O-linked glycosylation ensures the normal conformation of the autotransporter adhesin involved in diffuse adherence

The Escherichia coli adhesin involved in diffuse adherence (AIDA-I) is one of the few glycosylated proteins found in Escherichia coli. Glycosylation is mediated by a specific heptosyltransferase encoded by the aah gene, but little is known about the role of this modification and the mechanism involved. In this study, we identified several peptides of AIDA-I modified by the addition of heptoses by use of mass spectrometry and N-terminal sequencing of proteolytic fragments of AIDA-I. One threonine and 15 serine residues were identified as bearing heptoses, thus demonstrating for the first time that AIDA-I is O-glycosylated. We observed that unglycosylated AIDA-I is expressed in smaller amounts than its glycosylated counterpart and shows extensive signs of degradation upon heat extraction. We also observed that unglycosylated AIDA-I is more sensitive to proteases and induces important extracytoplasmic stress. Lastly, as was previously shown, we noted that glycosylation is required for AIDA-I to mediate adhesion to cultured epithelial cells, but purified mature AIDA-I fused to GST was found to bind in vitro to cells whether or not it was glycosylated. Taken together, our results suggest that glycosylation is required to ensure a normal conformation of AIDA-I and may be only indirectly necessary for its cell-binding function.     

Effect of glycosylation on the extracellular domain of the Ag43 bacterial autotransporter: enhanced stability and reduced cellular aggregation

Autotransporters constitute the biggest group of secreted proteins in Gram-negative bacteria and contain a membrane-bound beta-domain and a passenger domain secreted to the extracellular environment via an unusually long N-terminal sequence. Several passenger domains are known to be glycosylated by cytosolic glycosyl transferases, promoting bacterial attachment to mammalian cells. In the present study we describe the effect of glycosylation on the extracellular passenger domain of the Escherichia coli autotransporter Ag43alpha, which induces frizzy colony morphology and cell settling. We identify 16 glycosylation sites and suggest two possible glycosylation motifs for serine and threonine residues. Glycosylation stabilizes against thermal and chemical denaturation and increases refolding kinetics. Unexpectedly, glycosylation also reduces the stabilizing effect of Ca(2+) ions, removes the ability of Ca(2+) to promote cell adhesion, reduces the ability of Ag43alpha-containing cells to form bacterial amyloid and increases the susceptibility of the resulting amyloid to proteolysis. In addition, our results indicate that Ag43alpha folds without a stable intermediate, unlike pertactin, indicating that autotransporters may arrive at the native state by a variety of different mechanisms despite a common overall structure. A small but significant fraction of Ag43alpha can survive intact in the periplasm if expressed without the beta-domain, suggesting that it is able to adopt a protease-resistant structure prior to translocation across the membrane. The present study demonstrates that glycosylation may play significant roles in structural and functional properties of bacterial autotransporters at many different levels.     

Glycosylation of the self-recognizing Escherichia coli Ag43 autotransporter protein

Glycosylation is a common modulation of protein function in eukaryotes and is biologically important. However, in bacteria protein glycosylation is rare, and relatively few bacterial glycoproteins are known. In Escherichia coli only two glycoproteins have been described to date. Here we introduce a novel member to this exclusive group, namely, antigen 43 (Ag43), a self-recognizing autotransporter protein. By mass spectrometry Ag43 was demonstrated to be glycosylated by addition of heptose residues at several positions in the passenger domain. Glycosylation of Ag43 by the action of the Aah and TibC glycosyltransferases was observed in laboratory strains. Importantly, Ag43 was also found to be glycosylated in a wild-type strain, suggesting that Ag43-glycosylation may be a widespread phenomenon. Glycosylation of Ag43 does not seem to interfere with its self-associating properties. However, the glycosylated form of Ag43 enhances bacterial binding to human cell lines, whereas the nonglycosylated version of Ag43 does not to confer this property.     

Theoretical and experimental characterization of the scope of protein O-glycosylation in Bacteroides fragilis

Among bacterial species demonstrated to have protein O-glycosylation systems, that of Bacteroides fragilis and related species is unique in that extracytoplasmic proteins are glycosylated at serine or threonine residues within the specific three-amino acid motif D(S/T)(A/I/L/M/T/V). This feature allows for computational analysis of the proteome to identify candidate glycoproteins. With the criteria of a signal peptidase I or II cleavage site or a predicted transmembrane-spanning region and the presence of at least one glycosylation motif, we identified 1021 candidate glycoproteins of B. fragilis. In addition to the eight glycoproteins identified previously, we confirmed that another 12 candidate glycoproteins are in fact glycosylated. These included four glycoproteins that are predicted to localize to the inner membrane, a compartment not previously shown to include glycosylated proteins. In addition, we show that four proteins involved in cell division and chromosomal segregation, two of which are encoded by candidate essential genes, are glycosylated. To date, we have not identified any extracytoplasmic proteins containing a glycosylation motif that are not glycosylated. Therefore, based on the list of 1021 candidate glycoproteins, it is likely that hundreds of proteins, comprising more than half of the extracytoplasmic proteins of B. fragilis, are glycosylated. Site-directed mutagenesis of several glycoproteins demonstrated that all are glycosylated at the identified glycosylation motif. By engineering glycosylation motifs into a naturally unglycosylated protein, we are able to bring about site-specific glycosylation at the engineered sites, suggesting that this glycosylation system may have applications for glycoengineering.     

Glycosylation affects the rate of traffic of the Shaker potassium channel through the secretory pathway

We have examined the effect of glycosylation on the traffic of the voltage-gated Shaker potassium channel through the secretory pathway of mammalian cells. Shaker is glycosylated on two asparagines (N259 and N263) in the first extracellular loop. Electrophysiological experiments indicate that glycosylation is not necessary for channel integrity [Santacruz-Toloza et al. (1994) Biochemistry 33, 5607]. Consistent with this, we observe that unglycosylated N259Q+N263Q mutant channel forms oligomers as efficiently as the wild type and that this occurs in the endoplasmic reticulum. We have compared the kinetics of secretory traffic of the wild-type glycosylated and the N259Q+N263Q unglycosylated channels. Surface biotinylation of newly synthesized proteins indicates that the rate of delivery of the unglycosylated channel to the cell surface is slower than that of wild type. We have further dissected channel traffic using quantitative imaging. We observe that mutant channel traffics more slowly from the endoplasmic reticulum to the Golgi than wild type at 20 degrees C. This may contribute to the slowed delivery of the mutant to the cell surface. Neither the surface fraction at steady state nor the stability of Shaker is significantly affected by glycosylation in COS cells.     

N- and O-glycans are not directly involved in the oligomerization and apical sorting of GPI proteins

Oligomerization of glycosyl-phosphatidylinositol-anchored proteins (GPI-APs) into high molecular weight complexes is an essential step for their apical sorting in polarized epithelial cells. However, the mechanism by which apical GPI-APs oligomerize is still unclear. To investigate the possible role of N- and O-glycosylation, we have analysed the behaviour of two glycosylated GPI-anchored apical proteins, p75GPI and placental alkaline phosphatase (PLAP), and their glycosylation mutants. We found that both the N- and O-glycosylation mutants are apically sorted, associate to detergent-resistant microdomains and are able to oligomerize, like the wild-type proteins, suggesting that glycosylation does not have a direct role in GPI-AP oligomerization and apical sorting. Interestingly, when cells are depleted of cholesterol and treated with tunicamycin, treatments that by themselves do not affect PLAP sorting, PLAP is not able to oligomerize and is missorted to the basolateral surface, thus supporting an indirect role of N-glycosylation, possibly mediated by a raft-associated glycosylated interactor.     

Glycosylation of a membrane protein is restricted to the growing polypeptide chain but is not necessary for insertion as a transmembrane protein.

The membrane glycoprotein of vesicular stomatitis virus (VSV), synthesized in vitro in the presence of pancreatic microsomes, is glycosylated in two distinct steps while its polypeptide chain is nascent (Rothman and Lodish, 1977). We show here that unglycosylated glycoprotein, which accumulates in vivo following treatment of cells with tunicamycin and in vitro as a result of translation in the presence of detergent-treated microsomal membranes, is inserted normally as a transmembrane protein. This means that glycosylation, while normally occurring concurrently with insertion, is not required for insertion. Our experiments also show that the two steps in glycosylation correspond to the sequential transfer of preformed “core” oligosaccharides of typical structure to two Asn residues in the growing chain. The accumulation of unglycosylated glycoprotein in vitro is due to the fact that the completed transmembrane polypeptide cannot be glycosylated. The detergent treatment of microsomes impairs their rate of glycosylation so that chains are frequently completed before they can be glycosylated. This provides a simple explanation for certain types of heterogeneity often found in glycoproteins. We believe that the detergent treatment procedure results in the solubilization of the microsomal membrane followed by reconstitution. This is a prerequisite for the eventual purification of the membrane proteins and lipids involved in insertion and glycosylation of this model membrane protein.     

Laminin carbohydrates are implicated in cell signaling

We have examined how laminin carbohydrates participate in cellular responses and have focused upon cell spreading and neurite outgrowth. Our earlier studies showed that unglycosylated laminin fully supported cell adhesion but did not promote subsequent spreading of mouse melanoma cells or neurite outgrowth of rat pheochromocytoma cells (Dean et al. (1990): J Biol Chem 265:12553-12562). In the present experiments, we determined whether those cellular responses could be restored to adherent cells. When a mixture of unglycosylated and glycosylated laminins was used as a substratum for mouse melanoma cells, some cells began to spread when 30% glycosylated laminin was present. At least 65% glycosylated laminin was required to elicit a maximal spreading response by the majority of the cells. In separate experiments, we found that cell spreading was fully restored by a pronase digest of glycosylated laminin; a similar digest of unglycosylated laminin had no effect. These results indicate that laminin carbohydrates, rather than polypeptide sequences, were responsible for cell spreading. We also conclude that substrate attachment of the carbohydrate moieties was not essential. In other experiments, laminins containing immature oligosaccharides were produced using two glycosylation pathway inhibitors, swainsonine or castanospermine. When such laminins were used to study cell spreading or neurite outgrowth, laminin containing immature oligosaccharides was as effective as laminin which contains fully processed oligosaccharides. In contrast, laminin with partially processed oligosaccharides had incomplete activity. These composite reconstitution experiments show that laminin carbohydrates provide essential information to responsive cells, enabling them to progress from an adherent state to a spread form or to extend neurite processes.     

Presentation of MUC1 tumor antigen by class I MHC and CTL function correlate with the glycosylation state of the protein taken Up by dendritic cells.

We previously reported that the glycosylated MUC1 tumor antigen circulating as soluble protein in patients' serum is not processed by dendritic cells and does not elicit MHC-Class II-restricted T helper responses in vitro. In contrast, a long synthetic peptide from the MUC1 tandem repeat region is presented by Class II molecules, resulting in the initiation of T helper cell responses. Here we addressed the ability of dendritic cells to present various glycosylated or not glycosylated forms of MUC1 by MHC Class I. We found that three different forms of MUC1, ranging from glycosylated and underglycosylated protein to unglycosylated synthetic peptide, were able to elicit MUC1-specific, Class-I-restricted CTL responses. The efficiency of processing and the resulting strength of CTL activity were inversely correlated with the degree of glycosylation of the antigen. Furthermore, the more efficiently processed 100mer peptide primed a broader repertoire of CTL than the glycosylated protein.     

N-linked glycosylation affects the processing of mouse submaxillary gland prorenin in transfected AtT20 cells

Most mouse inbred strains carry two renin genes, Ren-1 and Ren-2, Renin-2, the product of the Ren-2 gene, is highly expressed in the submaxillary gland. It is a renin isoenzyme 96% similar to kidney renin-1, but unglycosylated. In order to investigate if glycosylation of prorenin affects its processing and/or secretion we have introduced two potential N-linked glycosylation sites into preprorenin-2 cDNA using site-directed mutagenesis. Expression plasmids were derived from wild-type and mutant renin-2 cDNA and were transfected into AtT20 cells. Both transfected cells, expressing glycosylated or unglycosylated forms, secreted prorenin and renin by the constitutive and regulated pathways, respectively. Prorenin was correctly processed to active renin but the second maturation site was not cleaved in AtT20 cells. The comparison of glycosylated and unglycosylated renin expression showed a diminished secretion of glycosylated active renin. Prevention of glycosylation with tunicamycin resulted in an improved secretion of active renin. Moreover, the efficiency of the trypsin activation in vitro was reduced for glycosylated prorenin and it was restored when the activation was performed on mutant renin secreted from tunicamycin-treated cells. It is proposed that the bulky carbohydrates attached to prorenin constitute a steric hindrance to proteolysis by maturation enzymes.     

Calreticulin associates with stress proteins: Implications for chaperone function during heat stress

Acute heat stress leads to the glycosylation of a “prompt” stress glycoprotein, P-SG67/64, identified as calreticulin. In the present study, we used immunoprecipitation to investigate the interactions of P-SG/calreticulin with other proteins during cellular recovery from heat stress. In heat-stressed CHO and M21 cells, both glycosylated and unglycosylated P-SGs interact with HSP90, GRP94, GRP78, and the other prompt stress glycoprotein, P-SG50, in an ATP-independent manner. Specificity of HSP-P-SG interactions was determined by chemical cross-linking with the homo-bifunctional agent DSP (3,3′-dithiobis[succinimidyl propionate]). Characterization of the cross-linked complexes involving calreticulin and heat shock proteins (HSPs) showed an average mass of 400–600 kDa by gel filtration chromatography. Overall, the consistent association of glycosylated and unglycosylated calreticulin with P-SG50 and unglycosylated HSPs suggests that P-SG/calreticulin is an active member of the cast of glycone/aglycone chaperones that cooperate to achieve cellular recovery from acute heat stress. J. Cell. Biochem. 69:30–43, 1998. © 1998 Wiley-Liss, Inc.     

N-Linked glycosylation of antibody fragments in Escherichia coli

Glycosylation is the predominant protein modification to diversify the functionality of proteins. In particular, N-linked protein glycosylation can increase the biophysical and pharmacokinetic properties of therapeutic proteins. However, the major challenges in studying the consequences of protein glycosylation on a molecular level are caused by glycan heterogeneities of currently used eukaryotic expression systems, but the discovery of the N-linked protein glycosylation system in the ε-proteobacterium Campylobacter jejuni and its functional transfer to Escherichia coli opened up the possibility to produce glycoproteins in bacteria. Toward this goal, we elucidated whether antibody fragments, a potential class of therapeutic proteins, are amenable to bacterial N-linked glycosylation, thereby improving their biophysical properties. We describe a new strategy for glycoengineering and production of quantitative amounts of glycosylated scFv 3D5 at high purity. The analysis revealed the presence of a homogeneous N-glycan that significantly increased the stability and the solubility of the 3D5 antibody fragment. The process of bacterial N-linked glycosylation offers the possibility to specifically address and alter the biophysical properties of proteins.     

Importance of glycosylation on function of a potassium channel in neuroblastoma cells

The Kv3.1 glycoprotein, a voltage-gated potassium channel, is expressed throughout the central nervous system. The role of N-glycans attached to the Kv3.1 glycoprotein on conducting and non-conducting functions of the Kv3.1 channel are quite limiting. Glycosylated (wild type), partially glycosylated (N220Q and N229Q), and unglycosylated (N220Q/N229Q) Kv3.1 proteins were expressed and characterized in a cultured neuronal-derived cell model, B35 neuroblastoma cells. Western blots, whole cell current recordings, and wound healing assays were employed to provide evidence that the conducting and non-conducting properties of the Kv3.1 channel were modified by N-glycans of the Kv3.1 glycoprotein. Electrophoretic migration of the various Kv3.1 proteins treated with PNGase F and neuraminidase verified that the glycosylation sites were occupied and that the N-glycans could be sialylated, respectively. The unglycosylated channel favored a different whole cell current pattern than the glycoform. Further the outward ionic currents of the unglycosylated channel had slower activation and deactivation rates than those of the glycosylated Kv3.1 channel. These kinetic parameters of the partially glycosylated Kv3.1 channels were also slowed. B35 cells expressing glycosylated Kv3.1 protein migrated faster than those expressing partially glycosylated and much faster than those expressing the unglycosylated Kv3.1 protein. These results have demonstrated that N-glycans of the Kv3.1 glycoprotein enhance outward ionic current kinetics, and neuronal migration. It is speculated that physiological changes which lead to a reduction in N-glycan attachment to proteins will alter the functions of the Kv3.1 channel.     

Glycosylation with heptose residues mediated by the aah gene product is essential for adherence of the AIDA-I adhesin

The diffuse adherence of Escherichia coli strain 2787 (O126:H27) is mediated by the autotransporter adhesin AIDA-I (adhesin-involved-in-diffuse-adherence) encoded by the plasmid-borne aidA gene. AIDA-I exhibits an aberrant mobility in denaturing gel electrophoresis. Deletion of the open reading frame (ORF) A immediately upstream of aidA restores the predicted mobility of AIDA-I, but the adhesin is no longer functional. This indicates that the mature AIDA-I adhesin is post-translationally modified and the modification is essential for adherence function. Labelling with digoxigenin hydrazide shows AIDA-I to be glycosylated. Using carbohydrate composition analysis, AIDA-I contains exclusively heptose residues (ratio heptose:AIDA-I approximately 19:1). The deduced amino acid sequence of the cytoplasmic open reading frame (ORF) A gene product shows homologies to heptosyltransferases. In addition, the modification was completely abolished in an ADP-glycero-manno-heptopyranose mutant. Our results provide direct evidence for glycosylation of the AIDA-I adhesin by heptoses with the ORF A gene product as a specific (mono)heptosyltransferase generating the functional mature AIDA-I adhesin. Consequently, the ORF A gene has been denoted 'aah' (autotransporter-adhesin-heptosyltransferase). Glycosylation by heptoses represents a novel protein modification in eubacteria.     

Proteolytic processing is not essential for multiple functions of the Escherichia coli autotransporter adhesin involved in diffuse adherence (AIDA-I)

The Escherichia coli adhesin involved in diffuse adherence (AIDA-I), like many other autotransporter proteins, is released in the periplasm as a proprotein undergoing proteolytic processing after its translocation across the outer membrane. The proprotein is cleaved into a membrane-embedded fragment, AIDAc, and an extracellular fragment, the mature AIDA-I adhesin. The latter remains noncovalently associated with the outer membrane and can be released by heat treatment. The mechanism of cleavage of the proprotein and its role in the functionality of AIDA-I are not understood. Here, we show that cleavage is independent of the amount of AIDA-I in the outer membrane, suggesting an intramolecular autoproteolytic mechanism or a cleavage mediated by an unknown protease. We show that the two fragments, mature AIDA-I and AIDAc, can be cosolubilized and copurified in a folded and active conformation. We observed that the release by heat treatment results from the unfolding of AIDA-I and that the interaction of AIDA-I with AIDAc seems to be disturbed only by denaturation. We constructed an uncleavable point mutant of AIDA-I, where a serine of the cleavage site was changed into a leucine, and showed that adhesion, autoaggregation, and biofilm formation mediated by the mutant are indistinguishable from the wild-type levels. Lastly, we show that both proteins can mediate the invasion of cultured epithelial cells. Taken together, our experiments suggest that the proteolytic processing of AIDA-I plays a minor role in the functionality of this protein.     

Autotransporter proteins: novel targets at the bacterial cell surface

Autotransporter proteins constitute a family of outer membrane/secreted proteins that possess unique structural properties that facilitate their independent transport across the bacterial membrane system and final routing to the cell surface. Autotransporter proteins have been identified in a wide range of Gram-negative bacteria and are often associated with virulence functions such as adhesion, aggregation, invasion, biofilm formation and toxicity. The importance of autotransporter proteins is exemplified by the fact that they constitute an essential component of some human vaccines. Autotransporter proteins contain three structural motifs: a signal sequence, a passenger domain and a translocator domain. Here, the structural properties of the passenger and translocator domains of three type Va autotransporter proteins are compared and contrasted, namely pertactin from Bordetella pertussis, the adhesion and penetration protein (Hap) from Haemophilus influenzae and Antigen 43 (Ag43) from Escherichia coli. The Ag43 protein is described in detail to examine how its structure relates to functional properties such as cell adhesion, aggregation and biofilm formation. The widespread occurrence of autotransporter-encoding genes, their apparent uniform role in virulence and their ability to interact with host cells suggest that they may represent rational targets for the design of novel vaccines directed against Gram-negative pathogens.     

A model vaccine exploiting fungal mannosylation to increase antigen immunogenicity

Ag mannosylation represents a promising strategy to augment vaccine immunogenicity by targeting Ag to mannose receptors (MRs) on dendritic cells. Because fungi naturally mannosylate proteins, we hypothesized that Ags engineered in fungi would have an enhanced capacity to stimulate T cell responses. Using the model Ag OVA, we generated proteins that differentially expressed N- and O-linked mannosylation in the yeast Pichia pastoris and compared them to their unglycosylated counterparts produced in Escherichia coli. We found that yeast-derived OVA proteins containing N-linkages, extensive O-linkages, or both were more potent than the unmannosylated Ags at inducing OVA-specific CD4+ T cell proliferation. This elevated response to fungal Ags was inhibited by mannan, suggesting involvement of MRs. However, the macrophage MR (CD206) was not essential, because macrophage MR-deficient dendritic cells were fully competent in presenting yeast-derived OVA Ags. Thus, the use of fungal glycosylation to provide N-linked and/or extensive O-linked mannosylation increased the capacity of the model Ag OVA to stimulate Ag-specific T cell responses in an MR-dependent manner. These data have implications for vaccine design by providing proof of principle that yeast-derived mannosylation can enhance immunogenicity.     

The effect of N-linked glycosylation on the substrate preferences of UDP glucuronosyltransferases

The presence of potential N-linked glycosylation sites (Asn-X-Ser/Thr) in two forms of UDP glucuronosyltransferase, designated UDPGTr-2 and UDPGTr-4, has been deduced from cDNA sequence data. These forms were glycosylated when synthesized from expression vectors transfected into COS cells and were converted to faster migrating species on SDS polyacrylamide gels when treated with endoglycosidase H. The role of glycosylation was investigated by determining the substrate specificities and stabilities of the glycosylated enzymes and their unglycosylated variants which were synthesized in the presence of tunicamycin. Analysis of the activities towards 13 different aglycones showed that the glycosyl moiety was not essential for catalytic activity and had no effect on the substrate preference of each form. The stabilities of the proteins were not adversely affected by the absence of this posttranslational modification. A possible effect of N-linked oligosaccharides on the catalytic properties of these two forms of UDP glucuronosyltransferase is discussed.     

Triple N-Glycosylation in the Long S5-P Loop Regulates the Activation and Trafficking of the Kv12.2 Potassium Channel

Mammalian voltage-dependent potassium (Kv) channels regulate the excitability of nerve and muscle cells. Kv12.2 features the longest S5-P loop among all known mammalian Kv channels with the most N-linked glycosylation sites (three sites). Despite its unique structural features, Kv12.2 is not well characterized. Because glycosylation plays important roles in the folding, trafficking, and function of various Kv channels, we focused on the N-glycosylation of Kv12.2. We show that Kv12.2 is N-glycosylated in Chinese hamster ovary (CHO) cells and in cultured neurons as well as in the mouse brain. As an effect of N-glycosylation on the function of Kv12.2, we demonstrate that removal of sugar chains causes a depolarizing shift in the steady-state activation without a significant reduction in current amplitude. Unlike the previously reported shift for Shaker-type Kv channels, this shift does not appear to be due to negatively charged sialic acid residues in the sugar chains. We next examined the trafficking in CHO cells to address whether the unglycosylated Kv12.2 channels are utilized in vivo. Although double mutants, retaining only one glycosylation site, are trafficked to the surface of CHO cells irrespective of the position of the glycosylated site, unglycosylated channels are not trafficked to the cell surface. Furthermore, we could not detect unglycosylated channels in the mouse brain. Our data suggest that only glycosylated Kv12.2 channels show proper voltage dependence and are utilized in vivo.     

Minimal structural and glycosylation requirements for ST6Gal I activity and trafficking

The influence of N-linked glycosylation on the activity and trafficking of membrane associated and soluble forms of the STtyr isoform of the ST6Gal I has been evaluated. We have demonstrated that the enzyme is glycosylated on Asn 146 and Asn 158 and that glycosylation is not required for the endoplasmic reticulum to Golgi transport of the membrane-associated form of the STtyr isoform. In addition, N-linked glycosylation may stabilize the protein but is not absolutely required for catalytic activity in vivo. In contrast, soluble forms of the protein consisting of amino acids 64-403, 89-403, and 97-403 are efficiently secreted and active in their fully glycosylated forms, but retained in the endoplasmic reticulum and inactive in their unglycosylated forms. These results suggest that membrane associated and soluble forms of the STtyr protein have different requirements for N-linked glycosylation. Elimination of the oligosaccharide attached to Asn 158 in the full length STtyr single and double glycosylation mutants generates proteins that are not cleaved and secreted but stably localized in the Golgi, like the STcys isoform of the ST6Gal I. This stable Golgi localization is correlated with the observation that these two mutants are active in in vivo assays but inactive in in vitro assays of membrane lysates. We predict that removal of N-linked oligosaccharides leads to an increased ability of the STtyr protein to self-associate or oligomerize which subsequently allows more stable retention in the Golgi and increased aggregation and inactivity when membranes are lysed in the in vitro activity assays.