Abstract
To establish functional circuitry, retinal neurons occupy spatial domains by arborizing their processes, which requires the self-avoidance of neurites from an individual cell, and by spacing their cell bodies, which requires positioning the soma and establishing a zone within which other cells of the same type are excluded1. The mosaic patterns of distinct cell types form independently and overlap. The cues that direct these processes in the vertebrate retina are not known2, 3. Here we show that some types of retinal amacrine cells from mice with a spontaneous mutation in Down syndrome cell adhesion molecule (Dscam), a gene encoding an immunoglobulin-superfamily member adhesion molecule4, 5, have defects in the arborization of processes and in the spacing of cell bodies. In the mutant retina, cells that would normally express Dscam have hyperfasciculated processes, preventing them from creating an orderly arbor. Also, their cell bodies are randomly distributed or pulled into clumps rather than being regularly spaced mosaics. Our results indicate that mouse DSCAM mediates isoneuronal self-avoidance for arborization and heteroneuronal self-avoidance within specific cell types to prevent fasciculation and to preserve mosaic spacing. These functions are analogous to those of Drosophila DSCAM (ref. 6) and DSCAM2 (ref. 7). DSCAM may function similarly in other regions of the mammalian nervous system, and this role may extend to other members of the mammalian Dscam gene family.
We have identified a spontaneous mutation in mice that creates a loss-of-function allele of Dscam, the Drosophila homologues of which function in both isoneuronal self-avoidance for dendrite arborization and heteroneuronal self-avoidance for axon tiling7, 8, 9, 10, 11, 12. The recessive mutation arose in the BALB/cByJ genetic background and caused an overt neurological phenotype. Mutant and wild-type mice are indistinguishable at birth, but are severely uncoordinated by postnatal day 3 (P3); as adults, the mutant mice have spontaneous seizures and kyphosis, but are fertile and long-lived (>24 months; see http://www.jax.org/research/media/wild_type.html for video). Through positional cloning (see Methods), a mutation was identified in Dscam. Sequencing of genomic and complementary DNA from affected mice revealed a 38-bp deletion in exon 17, causing a frame shift resulting in ten unique amino acids followed by a premature stop codon (Supplementary Fig. 1). This mutation truncates the protein in the second fibronectin repeat (Fig. 1a). Dscam messenger RNA levels in the brain were reduced by 70% in affected mice, consistent with nonsense-mediated decay (Fig. 1b).
Figure 1: Identification of a mouse Dscam mutation.
a, A schematic of the DSCAM protein domain structure. The extracellular portion of DSCAM consists of ten immunoglobulin-like repeats (Ig) and six fibronectin domains (FN). The Dscam deletion truncates the coding sequence in the second fibronectin domain (arrow), which is before the transmembrane and PAK1-interacting domains (PID). b, Northern blotting of mRNA purified from whole brains of wild-type (+/+), Dscam+/- and Dscam-/- mice revealed a 70% reduction in Dscam mRNA in the mutant sample. 
-actin (Actb) was used as a loading control. c–f, Haematoxylin and eosin stained sections of Dscam-/- and wild-type retinas from P0 and adult mice. The Dscam-/- retina is indistinguishable from that of the wild type during embryonic stages of development and at birth (c, d). INBL, innerneuroblast layer; ONBL, outerneuroblast layer; ONL, outernuclear layer; OPL, outerplexiform layer. e, f, In the adult Dscam-/- retina, the inner nuclear (INL), inner plexiform (IPL) and retinal ganglion (RGL) layers are disorganized compared to those of the wild-type retina, whereas other retinal layers are indistinguishable from that of the wild type. g, In situ hybridization with Dscam antisense probes (Dscam AS) revealed expression in a subset of cells in the inner nuclear and ganglion cell layers. h, Whole control P15 retina stained with antibodies to DSCAM and TH. i, A section of a control adult (6–10-week-old) retina labelled using antibodies to DSCAM and TH. j, A section of wild-type P15 retina stained with DSCAM antibodies. k, A section of the Dscam-/- retina stained with an antibody to DSCAM. Scale bars: c, d, 80 
m; e, f, 106 m; g, 160 
m; h, 45 m; i, 120 m; j, k, 65 m.
Despite the overt neurological phenotype of Dscam-/- mice, examination of the central nervous system of adults and developing embryos did not reveal any gross disorganization, with the exception of a caudal folium of the cerebellum (Supplementary Fig. 2), implying a subtle phenotype or a subpopulation of affected cells.
We did find disorganization of postnatal retinal anatomy (Supplementary Fig. 3). At birth or before, the mutant retina was indistinguishable from that of control mice (Fig. 1c, d). However, by P4, the ganglion cell layer and developing inner plexiform layer (IPL) were disorganized (not shown). This disorganization persisted into adulthood in amacrine and ganglion cells (Fig. 1e, f), and was confirmed with marker analysis (Supplementary Fig. 4). Within these cell populations, many subtypes are present13, 14, and Dscam was expressed in a subset of cells in these layers (Fig. 1g). Double in situ hybridization with markers of defined amacrine subtypes indicated that Dscam was expressed in dopaminergic amacrine cells (tyrosine hydroxylase (Th)-positive) and in bNOS-positive (Nos1-expressing) cells, but not in choline acetyltransferase (Chat)-positive starburst amacrine cells (SACs) or disabled 1-positive (Dab1) AII amacrine cells (Supplementary Fig. 5). Immunofluorescence with an antibody against the extracellular domain showed that DSCAM completely overlapped dopaminergic amacrine cell staining in >400 cells examined. In whole-mount retinas at P15, anti-DSCAM staining was distributed across the soma and processes of dopaminergic amacrine cells, but redistributed to the cell body in the adult retina (Fig. 1h, i). Additional immunoreactivity was detected in retinal cross-sections in lower, more vitreal layers of the IPL, below the strata of TH-positive processes (Fig. 1j). DSCAM immunoreactivity was absent from the mutant retinas, consistent with a protein-null mutation (Fig. 1k).
Wild-type dopaminergic amacrine cells were first labelled by anti-TH staining at around P6, when the neurons are still extending processes (Fig. 2a). The spacing of dopaminergic amacrine cell bodies was analysed by density recovery profiling (DRP)—a statistical measure of the cells' spatial autocorrelation, measuring the probability of encountering a homotypic cell body at varied distances from the reference cell. Regularly spaced cells have a zone of exclusion for other homotypic cells, indicated by a below-average cell density at short distances15 (Fig. 2b). Although their spacing was normal, the Dscam-/- dopaminergic amacrine cells showed abnormal morphology at P6 (Fig. 2c, d). Examining individual cell morphologies in 20 isolated dopaminergic amacrine cells in control and Dscam-/- retinas revealed defects in arborization in the mutant retina (Fig. 2e, f). Quantification indicated no differences in total length of processes or in the number of branches per unit length, but mutant cells had a significantly larger number of processes that self-crossed (Fig. 2g–j). This phenotype was more pronounced at P10 (not shown), and eventually large fascicles of TH-positive processes from many cells form in the mutant retina (Supplementary Fig. 6). Dopaminergic processes in the wild-type adult retina form a broad, uniform plexus, and the cell bodies are evenly spaced (Fig. 2k, l). The fasciculated adult Dscam-/- dopaminergic amacrine cells had abnormal positioning of their cell bodies, which associated with the fascicles and clumped significantly (Fig. 2m, n and Supplementary Figs 6 and 7).
Figure 2: Arborization and mosaic patterning of dopaminergic amacrine cells in control and Dscam-/- retinas.
a, A wild-type P6 retina stained with anti-TH. Dopaminergic amacrine cell processes have little proximal contact and rarely self-cross. b, DRP analysis of cell-body spacing identified an exclusion zone surrounding dopaminergic neurons in wild-type retinas. c, A P6 Dscam-/- retina stained with anti-TH. Individual dopaminergic amacrine cells frequently self-cross. Error bars in all figures represent s.d. d, DRP analysis of Dscam-/- dopaminergic amacrine cells indicates normal soma spacing at P6. e, f, Representative isolated P6 wild-type (e) and Dscam-/- (f) dopaminergic amacrine cells are shown; insets contain traces of the neurites. g–j, Analysis of 20 isolated dopaminergic amacrine neurons from either wild-type or Dscam-/- retinas. g, No significant difference in the total neurite length was observed between wild-type and Dscam-/- neurons (t-test, P = 0.55). h, Similarly, no significant difference in neurite branches per unit length was observed (P = 0.38). i, j, A significant increase in the number of self-crossings per length of neurite (i) or per number of branches (j) was observed for the Dscam-/- neurons over controls (P = 0.003 and 0.015, respectively). k–n, Whole adult (6–8 weeks) wild-type (k, l) or Dscam-/- (m, n) retinas stained with anti-TH. TH-positive neurite fascicles run throughout the Dscam-/- retina (m) in contrast to the wild-type retina (k), in which neurites arborize evenly. l, DRP analysis indicates that dopaminergic amacrine cells in the adult wild-type retina maintain and expand their mosaic spacing and exclusion zones. n, DRP analysis of adult Dscam-/- retinas indicates aggregation of dopaminergic amacrine neurons. o, Schematic of arborization and mosaic formation in wild-type (left) and Dscam-/- (right) retinas. At P6, both wild-type and Dscam-/- dopaminergic amacrine cells are organized in a mosaic pattern; however, wild-type neurites arborize, whereas Dscam-/- neurites self-cross. By P10, wild-type dopaminergic amacrine cells have an increased exclusion zone, whereas Dscam-/- dopaminergic amacrine neurons aggregate along bundled fascicles. Scale bars: a, c, e, f, 133 m; k, m, 388 m.
In the mutant retina, bNOS-positive amacrine cells, which normally express Dscam, also had marked differences in the cell morphology, with short bundled processes and randomly spaced cell bodies (Fig. 3a–d). Starburst and AII amacrine populations that do not express Dscam were normal in their cell-body distribution (Fig. 3e–l), maintaining an exclusion zone in the control and Dscam-/- retinas. The meshwork of processes in the IPL from starburst cells in the inner nuclear layer was preserved in the Dscam-/- retinas, with no abnormal fasciculation seen, and the short processes of AII cells were also normal. Cross-sections of control and Dscam-/- retinas were also examined, and all cell types reached their appropriate vertical layer in the retina (Supplementary Fig. 8).
Figure 3: Morphology and cell body spacing of other amacrine cell types.
a, b, Cell bodies of bNOS-positive cells, which also express Dscam in control retinas, are spaced and arborized in a manner similar to dopaminergic amacrine neurons. c, d, The bNOS-positive amacrine cells show fasciculated processes and randomly spaced cell bodies in Dscam-/- retinas. Starburst amacrine cells have an exclusion zone around the cell body and a meshwork of arborized processes in both control (e, f) and Dscam-/- (g, h) retinas. (Cell bodies are pseudo-colored red to distinguish them from the meshwork of processes in S2 of the IPL (green).) AII amacrine cells are also evenly spaced with very short processes in control (i, j) and Dscam-/- (k, l) retinas. m, Schematic depicting the stratification of bNOS-positive, SAC, DAC and AII neurites in the inner plexiform layer. Dopaminergic amacrine neurites arborize predominantly in S1, AII amacrine neurites arborize in between S1 and S2, bNOS-positive amacrine neurites arborize in S3, and starburst amacrine neurites arborize in S2 (soma in INL) or S4 (soma in RGL). n, Schematic summarizing mosaic formation in the Dscam-/- retina. DSCAM-positive cell types arborize and maintain mosaics in the wild-type retina, but fasciculate and fail to maintain mosaics in the Dscam-/- retina. Cell types in which DSCAM expression was not detected arborize and maintain mosaics in both the wild type and the Dscam-/- retina. Scale bars: a, c, 388 m; e, g, 62 m; i, k, 123 m. Different scales are used on the ordinates of DRP graphs for each cell type.
In normal retinas, the mosaic spacing of each cell type, such as AII and starburst cells, is independent, allowing dendritic overlap of functionally different cells16. If DSCAM-mediated recognition was required for this process, then cell types that express Dscam might influence the spacing of each other. We therefore examined the distribution of dopaminergic amacrine and bNOS-positive amacrine cells, relative to each other, in control and mutant retinas (Fig. 4 and Supplementary Fig. 9). These cell bodies were randomly spaced with respect to one another in both genotypes, indicating that DSCAM is necessary for maintaining mosaics in homotypic cells but is not sufficient to confer identifying signals between cell types. Also, despite the fasciculation of dopaminergic and bNOS-positive processes in the Dscam-/- retinas, no associations of processes across cell types were seen.
Figure 4: Co-existence of dopaminergic and bNOS-positive amacrine cells.
a, In wild-type retinas double-labelled for two Dscam-expressing cell populations (dopaminergic amacrine and bNOS-positive), the cell bodies and processes appear independent. b, This was confirmed by cross-DRP analyses comparing dopaminergic to bNOS-positive cells, which indicated no difference from average density (random spacing). c, In Dscam-/- mutant retinas, the fasciculation and cell-body-spacing defects are evident, but the spacing of cells remains independent and they do not aggregate, as indicated by cross-DRP analyses (d). In addition, the fascicles of processes also remain independent. e, Schematic depicting the independence of retinal neuron mosaic formation. Retinal neurons form organized mosaics that are independent of the spacing of other cell types. f, Schematic depicting dopaminergic amacrine and bNOS-positive neuron arborization in the wild-type and Dscam-/- retina. Wild-type dopaminergic amacrine neurons arborize in S1, whereas bNOS-positive neurons arborize in S3. In the Dscam-/- retina, the arborization of dopaminergic amacrine and bNOS-positive neurites becomes diffuse but, although neurites of both cell types fasciculate with homotypic neurites (dopaminergic amacrine with dopaminergic amacrine or bNOS-positive with bNOS-positive), they do not fasciculate with heterotypic neurites (dopaminergic amacrine with bNOS-positive). Scale bars in a and c represent 240 m.
Mammalian DSCAM is therefore required for isoneuronal self-avoidance and for heteroneuronal recognition within a cell type. In the absence of DSCAM, the processes of an individual dopaminergic or bNOS-positive amacrine cell fail to form proper arbors, and instead fasciculate with processes of other cells of the same type, secondarily pulling the cell bodies out of their mosaic position. However, different cell types expressing Dscam do not influence each other's spacing. A neurodevelopmental role for vertebrate DSCAM was not previously established beyond an early, non-neuron-specific, developmental phenotype seen in zebrafish17.
Our results are consistent with previous studies showing that retinal mosaics form very early, before neurons have extensive arbors, and that arborization and mosaic formation are largely independent18, 19. The normal spacing of cells in the P6 Dscam-/- retina indicates that the early stages of mosaic formation are intact. However, DSCAM-mediated heteroneuronal self-avoidance is needed to prevent fasciculation of homotypic cells, which destroys their mosaic pattern.
Decreased cell death may also indirectly contribute to the disruption of mosaics, based on the increase in dopaminergic and bNOS cell densities in the mutant retina. The absolute density of cells that would normally express Dscam was increased by 
250% for dopaminergic amacrine cells, and by 300% for bNOS-positive cells in the mutant retina (DRPs in Figs 2 and 3). In contrast, the densities of the ChAT-positive (starburst) and DAB1-positive (AII) cells were slightly decreased. Dopaminergic amacrine cells rely in part on cell death to sculpt their mosaic pattern20. We saw decreased TdT-mediated dUTP nick end labelling (TUNEL)-positive cells in the developing Dscam-/- retina, but saw no change in the number of proliferating cells (Supplementary Fig. 10); we also identified that clumped dopaminergic amacrine cells were not clonal, indicating that they are not the result of overproliferation from a progenitor, but consistent with lateral migration of cell bodies (Supplementary Fig. 11). DSCAM-mediated self-recognition may therefore be upstream of cell death in the wild-type retina, whereas non-Dscam-expressing cells may use other mechanisms for spacing21.
The fact that DSCAM mediates both isoneuronal and heteroneuronal self-avoidance in given amacrine cell types is interesting given recent work in Drosophila. In flies, Dscam is alternatively spliced to generate tens of thousands of protein isoforms6. The homophilic interaction of DSCAM is highly isoform-specific, and this specificity is determined by the variable immunoglobulin domains8, 22, 23. Therefore, Drosophila neurons use DSCAM for isoneuronal self-avoidance and arborization because each cell recognizes only its own processes, which express the same set of isoforms8, 9, 24. Reducing or eliminating the diversity of DSCAM isoforms perturbs Drosophila circuit organization25, 26. Dscam2 in Drosophila is not extensively alternatively spliced and mediates heteroneuronal cell avoidance in tiling the axons of L1 neurons in the medulla of the Drosophila visual system7. Mouse DSCAM functions in a highly analogous way, promoting self-avoidance for both arborization (isoneuronal) and the prevention of fasciculation (heteroneuronal). Although the vertebrate genes are not extensively alternatively spliced, they do undergo homophilic and paralogue-specific binding5, 27, 28. The coexistence of Dscam-expressing amacrine cell populations suggests that stratification in different IPL layers may allow reuse of DSCAM as a recognition signal. Alternatively, additional co-signals may distinguish distinct cell populations, consistent with the continued lack of interaction between bNOS and dopaminergic cells in the Dscam-/- retina. Additional complexity for self-recognition could also be introduced by other Dscam gene family members, such as Dscam-like 1 (Dscaml1), which is expressed in a similar but discrete population of retinal neurons27, 28(P.G.F. and R.W.B., unpublished) or the highly homologous Sidekick proteins29.
Methods Summary
Genetics and Dscam analysis
The Dscam mutation was mapped by breeding to C57BL6/J and establishing linkage in F2 mice with polymorphic markers, and the mutation was identified by positional cloning approaches. Mice were genotyped by PCR from genomic DNA spanning the 38-bp deletion (Dscam forward, CTTTGCGCGTTATGATCCT; Dscam reverse, GTGGTGTCGATACTGATG). Control mice were littermates of the Dscam-/- animals to control for genetic background effects and age.
Immunocytochemistry
Whole-mount dissected retinas or cryostat sections were stained using standard immunofluorescence protocols. Antibody sources are listed in the Methods. Fluorescence images were collected using a Leica SP5 confocal microscope or a Nikon epifluorescence microscope with a digital camera.
In situ hybridization
In situ hybridization was performed using digoxygenin- and fluorescein isothiocyanate (FITC)-labelled riboprobes, detected by HRP-conjugated secondary antibodies and TSA-Plus fluorescent substrates (Perkin Elmer).
Analysis of mosaics
The spacing of cell bodies was analysed using DRP as described previously, with the modification that the closest bin in the analysis was corrected to account for the diameter of the cell body15, 16.
Full methods accompany this paper.
SSC, 1
