Sugar-free frosting, a homolog of SAD kinase, drives neural-specific glycan expression in the Drosophila embryo

Essential aspects of neural development and function rely on the regulated expression of specific glycan structures. The involvement of glycans in neural cell function begins with neurectoderm differentiation, in which the elaboration of specific glycan structures on the Notch glycoprotein facilitates the differential ligand activation that drives cell fate selection (Okajima and Irvine, 2002). Subsequently, glycans expressed in developing and mature neural tissues impinge on other important cell signaling, migration, adhesion and synaptic functions (Matani et al., 2007). Several underlying molecular mechanisms for glycan function are well characterized, including the influence of N-linked polysialic acid on neural cell adhesion molecule (NCAM) interactions, the engagement of axonal glycosphingolipid glycans by a carbohydrate-binding protein on oligodendrocytes, and the activity-dependent deposition of glycosaminoglycans into perineuronal nets (Acheson et al., 1991; Matthews et al., 2002; Vyas et al., 2002). These instances, and others, reveal the prevalence of temporally and/or spatially regulated glycan expression. This is further supported by the existence of many anti-glycan antibodies (HNK-1, CAT, Jones, A2B5, FORSE-1, anti-HRP) that delineate subpopulations of neuronal or glial cells at various stages of development (Allendoerfer et al., 1999; Constantine-Paton et al., 1986; Jan and Jan, 1982; Kruse et al., 1984; Matthews et al., 2002; Sanes et al., 1986; Schnitzer and Schachner, 1982; Snow et al., 1987). Furthermore, aberrant glycan expression in humans is associated with various mild to severe pathologies (Dennis et al., 2009; Freeze, 2002; Lee et al., 2007; Michele and Campbell, 2003; Zak et al., 2002). Despite the overwhelming evidence for regulated glycan expression and for the significant consequences of disregulated glycan expression, the mechanisms that govern tissue-specific glycosylation are unknown, particularly during embryonic development.

We have previously reported that a mutation in an ectodermally expressed Toll-like receptor, Tollo, alters N-linked glycosylation in the developing Drosophila embryonic nervous system (Seppo et al., 2003). This defect was detected as loss of staining with antibodies that recognize a family of structurally related N-glycans, known as HRP epitopes, which are normally expressed in a restricted set of embryonic tissues (Jan and Jan, 1982; Snow et al., 1987). The Tollo mutation specifically abolishes HRP-epitope expression in neural tissue although Tollo is not expressed in neural cells that carry HRP epitopes. Rather, it is expressed and functions within non-neural ectodermal cells that surround differentiating neurons, establishing the basis for a transcellular paracrine signaling pathway that drives neuron-specific glycosylation (Seppo et al., 2003). Such transcellular signaling might drive cell-specific glycan expression through altered transcription of glycan biosynthetic genes (glycosyltransferases, glycan processing enzymes, etc.) or through mechanisms that modify trafficking through specific glycoprotein processing pathways. The relative contribution of altered transcription and altered cellular organization to tissue-specific glycan expression is completely unresolved in any biological context. To address this lack of knowledge and to identify the unknown components of the Tollo transcellular signaling mechanism, we undertook a random mutagenesis screen for genes that specifically affect HRP-epitope expression in the Drosophila embryo. Here, we describe an informative mutation recovered from this screen called sugar-free frosting (sff), the phenotype of which reveals the importance of regulated Golgi dynamics for neuron-specific glycan expression. Our sff mutation, which is the first described disruption of a Drosophila homolog of SAD kinase, interacts genetically with Tollo and modulates glycan complexity in neurons that are receptive to the Tollo transcellular signaling pathway (Crump et al., 2001; Inoue et al., 2006; Kishi et al., 2005). Our results lead us to propose a new paradigm in which tissue-specific glycan expression is sculpted by the relative activities of multiple protein kinases, each acting to facilitate flux through specific Golgi processing pathways.

Probes for immunohistochemistry and immunofluorescence used were: rabbit anti-HRP (1:2000 for embryos, 1:1000 for larvae), HRP-Concanavalin A (ConA; 1:100), HRP-conjugated goat anti-rabbit (1:1000) and goat anti-mouse (1:1000) antibodies from Jackson Laboratories; monoclonal antibodies 1D4 (anti-Fas2; 1:3), nC82 (anti-Brp; 1:100) and 22C10 (1:5) from the Developmental Studies Hybridoma Bank (DHSB, University of Iowa, IA, USA); biotin-conjugated PNA (peanut lectin; 5 μg/ml) obtained from Vector Laboratories; anti-GM130 (1:1000) monoclonal antibody obtained from Abcam; Alexa-conjugated secondary antibodies (Alexa 488, 568 and 633; 1:500), rabbit anti-GFP (cross-reactive with YFP; 1:5000) and PROLONG anti-fade obtained from Molecular Probes; TRITC-Phalloidin (1:100) obtained from Invitrogen. PNGaseA was from Calbiochem; trypsin and chymotrypsin were from Sigma.

Males of genotype w¹¹¹⁸; +; + were treated with 25 mM ethyl methanesulfonate and mated en masse to females of the genotype w¹¹¹⁸; Kr/CyO; D/TM6b. Individual male progeny heterozygous for the second and third chromosome markers were backcrossed to w¹¹¹⁸; Kr/CyO; D/TM6b females to create lines that carried mutagenized second or third chromosomes. Progeny from each individual line were sib-mated to give embryo collections that were stained with anti-HRP antibody. Lines showing altered HRP-epitope expression were subsequently stained with Concanavalin A. Lines that were deficient in anti-HRP staining but normal for Concanavalin A staining were maintained.

All mapping, balancer, deletion and marked chromosome stocks were obtained from the Bloomington Drosophila Stock Center at Indiana University. Two Tollo mutant lines were used. One is as previously described (Seppo et al., 2003), and the other (tollo^c5) was obtained from B. Charroux and J. Royet (IBDML, Marseille, France). For confocal characterization of the medial/trans Golgi, the YFP reporter carried by the Golgi-YFP line obtained from Bloomington was recombined onto the sff^B22 chromosome (third chromosome mutation). Therefore, for Golgi characterization, wild type refers to the parental w¹¹¹⁸; +; Golgi-YFP line. Otherwise, wild type refers to OreR.

Drosophila lines expressing Sff (CG6114) under Gal4/UAS control were generated by P-element-mediated insertional transgenesis. Vector containing CG6114 cDNA missing 209 bp of coding sequence (including the start codon) was obtained from Berkeley Drosophila Genome Project (clone GH13047 in pOT2). The insert was excised with EcoRI/XbaI and the missing coding sequence plus 97 bp of the 5′UTR was recovered by PCR from wild-type cDNA. Taking advantage of an endogenous EcoRI site, both sequences were ligated and cloned into the pBl(KS+) vector (Stratagene) and the integrity of the PCR-amplified region was verified by sequencing. The resulting full-length insert was subcloned into pUAST. Transformant lines were generated by co-injection of pUAST-CG6114 and wcΔ2-3 into w¹¹¹⁸ embryos by standard procedures (Brand et al., 1994). A second chromosome insertion was homozygosed into sff^B22 and crossed to ELAV-Gal4; sff^B22 homozygotes. Progeny embryos were stained with anti-HRP antibody to assess rescue of the sff phenotype.

Embryos from overnight collections were dechorionated, fixed, devitellinized and stained with antibodies using blocking conditions and wash buffers as previously described (Patel, 1994; Seppo et al., 2003). For histochemical probes (antibodies and ConA), Nomarski (differential interference contrast) and light micrographs were obtained on a Zeiss Axioskop microscope fitted with a Retiga 2000R CCD camera (Q Imaging, Surrey, Canada). Neuromuscular junction (NMJ) morphology was assessed in wandering third instar larvae raised at 18°C. Larval dissections were performed as previously described (Kaufmann et al., 2002). The NMJ at muscles 6 and 7 (abdominal segments 3 and 4) were imaged by laser scanning confocal microscopy (LSC, Olympus FV1000) with a 40× (N.A. 1.30) oil objective. Stacks of optical sections were collected in the z-dimension; step size, based on the calculated optimum, was 0.50 μm. Maximum intensity projections were assembled using NIH Image J. Golgi imaging was performed on an Olympus FV1000 LSC microscope with a 60× (N.A. 1.42) oil objective. Images were acquired in the z-dimension with step size of 0.45 μm. Laser intensity, slice thickness and all other acquisition parameters were identical for imaging wild-type and mutant embryos. Image analysis was accomplished offline with the Slidebook software package (Intelligent Imaging Innovations, Denver, CO, USA) on non-compressed z-stacks. Confocal stacks were adjusted for background and channel overlap using the software supplier's recommended algorithms. Total and colocalized fluorescence intensity for each marker was quantified within selected regions of interest.

To prepare the anti-sense probe for CG6114 (sff), wild-type genomic sequence corresponding to 2581 bp of the CG6114 3′ UTR was amplified by PCR and cloned into PCR 2.1 TOPO vector (Invitrogen). Clones were recovered bearing the insert in both sense and anti-sense orientations relative to the T7 promoter (Kopczynski et al., 1996). Digoxigenin-11-UTP-labeled RNA was prepared by in vitro transcription using T7 polymerase and DIG RNA Labeling Kit (Roche). In situ hybridization of anti-sense probes (see Table S2 in the supplementary material) to wild-type embryos from overnight collections (25°C) was detected by alkaline phosphatase with BCIP/NBT (5-bromo-4-chloro-3′-indolylphosphate/nitro-blue tetrazolium).

Total N-glycans were prepared from overnight embryo collections (25°C) by PNGaseA digestion of tryptic/chymotryptic peptides as previously described (Aoki et al., 2007). Following permethylation, total glycan profiles were acquired by mass spectrometry (MS) using nanospray ionization interfaced to a linear ion trap instrument (NSI-MSⁿ on an LTQ Ion Trap, Thermo-Fisher). The Total Ion Mapping (TIM) functionality of the instrument control and data acquisition software (Xcalibur, v2.0) was used to acquire full MS and MS/MS spectra (∼700 spectra per sample), which were manually interpreted to assign glycan structures and to quantify relative glycan prevalence. Ambiguities were resolved by acquisition of spectra at MS³-MS⁴ as needed.

Proteins from adult head lysates were reduced, alkylated, digested with sequencing grade trypsin (Promega) and subjected to proteomic analysis by LC-MS/MS (Lim et al., 2008). Two-dimensional RP/RP-HPLC (reverse phase/reverse phase high pressure liquid chromatography) was performed to separate the total peptide mixture into five fractions (F1-5) based on hydrophobicity. Fractionated peptides were loaded off-line onto a nanospray tapered capillary column emitter and LC-MS/MS analysis was performed on an LTQ Orbitrap mass spectrometer (Discovery, Thermo-Fisher) with a Surveyor MS pump plus (Thermo Scientific). As the five RP-fractions each contain peptide pools that possess different average hydrophobicities (increasing from F1-5), separate gradient elution conditions were employed for LC-MS/MS analysis of each fraction (see Fig. S6 in the supplementary material) in order to maximize peptide resolution. MS/MS spectra (36% normalized collision energy) were triggered upon detection of any mass-to-charge ratio (m/z) signal that corresponded to any entry into a parent mass list of theoretical CG6114 tryptic peptides, allowing for up to 2 missed cleavages, oxidized methionine, alkylated cysteine and charge states to +3. MS/MS data was searched against the Fly database (Drosophila melanogaster, 8-31-09) from Swiss Prot to which CG6114 was added, and against the reverse of this database to assess false discovery. SEQUEST parameters were set with a mass tolerance of 30 ppm, fragment ion cutoff of 0.1% and allowance for 2 missed cleavages.

Males that had been raised for 8 days at 25°C or 18°C were placed individually into empty plastic culture vials containing no media. Following dark adaptation for one hour (under red safelights) at 18°C, the test subjects were gently tapped to the bottom of the vial. The elapsed time for the subject to climb to a mark 3 cm up the side of the vial was recorded. For subjects that did not reach the pre-determined height, the trial was terminated after 2 minutes and the elapsed time was scored as 120 seconds.

Overnight embryo collections were dechorionated and total RNA was isolated using RNAeasy (Qiagen). cDNA was prepared from 5 μg of total RNA using the Superscript III First-Strand synthesis system (Invitrogen). Quantitative RT-PCR was performed as previously described, using SYBR green for detection (Nairn et al., 2007).

Antibodies generated against the plant glycoprotein horseradish peroxidase (HRP) cross-react with specific asparagine-linked oligosaccharide structures expressed in the nervous system of many arthropods (Jan and Jan, 1982; Snow et al., 1987). These N-glycans, collectively known as HRP epitopes, carry α3- and α6-linked fucose (Fuc) on the innermost n-acetylglucosamine (GlcNAc) of the chitobiose core (GlcNAcβ4GlcNAc, Fig. 1A). Although the recognized structures are difucosylated, α3-linked Fuc is the dominant antigenic determinant (Kurosaka et al., 1991). HRP epitopes are expressed throughout the entire Drosophila life cycle (Aoki et al., 2007; Fabini et al., 2001). In the embryo, anti-HRP staining was evident in the ventral nerve cord and peripheral nervous system, as well as in a limited set of non-neural tissues, including the garland gland and the hindgut (Fig. 1B-D).

The restricted distribution of HRP epitopes provides a platform for identifying genes that govern tissue-specific glycan expression. Therefore, we undertook a random chemical mutagenesis screen to harvest mutations that specifically alter expression of HRP epitopes without generally affecting N-linked glycosylation. Embryos from the F2 generation derived from mutagenized male founders were stained with anti-HRP antibody and examined for loss of HRP epitopes. Mutant lines that exhibited altered HRP-epitope expression were stained with Concanavalin A (ConA), which reports the presence of high-mannose oligosaccharides, to assess the integrity of the core glycosylation machinery.

One mutation recovered from the screen, initially designated B22, was a semi-viable third chromosome line with substantial loss of all neural HRP-epitope expression but with normal ConA staining (Fig. 1 and see Fig. S1 and Table S1 in the supplementary material). Residual staining was apparent in late stages within the axon scaffold of the ventral nerve cord. This neural staining was restricted to the dorsal aspect of the nerve cord and gave the appearance of a frosting-like layer of epitope, prompting us to name the mutation `sugar-free frosting' (sff). At reduced temperature (21°C), the onset of residual epitope expression was delayed until slightly later stages than that observed at 25°C (Fig. 1H-J). In non-neural tissues, hindgut staining was severely reduced although the garland gland retained HRP-epitope expression in sff^B22 embryos.

Recombination mapping placed sff distal to scarlet (st) on the left arm of the third chromosome. Deletion mapping in the interval between st and the left end of the chromosome identified a single deletion (Δbrm¹¹) that failed to complement the sff mutation. Embryos of the genotype sff^B22/Δbrm¹¹ displayed a more severe HRP phenotype than did sff homozygotes, and Δbrm¹¹ deletion homozygous embryos lacked all HRP epitopes, indicating that the sff^B22 allele is hypomorphic with regard to the induction of HRP-epitope expression (Fig. 2A-F). Loss of hindgut and garland gland HRP epitopes in the deletion homozygote implies a role for sff in the subset of non-neural tissues that normally express these glycans. We characterized the deletion breakpoints of the non-complementing Δbrm¹¹ deletion and of an overlapping, but complementary, deletion Δth¹⁰² (Mathew et al., 2009). Comparison of these breakpoints identified CG6114 as the only gene deleted by Δbrm¹¹ but not by Δth¹⁰² (see Fig. S2A in the supplementary material). Genomic sequencing of the CG6114 locus in the sff^B22 mutation identified a single nucleotide change (G to A) in a consensus AP-1 binding site 261 bp upstream of the predicted transcript start (see Fig. S2B in the supplementary material). The detected nucleotide change prevented the sequence from being recognized as an AP-1 binding site above threshold (85% confidence) by the TFSearch algorithm (v1.3) utilizing the TRANSFAC database (Heinemeyer et al., 1998). The CG6114 sequence is highly homologous to a serine-threonine protein kinase previously identified as Synapses of Amphids Defective (SAD-1 Kinase) in Caenorhabditis elegans and mammals (SAD-1/2) (see Fig. S3 in the supplementary material). All canonical amino acid residues essential for serine-threonine kinase catalytic activity are conserved in the CG6114 coding sequence (Hanks and Hunter, 1995). To verify that the reduction in HRP-epitope expression is due to decreased CG6114 function, we generated a transgenic line expressing CG6114 cDNA (UAS-CG6114) driven by the pan-neural driver, elav-Gal4. The UAS-CG6114 transgene was sufficient to rescue neuronal HRP-epitope expression in sff^B22 embryos (Fig. 2G,H).

By in situ hybridization, mRNA for sff was detected in the embryonic ventral nerve cord and peripheral sensory clusters in the ventral and lateral ectoderm (Fig. 3A-E) (Kopczynski et al., 1996). If present in the hindgut or garland gland, sff mRNA was below the current detection threshold for our in situ analysis. Therefore, despite complete loss of HRP epitopes in Δbrm¹¹ homozygotes (see Fig. 2E,F), additional tissue-specific regulatory mechanisms are likely to contribute to HRP-epitope expression in non-neural tissues. Consistent with our genetic characterization of the sff^B22 allele as a hypomorph, mRNA for sff was reduced, but not eliminated, in the mutant nervous system (Fig. 3B,E, quantified in Fig. 4). The cells expressing sff mRNA were neuronal as hybridization signal was detected within cells that were also positive for the neuron-specific mAb 22C10 (see Fig. S4 in the supplementary material). Expression of sff mRNA was first visible along the midline of the embryo at stage 11, prior to the initiation of synapse formation in the embryo and roughly coincident with the initial appearance of HRP epitopes and transcripts for FucTA, the α3-fucosyltransferase that synthesizes HRP epitopes (see Fig. S5 in the supplementary material) (Rendic et al., 2006; Rendic et al., 2010). As neurons continued to differentiate and extend axons, sff mRNA levels increased across the maturing nerve cord. In peripheral sensory clusters, mRNA was detected as early as stage 13, prior to significant accumulation of HRP epitopes and before synaptic contact had been made between sensory afferents and their targets in the ventral nerve cord, which generally begins during stage 16 (Martin et al., 2008; Merritt and Whitington, 1995). Consistent with reduced sff mRNA expression in sff^B22, proteomic analysis by LC-MS/MS demonstrated reduced Sff protein levels in sff^B22 adult heads, which also exhibited reduced HRP-epitope expression (see Table S3 and Fig. S6 in the supplementary material) (Lim et al., 2008).

HRP epitope loss in sff^B22 suggested that Sff might act in the transcellular signaling pathway defined by Tollo (Seppo et al., 2003). As heterozygous embryos, sff^B22 or Tollo mutations (over a wild-type chromosome) exhibited normal HRP-epitope expression. As homozygous embryos, the sff phenotype was not identical to the Tollo phenotype. Hindgut staining was absent and garland gland staining was reduced in sff^B22 but not in the Tollo mutant, whereas nerve cord staining was completely absent in Tollo mutants but still faintly present in sff^B22. Transheterozygous embryos of the genotype sff^B22, Tollo⁺/sff⁺, Tollo^– possessed aspects of both phenotypes; faint residual nerve cord staining was seen along with rescue of hindgut HRP epitopes (Fig. 4A-D). This composite phenotype is consistent with an interaction in which reduction of either gene dose results in sensitization for both phenotypes. Thus, sff^B22 enhanced partial loss of Tollo, and vice versa, indicating functional interactions between the two genes. Significantly, Tollo mutant embryos exhibited reduced sff transcripts, both in our previously described Tollo mutant (Fig. 4E,F) and in another mutant allele (tollo^c5) generated by FLP-FRT recombination (B. Charroux and J. Royet, personal communication), which also exhibits reduced HRP-epitope expression (Fig. 4G and see Fig. S7 in the supplementary material) (Parks et al., 2004). Therefore, ectodermal Tollo signaling influences neural glycosylation, at least in part, by driving expression of Sff in differentiating neurons.

Homozygous sff^B22 adults were reluctant to climb up the side of their culture vials, a phenotype previously described for Tollo mutants (Seppo et al., 2003). In general, sff^B22 adults righted themselves when knocked to the bottom of an empty culture vial and exhibited reasonable locomotion once movement was initiated, but failed to move in a directed fashion away from gravity (see Movie 1 in the supplementary material). We quantified this negative geotaxis defect by measuring the time required for individual adults to climb to a pre-designated height (Fig. 5A). Wild-type adults quickly reached the target height; 81% succeeded within 15 seconds at 25°C. However, sff mutants tarried and often never reached the designated height before the trial was ended at 2 minutes; only 36% succeeded within 15 seconds at 25°C. As with anti-HRP antibody staining, the geotaxis phenotype was more pronounced at reduced temperature.

Because motility requires a functional NMJ, we investigated the status of the NMJ in third instar larvae. In sff mutants, the NMJ was considerably less complex (Fig. 5B-E), with a significant reduction in both bouton and primary branch number. NMJ morphology is also disrupted in the C. elegans SAD-1 mutant, with poor alignment of synaptic vesicles at the active zone complex. Despite deficient vesicle tethering, the presynaptic specializations that make up the active zone are well structured in the C. elegans mutant (Crump et al., 2001). To assess in a similar manner the integrity of the active zone scaffold in our Drosophila sff mutant, we examined the distribution of the Bruchpilot protein (Brp), which participates in the formation of T-bars (the electron-dense projections upon which synaptic vesicles are tethered) at the NMJ (Wagh et al., 2006). The distribution of Brp within individual boutons was unaltered in sff mutant NMJs compared with wild type (Fig. 5F,G). Therefore, reduced Sff in Drosophila, similar to reduced SAD-1 in C. elegans, did not affect the structural integrity of the presynaptic scaffold required for vesicle release.

N-linked glycan biosynthesis and processing in Drosophila proceeds through steps that are well conserved across the animal kingdom (Fig. 6A). A major difference between Drosophila and vertebrate glycan processing is the activity of an n-acetylhexosaminidase, known as Fused Lobes (Fdl), which removes the GlcNAc residue added by n-acetylglucosaminyltransferase-I (GlcNAcT1). The Fdl enzyme drives the predominance of paucimannose structures in the wild-type Drosophila glycan profile by converting NM3N2 glycans to M3N2, thereby eliminating the precursor pool for hybrid or complex structures (Leonard et al., 2006). Limited expression of branching and terminal glycosyltransferases, such as GlcNAcT2, GlcNAcT4, GalT, GalNAcT or SiaT, also restrict the capacity for generating complex structures (Haines and Irvine, 2005; Koles et al., 2004; Repnikova et al., 2010; Sarkar et al., 2006). HRP epitopes represent less than 1% of all N-linked glycans in the Drosophila embryo (Aoki et al., 2007). Therefore, anti-HRP antibody staining reports the status of a limited set of N-glycan processing pathways.

To comprehensively identify glycan expression changes induced by the sff mutation, total N-linked glycan profiles of sff and wild-type embryos were analyzed by mass spectrometry. High mannose and paucimannose glycans together accounted for almost 90% of the total profile in wild-type (89.5%) and sff mutant (85.5%) embryos, indicating that the dominant glycan processing pathways are largely unaffected by the sff mutation. (see Table S4, structures 1-13, in the supplementary material). However, the prevalence of the M5N2 structure, an essential precursor for paucimannose, hybrid and complex glycan production, was reduced in the sff mutant (Fig. 6B). The decrease in M5N2 was mirrored by an increase in M3N2, a paucimannose glycan production of which requires Fdl, indicating robust activity of the processing hexosaminidase in sff embryos. Hybrid and complex glycans together made up 9.6% of the total profile in wild-type and 11.1% in sff embryos (see Table S4, structures 14-25 and 31-42, in the supplementary material). This modest increase in the prevalence of processed glycans, equivalent to a shift of 1.5% of the total glycan profile in the sff mutant, was more dramatic for a subset of individual complex structures (Fig. 6C). For example, the N2M3N2 glycan, a biantennary non-fucosylated structure, was increased by 30% and the N3M3N2F glycan, a triantennary core-fucosylated structure, was increased by 385% in the sff mutant. As expected, glycans bearing α3-linked Fuc were significantly reduced in sff embryos (54% reduction overall), accounting for 0.5% less of the total glycan profile than in wild type (Fig. 6D). However, one HRP epitope, the complex biantennary N2M3N2F2 glycan, was more prevalent in sff than in wild-type embryos. This difucosylated structure, like the non-HRP complex structures that were also increased in sff mutants, requires the activity of GlcNAcT2 or GlcNAcT4 (Fig. 6A), indicating that the sff mutation increases the flux of glycan processing towards greater complexity. Our previous structural characterization of the glycans expressed in Tollo mutant embryos also demonstrated increased prevalence of complex non-fucosylated glycans in addition to reduced expression of HRP epitopes, consistent with both genes acting within an intersecting pathway (Aoki et al., 2007).

To determine if the shifted glycan profile in sff results from altered transcription of genes encoding glycosyltransferases or glycan processing enzymes, we performed qRT-PCR in order to quantify transcript levels for genes encoding Fdl, FucTA, GlcNAcT1, GlcNAcT2 and a GlcNAcT4 candidate (CG9384) relative to GAPDH1 (Fig. 6E). FucTA transcription was modestly reduced in sff^B22 (30%), although the reduction in message was disproportionate to the larger reduction in HRP-epitope expression (54%). Furthermore, the increase in Fdl transcripts and the decrease in GlcNAcT2 transcripts were inconsistent with the shift in the total glycan profile towards greater complexity and reduced HRP-epitope production. Similar transcript abundances were detected in Tollo mutant embryos (see Fig. S7 in the supplementary material). The inability of transcriptional changes to account for the nature of the shift in glycan expression indicates that modulation of enzyme production is not the primary mechanism by which sff alters N-glycosylation.

Protein glycosylation reflects Golgi dynamics. Therefore, we hypothesized that Sff participates in the establishment of or trafficking through Golgi compartments. The Drosophila Golgi apparatus is striking for its lack of morphological similarity to classic conceptualizations of the ribbon-like stacked cisternae evident in many vertebrate cell types. Instead, the Drosophila Golgi exists as distributed puncta, forming multiple organelles within a single cell (Kondylis and Rabouille, 2009; Stanley et al., 1997). As the major change in sff N-glycan expression is a shift towards greater complexity, we focused first on late Golgi compartments using markers described previously to delineate the medial/trans and trans-Golgi in Drosophila (LaJeunesse et al., 2004; Yano et al., 2005). We assessed the relative distributions of Golgi-YFP (a medial/trans marker, YFP fused to the transmembrane localization domain of human β4GalT), Arachis hypogaea lectin (PNA) binding (a trans marker, Galβ3GalNAc specific lectin) and a neuronal protein (Fasciclin 2, Fas2) that bears an HRP epitope (Desai et al., 1994; Lin et al., 1994).

In triple-stained preparations (Golgi-YFP, PNA, Fas2), Golgi puncta were distributed throughout the nerve cord (see Fig. S9 in the supplementary material). Colocalized fluorescence intensity for each marker was quantified across all of the objects imaged within regions of interest in the ventral nerve cord (Fig. 7A-I and see Fig. S10 in the supplementary material). Colocalization of Fas2 with the trans marker demonstrates a striking change in the association of the HRP-bearing glycoprotein with this compartment of the Golgi (compare Fig. 7A with 7B, quantified in 7C). Essentially, all of the quantified Fas2 fluorescence intensity colocalized with the trans marker in sff mutants. The differential colocalization of Fas2 in sff mutants indicates that the cisternal association of this glycoprotein has shifted towards a later compartment, away from the earlier localization expected for enzymes that generate paucimannose-type glycans (including simple HRP epitopes) and towards the localization expected for later-processing enzymes involved in complex branching (GlcNAcT2, GlcNAcT4). Consistent with increased polarization between late Golgi compartments, the sff mutation caused greater segregation of trans and medial/trans markers (compare Fig. 7D with 7E, quantified in 7F).

To examine the integrity of the early Golgi in sff mutants, we assessed the distribution of GM130, a cis-localized Golgin (Nakamura et al., 1997). Overlap between the trans marker and GM130 was low in wild-type and sff mutant embryos. However, the mutant displayed a statistically significant increase in colocalization, indicating that this Golgi matrix component had expanded into the cisternal territory from which it is normally restricted (Fig. 7G-I).

We have presented data demonstrating that a mutation in Drosophila Sff alters glycan expression in the embryonic nervous system. The neural-specific N-linked glycans known as HRP epitopes are reduced in sff mutant embryos and the expression of a subset of complex glycans is increased. Protein kinases have previously been linked to altered glycosylation, based primarily on pharmacological manipulations in cultured cells (Yu and Bieberich, 2001). For example, treatment of NRK cells with activators of protein kinase A alters vesicular stomatitis virus G protein glycosylation in infected cells (Muniz et al., 1996; Muniz et al., 1997). Furthermore, protein kinase C activity modulates sialylation of glycosphingolipids in NG108-15 and of glycoproteins in SH-SY5Y, as well as polysialylation of NCAM in neuro-2A and PC-12 cells (Bieberich et al., 1998; Breen and Georgopoulou, 2003; Gallagher et al., 2000). However, the functional consequences of these pharmacological effects have not been validated in vivo, nor have molecular or cellular mechanisms underlying altered glycosylation been identified.

The expression of Drosophila HRP epitopes is not only tissue-restricted, it is also protein-restricted in wild-type embryos; only a subset of neuronal proteins bears α3-fucosylated glycans (Desai et al., 1994). The influence of α3-fucosylation on the function of any of the individual proteins that carry this specific post-translational modification is currently unknown; the glycan might influence protein stability or protein interactions, or serve as a recognition marker independent of the protein on which it resides. Regardless of its function, the tissue-specificity of HRP-epitope expression in the Drosophila embryo provides a platform for assessing the regulatory mechanisms that control glycosylation in vivo. In general, the regulation of glycosylation, and particularly of glycan diversity on specific proteins, is multi-dimensional. The coordinated expression of appropriate glycosyltransferases and target proteins alone is insufficient for achieving specific glycosylation. Target and enzyme must also be brought together within a permissive Golgi domain, which must contain other supporting activities, such as appropriate nucleotide sugar transporters and trimming exoglycosidases.

The multidimensionality of regulated glycan expression means that Sff might influence protein glycosylation at any of several control points. The kinase might modulate the expression of glycan modifying enzymes, but our analysis of glycogene transcript levels in the sff mutant indicates that the expression of glycosyltransferases and glycan processing enzymes cannot account for the observed shifts in glycan prevalence. Differential lysosomal degradation of specific protein glycoforms could also generate a shift in glycan profiles, but altered catabolism cannot explain the increase in complex glycans observed in the sff mutant. Finally, glycosyltransferase phosphorylation has been described in vitro, but evidence for a functional consequence of this modification is currently lacking (Gallagher et al., 2001; Hathaway et al., 2003; Ma et al., 1999). The hypothesis most consistent with our results is that Sff modulates glycan processing by influencing the compartmentation of glycan processing within the Golgi apparatus.

Very little is known regarding the coordinated Golgi dynamics that produce the confluence of components necessary for specific glycan expression in embryos or in mature tissues (Gerardy-Schahn et al., 2001; Nilsson et al., 2009), but a specific role for protein phosphorylation in controlling Golgi function is not without precedent. The cyclin-dependent kinase Cdk1-cyclin B initiates the process of Golgi disassembly during mitosis by phosphorylating Rab1 and Golgin GM130, which together recruit an ER-derived vesicle tethering factor (p115) to cis-Golgi membranes (Misteli and Warren, 1995). Other kinases, including polo-like kinases, extracellular signal-regulated kinases and mitogen-activated protein kinases, function along with Cdk1-cyclin B to dismantle the Golgi during mitosis, and additional kinases contribute to bulk Golgi trafficking during interphase or modulate vesicle coat assembly (Dirac-Svejstrup et al., 2000; Doray et al., 2002; Preisinger and Barr, 2005). In these examples of full Golgi collapse, Golgi partition during mitosis and vesicle coat disruption, kinase activities initiate dramatic Golgi remodeling events. The Sff function that we have described is more restricted, impacting a subset of relatively minor glycan structures primarily associated with the late Golgi. Thus, the relative dosage of specific kinases, within a larger ensemble of phosphorylating enzymes, might sculpt unique glycan expression profiles in specific cell types through targeted modulation of Golgi compartmentation.

Based on the function of SAD kinase at the mature synapse, where it facilitates tethering of vesicles to membrane-associated protein complexes, and on the narrowly distributed changes in the N-glycan profile of sff mutant embryos, we propose that Sff kinase activity modulates Golgi vesicle interactions within a subset of cisternal membranes (Fig. 8). Our glycan characterization indicates that appropriate targeting of these vesicles normally limits the expansion of N-linked glycan complexity and facilitates the generation of HRP epitopes in neural tissue. When sff activity is reduced, some glycoprotein substrates gain access to alternative Golgi processing enzymes and the loss in tethering specificity generates a shift in Golgi vesicle trafficking.

Vesicular capture at presynaptic membranes requires the interaction of specific Rab GTPase proteins and scaffolding/tethering factors (Inoue et al., 2006; Schoch et al., 2002; Sinka et al., 2008). Vesicular transport through the Golgi apparatus is also modulated by interactions between Rab proteins and tethering factors (Ren et al., 2009; Sztul and Lupashin, 2009). The Golgins, which exhibit compartment-specific distributions across the Golgi, are a component of the mechanism that imparts specificity to vesicle-target recognition (Short et al., 2005). Structural analysis predicts that the Golgins are extended molecules that reach into the surrounding cytoplasm from the Golgi membranes to which they are attached, providing a matrix for interaction and tethering of transport vesicles. The glycosylation phenotype that we have quantified in our Drosophila sff mutant indicates that the stability or functional integrity of Golgin-Rab-vesicle complexes might be differentially influenced by the activity of specific kinases. Therefore, Golgins, Rabs and, possibly, components of the retrograde transport machinery [conserved oligomeric Golgi (COG) and Golgi associated retrograde protein (GARP) complexes], are prime candidate substrates for phosphorylation by Sff. The modestly expanded distribution of a Golgin (GM130) in the nervous system of the sff^B22 mutant is consistent with the hypothesis that Sff modulates tethering factor function and could suggest that GM130 is a substrate for the enzyme. However, altered GM130 distribution might also result from aberrant modulation of Rabs or downstream tethering factors. Further genetic, biochemical and proteomic analysis will be necessary to ascertain the identity of the Sff substrates most relevant for controlling the Golgi dynamics that regulate tissue-specific glycosylation.

SAD kinases belong to the AMP-activated protein kinase (AMPK)-sucrose non-fermenting 1 (SNF1) family of serine-threonine kinases, many of which convey signals that regulate cell polarity and energy utilization (Barnes et al., 2007; Carling et al., 1994; Kishi et al., 2005). Neuronal polarization defects are not apparent in our sff mutant embryo. The lack of gross polarity defects in sff might simply reflect the hypomorphic nature of our sff^B22 allele, indicating that neuronal polarization requires less Sff protein than is necessary for establishing neuron-specific glycan expression. However, the sorting function of the Golgi apparatus is intimately associated with cell polarity. In some cases, the glycans on sorted proteins influence their sorting, linking the Golgi function that we have detected for Sff to cell polarity (Potter et al., 2006).

The reduction of sff mRNA in the Tollo mutant background implicates transcriptional control or message stability as a mechanism underlying neuron-specific glycan expression. Vertebrate SAD kinase expression is regulated through metabolic sensing pathways involving protein kinase B (AKT) inhibition of TSC1-TSC2 and activation of TORC1 (Wildonger et al., 2008). This paradigm nominates a range of survival/growth factors, hormones, neurotransmitters and other effectors of AKT or phosphoinositide 3-kinase (PI3K) activation as molecules capable of propagating the transcellular signal initiated by Tollo in Drosophila. In a broader context, our results provide a new mechanism for reinterpreting many years of empirical observations describing the impact of endocrine factors, tissue microenvironments and culture conditions on cellular protein glycosylation (Knezevic et al., 2009; Varki, 1993).

The authors gratefully acknowledge valuable consultations with L. Wells, R. Steet, P. Zhao, M. Ishihara, T. Katoh, A. Nairn and K. Moremen (CCRC, UGA). K. Habenicht, C. Hupman, M. Hirsu, R. Cantu, S. Lee and N. Price contributed essential technical support during implementation of the screen. We are extremely grateful to Bernard Charroux and Julien Royet (IBDM Luminy) for generously providing the tollo^c5 mutant. This work was supported by NIH/NIGMS (GM072839 to M.T.), a Cousins Foundation Fellowship (to S.B.) and a Toyobo Biotechnology Foundation Long-term Research Grant (to K.A.). Deposited in PMC for release after 12 months.