Constriction of the mitochondrial inner compartment is a priming event for mitochondrial division

Mitochondrial division is critical for the maintenance and regulation of mitochondrial function, quality and distribution. This process is controlled by cytosolic actin-based constriction machinery and dynamin-related protein 1 (Drp1) on mitochondrial outer membrane (OMM). Although mitochondrial physiology, including oxidative phosphorylation, is also important for efficient mitochondrial division, morphological alterations of the mitochondrial inner-membrane (IMM) have not been clearly elucidated. Here we report spontaneous and repetitive constriction of mitochondrial inner compartment (CoMIC) associated with subsequent division in neurons. Although CoMIC is potentiated by inhibition of Drp1 and occurs at the potential division spots contacting the endoplasmic reticulum, it appears on IMM independently of OMM. Intra-mitochondrial influx of Ca2+ induces and potentiates CoMIC, and leads to K+-mediated mitochondrial bulging and depolarization. Synergistically, optic atrophy 1 (Opa1) also regulates CoMIC via controlling Mic60-mediated OMM–IMM tethering. Therefore, we propose that CoMIC is a priming event for efficient mitochondrial division.

M itochondria continuously undergo fusion and division to regulate proper function, quality and distribution in response to changing cellular environments 1 . According to the endosymbiosis theory, the outer membrane (OMM) and inner membrane (IMM) of mitochondria have different chemical and molecular properties 2 . Thus, regulation of mitochondrial morphology requires the coordination of distinct machineries for OMM and IMM 3 . In mammalian cells, mitochondrial fusion is mediated by two sequential steps: OMM fusion by mitofusin 1/2 (Mfn1/2), and IMM fusion by optic atrophy 1 (Opa1). In contrast, mitochondrial division is known to be solely driven by dynamin-related protein 1 (Drp1). Drp1 translocates from the cytosol to OMM receptors including Fis1, Mff and MiD49/51 (ref. 4), and forms spiral-like structure by oligomerization 5 . Subsequently, this spiral structure constricts and divides a mitochondrion. However, structural biological studies have revealed that the Drp1 spiral is not sufficient for the initiation of mitochondrial constriction because mitochondrial diameter (0.5-1.0 mm) is much thicker than the Drp1 spiral (B100 nm) 5,6 , implying the requirement for an initial mitochondrial constriction step before Drp1 action. Recent reports have revealed that actin filaments promote initial mitochondrial constriction through direct contact of mitochondria with the endoplasmic reticulum (ER) 7,8 , and dynamin 2 collaborates with Drp1 for finalization of mitochondrial division 9 . This mode of action is similar to endocytosis mediated by dynamin 10 , and recent reports have demonstrated that Drp1 is involved in endocytosis as well as mitochondrial division 11,12 . These findings suggest that Drp1dependent mitochondrial division evolutionarily originated from cytosolic vesicle-scissoring machineries 2 .
Endosymbiosis theory proposes that the IMM is evolutionarily derived from a prokaryote-like endosymbiont, while the OMM is derived from the plasma membrane 2 . Prokaryotic division is executed by FtsZ, a tubulin-like protein, forming intracellular Z-ring 2 . Consistently, primitive eukaryotes, such as red algae, have mitochondrial orthologues of FtsZ, and they exhibit a Z-ring during IMM constriction in the mitochondrial matrix before Drp1-mediated ring formation 13 . This information implies that mitochondria require another step in constriction of the IMM before the activation of cytosolic Drp1. However, neither FtsZlike proteins nor Z-ring structures have been identified in mammals 2 . Nonetheless, constriction or division of the IMM in the absence of Drp1 has been observed in yeast 14 , C. elegans 15 and mouse fibroblasts 16,17 . Also, electron microscopy studies have shown that the IMM can divide in the absence of OMM division 18 . Although these reports support the notion that OMMindependent intra-mitochondrial event(s) may exist, its presence in mammalian cells and the underlying mechanisms have not been elucidated.
Numerous light-scattering experiments using isolated mitochondria have shown that intra-mitochondrial homoeostasis of cations, such as Ca 2 þ , K þ and Na þ , is important for the regulation of mitochondrial volume, causing changes in mitochondrial function 19 . In particular, K þ influx into the mitochondria induces swelling or bulging via osmosis. In addition, mitochondrial dysfunction by mitochondrial depolarization and Ca 2 þ overload induces Drp1-dependent mitochondrial fragmentation, which contributes to mitophagy or apoptosis 20 . In contrast, metabolically active mitochondria are elongated and are not readily fragmented by intracellular stressors, which is the result of enhanced fusion with Mfn1/2 and Opa1 and decreased activity of Drp1 (refs 21,22). Recent papers have reported that elongated mitochondria transiently form bead-on-a-string shapes with fluctuations in intramitochondrial metabolic status, membrane potential, ROS production and Ca 2 þ influx [22][23][24][25] . However, the significance of these morphological changes and their possible association with mitochondrial division has not yet been addressed.
In this study, we report that the transient changes in mitochondrial morphology to bead-on-a-string shape are primarily caused by intra-mitochondrial morphological alterations, referred to as Constriction of Mitochondrial Inner Compartments (CoMIC). While this intra-mitochondrial event occurs at ERmitochondrial contact sites, it appears to be initiated independently of Drp1 action and OMM constriction. We provide evidence that CoMIC is initiated by an influx of Ca 2 þ into the mitochondria, which mediates subsequent mitochondrial influx of K þ and Opa1 activation of the IMM. We propose that this regulatory mechanism can coordinate IMM morphological changes with OMM constriction for efficient mitochondrial division.

Results
Elongated mitochondria exhibit CoMIC before division. To explore the intra-mitochondrial events that take place during mitochondrial division, we performed time-lapse imaging in rat cultured cortical neurons expressing DsRed-mito, which labels the mitochondrial matrix 26 . Interestingly, a subset of long mitochondria in neuronal processes spontaneously exhibited repetitive and reversible beads-on-a-string-like structures accompanying focal constriction and bulging ( Fig. 1a and Supplementary Movie 1). On the other hand, co-expressed green fluorescent protein (GFP) signals were not altered, ruling out the possibility that the observed mitochondrial change is influenced by morphological alteration of neuronal processes. This phenomenon was also observed in other cell types, including 293T, HeLa and HepG2 cells ( Supplementary Fig. 1a), indicating that CoMIC is common to diverse cell types and species. Through kymographic analysis, we temporally and spatially analyzed the occurrence of CoMICs and mitochondrial division. The bulged and constricted sites were spatially conserved during repetitive CoMICs (Fig. 1b), and each mitochondrion exhibited unique and unsynchronized patterns of CoMIC spikes (Fig. 1c). These properties imply that CoMICs may be an intrinsic property of mitochondria. In young neurons (days in vitro 4, DIV4), the probability that a mitochondrion shows a CoMIC in a 10-min time period was 12.3%, and in mature neurons (DIV10), which have more elongated mitochondria 27 , the probability was 35.6% (Fig. 1d). Notably, mitochondria longer than 5 mm have a significantly higher probability of CoMIC (young neurons, 50%; mature neurons, 61.3%) than do shorter mitochondria (young neurons, 7.4%; mature neurons, 10.5%) (Fig. 1e). This indicates that elongated mitochondria are prone to CoMIC. In addition, mitochondria exhibiting CoMIC often underwent subsequent division at only one of multiple constricted sites (Fig. 1a-c). Further quantitative analyses revealed that 54.1 and 61.3% of mitochondria exhibiting CoMIC underwent subsequent division in young and mature neurons, respectively (Fig. 1e), and that larger proportions (61.5% and 81.7%, respectively) of dividing mitochondria exhibited prior CoMIC (Fig. 1f). These results suggest that CoMIC is closely correlated with mitochondrial division.
CoMIC is potentiated by suppression of Drp1. We further examined the association of CoMIC with mitochondrial division by overexpression of a GTPase-defective mutant of Drp1 (DN-Drp1), which inhibits the activity of endogenous Drp1 (ref. 28), resulting in marked elongation of mitochondria with a reduced rate of division ( Supplementary Fig. 2). By using other chemical and genetic dyes, we confirmed that the CoMIC was real but not DsRed-specific artifact ( Supplementary Fig. 3). Strikingly, DN-Drp1-expressing neurons exhibited potentiated frequency and duration of CoMIC compared with control neurons (Fig. 2a and Supplementary Movie 2), although the probability of CoMIC was not affected when compared with elongated mitochondria (Fig. 2b). This phenomenon was also found in other cell types, including 293T, HeLa and HepG2 cells ( Supplementary Fig. 1b), implying that the CoMIC-enhancing effect of DN-Drp1 is conserved in cells of diverse types and species. Kymographic analysis revealed that CoMIC was more frequent and prolonged in the mutant compared with control neurons (Fig. 2c). Further quantitative analyses showed significant increases in total duration, frequency and mean interval of CoMIC in DN-Drp1-expressing neurons compared with control neurons (Fig. 2d). In addition, mitochondrial bulging during CoMIC was also enhanced by DN-Drp1 (Fig. 2e,f). These results show that Drp1 relieves mitochondria from the tension caused by CoMIC by executing mitochondrial division. It is known that DN-Drp1 maintains strong oligomerization activity and affinity for the lipid membrane 29 . In addition, photo-damage or hypoxic stress may induce CoMIC-like morphology resulting from inhibition of Drp1 on the OMM 22,25 . Therefore, we wondered if the overexpression of DN-Drp1 promoted CoMIC via enhanced mitochondrial constriction with perturbation of division. However, knockdown of endogenous Drp1 with a small hairpin RNA (shRNA) similarly promoted CoMIC ( Supplementary  Fig. 4). Crucially, Drp1-knockout mouse embryonic fibroblasts also exhibited CoMIC (Fig. 2g). These data suggest that Drp1 is not required for the initiation of CoMIC, but may terminate CoMIC cycles by mitochondrial division.
ER-mitochondria contact site spatially specifies CoMIC spots. Recent studies revealed that physical association of ERmitochondria and actin filaments can trigger mitochondrial constriction before recruitment of Drp1 (refs 7,8). Our kymographic analyses showed that mitochondrial constriction occurs in spatially conserved sites throughout repetitive CoMIC   (Figs 1e and 2b), implying the presence of structural foci for CoMIC. We thus postulated that spatial association with the ER may specify the position at which CoMIC occurs. As expected, CoMIC and subsequent mitochondrial division occurred when mitochondria spatially intersected with or were enclosed by the ER, as assessed by labelling with GFP-Sec61b (Fig. 3a,b and Supplementary Fig. 5a). DN-Drp1-expressing neurons and 293T cells also exhibited CoMIC at ER-mitochondrial intersections (Fig. 3c,d, Supplementary Movie 3 and Supplementary Fig. 5b-d).
In addition, GFP-Mff, which marks putative sites of constriction and division in ER-mitochondria contact site 7 , localized on constriction sites during CoMIC in control and DN-Drp1expressing neurons (Fig. 3e,f). More recently, it has been demonstrated that replication of mitochondrial DNA (mtDNA) is spatially associated with mitochondrial division in ER-mitochondria contact site 30 . By using GFP-tagged Twinkle which is mtDNA helicase for replication 31,32 , we examined the positions where mtDNA replisomes localize during CoMIC. In off-phase, Twinkle-GFP signals were localized on the elongated mitochondria, and mitochondrial constrictions occurred juxtaposition of Twinkle-GFP during CoMIC (Fig. 3g,h). Therefore, we suggest that structural specification of CoMIC is associated with ER-mitochondrial contact site. Remarkably, however, during CoMIC, the spatial intersections between the ER and mitochondria seems to be maintained regardless of phase, and some intersections did not show any CoMIC (Fig. 3a-d and Supplementary Fig. 5b). However, depletion of Mfn2, which regulates ER-mitochondria contact 33,34 , did not modify DN-   Drp1-potentiated CoMIC ( Supplementary Fig. 6), similar to previous report 7 . These data indicate that contact with the ER per se does not induce CoMIC but most likely plays a role in the structural specification of CoMIC (Fig. 3i). We further examined whether actin filaments are involved in induction of CoMIC because actin polymerization triggers constriction of the OMM before Drp1-dependent division 8 . However, treatment with latrunculin B, which disrupts actin filaments, did not modify CoMIC ( Supplementary Fig. 7). Collectively, these data suggest that the cytosolic vesicle-forming machinery, such as actin filaments and Drp1, are not involved in the induction of CoMIC, while ER-mitochondrial intersections spatially specify the foci of CoMIC.
CoMIC is independent on OMM changes. Because cytosolic fission machinery is not involved in CoMIC, we wondered the behaviour of the OMM during CoMIC. Interestingly, a subset of OMM, which was labelled with Tom20-GFP, in elongated mitochondria exhibited neither constriction nor bulging during CoMIC in control neurons, but underwent subsequent division ( Fig. 4a,b nd Supplementary Fig. 8a). Such dissociation of CoMIC and OMM changes was also observed in 293T cells ( Supplementary Fig. 8b). In some cases, CoMIC and OMM constrictions were simultaneously observed, along with conservation of bulging spots. Similarly, neurons and 293T cells expressing DN-Drp1 exhibited no or marginal OMM bulging during potentiated CoMIC (Fig. 4c,d, Supplementary Movie 4 and Supplementary Fig. 8c). The quantification confirmed that the probability of focal OMM constriction was much lower than that of CoMIC (Fig. 4e). In addition, bulging of the OMM is weaker than that of CoMIC ( Fig. 4f,g), implying the dissociation of OMM and IMM at constriction sites during CoMIC. Supporting this, electron microscopy in shDrp1-infected neuron showed that some elongated mitochondria exhibit focal separation of IMM and OMM (Fig. 4h). To visualize the distance of OMM-IMM during CoMIC, we performed proximity assay by fluorescence resonance energy transfer (FRET) between mitochondria matrix and OMM, labelled by mito-BFP and Tom20-GFP, respectively (Supplementary Fig. 9). We expected that proximity between IMM and OMM may be represented by FRET signal, and found some hot and cold spots were shown on mitochondria ( Fig. 4i). Strikingly, among the hot spots, bulging of mitochondrial matrix occurred at some spots with enhanced proximity between OMM and IMM during CoMIC compared with off-phase ( Fig. 4i,j). After CoMIC, these hot spots returned to their basal level. On the other hand, the cold spots became striking at the constriction sites during CoMIC (Fig. 4i,j), indicating that constriction of mitochondrial matrix increases OMM-IMM distances. Notably, the hot and cold spots appeared to be spatiotemporally conserved with moderate dynamicity during offphase (Fig. 4k). Therefore, these data suggest that CoMIC induces transient disorganization of OMM-IMM contact.
Mitochondrial Ca 2 þ initiates and potentiates CoMIC. Spatial association of CoMIC with ER may indicate the Ca 2 þ -mediated functional linkage between CoMIC and ER. Especially, intra-mitochondrial Ca 2 þ entry is implicated in many mitochondrial functions, including respiration, apoptosis, reactive oxygen species production and mitochondrial transport 35,36 . To investigate involvement of intra-mitochondrial Ca 2 þ in CoMIC, we monitored the level of intra-mitochondrial Ca 2 þ during CoMIC by using CEPIA3mt (ref. 37). Notably, intra-mitochondrial Ca 2 þ was transiently increased just before the CoMIC in DN-Drp1expressing neuron, and returned to basal level in off-phase ( Fig. 5ac, mitochondrion #2). However, some mitochondria did not undergo CoMIC in spite of mitochondrial Ca 2 þ rise ( Fig. 5a-c, mitochondrion #1). It implies that mitochondrial Ca 2 þ signal may be necessary, but not sufficient, for CoMIC. Next, we examined the role of Ca 2 þ in CoMIC. Treatment with BAPTA-AM, a cell-permeable Ca 2 þ chelator, completely inhibited CoMIC in DN-Drp1-expressing neurons ( Supplementary Fig. 10a,b). However, ethylene glycol-bis(b-aminoethyl ether)-N,N,N 0 ,N 0tetraacetic acid (EGTA), a non-cell-permeable Ca 2 þ chelator, did not affect CoMIC ( Supplementary Fig. 10c). In contrast, A23187, which is a Ca 2 þ ionophore, induced robust CoMIC and eventual mitochondrial fragmentation ( Supplementary Fig. 10d). These data indicate that intracellular Ca 2 þ rise can induce CoMIC. In consistent, 2-APB, which is an inhibitor of IP 3 receptor releasing Ca 2 þ from ER, significantly declined the probability of basal CoMIC and shortened its total and average single durations (Fig. 5d,e), implying the contribution of intra-mitochondrial Ca 2 þ shuttling from ER to CoMIC. To more selectively explore the involvement of intra-mitochondrial Ca 2 þ entry in CoMIC initiation, we treated DN-Drp1-expressing neurons with Ru360, an inhibitor of mitochondrial Ca 2 þ uptake. Remarkably, treatment with Ru360 inhibited CoMIC as compared with vehicle (DMSO) treatment ( Fig. 5f,g). Quantitative analyses revealed that Ru360 significantly reduced the probability of CoMIC, and shortened the average duration of an individual CoMIC (Fig. 5j). Furthermore, Ru360 efficiently blocked CoMIC induced by thapsigargin (TG), which increases intracellular Ca 2 þ concentration by blocking calcium uptake via inhibition of sarco/ER Ca 2 þ ATPase ( Fig. 5hj). Similarly, depletion of Mcu by specific shRNA, which is mitochondrial uniporter, decreased the probability of CoMIC and shortened total duration of CoMIC (Fig. 5k,l), although it did not affect mitochondrial morphology ( Supplementary Fig. 11). Taken together, these data suggest that intra-mitochondrial Ca 2 þ entry is required for the induction and potentiation of CoMIC.
Mitochondrial Ca 2 þ promotes efficient mitochondrial division. Previous reports have shown that intra-mitochondrial Ca 2 þ entry is involved in mitochondrial division, although precise molecular mechanism was yet unclear [38][39][40][41] . Since CoMIC occurs before mitochondrial division and is mediated by intra-mitochondrial Ca 2 þ , we tested the Ca 2 þ -mediated functional link between CoMIC and mitochondrial division. First, we monitored the level of intra-mitochondrial Ca 2 þ during CoMIC and subsequent mitochondrial division in control neurons. Similarly to DN-Drp1expressing neurons, control neuron also exhibited spontaneous, transient and dramatic increment of intra-mitochondrial Ca 2 þ just before and during CoMIC, and it returned to basal level during off-phase (Fig. 6a-c). However, there was no remarkable fluctuation of intra-mitochondrial Ca 2 þ during mitochondrial division. These data indicate that intra-mitochondrial Ca 2 þ is involved in induction of CoMIC but not Drp1-mediated mitochondrial division. Next, we examined the effect of intramitochondrial Ca 2 þ on mitochondrial length. Treatment with Ru360 significantly increased mitochondrial length and attenuated TG-induced mitochondrial fragmentation (Fig. 6d,e). Consistently, depletion of Mcu also significantly decreased mitochondrial length (Fig. 6f,g). These data indicate that intra-mitochondrial Ca 2 þ entry promotes mitochondrial division as well as CoMIC.
Considering that mitochondrial division does not always accompany with fluctuation of intra-mitochondrial Ca 2 þ , we suggest that intra-mitochondrial Ca 2 þ indirectly promotes mitochondrial division by promoting CoMIC.
Mitochondrial bulging and depolarization is mediated by K þ . Mitochondrial bulging or swelling is induced by osmotic movement of water molecules into the mitochondrial matrix, and this Intensity Intensity process is mainly regulated by K þ influx controlled by the mitochondrial big-conductance Ca 2 þ -dependent K þ channel (mitoBK Ca ) and ATP-dependent K þ channel (mitoK ATP ), which finally leads to an increase in osmotic pressure between the cytoplasm and the mitochondrial matrix 19 . Considering that intra-mitochondrial Ca 2 þ induces CoMIC accompanied by mitochondrial bulging, we investigated the possible involvement of mitoBK Ca in CoMIC. Pre-treatment with paxilline (Pax), an inhibitor of mitoBK Ca , suppressed the probability of TG-induced, but not basal, CoMIC (Fig. 7a-c), but it did not affect the total duration, frequency or mean interval of CoMIC ( Supplementary  Fig. 12), implying partial involvement of induction/potentiation of CoMIC unlike Ca 2 þ . Instead, Pax efficiently prevented basal and TG-induced mitochondrial bulging during CoMIC (Fig. 6d,e). Previous studies have demonstrated that intramitochondrial influx of K þ induces depolarization of mitochondria as well as bulging 42,43 . Recently, it has been also demonstrated that CoMIC accompanies with mitochondrial depolarization 23,24 . Consistently with these studies, we also found that mitochondria are transiently depolarized during CoMIC (Fig. 6f,g). Remarkably, Pax efficiently alleviated mitochondrial depolarization during CoMIC (Fig. 6f,h). On the other hand, prolonged treatment of NS1619, an opener of mitoBK Ca , induced robust CoMIC and eventual mitochondrial fragmentation in DN-Drp1-expressing neuron ( Supplementary  Fig. 13), although the effect of NS1619 may not be a result from mitoBK Ca opening 44 . Collectively, these data indicate that mitoBK Ca is involved in mitochondrial bulging and depolarization during CoMIC.
Cleavage of Opa1 synergistically regulates CoMIC with Ca 2 þ . Recently, Opa1 has been reported as a coupler between mitochondrial morphology and transient mitochondrial depolarization during CoMIC 24 . Because we found that mitochondrial Ca 2 þ rise is required, but not sufficient for CoMIC (Fig. 5a), we wondered a possible role of Opa1 in CoMIC. First, we knocked down the expression of Opa1 with shRNA (shOpa1), and found that shOpa1 efficiently inhibits CoMIC without affecting DN-Drp1-induced mitochondrial hyper-elongation (Fig. 8a,b and Supplementary Fig. 14). Although the probability of CoMIC was not affected, the total duration and frequency of CoMIC were ARTICLE significantly decreased by shOpa1 (Fig. 8e). In addition, shOpa1 significantly reduced the TG-increased probability of CoMIC as well as total duration and average single duration (Fig. 8c-e).
These data indicate that Opa1 is required for induction and potentiation of CoMIC. We further examined whether other mitochondrial fusion machinery, Mfn1 and Mfn2, is also involved in CoMIC. Single depletion of Mfn1 or Mfn2 by shRNA affected neither DN-Drp1-potentiated CoMIC nor mitochondrial morphology ( Supplementary Fig. 6). Notably, double depletion of Mfn1 and Mfn2 strongly inhibited DN-Drp1-induced mitochondrial hyper-elongation, thereby reducing CoMIC, which occurs in only elongated mitochondria. However, elongated mitochondria also exhibited DN-Drp1-potentiated CoMIC, and it appears that mitochondrial fusion indirectly contributes to potentiation of CoMIC by elongating mitochondria. It has been known that IMM bound long Opa1 (L-Opa1) is cleaved by Oma1, which senses mitochondrial depolarization, resulting in the accumulation of soluble short Opa1 (S-Opa1) in the intermembrane space 45 . Thus, we postulated that Oma1 is upstream regulator of Opa1 in CoMIC. Very similarly to shOpa1, depletion of Oma1 by shRNA significantly reduced, but not probability, the total and average single duration of CoMIC without modifying DN-Drp1-induced mitochondrial hyper-elongation (Fig. 8f,g and Supplementary Fig. 15). It indicates that Oma1-mediated cleavage of Opa1 may play a role in the regulation of CoMIC. Recently, it has been demonstrated that S-Opa1 can induce mitochondrial fragmentation and its GTPasedefective mutant localizes to the foci of constriction and division with Drp1 at the ER-mitochondria contact site 46 . Thus, we further examined the differential involvement of the S-Opa1 and L-Opa1 in CoMIC. We co-transfected neurons with S-Opa1 fusing intermembrane space-localizing signal of apoptosisinducing factor or L-Opa1 lacking cleavage sites with DN-Drp1 ( Supplementary Fig. 16a). While S-Opa1 induced excessive mitochondrial fragmentation in neurons and 293T cells as shown in previous report 46 , its co-expression with DN-Drp1 did not affect mitochondrial hyper-elongation by DN-Drp1 ( Supplementary Fig. 16b,c), indicating that S-Opa1-induced mitochondrial fragmentation requires Drp1. Interestingly, in neurons co-expressing S-Opa1 with DN-Drp1, a greater proportion of mitochondria exhibited CoMIC-like structures (Fig. 8i). Quantitative analyses showed that the probability, total duration and average single duration of CoMIC were dramatically increased by S-Opa1 (Fig. 8h,i,n), indicating that the S-Opa1 sufficiently promotes induction/potentiation of CoMIC. In contrast, L-Opa1 induced mitochondrial elongation in control neurons, but it did not affect in DN-Drp1-expressing  neuron ( Supplementary Fig. 16b,c). Notably, L-Opa1 exhibited inhibitory effects on the induction of basal CoMIC and TGinduced CoMIC (Fig. 8l,m,o), indicating inhibitory role of L-Opa1 in CoMIC. Interestingly, the effect of S-Opa1 was efficiently blocked by ruthenium Red (RuR), an inhibitor of mitochondrial calcium uptake, or Pax (Fig. 8j,k,n). These data suggest that S-Opa1induced/potentiated CoMIC requires intra-mitochondrial influx of Ca 2 þ /K þ . Taken together, these data suggest that intramitochondrial influx of Ca 2 þ /K þ and cleavage of Opa1 are two interdependent pathways for the regulation of CoMIC.

Discussion
In this study, we found that a subset of elongated mitochondria underwent transient and repetitive CoMIC, which was closely associated with subsequent mitochondrial division. So far, CoMIC-like events have been proposed to be associated with mitochondrial physiologies, including mitochondrial uncoupling, ROS production and oxidative phosphorylation [22][23][24] . In this study, we suggest that CoMIC is strongly associated with mitochondrial division, based on that (1) mitochondria undergoing CoMIC have higher probability for subsequent mitochondrial division; (2) mitochondrial division occurs at the CoMIC foci; (3) suppression of Drp1 strongly influences the extent of CoMIC; and (4) conditions blocking CoMIC suppressed mitochondrial division. Furthermore, by inhibiting mitochondrial division, we dissected intra-mitochondrial events mediating the CoMIC; (1) intra-mitochondrial Ca 2 þ flux triggering CoMIC, (2) mitoK Ca -mediated mitochondrial bulging and depolarization and (3) Oma1-mediated accumulation of S-Opa1, which neutralizes Mic60-mediated OMM-IBM tethering (Fig. 10). Therefore, we propose that CoMIC is an intra-mitochondrial priming event for mitochondrial division.
Although we primarily focused on the mitochondria in neuronal processes, CoMIC was also found in multiple cell types including fibroblasts (mesoderm-derived cells) and HepG2 cells (endodermderived cells) as well as neurons (ectoderm-derived cells). Previous reports have also showed CoMIC-like mitochondrial constriction in other many mammalian cells under normal and pathologic conditions [22][23][24][25][49][50][51][52] . Furthermore, this event has also been observed in yeast 13 , C. elegans 14 and even plants 53 . Therefore, we suggest that the fundamental machinery triggering CoMIC may be conserved and operating in most, if not all, eukaryotic cells. Mitochondrial influx of Ca 2 þ plays an essential role in the induction of CoMIC. Interestingly, Ca 2 þ also promotes cellular constriction in prokaryotic cells by enhancing FtsZ polymerization and further bundling of FtsZ polymers 54 . Although mammalian FtsZ ring does not exist, this implies that Ca 2 þ -mediated CoMIC seems to be a fundamental and highly conserved machinery from prokaryotic to eukaryotic cells. Controversial data reported that there was no detectable change in intra-mitochondrial Ca 2 þ concentration during a CoMIC-like event 23 . However, we used more sensitive genetic dye CEPIA3mt (ref. 37), and could detect rise on intra-mitochondrial Ca 2 þ during CoMIC. In addition, Ru360 and depletion of Mcu inhibited DN-Drp1-potentiated CoMIC. Considering that elongated mitochondria can more efficiently absorb Ca 2 þ than short mitochondria in neuronal cells 55 , mitochondrial hyperelongation by DN-Drp1 may amplify intra-mitochondrial Ca 2 þ signals. When Drp1 is expressed normally and mitochondria is endogenously elongated, CoMIC is very transient and the effect of intra-mitochondrial Ca 2 þ influx on CoMIC may be small enough to be virtually undetectable. Furthermore, we showed that blockade of intra-mitochondrial Ca 2 þ entry inhibits efficient mitochondrial division. Consistently with our data, previous pharmacological studies have shown that Ru360 inhibits mitochondrial fragmentation by Ca 2 þ released from the ER by p20, which is a cleaved form of BAP31 by caspase 8 during Fasmediated cell death 39 . Furthermore, inhibition of mitochondrial Ca 2 þ uptake by NIM811 and carbonyl cyanide-ptrifluoromethoxyphenyl-hydrazone, which are a mitochondrial permeability transition inhibitor and a mitochondrial uncoupler, also blocks mitochondrial fragmentation by NMDA-mediated entry of extracellular Ca 2 þ in neurons 39,41,52,56 . Conversely, cells with mutations in the Micu1 gene, which regulates Mcu activity, exhibit excessively fragmented mitochondria with increased intramitochondrial Ca 2 þ (ref. 57). These reports indicate that intramitochondrial Ca 2 þ is critical for efficient mitochondrial division. However, considering that mitochondrial division did not directly accompany intra-mitochondrial Ca 2 þ rise, Ca 2 þmediated CoMIC is a prior event before mitochondrial division. In fact, intracellular Ca 2 þ activates cytosolic vesicle-forming machinery for mitochondrial division. Drp1 is upregulated by the Ca 2 þ -dependent kinase CaMKIa following neuronal depolarization 58 or the Ca 2 þ -dependent phosphatase calcineurin following mitochondrial damage and neuronal outgrowth 59,60 . In addition, Ca 2 þ can activate inverted formin-2, thereby promoting actin polymerization 61 . Therefore, Ca 2 þ mediates and coordinates both of cytosolic and intramitochondrial events for mitochondrial division. According to this model, a substantial increase in cytosolic Ca 2 þ can recruit Drp1 to the mitochondria to complete mitochondrial division, whereas CoMIC occurs when a lower level of Ca 2 þ is released from the ER and taken up into mitochondria.
Our data revealed that mitochondrial bulging during CoMIC is mediated by K þ via mitoBK Ca . Because the concentration of K þ in the cytosol is higher than that in mitochondria, as the IMM is impermeable to potassium ions, mitochondrial influx of K þ across the IMM requires specific channels, mitoBK Ca and mitoK ATP 19 . Opening of these channels induces mitochondrial influx of K þ , resulting in osmotic movement of water molecules into the mitochondrial matrix, thereby causing mitochondrial bulging or  swelling. Potential role of intra-mitochondrial K þ in mitochondrial volume and morphology has been elucidated. Intra-mitochondrial influx of K þ promotes the formation of doughnut-like mitochondria during mitochondrial adaptation to conditions of hypoxiareoxygenation 62 , and is also involved in formation of CoMIC-like structures by depletion of Letm1, which regulates homoeostasis of intra-mitochondrial Ca 2 þ and K þ (ref. 63). Furthermore, inhibition of aquaporin by HgCl 2 suppresses tunicamycin-induced CoMIC and further mitochondrial fragmentation 64 , indicating the involvement of osmosis in mitochondrial division. Notably, the Ca 2 þ -responsive regulatory domain of mitoBK Ca is likely to face the mitochondrial matrix rather than the cytosol 65 , supporting the notion that intramitochondrial Ca 2 þ , but not cytosolic Ca 2 þ , mediates mitochondrial bulging. In addition, we found that DN-Drp1 enhances mitochondrial bulging during CoMIC, and it appears to result from increased Ca 2 þ uptake by mitochondrial hyperelongation 55 . Although previous report demonstrated that H þ leak contributes to mitochondrial depolarization during CoMIC 30,31 , we found that inhibition of mitoBK Ca efficiently suppresses mitochondrial depolarization during CoMIC. Recent study demonstrated the transient mitochondrial pH flash by bulky H þ leakage, which could be induced by intra-mitochondrial Ca 2 þ /K þ influx 66 . Alternatively, the H þ leak may contribute to recovery from CoMIC to off-phase, because efflux of Ca 2 þ and K þ from the mitochondrial matrix are mainly mediated by a mitochondrial Na þ / Ca 2 þ exchanger and K þ /H þ exchanger driven by H þ leak 19 . However, we also found that long exposure to carbonyl cyanide m-chlorophenyl hydrazone (CCCP), which disrupts the proton gradient and induces bulky H þ leak, induces irreversible CoMIC with mitochondrial depolarization (Supplementary Fig. 18). Therefore, although we could not completely exclude that H þ leak may mediate CoMIC, it appears that intra-mitochondrial Ca 2 þ /K þ influx mainly mediates morphological and physiological change during CoMIC. Our data also suggested that Oma1-induced S-Opa1 induces and potentiates CoMIC. Recent studies have highlighted the role of S-Opa1 on mitochondrial division. Unlike L-Opa1, S-Opa1 sufficiently induced Drp1-dependent mitochondrial fragmentation in neurons, in agreement with a previous report 46 . Considering that potentiation of CoMIC in DN-Drp1expressing neurons was further strengthened by overexpression of S-Opa1, it is likely that S-Opa1 promotes CoMIC. However, shOpa1 did not influence the basal probability of CoMIC, although it efficiently suppressed the frequency and duration of CoMIC. Therefore, it is likely that S-Opa1 does not directly constrict the IMM, but is involved in the regulation of CoMIC. Based on our data with previous studies, S-Opa1 may have profission activity. However, S-Opa1 is required for proper mitochondrial fusion 67 , and in vitro IMM fusion can be promoted by S-Opa1 produced by i-AAA protease Yme1l under active oxidative phosphorylation and by Oma1 under mitochondrial depolarization 46,68 . This opposite effect of S-Opa1 could be mediated by disorganization of OMM-IBM contact via interplay between S-Opa1 and Mic60. The effect of S-Opa1 on the CoMIC was strongly modified by alterations of Mic60, which plays a role in tethering of OMM-IBM as core component of mitochondrial contact site and MIB complex 47,48 . Mic60 is evenly localized in IMM 69 , but S-Opa1 is preferentially localized in MIB complex and interacts with Mic60 (refs 47,48). Considering that S-Opa1 and L-Opa1 form oligomer 70 , this model may explain how L-Opa1 inhibits CoMIC: L-Opa1 may inhibit the S-Opa1and Mic60-mediated activation of CoMIC via neutralization of S-Opa1 as a dominant-negative molecule. Therefore, we propose that S-Opa1 is involved in the disorganization of OMM-IBM, whereas L-Opa1 antagonizes this process, thereby contributing to the CoMIC and IMM reorganization.
It is likely that intra-mitochondrial Ca 2 þ -K þ and Oma1-Opa1-Mic60 are two different axes for the control of CoMIC, and both are required for the efficient CoMIC. Therefore, inhibition of either cascades appear to inhibit CoMIC, while we do not entirely rule out the possibility that two cascades have significant interactions. For example, intra-mitochondrial influx of Ca 2 þ induces mild depolarization of mitochondria, leading to Oma1dependent cleavage of Opa1 (refs 40,46). On the other hand, Opa1 can also regulate intra-mitochondrial influx of Ca 2 þ . In digitonin-permeabilized cells, depletion of Opa1 increases intramitochondrial influx of Ca 2 þ in spite of a dissipation of the mitochondrial membrane potential 71 , indicating that Opa1 plays a role in intra-mitochondrial structure for proper calcium uptake. Therefore, mutually dependent interaction between Ca 2 þ and Opa1 could be an important aspect explaining the cyclic nature of CoMIC, and the blockade of any part of this cascade can disrupt the CoMIC cycle.

Methods
Cell culture and gene transfection. Cortices were carefully dissected from Sprague-Dawley rats (Orient Bio and Koatech, Korea) on embryonic day 17 in prechilled Hank's buffered salt solution (Gibco) supplemented with 6% glucose (Sigma) and 50 unit per ml penicillin-streptomycin (PS) (Gibco). Then, the samples were trypsinized and physically dissociated into single neurons. The dissociated neurons (10 5 cells per cm 2 ) were plated on glass coverslips, plates and dishes coated with poly-D-lysine (sigma), and incubated in neurobasal media (Gibco) containing 2% B27 supplement (Gibco), 0.5 mM L-glutamine (Gibco), 25 mM L-glutamate (Sigma) and 50 unit per ml penicillin/streptomycin (Gibco). The neurons were incubated in 5% CO 2 at 37°C. This experiment was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Korea University Institutional Animal Care and Use Committee. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Korea University (Permit Number: KUIACUC20110304-2). After 2 days in vitro (DIV2), we exchanged the medium for media containing no Lglutamate and maintained the cultured neurons in this medium. We transfected plasmids into cultured neurons on DIV2 or DIV8 using the Calphos mammalian transfection kit (Clontech) according to the manufacturer's protocol. After 24-48 h (DIV3-4 or DIV9-10), experiments were performed. Thapsigargin, Ru360 and latrunculin B were purchased from Calbiochem. Paxilline and CCCP were purchased from Abcam. Two-APB, ruthenium red and mito-TEMPO were purchased from Sigma. Drp1-knockout mouse embryonic fibroblasts were obtained from Dr Hiromi Sesaki 17 . 293T (ATCC CRL-3216), HeLa (ATCC CRM-CCL-2) and HepG2 (ATCC HB-8065) cells were incubated with Dulbecco's modified eagle medium (Gibco) supplemented with 10% foetal bovine serum (Gibco) and 50 unit per ml penicillin/streptomycin in 5% CO 2 at 37°C, and gene transfection was performed using Lipofectamine 2000 (Invitrogen, USA) and the Calphos mammalian transfection kit according to the manufacturer's protocols. Experiments were performed 24 h after transfection.
Live cell imaging and microscope image acquisition. Live cell imaging was mainly performed using an inverted fluorescence microscope (Carl Zeiss, Observer Z1) equipped with CoolLED (pE-2) as the light source, a definite focus module, a CCD camera (Photometrics, CoolSNAP fx), and a humidified chamber that maintained cells at 5% CO 2 and 37°C. Primary neurons and other cell lines were incubated in the appropriate culture media during live cell imaging. A Â 63 oil-immersion objective lens (Carl Zeiss, EC Plan-Neofluar, 1.25 NA) was used for image acquisition, and timelapse images were obtained every 5 s for 10 min using the Metamorph (version 7.7.5) imaging program (Molecular Devices). In addition, we used a confocal inverted laserscanning microscope (Carl Zeiss, LSM7 LIVE) equipped with a humidified incubator that maintained cells in 5% CO 2 at 37°C, and images were obtained with ZEN software (Carl Zeiss) with a Â 63 oil-immersion objective lens (Carl Zeiss, Plan-Apochromat, 1.4 NA). High-resolution live cell confocal images were obtained using an inverted laser-scanning microscope (Carl Zeiss, DE/LSM510 NLO) equipped with a humidified incubator that maintained cells in 5% CO 2 at 37°C. A Â 100 oilimmersion objective lens (Carl Zeiss, EC Plan-Neofluar, 1.3 NA) was mainly used with Â 3 digital zoom for image acquisition. Confocal serial sectional images were obtained every 10 s for 10 min using ZEN software (Carl Zeiss). Each time-lapse image consists of four or five sections obtained with a centred Z-section mode. For synchronized imaging of dual channels, we used two mixed lasers, 488 and 536 nm, for emission, and detected excitation using a splitter filter. To avoid excessive photo-damage caused by the LED or laser, we took time-lapse images from one slide within 1-1.5 h after drug treatment. The images were analyzed using the ImageJ program (USA National Institutes of Health). We linearly adjusted the intensity of the images, smoothed the images with 'Gaussian Blur' filtering and reconstructed three-dimensional images using the '3D viewer' module. The reconstructed images were recorded in 360°r otation. For measurement of mitochondrial length, we first fixed primary neurons expressing DsRed-mito with 4% paraformaldehyde in PBS, then stained their nuclei with 1 mg ml À l Hoechst 33342. Then, we captured images using an upright fluorescent microscope (Carl Zeiss Axioskop2 Plus) with a CCD camera (Photometrics, Cool-SNAP cf) using the Metamorph imaging programme. A Â 40 objective lens (Carl Zeiss, EC Plan-Neofluar, 0.75 NA) was used for image acquisition. In addition, we used a confocal inverted laser-scanning microscope (Carl Zeiss, LSM700). A Â 40 water-immersion objective lens (Carl Zeiss, LD C-Apochromat, 1.1 NA) was mainly used with Â 3 digital zoom for image acquisition. FRET images were acquired by confocal microscopy (Carl Zeiss, LSM700) using microscope filter combination for BFP and FRET: 405 nm excitation, 0-415 nm and 415-735 nm emission filters, respectively. Images were analysed using ImageJ software.
Kymographic and quantitative analyses of CoMIC. All images were modified with the ImageJ programme. For kymographic analysis, we modified time-lapse images of mitochondria in curved neuronal processes to straight images using the 'straighten' function. Then, the images were resliced to y axis and Z-projected with the 'reslice' and 'Z-project' modules. We manually generated constriction spikes (right diagram of kymography). Each incidence of CoMIC was labelled as one spike, and the duration of the CoMIC was expressed by the thickness of the spike. We performed quantitative analyses, including the total duration, frequency and mean interval, of CoMIC from all mitochondria longer than 5 mm in neuronal processes B200 mm from the soma. To measure the bulging ratio of foci, we selected two mitochondrial images of off-phase and maximum bulging from more than 10 CoMIC-containing mitochondria, and measured the ratio of the diameter of each foci in the off-phase and bulging site.
Measurement of mitochondrial length. Mitochondrial images captured from all neuronal processes (200 mm from the soma) were converted to binary images using the 'threshold' module in ImageJ. Then, the binary images were converted to images 1 pixel wide by the 'skeletonize' module, and mitochondrial length was assessed by the 'analyse particles' module.
Electron microscopy. The Drp1 knockdown viral constructs were transfected into 293gpg packaging cells using PEI (Polyscience, #23966). After transfection, media were changed to the neurobasal media containing 2% B27 supplement, 0.5 mM Lglutamine and 50 unit per ml penicillin/streptomycin. The supernatants containing viral particles were collected at 2-4 days after transfection, and infected primary cortical neurons. Drp1 knockdown cortical neurons were washed with 0.1 M phosphate-buffered saline (pH 7.4, PBS) and fixed with 10% paraformaldehyde, 2.5% glutaraldehyde in PBS at 4°C for overnight. After fixation, cells were washed with PBS and post fixed with 1% osmium tetroxide for 90 min. The fixed cells were dehydrated through ascending series of ethanol, and then embedded in Epon mixture. The embedding blocks were semi-thin sectioned (200 nm) using a Reichert-Jung Ultracut E ultramicrotome (Leica Microsystems), stained with Toluidine blue and trimmed for further observation. The trimmed sections were collected on 200-mesh cooper grid and stained with uranyl acetate and lead citrate. Observation of mitochondrial structure was performed using an H-7500 electron microscope (Hitachi) with 80 kV acceleration voltage.
Statistical analysis. Statistical analyses and graph plotting were conducted using the Sigmaplot 12.5 software. Centre values and error bars of all graph indicate mean value and s.e., respectively. To measure the probability of CoMIC, we separated mitochondria into those that showed CoMIC and those that did not, and assigned them values of 100% or 0%, respectively. Then, we averaged them and tested statistical significance with the Mann-Whitney rank sum test. To examine the statistical significance of the proportion of mitochondria undergoing subsequent division to that of mitochondria showing no or prior CoMIC, we generated contingency tables according to the presence of CoMIC and division. The contingency tables were tested with w 2 -tests with Yates's correction for continuity. Similarly, contingency tables for mitochondria showing prior CoMIC among nondividing or dividing mitochondria were also tested with a w 2 -test with Yates's correction for continuity. For statistical analysis of total CoMIC duration, frequency and mean interval and bulging ratio of the foci containing CoMIC, the Mann-Whitney rank sum test was used.
Data availability. The authors declare that all data supporting the findings of this study are available within the article and its Supplementary Information Files or from the corresponding author on reasonable request.