Identification of Genes Required for Neural-Specific Glycosylation Using Functional Genomics

Glycosylation plays crucial regulatory roles in various biological processes such as development, immunity, and neural functions. For example, α1,3-fucosylation, the addition of a fucose moiety abundant in Drosophila neural cells, is essential for neural development, function, and behavior. However, it remains largely unknown how neural-specific α1,3-fucosylation is regulated. In the present study, we searched for genes involved in the glycosylation of a neural-specific protein using a Drosophila RNAi library. We obtained 109 genes affecting glycosylation that clustered into nine functional groups. Among them, members of the RNA regulation group were enriched by a secondary screen that identified genes specifically regulating α1,3-fucosylation. Further analyses revealed that an RNA–binding protein, second mitotic wave missing (Swm), upregulates expression of the neural-specific glycosyltransferase FucTA and facilitates its mRNA export from the nucleus. This first large-scale genetic screen for glycosylation-related genes has revealed novel regulation of fucTA mRNA in neural cells.


Introduction
Neural cells require correct glycosylation patterns for their development, function, and viability. An example of this is the attachment of an a1,3-fucose moiety to an N-glycan core via a1,3linkage [1], a process which is prevalent in neural cells in Drosophila [2]. This a1,3-fucose moiety can be detected with an anti-horse radish peroxidase (HRP) antibody and therefore has also previously been referred to as an HRP epitope. The a1,3-fucose is thought to be essential for neural development, function, and behavior because a nac (neurally altered carbohydrate) Drosophila mutant that lacks this a1,3-fucose moiety exhibits deformation of the eyes [3], the misrouting of wing sensory neurons [4], and abnormal grooming behavior [5]. However, as it remains unclear that the nac mutation impairs only a1,3-fucosylation, the necessity of a1,3fucosylation for neural development and/or function in Drosophila has not been conclusively demonstrated.
The enzyme a1,3-fucosyltransferase (FucTA) [6], which is mainly expressed in neural cells, directly catalyzes a1,3-fucosylation. In addition to FucTA, other glycosylation-related proteins such as UDP-GlcNAc: a-3-D-mannoside-b -1,2-N-acetylglucosaminyltransferase I (Mgat1) [7], GDP-mannose 4,6-dehydratase (Gmd) [8], and a GDP-fucose transporter (Gfr) [9,10] are required for a1,3-fucosylation. Whereas Mgat1 provides a preferred substrate for FucTA by adding N-acetylglucosamine to the nonreducing end of an N-glycan, Gmd and Gfr are responsible for the synthesis and transport, respectively, of GDP-fucose, another substrate for FucTA. These genes, in contrast to the gene encoding FucTA, are widely expressed in various tissues and also utilized for other glycosylation processes such as O-fucosylation of Notch, a1,6-fucosylation of N-glycans, and formation of complex type N-glycans. Hence, the neural-specific expression of FucTA appears to account for the neural-specific regulation of a1,3fucosylation. However, the mechanisms regulating FucTA expression have remained largely unknown.
Forward genetic approaches have proven to be powerful methods of elucidating novel mechanisms. For example, the study of Drosophila genetics has yielded important contributions to our understanding of the developmental significance of proteoglycans [11,12] and Fringe-dependent Notch glycosylation [13]. Genetic screens for mutations affecting morphogenesis and growth factor signaling have now identified a number of genes involved in Notch glycosylation and/or proteoglycan formation. Most of these genes are conserved in mammals, suggesting that Drosophila is a useful model system for the study of glycosylation in metazoans. However, although previously performed screens of this nature have identified glycosyl enzymes and nucleotide sugar transporters, to date they have not been used to uncover regulators of these molecules.
To elucidate novel regulatory mechanisms underlying neuralspecific glycosylation, we performed a genetic screen in Drosophila and identified 109 genes required for glycosylation of a retinal neural cell-specific protein. These included 95 genes that are newly implicated in this process and 9 functional groups. Furthermore, 17 genes were identified to be specifically required for a1,3-fucosylation. Among these genes, we further analyzed the function of second mitotic wave missing (Swm), which contains an RNA-binding motif. Here, we show that Swm directly binds to fucTA mRNA, upregulates fucTA mRNA and protein levels, and facilitates the nuclear export of fucTA mRNA in neural cells. These results indicate that Swm is involved in neural-specific glycosylation in addition to the cell cycle, in which its involvement has been previously reported [14]. This report, the first large-scale screen for glycosylation in a multicellular organism, has thus identified a number of new genes directly or indirectly involved in glycosylation and unveiled a novel regulatory mechanism of neural-specific glycosylation.

Results
Biological significance of a1,3-fucosylation As a1,3-fucosylation has not proven to be essential for neural development and/or function in Drosophila, we first aimed to determine the neural function of a1,3-fucosylation. A phenotype of a piggyBac insertional mutant for the fucTA gene, fucTA f03774 , was examined. To validate this mutant, its central nervous system (CNS) was stained using anti-HRP antibody. The mutant CNS was negative for anti-HRP staining, suggesting that a1,3fucosylation was compromised (data not shown). CNS morphology was then compared between wild-type and fucTA mutant third instar larvae. The longitudinal length-to-width ratio of the ventral nerve cord (VNC) was significantly lower in fucTA mutants than in wild-type larvae ( Figure 1A and 1B), indicating that a1,3fucosylation plays an important role in neural development. However, the staining patterns of antibodies such as 22C10, BP102, anti-FasI, anti-FasII, and anti-FasIII were not noticeably different in the fucTA mutant (data not shown).

Screening strategy
To reveal novel regulatory mechanisms underlying neuralspecific glycosylation, a genetic screen was performed in adult Drosophila using the RNAi-mediated tissue-specific gene knockdown method. The procedure is summarized in Figure 2A. We chose eye-specific knockdown because eyes bear the neuralspecific glycan a1,3-fucose. Additionally, the ablation of genes essential for development and/or cell viability in eyes might not cause knockdown fly lethality, whereas their ablation in the central and/or peripheral nervous systems would be lethal. DsRNAs targeting different genes were expressed in Drosophila eyes by crossing the eye-specific Gal4 driver, GMR-Gal4, with the corresponding strains. These strains harbored inverted repeats of a portion of the corresponding cDNA downstream of the yeast upstream activating sequence (UAS), which is activated by the Gal4 protein (http://www.shigen.nig.ac.jp/fly/nigfly/about/ aboutRnai.jsp).
In our screen, the glycosylation of a model eye-specific glycoprotein was examined, as this allowed us to accurately measure the number of glycans added. We chose Chaoptin (Chp) [15] as our model glycoprotein based on the following advantages: (i) Chp is modified by neural-specific a1,3-fucosylation; (ii) detailed information is available on both the structures of the N-glycans and their attachment sites on Chp (described below in detail) [16]; and (iii) simple protein affinity purification is possible using an anti-Chp antibody (24B10). As shown in Figure 2B, Chp protein

Author Summary
Glycosylation plays crucial regulatory roles in various biological processes such as development, immunity, and neural functions. Accordingly, some glycans are generated in a stage-and tissue-specific manner. To address how such distinct glycosylation is regulated in different tissues, we performed a large-scale screen for genes involved in glycosylation of a neural-specific protein. We identified 109 genes, 95 of which are assigned for the first time as directly or indirectly involved in glycosylation. We further found that neural-specific glycosylation is regulated at the RNA level, which is a novel regulatory mechanism of tissuespecific glycosylation. could be purified as a single band detected by Coomassie Brilliant Blue (CBB) staining.
Chp is a Drosophila glycoprotein essential for rhabdomere formation in adult photoreceptor cells, which is mediated through cell adhesion activity [15,17]. N-glycosylation of Chp was recently shown to play a pivotal role in its stability, transport to the plasma membrane, and cell adhesion activity [18]. Chp N-glycan structures are classified as high-mannose type, pauci-mannose type, or complex type [16]. Glycan structures added to Chp are summarized in Figure S1. Interestingly, a1,3-fucose is added only at N1012, suggesting that Chp has a single difucosylated glycan at N1012. To detect these glycan structures, the binding of glycan probes such as lectins and anti-HRP antibody was examined. Immunopurified Chp was subjected to lectin blot or immunoblot analysis using ABA, ConA, DBA, DSA, LCA, Lotus, PHA-L4, PNA, SBA, UEA, and WGA lectins and anti-HRP antibody. The specificity of these lectins and anti-HRP antibody is summarized in Table S1. The binding abilities of these probes were quantified relative to the amount of Chp protein ( Figure 2C). These analyses suggested that the WGA, ConA, LCA, and DSA lectins and anti-HRP antibody would be useful as glycan probes in our assay system. As represented in Figure S1, WGA and ConA detect Nacetylglucosamine (GlcNAc) and mannose moieties of N-glycan core regions, respectively. LCA preferentially binds to N-glycans with a1,6-fucose and even weakly recognizes the core regions of Nglycans. Anti-HRP antibody binds to the a1,3-fucose moieties attached to the core [1,19]. DSA recognizes GlcNAc at the nonreducing end of N-glycans.
Chp protein purified from knockdown flies was subjected to lectin and immuno-dot blot analyses using the same glycan probes described above and an anti-Chp antibody. An example of the dot blot analysis is shown in Figure S2. The glycan moieties detected by these probes were quantified relative to Chp protein, and the zscores for each knockdown experiment were thus calculated. For verification of this procedure, a dsRNA that targets FucTA was expressed. The resulting knockdown of fucTA decreased the affinity of Chp to the anti-HRP antibody (66.2611.8% of the control, Figure 2D). Given the results of our previous mass spectrometry study showing that a1,3-fucose is added to the Chp N1012 site [16], Chp was purified from fucTA knockdown eyes, and mass spectrometry analysis was performed. The glycan structures attached to N1012 are summarized in Figure S3. The number of glycans with two fucoses present at N1012 (M3F2Gn2 and M3F2Gn3) selectively decreased in fucTA knockdown eyes ( Figure 2E). This finding is consistent with those of previous studies reporting that most single-or double-fucosylated N-glycans bear a1,6-fucose alone, or both a1,6and a1,3-fucose, respectively, in embryos [20] and adults [6].

Identification of glycosylation-related genes
A large-scale screen was performed in the current study using an RNAi library that was previously constructed in National Institute of Genetics, Japan (NIG, http://www.shigen.nig.ac.jp/fly/nigfly/ about/aboutRnai.jsp). With this library, the expression of 6923 Drosophila genes can be suppressed. From the primary screen followed by tests for reproducibility, 171 genes were identified as candidate genes, knockdown of which compromised glycosylation (Table S2). Since these candidate genes likely included falsepositives due to the off-target effects of RNAi [21], the results were validated by repeating the knockdown experiments using secondary sets of dsRNAs. During the construction of these secondary sets of dsRNAs, we ensured that the target regions did not overlap.
Of the 171 primary candidate genes, only 80 could be tested because, in most cases, the induction of the secondary sets of dsRNA resulted in lethal or severe eye malformation or cDNAs were too short to permit the design of secondary target regions. Of the 80 genes tested successfully, 57 were verified to be involved in glycosylation. Sources of the RNAi strains used for the verification of the 57 genes are listed in Table S2 (see ''Line ID or transformant ID used for validation''). When the validation results were compared with the calculated off-target probability scores (OTPS) provided by dsCheck software [22], 95.6% of the validated genes had scores of less than 3, whereas the scores of 60.7% of the genes yet to be validated were greater than 2. Thus, the genes with OTPS less than 3 were rescreened by testing whether knockdown of their suspected off-target genes led to glycosylation defects. When knockdown of the suspected off-target genes did not induce any glycosylation defects, the corresponding genes were classified as glycosylation-related genes (rank 2) along with the genes that were validated by secondary dsRNAs (rank 1) (Table S4).
To successfully identify additional glycosylation genes, we searched for genes that interacted with our newly identified genes in the yeast two-hybrid database BIOGRID (http://www. thebiogrid.org/) and in the genetic interaction data listed in FlyBase (http://flybase.bio.indiana.edu/). The RNAi fly strains harboring candidate genes that did not already exist in our library were obtained from the Vienna Drosophila RNAi Library Center (VDRC) [23] (Table S3). Among the 186 genes tested, 10 genes showed glycosylation defects. Further validation experiments identified 1 rank-1 and 5 rank-2 glycosylation-related genes among them. In total, 109 genes were eventually isolated and identified as glycosylation-related genes (Table S4). Significantly, although 14 of the 109 identified genes have already been shown to be involved in glycosylation, the remaining 95 genes were newly assigned to this category by our analysis.

Genes regulating neural-specific glycosylation
To identify genes specifically involved in neural-specific glycosylation, we selected genes whose knockdown induced only abnormal anti-HRP binding activity from among the 109 genes listed in Table S4. From this analysis, 17 genes showing a1,3fucosylation-specific defects were identified belonging to seven functional groups ( Figure 3B). In the RNA regulation group, swm (CG10084) [14], Pabp2 (CG2163) [24], and Hel25E (CG7269) are suggested to interact with each other in the Drosophila interaction database STRING (http://string.embl.de/) and literature [25].
Swm and Pabp2 are required for a1,3-fucosylation Since Swm and Pabp2 have been reported to physically interact with each other in BIOGRID, the roles of these genes were further analyzed in the context of glycosylation. To analyze the glycosylation defects induced by the knockdown of these genes in further detail, Chp glycans in the swm and Pabp2 knockdown lines were subjected to mass spectrometric analysis. The ratio of the a1,3-fucosylated form (M3F2Gn2) at N1012 was markedly decreased in both knockdown lines, whereas the other form (M3F2Gn3) was only slightly affected by swm or Pabp2 knockdown ( Figure 4A). These results suggest that Swm and Pabp2 are required for a1,3-fucosylation and that the M3F2Gn2 form might be more sensitive than the M3F2Gn3 form to decreases in Swm and Pabp2.
Since swm and Pabp2 mutants have been previously isolated [26] (Flybase), glycosylation defects were also examined in the amorphic and hypomorphic alleles of swm, swm F14 and swm F15 , respectively, and the hypomorphic allele of Pabp2, Pabp2 KG02359 . All cells homozygous for swm mutations were generated in whole Drosophila eyes using a modified FLP/FRT system [27]. Chp was isolated from these mutant eyes and detected by immunoblotting with anti-Chp and anti-HRP antibodies to quantify the amount of a1,3-fucose added to Chp. The a1,3-fucose added in swm F14 , swm F15 , and Pabp2 KG02359 cells decreased to 22.7% 6 4.8%, 28.3% 6 7.3%, and 59.5% 6 8.8% of wild-type levels, respectively ( Figure 4B). These results are further evidence from knockdown experiments that Swm and Pabp2 are somehow required for a1,3fucosylation.
Although Pabp2 is known to participate in polyA addition, the function of Swm is largely unknown. Thus, we focused on the role of Swm in glycosylation. Modification of a1,3-fucose is widely utilized for Drosophila neural proteins. Thus, we first examined whether swm depletion affects not only Chp but also a wide range of other proteins. Homogenates of adult eyes heterozygous or homozygous for swm F14 and swm F15 were analyzed by immunoblot analysis with an anti-HRP antibody ( Figure 5A). Several positive bands were detected, including a high-molecular-weight band corresponding to Chp (,150 kDa). Quantitative comparison between heterozygotes and homozygotes of the same swm alleles revealed that anti-HRP antibody affinity was reduced in most bands in the homozygotes, with the exception of an approximately 30-kDa product ( Figure 5A and 5B). The decrease was much larger in swm F14 than in swm F15 mutants, consistent with the severity of these mutations. Taken together, our data indicate that Swm is required for the a1,3-fucosylation of most proteins that are usually modified by a1,3-fucose. The observed requirement of Swm for most a1,3-fucosylation was also supported by the immunostaining of eyes with anti-HRP antibody. Adult photoreceptor cells from swm knockdown and control flies were simultaneously stained with anti-HRP and anti-Chp antibodies ( Figure 5C). Control photoreceptor cells appeared yellow in color because the intensities of green signal (anti-HRP) and red signal (anti-Chp) were nearly equal in this condition. In contrast, swm knockdown cells appeared relatively red in color, suggesting that the intensity of green signal was reduced compared to the red signal. These data clearly indicate that the signal strength of anti-HRP relative to anti-Chp was reduced in the knockdown photoreceptor cells compared to control cells.
Since a1,3-fucosylation is directly catalyzed by FucTA, the Swm-dependence of FucTA protein expression was examined. Because no anti-FucTA antibody was available, a knock-in fly was generated in which the fucTA locus was replaced by a Myc-tagged fucTA gene ( Figure S4A and S4B). When swm dsRNA was expressed by GMR-Gal4, FucTA-Myc protein levels were reduced to 47.562.4% of control values as revealed by immunoblot analysis (n = 2) ( Figure 6A and Figure S4C). The reduction of FucTA-Myc levels in swm mutant tissue was also examined. We costained the CNS of first instar larvae with anti-Myc and anti-GM130, a Golgi marker. The number of Golgi harboring FucTA-Myc protein was markedly decreased in swm F14 mutants (2.0% of total Golgi) compared to wild-type larvae (43.1% of total Golgi, Figure 6B). The different ratios of FucTA-Myc reduction observed in the biochemical and immunostaining experiments may be due to different levels of Swm depletion in knockdown and mutant flies.

Functional analyses of Swm
Although Swm has an RNA recognition motif (RNP-1) that is present in a wide variety of other RNA binding proteins [26], it remained undetermined whether Swm directly bound to mRNA.  Binding of Swm to fucTA mRNA was thus examined by expressing Flag-tagged Swm in the Drosophila neural cell line BG2-c6 [28] and by performing immunoprecipitation with an anti-Flag antibody. The quantity of endogenous fucTA mRNA that coprecipitated with Swm-Flag was higher in Swm-Flag-expressing cells than in controls in which Swm-Flag was not expressed or immunoprecipitation was performed using a control IgG ( Figure 6C). These results strongly suggest that Swm binds to fucTA mRNA in vivo.
The amount of fucTA mRNA under the control of Swm was next examined. dsRNA targeting swm was introduced into BG2-c6 cells, and the amount of endogenous fucTA mRNA was measured. In addition to a decrease in swm mRNA, to 27.3 6 1.9% of the control level (n = 3), the amount of fucTA mRNA also decreased to 60.6 6 10.5% of the control level ( Figure 6D). Furthermore, the stability of fucTA mRNA was examined in swm knockdown cells. The amount of fucTA mRNA was measured at 30, 60, and 90 min after total transcription was inhibited by addition of actinomycin D. Decay rates of fucTA were indistinguishable between swm knockdown and control cells (swm, t 1/2 = 97.72; control, t 1/2 = 100.48, Figure 6E). These data suggest that Swm is involved in the expression of fucTA mRNA but not in its stability.
Swm was recently reported to incorporate into the complex that regulates nuclear export of mRNAs [29]. The necessity of Swm for the nuclear export of fucTA mRNA was therefore examined in BG2-c6 cells. Following separation of cell lysates into nuclear and cytoplasmic fractions, fucTA mRNA levels were measured in each fraction of swm knockdown and control cells. The amount of fucTA mRNA in the nucleus increased by 9.6-fold in swm knockdown cells compared to control gfp knockdown cells ( Figure 6F). The nuclear export of actin (act5c) mRNA was also examined in swm knockdown cells. The amount of actin mRNA in the nucleus increased by 1.7-fold in swm knockdown cells compared to control cells. However, the fold change in nuclear accumulation of actin mRNA was significantly lower than that of fucTA mRNA in swm knockdown cells ( Figure 6F). These results suggest that relatively more Swm is required for nuclear export of fucTA mRNA than for nuclear export of actin mRNA.
Since nuclear export defects have been reported to result in an abnormally longer polyA length of mRNAs [29], the polyA length of fucTA mRNA was examined in swm knockdown cells. The polyA length of fucTA mRNA was clearly longer in swm knockdown cells than in control cells ( Figure 6G). Collectively, these results indicate that Swm positively regulates a1,3-fucosylation by promoting nuclear export of fucTA mRNA in neural cells.

Discussion
The current literature on glycosylation regulatory mechanisms is remarkably limited, and our current study is in fact the first report of a large-scale screen for glycosylation genes using a multicellular organism. We identified 109 genes in our screen, including 95 genes that have not been previously associated with glycosylation. In silico analyses using the available Drosophila databases suggested that these gene products participate in a variety of cellular processes (Figure 3), among which RNA regulation is, for the first time, shown to play an important role in the context of glycosylation. Moreover, the candidate genes belonging to the RNA regulation group were enriched by rescreening for a1,3-fucosylation-specific genes.

Biological significance of a1,3-fucosylation
In the current study, the fucTA mutant showed an imbalanced VNC. The VNC longitudinal length-to-width ratio was significantly shorter in fucTA mutants compared to wild-type larvae. To our knowledge, this is the first report of such shortened VNC in the loss-of-function mutant, although elongated VNC has already been reported in mutants for integrin [30], collagen IV [31], DHR3 [32], worniu [33] and C1GalTA [34]. Interestingly, C1GalTA encodes a core 1 galactosyltransferase that adds galactose to O-GalNAc attached to Ser or Thr protein residues. Taken together with our results, glycosylation patterns might affect the VNC shape. Although how a1,3-fucose affects VNC formation remains to be revealed, we think that two possible defects may result in a shortened VNC in fucTA mutants. (1) The aberrantly elongated VNC in the C1GalTA mutant is proposed to result from deglycosylation of laminin, a component of the extracellular matrix. In addition, overexpression of FucTA has recently been reported to enhance cell migration in the embryonic nervous system [35]. By analogy with the case of C1GalTA, cell migration in the larval VNC may be impaired in fucTA mutants because of aberrant interactions between migrating cells and extracellular matrix. (2) Another possibility is that neural activity may be impaired in fucTA mutants because neural activity has been reported to be necessary for VNC condensation in embryos [31]. This possible defect is consistent with the finding that one of the major a1,3-fucosylated proteins is a b-subunit of Na + /K + ATPase essential for the maintenance of neural cell membrane potentials [36]. It will be interesting to examine whether neural activity is controlled by glycosylation in the future.

Insight into the genes identified in this screen
Three types of genes were identified in this screen: (1) genes that specifically regulate glycosylation such as glycosyltransferases, (2) genes with multiple functions including regulation of expression, activity and/or localization of glycosyltransferases, and (3) genes with primary functions other than glycosylation that secondarily affect glycosylation. For example, knockdown of cytoskeletal regulation and mitochondrial function genes may secondarily affect glycosylation as a result of damage to cell structures and viability.
For several reasons, swm might be the second type of gene. Although swm has been previously reported to show involvement in the cell cycle [14], no glycosylation defects were observed with the knockdown of other known cell cycle regulators such as cdc2 (CG5363) or cdk4 (CG5072) ( Table S2 and S3). In addition, no cell cycle defects could be detected in the fucTA mutant. These data suggest that cell cycle and glycosylation are regulated independently. Therefore, swm would play at least two independent roles in cell cycle and glycosylation.
Traffic genes identified in this screen may be another example of the second type of genes. The conserved oligomeric Golgi (COG) complex has been reported to participate in the Golgi localization of glycosyltransferases and membrane traffic [37], but no involvement of the COG complex in glycosylation has been reported in Drosophila. Our screen identified Cog3 as a gene whose knockdown impaired a1,3-fucosylation. Thus, localization of FucTA protein was examined in cog3 knockdown photoreceptor cells. Expressed FucTA was hardly localized to the Golgi complex in knockdown photoreceptor cells (data not shown). Moreover, Golgi localization of FucTA was also preliminarily found compromised by knockdown of vps2 and vps20, other traffic genes (data not shown). These data raise the possibility that some traffic genes may be involved in glycosylation by localizing glycosyltransferases to appropriate Golgi and/or endoplasmic reticulum sites.

Preferential requirement of Swm in glycoforms
Our detailed analysis of glycan structures using mass spectrometry revealed that one (M3F2Gn2) of the two forms bearing an a1,3-fucose (M3F2Gn3 and M3F2Gn2) is more sensitive to the decrease of swm and Pabp2 than the other (M3F2Gn3). Two reasons may account for this difference, which are not mutually exclusive. (1) FucTA productivity differs between these two glycoforms. The structural difference between the glycoforms consists of a GlcNAc moiety attached to the nonreducing end of Nglycans. Given that this GlcNAc moiety has been shown to increase the reactivity of FucTA to the core of N-glycans [6], we assume that M3FGn3 is preferentially modified by FucTA to form M3F2Gn3. Therefore, under the condition of reduction, not complete loss, of FucTA by swm and Pabp2 knockdown, M3F2Gn2 might be preferentially lost. (2) The other possibility is that Swm or Pabp2 regulate Mgat1 or Fused lobes (Fdl) [38], the enzymes that add and remove the GlcNAc moiety, respectively, from the nonreducing end. In fact, the amount of mRNA encoding Mgat1, but not fdl, was slightly but significantly increased in swm knockdown cells ( Figure S5). This finding is consistent with the selective decrease of M3F2Gn2 induced by swm knockdown. However, it remains undetermined whether the preferential decrease of glycoforms is caused by an increase in Mgat1, the preferential productivity of FucTA, or both.

Regulation of fucTA mRNA by Swm
Expression of mRNA is generally regulated at transcription, splicing, 59-or 39-end processing, and nuclear export, and by the stability of the mRNA. The stability of fucTA mRNA was not affected by swm knockdown (Figure 6E). Given that 59-end processing is essential for mRNA stability, the 59-end processing of fucTA mRNA is not regulated by Swm. In addition, splicing of fucTA mRNA does not appear to depend on Swm activity since the misspliced form of fucTA mRNA in the swm knockdown cells could not be detected when we amplified fucTA mRNA using PCR primers that targeted different exons (data not shown). Furthermore, both transcription and nuclear export of mRNAs are regulated by the THO/TREX complex [39] and Sus1 [40]; thus, Swm may be a factor participating in both processes, with its depletion resulting in reduced expression and defective nuclear export of fucTA mRNA. We assume that abnormal extension of polyA in swm knockdown cells would be a secondary effect of nuclear export defects, as observed in mutants of genes required for nuclear export [29]. Moreover, Pabp2, Hel25E/UAP56 [41], and Nup358 [42], which participate in the nuclear export of mRNA, were also identified in our screen (Table S4), suggesting that nuclear export might be essential for glycosylation regulation.
Nuclear accumulation of polyA(+) RNA in swm knockdown S2 cells was confirmed in our in situ hybridization experiment, as previously reported [43]. However, in neural BG2-c6 cells, weak polyA(+) RNA signals in nuclei were comparable between control gfp knockdown and swm knockdown cells (data not shown), suggesting that only a small portion of mRNA might be exported by Swm. In addition, the ratio of nuclear to cytoplasmic fucTA mRNA was relatively higher than that of actin mRNA in swm knockdown BG2-c6 cells ( Figure 6F). Collectively, Swm may have preferred mRNAs for nuclear export. Moreover, the export efficiency and/or target RNA preference of Swm may differ between S2 and BG2-c6 cells.

Mammalian homologs of genes identified in this screen
Most of the genes identified in our present study have mammalian homologs (Table S5). Among these, Cog3 has been reported to be involved in glycosylation in organisms ranging from yeast to humans [37]. Interestingly, mutations of some COG components are causative for human diseases including congenital disorders of glycosylation (CDG) [44]. Further investigation of the genes identified here will likely provide additional insights into novel glycosylation regulatory mechanisms conserved in organisms ranging from Drosophila to humans and, possibly, into diseases involving glycosylation pathways.
The swm and Pabp2 mutant mosaic clones were generated using the FLP-FRT system [27] To investigate the localization of FucTA in RNAi-mutant flies, inducible Myc-tagged FucTA-expressing flies (UAS-fucTA-Myc) were generated. fucTA cDNA was a gift from S. Nishihara. A DNA fragment encoding the 3Myc sequence was ligated to the 39end of the coding region of fucTA and inserted into the pUAST vector. The vector was injected into w 1118 embryos for the production of transgenic flies.
To generate fucTA-Myc knock-in flies (fucTA-Myc KI ), the procedure was adopted from the ends-out knock in method [45]. Details are shown in Figure S4.

Primers
The primer sets used for the generation of transgenic flies are listed in Table S6. The primer sets used for analysis in cultured cells are listed in Table S7.

Quantification of ventral nerve cord shape
The central nervous systems were manually isolated from CS and fucTA f03774 third instar larvae and fixed in 4% paraformaldehyde/PBS at 4uC overnight. After several washes with PBS, brain hemispheres were manually removed and placed flatly onto slide glasses. The images were captured using an AxioCamequipped Zeiss AxioImager M1, and the length and width of VNCs were measured using AxioVision Rel. 4.7 software.

First screen
The first screen was carried out using a collection of transgenic fly lines carrying inducible dsRNA constructs. To drive expression of dsRNA for each gene in the eye, males from each RNAi fly line were crossed with GMR-Gal4 virgin females. Mating to inducible RNAi males and egg laying were carried out at 25uC for two days, and the resulting embryos were then incubated at 28uC. After 2 weeks, the hatched knockdown flies were collected and cultured for an additional 14 days at 28uC to enhance the effects of dsRNAs. More than 100 heads per line were isolated and homogenized in lysis buffer consisting of 10 mM sodium phosphate (pH 7.0), 100 mM NaCl, 0.5% Triton X-100, and complete protease inhibitor (Roche). The lysate was incubated with anti-Chp (24B10) covalently coupled with protein G-Sepharose (GE Healthcare) for 2 h. The Chp-bound Sepharose was subsequently washed with lysis buffer, and Chp was eluted with 50 mM triethylamine-acetic acid (pH 3.5).
Purified Chp was spotted onto Hybond ECL nitrocellulose membranes (GE Healthcare) using a Bio Dot slot format (Bio-Rad). After the membrane was blocked with 2% bovine serum albumin, immuno-or lectin blot analysis was performed with anti-Chp (N8A), anti-HRP, or HRP-conjugated ConA, LCA, DSA, and WGA. HRP-conjugated anti-mouse or HRP-conjugated antirabbit IgG was used as the secondary antibody. Detection was performed using Supersignal West Pico Chemiluminescent Substrate (Thermo), and quantification performed using a LAS-1000 Luminescent Image Analysis System (FujiFilm). An example of the dot blot is shown in Figure S2. Lectin affinity values per Chp values were converted to log values for normalization. On the basis of standard deviations from the values of GMR.GFP RNAi (as a control), genes with a z-score .3 were considered primary candidates. In the primary screen, we performed a single assay round. For the primary candidates, the lectin analysis was repeated (n = 3). Two-sided Student's t-tests were applied to the reproducibility test data between lectin values of knockdown and control flies. Genes with p,0.005 were considered to be reproducible findings.

Validation
Collection of the second transgenic fly stock lines was designed in such a way that there was no overlap between the primary construct in the original library and the new dsRNA sequence, and that the new dsRNA sequence had a low off-target probability score calculated by dsCheck [22]. Using the second lines, lectin analysis was independently repeated (n = 2), and the values between the knockdown and control (GMR.GFP RNAi ) were tested by two-sided Student's t-tests. Genes with p,0.05 were considered glycosylation-related genes (Rank 1). When no change in lectin values was observed solely using GMR-Gal4, the examination was repeated using GMR-Gal4; UAS-dicer2 (VDRC). RNAi flies from VDRC whose target sequence had no overlap with the primary construct were also used as second lines.
For candidate genes that were not evaluated with the second lines, first the number of 19mers with a perfect match to each Drosophila transcript was calculated using dsCheck. The highest number of 19mers with a perfect match to the off-target gene was defined as the ''off-target probability score''. The genes with a score below 3 were tested to determine whether knockdown of their suspected off-target genes led to glycosylation defects. When the knockdown of the suspected off-target genes did not lead to any evidence of glycosylation defects, the corresponding genes were added to the list of genes required for glycosylation (Rank 2).

GO analysis
Our final glycosylation hit list was annotated to include information on protein domain, molecular function, and protein family by cross-referencing InterPro (version 20.0 and 21.0) [47], Flybase, and Panther [48]. On the basis of the resulting annotation, genes were manually classified into 9 categories. These annotations are shown in Table S4.
Mass spectrometry (MS) analysis of glycopeptides and glycan structure MS analysis of glycopeptides was performed as previously described [16]. Purified Chp was digested with modified trypsin (Promega). Glycopeptides and peptide mixtures were separated by reversed-phase high-performance liquid chromatography (RP-HPLC) using an Inertsil C18 column (GL Science). The matrices for peptide analysis by matrix-assisted laser desorption/ionization time-of-flight MS (MALDI-TOF-MS) spectra were acquired using a Voyager mass spectrometer (Applied Biosystems). MS/MS measurements were performed using an Ultraflex TOF/TOF mass spectrometer with the LIFT-MS/MS facility (Bruker Daltonics GmbH).

Immunostaining
To investigate the amount of a1,3-fucosylation in vivo, frozen sections of adult eye demonstrating swm or gfp knockdown were prepared. Immunostaining was performed as previously described [49]. Fluorescent images were acquired with a laser scanning confocal microscope, LSM700 (Zeiss).
To construct an expression plasmid for C-terminal Flag-tagged Swm (Swm-Flag), cDNA containing the swm coding region was obtained from Canton-S total RNA by RT-PCR. A DNA fragment encoding the 3Flag sequence was ligated to the 39end of the swm coding region and inserted into the expression vector pRmHa.
For the transient expression of Swm-Flag, 1.5 mg plasmid was added to each well of a 24-well plate, in which 5610 5 BG2-c6 cells had been seeded. Plasmid expression was induced by incubating cells in the presence of 0.1-0.5 mM CuSO 4 for periods from 3 h to overnight.
For knockdown experiments, dsRNA for swm or gfp was produced using in vitro transcription T7 kit (Takara). A 3-mg aliquot of dsRNA was added to each well of a 24-well plate in which 5610 5 cells had been seeded. After 5 days, cells were harvested for the following assays.

Immunoblot
For protein analysis, equal amounts of protein extracted from fly eyes were subjected to SDS-PAGE. After transfer to PVDF membranes (Millipore), membranes were blocked in PBS containing 0.05% Tween 20 and 5% skim milk and incubated overnight with the primary antibody, followed by the secondary antibody. A monoclonal anti-a-tubulin antibody was used for normalization. Detection was performed using the Supersignal West Pico Chemiluminescent Substrate.

RNP-IP assay
Immunoprecipitation of protein-RNA complexes was performed according to a protocol by Niranjanakumari et al. [50]. BG2-c6 cells transfected with swm-Flag were harvested and crosslinked in PBS containing 0.2% PFA. Following sonication in RIPA buffer (50 mM Tris-HCl pH 7.5, 1% NP-40, 0.05% SDS, 1 mM EDTA, 150 mM NaCl) with protease inhibitor (Nakarai), each supernatant was mixed with a monoclonal anti-Flag M2 antibody or normal mouse IgG and then incubated with protein-G beads (GE Healthcare) at 4uC for 6 h. Sepharose beads were washed 5 times with RIPA buffer and incubated in 50 mM Tris (pH 7.0), 5 mM EDTA, 10 mM DTT, and 1% SDS at 70uC for 45 min to reverse cross-linking. RNA was extracted from the immunoprecipitates using Sepasol reagent (Nakarai) followed by DNase I treatment. The extracted RNA was used as a template to synthesize cDNA using Superscript III (Invitrogen) according to the manufacturer's protocol.

Real-time PCR
Total RNA was extracted using the RNeasy mini kit (Qiagen) or Sepasol reagent (Nakarai). Complementary DNAs were synthesized using Superscript III according to the manufacturer's protocol. Real-time PCR was carried out using a 7900HT Fast Real Time PCR system (Applied Biosystems) with Power Syber Green. Samples were normalized with Drosophila rpl32.

RNA degradation analysis
BG2-c6 cells transfected with gfp or swm dsRNA were grown for 5 days, and then actinomycin D (Nakarai) was added to a final concentration of 5 mg/mL to arrest de novo RNA synthesis. At 30, 60, and 90 min after actinomycin D treatment, the cells were harvested, and fucTA and rpl32 (control) mRNA was quantified by real-time PCR as described above.

PAT assay
PolyA measurement of fucTA mRNA was carried out as described [51]. PCR fragments from the PAT assay were separated on 2% agarose gels and visualized by Southern blot using DIG DNA Labeling and Detection kit (Roche).

Human homologs
The InParanoid database (version 7.0) was searched for mammalian homologs [53].  Figure S4 Knock-in of the fucTA gene by ends-out recombination. (A) The donor DNA was generated by FLP and I-SceI action from the X chromosome (top). Homologous recombination was used to insert the 3Myc sequence into the endogenous fucTA gene and the white gene into the region between fucTA and Msh6 genes (middle). The expected structure (bottom) was verified by PCR with two sets of primers (colored thick arrows). (B) Expected bands were amplified with two sets of primers (red, fucTA locus; blue, Msh6 locus) using chromosomal DNA from a fucTA knock-in fly (fucTA-Myc KI ) but not in white flies (w). (C) Myc-tagged FucTA protein (arrow) was detected by immunoblot using anti-Myc antibody in the adult eye extracts of fucTA knock-in flies (fucTA-Myc KI ) but not in white flies (w). The amount of Myc-tagged FucTA protein was reduced in swm knockdown eyes (swm KD) compared to the control eyes (gfp KD).