Transcriptional regulation of cytoskeletal functions and segmentation by a novel maternal pair-rule gene, lilliputian

Development of many multicellular animals begins with the formation of an epithelium by cleavage divisions. Early development of Drosophila embryos is characterized by 13 rapid and synchronous mitotic division cycles that result in a syncytium containing 6000 nuclei at its cortical cytoplasm (Foe et al., 1993). A major developmental transition occurs during cycle 14 in which the syncytial blastoderm is converted into a blastoderm epithelium. This transition includes a variety of cellular events, such as contraction of an actin-myosin network, microtubule-driven transport of organelles, formation of cell membranes, and a marked increase in zygotic transcription (Schejter and Wieschaus, 1993a). This transitional event ultimately results in the formation of individual cells at the cortex and thus is called cellularization.

Formation of the blastoderm epithelium during cellularization is driven by a contractile actin-myosin network that forms between individual nuclei and proceeds along the leading edges of the advancing furrow membranes. The zygotically active genes nullo, serendipity alpha (Sry α), and bottleneck (bnk) are required to maintain this cytoskeletal network during cellularization (Merrill et al., 1988; Wieschaus and Sweeton, 1988). These genes encode novel membrane- associated proteins that regulate different aspects of actin dynamics and cytoskeleton organization. These molecules have been proposed to play a direct role in regulating the dynamic reorganization and interaction of the contractile actin- cytoskeleton with the newly forming plasma membranes of cellularizing embryos (Schweisguth et al., 1990; Simpson- Rose and Wieschaus, 1992; Schejter and Wieschaus, 1993b; Hunter and Wieschaus, 2000). The nullo, Sry α, and bnk genes are expressed in a similar spatiotemporal pattern, with onset of expression at nuclear cycle 11, peak expression at cycle 13/early cycle 14, and downregulation in late cycle 14 (Schweisguth et al., 1989; Simpson-Rose and Wieschaus, 1992; Schejter and Wieschaus, 1993a; Schejter and Wieschaus, 1993b). Although the factors that regulate expression of these blastoderm-specific transcripts are not known, genetic evidence suggests that the onset of expression of the Sry α and nullo genes might depend on maternal gene products rather than a unique zygotically active regulator (Merrill et al., 1988; Wieschaus and Sweeton, 1988; Simpson-Rose and Wieschaus, 1992; Ibnsouda et al., 1995).

The cytoskeletal network is also required during the syncytial blastoderm stage for the correct localization of certain segmentation gene products that provide a framework for patterning the embryo (Nüsslein-Volhard and Wieschaus, 1980; Davis and Ish-Horowicz, 1991). Specifically, the transcripts of the zygotic pair-rule genes fushi tarazu (ftz) and hairy (h) are localized exclusively to the apical periplasm by a selective vectorial export mechanism (Hafen et al., 1984; Ingham et al., 1985; Davis and Ish-Horowicz, 1991; Lall et al., 1999). Pair-rule genes have been characterized by their mutant phenotypes demonstrating their requirement in the formation of alternate segments, by their expression pattern in seven stripes, and by their requirement in morphogenetic movements. Many zygotic pair-rule gene mutants exhibit defects in the extension of the germband during gastrulation (Irvine and Wieschaus, 1994). Germband extension represents the planar elongation and narrowing of the ectodermal cell layer during gastrulation (Costa et al., 1993). During this process the germband elongates twofold and narrows to about half its original width. Cell intercalation is observed during germband extension and has been proposed to be the driving force of the extension movements (Irvine and Wieschaus, 1994). Since all known pair-rule genes encode transcription factors, they cannot be the primary regulators of the extension movements.

Here we describe a novel pair-rule gene that is essential for proper cellularization, gastrulation and segmentation during embryogenesis. Since clonal analysis in the adult eye indicated that photoreceptor cells mutant for this gene are reduced in size (Neufeld et al., 1998b), we named this gene lilliputian (lilli). Mutations in lilli have been identified in a number of independent genetic screens as dominant suppressors of transgene-dependent phenotypes in the adult Drosophila eye (Neufeld et al., 1998b; Rebay et al, 2000), and we provide evidence that this suppression reflects a requirement of lilli in the proper transcription of these transgenes. In addition, in germline clone (GLC) embryos lacking all lilli gene product, the expression of a number of zygotic regulators required for cellularization and early patterning are defective, including Sry α, ftz and huckebein (hbz). Our results suggest that lilli directs early developmental events, such as cellularization and patterning, through transcriptional activation of a specific group of zygotic regulators.

Fly cultures and crosses were carried out according to standard procedures. Two lilli mutant alleles (lilli^XS575 and lilli^XS407) were recombined onto a 2L-FRT chromosome, mitotic clones were generated as described previously (Xu and Rubin, 1993), and germline clones were generated as described previously (Chou and Perrimon, 1996). All mutant analyses were performed using these alleles unless otherwise indicated. FACS analysis and area measurements of lilli mutant clones were performed as described (Neufeld et al., 1998a; Zhang et al., 2000).

Plasmid rescue was used to isolate genomic DNA adjacent to the P element lilli^l(2)00632. The genomic fragment was used as a probe to screen a Drosophila imaginal disc cDNA library (a gift from A. Cowman) and subsequently the Drosophila LP, GH and LD libraries (Rubin et al., 2000). One full-length lilli cDNA was isolated, sequenced and cloned into pUAS (Brand and Perrimon, 1993). Transgenic lines were established using P-element-mediated transformation and used in misexpression and rescue experiments. In addition, a large number of lilli cDNA species that may represent different splicing variants have been isolated. The full-length lilli cDNA isolated and characterized in this paper has the GenBank accession number AF289034.

Third instar wandering larvae were washed with phosphate-buffered saline, and resuspended in Passive Lysis Buffer (Promega). The chloramphenicol acetyltransferase (CAT) assays were performed in the linear range of the reaction on TLC plates as described previously (Courey and Tjian, 1988). Data from three to six experimental sets were quantitated on a Fuji phosphorimager and averaged.

For live observations and video timelapse analysis, the embryos were prepared as described by Wang et al. (Wang et al., 1999). For α-Nullo and α-Bnk double immunolabeling, embryos were heatfixed as described (Müller and Wieschaus, 1996). Immunohistochemical staining of embryos was performed as described (Müller et al., 1999). In situ hybridization to embryos and imaginal discs was carried out essentially as described by Tautz and Pfeifle (Tautz and Pfeifle, 1989). Scanning electron microscopy was performed as described by Kimmel et al. (Kimmel et al., 1990). Fixation, embedding and sectioning of adult eyes were performed as described previously (Wolff and Ready, 1991). Confocal microscopy was carried out using a Leica TCS-NT confocal microscope and images were processed with Adobe Photoshop 5.5 on a Macintosh computer.

lilli and lilli-Pten homozygous mutant clones induced by FLP/FRT- mediated recombination were marked by the absence of GFP expression using a ubiquitin-GFP FRT40A chromosome (kindly provided by Christina Martin-Castellanos). Fluorescence-activated cell sorting (FACS) of dissociated wing disc cells was carried out as previously described (Neufeld et al., 1998a).

Phyllopod (Phyl) is one of the most downstream nuclear components identified in the Sevenless receptor tyrosine kinase-RAS1 signaling pathway (Chang et al., 1995; Dickson et al., 1995). Using the eye-specific expression vector pGMR, which contains a multimerized binding site for the zinc-finger protein Glass placed upstream of the basal hsp70 promoter (Hay et al., 1994), we expressed Phyl in all cells posterior to the morphogenetic furrow during larval development, and in all cells except cone cells in the pupal eye. This resulted in a rough eye phenotype that we used to screen for dominant modifiers. The lilli gene corresponds to one of the complementation groups that strongly suppress the rough eye phenotype of GMR-phyl (Fig. 1A,B). Complementation analyses revealed that many lilli alleles have been identified as suppressors in a number of different GMR-based dominant modifier screens. For example, lilli alleles were isolated in a GMR-sina screen (also called SS2-1; Neufeld et al., 1998b) and in a GMR-Yan^ACT screen (also called SY2-1; Rebay et al., 2000). Mutations in lilli suppress the rough eye phenotypes generated by overexpression of either positive (Sina and Phyl) or negative (Ttk and Yan) components of the RAS1 signaling pathway under GMR control, as well as other GMR constructs from different signaling pathways (Table 1). In addition, lilli mutants dominantly suppress the rough eye phenotypes of many sE transgenes (Dickson et al., 1992; Dickson et al., 1996), in which the sevenless enhancer is placed upstream of the hsp70 basal promoter (Table 1).

These observations suggested that lilli was required, either directly or indirectly, for proper transcription from the GMR and sE expression constructs. Further supporting this hypothesis, we found that the levels of CAT activity from a GMR-CAT reporter construct (O’Neill and Rubin, unpublished reagents) were decreased by approx. 40% in third instar larvae heterozygous for lilli (Fig. 1C). Similar results were obtained when one copy of glass, a known activator of GMR transcription, was removed (O’Neill et al., 1995). These results suggest that lilli acts as a transcriptional regulator for GMR transgenes.

Previously, lilli was localized at 23C1-2 on the cytogenetic map, and the P element line l(2)0632 was shown to fail to complement lilli mutant alleles (Neufeld et al., 1998b). We found that the lethality of l(2)0632 could be reverted by P excision (77% (n=128) viable excisions recovered), indicating that the lethality corresponded to the P insertion site. Genomic fragments flanking the P element were isolated by plasmid rescue and used as probes to screen various cDNA libraries to isolate a large number of alternatively spliced cDNA clones, including one species that apparently encoded a full-length cDNA (Fig. 1D). Comparison of cDNA sequence to genomic sequence revealed a large transcription unit that spans approx. 68 kb (Fig. 1D). The gene contains 13 exons and the full-length lilli transcript is 8,516 bp in length. Conceptual translation of the open reading frame (ORF) yields a protein of 1,673 amino acids (Fig. 1D).

To determine whether this gene corresponded to lilli, genomic DNA was isolated from 4 EMS- and 19 X-ray- induced alleles, as well as 23 alleles induced by imprecise excision of l(2)0632. Southern blot analyses using the lilli ORF as a probe revealed polymorphisms in the coding region in several lilli mutant alleles, including two X-ray alleles, XS575 and XS407, which contained deletions leading to truncated proteins (Fig. 1D). We rescued lilli mutants using the GAL4/UAS system (Brand and Perrimon, 1993) to express the full-length cDNA. Homozygous lilli mutants, which normally die as late embryos, survived to adulthood when they carried one copy each of hs-GAL4 and UAS-lilli without heat shock treatment. No rescue was found in flies that carried hs-GAL4 alone. Based on mutational analysis of the gene, viable excisions of the P line, and cDNA rescue, we conclude that this gene corresponds to lilli.

The predicted Lilli protein contains a high mobility group (HMG1) motif and a C-terminal domain present in the AF4 proto-oncoprotein and the human fragile X mental retardation protein FMR2 (Fig. 1E,F). Many HMG-containing proteins are DNA-binding, nonhistone nuclear proteins that interact with the minor groove of DNA (McGhee and Felsenfeld, 1980; Reeves and Nissen, 1990; Grosschedl et al., 1994). The FMR2/AF4-related gene family encodes nuclear proteins implicated in mental retardation (FMR2), cancer (AF4), and lymphocyte differentiation (LAF4) (Gu et al., 1992; Gu et al., 1996; Ma and Staudt, 1996; Gecz et al., 1996; Gecz et al., 1997; Chakrabarti and Davies, 1997; Taki et al., 1999). The presence of the HMG1 domain and FMR2 homology is consistent with the suggestion that Lilli acts as a transcription regulator.

To gain insight into its role during embryogenesis, we examined the phenotype of embryos mutant for lilli. Most embryos lacking zygotic lilli failed to hatch and subsequently died, although a small percentage hatched and died as first or second instar larvae. Cuticle from the late embryos was normal, with three thoracic and eight abdominal segments (Fig. 2A). Loss-of-function lilli mutations were found to be allelic to a lethal P-element insertion, l(2)00632, that exhibited a pair-rule- like segmentation phenotype when the maternal component of the gene was removed (Perrimon et al., 1996). Since both RNA in situ hybridization (data not shown) and the l(2)00632 germline clone phenotype (Perrimon et al., 1996) suggested that lilli transcript is maternally contributed, we used the DFS- FLP technique to produce germline clones (GLC) that result in lilli null embryos that lack both maternal and zygotic Lilli activity (Chou and Perrimon, 1996).

lilli GLC embryos (lilli^XS575 and lilli^XS407) exhibited pair-rule segmentation defects more severe than those previously reported for lilli^l(2)00632 (Fig. 2B-D; Perrimon et al., 1996). Two classes of phenotypes were observed. Approximately half of the embryos (52%; n=137) were missing odd numbered segments, with the remaining denticle belts often fused (44%; n=66) (Fig. 2B,C). The other class of lilli GLC embryos failed to secrete cuticle properly (Fig. 2D). These two phenotypic classes appear to reflect variation inherent to the lilli loss-of- function phenotype, rather than partial rescue by a paternal copy of lilli, as they were similarly observed whether wild-type or heterozygous lilli males were used. To further characterize these segmentation defects, we examined expression of the Engrailed (En) protein, which is present in 14 stripes along the anterior-posterior axis of wild-type embryos and marks the parasegment boundaries (DiNardo and O’Farrell, 1987; Fig. 2E,G). In lilli GLC embryos, the even-numbered En stripes were missing (Fig. 2F,H). Similar defects in Wingless expression were also observed (data not shown). Together, these results show that lilli is required for the establishment of odd-numbered segments in the embryo.

The activity of other known pair-rule genes is not only required for segmental patterning, but also for germband extension. Similarly, we found that germband extension is affected by loss of lilli. About 90 minutes after onset of gastrulation, germband extension in wild type reached 60% of dorsal egg length. In contrast, the germband never extended beyond 25% of dorsal egg length in lilli GLC embryos (Fig. 2I,J). These results suggest that lilli is required for the convergent extension movements during germband extension, consistent with its function as a maternally provided pair-rule gene.

Given the potential role of lilli as a transcriptional regulator, we wished to examine whether the pair-rule phenotype in lilli GLC embryos corresponded to changes in the expression of early patterning genes. We therefore analyzed the spatiotemporal pattern of mRNA expression of several of these genes: the maternal coordinator gene bicoid (bcd); the gap gene hunchback (hb); the pair-rule genes fushi tarazu (ftz), even- skipped (eve), hairy (h), and runt (run); the segment polarity genes engrailed (en) and wingless (wg) and the terminal gap genes tailless (tll) and huckebein (hkb). The expression patterns of bcd, hb, eve, h and run mRNA appeared relatively normal (Fig. 3A,B,I-L; data not shown). In contrast, levels of ftz mRNA were significantly lower in lilli GLC embryos than wild-type embryos at the end of cellularization (Fig. 3E,F). ftz expression appeared normal prior to mid-cellularization (data not shown), after which its distribution became diffuse and uniform, and it rarely accumulated in stripes (Fig. 3F). Using ftz-lacZ and hb-lacZ transgenes to determine the level of such regulation, we found that expression of the ftz-lacZ transgene was markedly reduced in lilli GLC embryos, while that of the hb-lacZ transgene remained unimpaired (Fig. 3C,D,G,H). This suggests that Lilli regulates ftz gene expression at the transcriptional level. Since both ftz and lilli are required for even-numbered En stripes and odd-numbered segment formation (DiNardo and O’Farrell, 1987), this disruption of ftz expression may account for the pair-rule phenotype observed in lilli GLC embryos.

In addition, we found that hkb mRNA was either reduced or absent at the embryonic termini, while the expression of tll mRNA was largely unaffected in lilli GLC embryos (Fig. 3M-P). hkb establishes the anterior and posterior borders of the ventral furrow during gastrulation (Reuter and Leptin, 1994), and lack of hkb expression causes the mesoderm and ventral domain to extend to the poles; this domain is marked by expression of snail (sna). Accordingly, we found a small extension of the ventral domain that expresses sna and undergoes ventral furrow formation (data not shown).

The significant percentage of embryos that failed to properly secrete cuticle (see above) suggested that lilli GLC embryos had defects in addition to the patterning defects described above. We first examined cytoskeletal architecture integrity during cellularization. In wild-type embryos, early in cellularization the distribution of actin filaments changes from an apical cortical cap to an apical internuclear position (Schejter and Wieschaus, 1993a; Fig. 4A,B). Following this initial phase of cellularization, nuclei elongate and microtubules form characteristic arrays described as perinuclear inverted baskets, while actin filaments maintain a contractile regular network of hexagonal units surrounding the microtubular arrays (Fig. 4C-E). Toward the end of cellularization, the individual units of the actin network contract and the resulting cells retain thin connections, called yolk stalks, to the center of the embryo (Schejter and Wieschaus, 1993a; Fig. 4K,L).

lilli GLC embryos exhibited specific defects in the maintenance of the actin network during cellularization. The initial phase of cellularization occurred normally (Fig. 4F,G). However, during the second phase of cellularization, specific defects in the maintenance of the contractile actin network were observed, as the actin network began to contract and the furrow tips moved basally. The actin filaments became unevenly distributed between nuclei, ranging from abnormally large bundles to regions where the actin network was thin or absent (Fig. 4H-J), resulting in multinucleated cells (Fig. 4H,J). The microtubular baskets surrounding each nucleus appeared largely normal, even in regions where actin filaments were unevenly distributed (Fig. 4I). At the end of cellularization, the yolk stalks were irregular in shape and size, and large connections between cortical cells and the central yolk cell were frequently seen (Fig. 4M,N). Despite these defects, video-timelapse analysis revealed that the timing of cellularization and membrane formation was unimpaired in lilli GLC embryos (data not shown) and did result in an epithelial monolayer of cells with proper apical-basal polarity (Fig. 4O,P).

In addition to the failure in maintaining the actin network, lilli GLC embryos exhibited defects in transport of organelles during cellularization (Fig. 4Q). In wild-type embryos, lipid droplets move along microtubules in a bi-directional fashion and accumulate basally during cycle 14, near the plus ends of microtubules (Welte et al., 1998; Gross et al., 2000). As the cortical cytoplasm becomes depleted of lipid droplets, it appears transparent (Welte et al., 1998; Fig. 4Q). In lilli GLC embryos, this cortical clearing is perturbed, resulting in a ‘halo’ of non-cleared cytoplasm around the central yolk (Fig. 4Q). Living lilli GLC embryos were found to have abnormal distribution of lipid droplets during cellularization (Fig. 4Q) and about 80% (n=56) of the embryos failed to separate from the central yolk sac shortly after cellularization. To determine whether this failure to clear was caused by a general breakdown of cytoplasmic transport, we examined the transport of yolk vesicles and the integrity of the microtubule network. The distribution of yolk vesicles can be observed in fixed embryos following extraction of neutral lipid from the lipid droplets. In lilli GLC embryos, yolk vesicle movement was normal during cellularization (Fig. 5B), and the general distribution of microtubular arrays was largely unaffected (Fig. 4I). Thus, lilli does not induce general breakdown of microtubule-based transport, but rather is required specifically for the microtubule-based basal transport of lipid droplets.

Pan-genomic zygotic screens for genes that are required for proper function of the actin network during cellularization have identified three genes, nullo, Sry α and bnk (Merrill et al., 1988; Wieschaus and Sweeton, 1988). The cellularization phenotypes of lilli GLC embryos are similar to those observed for mutations in the blastoderm-specific genes nullo and Sry α (Schweisguth et al., 1990; Simpson and Wieschaus, 1990; Simpson-Rose and Wieschaus, 1992). In contrast, mutations in bnk disrupt the timing of microfilament rearrangement during cellularization (Schejter and Wieschaus, 1993b). We used antibodies against the Nullo and Bnk proteins to examine their distribution in lilli GLC embryos. In wild-type embryos, Nullo and Bnk proteins colocalize with filamentous actin at the leading edge of the invaginating furrows at mid-cellularization (Fig. 5E,F,I,J). In lilli GLC embryos, Nullo and Bnk were expressed and localized normally, although the vesicular Nullo staining in the basal periplasm was somewhat less pronounced (Fig. 5G,H,K,L). In grazing sections, the alterations observed in Nullo and Bnk distribution likely reflected the disruptions of the actin network (Fig. 5G,K). Thus, we conclude that the cellularization defects in lilli GLC embryos cannot be attributed to lack of Nullo or Bnk expression.

Sry α mRNA is normally expressed at low and uniform levels at cycle 13, and is then concentrated in two broad bands prior to its down regulation late in cycle 14 (Fig. 5A; Ibnsouda et al., 1993). Interestingly, we were unable to detect expression of the Sry α gene during cellularization in lilli GLC embryos (Fig. 5B). Expression of a Sry-lacZ transgene was likewise abolished in lilli GLC embryos (Fig. 5D), indicating that the defect in Sry α expression is at the transcriptional level. Since the mutant phenotype of Sry α is very similar to that of lilli GLC embryos, it seems likely that the cellularization defects observed in lilli GLC embryos are caused, at least in part, by a strong reduction in Sry α expression.

In previous studies, we found that retinal cells lacking lilli were significantly smaller than wild-type cells (Neufeld et al., 1998b). Recently, mutations in a number of components of a PI3K-dependent signaling pathway have been shown to reduce the size of a variety of cell types in Drosophila (Böhni et al., 1999; Chen et al., 1996; Montagne et al., 1999; Verdu et al., 1999; Weinkove et al., 1999; Zhang et al., 2000). The reduced cell size of these mutants reflects an underlying reduction in the rate of cellular growth, and thus inhibition of this pathway can lead to reductions in the size of mutant clones, organs, or animals (Conlon and Raff, 1999; Lehner, 1999; Edgar, 1999; Weinkove and Leevers, 2000). In contrast, despite the small size of individual lilli mutant adult cells, we found that we could generate relatively large clones of lilli mutant cells (Figs 6A and 7A), suggesting that lilli may affect cell size without changing the overall rate of growth. This was tested by comparing the size of individual lilli mutant clones to their wild-type twinspots in third instar eye and wing imaginal discs. We found that lilli clones induced either at 24-36 hours or 36- 48 hours after egg deposition (AED) were indistinguishable in size and number of cells from their wild-type twinspots. Of 75 individual clones examined at 96 hours after induction, the average area of lilli mutant clones was 1440 pixels, compared to 1400 pixels for corresponding wild-type twinspots (Fig. 6B). Thus, lilli mutant cells grew at 1.03 (±0.57) times the rate of wild-type cells, indicating that lilli is not required for normal rates of cell growth.

Interestingly, despite an approximately 50% reduction in the size of lilli photoreceptor cells and wing margin bristles in the adult (Fig. 7A,B), other cell types in the adult eye and wing were unaffected. For example, the surface of eyes containing lilli mutant clones appeared normal by SEM analysis (data not shown), suggesting that loss of lilli did not reduce the size of cone cells. The size of lilli cells in developing wing and eye imaginal discs appeared normal as well (Fig. 7C and data not shown). This was confirmed by FACS analysis of dissociated wing discs, which revealed no significant difference in size between lilli and wild-type cells (Fig. 7D). In addition, unlike mutations in components of the PI3K pathway, lilli mutant cells displayed a normal cell cycle profile (Fig. 7E). To test whether lilli is required for PI3K-mediated growth, we examined cells doubly mutant for lilli and Pten, an inhibitor of this pathway. Mutations in Pten increase cell size and advance G₁/S progression (Goberdhan et al., 1999; Huang et al., 1999; Gao et al., 2000); these effects are prevented by mutations in downstream components such as torso (tor/dTOR; Zhang et al., 2000). In contrast, loss of lilli did not prevent the cell enlargement or cell cycle changes caused by Pten mutation (Fig. 7F-7I and data not shown), indicating that lilli is not an essential element of the PI3K pathway. Interestingly, we noticed that ommatidia containing the enlarged lilli Pten photoreceptor cells were severely disorganized, and contained malformed rhabdomeres characteristic of cytoskeletal defects (Fig. 7H; Fischer-Vize and Mosley, 1994; Fan and Ready, 1997). Together, these results indicate that lilli affects the cell size through a growth-independent and PI3K-independent mechanism. We suggest that mutations in lilli may affect final cell size by disrupting the morphological changes that cells such as rhabdomeres and bristles, which are the specializations of photoreceptor and trochogen cells respectively, undergo during pupal development.

Four lines of evidence support the idea that lilli functions during embryogenesis as a maternally contributed pair-rule gene: lilli GLC embryos fail to establish even-number En stripes; they subsequently lack odd-number segments; they exhibit defects in germband extension, and these phenotypes are not seen in homozygous lilli embryos lacking zygotic Lilli function. Thus, our results demonstrate that Lilli is required maternally for normal segmentation during embryogenesis.

What is the mode of action of lilli in segmentation? Unlike zygotic pair-rule genes that are expressed in the trunk region of the embryo in seven stripes (St Johnson and Nüsslein- Volhard, 1992; Ingham and Martinez Arias, 1992), the expression pattern of lilli transcript is not segmental. Thus, lilli may play a role in the expression, localization or activity of other pair-rule genes. lilli probably does not act through the major gap genes knirps (kni), Krüppel (Kr), or giant (gt), because the striped pattern of the primary pair-rule genes eve, h and run are unaffected in lilli GLC embryos (Nüsslein- Volhard and Wieschaus, 1980; St. Johnson and Nüsslein- Volhard, 1992; Ingham and Martinez Arias, 1992).

One function of lilli is to regulate ftz transcription, since expression of both endogenous ftz mRNA and a ftz-lacZ transgene are markedly reduced in lilli GLC embryos. Although a low level of ftz mRNA remains in lilli GLC embryos, these transcripts never resolve into seven stripes. It is possible that lilli directly regulates ftz transcription in combination with other transcription and regulatory factors, or that lilli is required for the function or expression of such molecules. A candidate for such a factor is the product of the ftz-f1 gene, which was shown to be required maternally for ftz expression (Guichet et al., 1997; Yu et al., 1997).

While the developmental defects observed in lilli GLC embryos are largely consistent with ftz being the primary target of lilli, ftz mutant embryos do not exhibit defects in germband extension movements as strong as those observed in lilli GLC embryos (Irvine and Wieschaus, 1994). The only other known gene that is necessary for germband extension is that of the Drosophila 5HT₂-serotonin receptor (5HT₂-SR) (Colas et al., 1999). The function of 5HT₂-SR during germband extension is unclear, but since its segmental expression is indispensable for cell intercalation movements in the germband it may provide another possible target for lilli function. Further analysis of lilli targets is likely to result in the identification of genes that govern germband extension movements during gastrulation.

Genetic studies have indicated that cellularization specifically requires a relatively small number of zygotically active genes, including Sry α, nullo and bnk (Merrill et al., 1988; Wieschaus and Sweeton, 1988). Removal of the maternal contribution of lilli phenocopies the cellularization defects observed in zygotic mutation in either Sry α or nullo (Simpson and Wieschaus, 1990, Schweisguth et al., 1990). lilli thus represents the first identified gene that is required maternally for specific cytoskeletal functions at the onset of zygotic gene activity at mitotic cycle 14. Our results suggest that Sry α is another primary target of lilli function. First, in both lilli and Sry α mutant embryos Nullo protein distribution is normal, except where the actin network is disrupted (Postner and Wieschaus, 1994). Second, lilli regulates Sry expression at the transcription level, since both Sry mRNA expression and Sry-lacZ transgene expression are absent in the blastoderm in lilli GLC embryos, suggesting that this may be the primary cause for the cellularization defects observed in lilli GLC embryos.

Vesicle transport is a tightly regulated process during early development and depends on maternal and zygotic gene function (Welte et al., 1998). The basal movement of the lipid droplets during mitotic cycle 14 is mediated by a microtubule-dependent transport mechanism (Welte et al., 1998; Gross et al., 2000). Because lipid droplet transport is disrupted in lilli GLC embryos, lilli is also required for at least one microtubule-dependent transport process. Mutations in only one other locus, halo, cause a specific defect in cortical clearing similar to that of lilli. Since microtubule function and structure appears otherwise normal in lilli GLC embryos, lilli may be required for the expression and function of Halo, or other unidentified components of lipid droplet transport machinery. The molecular nature of Halo is not known and mutations are not available to examine potential genetic interaction with lilli.

One of the most striking aspects of the lilli mutant phenotype is the marked reduction in size of lilli mutant cells. Recently, a number of components of the PI3K-mediated signaling pathway have been shown to regulate cell size in Drosophila. Loss-of-function mutations in Insulin-like receptor (Chen et al., 1996), the insulin receptor substrate homolog chico (Böhni et al., 1999), PI3K subunits Dp110 and Dp60 (Weinkove et al., 1999), Dakt1 (Akt1; Verdu et al., 1999), tor/dTOR (Zhang et al., 2000), and dS6K (S6k; Montagne et al., 1999) each cause cell size changes similar to those which we observe in lilli mutant cells. This apparent phenotypic similarity is consistent with lilli regulating cell growth by modulating PI3K signaling, or by acting as a target of the PI3K pathway. However, lilli differs from these and other growth regulators in a number of significant ways. First, the reduction in cell size caused by loss of PI3K signaling is associated with an overall reduction in growth and proliferation, such that clones of mutant cells grow more slowly than their wild-type twinspots. In contrast, we found that lilli mutant cells proliferated to the same extent as wild-type cells, and could give rise to large clones encompassing a significant fraction of the adult eye. Second, whereas mutations in components of the PI3K signaling pathway reduce the size of both adult and imaginal disc cells, loss of lilli did not affect cell size in imaginal discs. Third, overexpression of Dp110, Dakt1 or dS6K leads to marked increases in cell and organ size, whereas overexpression of lilli actually decreased the size of retinal cells and bristles (data not shown). Together, these results argue that the reduced size of lilli mutant cells is not the result of defective growth, and that lilli does not function to regulate cell growth.

How then might lilli affect cell size? One plausible theory is suggested by the observation that lilli mutations have a disproportionate effect on cells with specialized structures such as rhabdomeres and bristles, while epithelial cells are largely unaffected. Rather than regulating cell growth, lilli may be required to organize such cellular structures. Given the extensive cytoskeletal rearrangements necessary to elaborate a rhabdomere or bristle, such a role would fit with the observed cytoskeletal defects in lilli GLC embryos. The phenotype of cells doubly mutant for lilli and dPTEN also supports this view: lilli was not required for the cellular hypertrophy caused by mutations in dPTEN, but the cytoarchitecture of the enlarged double mutant cells was disrupted. Thus, some of lilli’s functions may only be evident in cells in which great demands are placed on the cytoskeleton. Alternatively, lilli may be part of a mechanism that governs postmitotic cell size independently of growth signals.

Our genetic studies demonstrate that mutations in lilli disrupt the expression pattern of a subset of early patterning zygotic genes, including ftz, hkb, and Sry α. In addition, lilli mutants dominantly suppress expression from GMR and sE transgenes, both of which share a common promoter element. As most genes examined had largely normal expression in lilli GLC embryos, it is unlikely that Lilli is generally required for efficient transcription. A more likely explanation is that lilli specifically regulates the transcription of a limited subset of genes including ftz, hkb, and Sry α.

While both hkb and tll encode transcription factors that regulate terminal structure differentiation and are induced by the receptor tyrosine kinase Torso (Tor; reviewed in Lu et al., 1993; Duffy and Perrimon, 1994), hkb but not tll expression is strongly reduced in lilli GLC embryos. One difference between the two transcription factors is that low levels of Tor signaling induce tll expression, whereas higher levels are required for expression of hkb (Greenwood and Struhl, 1997). Thus one possibility is that lilli is required for translating the quantitative differences of Torso/Ras signaling into distinct expression patterns of these two transcription factors.

Based on our data, we propose that Lilli acts as a cofactor for zygotically expressed transcription factor(s). In such a model, Lilli would provide specificity to a more general zygotic factor, while this factor itself would provide the correct timing of transcriptional regulation. It is striking that genes like bnk, Sry α, and nullo, which have remarkably similar expression patterns and embryonic functions, exhibit different requirements for lilli action. While we have identified three potential targets of lilli function, there are defects in lilli GLC embryos that cannot be explained by misregulation of any known genes. The identification of new lilli targets will help to identify key regulatory genes involved in processes such as lipid droplet transport and germband extension movements, and may also provide functional cues to the family of the FMR2/AF4-like proteins.

Correspondence should be addressed to A.H.T. and H.A.M. We thank E. Hafen, F. Wittwer, and S. Newfeld for communication prior to publication. We thank C. Goodman, M. Frasch, S. Roth, E. Scheijter, and E. Wieschaus for antibodies. We thank Z. Zhou, H. Zhang, M. Brodsky, E. Schejter, A. Stathopoulos and A. Vincent for cDNA and RNA probes. The authors thank S. Mullaney for assistance with P-element mediated transformation, M. Evans-Holm and G. Tsang for assistance with DNA sequencing, S. Ruzin and A. Grimm for expert technical assistance with Zeiss Confocal microscopy. We are grateful to E. Wieschaus, M. Levine, K. M. C. Sullivan, T. Serano, B. A. Hay, M. Welte, T. Blankenship, E. Frise, A. Stathopoulos and A. Bailey for critical reading of this manuscript, and our colleagues, K. M. C. Sullivan, T. Serano, B. A. Hay, M. Therrien, F. D. Karim and D. Wasserman for helpful discussions. H. A. M. thanks E. Knust, E. Schejter, A. Vincent, M. Welte and A. Wodarz for discussions. A. H. T. is supported by a senior postdoctoral fellowship from the Leukemia and Lymphoma Society. H. A. M. is supported by the Deutsche Forschungsgemeinschaft.