Germ line stem cell differentiation in Drosophila requires gap junctions and proceeds via an intermediate state

Stem cells can self-renew and generate differentiated progeny. Both of these aspects must be tightly regulated for the correct development and maintenance of many tissues. Control of self-renewal or differentiation could be intrinsic, such as by unequal distribution of determinants, or extrinsic,by signals emanating from surrounding cells, termed `the niche'(Watt and Hogan, 2000). Communication between stem cells and their niche have been shown to require secretion of extracellular matrix signaling molecules, and interaction via membrane proteins (Spradling et al.,2001; Watt and Hogan,2000). Here we show that a gap junction protein is required for differentiation of female germ-line stem cells (GSCs) in the fruit fly Drosophila melanogaster.

The Drosophila ovary is composed of 16-18 units called ovarioles,containing a germarium at the anterior and a chain of developing egg chambers that extend posteriorly. GSCs reside in the anterior of the germarium and closely abut the somatic terminal filament, cap and inner-sheath cells(Fig. 1A), which comprise the GSC niche (Lin, 1998; Spradling, 1993; Xie and Spradling, 2000a). Several extrinsic factors, Decapentaplegic (Dpp), Piwi, Fs(1)Yb and Hedgehog,control GSC or somatic stem cell maintenance and are expressed in different subsets of niche cells (Cox et al.,1998; Cox et al.,2000; Forbes et al.,1996; King and Lin,1999; King et al.,2001; Xie and Spradling,1998; Xie and Spradling,2000b). Not much is known about how external signaling from the niche preserves GSC fate. One protein that is likely to play a role in GSC maintenance is Pumilio, a repressor of translation, which is required within GSCs for GSC maintenance (Forbes and Lehmann, 1998; Lin and Spradling, 1997; Murata and Wharton, 1995; Zamore et al.,1997). However, the targets and partners of Pumilio within GSCs are unknown.

Normally, GSCs divide asymmetrically, such that one daughter cell remains at the anterior, within the niche, and maintains GSC fate; the posterior daughter, one cell diameter removed from the niche(Fig. 1A), differentiates to form a cystoblast (Deng and Lin,1997). The process of GSC differentiation (here defined as the transition of GSCs to cystoblasts) remains largely unknown. In particular,unlike some mammalian stem cell systems, very few markers are available to follow the steps toward GSC differentiation. In wild-type females, GSCs can be distinguished from cystoblasts by two main criteria: position (GSCs are closer to cap cells), and expression of the cytoplasmic form of the Bag of marbles protein BamC, which appears in cystoblasts and dividing cysts, but not in GSCs. Two genes, bam and benign gonial cell neoplasm (bgcn)are necessary for the differentiation of the cystoblast(Gonczy et al., 1997; Lavoie et al., 1999; McKearin and Ohlstein, 1995; McKearin and Spradling, 1990; Ohlstein et al., 2000; Ohlstein and McKearin, 1997). However, the molecular function of these genes is unknown, and it remains unclear whether the bam/bgcn pathway is the only pathway necessary for GSC differentiation.

Recently, we have shown that the gap junction protein Zero population growth (Zpg) is required for an early step of both oogenesis and spermatogenesis (Tazuke et al.,2002). Gap junctions are intercellular, voltage-gated channels that allow cells to selectively share ions, metabolites and other messenger molecules. They have well-established roles in physiological and developmental processes such as cardiac muscle contraction and ovarian follicle maturation(Phelan and Starich, 2001; Simon and Goodenough, 1998; Simon et al., 1997; White and Paul, 1999). Here we show that GSCs lacking Zpg can divide, but the daughter cell destined to differentiate dies. Our results suggest that zpg may be necessary for the differentiation process itself, as well as for survival of differentiated germ cells, and that zpg probably acts in parallel to bamand bgcn. We further show that the differentiation of the GSC to a cystoblast is gradual, and that many of the germ cells in `stem cell tumors'caused either by strong mutations in bam or by overexpression of Dpp may be at an intermediate state between GSCs and cystoblasts.

For molecular identification of zpg as inx4, and characterization of zpg alleles zpg^z5352 and zpg^z2533see Tazuke et al. (Tazuke et al.,2002). Df(3L)CH20, Df(3L)CH12 uncover zpg (Hong and Hashimoto,1995). The zpg alleles used are most likely protein nulls, as they contain stop codons at positions 91 and 239, respectively(Tazuke et al., 2002). In newly eclosed flies (1-3 days old), both zpg alleles displayed the same phenotypes when homozygous, in trans to each other, or over either deficiency. In old females (three weeks of age), however, differences between flies that carry zpg point mutations and those that carry a point mutation over a deficiency were noted (data not shown). To avoid background interference, all our analyses were performed on young females, where the phenotypes of the zpg alleles were consistent and indistinguishable from that of the allele over the deficiency.

Three strong pum alleles were used in this study: In(3R)Msc, pum^ET7 and pum^ET1,described by Forbes and Lehmann (Forbes and Lehmann, 1998). All allelic combinations were tested and produced the described phenotypes. However, zpg,pum^ET1/zpg, In(3R)Msc produced mostly empty ovarioles, whereas zpg, pum^ET7/zpg, In(3R)Msc produced more ovarioles containing germ line than other allelic combinations. bgcn^QS2 was described by Ohlstein et al.(Ohlstein et al., 2000) and bam⁸⁶ by McKearin and Ohlstein(McKearin and Ohlstein, 1995). The hs-bam transgene on the X chromosome was described by Ohlstein and McKearin (Ohlstein and McKearin,1997). hs-dpp flies were generated by Ron Blackman. The JH1.1 strain carries eight copies of the hs-dpp transgene located at 16F-17A, 2B, 77, 96D-97C.

Rabbit anti-Vasa antibody was used at a dilution of 1:5000. Chicken anti-Vasa was a generous gift from Dr K. Howard, and was used at 1:5000. Rabbit anti-Zpg was used at 1:5000 and is described by Tazuke et al.(Tazuke et al., 2002). Rat anti-BamC was a gift of Dr D. McKearin and was used at 1:500. Rabbit anti p-Mad (antiphosphorylated SMAD1, PS1)(Persson et al., 1998) was a gift of P. ten-Dijke and was used at 1:1000. Rabbit anti-phospho-Histone H3 was from Upstate Biotechnology and was used according to the manufacturer's protocol. 1B1 monoclonal supernatant(Zaccai and Lipshitz, 1996)was from the Developmental Studies Hybridoma Bank and was used at 1:25. In all cases, secondary antibodies, coupled to FITC, Cy3 or Cy5, were purchased from Jackson ImmunoResearch Laboratories and were used at 1:500.

Newly eclosed flies were heat-shocked twice (hs-bam) or three times (hs-dpp) a day at 37°C, for 1 hour, for four days. Flies were dissected the following day and stained with the appropriate antibodies.

Fixation and immunostaining of ovaries were according to standard protocols. Imaging was performed on a Leica DM RBE confocal microscope, using the Leica TCS NT program.

Newly eclosed flies were dissected and double-stained with chicken anti-Vasa and rabbit anti-phospho-Histone H3 antibodies. Ovarioles were observed under a fluorescent microscope to score for cells that had anti-phospho-Histone staining. Only the cells at the anterior tip of the ovariole were defined as `GSCs'. The number of cells that double stained for Vasa and phospho-Histone H3 was divided by the total number of ovarioles scored for each genotype.

One- to 3-day-old flies of the genotype hs-flp¹²²; FRT82, armadillo-LacZ /FRT82 were heat shocked once for 15 minutes, at 37°C. Under these mild conditions, only 30-50% of the ovarioles, depending on the experiment,were marked, and marked ovarioles mostly had only one marked GSC. Flies were dissected 2-3 days after heat shock and stained with anti-Vasa and anti-β-Gal. Cells that carry no copies, one copy or two copies of the transgene can be distinguished by the intensity of anti-β-Gal staining.

Most ovarioles from newly eclosed female flies that carry a strong mutation in the gene zpg have only a few germ cells, located at the anterior tip of the ovariole; the others contain no germ cells(Tazuke et al., 2002). zpg ovarioles that were occupied by germ line harbored 3.8 germ cells on average, as determined by anti-Vasa antibody labeling to mark the germ line(s.d.=1.9, n=464). In comparison, wild-type germaria are filled with dividing cysts, the products of cystoblast divisions(Fig. 1C). The stem cell positions (close to cap cells) were occupied by zpg cells, but some single germ cells or clusters of germ cells were also located away from the tip (Fig. 1B,D). The single cells in the zpg mutant contained a round spectrosome(Fig. 1B,E), and could therefore be a mixed population of GSCs, cystoblasts or an intermediate between the two. To determine the developmental state of zpg germ cells, we double-labeled wild-type and zpg ovaries with anti-Vasa antibody and an antibody against cytoplasmic Bam protein (BamC), which stains cystoblasts and early cysts, but not GSCs(Ohlstein and McKearin, 1997)(Fig. 1C). Of the 169 zpg ovarioles that had germ cells(Fig. 1D), only 2 contained a single cell that was BamC-positive (∼1%). By contrast, 13 out of 50 wild-type ovarioles had a single cell stained with anti-BamC (26%; for further analysis of cystoblasts in wild type, see below). We conclude that most zpg ovarioles lack cystoblasts.

In some zpg ovarioles we observed clusters of 2-3 cells that stained positive with anti-BamC antibodies. We assume that these may be the developing cysts that give rise to the rare egg chambers observed in these mutants (Tazuke et al., 2002). As in wild type, the rare developing zpg cysts were interconnected by ring canals. However, the fusome, which normally spreads through the dividing cyst, was either aberrant or absent. The rare egg chambers were almost always composed of less than 16 germ line cells, and DAPI staining indicated degeneration of the chamber (data not shown). As the fusome controls the divisions of the cyst (de Cuevas et al.,1996; Lilly et al.,2000; Lin et al.,1994), the abnormal number of germ cells in the egg chamber is likely to be the result of a severed fusome in the mutants.

Using an antibody raised against the cytoplasmic tail of Zpg(Tazuke et al., 2002), we observed staining in germ cells of every stage of development from GSCs to budding cysts of wild-type germaria. The staining appeared to be at the membrane of the cells and was somewhat stronger in region 2(Fig. 1E). The presence of Zpg staining in GSCs and dividing cysts correlates with the early defect in germ cell differentiation in zpg mutants and suggests that Zpg acts in the germ line.

The low number of cystoblasts in zpg ovarioles could be the result of defects in GSC division or in survival of the GSC daughter cell destined to differentiate. To distinguish between these two possibilities, we stained zpg or wild-type germ cells for the mitotic marker phospho-Histone H3(Fig. 2A,B). The percentage of marked zpg cells at the stem cell position per ovariole was not statistically different from that of wild-type GSCs(Table 1). Similarly, the number of wild-type GSCs per ovariole in S phase (detected by BrdU labeling)was comparable to that of zpg cells (data not shown). Taken together,these data show that zpg and wild-type cells spend a similar proportion of time in M and S phase. Although we cannot rule out the possibility that zpg cells have an overall slower cell cycle, as compared to wild-type GSCs, our data show that zpg cells divide. The lack of differentiating cells in zpg ovarioles may therefore indicate that zpg germ cells die when they commence differentiation.

To test this hypothesis, we forced zpg germ cells to differentiate by inducing expression of Bam protein in these cells, using a heat shock construct (Ohlstein and McKearin,1997). In wild type, overexpression of Bam in GSCs induces the cystoblast differentiation program, resulting in a chain of egg chambers connected to a germarium depleted of germ line(Ohlstein and McKearin, 1997)(Fig. 2C). When Bam was induced in zpg flies, no germ cells could be observed with anti-Vasa staining(Fig. 2D). Heat shock itself was not the cause of germ cell death, as ovaries of heat-shocked zpganimals, which did not carry the hs-bam transgene, still possessed germ cells (Fig. 2D, inset). Thus, zpg germ cells die when forced to differentiate. This indicates that Zpg is required for the survival of differentiating germ cells. Taken together, our data indicate that zpg cells divide, and that the daughter cells that are destined to differentiate die.

To further explore the role of Zpg in early germ cell differentiation and survival, we recombined zpg alleles with mutant alleles of the gene pumilio (pum) (Forbes and Lehmann, 1998; Lin and Spradling, 1997). pum mutant ovaries exhibit a compound phenotype. Many ovarioles lack germ line completely(Fig. 3B,D), a defect that may be attributed to preoogenic defects(Asaoka-Taguchi et al., 1999; Forbes and Lehmann, 1998; Lin and Spradling, 1997; Parisi and Lin, 1999). Ovarioles with germ line have a germ line-depleted germarium connected to a few defective egg chambers (Fig. 3A). This phenotype suggests that Pum has roles in GSC maintenance(Forbes and Lehmann, 1998; Lin and Spradling, 1997). In ovaries from zpg, pum double-mutant females, many ovarioles were empty (see Materials and methods). This is consistent with the embryonic and larval requirement for pum (Fig. 3D). Ovarioles occupied by germ line exhibited a phenotype more similar to zpg than to pum mutants: few germ cells at the tip of the ovariole (Fig. 3C,D). Thus Zpg function is required for the differentiation of pum mutant germ cells. The apparent difference between the zpg,pum and hs-Bam; zpg phenotypes may reflect the different roles of Pum and Bam in GSC differentiation. Pum, as a translational repressor may permit GSC maintenance by repressing differentiation, which requires Zpg. By contrast, Bam may have a more instructive role in GSC differentiation, such that its overexpression can overrule GSC-maintenance cues emanating from the niche, independent of zpg(Ohlstein and McKearin, 1997)(this study).

In addition to defects in GSC maintenance, pum mutants also show defects in cyst development (Forbes and Lehmann, 1998; Parisi and Lin,1999) (Fig. 3A). This function was also evident in the zpg, pum phenotype. Although zpg, pum ovarioles occupied by germ line mostly resemble the zpg phenotype, we observed that they contained more germ cells, and had more dividing cysts and differentiating egg chambers, than those of the single zpg mutant (compare Fig. 3C with Fig. 3D). The double mutant had an average of 0.23 egg chambers per ovariole(n=290), compared with 0.02 (n=500) in flies homozygous for zpg and heterozygous for pum. The higher number of single cells and egg chambers in zpg, pum double mutants may indicate that Zpg function is less essential when cyst development is abrogated, as is the case in pum mutants.

Our analysis suggests that Zpg wild-type function is required for the differentiation of GSCs. At least two other genes, bam and bgcn, are required for early germ cell differentiation(Gonczy et al., 1997; Lavoie et al., 1999; McKearin and Ohlstein, 1995; Ohlstein et al., 2000). However, the phenotype of bam and bgcn mutant ovaries is strikingly different from that of zpg mutants. Ovaries mutant for bam or bgcn are filled with many undifferentiated single germ cells harboring a spherical spectrosome, which have been described as GSC tumors (Fig. 4A)(McKearin and Ohlstein, 1995; McKearin and Spradling, 1990; Ohlstein et al., 2000). By contrast, ovaries from zpg flies have only a few germ cells at the tip of the ovariole (Fig. 4B). To determine the functional relationship between these genes, we made flies double mutant for zpg with either bam or bgcn. Ovaries from newly eclosed females were stained to visualize the germ line and the spectrosomes. In the double mutant lacking both zpg and bam, only a few germ cells were detected at the ovariole tip(Fig. 4C). However, most double-mutant ovarioles had somewhat more germ cells than zpgovarioles (average for zpg, bam=6.4, s.d.=3.2, n=110;average for zpg=3.8, s.d.=1.9, n=464). Similar results were obtained with double mutants of zpg and bgcn(Fig. 4D).

As GSCs can survive in a zpg background, the predicted phenotype of a zpg, bam or bgcn; zpg double mutant would have been similar to a bam (or bgcn) phenotype (i.e. a germarium filled with undifferentiated GSCs). By contrast, the double-mutant phenotype more closely resembles the zpg phenotype. To test whether slow division of zpg cells accounts for the lack of tumors in young females, we analyzed older (1- to 2-week-old) females, and found a similar phenotype to that of young females (data not shown). Thus, wild-type Zpg function is required for the accumulation of bam or bgcnmutant germ cell tumors removed from the niche.

To investigate further a possible role for Zpg-mediated intercellular communication in the development of germ cell tumors, we analyzed the genetic interaction between dpp and zpg. It had been proposed that an increased Dpp signal induced over-proliferation of GSC-like cells(Xie and Spradling, 1998; Xie and Spradling, 2000b). We therefore reasoned that an increased Dpp signal could induce zpgcells to over-proliferate. To test this, we crossed flies carrying several insertions of a heat-shock dpp transgene (hs-dpp) into a zpg background. Flies were heat shocked, and then dissected and stained to reveal the germ line and spectrosomes. Control animals heterozygous for zpg, carrying a subset of the hs-dpp transgenes, had more single germ cells with spherical spectrosomes than did wild type(Fig. 5A). Homozygous zpg animals, which carried at least the same number of hs-dpp transgenes as the control, did not show an increase in germ cell number (Fig. 5B). To test whether zpg cells could respond to a Dpp signal, we stained ovaries of zpg animals with antibodies against phosphorylated Mad (p-Mad)(Persson et al., 1998). Mad is a component of the Dpp signal transduction pathway and is phosphorylated upon activation of the pathway (Massague,1998). In wild type, p-Mad staining could be detected in GSCs,cystoblasts and dividing cysts in region 1 of the germarium. The highest level of staining was observed in early germ cells; staining then gradually declined towards the posterior (Fig. 5C). p-Mad was also detected in the single cells that accumulate following Dpp overexpression (Fig. 5D). Similarly, p-Mad staining was detected in zpg germ cells (Fig. 5E,F), suggesting that the mutant cells are able to receive the Dpp signal. There may be several explanations for the failure of zpg cells to proliferate or survive in response to a Dpp signal. One is that the Dpp pathway is blocked downstream of Mad in zpg cells. Second, zpg cells may not be able to survive when unattached to the tip of the ovary. Third, proliferating cells in hs-dpp flies, which move away from the niche, are in a more differentiated state than the cells in the niche, and therefore die in a zpg background (see Discussion).

bam tumor cells and germ cells proliferating after Dpp overexpression (hs-dpp tumor cells) are considered to be GSCs because of their round spectrosomes and lack of BamC staining(McKearin and Ohlstein, 1995; Xie and Spradling, 1998; Xie and Spradling, 2000b). Yet, these cells do not accumulate in a zpg background. One possible explanation for this observation is that bam and hs-dppcells, as they move away from the niche, are at an intermediate state(pre-cystoblast) between a stem cell and a cystoblast, and that cystoblast development and survival requires Zpg (see Discussion). To determine whether an intermediate state between GSCs and cystoblasts exists in wild type, we triple-labeled ovarioles with anti-Vasa, 1B1 monoclonal antibody and anti-BamC, to mark the germ line, spectrosomes and cystoblasts, respectively(Fig. 6B). BamC antibody stained cysts of 4 or 8 cells strongly. Two-cell cysts had notably weaker staining. Only rarely did we observe cystoblasts, i.e. single cells, stained with anti-BamC [data not shown; similar observations were reported by Ohlstein and McKearin (Ohlstein and McKearin,1997)]. In many ovarioles, we could observe single cells with a spherical spectrosome that were removed from the stem cell position yet did not stain for the cystoblast marker BamC. We counted the number of cells that carried a spherical spectrosome and did not stain with anti-BamC. These cells would comprise the GSC population plus the presumptive intermediate population. Of 100 ovarioles scored, most had between 3 and 5 single cells that did not stain with anti-BamC (Fig. 6A). The average number of these cells was 3.9. This is a greater number than the average number of GSCs that populate an ovariole (between 2 and 3), as determined by cell-lineage analysis and electron microscopy(Carpenter, 1979; Margolis and Spradling, 1995; Wieschaus and Szabad, 1979). These data support the hypothesis that an intermediate state between a stem cell and a cystoblast exists in wild type (see also Ohlstein and McKearin,1997).

An intermediate cell population between slowly dividing stem cells and differentiating cells, described as transit amplifying cells, is a common feature in stem cell systems, including those giving rise to blood cells, skin and the gut epithelium. These are the products of stem cell division and have limited potential to divide prior to differentiation(Melchers and Rolink, 2000; Watt, 2000; Winton, 2000). In principle,the existence of a transit amplifying cell population in the Drosophila germ line could be suggested if the GSC division rate is too low to account for the number of cysts/egg chambers produced (including dying ones) in a set period of time. Acquiring this information, especially in region 1 of the germarium has proved to be difficult(Cox et al., 2000; Drummond-Barbosa and Spradling,2001; Lin and Spradling,1993; Smith et al.,2002). We therefore directly examined the products of possible division of cells at the transition state. If a pre-cystoblast cell amplifies,it should give rise to more than one egg chamber. By contrast, a cystoblast divides incompletely, giving rise to a two-cell cyst and, eventually, to one egg chamber. To distinguish between `intermediate-state' divisions and cystoblast divisions, we conducted mosaic analysis in wild-type germaria using the FLP-FRT marker system (Xu and Rubin,1993), in which cells that are produced by mitotic recombination are marked as twin-spots by the copy number of the gene (lacZ),encoding the marker β-Galactosidase (0 or 2 copies). Recombination in GSCs would result in a marked GSC, a string of marked cysts arising from its subsequent division, and one cyst that is the twin-spot of the original recombination event (Fig. 6C). A recombination event in the cystoblast would not be observed under our experimental conditions because marked cells within a cyst share the diffusable β-Gal. We reasoned that, if a cell at the transition state divides, we would observe twin-spot cysts even in germaria where GSCs are not marked (Fig. 6C). Such a situation could also arise if a marked GSC left the niche and differentiated. However, GSC loss is unlikely to affect our analysis, as the half-life of wild-type GSCs is 4.6 weeks (Xie and Spradling, 1998), whereas we dissected the females 2-3 days after induction of Flipase. Of 177 ovarioles scored, 104 contained no marked GSCs and no marked cysts (∼60%). 73 ovarioles contained marked cysts and a marked GSC (∼40%; Fig. 6D). We only observed ovarioles that contained marked cysts and marked GSCs. These results lead us to conclude that cells at the transition state in wild type do not divide at an appreciable rate but proceed to differentiate to a cystoblast. Thus, they do not constitute the equivalent of a transient amplifying population. By contrast, tumor cells in bam mutants, or when Dpp is overexpressed, continue to divide, as their differentiation is blocked.

zpg ovarioles contain single germ cells at the anterior tip. Most of these cells do not reach the cystoblast stage. As zpg cells are not arrested at a particular stage of their cell cycle, and can divide, we conclude that GSCs that lack zpg divide to give another stem cell and a daughter cell that dies at early stages of differentiation. Accordingly,overexpression of bam, which is necessary and sufficient to promote germ cell differentiation, kills zpg cells.

zpg encodes innexin 4 - one of eight innexins in Drosophila (FlyBase) (Phelan and Starich, 2001; Stebbings et al., 2002; Tazuke et al.,2002). Innexins are the functional homologues of the vertebrate connexins, or gap junction proteins(Phelan and Starich, 2001). In mammalian oogenesis, gap junctions have been implicated in cell-cell signaling. Early luteinization of granulosa cells is observed either when the oocyte is physically removed from immature wild-type follicles(el-Fouly et al., 1970), or in mice lacking connexin 37 (Simon et al.,1997), suggesting a gap junction-mediated signaling mechanism between the oocyte and granulosa cells.

Here we show that the Drosophila gap junction protein Zpg is required for the survival of the germ line stem cell daughter as it moves away from the niche and begins to differentiate. From its expression pattern, and the specificity of its function, it is clear that Zpg acts in a germ line autonomous way. However, we do not yet know whether Zpg facilitates communication between germ cells, or between germ line and soma. The germarium is a compact structure where early germ cells contact each other, the somatic cap cells contact GSCs, and inner-sheath cells contact GSCs and their daughters (King, 1970; Schulz et al., 2002). Further study is needed to clarify which cells communicate with germ cells through Zpg-gap junctions. Likewise, the nature of the requirement for zpg in GSC differentiation is still uncertain. Gap junctions may be used to supply the GSC daughter cell with nutrients, or with a survival factor required for its subsequent growth. Alternatively, gap junctions may be used to deposit a factor that is required for the differentiation process itself, rather than for survival, while an accessory mechanism eliminates cells that begin differentiating without that factor. The major argument in favor of a role for Zpg in differentiation at the stem cell stage comes from our phenotypic analysis of zpg, pum double mutants in which, unlike in pumsingle mutants, GSCs do not differentiate. Although Pum-mediated repression is removed in the double mutants, GSCs cannot differentiate as they may lack a differentiation signal provided by Zpg. It is harder to imagine how nutritional deficiency per se could block differentiation of the double-mutant cells because single-mutant zpg cells do begin to differentiate (and then die).

Recent studies suggest that the niche promotes stem cell fate through Dpp signaling (Xie and Spradling,1998; Xie and Spradling,2000b; Zhu and Xie,2003). This may be achieved through repression of bam(Chen and McKearin, 2003). GSCs are also tightly tethered to the niche via adherens junctions(Song et al., 2002). Other,as-yet unidentified mechanisms may be used by the niche to protect GSCs.

Once germ cells leave the niche they activate the differentiation pathway. We propose that differentiation of GCSs to cystoblasts is not direct but proceeds via an intermediate state (Fig. 6). We have shown that most wild-type cells, which by their position within the germarium were judged to be cystoblasts, are not stained with BamC antibody. Our finding concurs with Ohlstein and McKearin, who observed cytoplasmic Bam just before the cystoblasts divide to form a two-cell cyst (Ohlstein and McKearin,1997), and proposed the existence of an intermediate/pre-cystoblast state between GSCs and cystoblasts. In other stem cell systems, the intermediate cell population has a biological function,namely increasing the progeny of a single stem cell division. Our results indicate that in Drosophila females, cells at the intermediate state do not constitute a transit-amplifying cell population. However, the`pre-cystoblast state' may have a different biological significance. A vacant niche can be filled by a neighboring GSC that divides `sideways' instead of along the anteroposterior axis (Xie and Spradling, 2000b). An alternative for filling a vacant GSC position might be for a cell at the intermediate state to reoccupy the niche.

The suggested model, adding a transitory state between the stem cell and the cystoblast, raises an interesting question. Is the differentiation of a GSC to a cystoblast a continuous process or a discrete one? It is notable that none of the markers we currently use is specific to the stem cell, the cystoblast or the intermediate. Bam is present (in its fusomal form) already in the stem cell, and BamC gradually accumulates in the cystoblast(Ohlstein and McKearin, 1997)(this study). Pumilio protein and phosphorylated Mad are also detected from GSCs to cystoblasts and early cysts(Forbes and Lehmann, 1998)(this work). In other stem cell systems, including mammalian hematopoiesis,many intermediate cell types are known, and can be isolated by specific marker combinations (reviewed by Melchers and Rolink, 2000). Due to the relative lack of markers, the isolation of these `cell types' from Drosophila ovaries is currently impossible. The overlap of expression patterns of the proteins that are known to have a role in stem cell maintenance and differentiation may indicate that differentiation is gradual, and possibly reversible.

It has been suggested that in bam or bgcn mutant ovaries,or when Dpp is overexpressed, the germaria are filled with GSC tumor cells(Gonczy et al., 1997; Lavoie et al., 1999; McKearin and Ohlstein, 1995; Ohlstein et al., 2000; Xie and Spradling, 1998; Xie and Spradling, 2000b). Our findings of an intermediate cell population in wild type raises the possibility that GSC tumor cells share some properties with precystoblasts. Both these populations are single, do not stain for BamC, but exist outside of the niche. Some support to the analogy between pre-cystoblasts and `GSC tumors' is evident in the fact that the latter do not survive past the niche in a zpg background. Thus, the zpg double mutants allow us to distinguish two cell populations in `GSC tumors' - cells that are inside or outside of the niche. We suggest that when Dpp is overexpressed, or in bam/bgcn mutants, cells outside the niche cannot fully activate the differentiation program and are at an intermediate state between a GSC and a cystoblast. These cells die in a zpg background, whereas the tumor cells in contact with the niche survive. Our results thus suggest that beyond Zpg gap junctions and Dpp signaling, there must be additional signaling between GSCs and the niche, which helps maintain GSCs. Additional markers are needed to determine unequivocally whether bam tumors are similar to Dpp tumors, and whether they share properties with wild-type precystoblasts.

Differentiation of the stem cell daughter requires gap junctions. We assume zpg is acting in parallel to pum, dpp, bam and bgcnbecause the double mutants showed incomplete epistasis of zpg over each of these genes (Fig. 7). Although a role for gap junction proteins has been established in mammalian oogenesis (Ackert et al., 2001; Carabatsos et al., 2000; Juneja et al., 1999; Simon et al., 1997), to our knowledge, this is the first instance where a gap junction protein is shown to control stem cell differentiation. What passes through these gap junctions,and which cells are connected to GSCs through Zpg channels, is still unknown. One intriguing option is that Zpg forms part of the channels that connect GSCs to the surrounding somatic niche cells. If so, that would suggest that the niche in the Drosophila germarium is necessary, not only for stem cell maintenance, but also for stem cell differentiation.

In a recent paper, Kai and Spradling report observations of p-Mad staining in wild-type and Dpp overexpressing germaria similar to our own. In addition,they report that in bam mutant-ovaries tumor cells abutting niche cells stain with p-Mad, while cells away from the niche do not(Kai and Spradling, 2003).

We wish to thank members of the Lehmann laboratory, J. Treissman, C. Navarro and, especially, J. Morris for fruitful discussions and critical reading of the manuscript. We thank B. Wakimoto and D. Lindsley for providing zpg sterile lines; C. Hashimoto for the deficiency lines uncovering zpg; P. ten-Dijke for p-Mad antibody; D. McKearin for BamC antibody and hs-bam flies; and K. Howard for chicken anti-Vasa antibody. This work was supported by the Human Frontiers Science Program (L.G.),NIH-Reproductive Scientist Development Program (S.I.T.), Burroughs-Wellcome Fund (S.I.T.), NIH grant U54-HD31398 (S.I.T.), NIH grant P01-DK53074 (sub project #4) (M.T.F), and NIH grant R01 HD27549 (R.L.). R.L. is a Howard Hughes Medical Institute investigator.