klingon, a novel member of the Drosophila immunoglobulin superfamily, is required for the development of the R7 photoreceptor neuron

Neurogenesis represents a complex developmental problem; a cell is specified as neuronal, adopts a particular fate from a vast number of different neuronal cell types and realizes its function by becoming wired into the relevant axonal circuit. Like other developmental processes, the specification of neuronal cell fate involves the sequential and combinatorial activity of genes that gradually restrict the developmental potential of cell lineages during differentiation (reviewed by Goodman and Doe, 1993; Jan and Jan, 1993). However, an as yet unsolved question is how cells perform the complex functions unique to developing neurons. For example, a pathfinding axon must be able to respond to environmental cues as it navigates its way to its synaptic target. Neuronal cell bodies often have to migrate through multiple cell layers before they reach their final position. Consequently, it seems likely that there are molecules on the surface of neurons that act, via signal transduction pathways, to modulate gene expression and allow them to respond to their surroundings. There are a number of extracellular molecules on the surface of neurons which might regulate signaling pathways as well as providing a differential adhesive substrate. For example, the homophilic binding of vertebrate N-CAM (neural cell adhesion molecule) is thought to lead to the activation of a tyrosine kinase which results in neurite outgrowth (reviewed by Doherty and Walsh, 1994). NCAM is a member of the Immunoglobulin superfamily (IgSF); the distinguishing feature of this family is the Immunoglobulin (Ig) domain, which is thought to confer adhesive properties (reviewed by Williams and Barclay, 1988). A number of members of the IgSF have been isolated in Drosophila, of which Fasciculin II (FasII) is thought to be the Drosophila homologue of N-CAM (Grenningloh et al., 1991). All the Drosophila members of the IgSF are found on subsets of fasciculating axons in the developing nervous system and some are also expressed either in interacting glia or target musculature. This led to the proposal that, in the insect central nervous system, specific axonal pathways are made distinct by their differential labeling of surface recognition molecules (Grenningloh et al., 1990). However, with a few exceptions (Ramos et al., 1993), the genetic analyses of Drosophila IgSF genes have not yet uncovered a clear role for these molecules in axonal pathfinding. In contrast, fasII has been shown to regulate the expression of two proneural genes, achaete and atonal in the eye-antennal imaginal disc, suggesting that FasII can participate in setting up the field in which neuronal differentiation takes place (García-Alonso et al., 1995).

The Drosophila compound eye is an excellent model system in which to study questions of determination of neuronal fate and function. It has a small number of neuronal subtypes whose projections into the optic lobe can be easily visualized. The compound eye is composed of 800 unit eyes, or ommatidia. Each ommatidium contains about 20 cells of which eight are the photoreceptor neurons called R1 through R8. These photoreceptor neurons can be subdivided into three classes based on morphology, spectral sensitivity and the pattern of synaptic connections of the photoreceptor axons (reviewed by Hardie, 1986). Of the eight photoreceptor neurons, R7 is the only neuron sensitive to UV light. Its rhabdomere, the light gathering structure, has a smaller diameter than those of R1R6, and is positioned in the central apical region of the ommatidium, overlying the rhabdomere of the R8 neuron.

The eye develops from an undifferentiated epithelial sheet called the eye imaginal disc, half way through the third instar larval stage (reviewed by Wolff and Ready, 1993). A groove, called the morphogenetic furrow, appears at the posterior edge of the disc. The furrow moves anteriorly across the disc recruiting cells into ommatidial clusters in a sequential manner. Thus, the differentiation of developing ommatidial clusters occurs in a gradient across a disc, with clusters at the posterior edge being the most mature. Photoreceptor neurons within each ommatidial cluster differentiate in an invariant sequence which suggests that the cells are sequentially induced (reviewed by Tomlinson, 1988). The last photoreceptor to be induced is the R7 photoreceptor cell and the development of this photoreceptor has been the object of intense study. Neuronal differentiation of the R7 precursor is triggered by the activation of the Sevenless (Sev) receptor tyrosine kinase located in the membrane of the R7 precursor cell, by the Boss ligand on the R8 cell. This results in the activation of the RAS/MAPK signaling cascade in the R7 cell (reviewed by Zipursky and Rubin, 1994; Dickson and Hafen, 1993; Dickson, 1995). However, the RAS/MAPK signaling cascade is required not only for R7, but for all photoreceptor development (Dickson and Hafen, 1993). Consequently, it is probable that the Sev signaling pathway does not specify the unique properties necessary for the differentiation of a functional R7 neuron. Such properties are likely to be conferred by genes whose activities are specific to the R7 neuron.

Here we report on the identification and characterization of a gene, klingon (klg), which encodes a novel member of the Drosophila IgSF and is expressed in the R7 neuron throughout its development. klg was identified through screens for enhancer trap insertions that are expressed both in the eye imaginal disc and in the embryonic central nervous system (CNS). The expression pattern of the enhancer trap line, H214, has been described previously (Mlodzik et al., 1992), and this strain has been used extensively as a marker for the R7 precursor cell. H214 is thought to mark the R7 precursor cell independently of the Sev signaling pathway. In a sev⁻ background the R7 precursor is transformed to a non-neuronal cone cell. Expression of H214 in the R7 precursor is greatly reduced in sev larval eye discs, but in the sev pupal eye, expression resumes in one cone cell which presumably differentiated from the R7 precursor (Mlodzik et al., 1992). This suggests that the H214 expression pattern might reflect a commitment to R7 differentiation that is independent of the Sev signaling pathway. Thus, the gene associated with the H214 enhancer trap line was a candidate for a component of a parallel pathway that would provide some aspect of R7 identity or function independently of that provided by the sev signaling pathway. The gene identified through the H214 line, klg, encodes a new member of the Drosophila IgSF. Klingon is capable of mediating homophilic cell adhesion. Phenotypic analysis of loss-of-function alleles of klg has demonstrated a requirement for klg in the determination of the R7 neuron in a background where the level of signaling through Sev has been reduced. Although the loss of klg causes no detectable alteration in the number or structure of the photoreceptor neurons, ectopic expression studies show that klg can affect the position of the R8 rhabdomere within an ommatidium.

DNA sequences flanking the P-element insertion point in the H214 line were recovered by the plasmid rescue method as described by Mlodzik et al. (1990), and were used to initiate a chromosomal walk. EcoRI genomic restriction fragments isolated from a Drosophila genomic library (Maniatis et al., 1978) which cover a 20 kb region on either side of the P-element insertion, were used to screen a 9- to 12hour embryonic cDNA library (Zinn et al., 1988). Three cDNAs were obtained using a 8.9 kb EcoRI fragment, the largest of which was designated 25B. The three EcoRI restriction fragments of the 25B cDNA, 2.0 kb, 0.9 kb and 0.3 kb, were cloned into BlueScript (Stratagene) generating pSK-klg.1, pSK-klg.2 and pSK-klg.3 respectively. pSKklg.1 contains the entire klg open reading frame. Using the 2.0 kb EcoRI fragment of 25B as a probe, we isolated approximately thirty cDNAs from the embryonic cDNA library and two cDNAs from an eye imaginal disc cDNA library (constructed by Alan Cowman). However, sequencing of these cDNAs from their ends showed that all of them were contained within 25B. The 25B cDNA was completely sequenced on both strands by the chain termination method (Sanger et al., 1977) using Sequenase (US Biochemical Company). The predicted protein sequences were analyzed using the FASTA program. Using 25B as a probe, genomic clones beyond + 25 kb (Fig. 1) were obtained from a genomic library (Moses et al., 1989) and a cosmid library (Tamkun et al., 1992). Intron-exon boundaries were determined by sequencing genomic DNA clones using sequencing primers designed from the cDNA sequence. 5′-RACE was performed on mRNA from 8- to 12-hour embryos using the protocol from GIBCO/BRL. PCR primers were designed from sequences immediately internal to the 5′ end of the 25B cDNA. Other molecular biological techniques were performed using standard methods (Maniatis et al., 1989).

In situ hybridizations to whole-mount discs and embryos were performed as described by O’Neill and Bier (1994) and Tautz and Pfeifle (1989) respectively. Antibody staining of imaginal discs was performed as described Tomlinson and Ready (1987) except that the peripodial membrane of the eye disc was not removed. β -galactosidase protein was localized immunohistochemically using an anti-β galactosidase antibody purchased from Promega. Monoclonal antibodies 22C10, 1D4, BP102, BP104 and 24B10 were kind gifts from the laboratories of C. Goodman, A. Bieber and S. Benzer, and were used to examine the architecture of the CNS in E226 and E1432 mutant embryos. Sections of adult retina were made according to the method of Tomlinson and Ready (1987).

The H214 enhancer trap line contains an HZ element (Mlodzik and Hiromi, 1993) near the 5′ end of the klg gene. To isolate imprecise excisions of the P-element, the transposon was mobilized by supplying a stable genomic source of transposase in trans (Robertson et al., 1988) and stocks were established from 195 independent ry progeny. Southern analysis on DNA from these stocks, using genomic restriction fragments as probes, identified 6 deletions that extend greater than 5 kb from the insertion point. Two of these, E226 and E1432, were lethal over Df(3R)GR2, which deletes the region between 93F11-14 and 94E2-5 (Mohler, 1988). The extent of the deletions in E226 and E1432 was determined by analyzing RFLPs present in the parental chromosome and various wild-type or balancer chromosomes. E1432 deletes the ry⁺ marker from the HZ vector leaving the lacZ reporter intact and extends approximately 40 kb, terminating in the third intron. E226 deletes the entire klg ORF; the 5′ end-point is within 7 kb of the 5′ end of the H214 insertion and the 3′ end-point extends beyond the region covered by the walk. Consequently, it is possible that E226 may delete neighboring genes. Homozygous clones of E226 and E1432 were made in the adult eye by mitotic recombination using the FLP-FRT system (Golic and Linquist, 1991) and the Minute technique (Morata and Ripoll, 1975). The 75AE FRT insertion (Golic and Linquist, 1991; Hiromi et al., 1993) located at the base of 3R was used. Precise excisions of the H214 P-element insertion were generated in a similar manner to the imprecise excisions. The region encompassing the P-element insertion point was sequenced from a PCR-amplified fragment to ensure that the excision event restored the wild-type sequence.

A plasmid that allows the production of histidine-tagged Klg fusion protein was constructed by cloning a fragment of the klg cDNA encoding amino acids 106-528 of the Klg protein into the PstI site of the QE-9 vector (Qiagen). Histidine-tagged recombinant Klg protein (His-Klg) was induced and isolated on a Ni⁺-affinity column using the QIAexpress system (Qiagen). A New Zealand White rabbit (Mopsy) was immunized with 100 μg of His-Klg and boosted once a month. The serum (anti-Klg) recognized a 53 kDa protein seen only after the induction of S2 cell transfected with a metal-inducible Klg vector (see below).

For the expression of Klg in S2 cells, the complete klg ORF, contained in a 2.0 kb EcoRI fragment, was placed under the control of the metallothionein promoter using the pRMHa-3 vector (Bunch et al., 1988). The resulting plasmid, pRmHa-3-KlgA was cotransfected into S2 cells with the plasmid pPC4 (Jokerst et al., 1989) which confers α amanitin resistance. Transfections were carried out according to the method of Snow et al. (1989). For cell aggregation assays, 5 ml cell suspension at a density of 1-2× 10⁶/ml were allowed to grow overnight at room temperature under gentle agitation on a rotary shaker at 100 rpm. 5 μl of 700 mM CuSO₄ was added to induce the cells and agitation continued for 15-18 hours. For mixing experiments, cells were labeled with either a green (PKH2) or red (PKH26) fluorescent lipophilic dye (Sigma Immunochemicals) prior to aggregation. Confocal microscopy was performed using the Biorad MRC600 imaging head, mounted on a Nikon Optiphot II microscope and the pictures analyzed using the Confocal Assistant software (Microsoft).

Induced and uninduced S2 cells were separated from culture media by centrifugation (14,000 g for 2 minutes). Cell and medium fractions were subjected to a 12% SDS-polyacrylamide gel electrophoresis. The levels of Klg protein in eye imaginal discs were analyzed by dissecting 10 eye imaginal discs into 10 μl of sample buffer (0.0625 M Tris-HCl pH 6.8, 10% glycerol, 2.3% SDS, 0.5% β -mercaptoethanol) and boiling for 7 minutes before running on a SDS-polyacrylamide gel. Western blots were incubated with the anti-Klg serum at a dilution of 1:1000 and with a 1:2000 dilution of the secondary HRP-conjugated goat anti-rabbit IgG (Jackson) and detected using the Renaissance chemiluminescence kit (DuPont).

S2-Klg cells were induced and then incubated for 30 minutes at 37°C either with or without PI-PLC. (Oxford Glycosystems, 1U/ml in 50 mM Tris, pH 7.2). After centrifugation at 14,000 g for 10 minutes, the pellet (cell fraction) and supernatant were run on a western blot. The PI-PLC treatment led to the release of over 50% of the cellular Klg, whereas no Klg was released from cells that had not been treated with PI-PLC.

A P-element construct containing the klg ORF fused to UAS_G-linked promoter was made by cloning a 2.0 kb EcoRI fragment from pSKklg.1 into the EcoRI site of the pUAST vector (Brand and Perrimon, 1993). Germ-line transformation was performed using w¹¹¹⁸ as the host strain and pπ 25.7wc (Karess and Rubin, 1984) as the helper plasmid. Secondary jumps were made using a strain carrying a genomic source of transposase activity (Robertson et al., 1988). We established 20 independent transformant lines, including UAS_G-klg#1, 2, 3 and 6, each with an insert on chromosome 3 and UAS_G-klg#5 and 6, both with an insert on chromosome 2. Western blotting showed that, without GAL4, these lines have greater levels of Klg protein in eye discs than seen in wild-type. However, in situ hybridization experiments on eye imaginal discs from the six lines, using the 2.0 kb EcoRI fragment of 25B cDNA as a probe, showed that klg is still expressed in the endogenous pattern.

The H214 line was originally identified in an enhancer trap screen because of its expression during embryogenesis and eye imaginal disc development (Mlodzik et al., 1992). H214 expresses lacZ in a subset of neurons in the developing embryonic central and peripheral nervous system (CNS and PNS). The expression pattern of H214 in eye imaginal discs has been described previously in detail (Mlodzik et al., 1992). In brief, lacZ from the H214 insertion is expressed in the R7 photoreceptor cell throughout its development and in the R8 neuron, starting from pupal stages.

Genomic restriction fragments encompassing a 20 kb region flanking the H214 P-element insertion point were used to screen embryonic and eye disc cDNA libraries. We identified a partial 3.2 kb cDNA (25B) whose expression pattern in in situ hybridizations to whole-mount embryos and eye imaginal discs closely resembles that of the H214 line (Fig. 1). The gene identified by this cDNA will be called klingon (klg) because of its adhesive properties (see below). KLG (klg mRNA) is first expressed at embryonic stage 12 in a limited number of neurons in each segment of the developing CNS. By stage 14, KLG is expressed in a subset of neurons in each neuromere of the CNS (Fig. 1A) and also in all the neurons of the external sensory organs in the PNS, identical to the enhancer trap line (Fig. 1B). The high level of expression in a subset of neurons continues into stage 15/16 after the condensation of the CNS and persists into larval stages. Thus, KLG starts to be expressed as neurons are born and is expressed at high levels from stage 13 onwards when neurons are extending axons along the pathways that will form the segmentally repeated array of commissural and longitudinal axon bundles.

The expression of KLG in the eye imaginal disc is also very similar to that of the H214 enhancer trap line (Fig. 1D). KLG is expressed transiently in a band of cells three rows wide, about 4-5 rows behind the morphogenetic furrow. Posterior to this is a region of 2-3 rows in which KLG is undetectable. Expression reinitiates in multiple cells per ommatidial cluster which, within three rows, resolves to expression solely in the R7 cell. This is approximately four rows after the R7 precursor has joined the ommatidial cluster, at about the same time that the R7 cell begins to express neural antigens (Tomlinson and Ready, 1987). KLG expression in the R7 cell continues throughout the development of the eye. After pupariation, KLG also starts to be expressed in R8 cells (Fig. 1F). In the third instar larva, KLG is expressed in a subset of neurons in lamina, the region of the optic lobe to which the outer photoreceptor neurons project their axons (Fig. 1C).

We were interested in examining the expression pattern of KLG in a sev background to determine if the klg gene, like the amino acids characteristic of proteins that are attached to the H214 enhancer trap line, is expressed in the cone cell devel-membrane by a phosophotidylinositol (PI) anchor (Fig. 4B) oping from the R7 precursor. In a sev^d2 eye imaginal disc, klg (Ferguson and Williams, 1988). Klg has the same number of expression is indistinguishable from wild-type discs until about Ig domains and Fibronectin type III repeat as REGA-1, a row 14 when the KLG is found in multiple cells per omma-member of the IgSF which has recently been isolated in tidial cluster (Fig. 1E). However, this expression is not main-grasshopper (Fig. 4C; Seaver et al., 1996). The two proteins tained; there is no resolution to the R7 precursor cell and by are 31 to 43% identical in the corresponding Ig domains (Fig. the posterior edge of the disc, KLG is not detected (Fig. 1E). 4C), and share 27% identity over their entire length (data not In sev^d2 pupal eye discs, klg is expressed most strongly in one shown). cell at the focal plane of the ommatidial clusters, which we To determine if Klg is PI linked, S2 tissue culture cells that believe to be the R8 cell (Fig. 1H). In a focal plane above this, express Klg protein (see below) were incubated with PI-phos-KLG is expressed in a cell whose position is consistent with it pholipase C (PI-PLC). Upon PI-PLC treatment, more than 50% being a cone cell (Fig. 1I) while no expression is seen in wild-of the cellular Klg protein was released into the medium (Fig. type cone cells. The cone cell that expresses klg is likely to be 5A). No Klg protein is detected in the tissue culture medium derived from the R7 precursor that failed to differentiate as an without PI-PLC treatment, suggesting that Klg is not secreted. R7 neuron in the sev mutant. Thus, the klg gene marks the Hydropathy analysis reveals no hydrophobic segments of sufposition of the R7 precursor cell independently of the infor-ficient length to span the membrane (Fig. 4B). Thus, we believe mation that specifies its final identity.

Sequences 5′ of the 3.2 kb klg cDNA were isolated by extending the 25B cDNA 0.76 kb by 5′ RACE (Rapid Amplification of cDNA Ends). The P-element associated with the H214 enhancer trap line is inserted 4 bp upstream of the 5′ end of this extended 25B cDNA (Fig. 2). Northern blot analysis shows that the 25B cDNA hybridizes to two embryonic transcripts of 4.4 kb and 6.0 kb (data not shown). Since the 25B cDNA does not contain a poly A tail (Fig. 3), it is unclear which of the two transcripts 25B clone corresponds to.

The klg cDNA contains one long open reading frame which encodes a putative protein (Klg) of 528 amino acids (Fig. 3). The nucleotide sequence at the deduced start methionine matches the consensus sequence for Drosophila translational start sites (Cavener, 1987), and there are several in frame stop codons upstream of this methionine codon. Searches for sequence homology revealed that the putative Klg protein contains two different structural repeats that suggest Klg is a novel member of the Drosophila IgSF (reviewed by Williams and Barclay, 1988). Starting from the N terminus, there is a region which has 23-25% identity with the vertebrate neural cell adhesion molecules and contains three C2-type Ig domains (Fig. 4A). This is followed by one Fibronectin type III repeat, another structural motif common to IgSF members. At the C-terminus there is a run of moderately hydrophobic amino acids characteristic of proteins that are attached to the membrane by a phosophotidylinositol (PI) anchor (Fig. 4B) (Ferguson and Williams, 1988). Klg has the same number of Ig domains and Fibronectin type III repeat as REGA-1, a member of the IgSF which has recently been isolated in grasshopper (Fig. 4C; Seaver et al., 1996). The two proteins are 31 to 43 % identical in the corresponding Ig domains (Fig. 4C), and share 27 % identity over their entire length (data not shown).

To determine if Klg is PI linked, S2 tissue culture cells that express Klg protein (see below) were incubated with PI-phospholipase C (PI-PLC). Upon PI-PLC treatment, more than 50 % of the cellular Klg protein was released into the medium (Fig. 5A). No Klg protein is detected in the tissue culture medium without PI-PLC treatment, suggesting that Klg is not secreted. Hydropathy analysis reveals no hydrophobic segments of sufficient length to span the membrane (Fig. 4B). Thus, we believe the Klg protein encoded by the 25B cDNA to be entirely extracellular, GPI-linked to the membrane. The best match to the signal sequence (von Heijne, 1985) is a 13 amino acid sequence containing a stretch of hydrophobic amino acids starting at amino acid 48 (Fig. 2). Although there are examples of long signal peptides in viral envelope proteins (Pancino et al., 1994), we do not know whether this potential signal peptide of Klg is cleaved.

The klg gene is composed of at least 12 exons that are distributed over a region of about 60 kb. The intron-exon boundaries of exon 4 through 10 fall within and between the three putative Ig domains (Fig. 4C). Vertebrate members of the IgSF tend to have introns between each Ig domain, with N-CAM, in particular, also having introns within each Ig domain (Cunningham et al., 1987). klg is the only member of the Drosophila IgSF isolated so far that has intron placement similar to that of the vertebrate genes.

A number of members of the Drosophila IgSF have been shown to mediate the adhesion of isolated cells in vitro using the Drosophila tissue culture system (Elkins et al., 1990; Hortsch et al., 1995; Kania et al., 1993; Nose et al., 1992). In order to test if Klg protein can function as an adhesion molecule, we placed the klg ORF under the control of a metal-inducible promoter and transfected non-adhesive Drosophila S2 cells with plasmid DNA containing this chimeric gene. Four independently transfected lines (S2-Klg) were generated, each of which contains a mixed population of Klg-expressing cells. Western blot analysis showed that, upon induction, all of the S2-Klg lines express a 53× 10³M_r protein at high levels that is recognized by a polyclonal serum raised against recombinant Klg protein (Fig. 5A). This protein was not present in control S2 cells transfected with the parent vector or when Klg was not induced in S2-Klg cells. Three of the S2-Klg cell lines, when cultured overnight with gentle agitation in the presence of inducer, formed large aggregates consisting of hundreds of cells (Fig. 5C). Aggregation was not observed either with control S2 cells or in the absence of induction (Fig. 5B).

Cell adhesion can occur by either a homophilic or heterophilic mechanism. To determine how Klg mediates adhesion, we asked if S2-Klg cells can form aggregates with untransfected S2 cells. If S2-Klg cells adhered to S2 cells it would indicate that Klg could interact heterophilically with an endogenously expressed protein on the surface of the S2 cells. S2-Klg cells were labeled with a red fluorescent lipophilic dye and mixed in a 1:1 ratio with untransfected S2 cells labeled with a green fluorescent dye. When S2-Klg cells were induced as described above, the aggregates were formed entirely from S2Klg cells, the S2 cells remaining as single cells (Fig. 5E). This result indicates that Klg can function as a homophilic adhesion molecule. To test if Klg can mediate cell sorting, the S2-Klg cell line was mixed with a cell line (S2-Nrg) that, on induction, expressed another member of the Drosophila IgSF, Neuroglian (Bieber et al., 1989; Hortsch et al., 1995). Upon induction of both proteins, the aggregates formed were composed of either S2-Klg cells alone or S2-Nrg cells alone, but never both cell types (Fig. 5F). In a control experiment, two independently transfected S2-Klg cell lines were differentially labeled and mixed. The aggregates that formed after induction contained a mixture of both red and green cells showing that dye labeling per se does not cause sorting. These results indicate klg can mediate cell sorting in a homophilic manner. It does not eliminate the possibility that Klg can also adhere heterophilically, its receptor being neither Nrg nor present on S2 cells.

We carried out a genetic analysis of klg in order to understand its role in vivo. The P-element of the H214 enhancer trap line maps to position 94D on the third chromosome and its insertion does not cause any obvious phenotype. However, embryos and eye discs from the H214 line show at least a 10-fold reduction in the level of KLG transcript suggesting that H214 is a hypomorphic allele of klg (data not shown). A screen for imprecise excisions of the P-element generated two deletions (E226 and E1432) both of which are homozygous lethal. Neither complements a large deficiency of the region, Df(3R)GR2. E226 extends in both directions from the P-element insertion point and deletes the entire ORF and thus is a null mutation of klg. The deletion in E1432 starts in the H214 P-element and extends approximately 40 kb into the gene, including the first 3 exons that encode 74 amino acids of the putative protein. Transcripts that hybridize to the klg cDNA probe are still present in E1432 embryos.

In animals mutant for either of these alleles, homozygous mutant embryos hatch but die as second instar larvae. The overall structure of the embryonic CNS develops normally (data not shown). To analyze the function of klg during eye development, homozygous mutant clones of E226 and E1432 in the adult eye were generated by mitotic recombination. These clones show no alteration in the number or morphology of the photoreceptors (Fig. 6A).

Although klg is dispensable for R7 neuronal differentiation in an otherwise wild-type background, it is possible to show a requirement for the klg gene in the development of the R7 pho-toreceptor neuron in a background in which signaling through the Sev signaling pathway has been reduced. An assay has been developed using the combination of a gain-of-function Sos allele, Sos^JC2 and a loss-of-function sev allele, sev^E4 (Rogge et al., 1991, Karlovich et al., 1995). sev^E4 ommatidia have no R7 cells but, with the addition of one copy of Sos^JC2, the phenotype is rescued to an intermediate level whereby 18.5% of ommatidia contain R7 cells (Table 1). If the effectiveness of the Sev signaling pathway is then reduced by removing a single copy of a component required to activate Sev signaling, the number of ommatidia containing R7 cells is reduced. E226, E1432 and H214 were crossed into the sev^E4; Sos^JC2 background and the resulting progeny examined for the effect of heterozygous mutations in klg on R7 formation. E226, E1432, H214 or Df(3R)GR2 all cause at least a 3-fold reduction in the number of ommatidia containing R7 cells, compared to the figure for the ry⁵⁰⁶ parental chromosome (Table 1). Thus, the reduction of klg expression in a background of reduced Sev signaling has an effect on the specification of the R7 cell, suggesting that there is a requirement for klg in the determination of R7.

As described above, klg gene expression becomes localized to the developing R7 photoreceptor cell in the eye imaginal disc. To examine the functional significance of the klg expression pattern, we analyzed the consequences of ectopically expressing klg using the GAL4/UAS system developed by Brand and Perrimon (1993). We generated six independent UAS_G-klg lines and expressed klg in all photoreceptor neurons using the C155 driver, in which GAL4 is inserted near the elav gene leading to the expression of GAL4 in all neurons (Lin and Goodman, 1994). Tangential sections were made from adult eyes to examine the effect of expressing klg in all neurons on the stereotypical arrangement of cells within an ommatidium. All six lines showed minor disturbances in the morphology and number of photoreceptors. In particular, 1-2% of ommatidia in all eyes either gained or lost an R7 photoreceptor cell (Fig. 6B and C). More rarely, we observed rotated ommatidia or ommatidia that did not have the stereotypical arrangement of rhabdomeres (data not shown). Although only a small number of ommatidia were affected, the same range of phenotypes was consistently seen between the six lines.

The extra R7 cells seen when klg is ectopically expressed in all photoreceptor cells could be due to adhesive changes within the ommatidium which result in cells with the potential to become R7 (for example, a cone cell) fortuitously activating the Sev receptor. To test this, we asked if these additional R7 cells were dependent on the sev signaling pathway. The six C155; UAS_G-klg lines were crossed into a sev null background (sev^d2). When apical semi-thin sections were made from the retina from these six lines, there were a significant number of ommatidia that contained small rhabdomeres with the characteristic morphology of an R7 cell (Fig. 6F) when compared to the sev control retinae (Table 2). These ‘R7-like’ cells required the presence of the UAS_G-klg transgene, since they were rarely seen in a line in which the transgene had been excised (Table 2). To determine the identity of the cells with small rhabdomeres in the apical sections of C155 sev^d2; UAS_G-klg retinae, we examined the expression pattern of several R7 specific markers. The ‘R7-like’ cells in the retina do not express either of the fusions of lacZ to the promoters of the two R7-specific rhodopsins (Rh3-lacZ and Rh4-lacZ; Fortini and Rubin, 1990) nor is expression of H214 enhancer trap marker seen in the eye disc (data not shown). When serial sections were analyzed, the ommatidia containing cells with small rhabdomeres in apical sections did not have R8 rhabdomeres in the basal part of the retina (Fig. 6G). In wild-type ommatidia the R8 rhabdomere normally lies immediately below the R7 rhabdomere. The identity of the displaced R8 cells appears not to be altered, because molecular markers that label the R8 cell, for example, BB02 (Hart et al., 1990), are expressed normally in eye discs (data not shown). These results indicate that expressing klg in all photoreceptor neurons in a sev background causes the R8 rhabdomere to project more apically, assuming the position of the R7 rhabdomere (Fig. 6HJ). Although sev retinae also contain ommatidia in which the R8 rhabdomere has projected up from its basal position (Fig. 6D) the frequency of such ommatidia is significantly lower than individuals in which klg has been misexpressed (Table 2). We conclude that missexpression of klg can affect the position of the rhabdomeres within an ommatidium.

We have described the identification and characterization of the klg gene, a novel member of the IgSF, that has a restricted expression pattern in the R7 and R8 photoreceptor neurons of the compound eye of Drosophila melanogaster. Klg shares overall structural features with other neuronal IgSF members such as NCAM (Cunningham et al., 1987), L1 (Moos et al., 1988) and Contactin/F11 (Brümmendorf et al., 1989; Ranscht, 1988). In addition, the domain organization of Klg is identical to that of the grasshopper REGA-1 gene (Seaver et al., 1996). It is possible that REGA-1 and Klg form a new subfamily within the IgSF.

In vitro studies have implicated members of the IgSF in events such as neutrite outgrowth, growth cone guidance and axon fasciculation (reviewed by Dodd and Jessell, 1988; Edelman and Crossin, 1991; Sonderegger and Rathjen, 1992). Homologues (based on the number of structural repeats) of members of vertebrate IgSF have been identified in Drosophila, and have been shown to be expressed in the developing compound eye (Schneider et al., 1995; Ramos et al., 1993; Hortsch et al., 1990). However, all of these genes are expressed in all the photoreceptor neurons. Klg is the first IgSF member in Drosophila that is expressed in a specific subset of neurons in the developing eye. To address what the function of klg might be, we have analyzed the properties of klg in vitro, in tissue culture, and genetically by generating both loss-offunction and gain-of-function klg alleles.

We have shown that the induction of Klg protein in tissue culture cells can mediate the homophilic adhesion of S2 cells in a manner similar to FasII (Grenningloh et al., 1990), FasIII (Snow et al., 1989), Neuroglian (Hortsch et al., 1995) and Connectin (Nose et al., 1992). This demonstrates that Klg is a functional adhesion molecule. Klg is also able to mediate cell sorting. In a result similar to that seen on mixing cells expressing either FasI or FasIII (Elkins et al., 1990), a mixture of S2 cells expressing either Klg or Nrg sorted into aggregates that were homogenous for either Klg- or Nrg-expressing cells.

The recessive lethality of loss-of-function alleles of klg shows that klg is an essential gene. However, we have been unable to determine the cause of lethality. Because Klg is a member of the IgSF, we looked for defects in the structure of the embryonic CNS and for axonal guidance defects in the adult eye. No gross abnormalities were seen in either case, although we can not rule out the possibility that there were subtle problems not detected by the methods and probes used. In addition, there were no alterations in the number or morphology of the photoreceptors neurons in homozygous klg⁻ clones in the adult eye. The genetic analysis of other members of the Drosophila IgSF has been similarly complicated by the complex and apparent redundant nature of the nervous system. The null mutations of some genes, for example, fasII and neuroglian are homozygous lethal but the associated phenotype is either subtle (Bieber et al., 1989; Lin et al., 1994) or its analysis awaits more sophisticated probes for the central nervous system. However, it is possible to use sensitive genetic assays to determine if a gene is required for the specification of the R7 cell (Rogge et al., 1991; Simon et al., 1991). These assays can measure the effect of halving the dosage of a gene in a background where signaling through the Sev pathway is at a threshold level. In a sev^E4; Sos^JC2 background we were able to show at least a 3-fold reduction in the number of R7 cells when the animals are heterozygous for loss-of-function alleles of klg. This suggests that there is a modest requirement for klg in the formation of the R7 cell when the signaling through the sev pathway is compromised. It should be noted that klg expression is nearly abolished in mature ommatidial clusters in sev mutant eye discs (Fig. 1E, Mlodzik et al., 1992). Consequently, it is unlikely that the period during which klg participates in the specification of R7 occurs solely after klg expression has resolved to the R7 neuron. On the other hand, the expression of klg in earlier developmental stages during ommatidial assembly is unaffected by the sev mutation. It is possible that klg is required during this period for R7 neuronal determination.

A successful approach for analyzing the function of IgSF genes has been to misexpress the gene in a variety of cell types and thus generate gain-of-function phenotypes (Lin et al., 1994; Nose et al., 1994). Expression of klg in all neurons produced a range of phenotypes in the adult eye; the absence or loss of the R7 cell, the rotation of ommatidia and the misalignment of rhabdomeres within an ommatidial cluster. These phenotypes could occur if the ectopic expression of klg in the developing neurons brings about an alteration in the contacts between the cells within the ommatidial cluster. This might occasionally result in fate changes if the inductive process was disrupted. For example, in the case where an ommatidium lacks R7, the developing R7 cell might have lost contact with the R8 cell and not received its inductive signal. Similarly, a cell becoming ectopically juxtaposed to R8 at the critical signaling period might result in an ommatidium gaining an extra R7 cell. It is likely that this cell would be either a non-neuronal cone cell or mystery cell, because all the other photoreceptor cells form normally in the affected ommatidia. These alterations in cell fate are rare presumably because such disruptions must occur within in narrow time window to cause a transformation in cell fate. In agreement with this model, these additional cells appear to be dependent on signaling from Sev. A more dramatic phenotype is seen when klg is misexpressed in a sev background. A large fraction of the R8 cells project their rhabdomeres apically to assume the position of the R7 rhabdomere, showing that Klg is capable of altering the position of the R8 rhabdomere within the ommatidium (Fig. 6H-J). This phenotype is only seen in a sev background, presumably because when klg is misexpressed in an otherwise wild-type ommatidium, the R8 rhabdomere can not shift apically due to steric hindrance from the R7 rhabdomere. It is also possible that, in a wild-type ommatidium, the expression of klg in both the R7 and R8 cells prevents the R8 rhabdomere from moving apically.

In conclusion, these findings suggest a potential role for klg in mediating the correct positioning of photoreceptor precursor cells within the developing ommatidium. Although klg expression becomes restricted to the R7 neuron after differentiation of the R7 neuron has begun, klg is transiently expressed in multiple cells per ommatidial cluster during the period when sev signaling takes place. It is possible that the homophilic adhesion activity of klg is utilized at the critical period in the developing eye imaginal disc to ensure that the R7 precursor makes the correct contacts in order to receive inductive signals from the R8 cell. We still do not understand the significance of the restriction of klg expression solely to the R7 cell. This may reflect another function of klg, perhaps in guiding the R7 axon to its correct target in the optic lobe, in concert with the expression of klg in R8 and/or the laminar neurons.

We would like to thank colleagues in Princeton for stimulating conversations and advice. This manuscript was much improved by the comments of Steve West, Girish Deshpande, Susanne Kramer, Trudi Schupbach and Liz Gavis. We would also like to thank Alan Bieber and Helmut Kramer for S2 cell lines and advice on tissue culture work, Rosemary Reinke for help with the pseudopupil assay and Dan Barbash and Marek Mlodzik for their contribution in the initial stages of this work. The Mohler, Banerjee and Goodman laboratories kindly provided us with fly stocks. The Rubin and Schedl laboratories gave us cDNA and genomic libraries. We also want to thank Joe Goodhouse and Julie Waterbury for their help with the confocal imaging and Darren Hasara for his assistance with the production of antibodies. S. J. B. would like to thank Paul Lasko for his hospitality while this paper was drafted. This work was supported by the National Institute of Health Grant RO1 NS29662. Y. H. was a Pew scholar in Biomedical Sciences.