Cell, Volume 130
A Vast Repertoire of Dscam Binding
Specificities Arises from Modular
Interactions of Variable Ig Domains
Woj M. Wojtowicz, Wei Wu, Ingemar Andre, Bin Qian, David Baker,
and S. Lawrence Zipursky
Figure S1. Comparison of Binding Assays to Assess Dscam Binding Specificity
The binding properties of Dscam isoforms containing Ig2.7, Ig3.27 and highly-related Ig7 domains are
shown in four different binding assays. The table outlines the properties of each binding assay to indicate
1) sensitivity (as determined by whether heterophilic binding can be detected between isoforms 7.27.25
and 7.27.26), 2) whether the assay is readily quantitative, 3) whether the assay requires protein
purification, and 4) the number of binding experiments that can be conducted per day in our hands.
Representative binding data for these isoforms in each assay are shown (Matthews et al., 2007;
Wojtowicz et al., 2004)(J.J. Flanagan and S.L.Z, unpublished observations). F.L., full length.
Figure S2. Detection of Homophilic Binding Requires Clustering of Dscam Molecules.
(A) Immunoprecipitation assay to assess Dscam homophilic binding. Purified Dscam-Fc (bound to
protein G sepharose) was used as receptor protein to pull-down purified Dscam-His ligand protein. Pull-
downs were performed 1) in the absence (lanes 2-5) and presence (lanes 6-9) of Dscam-Fc receptor and 2)
in the absence (lanes 2-3, 6-7) and presence (lanes 4-5, 8-9) of α-His IgG-HRP which was used for
clustering Dscam-His ligand protein (Note that α-His IgG did not bind to protein G sepharose). (B)
ELISA-based binding assay to assess Dscam homophilic binding. Mouse α-Fc IgG was adsorbed to
ELISA plates and used to capture Dscam-Fc receptor protein from cell culture medium following
transient transfection. Dscam-AP ligand containing culture medium was added with or without α-AP IgG
for clustering Dscam-AP. Binding of Dscam-AP ligand to Dscam-Fc receptor was assessed by monitoring
AP activity following addition of AP substrate. This is a variation on the ELISA-based assay used
throughout this paper. AP activity was only linear over a 25-fold range while HRP activity was linear
over a 70-fold range. Additionally, more sensitive reagents are commercially available for HRP detection.
Therefore, the ELISA-based binding assay was optimized such that Dscam-AP is used as receptor and
Dscam-Fc is used as ligand with α-Fc IgG-HRP for clustering and detection of binding (see Figure 1 and
Figure S3. Swapping Only Surface Exposed Ig2 A’ β-Strand Residues Generates Promiscuous
Variable Domains. (A-B) Spacefill models of wild type variable Ig2 domains and A’ β-strand surface
swapped Ig2 domains are shown. Each variable Ig2 domain is represented by a different color. The
sequence alignment shows the A’ β-strand sequences (residues 105-114) for these Ig2 domains and the
surface exposed residues swapped. The binding properties of isoforms containing wild type and strand-
swapped Ig2 domains were tested using the ELISA-based binding assay. Binding is indicated as fold over
background by the number in each block. The average results of duplicate experiments are shown. (A)
Swapping residues between Ig2.4 and Ig2.7. (B) Swapping residues between Ig2.3 and Ig2.4. Potential
specificity residues were predicted by modeling different presumptive interfaces on the Ig2.1 interfaces in
the Ig1-4 crystal structure (Meijers et al., 2007).
Figure S4. The Ig2 Specificity Element can be Generalized to All Ig2 Variable Domains Tested.
(A) Spacefill models of wild type variable Ig2 domains (top row) and A’ β-strand swapped Ig2 domains
(bottom two rows). Each variable Ig2 domain is represented by a different color. Spacefill models of
swapping A’ β-strand from Ig2.1 (red) into Ig2.2-Ig2.12 (other colors) (Middle row). Spacefill models of
converse swaps of predicted A’ β-strand sequences of Ig2.4, Ig2.7 and Ig2.10 into Ig2.1 (Bottom row).
The boxed sequences on the right show an alignment of A’ β-strand sequences (residues 105-114) for all
12 variable Ig2 domains. (B) The binding properties of isoforms containing wild type and strand-swapped
Ig2 domains were tested using the ELISA-based binding assay. Binding is indicated as fold over
background (i.e. the number in each block). The unrelated control isoform 1.30.30 (not shown) was used
to provide a value for background binding. The average results of duplicate experiments are shown.
Specificity residues were predicted by modeling different presumptive interfaces on the Ig2.1 interfaces in
the Ig1-4 crystal structure (Meijers et al., 2007).
Figure S5. The Ig3 Specificity Element can be Generalized to Highly Diverse Ig3 Variable
(A) The A-A’ segment was swapped between pairs of Ig3 domains from distantly related regions of the
dendrograms (blue circles). (B-D) Spacefill models of wild type variable Ig3 domains and A-A’ segment
swapped Ig3 domains are shown. The sequence alignment shows the A-A’ segment sequences (residues
214-224) for these Ig3 domains and the residues swapped. The binding properties of isoforms containing
wild type and segment-swapped Ig3 domains were tested using the ELISA-based binding assay. Binding
is indicated as fold over background by a color scale and the number in each block. The unrelated control
isoform 1.30.30 (denoted “C”) was used to provide a value for background binding. The average results
of duplicate experiments are shown. (D) Note that heterophilic binding between Ig3.31 and Ig3.34 is
considerable though weaker for homophilic binding for each and, therefore, the color scale is changed to
aid visualization of specificity swaps. Despite strong heterophilic binding, the trend seen with the swaps
is the same as in B and C. Specificity residues were predicted by modeling different presumptive
interfaces on the Ig3.34 interface in the Ig1-4 crystal structure (Meijers et al., 2007).
Figure S6. The Ig7 Interface can be Generalized to Highly Diverse Ig7 Variable Domains. (A)
Ig7.20 and Ig7.25 differ at both residues 23 and 61, however, only one residue was swapped between
them. The binding specificity of these single-residue swapped Ig7 variants was tested using the ELISA-
based binding assay. Binding is indicated as fold over background by a color scale and the number in
each block. The unrelated control isoform 1.30.30 (denoted “C”) was used to provide a value for
background binding. The average results of duplicate experiments are shown. (B) Additional residues
corresponding to the positions in the modeled docking interfaces of Ig7.20 and Ig7.25 were swapped
between pairs of Ig7 domains from distant regions of the dendrograms (red circles). The table outlines the
Ig7 domains used for these studies and the residues that were swapped between them. The binding
properties of isoforms containing wild type and interface residue swapped Ig7 domains were tested using
the ELISA-based binding assay as in A. Some mutations resulted in a swap of binding specificity (left)
while others resulted in either a loss of binding specificity (i.e. exhibiting both homophilic and newly-
acquired heterophilic binding) or a reduction in homophilic binding (right).
Figure S7. Ig7 Docking Interface Modeling.
Distribution of binding modes resulting after symmetrical protein-protein docking of homology models of
Ig7.20 and Ig7.25. For each isoform 10 different homology models generated using Rosetta (Das et al.,
2007) were used as a starting point for symmetrical docking (Andre et al., 2007). The lowest energy
models from each docking experiment were collected and the position of Cα atoms of residue 23 (purple)
and 61 (orange) (in the center of the proposed binding region) in one monomer are shown. For illustrative
purposes, the positions of residues 59 (red) and 63 (blue) are also included. (A) Left panel, Unconstrained
docking models. Middle panel, Docking models constrained using a distance constraint of 8 Å between
residue 61 in both monomers. Right panel, Docking models additionally constrained to include models in
which residue 23 is also within 8.0 Å of itself in the other monomer. (B) Atom positions consistent with
two different docking orientations are shown: strict anti-parallel (large spheres) and criss-cross anti-
parallel (small spheres). Color code as in panel A.
Figure S8. Molecular Model for Binding Specificity at the Center of Symmetric Homodimer
Specificity switches at protein-protein interfaces frequently involve compensatory changes in either side-
chain size or polarity. For example, switches of a large-small residue interaction across an interface for a
small-large residue interaction, or a switch of a nonpolar-nonpolar residue interaction for a polar-polar
residue interaction are both expected to lead to specificity as the non-cognate pairs involve combinations
(big-big, small-small, hydrophobic-polar) likely to disfavor binding. However, these mechanisms cannot
explain specificity switches involving residues responsible for self-binding near the center of symmetric
homodimer interfaces, as residues in these domains interact with another copy of themselves. We have
generated molecular models that account for the specificity change observed by swapping residues 23 and
61 (Figures 5 and S5) and show illustrations for the Ig7.25 M61V mutant here. Spacefilling views of
residue 61 in molecular models of Ig7.25 are shown alongside a schematic representation of the Ig7
monomers (green ellipses) describing the effect of mutations on the interface. Docking models were
selected by screening for binding modes where the observed change in binding specificity fits the
experimental data. While, at first glance, this change appears to be in the big to small category (for which
homophilic specificity would not be expected) inspection of the docked complexes shows that
interdigitated M-M pairs and face-on-face β-branched and V-V residue pairs can pack quite well, while
the heterophilic M-V pair is less optimal as the β-branching prevents interdigitation of the long linear M
residue. M residues from opposing monomers interdigitate at the interface and allow homophilic binding.
Similarly, both V residues pack neatly in this space and permit homophilic binding. The same is true for
homophilic M-M and T-T, pairs at residue 23. The asymmetric substitution of M61V in only one
monomer leads to a clash between M and V that must be resolved by increasing the distance across the
interface or adopting an M conformation that points away. The same is true for heterophilic M-T pairs at
residue 23. This change in local packing can also propagate to adjacent residues thereby weakening other
interactions across the interface and preventing heterophilic binding. The idea behind the model for how
binding specificity is achieved at the center of symmetric homodimer interfaces is that long/thin matches
with long/thin while short/fat matches with short/fat.
Andre, I., Bradley, P., Wang, C., and Baker, D. (2007). Prediction of the Structure of Symmetrical Protein
Assemblies. PNAS in press.
Das, R., Qian, B., Raman, S., Vernon, R., Thompson, J., Bradley, P., Khare, S., Tyka, M. D., Bhat, D.,
Chivian, D. C., et al. (2007). Structure prediction for CASP7 targets using extensive all-atom refinement
with Rosetta@home. Proteins: Structure, Function and Bioinformatics in press.
Matthews, B. J., Kim, M. E., Flanagan, J. J., Hattori, D., Clemens, J. C., Zipursky, S. L., and Grueber, W.
B. (2007). Dendrite sef-avoidance is controlled by Dscam. Cell 129, 593-604.
Meijers, R., Puettmann-Holgado, R., Skiniotis, G., Liu, J.-H., Walz, T., Wang, J.-H., and Schmucker, D.
(2007). Structural Basis of Dscam Isoform Specificity. Nature submitted.
Wojtowicz, W. M., Flanagan, J. J., Millard, S. S., Zipursky, S. L., and Clemens, J. C. (2004). Alternative
splicing of Drosophila Dscam generates axon guidance receptors that exhibit isoform-specific homophilic
binding. Cell 118, 619-633.