MICAL3 Flavoprotein Monooxygenase Forms a Complex with Centralspindlin and Regulates Cytokinesis*

During cytokinesis, the antiparallel array of microtubules forming the central spindle organizes the midbody, a structure that anchors the ingressed cleavage furrow and guides the assembly of abscission machinery. Here, we identified a role for the flavoprotein monooxygenase MICAL3, an actin disassembly factor, in organizing midbody-associated protein complexes. By combining cell biological assays with cross-linking mass spectrometry, we show that MICAL3 is recruited to the central spindle and the midbody through a direct interaction with the centralspindlin component MKLP1. Knock-out of MICAL3 leads to an increased frequency of cytokinetic failure and a delayed abscission. In a mechanism independent of its enzymatic activity, MICAL3 targets the adaptor protein ELKS and Rab8A-positive vesicles to the midbody, and the depletion of ELKS and Rab8A also leads to cytokinesis defects. We propose that MICAL3 acts as a midbody-associated scaffold for vesicle targeting, which promotes maturation of the intercellular bridge and abscission.

Cytokinesis is the last step of cell division, during which the cytoplasm is partitioned between the two daughter cells. In animal cells, the major steps of cytokinesis are the formation of the actomyosin ring and its contraction, resulting in the ingression of the cleavage furrow, formation of an intercellular bridge, and its subsequent scission (1,2). At the onset of cytokinesis, microtubules at the spindle midzone arrange into an overlapping antiparallel array, the central spindle, which first forms a signaling platform important for the positioning of the cleavage plane and later organizes the midbody, an electron-dense structure that anchors the ingressed furrow and guides the assembly of abscission machinery (1-3).
One of the major players in the assembly and function of the central spindle is the centralspindlin complex, composed of the kinesin-6 family protein mitotic kinesin-like protein 1 (MKLP1) 4 (also known as KIF23 in mammals) and the Rho GTPase activating protein CYK-4 (also known as or MgcRac-GAP) (4). Centralspindlin participates in bundling antiparallel microtubules of the central spindle, activation of the Rho GTPase, which controls formation of the actomyosin ring, and the formation of the midbody (4).
Membrane trafficking, including both endosomes and secretory carriers, also plays an important role in cytokinesis (5)(6)(7). The small GTPase Rab8, known to label both endosomal and secretory vesicles (8 -11), prominently accumulates at the midbody (12)(13)(14), and its transport to the intercellular bridge is directed by cytoplasmic dynein and the doublecourtin domaincontaining protein-5 (12). The docking and fusion of Rab8positive secretory vesicles with the plasma membrane depends on the cortical complex, which contains a coiled-coil adaptor ELKS/ERC1 (15) and the flavoprotein monooxygenase enzyme MICAL3 (8). The members of the MICAL family, which in mammals includes MICAL1, MICAL2, and MICAL3, are best known for their ability to promote actin disassembly by oxidizing specific methionine residues (16 -18). In addition, MICAL1 and MICAL3 can bind to multiple members of the Rab GTPase family, including Rab8 (8,19), but the exact function of these interactions is poorly understood.
Here, we searched for new binding partners of MICAL3 and identified the two components of the centralspindlin complex among the highest hits. Because centralspindlin is a major player during cytokinesis, we hypothesized that MICAL3 also participates in this process. We tested this possibility using a combination of biochemical and cell biological experiments and found that MICAL3 is indeed required for cytokinesis. We showed that MKLP1 recruits MICAL3 to the central spindle, whereas MICAL3 participates in targeting to the midbody its binding partners ELKS and Rab8A, and the loss of both ELKS and Rab8A also caused cytokinesis defects. Interestingly, the cytokinetic function of MICAL3 appeared not to require its enzymatic activity. Our data suggest that MICAL3 plays a role in organizing membrane trafficking at the intercellular bridge and that it acts in this process not as an actin disassembly factor but as a protein-binding hub.

MICAL3
Interacts with MKLP1-To get insight into the function of MICAL3, we performed a mass spectrometrybased screen for its binding partners. We co-expressed in HEK293T cells the biotin ligase BirA together with MICAL3 carrying a bioGFP tag, which consists of GFP with an N terminally attached artificial peptide that can be efficiently biotinylated by BirA (20), and performed streptavidin pulldown assays from the lysates of these cells. BioGFP alone was used as a negative control. MICAL3 was the highest specific hit in this experiment, and the already known partners of MICAL3, ELKS (ERC1) and Rab8A, were also identified with high confidence (Fig. 1A). We also identified ␣and ␤-spec-  trins (SPTAN and SPTBN1; Fig. 1A), which are large cortical proteins involved in the control of mechanical stability of cellular membranes and assembly of cortical cytoskeleton and specialized membrane domains (21)(22)(23). Because MICAL3 can regulate actin disassembly, its binding to spectrins might play a role in controlling the organization of the cortical actin networks.
Among the other major hits in this screen were the two components of the centralspindlin complex, MKLP1 (KIF23) and CYK-4 (RACGAP) (Fig. 1A). The validity of the interaction with centralspindlin is supported by the fact that MICAL3 was previously reported as one of the potential partners of MKLP1 in HeLa cells in a high-throughput screen (24).
To investigate the interaction between MICAL3 and MKLP1 in more detail, we performed cross-linking mass spectrometry experiments. BioGFP-MICAL3 was co-expressed in HEK293 cells together with GFP-MKLP1 and BirA, biotinylated proteins were isolated by streptavidin pulldown, cross-linked on beads using disuccinimidyl sulfoxide (DSSO), and the crosslinked peptides were analyzed by mass spectrometry as described previously (25). We identified multiple cross-links within both MICAL3 and MKLP1. The cross-links within folded domains of known structure (MICAL1 monooxygenase, MKLP1 C terminus) were in good agreement with the DSSO distance constraint of 28.4 Å (23.4 Å for the maximal distance constraint of the DSSO cross-linker plus 5-Å tolerance for protein flexibility in solution), validating our cross-linking mass spectrometry approach (Fig. 1, B and C, and supplemental Table S1). We also found many cross-links between different domains of MICAL3, compatible with the previously proposed autoinhibitory folding of MICAL proteins ( Fig. 1B and supplemental Table S1) (8,26).
Specific cross-links were also observed between the second and third coiled-coil region of MICAL3 and the coiled-coil and the adjacent C-terminal domain of MKLP1 (Figs. 1C and 2A, and supplemental Table S1). Pulldown assays confirmed the cross-linking results and showed that the minimal MKLP1binding domain of MICAL3 encompasses its C-terminal coiled-coil region (MICAL3-CC4), whereas a stronger interaction was observed with a larger C-terminal fragment, which includes all the cross-linked sites (MICAL3-CC3, Fig. 2

, B-E).
Because the C termini of MICAL1 and MICAL3 are quite similar, we have also tested whether MICAL1 can interact with MKLP1, but found that this was not the case (Fig. 2D).
The minimal MICAL3-binding domain of MKLP1 was present in the strongly charged and likely unstructured region located at the C-terminal side of the coiled coil of MKLP1 (Figs. 2A and 3, A-C). By deletion mapping, we identified a region of 13 residues, which was essential for the interaction (mutant D7), whereas the deletion of the adjacent 15-amino acid region (mutant D8) did not perturb the binding (Figs. 1C and 3, A and D, and supplemental Table S1). Taken together, our results show that MKLP1 and MICAL3 primarily interact through the unstructured polypeptide region in MKLP1 and the C-terminal coiled-coil domain of MICAL3, with additional contacts provided by the second coiled-coil domain of MICAL3 and the coiled-coil domain of MKLP1.
MKLP1 Recruits MICAL3 to the Central Spindle and the Midbody-We next examined the localization of endogenous MICAL3 in mitotic cells and found that whereas no conspicuous localization of MICAL3 could be detected at the early cell division stages, MICAL3 was clearly enriched at the spindle midzone starting from anaphase and strongly accumulated at the central spindle and the midbody during cytokinesis (Fig.  4A). Fluorescently tagged MICAL3 colocalized with the endogenous MKLP1 at the central spindle and the midbody (Fig. 4B). This localization did not depend on the enzymatic activity of MICAL3 as it was observed with the enzymatically dead MICAL3 mutant (Fig. 4C). Importantly, in MKLP1-depleted cells, MICAL3 failed to localize to the spindle midzone, indicating that MKLP1 targets MICAL3 to this location (Fig. 4, D-G).
We next set out to test whether MICAL3 affects the localization of MKLP1. Because we did not manage to achieve an efficient siRNA-mediated knockdown of MICAL3, we used CRISPR/Cas9-mediated knock-out technology. Cells transfected with a vector co-expressing a puromycin resistance gene, a single guide RNA sequence specific for MICAL3 and Cas9 (27) were subjected to puromycin selection and analyzed at 6 -7 days after transfection to avoid the selection of cells, in which the potential cell division defects would be compensated by secondary mutations. This procedure led to highly efficient knock-out of MICAL3 protein expression (Fig. 4H). We found that the loss of MICAL3 had no effect on MKLP1 binding to the central spindle or the midbody (Fig. 4I). We conclude that MICAL3 is recruited to sites of centralspindlin accumulation but has no effect on its distribution.
MICAL3 Is Required for Cytokinesis-To determine whether the loss of MICAL3 affects cell division, we performed phasecontrast-based live imaging experiments and found that compared with the cells transfected with the vector alone, MICAL3 knock-out cells displayed a slight delay between the mitotic onset and the initiation of furrow ingression, and a very significant (more than 2-fold) increase in the time between the completion of furrow ingression and abscission (Fig. 5, A-C, and supplemental Movies S1 and S2). We observed no abnormalities during the formation or ingression of the cleavage furrow and detected no visible defects in the organization of the actin cytoskeleton during different mitotic stages, including cytokinesis (Fig. 6, A and B). The data indicate that the actin-depolymerizing activity of MICAL3 is unlikely to contribute to the dynamics of the actomyosin ring. We did, however, observe a significant increase in the percentage of cells that fused back after furrow ingression and the corresponding increase in the percentage of binucleated cells (from ϳ2 to ϳ10%) (Fig. 5, A, D, and E, and supplemental Movie S3). To test whether this effect of MICAL3 was specific, we performed rescue experiments with GFP-tagged MICAL3 expression constructs (Fig. 5F). Although no rescue was observed with GFP alone, we could completely rescue the binucleated phenotype by either the fulllength MICAL3 or its enzymatically dead mutant, but not with the MICAL3 deletion mutant lacking the MKLP1-binding C terminus (Fig. 5F). Thus, these data indicate that MICAL3 contributes to the stabilization and/or maturation of the intercellular bridge in a way that does not require its monooxygenase activity.
For comparison, we examined the phenotype of MKLP1 depletion and found that, as expected, its effect on cytokinesis was much more profound than that of MICAL3 knockout, with ϳ70% of binucleated cells (Fig. 5G). We observed a very significant although not complete rescue of this phenotype by expressing GFP-MKLP1 (Fig. 5G). The MKLP1 deletion mutant D7, which was unable to bind to MICAL3 and promote MICAL3 localization to the midbody (Figs. 3D and 5H), performed significantly less well in the rescue experiments than the wild type MKLP1 and the MKLP1-D8 mutant, which could restore MICAL3 accumulation at the midbody in MKLP1-depleted cells (Fig. 5, G and H).
Although we cannot exclude that the D7 mutant perturbs MKLP1 function in ways unrelated to MICAL3 binding, our results are fully compatible with a role of MKLP1-dependent recruitment of MICAL3 as a factor contributing to the successful completion of cytokinesis.
MICAL3 Promotes the Recruitment of ELKS and Rab8 to the Intercellular Bridge-Because our data indicated that the contribution of MICAL3 to cytokinesis does not require its actindestabilizing monooxygenase activity, we turned our attention to its binding partners. We first examined the distribution of ELKS, which was shown previously to bind to MICAL3 and participate in secretory vesicle docking (8). We found that ELKS prominently localized to the midbody in 90% of the cells, and that this localization was completely lost in MICAL3 knock-out cells (Fig. 7, A and B). Because MICAL3 is targeted to the midbody by MKLP1, these results suggest that MICAL3 might be able to connect ELKS to MKLP1. In support of this, mass spectrometry-based analysis of BioGFP-ELKS pull downs  from HEK293 cells identified MKLP1 (KIF23) as a potential interacting partner (Fig. 7C). We next tested whether the interaction between MKLP1 and ELKS is mediated by MICAL3 and found that co-expression of MICAL3 indeed enhanced the ability of MKLP1 to co-precipitate ELKS (Fig. 7D). We conclude that MICAL3 can form a triple complex with MKLP1 and ELKS and through the interaction with the former, target the latter to the midbody.
We have also examined the distribution of Rab8A-labeled membranes. Rab8A was present at ϳ75% of the midbodies (Fig.  8, A and B). The intercellular bridges lacking Rab8A-positive structures appeared to represent very late stages of cytokinesis, in line with previous publications showing that Rab8A-positive vesicles are removed from the intercellular bridge prior to abscission (13,14). The percentage of cells that exhibited at least some Rab8A signal at the intercellular bridge was significantly reduced in MICAL3 knock-out cells (Fig. 8, A and B), and the intensity of Rab8A staining in the midbody region was also reduced (Fig. 8C). In contrast, the localization of transferrin receptor-positive or Rab11-positive endosomes to the midbody was not perturbed by MICAL3 knock-out (Fig. 8, D and E). We conclude that MICAL3 specifically contributes to the recruitment of Rab8A to the midbody.
Mass spectrometry-based analysis of ELKS binding partners in HEK293 cells identified citron kinase (CRIK) and its binding partner kinesin-3 KIF14 (Fig. 7C), whereas these proteins were not present in the MICAL3 pull down (Figs. 1A and 7C). We have examined whether the localization of CRIK or KIF14 was perturbed in MICAL3 knock-out cells and found this not to be the case (Fig. 8F). Although these data do not rule out that ELKS, localized to the midbody through its interaction with MICAL3, somehow affects the function of CRIK-containing protein complexes, they indicate that MICAL3 does not participate in the targeting of this signaling protein complex to the intercellular bridge, in line with the data showing that CRIK directly interacts with MKLP1 (28).
ELKS and Rab8A Are Required for Cytokinesis-If the cytokinetic defect of MICAL3 knock-out cells occurs due to the inability to properly localize ELKS and Rab8A-postive membranes, then the depletion of these proteins can be expected to perturb cytokinesis as well. We examined this possibility by depleting ELKS and knocking down ELKS and Rab8A (Fig. 8, G and H); the two proteins were depleted by 75% and 95%, respectively. The intensity of Rab8A staining in the midbody region in ELKS-depleted cells was reduced to values close to background (Fig. 8, A and I). The depletion of either ELKS or Rab8A led to a significant increase of binucleated cells (ϳ10%, similar to MICAL3 knock-out) (Fig. 8, J and K). We conclude that the complex of MICAL3 and ELKS helps to target Rab8A to the intercellular bridge, and that all three proteins are required to the proper maturation of the bridge, which prevents the bridge recession and promotes timely abscission.

Discussion
In this study we identified MICAL3 as a new molecular player in cytokinesis and a binding partner of MKLP1. Previous work has shown that centralspindlin represents a major binding hub for different factors involved in cytokinesis (4), and our current work added MICAL3 to the list of MKLP1 interactors. Crosslinking experiments combined with mass spectrometry allowed us to show that the MICAL3-MKLP1 association is direct and to characterize the contact sites between the two proteins, illustrating the power of this approach. The MICAL3-binding region of MKLP1 is distinct from the MKLP1 sites known to bind to CYK-4, ARF6, or 14-3-3, but there might be some overlap with the binding sites for CRIK and CEP55 (Fig. 3A) (28 -32). The MKLP1 binding site of MICAL3 overlaps with that of Rab8A but is different from the region involved in the interaction with ELKS (8) (Fig. 2B), and in agreement with these data, we showed that MICAL3 can connect ELKS to MKLP1.
Our data indicate that MICAL3 is involved in stabilization and maturation of the intercellular bridge, because its loss led to increased frequency of cytokinetic failure and extended duration of abscission. Surprisingly, despite the well established role of MICAL enzymes in actin disassembly, we found no evidence that MICAL3 regulates actin cytoskeleton during cytokinesis. Instead, MICAL3 recruits ELKS and Rab8A to the midbody (Fig. 9), and the depletion of these two proteins caused an increase of binucleated cells, indicative of cytokinetic defect. The function of MICAL3 in cytokinesis thus appears to be mechanistically similar to that of the MICAL3-related protein MICAL-L1, which lacks an enzymatic domain, localizes to the midbody, and plays a role in organizing recycling endosomes at the intercellular bridge (33).
The next challenge is to identify the exact molecular role of Rab8A vesicles in cytokinesis. Because these vesicles are cleared from the intercellular bridge before the actual abscission (13)(14), we favor the idea that they transport molecules required for the maturation of the intercellular bridge and the midbody, such as some signaling factors. This function would be similar to that of Rab35, which contributes to actin depolymerization by delivering to the intercellular bridge the phosphoinositide phosphatase OCRL (34,35). The role of Rab8A vesicles in protein delivery to the midbody would be consistent with the observation that the enzymatic activity of MICAL3 is not essential for the cytokinetic function despite the fact that it promotes vesicle fusion with the plasma membrane.
The loss of MICAL3, ELKS, and Rab8A caused a similar, relatively mild cytokinetic defect, with ϳ10% binucleated cells, indicating that the absence of these proteins can be compensated by other factors, at least in HeLa cells, although the situation might be different in other systems. It is possible that different vesicular populations at the midbody have partially overlapping functions. Although Rab8A can be recruited to Rab11-positive recycling endosomes (11), the MICAL3-depen-    dent population of Rab8A vesicles at the midbody is distinct from Rab11-labeled endosomes, which were not affected by MICAL3 knock-out. The Rab8A vesicles studied here might overlap with the post-Golgi carriers, which were previously described to participate in cytokinesis (36,37). Golgi-dependent secretory trafficking was shown to be necessary for cytokinesis during development in invertebrates (5,7), and a similar function might be also important in mammals, constituting a potentially interesting avenue for future studies.
The target sequences of control and ELKS siRNAs were described previously (38). MKLP1 siRNA was described in Ref. 39. MICAL1 siRNA was synthesized by Ambion (ID s230028). Rab8A siRNA corresponded to the target sequence GGAA-AGCACAAATGAAGGA. Cells were transfected with 10 nM siRNAs using HiPerFect (Qiagen) and analyzed 3 days after transfection.
The target sequence for CRISPR targeting the human MICAL3 gene was CCCCAGTAAGGATAAGTCGA (Exon 3). Two complementary oligonucleotides with BpiI restriction sites for guide RNAs (gRNAs) were synthesized, and cloned into pX459 CRISPR/Cas9-Puro vector (27) (Addgene). HeLa cells were transfected with Px459 vector or pX459-MICAL3 gRNA and treated with 2 g/ml of puromycin for 2 days and analyzed within a 4-day period post-puromycin treatment.
To label proteins, cells were fixed with Ϫ20°C methanol for 10 min or 3.7% paraformaldehyde in PBS for 10 min at room temperature. Cells were permeabilized with 0.1% Triton X-100 in phosphate-buffered saline (PBS) for 10 min; subsequent washing and labeling steps were carried out in PBS supplemented with 2% bovine serum albumin and 0.05% Tween-20. At the end, slides were rinsed in 70 and 100% ethanol or airdried and mounted in Vectashield mounting medium (Vector Laboratories).

Mass Spectrometry
Sample Preparation-Samples were prepared from pulldown assays of biotinylated proteins from extracts of transfected HEK293T cells using streptavidin beads, and analyzed by subsequent mass spectrometry experiments as described previously (8). For the cross-linking experiment, the immunoprecipitated proteins were on-beads cross-linked using 1 mM DSSO cross-linker (41). The cross-linking reaction was carried out for 1 h at room temperature and quenched with 20 mM Tris-HCl (pH 7.8) for 30 min at room temperature. Subsequently, the on-beads cross-linked proteins were denatured with 2 mM urea, reduced with 4 mM dithiothreitol at 56°C for 30 min and alkylated with 8 mM iodoacetamide at room temperature for 30 min in the dark. Proteins were digested using trypsin at an enzyme-to-protein ratio of 1:20 (w/w) at 37°C for 2 h. The supernatant was removed from the beads and further digested at 37°C overnight. Protein digests were desalted using Sep-Pak C 18 cartridges (Waters), dried, and stored at Ϫ20°C for further use. The desalted digests were SCX fractionated as previously described (25), and only the late SCX fractions (predominantly containing higher charged species) were subjected to LC/MS analysis.
LC/MS/MS Analysis-Peptides and cross-linked peptides were analyzed using an ultra HPLC Proxeon EASY-nLC 1000 system (Thermo Fisher Scientific) coupled on-line to an ETD enabled LTQ Orbitrap Elite mass spectrometer (Thermo Fisher Scientific). For cross-link analysis, reversed-phase separation and MS data acquisition were performed as previously described and data analysis was accomplished using XlinkX software (25). Results were filtered by 1% false discovery rate and all reported cross-links were manually validated. For peptide identification, rapid CID was used for data acquisition. Peptide identification was performed using Proteome Discoverer 1.4 (Thermo Fisher Scientific). Results were filtered by 1% false discovery rate.

Image Acquisition
Images of fixed cells were collected with a Nikon Eclipse 80i upright fluorescence microscope and a CoolSNAP HQ2 CCD camera (Photometrics), using a Nikon Plan Apo VC ϫ60/ N.A.1.40 oil objective or Nikon Plan Apo VC ϫ100/N.A.1.40 ϫ oil objective.
Live-cell phase-contrast imaging was performed on inverted research microscope Nikon Eclipse Ti-E (Nikon) with the perfect focus system (Nikon), equipped with a Nikon Plan Apo-chromat ϫ20/0.75 phase-contrast objective (Nikon), a Cool-SNAP HQ2 CCD camera (Photometrics), and controlled with Micro-Manager Open Source Microscopy Software. To keep the cells at 37°C, a stage top incubator model INUBG2E-ZILCS (Tokai Hit) was used. Images were projected onto the CCD chip at a magnification of 0.3215 m/pixel. Round 25-mm coverslips were mounted in Attofluor Cell Chamber (Thermo Fisher) and maintained at 37°C and 5% CO 2 . Cells were imaged every 2 min for 6 -12 h.
Confocal 300-nm Z-stack images were acquired on a Leica SP8 DMI STED microscope equipped with a ϫ100 HC Pl Apo objective (Leica, NA 1.4. oil) and the Leica HyD detector controlled by the Leica Application Software X. Excitation of Alexa 488, Alexa 568, and Alexa 647 was performed with a SuperK EXTREME supercontinuum laser (NKT photonics) and DAPI was excited using a 405-nm diode laser controlled by a PDL800-B module (PicoQuant). Channels were imaged sequentially with a pixel size between 90 and 110 nm. Presented images are projections of 6 -10 Z-planes, processed in ImageJ.
Author Contributions-A. A. conceived and coordinated the study and wrote the paper. Q. L., F. L., K. L. Y., and A. S-M. designed, performed, and analyzed the experiments. L. C. K. and A. J. R. H. contributed to the coordination of the study, data analysis and writing the paper. R. T., I. G., and S. R. provided technical assistance with the experiments and their analysis and contributed to the preparation of the figures. All authors reviewed the results and approved the final version of the manuscript.