A mosaic genetic screen for novel mutations affecting Drosophila neuroblast divisions

Background The asymmetric segregation of determinants during cell division is a fundamental mechanism for generating cell fate diversity during development. In Drosophila, neural precursors (neuroblasts) divide in a stem cell-like manner generating a larger apical neuroblast and a smaller basal ganglion mother cell. The cell fate determinant Prospero and its adapter protein Miranda are asymmetrically localized to the basal cortex of the dividing neuroblast and segregated into the GMC upon cytokinesis. Previous screens to identify components of the asymmetric division machinery have concentrated on embryonic phenotypes. However, such screens are reaching saturation and are limited in that the maternal contribution of many genes can mask the effects of zygotic loss of function, and other approaches will be necessary to identify further genes involved in neuroblast asymmetric division. Results We have performed a genetic screen in the third instar larval brain using the basal localization of Miranda as a marker for neuroblast asymmetry. In addition to the examination of pupal lethal mutations, we have employed the MARCM (Mosaic Analysis with a Repressible Cell Marker) system to generate postembryonic clones of mutations with an early lethal phase. We have screened a total of 2,300 mutagenized chromosomes and isolated alleles affecting cell fate, the localization of basal determinants or the orientation of the mitotic spindle. We have also identified a number of complementation groups exhibiting defects in cell cycle progression and cytokinesis, including both novel genes and new alleles of known components of these processes. Conclusion We have identified four mutations which affect the process of neuroblast asymmetric division. One of these, mapping to the imaginal discs arrested locus, suggests a novel role for the anaphase promoting complex/cyclosome (APC/C) in the targeting of determinants to the basal cortex. The identification and analysis of the remaining mutations will further advance our understanding of the process of asymmetric cell division. We have also isolated a number of mutations affecting cell division which will complement the functional genomics approaches to this process being employed by other laboratories. Taken together, these results demonstrate the value of mosaic screens in the identification of genes involved in neuroblast division.


Background
The development of the nervous system of higher organisms requires the generation of an extraordinary cellular diversity. One mechanism by which this diversity can be established is the segregation of cell fate determinants to one specific daughter during cell division thereby generating progeny with different cellular identities. Neuroblasts, the Drosophila neural progenitors, have served as one of the major models for studying asymmetric division (reviewed in [1]). Neuroblasts divide along an apicalbasal axis, utilizing apical cues inherited from the neuroectoderm out of which they delaminate [2,3], to generate daughter cells with distinct identities. The large apical daughter cell retains its neuroblast identity and continues to divide while the small basal daughter cell, the ganglion mother cell (GMC), undergoes a single division to generate two postmitotic progeny of neuronal or glial identity.
The initial step in defining the asymmetry of neuroblast divisions is the establishment at the apical cortex of a multi-protein complex (reviewed in [4,5]) containing Inscuteable and two highly conserved signalling cassettes, the Par proteins -Bazooka (the Drosophila homologue of Par-3), Par-6 and atypical protein kinase C (DaPKC) -and the heterotrimeric G protein subunit Gαi together with the guanine nucleotide dissociation inhibitors Partner of Inscuteable (Pins) and Locomotion defects (Loco). The apical complex has several important functions during neuroblast asymmetric division including the correct orientation of the mitotic spindle along the apical-basal axis of the cell, the displacement of the spindle towards the basal cortex [6,7] and the establishment of a difference in spindle length between its apical and basal halves at anaphase [6,8]. This gives rise to a dramatic size asymmetry between daughter cells, with a small basal GMC budding from a large apical neuroblast. The apical complex is also essential for directing the localization of cell fate determinants to the neuroblast basal cortex. Phosphorylation of Lethal giant larvae (Lgl) by DaPKC appears to lead to the activation of Myosin II and the exclusion of Miranda from the apical cortex [9][10][11]. Myosin VI (Jaguar) is also required for basal localization of Miranda [12], although the mechanisms by which Miranda is transported and/or anchored to the basal cortex remain unknown.
Miranda functions as an adapter protein, localizing Staufen and Prospero (Pros) to the basal cortex [13][14][15]. Staufen, an RNA-binding protein required in the oocyte to localize bicoid mRNA [16], is employed in the neuroblast to anchor pros mRNA basally [17][18][19]. The segregation into the basal daughter cell of the homeodomain protein Prospero and its mRNA is the critical step in establishing GMC identity [20][21][22]. In the GMC, Pros translocates to the nucleus where it regulates gene expression, directing a drastic change in cellular identity [23][24][25].
Several molecules known to be involved in asymmetric neuroblast division have been identified in zygotic loss of function screens looking for embryonic phenotypes but the major limitation of this method is that maternal contribution of mRNA will mask the effects of the loss of many genes during embryogenesis. In support of this idea, animals lacking zygotic Gαi, Pins or Loco -three components of the apical complex identified biochemically or by a candidate gene approach -are viable, albeit with locomotion defects, and fertile, indicating that alleles of these genes would not have been found in a zygotic loss of function genetic screen [26][27][28][29].
A systematic germline clone screen might be an effective way to identify new components of the asymmetric cell division machinery. However, components such as Myosin II and Jaguar are required during oogenesis and do not give rise to fertilized eggs in germline clones [30,31], and we considered that such a screen would miss a number of the genes involved in neuroblast division.
To minimize the complications of maternal contribution or requirement of components of the asymmetry machinery in oogenesis, we decided to avoid embryonic neuroblasts entirely and switch to examining asymmetric division in third instar larval neuroblasts, which are known to employ most of the machinery used in embryos in an analogous manner [32]. Analysis of mutations in third instar larvae is possible directly where homozygous mutant animals survive until this stage and the phenotype of pins mutants has been examined in this way [33]. However, the majority of mutations in genes with important roles during development -including many of the molecules known to be required for neuroblast asymmetric division -are lethal before the third larval instar, and cannot be examined in this way. The classic method to circumvent early lethality in Drosophila is to generate postembryonic clones of cells homozygous for the mutation of interest, and we have used a variation of such a strategy in which clones are positively marked by the expression of GFP: the MARCM system (Mosaic Analysis with a Repressible Cell Marker; [34]), which has been used previously in a screen for phenotypes in mushroom body clones [35]. Core to the MARCM system is the use of the yeast GAL80 repressor, which blocks transcriptional activation by GAL4. Generation of somatic clones lacking GAL80 and homozygous for a mutation of interest in animals expressing GAL4 allows expression of UAS-CD8::GFP only within the clone. We considered an additional advantage to the MARCM approach. A number of molecules which have recently been implicated in neuroblast asymmetry, including Lgl, Myosin II, Myosin VI and Cdc2 [36], as well as Rab11 and Sec15 -components of the vesicular trafficking machinery [37,38] -have highlighted the importance of the general cellular machinery in asymmetric cell division. A clonal screen, therefore, has the advantage that mutations in components of the cell division machinery can be identified by a lack or an excess of proliferation within clones, which can then be examined with regard to markers of asymmetric cell division.
In this study, we describe the results of a MARCM screen on chromosome arm 3L, together with a screen of pupal lethal and semi-lethal mutations on chromosome 3. We identified 78 mutations affecting neuroblast division that fall into 48 complementation groups. The majority of these represent genes required for cell division, 12 of which correspond to previously described loci, and several of which also appear to have effects on asymmetric cell division. Although the bulk of the cell division mutants isolated in our screen do not have clear polarity phenotypes, we have deficiency mapped many of them to small genomic regions, reasoning that this would complement several RNAi-based screens currently being conducted to look for genes involved in cell division (for example [39]). In addition to the cell division complementation groups, we found new loci involved in neuroblast asymmetric division, with phenotypes affecting spindle orientation, localization of basal determinants and neuroblast cell fate.

Screen design and overview
To systematically identify genes involved in neuroblast division, while minimising the effects of maternal contribution, we have used a mosaic approach, employing the MARCM system [34]. To confirm the suitability of this system, we first generated clones of a wild type chromosome and examined the expression and localization of proteins known to be asymmetrically distributed in metaphase neuroblasts. As expected, within each clone we see a single large neuroblast with a variable number of smaller progeny labeled with CD8::GFP under the control of elavGAL4 C155 , which directs expression in neuroblasts as well as neuroblast progeny ( Figure 1B). As we anticipated from previous reports [32], the localization of known components of the asymmetric cell division machinery in metaphase neuroblasts appears to follow the model established from studies in embryos. Although larval neuroblasts do not have a clear apical-basal polarity with respect to the overlying epithelium, such as that defined in embryonic neuroblasts, known apical components localize in a cortical crescent opposite a crescent of known basal components, with the metaphase plate aligned in between ( Figure 1C). In contrast to previous findings [32], we also observe localization of Prospero to the basal cortex of >90% of metaphase neuroblasts ( Figure 1D). We do observe one minor difference between larval neuroblasts and those of the embryo. In embryonic neuroblasts, the basal marker Miranda is first localized to the apical cortex A screen for asymmetric division mutants in third instar lar-val neuroblasts Figure 1 A screen for asymmetric division mutants in third instar larval neuroblasts. (A) Crossing scheme used to establish mutant lines and generate MARCM clones of mutations lethal before the third instar stage. The asterisk indicates an EMS-induced mutation. (B-E) Expression of asymmetric cell division markers in third instar larval neuroblasts. DNA staining (blue); Miranda (red, B, C, E); Prospero (Red, D); CD8::GFP (green, B); Inscuteable (green, C, E). Bar: 10 μm. Wild type clones (B) contain a single large neuroblast, with a crescent of Miranda at metaphase, and a cluster of smaller Miranda negative progeny. Note that the CD8::GFP membrane marker outlines the mitotic spindle (arrowheads). (C) The apical and basal markers Inscuteable and Miranda are localized to opposite poles of third instar metaphase neuroblasts, while crescents of Prospero are seen in most neuroblasts (D); note that the neuroblast Pros staining is weaker than the nuclear staining in neighbouring cells. In contrast to embryonic neuroblasts, Miranda does not colocalize with Inscuteable in early prophase but rather is found both in the cytosol and in a cortical crescent at the opposite pole of the cell to Inscuteable (E).
in early prophase and then translocated to form a tight basal crescent at metaphase [14,15,19]. In larval neuroblasts we never observe co-localization of Miranda with Inscuteable. Rather, during early prophase Miranda is found both cytoplasmically and in a cortical crescent at the opposite pole of the cell to Inscuteable ( Figure 1E and data not shown). These observations confirm the validity of this clonal screening approach for neuroblast phenotypes, and suggest that the same or a largely overlapping repertoire of molecules is employed in larval and embryonic neuroblasts. As localization of the basal components is downstream of correct apical complex function, we chose to use an antibody against Miranda [40], together with the orientation of the metaphase plate and relative cell size in neuroblast clones, as final readouts of neuroblast asymmetric division.
To generate mutant lines, male flies isogenic for a chromosome carrying an FRT insertion at 79D-F (FRT2A) were mutagenized with ethyl methanesulfonate (EMS) and stocks established carrying mutations balanced over TM6B, Tb. In total we generated 1923 stocks carrying mutations causing lethality before the wandering third instar larval stage, as assayed by the absence of Tb + larvae, and approximately 350 pupal-lethal and semi-lethal chromosomes.
The 350 pupal-and semi-lethal lines giving rise to third instar larvae homozygous for the mutant chromosome were screened by antibody staining of non-Tb larvae. The crossing scheme for generating somatic clones of the remaining 1923 mutant lines is given in figure 1A. Females from our mutant lines were mated to males carrying tub-GAL80 on the FRT2A chromosome together with elavGAL4 C155 , hsp70Flp and UAS-CD8::GFP. Induction of mitotic recombination by 37° heatshocks at first and second instar stages resulted in the generation of clones within neural lineages that were homozygous for mutations on 3L and positively marked by the expression of CD8::GFP. Clones were readily identifiable by examination of intact larvae under UV light revealing that the great majority of Tb + female third instar larvae contained mutant clones in the brain. Nervous systems from these larvae were dissected and screened by antibody staining against Miranda and GFP and confocal microscopy. At least three metaphase or anaphase neuroblasts in at least two brains were examined looking for alteration or absence of the basal Miranda crescent or any misorientation of the metaphase spindle.

Pupal lethal screen
We initially screened our collection of approximately 350 third chromosome pupal-and semi-lethal mutations for those that showed a neuroblast phenotype in the larval brain, scoring at least five dividing neuroblasts in at least two brains. We identified two lines with neuroblast asymmetry defects ( Table 1: Asymmetric cell division defects), and four lines with defects in neuroblast cell division. All of these lines cause lethality at pupal or pharate adult stages, and complementation testing between them indicated that these six mutants represent five separate complementation groups ( Figure 2E-H; Table 1: Cell division defects).

Clonal screen
We screened our collection of 1923 lethal chromosomes for phenotypes in MARCM neuroblast clones, looking particularly for defects in the formation and localization of the Miranda crescent but also more generally for clones exhibiting cell division phenotypes. We identified two lines with defects in neuroblast asymmetric division ( An additional class of mutations failed to give rise to neuroblast containing clones, and we observed either no clones in these brains or clones of one or two neuronal cells, typical of neuron or GMC clones; we never observed clones larger than two cells not containing a neuroblast. Although in principle these small clones might reflect inappropriate neuroblast differentiation, the majority of these lines most likely carry cell lethal mutations, and were not examined further.

Asymmetric cell division defects PL26
In our pupal lethal screen, we identified a single mutation with a neuroblast polarity phenotype. In animals homozygous for the PL26 chromosome, we identified a misalignment of the mitotic spindle with respect to the Miranda crescent in a proportion of metaphase/anaphase neuroblasts. We scored misoriented spindles as those which form an angle of greater than 22.5° with the apicalbasal axis of the cell, defined by the position of the Miranda crescent -in wild-type metaphase/anaphase neuroblasts we never see this degree of misorientation. In PL26 brains, we observed 12/75 neuroblasts with >22.5°m isorientation compared to 0/63 for the FRT2A progenitor chromosome (Figure 2A, B), although it is presently unclear what proportion of these misoriented spindles will result in a failure to appropriately partition cell fate determinants into the GMC, and we have not investigated whether a telophase rescue of this misalignment -such as has been described in embryos [2,41] -might occur. We co-stained neuroblasts using an antibody against the Inscuteable protein, and found that the Miranda crescent forms opposite the Inscuteable crescent in all cases, but the metaphase plate is not correctly aligned with respect to the neuroblast apical-basal axis ( Figure 2C). Clones homozygous for the left arm of this chromosome do not show any phenotype (not shown) but it is possible that this lack of phenotype reflects a perdurance of the wild type gene product rather than necessarily localizing the PL26 mutation to 3R. We did identify a single deficiency in the Bloomington kit which fails to complement the PL26 lethality, but hemizygotes of PL26 with this deficiency do not show the spindle misalignment phenotype (not shown), and we have been unable to map the PL26 phenotype further.

PL17
One mutant, PL17, was initially identified in our pupal lethal screen as having a high mitotic index but overall reduction in the size of homozygous larval brains, suggesting a defect in cell cycle progression (described below). In addition to these defects we observed a frequent mislocalization of Miranda to discrete cytoplasmic regions, although localization of the apical marker Inscuteable appears unaffected ( Figure 2D, D'). Co-staining with the centrosomal markers Centrosomin (Cnn) or γtubulin (not shown) suggests that in these neuroblasts Miranda is located in a pericentrosomal region during mitosis. This phenotype is incompletely penetrant, and some neuroblasts exhibit normal Miranda crescents, while others have both crescents and pericentrosomal Miranda, presumably a reflection of differences in the perdurance of maternally provided protein between cells. MARCM clones of PL17 appear phenotypically wild-type (not shown), and we interpret this as a consequence of the perdurance of wild-type protein within clones. Deficiency mapping and complementation testing revealed that PL17 is allelic to imaginal discs arrested (ida), which encodes the Drosophila APC5 homologue, and a more detailed characterization of the defects observed in this line will be presented elsewhere (CS, PMO, R. Tuxworth and WC, manuscript in preparation).

J16
We isolated a single mutant line, J16, which appears to have defects in the size asymmetry of neuroblast divisions within mutant clones. Wild type clones invariably contain a single large neuroblast accompanied by a number of much smaller GMCs and neurons. Miranda expression is found in neuroblasts, and occasionally in GMCs, but never in post-mitotic neurons ( Figure 3A). In contrast to the wild type situation, clones of the J16 chromosome contain a number of large cells, many of which express Miranda, suggesting that the progeny of J16 neuroblast divisions fail to adopt the correct cell fate ( Figure 3B). Examining wild type clones we find that the neuroblasts are typically around 12 μm in diameter, with the neurons roughly 4 μm across. Cells within J16 clones are intermediate in size to that of neuroblasts and neurons, with sizes in the region of 7-9 μm ( Figure 3C, D).

Mutations identified in the pupal lethal screen
Lethality in this region is caused by a mutation in trio 9 A38, B10, B18 The majority of cells in J16 clones appear to be in metaphase, with Miranda localized to a cortical crescent, indicating a cell cycle arrest in J16 clones. Deficiency mapping using the Bloomington kit revealed a single lethal region containing the polo locus and subsequent complementation tests revealed that the J16 chromosome carries an allele of polo. Hemizygotes of J16 with a deficiency for the polo region (Df(3L)Exel9636) survive to the third larval instar, and brains of these animals contain appear to have a high rate of metaphase arrest in neuroblasts ( Figure 3E), but do not exhibit any phenotype suggesting a defect in asymmetric neuroblast division. This observation suggests that a second mutation in the J16 line is responsible for the cell size defects observed in clones.
To rule out any contribution of polo to the cell size phenotype observed in J16 clones, we introduced a polo genomic rescue fragment carried on the second chromosome into the J16 mutant background [42]. Clones of J16 with polo function thus restored do not exhibit the metaphase arrest phenotype, but still contain multiple similarly sized Miranda-expressing cells ( Figure 3F; arrow indicates a cell in anaphase). We have not undertaken a detailed analysis of asymmetric cell division in these clones, and it is not yet clear whether the phenotype is a reflection of a symmetric mode of division such as has been described in early larvae [32].
To investigate the consequence of these defects in neuroblast division we examined the expression of neuroblast and cell fate markers in homozygous J16 embryos. We do not see any defects in neurogenesis in J16 embryos, as assayed by the expression of the proneural marker Achaete first in proneural clusters and then neuroblasts (not shown). Although neuroblasts form correctly in J16, by stage 9 we observe a loss of expression of Worniu, a marker for all embryonic neuroblasts [28,43,44], at a low frequency ( Figure 3G, H). To determine whether this apparent loss of neuroblast identity affects the specifica- The J16 mutation leads to defects in neuroblast cell fate tion of cell fate in neuroblast progeny cells, we examined the expression of the neuronal marker Even-skipped (Eve) in stage 16 J16 embryos ( Figure 3I, J). In wild type embryos, Eve is expressed in ~ 20 neurons per hemisegment: the aCC/pCC, CQ, and RP2 neurons, and the EL neuron cluster, which are the respective progeny of four neuroblasts [45]. As the early loss of neuroblast identity would suggest, we find a loss of Eve-expressing neurons with a low frequency (~ 2% for CQ and RP2 neurons, 4% for EL neurons, n = 420). In all cases the entire progeny of an individual neuroblast are lost, suggesting that neuroblasts which lose Worniu expression in J16 embryos do not give rise to any of the appropriate progeny. As in larvae, introduction of a polo genomic rescue fragment was unable to rescue this J16 phenotype (not shown). The low penetrance of neuroblast defects in the embryo suggests the perdurance of maternal protein may be masking the embryonic phenotype. As we do not obtain fertilized eggs in J16 germline clones, even with the restoration of polo function, we have not explored this further.
The J16 mutation appears to lie outside the region uncovered by the Bloomington deficiency kit, and further investigation will be required to identify the genomic region responsible for these defects.

D76
We isolated a single allele, D76, in which dividing neuroblasts are almost never seen. On examining large numbers of clones we were able to determine that the rare neuroblasts which enter metaphase do not form a basal Miranda crescent, although Miranda is expressed and can be detected at low levels in the cytosol and Inscuteable is localized appropriately to the apical cortex ( Figure 4A, A', B, B'). In some neuroblasts we observe punctate spots of Miranda staining in the cytoplasm (indicated by an arrow in Figure 4C), the identity of which is unclear. Deficiency mapping uncovered a single lethal deficiency from the Bloomington kit, and complementation testing indicated that this corresponds to a hit in the trio gene. However, the phenotype of trio mutant larvae has been described and appears unlikely to be responsible for the phenotypes we observe in neuroblasts. Recombination mapping using a ru h th st cu sr e s ca chromosome revealed that the phenotype maps proximal to ru, excluding trio, and distal to th (data not shown). However, we have been unable to identify additional lethal deficiencies within this interval, and so have not mapped the D76 mutation further.

Cell division defects
In addition to the four lines with phenotypes during neuroblast asymmetric division, we identified a total of 76 mutations, in 46 complementation groups, which exhibit defects in cell division (Table 1: Cell division defects). We have categorized these as having defects in cell prolifera-tion or chromosome separation, or as giving rise to multinucleate or metaphase arrested cells; a final category contains lines which have a variety of membrane or vesicular defects.

Proliferation defects
In our pupal lethal screen we identified three lines in which the brains of homozygous larvae appear reduced in size (Table 1: Proliferation defects). Two of these mutations -PL13 and LVC73 -give similar phenotypes. The optic lobes of the brain are generally small, with fewer neuroblasts than wild type, and we rarely observe mitotic cells The D76 mutation affects cell proliferation and Miranda localization ( Figure 2E, F). Staining with an antibody against phospho-histone H3 (anti-PH3) revealed that the proportion of dividing neuroblasts in these brains is greatly reduced compared to wild type (for example 22% of neuroblasts in the central brain region stain with anti-PH3 in PL13 homozygous brains (n = 37), compared to 45% in PL13/ FRT2A heterozygotes (n = 60)), although when we do see neuroblasts in metaphase the localization of Miranda and orientation of the mitotic spindle appear normal. Despite their phenotypic similarities, these mutant lines do not show any lethality as transheterozygotes and their lethal regions map to distinct regions of the chromosome.
An additional line, PL17 -the Miranda localization phenotype of which is described above -gives larvae with brains in which the optic lobes are again reduced in size. However, the proportion of dividing neuroblasts in this line is significantly increased compared to wild type (76% of neuroblasts stain with anti-PH3 (n = 45)), suggesting a delay at meta-or anaphase ( Figure 2G). We identified two overlapping deficiencies in the Bloomington kit, Df(3L)HR119 and Df(3L)GN34, which fail to complement the lethality of PL17, and complementation testing revealed that PL17 is allelic to imaginal discs arrested.
In our clonal screen, we isolated nine mutations in which we see clones of only one or two cells but in which the neuroblasts do not show any gross morphological defects.
Because of the small numbers of mutant clone neuroblasts generated in our screening regimen we have not attempted to accurately measure the mitotic index in these mutations, but in eight out of nine cases the lack of mitotic neuroblasts observed suggests that the mitotic index is greatly reduced ( Figure 5A, B).
These eight mutations fall into five complemention groups which we have mapped to separate regions of 3L (Table 1: Proliferation defects). Testing known mutants in these regions revealed that the two mutations that fall into complementation group 7 are allelic to small-minded (smid). Furthermore, the phenotype observed in clones of these mutations reflects the described phenotype of smid homozygous larvae in which proliferation of postembryonic neuroblasts is reduced [46]. As described above, one of these complementation groups, D76, also gives rise defects in Miranda localization in the rare metaphase neuroblasts observed.
In addition to these lines with very low mitotic indices, we identified one line, E45, in which clone neuroblasts appear to be arrested in a metaphase-like state with condensed chromosomes, although without the formation of a clear metaphase plate ( Figure 5C). This mutation mapped to a region containing makos, the Drosophila cdc27 homologue, alleles of which are known to give a similar metaphase-like arrest phenotype [47], and testing against a known mutation indicated that our line is a new mks allele. Sequencing of the mks locus in our stock revealed a single nucleotide G→A transversion leading to the conversion of a tryptophan residue at amino acid 622 -before the conserved TPR repeat region -into a stop codon, suggesting that our mutation represents a loss of function allele.

Chromosome separation defects
A number of mutations were isolated which appear to show defects in sister chromatid separation at anaphase ( Table 1: Chromosome separation defects). These three complementation groups give rise to clone neuroblasts which appear to have an abnormally high DNA content but which do not contain multiple nuclei (Fig 5D, E, E').
Complementation testing with known cell division mutants on 3L identified one of these complementation groups, containing five mutations, as allelic to Kinesin-like protein at 61F (Klp61F; Figure 5D). A further group, with a single member C93 ( Figure 5E, E'), mapped to a deficiency containing separase (sse), previously known to be required for chromosome separation, and sequence analysis indicated that the C93 chromosome carries a G→A transversion at the exon 6 splice acceptor of sse, presumably leading to the formation of a truncated protein. Transheterozygotes of C93 with sse M13 or hemizygotes with Df(3L)Exel6106, which uncovers the sse locus, survive until late third instar stages and show phenotypes similar to those observed in C93 clones. These brains contain large cells which have an increased DNA content ( Figure  5F, F', F", G): we note that the phenotype in these animals is more severe than we observe in clones, suggesting -as we observed for the PL17 mutation -that these clones retain some wild-type sse protein.
Although no obvious Miranda localization defects in C93 clones were observed, we found that the large neuroblasts in C93/Df(3L)Exel6106 hemizygotes frequently contained several cortical crescents of Miranda instead of a single basal crescent. Interestingly, when we examined the localization of Inscuteable in these larvae, we found that neuroblasts with several Miranda crescents also had several Inscuteable crescents, and that these never overlapped ( Figure 5F, F', F"). In wild type neuroblasts the domains of Inscuteable and Miranda never abut precisely but are separated by a region of cortex containing neither protein, and in these mutant neuroblasts we see a similar region between the Inscuteable and Miranda crescents. To investigate the origin of the multiple crescents in C93 hemizygotes we stained neuroblasts with the centrosomal markers Centrosomin (Cnn) and γ-tubulin ( Figure 5H, H', I, I'), and found that these cells contain large numbers of centrosomes; presumably the failure to separate sister chromatids at anaphase leads to multiple rounds of cen-trosome and DNA duplication and failed cell division. It is possible that the presence of multiple centrosomes causes the formation of multiple crescents of Inscuteable, by a mechanism similar to that described for the centrosome-induced cortical polarity of the C. elegans embryo [48], although we have not attempted to establish a direct correspondence between individual centrosomes and Inscuteable crescents in these cells. The mechanism responsible for the basal positioning of Miranda in wild type neuroblasts could then lead to the formation of Miranda crescents in those regions of the cortex not occupied by Inscuteable.

Multinucleate cells
Two lines isolated in our pupal lethal screen, IV61 and CMV111, are allelic to one another and have a phenotype in larval brains consistent with a defect in cytokinesis: we frequently observe large cells which appear to have an unusually high DNA content ( Figure 2H). Complementation testing with known cell division mutants on chromosome 3 indicates that this complementation group is allelic to sticky, which encodes a serine/threonine kinase related to the mammalian citron kinase, and which has previously been shown to be required for cytokinesis in all Drosophila tissues [49].
In the clonal screen we identified a further allele of sti, D97 -which has a similar phenotype to the pupal lethal alleles -as well as 24 additional mutations, in 15 complementation groups, which also give rise to multinucleate neuroblasts (Table 1: Multinucleate cells). One of these mutations, C33, shows a severe cell division phenotype in neuroblasts and was mapped to the pebble locus, previously shown to be required for cytokinesis ([50]; Figure  6A). Mutant neuroblasts from the other 14 complementation groups typically contain two to four nuclei and may also represent a failure to complete cytokinesis ( Figure 6B, C, D). One of these is allelic to TBP-associated factor 4 (Taf4), and the multinuclear phenotype is presumably a downstream consequence of defects at the transcriptional level. The remaining complementation groups mapped to a number of chromosomal regions with the exception of groups 30 and 31 which complement the entire Bloomington deficiency kit and were not mapped further.

Membrane/vesicular defects
The final group of mutants we classified as having cell division phenotypes are those which give rise to small clones, typically of one or a few cells, in which the most striking defects are associated with the CD8::GFP which labels membranes (Table 1: Vesicular/membrane defects). Whereas in wild type neuroblast clones we observe the CD8::GFP outlining the cell and nucleus, and surrounding the mitotic spindle, in a number of these lines we frequently see multiple brightly labelled punctae within the cell, the identity of which is unclear ( Figure 6E, arrows). In other lines in this category we observe the CD8::GFP staining to fill the majority of the cell, suggesting an excess of membrane is present, but without necessarily concentrating into bright spots ( Figure 6F). In total we isolated 31 alleles of 18 genes, five of which we have identified by deficiency mapping and complementation analysis. We have isolated novel alleles of reptin, Taf-6, Int6 and Aatsile, which all encode proteins involved in general cellular processes at the level of transcription or translation. The phenotypes observed in mutant clones of these genes presumably reflect pleiotropic effects resulting from a disruption of these basal processes. We have also isolated a new allele of neurexin which is required for correct vesicle trafficking at the neuromuscular junction [51]. The remaining mutants in this category may therefore disrupt genes whose functions are required for membrane biosynthesis and/or vesicle targeting.

Discussion
We found a total of five complementation groups that affect the distribution of Miranda in dividing neuroblasts or the size asymmetry of the neuroblast division. At the time of writing, two have been mapped to a gene. One of these is separase which seems unlikely to be involved directly in the mechanism of asymmetric cell division. The second is ida which appears to disrupt the basal localization of Miranda leading to an accumulation of Miranda in a pericentrosomal compartment. A more detailed description of this phenotype, which suggests a novel connection between the APC/C and the localization of cell fate determinants, will be presented elsewhere (CS, PMO, R. Tuxworth and WC, manuscript in preparation). Several of the other mutations isolated in the screen are also likely to have a direct influence on the asymmetry machinery. In particular, we have identified a mutation affecting metaphase spindle orientation in neuroblasts, a process which is critical to the correct segregation of cell fate determinants, and another which appears to perturb the size asymmetry of the neuroblast division and the correct establishment of different identities in daughter cells.

Multinucleate cells in mutant clones
We did not find as many asymmetric division mutations as we might have expected given the scale of the screen: we estimate, using a Poisson approximation, that we have achieved approximately 80% saturation of 3L. Although it is possible that the design of our screen has prevented the detection of some genes, we believe that this is a reflection of the rather small number of genes involved exclusively in asymmetric cell division. Indeed, a similar screen of 3R has detected new alleles of several known players of neuroblast asymmetric division including miranda, prospero and scribbled, a tumour suppressor in the Lgl/Dlg pathway (RS-N, WGS and WC, unpublished data). This suggests that our screen methodology is effective at finding genes involved in neuroblast asymmetric division, and that it can in the future be applied to the whole genome to find further novel components of this machinery. The major limitation with the approach at present is that it is rather labour intensive, which necessarily leads to a low throughput compared to other screening methods. The use of live imaging methods, coupled with the GFP and RFP fusion reagents widely in use, would most likely relieve this bottleneck and allow a higher throughput. Similarly, improvements to the crossing schemes could be made. If chromosomes could be screened without the need to establish stocks of each one, perhaps with mutations being recovered from the siblings of the larvae examined, an F1 screen could be conducted which would enable a genome-wide saturating screen to be carried out much more rapidly.
In addition to finding several new polarity genes, we have isolated large numbers of mutations affecting cell division, and this seems to be a particular strength of such a clonal approach. Looking only at complementation groups which give rise to multinucleate neuroblasts we have found mutations in 13 regions not previously known to be involved in cytokinesis, and similar screens may prove beneficial to laboratories with a specific interest in aspects of cell division. As with other screening methods, in several cases mapping of the gene responsible for a cell division phenotype has yielded unexpected results: for example, mutations in taf4, involved in transcriptional initiation, appear to cause a failure in cytokinesis. Nonetheless, we have identified alleles of a number of previously reported cell division genes and we anticipate that a substantial number of the mutations described here will be directly involved in the processes of cell division.
Early studies of factors involved in asymmetric cell division described a number of genes with phenotypes that specifically disrupt this process and for which the establishment of neuroblast polarity is the primary or only function. It is now starting to become clear, however, that many of the molecules required for neuroblast asymmetry are also employed in a number of other roles within the cell, as well as in a range of tissues during development. For example, lethal giant larvae (lgl) and discs large (dlg), involved in the localization of basal components in neuroblasts, are both tumour suppressors, and in their absence larval brains and discs show a dramatic overgrowth (reviewed in [52]). Similarly, at least two Myosins, necessary for a range of cellular processes throughout development, are involved in Miranda localization [9,12]. This is probably why simple zygotic genetic screens looking for defects specifically in neuroblast asymmetric division are nearing their limit. In our clonal screen we found several mutations which adversely influence neuroblast proliferation but also disrupt the formation of the basal Miranda crescent. The molecules responsible for these phenotypes remain to be isolated, but may shed light on the connection between the cell cycle and localization of cell fate determinants. The differences in phenotypes observed in these lines will be of particular interest. For example, in D76 the Miranda crescent is entirely lost, while in PL17, affecting the Drosophila APC5 homologue, Miranda is observed to be strongly associated with the centrosome -the latter case is intriguing as Miranda has previously been found to be centrosomally localized in a cell cycle dependent manner, although how this relates to Miranda localization and function is unclear [19,53].
Aside from the clear utility of the MARCM system in screening for mutations affecting aspects of cell division, we consider that it could easily be adapted for use in a mis-expression screen. As clones are positively labelled by elavGal4 directing expression of GFP, the presence of an EP insertion [54], or one of the related UAS-containing transposable elements, not necessarily on the chromosome arm carrying the FRT site, would lead to transcription of genes downstream of this element specifically in labelled clones, and circumvent any early lethality caused by ectopic expression, a frequent limitation of such screens.

Conclusion
Previous screens for components of the asymmetric division machinery have focused on embryonic phenotypes, and are now reaching saturation. Here we have used a clonal approach to screen neuroblasts in the third instar larval brain, and have identified several novel mutations, the identification and further study of which will advance our understanding of the process of neuroblast asymmetric division.

Fly stocks and genetics
All Drosophila stocks were reared and maintained on standard yeast-cornmeal-agar medium [55] and all experiments were performed at 25°. To generate mutant lines, w flies carrying an FRT element inserted at polytene segment 79D-F (FRT 2A) were first isogenized for the third chromosome. Three to five day old males were then mutagenized by feeding with 1% sucrose solution containing 25 or 38 mM EMS as described previously [56]. Mutagenized males were crossed en masse to virgin females of the genotype TM8/TM6B, Tb, Ubi-GFP. Single male progeny were crossed to the balancer stock to establish stocks in which the mutagenized third chromosomes were balanced over TM6B, Tb, Ubi-GFP. Lines in which the mutagenized chromosome was homozygous viable were discarded.
Stocks in which animals homozygous for the mutagenized FRT2A chromosome were viable at the wandering third larval instar stage (350 lines), as assayed by the absence of the Tb marker, were screened by antibody staining of mutant brains. Lines lethal before the third instar (1923 lines) were screened using the MARCM system [34]. Females of each mutant stock were crossed to males of the MARCM driver line elavGAL4 C155 ,hsp70Flp/Y; UAS-CD8::GFP, UAS-LacZ; tub-GAL80, FRT2A (a gift from A. Gould and B. Bello). Crosses set up using females from the MARCM driver line and males from a mutant stock were significantly less productive, and these crosses were avoided except in cases in which few females of a mutant stock could be obtained.
After 24 hour periods of egg laying, progeny were heatshocked twice for 2 hours in a 37° water bath at first instar and second instar stages to induce mitotic recombination. Brains of female non-Tb wandering third instar larvae were dissected and screened by antibody staining.
Following screening, lines were placed in complementation groups by pairwise complementation testing and tested against alleles of the following candidate genes on 3L: pebble, encore, nuclear fallout, four wheel drive, pavarotti, fumble, polo, sticky and kinesin-like protein at 61F (Klp61F). This allowed us to identify alleles of pebble, polo, sticky and Klp61F. The remaining complementation groups were mapped initially by crossing to the third chromosome deficiency kit, provided by the Bloomington Drosophila stock centre. We performed further fine scale mapping with smaller deficiencies obtained from the Drosdel and Exelixis collections [57,58], as well as other deficiencies obtained from Bloomington, to define the minimal region containing each complementation group -deficiency breakpoints are described in Flybase [59]. Testing candidate genes in these regions allowed us to assign addi-tional groups as small-minded, TBP-associated factor 4 and -6 (Taf4 and -6), reptin, Isoleucyl-tRNA synthetase (Aats-ile), separase, makos, Int6 and neurexin. A number of complementation groups containing only a single allele failed to complement more than one region of 3L. In several of these cases we were able to place the phenotype in a single region by examination of hemizygous phenotypes or by meiotic recombination with appropriate markers from a ru h th st cu sr e s ca chromosome followed by clonal analysis. Several groups complemented the entire Bloomington deficiency kit for 3L and were not mapped further.
Stocks, other than those used to test candidate genes and available from the Bloomington or Szeged stock centres, were mks 1 [47], P{GFP-polo} [42] and sse M13 [60].
Secondary antibodies were obtained from Jackson labs (Cy3/HRP) or Molecular Probes (Alexa-488) and used at a concentration of 1:1000 (fluorescence) or 1:500 (HRP). DNA was visualized by the addition of ToPro-3-iodide (1:20,000, Molecular Probes) to one of the wash steps. Antibody staining of embryos was performed essentially as previously described [65]. Samples were viewed and images were taken using a Zeiss LSM 510 laser scanning confocal microscope or a Zeiss Axioplan 2 compound microscope. Images were processed using Adobe Photoshop.