Retinal Mosaics: Pattern Formation Driven by Local Interactions between Homotypic Neighbors

Individual types of retinal neurons are distributed across the retina as regular arrays of cells, commonly referred to as “retinal mosaics” (Wässle and Riemann, 1978). This regularity in their patterning is generally assumed to ensure that all locations across the retina are sub-served by every type of retinal neuron, each participating in their own unique way to the local computations carried out before transmission of the visual signal to higher visual centers in the brain. The extent to which each retinal cell type evidences spatial order varies, with some mosaics being only slightly more regular than random distributions of cells (Reese, 2008a). With the exception of certain photoreceptor mosaics however (particularly in non-mammalian retinas), the spatial order present within the more regular of these mosaics lacks any higher-order lattice-like patterning (Cook, 2003), and simulations based on local-spacing rules that minimize proximity between immediate homotypic neighbors are sufficient to account for the spatial order present (Eglen, 2006). Indeed, the order present within a mosaic is generally independent of that in other mosaics (Rockhill et al., 2000), suggesting that the composition of retinal clones and the spatial relationships between them cannot account for the patterning in these mosaics. Rather, cells within a mosaic are free to move tangentially as they space themselves out, establishing territories surrounding each cell where the probability of a homotypic cell is lower than at further distances from the cell (Reese and Galli-Resta, 2002). Functional retinal mosaics are commonly associated with the phenomenon of dendritic tiling, by which each cell in a mosaic extends its dendritic processes to “tessellate” the retinal surface by limiting further growth at the boundaries with homotypic dendritic fields, establishing a dendritic coverage of the retina of 1.0. Yet many regular retinal mosaics contain nerve cells with extensive dendritic overlap between their homotypic neighbors, for instance, the retinal horizontal cells and the starburst amacrine cells, having dendritic coverage factors 6–30 times greater, in the mouse retina (Reese, 2008b). Each of these cell types has been shown to disperse tangentially at a developmental stage presaging the emergence of the mature pattern (Galli-Resta et al., 1997; Raven et al., 2005), yet whether these cells move via active migration or by nuclear repositioning has been debated (Cook and Chalupa, 2000). Dendritic outgrowth at these stages is limited (Galli-Resta et al., 2002), but recent live imaging studies suggest these cell types may initially tessellate the retina with their dendrites before differentiating their mature, and vastly overlapping dendritic arbors (Huckfeldt et al., 2009). Ablating single horizontal cells at these early stages leads neighboring cells to extend their processes into the ablated territory, occasionally followed by somal repositioning nearer the center of the emerging dendritic field (Huckfeldt et al., 2009). These results suggest that homotypic interactions via developing dendritic arbors enable neighboring cells to space themselves apart, but the molecular means by which this interaction is achieved is unknown. A recent elegant study in the developing mouse retina reported in Nature now reveals two transmembrane proteins that participate in this homotypic recognition critical for intercellular repulsion in the mosaics of starburst amacrine cells and horizontal cells. Kay et al. (2012) identified, via microarray analysis of purified retinal cell types harvested at P6, two genes expressed by only these cell types (Megf10 and Megf11; the former was also expressed in Müller glial cells). They confirmed selective expression in vivo, with Megf10 preceding that of Megf11. Megf10 was detected prenatally in starburst amacrine cells as these nascent neuroblasts migrate into their mosaics in either the inner nuclear layer (INL) or ganglion cell layer (GCL). Megf11, by contrast, was not detected until after birth in both starburst amacrine cells and horizontal cells. Both exhibited, via immunofluorescence, a distribution of labeling expected for membrane-associated proteins, outlining the soma, and dendritic processes. Indeed, each homolog encodes a transmembrane protein containing multiple EGF-like domains and a cytoplasmic domain with multiple binding sites for signal transduction, though until now, these proteins were implicated only as engulfment receptors for clearing degenerating cells. Kay and colleagues show, via gene knockout analysis, a conspicuous phenotype on the spatial ordering in these retinal mosaics without affecting the mosaics of other neighboring cell types. The starburst amacrine cell mosaics in Megf10 retinas, but not Megf11−/− retinas, showed a striking loss of regularity, rendering them comparable to random simulations of cells matched in density and constrained by soma size. These mosaics lacked the minimal spacing constraint that generates their patterning. Curiously, the horizontal cell mosaics were only slightly affected in either the Megf10−/− or Megf11−/− retinas, but were conspicuously impacted in the double-knockout, again being comparable to random distributions of cells. The double-knockout retina did not, however, impact the starburst amacrine cell mosaic any more than did knocking out Megf10 alone. To investigate further the role of Megf10 in mosaic formation, Kay et al. (2012) electroporated plasmid DNA into dividing cells of the newborn mouse retina, transfecting primarily later-generated rod photoreceptors, bipolar cells, Müller glial cells, and some amacrine cells (all of the starburst amacrine cells and horizontal cells are generated prenatally, and so escaped transfection). This protocol creates large fields of transfected retina in which the concentration of Megf10-positive processes is elevated (albeit misexpressed in Retinal mosaics: pattern formation driven by local interactions between homotypic neighbors