Communication between neurons relies on precise patterns of interconnections between them. We are interested in understanding the molecular mechanisms by which these connections, referred to as synapses, are specified. This is a problem of daunting complexity. In the human brain there are about 1012 neurons linked together into a communication network by some 1015 synaptic connections. Even in the fruit fly Drosophila melanogaster, which we study, there are some 250,000 neurons and millions of synaptic connections. How do correct connections form during development? Presumably specific molecular labels on the surface of different neurons provide a basis for the cellular recognition that underlies this specificity. Identifying these labels and understanding how they work is the central goal of my laboratory.
Dscam Encodes a Large Family of Cell Surface Proteins That Regulate Neurite Self-Avoidance
We have studied a number of genes encoding cell surface proteins that are required for neuronal connections in the fly. Among these genes, Dscam is the most intriguing. Through a process called alternative splicing, the Dscam gene potentially encodes 38,016 different cell surface proteins. These include 19,008 different extracellular or ectodomains tethered to the cell membrane by two alternative transmembrane segments. All forms or isoforms of Dscam have the same overall domain structure or shape but different amino acid sequences within specific regions of the protein. Biochemical and genetic studies have coalesced to give a consistent picture of how Dscam diversity contributes to the assembly of neural circuits in the fly brain.
Studies in the early to mid-1980s led researchers studying the developing leech nervous system to propose that neurites, the axonal and dendritic processes of neurons, are endowed with the capacity to distinguish between their own processes or sister neurites and the neurites of all other cells. Interactions between self-neurites led to repulsion. They argued this would allow neurites to uniformly cover receptive fields, for instance, while allowing different neurons to share overlapping fields. More recent studies suggested that this might provide an important mechanism to ensure fidelity of branch segregation. Several lines of evidence support the view that Dscam proteins provide neurons with cell surface "identification tags," allowing neurons to distinguish between one another.
A series of biochemical and structural studies provided key insights into how Dscam functions as a cellular recognition molecule. Dscam ectodomains comprise an array of immunoglobulin-like or Ig domains. Three domains are variable, while the remaining domains are constant. Binding experiments with different isoforms support the view that more than 18,000 different ectodomains exhibit isoform-specific homophilic binding. That is, each isoform binds to itself but not to others. Binding was shown to require that each of the three variable Ig domains in one molecule precisely matches with its counterpart in the same isoform expressed on the surface of an opposing cell.
Structural studies in collaboration with Michael Sawaya and David Eisenberg (HHMI, University of California, Los Angeles) revealed that binding relies on the pairwise antiparallel interaction of discrete sequences within each variable Ig domain with their counterparts in their binding partner. Studies in vivo argue that this binding induces subsequent down-regulation of the receptor and activation of a cytoskeletal response, promoting the growth of processes away from each other.
Genetic studies established that Dscam is required for self-avoidance in many different developmental contexts, both within the central and the peripheral nervous systems. Each neuron appears to acquire a unique identity through stochastic expression of a set of isoforms. While thousands of isoforms are essential for the establishment of specific neural circuits in the fly to ensure identity, studies argue that the specific isoforms expressed by any neuron are unimportant. It is only important that the isoforms expressed by a neuron are largely different than those expressed by other neurons they encounter during development. (A grant from the National Institutes of Health provided support for the work on the role of different Dscam isoforms in mushroom body development.)
Formation of Neuronal Connections in the Fly Visual System
We have been studying the formation of the connections between photoreceptor neurons (R cells) and their targets in the brain. The compound eye of the fly contains some 800 simple eyes, or ommatidia, and each ommatidium contains eight R cells. These cells fall into three classes based on synaptic specificity. R1–R6 neurons connect to the first optic ganglion, called the lamina, and R7 and R8 neurons extend axons into and terminate within the medulla. Lamina neurons, designated L1–L5, also form connections in the medulla; three of these cells, L1–L3, are postsynaptic targets of R1–R6 neurons. The R7, R8, and L1–L5 neurons form a single optical processing unit, forming connections within specific layers, designated M1–M6, in the medulla. The connections of all but the L4 neurons are constrained to the same column. Genetic studies have provided some insights into the mechanisms underlying the formation of layer-specific connections and the mechanisms constraining connections to a single column.
N-cadherin, a homophilic cell adhesion molecule, plays a widespread role in regulating layer-specific connections for many but not all of these neurons. The dynamic spatial and temporal regulation of N-cadherin activity appears to underlie the reiterative use of N-cadherin in different cell types. N-cadherin plays an essential role in the targeting of R7 neurons to layer M6; in the absence of N-cadherin, R7 neurons terminate in layer M3, thus behaving as R8 neurons. This observation, among many others, underscores a close relationship between the genetic programs regulating R7 and R8 target layer specificity. Indeed, recent genetic studies argue that a key step in R7 targeting specificity is the active repression of an R8 targeting pathway in these cells. An intriguing finding arising from these studies argues that the same transcription factor plays a direct role in controlling the expression of rhodopsin and a cell surface protein controlling synaptic specificity, thereby coupling, in a direct way, sensory stimulus detection and processing of signals by the appropriate neural circuits.
Genetic studies defined three different mechanisms by which neurons restrict their connections to only a single column within a medulla layer. This process of restriction is often referred to as tiling. L1 neurons rely on homophilic recognition by Dscam2, a paralog of Dscam. Dscam2 promotes repulsion between the terminals and interstitial branches of L1 neurons in neighboring columns, thereby restricting their terminals to a single column within layers M1 and M5. By contrast, L5 neurons rely on N-cadherin–dependent adhesion between L5 processes and L2 terminals within the same column to restrict processes to a single column within layer M2. Finally, an autocrine mechanism requiring activin signaling to restrict motility of R7 terminal neurites and a paracrine interaction between neighboring R7 terminals, mediated by as yet unknown signals, act redundantly to restrict R7 neurites to a single column within layer M6. The autocrine mechanism appears to rely on nuclear translocation and activation of a distinct Smad-dependent transcription program.