Using the Drosophila visual system as a model, our laboratory is studying the molecular mechanisms by which neurons established specific connections during development. The fly retina contains three classes of photoreceptor neurons (R1-6, R7, and R8), each responding to a specific spectrum of light, and connecting to a specific layer in the brain (Fig. 24). We focus on the layer-specific targeting of R7 neurons, which are essential for visual function and, moreover, amenable to genetic manipulation. By using visu-al-driven behavior assays, we previously identi-fied N-cadherin (Ncad) and the receptor tyrosine phosphatase LAR because of their requirement for R7 target selection (Lee et al., 2001). Mosaic analysis showed that Ncad and LAR are required "cell-autonomously" in R7 neurons for selecting the proper target layer. To understand how Ncad regulates R7 target selection, we developed a genetic method to analyze single Ncad mutant R7 axons during development. Our data suggest that Ncad provides adhesive interactions between R7 growth cones and their synaptic partners. Furthermore, we have uncovered the striking molecular diversity of Ncad, which is generated by alternative splicing. We are testing the hypothesis that combinatory use of Ncad isoforms forms a synaptic code that matches pre- with post-synaptic partners.

FIGURE 24 Connectivity of Drosophila retina

Molecular Diversity of N-Cadherin
Lee, Chiba, Hsu
Known as the chemoaffinity theory as proposed by Roger Sperry 40 years ago, the possibility that surface receptors exist to serve as molecular tags directing neuronal connectivity has remained a matter of great interest to researchers. Given that the number of synaptic connections far exceeds the predicted number of surface receptors, it is unlikely that the correspondence between pre- and post-synaptic partners mediated by such surface receptors represents a simple one-to-one relationship. It has been proposed that the com-binatorial use of surface receptors and the dif-ferential regulation of receptor function operate as additional mechanisms for directing specific connections. Furthermore, the receptor repertoire can be expanded by alternative splicing, as ob-served in several receptors such as vertebrate protocadherin CNR (cadherin-related neuronal receptor) and insect Dscam (Down syndrome cell adhesion molecule). The identification of Ncad as the key regulator of R7 targeting and the discovery of multiple Ncad isoforms gener-ated by alternative splicing allow us to examine directly how these mechanisms contribute to connection specificity.

As removing Ncad results in only about 75 percent R7 axon mistargeting (Lee et al., 2001), we suspected the existence of an additional molecule(s) that is functionally redundant with Ncad. Genomic sequence analysis revealed a cadherin gene (designated Ncad2) located next to the Ncad gene (Fig. 25). The Ncad2 gene is smaller than the Ncad gene and appears to result from partial duplication of the Ncad gene. RNA transcript analysis indicated that Ncad2 is expressed in the developing eye disk. In addition, mature Ncad2 transcripts consist of two alternative spliced forms, one of which contains exon 9 and encodes a type I receptor that is similar to Ncad while the other form does not contain exon 9 and encodes a secreted molecule. The soluble form of cadherins has been shown to disrupt cell-adhesion. These two Ncad2 forms are present in about equal proportion in the eye disk mRNA pool; in the mRNA of the whole animal, the receptor type is predominant. Thus, this alternative splicing appears to be regulated in a tissue-specific fashion. We are currently generating Ncad2 mutants for analyzing Ncad2 function in vivo.

FIGURE 25 N-cadherin loci contain constant regions and variable regions.

We detected the striking molecular diversity of Ncad generated by alternative splicing. The genomic sequence analysis revealed that the Ncad locus contains three pairs of exons (exon7-exon7', exon13-exon13', and exon18-exon18') in modular arrangement (Fig. 25). RNA tran-script analysis indicated that these exon pairs are used in a mutually exclusive manner, i.e., each mature transcript contains only one of the two alternative exons. All six exons are used in developing eye tissue, although exons 7, 13, and 18' are used predominantly in the eye disk. Using these three "exon modules" in a combinatorial fashion, the Ncad locus is capable of encoding eight isoforms, which each share the same mo-lecular architecture but has a unique amino acid sequence. Similar molecular diversity generated by modular arrangement of alternative exons has been observed in other receptors and might be a general mechanism for neuronal diversity and connection specificity.

The notion that the different Ncad isoforms might have distinct functions was first suggested by protein sequence analysis and later verified by cell aggregation assay. Exons 7 and 7' encode the C-terminal half of the eighth cadherin repeat and N-terminal half of the ninth cadherin repeat, and exons 13 and 13' encode the eleventh and twelfth cadherin repeats in a similar manner. Exons 18 and 18' encode one and one-half of the EGF repeat and one-half of the transmembrane region. Interestingly, we found nonconservative amino acid changes in two regions that potentially mediate calcium binding and homophilic interactions, suggesting that these isoforms might have different adhesive activity. To test such a possibility, we expressed individual isoforms in S2 cells and assayed their ability to induce cell aggregation. Our preliminary data showed that the isoforms encoded by exon 7 but not exon 7' are capable of mediating homophilic interaction. In contrast, the region of Ncad encoded by exon13 and exon13' does not significantly affect Ncad's ability to induce cell aggregation. We are currently testing whether these isoforms can mediate heterophilic, or graded, interaction.

Cellular and Molecular Mechanism of R7 Target Selection
Lee, Khan
Like the vertebrate cortex, the medullar neu-ropil of Drosophila brains is laminated, with each layer (or lamina) receiving input from different afferents. R7 neurons connect to the deepest layer of the external medulla neuropil (designated M6) while the lamina interneurons and R8 connect to the more superficial layers, M1-5 and M3, respectively. To elucidate the mechanism by which R7 growth cones choose the appropriate target layer, we first characterized the cellular events occurring during the development of R7 target connections. To resolve the complex cellular interactions involved, we genetically ablated the cells of interest and examined the effects on R7 target selection. Finally, we performed developmental mosaic analyses on mutants with R7 targeting defects, allowing us to assign molecules to specific types of cells and cellular recognition events in which these molecules function. The analyses revealed that R7 target selection proceeds in three consecutive stages, each of which involves different cellular and molecular interactions.

During development, R8, R7, and laminal interneurons (LNs) sequentially innervate the medullar neuropil. R8s differentiate first and extend their axons into the medulla. Approximately 12 hours later, R7 axons project, along the R8 axon shafts, into the medulla, followed shortly thereafter by the LNs. The first stage of R7 target selection involves R7 growth cones defasciculating from R8 axons, thereby allowing R7 growth cones to advance even deeper into the medulla. The separation of R7 and R8 growth cones requires LNs whose axons follow R7 and project into the region between the R8 and R7 growth cones. Genetic ablation of LNs, using an eye-specific allele of hedgehog mutants (hh1), blocks the separation of R7 and R8 growth cones. Although the interaction between LNs and R7 (or R8) appears to be critical, the molecular nature of these afferent-afferent interactions is not currently understood.

In the second stage, it is likely that the R7 growth cones enter the target layer by recognizing envi-ronmental cues provided by their target neurons. This process requires Ncad function. Removing Ncad in single R7 neurons causes premature arrest of R7 growth cones at the intermediate layer (M4-5) instead of their entering the presumptive R7 recipient layer (M6). R7 target neurons sprout dendritic processes in this layer and express Ncad. In vitro, Ncad is capable of mediating homophilic interaction. We favor the view that Ncad provides adhesive interaction between R7 growth cones and their targets.

Once R7 growth cones reach the presumptive R7 recipient layer, they stop and undergo a conformational change from a spear-like structure into an expanded conformation. This morphological change signifies the beginning of the third stage, the stabilization phase. In LAR mutants, R7 axons project into the correct layer at earlier stages but later retract to the R8 recipient layer, suggesting that LAR is required for stabilizing the interactions between R7 and their targets (Clandinin et al., 2001). In vertebrates, LAR has been suggested to up regulate N-cadherin-mediated homophilic adhesion by dephosphorylating â-catenin. We envision that LAR stabilizes R7 target interaction by up-regulating Ncad activity in R7 growth cones.

Our analyses have dissected R7 target selection into a series of complex cellular and molecular interactions that involve afferent-afferent interactions as well as afferent-target interactions executed in a temporally and spatially controlled fashion. At the molecular level, the Ncad-based adhesion system appears to play a key role in R7 target selection. The targeting specificity, we believe, is achieved by both differential modulation of Ncad adhesive activity in different growth cones and combinatorial use of Ncad and Ncad2 isoforms as well as by other surface receptors. It is likely that other determinants of connection specificity remain to be discovered. The developmental analysis presented here provides a conceptual framework upon which the molecular details can be built.