Isoform-specific functions of Mud/NuMA mediate binucleation of Drosophila male accessory gland cells

Background In standard cell division, the cells undergo karyokinesis and then cytokinesis. Some cells, however, such as cardiomyocytes and hepatocytes, can produce binucleate cells by going through mitosis without cytokinesis. This cytokinesis skipping is thought to be due to the inhibition of cytokinesis machinery such as the central spindle or the contractile ring, but the mechanisms regulating it are unclear. We investigated them by characterizing the binucleation event during development of the Drosophila male accessory gland, in which all cells are binucleate. Results The accessory gland cells arrested the cell cycle at 50 hours after puparium formation (APF) and in the middle of the pupal stage stopped proliferating for 5 hours. They then restarted the cell cycle and at 55 hours APF entered the M-phase synchronously. At this stage, accessory gland cells binucleated by mitosis without cytokinesis. Binucleating cells displayed the standard karyokinesis progression but also showed unusual features such as a non-round shape, spindle orientation along the apico-basal axis, and poor assembly of the central spindle. Mud, a Drosophila homolog of NuMA, regulated the processes responsible for these three features, the classical isoform MudPBD and the two newly characterized isoforms MudL and MudS regulated them differently: MudL repressed cell rounding, MudPBD and MudS oriented the spindle along the apico-basal axis, and MudS and MudL repressed central spindle assembly. Importantly, overexpression of MudS induced binucleation even in standard proliferating cells such as those in imaginal discs. Conclusions We characterized the binucleation in the Drosophila male accessory gland and examined mechanisms that regulated unusual morphologies of binucleating cells. We demonstrated that Mud, a microtubule binding protein regulating spindle orientation, was involved in this binucleation. We suggest that atypical functions exerted by three structurally different isoforms of Mud regulate cell rounding, spindle orientation and central spindle assembly in binucleation. We also propose that MudS is a key regulator triggering cytokinesis skipping in binucleation processes. Electronic supplementary material The online version of this article (doi:10.1186/s12861-014-0046-5) contains supplementary material, which is available to authorized users.


Background
Most eukaryotic cells contain only a single nucleus because the karyokinesis in the M phase of the cell cycle is followed by cytokinesis. In certain cells, however, such as cardiomyocytes and hepatocytes, cytokinesis does not always occur, which results in cells containing two nuclei [1,2]. Sarcomere assembly is a possible factor repressing cytokinesis in cardiomyocytes [1,3], and insulin signaling plays a part in the generation of binucleate hepatocytes [2].
The production of binucleate cells is thought to result from the repression of certain phases of cytokinesis, such as formation of the contractile ring and ingression of the cleavage furrow. Cytokinesis occurs only when there is sufficient activation of Rho GTPase at the division plane. After chromosome segregation, a prominent bundle of microtubules, called the central spindle, forms between the spindle poles [4][5][6]. The centralspindlin complex, consisting of kinesin-6 and RhoGAP, moves toward the plus ends of the microtubules, corresponding to the cell equator, and associates with RhoGEF [7]. The RhoGEF thus specifically activates Rho GTPase at the division plane. Rho signaling activates effector proteins, such as diaphanous and Rho kinase, that in turn activate the formation of the actin contractile ring that completes cell division by pinching the daughter cells apart [8]. Loss of this cytokinesis machinery results in incomplete cytokinesis and produces binucleate cells [7,9,10]. There is, however, no solid evidence that normal binucleation events are regulated by inhibiting the functions of cytokinesis components, and little is known about the key regulators repressing the formation of cytokinesis machinery during binucleation. On the other hand, recent studies have shown a link between binucleation and inhibition of the cytokinesis machinery in cancer cells [11,12].
To investigate the mechanism by which cytokinesis is skipped during binucleation, we used as a model system the Drosophila male accessory gland, which produces seminal fluid proteins promoting reproductive success, such as the sex peptide Acp70A [13,14]. The exocrine epithelial cells in the male accessory gland, both the main cells and the secondary cells, are obviously binucleate ( Figure 1A) [15]. We previously showed that binucleation increases the plasticity of the cell shape, thereby enabling the volume of the accessory gland cavity to change [16], but the mechanisms of binucleation have remained unclear.
In the work reported here, we investigated the binucleation event in the accessory gland primordia, which was characterized by synchronous entry into the M phase after a cell-cycle-arrested interval during the mid-pupal developmental stage following standard cell proliferation in the early stage. We found that the binucleation results not from cell fusion but from mitosis without cytokinesis. We examined the mechanisms of binucleation by focusing on various morphological features different from those of standard dividing cells. We propose that isoform-specific functions of the microtubule binding protein Mud, a Drosophila homolog of NuMA, are the key regulators in binucleation of the Drosophila male accessory gland cells.

Results
Accessory gland epithelial cells are binucleated synchronously in the mid-pupal stage by mitosis without cytokinesis We first determined whether binucleation of the accessory gland epithelial cells is a result of skipping cytokinesis (as in cardiomyocytes). We observed the developmental stages and M-phase entry by using an antibody against phospho-histone H3 (P-H3), a marker for M-phase chromatin. Until 50 hours after puparium formation (APF), the accessory gland epithelial cells randomly entered the M phase but did not produce binucleate cells (Additional file 1: Figure S1A-E, A'-E' and J) ( Figure 1D). That is, standard cell division occurred. Subsequently, the cells arrested their cell cycle and delayed their M-phase entry for about 5 hours (50-55APF) (Additional file 1: Figure S1F and F') ( Figure 1D). The secondary cells then entered the M phase at 55 hours APF ( Figure 1B and D) (Additional file 1: Figure S1G and G'), and the main cells entered the M phase at 60 hours APF ( Figure 1C and D) (Additional file 1: Figure S1H and H'). We also found that the mitotic wave for binucleation in the main cell population initiated at the middle zone of the accessory gland lobe and propagated to the proximal and distal parts (Additional file 1: Figure S2). These results indicate a unique cell cycle regulation in this organ development. Importantly, the synchronous entries into the M phase accompanied the production of binucleate cells (Additional file 1: Figure S1K and Figure S2). No cytokinesis was evident in this M phase ( Figure 2F-J and F'-J'). After binucleation, the accessory gland epithelial cells did not enter a subsequent M phase (Additional file 1: Figure S1I and I', Figure S3) but showed a single round of the S phase, indicated by PCNA-GFP labeling (Additional file 1: Figure S3), indicating that endoreplication occurred ( Figure 1D). Thus the accessory gland epithelial cells, both secondary and main cells, became octaploid cells with two tetraploid nuclei. In the following section, we describe our examination of binucleation in the main cells. The secondary cells probably binucleated in the same way the main cells did.
Central spindle assembly and actin-contractile ring formation are inhibited during binucleation We next identified the cytological differences between standard cell division and binucleation in order to obtain clues to the mechanisms of cytokinesis skipping. For this purpose, we compared cell division in the early pupal accessory gland primordium and binucleation in the mid-pupal one ( Figure 1D). We focused on the following three differences. First, during the standard cell division in the early stage of accessory gland development, the M-phase cells were apically extruded and rounded ( Figure 2A-C and A'-C'), as is widely found in the standard epithelia [17]. During binucleation, however, cells were retained in the epithelial monolayer and did not show rounding ( Figure 2F-H and F'-H' compared with A-C and A'-C'). Second, the spindle orientation in standard mitosis was parallel to the epithelial plate ( Figure 2A-C and A'-C') [17]. The spindle formed during binucleation, in contrast, was always oriented perpendicular to the epithelial plate ( Figure 2F-H and F'-H'). Third, in standard cell division, the central spindles between segregated chromatids arose from anaphase to telophase, and then an actin contractile ring formed at the division plane ( Figure 2D,E,D' and E') [4][5][6]8]. During binucleation, in contrast, the central spindle was not properly assembled ( Figure 2I,J,I' and J'). Consequently, the subsequent formation of the contractile ring was also incomplete at telophase ( Figure 2J and J'), although the cleavage furrow was slightly formed at late anaphase ( Figure 2I and I'). On the other hand, other components of the mitotic spindle of binucleating cells looked normal, including the metaphase spindle ( Figure 2G and G' compared with B and B') [6] and kinetochore microtubules ( Figure 3C compared with A) [6,18]. Also apparently normal in the binucleating cells was the M phase progression indicated by the relationship between chromatid segregation (Figure 2G,H,G' and H' compared with B,C,B' and C') [18] and the decay of cyclin B ( Figure 3D and D' compared with B and B') [19]).
Forced activation of Rho GTPase in binucleation stage produces actin contractile ring Looking at the above three differences for clues to the mechanism of binucleation in the experiments intended to produce artificially rounded cells, we saw that cell extrusion and rounding did not seem to affect binucleation. In the binucleation stage neither overexpression of Sterile20-like kinase, which is related to cortical rigidity and cell rounding in standard cell division [20,21], nor knockdown of the adherens junction protein DE-cadherin encoded by shotgun [22], which maintains epithelial stability, led to cytokinesis (Additional file 1: Figure  S4A-D) ( Table 1). Binucleation also seemed to not be affected by the orientation of the mitotic spindle. This is because disruption of vertical spindle orientation in the binucleation stage by using a mutant for the centrosome protein Centrosomin or Sas-4 did not cause ectopic cytokinesis progression (Additional file 1: Figure S4E-H) ( Table 1) [23][24][25]. On the other hand, in centrosomin and Sas-4 mutants, weak assembly of the central spindle and an abnormal contractile ring could be observed at low frequencies (6% in centrosomin HK21 and 9% in Sas-4 S2214 ). However, we found that these induction-of-cytokinesis features were observed even in the cells showing normal (vertical) spindle orientation. Thus these results mean that contribution of the centrosome to the cytokinesis skipping during binucleation may be nonessential or even trivial.
Mis-assembly of the central spindle and incomplete formation of the contractile ring strongly induced skipping of cytokinesis during binucleation. The formation of the actin-contractile ring requires sufficient activation of Rho GTPase at the division plane [8]. To activate Rho GTPase encoded by Rho1, we temporarily elevated the level of the RhoGEF pebble around the binucleation stage. Pebble activates a Rho1 signaling cascade that phosphorylates MRLC (a regulatory light chain of non-muscle myosin II) encoded by spaghetti squash (sqh) to form the contractile ring [5,7,8]. Overexpression of pebble resulted in F-actin accumulation at the cleavage furrow, a sign of contractile ring formation, and in furrow ingression during the binucleation stage ( Figure 2K-O and K'-O') ( Table 1). We also found that overexpression of the activated form of sqh (sqh D20.D21 ) induced the formation of the contractile ring (Additional file 1: Figure S4I) ( Table 1). Furthermore, we tested a moderate overexpression of wild-type sqh or Septin-2 to see whether cells in the binucleation stage have a latent ability to create a contractile ring. Since the septin proteins are contractile ring components that act together with actomyosin and microtubules [26], we could easily see contractile ring formation in the telophase cells (Additional file 1: Figure S4J-M) ( Table 1). These results indicate that the level of central spindle assembly in accessory gland cells in telophase was too low to sufficiently activate Rho1 signaling for cytokinesis.

Mud regulates central spindle assembly, spindle orientation and cell rounding during binucleation
The insufficient activity of Rho1 signaling in binucleation is thought to be due to the insufficient assembly of the central spindle from anaphase to telophase, so we hypothesized that factors that repressed the central spindle assembly would be key regulators in skipping cytokinesis. Moreover, as stated above, the reduced central spindle assembly during binucleation should be accompanied by a non-round cell shape and orientation of the mitotic spindle along the apico-basal axis. Thus we examined candidate factorssuch as mitotic kinesins (kinesin-like protein at 10A, kinesin-like protein at 61 F, Pavarotti and no distributive disjunction) [4], microtubule-associated proteins (chromosome bows and Eb1) [4], microtubule-severing proteins (katanin 60, spastin) [27], Par proteins (bazooka, par-1) [28] and spindle orientation proteins (rapsynoid, G protein αi subunit and mushroom body defect (mud)) [29][30][31], with regard to their effects on cellular phenotypes of accessory gland cell binucleation. We found that loss of mud disrupted normal binucleation phenotypes, including   (Tables 2 and  3). mud encodes the Drosophila homolog of NuMA, which is associated with microtubules and plays a role in microtubule polymerization and determination of the spindle orientation [29][30][31]. In mud 4 hemizygotes and mud-knockdown cells, abnormally clear central spindle assembly and contractile ring formation were seen even in the binucleation stage ( Figure 4B-E and B'-E' compared with Figure 2I, J, I' and J') ( Table 2 and Table 3). Loss of mud was also associated with other morphological defects, such as abnormal spindle orientation and cell rounding ( Figure 4H-J and H'-J' compared with Figure 2H and H') ( Table 2 and Table 3). We thus confirmed that one copy of a chromosomal duplication encompassing the mud gene region ( Figure 4A) rescued these defects ( Figure 4F, G, F' and G') ( Table 2). These results suggest that Mud contributes to various cell morphologies during binucleation. We considered its repression of the central spindle assembly to be a major cause of binucleation because in mud 4 hemizygotes, we frequently observed cytokinesis progression even when neither horizontal spindle orientation nor cell rounding was observed ( Figure 4B-E and B'-E') (Additional file 2: Table S1).

Mud represses development of central spindle during binucleation
We showed that Mud repressed central spindle assembly and led to cytokinesis skipping. We made further observations of the central spindle assembly in wild-type and mud-knockdown cells in the binucleation stage. In particular, we used Pav::GFP (GFP-fusion of Pavarotti, a centralspindlin component) as a marker for the plus ends of the microtubules in the mitotic spindle [32,33]. Regarding the control results, during the standard cell division at the early pupal stage, Pav::GFP was localized to the midzone of the central spindle at telophase, as reported previously ( Figure 4K). During binucleation, in contrast, we never observed Pav::GFP-accumulated microtubules around the cell equator region ( Figure 4L). Detailed observation of Pav::GFP accumulation revealed that the impairment of central spindle assembly during binucleation was not due to impaired initiation of microtubule assembly. That is because the Pav::GFP accumulation around cell equator at anaphase was similarly observed in the case of dividing cells, implying that the central spindle precursor could develop at early anaphase (Additional file 1: Figure S5F and F' compared with A and A'). But during binucleation, impaired (5%, partial*) (5%, partial**) Induction of cell rounding fng-Gal4 (Tub-Gal80 TS ) > sterile 20-like kinase 0% (N = 34, 7) 0% (N = 34, 7) (12%, partial*) (18%, partial**) fng-Gal4 (Tub-Gal80 TS ) > shotgun.IR 0% (N = 37, 6) 0% (N = 37, 6) (8%, partial*) (14%, partial**) Disruption of centrosome*** centrosomin HK21 homozygote 6% (N = 36, 6) 6% (N = 36, 6) (11%, partial*) (8%, partial**) (22%, partial*) (42%, partial**) Numbers of cells and tissues (pairs of lobes) observed are shown as (N = cells, tissues). *A phenotype class with partial assembly of central spindle-like structure. **A phenotype class with partial accumulation of F-actin accompanied by furrow progression. ***Very low frequency of cells shows obvious misorientation of spindle axis. microtubule bundling or destabilization of the microtubules was apparent at late anaphase (Additional file 1: Figure S5G-J and G'-J' compared with C-E and C'-E'). As a result, during binucleation the development of microtubules was insufficient for their interdigitation at the spindle midzone. We also observed that the contractile ring component Peanut (Pnut), a Septin family protein, was localized around the cleavage furrow at anaphase but diffused during telophase (Additional file 1: Figure S6D-F and D'-F' compared with A-C and A'-C'). These results imply that the contractile ring begins to form during anaphase and is degraded during telophase. We also checked whether the loss of mud affected the localization of Pav::GFP during binucleation. In mudknockdown cells, the localization of Pav::GFP was restricted to the midzone of the central spindle ( Figure 4N). We also found that this localization was correlated with the central spindle formation. Cells with partially assembled central spindles showed a partial localization of Pav::GFP at the midzone ( Figure 4M), whereas cells with strongly assembled central spindles showed a clear localization ( Figure 4N). These results suggest that Mud represses microtubule polymerization so much that microtubule filaments are not targeted around the cell equator.

Isoform-specific functions of Mud regulate various traits in binucleation
Although Mud is known to promote polymerization of microtubules [34], our results showed that Mud may be a negative regulator of spindle formation during the  (45%, partial*) (33%, partial**) Numbers of cells and tissues (pairs of lobes) observed are shown as (N = cells, tissues). *A phenotype class with partial assembly of central spindle-like structure. **A phenotype class with partial accumulation of F-actin accompanied by furrow progression. ***Spindle orientated horizontally (0°± 45°) rather than vertically in metaphase. DC281: Chromosomal duplication encompassing the mud gene region. Table 3 Effects of knockdown for mud on binucleation and asymmetric cell division (asym. cell div.) RNAi target Frequencies of phenotypes in mitotic behaviors during binucleation asym. cell div.
Central spindle assembly Contractile ring formation Cell rounding Abnormal spindle orientation*** Multi-bristle binucleation process. The above results also indicate that unknown functions of Mud regulate central spindle assembly, spindle orientation, and cell rounding. Interestingly, mud generates various splicing variants that have different C-termini (Additional file 1: Figure S3A and referred to in http://flybase.org). To understand the relationship between these unknown functions of Mud and its structural isoforms that had not fully been analyzed, we compared the functions of three different splicing variants ( Figure 4A) (Additional file 1: Figure S7): a Pins binding domain (PBD)-containing isoform (Mud PBD ), a PBD-lacking longer isoform (Mud L ) and a PBD-lacking shorter isoform (Mud S ). Mud PBD is wellknown major variant that regulates spindle polarity in a Pins/LGN-dependent manner and promotes microtubule polymerization [29][30][31]34]. Mud L and Mud S , both of which lack the Pins/LGN-binding domain ( Figure 4A) (Additional file 1: Figure S7), have not yet been examined and their functions are unclear. We thus expected that mud L or mud S would exhibit novel functions that regulate central spindle assembly and spindle orientation during binucleation.
To identify isoform-dependent functions of Mud, we tested the ability of Mud PBD , Mud L and Mud S to rescue the various defects in binucleation in mud hemizygotes. Overexpression of mud PBD (Additional file 1: Figure S7 and Figure S8) rescued only their abnormal spindle orientation ( Figure 6A-C and A'-C'), reducing the frequency of abnormal spindle orientation from 40% in mud 4 /Y to 17% in mud 4 /Y + mud PBD ( Table 2). Overexpression of mud S (Additional file 1: Figure S7 and Figure S8) rescued their abnormally enhanced central spindle assembly ( Figure 6H-J and H'-J'), reducing the frequency of central spindle assembly from 82% in mud 4 /Y to 13% in mud 4 /Y + mud S ( Table 2). mud S also partially rescued the abnormal spindle orientation ( Figure 6G and G'), reducing the frequency of abnormal spindle orientation from 40% in mud 4 /Y to 20% in mud 4 /Y + mud S ( Table 2). In contrast, the overexpression of mud L (Additional file 1: Figure  S7 and Figure S8) effectively rescued the cell rounding phenotype found in mud hemizygotes ( Figure 6D and D'), reducing the frequency of cell rounding from 30% in mud 4 /Y to 12% in mud 4 /Y + mud L ( Table 2). mud L also partially rescued the abnormally enhanced assembly of central spindle ( Figure 6E, E', F and F'), reducing the frequency of central spindle assembly from 82% in mud 4 /Y to 30% in mud 4 /Y + mud L ( Table 2), but its ability was obviously less than that of mud S (compare the reduction to 30% by mud 4 /Y + mud L with the reduction to 13% by mud 4 /Y + mud S ). These results imply that Mud has the following isoform-dependent functions during binucleation: Mud L represses cell rounding and weakly represses central spindle assembly, Mud PBD and Mud S each play a role in orienting the spindle axis along the apico-basal axis, and Mud S also strongly represses central spindle assembly.

Mud S orients mitotic spindle along the apico-basal polarity and inhibits cytokinesis
The above overexpression results suggest that Mud S contributes to cytokinesis skipping during binucleation more than Mud L does, but the functions of Mud S during Drosophila development have not yet been reported. We therefore tried to determine whether endogenously expressed mud S actually regulates binucleation in the accessory gland. To do so, we performed a mud S -specific knockdown ( Figure 4A and Figure 6H-K and I'-K') (Additional file 1: Figure S7 and Figure S8) and found that the mud S -knockdown cells showed abnormally enhanced assembly of the central spindle ( Figure 6L, L', M and M') and abnormal orientation of the mitotic spindle ( Figure 6K and K') ( Table 3). These results strongly suggest that Mud S represses central spindle assembly and orients the spindle axis vertically but does not regulate cell rounding.
Mud PBD is known to regulate spindle orientation during asymmetric cell division, so we tried to determine whether the spindle orientation-regulating function of Mud S during binucleation is independent of that of Mud PBD . We tried to do that by determining whether a knockdown of Mud S affected asymmetric cell division in the sensory organ precursor cells (SOPs) on the adult wing margins ( Figure 5C). The spindle orientation of this asymmetric cell division is known to be regulated by Mud PBD in a Pins-and Gαi-dependent manner, and loss of this function results in a multi-bristle phenotype on the adult wing ( Figure 5C). In fact, the knockdown of all Mud isoforms caused abnormal spindle orientation and the multi-bristle phenotype (arrowheads in Figure 5E compared with D) (Table 3) [29,35,36]. The knockdown of mud S alone, in contrast, did not induce the multi-bristle phenotype ( Figure 5F) (Table 3), although it effectively caused spindle orientation changes during binucleation ( Figure 6I and I') ( Table 3). These results suggest that even though Mud S regulates spindle orientation during binucleation, it is not involved in the spindle orientation during asymmetric cell division of SOPs.
As shown above, Mud S seems to convert the mitotic morphologies from the cell-division type (horizontal spindle orientation and cytokinesis progression) to the binucleation type (vertical spindle orientation and cytokinesis skipping) ( Figure 6) ( Table 2 and Table 3). To examine this hypothesis, we tested the ability of Mud S to convert the mitotic morphologies of cells in the early pupal accessory gland primordia, in which standard cell division occurs. The results showed that the spindle orientation changed from horizontal to vertical ( Figure 8E compared with Figure 2D). Furthermore, binucleate cells appeared with some frequency (arrowhead in Figure 8F). This effect of overexpression of mud S on cell morphologies can be seen not only in the accessory gland but also in the wing imaginal disc ( Figure 8K, K', L and L'). Overexpression of mud PBD or mud L , in contrast, affected neither spindle orientation nor cytokinesis (Figure 8A-D, G-J and G'-J'). These results suggest that Mud S can convert mitotic morphologies from the cell-division type to the binucleation type.

Regulation of M-phase entry in binucleation of Drosophila male accessory gland
We characterized the final M-phase entry that contributed to binucleation of cells in the Drosophila male accessory gland ( Figure 1D) (Additional file 1: Figure  S1). We showed that the entry into the binucleation stage took place with a two-step cell cycle transition: cell cycle arrest for 5 hours and subsequent synchronous entry into the M phase ( Figure 1D) (Additional file 1: Figure S1). Thus, standard cell division and binucleation are separated by a 5-hour interval of cell cycle arrest (Additional file 1: Figure 1D) (Additional file 1: Figure   S1). This may indicate that standard mitosis and binucleation have very different regulations of the cell cycle and cytokinesis. In fact, although in most tissues in Drosophila standard mitosis can occur normally without the spindle checkpoint [37], we previously demonstrated that the knockdown of mad2, a spindle checkpoint component, frequently causes a defect in chromosomal segregation in accessory gland cells [16]. This is consistent with our proposal that binucleation is regulated by a system different from the one regulating standard mitosis.

Morphological features of mitotic cells during binucleation
We identified three morphological features characteristic of binucleation: the non-round shape of mitotic cells, the apico-basal orientation of the mitotic spindle and the poor assembly of the central spindle ( Figure 2F-J and F'-J' compared with A-E and A'-E'). The reduction of central spindle assembly could directly repress cytokinesis, but we did not find the significance of the vertical orientation of the spindle or the non-round shape of the cell. We can explain it as follows. If most cells of the epithelium synchronously enter the M phase with the cell rounding phenotype, the stability of the monolayer epithelium may be severely disrupted. In fact, we observed a severe defect in epithelial stability after the binucleation stage in a mud mutant in which the cells were rounded during the binucleation stage (data not shown). In addition, if the spindles in columnar cells are horizontally oriented, they will be less stable than they would be if they were vertically oriented because a spindle is aligned more stably along a longer axis than a shorter one [38]. Vertical orientation of the spindle and lack of cell rounding may thus be appropriate for synchronous binucleation of columnar cells, but these morphological features do not directly regulate cytokinesis skipping.

Mechanisms by which Mud regulates central spindle assembly, spindle orientation, and cell rounding during binucleation
We propose that Mud is a key factor in regulating binucleation. We demonstrated that Mud functions in a way that represses central spindle assembly, orients the mitotic spindle along the apico-basal axis and inhibits mitotic cell rounding during mitosis (Figure 7). We found a clue as to how Mud represses the central spindle assembly. In standard cell division, during late anaphase the microtubules of the central spindle are polymerized in order to target their plus ends at the cell equator [32,33], and we confirmed this in our experiments in which the plus ends were labeled with Pav::GFP ( Figure 4K). During binucleation, in contrast, the Pav::GFP marker did not sufficiently target the cell equator ( Figure 4L). This implies that the polymerization of the microtubules of the central spindle is insufficient to target them at the cell equator, and the central spindle therefore does not develop completely. We also showed that the Mud is needed in order to repress the growth of microtubules. In fact, the knockdown of mud promoted the growth of microtubules that targeted their plus ends at the cell equator even in the binucleation stage ( Figure 4M and N). However, the underlying molecular mechanisms of Mud in repressing polymerization of microtubules remain unclear. Moreover, the question of how Mud regulates the mitotic spindle orientation along the apico-basal axis and how it inhibits mitotic cell rounding are also unclear. The logical next step will be to relate the isoform of each Mud to various effector molecules regulating the orientation and the rounding.
Alternative splicing of mud/NuMA produces three types of structurally different proteins We showed that the three types of alternative splicing products Mud PBD , Mud L and Mud S have distinct functions from one another ( Figure 6, Figure 5, Figure 8 and Figure 7). It is known that the mud gene produces four isoforms (RH, RI, RJ and RL) that contain the Pins/LGN binding domain (PBD) and three isoforms (RF, RG and RK) that do not contain it ( Figure 5A) (Additional file 1: Figure  S7) (http://flybase.org). Moreover, the three PBD-lacking isoforms are structurally classified into the following two types: a longer isoform (RF) that simply skips the PBDencoding exons by alternative splicing, and two shorter isoforms (RG and RK) that contain a shorter isoformspecific exon instead of the PBD-encoding exons ( Figure 5A) (Additional file 1: Figure S7).
At least one of the PBD-containing isoforms (RL) is functional and known to regulate the spindle orientation in a Pins/LGN-dependent manner in asymmetric cell division [29][30][31]. In contrast, although mud L and mud S are actually transcribed [34] there is no evidence that Mud L and Mud S are functional and have Pins/LGN-independent functions. As in the Drosophila gene mud, splicing variants also occur in human NuMA1, and these variants encode a longer isoform (NuMA1-l), a shorter isoform (NuMA1-s) and a medium isoform (NuMA1-m) [39]. Although the functions of NuMA1-m and NuMA1-s are unclear, NuMA1-l has an LGN-binding domain in the C-terminal region and determines the spindle polarity in an LGN-dependent manner, the same as in the case of Drosophila Mud PBD ( Figure 5A) [40]. NuMA1-m and NuMA1-s, in contrast, like Drosophila Mud L and Mud S do not have an LGN-binding domain in the C-terminal region ( Figure 5A). Interestingly, we found sequence similarities between human NuMA1-s and Drosophila Mud S in their C-terminal domains, including the shorter isoform-specific regions ( Figure 5B). These similarities suggest that Mud S in Drosophila functions similarly to NuMA1-s in humans.

Isoform-dependent functions of Mud mediate various morphological changes of binucleating cells
We showed that the functions of mud PBD , mud L and mud S are independent during binucleation. The repression of mitotic cell rounding was a Mud L -specific function. In contrast, changing the orientation of the mitotic spindle along the apico-basal axis was controlled by both Mud PBD and Mud S (Figure 6 and Figure 7) ( Table 2 and  Table 3). Mud PBD was previously reported to be required for the spindle orientation during asymmetric cell division [29,35,36]. We showed, however, that Mud S is not associated with the spindle orientation during asymmetric cell division ( Figure 5F). On the other hand, the overexpression of mud S but not mud PBD reoriented the spindle along the apico-basal axis in dividing cells ( Figure 8A and E). These results suggest that Mud PBD and Mud S regulate the spindle orientation independently.
The function repressing central spindle assembly during binucleation was also shared by Mud L and Mud S , but we showed that Mud S contributed the most to repressing spindle assembly. In fact, the overexpression of mud S effectively rescued the mud 4 mutant phenotype, the abnormally enhanced assembly of the central spindle ( Figure 6I and J) ( Table 2). In addition, like the mud 4 mutant, a mud S -specific knockdown abnormally enhanced central spindle assembly in the binucleation stage ( Figure 6L and M) ( Table 3). Mud L , in contrast, only partially repressed the central spindle assembly during binucleation ( Figure 6F) ( Table 2). Moreover, overexpression of mud S , but not mud L , inhibited cytokinesis to produce binucleate cells in dividing cells such as the early pupal accessory gland cells ( Figure 8D and F) and the larval wing disc cells ( Figure 8J and L). This also suggests that mud S mainly contributes to the repression of central spindle assembly (Figure 7).

Conclusions
We described the binucleation event of the Drosophila male accessory gland during pupal development and analyzed the cellular mechanisms regulating this binucleation. We characterized a unique cell cycle regulation in the developing accessory gland: the M-phase entry for binucleation occurred synchronously at 55 APF after a cell cycle arrest for 5 hours. We also found that Mud, the Drosophila homolog of mammalian NuMA, regulated various features of the binucleating cells, such as a non-round shape, spindle orientation along the apico-basal axis, poor assembly of the central spindle and cytokinesis skipping. It is known that Mud binds Pins to determine the mitotic spindle orientation during the standard cell division or asymmetric cell division [29][30][31]41,42]. Interestingly, we found atypical functions of Mud that depended on three types of splicing isoforms, each differently regulating the above various features of binucleating cells. We concluded that Mud PBD , which is a well-known isoform having a PBD ( Figure 5A) (Additional file 1: Figure S7), oriented the spindle along the apico-basal axis. Mud L , one of the newly characterized isoforms and simply lacking a PBD, inhibited the mitotic cell rounding and weakly impaired the central spindle assembly (Figure 7). Mud S , another newly characterized isoform, containing a shorter isoform-specific domain instead of a PBD-containing domain ( Figure 5A) (Additional file 1: Figure S7), oriented the spindle along the apico-basal axis and strongly impaired the central spindle assembly (Figure 7). Importantly, overexpression of Mud S induced an ectopic binucleation even in the cell division stage, whereas overexpression of Mud PBD or Mud L did not (Figure 8). These results suggest that Mud S is an important regulator triggering cytokinesis skipping in binucleation. Abnormal expression of NuMA is known to be correlated with the production of cancer cells in mammals [43]. Our finding of atypical functions of Mud may contribute to the understanding of the relationship between NuMA and tumor progression.

Detection of formation of central spindle and contractile ring
We used microtubule bundles and actin filaments as markers for the central spindle and contractile ring, respectively. In central spindle assembly, microtubule filaments bundle together, forming a large structure that crosses the cell equator. We regarded microtubulerelated structures having these features (i.e., bundling and equator-crossing) with fewer microtubule filaments as partially assembled central spindles. However, if the filaments neither associated with each other nor crossed the cell equator, we did not regard the structures as central spindles.
The accumulation of actin filaments in the equatorial region of the cell membrane is usually associated with formation of a cleavage furrow, and we considered a contractile ring to have partially formed if either of these two features was observed.
Temporal expression of genes using the TARGET system The TARGET system [48] was used for the temporal expression of genes in the pupal accessory gland epithelium. To restrict the expression of the target genes by activation of Gal80 TS , flies were reared at a permissive temperature (19°C). To permit moderate expression of the target genes by weak activation of Gal80 TS , flies were reared at a semi-permissive temperature (26°C).

Construction of plasmids
pP-Acp70A-Stinger: The upstream enhancer of Acp70A (-477 to -34) was amplified from the genome DNA of Canton-S with a PCR (primer set #1 in Additional file 2: TableS2) and subcloned into T vector pMD20 (TaKaRa) by using TA cloning. The fragment for the Acp70A enhancer was digested with BglII and NotI and subcloned into pH-Stinger (Drosophila Genomics Resource Center (DGRC)) to construct pP-Acp70A-Stinger. pUAS-FLAG::mud PBD -attB (see Additional file 1: Figure S7, Figure S8 and Additional file 2: Table S1): The genomic fragment encompassing three subisoforms of mud PBD (mud-RH, mud-RI, and mud-RL in Additional file 1: Figure S7) (from just after the start codon to the stop codon) was amplified with a PCR (primer set #2 in Additional file 2: Table S2) from the BAC clone CH322-147E14 (P[acman] Resources) [58] and subcloned into T vector pMD20 (TaKaRa) by using TA cloning. The forward primer also included the Kozak sequence, start codon and FLAG tag sequence. The seventh intron (eighth intron in the case of mud-RJ), which included the mud S -specific exon, was removed from the fragment with an inverse PCR (primer set #3 in Additional file 2: Table S2) and ligated to generate a FLAG-tagged protein-coding region including mud-RH, mud-RI and mud-RL. This fragment was digested with NotI and KpnI and subcloned into pUASattB (FlyC31, http://www.flyc31.org/) [59] to generate pUAS-FLAG:: mud PBD -attB. pUAS-FLAG::mud L -attB (see Additional file 1: Figure S7, Figure S8 and Additional file 2: Table S1): The genomic fragment encompassing a mud L (mud-RF in Additional file 1: Figure S7) (from just after the start codon to the stop codon) was amplified with PCR (primer set #2 in Additional file 2: Table S2) from the BAC clone CH322-147E14 (P[acman] Resources) [58] and subcloned into T vector pMD20 (TaKaRa) by using TA cloning. The forward primer also included the Kozak sequence, start codon and FLAG tag sequence. The seventh intron, which included the PBD-encoding exons and mud Sspecific exon, was removed from the fragment with an inverse PCR (primer set #4 in Additional file 2: Table S2) and ligated to generate a FLAG-tagged Mud-RF-coding fragment. This fragment was digested with NotI and KpnI and subcloned into pUASattB (FlyC31, http://www.flyc31.org/) [59] to generate pUAS-FLAG::mud L -attB. pUAS-FLAG::mud S -attB (see Additional file 1: Figure S7, Figure S8 and Additional file 2: Table S2): The genomic fragment encompassing two subisoforms of mud S (mud-RG and mud-RK in Additional file 1: Figure S7) (from just after the start codon to the stop codon) was amplified with a PCR (primer set #5 in Additional files 2: Table S2) from the BAC clone CH322-147E14 (P[acman] Resources) [58] and subcloned into T vector pMD20 (TaKaRa) by using TA cloning. The forward primer also included the Kozak sequence, start codon and FLAG tag sequence. The FLAG-tagged fragment, which included both mud-RG and mud-RK, was digested with NotI and KpnI and subcloned into pUASattB (FlyC31, http://www. flyc31.org/) [59] to generate pUAS-FLAG::mud S -attB.
pUAS-mud S .IR-attB (see Additional file 1: Figure S7, Figure S8 and Additional file 2: Table S2): The genomic fragment including the mud S -specific exon and upstream intron with a splice donor and acceptor (Additional file 1: Figure S7) was amplified with a PCR (primer set #6 in Additional file 2: Table S2) from the genome DNA of Canton-S and subcloned into T vector pMD20 (TaKaRa) by using TA cloning (#1). The mud S -specific exon without the upstream intron was also amplified with a PCR (primer set #7 in Additional file 2: Table S2) from the genome DNA of Canton-S and subcloned into T vector pMD20 (TaKaRa) by using TA cloning (#2). The plasmid #1 was digested with NotI and XhoI, and the plasmid #2 was digested with EcoRI and NotI. Both of the digested fragments were combined in a single plasmid pUASattB (FlyC31, http://www.flyc31. org/) [59] to generate pUAS-mud S .IR.

RT-PCR to verify the expression of UAS-mud isoforms
Total RNA was extracted from third-instar larvae of the following four genotypes: w/Y;; hs-Gal4/UAS-FLAG::mud PBD , w/Y;; hs-Gal4/UAS-FLAG::mud L , w/Y;; hs-Gal4/UAS-FLAG::mud S , and w/Y;; hs-Gal4/+ (negative control). Before the extraction of RNA, larvae were heat-shocked twice at 37°C for 45 minutes and subsequently incubated at 25°C for 2 hours to reach the high level expression of UAS targeted genes. The cDNA of each genotype was synthesized using the oligo-dT primer with PrimeScript RT-PCR kit (TaKaRa). Two primer sets (#8 and #9 listed in Additional file 2: Table S2 of the supplementary material) that specifically amplify FLAG::mud transgenes but not endogenous mud genes (Additional file 1: Figure S8) were used for the PCR.
Semi-quantitative RT-PCR to verify isoform-specific knockdown by UAS-mud S .IR Total RNA was extracted from third-instar larvae of the following two genotypes: w/Y; Act5C-Gal4/UAS-mCherry. IR (control) and w/Y; Act5C-Gal4/+; +/UAS-mud S .IR (knockdown for mud S ). The cDNA of each genotype was synthesized using the oligo-dT primer with PrimeScript RT-PCR kit (TaKaRa). Expressions of genes were normalized by using Rpl32 as a reference gene. The three sets of primers #10, #11 and #12 listed in Additional file 2: Table S2 of the supplementary material (Additional file 1: Figure S7) were used for the PCR.

Animal ethics
All animals used in above genetic experiments were anesthetized with carbon dioxide before each mating. All procedures complied with guidelines of the Animal Ethics Committee of Gakushuin University.
Additional file 2: Tables S1 and S2 are included.