Glycosyltransferase complexes: the first signs

In eukaryotes, the majority of glycans are synthetized in specialized organelles, the endoplasmic reticulum (ER) and the Golgi apparatus. Together they harbor dozens of functionally distinct glycosyltransferases (and glycosidases) that sequentially add (or remove) a single sugar residue at a time to (or from) the growing oligosaccharide chain [1, 2]. As an example, Fig. 1 shows the schematic representation of the N-glycan processing steps that take place in the ER and the Golgi apparatus of eukaryotic cells. This generally accepted view, however, fails to explain how this sequence of enzymatic reactions is orchestrated to guarantee faithful synthesis of thousands of different glycans without any template, in the presence of enzymes that compete for the same substrate and/acceptor protein and also localize in the same Golgi sub-compartment [3]. Preservation of fidelity is important, as even a single change in the linkage type can have a drastic effect on glycan’s 3D structure and thus, also for its normal functions in a cellular context.

Fig. 1
figure 1

Left schematic representation of the secretory pathway and the main organelles involved in glycan synthesis. 1 The endoplasmic reticulum (ER), 2 the Golgi apparatus and 3 the plasma membrane. Note that some glycosylation (addition of N-acetylglucosamine) takes place also in the cell’s cytoplasm. Right a schematic cartoon showing the sequential processing of N-glycans by glycosyltransferases in the ER and the Golgi. The enzymes (glycosyltransferases and glycosidases) involved are traditionally thought to function separately one after the other by adding or removing sugar residues one at the time in a specified order to and from the growing oligosaccharide chain

Full size image

As of now, it turns out that one answer to that puzzle has been lying around for the last 50 years, but remained mainly unrecognized. Namely, several different studies carried out during the late 1960s and early 1970s already showed that glycosyltransferases tend to exist as enzyme complexes in the cells. One of the first examples of these was the observation that a soluble lactose synthase (LS, EC 2.4.1.22, for enzyme names and definitions, see Table 1) typically found in bovine milk consists of two protein components, GalT-I (encoded by the B4GALT1 gene) and α-lactalbumin [4–8]. This interaction lowered the Km of GalT-I transferase for glucose, enabling it to use glucose as an acceptor for lactose synthesis. Schwarz et al. [9, 10] were the first to identify complexes that consist of two dissimilar glycosyltransferases. By using immunoprecipitation, they showed that a xylosyltransferase (XylT) and a galactosyltransferase (GalT), enzymes that initiate the synthesis of chondroitin sulfate, interact directly with each other. Similar results were also obtained by Fishman [11] between N-acetylgalactosaminyl transferase (GalNAcT) and galactosyltransferase (GalT)) that synthetize neural glycolipids. Although no direct interaction between the two enzymes was demonstrated in this study, it was found that the endogenous acceptor, i.e., the product of the GalNAcT, was a far better substrate for the GalT than an exogenously added glycolipid. It also had an order of magnitude lower Km than the latter. In addition, Ivatt [12] suggested that N-acetyllactosamine synthesizing enzymes (GlcNAcT, GalT) form a complex upon their co-adsorption into the same liposome (in contrast to separate liposomes), as the endogenously generated reaction intermediate was used as the preferential substrate for the reaction. These observations were suggested to be consistent with the facilitated passage of the intermediate glycoprotein within the complex itself and also beneficial for glycosylation, as complex formation likely provides a means to increase the fidelity of glycan synthesis by preventing intervention by competing enzymes.

Table 1 Gene and enzyme names with their synonyms and definitions
Full size table

Since then, a number of other biochemical and cell biological studies including recent live cell imaging methods have provided compelling evidence for the existence of glycosyltransferase complexes in all eukaryotes and also bacteria. The current data now shows that such complexes are prevalent and exist in most—if not all—glycosylation pathways found in eukaryotic cells. In the next paragraphs, we will review the data gathered from yeast, plant and animal cells on these glycosyltransferase complexes one pathway at the time. We will focus mainly on their mutual interactions, rather than to their complexes with other proteins, unless the interaction is critical for the activity of a given enzyme. In the latter part of the review, we will also discuss both structural and biochemical data on the few complexes whose enzymatic activities change upon complex formation, or whose 3D structures have already been determined, just to provide a first glimpse on how these enzymes interact and how complex formation regulates the enzymatic activity of the complex constituents. We apologize the bacterial community for not including any bacterial data in this review, and also those whose data may have escaped our attention.

Complexes involved in N-linked glycans synthesis

Yeast N-glycosyltransferase complexes The yeast N-glycosylation pathway shows a high degree of pathway conservation with higher eukaryotes, involving similar oligosaccharide-dolichol precursor synthesis, glycan transfer to nascent proteins by an oligosaccharyltransferase (OST) complex, and removal in the endoplasmic reticulum (ER) of the three glucoses and the central-arm α1,2-linked Man from the newly transferred Glc3Man9GlcNAc2 [1, 2, 13–15]. The first steps to form GlcNAc2-PP-dolichol involves stepwise addition of two GlcNAcs by GlcNAc-1 phosphate transferase (ALGT7) and Alg13p/Alg14p UDP-GlcNAc-transferases (Fig. 1). These first three enzymes have been shown both by immunoprecipitation and gel filtration analyses [16] to form a hexameric complex with a stoichiometry of 2:2:2 and with native molecular weight of ∼200 kDa. Recently, Alg14 was shown to be the central unit and able to organize the formation of this three-enzyme glycosyltransferase complex [17].

The next enzymes in the row, the ER mannosyltransferases Alg1, Alg2 and Alg11 that add three more mannose residues to form the Man5GlcNAc2-PP-dol intermediate, have similarly been shown to form complexes with each other. Both genetic and biochemical evidence indicates that Alg1 interacts with itself, Alg2, or with Alg11 [18]. Thus, the two heteromeric Alg1-containing complexes differ from one another in that one complex contains Alg2 and the other contains Alg11. Both of these complexes were found to be functionally important, as missense mutations affecting the activity of Alg1, but not its assembly with Alg2 or Alg11, exhibited a dominant negative phenotype. Thus, dolichol-linked N-glycan precursor synthesis on the cytoplasmic side of the ER membrane appears to involve three main enzyme complexes, one formed by Alg7p/Alg13p/Alg14p and the other two either by Alg1/Alg2 or Alg1/Alg11. This arrangement likely ensures that each mannose residue will be linked correctly to the dolichol-linked precursor glycan structure even in the presence of several other competing mannosyltransferases on the cytosolic surface of the ER membrane.

After flipping to the ER lumen by the Rft1 protein [15], four additional mannose residues and three glucose residues are further added to Man5GlcNAc2-PP-dol intermediate. Enzymes responsible for these additions involve the Alg3/Alg9/Alg12 mannosyltransferases [19], which add the four mannoses to form the Man9GlcNAc2-dolichol structure. This is then followed by addition of three glucoses that are added by Alg6, Alg8 and Alg10 glucosyltransferases, but it is currently unclear whether these enzymes form complexes with each other as well. This core oligosaccharide is next attached to a specific asparagine residue (consensus motif Asn-X-Ser/Thr) in a polypeptide chain by the 8 subunit oligosaccharyl-transferase complex [20–24]. The Glc3Man9GlcNAc2 core structure is subject to trimming by α-glucosidases I and II, yielding Glc2Man9GlcNAc2 and Glc1Man9GlcNAc2 structures, the latter of which serves for ER quality control by ER chaperones Erp57, calnexin and calreticulin [25–27]. The glucose and the terminal mannose residue in the middle branch of the oligosaccharide chain are removed by α-glucosidase II and ER mannosidase I, respectively, before transport of the Man8GlcNAc2 to the Golgi.

Upon arrival in the yeast Golgi, up to 200 mannose residues are added to an outer chain of an N-glycan of secretory pathway proteins such as invertase, and the ‘mannan’ structural proteins of the cell wall. In contrast to mammalian N-glycan processing in the Golgi (Fig. 1), this ‘hypermannose’ structure consists of a long backbone of α1,6-linked residues with α1,2-linked branches which usually terminate in α1,3-linked residues [28]. After Och1p has attached the first α1,6-linked mannose to the core structure, two distinct polymerase complexes, M-Pol I (which consists of Van1p/Mnn9p) and M-Pol II (Hoc1p/Mnn11p/Mnn10p/Anp1p/Mnn9p) then synthetize the long mannan backbone [29], while the branching of this backbone is accomplished by the sequential actions of Mnn2p, Mnn5p and Mnn1p. Some of the branches also receive a phosphomannose that is added by Mnn4p and Mnn6p. All these proteins or their complexes were found to have mannosyltransferase activity in vitro.

N-glycosyltransferase complexes in mammals The processing of N-glycans in the Golgi apparatus of higher eukaryotes differs markedly from that of yeast. High mannose glycans synthetized in the ER are extensively processed to form hybrid and complex type N-glycans (Fig. 1) by a defined set of glycosyltransferases and glycosidases in each cell type (Stanley et al. Chap 8. In: [30]). Similar to the ER-localized glycosyltransferases, these Golgi enzymes also tend to form complexes with each other. The first example of these was GalT-I, which has been shown to form high molecular weight oligomers and/or homo(di)mers in isolated membrane preparations prepared from mammalian cells [31, 32]. Oligomerization was thought to have a role in Golgi retention [32]. Similar oligomers or homodimers have since been detected also with GlcNAcT-V, ST6Gal-I, FucT-I, FucT-III and FucT-VI [33–39]. By utilizing bimolecular fluorescence complementation (BiFC) and fluorescence resonance energy transfer (FRET) approaches, we have also recently confirmed that not only GalT-I but also the other main human Golgi N-glycosyltransferases tested (GlcNAcT-I, II, GalT-I, ST6Gal-I, ST3Gal-III) form homomers in live cells [40, 41].

The first heteromeric glycosyltransferase complexes were detected by using an ER-re-location assay developed by the Nilsson group [42, 43]. They showed that GlcNAcT-I, fused with a dibasic ER-retention signal derived from the cytoplasmic tail of the p33 invariant chain, was able to relocalize endogenous mannosidase II or GlcNAcT-II (but not GalT-I) from the Golgi to the ER, suggesting a direct interaction between these medial-Golgi enzymes. Based on their data, the authors suggested the new Golgi “kin recognition” model for medial-Golgi enzymes, which serves their correct targeting to the Golgi. Later on, the interaction between GlcNAcT-I and GlcNAcT-II has also been confirmed by using an immunoprecipitation approach [35]. Enzymes such as β3GnT-8 and β3GnT-2 involved in the elongation of specific branch structures of multi-antennary N-glycans with polylactosamine have also been shown to form a complex with each other [44, 45]. Moreover, cross-testing of all the potential interactions between the main N-glycosyltransferases by using BiFC and FRET approaches, we have shown that they all form heteromeric complexes with each other [40, 41]. The medial-Golgi enzymes GlcNAcT-I and GlcNAcT-II were found to form one such complex that may also contain other medial-Golgi GlcNAc transferases and mannosidase II. The other complexes consist of trans-Golgi enzymes, GalT-I and ST6Gal-I or GalT-I and ST3Gal-III. Cross-testing potential interactions between N- and O-glycosyltransferses [41] did not reveal any interacting partners, suggesting that complexes form only between sequentially acting enzymes within the same glycosylation pathway. Very recently, the BiFC approach was used to show interactions between tyrosylprotein sulfotransferases TPST1 and TPST2 [46] with the result that these enzymes formed complexes not only with themselves, but also with STGalT-I, producing either TPST1/ST6Gal-I or TPST2/ST6Gal-I heteromers. A different type of interaction takes place between GlcNAcT-I and GnT1IPL (Mgat4D) protein in that the latter has no known enzymatic activity and is expressed at high levels in the testicular germ cells [47]. The functional relevance of this inhibitory interaction has been shown to be in the down-regulation of the synthesis of hybrid and complex type N-glycans and thereby to increase attachment of developing sperm cells to nourishing Sertoli cells.

Plant N-glycosyltransferase complexes Complex type N-glycans in plants are essential for plant development and defense mechanisms. However, the glycans themselves are somewhat different from mammalian N-glycans. For example, the proximal core N-acetylglucosamine in plants is substituted by an α1,3 fucose (instead of α1,6 fucose in mammals) and the β-mannose of the core is substituted by a bisecting β1,2 xylose (a bisecting β1,4 N-acetylglucosamine in mammals). In addition, β1,3 galactose and α1,4 fucose linked to the terminal N-acetylglucosamine of plant N-glycans form the Lewis a (Le a) oligosaccharide structure instead of β1,4 linked galactose followed by a sialic acid in mammals. Recently, it has been shown [48, 49] that a number of plant N-glycosyltransferases also form complexes with each other. By using the two-photon FRET-FLIM approach, it was shown that Arabidopsis thaliana Golgi α-mannosidase I, Nicotiana tabacum β1,2-N-acetylglucosaminyltransferase I, Arabidopsis Golgi α-mannosidase II (ManII), and Arabidopsis β1,2-xylosyltransferase, form both homodimers and heterodimers, whereas the late-acting Arabidopsis β1,3-galactosyltransferase1 (GALT1) and Arabidopsis α1,4-fucosyltransferase, do not. However, GALT1 was found to interact with the medial-Golgi ManII. This observation may reflect slight differences in the organization of the Golgi cisternae between plants and mammalian cells, as the latter do not show any such interactions between medial- and trans-Golgi enzymes tested thus far [40, 41].

Collectively, these data show that most N-glycosyltransferases in yeast, plant and mammalian cells form both enzyme homomers and a variety of functionally relevant enzyme heteromers between sequentially acting glycosyltransferases in each glycosylation pathway (see Table 2 for a collected list the known enzyme complexes described in this review). Enzymes that function in different glycosylation pathways are expected to form distinct complexes with other enzymes in a pathway dependent manner. An important question that needs to be answered is whether heteromerization is an inherent property of all glycosyltransferases or only those that operate at critical points such as at initiation, branching, and termination of glycan chains.

Table 2 Known glycosyltransferase complexes, their functions, species identified and references
Full size table

Complexes involved in O-linked glycan synthesis

Mucin type O-GalNAc glycosylation O-glycosyltransferases catalyze addition of sugar residues to hydroxyl groups of serine or threonine amino acids in proteins. Most common O-linked glycans are mucins, and ~20 different GalNAcTs exist to initiate their synthesis using different protein acceptors. This is then followed by chain elongation to form 7 different core structures which are further processed to form a number of different glycan structures. Enzymes that form these core structures are well known and have also been tested for their ability to form complexes in live cells. As with N-glycosyltransferases, all the 9 core forming transferases and two sialyltransferases [41] were found to form homomeric complexes in the cells (see Table 2). This is consistent with the earlier notion that the core 2 β1,6 N-acetylglucosaminyl-transferase (C2GNT-I) forms disulfide bonded dimers [50]. In addition, it has been shown that the active C1GalT-1 (T-synthase, Core 1 synthase, β3-galactosyltransferase) is a dimer and does not bind its specific folding chaperone Cosmc, unlike the unfolded and inactive enzyme [51, 52]. Other potential homomeric and heteromeric interactions between O-glycosyltransferases themselves were recently tested by the dynamic FRET approach. Similarly to N-glycosyltransferases, it was found that all these 9 enzymes form enzyme homomers [41]. In addition, the initiating GalNAcT-6 and C1GalT required for the synthesis of the T(F)-antigen were shown to form a hetoromeric complex with each other. The same GalNAcT6 was also found to interact with the core 3 and core 6 forming glycosyltransferases (C3GnT-I, C2GNT-I). Cross-testing of other potential interactions between the nine O-glycosyltransferases tested did not reveal any other enzyme heteromers in the cells [41].

Other O-linked glycans Drosophila and mammalian Fringe proteins possess a fucose-specific α1,3 N-acetylglucosaminyltransferase activity that initiates elongation of O-linked fucose residues attached to epidermal growth factor-like sequence repeats of Notch. Although direct evidence is missing, it has been suggested that Fringe proteins function as dimers [53]. O-mannose containing proteins were originally identified in yeast already in the 1950s. Their typical tetra-saccharide structure consists of Siaα2–3Galβ1–4GlcNAcβ1–2ManSer/Thr. They are now known to constitute one-third of all O-linked glycans in the brain of higher organisms, but are also abundant in the muscle tissue, in which the protein α-dystroglycan is the main acceptor for O-linked mannose addition. A defect in its O-mannosylation/glycosylation is associated with Walker-Warburg syndrome, an autosomal recessive multisystem disorder characterized by complex eye and brain abnormalities with congenital muscular dystrophy [54, 55]. The enzymes responsible for adding the O-mannose and N-acetylglucosamine are catalyzed by O-mannosyltransferase 1 (POMT1) and 2 (POMT2). Neither POMT1 nor POMT2 have enzymatic activity alone, but when co-expressed, O-mannosyltransferase activity was recovered [56, 57]. This suggests that the activity of POMT1 and POMT2 requires physical interaction of these enzymes in vivo. It is not known whether POMGNT1, which adds β1,2-linked N-acetylglucosamine to O-mannose-containing glycoproteins and glycopeptides forms complexes with itself, later enzymes in the pathway or with POMTs themselves. In addition to mammalian enzymes, yeast O-mannosyltransferases have also been shown to form complexes with each other. Members of the PMT1 subfamily Pmt1p and Pmt5p were found to interact with members of the PMT2 subfamily, forming complexes between Pmt2p/Pmt3p, Pmt1p/Pmt2p or Pmt5p/Pmt3p [58]. On the other hand, Pmt4p (a member of the PMT4 subfamily) was found to form and function as a homomer. Finally, Proteins with collagen domains are modified by a disaccharide, Glcα1–2Galβ, which is assembled on hydroxylysine or hydroxyproline residues. The first step in the pathway is the obligate hydroxylation of either lysine or proline by lysyl- and prolylhydroxylases prior to addition of glucose (and galactose). Glycosylation takes place before the formation of the triple helix, and may control the rate of triple-helix formation and, thus, also the size of the collagen fibrils [59]. Mice lacking the lysylhydroxylase 3 isoform (LH3) fail to glycosylate collagen IV properly, causing embryonic lethality and deposition of misfolded collagen in the ER [60]. This single polypeptide-containing enzyme has all three enzyme activities needed for collagen glycosylation, and has been shown to exist as a homodimer on the luminal side of the ER membrane [61, 62].

Complexes involved in glycolipid synthesis

Glycosphingolipids form the main class of glycolipids. Their synthesis starts in the ER, and is completed by distinct glycolipid glycosyltransferases operating in the Golgi apparatus [30, 63, 64]. The first step in this series of reactions involves addition of either galactose or glucose to ceramide to form galactosyl- or glucosylceramide (GalCer, GlcCer). GalCer and other galactolipids are seldom extended to larger glycans. In more complex vertebrate glycosphingolipids, the GlcCer is typically extended first with β-linked galactose to give lactosylceramide (Galβ1-4GlcβCer) which is then further extended to generate a series of neutral “core” structures classified as ganglio-series (Galβ1-3GalNAcβ1-4Galβ1-4GlcβCer), neolacto-series (Galβ1-4GlcNAcβ1-3Galβ1-4GlcβCer), lacto-series (Galβ1-3GlcNAcβ1-3Galβ1-4GlcβCer), globo-series (Galα1-4Galβ1-4GlcβCer), and isoglobo-series (Galα1-3Galβ1-4GlcβCer). Of these, the ganglio-series are the most common, being highly concentrated in the brain and being rich in sialic acids, making them negatively charged and thus important in the processes of neural signaling [63]. The neolacto-series in turn are most common in the hematopoietic system; lacto-series are prominent in secretory organs while the globo-series are most abundant in erythrocytes [30].

The synthesis of ganglio-series glycosphingolipids involves first B4GalT6, which adds the β1,4-linked galactose to ceramide yielding lactosylceramide (LacCer i.e. Galβ1-4GlcβCer). This is then further modified mainly (with the exception of neutral GA2 “core” structure by B4GALNT1) to acidic gangliosides by SiaT9 and SiaT8A to form mono-, di- or tri-sialylated gangliosides (GM3, GD3 and GT3 gangliosides, respectively). Their synthesis is further elaborated to a-, b- and c-series by B4GALNT1 N-acetylgalactosaminyltransferase and by B3GalT4 galactosyltransferase. Several of these enzymes have been shown to form complexes with each other. For example, both GM2 and GD3 synthases have been shown to form homodimers [65, 66]. Enzyme activity measurements performed by Bieberich et al. [67] suggested an enzyme complex also between ST2 (Siat8A, also named ST8SIA1, Table 1) and GalNAcT (B4GALNT1) but not with ST1 (SiaT9). Complex formation between SiaT8A and B4GALNT1 was confirmed both by co-immunoprecipitation and by FRET measurements. Similarly, by using co-immunoprecipitation and FRET microscopy, Maccioni et al. were able to show that the GM3, GD3, and GT3 synthases (B4GalT6, SiaT9 and SiaT8A, respectively) form a multi-enzyme complex in the proximal Golgi [68–70], whereas the GM2, GD2 and GT2 synthase (B4GALNT1 and B3GalT4) form a distinct complex in the distal Golgi (TGN). Currently, it is not known whether similar complexes form also between enzymes that extend these sugar structures with sialic acid, fucose or GlcA. Some of these enzymes (e.g., Siat7, Siat4) may also operate in other glycosylation pathways as well.

Yeast and plant glycolipids Although lactosylceramide is the most common glycosphingolipid in vertebrates, it is rare in other eukaryotes. For example, it is substituted by Manβ1-4Glc-Cer and GlcNAcβ1-4Glc-Cer in invertebrates by inositol-1-O-phosphorylceramide in yeast, and by GlcNAcα1-4GlcAα1-2-myo-inositol-1-O-phosphorylceramide in plants. No direct evidence exists at the moment if the enzymes that synthesize these glycolipids form similar complexes to those described above for mammalian glycolipid synthesizing enzymes.

Complexes involved in proteoglycan synthesis

The Golgi apparatus is also responsible for the synthesis of cell surface and extracellular matrix proteoglycans. The polysaccharide side chains (glycosaminoglycan) added to various proteoglycans include keratin sulfate, chrondroitin sulfate, dermatan sulfate and heparin or heparan sulfate [30]. They all consist of disaccharide repeats made of either glucuronic acid (GlcA), iduronic acid (IdoA), GalNAc, GlcNAc or Gal. They are also variably sulfated, linear, and often up to 100 sugar residues long attached to serine residues of various proteins via a common tetrasaccharide primer (Xyl-Gal-Gal-GlcA). An exception is keratansulfate, which utilizes N- (KS I) or O-linked (KS II) sulfated poly-N-acetyllactosamine structures ([Galβ1-4GlcNAcβ1-3] n ) that are otherwise identical to those found on N- and O-linked glycoproteins. Thus, the enzymes responsible for keratan sulfate synthesis, synthesize other glycans as well.

By using electrophoretic mobility assays and cross-linking studies of the membrane-bound GlcAT-1, Ouzzine et al. [71] have shown that the membrane-bound GlcAT-I, an enzyme that adds glucuronic acid residue onto the above trisaccharide primer, forms enzymatically active disulfide-linked homodimers. In addition, McCormick et al. [72] have shown that EXT1 and EXT2 (both of which have GlcA- and GlcNAc-transferase activities and contribute to heparan sulfate biosynthesis [73] ) form not only enzyme homomers but also heteromers both in yeast and mammalian cells. Heteromerization also augmented their enzymatic activities in cell lysates. Moreover, of the five heparan sulfate biosynthetic enzymes, galactosyltransferase I and glucuronosyltransferase I (required for the formation of the linkage region), GlcNAc N-deacetylase/N-sulfotransferase 1, uronosyl 5-epimerase, and uronosyl 2-O-sulfotransferase), Pinhal et al. [74] were able to show that the last two of these (uronosyl 5-epimerase and uronosyl 2-O-sulfotransferase) interact with each other based on the ER relocation assay developed by Nilsson’s group [43]. Izumikawa et al. [75] also demonstrated multiple interactions between chondroitin synthase-1 (ChSy-1), ChSy-2 (chondroitin sulfate synthase 3, CSS3), and also between chondroitin-polymerizing factor (ChPF) and glucuronyltransferase II (CSGlcA-T).

Hyaluronan, a glycosaminoglycan defined by the disaccharide unit (GlcNAcβ1–4GlcAβ 1–3) n that is neither sulfated nor covalently linked to any protein, is rather exceptionally synthesized at the plasma membrane in various cell types by transmembrane enzymes called hyaluronan synthases. By using co-immunoprecipitation, FRET measurements and the proximity ligation assay, Tammi et al. [76] also recently were able to show that each of the three different hyaluronan synthases (HAS1, HAS2, HAS3) interact either with itself or with each other, forming heteromeric HAS1/HAS2, HAS1/HAS3 and HAS2/HAS3 complexes.

A vast array of other proteins is known to be modified with various glycans such as N-acetylglucosamine (GlcNAcazylation) and C-linked mannose. These modifications are in many cases also accomplished by glycosyltransferases complexed with either soluble or membrane proteins, which however, do not possess known transferase activity, and therefore, do not represent the main focus of this review.

Complexes involved in cell wall glycan and starch synthesis

Chitin synthesis In S. cerevisiae, the β1,4-linked N-acetylglucosamine polymer, chitin, is synthesized by a family of 3 specialized chitin synthases encoded by CHS1, CHS2 and CHS3 genes. Each of these have special role in the synthesis of septum, lateral cell walls and the bud neck. Whether all these enzymes form oligomers is currently mostly unclear, but it has been shown that at least the Chs3p isoform forms a complex with Chs4p/Skt5p protein [77]. In the tobacco hornworm (Manduca sexta), the CHS-2 has a molecular weight of 520 kDa, thrice the calculated molecular weight of the monomer, suggesting it exists as a trimer [78].

Cellulose synthesis Plant cell walls are composed mainly of high-molecular-weight polysaccharides, proteins, and lignins. Cellulose is the most abundant linear polysaccharide present and consists of up to thousands of β1,4-linked d-glucose units. These are synthesized in plants by a plasma membrane-localized cellulose synthase complex that forms 25–30 nm diameter symmetrical rosettes each with six subunits [79]. The core components of this complex in higher plants are a family of ten CESA proteins [80–82]. Genetic and biochemical evidence has shown that three unique CESA isoforms are required for both primary cell wall and secondary cell wall cellulose synthesis. In Arabidopsis thaliana, these are CESA1, CESA3, and CESA6 and CESA4, CESA7, and CESA8, respectively [83, 84]. The remaining CESAs (CESA2, CESA5, CESA9, and CESA10) are likely involved in tissue-specific processes and are partially redundant with CESA6 [85]. Intriguingly, these enzymes were also shown to homodimerize prior to formation of larger CESA oligomers [84], similar to N-glycosyltransferases found in mammalian cells [86]. Very recently, a split-ubiquitin membrane yeast two-hybrid system demonstrated interactions between the four primary CESAs (CESA1, CESA2, CESA3, CESA6) and three secondary CESAs (CESA4, CESA7, CESA8) but also between the primary CESAs and secondary CESAs in a limited fashion. Further functional analysis of transgenic lines showed that CESA1 could partially rescue irx1 (cesa8) null mutants, resulting in complementation of the plant growth defect and cellulose content deficiency [87]. The interactions between these enzymes were shown to utilize both disulfide bonds and non-covalent interactions [88].

Pectin and pectic arabinan synthesis Plant cell wall pectic polysaccharides are complex carbohydrates that are synthesized by low-abundance, Golgi membrane-bound biosynthetic enzymes. Arabidopsis galacturonosyltransferase (GAUT) 1 is an α1,4-galacturonosyltransferase (GalAT) that synthesizes homogalacturonan, the most abundant pectic polysaccharide. It has been shown that GAUT1 functions in a protein complex with the homologous GAUT7 protein [89]. In addition, Golgi localization of GAUT1 is dependent on the formation of the GAUT1/GAUT7 complex. Other yet unknown protein components were also found to co-immunoprecipitate with the GAUT1/GAUT7, but their functional relevance in pectin synthesis remains obscure. Similarly, pectic arabinan, consisting of α1,5-linked l-arabinofuranosyl residues to which other l-arabinofuranosyl residues are attached via α1,3- and α1,2-linkages, form a comb-like arrangement, and are synthetized by ARAD1 and its close homolog ARAD2. By using bimolecular fluorescence complementation, FRET and non-reducing gel electrophoresis, Harholt et al. [90] showed that ARAD1 and ARAD2 are localized in the same Golgi compartment and form intermolecular homo- and heterodimers when expressed transiently in Nicotiana benthamiana.

Xyloglucan synthesis Xyloglucan is the major hemicellulosic polysaccharide in the primary cell walls of dicotyledonous plants and has important structural and physiological functions in plant growth and development. In Arabidopsis, the Golgi-localized 1,4-β-glucan synthase, Cellulose Synthase-Like C4 (CSLC4), and three xylosyltransferases, XXT1, XXT2, and XXT5, are responsible for the synthesis of the xyloglucan backbone. By using bimolecular fluorescence, complementation and in vitro pull-down assays Chou et al. [91] have shown that at least two of these enzymes CSLC4 and XXT2 form homomeric complexes. Heteromeric complexes were also detected between XXT2/XXT5, XXT1/XXT2, and XXT5/CSLC4. The same authors also showed very recently that three additional enzymes (MUR3, XLT2 and FUT1) involved in the xyloglycan synthesis form complexes between FUT1 and MUR3, XLT2, XXT2 or XXT5. XLT2 also interacts with XXT5, but MUR3 does not [92]. They also showed that FUT1, XXTs and CSLC4 also form disulfide-linked enzyme homomers, while the formation of the heteromers does not involve covalent interactions. In vitro pull-down assays indicated that in the FUT1/MUR3 and FUT1/XXT2 interactions are mediated by the catalytic domains of these enzymes.

Starch synthesis Amylose is a major polysaccharide of starch, making up approximately 20–30 % of the structure. The other main component of starch is amylopectin, which is branched and makes up to 70–80 % of the structure. Previous studies have indicated that starch synthases form high molecular weight complexes [93–95]. Co-immunoprecipitation experiments and affinity chromatography assays with recombinant proteins showed that starch synthase I (SSI), SSIIa, and SBEIIb (starch branching enzyme) form complexes with each other. All interactions were enhanced by ATP and broken by alkaline phosphatase, indicating a role for protein phosphorylation in their assembly. The authors proposed that, during amylopectin biosynthesis, SSI and SSIIa form the core of a phosphorylation-dependent glucan-synthesizing protein complex, which then recruits SBEIIb. Differences in stromal protein complexes mirrored the complement of the starch synthesizing enzymes detected in the starch granules, suggesting that the complexes have a functional role in starch biosynthesis.

Protein domains involved in glycosyltransferase interactions

The vast majority of Golgi localized glycosyltransferases are type II membrane proteins (Fig. 2), and thus, have a short N-terminal cytoplasmic domain, a single-pass transmembrane domain (TMD), a stem domain and a C-terminal catalytic domain facing the Golgi lumen [30, 96, 97]. Based on current evidence, all these domains have been shown to link glycosyltransferases together. For example, homodimerization of many enzymes such as GalT-I and ST6Gal-I has been shown to be mediated by disulphide-bonds located in their transmembrane domains [32, 33, 37, 50, 71, 98]. Qian et al. [37] showed that a conserved Cys24 in the transmembrane region of rat ST6Gal-I is required for dimerization, as a Cys24Ala variant remained monomeric on SDS-PAGE. This does not, however, exclude the presence of other non-covalent interactions involved in homodimer formation. In fact, our domain swapping experiments combined with FRET measurements have shown that Cys24 deletion mutants still interact in the FRET assay via their catalytic domains [99]. Other Cys variants made (9 total) probably reflect misfolding rather than breakage of monomer–monomer contacts, given that the mutants localize into the ER (unpublished observations) or are used as sialyl motifs L and S within the catalytic domain of each ST6Gal-I monomer [100, 101]. Some plant glycosyltransferases also seem to utilize the same system for complex formation [48, 49, 90]. Similarly, glycolipid-synthesizing enzymes GalNAcT and GalT2 have also been shown to interact via their N-terminal cytosolic and transmembrane domains [68, 102].

Fig. 2
figure 2

Most of the Golgi localized glycosyltransferases are type II membrane proteins with a short N-terminal cytoplasmic domain, a ~20 amino acid α-helical TM-domain, a stem domain and a C-terminal globular catalytic domain in the lumen of the secretory pathway

Full size image

In addition to disulphide bonds, both the stem and catalytic domains of glycosyltransferases have been shown to be responsible for the heteromeric interactions between mammalian medial-Golgi enzymes GlcNAcT-I and ManII [35, 103]. Our previous domain swapping experiments showed that nearly all the main human N- and O-glycosyltransferase interactions are mediated by interactions between the catalytic domains of these enzymes [41], the only exception was GalT-I whose interactions either with itself or with ST6Gal-I seemed to involve mainly the cytoplasmic, transmembrane and/or the stem domain (120 N-terminal amino acids). Moreover, Arabidopsis cis- and medial-Golgi enzymes [48] seem also to utilize amino-terminal cytoplasmic-transmembrane-stem regions (CTS) for complex formation. Atmodjo et al. [89] in turn showed that Arabidopsis Golgi galacturonosyltransferases form heteromers via both covalent and non-covalent forces between their catalytic domains. By using BiFC and pull down assays, Chou et al. [92] have provided data recently to show that xyloglucan xylosyltransferases (XXT) not only form both homo- and heteromeric complexes via covalent and non-covalent bonds between the catalytic domains, but that the interaction surfaces are likely different depending on the XXT isoforms in the complex. The observed difference in the pH sensitivity of the N- and O-glycosyltransferase heteromers, relative to enzyme homomers [41], also suggests that the interaction surfaces needed for the self and non-self-binding are likely different.

Collectively, the observation that all different domains are utilized for complex formation depending on the enzyme(s) in question does not necessarily reflect contradictory findings, but rather, high sequence diversity of this class of enzymes and the use of distinct interaction surfaces in each case to link enzymes together. This would guarantee sufficient specificity for the interactions so that only relevant enzymes have a chance to interact. Sequence diversity at the amino acid level among glycosyltransferases may thus be an evolutionally driven phenomenon that likely has helped to organize glycosyltransferases into distinct and functionally relevant units.

Functional relevance of the glycosyltransferase complexes

Complex formation has been thought to be beneficial for glycosylation in several respects. Firstly, it has the potential to regulate the enzymatic activity of the complex constituents. Previous studies have shown that although enzymes can be active as monomers, homomers and heteromers [104], yet there is clear evidence that some enzymes also become activated upon complex formation. The best-known example of this is the heparin sulfate synthesis by EXT1 and EXT2 glycosyltransferases. When expressed alone, these enzymes localize predominantly to the ER and have only moderate activity [72]. However, when co-expressed, they localize to the Golgi and form more active enzyme heterodimers, as demonstrated by the increased levels of cell-associated GlcA and GlcNAc. Similarly, studies both in yeast and Cos-7 cells showed that while both EXT1 and EXT2 enzymes have GlcAT and GlcNAcT activity, their activities were markedly (3–7 fold, Table 3) augmented upon their co-expression [73]. The increase of glycosyltransferase activities was demonstrated also using only EXT1 and EXT2 catalytic domains, suggesting that they are sufficient to generate the heteromeric functional enzyme. Yet, there is still some controversy that relates to the exact functional role of EXT2, as it has very low activity. One possibility is that it may help folding and/or transport of EXT1 to the Golgi [105]. In addition, co-expression of polylactosamine synthesizing enzymes β3GnT-2 and β3GnT-8 has also been shown to increase the activity of both of these two enzymes in vitro [45]. Recent data from our laboratory has also shown that GalT-I and the ST6Gal-I activities are both increased by roughly 2.5 fold (relative to homodimers) upon their co-expression [41]. It is notable that in the case of GalT-I, the observed increase was not due to any trivial change in the amount of the enzyme protein itself nor of acceptor substrates, but rather to its interaction with the later enzyme in the same pathway. Furthermore, synthesis of GM3 and GD3 glycolipids by sialyltransferases Sial-T1 and Sial-T2 has been shown to result in near 2.5-fold higher Sial-T1 activity upon their co-transfection in CHO-K1 cells, relative to single transfected cells [106, 107]. This activity increase was also not found to be due to the appearance of Sial-T1 gene transcription activators or the stabilization of the Sial-T1 protein, but rather, to the activation of the Sial-T1 enzyme itself due to the formation of Gal-T1/Sial-T1/Sial-T2 multi-enzyme complex.

Table 3 Enzymatic activity of known glycosyltransferase homomers and heteromers
Full size table

Complex formation may also modulate the enzymatic activity of glycosyltransferases in a different manner. For example, ganglioside GM2 synthase [98] was found to be more active as a homodimer than as a monomer, whereas ST6Gal-I [33] monomer was more active than the homodimer. Unfortunately, however, it remained unclear to what extent, if any, heteromerization alters the activity of these enzymes (relative to enzyme homomers and monomers) in vivo and in vitro. Another example of an inhibitory interaction is the recently identified glycosyltransferase-like protein GnT1IP-L [47], which is known to interact specifically with the medial-Golgi enzyme GlcNAcT-I. Even though the GnT1IP-L protein does not seem to possess any measurable glycosyltransferase activity, its binding causes almost complete inhibition of GlcNAcT-I, which in turn down-regulates the synthesis of complex and hybrid N-glycans and helps developing sperm cells to remain attached to Sertoli cells and further differentiate into mature sperm cells. Similar interactions of glycosyltransferases with proteins that do not possess any enzymatic activity can modulate the enzymatic activity also by other means. As an example of this type of effect is the C1GalT-1 (T-synthase) and Cosmc, a specific folding chaperone that helps the enzyme to fold correctly [51, 52]. Another example is the DPM synthase, an enzyme that is needed for the synthesis of dolichol-phosphate-mannose (DPM), and consists of three non-homologous proteins, DPM1, DPM2 and DPM3. Of these, DPM2 and DPM3 are needed to keep the catalytically active DPM1 bound to the ER membrane and active, as otherwise it would move to the plasma membrane [108–112]. In addition, DPM2 may enhance the binding of the acceptor dolichol-phosphate and thereby keep DPM1 more active [109]. Moreover, cross-talk between DPM synthase and N-acetylglucosaminyl 1-phosphate transferase has been shown to increase the activity of both enzymes [111]. Complex formation between GPI-N-acetylglucosaminyltransferase and DPM2 has also been found to increase the transferase activity of the former by threefold [112]. Collectively, these observations suggest that the activity of the glycosyltransferases in many cases is modulated by complex formation. An increase in enzyme activity likely reflects co-operative functioning of the interacting enzymes in glycan synthesis and may involve substrate channeling through the complex and/or changes in the affinity of donor or acceptor substrates. Both of these are necessary prerequisites for faster processing and synthesis of a glycan chain. Complex formation may also be inhibitory to allow down-regulation of some glycosylation reactions for regulatory purposes. Other surprises are likely to come when more work are accomplished with other glycosyltransferase complexes.

Secondly, complex formation likely helps to preserve fidelity during glycan synthesis. Although direct evidence is still lacking, complex formation is generally thought to provide a means to prevent intervention by potentially competing enzymes, which otherwise would be able to use the same sugar as an acceptor for glycan extension. Thus, by linking relevant (sequentially acting) enzymes together, complex formation would guarantee that correct sugar and linkage type would be used to extend the glycan chain. In this scenario, enzyme complexes could act as a kind of “templates” for each specific glycosylation step. The mucin-type core O-glycosylation represents an example of this. It involves several core structure forming enzyme complexes that can potentially compete with other enzymes that use the same GalNAc as an acceptor, such as sialylation of the initiating GalNAc to form the Tn-antigen. Normally, the presence of a preformed heteromeric core complex such as that between ppGalNAcT-6 and C1GalT-I is expected to prevent such competing reactions from occurring, thus favoring the synthesis of a complex-specific core structure (e.g. T-antigen, the product of the ppGalNAcT-6/C1GalT-I complex). In the absence of such a complex, such as in cancer cells [41], competing enzymes such as the ST6GalNAc-I can function and prematurely sialylate the initiating O-linked GalNAc. Consistent with this scenario, cancer cells typically express both Tn- and sialyl-Tn antigen at much higher levels than non-malignant cells (see [41] and references therein). Moreover, we showed that a Golgi pH increase inhibits the formation of this heteromeric ppGalNAcT-6/C1GalT-I complex and also the normal mucin type O-glycosylation likely by affecting both the core 1 synthesis and its extension to more complex O-glycans. Nevertheless, direct proof is needed and requires the construction and/or testing mutant enzymes which preserve their activity but cannot interact with their relevant partner.

Thirdly, complex formation may also modulate glycan synthesis by altering enzyme’s substrate or acceptor specificity, or its polymerizing capacity. Lactose synthase complex is a good example for the first case. Lactalbumin binding to GalT-I enables the enzyme to use glucose as an alternate acceptor [113]. On the other hand, chondroitin sulfate polymerizing glucuronyltransferase (chondroitin synthase-3) is an example of the case where the polymerizing capacity of an enzyme is changed upon complex formation. This enzyme has been shown to form multiple enzyme complexes that consist of distinct chondroitin synthase family members [75]. Depending on the complex, distinct sized chondroitin sulfate chains were produced. In addition, they exhibited distinct, but overlapping acceptor substrate specificities towards two synthetic acceptor substrates. The same situation may also be true with hyaluronan synthesis where different complexes between HAS1, HAS2 and HAS3 have been shown to exist recently [76].

Fourthly, the observed pH sensitivity of the enzyme heteromers, but not of homomers [41], is important to notice, and reflects their formation in different cellular compartments. The latter were shown to form in the ER whereas the former are now known to assemble in the acidic Golgi environment [86]. This result emphasizes that homomers and heteromers are not competing enzyme species, but rather represent inter-dependent membrane constituents that undergo constant organelle micro-environment-dependent transitions between two physical states during their suggested recycling between the early secretory compartments (Fig. 3). Such transitions indeed take place between the ER and the Golgi [86], and provide a simple means to localize the most active enzyme species (heteromers) in the Golgi where these enzymes are known to operate, and thereby can increase the glycosylation potential of the Golgi. The absence of enzyme heteromers in acidification and glycosylation -defective cancer cells is in accordance with this view. Nevertheless, an important question that remains is whether divalent cations needed for the enzymatic activity of several glycosyltransferases have any role in complex formation. It needs also to be emphasized that not all enzymes have been shown to form such heteromers. Thus, the possibility that some enzymes may exist and also function as monomers and/or homomers cannot be excluded at this point.

Fig. 3
figure 3

A schematic model of the dynamic transitions between glycosyltransferase homomers and heteromers and their dependence on Golgi acidity (for details, see text)

Full size image

Finally, complex formation has been suggested to have a role also in the correct targeting of glycosyltransferases to the Golgi and to its various sub-compartments. This possibility was raised in the early 90s by showing that oligomerization correlated with the Golgi localization of various membrane proteins including coronavirus m1 protein and various glycosyltransferases [35, 42, 43, 114, 115] along with the “kin recognition” hypothesis [43]. These observations raised an idea that Golgi enzymes form complexes large enough not to fit into the forming transport carriers, leading to their retention in the organelle. However, the now generally accepted “cisternal maturation” model is not in accord with this oligomerization-mediated Golgi retention hypothesis, as it assumes that the Golgi enzymes continuously recycle between Golgi cisternae and the ER. In fact, it has been shown by using the FRAP (fluorescence recovery after photobleaching) approach that glycosyltransferases in general [116] as well as their homomers and heteromers [86] remain as mobile Golgi membrane constituents that recycle between the first two secretory compartments. Thus, even though some enzymes are mislocalized after experimental Golgi pH increase [99], it is currently not clear whether complex formation itself is a determinant for Golgi retention or retrieval. Further studies with explicitly monomeric and oligomeric glycosyltransferases will help answer to this pertinent question.

Structural aspects of glycosyltransferases and their complexes

There are approximately 250 known glycosyltransferases which, according to their DNA sequence, are categorized into 97 distinct families. Classification is based on amino acid sequence comparisons and differences in activated donor and specific acceptor molecules in forming glycosidic bonds. Crystal structures exist for glycosyltransferases representing 41 families including 11 families with at least one human enzyme structure solved. Yet, nearly all the existing crystallographic structures represent only the globular catalytic domains of enzyme monomers. They all also fall into one of the three main fold types (GT-A, GT-B and Cst II fold), depending on the relative positions of α-helices and a central β-sheet in each fold [97]. The main impact of all these structures is that they have helped to resolve both donor and acceptor glycan binding modes as well as their catalytic mechanisms (for excellent reviews, see [117–121] ). Here, instead, we focus on the few enzymes whose structures either appear in the crystals, or have been resolved, as complexes.

Volkers et al. [122] recently published apo- and ligand-bound crystal structures of human ST8SiaIII sialyltransferase active in polysialylation. The two enzyme monomers in the asymmetric unit are related by near two-fold symmetry (Fig. 4). Analysis by using the PISA server [123] showed that the dimer interface area ≈968 Å2 is well in the range typically observed for functionally relevant protein complexes and the presence of such a dimer was also experimentally detected in solution. ST8SiaIII structure represents a prime example of a glycosyltransferase homodimer with well plausible assembly: the dimer formation places both active sites on the same side of the dimer and within a distance from each other, which makes various functional scenarios possible.

Fig. 4
figure 4

Homomeric glycosyltransferase complex structures representing functionally relevant case studies. Top left ST8SiaIII sialyltransferase homodimer (gray cartoon model) [122] shows how the active sites come relatively close to each other and face the Golgi membrane. N-termini of the crystallized catalytic domains are situated on both ends of the dimer with the first amino acid residue (Trp90) indicated with space filling atoms and completed to a type II membrane protein model with schematic stem- and membrane spanning domains (black line and blue rectangle) included. Positions of the active sites are indicated with a catalytic residue His354. Top right The structure of α1,6-fucosyltransferase FUT8 [124] is shown. The first residues in the crystallized construct, Leu108, come close to each other suggesting a possible assembly of the stem- and membrane spanning domains. Active sites are indicated by catalytic residues Arg365. Bottom left Engineered dual-specificity blood group A and B antigen glycosyltransferase AA(Gly)B homodimer [127] also shows the active sites (represented by Trp181) on the ends of the dimer, but interestingly the N-termini are located on opposite sides of the dimer: the left one (Ser65) towards the viewer and the right one (Val64) on the back (indicated by a short dashed line and an arrow). Bottom right Human β1,3-Glucuronyltransferase I (GlcAT-I [129] and GlcAT-P appear to have their active sites (Gln281) buried deeper in the end of each monomer and N-termini (Met75) located parallel with them, perpendicular to the membrane. Protein Data Bank entries 5BO6, 2DE0, 3ZGF and 1FGG were used, respectively, to draw the figure. It should be noted that N-termini in these figures are those made visible in the structures, and that the stem- and transmembrane regions can be assembled in different ways, and are represented only schematically in the pictures

Full size image

The structure of human α1,6-fucosyltransferase, FUT8 [124], is another glycosyltransferase crystal structure exemplifying a homodimer. The FUT8 catalytic domain resembles the GT-B fold, but has an embedded SH3 domain (a common interaction domain) in the C-terminus, which does not take part in the binding interface, however. The crystallized protein structure contains also a large part of the stem domain, which is disordered in the crystal structure. The beginning of the catalytic domain forms two long α-helices (Fig. 4), which serve as the dimerization interface forming a compact four-helix bundle with a buried area of 3195 Å2. In this structure the catalytic centers of the two FUT8 monomers are also placed on the same side of the dimer, but not facing the membrane, and also on opposite ends of the dimer. Vicinity of the N-terminal ends of the catalytic domains would enable a scenario where the stem and the transmembrane domains contribute to dimer formation.

The other homodimeric complexes of glycosyltransferases present in crystal structures represent variations of the two structures discussed above. There are differences in the interaction surfaces, both in their volume and the structural details that characterize these interactions. Bovine β1,4-galactosyltransferase (GalT-I) [125] appears a dimer with the two catalytic domains making contact through eight hydrogen bonds within a total contact area of 424 Å2 [126]. The interaction area is sufficiently large to represent a functionally relevant interaction. Two structures of an engineered dual-specificity blood group A and B antigen glycosyltransferase AA(Gly)B by [127] were solved using two different crystals with different molecular packing (crystal contacts) and having either one or two molecules per asymmetric unit (Fig. 4). Despite these differences, the homodimer is assembled in either case exactly in the same way, indicating that the assembly represents a functionally relevant dimer possibly formed prior to exposing the protein sample to the crystallization screening. Human β1,3-Glucuronyltransferase I (GlcAT-I, Fig. 4) [128] and GlcAT-P [129] both represent physiologically relevant homodimers with 16 % of the total surface area being buried in the dimer interface (GlcAT-I) and nearly identical residues forming the interface also in the GlcAT-P homodimer. Interestingly, the homodimer assembly is highly different from that of the blood group A and B antigen glycosyltransferase AA(Gly)B complex [127]. While the latter has the active sites on opposite ends of the dimer, in GlcAT-I and GlcAT-P dimers they are next to each other on the same side of the homodimer.

The few cases discussed here show how little is known in structural detail on the formation and nature of glycosyltransferase complexes. From this data, it is apparent that such complexes can form in various ways and there is no one or two models that fit all. It is to be noted that all glycosyltransferase structures represent the catalytic domains with either complete or partial omission of the stem region from the crystallized constructs. Therefore the involvement of the stem region and possibly also the transmembrane domain in taking part in the dimerization cannot be excluded. These observations emphasize that structure determinations of also functionally important heterodimers at atomic detail are necessary for detailed understanding on how the complexes interact and how they co-operatively function in glycan synthesis. It is not clear if the catalytic domains—for which the structures already exist—are sufficient in each case to establish stable homo- or heterodimers. Attempts to resolve these issues are thus necessary and should take into account the following important considerations. Firstly, crystallization conditions may fail to simulate the conditions of the Golgi lumen important for heteromer formation. This is especially important for the pH sensitive interactions of some of the glycosyltransferase heteromers, though they have been shown to resist solubilization but only under proper experimental conditions [41, 86]. Secondly, it is clear that the homomeric and heteromeric complexes are transient and dynamic owing to their constant and microenvironment dependent transitions between less active homomers and more active heteromers during their recycling within the early secretory compartments. This may in fact be behind the difficulties in defining functionally relevant or irrelevant interactions. Thirdly, the trials in which the complex constituents are combined in a test tube does not necessarily guarantee that a complex will form. Rather, they may require conditions that exist only in the living cell. Fourthly, structures or structural models of any glycosyltransferase dimers or higher complexes must fulfill the expectation that the active sites are not buried within the interaction surface, and also that the active sites are situated in such a manner that the binding of donor and acceptor molecules is feasible.

Future directions

A wealth of information now has accumulated and shows that glycosyltransferases in general tend to form functionally relevant complexes in live cells, as in some cases, it has already been shown that the formation of such complexes, especially between sequentially acting enzymes, either augments or diminishes the enzymatic activity of the complex constituents. Yet, the main impact of their existence in most glycosylation pathways in all eukaryotes is that they expose a new level of regulation of glycan synthesis. They also seem to provide the “molecular template” for critical points in glycan synthesis by linking relevant enzymes together and thereby, preventing competing reactions from occurring. Evidence already exist that their absence in diseases such as cancers alters glycosylation of cell surface glycans, a phenomenon that often has fatal consequences on human wellbeing. Due to the individual nature of these complexes, future work should focus on detailed molecular characterization of the complexes, clarification of the interaction surfaces by mutagenesis and on resolving their 3D structures as well as functional consequences on glycan synthesis. This would help understand better how their interactions are regulated by organelle micro-environmental factors (such as pH) in both normal cells and cancerous cells in which such complexes are rare. Clarifying these issues will help uncover important cell and glycobiological questions related to the Golgi apparatus and its’ functioning as a glycosylation device and as a central station for protein delivery to post-Golgi compartments and beyond.