Two NEMO-like Ubiquitin-Binding Domains in CEP55 Differently Regulate Cytokinesis.

CEP55 regulates the final critical step of cell division termed cytokinetic abscission. We report herein that CEP55 contains two NEMO-like ubiquitin-binding domains (UBDs), NOA and ZF, which regulate its function in a different manner. In vitro studies of isolated domains showed that NOA adopts a dimeric coiled-coil structure, whereas ZF is based on a UBZ scaffold. Strikingly, CEP55 knocked-down HeLa cells reconstituted with the full-length CEP55 ubiquitin-binding defective mutants, containing structure-guided mutations either in NOACEP55 or ZFCEP55 domains, display severe abscission defects. In addition, the ZFCEP55 can be functionally replaced by some ZF-based UBDs belonging to the UBZ family, indicating that the essential function of ZFCEP55 is to act as ubiquitin receptor. Our work reveals an unexpected role of CEP55 in non-degradative ubiquitin signaling during cytokinetic abscission and provides a molecular basis as to how CEP55 mutations can lead to neurological disorders such as the MARCH syndrome.

Introduction

In animal cells, cytokinesis leads to the separation of the cytoplasm into two daughter cells. Defects during this process can lead to polyploidy and aneuploidy, situations that are frequently observed in cancers (Ganem et al., 2007). Cytokinesis begins in anaphase with the formation of an actomyosin ring, which constricts the plasma membrane of the cells, finally leading to the formation of an intercellular bridge (Mierzwa and Gerlich, 2014). Future daughter cells are connected by this intercellular bridge for several hours before individual assortment by a process termed abscission, which generally occurs in the G1 phase of the cell cycle (Gershony et al., 2014). At the center of the intercellular bridge is the midbody, a crucial component of the abscission process. It plays a role as the recruitment platform for abscission and acts as a stabilizer of the intercellular bridge (Mierzwa and Gerlich, 2014). After abscission, the midbody remnant (MR) is asymmetrically inherited by one of the two daughters cells and then degraded by autophagic or non-autophagic lysosomal pathways (Chen et al., 2013, Crowell et al., 2014).

Centrosomal protein 55 kDa (CEP55) is a key regulator of cytokinetic abscission. This protein is predominantly an α-helical coiled-coil protein, which is able to oligomerize (Zhao et al., 2006, Carlton and Martin-Serrano, 2007, Morita et al., 2007, Fabbro et al., 2005). CEP55 is essential for the recruitment of the Endosomal sorting complex required for the transport (ESCRT) machinery to the intercellular bridge (Morita et al., 2007, Carlton and Martin-Serrano, 2007, Carlton et al., 2008, Lee et al., 2008). CEP55 directly interacts with two components of the ESCRT machinery, ALG-2 interacting protein X (ALIX) and the ESCRT-I Tumor susceptibility gene 101 (TSG101), via its atypical α-helical coiled-coil dimeric region called EABR (ESCRT-and ALIX-binding region). Once recruited at the midbody through CEP55, the ESCRT complex orchestrates membrane remodeling events leading to the physical separation of the two daughter cells. In addition to its function in cytokinesis, CEP55 has also a central role in the midbody integrity maintenance (Zhao et al., 2006) and in the MR degradation (Ettinger et al., 2011, Kuo et al., 2011).

CEP55 is recruited to the midbody in late mitosis (Bastos and Barr, 2010). This recruitment depends on the Mitotic kinesin-like protein 1 (MKLP-1), a component of the Centralspindlin complex, which has been shown to co-immuno-precipitate with CEP55 under endogenous conditions as well as in reticulocyte lysates after in vitro transcription/translation of both proteins (Zhao et al., 2006, Bastos and Barr, 2010). It has been reported that CEP55 become phosphorylated during mitosis on its S425 and S428 residues by Cyclin-dependent kinase 1 (Cdk1) and Extracellular signal-regulated kinase 2 (Erk2), whereas its S436 residue is phosphorylated by Polo-like kinase 1 (PLK1) (Fabbro et al., 2005, Bastos and Barr, 2010). The PLK1 phosphorylation prevents CEP55 early recruitment to the mitotic spindle and its accumulation to the midbody, and its impairment leads to severe abscission defects and inhibition of the ESCRT machinery recruitment to the midbody (Bastos and Barr, 2010, Kamranvar et al., 2016). It has been proposed that PLK1 could indirectly interact with CEP55 via the formation of a quaternary complex, which involves Myotubularin-related proteins 3 and 4 (MTMR3 and MTMR4) (St-Denis et al., 2015). Both these myotubularins were shown to be important in the abscission process regulating CEP55 recruitment to the midbody (St-Denis et al., 2015). Importantly, a severe human embryonic pathology called MARCH (multinucleated neurons, anhydramnios, renal dysplasia, cerebellar hypoplasia, and hydranencephaly) was recently associated with a deletion of the last 40 amino acids in CEP55, leading to a severe defect of CEP55 recruitment to the midbody (Frosk et al., 2017).

The post-translational modification of proteins by mono-ubiquitin or poly-ubiquitin chains of variable length and/or different linkage types is involved in a vast range of cellular processes, including autophagy, DNA repair, protein turnover, and many signal transduction events such as the nuclear factor (NF)-κB signaling pathway (Komander and Rape, 2012). These chains of ubiquitin are specifically recognized by proteins that contain ubiquitin-binding domains (UBDs), allowing a large number of effector proteins to translate the modifications into specific outcomes. In the NF-κB cascade, NEMO (NF-κB Essential Modulator) contains two UBDs, which non-covalently bind to hybrid M1/K63 poly-ubiquitin chains to trigger the NF-κB activation (Emmerich et al., 2013). The NEMO UBD referred to as NOA/UBAN domain (also called NUB or CC2-LZ) forms a long parallel α-helical coiled-coil dimer, which selectively binds to a Methionine 1 (M1) di-ubiquitin chain (Rahighi et al., 2009). The second UBD is a classical CCHC-type zinc finger (ZF), which adopts a ββα fold and interacts with a weak preference to lysine 63 (K63) and M1 over lysine 48 (K48) poly-ubiquitin chains (Cordier et al., 2009, Ngadjeua et al., 2013).

During cytokinesis, the E2/E3 ubiquitin ligase BRUCE (BIR repeat containing ubiquitin-conjugating enzyme), the deubiquitinases USP8 (Ubiquitin specific peptidase 8), and AMSH (Associated molecule with the SH3-domain of STAM) were reported to be located at the midbody and play key roles in the cytokinetic abscission process (Pohl and Jentsch, 2008) (Mukai et al., 2008). However, despite an accumulation of mono-ubiquitin and different types of poly-ubiquitin chains at the midbody and intercellular bridge, little is known about the role of ubiquitin signaling during the abscission process (Pohl and Jentsch, 2008, Mukai et al., 2008, Matsumoto et al., 2010).

Here, we identified CEP55 as a member of the NEMO-related protein family. Indeed, we show that CEP55, like NEMO, contains two UBDs, namely, NOA and ZF, which differently regulate CEP55 functions. Both domains bind to poly-ubiquitin chains in vitro, but NOA is specifically involved in the abscission process, whereas we showed that ZF is critical for CEP55 recruitment to the midbody and is therefore crucial in CEP55 functions at the midbody. Altogether our data reveal a critical role for non-degradative ubiquitination linked to CEP55 function in the last steps of cell division.

Results

CEP55 Contains Two Ubiquitin-Binding Domains, which Are Structurally Similar to NOA and ZF UBDs of NEMO

The NEMO function depends on two UBDs in its C-terminal part, referred to as the NOA/UBAN domain (Laplantine et al., 2009, Wagner et al., 2008) and the ZF domain (Cordier et al., 2009, Ngadjeua et al., 2013). The NOA domain forms a long parallel α-helical coiled-coil dimer (Rahighi et al., 2009), whereas the ZF domain exhibits an UBZ (Ubiquitin-binding zinc finger) architecture based on a classical ββα fold (Cordier et al., 2009). Multiple sequence alignments including analysis of protein sequence databases, combined with the Paircoil2 algorithm, predict that the CEP55 protein contains in its C-terminal part two UBDs. We refer to these as NOA (residues 304–396, 38% similarity) and ZF (residues 435–464, 47% similarity), respectively, owing to their significant homologies with the same domains in NEMO, Optineurin and ABIN2 (Figures 1A, S1A, and S1B). To test this hypothesis, we first expressed in E.coli and purified to homogeneity a His-tag version of the human CEP55 NOA fragment (residues 304–396) (Figure 1B inset). Circular dichroism spectrum analysis in the far-UV showed that CEP55 His-NOA exhibits a high α-helical content of 80%, compatible with an α-helical coiled-coil structure (Figure 1B). To non-ambiguously determine the oligomeric state of CEP55 His-NOA in solution, we performed size exclusion chromatography experiments coupled in line with multi-angle light scattering (SEC-MALS), and showed that CEP55 His-NOA behaves mainly as monomers and dimers, with a very small amount of higher-order oligomers (3% of hexamers, Figure 1C). Based on these biochemical experiments, we generated a structural model of the NOACEP55 using NOANEMO as the structural template (Figure S1C) and observed that the C359 residue of one monomer is in close proximity with the C359 residue of the second monomer when NOACEP55 forms an α-helical coiled-coil dimer (Figure S1E). We utilized this ability of the CEP55 His-NOA construct to form an intermolecular disulfide bond at C359 to further confirm its dimeric state. As shown in Figure S1F, CEP55 His-NOA migrated at the dimeric molecular mass in the absence of the reducing agent dithioerythritol (DTE), whereas it migrated at the monomeric molecular mass upon treatment with DTE, further confirming that CEP55 His-NOA likely forms a long parallel α-helical coiled-coil dimer like that of NEMO. However, this isolated NOACEP55 domain appears to exhibit a much weaker dimerization affinity than that of NEMO NOA (6-fold less), which has been previously shown to have a dimerization constant around 30 μM (Hubeau et al., 2011).

Figure 1

CEP55 Contains NOA and ZF Domains

(A) Schematic representation of structural and functional domains of CEP55 and NEMO showing a similar C-terminal architecture encompassing two ubiquitin-binding NOA and ZF domains. Sequence numbering is given for the human proteins (NP_001093327.1 and NP_001120654.1). KBD, Kinase binding domain; CC1-2, Coiled-coil 1 and 2; HLX2, Helical domain 2; EABR, ESCRT and ALIX binding region; NOA, NEMO, Optineurin and Abin domain (also called UBAN, CoZi, CC2-LZ and NUB); ZF, Zinc Finger; UBD, Ubiquitin-binding domain. Both EABR and NOA domains form parallel coiled-coil dimers.

(B) Normalized far-UV circular dichroism spectrum of purified CEP55 His-NOA (0.03 mg/mL at 1°C), showing a high α-helix content compatible with a coiled-coil structure. In inset is depicted the SDS-PAGE analysis of CEP55 His-NOA after staining with Coomassie blue.

(C) Analysis of CEP55 His-NOA by size exclusion chromatography coupled with multi-angle light scattering (SEC-MALS). M, Monomer; D, Dimer; H, Hexamer.

(D) Fluorescence emission spectra of CEP55 ZF V438W peptide (2 μM, hereafter denoted as ZF (W)) in the presence (black) or in absence of zinc ion (gray) after excitation of both tyrosine (Y447 and Y461) and tryptophan (W438) residues at 280 nm. Note that the emission shoulder peak observed in the 300–315 nm region, which is only observed in absence of zinc ion, reflects tyrosine fluorescence emission (Y447, Y461). This shoulder peak disappears and vanishes to the profit of tryptophan emission after ion zinc addition, owing to an increase of Tyr to Trp internal FRET. Inset: model of CEP55 ZF (W) pointing out that the zinc-induced folding brings close together the two Y447, Y461 and W438 residues, leading to internal FRET.

(E) Fluorescence titration of CEP55 ZF (W) peptide (0.8 μM) using increasing zinc concentrations (0.1–1.4 μM) at excitation and emission wavelengths of 280 and 342 nm, respectively. Error bars represent the standard deviation of the fluorescence measurements of each titration point for 5 min.

To investigate whether the human CEP55 fragment containing the residues 435–464 adopts a zinc-finger fold, we generated a synthetic peptide mimicking this region and containing the V438W substitution (CEP55 ZF (W)). This substitution was previously described to act as a sensitive probe to monitor zinc-induced folding of the NEMO ZF without interfering with its overall structure (Cordier et al., 2009). We recorded the fluorescence emission spectrum of the CEP55 ZF (W) peptide at 300–450 nm using an excitation wavelength of 280 nm, allowing us to specifically irradiate the two tyrosines (Y447 and Y461) and tryptophan (W438) residues (Figure 1D). Upon zinc addition, CEP55 ZF (W) exhibits an increase of the fluorescence signal with a 7-nm blue shift of the tryptophan emission maximum (353–342 nm). Concomitantly, this leads to a decrease of the minor shoulder at 300–310 nm, which usually corresponds to fluorescence emission spectra of tyrosine residues. This clearly indicated that an internal FRET from tyrosine to tryptophan occurs only after the addition of zinc. In the presence and absence of zinc, the internal FRET efficiencies from Y to W were then determined and a 1.9 ± 0.06-fold increase was observed upon zinc addition (Table S1). Because internal FRET efficiency is highly sensitive to the distance between the acceptor and the donor, these data clearly indicated that zinc induces a spatial confinement of the two Y447 and Y461 to W438, reflecting a proper folding of the domain as a zinc finger. We took advantage of this zinc-dependent fluorescence signal at 342 nm to titrate CEP55 ZF (W) by zinc ion (Figure 1E). A strong binding affinity less than 20 nM was observed with a binding stoichiometry of 1:1. This binding stoichiometry was further confirmed using a colorimetric assay as described in Transparent Methods. Altogether, these data combined with multiple sequence alignment showed that the CEP55 fragment containing the residues 435–464 forms a CCHC-type ZF structure, which is likely built on a ββα fold. In a similar way to NOACEP55, a structural model of ZFCEP55 was generated using NEMO ZF as template, emphasizing the four residues (C440, C443, H458, and C462) that tetra-coordinate the zinc atom (Figure S1D) and that are strictly conserved among all CEP55 orthologues (Figure S1A).

NOACEP55 Preferentially Binds to Linear Di-ubiquitin Chain, whereas ZFCEP55 Binds to Non-degradative Linear and K63 Chains

To assess the ability of CEP55 His-NOA to bind to poly-ubiquitin chains, we next performed qualitative Ni-NTA pull-down experiments in the presence of increasing concentrations of purified K63- and M1-linked tetra-ubiquitin (Ub4) chains (Figures 2A and 2B). As negative control, we used a His-tagged recombinant protein (His-DARPin labeled Ctrl), which does not bind to any mono or poly-ubiquitin chains even at high concentration (up to 10 mM ubiquitin site concentration). As shown in Figures 2A and 2B, a specific binding of CEP55 His-NOA to K63- and M1-Ub4 was observed in a dose-dependent manner. Interestingly, similar pull-down experiments using different ubiquitin linkage types revealed that CEP55 NOA preferentially recognizes non-degradative M1 and K63-linked Ub4 chains over degradative K48- and K11-linked Ub4 chains (Figure 2C), indicating that CEP55 His-NOA prefers to interact with open and even extended conformations of Ub4. These results were surprising since the isolated NOA/UBAN domain of NEMO as well as its NEMO-related proteins OPTN and ABIN2 were reported to exhibit a strong linkage specificity for short M1 over K63 di-ubiquitin chains (Rahighi et al., 2009, Nakazawa et al., 2016, Lin et al., 2017). A possible explanation could be that pull-down experiments are only qualitative and performed under non-equilibrium conditions. For this reason, we next characterized the interaction between CEP55 His-NOA and K63- or M1-Ub4 chains by determining association (kon) and dissociation (koff) rate constants using biolayer interferometry (BLI) technology (Figures 2D and 2E). In this experiment, CEP55 His-NOA was first immobilized on anti-penta-His biosensors and incubated with different concentrations of K63- or M1-Ub4. A KD value of 27 μM for M1-linked Ub4 chains was observed (Figure 2E), which is 30-fold lower than that of K63-linked Ub4 (Figure 2D), indicating that CEP55 NOA preferentially binds to M1 poly-ubiquitin chains in vitro.

Figure 2

CEP55 NOA and ZF Form Specific Ubiquitin-Binding Domains, which Preferentially Bind to Non-degradative K63 and M1 Polyubiquitin Chains

(A and B) His pull-down experiments of CEP55 His-NOA (8 μM) with increasing concentrations of purified (A) K63 tetra-ubiquitin and (B) linear tetra-ubiquitin (1, 2, 4, 8, and 16 μM). Ctrl, negative control (His-DARPin).

(C) His pull-down experiment of CEP55 His-NOA (8 μM) with different linkages of tetra-ubiquitin chains (16 μM), including M1 (linear), K11, K48, or K63 tetra-ubiquitin as indicated. Ctrl, negative control (His-DARPin).

(D and E) Quantitative ubiquitin binding of CEP55 His-NOA by BioLayer Interferometry (BLI) using various concentrations of K63 (D) or linear (E) tetra-ubiquitin chains (2.5, 5, 10, and 20 μM). Anti-penta-His biosensors were used in each binding experiment to immobilize 8 μM of His-NOA (see Transparent Methods).

(F) Quantitative ubiquitin binding of Strep-tag II CEP55 ZF by BLI with a mixture of K63 Ub2-7 polyubiquitin chains. Biotinylated K63 Ub2-7 (0.2 mg/mL) was immobilized on a streptavidin biosensor before adding various concentrations of Strep-tag II CEP55 ZF (2.5, 5, 10, or 20 μM).

(G) Quantitative ubiquitin binding of Strep-tag II CEP55 ZF by BLI with an M1 (linear) di ubiquitin chain. M1 GST-Ub2 (20 μM) was immobilized on an anti-GST biosensor before the addition of Strep-tag II CEP55 ZF at different concentrations (1, 6.25, 12.5, or 25 μM).

(H) Binding comparison of the Strep-tag II CEP55 ZF (10 μM) by BLI between a mixture of K63-Ub2-7 or K48 Ub2-7 polyubiquitin chains (0.2 mg/mL). For the experiments shown in (F)–(H), the Strep-tag II CEP55 ZF was each time freshly prefolded with 1 mM ZnCl2, which was maintained at the same concentration in the binding buffer.

In a similar way, we next characterized the ability of CEP55 ZF to bind to K63 and M1 poly-ubiquitin chains using BLI assays. A mixture of biotinylated K63 poly-ubiquitin chains (Ub2-7) and GST-Ub2 M1 were immobilized on streptavidin (Figure 2F) and anti-GST antibody (Figure 2G) biosensors, respectively, before being incubated with increasing concentrations of a StrepII-tag version of CEP55 ZF peptides (STII-CEP55 ZF). CEP55 ZF similarly binds K63 Ub2-7 poly-ubiquitin chains and GST-Ub2 M1, since similar KD values of 2.1 and 1.6 μM were obtained, respectively. This ubiquitin-binding activity was specific because no binding was observed with a mixture of biotinylated K48 Ub2-7 chains (Figure 2H), indicating that CEP55 ZF preferentially recognizes K63 and M1 chains over K48 poly-ubiquitin chains. We also confirmed the specific binding of CEP55 ZF to mono- and K63 di-ubiquitin in solution by fluorescence polarization (FP) experiments using N-terminally fluorescein labeled CEP55 ZF and under more stringent buffer conditions (Figures S1G and S1H). Of note, similar binding affinities were observed with mono-ubiquitin and K63 di-ubiquitin in solution, but they were much lower than those obtained with BLI experiments. This is presumably because poly-ubiquitin chains were immobilized in BLI experiments, whereas they were in solution in FP experiments (see Discussion). Taken together, these data clearly demonstrated that CEP55 contains two UBDs in its C-terminal part: a UBZ-type (UBZCEP55) located at the C-terminal extremity that binds to non-degradative linear and K63 chains and an NOA domain (NOACEP55) that preferentially interacts with linear Ub chains.

Structure-Based Identification of UBZCEP55 Mutants Defective in Ubiquitin Binding

Generation of ubiquitin binding defective single-point mutants of UBZCEP55, without altering the overall ZF structure, could be a challenging task since recent crystal complexes of UBZ domains with ubiquitin revealed different modes of ubiquitin recognition (Figure S2). Nevertheless, we could classify at least three groups of UBZ domains, including FAAP20/Polη (Toma et al., 2015; PDB:3WWQ [Bomar et al., 2007]), RAD18/WRNIP1 (Hu et al., 2017; PDB:5VF0 [Suzuki et al., 2016], PDB:3VHT), and NDP52 (Xie et al., 2015; PDB:4XKL), based on their crystal structures in complex with Ub (Figures S2A and S2B). Structure-based sequence alignment analysis of UBZCEP55 from different organisms with different UBZ groups showed that UBZCEP55 displays the highest homology with the RAD18/WRINP1 UBZ group, suggesting a similar ubiquitin-binding mode (Figure S2C). In particular, Ub-interacting residues of WRINP1-UBZ in the β1-strand, β-loop (the loop between both antiparallel β-strands), and the α-helix of the UBZ ββα fold are conserved or replaced by functionally equivalent residues in CEP55-UBZ (Figure S2C). These residues are also strictly conserved in CEP55 from several organisms. We therefore generated a structural model of CEP55 ZF in complex with ubiquitin, using the crystal structure of the UBZ-GFPWRNIP1:ubiquitin complex as template (Figure 3). According to this model, the CEP55 residue E460, which is strictly conserved among all CEP55 orthologues (Figure S1A), forms a strong binding determinant of the complex by making a dual salt bridge with R42 and R72 of ubiquitin (Figure 3A). To validate our hypothesis, we generated an N-terminally fluorescein-labeled CEP55 ZF mutant (F-ZF E460A) as well as a StrepII-tag version of CEP55 ZF mutant (STII-CEP55 ZF E460A) containing the E460A single point mutation. F-ZF E460A displayed an almost 20-fold weaker affinity compared with WT for mono-ubiquitin when determined in solution by FP (Figure 3B). We confirmed this drastic loss of affinity of the E460A mutant by BLI using immobilized GST-Ub2 M1, since binding affinity of the E460A UBZ mutant was 26-fold reduced compared with the WT (Figure 3C).

Figure 3

Structure-Based Identification of ZF Single-Point Mutant Defective in Ubiquitin Binding

(A) Structural model of CEP55 ZF (green) in complex with mono-ubiquitin (orange) pointing out E460 as a critical residue that makes a specific salt bridge with R42 and R72 residues of ubiquitin.

(B) Mono-ubiquitin titration in solution monitored by fluorescence polarization of fluorescein-labeled WT and E460A mutant CEP55 ZF (10 μM) in the presence of 1 mM ZnCl2.

(C) Quantitative ubiquitin binding determined by BLI experiment of WT and E460A mutant CEP55 ZF with M1 (linear) di-ubiquitin chain. M1 GST-Ub2 was first immobilized on an anti-GST biosensor before adding various concentrations of WT and E460A mutant STII-CEP55 ZF (6.25, 12.5, and 25 μM). For the experiments shown in (B) and (C), all ZF peptides were freshly prefolded in the presence of 1 mM ZnCl2, which was maintained at the same concentration in the binding buffer.

We then verified whether the E460A mutation impedes the ability of UBZCEP55 to interact with ubiquitin and not the overall structure of UBZ domain. For this purpose, we investigated the zinc-induced folding of the E460A mutant (ZFCEP55 (W) E460A) by measuring its internal FRET efficiencies from tyrosine to tryptophan (Table S1). Similar fluorescence emission spectra of the WT and E460A mutant with or without the presence of Zn2+ indicated a similar folding (Figure S2D). Consistently, similar gains of internal FRET efficiencies from tyrosine to tryptophan were observed for the WT and E460A after the zinc addition (Table S1). Furthermore, Zn2+ titration experiments by fluorescence (Figure S2E) combined with a PAR colorimetric assay (data not shown) showed similar binding affinities and binding stoichiometries of the WT and E460A UBZ mutant. Thus, these data in agreement with our structural model indicated that the E460A mutation in UBZCEP55 is a crucial binding determinant for ubiquitin recognition but does not alter the proper folding of this ZF structure.

CEP55 ZF and Its Ubiquitin-Binding Ability Are Necessary for CEP55 Recruitment to the Midbody

To study the functional role of UBZCEP55 in the context of the full-length protein, we next carried out functional depletion/rescue experiments in HeLa cells. For this purpose, endogenous CEP55 was knocked down by small interfering RNA (siRNA) and transiently re-expressed with siRNA-resistant plasmids expressing either CEP55 WT (HA-CEP55 WT), a debilitating double mutant (HA-C440A/C443A), which is unable to tetracoordinate the zinc atom, or the E460A single-point mutant defective in the ubiquitin binding (HA-CEP55 E460A). As shown in Figure 4A, knockdown of the endogenous CEP55 by siRNA was very efficient, and ectopic expressions of the WT and its single and double mutants gave similar expression levels of CEP55 as judged by western blot. As previously reported in the literature, efficient knockdown of CEP55 leads to an increase of cells connected by an intercellular bridge (Figures 4A and S3A) as well as multinucleated cells (Figures 4B and S3A). These phenotypes were partially restored by the expression of the HA-CEP55 WT protein (Figures 4A, 4B, and S3A). We do believe that the incomplete rescue was due to technical issues of simultaneously reaching very high transfections efficiencies with both DNA and siRNA and to intrinsic properties of the CEP55 protein, which lead to mainly insoluble protein under transient expression conditions, as previously described (Kuo et al., 2011) (Figure S3C). In addition, it is likely that the insoluble form of CEP55 could trap the soluble and functional form of CEP55 and so contributes to an incomplete rescue. Nevertheless, this rescue was statistically significant and reproducible (n = 3 with more than 300 nuclei/cells analyzed by experiment) (Figures 4A and 4B). Strikingly, neither the CEP55 C440A/C443A double mutant nor the CEP55 E460A single-point mutant were able to functionally rescue normal cytokinesis as does the WT, as judged by the persistent increase of cells connected by intracellular bridges (Figure 4A) and the increase in the proportion of multinucleated cells (Figure 4B) induced by CEP55 knockdown. These data clearly indicated that CEP55 ZF and its ubiquitin-binding property are crucial for CEP55 function during abscission. Moreover, the C440A/C443A double mutant and the E460A single-point mutant, unlike the WT, were unable to functionally restore the accumulation of MRs induced by CEP55 depletion (Figure S3B), highlighting a critical role of UBZCEP55 in remnant midbody clearance.

Figure 4

CEP55 ZF Is Essential for CEP55 Recruitment to the Midbody and Ensuring Proper Cytokinetic Abscission

HeLa cells were concomitantly transfected with a non-targeting (NT) or CEP55 siRNA and with an empty vector or a CEP55 siRNA-resistant vector encoding HA-CEP55 WT, HA-CEP55 C440A/C443A, or HA-CEP55 E460A.

(A and B) Analyses of the ZF mutant's ability to rescue cytokinesis defect induced by CEP55 depletion. The percentage of cells connected by midbodies (A) and the percentage of multinucleated cells (B) were quantified. Images are represented in Figure S3A. Data represent the mean ± SD of three independent experiments, and the total number of nuclei (n) counted in the three experiments is indicated. The level of expression of CEP55 in the soluble fraction for the different conditions was controlled by WB.

(C) Recruitment to the midbody of HA CEP55 WT, HA CEP55 C440A/C443A, and HA CEP55 E460A. Cells were stained with DAPI (blue) and immunolabeled with HA (yellow), MKLP-1 (red), and α-tubulin (cyan) antibodies. Insets show an expanded view of the midbody.

(D) Relative quantification of CEP55 intensity at the midbody reporter in the different conditions. Cells were stained with DAPI, CEP55, MKLP-1, and β-tubulin antibodies. CEP55 intensity was normalized by MKLP-1 intensity at each detected midbody (MKLP-1 spot).

Analyses were conducted on three independent experiments, n represents the total number of midbodies. Scale bar, 10 μM. ***p < 0.001; *p < 0.05; ns, not significant (p ≥ 0.05).

Next, we assessed whether UBZCEP55 could be involved in the recruitment of CEP55 to the midbody. Because transient expression of the ectopic CEP55 could interfere with the endogenous CEP55 via the formation of hetero-oligomers, similar functional knockdown and rescue experiments of endogenous CEP55 were performed. Interestingly, although HA-CEP55 WT was recruited to the midbody, neither the C440/C443A double mutant nor the E460A single-point mutant were found to co-localize with the midbody (Figure 4C). Image quantification using MKLP-1 as midbody reporter confirmed the defect of both CEP55 mutants in their recruitment to the midbody compared with the WT (Figure 4D). In line with this, the ALIX endogenous protein, which is recruited to the midbody via CEP55 binding, also exhibited a clear defect in midbody localization with both CEP55 ZF mutants, but not the WT, using again MKLP-1 as midbody reporter (Figure S3D).

Taken together, these results reveal that the UBZCEP55 UBD and its ubiquitin-binding properties are crucial for CEP55 recruitment to the midbody, allowing CEP55 to ensure proper abscission and MR clearance.

Structure-Based Identification of CEP55 NOA Mutants Unable to Bind Ubiquitin

The precise mode of linear di-ubiquitin recognition by the NOA/UBAN domain originally found in the NEMO protein (Rahighi et al., 2009) was more recently confirmed by two other NEMO-like proteins called Optineurin (Nakazawa et al., 2016) and ABIN2 (Lin et al., 2017). Indeed, structural comparison of NEMO NOA in complex with an M1-linked di-Ub with the recent crystal structures of Optineurin with a linear di-Ub chain (Nakazawa et al., 2016) and ABIN2 with a linear tri-Ub chain (Lin et al., 2017) revealed a highly conserved binding mode of NOA with two M1-linked units of ubiquitin (Figure S4). Contrary to the UBZ family, the NOA UBD forms a long parallel α-helical coiled-coil dimer, in which one monomer mainly interacts with the distal Ub (second Ub), whereas the other monomer interacts with the proximal ubiquitin (first Ub). This explains why the NOA dimeric state is required for ubiquitin binding. Structure-based sequence alignment of NOA UBD family with CEP55 from different organisms reveal the presence of conserved or functionally equivalent ubiquitin-interacting residues with both proximal and distal ubiquitin. This sequence conservation suggests that CEP55 binds to a di-Ub chain in a similar manner to NOA family and the degree of conservation of each amino acid position was higher between CEP55 and ABIN2 NOA, especially when compared with residues interacting with proximal ubiquitin (Figure S4). We then generated a structural model of CEP55 NOA in complex with a linear di-Ub chain using the linear diubiquitin:NOA complex of NEMO as structural template (PDB: 2ZVO) (Figure 5A). According to the model, L351, Q354, and Q355 interact with the I44 centered hydrophobic patch of the distal ubiquitin (Figure 5A, left panel). L351 and Q355 residues are perfectly conserved in all the NOA family and CEP55 orthologs, whereas Q354 is conserved only in ABIN2. Moreover, D362 and F363 interact with the C-terminal tail of distal ubiquitin and are strictly conserved in all the NOA family (Figures 5A right panel and S4).

Figure 5

Structure-Based Identification of CEP55 NOA Mutants Defective in Ubiquitin Binding

(A) Modelization of CEP55 NOA (green) in interaction with linear di-ubiquitin pointing out critical residues in the recognition of the distal ubiquitin (orange) and proximal ubiquitin (cyan). Zoom on the interface between distal ubiquitin and CEP55 NOA. Left: residues involved in the recognition of the hydrophobic patch of the distal ubiquitin. Right: residues involved in the recognition of the C-terminal tail of the distal ubiquitin.

(B) Comparison of the association constants (KA) of the different His-NOA mutants for linear tetra-ubiquitin measured by BLI. Curves are represented in Figures S5A–S5C.

To validate the importance of these ubiquitin-interacting residues, we purified to homogeneity two His-NOA mutants containing either the L351A/Q354A/Q355A triple mutation or D362R/F363P double mutation and assessed their affinities with M1-linked tetra-ubiquitin by BLI (Figures 5B and S5A–S5C). A deletion mutant of the residues 343–364 corresponding to the minimal NOA site was also generated taking into account the nature of the heptad repeats in the coiled-coil dimer. For all of the three mutants, we observed drastic losses of binding affinity for M1-linked Ub4 compared with the WT (more than 500-fold), indicating that the mutated residues within the double and triple mutants correspond to critical ubiquitin-binding determinants in agreement with our structural model. To determine whether these mutations also caused global disruption of the dimeric coiled-coil structure, we next analyzed the mutants by SDS-PAGE in the absence of the reducing agent DTE (Figure S5D) and by SEC-MALS (Figure S5E). The L351A/Q354A/Q355A triple mutant and the deletion mutant form considerably less stable dimers than the WT, as no dimers with both mutants were detected by SDS-PAGE in the absence of DTE and SEC-MALS. For the D362R/F363P double mutant, a significant amount of dimers could be detected in SDS-PAGE without DTE, indicating that this mutant forms more stable dimers than the triple and deletion mutants. Nevertheless, the D362R/F363P mutant dimer was less stable than the WT, as judged by the dimer/monomer ratio of the mutant in SDS-PAGE without DTE, which is significantly lower than the WT. Furthermore, no dimer with this mutant could be observed with SEC-MALS. Rather, we could detect a second form of monomer more compact, which likely results from the replacement of Phe363 by Pro363 in this mutant. Therefore, all structure-guided mutants generated from the structural model displayed a total loss of ubiquitin binding activity, but they also displayed a defect in dimerization properties to varying extents.

CEP55 NOA Is Absolutely Required for Normal Cytokinetic Abscission but Not for CEP55 and ALIX Recruitment to the Midbody

To investigate the functional importance of CEP55 NOA UBD, we carried out functional depletion/rescue experiments in HeLa cells. To this end, we generated the D362R/F363P double mutant in the context of the full-length CEP55 protein. As shown in Figure 6A, the protein expression level of the D362R/F363P double mutant was similar to that of the WT. Nevertheless, it is worth noting that a significant portion of the WT and mutant proteins were found in the insoluble fraction of cell lysates, as we found for the CEP55 UBZ mutants (Figures S3C and S6C). Importantly, ectopic expression of the D362R/F363P double mutant was not able to rescue the cytokinetic defect induced by CEP55 knockdown, contrary to the WT (Figures 6A and S6A). On the other hand, the amount of multinucleated cells was almost similar in cells expressing either the double mutant or the WT (Figures 6B and S6A). Although an increase of cells connected by an intercellular bridge may reflect a delay in the abscission process, the appearance of multinucleated cells often results from more drastic abscission defects and/or an earlier event involved in the stabilization of the intercellular bridge and midbody integrity. In the case of CEP55, previous studies showed CEP55 plays crucial roles in midbody integrity (Zhao et al., 2006, Carlton et al., 2008) as well as in the abscission process (Morita et al., 2007, Carlton and Martin-Serrano, 2007, Lee et al., 2008). Consistent with previous studies, we conclude that the CEP55 NOA UBD plays an important role in CEP55 during abscission, albeit to a lesser extent than UBZCEP55, and does not appear to be involved in the stabilization of the intercellular bridge or seems less important than CEP55 ZF in the abscission process.

Figure 6

CEP55 NOA Is Not Involved in CEP55 Recruitment to the Midbody but Is Absolutely Required to Ensure Proper Abscission

(A and B) Effect of the double NOA mutant (D362R/F363P) on the ability to rescue cytokinesis defect induced by CEP55 depletion. HeLa cells were concomitantly transfected with a non-targeting (NT) or CEP55 siRNA and with an empty vector or a CEP55 siRNA-resistant vector encoding either HA-CEP55 WT or HA-CEP55 D362R/F363P. The percentage of cells connected by midbodies (A) and the percentage of multinucleated cells (B) were quantified. Representative images are shown in Figure S6A. Data represent the mean ± SD of three independent experiments, and the total number of nuclei (n) counted in the three experiments is indicated. The level of CEP55 expression in the soluble fraction were controlled by western blot as indicated.

(C) Recruitment to the midbody of HA-CEP55 WT and HA-CEP55 D362R/F363P. Same immunofluorescence experiments as (A). Cells were then stained with DAPI (blue) and immunolabeled with HA (yellow), MKLP-1 (red), and α-tubulin (cyan) antibodies. Insets show an expanded view of the midbody. Scale bar, 10 μM.

(D) Relative quantification of CEP55 intensity at the midbody as indicated. Immunofluorescence experiments were the same as (A) except for CEP55 labeling. CEP55 intensity was normalized by MKLP-1 intensity at the midbody upon MKLP-1 dots. The total numbers of midbodies (n) that result from three independent experiments are indicated.

***p < 0.001; *p < 0.05; ns, not significant (p ≥ 0.05).

Moreover, although the D362R/F363P double mutant exhibits a defect in the abscission process, it is normally recruited to the midbody, indicating that the detrimental effect of the mutant in cytokinesis occurs after its recruitment to the midbody (Figures 6C and 6D). Consistent with these data, we also observed that the D362R/F363P double mutant was able to recruit ALIX to the midbody as efficiently as the WT (Figure S6D), suggesting that the functionality of the EABR dimeric domain was not affected by the ubiquitin-binding deficient mutation in the NOA domain in the context of the full-length protein.

Finally, we quantified the percentage of the remnants by cell using knocked down HeLa cells of CEP55 expressing either the WT or the double mutant (Figure S6B). No difference was observed between the WT and the double mutant, indicating that CEP55 NOA is not involved in CEP55 MR clearance.

Altogether, our results showed that NOA UBD plays an important role in cytokinetic abscission, albeit to a lesser extent than UBZCep55 UBD, and is not involved in CEP55 nor in ALIX recruitment to the midbody neither in MR clearance.

UBZCEP55 Functions as a Specific Targeting Domain to the Midbody in a Ubiquitin-Binding-Dependent Manner

We next focused our studies on the UBZ-dependent recruitment of CEP55 to the midbody by examining the possibility that this domain alone could serve as a specific midbody localization signal. If UBZCEP55 forms an autonomous UBD domain for this targeting activity to the midbody, thus, any fusion proteins that only contain the UBZCEP55 domain should be specifically recruited to the midbody. For this purpose, we generated a GFP-UBZCEP55 construct and investigated by fluorescence microscopy its cellular localization. As shown in Figures 7A and 7B, GFP-UBZCEP55, but not GFP alone, co-localized with the midbody marker, MKLP-1, indicating that UBZCEP55 exhibits a specific targeting activity to the midbody. This was not due to the differences of protein expression level because GFP and GFP-UBZCEP55were expressed as similar levels in cells (Figure 7C). Because endogenous CEP55 could interfere with the GFP-UBZCEP55 recruitment to the midbody, we then performed experiments in CEP55 knock-downed HeLa cells, concomitantly expressing GFP alone or GFP-UBZCEP55 (Figures S7A and S7B). No significant difference in the targeting activity was observed in cells transfected with an siRNA control and CEP55-specific siRNA, demonstrating that the endogenous CEP55 was not necessary for the recruitment of GFP-UBZCEP55 to the midbody.

Figure 7

Ubiquitin-Binding Activity of CEP55 UBZ Is Required to Function as a Specific Cargo Receptor to the Midbody

(A) HeLa cells were transiently transfected with plasmids encoding for GFP alone or GFP-UBZCEP55 and then analyzed by immunofluorescence and fluorescence microscopy.

(B) Quantification of the percentage of co-localization between GFP or GFP-UBZCEP55 and MKLP-1 (midbody reporter) in transfected cells. Data represent the mean ± SD of six independent experiments. **p < 0.01.

(C) Expression levels were controlled by western blot.

(D) CRISPR-Cas9-mediated CEP55 knockout U2OS cell lines were transiently reconstituted following transfection with different plasmids encoding the WT, a debilitating double mutant unable to form a ZF structure (C440A/C443A), or the chimeric CEP55 proteins containing different ZF groups interacting with ubiquitin as indicated (left panel and Figure S2). Reconstituted cells were then analyzed by immunofluorescence and the levels of expression controlled by western blot before quantifying the percentages of co-localization between CEP55 and MKLP-1 (midbody) for the indicated chimeric CEP55 proteins.

Data represent the mean ± SE of two independent experiments; n represents the total number of midbodies. ***p < 0.001; **p < 0.01; *p < 0.05; ns, not significant (p ≥ 0.05).

We next addressed the question whether this CEP55 UBZ-dependent targeting activity depends on its ability to bind to ubiquitin and whether this ubiquitin binding depends on one type of M1 or K63 linkage of polyubiquitin chains. For these experiments, we generated CRISPR-Cas9-mediated CEP55 knockout U2OS cell lines to increase the sensitivity of our functional complementation assay (Figure S7C). We also generated different CEP55 chimeric proteins in which the CEP55 UBZ was replaced with other classes of UBD based on small ZF modules known to interact with ubiquitin such as UBZWRNP1, UBZRAD18, UBZNDP52, nZF2TAB2, and A20ZF7 (Figure 7D). UBZWRNP1 and UBZRAD18 were chosen because their crystal structures showed a binding mode similar to that of ubiquitin, which is likely close to that of UBZCEP55. In addition, UBZWRNP1 exhibits no ubiquitin linkage specificity, whereas UBZRAD18 exhibits a little preference for K63 over K48 polyubiquitin (4-fold) and an in vitro stronger polyubiquitin binding, owing to the extended LR motif (224–240) located at its C terminus (Thach et al., 2015). We also used the UBZNDP52, which is also built on a ββα fold structurally related to UBZWRNP1 and UBZRAD18. However, it uses a different mode of ubiquitin recognition (Figures S2B and S2C) and was reported to have no specificity for one type of ubiquitin linkage (Toma et al., 2015). By contrast, nZF2TAB2 and A20ZF7 display strong linkage specificity for K63 and M1 polyubiquitin chains, respectively, but they are likely structurally different from UBZCEP55. CEP55 knockout U2OS cells generated by the CRISPR-Cas9 technology were transiently reconstituted following transfection with different plasmids encoding for either the WT or the chimeric CEP55 proteins bearing the different UBZs mentioned before or a debilitating double mutant (C440A/C443A, unable to form a ZF structure). Reconstituted cells were then analyzed by immunofluorescence to quantify the percentages of co-localization between CEP55 and the midbody marker, MKLP-1. The level of expression of the different constructs was controlled by western blot in parallel experiments. As shown in Figure 7D, CEP55 ΔZF-UBZRAD18 totally restored the CEP55 recruitment to the midbody, similar to the WT, whereas the debilitating double mutant did not. All chimeric CEP55 proteins were expressed at a similar level in U2OS cells. The four other chimeric CEP55 proteins, CEP55 ΔZF-UBZNDP52, ΔZF-UBZWRNP1, ΔZF-nZF2TAB2, and ΔZF-A20ZF7, also complement the targeting activity of CEP55 but to in a lesser extent than the WT. Nevertheless, the CEP55 ΔZF-UBZNDP52 significantly displayed a better restoration than those of CEP55 ΔZF-UBZWRNP1, CEP55 ΔZF-nZF2TAB2, and CEP55 ΔZF-A20ZF7, which partially complement the CEP55 recruitment to the midbody at a similar level. Importantly, no significant difference was observed between CEP55 ΔZF-nZF2TAB2 and CEP55 ΔZF-A20ZF7, indicating that no predominant M1 or K63 linkage was required for the UBZ-dependent targeting activity of CEP55.

Next, we addressed the question of how the CEP55 UBZ contributes to cellular poly-Ub binding in the context of the full-length protein. To this end, His-tagged WT or mutant (C440/C443A) human full-length CEP55 purified from 293-F cells were incubated with whole-cell extracts containing free and anchored polyubiquitin chains. His-tagged WT and UBZ double mutant immobilized on Ni-NTA beads were then pulled down, and bound polyubiquitin chains were revealed by western blot using anti-Ub antibodies (Figure S7D). As negative control, we used a mutant of the NOA/UBAN ubiquitin binding domain of NEMO (D311G), which is defective in ubiquitin binding. We observed that the CEP55 UBZ significantly contributes to polyubiquitin binding since the C440A/C443A debilitating double mutant exhibits a slight, albeit significant decrease in polyubiquitin binding activity compared with the CEP55 WT. Note that this polyubiquitin binding activity of the UBZ double mutant was not totally abolished when compared with the NOA/UBAN mutant of NEMO. This suggests that CEP55 NOA, as well as the EABR domain, which interacts with UBD-containing ALIX and TSG101 proteins, could contribute to the polyubiquitin binding activity of CEP55. Moreover, and unsurprisingly, we observed that CEP55 and polyubiquitin chains specifically colocalize with each other and are localized to the midbody at the late stage of cytokinesis (Figure S7E).

Altogether, we conclude that the ubiquitin binding ability of CEP55 UBZ is required to specifically target the CEP55 protein to the midbody. In addition, no significant linkage preference for M1- or K63-linked chains was observed for this targeting activity, consistent with the ubiquitin-binding properties of CEP55 UBZ.

Discussion

NEMO is the prototype of a family of proteins including Optineurin and ABIN2, which contain the same type of UBDs. Although the UBD of these proteins are conserved, they exert different functions such as NF-κB regulation in the case of NEMO, membrane trafficking and autophagy in the case of Optineurin, and NF-κB and ERK MAP kinase regulation in the case of ABIN2. Here we identified the CEP55 protein, a regulator of cytokinesis, as a member of this family protein, which contains both NOA/UBAN and UBZ domains in the C-terminal part of the protein. Although we did not determine the tridimensional structures of these UBDs, we showed strong evidences by structure-guided sequence alignment analysis and biochemical and biophysical experiments that the predicted CEP55 NOA fragment (residues 304–396) adopts an α-helical coiled-coil dimer and in the C-terminal region (residues 435–464) a ZF/UBZ architecture, which are structurally similar to those of NEMO, Optineurin, and ABIN2 proteins. Based on these experiments, we generated structural models of both CEP55 NOA and UBZ domains, which are fully consistent with mutagenesis data and allowed us to identify several mutants defective in ubiquitin binding. These generated mutants in each domain and in the context of the full-length protein also permitted the demonstration of the functional importance of these NOA and UBZ domains for the CEP55 function in cell division and specifically in cytokinetic abscission.

Despite an extensive search of crystallization conditions and generation of different constructs, no high-quality diffraction crystals of CEP55 NOA or CEP55 UBZ with or without ubiquitin could be obtained. For NOA, this was likely due to its high tendency to form higher-order oligomers (mainly hexamers), which probably result from the assembly of three dimeric units. In the case of CEP55 UBZ, this was due to its poor solubility at high concentration as judged by dynamic light scattering experiments (data not shown). Moreover, this tendency of recombinant or synthetic CEP55 UBZ to aggregate and collapse when used at a concentration above 50–70 μM prevented us to determine its structure by NMR, contrary to NEMO UBZ for which we successfully solved the NMR structure (Cordier et al., 2009).

Although CEP55 NOA is structurally similar to NEMO and its related proteins, Optineurin and ABIN2, some characteristic features of CEP55 were observed. First, the nature of the typical seven-residue periodicity (abcdefg)n, known as the heptad repeat, significantly differs from that of NEMO in the N-terminal part of the NOA domain, also called the CC2 region in NEMO. Indeed, four hydrophobic residues at a and d heptad positions in NEMO (a: L260, L274 M295; d: A270), which are crucial for the dimer stability, are replaced by polar and charged residues in CEP55 (a: N311, S325, Q346; d: E321). The presence of these polar and charged residues could destabilize the hydrophobic packing of the α-helical coiled-coil dimer in the N-terminal region of CEP55 NOA, providing a possible explanation why the CEP55 NOA forms a dimer less stable than that of NEMO, with a dimerization constant that is about 6-fold lower than that of NEMO (Grubisha et al., 2010). In line with this, caution also should be taken on the apparent linkage selectivity for linear chains observed with the isolated NOA of CEP55. Indeed, previous crystal structures of NOA complexes with K63- and M1-linked di-ubiquitin chain revealed the interactions with the two ubiquitin moieties of M1-linked di-ubiquitin chain (distal and proximal), whereas only the distal ubiquitin moiety interacts with the K63-linked di-ubiquitin chain. These differences in ubiquitin-binding mode can lead to a more stabilizing effect on the NOA dimerization with the linear chain compared with K63 chains. Thus, the differences of affinity between linear and K63 chains for NOA CEP55 could be artificially enhanced by the dimerization effect induced by linear chain binding, possibly leading to an overestimation of linkage selectivity of CEP55 NOA for linear chains over K63 chains.

Even though size exclusion chromatography showed that the D362R/F363P eluates into two well-separated elution peaks, multi-angle light scattering analysis pointed out that the two peaks correspond to monomeric species with the same molecular mass, indicating that this NOA double mutant adopts two different monomeric forms in solution (Figure S5E, low panel). One monomeric species appears to be more elongated with a higher frictional coefficient compared with the second monomeric species, which displays the same elution profile as that observed with the WT and the L351/Q354A/Q355A triple mutant. Because the proline can adopt two cis and trans conformations in protein, the presence of this less compact second monomeric species is compatible with the F363P substitution in the NOA double mutant.

Moreover, the SDS-PAGE analysis under non-reducing conditions points out that the dimerization properties of the D363R/F363P double mutant are partially altered as compared with the WT, whereas those of the L351/Q354A/Q355A triple mutant are totally abolished (Figure S5D). Thus, the triple mutation induces a more destabilizing effect on CEP55 NOA dimerization than the double mutant. In the light of the structural model shown in Figure 5, it is tempting to speculate that each Q354 of one monomer makes a specific hydrogen bound with the sulfur atom of each M356 belonging to the second monomer, giving a possible explanation why a drastic destabilization of the CEP55 NOA dimeric state was observed with the L351A/Q354A/Q355A triple mutant. Notably, it is worth pointing out that a tight coupling between dimerization and ubiquitin binding was previously observed upon studies with the NOA/CC2-LZ domain of NEMO (Grubisha et al., 2010). Therefore, finding mutations in CEP55 NOA, which only alter ubiquitin binding without interfering with the dimerization process, could be a challenging task, which could be insurmountable at the current stage without prior knowledge of the crystal structures of the apoform and complexed form of CEP55 NOA with a ubiquitin chain.

When analyzing ubiquitin-binding properties of CEP55 UBZ, different binding affinities were observed when we determined dissociation constants in solution by fluorescence polarization and BLI experiments. Even if we did not strictly compare the same type of K63-linked chain in FP (K63-UB2) and BLI experiments (K63-Ub2-7), these differences in binding affinity cannot probably be due to the differences of the chain length. First, the mixture of K63-linked UB2-7 chains used in BLI experiments mainly contains di-ubiquitin and tri-ubiquitin chains and only a minor amount of Ub4-7 chains. Second, the experimental conditions used in solution and BLI experiments were different. Indeed, the DDM detergent was added at 0.1% in FP experiments because it was required to prevent the non-specific binding between ubiquitin and the fluorescein conjugated to CEP55 UBZ. This may considerably reduce the ubiquitin-binding constant by disrupting the hydrophobic interactions on the I44-centered hydrophobic interface of ubiquitin. Moreover, the ubiquitin chain used was in solution in FP experiments, whereas it was immobilized in BLI experiments. This could affect the binding entropy and may contribute to the difference of binding affinity observed between solution and BLI binding studies. Similar conflicting results between solution and pull-down binding studies were also reported with TAB2 NZF to assess the linkage selectivity between K48 and K63 Ub2 and Ub4 chains (Kulathu et al., 2009).

Strikingly, we observed that the CEP55 ΔZF-UBZRAD18 chimeric protein completely restored CEP55 midbody recruitment like the WT, whereas all the other chimeric proteins tested only partially compensate for the phenotype (Figure 7D). A possible explanation could be the binding mode of UBZRAD18 and its ubiquitin-linkage preference for K63 and linear chains over K48 chains, which appeared to be more similar to those of UBZCEP55. However, one cannot rule out the possibility that this best compensation effect was also due to the additional ELRM motif located at the C terminus of UBZRAD18 (Thach et al., 2015), which can confer a better binding affinity to polyubiquitin chains compared with all the other UBZs and nZFs tested.

Importantly, recent studies reported that CEP55 mutations are responsible for two human pathologies named MARCH (Frosk et al., 2017) and Meckel-like syndrome (Bondeson et al., 2017, Rawlins et al., 2019), which exhibit some similar clinical signs. Hypomorphic CEP55 mutations were found to be associated with MARCH, which is a severe and autosomal-recessive syndrome that affects neuronal mitosis. The developmental features of MARCH include hydranencephaly, anhydramnios, renal dysplasia, and cerebellar hypoplasia. Amorphic CEP55 mutations were reported to be associated with the Meckel-like syndrome and with in utero lethality. This syndrome is linked to cilia-related disorders, called ciliopathies, and some clinical signs detected in the affected fetus are similar with MARCH, including aberrant developments of brain and cerebellum, renal cysts, and oligohydramnios. Importantly, the MARCH-related p.S425X mutation (c.1274C > A), which has a similar CEP55 expression level as that of a healthy patient control (Frosk et al., 2017), corresponds to a deletion of the UBZ. Strikingly, this p.S425X mutation exhibits a defect in CEP55 recruitment to the midbody like the C440A/C443A debilitating double mutant investigated in the present study. Therefore, it is tempting to speculate that all mutations of CEP55 that alter the UBZ structure by modifying the zinc-ion-coordinating residues (C440, C443, H458, and C462) or that impair its ubiquitin-binding properties like the E460A single mutant may also lead to the MARCH syndrome. It is also tempting to predict that the extent and severity of the clinical signs found in MARCH and Meckel-like syndromes could be correlated with mutations found in the CEP55 NOA and UBZ domains, taking into account our findings that mutations located in CEP55 UBZ lead to a more severe defect in cytokinetic abscission than those contained in NOA.

Live-cell imaging with super resolution microscopy using live PALM experiments supports the view that NEMO induces dynamic condensates with liquid-like properties (Scholefield et al., 2016), which are reminiscent of the formation of membrane-less compartments induced by liquid phase separation (Banani et al., 2017). We previously showed that these NEMO condensates are mediated by the C-terminal bipartite UBD of NEMO and the multivalent interactions with M1- and K63-linked polyubiquitin chains. Like NEMO, and its Optineurin and ABIN2 related proteins, CEP55 likely forms an elongated coiled-coil structure with two similar folded UBDs in its C-terminal part that are interspersed by a disordered linker of 30–50 amino acids rich in hydrophobic Pro residues. Such a bipartite domain and CEP55 oligomerization could favor multivalent-multivalent interactions with extended conformations of M1- and K63-linked polyubiquitin chains, facilitating formation of membrane-less biomolecular condensate through liquid phase separation. These multivalent interactions could guide the assembly with midbody-associated proteins required for cytokinesis such as MKLP-1 and ALIX proteins. Consistent with this hypothesis, we observed impaired interactions of the C440A/C443A debilitating UBZ mutant of CEP55 with MKLP-1 and ALIX in preliminary experiments. Further characterization of CEP55 and its ubiquitin-mediated protein assembly should provide deeper insights into the mechanism by which CEP55 is tightly regulated in cytokinesis and may contribute to open therapeutic strategies for cancers.

Limitations of the Study

In this study, we have clearly shown that CEP55 contains two NEMO-like UBDs in its C-terminal part, which are both crucial for cytokinetic abscission in a ubiquitin-binding-dependent manner. However, neither the ubiquitinylated protein, which can be CEP55 itself, nor the ubiquitin ligase(s) that contribute to CEP55 recruitment to the midbody were identified in this study but are worth investigating in the future. Moreover, although structural models of CEP55 NOA and UBZ domains were validated by biophysical and biochemical data, determination of their true structures by NMR and/or crystallization, which proved more difficult than expected, is lacking and requires further investigation in the future.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.