View of the Dscam₁₋₈ dimer from the front (A), side (B), bottom (C), and top (D). The dimer forms a two-fold symmetric double “S.” Schematics of the eyes in (A) domain key show the perspective illustrated in (B)–(D). Left: Schematics illustrate the orientation of the eight Ig domains within each molecule. Right: Structures of the dimer in stereo view. Ig8 in molecule B is modeled with coordinates from model A. Coloring is as in Figure 2. Front view (A) shows view of dimer perpendicular to the two-fold symmetry axis. Note that the intermolecular contacts between molecules A and B occur between opposing Ig2-Ig2, Ig3-Ig3, and Ig7-Ig7 domains. Black ovals (C and D) indicate the symmetry axis.

Top: Electron micrographs suggest that Dscam monomers exhibit the Ig1–Ig4 horseshoe, whereas the remainder of the protein exhibits various conformations (Meijers et al., 2007). This lack of a precise topology is incompatible with dimer formation, as suggested previously (Schmucker, 2007). Bottom: We propose that matching at all three variable domains and Ig5:Ig6 intramolecular interactions that facilitate a sharp bend between Ig5 and Ig6 lead to the formation of the double “S” upon homophilic binding. This proposed conformational change upon homophilic binding may play a role in transducing the signal of homophilic binding to the cytoplasmic domain (orange strarburst), which initiates repulsive signaling leading to downregulation of the homophilic binding complex and cytoskeletal rearrangements, causing neurites A and B to be repelled from one another.

Dscam₁₋₈ forms a dimer (see Figure 3) containing molecules A and B, shown separately. Left: Schematics illustrate orientation of Ig domains within each molecule. Molecules A and B adopt an S shape. Molecule B is flipped 180 degrees relative to molecule A, revealing the back face. The domains are colored as in Figure 1. Molecule A is represented as a lighter version of these colors. Right: Crystal structures of molecules A and B shown in cartoon. Ig8 in molecule B is modeled with coordinates from model A. Sulphur atoms in disulfide bonds are drawn as spheres. Numbers on the structures indicate amino acid residues at Ig domain boundaries. N terminus is indicated. The remaining eight C-terminal domains of Dscam (not present in the crystal structure) would presumably lead to the two opposing cells (left for molecule A and right for molecule B). Domains Ig1–Ig4 comprise the “horseshoe” in the top half of the “S.” Domains Ig5–Ig8 comprise the bottom half of the “S.”

(A) Structure of the Ig7.30-Ig7.30 interface in the Dscam₁₋₈ dimer. Left: View of Ig7-Ig7 interface down the two-fold symmetry axis (black oval). Molecule A is in lighter colors. The Ig7.30-Ig7.30 interface is two-fold symmetric and formed between identical segments along the ABED strands comprising one face of the Ig domain. The two domains are aligned so that their A, B, E, and D strands are nearly antiparallel. The residues represented in sticks (light orange) have been shown to affect binding specificity (Wojtowicz et al., 2007). Right: Slices of the Ig7.30-Ig7.30 interface taken from four different depths illustrate surface complementarity between subunits. Intermolecular H bonds are drawn as dashed yellow lines.
(B) Ig7 docking models. Binding properties are based on binding studies (Wojtowicz et al., 2007). Left: Ig7.20-Ig7.20 homophilic interface exhibits complementarity across all of the ABED strands. Middle: Despite several residue differences (orange), the Ig7.20-Ig7.19 heterophilic binding interface exhibits complementarity across all of the ABED strands. Lower levels of binding between Ig7.20-Ig7.19 as compared with the self-binding exhibited by each suggest these residue differences decrease the complementarity each exhibits with itself. Right: Multiple residue differences (orange) at the Ig7.20-Ig7.4 heterophilic nonbinding interface lead to electrostatic and shape noncomplementarity (yellow starbursts) between residues on the B and E strands. Given this noncomplementary, it is not surprising these domains do not bind to each other.

(A) Dscam encodes cell surface molecules comprising immunoglobulin (Ig) domains (ovals), fibronectin type III domains (rectangles), a transmembrane domain (yellow line), and a cytoplasmic tail. The Dscam gene contains four variable exon blocks. Isoforms contain one variant from each block. These encode the N-terminal halves of Ig2 (red; 12 alternatives) and Ig3 (blue; 48 alternatives), all of Ig7 (green; 33 alternatives), and the transmembrane domain (yellow; two alternatives). Dscam proteins engage in isoform-specific homophilic binding. Biochemical studies led to the proposal that 18,048 of the 19,008 ectodomains encoded by the locus mediate isoform-specific homophilic binding (Wojtowicz et al., 2007). The minimal region of the ectodomain required to support homophilic binding includes the N-terminal eight Ig domains (Dscam₁₋₈) (Wojtowicz et al., 2004).
(B) Schematic illustration of the principles underlying Dscam isoform-specific homophilic binding. Top: The homophilic binding regions of Dscam proteins expressed on opposing cell surfaces. Bottom: Only the Ig2, Ig3, and Ig7 variable domains in two opposing isoforms are shown. “Matching” (as represented by the same shape) at all three variable domains is required for protein binding (with rare exceptions).

(A) Left: Schematic of molecule A. Right: The Ig5:Ig6 intramolecular interface. R496 in Ig6 makes significant contributions to the Ig5:Ig6 interface by forming H bonds with the backbone oxygen atoms of N465 and G467 in Ig5. Sequence alignment of 94 vertebrate and invertebrate Dscam proteins (see Figure S9) revealed that R496 and other residues are highly conserved.
(B) Flies containing the Dscam^R496W mutation exhibit phenotypes in neuronal patterning. Genotypes are as indicated. Left: Mushroom bodies within the central brain are highlighted in red. Upper middle and right: Schematics of mutant and wild-type. Four neurons within each mushroom body are shown. The cell body (oval) of each neuron resides in the Calyx. During development, axons branch at the base of the peduncle (P), giving rise to two sister neurites. Homophilic binding between identical sets of Dscam isoforms expressed on each sister neurite leads to repulsive interactions between sister neurites. Repulsion causes one sister neurite to extend into the dorsal lobe (D) and the other to extend into the medial lobe (M). Because sister branches share few, if any, isoforms with branches of other neurons, they selectively repel one another, leading to sister branch segregation with high fidelity. Gray box outlines region stained in lower panels. Lower middle and right: Axons in the D and M lobes are highlighted by anti-FasII staining. In Dscam^R496W flies, the D lobe is missing and the M lobe fails to extend to the midline (dotted line).
(C) Dscam^R496W mutant protein shows reduced homophilic binding. Left: Schematic illustration of ELISA-based binding assay utilized to assess homophilic binding is shown. Dscam ectodomain (blue) fused to alkaline phosphatase (Dscam-AP) is captured on a 96-well plate with anti-AP antibody. Dscam ectodomain fused to the Fc region of IgG (Dscam-Fc) is added and binding between Dscam-AP and Dscam-Fc is detected with anti-Fc antibody conjugated with horseradish peroxidase (HRP). Right: Homophilic binding levels of Dscam and Dscam^R496W protein are shown. The isoform used for binding is the same isoform used for crystallization studies of Dscam₁₋₈. Binding was quantified by measurement of HRP enzymatic activity. A control isoform differing at all three variable domains was used to provide a value for background binding. Data are represented as average ± SD.

(A) Superimposition of horseshoe domains from eight different Dscam structures reveals lack of variability in domain-domain interactions, suggesting conformational rigidity in Ig1–Ig4. Structures superimposed (listed from dark to light) are the following: Dscam₁₋₈, molecules A, B, and C; Dscam_{1−4 (1.34)} chains A and B (PDB ID 2V5M and 2V5R); and Dscam_{1−4 (9.9)} chains A and B (PDB ID 2V5S). RMS deviations in αCs are under 1.3 Å for all pairwise comparisons. Red and blue colors represent variable regions of Ig2 and Ig3, respectively.
(B) Rigidity of Ig1–Ig4 horseshoe originates from tight intramolecular packing of Ig2 and Ig3 within each molecule as illustrated via cross-section of the Ig1–Ig4 dimer. Light and dark shading distinguishes molecules in the dimer.
(C) Structure of Ig2.1-Ig2.1 interface in Dscam₁₋₈ dimer. The Ig2-Ig2 interface is viewed down the two-fold symmetry axis. Molecule A is in lighter colors. The Ig2.1-Ig2.1 interface is two-fold symmetric and formed between identical segments along the A and A′ strands (residues 107−114). Opposing N111 residues pack against one another at the symmetry axis (black oval). Left and right networks flank the symmetry axis. K112 forms hydrogen bonds (H bonds) with E107 and D109. D109 additionally forms H bonds with H114. Intermolecular H bonds are drawn with yellow dashed lines. The Ig2.1 interface was previously described (Meijers et al., 2007).
(D) Ig2 interface docking models. Binding properties based on previous binding studies (Wojtowicz et al., 2007). Coloring is as in (A). Schematics illustrate presence or absence of electrostatic and shape complementarity at interface. Top: Ig2.3-Ig2.3 and the Ig2.10-Ig2.10 homophilic interfaces are shown. Opposing N111 and I111 residues, respectively, pack against one another at the symmetry axis (black oval). Flanking left and right networks comprise unique networks that exhibit electrostatic and shape complementarity. Bottom left: Ig2.1-Ig2.3 binding heterophilic interface is shown. These isoforms engage in heterophilic binding, albeit at levels lower than the homophilic binding of each. Lower levels of binding may be attributed to the left network, which is bulkier than its wild-type counterpart (D107-Y114 versus E107-H114), and may introduce steric constraint and to the right network, which is less bulky than its wild-type counterpart (D107-H114 versus D107-Y114) and may destabilize the H bonds. Bottom right: Ig2.1-Ig2.10 nonbinding heterophilic interface is shown. Electrostatic and shape noncomplementarity (yellow starbursts) are observed at the left and right networks. The left network contains three negatively charged residues (E107, D109, and D112). The right network contains steric clash between positively charged residues (K112 and R109).
(E) Structure of Ig3-Ig3 interfaces in Dscam₁₋₈ and Dscam_{1−4 (1.34)} dimers. View of Ig3-Ig3 interface down the two-fold symmetry axis (black ovals) is shown. Ig3.30-Ig3.30 (left) and Ig3.34-Ig3.34 (right; [Meijers et al., 2007]) dimer interfaces are shown. Molecule A is in lighter colors. Interface is formed between the identical transition segment between the A and A′ strands in opposing domains. In Ig3.34, this region forms a single turn of a helix; in Ig3.30, it forms a β strand. Both interfaces exhibit electrostatic and shape complementarity. Given the differences in structure, it is not surprising that these domains do not bind to each other (Wojtowicz et al., 2007).