Genetic circuitry of Survival motor neuron, the gene underlying spinal muscular atrophy

Significance

Spinal muscular atrophy (SMA), the leading genetic cause of infant mortality, is a devastating neurodegenerative disease caused by reduced levels of Survival Motor Neuron (SMN) gene activity. Despite well-characterized aspects of the involvement of SMN in small nuclear ribonucleoprotein biogenesis, the gene circuitry affecting SMN activity remains obscure. Here, we use Drosophila as a model system to integrate results from large-scale genetic and proteomic studies and bioinformatic analyses to define a unique SMN interactome to provide a basis for a better understanding of SMA. Such efforts not only help dissect Smn biology but also may point to potential clinically relevant targets.

Abstract

The clinical severity of the neurodegenerative disorder spinal muscular atrophy (SMA) is dependent on the levels of functional Survival Motor Neuron (SMN) protein. Consequently, current strategies for developing treatments for SMA generally focus on augmenting SMN levels. To identify additional potential therapeutic avenues and achieve a greater understanding of SMN, we applied in vivo, in vitro, and in silico approaches to identify genetic and biochemical interactors of the Drosophila SMN homolog. We identified more than 300 candidate genes that alter an Smn-dependent phenotype in vivo. Integrating the results from our genetic screens, large-scale protein interaction studies, and bioinformatic analysis, we define a unique interactome for SMN that provides a knowledge base for a better understanding of SMA.

Spinal muscular atrophy (SMA), the leading genetic cause of infant mortality, results from the partial loss of Survival Motor Neuron (SMN) gene activity (1). Numerous studies indicate that SMN functions as a central component of a complex that is responsible for the assembly of spliceosomal small nuclear ribonucleoproteins (snRNPs) (reviewed in ref. 2). SMN is also reported to play additional roles, including mRNA trafficking in the axon (3). In humans, SMN is encoded by two nearly identical genes, SMN1 and SMN2, which are located on chromosome 5 (4). SMN2 differs from SMN1 in that only 10% of SMN2 transcripts produce functional SMN due to a single-nucleotide polymorphism that results in inefficient splicing of exon 7 and translation of a truncated, unstable SMN protein (1, 5, 6). The clinical severity of SMA correlates with the SMN2 copy number, which varies between individuals (7). As the small amount of functional SMN2 protein produced by each copy of the gene is capable of partially compensating for the loss of the SMN1 gene function, higher copy numbers of SMN2 typically result in milder forms of SMA. Therefore, genetic modifiers capable of increasing the abundance and/or specific activity of SMN hold promise as therapeutic targets.

The Drosophila genome harbors a single, highly conserved ortholog of SMN1/2, the Smn gene. SMN is essential for cell viability in vertebrates and Drosophila (8, 9). In Drosophila, zygotic loss of Smn function results in recessive larval lethality (not embryonic as might be expected) due to the rescue of early development by maternal contribution of Smn. The larval phenotype includes neuromuscular junction (NMJ) abnormalities that are reminiscent of those associated with the human disease, rendering this invertebrate organism an excellent system to model SMN biology (9–11). We previously described a genetic screen for modifiers of the lethal phenotype resulting from a complete loss-of-function Smn allele (12). This screen, although it probed half of the Drosophila genome, identified only a relatively small number of genes that affected the NMJ phenotype associated with Smn loss of function (12). In particular, it did not identify genes involved in snRNP biogenesis, the molecular functionality that is most clearly associated with SMN.

As the human disease state results from partial loss of SMN function, we reasoned that a screening paradigm using a hypomorphic Smn background (as opposed to a background that completely eliminates SMN function) would more closely resemble the genetic condition in SMA. Such a screen would therefore enhance our ability to detect elements of the Smn genetic network and, consequently, add significantly to our efforts to both dissect the Smn genetic circuitry and identify clinically relevant targets with mode of action.

This complementary screen proved to be more sensitive than our previous screen and led to the identification of over 300 genetic interactors. Taking advantage of the recently established Drosophila Protein Interaction Map (DPiM) (13), we related the newly identified genetic interactors to the SMN protein interactome, producing an integrated Drosophila SMN biological network. Finally, the Drosophila SMN network was evaluated for its relevance to human biology by mapping Drosophila SMN network genes to their human homologs and analyzing the human network using computational biology tools. The projection of the Drosophila SMN network derived from this study onto the human network derived from prior knowledge provides a rational basis for SMN functional hypotheses and network intervention points that carry potential for so-far-unexplored clinical applications.

Results

Genetic Screen for Modifiers of Smn-Dependent Lethality.

We examined several Smn-RNAi strains under the control of the yeast upstream activation sequence (UAS) to identify a hypomorphic Smn allele that could be used to model SMA in Drosophila more faithfully than alleles that completely abolish Smn function. We identified a transgenic strain, UAS-Smn-RNAi^FL26B (FL26B), that displays a less severe phenotype than the allele used in our previous screen (12). Specifically, expression of FL26B under the control of tubulinGAL4 (tubGAL4::FL26B) results in late pupal lethality whereby ∼50% of the pupae reach a more mature (pigmented) developmental stage before death than their less mature, unpigmented siblings (Fig. 1A).

Fig. 1.

Genetic modifiers of Smn using pupal lethality to screen the Exelixis collection of transposon insertions and their functional roles. (A) tubGAL4-directed expression of an inducible Smn-RNAi construct (UAS-Smn-RNAi^FL26B) leads to a fully penetrant pupal lethality where ∼40% of the pupae reach a pigmented developmental stage (Control). The remaining 60% die at an earlier unpigmented developmental stage. Introduction of an Smn deficiency into this background causes the entire population of pupae to die at the unpigmented stage (Smn deficiency), whereas ectopic Smn expression leads to survival to adulthood of the vast majority of pupae (Smn rescue). Introduction of previously isolated enhancers (d02492 and d09801) and suppressors (f05549 and c05057) of Smn (12) leads to quantitative changes in the fractions of pigmented vs. unpigmented pupae. (B) The screening strategy to identify genetic modifiers of the Smn pupal lethality phenotype using the Exelixis collection (illustrated for insertions on the third chromosome). The lethal phase for all Smn Tb+ transposable element (TE) (14) pupae in individual test crosses are scored and compared with those observed in control crosses (more survival = enhancers, more lethality = suppressors). (C) Drosophila functional categories overrepresented in the genetic modifier list. GO biological functions with the highest significance relate to known Smn functions such as alternative splicing or SMA-affected processes (neuronal and muscular). Enrichment significance is expressed as the –log₁₀ (P values).

We determined that this phenotype, measured by the ratio of pigmented to unpigmented pupae, is sensitive to Smn gene dosage, as reducing or increasing Smn copy number in the tubGAL4::FL26B genetic background resulted in enhancement or suppression, respectively (Fig. 1A). In addition, the ability of wild-type Smn (expressed by a UAS-Smn-GFP transgene) to rescue the lethality indicates that this phenotype does not result from off-target RNAi effects. These results were corroborated using an independent Smn RNAi strain. Finally, we demonstrated that previously identified Smn modifiers altered the Smn RNAi phenotype in the expected fashion (Fig. 1A). Together, these results demonstrate that the tubGAL4::FL26B phenotype is useful in detecting changes in Smn functional activity and is thus a suitable assay on which to base a modifier screen that will define and dissect the Smn genetic network.

Using this assay, we screened the Exelixis collection of genome-wide insertional mutations (http://drosophila.med.harvard.edu) (14, 15) for dominant modification of the lethality associated with the tubGAL4::FL26B strain (see Fig. 1B for scheme) and identified nearly 1,600 candidate strains. To eliminate false positives, all interacting insertions were retested. Only those that were not lethal in trans with tubGAL4 (12) and for which results could be clearly repeated were finally designated as modifiers. From this analysis, we identified 303 modifying strains (129 enhancers and 174 suppressors), which represents nearly 2% of the collection and a greater than 10-fold increase in hit recovery in comparison with our previous screening method. As the genomic location of some insertions in the Exelixis collection may be near multiple coding regions, unambiguous gene assignments are not always possible. Given this consideration, we determined that these 303 insertions potentially affected 340 Drosophila genes. In most cases, single genes were affected by single insertions, although 14 genes scored in the screen were represented by two or more alleles. No gene could be assigned for 36 of the 303 insertions. Careful human-to-Drosophila homology mapping using a combination of several prediction algorithms shows that, of the 340 Drosophila genes, at least 229 have human orthologs. Because a fraction of these genes are represented by multiple paralogs, we identified a total of 322 human genes corresponding to the 229 Drosophila modifiers (Materials and Methods and Dataset S1).

To assess the biological space covered by the Smn modifiers, the modifiers were evaluated using the Database for Annotation, Visualization and Integrated Discovery (DAVID) (http://david.abcc.ncifcrf.gov) (16, 17), which identifies biological processes statistically overrepresented in the set of genetic modifiers. Analysis of the Drosophila SMN network with DAVID identified known SMN molecular activity (alternative splicing) and SMN-dependent processes (“neuron differentiation,” “axon guidance,” “axonogenesis,” “muscle organ development,” and “dendrite morphogenesis/development”) (Fig. 1C) (12, 18). The predominant processes enriched in the modifier set reflect broad effects on morphogenesis and development, reinforcing the notion that Smn depletion has pleiotropic consequences.

Integration of Drosophila Genetic and Proteomic Interactors.

To determine whether the genetic modifiers are interconnected through physical interactions, we placed them in the context of the recently generated Drosophila Protein Interaction Map (DPiM) (13). We first retrieved the set of proteins that copurify with Drosophila Smn in DPiM and asked whether any of the modifiers belong to this Smn subnetwork. From all Drosophila proteins tested, eight form protein complexes with Smn and, interestingly, one, NAD-dependent methylenetetrahydrofolate dehydrogenase (Nmdmc), is also a genetic modifier. Expansion of the Smn subnetwork to include proteins that form complexes with each of the eight members in turn identified 35 additional proteins. Some of these proteomic interactors have been previously associated with Smn function [e.g., gem (nuclear organelle) associated protein2 (Gemin2), Gemin3, and several snRNPs] and nearly half (20/43) are known to be involved in mRNA processing, a functionality closely linked to the documented biochemical role of SMN (Fig. S1 and Dataset S2).

To better understand the biological functions identified by the Smn modifiers, we extracted the Drosophila protein complexes, which include the Smn genetic modifiers in DPiM. We found that 62 of the proteins corresponding to genetic modifiers passed the stringent statistical criteria necessary for inclusion in the DPiM interactome (Materials and Methods), which includes only the top 5% of the total interactions scored in the co-affinity purification coupled to mass spectrometry analysis. These 62 proteins were associated with 50 separate Markov clusters, a statistical definition of significantly associated proteins (19), each of which may define a functional protein complex (see Fig. 2 for the subnetwork of complexes identified by this analysis). Of these 50, we focused our attention on the 24 that are enriched for specific biological functions based on Gene Ontology (GO) terms—a system that provides a controlled vocabulary of terms for describing gene cellular and molecular functions (20). The majority of these clusters harbor a single modifier, but nine contained two or more (Dataset S3).

Fig. 2.

Extended Drosophila genetic subnetwork. The subnetwork of proteins connected to Smn and its genetic modifiers in the DPiM. A total of 62 Smn genetic modifiers (diamonds with red border) are directly connected to 361 other proteins (circles), also known as first-degree neighbors through 3,800 interactions. The thickness of the gray lines connecting the proteins is proportional to the interaction score in the DPiM. Proteins belonging to individual clusters with GO term enrichment are shown with different colors. Proteins colored gray are part of clusters that are not enriched for any specific GO terms. Smn protein (indicated by an arrowhead) itself is connected to only Nmdmc and shown as an interacting pair at the bottom.

Inspecting the GO biological function terms of these 24 clusters (annotated in Fig. 2), we find that many contain annotated functions previously linked to Smn activity, including RNA metabolism (“RNA splicing,” “mRNA binding”) (reviewed in ref. 3), translation control (“eIF3 complex”) (21–24), endocytosis (“Snap/SNARE complex”) (25, 26), and protein transport (“flotillin complex”) (27–29). Importantly, several genetic modifiers fall within protein complexes whose functions have not been previously associated with Smn function. These include complexes with phosphatase and kinase activities as well as those involved in intracellular signaling (“Toll-signaling pathway” and “Hedgehog signaling”) (Fig. 3). Two independent loss-of-function alleles of the ectoderm-expressed 4/Drosophila sterile alpha and armadillo motif (Ect4/dsarm) gene suppress Smn-dependent lethality. The recovery of Ect4/dsarm may provide additional evidence linking the Toll-signaling pathway to Smn activity as it encodes a Toll/interleukin-1 receptor homology domain. Intriguingly, loss-of-function Ect4/dsarm mutations also suppress Wallerian degeneration phenotypes observed in Drosophila and mouse models (30). Together, these data suggest that the Wallerian degeneration pathway may also affect Smn pathobiology, an effect that may be mediated through Toll signaling. Hence, this approach both confirmed and expanded the functional categories and pathways associated with SMN.

Fig. 3.

Human Smn genetic modifiers network. Ingenuity Pathway Analysis indicates that about one-third of the human orthologs of Drosophila Smn genetic modifiers (103 genes, green circles) are connected in a network involving 282 interactions with other modifiers and SMN1/SMN2 (yellow). Different types of interactions are indicated with distinct colored lines. A small number of modifiers (19 genes) have interactions only with other modifiers but not with SMN1/SMN2 (four pairs are two human orthologs of the same Drosophila gene). Nearly two-thirds of the modifiers (177 genes, not shown) have no interactions that would connect them to SMN1/SMN2 or other modifiers.

To further explore the relationships of the 62 proteins and their functional context within DPiM, we carried out a first-degree neighbor analysis to identify other proteins directly connected in the network that may represent potential biochemical interactors. This retrieved 361 additional proteins that are linked to the 62 Smn modifier interactors (Fig. 2 and Dataset S3). These 361 proteins include 128 that are directly linked to at least 2 of the 62 modifiers (Dataset S3). A GO term (i.e., functionalities) analysis of these proteins reveals additional connections to the spliceosome, RNA binding, and Snap/SNARE functions. Thus, considering genetic modifiers in the context of the DPiM provides us with a perspective on the diverse molecular functions that can modulate SMN activity in vivo.

Overlaying the Genetic Modifiers on the Human Interactome.

To study the Smn genetic circuitry in the human context, we generated a human view of the genetic Smn interactome, taking advantage of the manually curated source of human molecular interactions from the Ingenuity Pathway Analysis (IPA) database (Ingenuity Systems, www.ingenuity.com). This database integrates human gene relationships derived from a variety of experimental approaches, including proteomic studies. Using the human Smn proteins and the 322 human genes corresponding to the genetic modifiers identified in Drosophila (see above and Materials and Methods), we used the IPA knowledge base to derive a human SMN interaction network. Unlike DPiM, IPA is not limited to physical interactions, thus allowing consideration of other functional interactions including, for example, expression, localization, modification, and regulation. Such an approach allowed us to evaluate potential indirect relationships between the modifiers and SMN and to uncover molecular functions beyond the canonical role of SMN in the SMN complex.

Based on the generated network, we found that orthologs of the five modifiers heterogeneous nuclear ribonucleoprotein R (HNRNPR), small nuclear ribonucleoprotein D1 (SNRPD1), synaptotagmin binding, cytoplasmic RNA interacting protein (SYNCRIP), transformer 2 beta homolog (TRA2B), zinc finger protein 259 (ZNF259) are directly related to SMN1/2 (Fig. 3). HNRNPR, SNRPD1 and SYNCRIP proteins physically interact with SMN1 and -2 and have a role in RNA splicing (31). TRA2B was shown to regulate SMN2 protein levels by being a potent splicing enhancer (32). Finally, ZNF259 was shown to be necessary for the localization of Smn1 to nuclear bodies (33) and more recently emerged as a key modifier of SMA pathology in patients (34). These findings support the relevance of the identified Drosophila modifiers in understanding the human pathways underlying SMA pathology.

Furthermore, 98 modifiers are indirectly related via these five interactors to human SMN. Together, these 103 proteins representing one-third of the identified modifiers are interconnected in a human IPA database. In addition, we find another group of 19 proteins that make pair-wise functional interactions with other SMN genetic modifiers, but do not connect to the human Smn interactome. The remaining 177 proteins that are not connected in the human interactome (and the 19 that have pair-wise connections, 4 pairs of which are between two orthologs of the same Drosophila gene) potentially represent functions that have not been linked to SMN biology in human studies so far.

Expansion of the human SMN interactome beyond the 103 modifiers, by incorporating first-degree neighbor proteins of SMN in the database, connects an additional 48 modifiers (Fig. 4). This expanded human SMN network contains intermediates that are known to associate physically with SMN (GEMINs, HNRNPRs, and other small ribonucleoprotein family members) (3, 13) and signaling pathway elements known to affect SMN activity [fibroblast growth factor 2 (FGF2), glycogen synthase kinase 3 beta (GSK3B), mitogen-activated protein kinase kinase kinase 5 (MAP3K5)] (18, 35–38).

Fig. 4.

Extended human Smn genetic subnetwork. Adding SMN1/2 first-degree neighbors to the network shown in Fig. 3 generated an extended subnetwork. In this interactome, 151 human orthologs of Drosophila Smn genetic modifiers (green circles) directly or indirectly connected to SMN1/SMN2. A total of 48 modifiers are connected to SMN1/SMN2 through 71 additional intermediate proteins (pink circles) from the literature; 7 among them (blue circles) were also identified in replicate SMN bait purifications in Drosophila. Different types of interactions are indicated with distinct colored lines.

Validation of Genetic Modifiers at the Larval NMJ.

We chose to prioritize the modifiers for further functional characterization by using membership within both the Smn modifier network in DPiM (Fig. S1) and the expanded IPA human SMN network (Fig. 4) as the primary criterion. A total of 36 genes are shared between these interactomes (Fig. 5 and Dataset S4). The list includes 4 previously analyzed modifiers (Actn, Moes, Fim, cut up) (12, 26), 13 enhancers (Sod, Hsp68, Hsf, step, CG17838, ns1, shrb, VhaSFD, Rel, Hexo2, osa, CG13902, cathD), and 19 suppressors (CG30194, Nedd4, Pka-R2, Rho1, Tango7, Argk, 14–3-3-epsilon, Zpr1, CG9769, cenG1A, flw, comt, CG9062, l(3)72Ab, Karybeta3, HmgZ, Rbsn-5, sel, Paip2).

Fig. 5.

Network of 36 high-priority genetic interactors of Smn. The network shows 36 human orthologs of Drosophila Smn genetic modifiers (green circles) connected to SMN1/SMN2 in humans. These 36 modifiers are present in Drosophila DPiM as well as in the human IPA-based network and were selected for functional validation in NMJs. The intermediate proteins (pink circles) shown provide the shortest path to connect the modifiers to SMN1/SMN2. Different types of interactions are indicated with distinct colored lines.

Our previous analyses (12) indicated a strong correlation between the strength of the lethal Smn phenotype with the severity of NMJ abnormalities. Therefore, examination of the effects of Drosophila modifiers on the Smn NMJ phenotype was used to validate their roles in Drosophila and prioritize the corresponding orthologs for further investigation in vertebrate model systems. We used NMJ assays (12, 18) to sample the ability of a subset of these modifiers to alter the tubGAL4::FL26B NMJ phenotype. Examination of third instar larvae carrying a combination of tubGAL4::FL26B and each of 20 modifier strains revealed that 11 of the 20 (55%) strains show a statistically significant change in the number of synaptic boutons and are modifiers of the Smn NMJ phenotype (Fig. 6).

Fig. 6.

Genetic modifiers of Smn regulate NMJ morphology. (A) An NMJ derived from muscle 6/7 of a tubulinGAL4::UAS-Smn-RNAi^FL26B (tubGAL4::FL26B) third instar larva. (B and C) A reduction in NMJ size is observed upon introduction of enhancers c024569 (B) and d02738 (C) into the tubGAL4::FL26B background. (D–F) Introduction of suppressors c05501 and c06705 into this screening background leads to an increase in NMJ size, whereas suppressor d05711 (F) does not result in significant modification of the NMJ. (G) Quantitation of bouton numbers/muscle in individuals of indicated genotypes, which include enhancers (red) and suppressors (green) and normalized per muscle surface area (MSA), expressed as the percentage change compared with Tub Gal4:Smn RNAi alone. The blue bar represents the percentage of bouton number/MSA in normalized controls. ANOVA multiple comparison test was used for statistical analysis of the bouton numbers/muscle. Significance: P ≤ 0.05. (Scale bar, 50 mm.) n = 20. All preparations were stained with anti-HRP (red) and anti-discs large (Dlg) (green). The muscle nucleus was labeled using DAPI.

Effect of Genetic Modifiers on Smn Protein Levels and Localization.

Given that the severity of the disease phenotypes, in both patients and Drosophila models, correlates with SMN protein levels, we examined whether the prioritized genes affected SMN levels in Drosophila Schneider 2 receptor plus (S2R+) cells (39), the same cell line used to generate DPiM. We used an image-based analysis (36) to quantify SMN protein levels in S2R+ cells expressing inducible FLAG-HA tagged constructs corresponding to 21 Smn-modifying genes available from Universal Proteomics Resources (40). Untransfected cells within the same wells were used as controls. Surprisingly, we found that none of these ectopically expressed modifier genes altered total Smn protein expression significantly (Fig. S2A). Because Smn is localized in both the cytoplasm and the nucleus, we also used this assay to evaluate whether any of these modifiers altered its distribution between these two compartments. We found that seven modifiers significantly increased the nuclear Smn levels (Fig. S2B and Dataset S5), consistent with the notion that some modifiers from the screen, which affect Smn lethality and NMJ phenotype, may directly affect Smn distribution between the nucleus and cytoplasm. It is worth noting that a recent study (41) showed that mutant superoxide dismutase-1 (SOD1), known to cause familial amyotrophic lateral sclerosis, alters the subcellular localization of the SMN protein and disrupts its recruitment to Cajal bodies, thereby preventing the formation of nuclear “gems.” Sod was identified in our screen as an enhancer and was also shown to affect NMJ phenotype (Fig. 6). A subset of modifiers does not alter either Smn levels or Smn’s localization. How these modifiers affect the functional Smn remains to be explored. Given these results, however, it is important to note that small changes in SMN function may have an important biological impact, given that the severity of clinical manifestation in SMA patients correlates with small changes in SMN expression (1).

Discussion

Different animal models for SMA-associated neuromuscular defects have contributed significantly to a better understanding of spinal muscular atrophy etiology and genetics over the past few years. However, despite the well-characterized role for SMN in snRNP biogenesis, the links between its molecular function and the defects observed in SMA patients remain unclear. One of the key features of SMA is that the severity of the disease is dependent on SMN dosage, prompting the development of therapeutic strategies designed to increase SMN protein levels in patients. Still, it is essential to identify alternative approaches to modulate SMN activity. For this purpose, genetically tractable invertebrate systems may help to identify so-far-undiscovered elements of the SMN genetic circuitry. In particular, these organisms provide more flexible avenues to investigate the poorly understood role of SMN at the NMJ.

We have used Drosophila as our experimental system and previously described a genetic screen that uncovered a small number of Smn modifiers (12) of a strong loss-of-function mutant phenotype. In this screen, we identified functional links between Smn and the FGF pathway (12, 18), a relationship corroborated and extended by recent evidence in a severe mouse model of SMA that demonstrated widespread alterations of the FGF-system in both muscle and spinal cord (37).

The relatively small number of modifiers recovered suggested that a more sensitive genetic screen could provide extended information about the Smn genetic network. Our assessment of the lethal phase, exhibited by a mild loss-of-function Smn RNAi allele that more closely resembles the SMA hypomorphic condition, provided us with a more sensitive and quantifiable assay for genetic interaction. In comparison with our previous results, the RNAi-based screen described here provided us with a broader spectrum of modifiers, including those related to the canonical role of Smn in snRNP biogenesis as well as additional elements of FGF and bone morphogenetic protein (BMP) signaling (12). Our careful mining of the screening modifier list based on functional term enrichment and interactome analysis both in Drosophila and humans suggest that loss of Smn function may impact a range of developmental and maintenance-related programs of the whole neuromuscular system, including synaptic vesicle recycling, ion channels, and signaling pathways that regulate intrinsic cellular functions. Finally, this analysis also uncovered biological processes not previously associated with Smn.

Among the newly recovered genes, many are associated with RNA metabolism; however, the majority is not involved with canonical SMN activity of snRNP biogenesis and includes factors involved in transcription, posttranscriptional modifications, RNA transport, and translation regulation. Intriguingly, CG17838 is the Drosophila homolog of two closely related vertebrate RNA-binding proteins, HNRNPR and SYNCRIP, both of which bind to SMN in a yeast two-hybrid assay (31) and localize to mRNA-containing granules that are transported in cultured neurons (27, 31, 42). Because both SMN and HNRNPR affect localization of mRNA in axons (43, 44), this could have profound consequences on local translation in neurons (44).

Given the complexity of motor neuronal subcellular domains and their distance from the neuromuscular synapses, local regulation of the translation of synaptic proteins is likely to be important in synaptic plasticity and neurological diseases. In fact, many mRNA-binding proteins that function as key regulators of local RNA translation are associated with neurological diseases, including fragile X mental retardation 1 (FMRP) in fragile X syndrome, ataxin 2 (ATXN2) in spinocerebellar ataxia, and TAR DNA binding protein (TDP-43), fused in sarcoma (FUS), angiogenin (ANG), and ATXN2 in amyotrophic lateral sclerosis. Consistent with a possible role for Smn in affecting local translation, we recovered pumilio and eIF-4E, which are thought to be a part of the local translational apparatus in neuromuscular synapses (45). Furthermore, we recovered another translation regulator, eukaryotic initiation factor 4A (eIF-4A), which negatively regulates BMP-signaling components. Components of the BMP-signaling pathway have been shown to play a role in retrograde signaling in the NMJ (46, 47). Our results support the relationship between Smn and local translation and also provide an additional link to the retrograde signaling present in the neuromuscular system. Interestingly, perturbation of RNA translational control may result in defects in endocytosis (48, 49), a process that has been suggested to play a key role in neurodegenerative diseases, including Alzheimer’s (50) and Huntington diseases (51). Consistent with this notion, aberrant synaptic vesicle release at the NMJs in severe SMA mice may be evidence of impaired synaptic vesicle dynamics and/or abnormal active zone architecture (52). Further supporting a link between endocytosis and Smn (26), we identified Synaptotagmin1, Synaptotagmin-alpha, Syntaxin4, and comatose, the Drosophila homolog of N-ethylmaleimide sensitive factor, which are core components of synaptic vesicle recycling. We also recovered genes that are directly required for synaptic transmitter release, such as methuselah, or indirectly required, such as bruchpilot, which plays a role in constructing the active zone.

Although many of the recovered genes broadly impact the neuromuscular system, a subset includes the Drosophila homologs for several disease-related genes, including Neurexin 1 (schizophrenia and autism spectrum disorders) (54, 54), Dystrophin (Duchenne muscular dystrophy) (55, 56), Superoxide dismutase (41, 57), ras homolog family member A (amyotrophic lateral sclerosis) (58), and Ect4/dSarm (Wallerian degeneration) (30). Our recovery of these genes suggests that the genetic network identified by our screen may overlap, perhaps significantly, the genetic networks impacted by other human neurological disorders. If true, the use of Drosophila to explore other neurological disease networks via genetic screens, combined with the integration of additional genome-wide approaches, could identify common therapeutic targets that could potentially be tested in other disease models.

Because such genetic modifier screens are very sensitive and are able to recover a large number of modifiers that span a broad range of molecular functions, prioritization of candidates for further validation is essential. Here, bioinformatics mining allowed us to assemble a list of 36 Drosophila genes with human homologs for continued investigation. The majority of these tested genes showed a functional role in the structure and/or development of the NMJ in Drosophila, and some can alter the distribution of Smn in S2R+ cells, making them good candidates to pursue in vertebrate models of SMA. In addition, candidates may be drawn from a pool of modifiers that include members of signaling pathways such as G protein coupled receptors, kinases, and proteases, which are considered to be plausible small-molecule targets, or secreted or membrane proteins, which may be targeted by antibodies. Our results thus provide an extensive list of genes and pathways that have been functionally linked to an Smn-dependent phenotype and therefore represent potential therapeutic targets.

Materials and Methods

Drosophila Stocks and Culture.

All Drosophila stocks were maintained on standard Drosophila medium at 25 °C. The generation of the Smn alleles and constructs used in this study (Smn^X7, UAS-Smn-RNAi^FL26B, UAS-Smn-GFP) were originally described in ref. 12. The tubulinGAL4 and TM6B, Tb Hu tubulinGAL80 chromosomes used to generate the screening stock were obtained from the Bloomington Drosophila Stock Center.

Genetic Modifier Screen.

Individual strains from the Exelixis Collection (housed in the Artavanis-Tsakonas laboratory in the Department of Cell Biology, Harvard Medical School, Boston) were tested for the ability to genetically modifiy the tubGAL4-induced UAS-Smn-RNAi^FL26B pupal lethal phenotype by mating three to five males of the strain to three females of the w; UAS-Smn-RNAi^FL26B; tubGAL4/TM6B, Humeral (Hu), Tubby (Tb) tubGAL80 screening stock. After 2 d, adults were transferred to a fresh vial to create a duplicate cross and to maintain optimal culture density. Accordingly, adults were discarded from the duplicate after an additional 2 d had passed. Fifteen days after being initiated, crosses were scored by counting the number of pigmented and unpigmented Tb+ pupae along with any Hu+ adult escapers. The ratio of unpigmented pupae to the total of pigmented pupae plus escaper adults was compared with that derived from control crosses using males from the isogenic strain in which the Exelixis Collection was generated. Two control crosses were performed for each set of ∼100 strains that were tested. Crosses that failed to produce 40 experimental animals were repeated as above. A change in the ratio of unpigmented to pigmented individuals of 20% corresponded to an approximate 1.5 SD from the mean. Enhancers were defined as those mutations causing a reduction in this percentage (≤30%), whereas those that increased this percentage (≥70%) were classified as suppressors.

Gene Assignments for the Exelixis Collection of Transposon Insertions.

Data for Drosophila genes and Exelixis transposon insertion sites were obtained from FlyBase version 5.39, which was current as of August 2011. Of the 15,952 Exelixis stocks screened, 14,621 stocks were mapped in FlyBase to 15,326 transposons with specific insertion sites within the Drosophila genome. To determine the coordinates of insertion sites of transposons present in the remaining 331 strains, sequences from the region flanking the insertion sites (15) were searched against the Drosophila genome using the blastn program of the Basic Local Alignment Search Tool (BLAST) (59). The insertion sites of the transposons were then used to create gene assignments according to the following criteria: a transposon was considered to map to a particular gene if its insertion site were located in the transcription unit of the gene itself or within either 1 kb upstream of the transcription start site or 100 bp downstream of the transcription termination site.

Mapping Drosophila Genes to Human Orthologs.

FlyBase version v5.39 identifies 15,233 Drosophila genes, which have been iteratively mapped to human orthologs using predictions made by several prediction algorithms. Multiple predictions were combined into a single prediction by ordering the algorithms based upon lowest false-positive and highest false-negative rates (60) and choosing the first prediction. The methods used (in order) were inParanoid version 7 (61), orthoMCL version 5 (62), Homologene build 65 (63), and orthoMCL version 2 (64). The inParanoid predictions were selected using a probability score of 0.4. As a result, of the 15,233 Drosophila genes considered, 6,821 could be mapped to 6,703 human gene ids. This dataset was used to assign the human orthologs of Drosophila Smn modifiers as shown in Dataset S1.

Function and Network Analysis.

The functional enrichment GO terms of the Drosophila genetic modifiers was assessed using Expression Analysis Systematic Explorer (EASE) statistics available through DAVID Bioinformatics Resources using the Exelixis collection as a reference set (16, 17). Human and Drosophila protein–protein and genetic interactions were visualized and analyzed (first neighbors) using Cytoscape (65). Cytoscape BinGo plugin was used to evaluate the functional categories of the retrieved clusters in the Drosophila subnetworks. The human network was generated through the use of IPA (Ingenuity Systems, www.ingenuity.com) and further visualized and mined in Cytoscape.

Neuromuscular Junction Analyses.

Third instar larvae were dissected in cold 1× PBS and fixed at room temperature (RT) for 20 min in 4% paraformaldehyde. The samples were washed in 0.1% Triton-X100 in PBS (PTX) and incubated overnight at 4 °C with primary antibody. The primary antibody was washed off with PTX at RT. The samples were incubated at RT with secondary antibody for 90 min. This was followed by a PTX wash, and the tissues were mounted in Vectashield Mounting Media with DAPI (Vector Laboratories). Bouton numbers were counted using a Nikon TE2000 microscope, based on the Discs large protein (Dlg) and anti-HRP staining in the A3 segment muscle 4 as indicated. The muscle area for every animal was measured, and no significant difference was observed among different genotypes. At least 20–25 animals of each genotype were dissected for the bouton analysis. The ANOVA multiple comparison test was used for statistical analysis of the bouton number/muscle.

Microscopy.

All images were collected with a Nikon C1si spectral point scanning confocal connected to a Nikon TE2000 inverted microscope equipped with differential interference contrast, phase, and epifluorescence optics, a 40× Plan Fluor N.A. 1.4 objective lens, and the Perfect Focus System for continuous maintenance of focus. Mercury arc lamp (100 mW) illumination was used for viewing fluorescence by eye, and confocal scanning used the following Melles Griot solid-state diode lasers: 405 nm, 488 nm (10 mW), and 561 nm (10 mW). The image acquisition software used was Nikon EZ-C1. All samples were mounted and imaged in Vectashield mouting medium with DAPI (Vector Laboratories) at RT. Adobe Photoshop CS5 was used to pseudocolor images.

Analysis of Smn Levels in S2R+ cells.

Drosophila S2R+ cells (39), a derivative of Schneider S2 cells, were cultured in Schneider’s Drosophila medium (Gibco) with 10% (vol/vol) FBS, 100 U/mL of penicillin, and 100 μg/mL of streptomycin at 25 °C. FLAG-HA constructs (13) were transfected using TransIT-2020 (Mirus). One day after transfection, plasmid expression was induced with 0.35 mM CuSO₄ overnight. Harvested cells were plated on Con A-coated (0.5 mg/mL; Sigma) plate and fixed at room temperature for 20 min in 4% (wt/vol) paraformaldehyde (Electron Microscopy Sciences). The cells were washed in PBS-DT (0.3% sodium deoxycholate, 0.3% Triton X-100 in PBS) and incubated overnight at 4 °C with rabbit anti-Smn (1:2,000) (18) and mouse anti-Flag (1:1,000, Sigma). After washes in PBS-DT, the cells were incubated with mouse Alexa 488- and rabbit Alexa 568-conjugated secondary antibodies (1:500; Molecular Probes), followed by washing in PBS-T (0.1% Triton X-100 in PBS). The samples were mounted in Vectashield mounting medium with DAPI (Vector Laboratories).