Slow as Molasses is required for polarized membrane growth and germ cell migration in Drosophila

Cell migration is an essential element of metastasis, inflammation, infection and embryonic development. The study of germ cell migration provides an excellent opportunity to gain insights into these processes, as germ cell migration shares with them key features such as invasion through an epithelium, cell-cell adhesion and directed migration to a target tissue. Germ cell migration is a well-conserved process in species such as human, mouse, fish and fly. Drosophila is an ideal animal in which to study the genetics of germ cell migration because it is easy to visualize migrating germ cells and to identify genes that control their migration.

Drosophila germ cells are the first cells to form in the embryo. During the first three hours of embryogenesis, nuclei divide synchronously without cytokinesis to form a syncytium. During the eighth and ninth cell cycle, nuclei migrate to the periphery of the embryo. Those nuclei that reach the posterior pole are surrounded by cell membrane as they bud off of the single-celled embryo and become primordial germ cells (for a review, see Williamson and Lehmann, 1996). The somatic cells of the embryo form later, when an apical to basal cell membrane growth encapsulates each nucleus. Owing to the two modes of cell formation in the early embryo, two cell layers are formed at the posterior pole: the germ cells, which form by budding at nuclear cycle 10, and the posterior somatic blastoderm cells, which form by polarized cellularization at nuclear cycle 14 (for reviews, see Müller, 2000; Schejter et al., 1992). The posterior somatic blastoderm cells give rise to the posterior midgut. The germ cells adhere to the posterior midgut primordium as it moves dorsally and involutes during gastrulation to form a pocket inside the embryo. By the end of gastrulation, the primordial germ cells are therefore located inside the pocket of the posterior midgut primordium. Subsequently, germ cells migrate through the epithelium of the posterior midgut, crawl dorsally along the midgut, and move off the midgut, into the mesoderm, to contact somatic gonadal precursor cells (SGPs), which form in parasegments 10-12 (Fig. 1A,B) (Boyle and DiNardo, 1995). Germ cells maintain contact with SGPs as they coalesce and form the embryonic gonad at stage 14 (Fig. 1C) (for a review, see Starz-Gaiano and Lehmann, 2001).

Several molecules have been identified that are known to guide germ cells. wunen and wunen 2 encode phospholipid phosphatases and are expressed in the posterior midgut. Wunen phosphatase activity was shown to repel germ cells as they move on the midgut epithelium (Starz-Gaiano et al., 2001; Zhang et al., 1997). HMG-CoA reductase (Hmgcr) is highly expressed within the developing SGPs and is required to attract germ cells into the gonadal mesoderm. In Hmgcr mutants, most germ cells remain on the surface of the midgut and fail to enter the mesoderm. Hmgcr has also been shown to be sufficient to attract germ cells at a distance. When Hmgcr is misexpressed in the epidermis or in the nervous system, germ cells are attracted towards the areas of high HMG-CoA reductase (Van Doren et al., 1998b). Although Hmgcr mutants have consistent and fully penetrant germ cell migration defects, some germ cells do enter the mesoderm, associate with SGPs and migrate correctly to the gonad. This suggests that additional germ cell guidance factors direct germ cells from the midgut to the mesoderm.

We report the identification of slow as molasses (slam), a new gene involved in germ cell migration. Embryos homozygous for point mutations in slam have a strong and penetrant germ cell migration defect, which resembles that of Hmgcr mutants. Our genetic analysis suggests that Slam acts independently of Hmgcr and we propose that Slam acts in the midgut and is required for germ cells to transit from the midgut to the mesoderm. In addition to the germ cell migration defect, we observed cellularization defects in mutant embryos after removal of the maternal and zygotic contribution of slam. Molecular characterization of slam reveals that slam is one of the first genes transcribed in the embryo. Consistent with a role in polarized cell growth during blastoderm formation, Slam protein localizes to the growing basal-lateral membrane during nuclear cycle 14. slam RNA and protein are no longer expressed at the time when we observe a defect in germ cell migration. Based on the mutant phenotype and the analysis of the protein distribution, we propose that Slam mediates the localization of a germ cell guidance signal to the basal side of posterior midgut cells.

waldo¹ and waldo² alleles of slam were identified in a zygotic mutagenesis screen for germ cell migration genes on the second chromosome (Broihier, 1998). The two alleles have similar germ cell migration defects. slam^waldo; Hmgcr double mutants were made with slam^waldo1 and Hmgcr^clb1 (Van Doren et al., 1998b). To test for dominant suppression of Hmgcr ectopic germ cell attraction by slam loss of function, elav-gal4, UAS-Hmgcr males were crossed to virgins with germline clones of slam^waldo; the progeny would therefore be completely lacking maternal slam and zygotically heterozygous for slam (M^–Z^–/+slam^waldo). Germline clones of slam^waldo were made by crossing slam^waldo1 FRT 40A / CyO virgins to hs flp; ovo^D FRT 40A / CyO males and heat shocking the progeny at 37°C for 2 hours on 2 successive days during first and second larval instar. To test the phenotype of M^–Z^–slam^waldo mutants, hs flp; slam^waldo1 FRT 40A / ovo^D FRT 40A virgins that contained slam^waldo1 germline clones were crossed to slam^waldo2/CyO, P(ftz-lacZ) males.

Meiotic recombination mapping placed slam between dp and b. The following P elements were used for male mitotic recombination: l(2)k10004 (25B), k13720 (26C), l(2)k07502b (26D1-2), k04917(26D6-8), k03201(26D6-9), k14206 (26F), l(2)k00605 (27A), EP2369 (27F), EP2625 (28B), l(2)rL220 (28C), 02496 (28D1-2), P 1478 (28D3-4), l(2)k11101(28D10-11), l(2)05836 (28E1-2) and l(2)k13638 (28E3-4). slam mapped between k13720 and l(2)k07502b. The mapped region included 16 candidate genes, two of which were uncovered by the deficiency Df(2L)tig^X (a gift from T. Bunch) (Bunch et al., 1998), which complements slam, leaving 14 candidate genes. In situ probes were constructed for each of the 14 candidate genes either from an EST (Research Genetics) or an amplified genomic fragment, subcloned into pCR 4Blunt-TOPO using the Zero Blunt TOPO PCR cloning kit (Invitrogen). The following probes were made from ESTs: Pez (LD 11612), CPR (LD 45615 and LD 46590), CG9498 (GH 06262), and CG9505 (GH 11680 and GH 10925). The following probes were made from subcloned genomic fragments: CG9491, CG9497, CG9499, CG9501, CG9500, CG9506, CG13981, CG9507, CG9508 and CG9511. To transcribe in situ hybridization probes from ESTs, the plasmids were linearized with EcoRI and transcribed with Sp6 polymerase for antisense RNA probes; sense RNA control probes were transcribed with T7 polymerase after linearization with XhoI. For genomic fragments, RNA probes were transcribed with T3 polymerase after linearization of the subcloned fragments with NotI. T7 polymerase was used to transcribe probes after linearization with PmeI or SpeI. Transcription reactions were carried out with the digoxigenin RNA labeling kit (Roche).

Pez, CPR, and CG9506 were sequenced by PCR amplification of genomic DNA from both slam mutant alleles. Genomic DNA from mutant progeny of slam^waldo1 / CyO P(Kr-GFP) or slam^waldo2 / CyO P(Kr-GFP) was isolated after selecting homozygous mutant embryos (those that did not contain GFP). The PCR products were gel-purified with the Qiaquick kit (Qiagen), sequenced at the Rockefeller University DNA sequencing lab and analyzed with the SeqManII program.

Rabbit α-Slam was raised against the peptide ETPKDDLQTGEFDFAKP in the second exon. The following antibodies were used for immunostaining of embryos: rabbit α-Vasa (1/2500), rabbit α-β-galactosidase (Cappel, 1/20,000), rabbit α-Slam (1/3000 – 1/5000), mouse α-Neurotactin (BP106 Hybridoma bank, 1/50) and rabbit α-Myosin (1/750, gift from C. Field). Antibody detection was performed using a biotinylated secondary antibody (Jackson ImmunoResearch) and the ABC Elite Kit (Vector Labs), or with a directly conjugated Alexa 488 (Molecular Probes), Cy3 or Cy5 secondary antibody (Jackson ImmunoResearch). For staining with α-Vasa and α-β-galactosidase, embryos were fixed and devitellinized according to the method described by Gavis and Lehmann (Gavis and Lehmann, 1992), with the modification that 1×PBS was used in place of PEMS during the fixation. For staining with α-Slam, α-Neurotactin and α-Myosin, embryos were heat fixed by immersion for 5 seconds into boiling 0.03% Triton-X, 68 mM NaCl, followed by devitellinization in methanol. Embryos were rehydrated and subjected to antibody staining as described by Eldon and Pirotta (Eldon and Pirotta, 1991). Fluorescently labeled embryos were mounted in Aqua Poly/Mount (Polysciences) and analyzed with a Leica TCS/NT confocal microscope. For whole-mount analysis of immunohistochemically labeled embryos, mountings were made in PolyBed812 (Polysciences) according to Ephrussi et al. (Ephrussi et al., 1991) and analyzed with a Zeiss Axiophot using Nomarski optics. Section analysis was performed as described by Broihier et al. (Broihier et al., 1998). Sections were 2 μm. For double labeling of embryos with an antibody and in situ hybridization probe, embryos were first antibody stained as described above (except washes were made in 1×PBS, 0.1% Tween-20, 50 μg/ml heparin and 250 μg/ml tRNA), and then subjected to in situ hybridization. In situ hybridization was performed as described by Lehmann and Tautz (Lehmann and Tautz, 1994). Embryos were incubated with DIG-labeled RNA probes at 55°C overnight. Probe hybridization was visualized with an alkaline phosphatase-conjugated anti-DIG antibody, followed by treatment with NBT and BCIP.

We identified two alleles of slam in an EMS mutagenesis screen for zygotically acting genes that affect germ cell migration (Broihier, 1998). We refer to these two mutants as the waldo alleles of slam, or slam^waldo alleles, to indicate their origin (see Materials and Methods). Both alleles have similar germ cell migration phenotypes. In slam^waldo mutant embryos, germ cells migrate correctly through the midgut epithelium and along the midgut towards the dorsal side of the embryo (Fig. 1D), but fail to enter the mesoderm to contact the SGPs (Fig. 1E). Entering the mesoderm at this stage is crucial for germ cells to become incorporated into the embryonic gonad later. Thus, slam^waldo mutant germ cells that fail to contact the SGPs at stage 11 will be located outside the gonads at stage 14 (Fig. 1F).

Detailed comparison between wild-type and mutant development suggests that germ cells in slam^waldo mutants are delayed in their movement from the gut to the mesoderm during stage 11 (compare Fig. 1A,B with 1D,E). As a result, some germ cells remain on the midgut throughout embryogenesis and still are found on the midgut at stage 14 (Fig. 1F, arrowheads). Other germ cells move off the midgut but seemingly too late to reach the most posterior cluster of SGPs. Because the germ band has begun to retract, delayed germ cells move into a more posterior region of the mesoderm, which lacks SGPs. These germ cells are found outside and posterior to the embryonic gonad at stage 14 (Fig. 1F, arrows). The remaining germ cells are found associated with SGPs in the embryonic gonad at the end of embryogenesis. On average, 50% of germ cells are lost per embryo in slam^waldo1 and slam^waldo2, compared with fewer than 10% in the control (slam/+) embryos.

The germ cell migration phenotype in slam embryos could be due to a germ cell autonomous defect; however, Slam RNA and protein are not expressed in the germ cells (see Fig. 4D, inset). Alternatively, the phenotype could be due to a defect in either the specification or differentiation of the somatic tissues that contact germ cells during their migration, or more directly in the production or localization of a guidance cue. As the migration defect is first observed when the germ cells leave the midgut, we analyzed the development of the midgut and the gonadal mesoderm in mutant embryos. In situ hybridization with RACE, an enzyme expressed in the posterior midgut primordium, reveals a delay in the transition from an epithelium to a mesenchyme in mutant embryos. At stage 11, when in wild-type embryos the midgut primordium has already started to flatten out and extend along the mesoderm, the midgut remains more compact in the mutant. The length of the posterior midgut is visibly shorter in the mutant (Fig. 1D). This defect is transient, as later midgut markers, such as dpp-lacZ, reveal that a continuous digestive tract is formed; however, in some embryos the second midgut constriction fails to form (Broihier, 1998). FasIII staining reveals that the visceral mesoderm forms normally and surrounds the midgut (Broihier, 1998). Most mutant embryos become crawling larvae and are capable of passing colored yeast normally through their digestive tracts (data not shown). Even if a mild midgut defect exists, this cannot explain the germ cell migration defect of slam^waldo mutants. The midguts of embryos that lack maternal and zygotic integrin function also fail to undergo an epithelial to mesenchymal change at stage 11 (Martin-Bermudo et al., 1999) and appear morphologically very similar to those of slam^waldo, but the integrin mutants have normal germ cell migration (data not shown). This implies that timing of the epithelial to mesenchymal transition in the posterior midgut is not crucial for germ cells to move from the midgut into the mesoderm, and suggests that the slam^waldo mutant germ cell migration defect is not simply a result of the altered midgut morphology.

For the analysis of the mesoderm, we used two markers: RNA expression of the retrotransposon 412 and antibody staining for Zfh-1 protein, which mark the lateral mesoderm at stage 10-11 as well as the gonadal mesoderm, later. Both markers are expressed in slam^waldo mutants, and the number and distribution of lateral mesoderm cells in mutant embryos seems similar to that of wild type during stages 10-11 (Fig. 1B,C,E,F). However, later when gonadal mesoderm cells move towards each other to align and eventually coalesce to form the embryonic gonad, some mutant embryos show a reduction in the number of SGPs. In abdA mutants, lateral mesoderm forms correctly, but no SGPs are specified. Germ cells in these mutants still migrate off the midgut into the lateral mesoderm, but are lost later in embryogenesis (Moore et al., 1998). This suggests that lateral mesoderm, not later SGP formation, is necessary for germ cells to move off the midgut into the mesoderm. Therefore the reduction in SGP number in slam^waldo mutants is not responsible for the slam^waldo germ cell migration defect. Taken together, these data suggest that the defect in slam^waldo mutants is more likely to be due to a defect in midgut signaling or guidance than to a failure in gonadal mesoderm specification or midgut development.

The phenotype of slam^waldo mutant embryos is strikingly similar to that of Hmgcr mutants. In both mutants, germ cells fail to move off the midgut and do not associate with SGPs in the mesoderm at stage 11-12. We also see some germ cells correctly migrating in both single mutants, even in null alleles of Hmgcr (Fig. 2). An instructive role as a germ cell attractant was demonstrated for Hmgcr by the finding that mis-expression of Hmgcr leads to attraction of germ cells to the ectopic site independent of SGP differentiation (Van Doren et al., 1998b).

To analyze a possible interaction between the two genes, we analyzed Hmgcr expression and function in slam mutant embryos. Hmgcr RNA is properly expressed in slam^waldo mutants, indicating that slam is not required upstream of Hmgcr for its RNA expression (data not shown). To test whether slam acts downstream of Hmgcr, we reduced the levels of Slam activity after ectopic expression of Hmgcr. We found that ectopic Hmgcr expression is still capable of attracting germ cells in a slam^waldo heterozygous mutant background (see Materials and Methods), suggesting that Slam is not required for Hmgcr-mediated germ cell attraction to the mesoderm. Together, these data make it less likely that Slam and Hmgcr act within the same pathway and favor the hypothesis that Hmgcr and slam act independently and provide separate guidance cues. This conclusion is further supported by the analysis of slam^waldo; Hmgcr double mutants. The germ cell migration phenotype in double mutant embryos was stronger than either single mutant, as no germ cells move off the midgut. However, the specificity of this defect is unclear as the double mutant embryos were poorly differentiated, a phenotype that is not found in collections from either single mutant (compare Fig. 2D with 2B and 2C). The fact that the double mutant has a novel phenotype suggests that, while the activity of Hmgcr and slam may not directly rely on each other’s function, the two genes may act in parallel and may regulate common downstream pathways.

There are several explanations for why not all germ cells in slam^waldo mutants are lost. One is that the function of slam is partially redundant with other germ cell guidance genes, such as Hmgcr (see above). Another possibility is that slam^waldo mutants are not complete null alleles and/or that maternal contribution of slam partially rescues the germ cell migration phenotype. To address the second hypothesis, we generated germline clones homozygous mutant for slam^waldo using the flp-FRT-ovo^D system. Removal of maternal and zygotic slam (M^–Z^–slam^waldo) revealed a new and unexpected phenotype: M^–Z^–slam^waldo mutant embryos fail to cellularize during nuclear cycle 14 (Fig. 3). Interestingly, germ cells form normally during cycle 10. The germ cells have no soma to move through, so germ cells are found distributed throughout an otherwise unstructured embryo.

In wild type, nuclei migrate to the periphery of the embryo during the ninth nuclear cycle. During each of the following four nuclear divisions, these nuclei and their associated centrioles induce structural changes in the actin cytoskeleton (actin cap) and the cortical membrane. After the 14th nuclear division, the membrane grows and invaginates between each peripheral nucleus to create an epithelium of somatic cells. This membrane invagination proceeds first slowly for 35-40 minutes (slow phase) and then rapidly for 15-20 minutes (fast phase) until each nucleus is surrounded by a cell membrane (Müller, 2000; Schejter et al., 1992).

Nuclei reach the periphery normally in M^–Z^–slam^waldo mutants, the primordial germ cells bud normally, and the somatic nuclei continue to divide until cycle 14. The phases of cellularization during cycle 14 were followed by live observation and by observing the progression of the cellularization front (furrow canal) using an anti-Myosin antibody in wild-type and M^–Z^–slam^waldo mutants (Fig. 3; data not shown). The first deviation from wild type is seen during the slow phase of cellularization. As the embryo proceeds through cycle 14, the nuclei elongate and the membrane invaginates. However, compared with wild type, membrane invagination in slam mutants is delayed. At the time when membranes normally enclose each nucleus basally, the incompletely invaginated membranes of M^–Z^–slam^waldo mutants entrap the nuclei as they pinch off basally (Fig. 3H). In wild type, after cellularization is complete, the complex movements of gastrulation lead to the invagination of the midgut primordia and the mesoderm and to the extension of the germ band. M^–Z^–slam^waldo mutants do attempt to gastrulate, but the vigorous movements of gastrulation disrupt the incompletely formed somatic cells; the embryos look like they fall apart and fail to develop further.

This cellularization phenotype is only seen in M^–/–Z^–/–slam mutants (Fig. 3I). A single zygotic copy of slam (M^–/–Z^+/–slam) restores both cellularization and germ cell migration (Fig. 3J,K), and a single maternal copy (M^+/–Z^–/–slam) rescues the cellularization defect, but not the germ cell migration defect (Fig. 1; data not shown). In an independent study, Lecuit and Wieschaus identified slam as the zygotic cellularization gene in the 26BF genomic interval (Lecuit et al., 2002). Their and our data suggest that the cellularization defect is the null phenotype because it is the phenotype of embryos that lack maternal and zygotic slam^waldo alleles (this study), embryos deficient for the genomic region containing slam and embryos injected with slam RNAi (Lecuit et al., 2002). Our slam^waldo mutants are presumably hypomorphic alleles, such that maternally provided slam activity is sufficient for cellularization and thereby reveal a second function for Slam in germ cell migration.

To identify the gene responsible for the slam^waldo mutant phenotypes, we used male mitotic recombination (Chen et al., 1998) to map slam to a 100 kb genomic region between two P elements (k13720 at 26 C2-3 and l(2)k07502 at 26 D1-2). The Berkeley Drosophila Genome Project had identified the position of these P elements in the genome and predicted 16 genes in this interval (Fig. 4A). Two of these genes, tiggrin and CG13982, were removed in Df(2L) tig^X, a deficiency strain that complemented the slam^waldo mutants; this eliminated them as candidates. An important clue for the molecular identification of the slam gene came from the observation that the M^–Z^–slam^waldo cellularization phenotype was rescued zygotically. As expression of the zygotic genome is barely initiated during the syncytial blastoderm stage, we reasoned that the slam gene had to be one of the earliest zygotically transcribed genes and that its expression pattern may resemble that of other zygotic genes required for cellularization such as nullo, serendipity α and bottleneck (Schejter et al., 1992). With this information in mind, we determined the RNA expression pattern of the remaining 14 candidate genes by in situ hybridization and found three candidates that were expressed during blastoderm formation (Pez, CPR and CG9506, Fig. 4, data not shown). The predicted protein coding regions of these three genes were sequenced from DNA isolated from homozygous mutant embryos for each of the slam^waldo alleles. We found single base pair changes for both alleles in CG9506: Q722stop (slam^waldo1) and G879K (slam^waldo2) (Fig. 4B). Based on the correlation between the cellularization phenotype and the expression pattern, and the fact that the mutant alleles carry a nonsense and a missense mutation in the gene, respectively, we conclude that CG9506 is slam.

CG9506/slam encodes a protein of 1173 amino acids (Fig. 4B). Despite application of a large number of sequence analysis programs [BLAST (Altschul et al., 1997), PROSITE (Hofmann et al., 1999), PFAM (Bateman et al., 2002) and SMART (Schultz et al., 2000)], we were unable to find any homology between Slam and any proteins or protein motifs in the existing databases, except for a short 30 amino acid coiled-coil region at residues 493-523, predicted by PFAM and SMART. Thus, at this point, the two slam alleles provide the only evidence regarding regions of the protein important for Slam function. The slam^waldo1 allele is predicted to produce a truncated protein, while the slam^waldo2 allele has a single amino acid glycine to lysine change in the region truncated by slam^waldo1 (Fig. 4B).

Detailed analysis of the RNA expression pattern during embryogenesis revealed low levels of uniformly distributed slam RNA, which is most likely of maternal origin (Fig. 4C). During nuclear cycle 10 we first detect an increase in slam RNA, presumably because of new zygotic transcription. During nuclear cycle 14 as cellularization begins, slam RNA is highly expressed and this expression is maintained throughout cellularization (Fig. 4D). slam RNA is localized basally within the cytoplasm surrounding each nucleus (data not shown). Slam expression is due to transcription from the embryonic genome as it is only observed in somatic nuclei but not in the germ cells, which are transcriptionally repressed during early embryogenesis (Seydoux and Dunn, 1997; Van Doren et al., 1998a; Zalokar, 1976) (Fig. 4D and insert). At the end of cellularization, slam RNA is limited to three stripes (Fig. 4E); the RNA levels then rapidly decline and slam RNA can no longer be detected after gastrulation (stage 8 onwards) (Fig. 4F).

To analyze the time course and distribution of Slam protein, we raised an antibody against a peptide of the Slam protein. We used a peptide competition experiment to test the specificity of the Slam antibody. Pre-incubation of the Slam antibody with the Slam peptide blocks Slam staining in embryos (Fig. 5F). Pre-incubation with a different peptide from an unrelated protein (Nanos) (Fig. 5G) did not block Slam staining. This result shows that anti-Slam specifically recognizes the Slam protein.

Slam protein first can be detected at the plasma membrane of embryos at nuclear cycle 11, before cellularization begins, but after nuclei have migrated to the cortex (Fig. 5A). At this stage, we observe Slam also in the cytoplasm basal to the peripheral nuclei; the position of this staining suggests that we are detecting Slam in the Golgi (Fig. 5A, arrowhead) (Sisson et al., 2000). During cellularization, Slam is localized to the most advancing membrane as it invaginates between each nucleus (Fig. 5A-C). Confocal microscopy of embryos doubly labeled for Slam protein and Neurotactin, which outlines the cell membranes, reveals that Slam is localized to the furrow canal, which marks the most basal part of the growing membrane (Fig. 6A-D,G). The germ cells do not contain any Slam protein (Fig. 6E,F).

Once the embryo begins to gastrulate, we no longer see Slam protein at the membrane of invaginating cells; however, it is maintained in the cell membranes of the rest of the epithelium (Fig. 5D, Fig. 6D,H white arrow versus black arrow). We analyzed Slam staining in thin sections, where localization to the basal-lateral membrane is clearly evident. Slam protein seems to be associated with the membrane. In thin sections of gastrulating embryos, we observe Slam protein first lost in the membranes of invaginating cells (Fig. 5D, Fig. 6H). Slam protein can no longer be detected in embryos by the end of gastrulation (Fig. 5E). The RNA and protein expression profile of slam correlated well with the slam cellularization phenotype. No RNA or protein was detected, however, during the second phase of Slam function, germ cell migration, suggesting that early Slam expression is not only necessary for cellularization but also later for germ cell migration.

We have identified a novel gene, slam, which is required for the formation of the cellular blastoderm and for germ cell migration in a pathway independent of Hmgcr. Consistent with the cellularization phenotype, slam RNA is expressed at high levels in the early embryo just before cellularization and Slam protein localizes to the basal membrane of newly forming cells. Slam expression is transient and by the time of germ cell migration we can no longer detect slam RNA or protein. Our data suggest that signals required for the migration of germ cells during later stages of development are deposited into the cell membrane during blastoderm formation.

We identified slam in a zygotic screen for germ cell migration defects and showed that lack of maternal and zygotic slam function causes a failure in blastoderm cellularization. slam was identified independently as one of a small number of genes with a zygotic cellularization phenotype (Lecuit et al., 2002; Merrill et al., 1988; Wieschaus and Sweeton, 1988). In a genome-wide screen using deletions of large genomic regions, seven regions of the genome, including the slam region, were identified as crucial for cellularization of early embryos (Merrill et al., 1988; Wieschaus and Sweeton, 1988). Other genes accounting for the cellularization phenotype in three of the genomic regions were identified as nullo, serendipity α and bottleneck (Schejter et al., 1992). These three genes share the following characteristics: first, they are expressed transiently at high levels only during blastoderm stage; second, they encode novel, small proteins, which share no homology to other genes in the Drosophila genome or in other genomes; third, the phenotype was only revealed in embryos deficient for that genomic region; there were no known point mutant alleles. Despite their dramatic effect on embryogenesis, these genes had escaped detection in the large-scale mutagenesis screens for zygotic embryonic lethal mutations carried out by Nüsslein-Volhard and Wieschaus (Jürgens et al., 1984; Nüsslein-Volhard et al., 1984; Wieschaus et al., 1984), presumably because of the difficulty in identifying point mutations in these genes.

Several findings identify slam as a fourth member of this group of cellularization genes. First, slam maps to region 26D, which when deleted causes a cellularization defect; second, slam is expressed transiently and at high levels at the cellular blastoderm stage; and third, slam encodes a novel protein. In addition, in a parallel study, Lecuit and Wieschaus showed that injection of slam RNAi produces a cellularization phenotype similar to that of a genomic deletion of the region (Lecuit et al., 2002). However, in contrast to nullo, serendipity α and bottleneck, and the genomic deletion of the 26D region, our genetic analysis showed that the cellularization phenotype was produced only after removal of both maternal and zygotic slam. Embryos treated with slam RNAi or embryos missing large genomic regions containing slam fail to cellularize, and look like the maternal^–zygotic^–slam^waldo mutants. This suggests that our slam^waldo alleles are hypomorphs, such that the maternal contribution is sufficient to rescue the cellularization defect of homozygous mutant slam^waldo embryos but not that of embryos deleted for the genomic region that contains slam. Simultaneous removal of the maternal and zygotic slam^waldo (M^–Z^–) contribution reduces gene activity below a threshold required for cellularization. While this scenario offers a plausible explanation for the slam phenotypes, it remains puzzling that slam is the only cellularization gene with point mutant alleles, and that these point mutations were identified on the basis of a germ cell migration phenotype and not by the cellularization phenotype.

During cellularization the membrane surface of the embryo increases 25-fold. To account for this impressive membrane growth, it has been proposed that new synthesis of membrane and controlled vesicle fusion along the apical and lateral membrane contribute to most of the observed membrane growth. Gene products known to be associated with Golgi vesicles, such as Lava Lamp, or known components of the vesicle fusion machinery, such as Syntaxin, have been implicated in cellularization (Burgess et al., 1997; Sisson et al., 2000). We detect Slam protein in vesicles close to the membrane as well as in the membrane. Furthermore, detailed analysis of the slam cellularization defect observed after RNAi injection suggests that slam mutants are defective in directed transport of membrane vesicles (Lecuit et al., 2002).

In addition to its role in cellularization, slam is also required for germ cell migration. Our data strongly suggest that for germ cell migration, like for cellularization, Slam function is needed in the somatic cells and not in the germ cells. We detect slam RNA and protein only during blastoderm and the onset of gastrulation (stage 4-5, 2-3.5 hours after fertilization) but not later (stage 10-11, 5-7 hours after fertilization) during germ cell migration. We did not detect maternally loaded Slam protein in germ cells and zygotic expression of slam RNA was absent from germ cells. Furthermore, we found that the germ cell migration phenotype is only observed in homozygous mutant embryos and that a paternal slam⁺ allele rescues both the cellularization and the germ cell migration defect. As germ cells are transcriptionally repressed until they initiate migration, the genetics of the slam phenotype and the slam RNA and protein expression pattern argue against a role of slam in germ cells.

slam RNA and protein expression profiles lead us to conclude that expression of Slam in the soma during blastoderm formation must be necessary later in development for germ cell migration. One possible explanation for the role of slam in germ cell migration could be that proper cellularization is required for germ cell migration and that zygotic slam^waldo mutants have a subtle cellularization defect that later causes a germ cell migration defect. We did not detect any obvious cellularization defects in homozygous slam^waldo mutant embryos (data not shown). Thus, as an alternative, we propose that slam-targeted membrane vesicles may not only accomplish rapid membrane growth but may also be used to insert a germ cell guidance signal into the growing membrane. This hypothetical guidance signal would be deposited before or during gastrulation, when Slam is present, but act after gastrulation when germ cells are migrating off the midgut (Fig. 7). Our mutant analysis suggests that Slam function in the midgut primordium is crucial for germ cell migration. We observed a delay in germ cells migrating from the midgut to the mesoderm and a delay in the epithelial to mesenchymal transition of the midgut primordium in mutant embryos. A possible model is that during blastoderm formation Slam is needed for the deposition of a germ cell guidance factor into the basal-lateral membrane. This factor could be present in all cells, or just in the posterior midgut. During gastrulation, the germ cells migrate through the posterior midgut primordium from the apical to the basal side (Fig. 7C). Once they reach the basal side of the epithelium, germ cells require this guidance factor to move quickly off the midgut and into the mesoderm (Fig. 7D). The nature of this guidance factor is unknown, it could be a factor very specific to germ cells or it could be a component of the basement membrane, which mediates cell migration in many systems. The guidance factor may be a germ cell repellant or a protein that promotes germ cell movement. It is likely to be independent of Hmgcr based on the slam^waldo;Hmgcr double mutant phenotype. It is also likely independent of wun and wun 2. The wun/wun 2-dependent germ cell reorientation on the midgut appears normal in slam^waldo mutant embryos, suggesting that slam^waldo mutations affect a later step in germ cell migration than wun/wun 2.

Blastoderm-specific genes, like slam, nullo, bottleneck and serendipity-α only affect formation of cells at the blastoderm stage. These genes are not required for germ cell formation, which happens at the same developmental stage, and are not expressed during cytokinesis at other stages of development. This suggests that blastoderm cellularization is a highly specialized form of cytokinesis. Blastoderm formation has been used as a model system to study the generation of polarized epithelia. In support of this idea, blastoderm cells form apical junctions and contain proteins in a polarized distribution typical for epithelial cells. However, it has been difficult to establish a functional link between the polarity established during the blastoderm stage and the polarity observed in epithelial cells at later stages of differentiation, because embryos that fail to form cells in the first place cannot be analyzed later. slam^waldo mutations cause defects in germ cell migration and midgut morphology, but not in blastoderm cellularization. Nevertheless, the blastoderm-specific expression of Slam suggests that Slam function at this early stage is required for later development. Our findings, therefore, provide evidence that Slam may not only generate the cells from which the embryo develops but may also establish spatial cues within those cells needed for later morphogenesis.

We thank our colleagues and the Bloomington Stock Center for providing fly stocks and antibodies as listed in the text. We thank Hélène Zinszner, Adriana Lafaille and Anja Socjzewska for technical assistance. We are grateful to Jessica Treisman for suggesting the analysis of slam^waldo M^–Z^– mutants, and to Eric Wieschaus and Thomas Lecuit for sharing unpublished results with us. We thank Evan Stein for computational assistance with sequence analysis. Thanks also to members of the Lehmann laboratory for technical suggestions and for critically reading the manuscript. J. A. S. is supported by RO1 HD41900 from the NIH. H. T. B was supported by a training grant from the NIH. L. A. M. was supported by a pre-doctoral fellowship from the NSF. R. L. is an HHMI investigator.