Glial cells missing: A binary switch between neuronal and glial determination in drosophila

Cell, Vol. 82, 1025-1036, September 22, 1995, Copyright © 1995 by Celi Press

glial cells missing: A Binary Switch between Neuronal and Glial Determination in Drosophila Toshihiko Hosoya, * Kazunaga Takizawa, Koushi Nitta, and Yoshiki Hotta Department of Physics Graduate School of Science University of Tokyo Tokyo 113 Japan

Summary In the Drosophila CNS, both neurons and glia are derived from neuroblasts. We have identified a gene, glial cells missing (gcm), that encodes a novel nuclear protein expressed transiently in early glial cells. Its mutation causes presumptive glial cells to differentiate into neurons, whereas its ectopic expression forces virtually all CNS cells to become glial cells. Thus, gcm functions as a binary switch that turns on glial fate while inhibiting default neuronal fate of the neuroblasts and their progeny. Similar results are also obtained in the PNS. Analyses of the mutant revealed that "pioneer neurons" can find correct pathways without glial cells and that neurons and glia have a common molecular basis for individual identity. Introduction One of the major challenges in developmental biology is to understand how cellular diversity is created. A theoretically possible mechanism would be to determine two cell fates by the presence or absence of a single switch molecule. The nature of such a gene was clearly foretold by Waddington (1940). To define such a switch molecule experimentally, we require three pieces of evidence. First, the presence of the molecule is restricted to the cells that are to be "turned on." Second, in the absence of the molecule, all cells follow the "default" fate. Third, the ectopic presence of the molecule turns on all cells to the "alternative" fate. Under these criteria, only a few molecules have been shown to act as binary switches in major cell fate decisions. Here we report our findings of an additional gene in this class in Drosophila, one which functions as a binary switch to determine whether cells in the nervous system follow one of the two differentiation pathways, neuronal or glial. The Drosophila nervous system is a highly sophisticated model system for developmental neurobiology, with its well-characterized anatomy and powerful genetic techniques (reviewed by Goodman and Doe, 1993). The neurogliogenesis of the embryonic central nervous system (CNS) begins in the neuroectoderm with the invagination of about 29 identified neuroblasts and one glioblast (the

*Present address: Molecular Genetics Laboratory, University of Tokyo, Tokyo 113, Japan.

longitudinal glioblast [LGB]) per hemisegment. The neuroblasts generate several ganglion mother cells, each of which then divides to become two neurons. It was recently shown that some neuroblasts also produce glial cells. The neuroblast NBI-1 first generates two identified neurons, aCC and pCC. In abdominal segments, NBI-1 then generates three glial cells, including the A and B glial cells plus a cluster of neurons, indicating that NBI-1 is a neuroglioblast (Udolph et al., 1993). It is also strongly suggested that the abdominal N B6-4 divides symmetrically to produce the VUM support glial cell and a cell body gila (S. Higashijima et al., personal communication). In addition to these gilaproducing neuroblasts, the LGB divides several times to produce six to eight glial cells that constitute majority of the longitudinal glial cells (Doe et al., 1988; Jacobs et al., 1989; Goodman and Doe, 1993; Ito et al., 1995). The origins of other glial cells are mostly unknown, but they are likely to be derived from the ectodermal neuroblasts. In addition to these ectodermal cells, the mesectodermal precursors generate midline gila and additional neurons. The mature em bryonic CNS consists of about 200 neurons and 24-28 ectodermal glial cells per hemisegment plus a small number of mesectodermal midline gila. Many neurons and most of the glial cells have been identified by their locations, morphology, and gene expression patterns (e.g., reviewed by Goodman and Doe, 1993; for glial mappings and nomenclature, see also Kl&mbt and Goodman, 1991 ; Ito et al., 1995). Extensive genetic analyses have identified a number of genes involved in cell fate specification in the CNS. Among them, several genes, including elav, pointed, and repo (also known as rk2), are expressed either in neurons or in glial cells exclusively and have functions in differentiation steps specific for the corresponding cell types (e.g., Robinow and White, 1988; Kl&mbt, 1993; Klaes et al., 1994; Xiong et al., 1994; Campbell et al., 1994; Halter et al., 1995). However, none of the previously identified genes have been shown to be involved in neuronal/glial determination. The mutant we isolated is named glial cells missing (gcm) based on the initial finding that ectodermal glial cells are entirely absent. The gcm gene encodes a novel nuclear protein. First, we have shown that gcm is expressed exclusively in all ectodermal glial cells at the earliest stages. Second, in gcm loss-of-function mutant, the presumptive glial cells differentiate into neurons. Third, under gcm ectopic expression, virtually all CNS cells differentiate into glial cells. These results indicate that gcm functions as a binary switch in neuronal/glial determination. We also obtained consistent results in the peripheral nervous system (PNS), showing glial/neuronal determination in the PNS is also governed by the gcm gene. We obtained further insights in differentiation and functions of glial cells by examining the gcm phenotype. We have shown that the "pioneer neurons" can find their correct pathways independent of the presence of glial cells, while other follower axons show defects in fasciculation

Cell 1026

and pathfinding in the mutant nervous system. We also found that the presumptive glial cells that had transformed into neurons in g c m mutants behave in a stereotyped and reproducible manner, suggesting that each individual glial cell is transformed to a neuron with unique characteristics. Thus, we would like to suggest that neurons and glial cells have a common molecular mechanism for the determination of individual identity. During the preparation of our manuscript, we noticed that groups from the Bellen and Goodman laboratories had also identified the same gene. We have agreed to use the name glial cells missing (gcm). These two groups present results essentially consistent with ours supporting validity of our conclusions (Kania et al., 1995; Jones et al., 1995 [this issue of Cell]).

Results Isolation of gcm Mutants with Defects in the Embryonic Nervous System

C

D

Figure 1. The gcm Mutants Fail to Expressthe gcm Gene and Have Defects in the Embryonic Nervous System (A and B) The gcm mutants have defects in the formation of axonal tracts. Normal (A) and gcm' mutant (B) embryonicCNSs at stage 16 were stained with MAb BP102, which is specific for CNS axons. In a normal embryo (A), BP102 stains the ladder-likeCNS axon bundles. In a gcm 7 mutant (B), the longitudinal tracts have often much fewer axons. The transverse commissuresare sometimes fuzzy. Anterior is up. (C and D) The gcm transcriptswere not detectedin the mutants. Normal gcm~len~CyO heterozygousembryos(C) and gcm ~homozygous mutant embryos(D) at early stage 11 were stained for gcm rnRNA by in situ hybridization.The transcriptsweredetectedin CNS neurogenic regions in normal embryos(brownspots in [C]). Blue stripes in (C) are generatedby the lacZ gene on the balancerchromosome,indicating heterozygosity.In contrast,gcm mRNAwas not detectedin gcm 7mutants (D). Anterior is up, ventral to left. Scale bar is 10 p.m in (A) and (B) and 60 p.m in (C) and (D).

The mutant g c m was isolated in our enhancer detector screening designed to identify genes involved in cell fate determination in the embryonic CNS. We generated 100 lethal strains by insertion of P elements carrying the enhancer detector gene, lacZ. We screened these strains first for lacZ expression in the CNS to isolate candidate genes involved in CNS development. From these isolated strains, we screened for defects in CNS development by staining the axonal tracts with an antibody against horseradish peroxidase (HRP), which stains all neuronal cell surfaces, including axons (Jan and Jan, 1982). The original g c m allele, g c m P, showed serious defects in the formation of CNS axonal tracts. In vivo excisions of the P element generated three revertants and 17 lethal alleles with varied severity in axonal phenotype, indicating that the g c m chromosome has no other lethal mutations and that the g c m gene is in the vicinity of the P element insertion. Among them, g c m I had the most severe axonal defects. The longitudinal tracts of this allele have much fewer axons than normal in approximately 40% of the hemisegments (compare Figure 1A with Figure 1B). The transverse commissures are sometimes fuzzy. The g c m locus was mapped to cytogenetic position 30C on the second chromosome by in situ hybridization to polytene chromosomes. The mutant did not complement the lethality of Df(2L)3OA;C, a chromosome that has deficiency at 30A-30C (Lindsley and Zimm, 1992). To clone the g c m gene, we first isolated a genomic fragment adjacent to the P element insertion from the mutant genomic library and then cloned the wild-type genomic DNA around it (Figure 2A). By in situ hybridization and by Northern blot analysis, we searched a 15 kb genomic DNA for sequences transcribed in the embryonic CNS. Both analyses detected only one transcription unit adjacent to the insertion point (Figure 2A). Screening of a cDNA library of an early embryonic stage using this region as a probe resulted in the isolation of several cDNA clones. The longest clone was 2 kb, consistent with the results of Northern blot analysis. The 5' end of the cDNA was mapped 5 bp downstream of the insertion point (Figure 2A). Genomic

glial cells missing DeterminesGlial versus Neuronal Fate 1027

A

--~

~

~

gcml

gcmP

1Kb

D

gcm 5' ~

3'

B i Z~VLNGHPITi~PVPMPVPMPVPSPPATKSRV : : : : : : : : : : : : : 31 AIDWDINDSKMPSVGEFDDFNDWSNGHCRL 61

ZYSVQSDEARKHASGWAMRNTNNHNVNILK

91 KSCLGVLLCSAKCKLPNGASVHLRPAICDK 121

C

ARRKQOGKQCPNRNCNGRLEIQACRGHCGY

151

PVTHFWRRDGNGIYFQAKGTHDHPRPEAKG

181

STEARRLLAGGRRVRSLAVH
211

LSSLRPTKRRQAKTQSIQESKRRRMGASDVL

241

ETKp~=~~SS=H=STNFNQSQG

271

SYVPAGQGSVISQWNREIHYETEDPCYANG

301

MYSYDMLHSPLSAHSSTGSYYQENKPQQLQ

331

HSQYQgQLSPQQHVPVSYDPSQPI$SALQC

361

GMPSYEICDDTSSLTSSSGYCSEDYGYYNG

391

YLPNSLDVSNGSQSQNLSQDASIYTTSSEI

421

FSVFESTENGG~TSGVDLIYDEATAYQQH~

451

QQGTFPHLTNYQQEPQDQMQSADYyYSNTG

481

VDNSWNIQMDATYHPVNSTDPIYC*

1

100

200

300

400

500

+l I1[ IH/llllllllllllllfllllll III Illll II [I -I IIIII IIInllE[lll IIII IIIIIIIIII IIIIirl Figure 2. StructureofthegcmGeneandtheConceptuaIGcmProtein (A) Genomic map of gcm locus. About 12 kb of the genomic map near the P element insertion site is shown. The P element insertion is depictedby a triangle, with an arrow indicatingthe lacZ gene. Open and closed rectangles indicate the coding and noncoding regionsof the gcm exons, respectively.The intron is 75 bp long. The deletion in gcm' is shown at the top. Restrictionenzymes:B, BamHI;E, EcoRI; H, Hindlll. (B) Deducedamino acid sequenceof gcm. The conceptualaminoacid sequenceencodedby the gcm cDNA clone9 is shown.The gcm cDNA is 1960 bp long with an open reading frame of 504 amino acids. The predictedamino acid sequenceshowed no significant homologywith any known protein. It has two candidatemethioninesfor the initiation codon (italicized).The nucleartargeting signal is indicatedby double underlining. Single underlining indicates the basic cysteine-rich region. The MPVP repeats and the PEST-like sequence are indicated by broken double underlining. The homopolymeric amino acid stretches are depicted by broken single underlining. (C) Characteristicfeaturesof the primarystructureof the gcm protein. The N-terminalis to the left. The doubleand singleunderliningindicate the nuclear targeting sequence and the basic cysteine-rich region, respectively.Small linesin the upperand the lowerhalf of the rectangle indicate basic and acidic residues, respectively. deletion in the gcm I chromosome is schematically shown in Figure 2A. By in situ hybridization to embryos, we examined whether gcm mutants express the isolated gene. In normal

embryos heterozygous for gcm' and a balancer chromosome, mRNA of the gene is detected in the CNS neurogenic region (brown spots in Figure 1C), in addition to the blue stripes that are the reaction product generated by the lacZ gene on the balancer chromosome indicating heterozygosity. In contrast, mRNA of the gene was not detected in homozygous gcm' embryos (Figure 1 D). We also found that embryos homozygous for gcm Pexpress no detectable mRNA of the gene, although they have a slightly milder axonal phenotype than g c m t These results indicate that the isolated gene is gcm. Because these two alleles, gcm ~and gcm p, express no detectable gcm mRNA, we used them for further phenotype analysis. Since both alleles have intact gcm coding regions, we cannot exclude a possibility that these alleles express very small amount of normal gcm protein.

The gcm Gene Encodes a Novel Nuclear Protein The 1960 bp gcm cDNA contains an open reading frame of 504 amino acids (Figure 2B). The predicted amino acid sequence of the gcm cDNA has a distinct nuclear targeting sequence, (KJR)2-Xlo-(K/R)3_s(Dingwall and Laskey, 1991), shown by double underlining in Figures 2B and 2C. Other than the nuclear targeting sequence, gcm showed no significant homology with any known proteins. Thus, gcm is most likely a nuclear protein of a novel class. One remarkable feature of the protein is that most basic residues are located in the N-terminal half, comprising several basic regions (Figure 2C). On the other hand, acidic residues are distributed evenly throughout the protein (Figure 2C). Gcm has a highly basic cysteine-dch region (single underlining in Figures 2B and 2C). Gcm also has one PEST-like sequence, which may function as a signal for rapid degradation (Rogers et al., 1986), and three repeats of a sequence MPVP in the N-terminal region (both indicated by broken double underlining in Figure 2B). Homopolymeric stretches of glutamine, serine, glycine, and tyrosine residues are found in the C-terminal half of the protein (broken single underlining in Figure 2B). The open reading frame has two candidate initiation methionines (italicized in Figure 2B), whose preceding nucleotide sequence matches the proposed translation initiation sequence, (C/A)AA(A/C) (Cavener, 1987). The 3' untranslated region carries an mRNA instability motif, AUUUA (Shaw and Kamen, 1986).

Expression of gcm in Glial Cells at Early Stages We investigated gcm expression by in situ hybridization to embryos. In the embryonic CNS, we found that the expression of gcm mRNA is located in glial precursors and immature glial cells during a short period of gliogenesis. In the CNS, gcm mRNA is first detected in one cell per hemisegment at early stage 11 (Figure 3A; staging is according to Campos-Ortega and Hartenstein, 1985). This cell is the LGB, located in the most lateral position in the neurogenic region medial to the segmental furrow. At midstage 11, LGB divides symmetrically (thin arrowheads in Figure 3B); then, the progeny migrate medially. Two to three other cells per hemisegment express gcm mRNA at this stage. One is NB6-4 (closed arrows in Figure 3B),

Cell 1028

Figure 3. Expression of gcm mRNA and the Enhancer Detector lecZ (A-C) gcm mRNA is expressed in the glial pre~ i ~ ~ == cursors and in the immature glial cells. CNS neurogenic regions of normal embryos were stained for gcm mRNA by in situ hybridization. gcm rnRNA is first detected in the LGB in each hemisegment at early stage 11 (A). At midstage 11 (B), NB6-4 (closed arrow) and one additional cell (open arrow) start to express gcm. The thin arrowhead indicates the LGB, which has just divided symmetrically. At early stage 12 (C), gcm mRNA is detected in the medially migrating VU M support glial cell (closed arrow), which is one of the progeny of NB6-4. Expression pattern in other cells at this stage also matches that of developing glial cells. Throughout embryonic development, gcm mRNA is not detected in the midline region (thick arrowheads in [A]-[C]), indicating gcm is not expressed in midline gila. Small bars and thick arrowheads indicate segmental furrows and midline, respectively. Anterior is up. (D) gcm mRNA is also detected in the PNS neurogenic region of each segment at stage 10. This expression disappears at the beginning of stage 11. Anterior is up, ventral to the left. (E) The enhancer detector lacZ is expressed in glial cells in the same pattern as gcm mRNA. Normal gcmPl+heterozygous embryos at midstage 11 were stained with anti-lacZ antibody. At this stage, lacZ is detected in the progeny of the LGB, NB6-4, and in one additional cell per hemisegment (compare with [B]). (F) gcmPIgcmP mutant embryos at mid-stage 11 were stained with anti-lacZ antibody. The number and locations of the lacZ-positive cells are almost identical to those in normal gcmPl+ heterozygous embryos (compare with [E]), indicating that the glial cell divisions are normal in the mutant• Small bars, thick and thin arrowheads, and closed and open arrows in (E) and (F) indicate the same as those in (A)-(C). Scale bar is 10 ~m in (A)-(C), (E), and (F) and 100 I~m in (D).

A

A D

A

l

E

F

IA

A

judged by its location and behavior of the progeny. Just after it expresses gcm, NB6-4 divides symmetrically; then, one of the progeny, the presumptive VUM support glial cell, migrates medially, gcm mRNA is detected in the migrating presumptive VUM support glial cell (closed arrows in Figure 3C). The other one or two gcm-positive cells located around the anterior border of the segment (open arrow in Figure 3B) correspond to the immature glial cell(s) reported by Halter et al. (1995) (see Figure 4B in Halter et al., 1995). At early stage 12, gcm mRNA is detected in several other cells (see Figure 3C). Expression patterns in those cells match that of developing glial cells (Xiong et al., 1994; Campbell et el., 1994; Halter et al., 1995). The expression fades during stage 12 and is hardly detected at stage 13, when most of the glial cells have become postmitotic (Campbell et ai., 1994; Halter et al., 1995). We traced these gcm-positive cells to later developmental stages by marking them with the enhancer detector lacZ on the gcm p chromosome. The lacZ protein of the gcm p chromosome, diffuses into whole cell bodies to make entire cell morphology visible. We found that the gcmPl+ embryos express lacZ essentially in the same pattern as gcm mRNA (compare Figure 3E with Figure 3B), indicating that gcm-positive cells coexpress lacZ on the gcm p chromosome. Since lacZ protein is stable in Drosophila em-

bryos, it is detected in virtually all the ectodermal glial cells at a later stage (Figure 4C). We could detect neither gcm mRNA nor lacZ expression in the midline region (large arrowheads in Figures 3A-3C and 3E), indicating that neither is expressed in the midline glial cells, which are derived from the mesectoderm. These results indicate that gcm is expressed in all glial precursors and immature glial cells except those derived from the mesectoderm. In the brain neurogenic region, we also detected gcm mRNA expression during stage 11 (data not shown). In the larval brain, gcm enhancer trap lacZ is expressed in the optic lobe from late second instar to third instar (data not shown), when gliogenesis takes place (Winberg et al., 1992). It is very likely that these gcm- or lacZ-positive cells in the brain are glial precursors and immature glial cells. At stage 10, gcm mRNA is detected in the PNS neurogenic region of each segment (Figure 3D). This expression ceases at the beginning of stage 11. In gcmPl+ embryos, lacZ protein is detected in PNS neurogenic region in the same pattern as that of gcm mRNA. At later stages in the PNS, lacZ is detected in a subset of glial cells such as ligament cells of the chordotonal organ, support cells of dorsal bipolar dendritic cells, and peripheral glial cells (data not shown). Besides the neurogenic regions, gcm is also expressed

glial cells missing Determines Glial versus Neuronal Fate 1029

in the most anterior region of the presumptive mesoderm beginning at the cellular blastoderm stage (data not shown). Expression of gcm in this region fades during stage 11.

Mutation of gcm Transforms CNS Glial Cells into Neurons

Figure 4. Mutant Glial Cells Differentiate into Neurons with Unique Individual Characteristics (A and B) gcm mutation disrupts glial differentiation. Normal (A) and gcm' homozygous mutant (B) embryos at stage 15 were stained with the anti-repo antibody• In normal embryos (A), the antibody stains nuclei of most of the glial cells, except midline gila. Only glial cells on the dorsal surface are visible in this focal plane• In gcm mutants (B), virtually no cells express repo. (C) In gcmPl+ normal heterozygous embryos, the lacZ-positive early glial cells differentiate into mature glial cells• gcmPl+ normal heterozygous embryos at early stage 14 were stained with anti-lacZ antibody. LacZ is detected in virtually all glial cells, except in midline gila. In this picture, only glial cells on the dorsal surface of the CNS are in focus. The array of longitudinal glial cells is visible• A and B glial cells are indicated as A and B, respectively• (D) In gcm mutant embryos, the lacZ-positive presumptive glial cells differentiate into neurons. The CNS of gcmPlgcmPmutant embryos at early stage 14 were stained with anti-lacZ antibody. The number of lacZ-positive cells is about 25 per hemisegment, about the same as that of normal ectodermal glial cells (24-28). These lacZ-positive cells migrate to positions different from the normal pattern (C), and most of them have a neuron-like round shape. We have confirmed that at least 21 of them send out axons, strongly suggesting that they have differentiated into neurons• The location and axonal extension pattern of these lacZ-positive cells are highly stereotyped. In abdominal segments, seven to nine round-shaped cells located lateral to the anterior commissure send out axons medially. These axons fasciculate with each other to form a thick bundle and cross the midline along the posterior edge of the anterior commissure (closed arrow). Four to six cells located posterolateral to the posterior commissure send out axons anteromedially, which extend in the posterior commissure (open arrow). Six to eight cells located anterior of the anterior commissure send out axons posteromedially, which presumably join the anterior commissure (not in focus). At later stages, about four cells located medially (asterisk), just anterior to the anterior commissures, extend axons posteriorly (axons are not visible at this stage). In thoracic segments, the anterior commissure has two bundles of lacZ-positive axons

Since the expression pattern suggested an involvement of gcm in early glial differentiation, we first investigated glial cell divisions in the mutant. We used lacZ protein produced by the gem Pc h r o m o s o m e to visualize glial cells in the mutant. In the gcmPlgcm p mutant embryos, in which gcm mRNA expression is hardly detectable, the initial lacZ pattern is almost normal (compare Figure 3F with 3El. The final n u m b e r of lacZ-positive cells is about 25, almost the s a m e as that of normal ectodermal glial cells (24-28). Until early stage 12, the locations of these lacZ-positive cells in the mutant are very similar to the locations of lacZ-positive glial cells in normal gcmPl+ embryos. Thus, in gem mutants, the initial divisions of the presumptive glial cells are essentially normal, and these cells express the enhancer detector lacZ as in normal embryos. To e x a m i n e differentiation of mutant glial cells, we studied expression of a glial-specific protein, repo. Repo is a h o m e o d o m a i n protein, required for glial differentiation, that is expressed in all CNS glial cells except in the mesectodermal midline gila (Xiong et al., 1994; Campbell et al., 1994; Halter et al., 1995). Repo expression begins at early stage 11 in the glial progenitors and immature glial cells and then lasts to the end of e m b r y o n i c development. Figures 4A and 4B show photographs of late e m b r y o s stained with anti-repo antibody. In normal embryos, most of the glial cells e x c e p t the midline gila express repo (Figure 4A). In gcm ~ embryos, virtually no cells express repo (Figure 4B). In gcmPlgcm p mutant embryos, only a few of the lacZpositive cells express repo in each hemisegment. This leakage expression is rather stochastic, but it is apparently all or none. The mutant glial cells do not express repo even at earlier stages. On the other hand, gcm m R N A expression is normal in repo 3 mutant, a putative null allele of repo (Xiong et al., 1994; Halter et al., 1995), indicating that gcm is epistatic to repo (data not shown). We also e x a m i n e d expression of a putative transcription factor protein, prospero (pros). In normal embryos, pros is expressed in a subset of neurons and a subset of glial cells, including the longitudinal gila (Doe et al., 1991). In gcm 1 embryos, pros expression in neurons is almost normal, whereas its expression in glial cells is absent (data not shown). We further investigated the fate of the mutant glial cells. In normal gcm~l+ embryos, the lacZ-positive presumptive

(arrowhead). Two lacZ-positive cells in a medial region of the thoracic segments (circle) send out axons posterolaterally (slightly out of focus). These two axons do not fasciculate with each other. Small bars in (C) and (D) indicate the segmental boundaries. One thoracic segment (T3) and four abdominal segments (A1-A4) are shown. Anterior is up. Scale bar is 30 ~m in (A) and (B) and 20 p.m in (C) and (D).

Cell 1030

:s, : ~"

•;

•72 •

,•,•>:•"

g

""•"

:.,

F

,+-

,"

.

'

7

,

BIB

Figure 5. gcm Mutation Transforms Glial Cells to Neurons in the PNS (A and B) gcm mutation disrupts repo expression in the PNS glial cells. The PNSs of normal (A) and gcm ~ mutant (B) embryos were stained with anti-repo antibody. In the normal PNS (A), repo is expressed in several glial cells, such as in ligament cells of the chordotonal organ (open arrowhead), the support cell of dorsal bipolar dendritic cell (asterisk), and PC3 (circle), one of the peripheral glial cells. Repo expression in these cells is not detected in the gcm mutant (B). (C and D) The mutant ligament cells have neuron-like morphology• Nuclei of all PNS cells in normal (C) and in gcm ~ mutant (D) embryos were stained with the antibody against cpo. In normal embryos (C), five round-shaped nuclei of neurons (closed arrow) and five thin nuclei of ligament cells (open arrowhead) are stained in two rows in the ventral half of the chordotonal organ• In gcm mutants (D), there are 10 round cells (open arrow) at the ventral part of the chordotonal organ, indicating that the mutant ligament cells have a neuron-like shape. (E and F) The gcm mutants have an increased number of neurons in the chordotonal organ. Normal (E) and gcm I mutant (F) embryos were stained with the antibody against a neuron-specific protein, elav. In

glial cells differentiate into mature glial cells (Figure 4C). In g c m P I g c m P mutant embryos, the lacZ-positive presumptive glial cells fail to migrate to normal positions (Figure 4D). Their cell bodies have a round shape typical of neurons. We found that most of these lacZ-positive cells send out axons. At stage 14, lacZ-positive axon bundles running across the midline are visible in every segment (open and closed arrows and arrowhead in Figure 4D). This axonal lacZ expression is not simply caused by the increase of l a c Z gene dosage in g c m P I g c m P, because the same results were obtained with embryos heterozygous for g c m p and g c m 1, the latter of which express no detectable lacZ. By examining a number of hemisegments, we obtained the following results. First, the number of lacZ-positive cells in the mutant is about 25 per hemisegment, almost the same as that of the normal ectodermal glial cells (2428). Second, at least 21 lacZ-positive cells send out axons. Third, the axonal extension pattern is highly stereotyped and reproducible, not only from individual to individual, but from segment to segment, with minor differences between thoracic and abdominal segments (Figure 4D; see the legend for details). Thus, it is strongly suggested that in g c m mutants the lacZ-positive presumptive glial cells differentiate into neurons with unique characteristics. The g c m mutants also show general phenotype common to glial mutants. Normal CNS neuromere shortens at the end of the embryonic development, whereas the mutant CNS remains elongated (data not shown). This condensation defect has also been seen in other mutants with glial defects, such as p r o s and r e p o (Doe et al., 1991; Xiong et al., 1994; Campbell et al., 1994; Halter et al., 1995). The contour of the mutant neuromere is irregular compared with normal embryos (Figures 4A and 4B). The r e p o mutant shows a similar phenotype (Halter et al., 1995). The mesectodermal midline gila express neither g c m nor repo. We investigated their cell fate using an enhancer trap strain AA142, which expresses l a c Z in midline gila (Kl&mbt et al., 1991). AA142 stained normal number of cells in g c m ~ background, indicating that g c m mutation does not affect midline gila (data not shown).

Mutation of g c m Transforms PNS Glial Cells into Neurons In the normal PNS, repo is expressed in g c m - p o s i t i v e glial cells: the ligament cells of the lateral pentascolopidial chordotonal organ (LCH5), the support cells of the dorsal bipolar dendritic cell, and the peripheral glial cells (Figure 5A; Xiong et al., 1994; Campbell et al., 1994; Halter et al., 1995). In g c m mutants, repo expression in these PNS glial cells is not detected (Figure 5B), indicating g c m is required for differentiation of these cells.

normal embryos (E), elav is expressed in five neurons (closed arrow), while ligament cells do not express elav. In gcm I embryos, elavpositive cells increase in 15% of the chordotonal organs examined, indicating an increase in neurons• (F) shows an extreme case with 10 elav-positive cells (open arrow). All embryos are at stage 16. Anterior is left, dorsal up. Scale bar, 10 ixm.

glial cells missing DeterminesGlial versus NeuronalFate 1031

Figure 6. Ectopic Expression of gcm Promotes Glial Differentiation While Inhibiting Neuronal Differentiation Ectopicexpressionof gcm in the CNS was performed by generatingscaGAL; UAS-gcm embryos. (A) The CNS of an embryo with gcm ectopic expression, stained with anti-repo antibody. Under gcm ectopicexpression,80%-90% of the CNS cells express repo (comparewith a normalembryo,shownin Figure4A).The total numberof the CNS cells is not significantlyaltered. Most of these repo-positiveceils have elongated or irregular morphologytypical of glial ceils.Only t0%-20% of the CNS cellsare repo negativeand have a neuron-likeround shape. (B and C) CNS of a normalembryo(B) and an embryo with gcm ectopicexpression(C), stained with an antibodyagainsta neuron-specificprotein, elav. In normalembryos,all neurons,about 200 per hemisegment,expresselav (B). Under gcm ectopic expression,the numberof elav-positivecells is dramaticallyreducedto 10-30 per hemisegment(C). Mostof the elav.negativecells in (C) have a gila-likeelongatedor irregularshape,whereasall the elav-positivecells in (C) have a neuron-likeround shape. (D) CNS axons of an embryowith gcm ectopic expressionwas stainedwith BP102antibody.The numberof axons are much reducedcompared with a normalembryo(shown in Figure 1A). All embryosare at stage 15. Anterioris up. Scale bar, 30 I~m.

A

B

C

D

We further investigated fate of the ligament cells. The LCH5 is comprised of five chordotonal organs. Each chordotonal organ is comprised of one neuron and three glial cells: the ligament cell, cap cell, and scolopale. These four cells are derived from a single sensory organ precursor (Bodmer et al., 1989). We stained all cell nuclei of the PNS with an antibody against the couch potato (cpo) protein (Bellen et al., 1992). In normal embryos, five roundshaped nuclei of neurons (closed arrow in Figure 5C) and five thin-shaped nuclei of ligament cells (arrowhead in Figure 5C) are stained in two rows in the ventral half of the chordotonal organ. In gcm mutants, 10 round cells are visible at the ventral part of the chordotonal organ (open arrow in Figure 5D), indicating that the mutant ligament cells are transformed to cells indistinguishable from neurons. In normal LCH5, the five neurons express a neuronspecific protein, elav (Figure 5E; Robinow and White, 1988). In gcm ~ embryos, 15% (7 of 48) of chordotonal organs showed increase in elav-positive cells. Figure 5F shows an extreme case in which 10 elav-positive cells are visible, indicating that all the five ligament cells are transformed into elav-positive neurons. These results indicate that gcm mutation transforms repo-positive glial cells to neurons in the PNS as well as in the CNS.

Ectopic Expression of gem Transforms Neurons into Glial Cells To investigate the effect of gcm ectopic expression in neuronal cells, we used the GAL4-UAS system (Brand and Perrimon, 1993). The gcm cDNA was subcloned into the polycloning site of the pUAST vector and was then introduced into flies by germline transformation to establish UAS-gcm flies. The UAS-gcm flies were crossed to the scaGAL strain (Klaes et al., 1994), which carries a transgenic GAL4 gene driven by the promoter of a neuro-

genic gene, scabrous (sca) (Mlodzik et al., 1990). In scaGAL; UAS-gcm embryos, GAL4 protein is expressed in all the neuroblasts and their progeny (Klaes et al., 1994) and then binds to the upstream activation sequences (UASs) in the pUAST vector to activate the recombinant gcm gene. We first found that in these scaGAL; UAS-gcm embryos, all the neuroblasts become repo positive (data not shown). At stage 15, approximately 8 0 0 - 9 0 % of CNS cells express repo (Figure 6A; compare with a normal embryo shown in Figure 4A). Most of these repo-positive cells have elongated or irregular cell shapes typical of glial cells (e.g., Ito et al., 1995). Only 1 0 0 - 2 0 % of the CNS cells are repo negative, and as far as we could examine, all of them have round shapes typical of neurons. Thus, the number of glial cells increases significantly under gcm ectopic expression. We further investigated expression of a neuron-specific protein, elav. In normal embryos, all neurons, about 200 per hemisegment, express elav (Figure 6B; Robinow and White, 1988), whereas glial cells do not (Xiong et al., 1994; Campbell et al., 1994). Under gcm ectopic expression, only 10-30 cells per hemisegment, about 5%-15% of the normal number, are elav positive (Figure 6C). Thus, the number of elav-positive cells is reduced by 85%-95%. Most of the elav-negative cells have gila-like elongated or irregular shapes, whereas all the elav-positive cells have neuron-like round shapes. Embryos with gcm ectopic expression have much fewer axons (Figure 6D) compared with normal embryos (see Figure 1A). Those remaining neurons in embryos with gcm ectopic expression (Figure 6C and 6D) are likely those that have divided and differentiated before effective amount of gcm has accumulated. Because gcm ectopic expression do not significantly alter the total number of CNS cells, the increase of glial cells must be at the expense of neurons. Thus, gcm ectopic expression inhibits neuronal differentiation while promoting glial differentiation.

Cell 1032

Pioneer Neurons Can Find Correct Pathways in gcm Mutant

Figure 7. Pioneer Neurons Can Find Correct Pathways in the gcm Mutant, Whereas Other Follower Axons Have Serious Defects in Tract Formation Normal (A and C) and gcm' mutant (B and D) embryos at early stage 13 (A and B) and stage 16 (C and D) were stained with anti-fasciclin II antibody. (A and B) In stage 13 normal embryos (A), the anti-fasciclin II antibody stains anteriorly extending pCC axons (arrows in [A]) and laterally extending aCC axons (arrowheads in [A]), which pioneer the longitudinal pathway and the intersegmental nerve, respectively. In gcm 7mutants (B), axons of these pioneer neurons appear normal in most segments, indicating that they can find their correct pathway without the help of glial cells. (C and D) The gcm mutants show defects in axonal tract formation at late embryonic stages. At stage 16, when CNS development is almost complete, the anti-fasciclin II antibody stains three discrete bundles in the longitudinal connectives (C). The pCC axons are in the medial-most MP1 pathway. In gcm mutants (D), the three fasciclin II-positive bundles are often interrupted and appear defasciculated. The second medial-most FN3 pathway often show misroutings to lateral directions (open arrows in [D]). In normal embryos, axons of the motoneurons exit CNS in two nerve bundles, segmental nerve (closed arrow in [C]), and intersegmental nerve (arrowhead in [C]). These two nerves normally make contact at the border of CNS (exit junction; asterisk in [C]). In gcm mutant embryos (D), almost normal number of axons leave the longitudinal tracts. But they often exit the longitudinal tracts at abnormal positions in many thin bundles (open circles in [D]) or in one thick bundle (closed circle in [D]). Even when two nerves are formed,

In the Drosophila CNS, many glial cells are associated with axons and have been implicated to function in axonal guidance (e.g., reviewed by Goodman and Doe, 1993). To elucidate glial functions in axonal tract formation, we investigated axonal behavior in g c m mutant. The longitudinal glial cells have been proposed to have functions in guiding the pioneer neurons, which extend the first axons in the longitudinal pathway (e.g., Jacobs and Goodman, 1989; Jacobs, 1993; also reviewed by Goodman and Doe, 1993). We investigated axonal behavior of pioneer neurons using the anti-fasciclin II antibody (G. Helt and C. S. Goodman, personal communication). At early stage 13 in normal embryos, the anteriorly extending axon of a fasciclin II-positive pioneer neuron, pCC, reaches the adjacent anterior segment (arrows in Figure 7A). Slightly later, the MP1 neuron, another fasciclin II-positive pioneer neuron, extends its descending axon in the longitudinal pathway (data not shown). In g c m mutants, longitudinal glial cells do not exist since the lacZ-positive presumptive glial cells (see Figure 3F) fail to migrate to form the longitudinal array (see Figure 4D). However, pCC axons reach the next segments in most cases (arrows in Figure 7B). The MP1 neurons also extend their axons normally slightly later than the pCC axons (data not shown). Thus, these pioneer neurons can find their correct pathways even in the absence of the longitudinal glial cells. Since insect glial cells have also been implicated for motor nerve formation (Bastiani and Goodman, 1986; also reviewed by Goodman and Doe, 1993), we investigated the behavior of the axons that pioneer motor nerves. In a normal embryo, axons of the motoneurons exit CNS in two nerves, the segmental nerve (arrow in Figure 7C) and the intersegmental nerve (arrowhead in Figure 7C). The anti-fasciclin II antibody stains many of the motor axons, including the axons of the aCC neurons, which pioneer the intersegmental nerve (arrowheads in Figure 7A). We found that the aCC axons extend almost normally in the mutant (arrowheads in Figure 7B), indicating they can find their correct pathway without glial cells. Although pathfindings of the pioneer neurons appear normal, g c m mutants have serious defects in axonal tract formation at later stages. At stage 16, when CNS development is almost complete, the anti-fasciclin II antibody stains two to three discrete bundles in the longitudinal connectives (Figure 7C). The pCC axons are included in the most medial MP1 pathway. In the g c m mutant, the fasciclin II-positive bundles are often interrupted and appear defasciculated (Figure 7D). The second medial-most FN3 pathway often show misroutings to lateral directions (open arrows in Figure 7D). The g c m mutants also show defects in the formation of motor nerves. In g c m mutant embryos (Figure 7D), almost

they fail to make contact at the correct position and go up the body wall independently (closed arrow and arrowhead in [D]). Anterior is up. Scale bar is 20 I~m in (A) and (B) and 30 ~Lm in (C) and (D).

glial cells missing Determines Glial versus NeuronalFate 1033

normal number of axons leave the longitudinal tracts. But they often exit the longitudinal tracts at abnormal positions as many thin bundles (open circles in Figure 7D) or as only one thick bundle(closed circle in Figure 7D). In normal embryos, the segmental nerve and the intersegmental nerve make a contact at the margin of the neuromere (exit junction; asterisk in Figure 7C). In gcm mutants, even when two nerves are formed, they often do not make a contact; instead, they go up the body wall independently (arrow and open arrowhead in Figure 7D). Even with these abnormalities, almost the same number of axons finally leave the CNS and go up the lateral body wall as in normal embryos. Discussion Gcm Functions as a Binary Switch between Neuronal and Glial Determination We have isolated a Drosophila mutant, gcm, that showed serious defects in axon bundle formation. The isolation and sequencing of gcm cDNA revealed that gcm encodes a novel protein with a distinct nuclear targeting signal. We showed that gcm is expressed transiently in all glial cells except in the mesectodermal midline gila. Expression of gcm mRNA coincides closely with gliogenesis as it begins at early stage 11 and then fades during stage 12. We investigated glial cell fate in the mutant using the enhancer detector gene lacZ on the gcm p mutant chromosome, which is expressed in the glial cells in normal heterozygous embryos. In homozygous mutant embryos, the number and initial locations of the lacZ-positive cells are normal, indicating glial cell divisions are almost normal. But most of these mutant glial cells fail to express a gilaspecific protein, repo, indicating that they have defects in glial differentiation. They also show defects in cell migration. To our surprise, these mutant cells begin to send out axons in a stereotyped manner. These results strongly suggest that in gcm mutants, presumptive glial cells differentiate into neurons. We investigated the effect of gcm ectopic expression in the CNS using the GAL4-UAS system that is driven by the sca gene promoter. Under gcm ectopic expression, most of the CN S cells became repo positive, while number of elav-positive neurons decreased at least by 85% with concomitant reduction of axons. Thus, gcm ectopic expression promotes glial differentiation while inhibiting neuronal differentiation. From these results, we conclude that glial/neuronal determination in Drosophila is governed by a single molecule, gcm. All CNS cells are originally bipotential so that they can differentiate into both neurons and glial cells, with neuronal differentiation as a default. They differentiate into neurons when gcm is absent, while they differentiate into glial cells when gcm is present. Under gcm ectopic expression, most CNS cells differentiate into glial cells. We consider that virtually no cells have intermediate characteristics between neurons and gila from the following reasons. First, cells negative for both elav and repo must be very small in number, if there are any at all, since repo-negative cells are only 100--20%

in number, and they all have neuron-like round shapes, whereas most of elav-negative cells have gila-like morphology. Second, similarly, cells positive for both elav and repo must be very small in number, if there are any at all, since elav-positive cells are only 5%-15% in number, and they all have neuron-like round shapes, whereas most of repo-positive cells have gila-like morphology. These results strongly suggest that gcm essentially functions in an all-or-none manner. Thus, we propose that gcm acts as a binary switch in neuronal/glial determination. Differentiation of Individual Glial Cells Once determined to either a neuronal or glial fate, each cell differentiates into a cell with unique individual characteristics, e.g., cell shape, location, or axonal extension pattern. In gcm mutants, presumptive glial cells differentiate into neurons. By examining the behavior of more than 21 cells among 25 transformed cells per hemisegment, we conclude that patterns of their cell migration, axonal extension, and fasciculation are highly reproducible not only from individual to individual, but also from segment to segment with minor difference between thoracic and abdominal segments. This indicates that each individual glial cell is transformed into a neuron with unique characteristics. This can be explained by assuming that mechanisms that specify glial identity can also control cellular characteristics under abnormal differentiation into neurons. Thus, we would like to suggest that neurons and glial cells share a common molecular mechanism for their identity determ ination and th at this mechanism is independent of glial/neuronal determination mediated by gcm. Several genes, such as orthodenticle (otd), fushi tarazu (ftz), and pros, are expressed in the subsets of both neurons and glial cells and are required for their development (Finkelstein et al., 1990; Doe et al., 1988, 1991). These genes are good candidates for the components of such a common identity determination mechanism. This mechanism may also be shared by m uscles, since several molecules are expressed in specific subsets of both neurons and muscles to determine their specific interactions (Chiba et al., 1995; Nose et al., 1994). In the PNS, we obtained results highly consistent with those in the CNS. The two genes gcm and repo are expressed in the same PNS glial cells, such as ligament cells, the support cells of dorsal bipolar dendritic cell, and peripheral glial cells (Xiong et al., 1994; Campbell et al., 1994; Halter et al., 1995; this paper). In gcm mutants, repo expression in these cells is not detected. The mutant ligament cells have round shapes typical of neurons, and they often express a neuron-specific protein, elav, indicating their transformation into neurons. These results indicate that gcm also controls neuronai/glial determination in the PNS in a similar manner as in the CNS. In the chordotonal organ, gcm is expressed in the ligament cells, but is not expressed in the other two glial cells, the cap cell and the scolopale. Thus, cell fate determination of these cells must depend on other molecules. A gene called sanpodo is shown to be involved in the determination between scolopale and neuron (Salzberg et al., 1994). In the chordotonal organ, gcm most likely functions

Cell 1034

in neuronal/glial decision or glial identity specifications (or both) in combination with other genes. Involvement of other genes in neuronal/glial decision may explain the lower frequency of gila-to-neuron transformation in the mutant PNS than in the CNS. Among glial cells in the CN S, midline glial cells are different from other glial cells in many aspects. They are derived from mesectoderm and express neither g c m nor repo. In g c m mutants, midline glial cells appear normal, indicating gcm is not essential for their development. Thus, the mesectoderm-derived cells have different mechanisms for neuronal/glial determination, which is independent of gcm. This mechanism may be controlled, at least partly, by a segmentation gene engrailed(en), since en is required for correct glial/neuronal determination in grasshopper midline cells (Condron et al., 1994). At the postembryonic stage, we detected the expression of g c m enhancer detector lacZ in the larval optic lobe primordia, suggesting that gcm also functions in gliogenesis in the larval brain. Gcm may also function in glial/neuronal determination in the adult PNS, since neurogenesis and gliogenesis appear to be coupled in the wing discs (Giangrande et al., 1993; Giangrande, 1994). Glial Functions in Axonal Tract Formation In the Drosophila CNS, many glial cells are associated with axon bundles and have been implicated to be involved in nerve tract formation (e.g., reviewed by Goodman and Doe, 1993). In particular, the longitudinal glial cells have been considered to be important for axonal extension and guidance, since the pioneer neurons extend along them (Jacobs and Goodman, 1989). However, we have shown that pCC and MP1 pioneer neurons extend their axons apparently normally in the longitudinal direction in g c m mutants. Thus, some guidance mechanisms other than the longitudinal glial cells must lead these pioneer neurons to the correct longitudinal pathway. Glial cells have also been considered to be required for the pathfinding of motoneurons (Bastiani and Goodman, 1986). However, in g c m mutants, the pioneer axons for the intersegmental nerve leave the CNS apparently normally and then extend to the lateral direction. At later stages, almost normal numbers of axons leave the CNS to go up the lateral body wall. Therefore, it is strongly suggested that some guidance mechanisms other than glial cells lead these motoneurons to the lateral direction. Although pioneer neurons behave apparently normally, the mutant shows serious axonal defects at later stages. Both CNS axonal tracts and motor nerves show defasciculation and misroutings. These results suggest that although glial cells are dispensable for the early pathfinding of a small number of pioneer neurons, they have essential supportive functions in the correct tract formation of the majority of axons. Since these nerve tracts are associated with a number of glial cells in normal embryos, their defects in the mutants can be caused by loss of pathfinding cues or extension support supplied by the glial cells. On the other hand, the axonal defects can also be caused by the secondary abnormality in neurons, since glial cells have functions in development and maintenance of neurons

(Klaes et al., 1994; Buchanan and Benzer, 1993). To elucidate glial functions in more detail, it is necessary to analyze the phenotypes with individual glial cell eliminations. Gcm Is a Novel Protein That Controls Major Cell Fate Decision We have demonstrated the remarkable functions of gcm gene, but we do not know its mode of action. Gcm has no known motifs other than the nuclear targeting sequence. Its activity must lead to transcriptional regulation of a number of genes since gcm determines cell fate between two distinct cell types, neuron and glia. In this paper we have shown that gcm ectopic expression induces expression of repo, indicating that gcm activity can lead to transcriptional activation. We have also shown that g c m ectopic expression inhibits elav expression and the formation of axons, suggesting that gcm activity can lead to transcriptional inhibition. The mechanism of the transcriptional regulation by gcm can be direct or indirect. Since gcm is most likely a nuclear protein, we prefer the possibility that gcm is involved directly in transcriptional regulation. Two possibly functional features are found in gcm. One is the cysteine-rich region, which may bind metal to constitute a DNA-binding domain. The other is the highly basic region in the N-terminal half of the protein, which may function in interactions with nucleic acids. Supposing that gcm is a direct transcriptional regulator or a transcription factor, repo and elav are good candidates for its target genes. Gcm is a novel regulatory protein that functions as a binary switch in a critical bifurcation point in the nervous system development. A similar or homologous molecule may also play an essential role in the vertebrate nervous system, where such binary cell fate decisions also take place, including glial/neuronal determination from a common precursor. Experimental Procedures Enhancer Detector Screening In our laboratory100 lethal mutantstrainswere establishedby insertion of the P elementsHZ and PZ (Mlodzikand Hiromi, 1992)into the second and the third chromosomesof ry strain with the wild-typeCS background using the standardprotocol(Bier et al., 1989; Bellenet al., 1989).Theywere first balancedwith balancerchromosomesCyO and TM3. Embryosof these balancedstocks were stained for lacZ activity. Amongthem, gcmp was shown to expresslacZ activity in a small subsetof cells in the embryonicCNS. Priorto phenotypeanalysis, gcmPwas balancedon a balancerchromosomeen'CyO, which carriesa/acZ geneexpressedin the patternof segmentpolaritygene wing/ess. This lacZ gene was used to distinguish embryoswith the balancersfrom homozygousmutantembryosin subsequentanalyses. In vivo excisionof the P elementwas carriedout as described(Bellen et al., 1989). Antibody Staining Antibody stainingwas done as describedby Ito et al. (1995),except that stainedembryosweremountedin 90% glycerol.Stainingfor lacZ activitywas carriedout essentiallyaccordingto the protocolof Kl&mbt et al. (1991). Working concentrationsand referencesor sourcesof the primary antibodies were anti-HRP antibody (1:100; Cappel), MAb BP102 (1:100; generatedby A. Bieberand N. Patelin the Goodmanlaboratory), antibodyagainst RK2 (repo) (1:3000; Campbell et al., 1994), anti-pros antibody MR1A (1:100, generatedin the Doe laboratory),

glial cells missing Determines Glial versus Neuronal Fate 1035

anti-I~-galactosidase (1:1000; Cappel), anti-elav antibody 9F8A9 (1:5; O'Neill et al., 1994), anti-cpo antibody (1:100; Bellen et al., 1992), anti-fasciclin II antibody MAb 1 D4 (1:100; generated by G. Helt in the Goodman laboratory).

Ministry of Education, Science, and Culture. Part of this work was carried out at Molecular Genetics Laboratory of the University of Tokyo.

Molecular Cloning

References

All procedures for molecular cloning were essentially according to the standard protocols (Sambrook et al., 1989). For in situ hybridization to polytene chromosomes, see Masai et al. (1993) and references therein. Genomic libraries of the CS strain and the gcmP/CyO strain were constructed with the ZFIXII kit (Stratagene). From the library of gcm~/ CyO strain, a genomic DNA fragment adjacent to the insertion point of the P element was obtained using the 5' half of the/acZ gene as a probe. This fragment was used to screen the wild-type genomic library to obtain several clones, spanning the 20 kb around the P element insertion. They were digested with Hindlll or Xbal to be subcloned into the polycloning site of a plasmid vector pBluescript II SK(-) (Stratagene). These clones were further divided by digestion and used as probes to test embryonic expression both by Northern blot analysis and in situ hybridization. Both analysis detected only one transcription unit adjacent to the P element insertion. This region was used as a probe to screen 2 × 10s pfu of an embryonic cDNA library (3-12 hr AEL; made in Kornberg laboratory) to obtain 11 clones. One genomic clone and the two longest cDNA clones, clone 1 and clone 9, were sequenced in both directions using the Exolll deletion procedure and Sequenase II kit (U. S. Biochemicals). Homology search was carried out with the FASTA program (Pearson and Lipman, 1988) in GenBank.

In Situ Hybridization to Embryos In situ hybridization with digoxigenin-labeled antisense RNA probes was done essentially as described by Masai et al. (1993). Genomic DNA fragments and cDNA clones were subclonad into pBluescript II SK(-) (Stratagene) and then used as templates for RNA probe production using the DIG-RNA labeling kit (Boehringer Mann heim). For mRNA expression assay in the mutant, eggs of gcmVen~CyO flies were collected and stained for lacZ activity as described by Kl~mbt et al. (1991). Embryos were then devitellinated and subjected to further steps o1 in situ hybridization. Mutant homozygous embryos (gcml/gcm ') were distinguished from embryos with balancers (gcmVen'CyO, en~lCyO/ en'~CyO) by the absence of/acZ expression.

Ectopic Expression Using the GAL4-UAS System The gcm gene was ectopically expressed in all the embryonic neuroblasts and their progeny using the GAL4-UAS system (Brand and Perrimon, 1993). The MluI-Aflll fragment of gcm cD NA clone 9, which contains the entire coding region plus about 100 bp of 5' and 200 bp of 3' untranslated regions, was blunt-ended and inserted into the bluntended EcoRI site of the pUAST vector to form the P[UAS-gcm] construct. This construct was introduced into white flies by germline transformation according to the standard protocol (Spradling, 1986) to establish six UAS-gcm lines. These lines were either balanced or made homozygous and then crossed to scaGAL lines (Klaes et al., 1994).

Acknowledgments We thank Drs. B. W. Jones and C. S. Goodman for sharing their unpublished data and for fruitful discussions. We also acknowledge Dr. H. J. Bellen for allowing us access to his experimental data prior to publication. We thank Drs. A. Bush, A. Chiba, C. Q. Doe, Y. Hiromi, K. Ito, F. Matsuzaki, and H. Okano for thoughtful discussions. We would like to acknowledge Drs. B. W. Jones, K. Ito, and K. Mogami for critically reading the manuscript. We appreciate generous help from the many Drosophila scientists and institutes who provided us with antibodies and other materials. Among them are Drs. H. J. Bellen, C. S. Goodman, C. Kl&mbt, N. Perrimon, A. Tomlinson, and the Developmental Studies Hybridoma Bank of the University of Iowa. We acknowledge Ms. A. Ishiwa, Y. Fujioka, and M Seki for their technical assistance. We thank all the members of our laboratory for constant and encouraging discussions during the course of this study. This work was supported by a project grant to Y. H. from the Human Frontier Science Program Organization and by Grants-in-Aid for Priority Areas from the Japanese

Received August 15, 1995; revised August 29, 1995,

Bastiani, M. J., and Goodman, C. S. (1986). Guidance of neuronal growth cones in the grasshopper embryo. Ill. Recognition of specific glial pathways. J. Neurosci. 6, 3542-3551. Bellen, H. J., O'Kane, C. J., Wilson, C., Grossniklaus, U., Pearson, R. K., and Gehring, W. J. (1989). P-element-mediated enhancer detection: a versatile method to study development in Drosophila. Genes Dev. 3, 1288-1300. Bellen, H. J., Kooyer, S., D'Evelyn, D., and Pearlman, J. (1992). The

Drosophila Couch potato protein is expressed in nuclei of peripheral neuronal precursors and shows homology to RNA-binding proteins. Genes Dev. 6, 2125-2136. Bier, E., Vaessin, H., Shepherd, S., Lee, K., McCall, K., Barbel, S., Ackerman, L., Carretto, R., Uemura, T., Grail, E., Jan, L. Y., and Jan, Y. N. (1989). Searching for pattern and mutation in the Drosophila genome with a P-lacZ vector. Genes Dev. 3, 1273-1287. Bodmer, R., Carretto, R., and Jan, Y. N. (1989). Neurogenesis of the peripheral nervous system in Drosophila embryos: DNA replication patterns and cell lineages. Neuron 3, 21-32. Brand, A. H., and Perrimon, N. (1993). Targeted gene expression as a means of altering cell fates and generating dominant phenotypes. Development 118, 401-415. Buchanan, R. L., and Benzer, S. (1993). Defective glia in the Drosophila brain degeneration mutant drop-dead. Neuron 10, 839-850. Campbell, G., G6ring, H., Lin, T., Spana, E., Andersson, S., Doe, C. Q., and Tomlinson, A. (1994). RK2, a glial-specific homeodomain protein required for embryonic nerve cord condensation and viability in Drosophila. Development 120, 2957-2966. Campos-Ortega, J. A., and Hartenstein, V, (1985). The Embryonic Development of Drosophila melanogaster (Berlin: Springer-Verlag). Cavener, D. R. (1987). Comparison of the consensus sequence flanking translational start sites in Drosophila and vertebrates. Nucl. Acids Res. 15, 1353-1361. Chiba, A., Snow, P., Keshishian, H., and Hotta, Y. (1995). Fasciclin III as a synaptic target recognition molecule in Drosophila. Nature 374, 166-168. Condron, B. G., Patel, N. H., and Zinn, K. (1994). engrai/ed controls glial/neuronal cell fate decisions at the midline of the central nervous system. Neuron 13, 541-554. Dingwall, C., and Laskey, R. A. (1991). Nuclear targeting sequencesa consensus? Trends Biochem. Sci. 16, 478-481. Doe, C. Q., Hiromi, Y., Gehring, W. J., and Goodman, C. S. (1988). Expression and function of the segmentation gene fushi tarazu during Drosophila neurogenesis. Science 239, 170-175. Doe, C. Q., Chu-LaGraff, Q., Wright, D. M., and Scott, M. P. (1991). The prospero gene specifies cell fates in the Drosophila central nervous system. Cell 65, 451-464. Finkelstein, R., Smouse, D., Capaci, T. M., Spradling, A. C., and Perrimon, N. (1990). The orthoclentic/e gene encodes a novel homeo domain protein involved in the development of the Drosophila nervous system and ocellar visual structures. Genes Dev. 4, 1516-1527. Giangrande, A. (1994). Glia in the fly wing are clonally related to epithelial cells and use the nerve as a pathway for migration. Development 120, 523-534. Giangrande, A., Murray, M. A., and Palka, J. (1993), Development and organization of glial cells in the peripheral nervous system of Drosophila me/anogaster. Development 117, 895-904. Goodman, C. S., and Doe, C. Q. (1993). Embryonic development of the

Drosophila central nervous system, In The Development of Drosophila me/anogaster, Volume 2, M. Bate and A. Martinez Arias, eds. (Cold Spring Harbor, New York: Cold Spring Harbor Laboratory Press), pp. 1131-1206.

Cell 1036

Halter, D. A., Urban, J., Rickert, C., Ner, S. S., Ito, K., Travers, A. A., and Technau, G. M (1995). The homeobox gene repo is required for the differentiation and maintenance of glia function in the embryonic nervous system of Drosophila melanogaster. Development 121,317332. Ito, K., Urban, J., and Technau, G. M. (1995). Distribution, classification, and development of Drosophila glial cells in the late embryonic and early larval ventral nerve cord. Roux's Arch. Dev. Biol. 204, 284307. Jacobs, J. R. (1993). Perturbed glial scaffold formation precedes axon tract malformation in Drosophila mutants. J. Neurobiol. 24, 611-626. Jacobs, J. R., and Goodman, C. S. (1989). Embryonic development of axon pathways in the Drosophila CNS. I. A glial scaffold appears before the first growth cones. J. Neurosci. 9, 2402-2411. Jacobs, J. R., Hiromi, Y., Patel, N. H., and Goodman, C. S. (1989). Lineage, migration, and morphogenesis of longitudinal glia in the Drosophila CNS as revealed by a molecular lineage marker. Neuron 2, 1625-1631. Jan, L. Y., and Jan, Y. N. (1982). Antibodies to horseradish peroxidase as specific neuronal markers in Drosophila and in grasshopper embryos. Proc. Natl. Acad. Sci. USA 79, 2700-2704. Jones, B. W., Fetter, R. D., Tear, G., and Goodman, C. S. (1995). glial cells missing: a genetic switch that controls glial versus neural fate. Cell 82, this issue. Kania, A., Salzberg, A., Bhat, M., D'Evelyn, D., He, Y., Kiss, I., and Bellen, H. J. (1995). P-element mutations affecting embryonic peripheral nervous system development in Drosophila melanogaster. Genetics 139, 1663-1678. Klaes, A., Menne, T., Stollewerk, A., Scholz, H., and Kl&mbt, C. (1994), The Ets transcription factors encoded by the Drosophila gene pointed direct glial cell differentiation in the embryonic CNS. Cell 78, 149160. Kl&mbt, C. (1993). The Drosophila gene pointed encodes two ETS-like proteins which are involved in the development of the midline glial cells. Development 117, 163-176. Kl&mbt, C., and Goodman, C. S. (1991). The diversity and pattern of glia during axon pathway formation in the Drosophila embryo. Glia 4, 205-213. Kl~mbt, C., Jacobs, J. R., and Goodman, C. S. (1991). The midline of the Drosophila central nervous system: a model for the genetic analysis of cell fate, cell migration, and growth cone guidance. Cell 64, 801-815. Lindsley, D. L., and Zimm, G. G. (1992). The Genome of Drosophila melanogaster (San Diego, California: Academic Press). Masai, I., Okazaki, A., Hosoya, T., and Hotta, Y. (1993). Drosophila retinal degeneration A g e n e encodes an eye-specific diacylglycerol kinase with cysteine-rich zinc-finger motifs and ankyrin repeats. Proc. Natl. Acad. Sci. USA 90, 11157-11161. Mlodzik, M., and Hiromi, Y. (1992). Enhancer trap method in Drosophila: its application to neurobiology. Meth. Neurosci. 9, 397-414. Mlodzik, M., Baker, N. E., and Rubin, G. M (1990). Isolation and expression of scabrous, agene regulating neurogenesis in Drosophila. Genes Dev. 4, 1848-1861. Nose, A., Takeichi M., and Goodman, C. S. (1994). Ectopic expression of connectin reveals a repulsive function during growth cone guidance and synapse formation. Neuron 13, 525-539. O'Neill, E. M., Rebay, I., Tjian, R., and Rubin, G. M. (1994). The activities of two Ets-related transcription factors required for Drosophila eye development are modulated by the Ras/MAPK pathway, Cell 78, 137147. Pearson, W. R., and Lipman, D. J. (1988). Improved tools for biological sequence comparison. Proc. Natl. Acad. Sci. USA 85, 2444-2448. Robinow, S., and White, K. (1988). The locus elav of Drosophila mela. nogaster is expressed in neurons at all developmental stages. Dev. Biol. 126, 294-303. Rogers, S., Wells, R., and Rechsteiner M. (1986). Amino acid sequences common to rapidly degraded proteins: the PEST hypothesis. Science 234, 364-368.

Salzberg, A., D'Evelyn, D., Schulze, K. L., Lee, J.-K., Strumpf, D., Tsai, L., and Bellen, H. J. (1994). Mutations affecting the pattern of the PNS in Drosophila reveal novel aspects of neuronal development. Neuron 13, 269-287. Sambrook, J., Fritsch, E. F., and Maniatis, T. (1989). Molecular Cloning: A Laboratory Manual, Second Edition (Cold Spring Harbor, New York: Cold Spring Harbor Laboratory Press). Shaw, G., and Kamen, R. (1986). A conserved AU sequence from the 3' untranslated region of GM-CSF mRNA mediates selective mRNA degradation. Cell 46, 659-667. Spradling, A. C. (1986). P element-mediated transformation. In Drosophila: A Practical Approach, D. B. Roberts, ed. (Oxford: IRL Press), pp. 175-197. Udolph, G., Prokop, A., Bossing, T., and Technau, G. M. (1993). A common precursor for gila and neurons in the embryonic CNS of Drosophila gives rise to segment-specific lineage variants. Development 118, 765-775. Waddington, C. H. (1940). Organisers and Genes (Cambridge: Cambridge University Press). Winberg, M. L., Perez, S. E., and Steller, H. (1992). Generation and early differentiation of glial cells in the first optic ganglion of Drosophila melanogaster. Development 115, 903-911. Xiong, W.-C., Okano, H., Patel, N. H., Blendy, J. A., and Montell, C. (1994). repo encodes a glial-specific homeo domain protein required in the Drosophila nervous system. Genes Dev. 8, 981-994. GenBank Accession Number

The accession number for the sequence reported in this paper is D64040.