The Problem: The assembly of neural circuits

The human brain contains 100 billion neurons and each neuron, on average, elaborates on the order of 1,000 synaptic connections. Even in the fruit fly, there over 100,000 neurons interconnected by millions of synapses. Where these connections have been analyzed in detail at the level of individual identified neurons, one is struck by their incredible specificity.

Roger Sperry proposed that brain wiring would rely on many different cell recognition molecules with different binding specificities. He stated in a landmark paper in PNAS in 1963 that "cells must carry some kind of identification tags... by which they are distinguished one from another almost at the level of single neurons." The central goal of my laboratory is to identify the recognition molecules that allow neurons to distinguish between one another and to understand the overall logic by which connections are patterned through these interactions.

Our work over the past 15 years has uncovered the function of a large family of molecular tags in Drosophila encoded by the Dscam1 gene. This gene encodes on the order of 20,000 different ectodomains of a transmembrane protein through alternative splicing of a common transcript. These ectodomain isoforms share the same domain structure with both constant and variable immunoglobulin (Ig) domains. There are three variable domains and they determine binding specificity. Isoforms sharing the same three variable domains bind to each other, but not to isoforms differing at any one of the three variable domains. This binding specificity provides the molecular basis for self-avoidance, the tendency of neurites of the same cell type to repel one another. This involves two processes, self-recognition and the ability of neurites to discriminate between self and non-self. Each neuron expresses a unique combination of isoforms and this pattern arises through a probabilistic mechanism. This provides each neuron with a unique cell surface identity. Self-neurites recognize one another as self and this elicits a repulsive response. By contrast, the difference in expression between isoforms on the neurites of different cells allows them to discriminate between self and non-self, that is to discriminate between one another. Genetic studies reveal that thousands of isoforms are necessary to provide a robust system of self-avoidance.




Alternative splicing of the Dscam1 gene gives rise to a vast family of Ig superfamily cell recognition molecules. Identical isoforms comprising the same variable Ig2, Ig3 And Ig7 domains binds homophilically between cell surfaces. By contrast, isoforms differing by even one of the three domains do not bind.

Recent studies in mammalian brain development have uncovered a role for clustered protocadherins in regulating self-avoidance. They promote this process in an analogous way to Dscam1 proteins in Drosophila. In both systems, diverse cell surface proteins endow each neuron with a unique cell surface identity and this allows neurites of different cells to discriminate between self and non-self. Thus, a common strategy for self-avoidance has arisen via convergent evolution of two prominent families of cell recognition molecules.

The primary focus of our current research is to uncover the cellular recognition mechanisms by which neurons discriminate between appropriate and inappropriate synaptic partners. We have approached this problem in the Drosophila visual system. In recent work we have also approached this question in the developing mouse retina.

Wiring the Fly Visual System

Ramon y Cajal used the Golgi stain to uncover the extraordinary diversity of neuronal cell types in many different organisms. The insect visual system was a particular favorite of Cajal's. This was, in part, because of the extraordinary diversity of neuronal cell types (>100) in what he envisioned when he began his studies to be a simple system. Indeed, he recognized the complexity and organization of fly compound eye and the first two synaptic relays the lamina and medulla, were similar to the vertebrate retina, comprising the photoreceptor layer, and the two analogous synaptic regions the outer and inner plexiform layers, respectively. We now know the detailed pattern of synaptic connections between neurons within the lamina and medulla through serial EM reconstruction experiments. These patterns are complex and highly specific.



Genetic tools developed by the fly community provide a remarkable opportunity to explore the genetic programs and the specific cellular recognition molecules regulating assembly of circuits in the fly visual system. There are markers for virtually every cell type and methods for genetically manipulating them. Together, these methods allow one to screen for mutations, characterize gene expression through high-throughput methods and to analyze gene function at the level of single mutant neurons of a particular type in an otherwise wild animal. 



Over a century ago Cajal and Sanchez described the complex and exquisite morphology of diverse neurons in the insect visual system marking the birth of fly visual system research. 

Molecular mechanisms of brain wiring

Our work has uncovered a role for numerous cell surface proteins in regulating assembly of neural circuits in the fly visual system. These include, a classical cadherin (N-cadherin), a protocadherin (Flamingo), a receptor tyrosine phosphatase (Lar), the Dscam family of proteins and, most recently, the Dpr and DIP protein families. We have been particularly interested in uncovering the function of families of proteins with different recognition specificities, as we anticipate that these would provide insight into the molecular logic by which the axons and dendrites of large numbers of neurons discriminate between one another.

Identification of Dpr and DIPs for candidates of synaptic specificity 

We have taken genetic and molecular approaches to identifying cell surface proteins regulating neural development. With the advent of high throughput RNA-sequencing and methods for isolating highly purified cell types, we recently set out to identify cell surface proteins expressed just prior to the onset of synapse formation for a population of some seven different neurons which form synapses in different layers and with different cells within the medulla. We discovered that two interacting protein families, the Dprs and DIPs, each comprising Ig domains, are expressed in different pairs of synaptic partners. These expression patterns are consistent with a model in which they control synaptic specificity. Using CRISPR technologies, we have generated null mutations in several interacting pairs. Similar defects are seen in these mutants arguing that different ligand-receptor pairs regulate a similar process in different synaptic partners. Detailed developmental analyses are in progress to uncover the mechanisms by which these proteins work.


Different synaptic partners  (red and green) in the medulla express different Dpr/DIP pairs. As our maps of Dpr and DIP expression in neurons are only partial in the end all synaptic partners may have a matching Dpr/DIP pairs.

Patterning of synapses in the lamina

Dscam1, Dscam2 and Dscam4 act in different combinations and in different ways to pattern dendrites. Throughout the fly visual system synapses are of the multiple contact type. Each synapse comprises a single presynaptic active zone and multiple, postsynaptic elements, from two to five. The best characterized of these are tetrads. Each photoreceptor (R cell) axon terminal forms some 50 tetrads with dendrites of lamina neurons (L1-L3). These synapses contain constant and variable postsynaptic element compositions.

All contain one L1 and one L2. The other two sites at the tetrad are variable. Dscam1 and Dscam2 act in a redundant fashion to regulate the composition of postsynaptic elements at tetrads. L1 and L2 each express a unique combination of Dscam1 isoforms and L1 expresses Dscam2B and L2 expresses Dscam2A. The invariant matching of L1 and L2 at tetrads relies on homophilic repulsion between processes of the same cell mediated in parallel by Dscam 1 and Dscam 2.

In a separate series of experiments, we discovered that Dscam2 acts in combination with Dscam4 to regulate targeting of L4 dendrites. They act within the same pathway to promote a binary decision to associate with lamina rather retinal axons. Here Dscam2 and Dscam4 promote this interaction through an adhesive mechanism.

Although the area of contact between L4 dendrites and the dendrites of L1 and L2 are similar, L4 shows a marked preference for synapses on L2. Through RNA sequencing of all three neurons just prior to the onset of synapse formation, we have identified some 15 pairs of interacting proteins. These are candidates for preventing synaptic connections between L4 and L2, promoting synapses between L4 and L1, or both. Genetic experiments are in progress to uncover the function of these pairs of recognition molecules in controlling synaptic specificity.

Reconstruction of the dendrites of wild type (red) and Dscam2 mutant (green) L4 neurons. Arrowheads indicate axons passing through the lamina and the red and green "A" indicate the dendrites (double area indicates ectopic dendrite of the mutant neuron).

Genetic program for presynaptic differentiation

Presynaptic differentiation of all three major classes of photoreceptor neurons or R cells occurs synchronously. To explore gene expression during this process we carried out RNA sequencing of transcripts isolated at seven different type points prior to, during and following presynaptic differentiation. Time points were clustered during the transition between these phases. To isolate RNA in these experiments, we modified the ribosome trap technology developed by Heintz and colleagues. Here, we tagged the N-terminus of the large ribosomal subunit protein Rpl10 with two epitope tags separated by a protease site. This double-tag markedly improved the purity of the mRNA isolated. Dynamic changes in the expression of mRNAs was observed during the transition to presynaptic differentiation, with more transcripts changing then observed for transcripts isolated from sorted cells. Strikingly, changes in mRNAs encoding membrane and secreted proteins were most prominent. By contrast, and surprisingly, changes in the levels of transcripts encoding presynaptic proteins were modest. These transcripts, however, showed dramatic changes with selective lengthening of their 3'-UTRs. In turn, the extended UTRs showed marked enrichment of binding sites for RNA binding proteins known to regulate translation, mRNA transport and mRNA stability. This system, the synchrony of development and the large number of cells in each animal, and the ease of genetic and molecular manipulation, make it particularly well suited to a combined genetic, biochemical and cell biological analysis of synaptogenesis.

Live imaging in the intact developing pupal visual system

To gain a deeper and more complete understanding of the mechanisms underlying assembly of circuits, we have developed a non-invasive live imaging protocol to follow in detail specific steps in the assembly of the fly visual system. The long-term goal here is to integrate genetic, biochemical, protein localization and live imaging data to provide an in-depth understanding of specific steps in wiring the brain.

Here, we applied this approach to characterize a specific step in the guidance of R8 axons from a temporary target at the surface of the medulla neuropil to its target layer buried some 12 microns away within the neuropil. Previous studies from Iris Salecker and her colleagues at the Mill Hill Laboratories in London demonstrated that netrin produced in the incipient M3 layer was essential for R8 to target to M3. This targeting required the Netrin receptor Frazzled (FRA; the fly homologof mammalian DCC). This configuration, a Netrin expressing cell some distance from the growth cone it influences, is consistent with the canonical view of Netrin acting at a distance to determine the growth cone polarity or guidance to the target layer.

By comparing the behavior of wild type and frazzled (fra) mutant R8 growth cones, lacking a receptor for Netrin, in the same preparation we discovered that that all fra mutant growth cones reach and recognize the M3 layer but fail to stably adhere to it. This conclusion could have been reach through analyzing fixed preparations.

These studies underscore the need to combine genetic and biochemical studies with live imaging to understand wiring at a mechanistic level. This approach can be applied to different neurons in the visual system, exploring the dynamics of protein localization in growth cones and to different genes and combinations of them. We anticipate that integrating these approaches will provide new insights into the cell biology of neuronal morphogenesis, growth cones movement and synapse formation.

Searching for determinants of synaptic specificity in the mouse visual system

In the mouse, rod photoreceptor neurons form synapses with rod bipolar cells and cone photoreceptor neurons form synapses with cone bipolar cells. The cellular recognition molecules that determine this specificity remain unknown. We set out to identify candidate cell surface molecules regulating specificity by using RNA sequencing of all four neuronal cell types prior to, during and after synapse formation.

We have identified many differentially expressed cell surface and secreted protein with the potential to promote recognition specificity. Using electroporation of retina at postnatal stages with CRISPR guide RNAs and Cas9 to induce knock out mutations, we have uncovered phenotypes in mutant animals. These phenotypes are being studied in greater detail to understand how they regulate the interactions between rod and cone terminals and dendrites of bipolar neurons. We have pursued these studies in collaboration with Joshua Sanes at Harvard University.

Rods and cones (blue nuclei in top part of figure) form synaptic connections with the dendrites of rod bipolar neurons (green) and cone bipolar neurons (red). We are searching for the cell recognition molecules that control this specificity.