The role of protein glycosylation in Alzheimer disease

Transcription

1 REVIEW ARTICLE The role of protein glycosylation in Alzheimer disease Sophia Schedin-Weiss, Bengt Winblad and Lars O. Tjernberg Karolinska Institutet Alzheimer Disease Research Center (KI-ADRC), Novum, Stockholm, Sweden Keywords b-secretase; Alzheimer disease; amyloid b-peptide; amyloid precursor protein; glycan; glycosylation; nicastrin; tau; c-secretase Correspondence S. Schedin-Weiss, Department of Neurobiology, Care Sciences and Society, Karolinska Institutet Alzheimer Disease Research Center (KI-ADRC), Novum, Stockholm, Sweden Fax: Tel: sophia.schedin.weiss@ki.se (Received 28 August 2013, revised 25 October 2013, accepted 28 October 2013) doi: /febs Glycosylation is one of the most common, and the most complex, forms of post-translational modification of proteins. This review serves to highlight the role of protein glycosylation in Alzheimer disease (AD), a topic that has not been thoroughly investigated, although glycosylation defects have been observed in AD patients. The major pathological hallmarks in AD are neurofibrillary tangles and amyloid plaques. Neurofibrillary tangles are composed of phosphorylated tau, and the plaques are composed of amyloid b-peptide (Ab), which is generated from amyloid precursor protein (APP). Defects in glycosylation of APP, tau and other proteins have been reported in AD. Another interesting observation is that the two proteases required for the generation of amyloid b-peptide (Ab), i.e. c-secretase and b-secretase, also have roles in protein glycosylation. For instance, c-secretase and b-secretase affect the extent of complex N-glycosylation and sialylation of APP, respectively. These processes may be important in AD pathogenesis, as proper intracellular sorting, processing and export of APP are affected by how it is glycosylated. Furthermore, lack of one of the key components of c-secretase, presenilin, leads to defective glycosylation of many additional proteins that are related to AD pathogenesis and/or neuronal function, including nicastrin, reelin, butyrylcholinesterase, cholinesterase, neural cell adhesion molecule, v-atpase, and tyrosine-related kinase B. Improved understanding of the effects of AD on protein glycosylation, and vice versa, may therefore be important for improving the diagnosis and treatment of AD patients. Introduction Alzheimer disease Alzheimer disease (AD) is the most common neurodegenerative disease, affecting ~ 30 million people. Age is the most important risk factor, and as the life span of the population increases, there will be a steep increase in the number of AD cases in the coming years. Presently, only symptomatic treatment is available, but there is an ongoing massive research effort aimed at finding disease-modifying drugs. Neuropathologically, the disease is characterized by extracellular deposits composed of fibrils formed by amyloid b-peptide (Ab), and intraneuronal tangles composed of Abbreviations Ab, amyloid b-peptide; AChE, acetylcholinesterase; AD, Alzheimer disease; APP, amyloid precursor protein; BACE-1, b-site amyloid precursor protein-cleaving enzyme 1; CHO, Chinese hamster ovary; Con A, concanavalin A; CSF, cerebrospinal fluid; dolichyl-pp, dolichylpyrophosphate; Endo H, endoglycosidase H; ER, endoplasmic reticulum; Fuc, fucose; GFAP, glial fibrillary acidic protein; HEK, human embryonic kidney; Man, mannose; MBL, mannan-binding lectin; MEF, mouse embryonic fibroblast; MNJ, mannosidase type 1 inhibitor; NCAM, neural cell adhesion molecule; NeuAc, N-acetylneuraminic acid; OST, oligosaccharyl transferase; PHF, paired helical filament; PS, presenilin; P-tau, hyperphosphorylated tau; sappa, soluble amyloid precursor protein a; sappb, soluble amyloid precursor protein b; ST, sialyltransferase; Tf, transferrin; TM, transmembrane; TPM, tropomyosin; WGA, wheat germ agglutinin. 46 FEBS Journal 281 (2014) ª 2013 FEBS

2 S. Schedin-Weiss et al. Glycosylation and AD hyperphosphorylated forms of the microtubule-associated protein tau. The tangles contain paired helical filament (PHF) structures. Since the discovery of Ab 30 years ago, there has been intense research on this ~ 40-residue peptide, and several lines of evidence suggest that the polymerization of Ab into neurotoxic aggregates is a key event in the pathological cascade that results in AD. Ab is derived from its type 1 transmembrane (TM) precursor, amyloid precursor protein (APP), by the proteolytic action of b-secretase and c- secretase. b-secretase, or BACE-1 (b-site APP-cleaving enzyme 1), mediates the initial cleavage that generates soluble APPb and a C-terminal membrane-bound fragment, C99, which is the immediate substrate for c-secretase. Whereas BACE-1 consists of a single protein, c-secretase is an assembly of at least four different proteins: presenilin (PS)1 or PS2, nicastrin, anterior pharynx-defective 1, and PS enhancer 2. c-secretase cleaves its substrate within the TM region, thereby releasing the APP intracellular domain, which could possibly be involved in transcription. The 48-residue or 49-residue C-terminal stub left in the membrane is thereafter further processed into different Ab variants. The 40-residue variant, Ab40, is the most common product, but it is the two residue longer Ab42 that mediates most of the toxicity [1]. Interestingly, the mutations that lead to familial forms of AD with an early onset (< 65 years of age) are found in APP or PS, and lead to increased Ab production or an increased Ab42/Ab40 ratio [2]. Many of the drugs tested in clinical trials are aimed at lowering the Ab levels in the brain, and the most popular strategies have been to reduce Ab production by inhibiting b-secretase or c-secretase, or to increase Ab clearance with immunological approaches. Most of the trials have shown poor results, and it has been suggested that Ab could be the wrong target, that the drug concentration in the central nervous system is too low, or that the clinical trials should start before the patients show clinical symptoms. In support of the last of these, recent data from a passive vaccination trial showed significant improvement in mild, but not in moderate, AD patients [3]. Thus, there is an immediate need for an early biomarker of AD that could be used for selecting presymptomatic cases for clinical trials, and, when there is a drug on the market, to enable treatment to be started at an early stage of the disease. Furthermore, we need to expand our knowledge of the disease process in order to find complementary or alternative treatment strategies. One line of research that could be more thoroughly investigated is the role of post-translational protein modifications, in particular protein glycosylation. It is estimated that > 50% of all proteins are glycosylated, and glycoproteins usually exist in many glycosylation variants, giving rise to multiple gene products from one gene. Thus, glycans contribute significantly to proteome expansion in higher organisms and are vital for brain functions, including memory and learning [4]. As several studies have suggested that protein glycosylation is altered in AD, this relatively unexplored topic deserves more attention, and could potentially be highly important for the development of improved biomarkers and treatment methods for AD. In the present review, we will summarize the current knowledge on the role of glycosylation in AD. Protein glycosylation We will give a brief summary of protein glycosylation, but, for a comprehensive description of the topic, we refer to textbooks, e.g. Essentials of Glycobiology [5]. Glycans are referred to here as monosaccharides, oligosaccharides, or polysaccharides, either free or bound to glycoproteins. The smallest building blocks of glycans are cyclic forms of monosaccharides. Two stereoisomeric foms, denoted anomeric forms, can result upon cyclization of monosaccharides, generating the isomers a or b (Fig. 1A). In oligosaccharides and polysaccharides, monosaccharide units are held together by glycosidic linkages, i.e. covalent bonds formed between the anomeric carbon of one monosaccharide and an OH-group of another monosaccharide. Glycans have a nonreducing end and a reducing end, which has an anomeric carbon that is not involved in a glycosidic bond (Fig. 1B). The glycosidic bonds can exist in two stereoisomeric forms. These are named according to the position of the anomeric carbon in the nonreducing end monosaccharide followed by the isomeric form and the position of the OH-group to which the anomeric carbon is attached in the reducing end monosaccharide (Fig. 1B). Many monosaccharide units in glycans are five-carbon or six-carbon units, but more complicated monosaccharide units, such as sialic acid, also exist (Fig. 1C). The functional properties of glycans in glycoproteins can be broadly divided into two categories: (a) structural and modulatory functions; and (b) recognition of glycans by other molecules. Glycans thus modulate cell cell, cell matrix and cell molecule interactions, and are involved in many important processes in complex organisms, including the assembly and development of multicellular organs. They act as signals that determine how the glycoproteins are processed within the cell and which compartments they are targeted to. Glycans are also important in immunology, as the FEBS Journal 281 (2014) ª 2013 FEBS 47

3 Glycosylation and AD S. Schedin-Weiss et al. A B C Fig. 1. Examples of sugar structures. (A) Anomeric forms of Glc. In the a-isomer, the OH-group of the anomeric carbon and the CH 2 OH group at C5 point towards different sides of the plane of the ring, whereas in the b-form, they point to the same side. (B) Maltose, which is a disaccharide containing two a-d-glc residues joined by an a-1,4-glycosidic bond. The reducing end is the side that is attached to proteins. (C) The structures of other monosaccharides that are important building blocks in glycans, including a-d-man, b-d-gal, a-l-fuc, a-d-glcnac, and a-d-neuac, the sialic acid form present in humans. The image was drawn with ACCELRYSDRAW 4.1. blood group antigens are determined by glycan epitopes. In infections, glycans mediate pathogen host cell interactions. Specialized forms of intracellular glycans, which are rapidly added and removed, function as molecular switches. However, owing to the complexity of glycans and the resulting challenges in studying them, it is likely that many glycan functions remain to be revealed. Glycans are typically covalently linked to either an asparagine (N-glycans) or serine/ threonine (O-glycans) residue on the glycoproteins. N-glycosylation begins in the endoplasmic reticulum (ER) with the addition of a precursor oligosaccharide, Glc 3 Man 9 GlcNAc 2 (Fig. 2), which is transferred from the lipid dolichyl-pyrophosphate (dolichyl-pp) to the luminal side of a polypeptide chain. This usually occurs on a growing polypeptide during protein translation. The sugar is typically attached to the asparagine within the sequence Asn-X-Thr/Ser, where X can be any amino acid except proline. A TM enzyme complex in the rough ER called the oligosaccharyl transferase complex (OST) catalyzes the transfer of the precursor oligosaccharide to the protein [6]. Human OST is composed of the seven protein subunits ribophorin I, ribophorin II, defender against apoptotic cell death 1, N33/IAP, OST4, STT3, and OST48. STT3 is the catalytic subunit, and exists in two isoforms, A and B, which show somewhat different substrate selectivities. The sugar chain is then processed in the ER lumen by the sequential removal of glucose residues by a-glucosidases I and II. Many glycoproteins are further processed by ER a-mannosidase I, which removes the terminal mannose (Man) from Man 9 GlcNAc 2 to yield Man 8 GlcNAc 2. Thus, most glycoproteins enter the cis-golgi carrying eight or nine Man residues. These structures are termed high-man structures (Fig. 3A). In the cis-golgi, Man 5 GlcNAc 2 is formed by the action of a1,2-mannosidases IA, IB, and 1C. This is the intermediate for the generation of complex (Fig. 3B) and hybrid (Fig. 3C) types of N-glycan, which is initiated in the intermediate Golgi by an N-acetylglucosaminyl transferase; this is followed by processing by Golgi-resistent mannosidases and GlcNAc transferases. The Golgi complex also harbors enzymes to generate N-glycans with more than two branches (Fig. 3D,E) and bisecting GlcNAc, i.e. a GlcNAc attached to the first Man from the core (Fig. 3F). Further elongation of glycans during processing into mature N-glycans occurs mainly in the trans-golgi network by the action of galactosidases and GlcNAc transferases to add GlcNAc and Gal residues. Some glycans are also processed by fucosidases, P P dolichol Fig. 2. Dolichyl-PP oligosaccharide: the structure of the dolichyl- PP-linked oligosaccharide, the precursor in the biosynthyesis of N- glycans. After the glycan has been attached to a polypeptide, it is processed while being transported through the ER Golgi system to generate many different N-glycan structures. Blue square: GlcNAc. Green circle: Man. Blue circle: Glc. The image was drawn with GLYCOWORKBENCH. 48 FEBS Journal 281 (2014) ª 2013 FEBS

4 S. Schedin-Weiss et al. Glycosylation and AD A B C D E F G Fig. 3. Nomenclature of some N-linked glycans. Glycans are broadly categorized into three major groups: (A) high-man glycans, carrying only Man in addition to the core GlcNAc units; (B) complex glycans, which have more than one type of additional monosaccharide unit; and (C) hybrid glycans, carrying a mixture of high-man and hybrid antenna. The glycans shown in (A) (C) are called bi-antennary, as they have two branches. (D, E) Examples of tri-antennary and tetra-antennary glycans. (F) A glycan with a bisecting GlcNAc. (G) A core glycosylated glycan, with a Fuc attached to the innermost GlcNAc. Blue square: GlcNAc. Green circle: Man. Yellow circle: Gal. Purple diamond: NeuAc. Red triangle: Fuc. The image was drawn with GLYCOWORKBENCH. which add a fucose (Fuc) residue in an a1,6-linkage to the GlcNAc adjacent to asparagine in the core (Fig. 3G). The most important capping or decorating reactions, i.e. alteration of the terminal ends of glycans, involve the addition of sialic acid, Fuc, Gal, and GalNAc. Some molecular tools for studying glycoproteins are as follows. Inhibitors of glycosidases/glycosyltransferases: the functions of protein-linked glycans can be studied by treatment of cells or animal models with inhibitors of glycosidases or glycosyltransferases. One commonly used inhibitor is tunicamycin, which blocks the addition of N-linked glycans to glycoproteins by inhibiting the first step in the generation of dolichyl-pp oligosaccharide. Deoxynojirimycin is a glucosidase I inhibitor that prevents the removal of Glc residues, and thereby increases the ER retention time. Mannosidase type 1 inhibitor (MNJ) inhibits the trimming of high-man oligosaccharides, further preventing processing into complex oligosaccharides. Enzymatic removal of glycans: peptide N-glycosidase F is a commonly used enzymatic tool that cleaves N-linked glycans between the innermost GlcNAc and the asparagines of high-man, hybrid and complex oligosaccharides, leaving the entire released glycan intact. Endoglycosidase H (Endo H) cleaves between the two innermost GlcNAc residues, leaving one GlcNAc residue attached to the asparagine for oligomannose and most hybrid types of N-linked glycans, whereas complex glycans are not released. Neuraminidases (also called sialidases) catalyze the hydrolysis of terminal sialic acid residues, leaving the remaining glycan chain without this negatively charged monosaccharide. Lectins: lectins are carbohydrate-binding proteins with high specificity for various sugar structures. They can thus be used for purification, ELISA, fluorescence microscopy, and many other methods. For example: concanavalin A (Con A) binds to a-d-man and a-d-glc residues; wheat germ agglutinin (WGA) binds to GlcNAc and sialic acid; Ricinus cummis agglutinin binds to Gal; and Lens culinaris agglutinin binds to a-d-man. O-glycosylation is more difficult to predict than N-glycosylation, although databases exist that determine the probability of a serine or threonine being O-glycosylated (e.g. NetOGlyc). In contrast to N-glycosylation, O-glycosylation occurs only after protein translation. Many O-linked glycans are linked via a GalNAc moiety to the protein. The first step in this type of O-glycosylation is the transfer of GalNAc from UDP-GalNAc to the amino acid, which is catalyzed by a polypeptide-n-acetyl-galactosaminyltransferase. Many different forms of polypeptide- N-acetyl-galactosaminyltransferase exist. The addition of the next glycan determines the core structure of the FEBS Journal 281 (2014) ª 2013 FEBS 49

5 Glycosylation and AD S. Schedin-Weiss et al. Fig. 4. Core structures in O-glycans. The eight different core structures shown are attached to the OH-groups of Ser/Thr of glycoproteins (and in rare cases to Tyr). These core structures are further substituted and processed to form a huge variety of glycan structures. Yellow square: GalNAc. Blue square: GlcNAc. Yellow circle: Gal. The image was drawn with GLYCOWORKBENCH. O-linked glycan. There are eight core structures (Fig. 4), which may be further substituted by other sugars. Thus, O-glycans differ from N-glycans in not having a common core structure. Many additional enzymes are involved in generating the final O-glycan structures, which are highly variable in size and composition. Some of the enzymes involved are common for N-glycans and O-glycans, and even glycolipids. Another, specialized, form of O-glycosylation is O-GlcNAcylation, which differs from other types of glycosylation by the product not being further processed after the addition of a single GlcNAc to serine or threonine. It also differs from other types of glycosylation by occurring on nuclear and cytoplasmic proteins, and by being more dynamic. O-GlcNAcylation is of importance in signal transduction, and addition of O-GlcNAc residues to a specific site on a protein and their release are regulated by the concerted action of O-GlcNAc transferase and O-GlcNAcase. O-GlcNAc and protein phosphorylation often compete for binding to either the same or proximal attachment sites, and thus interact in their signaling. Apart from the classes described above, there are additional types of O-glycans, e.g. in proteoglycans, that contain glycosaminoglycan chains with up to 200 monosaccharides, but those glycans will not be covered in this review. Sialic acids comprise a group of monosaccharide units that are often found in the terminal position of the oligosaccharides of proteins. They are typically negatively charged at physiological ph, and are important recognition molecules for many cellular functions. The most common sialic acid found in mammalian cells is N-acetylneuraminic acid (NeuAc) (Fig. 1C). The enzymes that catalyze the transfer of sialic acids to glycans, STs, are membrane-bound proteins in the Golgi apparatus. Several forms with different specificities exist, and cleaved forms have been found in body fluids, e.g. serum and milk. Human STs typically add sialic acid to the nonreducing terminal position (Fig. 1B) of glycans in a-2,3-linkage or a-2,6-linkage to a Gal residue, or a-2,6-linkage to GalNAc or Glc- NAc. Sialic acids are also found a-2,8-linked to sialic acid residues in gangliosides and in polysialic acid, which is expressed on, for instance, neural cell adhesion molecule (NCAM). ST6GalI is a member of the human ST family, and transfers sialic acid with an a- 2,6-linkage to a terminal Gal residue of Gal-1-4Glc- NAc disaccharide found as a free disaccharide or as a terminal disaccharide of N-linked or O-linked oligosaccharides. AD and protein sialylation A number of reports indicate that protein sialylation is altered in AD. A significant decrease in soluble sialyltransferase (ST) activity in serum was reported in a study comparing 12 AD patients with 12 age-matched controls [7]. Subsequently, it was reported that ST activity was decreased in membrane and soluble fractions from AD and control postmortem brains [8]. The activity was decreased in both the soluble and membrane fractions of the frontal lobe and temporal lobe, but not in the hippocampus. The decrease in ST activity seemed to be attributable to decreases in both a- 2,3-ST and a-2,6-st activities. Differences in sialylation between AD and healthy individuals have also been indicated from lectin blotting analysis of cerebrospinal fluid (CSF) proteins from AD, probable AD and non-ad patients [9]. Biotinylated forms of WGA, Con A, R. cummis agglutinin and L. culinaris agglutinin were used. Staining with WGA was significantly lower in patients with AD, whereas staining with other lectins was not. As WGA binds to GlcNAc and sialic acid-containing saccharides, whereas Con A binds to GlcNAc but not to sialic acid, the results indicated that the CSF proteins were less sialylated in AD patients than in non-ad patients. Another WGA-binding study confirmed the reduced binding to this lectin of several glycoproteins in CSF from AD patients as compared with healthy individuals, and pointed to transferrin (Tf) as one of these glycoproteins [10]. Owing to the cross-reactivity of lectins and the limited sample size, however, further studies with other methods and larger populations are necessary to evaluate the significance of these observations. An interesting notion in the context of sialylation alterations in AD is that the gene for a sialic acidbinding receptor, cluster of differentiation 33, has been 50 FEBS Journal 281 (2014) ª 2013 FEBS

6 S. Schedin-Weiss et al. Glycosylation and AD found to be associated with late-onset AD in several recent genome-wide association studies [11 14]. Cluster of differentiation 33 is also called Siglec 33, and is a member of the sialic acid-binding immunoglobulin-like lectin superfamily. It is a cell surface immune receptor that binds to extracellular sialylated glycans and signals via its cytoplasmic domain. The receptor has been studied primarily in the peripheral immune system, where it is expressed on myeloid progenitors and monocytes. However, it is also expressed in the brain, where its role is still unknown. The importance of Siglec 33 and sialylated proteins/glycans in protein protein and protein cell interactions for brain functions will therefore be an important future research topic. APP and glycosylation Several studies have shown that N-glycans are functionally important for APP. APP has two potential N-glycosylation sites, at Asn467 and Asn496. Deletion of these residues in an in vitro rabbit reticulate system and in vivo in Chinese hamster ovary (CHO) cells showed that Asn467 was N-glycosylated, whereas N- glycosylation of Asn496 has, to our knowledge, not yet been confirmed. The N-linked glycan was completely resistant to Endo H, suggesting that it was subjected to complex glycosylation [15]. This was confirmed by Sato et al., who showed that APP in CHO cells has bi-antennary and tri-antennary complex-type N-glycans with fucosylated and nonfucosylated trimannosyl cores [16]. In another study in CHO cells, deletion of Asn467 and Asn496 was found to result in decreased secretion and decreased microsomal localization of APP, indicating that the glycans affect the intracellular sorting of the protein [17]. APP in CSF is, indeed, both N-glycosylated and O-glycosylated [18]. The importance of N-glycans in the sorting and secretion of APP has also been shown in other studies. In vivo studies in hamsters showed that treatment with an a-mannosidase inhibitor that specifically blocks the formation of hybrid and complex types of N-glycan (deoxymannojirimycin) reduced the transport of APP and other synaptic glycoproteins to the synaptic membranes [19]. The interference with the formation of complex glycans by treatment with mannosidase inhibitors also leads to decreased secretion of APP [20]. Tienari et al. showed, by deletion of the carbohydrate domain of APP, as well as with tunicamycin treatment in hippocampal neurons, that N-glycans are required for the proper axonal sorting and secretion of APP [21]. An increase in the degree of sialylation of the N-linked glycans of APP has been reported to enhance the secretion of both APP and its metabolites [20,22]. For a schematic illustration, see Fig. 5 (upper panel). The Swedish and London mutations of APP lead to an increased total amount of Ab and an increased Ab42/Ab40 ratio. Both of these mutations also give rise to altered N-glycosylation of APP, with an increased content of bisecting GlcNAc residues (for an explanation of bisecting GlcNAc, see Fig. 3F) [23]. In accordance with this, mrna expression of the responsible enzyme, GlcNAc transferase III, is increased in AD brains [24]. Several O-glycosylation sites have been defined on APP, and O-glycans have also been reported to affect the functions of APP. Perdivara et al. showed that Thr291, Thr292 and Thr576 are O-glycosylated in APP695 expressed in CHO cells (numbering as in fulllength APP695) [25]. Kituzame et al. showed that one additional O-glycan is present in the longer splice variant, APP770, in COS cells [26]. Several additional O-glycosylation sites in APP have recently been identified in human-derived CSF: Ser597, Ser606, Ser611, Thr616, Thr634, Thr635, Ser662, and Ser680 (numbering as in the canonical sequence, APP770, without the signal peptide) [27,28]. The roles of these O-glycans are elusive, although it has been proposed that APP processing by a-secretase, b-secretase and c-secretase occurs after O-glycosylation of APP, and that O-glycosylated APP is preferentially secreted [26,29]. Interestingly, a recent report demonstrated a new type of tyrosine O-glycosylation on short (Ab 1 15 to Ab 1 20 ) but not on full-length (Ab 1 38 to Ab 1 42 )Ab fragments [27]. In a study using CSF from AD patients and nondemented controls, an increase in the short Ab fragments carrying the tyrosine-linked glycan was observed in AD patients. APP is also O-GlcNAcylated [30] and it was recently suggested that O-GlcNAcylation affects APP processing, resulting in increased levels of soluble APPa (sappa) and decreased Ab secretion [31]. It should be noted, however, that other proteins involved in AD pathology, e.g. nicastrin and tau, are also O- GlcNAcylated. Therefore, the effects of treatments that decrease or increase the extent O-GlcNAcylation in cell cultures or animal models may be caused by more than one protein, and not only the proteins that affect Ab. BACE-1 and glycosylation Interestingly, the APP-processing enzyme BACE-1 can affect protein sialylation. In addition to APP, BACE-1 has several other substrates, one of which is ST6Gal1, which is found as an intracellular, membrane-bound form present in the Golgi complex, and as a soluble, FEBS Journal 281 (2014) ª 2013 FEBS 51

7 Glycosylation and AD S. Schedin-Weiss et al. APP APP Tau ER Golgi Vesicle 1 2 APP Tau 3 1 P-tau 4 2 Cytoskeleton 5 PHF 6 APP sappα Aβ sappβ Fig. 5. Schematic overview of the roles of glycosylation in the processing of APP and tau. Correct glycosylation is required for axonal sorting and processing of APP (upper panel). An alteration in N- glycosylation results in protein accumulation within the perinuclear region of the cell (1). Inhibition of the formation of N-glycans or complex N-glycosylation or sialylation of APP interferes with axonal sorting (2) and secretion (3) of APP, as well as the secretion of sappa (4), sappb (5), and Ab (6). The formation of P-tau and PHFs is affected by glycans (lower panel). N-glycans promote the formation of P-tau (1) and PHF (2) and stabilize these structures, whereas O-GlcNAcylation (O-GlcNAc) has the opposite effect, preventing the formation of P-tau (3) and PHFs (4). Hyperphosphorylation of tau and the resulting formation of PHFs disrupts microtubuli (5) and impairs axonal transport (6). secreted form. Several studies have shown that BACE-1 processsing of ST6Gal1 is necessary for the generation of the soluble form of this ST [32 35]. The findings were supported by studies on BACE-1 gene knockout mice, which only had one-third of the level of plasma ST6Gal1 in control mice, as well as BACE-1 transgenic mice, which had increased plasma levels of ST6Gal1 [36]. As most glycosyltransferases show Golgi localization, and many of these are cleaved and secreted from the cell, Kitazume et al. hypothesized that other glycosyltransferases may also be BACE-1 substrates [37]. They focused on a series of STs as candidates for BACE-1 substrates, and found that BACE-1 cleaves the polysialyltransferase ST8Sia IV in vitro. They also found that BACE-1 overexpression in COS cells enhances the secretion of ST3Gal I, ST3Gal II, ST3Gal III, and ST3Gal IV, although cleavage of these could not be detected in vitro. Thus, it is possible that BACE-1 modifies the secretion of some STs via mechanism(s) other than cleavage. One possibility is that BACE-1 activates one or several other proteases responsible for the cleavage and secretion of ST3Gal proteins. Alternatively, BACE-1 could inactivate the machinery for retention of ST3Gal proteins in the Golgi [37]. Although BACE-1 affects the sialylation of secreted glycoproteins, it does not seem to have any effect on cell surface glycoproteins [38]. Sugimoto et al. discussed a possible mechanism for this appearance, whereby soluble ST6Gal1 can move more freely, owing to the loss of its membrane anchor region, which could improve the catalysis of soluble glycoproteins in the trans-golgi network or secretory vesicles [38]. Nakagawa et al. reported that overexpression of ST6Gal1 in Neuro2a cells enhanced a-2,6-sialylation of endogenous APP and increased the extracellular levels of its metabolites (two-fold increase in Ab, threefold increase in soluble APPb (sappb), and 2.5-fold increase in sappa) [22]. Enhanced secretion of sappb was also confirmed in wild-type CHO cells upon ST6Gal1 overexpression. Sialylation-deficient mutant CHO cells secreted half as much Ab as wild-type CHO cells [22]. Thus, there may be a link between sialylation of APP, the metabolic turnover of APP, and AD pathology. PS and glycosylation PS does not have any known glycosylation sites. Instead, several studies have suggested that PS1 affects 52 FEBS Journal 281 (2014) ª 2013 FEBS

8 S. Schedin-Weiss et al. Glycosylation and AD N-glycosylation of proteins during processing in the ER Golgi system. One such protein is nicastrin, which will be described in more detail below; others will be described here. One study showed that overexpression of wild-type PS1 or PS1 with a familial AD mutation (M146L) in the neuroblastoma cell line SH-SY5Y resulted in decreased sialylation of NCAM [39]. This cell surface protein plays a key role in brain function, and has been implicated in cell cell adhesion, neurite outgrowth, synaptic plasticity, learning, and memory. NCAM can be glycosylated by the addition of polysialic acid, a large oligosaccharide composed of multimeric chains of sialic acid residues joined in an a-2,8 linkage, which is attached to the penultimate Gal of the core oligosaccharide by an a-2,3 linkage. Overexpression of the PS1 variants reduced the cell surface expression of a-2,3-sialoglycoproteins. A change in NCAM molecular size was in agreement with the loss of sialic acid, and immunocytochemistry suggested that the subcellular location of NCAM was altered by PS1 overexpression, as the staining was diffuse and intracellular rather than membranous. These results suggest that overexpression of either wild-type or mutant PS1 disturbs glycoprotein processing within the Golgi. The glycosylation, maturation and subcellular location of tyrosine-related kinase B has been reported to be defective in PS1 / mouse primary cortical neurons [40]. The receptor mediates the effects of neurotrophins, e.g. brain-derived neurotrophic factor, including neuronal differentiation and survival. The effect of PS1 on N-glycosylation was also observed in another study examining the effects of PS1 deletions or mutations on lysosomal proteolysis and autophagy [41]. Autophagy, the major pathway for lysosomal degradation in cells, is defective in AD [42]. Lee et al. have shown that autophagosomal functions, including autolysosome acidification and cathepsin activation, require PS1, and that transfection of PS1/PS2-null murine blastocysts with human PS1 restores those functions [41]. Several pieces of evidence suggest that the deficits observed in PS1-null mice are caused by failed PS1- dependent trafficking of the v-atpase V0a1 subunit to the lysosomes. Coimmunoprecipitation verified that PS is associated with components of the OST (Sec61a), and the authors proposed a model in which N-glycosylation of the V0a1 subunit is essential for the translocation of v-atpase from the ER to the lysosomes, which, in turn, requires that PS1 interacts with this complex. However, this model could not be supported by Zhang et al. [43], who were unable to observe alterations in the N-glycosylation of V0a1 in mouse embryonic fibroblasts (MEFs) deficient in PS1 and PS2. They were also unable to find evidence that the turnover of autophagic substrates, vesicle ph or lysosome function was altered in cells lacking PS1 and PS2, and these issues therefore need to be further examined. PS1 deficiency also alters the subcellular distribution and turnover of telencephalin, APP, and APP-like protein 1 [40,44]. APP-like protein 1 is, like APP, cleaved by the secretases, and is a regulator of insulin and glucose homeostasis. Telencephalin is also called ICAM5, and belongs to the intercellular adhesion molecule subfamily. Telencephalin is a type 1 TM glycoprotein that promotes dendritic outgrowth and contributes to longterm potentiation, a long-lasting enhancement in signal transmission between neurons that is believed to be a correlate for learning and memory. Notably, in vitro and in vivo studies employing a yeast two-hybrid system as well as coimmunoprecipitation showed that PS1 and PS2 interact with telencephalin [44]. Altogether, these data indicate that PS1 does affect the N-glycosylation of several proteins. This effect may be obtained by PS1 affecting the subcellular location of the proteins, which, in turn, affects the N-glycosylation process. Alternatively, PS1 may have a direct effect on N-glycosylation, which, in turn, affects the subcellular location of the proteins. These issues need to be elucidated in future investigations. Nicastrin and glycosylation Nicastrin has as many as 16 potential N-glycosylation sites, but the functions of the N-linked glycans on nicastrin are not yet clear. The presence of N-glycans on nicastrin was confirmed at the same time as it was found to be a component of c-secretase [45]. Western blotting of human embryonic kidney (HEK)293 cells expressing V5-tagged nicastrin gave a band with a molecular mass of ~ 110 kda that was reduced to ~ 80 kda after treatment with Endo H, suggesting that N-glycans could be removed by glycosidase treatment. Several studies of nicastrin expressed in cell lines have shown two nicastrin bands in western blots. The lower molecular mass band ( kda, depending on the cell type and construction of the expressed protein) is called immature nicastrin, and carries N-glycans that have not been subjected to complex glycosylation, whereas the higher molecular mass band ( kda), which carries complex N-linked glycans, is called mature nicastrin. Studies on several cell types have shown that mature nicastrin contains a mixture of high-man, hybrid and complex N-glycans. Mature nicastrin is partially sensitive to Endo H treatment, whereas immature nicastrin is fully sensitive to the same treatment [46,47], in agreement with the presence of only high-man glycans on immature nicastrin. FEBS Journal 281 (2014) ª 2013 FEBS 53

9 Glycosylation and AD S. Schedin-Weiss et al. In primary neurons, it has been suggested that only the mature form is present, and neuraminidase treatment and lectin-binding studies have shown that neuronal nicastrin contains sialic acid [47]. Also in human and rat brain, the mature form is mainly found [48,49], although differences in the proportions of mature and immature nicastrin have been reported during rat brain development [50]. Several studies have shown that complex glycosylation of nicastrin is dependent on PS. Two different studies have reported that nicastrin produced in MEFs derived from PS1 / and PS2 / mouse embryos fails to become mature nicastrin [47,51]. Strong downregulation of the level of mature nicastrin has also been observed in cells that lack only PS1 [47,52], whereas only a small decrease in the amount of mature nicastrin was observed in cells that lacked only PS2 [47]. Coimmunoprecipitation studies with stable Neuro2a cells, HEK293 cells and SHSY-5Y cells have shown that PS has a strong preference for the mature over the immature form of nicastrin [46,52,53]. Furthermore, the mature form has a longer half-life than the immature form [53]. As N-glycosylation of proteins begins in the ER with the production of high-man N-glycans, which are subsequently processed into complex glycans in the Golgi compartment, it has been suggested that PS is required for nicastrin trafficking through the secretory pathway [53]. In support of this hypothesis is the observation that nicastrin produced in fibroblasts deficient in PS1 and PS2 accumulates in the ER and fails to reach both the medial Golgi compartment and the cell surface [51]. Several nicastrin mutants with loss-of-function deletions fail to become resistant to both Endo H and trypsin treatment. As mature nicastrin is partly resistant to trypsin treatment, these findings indicate that a conformational change of nicastrin, which is believed to occur normally during assembly of the c-secretase complex, does not occur in these variants [54]. Tomita et al. studied nicastrin mutants expressed in a cellular system based on a stable Neuro2a cell line [53], and found that nicastrin deletion mutants lacking a large conserved region of the ectodomain or C-terminal (cytoplasmic) residues did not produce any mature nicastrin. A missense mutation replacing the conserved Asp336 and Tyr337 led to a reduced amount of mature nicastrin [53]. PS1 with mutations in the catalytic site residues Asp257 and Asp385 expressed in MEFs did not show any differences with respect to migration of nicastrin in SDS/PAGE, showing that the catalytic site residues of PS are not involved in the PS-mediated complex glycosylation of nicastrin [55]. Site-directed mutagenesis of three functionally important residues in nicastrin, C248S, E333Q, and G339A, in HEK/APP cells led to reduced interaction with PS1CTF, PS2CTF, and presenilin enhancer 2, as well as reduced formation of complex glycans and reduced trypsin resistance [56]. Inhibition of complex glycosylation by kifunensine or MNJ, which prevent processing to complex oligosaccharides, did not affect trypsin resistance or Ab production [47,56]. Surprisingly, however, under such conditions, immature nicastrin coimmunoprecipitated with PS1, indicating that complex oligosaccharides are not required for the binding of nicastrin to PS [47]. Production of Ab and Notch intracellular domain in MNJ-treated fibroblasts was largely unaffected. In conclusion, association with PS appears to be necessary for nicastrin to leave the ER and to be subjected to complex glycosylation. Altogether, these studies indicate that nicastrin needs to associate with PS in order to be normally N-glycosylated, and, although normal glycosylation of nicastrin may not be required to generate active c-secretase, the properties of the glycans attached to nicastrin may affect the subcellular location and/or substrate selectivity of c-secretase. It remains to be determined whether it is the presence of PS or the presence of complex glycans formed only in the presence of PS that is necessary for targeting nicastrin to the correct cellular location. In addition to N-glycosylation, nicastrin has recently been reported to be O-GlcNAcylated [57]. The O-Glc- NAcylation site was reported to be at Ser708, i.e. at the C-terminal cytosolic side, and inhibition of O- GlcNAcase reduced c-secretase activity, both in a cellular system and in a mouse model with five familial AD mutations (overexpressing human APP695 with three mutations and PS1 with two mutations, all of which increase the formation of Ab42) [57]. Inhibition of O-GlcNAcase also reduced the amount of Ab plaques and improved memory impairment in the mice. It should be noted, however, that some other proteins involved in AD pathogenesis are O-GlcNAcylated, e.g. APP and tau (see below). Thus, it is difficult to ascertain whether the effects of treatments with O-GlcNAcase inhibitors in cell culture are attributable to effects on specific proteins. Tau and glycosylation N-glycosylation of proteins generally occurs only on extracellular proteins or the extracellular domains of membrane-bound proteins. It is therefore intriguing that tau, which is a cytosolic protein, has been found to be N-glycosylated in AD but not in control brain. Tau has three potential N-glycosylation sites, and the 54 FEBS Journal 281 (2014) ª 2013 FEBS

10 S. Schedin-Weiss et al. Glycosylation and AD presence of N-linked glycans on tau has been confirmed by a combination of lectin staining, monosaccharide composition analysis, and N-glycosidase F treatments [58 60]. High-Man and truncated structures, as well as complex, sialylated N-glycans, have been identified on hyperphosphorylated tau (P-tau) and PHF-tau, and the proportion of truncated glycans was higher in PHF-tau than in P-tau [59]. Although potential explanations have been suggested, it is not clear why tau is glycosylated in AD but not in control brain [59]. One explanation could be that the altered subcellular location of tau in AD results in altered accessibility to N-glycosylation enzymes. A second possibility is that OST activities are increased in dystrophic neurons in AD brain. A third explanation could be that the activity of a so far unidentified cytosolic N-glycosidase, which normally removes N-glycans, is downregulated in AD brain. In addition to affecting the formation of P-tau and PHFs, N-linked glycans also appear to be involved in the maintenance of PHFs [58], as N-glycosidase F treatment destroys these structures. The aberrant N-glycosylation of tau was found to make tau more susceptible to phosphorylation and less susceptible to dephosphorylation, and may thus precede the hyperphosphorylation of tau [60,61]. Tau is also multiply O-GlcNAcylated [62], and the level of O-GlcNAcylation of tau is decreased in AD brain as compared with control brain [63]. In contrast to N-glycosylation, O-GlcNAcylation appears to protect against aberrant phosphorylation in AD [64,65]. O-GlcNAcylation is dependent on the Glc concentration, which is typically decreased in AD patients. As O-GlcNAcylation and phosphorylation occur reciprocally, it is possible that a decrease in O-GlcNAcylation precedes the hyperphosphorylation of tau in AD brain [66,67]. Although many O-GlcNAcylation sites are believed to exist in tau, only a few have been experimentally identified [68 70]. For a schematic illustration, see Fig. 5 (lower panel). Tf and glycosylation Tf is a monomeric ~ 80-kDa glycoprotein that binds to and transports iron in blood. It consists of 679 amino acids folded into two domains, each of which contains a metal ion-binding site. Some studies have indicated that there is a correlation between a genetic variant of Tf, carrying a P570S mutation called TfC2, and AD [71 73]. Tf is N-glycosylated at Asn413 and Asn611, and, owing to heterogeneity in the glycan structures, several glycosylation variants exist. The most common form in normal blood plasma is biantennary, in which each antenna is capped with two sialic acid residues (tetrasialo-tf). It has been found that the degree of sialylation of Tf is increased in blood plasma from AD patients as compared with healthy individuals, with a relative increase in pentasialo-tf and hexasialo-tf [74,75]. However, there was a significant reduction in pentasialo-tf and hexasialo-tf in blood plasma from TfC1/TfC2 heterozygous patients as compared with TfC1 homozygous AD patients [75]. In contrast, the degree of sialylation of Tf is decreased in CSF. One report on CSF blots probed with the lectin WGA from AD patients, nondemented AD patients and nondemented controls revealed a lower degree of WGA-binding capacity in the CSF from AD patients, despite the fact that the levels of Tf protein were unaltered [10]. As WGA binds to sialic acid and GlcNAc, these data indicate altered glycosylation. Futakawa et al. [76] identified two Tf variants in CSF with different SDS/PAGE migration properties. One was determined to be bi-antennary asialo, agalacto complex-type N-glycan that carries bisecting b-1,4-glcnac and core a-1,6-fuc (Tf1), and one was determined to carry a-2,6-sialyl glycans such as serum Tf (Tf2). The Tf2/Tf1 ratio was not increased in AD patients, contrasting with the results of Taniguchi [10], who suggested that Tf detected with WGA lectin is a biomarker for AD. The differences could be attributable to the difference in the methods used, as Futakawa used a Tf antibody instead of lectin to detect the Tf1 band. Notably, CSF contains more than two different glycosylation variants [10], suggesting that Tf1 and Tf2 bands in SDS/ PAGE each contain several glycosylation variants. In a later study, ELISAs employing lectins specific for Tf1 and Tf2, respectively, were developed for detection of the two Tf variants. Such ELISA assays may be used to investigate the possibility of using the Tf2/Tf1 ratio as a biomarker for AD in future studies [77]. Other proteins of relevance to AD In addition to the proteins described above, alterations in the glycosylation of other proteins have been reported in AD. Acetylcholinesterase (AChE), also known as acetylhydrolase, is a serine protease that degrades the neurotransmitter acetylcholine and terminates synaptic transmission at cholinergic brain synapses. The protease has three potential N-glycosylation sites, and can exist as different isoforms, of which the tetramers are considered to be the major form, and believed to exert the AChE activity. The role of AChE in AD has recently been reviewed [78]. The monomer form is increased in AD, and, on the basis of lectin FEBS Journal 281 (2014) ª 2013 FEBS 55

11 Glycosylation and AD S. Schedin-Weiss et al. binding, altered glycosylation of the protein has been reported, with an increase in the proportion of a form that does not bind to Con A [79,80]. This alteration in glycosylation of AChE has been reported both in brain tissue (frontal cortex) and in CSF in AD patients [81,82]. Likewise, the related compound butyrylcholinesterase is differently glycosylated in AD, and both of these cholinesterases have been suggested to be biomarkers for AD [83,84]. The changes in glycosylation of these proteins, however, occur relatively late in the course of disease progression, and are therefore not likely to be early biomarkers [85]. In fact, it has been suggested that both Ab and tau induce the altered glycosylation of the cholinesterases [78]. It has been reported that PS1 interacts with AChE, and affects the enzymatic activity and glycosylation of the enzyme [86]. Reelin is another protein involved in regulating synaptic function and plasticity that has altered glycosylation properties in AD. The glycosylation of reelin has been investigated with lectin-binding studies, which have shown a reduced extent of binding to Con A in CSF from AD patients [87]. Similarly, the fraction of reelin that does not bind to Con A is decreased in the frontal cortex from AD patients, and, as for AChE, Ab appears to alter the glycosylation of reelin [88]. Mannan-binding lectin (MBL) binds in a Ca 2+ - dependent manner to carbohydrates, with specificity for Man-containing or GlcNAc-containing structures. This lectin can activate the classical complement pathway, and proteins of this pathway are found in the brains of AD patients. The potential involvement of MBL in AD was studied by immunohistochemistry in the brains of AD patients, and the levels of MBL were determined in CSF and serum from AD patients [89]. The study showed that MBL is associated with blood vessels in both AD patients and controls, and a lowered level of MBL in CSF was observed in AD patients. A recent study further showed that MBL binds to Ab, and, as this interaction affected the inflammatory response with peritoneal macrophages, the authors proposed that MBL may affect AD pathogenesis by playing a role in Ab clearance [90]. Glycoproteomics and AD One way to study glycoproteomics in AD is by IEF/ 2D gel electrophoresis combined with identification of the protein peptides digested from the spots by mass spectrometry (MS). In this approach, the samples and controls are differently labeled with fluorescent dyes before separation, and spots where the intensity differs can be punched out from the gels and analyzed by MS. Such an approach has been used to study alterations in the expression and glycosylation levels of glycoproteins in serum [91] and CSF [92,93]. Alterations in CSF glycoproteins can reflect the ongoing disease progress in the brain, and the CSF studies revealed that both the expression and the glycosylation of a 1 - antitrypsin are decreased in AD patients. Glycoproteomics studies in CSF are, however, problematic, owing to the very high levels of proteins such as albumin that tend to make analysis of other, low-abundance, proteins difficult. Although depletion methods are possible, these methods may introduce other experimental errors, e.g. loss of other proteins. Kanninen et al. [94] used 2D gel electrophoresis to study the degree of glycosylation of cytosolic proteins in the frontal cortex of brains from AD patients and nondemented controls. Relative protein amounts and the degree of glycosylation were detected with protein staining and glycoprotein staining, respectively, and the glycoproteins were identified with MS. As N-glycosylation is generally considered to be restricted to the extracellular portion of membrane-bound proteins and secreted proteins, the cytosolic proteins are presumably O-glycosylated. The major finding reported from that study was a reduction in the glycosylation of collapsin response mediator protein 2, a protein that regulates the assembly and polymerization of microtubules, and is associated with neurofibrillary tangles in AD. In contrast, the degree of glycosylation of glial fibrillary acidic protein (GFAP) and intermediate filament protein (which is expressed in, for instance, astrocytes, and is believed to be involved in the maintenance of astrocytic mechanical strength) was increased in AD. Another method that has been used to study alterations of glycoproteins in AD is enrichment of glycoproteins with lectin chromatography followed by protein identification with MS [95]. WGA and Con A were used to study glycoproteomics in the hippocampus and the inferior parietal lobe in the brains of AD patients, patients with mild cognitive impairment, and nondemented controls [96,97]. Alterations in lectin affinity for proteins involved in, for instance, Glc metabolism (a-enolase, c-enolase, and glutamate dehydrogenase), chaperone functions (Glc-regulated protein 78, heat shock protein 90, protein disulfide isomerase, and Glc-regulated protein 96), cytoskeletal maintenance [GFAP, tropomyosin (TPM)1, TPM2, TPM3, c,e,f, gelsolin, and calreticulin], synaptic function (dihydropyrimidase 2, Rab GDP dissociation inhibitor XAP4, and b-synuclein) and cell signaling (protein phosphatase-related protein SDS22), and calmodulin) were identified. Similarly, a recent proteomics study on dissected hippocampal neurons revealed alterations in proteins involved in, for instance, tran- 56 FEBS Journal 281 (2014) ª 2013 FEBS