Metaxins are core components of mitochondrial transport adaptor complexes

Trafficking of mitochondria into dendrites and axons plays an important role in the physiology and pathophysiology of neurons. Mitochondrial outer membrane protein Miro and adaptor proteins TRAKs/Milton link mitochondria to molecular motors. Here we show that metaxins MTX-1 and MTX-2 contribute to mitochondrial transport into both dendrites and axons of C. elegans neurons. MTX1/2 bind to MIRO-1 and kinesin light chain KLC-1, forming a complex to mediate kinesin-1-based movement of mitochondria, in which MTX-1/2 are essential and MIRO-1 plays an accessory role. We find that MTX-2, MIRO-1, and TRAK-1 form another distinct adaptor complex to mediate dynein-based transport. Additionally, we show that failure of mitochondrial trafficking in dendrites causes age-dependent dendrite degeneration. We propose that MTX-2 and MIRO-1 form the adaptor core for both motors, while MTX-1 and TRAK-1 specify each complex for kinesin-1 and dynein, respectively. MTX-1 and MTX-2 are also required for mitochondrial transport in human neurons, indicative of their evolutionarily conserved function. Mitochondrial trafficking is carefully regulated and functionally important in neurons. Here, the authors screen C. elegans for abnormal mitochondrial distribution and identify metaxin-1 and -2, which bind Miro and microtubule motor proteins to promote mitochondrial trafficking.

N euronal mitochondria play important roles in neuronal physiology. As the cellular power plants of neurons, mitochondria provide ATP to sustain the firing of action potentials and continuous synaptic transmission 1,2 . As a cellular calcium reservoir, mitochondria regulate axonal cytosolic Ca 2+ level 3 . In addition, mitochondria participate in producing many biosynthetic intermediates, as well as cellular stress responses such as autophagy and apoptosis 4,5 . The extreme size of neurons creates challenges for ATP diffusion 6 . Hence, neurons have evolved specialized mechanisms to efficiently transport mitochondria to distal processes where energy is in high demand 2,7-9 . Impairment of mitochondrial trafficking in neurons results in neuronal development and maintenance defects, and human neurodegenerative diseases [10][11][12] .
Mitochondrial transport in axons and dendrites is mediated primarily by the microtubule-dependent motor systems, including both kinesin and dynein 13 . Kinesin drives mitochondria towards the plus-ends of microtubules, while dynein moves them in the opposite direction. Among the kinesin superfamily, KIF5/ kinesin-1 and KIF1B/kinesin-3 are the primary kinesins that mediate mitochondrial transport in neurons 14 . Mitochondria recruit molecular motors through membrane adaptor proteins, which not only mediate the movement of mitochondria, but also dictate how neuronal activity and extrinsic signals regulate the abundance of mitochondria in axons 2,15 . Previous studies have shown that the adaptor complex contains two components: the mitochondrial outer membrane atypical Rho GTPase Miro 16 , and the adaptor protein Milton in Drosophila (TRAKs or OIP106 and GRIF-1 in mammals) 17,18 . Importantly, the Miro-TRAK complex plays critical roles in linking mitochondria to both kinesin and dynein motors 2,9 . In mammalian cells, two TRAK proteins function differently to steer polarized transport of mitochondria: TRAK1, which is mainly distributed in axons, binds to both kinesin-1 and dynein, and is required for axonal localization of mitochondria, whereas TRAK2, which is mainly localized in dendrites, primarily interacts with dynein and is responsible for the dendritic distribution of mitochondria 19 .
Miro is localized to the outer mitochondrial membrane through the C-terminal transmembrane domain and contains two GTPase domains that are distantly related to the small GTPases of the Ras superfamily 2 . In Drosophila, dmiro mutants showed dramatic defects in mitochondrial trafficking to axons and dendrites 16 . Further studies have shown that Miro anchors mitochondria to microtubule motors through binding to Milton/ TRAK 20,21 . Besides, Miro might also bind to KIF5/kinesin-1 directly 22 , and recruits the mitochondrial myosin Myo19 to link mitochondria to the actin cytoskeleton 23 . Miro contains a pair of Ca 2+ -binding EF-hands, which sense neuronal activity or oxidative stresses to arrest mitochondrial trafficking 24 . Miro is also the target of the Parkinson's disease-associated gene PINK1, which phosphorylates Miro and triggers its degradation through the ubiquitin ligase Parkin 25 .
Is Miro the only outer membrane protein that links mitochondria to motor proteins? Several pieces of evidence suggest that this is unlikely. First, loss of Drosophila Miro (dMiro) cannot fully block mitochondrial movement: small amounts of mitochondria are still localized in neurites 16,26 . In rodent hippocampal neurons, the distribution of mitochondria along the axon is not dependent on Miro1, although fast axonal mitochondrial trafficking is severely affected in Miro1 knockout neurons 11 . The TRAK1 and TRAK2 adaptor proteins are still recruited to the outer mitochondrial membrane to drive mitochondrial trafficking in Miro1/2 double-knockout cells 23 . These data indicate that the previously unknown adaptors, probably independent of Miro, exist on the mitochondrial outer membrane to recruit TRAK or other proteins to mediate mitochondrial trafficking.
To identify potential adaptor proteins, here we establish an in vivo system to investigate mitochondrial distribution and trafficking in axons and dendrites using the PVD sensory neuron of C. elegans. The PVD neuron elaborates a single axon, one anterior and one posterior dendrite with highly ordered branches 27 . Our previous work has shown that microtubule organization is uniform along the anterior-posterior (A-P) axis of PVD 28 . Microtubules in the anterior primary dendrite are predominantly oriented with their minus-ends pointing towards the distal dendrite. In contrast, the posterior primary dendrite and the axon mainly contain plus-end out microtubules. Thus, dynein transports cargo towards the tip of the anterior primary dendrite, whereas kinesins transport cargo towards the distal posterior dendrite and axon. Using cell type-specific promoters, we have visualized mitochondria in PVD neurites with single-cell resolution.
During a forward genetic screen for mutants with abnormal PVD dendritic mitochondrial distribution, we isolated the mtx-2 mutant, which abolished dendritic mitochondria. We further show that the known binding partner of MTX-2, MTX-1, is also required for mitochondrial trafficking. The metaxin complex consisting of metaxin1 (called SAM37, MAS37, or TOM37 in yeast) and metaxin2 (called SAM35, TOM38, or TOB38 in yeast), functions with SAM50 to form the mitochondrial sorting and assembly machinery to fold and insert β-barrel proteins onto the mitochondrial membrane 29 . Our data show that the metaxin complex directly binds to the kinesin motor and MIRO-1 to control mitochondrial trafficking. Our genetic and biochemical data suggest that there are two adaptor complexes: an MTX-2/ MIRO-1/MTX-1/KLC-1 complex responsible for kinesinmediated mitochondrial movement and an MTX-2/MIRO-1/ TRAK-1 complex responsible for dynein-mediated mitochondrial trafficking. Finally, MTX-1 and MTX-2 homologs mediate mitochondrial trafficking in human iPS-cell derived neurons, revealing that their functions in mitochondrial trafficking are evolutionarily conserved.

Results
MTX-2 and MIRO-1 are essential for mitochondrial localization in PVD axon and dendrites. To study trafficking mechanisms of mitochondria in axons and dendrites, we visualized mitochondria by tagging the mitochondrial outer membrane protein TOMM-20 with GFP in the C. elegans sensory neuron PVD (PVD > TOMM-20 (1-54AA)::GFP), together with PVD > mCherry to label the PVD neurites (Fig. 1a). In addition to the intense GFP signal in the PVD soma, round, oval, and occasionally rod-shaped mitochondria were detected as individual punctum throughout the highly elaborate anterior and posterior dendrites (Fig. 1a, b) as well as in the axon (Fig. 1a). Additional mitochondria markers showed a similar pattern as the TOMM-20 marker ( Supplementary Fig. 1a, b).
A large body of evidence has shown that the conserved mitochondrial protein Miro is required for mitochondrial trafficking. We used Cas9 to generate a null mutation in miro-1, the C. elegans orthologue of Miro 30 . In the miro-1 mutant, PVD dendrites were devoid of mitochondria ( Fig. 1b-d), confirming the requirement of the Miro homologue for proper mitochondria trafficking in C. elegans. To find other molecules important for mitochondrial trafficking, we carried out an unbiased forward genetic screen using PVD dendritic mitochondria as a system. From a visual screen covering 4000 haploid genomes, the mutation with the largest defect in mitochondrial distribution, wy50256, showed nearly complete absence of mitochondria from both anterior and posterior dendrites, a phenotype that is indistinguishable from that of miro-1 mutants.
With genetic mapping and sequencing, we found that wy50256 caused a premature stop codon at Q9 of mtx-2, which encodes one component of the mitochondrial outer membrane protein complex metaxin, is responsible for this trafficking phenotype ( Fig. 1b-d). This phenotype might be due to the lack of mitochondrial trafficking into the neurites or defects in the anchoring or maintenance of mitochondria in neurites. To distinguish between these possibilities, we performed time-lapse imaging experiments to record the dynamics of mitochondrial trafficking. In wild-type worms, the somatic mitochondrial network gave rise to elongated tubules. As the tubule lengthens ("extension"), a distal tubular segment broke off through a distinct fission event ("fission"). The rod-shaped mitochondrion then migrated away progressively ("transport") ( Fig. 1e  b Representative confocal images showing dendritic morphology and mitochondrial distribution of PVD neuron in wild type, miro-1, and mtx-2 mutants. Magenta: PVD > mCherry. Green: PVD > TOMM-20(1-54AA)::GFP. Scale bar: 10 μm. c, d Quantification of the number of mitochondria in anterior (c) and posterior (d) dendrites of PVD. Data are shown as mean ± SEM. One-way ANOVA with Tukey's multiple comparisons test (95% CI). ****p < 0.0001. (exact p values and sample size: wild-type vs. miro-1, p < 0.0001; wild-type vs. mtx-2, p < 0.0001. n = 72 animals for each genotype). e Representative timelapse images showing "extension", "fission", and "transport" behaviors of mitochondria in the anterior dendrite of PVD. Scale bar: 10 μm. f A schematic diagram showing mitochondrial distribution in axon and dendrite of DA9. Magenta: DA9 morphology. Green: mitochondria. Anterior is to the left and dorsal is up. g Representative confocal images showing the mitochondrial distribution of DA9 neuron in wild type, miro-1, and mtx-2 mutants. Green: DA9 > TOMM-20(1-54AA)::GFP. Scale bar: 10 μm. h, i Quantification of mitochondria number in dendrite and axon of DA9 in wild type, miro-1, and mtx-2 mutants. Data are shown as mean ± SEM. One-way ANOVA with Tukey's multiple comparisons test (95% CI). ****p < 0.0001. (exact p values and sample size: h wild-type vs. miro-1, p < 0.0001; wild-type vs. mtx-2, p < 0.0001. wild-type, n = 23; miro-1, n = 24; mtx-2, n = 24 animals. i wild-type vs. miro-1, p < 0.0001; wild-type vs. mtx-2, p < 0.0001. wild-type, n = 15; miro-1, n = 20; mtx-2, n = 23 animals). Source data are provided as a Source Data file.  Supplementary Fig. 1c, d, and Supplementary Movie 1). Within PVD dendrites, mitochondria were transported both away from the cell body (anterograde) and towards the cell body (retrograde) while many remained stationary during the course of our imaging experiments ( Supplementary Fig. 1e-h). Compared with wild type, both miro-1 and mtx-2 mutants showed diminished "extension" in both anterior and posterior dendrites (Supplementary Fig. 1i-l and Supplementary Movie 2). In wild-type controls, about a third of the "extension" events lead to "fission" and subsequent transport. In miro-1 or mtx-2 mutants, we did not observe a single "fission" and "transport" events during a total of 14 and 17.5 h of time-lapse recordings, respectively (Supplementary Fig. 1i-l). These data suggest that mtx-2 functions in transporting mitochondria to PVD dendrites.
To test if mtx-2 is also required for mitochondrial trafficking in other neurons, we examined mitochondrial localization in the motor neuron DA9. Both miro-1 and mtx-2 mutants showed severely reduced mitochondria in both the axon and dendrite of DA9 ( Fig. 1f-i), indicating mtx-2 and miro-1 are also required for mitochondrial localization into DA9 axons and dendrites. On the other hand, the distribution of early endosome (GFP::RAB-5) and late endosome (GFP::RAB-7) in PVD neurites is indistinguishable in wild type, miro-1, and mtx-2 mutants, further confirming that mtx-2's effect on mitochondrial localization is specific (Supplementary Fig. 1m-p).
The metaxin/SAM (sorting and assembly machinery) complex is thought to be localized on the mitochondrial outer membrane and helps to insert β-barrel proteins into the mitochondrial membrane 29 . We generated PVD-specific expression of a singlecopy transgene of GFP fused to MTX-2 to monitor its subcellular localization. We detected GFP signal in the cell body and fluorescent puncta along PVD neurites. Co-expression of TOMM-20::mCherry demonstrates that GFP::MTX-2 signal is co-localized with TOMM-20 ( Fig. 2a, b), further confirming that MTX-2 in the worm is localized to mitochondria of neurons.
MTX-2 can substitute for part of MIRO-1's function when overexpressed. The similarity between the mtx-2 and miro-1 mutant phenotypes argues that MTX-2 might function together with MIRO-1. To understand the relationship between mtx-2 and miro-1, we constructed double mutants between mtx-2 and miro-1. In PVD, mitochondria were completely absent from both anterior and posterior dendrites in the double mutants (Fig. 2c). The double mutant phenotype was not statistically different from those of single mutants because the single mutant phenotypes were already severe (Fig. 2d, e). In DA9 dendrites, the mitochondria were absent in the double mutant as well as in both single mutants ( Supplementary Fig. 2a, b). These results suggest that both miro-1 and mtx-2 are essential for mitochondrial transport into dendrites. Interestingly, miro-1 and mtx-2 single mutants showed a reduced number of mitochondria in the DA9 axon, with about 30% of mitochondria remaining in the axons. In the double mutants, mitochondria were completely absent from DA9 axons ( Supplementary Fig. 2a, c), suggesting that MTX-2 can at least carry out some MIRO-1 independent functions.
To further test the genetic interaction between these two genes, we overexpressed each gene in either miro-1 or mtx-2 mutants. Expression of miro-1 or mtx-2 genomic DNA constructs in PVD in the corresponding miro-1 or mtx-2 mutant fully rescued the mitochondria phenotype (Fig. 2c, f, g), indicating that both genes function cell autonomously to transport mitochondria. While overexpression of miro-1 did not rescue any aspect of the mtx-2 phenotype (Fig. 2c, f, g), overexpression of mtx-2 completely rescued mitochondrial localization in the posterior dendrites of miro-1 mutants, but did not in the anterior dendrites ( Fig. 2h-j).
These results argue that although both MTX-2 and MIRO-1 are required for mitochondrial transport under physiological conditions, MTX-2 can substitute for MIRO-1 in certain contexts, but not vice versa.
Despite the fact that MIRO-1 is a tail α-helical anchor protein and not a β-barrel protein, the known role of the metaxin complex makes it plausible that it might help to localize MIRO-1 into the mitochondrial outer membrane. To test this idea directly, we ask if mtx-2 is required for the mitochondrial localization of MIRO-1 with a strain where GFP was inserted in frame into the endogenous miro-1 locus using Cas9 ( Supplementary Fig. 3a). The wild type GFP::MIRO-1 showed a punctate fluorescence pattern throughout the cytosol of embryos that co-localized with TOMM-20::mCherry ( Supplementary Fig. 3a), confirming its localization in mitochondria. The pattern and intensity of GFP:: MIRO-1 fluorescence were unaffected in mtx-2 mutants (Supplementary Fig. 3b, c), indicating that MTX-2 does not function to recruit MIRO-1 to mitochondria. To further support this idea, we performed western blot analysis using the GFP::MIRO-1 knockin strain. Detected by the GFP antibody, MIRO-1 levels were indistinguishable in mtx-2 mutants and the wild-type control, indicating that the steady-state level of MIRO-1 is not affected in mtx-2 mutants ( Supplementary Fig. 3d, e). This result is consistent with a previous study that gem1, the ortholog of miro-1 in S. cerevisiae, is independent of metaxin/SAM50 complex for its mitochondrial localization 31 . We also tested if miro-1 was required for the mitochondrial localization of MTX-2.
In the miro-1 mutant, GFP::MTX-2 was still localized to puncta along the posterior dendrites of PVD (Fig. 2h), suggesting that miro-1 is not required for MTX-2's mitochondrial localization. Hence, MTX-2 and MIRO-1 are localized to mitochondria independent of each other.
MTX-2 directly binds to MIRO-1. Next, we co-expressed tagged versions of MTX-2 and MIRO-1 in HEK293T cells and used coimmunoprecipitation assay to test if they interact with each other biochemically. This experiment showed that FLAG-MTX-2 could co-precipitate with HA-MIRO-1 and vice versa. DRP-1, a mitochondrial protein does not co-precipitate with MTX-2 or MIRO-1 ( Supplementary Fig. 3f-i). To further test if MIRO-1 and MTX-2 bind directly to each other, we expressed tagged proteins in E. coli. GST pull-down experiments using bacterial extracts of tagged proteins showed that, compared with the GST control, GST-MIRO-1 could pull down Trx-MTX-2 but GST could not ( Fig. 2l and Supplementary Fig. 3j), suggesting that MIRO-1 can directly binds to MTX-2.
To further characterize the MIRO-1/MTX-2 complex, we copurified His-MIRO-1 and Trx-MTX-2 and examined the protein complex by analytical gel-filtration (Fig. 2k). Trx-MTX-2 and His-MIRO-1 were co-eluted as one peak, confirming that these two proteins directly bind to each other and form a complex in solution. Taken together, these data support a model in which MTX-2 and MIRO-1 form a complex on the mitochondrial outer membrane to mediate its intracellular transport (Fig. 2m).
MTX-1 plays an important role in mitochondrial trafficking in neurons. Mammalian metaxin1 and metaxin2 function as a complex on the mitochondria outer membrane 32 . The worm orthologue of metaxin1, MTX-1, could be co-immunoprecipitated by MTX-2 when they were co-expressed in HEK293T cells and vice versa ( Supplementary Fig. 4a, b). To test if MTX-1 also functions in mitochondrial trafficking, we constructed a null mutant of mtx-1 using CRISPR-Cas9. Interestingly, mtx-1 mutants were devoid of mitochondria in posterior PVD dendrites but showed a slightly enhanced number of mitochondria in anterior ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-20346-2 dendrites (Fig. 3a, b). Time-lapse imaging experiments showed that mtx-1 mutants showed dynamic behavior defects only in posterior dendrites ( Supplementary Fig. 4c, d). Western blot analysis showed that the level of MIRO-1 was not affected by the mtx-1 mutation ( Supplementary Fig. 3a-e).
The critical difference between anterior and posterior dendrites in PVD is that anterior dendrites contain exclusive minus-end-out microtubule arrays (MTs), similar to other C. elegans dendrites, including the DA9 dendrite. However, posterior PVD dendrites contain plus-end-out microtubules similar to axons in other worm neurons (Fig. 3c) 28 . Consistently, mtx-1 mutants showed reduced mitochondrial localization in the DA9 axon (plus-end-out MTs), but normal mitochondrial localization in the DA9 dendrite (minus-end-out MTs) (Fig. 3e-h). Together, these  data prompt us to propose that MTX-1 is required for mitochondrial transport by the plus end-oriented motor kinesin, but not by dynein.
To test this hypothesis, we examined a partial loss of function allele of the kinesin-1 heavy chain/UNC-116, a conditional unc-116 knockdown strain, as well as a loss-of-function allele of a kinesin light chain/klc-1. Consistent with our hypothesis, all three strains showed a reduced number of mitochondria in posterior dendrites (Fig. 3i, k). These strains also showed reduced mitochondria in anterior dendrites (Fig. 3i, j), which is likely due to the reversal of microtubule polarity in anterior dendrites of kinesin mutants. We have previously shown that in unc-116 mutants, anterior PVD dendrites change their MTs polarity and contain mostly plus-end-out microtubules, while posterior dendrites preserve their plus-end-out MTs (Fig. 3c, d) 28 .
Similar to the loss of mitochondria in PVD, the unc-116 mutation completely blocked mitochondrial transport into both the DA9 axon and dendrite ( Fig. 3e-h). This is not surprising since the same mutation also reverses MTs polarity in the DA9 dendrite to plus-end-out MTs 33 . klc-1 mutants did not show a phenotype in DA9 ( Fig. 3e-h), which might be due to the redundancy between KLC-1 and KLC-2. KLC-1 and KLC-2 are two closely related kinesin light chains sharing 46% sequence identity. klc-2 null mutations cause larval lethality, precluding phenotypic analysis in neurons. However, we circumvented this complication by utilizing another weak allele of klc-2, which did show reduced mitochondrial number in the DA9 axon and dendrite ( Fig. 3e-h). Together, these data suggest that MTX-1 is only required for kinesin-1-mediated mitochondrial transport, while MTX-2 and MIRO-1 are required for both kinesin-and dynein-mediated mitochondrial transport.
MTX-1 functions together with MTX-2 and MIRO-1. To understand the relationship between mtx-1 and mtx-2, we ask if they rely on each other to be localized to mitochondria. A singlecopy GFP::MTX-1 transgene showed a mitochondrial localization pattern in PVD (Fig. 4a). This localization was completely lost in the mtx-2 mutant (Fig. 4a), suggesting that mitochondrialocalized MTX-1 needs MTX-2 for targeting or stabilization. On the other hand, MTX-2 was localized to neuronal mitochondria normally even in the absence of MTX-1 (Fig. 4b).
These results are surprising because metaxin1 was shown to recruit metaxin2 to mitochondria in cultured mammalian cells 32 . Unlike in the mtx-2 mutant, MTX-1 still localized to mitochondria in miro-1 mutants (Fig. 4c).
Next, we created compound mutants between mtx-1, mtx-2, and miro-1 to dissect the relationship between them. As predicted, mtx-1 mtx-2 double mutants completely lacked mitochondria in both anterior and posterior dendrites in PVD and DA9 dendrites (Supplementary Figs. 4e-g and 2a-c). Similarly, mtx-1 miro-1 double mutants showed a very similar phenotype to that of the mtx-1 mtx-2 double mutants in both PVD and DA9 dendrites (Supplementary Figs. 4e-g and 2a-c). Since single mutations only caused a partial reduction of mitochondria in the DA9 axon, we used the enhancement of phenotypes to study the genetic interactions between these three genes. Interestingly, mtx-1 mtx-2 double mutants did not show an enhanced phenotype compared to single mutants (Supplementary Fig. 2a-c), suggesting that they function in the same genetic pathway. Both mtx-1 miro-1 and mtx-2 miro-1 completely lacked mitochondria in the DA9 axon, indicating that MIRO-1 functions at least partially in parallel to MTX-1/2 ( Supplementary  Fig. 2a-c). The triple mutants also showed complete transport defects in the DA9 axon and dendrite ( Supplementary Fig. 2a-c).
can substitute for MIRO-1 for mitochondrial transport by kinesin but MIRO-1 is absolutely required for trafficking towards the minus end of microtubules by dynein.
To further test if MIRO-1 interacts with Mtx1 and Mtx2 on mitochondria, we isolated mitochondria from HEK293T cells and performed the pull-down experiment using a Miro1 antibody. Both Mtx1 and Mtx2 could be detected in the immunoprecipitate, further supporting the notion that these three proteins interact with each other on the mitochondrial outer membrane (Fig. 4h).
The MTX-2/MIRO-1/MTX-1 complex links KHC/UNC-116 through light chain KLC-1. Miro is coupled to kinesin-1 either through direct binding to its kinesin heavy chain or indirectly through TRAKs 21,22 . Given that both MTX-2/MTX-1 and KLC-1 are important for mitochondria transport in PVD, we tested the direct binding between the MTX-1/MTX-2 complex, MIRO-1, and KLC-1, which functions as the cargo-binding subunit for kinesin-1 for many cargoes 34 . First, GST-MIRO-1 could pulldown the MTX-1/2 complex (formed by co-expression of Trx-  Supplementary Fig. 3j). To further prove that these three proteins form a complex, we used analytical gel-filtration analysis to show that His-MIRO-1 could be coeluted with the MTX-1/2 complex (Trx-MTX-1 and Trx-MTX-2) in a single major peak that contained all three proteins (Fig. 4i). These results strongly suggest that MIRO-1 binds to the MTX-1/2 complex directly. Moreover, in the additional pull-down assay with changing the Trx-MTX-1 dosage, no visible difference in the amount of Trx-MTX-1 was observed in the complex, confirming that these three proteins indeed exist in the same complex ( Supplementary Fig. 4i).
Next, with the same type of pull-down experiments and gelfiltration analysis, we showed that the MTX-1/2 complex could directly bind to the TPR domain of KLC-1, which is the primary cargo-binding domain of kinesin light chains, and the three proteins (Trx-MTX-1, Trx-MTX-2, and His-KLC-1-TPR) could be co-eluted as one major peak in gel-filtration (Fig. 4k, l and Supplementary Fig. 3j). Lastly, GST-MIRO-1 could pull-down the MTX-1/MTX-2 complex together with KLC-1-TPR and all four proteins (Trx-MTX-1, Trx-MTX-2, His-MIRO-1, and His-KLC-1-TPR) could be co-eluted together as a single peak in our gelfiltration analysis (Fig. 4m, n and Supplementary Fig. 3j). These biochemical data using purified proteins demonstrate that the metaxin complex can form a complex with MIRO-1 and a kinesin light chain. Together with the strong loss-of-function phenotype of these genes, we propose that the MIRO-1/metaxins tri-protein complex functions as an essential adaptor to couple mitochondria to kinesin-1 for trafficking to neurites (Fig. 4o).
The MTX-2/MIRO-1/TRAK-1 complex mediates mitochondrial transport by dynein. To further understand mitochondrial trafficking to the anterior dendrite, which contains predominantly minus-end-out microtubules, we specifically ask whether dynein is involved and how metaxins link to dynein. The dynein complex is composed of a heavy chain, a light chain, a light intermediate chain, and intermediate chains. We obtained a dli-1 (dynein light intermediate chain) mutant in our screen, which blocked mitochondrial localization in PVD anterior dendrites but had no effect on mitochondria in posterior dendrites (Fig. 5a-c). In mammalian cells, Miro and Milton/TRAK also couple mitochondria to dynein. We then generated a null mutant of trak-1, the sole orthologue of Milton/TRAK in worm, and found that this mutation showed a very similar effect as the dli-1 mutation in the distribution of mitochondria in anterior dendrites ( Fig. 5a-c). This result suggests that while MTX-1 is specifically required for plus-end-out movements, trak-1 is required for minus-end-out movement of mitochondria. Our finding is consistent with experiments in DA9 ( Fig. 5d-f), showing that trak-1 only blocked movement of mitochondria to the dendrite but not the axon. Time-lapse imaging of trak-1 showed that "extension" events were reduced and no "transport" events were observed during a total of 14.5 h of recordings in the anterior dendrite, while it was normal in the posterior dendrite, indicating it functions solely in the trafficking of mitochondria to the anterior dendrite ( Supplementary Fig. 5j). These data suggest that TRAK-1 in C. elegans is only required for dynein-mediated mitochondrial transport, but not required for kinesin-1-mediated transport, while MTX-1 does the opposite. Consistent with this model, trak-1 mtx-1 double mutants nearly completely lost mitochondria in both PVD anterior and posterior dendrites ( Supplementary Fig. 5a-c). In DA9 axons, a trak-1 mutation could not enhance mtx-1 or mtx-2 mitochondria phenotype, which is due to reduced kinesin-1-mediated transport (Supplementary Fig. 5d, f).
In mammalian and Drosophila neurons, TRAK bridges Miro and motor proteins. To test if C. elegans TRAK-1 performs a similar function, we performed co-immunoprecipitation experiments by co-expressing HA-MIRO-1 and FLAG-TRAK-1 in HEK293T cells. We found that HA-MIRO-1 could indeed be efficiently co-precipitated by FLAG-TRAK-1 and vice versa ( Supplementary Fig. 5g, h), indicating that this interaction is conserved. Since MTX-2 is also required for trafficking of mitochondria to anterior dendrites, we also tested potential interaction between TRAK-1 and MTX-2 with the same method. We found that HA-MTX-2 could be co-precipitated by FLAG-TRAK-1 and vice versa ( Supplementary Fig. 5g, i), albeit not as strong as MIRO-1. This interaction is likely to be specific because the FLAG tag alone or FLAG-DRP-1 did not pull-down any HA-MTX-2 or HA-TRAK-1 ( Supplementary Fig. 3f, g). Together, these data suggest that MTX-2, MIRO-1, and TRAK-1 form a complex to link mitochondria to the dynein motor (Fig. 5k). However, the unstable nature of TRAK-1 as a purified protein prevented us from testing this hypothesis using analytical gelfiltration analysis and GST pull-down assays.
Impairment of mitochondrial trafficking leads to dendrite degeneration. Mitochondria play a pivotal role in neuronal function by producing ATP and buffering intracellular calcium.  To understand the physiological significance of dendritic mitochondria, we examined the effect of mitochondrial trafficking mutants on PVD dendritic integrity. Strikingly, miro-1, mtx-2, and trak-1 mutants, which all lacked mitochondria in the anterior dendrite, displayed age-dependent degeneration of distal secondary and quaternary anterior dendrites (Fig. 6a-d). Minor morphological phenotypes were observed in day 1 adults, indicating that dendritic outgrowth is largely normal. However, maintenance of the PVD dendrites appears to require local mitochondria in the dendrites. Degeneration occurred only in anterior dendrites without affecting posterior dendrites, correlating with the specific loss of mitochondria from anterior dendrites in these three mutants. Consistently, we did not observe any degeneration of anterior dendrites in the mtx-1 mutant (Fig. 6b-d). We also did not observe any degeneration of posterior dendrites in the mtx-1 mutant or any other mutants. We speculate that the short length of the posterior dendrite allows ATP or other mitochondria-derived molecules to efficiently diffuse from the soma to dendrites, and therefore posterior dendrites are less susceptible to mitochondrial trafficking defects. Consistent with this notion, even the degeneration of the anterior dendrite only occurred in the distal dendrite covering roughly 2/3 of the normal dendritic arbor. Together, these data suggest that dendritic mitochondria play critical roles in locally maintaining dendrite morphology during the aging process.
Loss of the metaxin complex impairs mitochondria transport in human neurons. Thus far, we have shown that MTX-1 and MTX-2 form a vital adaptor complex to mediate mitochondrial transport into the dendrites and axons of neurons in C. elegans.
To test if their functions are conserved in human neurons, we knocked down Mtx1, Mtx2, Trak1, and Trak2 using RNAi in induced pluripotent stem cell (iPSC)-derived neurons from a healthy human subject ( Fig. 7a and Supplementary Fig. 6a-d).
We transfected those neurons with mito-DsRed to label mitochondria and recorded live mitochondria movements in axons. Consistent with previous reports, Trak1, and Trak2 knockdown reduced mitochondrial movements and increased stop frequency in both anterograde and retrograde directions (Fig. 7a-d), suggesting that the TRAK adaptors are important for both kinesin and dynein-mediated mitochondria transport. Notably, knockdown of Mtx1 or Mtx2 significantly compromised mitochondrial movement in both directions, as evidenced by the significant reduction in the percentage of time that mitochondria spent in motion, the increase in stop frequency, and the decrease of turn back frequency (Fig. 7a-e). These results demonstrate that the function of metaxin in mitochondria trafficking is conserved in human neurons.
In summary, we have identified an evolutionarily conserved protein complex that acts as a crucial adaptor necessary for linking mitochondria with microtubule motor proteins. This complex functions together with the known MIRO-1 adaptor protein and likely generates two separate adaptor complexes to accommodate movement by kinesin and dynein motors.

Discussion
To better understand the molecular mechanisms of mitochondrial trafficking in neurons, we established an in vivo system using the PVD sensory neuron in C. elegans. Several features of PVD are distinct from existing Drosophila and mammalian neuronal culture systems, both of which have made important contributions to our understanding of mitochondrial trafficking. First, the clear mitochondrial distribution within the elaborate PVD dendritic arbors makes it possible to visualize and study mitochondria trafficking in dendrites with single-cell resolution. Second, the distinct microtubule polarity in anterior and posterior dendrites is useful for distinguishing specific motor programs that are affected by mutations. Third, our ability to perform dynamic imaging of mitochondrial "extension", "fission" and "transport" from soma to neurites helps to pinpoint the precise molecular defects in mutants. Last, it appears that mitochondrial trafficking mutants that we have isolated so far do not affect organismal viability, making the isolation of homozygous mutants possible. In a forward visual genetic screen covering 4000 haploid genomes, we recovered alleles of several known genes for mitochondrial trafficking and fission including miro-1, dli-1, and drp-1 (dynamin-related protein 1).
Mammalian kinesin-1 is composed of two motor subunits (heavy chains, KHCs) and two light chains (KLCs) 34 . Usually, KLCs but not KHCs bind to cargo vesicles and proteins 34 . However, strong genetic evidence indicates that KHC can transport mitochondria in a light chain-independent manner 21 , although kinesin light chains are found on mitochondria by several studies [38][39][40] . Indeed, KLC antagonizes KHC's ability to bind to Milton and is absent from the Milton-KHC complex 21 . C. elegans only contains a single Milton like molecule, TRAK-1, but the trak-1 mutant does not show any defects in the mitochondrial distribution in PVD posterior dendrites or the DA9 axon ( Fig. 5a-d and Supplementary Fig. 5f), which are both dependent on kinesin-1-mediated transport. Together, these data suggest that mitochondria might be transported by two different kinesin complexes, one of which contains Milton/TRAK but without KLC, the other contains KLC and both metaxins. The PVD and DA9 neurons rely on the latter complex.
For dynein-mediated transport, miro-1, mtx-2, trak-1, and dli-1 are required. Similar to Drosophila and mammalian cells, the importance of MIRO-1 and TRAK-1 in dynein-mediated transport is supported by the strong loss of mitochondria from anterior PVD dendrites as well as the DA9 dendrite, both of which contain nearly pure minus-end-out microtubules. Interestingly, MTX-2 but not MTX-1 is also essential for dynein-mediated transport. We can detect a clear direct interaction between MIRO-1 and MTX-2 (Fig. 2k, l and Supplementary Fig. 3j), suggesting that these two mitochondrial proteins might exist in a complex. In contrast to the kinesin-1 adaptor complex, for dynein, MIRO-1's function is essential because overexpression of MTX-2 could not rescue the miro-1 phenotype. Consistent with this result, Miro1 is required for TRAK2-dependent retrograde redistribution of mitochondria in mammalian neurons 23 .
We further investigated if MTX-1 and MTX-2's function in mitochondria transport is conserved in vertebrate neurons. The fact that we have also detected mitochondrial trafficking defects in both anterograde and retrograde directions in human neurons suggests that MTX-1 and MTX-2 homologs are likely to play important roles in both kinesin-1-and dynein-mediated mitochondrial transport (Fig. 7a-e). Compared with the null mutant alleles in C. elegans, the RNAi approach likely represents a partial loss of MTX-1 and MTX-2. We have chosen the sparse transfection method in iPSC-derived neurons in order to track individual mitochondrial movement within defined axons and determining the directionality of movements 24,41 , although we recognize that the CRISPR/Cas9 method potentially can generate more consistent knockdown and represents a more reliable way to assess protein function. Detailed analyses of MTX-1 and MTX-2 in vertebrate neurons will be needed to define the adaptor complexes responsible for anterograde and retrograde trafficking.
There are likely differences between the mammalian cells and worm neurons in terms of the exact motor-adaptor complexes responsible for mitochondria transport for several reasons. First, the mammalian TRAK proteins are required for both kinesinand dynein-mediated transport. However, our data showed that trak-1 is likely only required for dynein-but not kinesinmediated mitochondria transport in C. elegans neurons. Second, the mechanism for localizing MTX-2 to mitochondria is likely different between worm neurons and mammalian cells. In worms, MTX-2 localizes to mitochondria independent of MTX-1 (Fig. 4b). On the contrary, one study showed that MTX-1 recruits MTX-2 to mitochondria in mammalian cell 32 . It is therefore plausible that knocking down MTX-1 in iPSC-derived neurons might indirectly affect MTX-2's localization to mitochondria, and thus affect dynein-mediated retrograde transport. Third, while C. elegans only have two metaxins, the human genome contains three metaxins.
The physiological significance of local mitochondria in neurites has been mostly studied in the axon. Mitochondria provide energy for efficient action potential firing, participating in synaptic transmission and regulating synaptic Ca 2+ levels 1,42 . However, the function of local mitochondria in dendrites has not been systematically studied. Mitochondrial trafficking mutants that reduce anterior PVD dendrites show a striking agedependent degeneration. The phenotype is particularly strong in the distal half of the dendrite, suggesting that distal dendrites might suffer from depleted ATP due to the inefficient diffusion of ATP in neurites. Such intracellular energy gradients have been reported in mouse embryonic fibroblasts (MEFs), which are indeed shaped by mitochondrial distribution 43 . The exact mechanisms for such degeneration remain to be studied.
In summary, we have identified previously unknown mitochondrial adaptor proteins that play essential roles in connecting mitochondria to molecular motors. MTX-2 and MIRO-1 are core components of both kinesin-1 and dynein adaptor complexes, whereas MTX-1 and TRAK-1 specify the adaptors for kinesin-1 and dynein, respectively.
CRISPR/Cas9-assisted knockout and knockin worms were generated 46 . The guide RNA target sequence was selected according to the design tool (http:// crispor.tefor.net/). For GFP::MIRO-1 knockin animals, gfp was inserted between miro-1 promoter and the coding sequence. gfp fragment was inserted betweeñ 1.3 kb upstream and~1.3 kb downstream homologous sequence amplified from the N2 genomic DNA, and then assembled into pPD95.77 backbone as gfp knockin repair template plasmid. Synonymous mutations were introduced to Cas9 target site to prohibit the cleavage of repair template by Cas9. The gfp knockin repair template plasmid (30 ng/μl), Cas9-gRNA plasmid (50 ng/μl), and selective marker Pord-1 > gfp (30 ng/μl) were co-injected into N2 worms. The gfp knockin worms were identified and confirmed by PCR and Sanger sequencing.
All of the strains and the plasmids used are listed in Supplementary Tables 1 and 2, respectively. Primers used for genotyping and plasmids construction are listed in Supplementary Tables 3 and 4, respectively.
Mitochondrial isolation, immunoprecipitation, and western blotting. Cell mitochondria isolation kit (Beyotime) was used for mitochondrial isolation from HEK293T cells. The experiment was carried out according to the instructions of the kit. In brief, the cultured cells were collected and mechanically homogenized. After centrifugation at 1000 × g for 10 min at 4°C, crude supernatant was spun at 3500 × g for 10 min at 4°C to pellet intact mitochondria. The acquired pellet was resuspended in lysis buffer (Beyotime) with 0.25 mM PMSF and protease inhibitors as the mitochondrial fraction. Then the supernatant was collected carefully and centrifuged at 12,000 × g for 10 min at 4°C. The new supernatant was regarded as the cytosolic fraction. Both mitochondrial fraction and cytosolic fraction were incubated with 50 µl mouse anti-Miro1 antibody (ab188029, AbCam) at 4°C overnight, respectively, and then were mixed with 200 µl washed Pierce Protein A/ G Magnetic Beads (88802, Thermo Fisher Scientific) for 2 h at 4°C, respectively. Beads were then washed four times with wash buffer (150 mM NaCl, 50 mM Tris-HCl) with 0.25 mM PMSF and protease inhibitor cocktail. Then the beads were resuspended in SDS loading buffer and boiled for 10 min before the samples were loaded into SDS-PAGE gel. After electrophoresis, proteins were transferred onto a PVDF membrane. Transferred membranes were blocked for 1 h in Tris Buffered Saline Tween (TBST) containing 5% fat-free milk, and then incubated with the following primary antibodies: mouse anti-Miro1 (WH0055288M1, Sigma-Aldrich) at 1:1000, mouse anti-Metaxin1 (sc-514469, Santa Cruz Biotechnology) at 1:500, mouse anti-Metaxin2 (sc-514231, Santa Cruz Biotechnology) at 1:500, mouse anti-Tom20 at 1:500 (sc-17764, Santa Cruz Biotechnology), at room temperature for 4 h. After washing three times using TBST, 10 min each, the membranes were incubated with the HRP-conjugated Affinipure Goat Anti-Mouse IgG(H + L) (SA00001-1, Proteintech) at 1:2500 at room temperature for 2 h. After three times washing of TBST, pictures were taken.
Co-immunoprecipitation assay. Co-immunoprecipitation assay was used to detect the protein-protein interaction in vivo. The plasmids expressing Flag-tagged bait proteins and HA-tagged prey proteins were cloned into pNTAP vector individually. Each bait and prey pair plasmids were cotransfected into HEK293T cells and the transfected cells were harvested after 48 h. Then the cells were lysed using NP40 Cell Lysis Buffer (Thermo Scientific™ Catalog #FNN0021) with protease inhibitors (Sigma Catalog #P8340) on ice for 30 min. The lysates were pre-cleared with centrifugation at 12,000 × g for 10 min at 4°C. The supernatant of each sample was mixed with 50 μl ANTI-FLAG® M2 Affinity Gel (Merck Catalog #A2220). The mixture was incubated for 4 h at 4°C with gentle rotation. Before adding to the mixture, the ANTI-FLAG® M2 Affinity Gel was washed with NP40 Cell Lysis Buffer three times. After rotation, the beads were washed by NP40 Cell Lysis Buffer with protease inhibitors four times. The bound proteins were eluted with SDS-PAGE buffer by incubating the sample for 10 min at 98°C. The immunoprecipitates were subjected to Western blotting using the indicated antibodies.
GST pull-down assay. All proteins used in the GST pull-down assay were expressed in E. coli. The cell lysates containing GST or GST-tagged MIRO-1 proteins were mixed with cell lysates containing His-tagged or Trx-tagged proteins (Trx-MTX-2, co-expressed Trx-MTX-1/Trx-MTX-2, co-expressed Trx-MTX-1/ Trx-MTX-2 with additional Trx-MTX-1, or co-expressed Trx-MTX-1/Trx-MTX-2 and His-KLC-1-TPR), respectively, in each experiment. The cell lysates containing co-expressed GST/Trx-MTX-2 or co-expressed GST-MTX-1/Trx-MTX-2 proteins were mixed with cell lysates containing His-tagged KLC-1-TPR proteins, respectively. After centrifugation, supernatants in PBS were incubated with Glutathione Sepharose 4 Fast Flow (GE Healthcare) beads at 4°C overnight. After washing the beads five times with PBS, the proteins captured by the beads were eluted with PBS containing 20 mM glutathione and detected by SDS-PAGE with Coomassie-blue staining.
Neuronal derivation from iPSCs and transfection. The iPSC line in this study has been fully characterized by the previous studies 49,50 . iPSCs were derived to midbrain dopaminergic neurons 49,[51][52][53][54] . In brief, neurons were generated using an adaptation of the dual-smad inhibition method with the use of smad inhibitors dorsormorphin (Sigma) and SB431542 (Tokris), and the addition of GSK3β inhibitor CHIR99021 (Stemgent). SHH was replaced with the smoothened agonist SAG. To gain a higher purity of neural precursor cells, 12 days after neural induction, rosette-forming neuroectodermal cells were manually lifted and detached en bloc, and then cultured in suspension in a low-attachment dish (430589, Corning Inc.) with N2 medium with 20 ng/ml BDNF, 200 µM Ascorbic Acid, 500 nM SAG, and 100 ng/ml FGF8a. On day 17, neurons were transferred onto poly-ornithine and laminin-coated glass coverslips in a 24-well plate. On day 18, medium was switched to N2 medium supplemented with 20 ng/ml BDNF, 200 µM Ascorbic Acid, 20 ng/ml GDNF, 1 ng/ml TGFβ3, and 500 µM Dibutyryl-cAMP for the maturation of dopaminergic neurons. Neurons were used at day 21-22 after neuronal induction, when about 80-90% of total cells expressed the neuronal marker TUJ-1, and 12% of total cells expressed TH and markers consistent with ventral midbrain neuronal subtypes 50,55 .
For transfection, on day 18-19 after neuronal induction, the culture medium was replaced with Opti-MEM (Gibco) prior to transfection. 100 nM RNA, 0.5 μg EGFP, and 0.5 μg mito-dsRed, or 5 μl Lipofectamine 2000 was diluted in Opti-MEM at room temperature (22°C) to a final volume of 50 μl in two separate tubes, and then contents of the two tubes were gently mixed, incubated for 20 min at room temperature, and subsequently added onto neurons. After transfection for 6 h, Opti-MEM containing RNA-DNA-Lipofectamine complexes was replaced with regular N2 medium. After transfection for 3 days, neurons were imaged.

Live image acquisition and quantification
Worm. C. elegans were anesthetized using 1 mg/ml levamisole in M9 buffer, and then mounted on 3% (w/v) agar pads. All images for control and the corresponding mutants were taken by the spinning disk confocal imaging system (Yokogawa CSU-X1 Spinning Disk Unit) at the same condition. For most images showed in this study, late L4 stage animals were imaged using a 100×/1.45 NA objective, and then processed by Image J. For GFP::MIRO-1 expression level quantification, gastrulation stage embryos were imaged, and fluorescence intensity of the single layer of the largest cell number was quantified by Image J. Mitochondrial dynamics were recorded around the cell body or primary dendrites of L3 or early L4 stage worms. For movies around the cell body region, dynamic images were recorded continually with 6-s interval for 30 min. Mitochondrial dynamics were classified into three events: "extension", "fission", and "transport". The frequency of events was calculated as the ratio between the worms displaying an event within 30 min of recording and the total number of worms recorded. In primary dendrite, time-lapse movies were taken with 3-s interval for 10 min. Mitochondrial behaviors were characterized by state (mobile or stable), the direction of motion (anterograde or retrograde), velocity. The velocity of mitochondrial movement was quantified by Imaris (Bitplane). For mitochondrial distribution quantification, worms at day 1 stage were examined under 63×/1.4 NA objective using an Axio Imager M2 microscope (Carl Zeiss).
Human cell. Neurons on glass coverslips were placed in a 35-mm petridish containing the Hibernate E low-fluorescence medium (BrainBits) on a heated stage of 37°C, and imaged with a 63×/N.A.0.9 water-immersion objective with excitation at 561 nm or 488 nm. We imaged axons positive for both mito-dsRed and EGFP which likely took up RNA duplex. Axons longer than 50 μm were selected for recording. Time-lapse movies were obtained continually with 5-s intervals for 3-5 min. For quantification, kymographs were generated from time-lapse movies by ImageJ, representing a 100-s period. Each kymograph was then imported into a macro written in Labview (NI, TX), and individual mito-dsRed puncta were traced using a mouse-driven cursor at the center of the mito-dsRed object. Using Matlab (The MathWorks, MA), we determined the following parameters: (1) the instantaneous velocity of each mitochondrion, (2) the average velocity of those mitochondria that are in motion, (3) the percent of time each mitochondrion is in motion, (4) stop frequency, and (5) turn back frequency.
Statistics and reproducibility. Statistical analyses were performed using Graph-Pad Prism 9.0.0.121. Two-tailed unpaired Student's t test, one-way ANOVA with Tukey's multiple comparisons test, or Fisher's exact test was used to calculate the p values. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001 are considered significant. The types of the statistical tests, the sample size, exact p values, and statistical significance are reported in the figures and corresponding figure legends. Data shown in graphs represent the mean ± standard error of the mean (SEM), and individual data points are plotted whenever possible. Statistical analysis is conducted on data from at least three biologically independent experimental replicates excepted when otherwise stated. Micrographs and biochemical images shown in the figures are representative of three or more independent experiments with similar results. The source data for statistical analyses can be found in the source data file.