The tethering function of mitofusin2 controls osteoclast differentiation by modulating the Ca2+–NFATc1 axis

Dynamic regulation of the mitochondrial network by mitofusins (MFNs) modulates energy production, cell survival, and many intracellular signaling events, including calcium handling. However, the relative importance of specific mitochondrial functions and their dependence on MFNs vary greatly among cell types. Osteoclasts have many mitochondria, and increased mitochondrial biogenesis and oxidative phosphorylation enhance bone resorption, but little is known about the mitochondrial network or MFNs in osteoclasts. Because expression of each MFN isoform increases with osteoclastogenesis, we conditionally deleted MFN1 and MFN2 (double conditional KO (dcKO)) in murine osteoclast precursors, finding that this increased bone mass in young female mice and abolished osteoclast precursor differentiation into mature osteoclasts in vitro. Defective osteoclastogenesis was reversed by overexpression of MFN2 but not MFN1; therefore, we generated mice lacking only MFN2 in osteoclasts. MFN2-deficient female mice had increased bone mass at 1 year and resistance to Receptor Activator of NF-κB Ligand (RANKL)-induced osteolysis at 8 weeks. To explore whether MFN-mediated tethering or mitophagy is important for osteoclastogenesis, we overexpressed MFN2 variants defective in either function in dcKO precursors and found that, although mitophagy was dispensable for differentiation, tethering was required. Because the master osteoclastogenic transcriptional regulator nuclear factor of activated T cells 1 (NFATc1) is calcium-regulated, we assessed calcium release from the endoplasmic reticulum and store-operated calcium entry and found that the latter was blunted in dcKO cells. Restored osteoclast differentiation by expression of intact MFN2 or the mitophagy-defective variant was associated with normalization of store-operated calcium entry and NFATc1 levels, indicating that MFN2 controls mitochondrion–endoplasmic reticulum tethering in osteoclasts.

Mitochondria exist within a highly regulated and dynamic network in most cell types. Termed mitochondrial dynamics, the linked processes of fission, fusion, and mitophagy play an important role in mitochondrial homeostasis. Changes in mitochondrial networks are frequent, and this architecture varies depending on the developmental or metabolic needs of a cell (reviewed in Refs. 1,2). Mitofusin1 (MFN1) 2 and Mitofusin2 (MFN2), transmembrane GTPases on the outer mitochondrial membrane, control mitochondrial architecture by tethering adjacent mitochondria and initiating fusion. MFN2 additionally mediates tethering of mitochondria to the ER, promotes mitophagy, and allows mitochondrial transport down axons. MFN1 and MFN2 are necessary for survival, as global deletion of either protein results in embryonic lethality prior to embryonic day 12.5 in mice (3). Double deletion of MFN1 and MFN2 leads to even earlier mortality and, in mouse embryonic fibroblasts, causes severe fragmentation of mitochondrial networks (3).
In humans, the nervous system is highly impacted by MFN2 defects, and MFN2 mutations are linked to neuropathies (reviewed in Refs. 4,5). The human syndrome most associated with mutations in MFN2 is Charcot-Marie-Tooth type 2A (CMT2A), a neurodegenerative disorder in which motor and sensory defects develop within the first two decades of life (6). Mice with homologous Mfn2 mutations have sensory and/or motor defects and axonal degeneration accompanied by altered mitochondrial dynamics (7)(8)(9), and cultured human cells from CMT2A patients display decreased electrical properties and reduced mitochondrial membrane potential (10, 11). Models lacking mitofusins also yield conditions of myocardial hypertrophy, heart tube contractile dysfunction, vascular smooth muscle cell proliferation, and skeletal muscle atrophy, to name a few (12)(13)(14)(15). To our knowledge, effects on the skeleton have not been reported in CMT2A mouse models or human patients. Additionally, no human syndromes with MFN1 mutations have been described.
Bone homeostasis relies on the coupled activity of osteoblast and osteoclast (OC) cells that build and resorb the bone matrix, respectively. OCs are rich in mitochondria, and differentiation from bone marrow macrophage (BMM) precursors to OCs is associated with a significant increase in mitochondrial DNA copy number, oxidative phosphorylation subunit protein levels, and oxygen consumption rates (16). Previous investigations of mitochondria in the OC lineage have largely focused on the role of mitochondrial biogenesis and oxidative phosphorylation in supporting OC function (17)(18)(19). We were prompted to explore the role of mitochondrial dynamics in OCs because this specialized cell type is mitochondrion-rich and undergoes vast morphological changes to become a mature multinucleated cell. We hypothesized that mitochondrial fusion might accompany OC precursor fusion during osteoclastogenesis and that tethering of mitochondria to each other or the ER is important for OC formation and/or function. To test this, we generated a mouse model in which MFN1 and MFN2 are conditionally deleted in the OC lineage with LysM-cre. We found that knockout of both homologs is associated with increased bone mass in female animals and blunted osteoclastogenesis in culture. Rescue experiments in vitro revealed that MFN2 more strongly induces osteoclastogenesis than MFN1 and that in vivo, depletion of MFN2 is protective against age-and RANKL-induced bone loss. Last, we demonstrate mitochondrial tethering to be the predominant function of MFN2 in OCs, associated with changes in Ca 2ϩ -NFATc1 signaling.

Bone mass is increased in female mice lacking MFN1 and MFN2 in the OC lineage
Because OCs undergo vast morphological changes to become mature multinucleated cells, we hypothesized that alterations in mitochondrial dynamics, and fusion in particular, accompany and modulate this event. To prevent mitochondrial fusion, we employed mice with floxed alleles for both Mfn1 and Mfn2, directing their excision to the OC lineage with LysM-cre. All of our mice were mated onto a homozygous cre background (LysM c/c ) to increase recombination efficiency. We intercrossed double heterozygous (LysM c/c ; (Mfn1/2) fl/ϩ ϭ ctrl) mice to generate double knockout (LysM c/c ; (Mfn1/2) fl/fl ; dcKO) littermates, which were born at normal Mendelian ratios. Subsequently, these dcKO progeny were mated to the double-heterozygous ctrl mice to produce our experimental groups. In parallel, double heterozygous ctrl mice were bred to cre-only (LysM c/c ;(Mfn1/2) ϩ/ϩ ) mice to generate a cohort of ctrl and cre-only littermates. Importantly, there was no difference in bone mass parameters between ctrl and creonly groups (Fig. S1, A-N), indicating that ctrl doubleheterozygous animals are an appropriate control group for dcKO mice.
dcKO mice showed no obvious gross phenotypes and had weights identical to ctrl littermates (Fig. S2, A and B). Male and female animals were sacrificed at 2 months of age, a time of high OC activity, and basal bone morphology was assessed by CT of dissected femora. We found a strong bone phenotype in female animals at this age, with all trabecular and cortical parameters consistent with a high bone mass (significantly increased bone volume (BV) per total volume (TV), bone mineral density (BMD), trabecular number (Tb.N), and cortical thickness (Cort.Th)) ( Fig. 1, A and B, and Fig. S2, C-F). Interestingly, males showed no significant differences between ctrl and dcKO groups in any of these parameters (Fig. 1, C and D, and Fig. S2, G-J). Serum CTX-1 levels were decreased in dcKO females, indicating a decrease in bone catabolism ( Fig.  2A). Histomorphometry showed that OC surface and number in dcKOs trended lower than in ctrls in female mice (Fig.  2, B-E). Osteoblast-OC coupling was not affected by loss of MFN1 and MFN2 in OCs; indices of bone formation, such as serum P1NP, mineralizing surface, mineral apposition rate, and bone formation rate, were unchanged between groups (Fig. 2, F-K).

Osteoclastogenesis is inhibited in MFN1/2-deficient cultures
To more closely characterize the impact of mitofusin loss in the OC lineage, BMMs were expanded from bone marrow of ctrl and dcKO mice. Compared with cre-only OCs with WT mitofusin expression, double-heterozygous ctrls produced less MFN1 and MFN2 protein, which was further diminished in dcKO cells (Fig. 3A). dcKOs were unable to respond to RANKL and differentiate into OCs on plastic wells, whereas differentiation of ctrls was intact (Fig. 3B). When cultured on bone, resorption was also minimal (Fig. 3C). Consistently, levels of the OC markers NFATc1, DC-STAMP, and CTSK were reduced in dcKOs at the mRNA level (Fig. 3, D-F). Western blots also demonstrated lower levels of Cathepsin K (CTSK) and C-SRC proteins, which supports a defect in OC differentiation (Fig. 3G).
The levels of protein and mRNA for both MFN homologs increased throughout osteoclastogenesis in WT cells (Fig. 4, A  and B). Therefore, we sought to determine whether MFN1 or MFN2 was dominant in contributing to the dcKO phenotype and attempted to rescue the defect in OC formation by retrovirally expressing MFN1 or MFN2 in dcKO BMMs. We found that overexpression of MFN2, but not MFN1, restored full OC differentiation, suggesting that MFN2 is the dominant mitofusin homolog that supports osteoclastogenesis (Fig. 4, C-E).

MFN2 plays a role in OCs in vivo
To assess whether loss of MFN2 alone in OC lineage cells would have an effect on bone mass in vivo, we generated litter-

MFN2 directs osteoclastogenesis
mate cohorts of mice with WT Mfn2 (LysM c/c ; Mfn2 ϩ/ϩ ; creonly) and conditionally targeted Mfn2 (LysM c/c ; Mfn2 fl/fl ; Mfn2 cKO). We confirmed depletion of MFN2 by evaluating protein and mRNA expression (Fig. 5, A and B). Although we observed upward trends in female Mfn2 cKOs at 2 and 4 months of age, the differences in BV/TV and BMD did not become statistically significant until 12 months of age; there was no difference in bone mass between male cKO and male cre-only mice at any age examined (Fig. 5, C-F, Table 1, and Table S1). The percentage of trabecular bone retained from 4 months to 12 months was 24% for ctrls versus 46% for cKO females (64% versus 63% for males). Tartrate resistant acid phosphatase (TRAP)-stained 2-month-old tibia sections show a trend toward lower OC number and surface per bone surface (Oc.N/BS and Oc.S/BS) in female cKO mice (Fig. S3). Thus, loss of MFN2 has limited effects on basal bone mass in young animals, whereas it protects female mice from age-related bone loss.
Our group and others have previously observed that challenging animals with a stimulus for pathological bone loss can unmask a robust OC phenotype even when basal changes are not prominent (20 -23). We hypothesized that this may be the case for Mfn2 cKO animals and therefore induced acute osteolysis in 2-month-old animals. Male and female cre-only and cKO mice were challenged with intraperitoneal RANKL injection, which stimulates OC activity and induces significant loss of trabecular bone in 2 days. Although cre-only males, male cKOs, and cre-only females lost 30%-45% of trabecular bone fraction following RANKL injection, female cKOs were protected from induced osteolysis, showing no significant change (Fig. 5, G-I, and Fig. S4). Therefore, even with a strong OC stimulus, male mice are not affected by loss of MFN2, whereas female mice require MFN2 in the OC to respond.

Mitochondrial tethering is the dominant MFN2 mechanism contributing to osteoclastogenesis in vitro
MFN2 has two primary molecular functions, tethering/fusion and mitophagy, that can be selectively disabled via mutation of threonine and serine residues to constitutively activate or ablate the signal for mitophagy (24). MFN2 T111E-S442E (MFN2-EE) mimics phosphorylation by PINK1 kinase, driving spontaneous mitophagy and preventing fusion, leading to a highly fragmented mitochondrial network. In contrast, MFN2 T111A-S442A (MFN2-AA) has the opposite effect; the nonphosphorylatable MFN2 residues eliminate mitophagy but allow normal tethering and fusion ( Fig. 6A) (24, 25). To examine the ability of either MFN2 function to impact OCs, we retrovirally overexpressed each mutant, along with empty vector (V) and MFN2-WT controls, in female dcKO BMMs that were defective in producing OCs (Fig. 6B). To verify that mutants were functioning properly, we stained transduced BMMs with MitoTracker Green and LysoTracker Red prior to live-cell con-

MFN2 directs osteoclastogenesis
focal microscopy (Fig. 6C). Mitochondrial colocalization with lysosomes, an indication of mitophagy, was minimal under all conditions. Similar to previous studies with these constructs, MFN2-WT and MFN2-AA increased tethering and fusion, as manifested by increased mitochondrial aspect ratios (mitochondrial length/mitochondrial width) compared with less fused V and MFN2-EE samples (Fig. 6D). Following 6 days of RANKL exposure, cultures were fixed and stained for TRAP (Fig. 6, E and F). We found that transduction of dcKO BMMs with MFN2-AA restored osteoclastogenesis to a level comparable with MFN2-WT, indicating that the mitophagy function of MFN2 is dispensable for osteoclastogenesis. In contrast, transduction with MFN2-EE had only a minimal effect on osteoclastogenesis, suggesting that tethering/fusion by MFN2 is necessary for osteoclastogenesis to occur efficiently ( Fig. 6, C-F). Because MNF1 was absent from these cells, this reveals that mitochondrial tethering by MFN2 contributes to OC formation in a way that MFN1 cannot.

MFN2 directs osteoclastogenesis Ca 2؉ signaling is perturbed in pre-OCs with impaired MFN2mediated tethering
To further explore the mechanism through which MFN2 controls osteoclastogenesis, we assayed cytoplasmic Ca 2ϩ levels with Fura-2 basally, following ATP stimulation (to release ER calcium stores) and after addition of extracellular calcium (2 mM) to activate store-operated calcium entry (SOCE). In pre-OCs generated with 48 h of RANKL exposure, dcKOs had higher basal Ca 2ϩ levels and a similar ATP-mediated spike from baseline but a muted intake of Ca 2ϩ following SOCE activation compared with ctrls, as assessed by area under curve quantification (Fig. 7, A-C). We used the same assay in dcKO BMMs overexpressing MFN2-WT, -EE, and -AA. Compared with the vector control, MFN2-WT and MFN2-AA increased responsiveness to extracellular Ca 2ϩ whereas MFN2-EE did not, suggesting that the tethering function of MFN2 regulates SOCE (Fig. 7, D-F). Congruent with restoration of osteoclastogenesis, cells with improved SOCE (MFN2-WT and -AA) had increased NFATc1 levels compared with dcKOs with V or MFN2-EE transduction,

Discussion
An abundance of mitochondria has been deemed important for function of the OC, and increases in mitochondrial proteins are associated with differentiation from myeloid precursors (16,26,27). Mitochondrial biogenesis, mediated by PGC1␣ and PGC1␤, plays an important role, as global deletion of either of these factors leads to increased trabecular bone mass (17,28). Further, conditional PGC1␤ deletion in OCs results in im-paired cytoskeletal organization and bone resorption (19). OCs have been reported to utilize oxidative phosphorylation and aerobic glycolysis, and it has been suggested that the primary function of mitochondria may not be ATP generation at the ruffled border (29).
To better understand the role of mitochondria and put into context the elongated organelles observed in human OCs (27), we aimed to perturb mitochondrial network organization. We found that dual depletion of the key mitochondrial dynamics proteins MFN1 and MFN2 in the OC lineage leads to increased bone mass in female mice in vivo, accom-  Table 1 and Table S1. *, p Ͻ 0.05; **, p Ͻ 0.01; ns, not significant; unpaired t test with Welch's correction in B-F and ordinary two-way ANOVA with multiple comparisons in G and H.

MFN2 directs osteoclastogenesis
panied by a decrease in serum CTX-1 levels. This corresponds to an inability of OCs derived from these dcKO animals to form in culture. Previous work in the field has focused primarily on how changes in mitochondrial content and oxidative phosphorylation impact OC resorption (16,19,30,31). Our model expands the relationship between mitochondria and OCs by linking disruption of mitochondrial dynamics to inhibition of OC formation.
To determine whether MFN1 and MFN2 contribute differentially to OC formation, we overexpressed each homolog in dcKO BMMs in culture and found that osteoclastogenesis was much more efficient with MFN2 than MFN1 addition. Further, loss of MFN2 alone led to increased bone mass in vivo, albeit delayed compared with the doubly deficient cohort. Similarly, work with heart and skeletal muscle has revealed that loss of MFN1 and MFN2 produces more severe phenotypes than single knockout of either mitofusin (14,32). It is likely that the mitochondrial fusion function of MFN2 plays some role in OCs, which allows compensation by MFN1 to reduce the severity of bone accrual in MFN2 singly deficient mice, although at least one MFN2-unique role is also important.
MFN2, but not MFN1, functions in mitophagy, mitochondrial transport in axons, and ER-mitochondrion interactions. Phenotypic effects of specific mutations reveal significant cell type specificity in these various roles. For example, fibroblasts derived from CMT2A patients show differences in ER-mitochondrion spacing that correlate with disease severity, suggesting that alteration of ER-mitochondrion tethering contributes to pathogenicity in humans (33). In contrast, mitophagy is required for metabolic reprogramming of mammalian hearts during perinatal development, but tethering by MFN2 is dispensable for this process (24). We utilized point mutants separating tethering and mitophagy functions to probe which MFN2 function is important in the OC lineage and found that the tethering function of MFN2 is necessary for osteoclastogenesis.
The master OC transcription factor NFATc1 depends on cytoplasmic Ca 2ϩ flux and requires dephosphorylation by calcineurin for translocation to the nucleus (34), leading to amplification because of binding at its own promoter. We find that NFATc1 mRNA levels are diminished in our dcKO cells, suggesting alterations in Ca 2ϩ handling. The ER-mitochondrion juxtaposition, regulated by the tethering function of MFN2 but not MFN1, is known to impact Ca 2ϩ signaling (9,33,35). We therefore examined cytoplasmic Ca 2ϩ levels and found that dcKOs had a significantly reduced magnitude of SOCE. Similar blunting of SOCE was found in C2C12 myoblasts with MFN2 knockdown (36). Interestingly, our dcKOs showed higher basal cytoplasmic Ca 2ϩ , whereas MFN2-depleted C2C12 cells and BMMs had lower basal cytoplasmic Ca 2ϩ than controls (36,37). It is possible that the presence of MFN1 in the experiments with MFN2 knockdown plays a role in basal levels, whereas MFN2 is more important for SOCE regulation. On our MFN1/MFN2deficient background, we further found that the tethering-competent/mitophagy-deficient MFN2-AA, but not the tetheringdeficient/mitophagy-activated MFN2-EE, restored SOCE along with NFATc1 induction and osteoclastogenesis. Previously, we found that cells lacking TMEM178, which show increased NFATc1 and osteoclastogenesis, have elevations in basal cytoplasmic Ca 2ϩ and SOCE (38,39). Thus, it is likely that the magnitude of Ca 2ϩ excursions, rather than the basal level, is more important for osteoclastogenesis. Mechanistically, we propose that MFN2-mediated restoration of ER-mitochondrion contacts, which regulate mitochondrial Ca 2ϩ uptake, are responsible for the ability of cells to generate the large swings in cytoplasmic Ca 2ϩ (i.e. high-amplitude oscillations) that drive NFATc1 activation.
We find striking sexual differences in our bone phenotypes, and to our knowledge, this is the second report of differential male versus female responses to mitofusin loss. Conditional ablation of MFN2 in brown adipose tissue leads to sex-specific remodeling of mitochondrial function; females displayed increased ATP-synthesizing fatty acid oxidation, whereas males experienced increased glycolytic capacity (40,41). Although loss of MFN2 in brown adipose tissue caused distinct abnormalities in males and females, we only saw changes in females with loss of mitofusins in the OC lineage. We previously found higher OC numbers in the trabecular compartment of female mice, possibly explaining the greater impact on bone mass with more abundant OCs in this sex (42). One explanation could be that there is a threshold effect for mitofusins on OC formation and that females are more sensitive to mitofusin depletion than males. Further, the high bone mass phenotype may be more severe in female dcKOs compared with single Mfn2 cKOs because residual MFN1 expression partially compensates for loss of MFN2. Consistent with this threshold hypothesis, we see no differences in bone mass between creonly and double heterozygous ctrl animals despite protein levels differing by about half. Because residual mitofusin proteins were detected in our lysM-cre models, greater effects, including increased bone mass in male mice, might be achieved with more complete recombination in the OC lineage. Future investigations can utilize RANK-cre to target a similar early devel-  (43,44), and our work opens the possibility that it may also preserve bone mass in women.
Mice were housed communally with ad libitum access to fresh chow and water in a pathogen-free, temperature-controlled barrier facility. Daily observation and weekly cage changes were provided by staff of the Division of Comparative Medicine. Laboratory members coordinated animal husbandry and breeding, and the Division of Comparative Medicine veterinarian assessed all health concerns. All protocols were approved by Washington University School of Medicine's Animal Studies Committee (protocol 20170025). . D, quantification of mitochondrial aspect ratio (mitochondrial length/width). E, BMMs were TRAP-stained following 6 days of RANKL exposure, revealing robust osteoclastogenesis with MFN2-AA but not MFN2-EE. Scale bars ϭ 400 m. F, quantification of OC numbers in three biological replicates. ****, p Ͻ 0.0001, ordinary one-way ANOVA.

Micro-computed tomography
Following dissection of right femora from 2-, 4-, and 12-month-old animals, bones were fixed in 10% neutral buffered formalin (Di Ruscio & Associates, Inc., Fenton, MO) for 24 h and then stored in 70% ethanol. Trabecular and cortical scans were acquired proximal to the distal growth plate and at the mid-shaft of femora, respectively (CT40, Scanco, Brüttisellen, Switzerland; 10-m resolution, 55 kilovoltage peak, 145 A, 300-s integration time). Thresholds were determined in a blinded manner and analyzed with reference to accepted guidelines (47,48).

Histomorphometry
2-month-old mice were injected i.p. with 10 mg/kg calcein (C0875, Sigma) and 30 mg/kg Alizarin Red (A3882, Sigma) 7 and 2 days prior to sacrifice, respectively. Intact femora/ tibiae were fixed in 10% neutral buffered formalin for 24 h prior to storage in 70% ethanol. Tissues were embedded in methylmethacrylate and sectioned sagittally by the Washington University Musculoskeletal Histology and Morphometry Core. Unstained and TRAP-stained (Sigma) slides were imaged at ϫ20 high resolution using a NanoZoomer 2.0 with bright field and FITC/TRITC (Hamamatsu Photonics). Images were then analyzed via Bioquant Osteo software according to the manufacturer's instructions and published standards (v18.2.6, Bioquant Image Analysis Corp., Nashville, TN) (49).
Expanded BMMs were plated at a density of 9,000 and 300,000 cells/well for 96-and 6-well plates, respectively. Purified GST-RANKL at 30 ng/ml was added to ␣-MEM containing 10% FBS, 100 IU/ml penicillin/streptomycin, and a 1:50 dilution of CMG 14-12 cell supernatant. The medium was changed every alternate day until control wells were filled with multinucleated OCs. TRAP stains were employed following 10-min fix-

Resorption assay
Expanded BMMs were seeded on bovine bone slices and differentiated to OCs with purified GST-RANKL as above. At maturity, bone slices were fixed for 10 min with 4% paraformaldehyde and 0.1% Triton X-100 in PBS. Bone slices were incubated in 0.5 NaOH for 30 s and kept hydrated in PBS. Cells were scraped off bone slices with a cotton swab, and bone slices were incubated with 20 g/ml peroxidase-conjugated wheat germ agglutinin in PBS (Sigma, 61767) for 30 min. Subsequently, slices were washed three times in PBS and incubated with a DAB Chromogen Kit (Biocare Medical, BDB2004H) at 37°C for 30 min. Slices were left to air-dry and imaged under a bright field at ϫ20.

Induced osteolysis
2 month old ctrl and Mfn2 cKO animals were injected i.p. with 1 mg/kg of body weight GST-RANKL twice, 24 h apart, as published previously (53). 50 h after the first injection, animals were sacrificed, and femoral trabecular bone was evaluated by CT.

MFN2 directs osteoclastogenesis
Washington University Hope Center and the Center for Investigation of Membrane Excitability Diseases Live Cell Imaging Facility. 100 M ATP and 2 mM CaCl 2 were used as stimuli as indicated.

Statistics
Two-way ANOVA with Tukey's multiple comparisons and one-or two-tailed unpaired t tests with Welch's correction were performed with GraphPad Prism built-in statistical analysis (GraphPad Software, Inc., La Jolla, CA). All data are represented as mean Ϯ S.D. with three or more biological replicates. p values are designated as follows: *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001; ****, p Ͻ 0.0001.

Data availability
All data are contained within the manuscript.