is capitalized). Amplified constructs were combined (resulting in the plasmid ‘AS Dnm1+mCherry’).

For Dnm1 and Dnm1–mCherry expression, pGex6p-1Dnm1, pGex6p-1Dnm1mCherry and pGex6p-1Dnm1k39AmCherry were introduced into the Escherichia coli XL-1 Blue strain. Transformed XL-1 Blue strain cells were cultured at 37°C for 12 h in 100 ml of Luria Bertani (LB) medium and scaled up to 1 l of LB medium, and further incubated at 37°C for 2 h and at 18°C for 1 h. Isopropyl β-D-1-thiogalactopyranoside (IPTG) was added to a final concentration of 0.1 mM, and cells were harvested after a further 8 h incubation at 18°C by centrifugation at 1000 g for 10 min. Cell pellets were resuspended in 100 ml of HEPES dynamin buffer (HDB150) containing 150 mM NaCl, 20 mM HEPES-KOH pH 7.5, 2 mM EGTA, 1 mM MgCl₂, 1 mM dithiothreitol (DTT) and cOmplete protease inhibitor cocktail (Roche). After homogenization by sonication, the supernatant was filtered and rotated at 4°C with 1 ml of glutathione Sepharose 4B beads (GE Healthcare) for 1 h. The beads were collected and resuspended in 10 ml of HDB100 (containing 100 mM NaCl). Then, beads were loaded onto a 10 ml Poly-Prep Chromatography column (BioRad) and washed with HDBs [10 ml of HDB100, 20 ml complete-free HDB75 (containing 75 mM NaCl) and 20 ml of complete-free HDB50 (containing 50 mM NaCl)]. After washing with HDBs, Sepharose beads were treated with PreScission Protease (GE Healthcare) and eluted.

Assays of GTPase activity were performed at 37°C in reaction mixtures containing 0.5 μM Dnm1, Dnm1–mCherry or Dnm1-K39A–mCherry, and 20 mM HEPES, pH 7.5, 30 mM KCl, 1 mM MgCl₂ and 1 mM DTT with various GTP concentrations (0, 0.1, 1, 5, 10, 100, 200, 300 and 400 μM). Released phosphate was detected by Malachite Reagent (BioAssay Systems), and the absorbance at 600 nm was measured using a microplate reader (Benchmark Plus).

Isolated peroxisomes were dissolved in 200 µl of dynamin-based ring assembly buffer (DAB) (20 mM HEPES pH 7.5, 50 mM KCl, 2 mM EGTA, 1 mM MgCl₂, 1 mM DTT and complete EDTA-free protease inhibitor). 20 µl of the isolated peroxisome solution was incubated at 37°C for 5 min with 0.5 µM Dnm1–mCherry or Dnm1-K39A–mCherry. The reaction was stopped by the addition of 50 mM EDTA at 4°C and was centrifuged at 3000 g for 5 min to separate the pellet (isolated peroxisomes) and the supernatant fractions. For the semi-in-vivo assay for dynamin-based ring formation, the pellet was dissolved in DAB containing 500 mM n-octyl-β-D-glucopyranoside and layered onto 50% Percoll. After centrifugation, the 50% Percoll surface layer was collected and incubated with primary antibodies as used in immunoblotting analysis at 4°C overnight. Then samples were incubated at 4°C for 30 min with secondary antibodies – Alexa-Fluor-488- and Alexa-Fluor-555-conjugated goat anti-rabbit or anti-mouse IgG (ThermoFisher).

We thank Drs. O. Misumi (Yamaguchi University) and F. Yagisawa (Ryukyu University) for C. merolae 10D; and Human Proteome Research Center Kyushu University for LC-MS/MS analysis of POD machinery.