Steroid hormone-dependent transformation of polyhomeotic mutant neurons in the Drosophila brain

Development of nervous systems involves derivation of numerous distinct subtypes of neurons from limited numbers of precursors (reviewed by Doe and Skeath, 1996). Cells imprint histories of development in their genomes; and differential imprinting of genomes is fundamental to cell differentiation (reviewed by Orlando, 2003). The Polycomb group (PcG) of nuclear proteins, including Polyhomeotic (Ph), is known to modulate gene expression globally in the genomes according to their patterns of imprinting (reviewed by Pirrotta,1998; Beuchle et al.,2001). However, little progress has been made to elucidate roles of PcG proteins in neuronal development since the initial characterization of the neural phenotypes of ph mutants 17 years ago(Smouse et al., 1988).

Neural tissues differentiate step by step. First, distinct neuronal precursors, characterized with different transcriptional codes, are specified during patterning of neuroectoderm (Briscoe et al., 2000; Urbach et al.,2003). Second, distinct precursors further give rise to different characteristic sets of multiple subtypes of neurons during neurogenesis(Lee et al., 1999; Schmid et al., 1999; Jefferis et al., 2001). Both processes of cell diversification involve spatial and/or temporal patterning of tissues (reviewed by Jacob and Briscoe,2003; Zhong,2003). Tissue patterning permits acquisition of different gene expression profiles in originally equivalent cells, while maintenance of cell type-specific transcription programs depends on various mechanisms of epigenetic functions (Orlando,2003).

Much of our knowledge about roles of epigenetic imprint in cell differentiation comes from characterization of several PcG genes that were originally identified through mutant screens based on derepression of homeotic(Hox) genes in Drosophila embryos (e.g. Lewis, 1978; Duncan, 1982; Jurgens, 1985). Proper patterning of Drosophila embryos requires expression of distinct Hox genes in different spatially restricted regions along the anteroposterior axis(e.g. McGinnis and Krumlauf,1992). Expression of Hox genes is initially controlled by the Gap proteins, such as Hunchback and Kruppel, which set the limits of Hox gene expression by repressing transcription during early embryogenesis (reviewed by Bienz and Muller, 1995). Interestingly, such repression of Hox genes lasts through cell divisions and in the absence of the Gap repressors. Heritable silencing of Hox genes at the later developmental stages requires PcG proteins (e.g. Lewis, 1978; Duncan, 1982; Jurgens, 1985; Bienz and Muller, 1995). Both the Hox genetic system and its late epigenetic maintenance exist in higher organisms (e.g. Gould, 1997; Forlani et al., 2003). PcG proteins, thus, constitute a widely conserved cell memory system that prevents changes in cell identity by maintaining transcriptional repression of previously suppressed genes throughout development and in adulthood.

PcG proteins are thought to maintain gene silencing by controlling chromatin accessibility (e.g. Boivin and Dura, 1998; Zink and Paro,1995; Fitzgerald and Bender,2001). In vivo, different PcG proteins form at least two distinct multimeric chromatin silencing complexes(Ng et al., 2000). The PRC1 complex, containing Polycomb (Pc) and Polyhomeotic (Ph) among others, appears physically associated with the chromatin of specific cis-regulatory sequences in Hox genes, called Polycomb response elements (PREs) (e.g. Shao et al., 1999; Saurin et al., 2001; Horard et al., 2000; Bloyer et al., 2003; Ringrose et al., 2003). Meanwhile, experiments with DNA-tethered PcG proteins, such as GAL4-Pc, have provided evidence that PcG proteins function as potent transcriptional repressors (Muller,1995). But detailed molecular links remain missing from binding of PcG complexes with PREs to chromatin remodeling and transcriptional silencing. It is also unclear how transiently expressed factors help chromatin silencing complexes find their way onto targets and how these complexes retain characteristic chromatin-binding patterns from one cell generation to the next. In addition, evidence accumulates to challenge our conventional views on the biological functions of PcG. First, several PcG proteins seem to have a dual role in both repression and activation of transcription, depending on the locus and genetic context (Brock and van Lohuizen, 2001). Second, distinct PcG proteins function in different complex manners. Some PcG proteins are differentially distributed from others on polytene chromosomes(DeCamillis et al., 1992; Franke et al., 1992; Ng et al., 2000); and the phenotypes of different PcG mutants are distinct (e.g. Campbell et al., 1995; Narbonne et al., 2004). Finally, derepression of Hox genes is variably involved in different mutants and in distinct tissues (e.g. Beuchle et al., 2001; Dura and Ingham,1988; Simon et al.,1992; Choi et al.,2000).

Although global misrouting of CNS axons has been well demonstrated in ph mutant embryos (Smouse and Perrimon, 1990), little is known about roles of PcG complexes in specific neuronal developmental processes. Here, we report recovery of a new ph recessive lethal mutation from genetic mosaic screens in adult fly brains. Loss of subtype identity was evident in ph mutant clones within otherwise phenotypically wild-type brains. Through metamorphosis, all ph mutant neurons were transformed into cells with unidentifiable projection patterns and indistinguishable gene expression profiles. But postembryonic-born ph mutant neurons were never transformed without experiencing the prepupal ecdysone peak. In addition, we detected limited derepression of Hox genes in ph mutant neurons and requirement of distinct PcG proteins for different aspects of neuronal development. Taken together, we demonstrate that Drosophila Ph plays an essential role in maintaining neuronal diversity through metamorphosis, showing possible two-way interactions between steroid hormone signaling and the epigenetic functions of PcG.

Creation of l(X)MB342 MARCM clones involves: (1) FRT19A,hs-FLP,tubP-GAL80;UAS-mCD8GFP;GAL4-OK107; (2) UAS-mCD8GFP,GAL4-NP225; (3) ato-GAL4(Hassan et al., 2000); and (4) elav-GAL4 (Luo et al.,1994). (5) FRT19A, UAS-mCD8GFP was used to create wild-type MARCM clones. (6) Pc^XT109,FRT2A/TM6C and (7) E(z)⁷³¹,FRT2A/Tm6C were used together with (8) hs-FLP,UAS-mCD8GFP;FRT2A,tubP-GAL80;GAL-OK107, (9) GAL4-C155,UAS-mCD8GFP,hs-FLP and (10) FRT2A,tubP-GAL80 to create Pc and E(z) MARCM clones. Fly stocks used for mapping the mutations include (11) ec,cv,ct,t/C(1)DX (BL-1163), (12) y,ct,ras,f (BL-4362), (13) Df(1)64c18,g,sd/Dp(1;2;Y)w[+]/C(1)DX (BL-936), (14) Dp(1;f)R/y,dor (BL-761), (15) Df(1)w258-45/C(1)DX;Dp(1;3)w[vco] (BL-1527), (16) Df(1)Pgd-kz (BL-1902), (17) Df(1)Pgd35 (BL-1986), wapl² (BL-5741), (18) ph⁶⁰² (BL-5444), (19) Pgd^G0385(BL-11998) and (20) l(1)G0458 (BL-10112). Phenotypic rescue was carried out with (21) P{ph-d⁺} transgene that comprises the ph-d transcription unit(Randsholt et al., 2000).

Our ongoing genetic mosaic screens have been reported before(Lee et al., 2000; Wang et al., 2002; Zheng et al., 2003). Briefly,MARCM clones of MB neurons, which are homozygous for EMS-mutagenized various chromosome arms, were created and screened for abnormal neurogenesis or neuronal morphogenesis. Mutant chromosomes with interesting phenotypes were then recovered for future analysis.

After learning that the l(X)MB342 line was homozygous lethal, we first mapped the lethal mutation(s) using linkage analysis. Homologous recombination was induced between the FRT19A, UAS-mCD8-GFP mutant chromosome and an X chromosome carrying multiple visible mutations. Hemizygous male progeny were collected and analyzed for viable recombination patterns. Based on the recombination patterns, l(X)MB342 mutation(s) should be between y (1A5) and w (3C1). Three duplications, Dp(1;f)R,Dp(1;2;Y)w[+], and Dp(1;3)w[vco], were found to rescue the lethality caused by homozygous l(X)MB342. Dp(1;2;Y)w[+] rescued l(X)MB342 males were used to conduct complementation tests with related deficiency or mutant lines.

Following induction of mitotic recombination at selected stages, MARCM clones of GAL80-minus cells were created from heterozygous precursors and examined at various later stages. Organisms were dissected in cold phosphate-buffered saline and their brains were fixed and immunostained, as previously described (Lee and Luo,1999). MARCM clones were detected by the rat anti-mCD8 mAb (1:100,Caltag). Immunofluorescent signals were collected by confocal microscopy and then processed using Adobe Photoshop.

Fly brains were fixed and subjected to immunostaining following the procedures as described previously (Lee et al., 1999). Rabbit polyclonal antibodies against Lab (1:100), Pb(1:100) and Dfd (1:30) were kindly supplied by T. Kaufman. The mouse monoclonal antibodies anti-Abd-A (1:400) and anti-Ubx (1:30) were gifts from I. Duncan and J. Müller, respectively. Other mouse monoclonal antibodies used in this study, including anti-Abd-B (1A2E9) (1:100), anti-Antp (8C11)(1:100) and anti-Scr (6H4.1) (1:100), were obtained from the Developmental Studies Hybridoma Bank developed under Department of Biological Sciences, Iowa City, IA 52242.

The culture protocol was adopted from Gibbs and Truman(Gibbs and Truman, 1998). Briefly, the CNSs were dissected from mid-third instar larvae and, following removal of the ring glands, cultured in 200 μl of the Shield & Sang M3 Drosophila medium (Sigma) containing 7.5% heat-inactivated fetal calf serum (Gibco) and 1% of penicillin (10,000 units/ml)-streptomycin (10 mg/ml)solution (Sigma). Cultures were kept in a 25°C humidified culture incubator and aerated with a mixture of 95% air and 5% CO₂. Culture medium was changed every 48 hours. 20-hydroxyecdysone (20E) (Sigma) was dissolved in isopropanol with a concentration of 10 mg/ml to serve as a storage solution. A final concentration of 1 μg/ml 20E was achieved by adding this storage solution directly into culture medium. The same amount of isopropanol was added into the culture medium in the control experiments. After culture, the CNS was fixed in 4% paraformaldehyde in PBS.

With various GAL4 drivers, MARCM technologies allow for labeling of distinct specific clones of GAL80-minus neurons in otherwise GAL80 heterozygous brains (Lee and Luo,1999). For example, using GAL4-OK107(Connolly et al., 1996) in MARCM, one can selectively label clones of GAL80-minus neurons in the Drosophila olfactory learning/memory center, the mushroom bodies(MBs) (e.g. Lee et al., 1999). The MBs are paired neuropils (Fig. 1A). One MB is derived from four neuroblasts (Nbs), each of which continuously undergoes asymmetric divisions to produce a similar set of MB neurons through different developmental stages until fly eclosion(Ito et al., 1997; Lee et al., 1999). In the presence of GAL4-OK107, when mitotic recombination results in loss of GAL80 from one MB Nb, one would observe labeling of all the MB neurons subsequently derived from the GAL80-minus MB Nb(Fig. 1B). One MB Nb clone,generated in newly hatched larvae, consists of about 400 neurons at the adult stage and exhibits all basic MB morphological features. MB cell bodies are clustered on the posterior dorsal surface of the protocerebrum. Their dendrites elaborate immediately below cell bodies and constitute the calyx. Axons then project through the peduncle into various subsets of MB lobes(Fig. 1B).

To identify genes governing various aspects of neuronal development, we have been generating MARCM clones of MB neurons that are homozygous for various mutations in otherwise largely heterozygous organisms(Lee et al., 2000; Wang et al., 2002; Zheng et al., 2003). Following chemical mutagenesis, we screened about 800 X chromosomes(Fig. 1C) and recovered one mutated X chromosome that drastically altered the general patterns of MARCM clones (Fig. 1D). We no longer observed typical MB clones. Instead, there were multiple unrecognizable clones of cells in the mosaic brains. It appears that clones of homozygous mutant neurons undergo very rudimentary morphogenesis and fail to acquire any specific projection patterns (Fig. 1D). In addition, homozygous mutant cells might constitutively express GAL4 (see below), resulting in ubiquitous labeling of GAL80-minus clones despite use of a MB-selective GAL4 driver.

The mutant stock is homozygous lethal and probably carry one lethal mutation, l(X)MB342, responsible for most of the abnormal phenotypes. To determine how many lethal hits actually exist and roughly map where l(X)MB342 might be located, we conducted linkage analysis based on recombination patterns and frequencies between the hemizygous lethal phenotype and multiple X-chromosome recessive visible markers (see Materials and methods). We detected only one X-chromosome lethal hit around the 2 to 3 cytogenetic region. Presence of l(X)MB342 in the 2 to 3 cytogenetic region was further confirmed by rescuing hemizygous males with several duplicated genomic fragments spanning the 2 to 3 cytogenetic domain. Complementation experiments were then made possible between the rescued l(X)MB342 hemizygous male flies and various deficiency lines. We thus mapped the l(X)MB342 mutation to the interval between 2D3 and 2E1(Fig. 2A). Saturated mutagenesis followed by detailed genetic analysis has previously revealed five lethal complementation groups around this region(Perrimon et al., 1985). Interestingly, l(X)MB342 fails to complement with three of the five lethal groups. It appears that ph, PgD and wapl, which are located within a 40 kb genomic segment, are all mutated in the l(X)MB342 mutant X chromosome(Fig. 2A). However, no deletion could be detected in any of these open reading frames (data not shown). Furthermore, we found that a lethal P-element line, l(1)G0458(BL-10112), that carries a P-element insertion within the promoter region of ph-D (Peter et al.,2002), also fails to complement with ph, PgD and wapl mutations, suggesting possible presence of a local cis element essential for normal expression of several contiguous genes.

To identify which gene(s) are primarily involved in causing l(X)MB342 phenotypes, we conducted MARCM analysis for various known mutations, including ph⁶⁰², PgD^G0385 and wapl². Among them, only ph⁶⁰² allows us to obtain multiple unrecognizable clones of GAL4-positive and GAL80-minus cells after transient induction of mitotic recombination in newly hatched larvae. MARCM clones of ph⁶⁰² mutant cells grossly phenocopy l(X)MB342 mutant clones(Fig. 2B). Ph gene extends over a 28.6 kb genomic fragment and comprises two independent transcription units, ph proximal (ph-p) and phdistal (ph-d), which are direct tandem duplication(Dura et al., 1987). ph⁶⁰² is a null ph allele that carries a 2 kb deletion in ph-d and a complex genomic DNA rearrangement involving both deletion and inversion in ph-p(Boivin et al., 1999). Interestingly, a transgene that specifically encodes Ph-d, P{ph-d⁺} can largely rescue the mutant phenotypes of l(X)MB342 clones (Fig. 2C). This is not only consistent with the notion that Ph-P and Ph-D are functionally redundant (Dura et al., 1987; Deatrick et al.,1991; Saget and Randsholt,1994), but also suggest further that loss of Ph functions accounts for most, if not all, of the observed l(X)MB342 phenotypes. To examine specifically roles of Ph proteins in development of distinct neurons,we selectively focused phenotypic analysis on clones of ph⁶⁰² mutant neurons hereafter.

In insect brains, postmitotic neurons of the same lineages often remain associated; and distinct clones of neurons normally acquire different characteristic projection patterns (e.g. Fig. 1A,B; Fig. 3A,D,I,L). One can, thus,describe different clones of neurons based on the cell body positions and neurite trajectories of individual clones. Although most GAL4-OK107-labeled ph mutant neurons exist in discrete clones, we are no longer able to determine clonal identity because of a lack of recognizable neuropil structural features in the ph mutant clones. In addition, without Ph, the MB GAL4 driver, GAL4-OK107, appears labeling more GAL80-minus clones than it usually does(Fig. 3N, compared with 3J). In wild-type mosaic brains, GAL4-OK107 labels about 0.75±0.23 neuroblast clones per brain; and the majority of these clones consist of MB neurons. By contrast, there are 8.56±2.44 clones of ph mutant neurons per brain that could be labeled by GAL4-OK107 after comparable induction of mitotic recombination.

To gain additional insights into the neuronal functions of Ph, we repeated MARCM analysis of ph⁶⁰² using various GAL4 drivers. Different GAL4 drivers normally label different types of MARCM clones,depending on GAL4 expression patterns. Thus, while the pan-neuronal elav-GAL4 (Luo et al.,1994) allows for visualization of all neuronal clones, GAL4-OK107 (Connolly et al.,1996), GAL4-NP225 (gift from K. Ito), ato-GAL4(Hassan et al., 2000) and GAL4-EB1 (Wang et al.,2002) selectively label clones of MB neurons, antennal lobe projection neurons (PNs), Atonal-positive dorsal cluster (DC) neurons and specific ellipsoid body (EB) neurons, respectively, in wild-type mosaic brains(Fig. 3I-L; data not shown for GAL4-EB1). Interestingly, when such distinct GAL4 drivers were individually used to label ph mutant clones, we detected either all or none of ph mutant clones (Fig. 3M-P). First of all, similar numbers of elav-GAL4-labeled clones exist between wild-type (8.85±3.15) and ph(9.12±2.79) mosaic brains, suggesting that ph mutation does not affect patterns of mitotic recombination or initial neurogenesis (compare Fig. 3I with 3M). Second, GAL4-OK107 and GAL4-NP225 label similar numbers of ph mutant clones (8.56±2.44 and 7.87±2.62,respectively) as the pan-neuronal elav-GAL4 does, probably owing to ectopic activation of these GAL4 drivers in all ph mutant neurons(compare Fig. 3N and 3O with 3M). Third, ato-GAL4 and GAL4-EB1, by contrast,become suppressed and fail to label any ph mutant cells in adult mosaic brains (Fig. 3P). In summary, we no longer detect differential GAL4 expression in phmutant neurons of various origins. Moreover, instead of acquiring different characteristic projections, distinct clones of ph mutant neurons are typically found with similar simple bundles of neurites(Fig. 3M-O). Endogenous genes may also display either ubiquitous or no expression in ph mutant clones, as evidenced by lack of fasciculin II(Lin and Goodman, 1994)expression (based on immunostaining with the 1D4 monoclonal Ab) in all the examined ph mutant clones (data not shown). These phenomena collectively suggest that ph mutant neurons lose their individual identities and are uniformly transformed into indistinguishable abnormal cells.

Regulation of gene expression by PcG complexes is thought to occur via changes in local chromatin structures (reviewed by Levine et al., 2004). We wondered whether derepression versus inactivation of a given GAL4 driver depends on its specific genomic locus. Jumping an existing ato-GAL4transgene out of its original chromosome, we isolated 10 additional independent ato-GAL4 drivers that maintain strong GAL4 expression in Ato-positive CNS neurons but probably have the ato-GAL4 transgene inserted in 10 different genomic domains. We observe that none of these independent ato-GAL4 drivers is capable of labeling phmutant neurons in adult mosaic brains, yielding no evidence for involvement of local chromatin structures in constitutive silencing of atonal. Instead, silencing of ato-GAL4 in ph mutant neurons possibly occurs following derepression of some ato-GAL4 repressor(s).

Remarkably, the aforementioned ph mutant phenotypes are not observed prior to pupal formation. First, in mosaic larval brains, most subtype-specific GAL4 drivers label different characteristic phmutant clones. For example, although GAL4-NP225 failed to label anything (Fig. 3G), GAL4-OK107 and ato-GAL4 selectively labeled MB and DC clones, respectively (Fig. 3F,H). These observations suggest that derepression of GAL4-OK107 and inactivation of ato-GAL4 in phmutant neurons must occur later. Second, despite various degrees of morphological defects, distinct mutant clones roughly retained their normal patterns of projections at the wandering larval stage(Fig. 3E-H compare with 3A-D). Thus, ph mutant neurons remain largely recognizable until sometime after pupal formation. What a coincident for the loss of neuronal subtype identity to take place during metamorphosis when the prepupal ecdysone peak activates various transcriptional hierarchs in distinct types of cells.

If ph mutant neurons gradually transform regardless of ecdysone-mediated metamorphosis, one would observe similar phenotypes in all aged ph mutant clones no matter whether they were induced before or after the prepupal ecdysone peak. To examine possible involvement of ecdysone signaling in Ph-dependent maintenance of neuronal diversity, we generated clones of ph mutant neurons at late larval versus early pupal stages and checked their phenotypes 2 days and 2 weeks after eclosion. As expected,we obtained many clones within the optic lobes upon induction of mitotic recombination at the mid-3rd instar stage when optic lobe (OL) precursors were actively dividing (Fig. 4A). Distinct GAL4 drivers were again used to probe neuronal cell fates. For example, all GAL4-OK107-labeled OL cells are negative for GAL4-NP225 expression in wild-type mosaic brains (compare Fig. 4A with 4B). By contrast,OL clones of ph mutant neurons are positive for both GAL4-OK107 and GAL4-NP225 expression(Fig. 4C,D). Additional phenotypes are observed in the morphologies of ph mutant OL clones. For example, most mutant OL neurons are aberrantly aggregated and their bundled neurites fail to defasciculate into any recognizable optic lobe neuropils (Fig. 4C,D). We thus conclude that all the ph mutant neuronal clones generated prior to pupal formation have lost their original identities and evolved into undistinguishable clones at the adult stage. We then generated MARCM clones 1 day after pupal formation and, by great contrast, detected no indication for similar cell fate transformation even in much aged ph mutant neurons. MB Nbs, different from other Nbs, continue producing post-mitotic neurons through the entire pupal life and can be subject to mitotic recombination until eclosion (Lee et al.,1999). Unlike ph mutant OL clones, pupal-born MARCM clones of ph mutant MB neurons are strongly labeled by GAL4-OK107 while remaining negative for GAL4-NP225expression, even in 2-week-old adults (compare Fig. 4G with 4H). In addition,we see no problem in identifying pupal-born ph mutant MB neurons. They acquired basic MB neuronal trajectories despite subtle morphogenetic defects (Fig. 4G), reminiscent of larval-born ph mutant MB neurons examined before pupal formation(Fig. 3F). Taken together,these observations suggest that ph mutant neurons lose their subtype-characteristic gene expression and neurite projection patterns specifically during metamorphosis.

To examine further when distinct ph mutant neurons acquired the final uniform gene expression profile, we characterized both control and ph mosaic brains for patterns of MARCM clones at various pupal stages. When GAL4-NP225 was used to label MARCM clones, we detected only clones of PNs in wild-type mosaic brains but started to observe multiple unrecognizable GAL4-positive clones in ph mosaic brains around 1-2 days APF (Fig. 5). We observe similar phenomena in the clones of ph mutant neurons that were induced at the mid-3rd instar stage (data not shown). All these results suggest that loss of neuronal diversity in ph mutant clones specifically occurs during early metamorphosis and probably because of induction of unpatterned ecdysone-mediated transcriptional hierarchies (see Discussion).

To determine directly the role of ecdysone in the transformation of ph mutant neurons during early metamorphosis, we subsequently examined whether and how synthetic 20-hydroxyecdysone affects MARCM labeling of ph mutant clones in cultured fly brains. We first found that loss of neuronal diversity, as evidenced by appearance of multiple unrecognizable MARCM clones, inevitably occurred in ph mosaic brains that were cultured from the late wandering larval stage (empty gut stage) (data not shown). By contrast, transformation of ph mutant neurons depended on the availability of 20-hydroxyecdysone when culture of mosaic brains started before larval wandering (Fig. 6). We cultured ph mosaic brains for 4 days and detected no sign for loss of neuronal diversity in the absence of 20-hydroxyecdysone(Fig. 6A). But adding 20-hydroxyecdysone into the media from the beginning or even 4 days later efficiently led to transformation of ph mutant clones and appearance of multiple bizarre GFP-positive clones in cultured ph mosaic brains(Fig. 6B,C). Such ecdysone-dependent changes in the patterns of MARCM clones were never observed in wild-type mosaic brains that were cultured in parallel (data not shown). These results provide direct evidence for involvement of ecdysone signaling in transforming ph mutant neurons of various origins into a homogeneous population of abnormal cells.

Ph is best known for its involvement with other PcG proteins in maintaining the silent state of homeotic (Hox) genes (e.g. Beuchle et al., 2001; Choi et al., 2000). To determine roles of general PcG functions in neuronal development, we conducted loss-of-function mosaic analysis for two other PcG proteins, Polycomb (Pc) and Enhancer of zeste [E(z)]. Two distinct multimeric PcG complexes, which each contains different PcG proteins, have been identified(Levine et al., 2004). Multiple lines of evidence reveal that Ph forms complexes with Pc but not with E(z) (e.g. Franke et al.,1992; Shao et al.,1999). However, neither Pc nor E(z) mutant clones phenocopy Ph loss-of-function phenotypes(Fig. 7). First, we no longer observe ectopic expression of GAL4-OK107 in the neuronal clones, that are normally negative for GAL4-OK107 expression, within Pcor E(z) mosaic brains. This did help us identify variably deformed MB clones. Second, MB clones homozygous for these different PcG mutations exhibit distinct morphological anomalies. At the wandering larval stage, we specifically observe exuberant disorganized `dendrites' in Pc Nb clones (arrow in Fig. 7A). Preferential involvement of Pc in confining dendritic growth is further evidenced by acquisition of over-elaborated dendrites, but not axons, in the single-cell clones of Pc mutant adult MB neurons(Fig. 7F compared with 7E). By contrast, no specific gross defect could be detected in either larval E(z) mutant Nb clones (Fig. 7B) or adult E(z) mutant single-cell clones(Fig. 7G). In addition,although both Pc (Fig. 7C) and E(z) (Fig. 7D) mutant MB Nb clones exhibit many ectopic neurites at the adult stage, the presence of calyx-like structures in the E(z) (inset in Fig. 7D), but not Pc,clones, together with significant reductions in the numbers of normal-projecting axons strongly suggest that the wide-spreading neurites are mostly misguided axons, but not overshooting exuberant dendrites, in the adult E(z) mutant Nb clones (Fig. 7D). Interestingly, such defasciculation of neurites is in great contrast with the aberrantly aggregated neuronal bundles of Ph mutant clones (compare Fig. 7C,D with Fig. 1D). Finally, we observe ectopic labeling of glia-like cells only in E(z) mutant mosaic adult brains (Fig. 7H). This phenomenon suggests selective involvement of E(z) in governing glial identity. Taken together, distinct PcG proteins are differentially involved in regulating various aspects of brain development, suggesting specific roles of Ph in maintaining neuronal diversity especially through metamorphosis.

Derepression of Hox genes has been implicated in the acquisition of abnormal cell fates by PcG mutant cells outside the nervous system (e.g. Choi et al., 2000; Beuchle et al., 2001). We wondered whether similar mechanisms underlie abnormal development of PcG mutant neurons. Using immunohistochemistry, we characterized expression of eight Hox genes, including abd-A, abd-B, ubx, antp, pb, dfd, lab and scr, in various PcG mutant clones at the wandering larval stage. Distinct Hox genes are normally expressed in different characteristic segmental domains. For example, expression of abd-B and ubxis restricted to the terminus and one middle segment of the ventral ganglion,respectively (Fig. 8A,H). Examining whether any Hox gene became ectopically expressed in mosaic larval CNSs, we found that only abd-B was misexpressed in ph mutant clones, while misexpression of several Hox genes, including abd-A,abd-B and antp, occurred in Pc mutant clones(Fig. 8; Table 1). Furthermore, when we ectopically expressed abd-B in wild-type MARCM clones using GAL4-C155/UAS-abd-B, no obvious defects were found in the axonal projection patterns of various neurons (J.W., C.-H.J.L. and T.L.,unpublished). These results suggest little involvement of Hox genes in the unanimously abnormal responses of ph mutant neurons to ecdysone signaling during early metamorphosis. In addition, only antp became derepressed when E(z) was knocked out; and derepression of antp was restricted to about 20% of E(z) mutant clones(Table 1). Taken together,derepression of distinct Hox genes occurs in different PcG mutant clones,further suggesting differential involvement of distinct PcG proteins in governing neuronal identities through development.

Ph is well implicated in maintaining cell fates via controlling transcription of genes in distinct cell type-characteristic manners (e.g. Dura and Ingham, 1988). Deregulation of multiple genes aberrantly occurs in ph mutant tissues(e.g. Dura and Ingham, 1988; Beuchle et al., 2001). A similar mechanism probably underlies most of the abnormalities in phmutant neurons. In particular, there are multiple lines of evidence suggesting mal-expression of various subtype-specific GAL4 drivers in ph mutant clones. First, with respect to GAL4-OK107, GAL4-NP225 and elav-GAL4, the use of various GAL4 drivers results in labeling of similar numbers of clones(Fig. 3M-O). Second, we observe clones in the central brain versus the optic lobe, depending on when mitotic recombination is induced (compare Fig. 3N-O with Fig. 4C-D), which is the same as in wild-type mosaic brains (compare Fig. 3B with Fig. 4A). Third, ato-GAL4 and GAL4-EB1 fail to label any clone (Fig. 3P, data not shown for GAL4-EB1), arguing against constitutive expression of UAS-transgenes in mutant clones. Finally, examining clones through development reveals no evidence for derivation of some clones from other clones; and, instead, we constantly observe sudden labeling of full-sized clones shortly after a big ecdysone pulse(Fig. 5). Apparently, loss of Ph function alone is short of causing the full spectrum of abnormalities. Mass ecdysone is required for the pathological transformation of ph mutant neurons in the Drosophila brain, raising several interesting possibilities about mutual involvement between the epigenetic function of PcG and the global nuclear signaling of steroid hormones.

Distinct wild-type cells respond differentially to ecdysone (e.g. Thummel et al., 1990; Truman et al., 1994; Bender et al., 1997; Lee et al., 2000), but ph mutant neurons of distinct origins become no longer distinguishable after ecdysone signaling. Ecdysone mediates diverse biological activities partially via binding to different heterodimeric receptors (e.g. Schubiger et al., 1998; Lee et al., 2000; Cherbas et al., 2003). Its conventional receptors consist of the nuclear receptor superfamily members ecdysone receptor (EcR) and Ultraspiracle (USP; the Drosophila RXR)(Yao et al., 1992; Yao et al., 1993). There are three documented EcR isoforms (Talbot et al., 1993); and cells that express different EcR isoforms have been shown to undergo different changes in response to the prepupal ecdysone peak (Truman et al., 1994). For example, abundant EcR-B1 exists selectively in the neurons that remodel projections during early metamorphosis(Schubiger et al., 1998; Lee et al., 2000). As we detected no change in EcR expression patterns in ph mutant neurons(J.W., C.-H.J.L. and T.L., unpublished), it is unlikely that the aberrant responses of ph mutant neurons to the prepupal ecdysone peak occur as a result of derepression of specific EcR isoforms. In addition, derepression of multiple Hox genes appears not involved either. Nevertheless, given the involvement of Ph in silencing transcription, it remains possible that derepression of other unidentified genes directly re-programs ecdysone-induced transcriptional hierarchies, leading to transformation of ph mutant neurons. Alternatively, it is possible that loss of the epigenetic function of Ph may permit diffuse activation of prohibited genes by normal transcriptional hierarchies. Moreover, massive steroid hormone signaling might directly modify genomic imprinting when PcG functions are compromised.

It is generally thought that nuclear signaling of steroid hormones occurs routinely through the life of an organism without affecting the memories of most cells. However, high levels of sex hormones during pregnancy possibly alter gene expression patterns permanently in the mammary gland even after involution (Ginger et al.,2001). Exposure to some hormonally active reagents during early development also has the potential for imprinting long-lasting changes on the action of related hormones (Mena et al.,1992). In mammalian neurons, the estrogen receptor-α was further found to silence gene expression in an epigenetic fashion and via hypermethylation of the involved promoters(Zschocke et al., 2002). All these phenomena argue for the abilities of steroid hormones to modulate genomic imprint, at least, under certain circumstances. Ecdysone-dependent transformation of ph mutant neurons, thus, provides a possible model system for characterizing the epigenetic functions of steroid hormones in genetically malleable organisms. In addition, our demonstration of the unusual potent epigenetic effects of ecdysone in ph mutant neurons suggests complex mechanisms may underlie pathogenesis of other documented PcG loss-of-function phenotypes.

Both derepression and inactivation of genes occur in transformed ph mutant neurons, characterization of which offers some molecular insights into this status of transformation. First, we no longer detected the fine-tuning of gene expression in transformed cells; and all the examined drivers appeared either fully on or completely off. Second, on or off could not be simply attributed to the genomic locations of drivers, as evidenced by constitutive silencing of the multiple independently inserted atonal-GAL4 transgenes. Third, transformed cells retained neuron-type morphologies and remained positive for the neuron-specific gene elav;and ph mutant neurons had been earlier reported to acquire normal-looking neurites in culture (Smouse and Perrimon, 1990). Taken together, the transformation leads to loss of subtype identity without affecting basic neuronal fates, abolishes the genomic imprints governing fine controls over gene expression, and locks gene expression in `on' or `off' possibly in a promoter-autonomous manner (largely independent of its chromatin environment).

Finally, loss of Ph, Pc, versus E(z) results in distinct phenotypes in the developing fly brain. Differences in their underlying pathological mechanisms are well exemplified by differential derepression of distinct Hox genes in different PcG clones. In addition, for a given PcG mutation, patterns of Hox gene derepression vary from neural clones to wing disc clones(Beuchle et al., 2001) and visceral mesoderm (Choi et al.,2000). It remains to be elucidated how distinct PcG functions are governed in diverse cell type-characteristic manners.

We are grateful to J. Müller for Pc and E(z)mutations, to N. B. Randsholt for P{ph-d⁺}transgene, to K. Ito and H. Bellen for GAL4-NP225 and ato-GAL4, and to T. Kaufman, I. Duncan and J. Müller for antibodies against Hox proteins. We also thank P. Jones for critically reading the manuscript. This work is supported by the National Institutes of Health. T.L. is a Klingenstein Fellow.