Centralspindlin: at the heart of cytokinesis

Abstract

The final step in the cell cycle is the formation of two genetically identical daughter cells by cytokinesis. At the heart of cytokinesis in animal cells is the centralspindlin complex which is comprised of two proteins, a kinesin-like protein, MKLP1, and a Rho GTPase activating protein, CYK-4. Through its targeted localization to a narrow region of antiparallel microtubule overlap immediately following chromosome segregation, centralspindlin initiates central spindle assembly. Centralspindlin has several critical functions during cell division including positioning of the division plane, regulation of Rho family GTPases, as well as midbody assembly and abscission. In this review we will examine the biochemistry of centralspindlin and its multiple functions during cell division. Remarkably, several of its critical functions are somewhat unexpected. Although endowed with motor domains, centralspindlin has an important role in generating stable, antiparallel microtubule bundles. Although it contains a Rho family GAP domain, it has a central role in the activation of RhoA during cytokinesis. Finally, centralspindlin functions as a motor protein complex, as a scaffold protein for key regulators of abscission, and as a conventional RhoGAP. Because of these diverse functions, centralspindlin lies at the heart of the cytokinetic mechanism.

Introduction

The final step in the cell cycle is the formation of two genetically identical daughter cells by cytokinesis. At anaphase onset in animal cells, the bipolar mitotic spindle undergoes a dramatic reorganization to form a structure called the central spindle or spindle midzone (Figure 1) [1]. The central spindle assembles between the segregating chromosomes during anaphase through the actions of microtubule-associated proteins and kinesin family motors [1]. Importantly, the central spindle can dictate the position of the cleavage furrow and it plays an important role in the completion of cytokinesis [2–6].

Figure 1.

Centralspindlin localizes to the central spindle in early anaphase, throughout cytokinesis where it facilitates key cytokinetic events. Schematic depicting the localization of centralspindlin during cytokinesis. Arrows indicate likely site of RhoA activation. Micrographs of an early C. elegans embryo in which the CYK-4 component of centralspindlin (green) and the plasma membrane (red) are labeled.

At the heart of central spindle assembly is the centralspindlin complex which is comprised of two proteins: a kinesin-like protein, MKLP1, and a Rho GTPase activating protein (RhoGAP), CYK-4 [7]. Here we will examine the biochemistry of centralspindlin and its multiple functions during cell division including assembly of the central spindle, regulation of Rho family GTPases, and midbody assembly and abscission.

Identification and structure of the kinesin-like protein MKLP1

Mitotic kinesin-like protein 1 (MKLP1) was first identified as a protein associated with the mitotic spindle during cytokinesis in mammalian cells [8, 9]. The protein was initially referred to as the CHO1 antigen, and its localization during mitosis was striking; it diffusely localized in the cytoplasm surrounding the mitotic spindle at metaphase, and then rapidly and stably accumulated at the center of the anaphase spindle and midbody during cytokinesis [9]. CHO1 appeared to be a microtubule-associated protein that could be involved in bundling the microtubules that comprise the central spindle [9]. Cloning of the CHO1/MKLP1 gene indicated the N-terminus of the molecule contained a domain with high sequence similarity to the motor domain of kinesin-like proteins (Figure 2) [10]. Preliminary functions proposed for MKLP1 during mitosis included cross-linking and sliding of antiparallel microtubules in the anaphase spindle, and force generation to drive spindle elongation during anaphase B [10].

Figure 2.

Centralspindlin and its interaction partners. Domain structure of MKLP1 and CYK-4 and the regions that mediate interactions with their diverse partners.

Orthologs of MKLP1 were subsequently identified in both Drosophila [6] and in C. elegans [11, 12]. Characterization of Pavarotti (Pav), the Drosophila ortholog of MKLP1, first established a specific requirement for MKLP1 during cytokinesis [6]. Zygotically expressed Pav is required for cytokinesis in early Drosophila embryos [6], following the process of cellularization of the syncytial blastoderm. Interestingly, in Pav mutants, anaphase proceeds as in wild-type embryos, however dramatic defects in antiparallel microtubule bundling arise during central spindle assembly. Additionally, Pav mutant cells fail to initiate contractile ring assembly which results in cytokinesis failure [6]. Similar to Drosophila Pav mutants, null mutants of ZEN-4 [12], the C. elegans ortholog of MKLP1, or depletion of ZEN-4 by RNAi [11]. Interestingly, rather than driving spindle elongation as might be expected for a plus end-directed motor, ZEN-4 actually limits the extent of spindle elongation during anaphase [2]. Additionally, although cleavage furrowing does occur in the absence of ZEN-4, a stable midbody structure never develops and the furrow regresses causing a late stage cytokinesis failure [11]. Collectively, these observations provide the first indication that MKLP1 family members function specifically during cytokinesis.

Identification of the novel RhoGAP CYK-4

CYK-4 was first identified in human cells as a new Rho family GTPase activating protein (GAP) [13]. The protein was called MgcRacGAP (Male Germ Cell RacGAP) since it appeared to be predominantly expressed in the male germ cells [13]. The connection to cytokinesis was forged by forward genetics in C. elegans. Specifically, characterization of a temperature-sensitive allele of CYK-4, the C. elegans ortholog of MgcRacGAP, revealed that CYK-4 is essential for cytokinesis [14]. Interestingly, the phenotypes of CYK-4 mutant embryos and embryos deficient in the kinesin-like protein ZEN-4, are identical [14]. Additionally, ZEN-4 and CYK-4 colocalize in the cytoplasm during metaphase, and on the central spindle during anaphase and cytokinesis (Figure 1) [14]. Importantly, ZEN-4 and CYK-4 are interdependent for their proper localization to the central spindle at anaphase [14]. Collectively, these data suggest a functional interaction between CYK-4 and ZEN-4 in vivo and that the two proteins could cooperate to bundle antiparallel microtubules to form the central spindle [14].

Molecular organization of centralspindlin

Biochemical analysis of the C. elegans proteins demonstrated that ZEN-4 and CYK-4 directly interact both in vitro and in vivoforming a high-affinity, stoichiometric 2:2 tetramer, and that these properties are conserved in the human proteins [7]. This stable complex was named centralspindlin [7]. Hereafter, we will use the terms MKLP1 and CYK-4 to refer to these molecules in general, and will use the species-specific names otherwise (Hs MKLP1, Dm Pavarotti, Ce ZEN-4; Hs MgcRacGAP/Cyk4, Dm RacGAP50C, Ce CYK-4).

MKLP1 is comprised of an N-terminal motor domain, an atypically long linker region, a parallel coiled-coil, and a C-terminal globular tail domain. CYK-4 has a short, N-terminal region followed by a parallel coiled-coil, a central C1 domain, and a C-terminal RhoGAP domain (Figure 2) [14]. MKLP1 and CYK-4 form homodimers through their respective coiled coils. These homodimers then form a heterotetramer. The binding interface between the two subunits of centralspindlin is formed by the N-terminal region of CYK-4 and an 85 residue linker domain in MKLP1 that forms the connection between the motor domain and the coiled-coil (Figure 2) [7, 15]. While the interaction between MKLP1 and CYK-4 is conserved, the sequences are not [7]. Intriguingly, the 85 residue linker domain in MKLP1 starkly contrasts with the corresponding region in all other identified kinesins with N-terminal motor domains [7]. In particular, the linker domain in most kinesins is 13–15 amino acids and the sequence is highly conserved [16]. In conventional kinesin, this critical region undergoes nucleotide-dependent conformational changes that are critical for plus end-directed motility [17, 18]. By analogy, it is possible that the divergent, CYK-4 binding linker region interacts with the motor core and modulates its motility.

Centralspindlin functions

Central spindle assembly

Centralspindlin is a major regulator of central spindle assembly in animal cells. At anaphase onset, centralspindlin precisely localizes to the plus ends of antiparallel microtubules where it initiates microtubule bundling and central spindle assembly (Figure 1) [7,19]. Neither CYK-4 nor MKLP1 alone can bundle microtubules, but the intact centralspindlin complex can induce extensive microtubule bundling in vitro [7]. Centralspindlin does not act alone in central spindle assembly, but rather it cooperates with additional motors and microtubule-associated proteins. The interplay between these factors has been recently reviewed [1].

Models for the molecular mechanism of microtubule bundling by centralspindlin

Although the molecular composition of centralspindlin has been characterized, the mechanistic details of how centralspindlin bundles microtubules are not adequately understood. Importantly, we do not yet know whether CYK-4 directly interacts with microtubules or whether it may alter the structural and/or biochemical properties of MKLP1. There are several attractive models for how centralspindlin could bundle microtubules. First, CYK-4 may directly bind microtubules through a binding site that has not been determined. Alternatively, CYK-4 binding may induce a structural rearrangement in MKLP1 to allow centralspindlin to preferentially bundle microtubules in an antiparallel configuration. CYK-4 could also promote discrimination (or assist MKLP1 in discriminating) between parallel and antiparallel microtubules. Finally, because CYK-4 binds within the neck linker region of MKLP1, a critical structure for the motility of other N-kinesins, CYK-4 binding may alter the motor behavior of MKLP1 to facilitate the precise and stable accumulation of centralspindlin on the plus ends of the antiparallel microtubules at the center of the central spindle. Because centralspindlin is a master regulator of cell division, it is critical to define the molecular requirement for CYK-4 in microtubule bundling and elucidate the mechanism by which the interaction between MKLP1 and CYK-4 allows centralspindlin to form the central spindle.

Centralspindlin clustering contributes to microtubule bundling

How does centralspindlin localize to the plus ends of antiparallel microtubules at the center of the central spindle? Mutation of critical residues within the nucleotide-binding region of MKLP1 disrupts the localization of centralspindlin to the central spindle indicating that the localization of centralspindlin depends on the catalytic activity of MKLP1 [20]. Interestingly, in vitro analysis of the motility behavior of ZEN-4 has shown that as a dimer, the motor is non-processive on single microtubules [21]. This non-processive behavior is unusual for kinesin motors, but, intriguingly, ZEN-4 and MKLP1 contain a domain that mediates oligomerization, allowing it to self-associate into higher order clusters (Figure 2). In contrast to dimers of ZEN-4, clusters of ZEN-4 are highly processive [21]. Clustering of centralspindlin is also required for central spindle assembly and cytokinesis in vivo [21]. A novel mechanism regulates the clustering of centralspindlin; this involves Aurora B kinase, a component of the chromosomal passenger complex, and 14-3-3, a highly conserved regulatory protein that associates with many different phosphoproteins [22]. MKLP1 can be phosphorylated on a conserved serine residue, S710, to which 14-3-3 binds (Figure 2), thereby inhibiting oligomerization [22]. Aurora B primarily phosphorylates MKLP1 at the highly conserved site S708 [23]. Phosphorylation at this site antagonizes 14-3-3 binding to phospho-S710 [22]. Thus, by inhibiting a negative regulator of centralspindlin, Aurora B promotes clustering of centralspindlin enabling it to stably accumulate on the plus ends of antiparallel microtubules [22].

Regulation of Rho family GTPases

Following central spindle assembly, the next role that centralspindlin performs during cytokinesis is to promote activation of the small GTPase RhoA (Figure 1). Because RhoA directly activates several effectors that are important for contractile ring assembly and cytokinesis, centralspindlin plays a critical role in directing cleavage furrow formation [24]. The accumulation of centralspindlin at the center of the spindle midzone, midway between the segregated chromosomes, places this complex in a reasonable position to execute this function. Nevertheless, this is a surprising activity for this complex given that the CYK-4 subunit of centralspindlin contains a RhoGAP domain (Figure 2) that would be predicted to inactivate RhoA, not to activate it.

RhoA activation during cytokinesis requires, and is likely mediated by, the RhoGEF ECT-2 [25] which directly interacts with CYK-4 (Figure 2) [26, 27]. The non-catalytic N-terminus of ECT-2 contains tandem BRCT domains that form a phosphopeptide binding module. This domain binds to the N-terminus of CYK-4 once it is phosphorylated by Polo-like kinase Plk1 [28, 29]. In most cell types studied to date, depletion of CYK-4 or mutation of its key Plk1 phosphorylation sites abrogates activation of RhoA [27, 30–32]. One exception occurs in C. elegans embryos. In these embryos, although CYK-4 is in fact required for central spindle-directed RhoA activation, a second pathway functions in parallel to CYK-4 to direct furrow formation and RhoA activation [2,4]. CYK-4 binding likely relieves ECT-2 autoinhibition [33] and directs recruitment of ECT-2 to the central spindle during anaphase where it is assumed to direct local activation of RhoA [28,29].

What is the function of the GAP domain of CYK-4?

The function of the GAP domain of CYK-4 has been studied in a wide range of contexts and although the results are somewhat system-dependent, they fall into 3 major categories. In the first group of experiments, mutation of the catalytic arginine of the GAP domain has no discernible effect on cytokinesis [34,35]. The second group indicates that the GAP domain performs its conventional function: limiting the activity of RhoA and/or other Rho GTPases [32]. The third group indicates that the GAP domain contributes to RhoA activation [34, 36–38].

How can these diverse conclusions concerning the function of the GAP domain of CYK-4 be explained? One issue is the nature of the mutations used in these various studies. These include a point mutation of the catalytic arginine, large and small deletions, and substitution mutations in the core of the GAP domain. These diverse perturbations may differentially affect potential activities of the GAP domain. For example, if the GAP domain modulates the activity of ECT-2 through an allosteric mechanism, point mutations that disrupt GAP activity may not necessarily impair allosteric activation of ECT-2, whereas less subtle mutations would be predicted to impair multiple functions of the GAP domain.

Interestingly, mutation of the catalytic arginine of the GAP domain has no cytokinetic phenotype in two of the three cells types examined, specifically chicken DT-40 cells [34] and the Drosophila nervous system [35]. In contrast, in Xenopus embryos, point mutation of this residue results in an increase in active RhoA and a broadening of the zone of active RhoA [32]. Large and small deletions within the GAP domain, as well as more severe substitution mutations, invariably have more dramatic phenotypes. In particular, in Drosophila embryos, a small deletion around the catalytic site appears to prevent contractile ring assembly [38]. In chicken DT-40 cells, deletion of the entire GAP domain also blocks cytokinesis, though the cytokinetic phenotype was not characterized in detail [34]. In C. elegans embryos, two independent substitution mutations in the GAP domain have been identified that abrogate the ability of CYK-4 to promote RhoA activation [36,37]. Collectively, these findings are consistent with a requirement for the GAP domain of CYK-4 for activation of RhoA. Interestingly though, in Xenopus embryos, deletion of the entire GAP domain actually results in an increase in the amount of active RhoA, although the localization of active of RhoA is unstable and appears to oscillate in the plane of the membrane. These oscillations were not seen however when the catalytic arginine was mutated [32].

Why does the specific loss of GAP activity lead to an increase in RhoA activity in some cases (Xenopus embryos) but not others (chicken B-lymphocytes, Drosophila embryos, C. elegans embryos)? The presence of other regulators of RhoA may explain these differences. For example, C. elegans embryos contain two other RhoGAPs, RGA-3/4, that negatively regulate RhoA activity [39,40]. Redundant RhoGAPs may also exist in DT-40 cells and Drosophila embryos, but not Xenopus embryos. Alternatively, if RhoA is only weakly activated under wild-type conditions in a particular cell type, loss of CYK-4 RhoGAP activity could be inconsequential. Although it is surprising that a GAP domain would contribute to the activation of RhoA, CYK-4 and ECT-2 interact through their N-termini (Figure 2), which could facilitate allosteric activation or localization of ECT-2 in a GAP domain-dependent manner. Overall, these results indicate that the function of the GAP domain may not be identical in all cell types. Furthermore, as will be discussed below, it is possible that the GAP activity of CYK-4 may be evolutionarily conserved due to cytokinesis-independent roles of this multifunctional protein.

Does RhoA activation require CYK-4 alone or the intact centralspindlin complex?

Whereas loss of CYK-4 generally results in either a reduction or elimination of active RhoA, depletion of MKLP1 has consequences of variable severity depending on the cell type. In human cells, depletion of MKLP1 does not prevent activation of RhoA, though it does impair its spatial regulation [20,27]. In contrast, in Drosophilaloss of the MKLP1 ortholog Pavarotti prevents all signs of contractile ring assembly [6]. In C. elegansa similar perturbation prevents central spindle-directed furrowing and reduces the accumulation of RhoA effectors, although central spindle-independent furrowing is unaffected [4]. These results suggest that CYK-4 must be properly localized in order to efficiently activate RhoA. Pavarotti may enhance the membrane recruitment of CYK-4, thereby promoting its ability to activate RhoA at the equatorial cortex. This is consistent with the membrane localization of Pavarotti [6] and the finding that a membrane-tethered form of CYK-4 induces hypercontractility in Drosophila S2 cells [41]. Alternatively, depletion of MKLP1 could cause destabilization of CYK-4, in a cell-type specific manner. Thus, while MKLP1 can facilitate CYK-4-dependent RhoA activation, there is not an absolute requirement for the intact centralspindlin complex for the activation of RhoA.

Is centralspindlin the molecular counterpart of Rappaport's astral-stimulating activity?

Because one of the two subunits of centralspindlin is implicated in RhoA activation, and the other is a kinesin family motor protein, centralspindlin is a good candidate for the furrow-inducing activity championed by Ray Rappaport [42]. He posited that a furrow-stimulating activity would travel along microtubules and promote furrowing at the cortical site that receives the highest level of stimulus, usually at the equator due to the combined action of the two asters. The ability of centralspindlin to translocate to the plus end of microtubules and to activate RhoA fulfill the two most important characteristics of this proposed furrow-stimulating activity. While this is attractive notion, additional analysis is required to confidently ascribe a molecular identity to this functional activity.

A positive signal from the spindle cannot explain all cases of furrow induction. In a variety of cell types at least two mechanisms are operable [2,3,43,44]. A mode independent of microtubules has even been documented in Drosophila neuroblasts [45]. Nevertheless, the positive signal from the spindle constitutes a dominant mechanism and the furrow resulting from this cue usually ingresses to completion. Therefore, the question of whether or not centralspindlin is the furrow-inducing stimulus should be restricted to cases where the central spindle directs furrowing (Figure 3). Indeed, in such cases, centralspindlin is essential. Interestingly, the rate of transport of the cleavage stimulus was estimated to be ~6 µm/min [46]. In vitrocentralspindlin oligomers travel at ~ 2 µm/min [21], which is not a perfect fit, but as this is a comparison between in vivo and in vitro measurements involving activities from different species, this difference should not be considered disqualifying. In sum, centralspindlin has many of the characteristics that are expected for an activity that communicates the position of the spindle to the cell cortex.

Figure 3.

Possible models for the pools of centralspindlin involved in furrow positioning. In the aster and central spindle model, centralspindlin concentrates on plus ends of all microtubules. In the central spindle model, centralspindlin concentrates primarily on overlapping plus ends. Only the latter model is consistent with the finding that midzone ablation eliminates furrow positioning (see text for details).

Importantly, there is also a distinction between centralspindlin and the proposed aster-stimulating activity. Centralspindlin accumulates dramatically on the antiparallel microtubules of the spindle midzone, and to a much lesser extent on astral microtubules (Figure 1) [6,7,9,11,14,19,47]; though astral microtubule localization has been detected [31,48]. The dramatic accumulation of centralspindlin on the spindle midzone raises the question of whether these pools of centralspindlin differ in their ability to induce furrows. Although this is an important question, it is a challenging one. It cannot be easily addressed by physical micromanipulation experiments. Repositioning the spindle moves both the asters and the midzone, large blocks placed in the equatorial region likely affect both the astral- and the central spindle-mediated pathways for furrowing [49], and nearby paired asters can form antiparallel bundles generating ectopic spindle midzones that could influence the ability of the spindle to signal [42,50,51]. Genetic perturbations that disrupt the spindle midzone without disrupting the asters provide another approach, but only if such perturbations of the midzone do not also perturb cortical stimulation, as is the case when centralspindlin is mutated. Laser ablation of the midzone of could distinguish between these models, and this has been tested in C. elegans embryos [52]. However, because these embryos furrow through two distinct pathways, it is necessary to perform this analysis in embryos in which the polar inhibition pathway is inhibited [2]). Laser ablation of the spindle midzone blocks furrow induction [52], suggesting that midzone-localized components are required to promote furrow induction and that astral-localized centralspindlin is not sufficient to do so, at least in this cell type (Figure 2).

This question bears further analysis in other cell types, because centralspindlin can associate with astral microtubules and there are observations that implicate this pool in inducing cortical contractility. Purified ZEN-4 accumulates at the tips of single microtubules in in vitro motility assays [21]. Consistent with this, in monopolar cells, centralspindlin accumulates at the ends of what appear to be parallel microtubule bundles that induce furrows [53,54]. However, single asters do not always have furrow inducing activity. For example, an aster will not induce furrowing in a spherical marine invertebrate embryo, but it will if the cell is deformed into a cylinder, likely as a consequence of symmetry breaking due to non-uniform aster-cortex interactions [55,56].

The accumulation of centralspindlin on asters is difficult to detect in cells with bipolar spindles (Figure 1), indicating that centralspindlin has a higher affinity for antiparallel microtubule bundles than parallel bundles. Interestingly, the accumulation of centralspindlin on the spindle midzone has distinct requirements from its accumulation on astral microtubules. In C. elegans embryos and in human cells with bipolar spindles, centralspindlin accumulation at the spindle midzone is independent of RhoA activation [19,57]. In monopolar cells, by contrast, RhoA activation is required for accumulation of centralspindlin on microtubule bundles [54]. The latter case is reminiscent of observations in Drosophila cells indicating an interdependence between central spindle assembly and contractile ring assembly [6,58,59]. Thus, there may be at least two distinct modes of centralspindlin accumulation on microtubules: a stable, RhoA-independent mode on overlapping, antiparallel plus ends, and a transient accumulation on parallel plus ends that is stabilized by RhoA activation.

In summary, the central spindle appears to be the primary site where centralspindlin accumulates and directs furrow formation. Indeed, centralspindlin-directed accumulation of ECT-2 at the central spindle is required for furrow induction [28,29]. Centralspindlin can associate with astral bundles and play a secondary role that can suffice under certain circumstances. Thus, while centralspindlin has the furrow inducing activity and the microtubule-based motility proposed for Rappaport’s aster-stimulating activity, its localization pattern and primary site of action differs substantially from the hypothesis.

However, it is still not clear exactly how and where CYK-4 facilitates RhoA activation. Although the spindle midzone appears to be the key site for furrow induction and ECT-2 activation, RhoA is likely to be activated while membrane-bound, due to the overlapping binding sites on RhoA between RhoGEFs and RhoGDI, a protein that allows RhoA to dissociate from membranes [60]. Alternatively, ECT-2 may be activated by CYK-4 on the central spindle after which ECT-2 could diffuse to the cortex and activate RhoA [28,29]. Finally, a complex of ECT-2 and centralspindlin may travel to the cortex and activate RhoA. In some cell types, the antiparallel microtubule bundles containing both centralspindlin and ECT-2 may lie in the sub-cortical region [27,31,61]. This question clearly requires further study.

Midbody assembly and abscission

The final process that centralspindlin mediates during cytokinesis is abscission, the resolution of a single membrane that envelopes the two nascent daughter cells into two separate membranes (Figure 1). This is a complex process involving maturation of the spindle midzone into a stable midbody, partial disassembly of the dense microtubule bundles, recruitment of proteins that directly promote abscission, and finally abscission per se.

Once the cleavage furrow has fully ingressed, the midbody proper forms. The midbody is a bulge that lies at the center of the cytoplasmic bridge connecting the two nascent daughter cells. The bulge contains highly compacted microtubules decorated by amorphous, electron-dense material. Centralspindlin remains highly concentrated in this region, although instead of concentrating on the overlapping microtubule plus ends, centralspindlin appears to undergo a structural rearrangement such that it is ordered into a small, dense, cortically associated ring surrounding the midbody [62]. Though some epitopes, notably tubulin, are masked in the midbody [63], this is not the basis for the absence of MKLP1 signal at the center of the midbody, as GFP fusions also indicate localization in a cortical ring. The mechanism underlying the transition from the microtubule bundles to the small cortical ring is not clear, nor has the link between centralspindlin and the plasma membrane been identified. Though the MKLP1 interacting protein ARF6 is an attractive candidate, as it has a lipid tail and binds to MKLP1 [64], it cannot be the sole link because MKLP1 localizes normally in ARF6 knockout cells [64]. However, it may contribute to midbody stability [65]. Anillin is also concentrated at the midbody. By virtue of its abilities to bind to centralspindlin, RhoA, ECT-2 and the plasma membrane, anillin could also facilitate membrane association of centralspindlin [43,66,67].

The first step towards abscission involves binding of CEP55 to centralspindlin (Figure 2) [68]. CEP55 is a dimeric coiled-coil protein that forms a bridge between centralspindlin and the factors that are directly responsible for membrane abscission. Specifically, CEP55 binds to both ALIX and the TSG101 subunit of the ESCRT-I subunit [69,70]. These components recruit, in turn, the ESCRT-III complex. All of these factors are required for completion of cytokinesis. The ESCRT-III complex is known to function in membrane scission events in which the neck of the scissile membrane is topologically extracellular, such as in viral budding, cytokinesis, and multivesicular body formation [69]. In all of these cases, the scissile event is mediated by proteins located on the cytoplasmic face of the narrow neck. Purified, recombinant ESCRT-III complex is sufficient to drive this class of membrane remodeling events. ESCRT-III subunits are found in membrane-associated filaments at constriction sites on either side of the mature midbody [71,72]. Therefore, ESCRT-III is probably directly responsible for abscission.

A sequence of direct protein-protein interactions links centralspindlin to the ESCRT-III complex. However, these interactions are subject to additional levels of regulation. For example, during mitosis, CEP55 concentrates on centrosomes and then relocalizes to the central spindle during the process of cytokinesis. The most recent data indicates that this recruitment occurs at the midbody stage due to inhibition by Plk1 [73]. The adaptor protein ALIX and the core ESCRT-III proteins likewise accumulate on mature midbodies [69–72], but they are not always co-recruited [73], indicating that additional regulatory mechanisms are at play. ESCRT-III complex subunits in turn recruit the microtubule severing complex spastin that facilitates abscission by promoting disassembly of the microtubule bundles in the midbody [74].

Following membrane abscission, centralspindlin rings can remain associated with the membrane, forming division remnants of variable stability. Their stability appears to be dependent on intact centralspindlin, as pre-formed remnants dissociate when embryos with a temperature sensitive allele of MKLP1 that disrupts binding to CYK-4 are shifted to the non-permissive temperature [14]. It is not currently known whether centralspindlin has a well-defined function in these rings following abscission. In C. elegans embryos, the division remnant correlates with the site of capture for astral microtubules which drives reorientation of the spindle in the posterior blastomere of the 2-cell embryo [14,75]. However direct evidence for centralspindlin involvement in this process is lacking. Eventually, these division remnants detach from the membrane and are ultimately degraded through autophagic mechanisms [76]. Intriguingly, the division remnants seem to persist in stem cells and transformed cells, though their roles in these cells are only beginning to be analyzed [76]. Studying this particular function of centralspindlin is challenging, as the complex must perform numerous essential functions prior to abscission.

Additional functions of centralspindlin

Though centralspindlin mediates a plethora of events during cytokinesis, the complex is also implicated in several biological processes that are unrelated to cytokinesis. These non-cytokinetic roles must be considered when interpreting centralspindlin mutant phenotypes. In particular, both CYK-4 and MKLP1 are highly expressed in the cerebellum as well as in the hippocampus during rodent development [77,78]. CYK-4 in particular is highly expressed in Purkinje cells which are notable for containing elaborate dendritic branches of mixed polarity microtubule bundles. Depletion studies implicate MKLP1 in dendrite formation [79]. Thus, the antiparallel microtubule bundling activity of centralspindlin may also be relevant in non-dividing cells. Increased dendritic branching and inappropriate axonal elongation have been observed in Drosophila embryos with a mutation in the tumbleweed gene, encoding Drosophila CYK-4. Interestingly, this function, but not cytokinesis, appears to require the GAP activity of CYK-4 [35]. Centralspindlin also has a role in controlling microtubule organization in Drosophila myotubes [80].

Quite distinctly, CYK-4 has been implicated in the regulation of mammalian kinetochores [81]. CYK-4 stabilizes a pool of CENP-A that is recruited to kinetochores during G1 of the cell cycle. Consistent with this function, CYK-4 can be detected at kinetochores during late G1. Evidence points to CYK-4 exerting this function through the cycling of the Rho-related GTPase Cdc42. However, this role of CYK-4 is not associated with chromosome segregation defects. Likewise, cells lacking Cdc42 are viable [82], indicating other mechanisms for CENP-A deposition must suffice to maintain the function of the centromere.

Finally, centralspindlin has also been implicated in several developmental processes. In particular, a hypomorphic allele of ZEN-4 in C. elegans causes abnormal pharynx development. In these mutants, a specific population of pharyngeal cells fail to acquire epithelial characteristics [83]. This defect is independent of previous cytokinetic defects, because the affected cells are mono-nucleated and because the defect can be rescued with a transgene that is expressed post-mitotically. Likewise, centralspindlin subunits can function as transcriptional regulators [84,85]. Lastly, MKLP1 and other central spindle-associated proteins concentrate adjacent to basal bodies in ciliated cells and can impact cilia growth [86].

These diverse, non-cytokinetic functions of centralspindlin raise the possibility that some domains in centralspindlin could be evolutionarily conserved not because of their role in cytokinesis, but rather because of the other important functions the complex executes. Similar considerations could also explain the transforming activity of the RhoGEF ECT-2. ECT-2 has documented roles in promoting cell migration and the epithelial-to-mesenchymal transition [87–90], and these functions of ECT-2 may underlie its ability to promote cell transformation, as opposed to its role in promoting cytokinesis.

Concluding Remarks

Here we have explored the myriad functions that the centralspindlin complex performs during animal cell cytokinesis. Remarkably, several of its critical functions are somewhat unexpected given the prominent conserved domains in this complex. Although endowed with motor domains, centralspindlin has an important role in generating stable, antiparallel microtubule bundles. Although it contains a Rho family GAP domain, it has a central role in the activation of RhoA during cytokinesis. The complex also has additional functions in cytokinesis as a motor protein complex, as a scaffold protein for key regulators of abscission, and as a conventional RhoGAP. Because of these diverse functions, centralspindlin lies at the heart of the cytokinetic mechanism.