LAB RESEARCH

Motor axon guidance and muscle targeting.The Drosophila motor axon network has provided one of the best systems in which to study axonal pathfinding mechanisms. The network is simple: 32 motoneurons innervate 30 body wall muscle fibers in each abdominal segment. Each motoneuron axon is targeted to a specific muscle fiber, and very few projection errors are made during normal development. Thus, the motor axon network is a genetically hard-wired map, and is an ideal system in which to study how genes control the formation of specific synaptic connections. In much of our work, we have focused on the roles of tyrosine phosphorylation in regulating motor axon guidance decisions. We are now conducting genetic screens to determine the mechanisms by which cell surface proteins label specific muscle fibers for recognition by motor axon growth cones.

Targeting of motor axons to specific muscle fibers. Despite the advances in characterizing molecules that regulate motor axon pathfinding, we still understand little about how specific muscle fibers are recognized as targets for synapse formation by these axons. Many mutations affect pathfinding decisions, leading to aberrant wiring of the neuromuscular system, but no single loss-of-function (LOF) mutations are known that block recognition of specific muscle targets. These results are most easily explained by invoking genetic redundancy in target labeling. If each muscle fiber were defined by a combination of several cell surface labels, removing one of the labels might not have a major effect on targeting of axons to that fiber. This would explain why targeting molecules have not been identified in conventional LOF genetic screens.
Studies of gain-of-function (GOF) phenotypes by other groups are consistent with the redundancy hypothesis. For example, the homophilic cell adhesion molecule Fasciclin III (Fas III) is expressed on only two muscle fibers, 6 and 7, and on the growth cone of the RP3 neuron that innervates these two fibers. Fas III appears to be a functional target label, because when it is ectopically expressed on other muscle fibers near 6 and 7, the RP3 neuron makes abnormal synapses on these Fas III-expressing fibers. However, when Fas III is removed by a LOF mutation, there is no effect on targeting of RP3 to 6 and 7. These results imply that Fas III can be used for muscle targeting, but that targeting of 6 and 7 can still proceed in its absence, presumably because these fibers are also labeled by other surface molecules that can be recognized by the RP3 growth cone when Fas III is not present.
These findings suggest that cell-surface proteins that label specific targets in the motor axon system might be identifiable by a GOF genetic screen in which candidate labels are ectopically expressed on all muscle fibers. If these proteins are functional labels, their misexpression might produce alterations in target recognition, as observed in the Fas III experiments described above. By identifying genes encoded in the Drosophila genome that can confer GOF phenotypes in which targeting of specific muscle fibers is altered, we will acquire the tools to understand the mechanisms involved in target recognition in this system. This type of screen should allow us to overcome the redundancy problem. For example, suppose one could identify three different cell-surface proteins that are normally expressed on a specific muscle fiber, but whose misexpression on other muscle fibers produces targeting errors. One might then predict that removing all three of these proteins by making a triple LOF mutant (through conventional or RNAi techniques) would now prevent targeting of this muscle fiber. Through these kinds of experiments, we could begin to understand the combinatorial code for muscle targeting. Insights into the motor axon targeting code would be likely to facilitate an understanding of targeting in other neuronal systems (e.g. the antennal lobe, optic lobe, and mushroom body), since candidate target labels are usually expressed by a variety of neuronal and non-neuronal cell types.
To conduct this GOF screen, we first created a database of all cell-surface and secreted (CSS) proteins in Drosophila that are likely to be involved in specific cell-cell interactions. The database was generated by database mining and reiterative computational screening. We defined all fly genes encoding proteins that contain domains known to be present in CSS proteins in other eukaryotes (including all of the 240 domains in the 'extracellular' portion of the SMART database, http://smart.embl-heidelberg.de/browse.shtml, that are represented in flies). We then eliminated several hundred genes that we thought were unlikely to be important for cell recognition, and defined a CSS cell recognition candidate collection of about 1000 genes.
To drive expression of these genes in muscles, we used the ‘EP’ system, in which a P element containing a block of UAS sequences that are responsive to the yeast transcription factor GAL4 is jumped around the genome. Like other P elements, EPs usually land upstream of genes. If a line bearing an EP upstream of a gene is crossed to a ‘driver’ line expressing GAL4 in all muscle fibers, the gene will now be expressed at high levels in muscles in the resulting progeny embryos and larvae. To find EP-like elements upstream of the CSS genes, we searched through about 40,000 different insertions that have been maintained in collections of Drosophila lines. These include the original EP set generated by Pernille Rorth, the EY insertion lines generated in the Bellen lab, the GS lines developed in Japan, insertions generated by Exelixis, Inc. and maintained at Harvard, and the GE lines developed by GenExel, Inc. We were able to identify insertions that can confer expression of 421 of the 1005 CSS genes in our database, representing about 40% of the repertoire and including members of all CSS protein families.
To screen for genes encoding potential targeting molecules, we crossed each of these insertions to a strong pan-muscle GAL4 driver and visualized motor axons and neuromuscular junction synapses in the resulting F1 progeny larvae by immunostaining. We have already identified over 30 genes that cause synaptic mistargeting on muscles 12 and 13, and ~160 genes that cause synaptic morphology phenotypes.
We have focused initially on the analysis of the mistargeting genes, as this is our primary interest and there are fewer of these to consider. We began by confirming the phenotypes with other GAL4 drivers. We then evaluated the normal expression patterns of the mRNAs encoding these genes, using in situ hybridization and antibody staining. This showed that many of the mistargeting genes are normally expressed in muscles or in regions of the periphery that are contacted by motor axons during axon outgrowth. We then studied LOF phenotypes for the genes by obtaining insertion mutations in or near the genes that reduce their expression, and by knocking down expression using transgenic RNAi techniques. We have already identified at least six genes for which reducing expression produces embryonic or larval phenotypes. The data thus far suggests that one gene encodes an epidermal protein that is necessary to drive motor axons away from the epidermis and toward their muscle targets. Another gene encodes a neural receptor necessary for ISNb guidance. Four genes may encode the type of proteins we were searching for in the screen: cell surface proteins that selectively regulate targeting of motor axons to specific muscle fibers.
We are continuing to examine these and other genes identified by the screen. Our long-term goal is to establish how the members of the gene families represented in the mistargeting gene collection work together to confer an accurate pattern of motor axon targeting. This will require making double and multiple mutants (or RNAi knockdowns that affect more than one protein expressed on an individual muscle fiber (M. Kurusu, A. Cording, et al., manuscript in preparation).

Neural receptor tyrosine phosphatases. In the 1990s, we showed that receptor-linked protein tyrosine phosphatases (RPTPs) are selectively expressed on CNS axons and growth cones in the Drosophila embryo, and that these RPTPs regulate motor and CNS axon guidance during embryonic development. RPTPs directly couple cell recognition via their extracellular domains to control of tyrosine phosphorylation via their cytoplasmic enzymatic domains. The extracellular regions of the fly RPTPs all contain immunoglobulin-like (Ig) and/or fibronectin type III (FN3) domains, which are usually involved in recognition of cell-surface or extracellular matrix ligands. Their cytoplasmic regions contain either one or two PTP enzymatic domains. The fly genome encodes six RPTPs (LAR, PTP10D, PTP69D, PTP99A, PTP52F, PTP4E), and we have generated or obtained mutations in all six of the genes encoding these proteins.
We have now performed a detailed characterization of the genetic interactions among all six RPTPs. We find that each growth cone guidance decision in the neuromuscular system has a requirement for a unique subset of RPTPs; thus, in a sense, there is an "RPTP code" for each decision. In some cases, the RPTPs work together, so that defects are only observed when two or more are removed. In other cases, however, phenotypes produced by removal of one RPTP are suppressed when a second RPTP is also absent. Our results provide evidence for three types of relationships among the RPTPs: partial redundancy; collaboration; and competition. Our most recent work will be described in a manuscript to be submitted soon that analyzes the complete pairwise genetic interaction matrices for control of CNS longitudinal tract and motor axon guidance mediated by all six of the RPTPs (M. Jeon et al., submitted). Our major efforts in this area are now directed toward understanding these relationships at the biochemical level, through definition of upstream (ligands) and downstream (substrates) components of RPTP signaling pathways (see below).

A genetic approach to identification of RPTP ligands. The ligands recognized by RPTPs in vivo have not been identified in any system. In order to understand how RPTPs regulate axon guidance, it is essential to know when and where they engage ligands, and how ligand binding affects enzymatic activity and/or localization.
One of our current approaches to identifying ligands is based on our observation that fusion proteins in which the extracellular domains of RPTPs are joined to human placental alkaline phosphatase (AP) can be used to stain live-dissected Drosophila embryos. Each of four fusion proteins (LAR-AP, PTP69D-AP, PTP10D-AP, PTP99A-AP) binds in a specific manner. Each fusion protein stains a subset of CNS axons and also binds to other cell types in the periphery. To identify the genes encoding the RPTP ligands, we are screening deficiency mutations that remove specific portions of the genome. We began by screening the Bloomington ‘deficiency (Df) kit’ of 266 fly lines. Each Df line was crossed to GFP balancers so that Df/Df embryos could be identified, and we then stained these embryos with each of 4 fusion proteins (LAR-HS2-AP, 69D-AP, 10D-AP, 99A-AP). Since each Df lacks a specific region of the genome, if homozygous Df embryos don't stain with a fusion protein, this indicates that this genomic region contains a gene required for ligand expression. Overlapping Dfs and point mutants can then be screened in order to identify the relevant gene within the Df.
Using this screen, we found a Df that contains a gene encoding a ligand that binds to LAR-AP, and have identified this ligand as Syndecan (Sdc). This work has been published (Fox, A.N., and Zinn, K. (2005) The heparan sulfate proteoglycan Syndecan is an in vivo ligand for the Drosophila LAR receptor tyrosine phosphatase. Current Biology 15, 1701-1711.) Sdc is a heparan sulfate proteoglycan (HSPG). Our results show that LAR binds to the glycosaminoglycan side chains of Sdc with nanomolar affinity, and that Sdc is required for DLAR-mediated axon guidance. We can generate motor axon guidance errors by overexpressing LAR on neurons, and find that the same errors are generated by ectopically expressing Sdc on muscles. This Sdc GOF phenotype is suppressed by LOF mutations in the Lar gene, indicating that LAR is epistatic to (downstream of) Sdc. This result shows that muscle Sdc can function in trans as a ligand for LAR on neuronal growth cones, and suggests that binding to Sdc increases LAR's signaling activity.
We have continued the Df screen, and have identified 4 regions required for 99A-AP staining. Ashley Wright is now screening overlapping Dfs and point mutations to find the responsible genes. Our results thus far already indicate that a novel glial-neuronal interaction is required to specify expression of the 99A ligand.
Our approach is general, and can be used to identify ligands for any 'orphan receptor' that has a Drosophila ortholog. We also used the method to define genomic regions required for expression of selected cell surface antigens, including those recognized by the 1D4 and BP102 monoclonal antibodies (mAbs). As part of the analysis, we have defined a new Df kit for embryonic screening, which uses alternative Bloomington Dfs to allow screening of regions of the genome whose removal in the normal Df kit causes early developmental failure. This new kit contains about 450 lines, and covers about 89% of polytene chromosome bands. It can be used to analyze any region of the genome for the desired embryonic phenotype. We have already analyzed about half of the genome for regions necessary for motor axon guidance by staining Df embryos with 1D4.

A gain-of-function screen for RPTP ligands. Despite the success of the Df screen (an LOF approach), it is clearly not capable of identification of all RPTP ligands, and may not even be capable of finding most of them. First, about 11% of the genome still cannot be screened, either because no Dfs exist there or because embryos homozygous for those regions do not develop. Second, and most important, the four RPTP-AP probes all stain subsets of CNS axons, in addition to other patterns outside the CNS. If multiple ligands for an RPTP were all expressed on CNS axons, removal of one ligand gene by a Df might not perturb staining enough to detect a difference from wild-type. We already know that this is the case for LAR: Sdc is expressed both on CNS axons and in the periphery, but only peripheral staining is eliminated in an Sdc mutant. CNS axons in Sdc mutants continue to stain with LAR-AP, and are also stained by a mutant version of LAR-AP that cannot bind to Sdc. These data show that there is at least one non-HSPG ligand for LAR that is expressed on CNS axons together with the HSPG ligand Sdc. Because of these limitations, we have developed a new GOF approach to ligand screening that allows direct identification of proteins that bind in embryos to an RPTP probe, regardless of whether such proteins are normally expressed in patterns that overlap with those of other ligands. This approach is also general and can be applied to any orphan receptor of interest that has Drosophila orthologs. It is based on observations made by Fox and Zinn (2005), who showed that when Sdc is ectopically expressed on muscle fibers, this produces ectopic muscle staining with LAR-AP, which normally does not bind to muscles. Thus, if one were able to express ligand genes in new patterns in the embryo, one would expect to be able to see additional staining with RPTP-AP probes and identify ligands in this manner.
Our approach is a directed EP screen. It uses the collection of EP element lines described above to ectopically express CSS proteins in new patterns in the embryo. To screen for new RPTP ligands, we are crossing each line in our CSS EP collection to GAL4 driver lines that confer ectopic gene expression in cells that normally do not stain with RPTP-AP fusion proteins. If I detect new staining patterns in embryos derived from such a cross, this may indicate that the gene driven by that EP-like element encodes a protein that can bind to the RPTP. We have already found a number of such lines, as described below. This screen is ongoing.

Searching for RPTP substrates. It is difficult to identify PTP substrates biochemically because PTPs usually do not display strong specificity in vitro. To find substrate candidates, we performed yeast two-hybrid screens with 'substrate-trap' mutant versions of PTP10D, PTP69D, PTP52F, and PTP99A. These 'trap' proteins form stable complexes with tyrosine-phosphorylated substrates because they bind normally but cannot catalyze dephosphorylation. We introduced a constitutively activated chicken Src tyrosine kinase into yeast together with the PTP trap constructs and the cDNA library, in the hope that it would phosphorylate relevant substrate fusion proteins made from cDNA library plasmids. We identified several classes of clones whose interactions with the substrate-trap RPTPs are dependent on coexpression of the tyrosine kinase, suggesting that they may be substrates. These include a cell-surface receptor identified with PTP52F, as well as some intracellular signaling proteins. For the receptor, we expressed a GFP-tagged version in transfected Drosophila S2 cells, and showed that this tagged protein selectively interacts with a cotransfected tagged substrate-trap PTP52F construct. Importantly, association in the transfected cell system requires simultaneous coexpression of the Src tyrosine kinase, and association is not observed if an enzymatically active version of PTP52F is used instead of the trap mutant. Both of these characteristics imply that the receptor is indeed a genuine PTP52F substrate. We are currently mapping the tyrosine(s) that are substrates for PTP52F dephosphorylation using this trap cotransfection assay.

Tracheal development: localization of apical proteins to the tracheal lumen is controlled by RPTPs. In the process of examination of double mutants lacking expression of the closely related proteins PTP4E and PTP10D, we noticed that the tracheal network exhibits severe defects in these embryos. We have shown that these defects, which include formation of huge 'bubbles' along the tracheal tubes, arise from mislocalization of apical proteins to intracellular vesicle compartments. Localization of basolateral proteins is unaffected. This mislocalization phenotype also involves dysregulation of Rho GTPases and receptor tyrosine kinases. Our current work is directed toward a mechanistic understanding of the relationships among the proteins whose perturbation gives rise to these apical targeting phenotypes (M. Jeon and K.Z., manuscript in preparation).

Genes controlling synaptogenesis in the larval neuromuscular system. Motor growth cones reach their muscle targets during late embryogenesis and then mature into presynaptic terminals that are functional by the time of hatching. The pattern of Type I neuromuscular junction (NMJ) synapses in the larva is simple and highly stereotyped, with boutons restricted to specific locations on each muscle fiber. These synapses continue to expand and change as the larva grows, because their strengths must be matched to the sizes of the muscle fibers they drive. This growth represents a form of synaptic plasticity, because it is controlled by feedback from the muscle to the neuron. Studies of NMJ synapses in flies are relevant to an understanding of synaptic plasticity in the mammalian brain, because the fly NMJ is a glutamatergic synapse, organized into boutons, that uses ionotropic glutamate receptors homologous to vertebrate AMPA receptors.

Control of synaptic local translation by Pumilio and Nanos. Our recent work on synapses has focused on control of synaptic protein translation. Local translation at synapses has been studied in Aplysia, mammalian, and arthropod systems. It has attracted interest because it is a mechanism that allows neurons to separately adjust the strengths of individual synapses.
To identify genes involved in synaptogenesis in larvae, including those that regulate local translation, we devised and executed a GOF screen of live third instar larvae. In the screen, we identified pumilio (pum), which encodes an RNA-binding protein that shuts down translation of specific mRNAs by binding to their 3' untranslated regions. Translational repression by Pum controls posterior patterning during embryonic development. In a 2004 paper, we showed that Pum is an important mediator of synaptic growth and plasticity at the NMJ. Pum is localized to the postsynaptic side of the NMJ in third instar larvae, and is also expressed in larval neurons. Neuronal Pum regulates synaptic growth. In its absence, NMJ boutons are larger and fewer in number, while Pum overexpression increases bouton number and decreases bouton size. Postsynaptic Pum negatively regulates expression of the essential translation factor eIF-4E (the cap-binding protein) at the NMJ, and Pum binds selectively to the 3’UTR of eIF-4E mRNA. These data suggest that Pum is a direct regulator of local eIF-4E translation, and that eIF-4E (which is normally limiting for translation) in turn switches on translation of other synaptic mRNAs. Pum also directly regulates the GluRIIa glutamate receptor. These results, together with genetic epistasis studies, suggest that postsynaptic Pum modulates synaptic function via direct control of local synaptic translation.
In our current work, we have identified the Pum cofactor Nanos, which works together with Pum to repress translation in the early embryo, as a participant in Pum regulation of targets at the NMJ. In nos mutants (or transgenic nos RNAi larvae), GluRIIa is upregulated as in pum mutants, and there are changes in presynaptic terminal morphology at the NMJ. Interestingly, eIF-4E is not strongly upregulated in nos mutants, consistent with the observation that Pum-eIF-4E mRNA complexes are not able to recruit Nos. Pum also regulates Nos levels at the NMJ; this may be an important autoregulatory control mechanism (K. Menon et al., manuscript in preparation).

Assembly of Pumilio into ordered aggregates as a regulatory switching mechanism. We are also studying Pum in another context: its potential role as a prion-like switch that could control synaptic translation via regulated assembly into an ordered aggregate. This project emerged from a computational search we performed to identify switch proteins that might have the capacity to form ordered aggregates. This is relevant to human disease as well, since proteins involved in many human neurodegenerative diseases share a propensity to form amyloid aggregates. One class of sequences that can form amyloids are domains rich in glutamine (Q) and asparagine (N). These are present in many metazoan proteins, including ~450 in Drosophila. Q/N domains are found in all yeast prions, and these domains have been positively selected during evolution, perhaps in order to allow reversible switching of the functional domain of the prion into an inactive aggregated state. We wondered this type of selection might also maintain Q/N domains in metazoans. To examine this question, we devised a computational search strategy to identify candidates for nucleic-acid binding prion switches in metazoan proteomes.
One of the two strong Drosophila candidates identified in this search is Pum. As described above, work by our group had shown that Pum is localized to the postsynaptic side of the larval NMJ, where it acts as a regulator of local mRNA translation. We found that a Q/N-rich domain (denoted NQ1) from Pum exhibits prion-like behavior in budding yeast, including heritable phenotypic switching and reversibility by guanidine hydrochloride. NQ1 purified from E. coli converts in vitro to an aggregated form that exhibits amyloid-like characteristics, including formation of fibers and Congo Red birefringence. To test whether NQ1 aggregate formation can perturb Pum's function in the nervous system, we created transgenic fly lines in which NQ1 expression is driven by GAL4. Our results show that postsynaptic NQ1 expression generates alterations in the NMJ that phenocopy the pum loss-of-function phenotype and interact genetically with pum mutations. Postsynaptic Pum overexpression is lethal, but co-overexpression of NQ1 rescues this lethality, suggesting that NQ1 can inactivate endogenous Pum. We are currently investigating whether amyloid formation is a pathological state or a normal regulator of Pum activity in vivo (A. Salazar, E. Silverman, submitted).