The Saccharomyces cerevisiae Protein Mnn10p/Bed1p Is a Subunit of a Golgi Mannosyltransferase Complex*

In the yeast Saccharomyces cerevisiaemany of the N-linked glycans on cell wall and periplasmic proteins are modified by the addition of mannan, a large mannose-containing polysaccharide. Mannan comprises a backbone of approximately 50 α-1,6-linked mannoses to which are attached many branches consisting of α-1,2-linked and α-1,3-linked mannoses. The initiation and subsequent elongation of the mannan backbone is performed by two complexes of proteins in the cis Golgi. In this study we show that the product of theMNN10/BED1 gene is a component of one of these complexes, that which elongates the backbone. Analysis of interactions between the proteins in this complex shows that Mnn10p, and four previously characterized proteins (Anp1p, Mnn9p, Mnn11p, and Hoc1p) are indeed all components of the same large structure. Deletion of either Mnn10p, or its homologue Mnn11p, results in defects in mannan synthesisin vivo, and analysis of the enzymatic activity of the complexes isolated from mutant strains suggests that Mnn10p and Mnn11p are responsible for the majority of the α-1,6-polymerizing activity of the complex.

The glycoproteins of the cell wall of the yeast Saccharomyces cerevisiae are modified with both N-linked and O-linked glycans. The O-linked structures are chains of 4 -5 mannoses attached to serines and threonines, whereas the N-linked glycans comprise the conventional core structure to which can be attached a long outer chain structure of up to 200 mannose residues (1,2). This "mannan" modification is so extensive that the mannoproteins contribute up to 40% of the dry weight of the yeast cell wall (3). Mannans provide an external layer to the wall, which is believed both to contribute to its structural integrity and serve to exclude hydrolytic enzymes. Analysis of the structure of the S. cerevisiae mannan shows that it consists of a long ␣-1,6-linked backbone of about 50 residues to which short branches of 3-4 residues in length are attached (1,4,5). The first 2 mannoses of these branches are ␣-1,2-linked, and the final mannose is ␣-1,3-linked with some branches being additionally modified by the addition of a phosphomannose residue. A similar type of mannan structure comprising a long backbone with side branches is a feature of the cell wall of all other yeast and fungi, although the precise linkages, and even sugar residues, vary extensively even between quite closely related species (6 -10). This diversity may reflect a selective pressure to evade hydrolytic enzymes or immune responses that recognize a particular combination of residues and linkages.
The synthesis of the mannan structure in S. cerevisiae has been extensively investigated by both genetic and biochemical means. The mannan outer chains are only attached to the N-linked glycans of proteins destined for inclusion in the cell wall or residence in the periplasmic space. The synthesis begins in the Golgi apparatus where the first ␣-1,6-linked mannose residue of the mannan backbone is attached to the core by the Och1p mannosyltransferase (11). This mannose is attached to all N-linked glycans, but then only on a subset is it extended further to form the mannan backbone. On the remaining glycans a single ␣-1,2-linked residue is added and then an ␣-1,3linked residue is added to this, and to other points on the core structure (4). The extension of the mannan backbone is performed by two recently identified complexes of proteins in the cis Golgi. The first contains the proteins Mnn9p and Van1p, and the second has been shown also to include Mnn9p as well as the proteins Anp1p, Hoc1p, and Mnn11p (12). These two complexes possess ␣-1,6-transferase activity in vitro, and mutations in their components can result in either complete loss of the mannan backbone (mnn9 and van1) or mannan chains with a short backbone of 10 -15 residues (anp1), suggesting that the two complexes act sequentially to initiate and then extend the backbone. All of these proteins are type II membrane proteins, with a single transmembrane domain, a topology typical of Golgi glycosyltransferases (13,14). The branches on the mannan backbone are initiated and then extended by two ␣-1,2mannosyltransferases encoded by MNN2 and MNN5 (15). These two membrane proteins are not components of the Van1p-or Anp1p-containing complexes, which make the backbone. Finally the ␣-1,3-linked mannose and the mannose phosphate are added by Mnn1p and Mnn6p, respectively (16 -18).
Studies on mannan synthesis in S. cerevisiae are facilitated by mannan not being essential for viability. Survival without mannan is, however, dependent on cells being able to sense the cell wall defect, presumably to allow up-regulation of other cell wall components especially chitin (19). Thus mutations that result in defective mannans show synthetic lethality with components of the protein kinase C pathway, which regulates the expression of cell wall components (15,20,21). Cell wall integrity is likely to be particularly critical during budding, and again survival without mannan seems also to depend both on the actin cytoskeleton being well organized and on a functional mitotic check point, which allows bud formation to be slowed (22)(23)(24). In this study, we investigate a Golgi membrane protein of previously unknown function that was initially identified as corresponding to a bud emergence delay mutation bed1 (22). It was subsequently shown to correspond also to a previously identified mannan synthesis mutant mnn10 (25,26). Because Mnn10p is distantly related to Mnn11p, one of the components of the Anp1p-containing complex, we investigated its relationship to this complex. We show here that it is a further component of the complex that seems to comprise five proteins. This large structure does not appear to contain further proteins. Investigation of the activity of the complex in wild type and mutant strains indicates that the Mnn10p and Mnn11p proteins are responsible for the majority of the ␣-1,6polymerizing activity of the complex.

EXPERIMENTAL PROCEDURES
Yeast Strains and Plasmids-S. cerevisiae strain SEY6210 (Mat␣ ura3-52 his3⌬200 leu2-3, 112 trp-⌬901 lys2-801 suc2-⌬9) and derivatives were used throughout (27). Genes were disrupted by replacement of the open reading frame, or tagged at the exact C terminus using homologous recombination with polymerase chain reaction products containing Schizosaccharomyces pombe HIS5 or Kluyveromyces lactis URA3 genes, and integrants checked with polymerase chain reaction (28,29). For protein A fusions cleavable by the tobacco etch virus (TEV) 1 protease, plasmid pZZ-HIS5 was used as described previously (15). For hemagglutinin (HA) epitope tagging, polymerase chain reaction products were amplified from plasmid p3xHA-HIS5 that encodes a linker of amino acid sequence GAGAGA, followed by three copies of the sequence YPYDVPDYA, with the first and third repeats followed by a G residue. For Myc-tagging of open reading frames, a similar plasmid with nine copies of the Myc tag was used 2 except for Myc-tagged invertase, which was expressed in strains from a constitutive promoter using an integration plasmid pTi Invmyc (26).
Immunoblotting and Immunoprecipitation-Proteins separated by SDS-polyacrylamide gel electrophoresis were electrophoresed onto nitrocellulose and probed with antibodies in phosphate-buffered saline, 2% dried milk, 0.5% Tween 20. Immunoblotting and immunoprecipitations were performed with monoclonals against the Myc epitope (9E10), the HA epitope (12CA5), or ribosomal protein Tcm1p (kindly provided by Jon Warner) or with rabbit antisera against both tags (Santa Cruz Biotechnology) or Anp1p (12). Peroxidase-conjugated anti-rabbit and anti-mouse secondary antibodies were detected by chemiluminescence (ECL, Amersham Pharmacia Biotech). Total protein samples were prepared by resuspending log-phase yeast at one A 600 unit per 20 l of SDS buffer, bead beating for 5 min at 4°C (425-600-m glass beads, Sigma), and denaturing at 65°C for 5 min. Immunoprecipitates were prepared from unlabeled cells as described previously (15). Precipitation of protein A fusions was performed essentially as described previously but adapted for metabolically labeled cells. Thus four A 600 units of log-phase cells were labeled with 200 Ci of Tran 35 S-label (1,170 Ci/mmol, ICN) in 0.5 ml for 30 min at 30°C, spheroplasted as for the unlabeled cells, and resuspended in 0.5 ml of T/D/I buffer (10 mM triethanolamine, pH 7.5, 150 mM NaCl, 1% digitonin, 2 mM EDTA supplemented by protease inhibitors). After centrifugation at 14,000 ϫ g for 10 min to remove cell debris, the supernatant was added to 15 l of IgG-Sepharose (Amersham Pharmacia Biotech), rocked gently overnight, and the beads washed five times for 1 h in 0.4 ml T/D/I, before cleavage overnight in 20 l of T/D/I with 5 units of TEV protease (Life Technologies, Inc.), all steps after spheroplasting being at 4°C. The supernatant from the beads was removed, divided in two and treated with, or without, 500 units of endoglycosidase H (Endo H) according to the manufacturer's instructions (New England Biolabs). Protein was precipitated by methanol/chloroform extraction, resuspended in SDS sample buffer, separated by SDS-polyacrylamide gel electrophoresis, and the gel was treated with Amplify (Amersham Pharmacia Biotech) and fluorographed.
Gel Filtration-Protein complexes prepared by IgG-Sepharose precipitation and TEV protease cleavage from 1,000 A 600 units of cells, or 14,000 ϫ g supernatants from spheroplasts lysed in T/D/I buffer, were fractionated on a 34 ϫ 1.5-cm column of Sephacryl S400 (Amersham Pharmacia Biotech). The column was run in 10 mM triethanolamine, pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.4% digitonin at 15 ml/h and 1 ml of fractions were collected. 100 l of fractions were precipitated by methanol/chloroform extraction with soybean trypsin inhibitor as a carrier, resuspended in SDS buffer, and analyzed by immunoblotting. For standards, thyroglobulin, catalase, and ferritin (Amersham Pharmacia Biotech) were run on the same column under identical conditions and analyzed by SDS-polyacrylamide gel electrophoresis and Coomassie Blue staining.
Mannosyltransferase Assays-Protein A fusions or Anp1p-containing complexes were isolated and used on the beads for mannosyltransferase reactions as described previously (12,15). The reaction products were fractionated by gel filtration on a 14 ϫ 1.0-cm Sephadex G10 column (Amersham Pharmacia Biotech) run in deionized water. The radioactive peak fractions were pooled, concentrated by lyophilizing, treated with mannosidases, and the products reacted with 8-aminonaphthalene-1,3,6-trisulfonic acid (ANTS) for separation by electrophoresis as described previously (12,30) with the reaction products from 50 to 100 A 600 units of cells loaded per lane. The resulting gels were exposed to a PhosphorImager screen for 3 days (or 2 weeks for the ⌬mnn11 and ⌬mnn10 experiments) and analyzed with a Molecular Dynamics Phos-phorImager and ImageQuant software.

Mnn10p Is a Component of the Anp1p-Mnn9p
Complex of the cis Golgi-To determine whether Mnn10p was a component of one of the two Mnn9p-containing complexes, the same co-immunoprecipitation approach was followed that was originally used to reveal the interactions between Mnn9p and the other members of the complexes (12). Thus a triple HA-epitope tag was inserted at the end of the MNN10 open reading frame by homologous recombination. Protein blotting revealed that the tagged protein migrated as a single band with an apparent molecular mass of 46 kDa, reasonably close to that predicted from the sequence of the protein and the epitope tag (49 kDa, Fig. 1A). An anti-HA monoclonal antibody was used to precipitate the protein from yeast solubilized with the mild detergent digitonin. The immunoprecipitates were then blotted with antibodies to Anp1p and Van1p, and Fig. 1B shows that Mnn10p is associated with Anp1p and not with Van1p. Moreover, blotting of the supernatant following precipitation revealed that the Anp1p was quantitatively precipitated by the tagged Mnn10p, indicating that all of the Anp1p is associated with complexes that contain Mnn10p (Fig. 1B).
Organization of the Anp1p-Mnn9p Complex-Our previous analysis of the Mnn9p-containing complexes had shown that Anp1p is associated not only with Mnn9p but also with two other membrane proteins, Hoc1p and Mnn11p. As with Mnn10p, these proteins were not found associated with the Van1p-Mnn9p complex. This raises the question of how these five proteins are organized. In each case it is known that the protein associates with Anp1p. One possibility is that all five proteins are in a single complex. Alternatively there could be distinct complexes containing Mnn9p-Anp1p and one or two of the other components. We addressed this issue in two ways. First, we constructed strains with different epitope tags at the C terminus of pairs of the proteins. Immunoprecipitates were then performed using one anti-tag antibody, and the precipi- tated proteins probed on a blot for the second tagged protein. A representative experiment is shown in Fig. 2A for strains expressing Mnn10p and Mnn11p, tagged with Myc and HA in both combinations. In both cases, not only could the HA-tagged member of the pair be precipitated with the anti-HA monoclonal antibody as expected, but it could also be precipitated with the anti-Myc monoclonal antibody. This indicates that Mnn10p and Mnn11p are present in the same complex. Likewise the HOC1 open reading frame was tagged with the Myc epitope in strains in which either Mnn10p or Mnn11p were tagged with HA. In both cases the Myc-tagged Hoc1p could be precipitated with antibodies to the Myc or the HA tags, and also with a rabbit polyclonal against Anp1p. Blotting of the digitonin lysate of the Hoc1p-Myc-expressing cells revealed two bands, of which only the upper is present in blots of cellular proteins solubilized directly in SDS-containing sample buffer ( Fig. 2B and data not shown). Because the tag is at the C terminus, this suggests that proteases in the digitonin lysate remove about 5 kDa from the NH 2 terminus of the protein.
Such clipping in the region after the amino-terminal transmembrane domain has been seen for many Golgi enzymes in both yeast and mammals and is proposed to reflect this part of the enzymes being a flexible stalk region (31,32). Interestingly, this clipped form does not co-precipitate with either Mnn10p, Mnn11p or Anp1p, which may reflect removal by the protease of a region required for association in the complex (Fig. 2B).
The only exception to this co-precipitation of the various Anp1p-associated proteins was observed when tagged Hoc1p was precipitated from a strain also expressing tagged Mnn10p. Although precipitation of HA-tagged Mnn10p with anti-HA resulted in the co-precipitation of Myc-tagged Hoc1p, only low levels of Mnn10p-HA were precipitated when the precipitation was performed the other way round with anti-Myc (Fig. 2B). One possibility is that the C-terminal tag on Hoc1p lies close to the interface between Hoc1p and Mnn10p, and that binding of the anti-Myc monoclonal antibody disrupts the interaction.
Precipitation of the Anp1p-containing Complex from Metabolically Labeled Cells-To investigate further the composition of the Anp1p-containing complex and examine the possibility that it contains yet further components, a second approach was used to address the composition of the complex. For this a protein A tag was attached to the C terminus of each of the components of the complex in separate strains. The spacer between the protein and the tag contained a cleavage site for the sequence-specific protease from TEV. This allows the tagged proteins to be captured on IgG-Sepharose, and then proteolytically eluted under native conditions with minimal background (15). The isolation was initially performed with strains expressing either protein A-tagged Anp1p or tagged Van1p, or no tagged protein. Fig. 3A shows that the two tagged proteins precipitate a distinct pattern of bands, consistent with their being in two distinct complexes that share only the component Mnn9p. This analysis was then extended to the other components of the Anp1p-containing complex, and Fig. 3B shows the results of such a precipitation and elution performed on 35 S-labeled strains in which the different components of the complex had been tagged. A thirteen residue spacer including the TEV cleavage site is left on the protein following protease treatment, and this results in a small reduction of mobility compared with the nontagged version. In most cases all the other proteins can be identified as co-precipitating with a particular component. Analysis of the gels is complicated by the fact that Mnn10p and Hoc1p apparently co-migrate (consistent with their predicted molecular weights differing by only 0.4 kDa). However, when Mnn10p is shifted following the addition of the tag, a band the size of Hoc1p remains. Interestingly in the converse case, when the tag is attached to Hoc1p, only a very faint band in the position of Mnn10p is observed. This is consistent with the immunoprecipitation results described above, and with the notion that the C terminus of Hoc1p may be close to the interface with Mnn10p.
Taken together, these results strongly suggest that there are two different Mnn9p-containing complexes in the yeast cis Golgi. The first comprises Van1p and Mnn9p, and the second Anp1p, Mnn9p, Hoc1p, Mnn10p, and Mnn11p. Careful examination of the autoradiograms did not reveal any further unidentified bands in the Anp1p complex suggesting that these five proteins are the major constituents of the complex. However it remains possible that there are further proteins present at stoichiometry too low for detection by this approach. We have previously shown that both these complexes have ␣-1,6mannosyltransferase activity in vitro, attaching multiple residues to make polymeric structures. To simplify further discussion, we propose to refer to these two complexes as mannan polymerases (M-Pol) I and II, respectively.
Analysis of M-Pol I and M-Pol II by Gel Filtration-The above analysis indicates that M-Pol II contains three more components than M-Pol I. As such it would be expected to be a larger structure, and this prediction was tested using gel filtration. The protease elution method described above allows the complexes to be released from the beads in an assembled state, and such released complexes were applied to a gel filtration column. Fig. 4A shows that the two complexes do indeed elute from such a column with very different profiles. The ]methionine/cysteine. The yeast strains were either SEY6210, or the same with a protein A tag attached to the C terminus of either Anp1p or Van1p. The proteins were eluted from the beads using TEV protease, which cleaves before the protein A tag, and then digested with Endo H before polyacrylamide gel separation. B, as shown in A except that the cleavable tag was attached to the indicated protein, and the protease eluates run after treatment either with or without Endo H. A 13-residue spacer remains after tag cleavage resulting in the protein having a reduced mobility, and for the M-Pol II components, this version of the protein is indicated by an arrowhead. Mnn11p migrates as a slightly fainter and more fuzzy band than the others, the former perhaps reflecting its having fewer methionines than all but Hoc1p (12). Except for Van1p, none of the other proteins is affected by Endo H digestion, consistent with there being no predicted N-linked glycosylation sites in the luminal domains of the five proteins (the site present, 8 residues into the lumen from the TMD of Hoc1p, is predicted to be too close to the membrane to be modified (46)). membrane proteins the column was run in the presence of digitonin, which forms micelles of 70 kDa in 0.1 M NaCl (33). Although this sort of gel filtration analysis cannot be expected to give very accurate sizes, it seems likely that at least some of the proteins are present in the M-pol complexes at a greater than monomeric stoichiometry.
Deletion of Mnn10p and Mnn11p Results in Similar Defects in N-linked Glycosylation-It has been previously shown that M-Pol I and M-Pol II have both ␣-1,6and ␣-1,2-mannosyltransferase activity in vitro, raising the question of what contribution the individual components of the complex make to this activity (12). To address this issue, we initially examined the effect of deleting individual subunits on mannan synthesis in vivo. It is known that mutations in Mnn10p result in reduction in the length of mannan chains, but the effect of loss of Mnn11p has not previously been reported. The protein invertase is modified with mannan chains on 8 -10 of its N-linked glycans, and Fig. 5A shows that the gel mobility of invertase is increased in yeast in which either Anp1p, Mnn10p or Mnn11p is absent. Consistent with this the ⌬mnn11 strain also shows increased sensitivity to hygromycin, a property of other mannan defective mutants including anp1 and mnn10 (data not shown). This demonstrates that Mnn11p is required for the synthesis of full-length mannan chains in vivo, and justifies our inclusion of the gene in the MNN class as defined by Ballou and co-workers (34,35).
It has been observed previously that removal of the Mnn9p component of M-Pol II results in the destabilization of Anp1p (12,36). This presumably reflects a requirement for association between the two proteins for correct folding. In contrast, deletion of the other members of the complex does not result in the complete loss of Anp1p (Fig. 5B). However, the levels are reduced in some cases, and normalization to a control ribosomal protein (Tcm1p) shows that in the ⌬mnn10 and ⌬mnn11 strains the level of Anp1p is reduced by 85 and 75%, respectively, whereas deletion of HOC1 has no effect. This suggests that Mnn10p and Mnn11p contribute to the folding or stability of M-Pol II.
Deletion of Mnn10p or Mnn11p Alters the in Vitro Mannosyltransferase Activity of M-Pol II-Because the level of Anp1p is reduced in the ⌬mnn10 and ⌬mnn11 strains, it is possible that the defect in mannan synthesis seen is not because the proteins are directly involved in mannan synthesis, but rather because they are required for the stability or integrity of other proteins in the complex, which are actually responsible for mannan synthesis. To address the catalytic role of Mnn10p and Mnn11p, the Anp1p-protein A fusion described above was used to isolate M-Pol II for examination of its enzymatic activity in vitro. We initially examined the activity of the M-Pol II from wild type cells, and compared it with that of M-Pol I. Thus Anp1p and Van1p protein-A fusions were precipitated from cells and incubated on the beads with labeled GDP mannose and ␣-1,6-mannobiose as an acceptor. The products of the transferase reactions were divided into aliquots and digested with combinations of mannosidases. Finally, the charged fluorophore ANTS was attached to the reducing ends of the products to allow analysis by fluorophore-assisted carbohydrate electrophoresis (30). The gels were autoradiographed and examined by fluorescence to examine the reaction products. Fig.  6 shows that both M-Pol I and M-Pol II produced a ladder of mannosylated products as previously observed (12). Mannosidase digestion revealed that the product of M-Pol II is almost completely digested to monomeric mannose by ␣-1,6-mannosidase, suggesting that this is the major linkage made by the complex. In contrast, for M-Pol I apparently only about half of the products are ␣-1,6-linked.
The next step was to examine the activity of M-Pol II in the various deletion strains. The amount of mannosyltransferase activity precipitated was greatly reduced in the ⌬mnn10 and ⌬mnn11 strains (ϳ5-10% of wild type), but by using twice as many cells and longer exposure times it was possible to perform a digestion analysis. This revealed that in contrast to the case for the intact complex, the reaction products had received just a single mannose, and this was almost entirely resistant to ␣-1,6-mannosidase, even in conditions where the mannobiose substrate was completely digested. Fig. 6 shows the results obtained with M-Pol II from the ⌬mnn11 strain, and a similar result was obtained with the ⌬mnn10 strain (data not shown). In contrast, deletion of HOC1 did not alter the total activity of M-Pol II, nor the size or digestion sensitivity of the reaction product (data not shown). This result suggests that the reduction in mannan synthesis seen in the ⌬mnn10 and ⌬mnn11 strains is not simply a consequence of a general reduction in the level of the M-Pol II, but rather reflects the loss of the major ␣-1,6-mannosyltransferase activity of the complex. DISCUSSION In this paper we have examined the product of the MNN10/ BED1 gene. The gene was originally cloned as corresponding to the bud emergence delay mutant bed1 (22). It was subsequently shown to correspond to the previously identified mannan synthesis mutation mnn10 (26). Mnn10p is distantly related to a protein Mnn11p, which we recently found to be a component of one of the two protein complexes in the cis Golgi that are involved in the synthesis of the backbone of yeast mannan. Thus we examined its relationship to these two complexes, and report here that it is a component of the Anp1pcontaining complex that we refer to as M-Pol II. It was initially reported that epitope-tagged Bed1p was in the endoplasmic reticulum, based on overexpression of the protein, but it was subsequently found to have a Golgi localization when expressed as a single copy gene (22,26). We have previously observed a similar phenomenon with another component of M-Pol II, Anp1p, suggesting that both proteins need to be associated with other components of the complex to fold correctly and exit the endoplasmic reticulum (12,37). Indeed deletion of MNN10 results in a decrease in the levels of Anp1p.
The identification of Mnn10p as a component of M-Pol II brings the number of proteins found in the complex to five. Co-immunoprecipitation suggests that all five are in the same complex, and indeed the apparent size of the complex on gel filtration (ϳ1,300 kDa) suggest that the components are pres- A, immunoblot of total protein from yeast strain SEY6210 expressing a Myc-tagged version of invertase (wild type), or the same strain in which the indicated genes were deleted, or SEY6210 without invertase (Ϫ). B, immunoblots of total proteins from strains, as shown in A, probed with antisera to Anp1p, or with a monoclonal to the ribosomal protein L3 (Tcm1p) as a control for protein recovery. ent in more than one copy per complex. This raises the question of the exact function of the whole complex, and how this is shared between the individual components. Previous work has shown that deletion of either ANP1 or MNN1O results in defective mannan with a backbone reduced to 10 -15 residues (ANP1 being MNN8) (38,39). 3 We show here that deletion of MNN11 results in an alteration in invertase mobility similar to that seen with the deletion of these other two genes. In contrast, it is known that deletion of MNN9, or of VAN1, results in no mannan backbone at all beyond the single ␣-1,6-linked residue added by Och1p (38,40). Thus it seems most likely that the Mnn10p-containing M-Pol II elongates the mannan backbone after it is initiated by the Mnn9p-Van1p-containing M-Pol I (Fig. 7). In this model, M-Pol I would perform the initial extension of the ␣-1,6-linked backbone beyond the first ␣-1,6linked residue, which is added to all N-linked glycans by Och1p, consistent with the phenotypes of mnn9 and van1 mutants. However, it should be noted that we found that about half the mannose transferred in vitro by M-Pol I is in a linkage that is resistant to ␣-1,6-mannosidase, but sensitive to ␣-1,2/ 3-mannosidase. This suggests that either the isolated M-Pol I is associated with an additional protein with ␣-1,2/3-mannosyl-3 Neta Dean, personal communication.

FIG. 7. A model for the modification of N-linked glycans in the Golgi of S. cerevisiae.
All N-linked glycans have the same Man 8 GlcNAc 2 structure when they arrive at the cis Golgi from the endoplasmic reticulum, and Och1p adds a single ␣-1,6-linked mannose to them all. However the next steps are protein dependent. Many, but not all, of the proteins destined for incorporation in or under the cell wall have a mannan structure attached. The first step that commits them to this pathway is for M-Pol I to attach a short ␣-1,6-linked mannose polymer to the mannose added by Och1p. This is then extended further to a length of 40 -60 mannoses by M-Pol II. The extension of the backbone may then be terminated by a capping ␣-1,2-linked mannose. The rest of the backbone is then branched with ␣-1,2-linked mannoses by Mnn2p and Mnn5p, and the chains terminated by an ␣-1,3-linked mannose added by Mnn1p, and phosphomannoses added by Mnn6p. In contrast, the proteins of the vacuole and the internal membranes, as well as some secreted proteins, do not have mannan attached, but rather a single ␣-1,2-linked mannose is added to the Och1p product by an as yet unidentified ␣-1,2-mannosyltransferase termed ␣-1,2-MTII (47). As discussed in the text, this activity could lie within M-Pol I itself. Mnn1p attaches ␣-1,3-linked mannoses to this and to the core. What determines which proteins receive which modification is at present unclear. The reaction products were digested with the indicated mannosidases, modified with the charged fluorescent molecule ANTS, and then separated on polyacrylamide gels. The resulting gels were exposed to a PhosphorImager screen to identify the radioactive products (upper panels) or visualized by fluorescence (lower panels). The unmodified ␣-1,6-mannobiose substrate appears as a prominent band in the fluorescent images and is completely digested by the ␣-1,6-mannosidase. A glucose oligomer standard ladder is visible in the fluorescent images. For the mannosidasetreated samples the percentage of radioactive sugar released as monomer was determined from the phosphorimage and is indicated.
transferase activity (although we have not found association between M-Pol I and the ␣-1,2-mannosyltransferases Mnn2p, Mnn5p and Mnt1p (12,15)), or the complex itself can add ␣-1,2-mannose, as well as synthesizing the start of the ␣-1,6linked backbone. One interesting explanation would be that the ␣-1,2-mannosyltransferase activity that is proposed to modify the N-linked glycans that do not receive mannan, (␣-1,2-MTII in Fig. 7) could also reside in M-Pol I, a possibility that we are currently investigating.
Although it seems likely that M-Pol II performs the extension of the mannan backbone after it is initiated by M-Pol I, the question remains of the functions of the individual components of M-Pol II. Mnn10p and Mnn11p seem clear candidates to be ␣-1,6-mannosyltransferases. First, they contain a "DXD" motif that is conserved in many of their homologues that we have recently shown is a common feature of many families of nucleoside-diphosphate-sugar using glycosyltransferases (17). Second, when assayed in vitro the isolated M-Pol II catalyzes primarily the formation of ␣-1,6-linkages with only a small percentage of the product being susceptible to ␣-1,2/3-mannosidase, but when either Mnn10p or Mnn11p is removed there is relatively much less ␣-1,6-transferase activity in the complex. It may be that having two such transferases, possibly present in multiple copies, in a large complex facilitates the rapid synthesis of a long ␣-1,6-linked backbone. One objection to this model might be that Mnn10p is quite closely related to an ␣-1,2-galactosyltransferase from S. pombe (GM12) (41). The conserved portion of Mnn10p (the C-terminal 280 residues) shows 27% identity to GM12 and only 21% identity to Mnn11p, when aligned with PSI-BLAST (42). However it is very unlikely that Mnn10p is a galactosyltransferase. Galactose is not normally found in S. cerevisiae glycoproteins, and indeed when the S. pombe GM12 is expressed in the cells, galactose is now incorporated into the mannan (43). The relatedness of the two genes may reflect evolutionary pressure to vary mannan structure. Branched polysaccharides containing either mannose alone, or mannose and galactose (galactomannans) are a common feature of the cell walls of yeast and fungi (44,45). It is possible that the pressure for variation of cell wall structure has caused the enzymes to shift between using galactose and mannose, and to adding either the backbone or the branches with different linkages. The presence of homologues of Mnn10p in plants and Caenorhabditis elegans raises the possibility that related complexes of glycosyltransferases may be involved in making large oligosaccharide structures in other organisms (12,26).
If Mnn10p and Mnn11p are ␣-1,6-mannosyltransferases, then what can be said of the function of Anp1p, Mnn9p, and Hoc1p? All have conserved DXD motifs, and Hoc1p is homologous to the Och1p ␣-1,6-mannosyltransferase. Although they may all contribute to ␣-1,6-mannosyltransferase activity, we also found that M-Pol II can form linkages resistant to ␣-1,6mannosidase but sensitive to an ␣-1,2/3 mannosidase. This is unlikely to represent the initiation of the ␣-1,2-linked branches as we have recently identified the product of the MNN2 gene as being responsible for this, and it is not a component of M-Pol II (15). However, in mnn2 strains, a single ␣-1,2-linked mannose is still found at the very end of an otherwise unbranched chain, and it has been suggested that it might serve a capping function (39). This capping mannose could be added by one of these remaining components and it may serve to regulate mannan size, or prevent re-elongation of the products of M-Pol II. It has also been argued that Hoc1p has a regulatory function as removal of the gene has little effect on mannan size, but overexpression can suppress defects in protein kinase C (20). It has long been suggested that the size of mannans is regulated in a cell cycle and growth state-dependent fashion. Now that a clearer picture of the composition of M-Pol I and M-Pol II is emerging it will hopefully be possible to understand the precise structure, functioning, and regulation of these complexes, which may reveal principles relevant to the synthesis of large oligosaccharides in other eukaryotes.