Lobe mediates Notch signaling to control domain-specific growth in the Drosophila eye disc

The Drosophila eye is consisted of approximately 800 ommatidia and is subdivided into symmetrical dorsal and ventral domains. Its larval precursor, the eye imaginal disc, grows by cell division, increasing in number from about 130 cells at the end of the first instar to 9700 cells at the third instar (Wolff and Ready, 1993). Signaling events that occur at the DV boundary are required to coordinate this extensive amount of growth.

Dorsal and ventral domains are distinguished by the asymmetrical expression of dorsal-specific wingless (wg) and ventral-specific decapentaplegic (dpp) in early eye disc (Cho et al., 2000). Dorsal wg activates the dorsal-specific iroquois (iro) gene complex whose products restrict fringe (fng) expression to the ventral domain (Heberlein et al., 1998; Cho and Choi, 1998; Dominguez and de Celis, 1998; Yang et al., 1999; Maurrel-Zaffran and Treisman, 2000). Opposing fng⁺ and fng^– cells across the DV boundary establish the molecular mechanism that functionally defines the DV boundary. Fng promotes ventral cells along the DV boundary to express a N ligand, Ser, which in turn initiates a Ser-N-Delta (Dl) positive-feedback loop that activates N signaling (Kim et al., 1995; Huppert et al., 1997). Furthermore, Fng inhibits the Ser-dependent N activation in the rest of the ventral domain and potentiates N-Dl interactions at the DV boundary (Fleming et al., 1997; Panin et al., 1997). It does so by directly binding N (Ju et al., 2000) and enzymatically modifying N (Bruckner et al., 2000; Munro and Freeman, 2000). Thus, although N receptors are ubiquitously present in the eye disc, only those at the DV boundary are activated.

N activation at the DV boundary is required for the subsequent eye growth. Loss of N results in small-eye or no-eye (Cagan and Ready, 1995; Dominguez and de Celis, 1998; Papayannopoulos et al., 1998), and N gain-of-function induces non-autonomous overgrowth (de Celis et al., 1998; Go et al., 1998). The non-autonomous effect of midline N activation suggests the presence of a diffusible growth factor that is emitted from the DV boundary, but the identity of this factor and its downstream mediators remain unknown.

To better understand how midline N activation is coupled to the subsequent eye growth, we studied a Drosophila mutant with a small eye, L. The gene was first reported in 1925 and has been commonly used as a second chromosome dominant marker (Morgan et al., 1925). However, mechanisms that underlie its growth defect are little understood. We report the cloning of L and characterization of its functions. Remarkably, L functions are domain specific. L mediates the proliferative effect of N signaling only in the ventral domain, and loss of L specifically abolishes the ventral eye growth without affecting dorsal eye development. We also show that ventral-specific N ligand, Ser, plays dual functions in eye development. In addition to its role in DV boundary formation, Ser is required for the growth of the ventral eye disc, and its ventral expression is regulated by L. We propose that the eye disc is partitioned into dorsal, medial and ventral domains, the growth of which is independently regulated.

We used UAS-Ser^DN (R. Fleming), UAS-fng, UAS-N^intra, UAS-N^DN, Ser-lacZ, Ser^rev2-11FRT82B (S. Artavanis-Tsakonas), ey-GAL4, dpp-GAL4, N^264-43, UAS-mirr, L¹, L², L⁴, L⁵, L^si, L^rv5=In(2LR) 26F; 51F2-51A8, L^D, phyl²²¹ (B. J. Dickson), cpsf mutants (H. W. Brock), Asx¹ and Asx^P1 (D. Sinclair), l(2)03563, and two deficiency chromosomes that uncover L [Df(2R)trix 51A1-6; 51B6 and Df(2R)L4 51A2; 52A12-B1], which have been described by FlyBase.

Overlapping cosmids covering the genomic region between Asx and phyl were isolated from a genomic cosmid library (Tamkun et al., 1992). Additional cosmids covering the 51A region were obtained from the European Drosophila Genome Project. Standard germline transformations were carried out to generate transgenic lines that rescued L mutations (Rubin and Spradling, 1992). To generate antibody, cDNA coding the C-terminal 370 amino acids of L protein was cloned into pGEX vector (Invitrogen) and the fusion protein was used to immunize mice at Caltech monoclonal facility (Ou et al., 1993). Sequences were deposited with GenBank – Accession Number AF522076.

P-element mobilization was carried out by crossing P1201(ry⁺)/+; ry⁵⁰⁶[Δ2-3 ry⁺] Sb/+ mosaic male with L^si/L^si; ry females. L^si/+ flies have an anterior nick in the eye or wild-type eyes and L^si/Df(2R)trix flies have half-eyes. We selected P1201*/ L^si flies that had half-eyes and established multiple lines (*indicates a new insertion). P-element insertion site was then determined by sequencing the flanking genomic DNA isolated from inverse PCR. The P1201 line used for local transposition was reported to be homozygous lethal and have a P insertion at base 162 of CG10109, but it unexpectedly complements the deficiency chromosome of the region, Df(2R)trix (Spradling et al., 1999). We had re-identified the P insertion site of P1201 to be within the first intron, at 956 bp from the exon/intron junction.

L loss-of-function alleles were isolated from a screen for revertant of the L^D dominant half-eye phenotype. L^D/+ male flies were irradiated with 4000 rads γ-ray and crossed to wild-type females. Four revertants were isolated from a screen of approximately 40,000 progenies. One of which, L^rev6-3, failed to complement Df(2R)trix or L alleles but complemented multiple alleles of Asx, cpsf and l(2)03563, known mutations adjacent to L. Hetero-allelic combination of L^rev6-3 and other eye-specific L alleles generated ventral-reduced eyes that were indistinguishable from those of eye-specific L alleles over Df(2R)trix.

yw hsFLP122; L^rev6-3 FRT42D/arm-lacZ FRT42D and yw hsFLP122; Ser^rev2-11FRT82B/arm-lacZ FRT82B larvae were heat shocked at 37°C for 1 hour either at 24 hours after egg laying (AEL), during the first instar stage or 48 hours AEL (during the second instar stage), and dissected after a 24 hour interval or at the wandering larvae stage. Mutant clones were identified by the absence of β-galactosidase protein.

UAS-L construct was made by cloning full-length L cDNA into the transformation vector pUAST (Pirrota, 1988). To generate clones of cells ectopically expressing L or Ser^DN, first instar larvae from the cross between UAS-L (or other UAS flies); UAS-lacZ males and yw P[Actin>CD2>GAL4; w]; hs-Flp,MKRS/TM6Tb females were heat shocked at 37°C for 1 hour at 0 or 24 hour after hatching. The larvae were dissected either 24 hours after the heat shock or at the wandering larval stage, in order to visualize early first instar clones and late clones. Expression clones were identified by the presence of β-galactosidase protein.

Dissected imaginal discs were stained with either diaminobenzidine (DAB) or immunofluorescent markers as described (Choi et al., 1996). Primary antibodies used were mouse anti-L (1:20), mouse anti-β-galactosidase (1:250; Promega), rabbit anti-β-galactosidase (1:200; Cappel), rabbit anti-Dlg (1:1000; K. Cho), guinea pig anti-Dlg (P. Bryant), rabbit anti-Ser (1:20; E. Knust) and mouse anti-Elav. Secondary antibodies were horse radish peroxidase-conjugated anti-rabbit IgG, anti-rabbit fluorescein isothocyanate (FITC) and anti-mouse CY3 (Jackson Laboratory). Fluorescent images were taken with Zeiss LSM laser-scanning confocal microscope.

With variable severities and penetrances, all L alleles examined (L¹, L², L⁴, L⁵ and L^si) as heterozygotes exhibit a nick near the anterior midline of the eye, and the overall size of the eye is slightly reduced. As homozygotes, eye size is greatly reduced, primarily in the ventral domain. L^si allele typifies the observed phenotypes with the highest penetrance. Homozygous L^si animals show a preferential loss of the ventral eye with complete penetrance (Fig. 1B), and ∼70% of L^si/+ animals have an anterior nick in one or both eyes (Fig. 2G). The preferential loss of the ventral eye is also apparent in the eye imaginal disc morphology (Fig. 1D). Importantly, as homozygous L^si mutants are viable and its half-eye mutant phenotype is indistinguishable from that of L^si over deficiency chromosomes, this suggests that L^si is a strong eye-specific allele that minimally affects the development of other tissues.

To assess the extent of ventral eye loss in L^si homozygotes, we used an enhancer trap line, mirr^B1-12 (mirr-lacZ), which has a dorsal-specific expression of white (w) reporter gene (Fig. 1E) (Choi et al., 1996). In w^–; L^si/L^si; mirr^B1-12/+ flies, the overall eye size was reduced, and all but one or two rows of remaining ommatidia were w⁺, suggesting that most, if not all, of the ommatidia were dorsal (Fig. 1F). The dorsal polarity was confirmed in adult-eye sections (data not shown). Eye imaginal discs from staged w^–; L^si/L^si; mirr^B1-12/+ larvae showed ventral domain reduction starting at early second instar (Fig. 1H), indicating that L functions are required for early eye development.

To further study the wild-type function of L, we generated clones of L^– cells by mitotic recombination (Xu and Rubin, 1993), using a loss-of-function allele, L^rev6-3 (see Materials and Methods). Homozygous L^rev6-3 embryos with no detectable L protein expression fail to complete germ-band retraction and show no cuticle formation. L^rev6-3 clones caused distinct eye phenotypes depending on the time of clone induction and the location of the clones. In one scheme, clones were induced at the first instar stage. In the resulting third instar eye discs that contained ventral L^rev6-3 clones, the ventral eye disc was greatly reduced, and to a large extent the L^rev6-3/+ and +/+ tissue disappeared together with the L^rev6-3/L^rev6-3 tissue (Fig. 1I). By contrast, dorsal clones of considerable size did not cause obvious size reduction in the dorsal eye, nor did the mutant clones significantly affect the ensuing photoreceptor differentiation and polarity determination (Fig. 1I). Consistent with the eye disc phenotype, adult mosaic eyes had a relatively normal appearing dorsal domain, while most of the ventral region was replaced by the cuticle (Fig. 1J). In related experiments, mitotic recombination induced at late second and third instar stages generated multiple clones in both dorsal and ventral domains but did not result in any obvious eye defects (data not shown), suggesting an early, transient requirement of L. All together, studies of L^si phenotype and L^rev6-3 clonal phenotype indicate that L is non-autonomously required for the ventral eye growth but not so in the dorsal, and its functions are required during early stages of eye development.

Previous studies have shown that N activation at the DV boundary is vital for eye disc growth (Papayannopoulos et al., 1998; Cho and Choi, 1998; Dominguez and de Celis, 1998). As L is required specifically for ventral growth, it raises the possibility that L may mediate the proliferative effect of midline N signaling in the ventral eye. We used the Gal4-UAS system (Brand and Perrimon, 1993) to test this hypothesis. Overexpression of a constitutively active N (N^intra) (Fortini et al., 1993) by the dpp-Gal4 driver, which drives expression along the posterior edge of the eye disc, caused gross overgrowth of the eye in both dorsal and ventral domains (compare wild type in Fig. 2A with 2B). Reducing L gene dose strongly suppressed the ventral overgrowth but had much less of an effect on dorsal overgrowth (Fig. 2D). This ventral-specific suppression of N gain-of-function phenotype suggests that L acts downstream of N.

In contrast to N-induced overgrowth, eliminating N signaling by expressing a dominant-negative form of N (N^DN) (Klein et al., 1997) using the eyeless (ey)-Gal4 driver consistently resulted in small-eye or no-eye (Fig. 2E). ey-Gal4 drives Gal4 expression in early eye discs and anterior to the furrow in the third instar discs. Co-expression of L and N^DN partially suppressed this N^DN overexpression phenotype in the ventral domain: ventral eye was selectively restored in close to 20% of ey-Gal4/UAS-N^DN UAS-L animals (19/86 animals; Fig. 2F). The size of the restored ventral eye was either smaller or equal to the reduced dorsal eye, and in no instances did we detect ventral tissue without the presence of at least some dorsal tissue. The presence of residual dorsal eye indicates that N^DN overexpression may not completely eliminate endogenous N functions. It also suggests that N activity, even at a low level, is a prerequisite for L to induce ventral proliferation.

Given that the requirement of L functions is early and transient, the suppression by L of N^DN phenotype may be specific to undifferentiated cells. That is indeed the case. We overexpressed N^DN using GMR-Gal4 that induces Gal4 expression in all cells posterior to the furrow. GMR-Gal4/UAS-N^DN animals showed a rough eye phenotype with a relatively normal eye size, and the eye roughness was not suppressed by overexpression of L in GMR-Gal4/UAS-N^DN UAS-L animals (data not shown).

If L and N act in the same pathway, transheterozygous mutations of these two genes may result in enhanced phenotypes. L^si/+ flies have nicks at the anterior edge of the eye (Fig. 2G), but the defect is not so severe to result in half-eyes. Loss of one copy of N does not cause visible eye defects (Grimwade et al., 1985). Transheterozygote N^264-47/+; L^si/+ adults, however, had half-eyes in one or both eyes with approximately 50% penetrance (Fig. 2H). We had also observed similar enhancement of L phenotype by mutations in Enhancer of split, a major downstream effector of N signaling (data not shown). In summary, genetic interactions between L and N support the hypothesis that L mediates the proliferative effect of N signaling specifically in the ventral domain.

It is likely that L interacts with other ventral-specific genes, and one candidate gene is Ser, a N ligand, whose expression is ventrally enriched in the wild-type first instar eye disc (Fig. 3A) (Cho et al., 2000). Ser is required for eye development as Ser loss-of-function mutants have small eyes (Speicher et al., 1994; Go et al., 1998). To understand possible regulatory relations between L and Ser, we first examined Ser expression in first instar eye discs from L^si homozygotes. In these eye discs, ventral Ser expression was greatly reduced, but notably the expression along the DV boundary was not affected (Fig. 3B, bracket). We used both a Ser-lacZ enhancer trap line and an anti-Ser antibody to detect Ser expression in these L^si homozygotes eye discs, and similar Ser expression patterns were observed (data not shown). This observation suggests that L might promote ventral Ser expression except in regions near the DV boundary. In addition, as L^si homozygotes eventually lose its ventral eye, loss of ventral Ser expression in these mutants suggests that Ser may positively regulate ventral eye growth.

We induced L null clones to further test the hypothesis that L is a region-specific, positive regulator of Ser expression in first instar eye discs. L^rev6-3 clones altered Ser expression in a position-dependent way. Ventral L^– clones away from the DV boundary showed decreased Ser expression within the clone (four clones scored; Fig. 3D-F, white arrow), but clones near the posterior DV boundary had no such effects (three clones scored; Fig. 3D-F, yellow arrow). The effect of L^– cells on Ser expression appeared to be restricted within the clone as Ser expression outside of the clone was not visibly affected (Fig. 3D-F, white arrow). To test if L positively regulates Ser expression, we used the flp-out system (Pignoni and Zipursky, 1997) to generate clones of cells that overexpress L. In dorsal and ventral domains of first instar eye discs, L flp-out clones either induced or upregulated Ser expression, respectively (Fig. 3G-I). In summary, L is crucial in maintaining ventral Ser expression levels.

To understand the role of Ser in eye growth, Ser-null (Ser^rev2-11) mutant clones were examined in eye discs and adult eyes. We did not find obvious defects in the size of the eye disc or photoreceptor differentiation (Fig. 4A), consistent with previous reports (Sun and Artavanis-Tsakonas, 1996; Papayannopoulos et al., 1998). This lack of Ser-null clone phenotype implies that either the Ser protein is diffusible, or Ser is functionally redundant. In order to remove more of the wild-type Ser functions, we generated flp-out clones expressing a diffusible, truncated form of Ser (Ser^DN). Ser^DN consists of Ser extracellular domain but lacks the transmembrane domain and is capable of antagonizing wild-type Ser functions (Hukriede et al., 1997).

Eye discs that contained Ser^DN flp-out clones were variably reduced. Three kinds of phenotypes were observed and might be attributed to slight variations in the timing of clone inductions and the location of the clones:

(1) In 19 eye discs that contained 52 dorsal clones, there were only two small ventral clones found (one such eye disc is shown in Fig. 4B). This suggests that while the expression of Ser^DN had little effects on dorsal and medial cell viabilities, ventral cells expressing Ser^DN failed to proliferate and/or died.

(2) The most severely affected eyes were rudimentary, probably owing to clones near the DV boundary that disrupted the wild-type Ser functions in DV boundary formation (Papayannopoulos et al., 1998; Cho and Choi, 1998; Dominguez and de Celis, 1998). Similar phenotype was also observed in ey-Gal4/UAS-Ser^DN eye discs that lacked a DV boundary (Fig. 4C) (Kumar and Moses, 2001).

(3) In some cases, a preferential reduction of the ventral eye occurred (Fig. 4D), resulting in a phenotype that was strikingly similar to that of L^– mosaic clones (Fig. 1I,J).

Taken together with the regulatory relations between L and Ser, these results suggest that loss of ventral Ser expression probably contributes to L ventral eye loss.

L was placed in the 51A2-B1 region based on the breakpoints of Df(2R)L4, a chromosomal deficiency uncovering L (Baker and Ridge, 1980). Multiple overlapping cosmids spanning the 150 kb L region were isolated, and genetically transformed lines containing these rescue constructs were generated (see Materials and Methods). One of these lines, cosD3, contains ∼25 kb genomic DNA and was able to rescue eye phenotypes of the eye-specific L^si allele, reducing both severity and penetrance of mutant phenotypes (Fig. 5A, Table 1). The rescue of L^si/+ eye phenotypes by increasing wild-type copies of the gene suggests that L^si mutation is dominant negative in nature (Table 1). cosD3 also rescued the recessive lethality of L^P17 allele (see below).

The rescue of L associated phenotypes and lethality by cosD3 identifies the genomic region that encodes the L gene. Furthermore, a 8.5kb EcoRI fragment of cosD3 detected a restriction polymorphism on the Southern blot of an L inversion allele, L^rv5 (Baker and Ridge, 1980) (data not shown). Using this 8.5 kb fragment as a probe, multiple overlapping cDNA clones from an embryo cDNA library were isolated. Sequence analysis of multiple cDNA clones revealed a 1.6 kb open reading frame (ORF) and 5′ and 3′ untranslated sequences of 308 and 565 bp, respectively. The ORF agrees with the CG10109 sequence deposited by the Drosophila Genome Project. A representative 2.6 kb cDNA probe detected a single transcript of similar size on a northern blot made from larvae of all stages, suggesting that it represents the full-length transcript (Fig. 5D).

To verify that CG10109 is indeed the L gene, we identified the P-element insertion site of L^P17 (see Materials and Methods). L^P17 is recessive embryonic lethal and is phenotypically allelic to L alleles. L^P17/L^si animals showed loss of the ventral eye that is indistinguishable from L^si/L^si half-eye phenotype. Approximately 40% of L^P17/+ heterozygotes had defects at the anterior edge of the eye, similar to L^si/+ heterozygotes. Precise excision of the P-element restored the wild-type eye as well as viability, demonstrating that it was responsible for the mutant phenotypes. The P-element of L^P17 was inserted 163 bp downstream to the start codon in the first exon of CG10109 (Fig. 5A,B). The recessive lethality of L^P17 was partially rescued by cosD3: the viability of L^P17/L^P17; cosD3/+ animals was 80% less than the L^P17/+; cosD3/+ animals. Finally, full-length UAS-L cDNA construct under the control of hs-Gal4 driver was able to rescue lethality associated with the L^P17 allele when given daily heat shock throughout embryonic and larval stages. The rescued animal restored the eye fully (∼70% of flies) or partially (30% of flies). The rescue of lethality and eye defects by L genomic DNA and cDNA positively identifies CG10109 as the L gene.

The putative L protein is 562 amino acids long and contains a poly-glutamine rich region and a C terminus that shares significant sequence similarities with novel insect, mouse and human proteins of unknown functions (Fig. 5C). A polyclonal antibody was generated, and it detected a single band of ∼60 kDa on a western blot prepared from third instar eye imaginal discs (data not shown). Furthermore, this antibody revealed no detectable level of L protein expression in null animals (Fig. 5E). To examine L expression pattern, L transcript and protein were detected by mRNA in situ hybridization and by the polyclonal antibody. L is ubiquitously expressed in first instar eye discs (data not shown). In the third instar disc, the RNA transcripts were detected in the antenna disc and in undifferentiated cells anterior to the furrow in the eye disc (Fig. 5F). By comparison, the protein could be readily detected in the antenna and anterior to the furrow, but a much lower level of expression is also present posterior to the furrow (Fig. 5G).

The role of N signaling in growth is well documented in many organisms (for a review, see Artavanis-Tsakonas et al., 1999). For example, constitutive N activation in C. elegans causes germline tumor formation (Berry et al., 1997), and in humans, expression of human Ser homolog, JAG1 (jagged 1), induces the expansion of hematopoietic stem cells (Varnum-Finney et al., 1998; Karanu et al., 2000). In Drosophila, N activation at domain boundaries is required for the growth of wing and eye imaginal discs. In developing wing discs, activated N along the DV boundary upregulates the expression of Wg, which in turn induces wing outgrowth (Diaz-Benjumea and Cohen, 1995). In the eye disc, the proliferative effect of N is also likely to be mediated by diffusible factors emitted from midline, although their identities are not known at this time.

We have presented genetic evidence that L could be a component of an intracellular pathway that transduces N signaling in cells of the ventral eye (Fig. 2). The conserved L sequences suggest that its mammalian counterparts may play similar functions in mediating N signaling, although the precise function of the conserved domain is not known. The striking domain specificity of L-mediated growth may be the result of various mechanisms. It is possible that signaling molecules sent to the ventral domain are different from the ones sent to the dorsal. Signaling molecules may be selectively sequestered in one domain but not the other, as in the case of dpp signaling in the wing disc (Teleman and Cohen, 2000). L may cooperate with other ventral-specific genes to transduce the N signaling, or the expression of L may be transiently ventral specific in the early eye disc. More than one mechanism may contribute to the domain specificity of L functions.

Previous experiments have shown that growth of neighboring, symmetrical domains may be independently controlled. In Drosophila wing discs, increased expression of hedgehog along the anteroposterior boundary causes anterior wing overgrowth but has no effects on the posterior wing (Tabata and Kornberg, 1994; Johnson et al., 1995); ubiquitous overexpression of Ser increases the ventral wing tissue but not the dorsal (Speicher et al., 1994). Additionally, there are other Drosophila eye mutants that show preferential reductions of the ventral eye, such as wg mutants (Ma and Moses, 1995) and dpp mutants (Treisman and Rubin, 1995). However, dominant mutation Rough eye suppresses these ‘furrow stop’ mutant phenotypes of dpp and wg but not the L phenotype (Heberlein et al., 1993), suggesting fundamental differences in nature and function between L and furrow stop mutants.

Clonal study shows that L null clones have striking domineering non-autonomous effects, such that the viability of wild-type tissue immediately adjacent to L mutant clones are severely reduced (Fig. 1I). Nevertheless, this non-autonomous deleterious effect is limited to ventral domain, as clones abutting the DV boundary do not seem to affect dorsal cell viability. This domineering non-autonomous effect may be the result of interspersed L null clones disrupting the physical integrity of the imaginal disc epithelium, causing the disc to fall apart. It is also possible that L is redundant in the dorsal domain, thus loss of L can be compensated by another dorsal-specific gene. Another possibility that we favor incites a failure of the clone cells to send out a locally acting growth signal. As our data indicate that L is a regulator of Ser expression (Fig. 3), could Ser be this local-acting, diffusible factor?

Homozygous Ser mutants have small eyes, indicating that Ser is required for proper eye growth. However, removing Ser in clones of cells does not result in mutant eye phenotypes, as shown in this report (Fig. 4A) and others (Sun and Artavanis-Tsakonas, 1996; Papayannopoulos et al., 1998). These observations are consistent with Ser being a diffusible factor, but other possibilities exist.

First, there may be other functionally redundant Ser-like molecules, but no candidates have yet been identified. Second, in Ser-expressing cells, Ser may autonomously induce the expression of diffusible signaling molecules that act non-autonomously. Ligand and receptor interactions within the membrane of the same cell have been demonstrated in the case of N and Dl. N and Dl can physically associate within the membrane of a single cell (Fehon et al., 1990), and the expression level of Dl in a cell can modulate its own N response (Jacobsen et al., 1998). In this manner, Ser-N interaction may lead to the expression of diffusible factors that rescue clones of Ser^– cells. Third, the ability of adjacent wild-type cells to rescue Ser^– cells suggests that Ser protein may be diffusible. This is consistent with low levels of Ser expression that we observed in L^– clones; and in our anti-Ser antibody staining, we consistently observed intense, dotty cytoplasmic staining, possibly secretory vesicles (Fig. 3). However, the presence of secreted Ser is yet to be confirmed, even though diffusible Dl has been detected in Drosophila extract (Qi et al., 1999).

The domain-specificity of L phenotype indicates that the eye disc is partitioned, and the growth of individual domain is differentially regulated. Loss of the ventral eye in L mutants does not seem to affect DV boundary formation or the associated midline N activation, because disruptions of either of these events would result in abnormal dorsal growth. Additional data suggest that L does not affect the initial DV domain specification: (1) L is functionally downstream to N activation (Fig. 2); (2) L mutation does not affect Ser expression at the DV boundary (Fig. 3); and (3) domain-specific expression patterns of dpp, fng and wg are not affected in the first instar L mutant eye discs (data not shown).

Consistent with this model, we propose that in the seemingly homogenous Ser-expressing, first instar ventral domain, there are actually two distinct groups of Ser-expressing cells: ventral midline cells (Fig. 6A, yellow outline) abutting the dorsal midline cells, and the rest of the ventral cells (Fig. 6A, green outline). Their putative functions are different and their Ser expression is independently regulated. In the ventral midline cells, Ser is involved in setting up the DV boundary, and its expression is regulated by the Ser-N-Dl positive-feedback loop (Fig. 6B, yellow cells). The midline Ser expression can be further modified by Fng and Hedgehog, both of which can induce Ser expression only near the DV boundary but not elsewhere in the eye field (Papayannopoulos et al., 1998; Cho et al., 2000), emphasizing again the distinctiveness of these midline cells.

By comparison, in the rest of the ventral domain, Ser is directly involved in controlling local growth. Loss of Ser in the ventral domain causes ventral-specific growth defects similar to the loss of L (Figs 1, 4). Ser expression in the ventral domain may not be sustained by the Ser-N-Dl loop, as ventral Fng inhibits potential Ser-N interaction which is necessary to initiate the positive feedback loop (broken line in Fig. 6B) (Papayannopoulos et al., 1998). Instead, ventral Ser expression is regulated by L (Fig. 3).

Our data suggest the eye primordium is partitioned into dorsal, midline and ventral domains with different gene expression and growth properties. It highlights the importance of local cellular context in interpreting signals released from the domain boundaries and shows that the growth of symmetrical domains may be asymmetrically regulated. Our model may also be applicable to the development of other imaginal discs as well as other developmental systems.

We are grateful to K.-O. Cho for initial observation of ventral specificity of L mutant. We thank H. Bellen, K.-O. Cho, H. Sun and S. Izaddoost for comments and discussion, and Z. H. Chen for technical assistance. We thank the Indiana Stock Center for providing flies and Caltech monoclonal facility to make the L antibody. We thank K.-O. Cho for Dlg antibody. Confocal microscopy was supported by a Core Grant for Vision Research from the NIH to D. B. Jones. J. C. is supported by NRSA and K.-W. C. is supported by the NIH.