The retinal determination gene, dachshund, is required for mushroom body cell differentiation

Mushroom bodies (MB) are insect brain structures that are required for learning and memory. Flies without MB or with abnormal MB exhibit defects in associative learning paradigms such as olfactory classical conditioning (de Belle and Heisenberg, 1994; Heisenberg et al., 1985). Several proteins required for normal olfactory learning are preferentially expressed in the MB. These include the products of dunce, rutabaga and DCO, genes that encode components of the cAMP/PKA pathway (Han et al., 1992; Nighorn et al., 1991; Qiu and Davis, 1993; Skoulakis et al., 1993). These lines of evidence, along with results from parallel studies in other insects (Hammer and Menzel, 1995; Meller and Davis, 1996), emphasize the importance of MB for learning and memory in Drosophila.

The MB have a unique architecture. The cell bodies are located in the dorsal posterior brain above the neuropil areas housing their dendritic processes, which are known as the calyces. These unipolar neurons project anteriorly through the brain, sending dendritic branches into the calyces and continuing anteriorly as a nerve termed the peduncle. The peduncle diverges in the anterior brain into five discrete areas of neuropil known as lobes. The a and n’ lobes are columns of neuropil oriented vertically in the brain and the 0, 0’ and y lobes are columns oriented horizontally (Crittenden et al., 1998). This basic structure, in which the peduncle gives rise to vertically and horizontally oriented lobes, first emerges late in embryogenesis (Tettamanti et al., 1997).

The adult MB develop from four neuroblasts per brain hemisphere by the sequential generation of three classes of neurons (Crittenden et al., 1998; Ito and Hotta, 1992; Ito et al., 1997; Lee et al., 1999). The y neurons, which are unbranched and project their axons into the y lobe, are born from embryogenesis through the third larval instar (Armstrong et al., 1998; Lee et al., 1999). The o<0’ neurons are branched and project axon collaterals into both the n’ and 0’ lobes. These neurons are born from the third larval instar through puparium formation (Lee et al., 1999). The a0 neurons are also branched and project axon collaterals into the a and 0 lobes. These neurons are born after puparium formation (Lee et al., 1999). The larval MB are simpler in structure than adult MB in having only two lobes: a vertically oriented, a-type lobe and a horizontally oriented, 0-type lobe. Although y neurons in the adult are unbranched and project only into the y lobe, larval y neurons send projections into both the a-type and 0-type lobes (Lee et al., 1999). The larval y axons then degenerate early in pupariation and subsequently regrow as unbranched projections to form the adult y lobe (Lee et al., 1999; Technau and Heisenberg, 1982).

Four genes are known to be responsible for the determination of the Drosophila retina. Two of these genes, eyeless (ey) and sine-oculis (so), encode homeobox proteins. Two others, dac and eyes absent (eya), encode nuclear proteins that lack protein motifs suggestive of specific function (Bonini et al., 1993; Cheyette et al., 1994; Mardon et al., 1994; Quiring et al., 1994). Protein interaction studies have shown, however, that SO binds EYA, and EYA binds DAC (Chen et al., 1997; Pignoni et al., 1997). Furthermore, DAC and EYA are capable of activating transcription in yeast if fused to the GAL4 DNA binding domain (Chen et al., 1997). These studies have therefore suggested that EYA and DAC are cofactors for the transcription factor, SO.

A fascinating insight into the function of these genes was obtained from ectopic expression studies. Ectopic expression of EY, EYA or DAC produces ectopic eyes on the fly (Bonini et al., 1997; Chen et al., 1997; Halder et al., 1995; Shen and Mardon, 1997), and this capacity for eye determination is enhanced when most combinations of the four genes are combinatorially expressed (Chen et al., 1997, 1999; Pignoni et al., 1997). Furthermore, ectopic expression of a single retinal determination gene induces the expression of others (Bonini et al., 1997; Chen et al., 1997; Halder et al., 1998; Pignoni et al., 1997; Shen and Mardon, 1997). These previous findings have suggested a model in which EY functions earlier than the three other retinal determination genes, but through multiple feedback loops in which each protein can induce the expression of the others.

We show here that DAC is also required for the normal development of the mushroom bodies. The MB in dac mutants are disorganized and their processes demonstrate a failure in normal targeting. Our findings indicate that the retinal determination gene, dac, serves the differentiation of neurons in the central brain.

Frontal cryosections of adult heads were prepared and stained for lacZ as described by Han et al. (1992). For immunohistochemistry, flies were fixed in Carnoy’s fixative for 4 hours, dehydrated, cleared in methylbenzoate overnight and embedded in paraffin. Paraffin sections were dried overnight at 42°C. Paraffin was removed by immersing the slides in xylenes, followed by rehydration through a graded ethanol series and 1o minutes in PBHT (0.02 M PO₄, 0.5 M NaCl, 0.2% Triton X-100, pH 7.4). Slides were blocked in PBHT+NGS (PBHT + 5% normal goat serum) for 2-5 hours. Primary antibody incubations (overnight) were carried out in PBHT+NGS at the following dilutions: mouse anti-DAC^MAb1”¹ at 1:30 and rabbit anti-LEO (Skoulakis and Davis, 1996) at 1:2500. Horseradish peroxidase-based antibody detection was carried out with the Vectastain Elite ABC kit (Vector labs). For fluorescent detection, primary antibody incubation was in PBHT+NGS at the following dilutions: mouse anti-DAC^MAb1”¹ at 1:10, rabbit anti-LEO at 1:1000 and rabbit anti-DCO at 1:300. Secondary antibodies (FITC-conjugated goat anti-mouse and Texas Red®-conjugated goat anti-rabbit; from Rockland Antibodies) in PBHT+NGS were added at 1:200. Slides were mounted in Vectashield (Vector labs). For whole-mount immunohistochemistry, the CNS of larvae (or brains of adultsIpupae) were dissected in PP (1x PBS, 4% formaldehyde, 0.1% Triton X-100) and fixed for the following times: 20 minutes for L1, 30 minutes for L3, and 45 minutes for adultIpupal brains. After fixation, L3 CNS and brains were digested for 5-10 minutes in enzyme (collagenase, dispase and hyaluronidase, each at 1 mgIml in 1x PBS). The brains were blocked for 30 minutes in PBT (1x PBS, 0.1% Triton X-100, 0.1% BSA)+NGS, followed by primary antibody incubation overnight at 4°C. For fluorescent detection, mouse anti-DAC^MAb1”¹ was used at 1:10, rabbit anti-DCO at 1:500 and rat anti-BrdU at 1:10 (Harlan Sera-Lab). Fluorescent secondary antibodies (above) were used at 1:1000 and Texas Red®-conjugated goat anti-rat (Rockland) was used at 1:300. Staining of embryos was carried out using procedures described by Patel (1994) and alkaline phosphatase detection.

The dac Drosophila mutants were identified by their yellow body or mouthhook color from yw; daC^nullICyO^y+ stocks. Mutants were dissected from their pupal cases at the approximate time of normal eclosion by monitoring for wing darkening and the presence of abdominal bristles. Control animals of the same age were of the genotypeyw;+/+. The dac⁴ allele is a deletion of most of cytological band 36A and thus removes the entire dac locus. The eye and leg phenotypes of dac¹ and dac³ homozygotes are identical to those of dac⁴ homozygotes, indicating that all three alleles are phenotypic nulls (Mardon et al., 1994). In addition, these animals are protein nulls by immunohistochemistry.

Stocks for MARCM experiments were obtained from L. Luo (Stanford University). Animals of the genotype GAI,4^C1-⁵⁵. UAS-mCD8-GFP, hs-FLP; dac^nul1 FRT^4(MIGAL80 FRT⁴⁰-⁴ or GAI 4 ‘•’” UAS-mCD8-GFP, hs-FLP; dac⁺ FRT^4(MIGAL80 FRT⁴⁰-⁴ were collected within 2 hours of larval hatching and heat shocked for 30 minutes at 38°C.

P element lines for the enhancer detector screen were obtained from the Bloomington Drosophila Stock Center (Indiana University, Bloomington, USA), Berkeley Drosophila Genome Program (Torok et al., 1993), D. Glover (Deak et al., 1997), N. Perrimon (Harvard University), C. Goodman (University of California, Berkeley) and A. Schneiderman (Cornell University). DNA flanking the PZ element of the dac^P line was identified by inverse PCR and plasmid rescue.

First instar larvae collected within 2 hours of hatching were fed on sugar food containing 0.1 mgIml BrdU for 45 minutes, then immediately dissected.

Because of the importance of MB for olfactory learning, we screened 1643 enhancer detector lines for preferential expression of the reporter in these cells. The dac^P line was one selected in the screen (Fig. 1B). DNA flanking the enhancer detector element was isolated by inverse PCR and plasmid rescue, which revealed that the insertion occurred 11 base pairs downstream of the putative dac transcription start site (Fig. 1A). To confirm that the enhancer detector was accurately reflecting DAC expression, we stained adult fly heads with an anti-DAC monoclonal antibody. The antibody revealed robust DAC expression in the nuclei of MB neurons (Fig. 1C) along with scattered neurons throughout the brain. This robust expression in the MB was also observed for D. virilis (Fig. 1D), suggestive of a conserved function over 60-80 million years of evolution.

To gain insights into the role of DAC in MB development or function, we stained animals at various stages. DAC expression was evident in the MB nuclei at 48, 72 and 96 hours after puparium formation (Fig. 2A,B). It was also observed in the area of the MB in larvae at all stages (Fig. 2C) and embryos as young as stage 9 (Fig. 2D and not shown). The time of expression in embryos coincides with when MB neuroblasts (NB) segregate from the neuroepithelium and begin dividing to produce neurons (Younossi-Hartenstein et al., 1996). To conclusively identify the DAC-positive nuclei as those of MB neurons, we stained third instar larval brains for DAC along with DCO, a cytoplasmic marker for MB (Skoulakis et al., 1993). Optical sections through the brains revealed DCO in the MB soma, axons and dendrites, and DAC in the nuclei of these DCO-labeled cells (Fig. 2E). Thus, DAC expression in the MB begins in embryogenesis at about the time that MB neuroblasts segregate and begin dividing. This preferential expression remains into adulthood.

Because DAC plays a role in retinal determination, we questioned whether DAC was expressed in MB neuroblasts, their immediate descendants (ganglion mother cells, GMC), or in MB neurons. The MB NB are the only cells in the dorsal region of the brain that divide during the first 8 hours after larval hatching, so that BrdU is specifically incorporated into the MB NB and their progeny during this time (Ito and Hotta, 1992). We examined DAC expression in MB NB, GMC and neurons after a 45 minute feeding period. This time is sufficient for the MB NB, GMC and a few MB neurons (not shown) to incorporate BrdU (Fig. 2H). The NB were identified by their large and diffuse nuclei; the GMC by their smaller size and location next to an NB; and the MB neurons by their yet smaller size and proximity to an MB NB. None of the BrdU-positive cells, including the NB, GMC or young MB neurons, were labeled strongly with the anti-DAC antibody (Fig. 2F). Only MB neurons born during embryogenesis showed strong DAC expression. Therefore, DAC is strongly expressed in MB neurons that are 8-10 hours old (not shown), but not in NB, GMC or newly born MB neurons. This is consistent with a role for DAC in mature MB neurons, rather than in the developmental events such as cell division that lead to their birth.

The preferred expression of DAC in the post-mitotic MB neurons during embryonic, larval and pupal stages suggested that DAC might function in their differentiation. To test this idea, we examined the morphology of MB neurons in dac null mutants. The structure of the MB neurons in the mutants was grossly altered. The calyx and peduncle are normally well defined (Fig. 3C,H), but in the mutants these neuropil areas exhibited irregular edges due to a failure to contain MB processes to their normal domains (Fig. 3D,I). In addition, aberrant projections were often observed emanating from the calyx in the direction of the tips of the a lobes (Fig. 3D,I). In the wild type, the a lobe is thick and straight and is wrapped by the n’ lobe (Fig. 3J). In dac mutants, the dorsal projection was severely reduced in size and exhibited a posterior slant, giving the impression that either a or n’ was missing (Fig. 3K). Of the horizontally oriented lobes, g is the most dorsal and anterior, 0 is the most ventral, and 0’ is sandwiched between the two (Fig. 6B). In wild-type animals, these lobes are readily distinguishable (Fig. 3A). In dac null animals, the horizontal lobes appeared grossly disorganized (Fig. 3B). The 0, 0’ and y lobes were not distinguishable as individual lobes in either frontal or sagittal sections. In more anterior sections where only the y lobe is present, the dac mutants exhibited irregularities of the neuropil edges, again suggestive of a disregard to boundaries (3F,G). The disorganization of the horizontal lobes and the reduced vertical lobes are completely penetrant phenotypes, others are partially penetrant (Table 1).

The MB were not the only brain region disrupted in dac mutants; other regions of the brain were affected as well. The antennal lobe and much of the central complex appeared grossly normal in dac mutants, but the ellipsoid body (EB) of the central complex was notably disrupted. This structure appears as a circular hub with distinct lateral projections in wild-type animals (Fig. 5D). In dac mutants, the hubs were misshapen and seemingly fused with their lateral projections (Fig. 5F). The optic lobes, which consist of the lamina, medulla, lobula and lobula plate, are normally organized into a highly ordered array. In the mutants, the optic lobes were small and disorganized. Some of these effects may be secondary to the failure in retinal development and brain innervation.

The growth of MB axons occurs in two phases. The first is the initial outgrowth of y and a70’ axons and occurs throughout embryonic and larval development. The second begins early in pupariation when a/0 neurons initiate axon outgrowth. Also during this second period, y neurons retract their axons, which then regrow to help establish the adult MB morphology (Lee et al., 1999). We analyzed the MB of late stage dac larvae to determine whether DAC was required during both phases or just one. Only 5-10% of the dac larval brain hemispheres showed a marked size reduction of the a-type lobe, and the in the MB of dac null mutants (Brand and Dormand, 1995). If the MB defects were secondary to other brain defects, then expression of DAC in the MB would not be expected to provide rescue. Conversely, rescue of the MB phenotype would suggest that dac function in the MB is required for their proper development. GAL4^OK107 is an enhancer detector line that exhibits highly preferred expression in larval and pupal MB as well as in the median bundle (Fig. 5G). When the enhancer detector was used to drive the expression in a dac mutant background of either of two different dac cDNAs, UAS-dac-7C1 or UAS-dac-21 (Fig. 1A), the MB exhibited morphology very close to wild type (Fig. 5). This rescue was dependent upon temperature, reflecting the temperature-sensitivity of GAL4 activity (Table 3). Nevertheless, the rescue was not complete: the anterior part of the horizontal lobes was slightly disorganized. The incomplete rescue could be due to an improper level of DAC expression, inappropriate developmental onset of GAL4^OK107, or a requirement for more than one splice variant of dac. The dac mutant animals with defect was frequently unilateral (Fig. 4, Table 2). Not only was the phenotype dramatically less penetrant in larvae than in adults, it was also less severe. No aberrant projections were evident from the 0-type lobe, the peduncle or the calyx. Thus, the MB phenotype in larvae at late stages is both less penetrant and less severe than in adults of the same genotype. This indicates that dac is required during the second phase of axonal growth.

The presence of brain phenotypes outside the MB in dac mutants prompted us to ask whether DAC is required in the MB for their development or whether the MB phenotypes could be secondary to other defects. We tested this in two ways. First, we used the GAL4-UAS system to drive DAC expression only the UAS-dac transgene showed an MB phenotype equal in character and severity to that of dac null mutants (Table 3). Interestingly, the MB defects were corrected by GAL4-driven dac expression in the MB, other brain defects remained. The optic lobes were small and disorganized and the ellipsoid body was misshapen as in dac mutants (Fig. 5D-F). Thus, dac expression in the MB rescues the MB defects, but not other structural defects in the brains of dac mutants.

In a second approach, we removed dac expression from the MB in an otherwise wild-type brain by creating mosaic animals. The mutant cells were positively labeled using the MARCM system (Lee and Luo, 1999; Lee et al., 1999). For this analysis, a P element in which the tubulin promotor drives GAL80 (tubP-GAL80) was placed in trans to either a dac null allele or a wild-type chromosome (Fig. 6A). GAL80 represses the transcriptional activation of the UAS-mCD8-GFP marker that would normally be driven in all neurons by the neuronspecific driver, GAL4^C1JJ. Upon induction of a heat-inducible flp recombinase, mitotic recombination can result in the loss of GAL80 from one of the daughter cells, allowing the expression of the UAS-mCD8-GFP marker specifically in the cells of the clone (Fig. 6A). An MB neuroblast clone induced in early first instar larvae will produce all three types of MB neurons and represent approximately one quarter of the MB, since each MB develops from four NB and each contributes MB neurons that become part of all lobes (Ito et al., 1997; Lee et al., 1999).

We generated NB clones in the MB of dac heterozygotes by heat shock just after hatching, as depicted in Fig. 6A. The brains of the resulting adult animals were then stained with anti-mCD8 to reveal the dac mutant axons and anti-DCO to visualize the overall MB neuropil, but primarily the tips of the cd cd lobes where DCO staining is highest. We generated clones in control animals in a similar way, substituting the dac chromosome with a normal second chromosome. In each of the 11 control MB NB clones that we examined, the clonal axons were found distributed among all five lobes of the MB (Fig. 6C). Thus, the descendents of a normal MB NB invade and fill all neuropil compartments of the MB rather homogeneously. In contrast, dac mutant axons failed to fill the a lobes in 28 dac clones. This defect was not due to death of the oc/p neurons, since control and mutant neuroblast clones contained the same number of neurons as determined by cell counts (control=313±100, n=5; dac mutant=354±43, n=9). The four examples shown (Fig. 6D-G) illustrate that despite the a lobe defect, the axons do contribute to the cd lobe. In addition, in contrast to the homogeneous filling of the horizontal lobes by normal axons (Fig. 6C), the dac axons fill these neuropils in a disorganized way with substantial variability between clones (Fig. 6D-G). Therefore, dac mutant axons fail to become properly organized in the horizontal lobes and to fill the a lobe properly.

Because ey is upstream of dac in eye development and is also preferentially expressed in the MB (P. Callaerts and V. Hartenstein, personal communication), we hypothesized that dac and ey might function in the same pathway for the development of MB. If so, a synergistic effect on the phenotype might be detected when both genes are mutant.

The dac^P allele exhibits a mild MB phenotype in homozygous adults and the ey² allele produces no apparent MB phenotype in homozygous adults. No difference between the double homozygotes of genotype dac^P/dac^P; eyfley² and dac^P homozygotes was apparent (Fig. 7A-I). Nevertheless, the double mutants did show a genetic interaction in the eye. The eyes of ey² animals are variably reduced whereas dac^P animals have a consistent but less severe reduction in eye size. The eyes of the double mutants were consistently very small or absent (data not shown). We also examined dac^P homozygotes in combination with the more severe and lethal ey alleles, ey^D1Da and ey^JD. The MB of the dac^P/ dac^P; eyD^1Da/+ animals and dac^P / dac^P; ey^JD/+ animals exhibited an MB phenotype similar in severity to dac^P homozygotes. In addition, we examined double heterozygotes of the severe ey alleles ey^D1Da and ey^JD in combination with each of the three dac null alleles. The animals of each genotype had normal MB and eyes. Thus, we found no evidence for genetic interaction of dac and ey in MB development.

Since ey is genetically upstream of dac for eye development, we hypothesized that ey might be required for dac expression in the MB. To test this hypothesis, we stained sections of ey²²⁰ and ey°^1Da homozygotes with anti-DAC and anti-DCO antibodies. Although the number of MB neurons was reduced in ey^2D and ey^D1Da homozygotes, the level of DAC immunofluorescence in each remaining MB neuron appeared equivalent to that of Canton-S animals (Fig. 7J,K).

Because ey directs so transcription in the ey disc and so is required for dac expression (Chen et al., 1997; Halder et al., 1998), an absence of so from the MB might explain the apparent inability of ey to direct dac expression. We examined the expression of a so reporter, so⁷, in the central brain, using anti-b -gal immunofluorescence. The so⁷ reporter was expressed in cells just anterior to, but not in, the MB (Fig. 7L). The staining of so⁷ adult brains revealed the same pattern of P-gal expression, in neurons immediately anterior to the MB (data not shown).

Collectively, these results point to the likelihood that dac and ey participate in different aspects of MB development, unlike their unified role in retinal determination in the eye. This occurs in spite of the fact that ey, like dac, is preferentially expressed in MB and mutation of either results in abnormal MB development.

Our results describe a novel function for the retinal determination gene, dachshund, in mushroom body cell differentiation. Our data demonstrate that DAC is required autonomously within the MB for three major aspects of MB cell differentiation. First, DAC is required for MB neurites to respect their normal neuropil borders. Second, it is required for axons to arrange themselves properly within the horizontal lobes. Third, and most dramatically, DAC is required for MB axons to be able to fill the a lobe neuropil properly.

The mechanisms for dac involvement in these three aspects of MB cell differentiation are unknown, but we offer several hypotheses to account for the observations. DAC likely functions in the MB to direct the transcription of genes involved in these processes. This idea is consistent with our data showing that DAC is expressed abundantly in the nuclei of post-mitotic MB neurons, but not in the MB NB nor the GMC. In addition, the role ascribed to DAC in retinal determination is that of a cofactor for transcriptional activation. These observations support a conserved biochemical role for DAC as a transcriptional modulator in MB neurons.

But how do the cellular phenotypes arise? The failure of dac MB neurites to remain confined to their normal territory suggests that DAC may regulate the expression of cell surface molecules involved in territory recognition. These could be cell adhesion molecules that sequester the processes and prohibit excessive growth. Or, they could be cell surface receptors involved in the recognition of repulsive signals from boundary neuropil or glia. Similar ideas underlie the possibilities for the disorganization of axons within the medial lobes. The normal organization presumably involves attractive and repulsive interactions among the individual axons and their terminals and the postsynaptic elements. An imbalance in these forces could account for the failure of dac MB axons to fill the horizontal lobes homogeneously. The loss of axons from the a lobe in dac mutants suggests more specific ideas. The neurons that send projections to the a lobe may never extend neurites, or may retract them in the absence of DAC function. Or, the cause may be within the realm of appropriate targeting and pathfinding. The neurites destined to fill the a lobe may take the wrong turn without appropriate guidance cues and invade the medial lobes. Irrespective of which mechanisms ultimately prove to be true, the identification of dac as an important regulator of MB cell differentiation offers a foothold through which to probe the molecular events underlying MB development.

When is DAC required for normal mushroom body cell differentiation? Although dac null adults had severe MB phenotypes, most mutant larvae had no discernable phenotype. The 6% of mutant larvae that showed a phenotype had a reduced a-type lobe but no other defects. From these data we conclude that the phenotypes in dac larvae are both less severe and less penetrant than in adults. Furthermore, because the larvae were dissected within 5 hours of puparium formation, the majority of the MB structural defects must occur during pupariation. At puparium formation, the MB consists of both g and a’0’ neurons; the cd0 neurons are added after pupariation (Lee et al., 1999). The lack of severity and penetrance of the larval MB phenotype along with the inability of dac axons to contribute to the a lobe lead to the attractive proposition that the 010 neurons are particularly sensitive to the loss of DAC function. This sensitivity explains the onset of the phenotype as well as the failure to observe a lobe projections in the mutants. Nevertheless, dac mutant larvae do exhibit an MB phenotype with low penetrance, so the gene must have some role, albeit minor, in the development of g and a’/0’ neurons.

It is intriguing that a second retinal determination gene, ey, is also expressed preferentially in the MB and is required for their normal development. Although these genes operate in the same genetic pathway in the retina, we found no evidence for genetic interaction in mushroom body development. Indeed, their respective phenotypes are quite different. The ey mutations produce a dramatic reduction in the number of MB cells, potentially indicating a role for EY in MB cell survival, division or fate. In contrast, dac mutations have no discernable effect upon the number of MB cells produced but affect their post-mitotic differentiation. These observations suggest that dac and ey function independently in MB development. This conclusion is tempered by the fact that the ey alleles available are hypomorphs. However, consistent with independent roles, the two other retinal determination genes (eya and so) in the ey/dac pathway appear not to be expressed in the MB (Fig. 7L and Bonini et al., 1998).

Only a few genes have been identified so far that affect the overall development of mushroom bodies. The gene products of the mutants mushroom body defect, mushroom bodies deranged, mushroom bodies miniature and short stop, remain unknown (Lee and Luo, 1999; Heisenberg, 1985). The linotte gene has been shown to encode a receptor tyrosine kinase and mushroom bodies tiny, a protein related to P-21 activated kinase (Melzig et al., 1998; Moreau-Fauvarque et al., 1998). How these genes fit within the genetic cascade underlying MB development remains unknown. The identification of transcription factors and their accessory proteins, such as DAC and EY, that function early in the developmental hierarchy underlying MB development provides a starting point to dissecting the development of these important neural centers.

We would like to thank Saira Ahmed and Bradley Schroeder for technical assistance, and Ketia Endo for critical reading of the manuscript. We thank the Bloomington Stock Center, the Berkeley Drosophila Genome Project, D. Glover, N. Perrimon, P. Callaerts, C. Goodman and A. Schneiderman for fly stocks. Liqun Luo (Stanford) generously provided fly stocks and information about the MARCM technique. This work was supported by a training grant from the National Eye Institute and a predoctoral fellowship from the National Institutes for Mental Health (S. R. M.), a Damon Runyon-Walter Winchell postdoctoral fellowship (G. R.), grants from the NIH to G. M. and R. L. D. and the R. P. Doherty-Welch Chair in Science (R. L. D.).