Resolving embryonic blood cell fate choice in Drosophila:interplay of GCM and RUNX factors

Understanding how cell lineages are established remains a central task in developmental biology and particularly in the field of hematopoiesis. Strikingly, it has become clear over the past few years that several aspects of hematopoiesis have been conserved between Drosophila and man(Evans et al., 2003). Thus, Drosophila may provide a valuable model system with which to gain insight into the mechanisms of blood cell lineage segregation in vivo. As in vertebrates, hematopoiesis in the fruit fly occurs in two waves: blood cell progenitors arise from the head mesoderm in the early embryo and from a specialised organ, the lymph gland, in the larva(Holz et al., 2003). These progenitors (prohemocytes) give rise to three terminally differentiated cell types (collectively called hemocytes), related to the vertebrate myeloid lineages: plasmatocytes, crystal cells and lamellocytes(Meister, 2004). Approximately 95% of hemocytes are plasmatocytes, which function as macrophages, engulfing apoptotic cells and small pathogens such as bacteria(Sears et al., 2003; Tepass et al., 1994). Crystal cells constitute a smaller population of blood cells (about 5%) that participate in melanisation, an insect-specific process involved in wound healing and the encapsulation of foreign invaders(Rizki et al., 1980). Finally,the lamellocytes encapsulate foreign bodies too large to be engulfed by macrophages (Crozatier et al.,2004). They are only generated during larval hematopoiesis, under specific conditions such as parasitisation of the larvae by wasp eggs.

Despite the evolutionary distance between Drosophila and vertebrates, many of the molecular pathways governing hematopoiesis have been conserved (Evans et al., 2003). In particular, transcription factors of the GATA, FOG and RUNX families, which regulate several steps of hematopoiesis in vertebrates, also control Drosophila hematopoiesis (Fossett et al., 2001; Lebestky et al.,2000; Rehorn et al.,1996). In the embryo, all prohemocytes express the GATA transcription factor Serpent (SRP), which is required for blood cell precursor specification and maintenance (Rehorn et al., 1996). From this pool of prohemocytes, two populations of hematopoietic cells will emerge, plasmatocytes and crystal cells(Lebestky et al., 2000). SRP participates in their differentiation and remains expressed in the differentiated blood cells (Fossett et al., 2003; Waltzer et al.,2002). Furthermore, key lineage-specific factors are required for the differentiation into these two cell types. On the one hand, the RUNX transcription factor Lozenge (LZ) forms a functional complex with SRP to induce crystal cell formation (Waltzer et al., 2003), and in a lz mutant no crystal cells appear(Lebestky et al., 2000). On the other hand, the two related transcription factors Glial cells missing(GCM) and GCM2 (also known as Glide and Glide2) are jointly required for plasmatocyte differentiation (Alfonso and Jones, 2002; Bernardoni et al.,1997; Kammerer and Giangrande,2001). In the absence of both gcm and gcm2(gcm/gcm2), plasmatocytes do not differentiate normally and their number is strongly reduced, whereas crystal cell formation appears unaffected (Alfonso and Jones,2002). In addition, enforced expression of GCM in crystal cells converts them into plasmatocytes (Lebestky et al., 2000). By contrast, ectopic expression of LZ in plasmatocytes induces crystal cell marker expression but does not repress plasmatocyte cell fate (Waltzer et al.,2003).

It is proposed that, for their differentiation, crystal cells must express lz but not gcm/gcm2, whereas plasmatocytes have to express gcm/gcm2 but not lz(Lebestky et al., 2000). Interestingly, gcm/gcm2 expression is detected from stage 5 in the hematopoietic primordium (Alfonso and Jones, 2002; Bernardoni et al., 1997), whereas the onset of lz expression is only detected later (at stage 10) (Lebestky et al., 2000). These observations raise several questions: (1) do the crystal cell and the plasmatocytes precursors emerge from the same pool of prohemocytes; (2) when do the two populations segregate; and (3) what is the relationship between lz and gcm/gcm2 during blood cell lineage choice?

To address these questions, we have undertaken an analysis of the mechanism of segregation of the two embryonic blood cell lineages. We find that gcm is expressed early on in all prohemocytes but is rapidly downregulated in the anterior-most cells of the hematopoietic primordium,which initiate lz expression by stage 7. Our results suggest that the coordinated repression of gcm and activation of lz in the anterior row of prohemocytes is a key step in the regulation of blood cell lineage choice. In the absence of both gcm and gcm2, we observe a striking increase in the number of crystal cells, indicating that gcm/gcm2 actually regulates crystal cell development. We further show that gcm/gcm2 inhibits crystal cell formation in two steps: first by regulating the number of cells that initiate lz expression, and second by interfering with lz maintenance in these cells. Furthermore, contrary to what has been reported during larval hematopoiesis (Duvic et al.,2002; Lebestky et al.,2003), we demonstrate that Notch signalling is neither sufficient nor required for crystal cell formation. Finally, our results indicate that prohemocytes are bipotent progenitors, and that the interplay between gcm/gmc2 and lz expression dictates the cell fate choice.

Most Drosophila melanogaster lines were obtained from the Bloomington Drosophila Stock Center. The following strains were kindly provided by different laboratories: uas-gcm, uas-gcm2, gcm-lacZ(gcm^rA87), Df(2L)200 and Df(2L)Gcm2 (B. Jones,University, MS, USA); uas-lz and lz^R1 (U. Banerjee, Los Angeles, CA, USA); stg² (M. Crozatier,Toulouse, France); N^55e11 FRT10.1 (A. Martinez-Arias,Cambridge, UK). uas-srp, srp-gal4 and pg33 have been previously described by Waltzer et al.(Waltzer et al., 2002). To generate gcm- and gcm2-deficient germ-line clones, males carrying hs-flp; FRT40A ovo^D1/CyO were crossed to Df(2L)200 FRT40A/CyO females. Larvae from this cross were heat shocked daily for 1 hour at 37°C for 3 days, and the emerging Df(2L)200 FRT40A/FRT40A ovo^D1 adult females were crossed to Df(2L)200/CyO (twi-lacZ) males. Notch germ-line clone mutants were generated according to a similar protocol, using the N^55e11 FRT10.1 and the ovo^D1 FRT10.1;hs-flp stocks.

Unless specified, crosses and embryo collections were performed at 25°C. To induce transient expression of GCM, hs-gal4; uas-gcmembryos were collected at 18°C for 3 hours, aged at 18°C for 6, 9 or 12 hours, heat shocked twice at 37°C for 20 minutes, and aged at 18°C for 16, 13 or 10 hours, respectively, before being processed for analysis.

The 1.5 kb upstream regulatory region of lz (nucleotides 234118 to 235562 on the genomic scaffold AE003446) was cloned into pCasper-hs43-lacZ to generate pLZ-lacZ. The corresponding P{lz-lacZ} transgenic lines were generated by standard P-element-mediated transformation into w¹¹¹⁸ flies.

The in situ hybridization technique and probes used have been previously described by Waltzer et al. (Waltzer et al., 2003).

For double fluorescent staining, the following antibodies were used: rabbit anti-Serpent antibody (1/500) (Reuter,1994), rabbit anti-β-GAL antibody (1/500; Cappel Pharmaceutical), goat anti-rabbit Alexa Fluor 488 (1/400; Molecular Probes),sheep anti-DIG antibody (1/500; Roche), donkey anti-sheep antibody Alexa Fluor 488 (1/400; Molecular Probes) and/or anti-fluorescein AP (1/1000; Roche),revealed with Fast Red substrate (Vector).

As a first step to elucidate the mechanisms underlying embryonic blood cell fate choice, we compared the expression patterns of the key transcription factors srp, gcm and lz from the earliest stage of hematopoiesis, i.e. from stage 5, when srp expression defines the hematopoietic anlage in the mesoderm(Rehorn et al., 1996).

GCM and GCM2 are co-expressed transcription factors that act redundantly to induce plasmatocyte differentiation(Alfonso and Jones, 2002) and that have been suggested to be plasmatocyte specific(Alfonso and Jones, 2002; Lebestky et al., 2000). We investigated whether gcm is expressed only in a sub-population of prohemocytes that will give rise to plasmatocytes or in all prohemocytes,including the prospective crystal cells. As shown Fig. 1, at stage 5 (2 hours and 10 minutes to 2 hours and 50 minutes after egg laying; AEL), gcm and srp transcripts co-localise in all of the cells that constitute the hematopoietic primordium (Fig. 1A). gcm2 is also expressed in the hematopoietic anlage from stage 5, but at a much lower level than gcm(Alfonso and Jones, 2002),therefore precluding the precise analysis of its expression domain by fluorescent in situ hybridization. Interestingly, at stage 6 (2 hours and 50 minutes to 3 hours AEL), gcm transcripts are no longer detected in the anterior-most row of srp-expressing cells(Fig. 1B). We extended this observation by comparing the localisation of the gcm transcripts with that of the SRP protein (Fig. 1C). Thus, although gcm is initially expressed in all prohemocytes, its expression is subsequently downregulated in a sub-population of prohemocytes.

To confirm that gcm is also expressed in the crystal cell precursors, we analysed the expression of the P{lacZ} insertion in gcm (gcm^rA87), previously shown to recapitulate gcm expression (Bernardoni et al.,1997). Taking advantage of the long half-life of β-gal, we observed that almost all of the crystal cells (DoxA3-positive cells)co-expressed β-gal by stage 11 (Fig. 1K). However, β-gal staining diminished progressively in the crystal cell population at later stages(Fig. 1L,M). For comparison,SRP (Lebestky et al., 2000) or srp-gal4/uas-lacZ expression is maintained in crystal cells throughout embryogenesis (Fig. 1N). Hence, although gcm is initially expressed in all prohemocytes, it is rapidly repressed in the crystal cell precursors. Thus,the transcriptional downregulation of gcm in the presumptive crystal cell progenitors (and its maintenance in the remaining prohemocytes) is the earliest known manifestation of a blood cell-lineage choice.

lz activity is required for crystal cell formation(Lebestky et al., 2000). Using direct in situ hybridization and immunostaining, it was shown that lzis weakly expressed in the crystal cell precursors from stage 10 (4 hours and 20 minutes to 5 hours and 20 minutes AEL). However, our results with gcm suggest that the initial blood cell lineage choice takes place at around stage 6, one hour earlier than the reported lz expression. Because low levels of expression might hinder lz detection during the early stages of hematopoiesis, we re-assessed the onset of lzexpression by different means. First, we made use of the lz-gal4enhancer-trap line that recapitulates lz expression(Lebestky et al., 2000). By generating an auto-amplification loop with three copies of uas-gal4transgenes (Hassan et al.,2000), we observed lacZ expression as early as stage 7 (3 hours to 3 hours and 10 minutes AEL; Fig. 1G). In parallel, we generated transgenic lines containing different regulatory regions of lz cloned upstream of a lacZreporter gene. Similarly, we found that a 1.5 kb-long lz-upstream region drove lacZ expression from stage 7 in a row of prohemocytes and, subsequently, in differentiated crystal cells(Fig. 1F,O). Remarkably, the lz-lacZ-expressing cells are not homogeneously distributed among the prohemocyte population but comprise the anterior-most row of cells in which gcm transcripts are no longer detected(Fig. 1D,E). It thus appears that lz expression is induced by stage 7 in a specific domain of the hematopoietic anlage that corresponds to the domain where gcmexpression is lost.

Crystal cell precursors differentiate into two bilaterally symmetrical groups of cells that remain localised around the proventriculus(Lebestky et al., 2000). In the course of our experiments with the uas-gal4 amplification system,we noticed that some lz-gal4/uas-lacZ-positive cells were localized outside the crystal cell clusters(Fig. 1H,I) and did not express DoxA3 (Fig. 3E),suggesting that not all the cells that initiate lz expression differentiate into crystal cells. This enticed us to investigate the fate of all of the cells that initiate lz expression (so-called lz⁺ progenitors). As shown in Table 1, by stage 14, 60% of the marked cells are clustered along the proventriculus and co-express the crystal cell marker DoxA3. The remaining 40% are scattered throughout the embryo, lack the expression of crystal cell-specific markers and morphologically resemble macrophages (Fig. 1J). Therefore, only a fraction of the lz⁺progenitors differentiates into crystal cells, while the rest become plasmatocytes. These data, together with the observation that gcm is initially expressed in all prohemocytes and is required for plasmatocyte differentiation (Alfonso and Jones,2002; Kammerer and Giangrande,2001), led us to re-assess the role of gcm/gcm2 during blood cell lineage choice.

To address the possible role of gcm and gcm2 during crystal cell development, we first analysed the phenotype of embryos carrying a deficiency that removes both genes (Df(2L)200)(Alfonso and Jones, 2002). In Df(2L)200 mutant embryos, we observed a striking increase in the size of the crystal cell clusters and the presence of numerous ectopic DoxA3-expressing cells scattered throughout the embryo (Fig. 2B,D compare with 2A,C). Similar results were obtained when we analysed the expression of other crystal cell markers, such as CG5579 and CG8193. These results were confirmed with a smaller deficiency (Df(2L)Exel7042) that also removes gcmand gcm2 (data not shown). Morphological observation strongly suggested that these DoxA3⁺ cells are crystal cells(Fig. 2F), despite the ectopic localisation of some of them. Furthermore, these cells efficiently melanised in a Bc mutant context and were not observed in a lz mutant background (Fig. 2G,I). Thus,in sharp contrast to previous reports(Alfonso and Jones, 2002; Lebestky et al., 2000), our data indicate that gcm and gcm2 normally inhibit crystal cell development.

To ensure that the above phenotypes were due to the lack of gcmand/or gcm2 but not to another gene deleted by the deficiencies, we also examined the phenotypes of gcm or gcm2 single mutants. Lack of gcm resulted in a marked increase in the number of crystal cells (Fig. 2L, Table 1), whereas lack of gcm2 alone did not significantly affect their development(Fig. 2J). The phenotypes were not enhanced when a gcm or gcm2 mutation was combined with Df(2L)200 (Fig. 2K,M). Nonetheless, the strongest phenotype was observed when both gcm and gcm2 were deleted (Fig. 2D, Table 1). These results indicate that gcm is primarily responsible for inhibiting crystal cell development, and that its lack of function can be partially compensated for by gcm2. To verify that gcm2 is also able to inhibit crystal cell fate, we overexpressed gcm2 in crystal cell precursors. Similar to what was shown for GCM(Lebestky et al., 2000)(Fig. 2N,),lz-gal4-driven expression of GCM2 inhibited crystal cell formation(Fig. 2O). Thus, both gcm and gcm2 are capable of inhibiting crystal cell formation.

Comparison between the expression patterns of the crystal cell-specific marker DoxA3 and the panhemocyte marker peroxidasin(Pxn) (Nelson et al.,1994) indicated that most hemocytes do not acquire the crystal cell fate in gcm/gcm2 mutant embryos (Fig. 2D, compared with 2R). Thus, in a Df(2L)200 mutant embryo, it was shown previously that there are approximately 250 PXN-labeled cells(Alfonso and Jones, 2002),whereas we observed about 90 crystal cells, and 40% of them were mislocated(Table 1). Because some gcm activity is maternally contributed(Bernardoni et al., 1997), we wondered whether this might explain why only some prohemocytes adopt a crystal cell fate or some crystal cells migrate. However, even in embryos derived from Df(2L)200 germline clones, ectopic crystal cells were still present(Fig. 2P) and most hemocytes did not differentiate as crystal cells(Fig. 2S). Thus, although gcm and gcm2 inhibit crystal cell development, their absence is not sufficient to cause a complete switch in the fate of the hematopoietic precursors from plasmatocytes to crystal cells.

Next, we tried to understand how gcm and gcm2 control crystal cell fate. As downregulation of gcm in the anterior row of prohemocytes shortly precedes lz activation, we investigated whether lz activation was modified in the absence of gcm and gcm2. On monitoring lz-lacZ expression in stage 7 embryos(i.e. before proliferation resumes), we observed more lz⁺progenitors in the Df(2L)200 mutant embryos than in wild type(Fig. 3B). Interestingly, all the lz⁺ progenitors remained localised to the anterior of the hematopoietic domain, indicating that lz induction is still spatially restricted. Thus, in the absence of gcm/gcm2 there is an increase in the size of the prohemocyte subpopulation that initiates lz expression.

We next analysed the fate of the lz⁺ progenitors. In stage 14 embryos, we observed a large increase in the number of lz-gal4/uas-lacZ-expressing cells in the absence of gcm/gcm2 (Table 1, Fig. 3D). Double labelling indicated that almost all of the lz-gal4/uas-lacZ-expressing cells, even those scattered throughout the embryo, co-expressed the crystal cell marker DoxA3(Table 1, Fig. 3F). This is in marked contrast with the wild-type situation, where only the lz-gal4/uas-lacZ cells located around the proventriculus express DoxA3 (Table 1, Fig. 3E). Thus, in the absence of gcm/gcm2, all the lz⁺ progenitors,both those that remain near the proventriculus and those that migrate to distant positions, differentiate into crystal cells.

To ensure that the increase in crystal cells was directly related to a rise in the number of lz⁺ progenitors and not to a modification of the proliferation program, we determined the number of crystal cells formed in the absence of cell proliferation. Accordingly, we introduced the string (Drosophila cdc25) mutation, which blocks all postblastodermal cell divisions (Edgar and O'Farrell, 1990), and monitored the number of crystal cells formed in wild-type or in Df(2L)200 mutant embryos. Even in a string mutant context, lack of gcm/gcm2 induced a twofold increase in the number of crystal cells(Fig. 3, compare G with H).

In summary, the absence of gcm/gcm2 results in an increase in the number of lz⁺ progenitors and allows all of them to differentiate into crystal cells. These data suggest that gcm and gcm2 inhibit crystal cell differentiation by a two-step process. First, they limit the induction of lz expression to a subset of prohemocytes, thereby regulating the number of crystal cell precursors. Second, they inhibit the acquisition of the crystal cell fate in 40% of the lz⁺ progenitors, which take on a plasmatocyte fate instead.

The preceding results suggested that gcm represses lzexpression. To test this hypothesis, we performed a gain-of-function analysis using the uas/gal4 system. Interestingly, GCM (and GCM2, data not shown) repressed lz when it was expressed throughout the mesoderm(twi-gal4), in the entire hemocyte population (srp-gal4) or in the presumptive crystal cells (lz-gal4)(Fig. 4B-D). Similar results were obtained when we monitored the more robust lz-lacZ expression instead of lz. (Fig. 4G). By contrast, lz expression was not affected when GCM was expressed under the control of the plasmatocyte-specific driver pg33 (Fig. 4E)(Waltzer et al., 2003). Thus gcm and gcm2 can repress lz expression in a cell-autonomous manner.

All of the cells initiating lz expression differentiate into crystal cells in the absence of gcm/gcm2, whereas 40% of them do not in a wild-type situation. To address the possibility, that gcm also inhibits the maintenance of lz, we induced gcm overexpression at different times during embryogenesis using a hs-gal4 driver (see Materials and methods for details). Interestingly, transient ectopic expression of GCM around stage 9, 10 or 11 still repressed crystal cell formation(Fig. 4L-N). Therefore gcm probably also inhibits the maintenance of lzexpression.

As lz expression must be maintained for crystal cell differentiation (Lebestky et al.,2000), we wondered whether lz expression might be auto-activated. Reminiscent of its capacity to induce ectopic crystal cell markers (Waltzer et al.,2003), twi-gal4-driven LZ expression activated lz-lacZ in the srp-expressing domains(Fig. 4I). Indeed, SRP and LZ form a functional complex to induce crystal cell development(Waltzer et al., 2003). Therefore we ascertained whether they also cooperate to regulate lzexpression. Whereas we observed restricted ectopic activation of lz-lacZ by SRP or LZ alone (Fig. 4H,I), lz-lacZ was strongly activated throughout the mesoderm when SRP and LZ were co-expressed(Fig. 4J). Thus SRP and LZ cooperate to maintain lz expression via a positive-feedback loop.

If gcm (and gcm2) antagonizes crystal cell development only by repressing lz expression, we surmised that we might rescue crystal cell formation by uncoupling lz expression from gcmregulation. Accordingly, we co-expressed LZ and GCM under the control of the srp-gal4 driver and monitored crystal cell marker expression. As shown in Fig. 4, LZ induced DoxA3 expression to a similar extent in the presence or in the absence of overexpressed GCM, indicating that GCM does not inhibit LZ function(compare Fig 4Q with 4R). All together, these data demonstrate that gcm and gcm2 repress crystal cell formation by inhibiting both the induction and the maintenance of lztranscription.

During larval hematopoiesis, Notch signalling in the lymph gland is required and is sufficient to induce crystal cell formation(Duvic et al., 2002; Lebestky et al., 2003). Furthermore, the number of crystal cells is decreased in a Notchmutant embryo, suggesting that Notch signalling might also be required for embryonic crystal cell formation (Lebestky et al., 2003). Since the remaining crystal cells in a Notch mutant might reflect Notch maternal contribution(Lebestky et al., 2003), we generated Notch mutant germ line clones. We still observed crystal cells, both in embryos derived from Notch germline clones(12.8±2.6 crystal cells; n=21) and in Notch zygotic mutants (15.9±3.5 crystal cells; n=19; Fig. 5B,C). Therefore, although Notch participates in embryonic crystal cell development, it is clearly not required. To test whether ectopic Notch signalling activates crystal cell formation in the embryo, we expressed an activated form of Notch in all of the prohemocytes using the srp-gal4 driver or in plasmatocytes using the pg33 driver. In neither case did we observe more crystal cells(Fig. 5D,E). Therefore,contrary to the previous suggestion(Lebestky et al., 2003), we show that Notch signalling is neither sufficient nor required to induce crystal cell formation in the embryo.

In a gcm/gcm2 mutant background, as in wild type, only a fraction of the prohemocytes develops into crystal cells. In addition, unlike gcm, which can convert crystal cells into plasmatocytes(Lebestky et al., 2000),ectopic expression of lz induced crystal cell marker expression without repressing plasmatocyte differentiation(Waltzer et al., 2003)(Fig. 6C,G,K). This raised the question: do all prohemocytes have the ability to develop as bona fide crystal cells? To address this issue, we expressed lz in all of the prohemocytes in a gcm/gcm2 mutant background with the srp-gal4 driver and monitored hemocyte differentiation. Under these conditions, almost all of the hemocytes remained around the proventriculus,judging by the expression of the hemocyte marker Pxn(Fig. 6H). Furthermore, all the hemocytes expressed high levels of crystal cell-specific markers and morphologically resembled differentiated crystal cells(Fig. 6D,L). Thus, in the absence of gcm/gcm2, lz is capable of inducing differentiation of all the prohemocytes into crystal cells. In conclusion,these results strongly support the hypothesis that prohemocytes are bipotential hematopoietic progenitors that can differentiate either into plasmatocytes or into crystal cells depending on the respective activity states of gcm/gcm2 and lz.

We have taken advantage of the relatively simple model provided by Drosophila embryonic hematopoiesis to attempt to unravel the mechanisms that underlie the choice of two blood cell fates in vivo. Our data indicate that crystal cells and plasmatocytes develop from a pool of bipotential hematopoietic progenitors. We show that the earliest detectable manifestation of the segregation of the two blood cell lineages occurs in the anterior row of prohemocytes with the downregulation of gcm and the induction of lz. Furthermore, we demonstrate that the number of lz-expressing precursors, and their final differentiation into crystal cells or plasmatocytes, is regulated by gcm/gcm2activity, which inhibits lz induction and maintenance. Thus,embryonic blood cell lineage segregation is revealed to be a highly dynamic process in which the interplay between the transcription factors gcm/gcm2 and lz plays a crucial role.

It was shown previously that gcm and gcm2 are required for the proper differentiation of plasmatocytes, and GCM and GCM2 were claimed to be plasmatocyte-specific lineage transcription factors that are not involved in crystal cell development(Alfonso and Jones, 2002; Lebestky et al., 2000). By contrast, our results clearly demonstrate that gcm and gcm2inhibit crystal cell formation. Furthermore, we detected the expression of gcm in all of the prohemocytes. including the prospective crystal cell precursors, at stage 5, a result confirmed by tracing gcm-lacZexpression into early differentiating crystal cells. Thus, gcm and gcm2 participate in blood cell fate segregation by regulating not only plasmatocyte development but also that of crystal cells.

gcm and gcm2 have been most intensively studied during neurogenesis, where they are required to promote glial cell development at the expense of neuronal cell fate (Van De Bor and Giangrande, 2002). We show here that they also regulate a binary cell fate choice during hematopoiesis. However, although their expression is restricted to glial precursors during neurogenesis(Alfonso and Jones, 2002; Kammerer and Giangrande, 2001; Vincent et al., 1996), they are initially expressed in all prohemocytes irrespective of their subsequent fate. Furthermore, in the absence of gcm/gcm2, whereas almost all presumptive glial cells are transformed into neurons(Alfonso and Jones, 2002; Kammerer and Giangrande, 2001; Vincent et al., 1996), only a small proportion of the presumptive plasmatocytes adopts a crystal cell fate. Therefore, the function and mechanism of action of gcm/gcm2in regulating cell fate choice during neurogenesis and hematopoiesis are different.

We have deduced that gcm and gcm2 control crystal cell formation by a two-step process. First, gcm/gcm2 determines the number of crystal cell precursors by restricting the initiation of lz expression in the prohemocyte population(Fig. 7). In the absence of gcm/gcm2, we observed more lz⁺ progenitors,correlating with a greater number of differentiated crystal cells at later stages. Our data indicate that gcm is expressed early in the entire hematopoietic primordium but is rapidly downregulated in the prospective lz expression domain. Maintaining GCM or GCM2 expression in the lz⁺ progenitors inhibited crystal cell differentiation. Thus, repressing gcm/gcm2 expression in the anterior population of prohemocytes is most probably a prerequisite for the emergence of crystal cells.

Second, gcm and gcm2 regulate the proportion of lz⁺ progenitors that ultimately differentiate in crystal cells: whereas 40% of these cells differentiate into plasmatocytes in wild-type embryos, all of them become crystal cells in the absence of gcm/gcm2. Lebestky et al. already noted that some lz-expressing cells differentiate into plasmatocytes and suggested that this might be due to the de novo activation of gcm expression in these cells (Lebestky et al.,2000). Our results extend their observations and demonstrate that gcm participates in this process, although it is not re-expressed in the lz⁺ cells. Our data further suggest that the residual gcm activity present in the lz⁺ progenitors may contribute to the relative plasticity in the fate of these progenitors,allowing some of them to differentiate into plasmatocytes. In summary, we provide compelling evidence that gcm and gcm2 play a key role in regulating cell fate choice in prohemocytes and lz⁺ progenitors.

Our study yields new insight into the regulation and mode of action of lz during embryonic crystal cell development. Although plasmatocytes migrate through the embryo, crystal cells gather around the proventriculus. Strikingly, we showed that in the absence of gcm/gcm2,srp-driven high-level expression of lz induced the transformation of all of the hemocytes to authentic crystal cells that remain clustered. By contrast, when lz is expressed under the control of its own promoter, 40% of lz⁺ cells migrate through the embryo whether or not they express gcm/gcm2. Hence, our data suggest that high levels of lz are required for crystal cell clustering and that lz induction in prohemocytes is heterogeneous. Below a certain threshold, lz⁺ progenitors retain the default migratory behaviour of hemocytes and, in the presence of gcm/gcm2, can differentiate into plasmatocytes. It is noteworthy that gcm/gcm2 participate in (but are not required for) hemocyte migration (Alfonso and Jones, 2002; Bernardoni et al., 1997). Thus, lz and gcm/gcm2appear to have opposite effects on blood cell migration, with gcm/gcm2 promoting a migratory behaviour that dominates the inhibitory effect of lz.

It has been shown that lz function is continuously required to promote crystal cell development (Lebestky et al., 2000). Here, we have identified an enhancer of lzthat is transactivated by the SRP/LZ complex. This observation suggests that,once initiated, lz expression can be maintained by a positive autoregulatory feedback loop, thereby providing a simple mechanism to stabilise crystal cell lineage commitment. This enhancer contains several RUNX-binding sites and we are currently investigating the role of these sites in lz autoregulation. Interestingly, the three mammalian homologues of the RUNX factor LZ contain several conserved RUNX-binding sites in their promoters (Otto et al., 2003). Furthermore, RUNX2 maintains its own expression through an auto-activation loop in differentiated osteoblasts (Ducy et al., 1999), whereas RUNX3 inhibits RUNX1 expression in B lymphocytes (Spender et al.,2005). Thus, auto- or cross-regulation might be a common feature of the RUNX family. In addition, we showed that GCM/GCM2 repress lz expression. However, no consensus GCM-binding sites are present in the lz crystal cell-specific enhancer. Interestingly, it was recently shown that zebrafish gcmb is expressed in macrophages(Hanaoka et al., 2004). Yet,the putative functions of the two gcm homologues and their possible interplays with RUNX factors have not been investigated during vertebrate hematopoiesis.

Because gcm is expressed early in the entire hematopoietic anlage,it is tempting to speculate that prohemocytes are primed to differentiate into plasmatocytes (i.e. macrophages). Thus, it appears likely that Drosophila blood cells progenitors are not `naïve'. Similarly,mammalian stem and progenitor blood cells express low levels of lineage-affiliated genes and it has been suggested that they are primed for differentiation (Graf, 2002). Furthermore, from an evolutionary perspective, macrophages are certainly the oldest and most pervasive blood cell type(Lichanska and Hume, 2000),and it is remarkable that another hematopoietic cell type may have evolved from this lineage through the use of a conserved RUNX transcription factor.

Acquisition of crystal cell fate involves both the repression of the primary fate (i.e. repression of gcm) and the activation of lz. Our data show that these two steps are coordinated in space and time. Nonetheless, the induction of lz is not the mere consequence of relieving gcm/gcm2-mediated repression of lz but requires an active and localised process. How gcm transcription is repressed and lz activated in the anterior row of prohemocytes is currently unknown. In the lymph gland, Notch/Serrate signalling is necessary and sufficient to induce crystal cell formation by activating lzexpression (Duvic et al., 2002; Lebestky et al., 2003). However our results demonstrate that, contrary to the situation in larvae,Notch is not required for crystal cell formation in the embryo. In this respect, it is interesting to note that neither gcm nor gcm2is expressed in the lymph gland (B.A., unpublished). Hence, the process that segregates crystal cells from plasmatocytes relies on different mechanisms in the embryo and in the larval lymph gland. Similarly, in vertebrates, primitive and definitive hematopoiesis may also depend on partially distinct programs(Shepard and Zon, 2000). For instance, in mouse, the transcription factor PU.1 plays an essential role in the emergence of definitive macrophages but does not seem to be required for the formation of primitive macrophages in the yolk sac(Lichanska et al., 1999).

The coincident repression of gcm and activation of lzbetween stages 6 and 7 in a row of prohemocytes is remarkable, as it suggests that the head mesoderm is delicately patterned at this early stage of development. The hematopoietic primordium is located in the posterior head region, whose patterning involves several genes including buttonhead,empty spiracles, orthodenticle and collier(Crozatier et al., 1999). However, mutations of these genes do not specifically suppress crystal cell or plasmatocyte development (L.B., unpublished). Further work will thus be required to understand the coordination permitting the silencing of gcm and the activation lz that triggers the choice of one fate at the expense of the other.

It was shown that gcm can induce the differentiation of all of the prohemocytes into plasmatocytes (Lebestky et al., 2000). The data presented here demonstrate that, in the absence of gcm/gcm2, lz can transform all of the hemocytes to crystal cells. Thus, Drosophila prohemocytes are bipotent progenitors. However, the incapacity of lz to repress gcm(and thereby plasmatocyte fate) implies that the resolution of cell fate choice does not rely on reciprocal antagonism between two `lineage-specific'transcription factors like between GATA1 and PU.1 during myeloid/erythroid cell fate choice in vertebrates (Galloway et al., 2005; Graf,2002; Rhodes et al.,2005). Instead, we propose that Drosophila embryonic blood cell fate segregation is a process that can be divided into two consecutive phases (Fig. 7). A local cue triggers the process by downregulating gcm and activating lz in the anterior population of prohemocytes, whereas gcmexpression is maintained in the remaining cells, which differentiate into plasmatocytes. Then, in the lz⁺ progenitors, the relative levels of LZ and GCM will dictate lineage choice. If the ratio of LZ to GCM is high enough to overcome GCM-mediated repression of lz expression, LZ can elicit its autoregulatory activation loop and the progenitor will differentiate into a crystal cell. If not, GCM inhibits lzautoactivation and the progenitor differentiates into a plasmatocyte. Such a mechanism of segregation could provide some plasticity, because the size of a population may be regulated at different times by physiological cues influencing either the initiation event or the feed-back loop required for its development.

In conclusion, our data shed light on the transition in vivo from bipotent hematopoietic progenitors to lineage-restricted precursors. Interestingly, the embryonic Drosophila cell fate choice occurs though an original mechanism distinct from that observed during larval hematopoiesis. Moreover,this process does not seem to involve reciprocal negative regulation between two `lineage-specific' transcription factors. Hence, the mechanisms leading to the resolution of hematopoietic lineages in vivo appears to be more complex and diverse than expected.

We are grateful to J. Smith and members of the CBD for critically reading the manuscript, and to M. Tauzin for her help. We thank B. Ronsin and the IFR109 imaging platform. We also thank U. Banerjee, R. Reuter, A Giangrande and B. Jones for reagents and fly stocks. This work was supported by grants from the CNRS and ARC. G.F. is supported by a post-doctoral fellowship from the ARC.