AU2012227297B2 - Combinatorial DNA Library for Producing Modified N-Glycans in Lower Eukaryotes - Google Patents

Abstract

The present invention relates to eukaryotic host cells having modified oligosaccharides which may be modified further by heterologous expression of a set of glycosyltransferases, sugar transporters and mannosidases to become host-strains for the production of mammalian, e.g., human therapeutic glycoproteins. The invention provides nucleic acid molecules and combinatorial libraries which can be used to successfully target and express mammalian enzymatic activities such as those involved in glycosylation to intracellular compartments in a eukaryotic host cell. The process provides an engineered host cell which can be used to express and target any desirable gene(s) involved in glycosylation. Host cells with modified oligosaccharides are created or selected. N-glycans made in the engineered host cells have a Man5GlcNAc2 core structure which may then be modified further by heterologous expression of one or more enzymes, e.g., glycosyltransferases, sugar transporters and mannosidases, to yield human-like glycoproteins. For the production of therapeutic proteins, this method may be adapted to engineer cell lines in which any desired glycosylation structure may be obtained. Man ctl,2-Man cxl,6 2/20 ER Man al,2-Man al,3-.Yna I Man 11,4-GIcNAc P1,4-GlCNAW-lAsn Man a1,2-Man al,2-Man czl,3 IMafl9GIc7Nk2 "Man~a1,2-Mn~aj, a-1,2 man nosidlase Manal3 ---Man al 16\ Man a1,3-' ~/ M an 131 .-G INAc 1 ,4-G IcNAcpIAsn Man oQ1,2-Man al,24vMan al,3 Man 8GlcNAc2 Man ai,6 ~ ral.2 mannosidase (IA. 18. IC) Manal, -- Mn a, .6/Man P1,4-GlcNAc pI1 4-GlcNAcJ31.Asn Man cxl,3 / Man5GlcNWC2 Man al,6 04.(1.2 GnTI Man a1l,3--- Ma )6Man pi1 4-GloNAc P31,4-GIcNAcI3I-Asn GlcNAc 13,2-Man zl ,3~ GlN 3ln cN~c Man al 16 \ 4mannosldase IE GOLGI G~oN~c f31,2 Man 13 01,4..GIcNAc 114-GIcNAcI3I-Asn GLIGIcNAc 31,2-Ma 4,.3 (3. nU.B14G lNT I f-14nT lc GIcNAc 1 ,24 Man al.6\ 12TI014 TZ -,GT GcNAc 031,4 /Man 131,4-GcNAc 131,4-GlcNAcJpl-An GIcNAc P1,2 -Man al,3 Cornex cati Gl 14cNAc 1,2w 1.Ga Gal 131,4-G cNAc P1 2 1,4 a a Gal P1,4-GIcNAc 13,4... \Man p31,4-GlcNAc 13,4-GIcNAcp1-Asn Gal 131,4-GlcNAc 01,2X' NANA Gal 1,4-GIcNAc 1 , 2 N a2,3 ST. 2,ST NANA Gal P 1,4-GlNAc P 1,4 ~,.Man a a 14Gc~ 14Gc~p-s NANA Gal 13,-lNc 13 'M 1 aI3,4-GIcNAc 131,4-GcN P1.As (NANA Gal 13,4-GlcNAc 13,2~ nCmlxGyorti FIG. lB

Description

COMBINATORIAL DNA LIBRARY FOR PRODUCING MODIFIED N-GLYCANS IN LOWER EUKARYOTES 100011 The following statement is a full description of this invention, including the best method of performing it known to the applicant: CROSS-REFERENCE TO RELATED APPLICATIONS [0002] This application is a divisional of Australian application no. 2010200666, which is a divisional of Australian application no. 2004213869 and is related to International application no. PCT/US2004/005244 filed on 20 February, 2004 claiming priority to United States application Serial No. 10/371,877 which is a continuation-in-part of United States Application Serial No. 09/892,591, filed on 27 June, 2001, in which priority is claimed to U.S. Provisional Application Serial No. 60/214,358, filed on 28 June, 2000, U.S. Provisional Application Serial No. 60/215,638, filed on 30 June, 2000, and U.S. Provisional Application No. 60/279,997, filed on 30 March, 2001; each of which is incorporated herein by reference in its entirety. [Text continues on page 2.] 1 FIELD OF THE INVENTION [00031 The present invention is directed to methods and compositions by which non-human eukaryotic host cells, such as fungi or other eukaryotic cells, can be genetically modified to produce glycosylated proteins (glycoproteins) having 5 patterns of glycosylation similar to those of glycoproteins produced by animal cells, especially human cells, which are useful as human or animal therapeutic agents. BACKGROUND OF THE INVENTION 10 Glycosylation Pathways in Humans and Lower Eukaryotes [0004] After DNA is transcribed and translated into a protein, further post translational processing involves the attachment of sugar residues, a process known as glycosylation. Different organisms produce different glycosylation enzymes (glycosyltransferases and glycosidases), and have different substrates (nucleotide 15 sugars) available, so that the glycosylation patterns as well as composition of the individual oligosaccharides, even of the same protein, will be different depending on the host system in which the particular protein is being expressed. Bacteria typically do not glycosylate proteins, and if so only in a very unspecific manner (Moens and Vanderleyden, 1997 Arch Microbiol. 168(3):169-175). Lower 20 eukaryotes such as filamentous fungi and yeast add primarily mannose and mannosylphosphate sugars. The resulting glycan is known as a "high-mannose" type glycan or a mannan. Plant cells and insect cells (such as Sf9 cells) glycosylate proteins in yet another way. By contrast, in higher eukaryotes such as humans, the nascent oligosaccharide side chain may be trimmed to remove several 25 mannose residues and elongated with additional sugar residues that typically are not found in the N-glycans of lower eukaryotes. See, e.g., R.K. Bretthauer, et al. Biotechnology and Applied Biochemistry, 1999, 30, 193-200; W. Martinet, et al. Biotechnology Letters, 1998, 20, 1171-1177; S. Weikert, et al. Nature Biotechnology, 1999, 17, 1116-1121; M. Malissard, et al. Biochemical and 30 Biophysical Research Communications, 2000, 267, 169-173; Jarvis, et al., Current Opinion in Biotechnology, 1998, 9:528-533; and M. Takeuchi, 1 Tends in Glycoscience and Glycotechnology, 1997, 9, S29-S35. 2 10005] Synthesis of a mammalian-type oligosaccharide structure begins with a set of sequential reactions in the course of which sugar residues are added and removed while the protein moves along the secretory pathway in the host organism. The enzymes which reside along the glycosylation pathway of the host 5 organism or cell determine the resulting glycosylation patterns of secreted proteins. Thus, the resulting glycosylation pattern of proteins expressed in lower eukaryotic host cells differs substantially from the glycosylation pattern of proteins expressed in higher eukaryotes such as humans and other mammals (Bretthauer, 1999). The structure of a typical fungal N-glycan is shown in Fig. 1A. 10 [0006] The early steps of human glycosylation can be divided into at least two different phases: (i) lipid-linked Glc 3 MangGlcNAc 2 oligosaccharides are assembled by a sequential set of reactions at the membrane of the endoplasmic reticulum (ER) and (ii) the transfer of this oligosaccharide from the lipid anchor dolichyl pyrophosphate onto de novo synthesized protein. The site of the specific transfer is 15 defined by an asparagine (Asn) residue in the sequence Asn-Xaa-Ser/Thr where Xaa can be any amino acid except proline (Gavel, 1990). Further processing by glucosidases and mannosidases occurs in the ER before the nascent glycoprotein is transferred to the early Golgi apparatus, where additional mannose residues are removed by Golgi specific alpha (cc)-1,2-mannosidases. Processing continues as 20 the protein proceeds through the Golgi. In the medial Golgi, a number of modifying enzymes, including N-acetylglucosaninyl Transferases (GnT, GnTII, GnTIII, GnTIV and GnTV), mannosidase II and fucosyltransferases, add and remove specific sugar residues. Finally, in the trans-Golgi, galactosyltranferases (GalT) and sialyltransferases (ST) produce a glycoprotein structure that is released 25 from the Golgi. It is this structure, characterized by bi-, tri- and tetra-antennary structures, containing galactose, fucose, N-acetylglucosamine and a high degree of terminal sialic acid, that gives glycoproteins their human characteristics. The structure of a typical human N-glycan is shown in Fig. 1B. [0007] In nearly all eukaryotes, glycoproteins are derived from a common lipid 30 linked oligosaccharide precursor Glc3MansGlcNAc 2 -dolichol-pyrophosphate. Within the endoplasmic reticulum, synthesis and processing of dolichol pyrophosphate bound oligosaccharides are identical between all known eukaryotes. 3 However, further processing of the core oligosaccharide by fungal cells, e.g., yeast, once it has been transferred to a peptide leaving the ER and entering the Golgi, differs significantly from humans as it moves along the secretory pathway and involves the addition of several mannose sugars. 5 [0008] In yeast, these steps are catalyzed by Golgi residing mannosyl transferases, like Ochip, Mntlp and Mnn1p, which sequentially add mannose sugars to the core oligosaccharide. The resulting structure is undesirable for the production of human-like proteins and it is thus desirable to reduce or eliminate mannosyltransferase activity. Mutants of S.cerevisiae, deficient in mannosyl 10 transferase activity (for example ochi or mnn9 mutants) have been shown to be non-lethal and display reduced mannose content in the oligosaccharide of yeast glycoproteins. Other oligosaccharide processing enzymes, such as mannosylphosphate transferase, may also have to be eliminated depending on the host's particular endogenous glycosylation pattern. 15 Sugar Nucleotide Precursors 100091 The N-glycans of animal glycoproteins typically include galactose, fucose, and terminal sialic acid. These sugars are not found on glycoproteins produced in yeast and filamentous fungi. In humans, the full range of nucleotide 20 sugar precursors (e.g. UDP-N-acetylglucosamine, UDP-N-acetylgalactosamine, CMP-N-acetylneuraminic acid, UDP-galactose, GDP-fucose, etc.) are synthesized in the cytosol and transported into the Golgi, where they are attached to the core oligosaccharide by glycosyltransferases. (Sommers and Hirschberg, 1981 J. Cell Biol. 91(2): A406-A406; Sommers and Hirschberg 1982 J. Biol. Chem. 257(18): 25 811-817; Perez and Hirschberg 1987 Methods in Enzymology 138: 709-715). [0010] Glycosyl transfer reactions typically yield a side product which is a nucleoside diphosphate or monophosphate. While monophosphates can be directly exported in exchange for nucleoside triphosphate sugars by an antiport mechanism, diphosphonucleosides (e.g. GDP) have to be cleaved by phosphatases (e.g. 30 GDPase) to yield nucleoside monophosphates and inorganic phosphate prior to being exported. This reaction is important for efficient glycosylation; for example, GDPase from Saccharomyces cerevisiae (S.cerevisiae) has been found to be 4 necessary for mannosylation. However that GDPase has 90% reduced activity toward UDP (Berninsone et al., 1994 J. Biol. Chem. 269(1):207-21 1). Lower eukaryotes typically lack UDP-specific diphosphatase activity in the Golgi since they do not utilize UDP-sugar precursors for Golgi-based glycoprotein synthesis. 5 Schizosaccharomyces pombe, a yeast found to add galactose residues to cell wall polysaccharides (from UDP-galactose) has been found to have specific UDPase activity, indicating the potential requirement for such an enzyme (Berninsone et al., 1994). UDP is known to be a potent inhibitor of glycosyltransferases and the removal of this glycosylation side product may be important to prevent glycosyl 10 transferase inhibition in the lumen of the Golgi (Khatara et al., 1974). See Berninsone, P., et al. 1995. J. Biol. Chem. 270(24): 14564-14567; Beaudet, L., et al. 1998 Abc Transporters: Biochemical, Cellular, and Molecular Aspects. 292: 397-413. 15 Sequential Processing of N-glycans by Compartmentalized Enzyme Activities [00111 Sugar transferases and glycosidases (e.g., mannosidases) line the inner (luminal) surface of the ER and Golgi apparatus and thereby provide a "catalytic" surface that allows for the sequential processing of glycoproteins as they proceed through the ER and Golgi network. The multiple compartments of the cis, medial, 20 and trans Golgi and the trans-Golgi Network (TGN), provide the different localities in which the ordered sequence of glycosylation reactions can take place. As a glycoprotein proceeds from synthesis in the ER to full maturation in the late Golgi or TGN, it is sequentially exposed to different glycosidases, mannosidases and glycosyltransferases such that a specific carbohydrate structure may be 25 synthesized. Much work has been dedicated to revealing the exact mechanism by which these enzymes are retained and anchored to their respective organelle. The evolving picture is complex but evidence suggests that stem region, membrane spanning region and cytoplasmic tail, individually or in concert, direct enzymes to the membrane of individual organelles and thereby localize the associated catalytic 30 domain to that locus (see, e.g., Gleeson, P.A. (1998) Histochem. Cell Biol. 109, 517-532). 5 [0012] In some cases, these specific interactions were found to function across species. For example, the membrane spanning domain of ca2,6-ST from rats, an enzyme known to localize in the trans-Golgi of the animal, was shown to also localize a reporter gene (invertase) in the yeast Golgi (Schwientek, 1995). 5 However, the very same membrane spanning domain as part of a full-length c2,6 ST was retained in the ER and not further transported to the Golgi of yeast (Krezdorn, 1994). A full length GaIT from humans was not even synthesized in yeast, despite demonstrably high transcription levels. In contrast, the transmembrane region of the same human GaIT fused to an invertase reporter was 10 able to direct localization to the yeast Golgi, albeit it at low production levels. Schwientek and co-workers have shown that fusing 28 amino acids of a yeast mannosyltransferase (MNTI), a region containing a cytoplasmic tail, a transmembrane region and eight amino acids of the stem region, to the catalytic domain of human GalT are sufficient for Golgi localization of an active GalT. 15 Other galactosyltransferases appear to rely on interactions with enzymes resident in particular organelles because, after removal of their transmembrane region, they are still able to localize properly. [00131 Improper localization of a glycosylation enzyme may prevent proper functioning of the enzyme in the pathway. For example, Aspergillus nidulans, 20 which has numerous a-1,2-mannosidases (Eades and Hintz, 2000 Gene 255(1):25 34), does not add GlcNAc to MansGlcNAc 2 when transformed with the rabbit GnTI gene, despite a high overall level of GnTI activity (Kalsner et al., 1995). GnTI, although actively expressed, may be incorrectly localized such that the enzyme is not in contact with both of its substrates: UDP-GIcNAc and a 25 productive Man 5 GlcNAc 2 substrate (not all Man 5 GlcNAc 2 structures are productive; see below). Alternatively, the host organism may not provide an adequate level of UDP-GlcNAc in the Golgi or the enzyme may be properly localized but nevertheless inactive in its new environment. In addition, Man 5 GlcNAc 2 structures present in the host cell may differ in structure from 30 Man 5 GlcNAc 2 found in mammals. Maras and coworkers found that about one third of the N-glycans from cellobiohydrolase I (CBHI) obtained from T.reesei can be trimmed to MansGlcNAc 2 byA.saitoi 1,2 mannosidase in vitro. Fewer than 1% 6 of those N-glycans, however, could serve as a productive substrate for GnTI. The mere presence of Man 5 GlcNAc 2 , therefore, does not assure that further processing to MansGlcNAc 2 can be achieved. It is formation of a productive, GnTI-reactive MansGlcNAc 2 structure that is required. Although MansGlcNAc 2 could be 5 produced in the cell (about 27 mol %), only a small fraction could be converted to MansGlcNAc 2 (less than about 5%, see Chiba WO 01/14522). 100141 To date, there is no reliable way of predicting whether a particular heterologously expressed glycosyltransferase or mannosidase in a lower eukaryote will be (1), sufficiently translated (2), catalytically active or (3) located to the 10 proper organelle within the secretory pathway. Because all three of these are necessary to affect glycosylation patterns in lower eukaryotes, a systematic scheme to. achieve the desired catalytic function and proper retention of enzymes in the absence of predictive tools, which are currently not available, would be desirable. 15 Production of Therapeutic Glycoproteins 10015] A significant number of proteins isolated from humans or animals are post-translationally modified, with glycosylation being one of the most significant modifications. An estimated 70% of all therapeutic proteins are glycosylated and thus currently rely on a production system (i.e., host cell) that is able to glycosylate 20 in a manner similar to humans. Several studies have shown that glycosylation plays an important role in determining the (1) immunogenicity, (2) pharmacokinetic properties, (3) trafficking, and (4) efficacy of therapeutic proteins. It is thus not surprising that substantial efforts by the pharmaceutical industry have been directed at developing processes to obtain glycoproteins that are as 25 "humanoid" or "human-like" as possible. To date, most glycoproteins are made in a mammalian host system. This may involve the genetic engineering of such mammalian cells to enhance the degree of sialylation (i.e., terminal addition of sialic acid) of proteins expressed by the cells, which is known to improve pharmacokinetic properties of such proteins. Alternatively, one may improve the 30 degree of sialylation by in vitro addition of such sugars using known glycosyltransferases and their respective nucleotide sugars (e.g., 2,3 sialyltransferase and CMP-sialic acid). 7 [0016] While most higher eukaryotes carry out glycosylation reactions that are similar to those found in humans, recombinant human proteins expressed in the above mentioned host systems invariably differ from their "natural" human counterpart (Raju, 2000). Extensive development work has thus been directed at 5 finding ways to improve the "human character" of proteins made in these expression systems. This includes the optimization of fermentation conditions and the genetic modification of protein expression hosts by introducing genes encoding enzymes involved in the formation of human-like glycoforrns (Werner, 1998; Weikert, 1999; Andersen, 1994; Yang, 2000). Inherent problems associated with 10 all mammalian expression systems have not been solved. [0017] Fermentation processes based on mammalian cell culture (e.g., CHO, murine, or human cells), for example, tend to be very slow (fermentation times in excess of one week are not uncommon), often yield low product titers, require expensive nutrients and cofactors (e.g., bovine fetal serum), are limited by 15 programmed cell death (apoptosis), and often do not enable expression of particular therapeutically valuable proteins. More importantly, mammalian cells are susceptible to viruses that have the potential to be human pathogens and stringent quality controls are required to assure product safety. This is of particular concern as many such processes require the addition of complex and temperature 20 sensitive media components that are derived from animals (e.g., bovine calf serum), which may carry agents pathogenic to humans such as bovine spongiform encephalopathy (BSE) prions or viruses. Moreover, the production of therapeutic compounds is preferably carried out in a well-controlled sterile environment. An animal farm, no matter how cleanly kept, does not constitute such an environment, 25 thus constituting an additional problem in the use of transgenic animals for manufacturing high volume therapeutic proteins. [00181 Most, if not all, currently produced therapeutic glycoproteins are therefore expressed in mammalian cells and much effort has been directed at improving (i.e., "humanizing") the glycosylation pattern of these recombinant proteins. Changes in 30 medium composition as well as the co-expression of genes encoding enzymes involved in human glycosylation have been successfully employed (see, for example, Weikert, 1999). 8 Glycoprotein Production Using Eukaryotic Microorganisms [00191 The lack of a suitable mammalian expression system is a significant obstacle to the low-cost and safe production of recombinant human glycoproteins 5 for therapeutic applications. It would be desirable to produce recombinant proteins similar to their mammalian, e.g., human, counterparts in lower eukaryotes (fungi and yeast). Production of glycoproteins via the fermentation of microorganisms would offer numerous advantages over existing systems. For example, fermentation-based processes may offer (a) rapid production of high concentrations 10 of protein; (b) the ability to use sterile, well-controlled production conditions; (c) the ability to use simple, chemically defined (and protein-free) growth media; (d) ease of genetic manipulation; (e) the absence of contaminating human or animal pathogens such as viruses; (f) the ability to express a wide variety of proteins, including those poorly expressed in cell culture owing to toxicity etc.; and (g) ease 15 of protein recovery (e.g. via secretion into the medium). In addition, fermentation facilities for yeast and fungi are generally far less costly to construct than cell culture facilities. Although the core oligosaccharide structure transferred to a protein in the endoplasmic reticulum is basically identical in mammals and lower eukaryotes, substantial differences have been found in the subsequent processing 20 reactions which occur in the Golgi apparatus of fungi and mammals. In fact, even amongst different lower eukaryotes there exist a great variety of glycosylation structures. This has historically prevented the use of lower eukaryotes as hosts for the production of recombinant human glycoproteins despite otherwise notable advantages over mammalian expression systems. 25 [0020] Therapeutic glycoproteins produced in a microorganism host such as yeast utilizing the endogenous host glycosylation pathway differ structurally from those produced in mammalian cells and typically show greatly reduced therapeutic efficacy. Such glycoproteins are typically immunogenic in humans and show a reduced half-life (and thus bioactivity) in vivo after administration (Takeuchi, 30 1997). Specific receptors in humans and animals (i.e., macrophage mannose receptors) can recognize terminal mannose residues and promote the rapid clearance of the foreign glycoprotein from the bloodstream. Additional adverse 9 effects may include changes in protein folding, solubility, susceptibility to proteases, trafficking, transport, compartmentalization, secretion, recognition by other proteins or factors, antigenicity, or allergenicity. [0021] Yeast and filamentous fungi have both been successfully used for the 5 production of recombinant proteins, both intracellular and secreted (Cereghino, J. L. and J. M. Cregg 2000 FEMS Microbiology Reviews 24(1): 45-66; Harkki, A., et al. 1989 Bio-Technology 7(6): 596; Berka, R. M., et al. 1992 Abstr.Papers Amer. Chem.Soc.203: 121-BIOT; Svetina, M., et al, 2000 J. Biotechnol. 76(2-3): 245 25 1). Various yeasts, such as K lactis, Pichia pastoris, Pichia methanolica, and 10 Hansenula polymorpha, have played particularly important roles as eukaryotic expression systems because they are able to grow to high cell densities and secrete large quantities of recombinant protein. Likewise, filamentous fungi, such as Aspergillus niger, Fusarium sp, Neurospora crassa and others, have been used to efficiently produce glycoproteins at the industrial scale. However, as noted above, 15 glycoproteins expressed in any of these eukaryotic microorganisms differ substantially in N-glycan structure from those in animals. This has prevented the use of yeast or filamentous fungi as hosts for the production of many therapeutic glycoproteins. 100221 Although glycosylation in yeast and fungi is very different than in 20 humans, some common elements are shared. The first step, the transfer of the core oligosaccharide structure to the nascent protein, is highly conserved in all eukaryotes including yeast, fungi, plants and humans (compare Figs. 1A and 1B). Subsequent processing of the core oligosaccharide, however, differs significantly in yeast and involves the addition of several mannose sugars. This step is 25 catalyzed by mannosyltransferases residing in the Golgi (e.g. OCHI, AMNT1, MNN1, etc.), which sequentially add mannose sugars to the core oligosaccharide. The resulting structure is undesirable for the production of humanoid proteins and it is thus desirable to reduce or eliminate mannosyltransferase activity. Mutants of S.cerevisiae deficient in mannosyltransferase activity (e.g. ochi or 0nn9 mutants) 30 have shown to be non-lethal and display a reduced mannose content in the oligosaccharide of yeast glycoproteins. Other oligosaccharide processing enzymes, such as mannosylphosphate transferase, may also have to be eliminated 10 depending on the host's particular endogenous glycosylation pattern, After reducing undesired endogenous glycosylation reactions, the formation of complex N-glycans has to be engineered into the host system. This requires the stable expression of several enzymes and sugar-nucleotide transporters. Moreover, one 5 has to localize these enzymes so that a sequential processing of the maturing glycosylation structure is ensured. [00231 Several efforts have been made to modify the glycosylation pathways of eukaryotic microorganisms to provide glycoproteins more suitable for use as mammalian therapeutic agents. For example, several glycosyltransferases have 10 been separately cloned and expressed in S cerevisiae (GaiT, GnTI), Aspergillus nidulans (GnTI) and other fungi (Yoshida et al., 1999, Kalsner et al., 1995 Glycoconj. J. 12(3):360-370, Schwientek et al,, 1995). However, N-glycans resembling those made in human cells were not obtained. [00241 Yeasts produce a variety of mannosyltransferases (e.g., 1,3 15 mannosyltransferases such as MNNI in S. cerevisiae; Graham and Emr, 1991 J Cell. Bio. 114(2):207-218), 1,2-mannosyltransferases (e.g. KTR/KRE family from S.cerevisiae), 1,6-mannosyltransferases (e.g., OCH! from S.cerevisiae), mannosylphosphate transferases and their regulators (e.g., MNN4 and MNN6 from S.cerevisiae) and additional enzymes that are involved in endogenous 20 glycosylation reactions. Many of these genes have been deleted individually giving rise to viable organisms having altered glycosylation profiles. Examples are shown in Table 1. Table 1. Examples of yeast strains having altered mannosylation Strain N-glycan (wild type) Mutation N-glycan (mutant) Reference S. pombe Man.,GlcNAc 2 OCHI MangGlcNAc 2 Yoko-o et al., 2001 FEBS Lett. 489(1):75-80 S.cerevisiae Ma,GlcNAc 2 OCH1/MNN1 MansGlcNAc 2 Nakanishi-Shindo et A. 1993 J. io. Chem. 268(35):26338 26345 S.cerevisiae Man>,GlcNAc 2 OCH/MNN1I/MNN4 ManaGlcNAc 2 Chiba et al., 1998 J. Biol. Chem. 273, 26298-26304 P.pastoris Hyperglycosylated OCHI (complete Not Welfide, Japanese deletion) hyperglycosylated Application Publication No. 8 11 336387 P.pastoris Man>RGlcNAc 2 OCHI (disruption) Man>sGIcNAc 2 Contreras et al. WO 02/00856 A2 [0025] Japanese Patent Application Publication No. 8-336387 discloses the deletion of an OCHJ homolog in Pichia pastoris. In S. cerevisiae, OCHJ encodes a 1,6-mannosyltransferase, which adds a mannose to the glycan structure 5 MansGlcNAc 2 to yield MangGlcNAc 2 . The MangGlcNAc 2 structure, which contains three 1,6 mannose residues, is then a substrate for further 1,2-, 1,6-, and 1,3- mannosyltransferases in vivo, leading to the hypermannosylated glycoproteins that are characteristic for S.cerevisiae and which typically may have 30-40 mannose residues per N-glycan. Because the Ochlp initiates the transfer of 1,6 10 mannose to the ManaGlcNAc 2 core, it is often referred to as the "initiating 1,6 mannosyltransferase" to distinguish it from other 1,6 mannosyltransferases acting later in the Golgi. In an ochl mnnl mnn4 mutant strain of S.cerevisiae, proteins glycosylated with MansGIcNAc 2 accumulate and hypermannosylation does not occur. However, ManGlcNAc 2 is not a substrate for mammalian 15 glycosyltransferases, such as human UDP-GlcNAc transferase I, and accordingly, the use of that mutant strain, in itself, is not useful for producing mammalian-like proteins, i.e., with complex or hybrid glycosylation patterns. [0026] Although Japanese Patent Application Publication No. 8-336387 discloses methods to obtain an och1 mutant of P.pastoris displaying a reduced 20 mannosylation phenotype, it provides no data on whether the initiating 1,6 mannosyltransferase activity presumed to be encoded by OCHI is reduced or eliminated. It is well-established in the field of fungal genetics that homologs of genes often do not play the same role in their respective host organism. For example, the Neurospora rca-I gene complements an AspergillusflbD sporulation 25 mutant but has no identifiable role in Neurospora sporulation. Shen, W.C. et al., Genetics 1998;148(3):1031-41. More recently, Contreras (WO 02/00856 A2) shows that, in an ochi mutant of P. pastors, at least 50% of the cell wall glycans cannot be trimmed to Man 5 GlcNAc 2 with a Trichoderma reesei a'-1,2-mannosidase (see Fig. I1 of WO 02/00856 A2). As the wild-type displays a very similar 30 glycosylation pattern (Fig. 10, Panel 2 of WO 02/00856 A2), it appears that the OCHI gene of P.pastoris may not encode the initiating 1, 6 -mannosyltransferase 12 activity and is thus different from its genetic homolog in S.cerevisiae. Thus, to date, there is no evidence that initiating a-1, 6 -mannosyltransferase activity is eliminated in och1 mutants of P. pastoris, which further supports the notion that the glycosylation pathways of S.cerevisiae and P.pastoris are significantly 5 different. [0027] Martinet et al. (Biotechnol. Lett. 1998, 20(12), 1171-1177) reported the expression of a-1,2-mannosidase from T. reesei in P. pastoris. Some mannose training from the N-glycans of a model protein was observed. However, the model protein had no N-glycans with the structure MansGlcNAc 2 , which would be 10 necessary as an intermediate for the generation of complex N-glycans. Accordingly, that system is not useful for producing proteins with complex or hybrid glycosylation patterns. [0028 Similarly, Chiba et al. (1998) expressed a-1,2-mannosidase from Aspergillus saitot in the yeast Saccharomyces cerevisiae. A signal peptide 15 sequence (His-Asp-Glu-Leu) (SEQ ID NO: 5) was engineered into the exogenous mannosidase to promote its retention in the endoplasmic reticulum. In addition, the yeast host was a mutant lacking enzyme activities associated with hypermannosylation of proteins: 1,6-mannosyltransferase (ochl); 1,3 mannosyltransferase (mnnl); and a regulator ofmannosylphosphate transferase 20 (mnn4). The N-glycans of the triple mutant host consisted primarily of the structure Man8GlcNAc 2 , rather than the high mannose forms found in wild-type S.cerevisiae. In the presence of the engineered mannosidase, the N-glycans of a model protein (carboxypeptidase Y) were trimmed to give a mixture consisting of 27 mole % MansGIcNAc 2 , 22 mole % Man 6 GlcNAc 2 , 22 mole % Man 7 GlcNAc 2 , 25 and 29 mole % Man8GlcNAc 2 . Trimming of cell wall glycoproteins was less efficient, with only 10 mole % of the N-glycans having the desired MansGlcNAc 2 structure. [0029] Even if all the Man 5 GlcNAc 2 glycans were the correct MansGlcNAc 2 form that can be converted to GlcNAcMansGlcNAc 2 by GnTI, the above system 30 would not be efficient for the production of proteins having human-like glycosylation patterns. If several glycosylation sites are present in a desired protein, the probability (P) of obtaining such a protein in a correct form follows the 13 relationship P=(F)", where n equals the number of glycosylation sites, and F equals the fraction of desired glycoforms. A glycoprotein with three glycosylation sites would have a 0.1% chance of providing the appropriate precursors for complex and hybrid N-glycan processing on all of its glycosylation sites. Thus, using the 5 system of Chiba to make a glycoprotein having a single N-glycosylation site, at least 73 mole % would have an incorrect structure. For a glycoprotein having two or three N-glycosylation sites, at least 93 or 98 mole % would have an incorrect structure, respectively. Such low efficiencies of conversion are unsatisfactory for the production of therapeutic agents, particularly as the separation of proteins 10 having different glycoforms is typically costly and difficult. [00301 Chiba et al. (WO 01/14522) have shown high levels of MansGlcNAc 2 structures on recombinant fibroblast growth factor (FGF), a secreted soluble glycoprotein produced in S.cerevisiae. It is not clear, however, that the detected MansGlcNAc 2 was produced inside the host cell (i.e. in vivo) because the a-1,2 15 mannosidase was targeted by fusion with an HDEL (SEQ ID NO: 5) localization tag, a mechanism, which is known to be leaky (Pelham H.R. (1998) EMBO J. 7, 913-918). It is more likely that FGF was secreted into the medium, where it was then processed by ct-1,2 mannosidase which had escaped the HDEL (SEQ ID NO: 5) retrieval mechanism and leaked into the medium. As mentioned above, an 20 intracellular protein (CPY), expressed in the same strain, contained mostly glycans (more than 73%) that were ManGlcNAc 2 and larger. The majority of the MansGlcNAc 2 structures on FGF are, thus, likely to have been produced ex vivo. It is further unclear whether the Man 5 GlcNAc 2 structures that were produced were productive substrates for GnTI. 25 [0031] As the above work demonstrates, one can trim Man8GlcNAc 2 structures to a MansGlcNAc 2 isomer in S.cerevisiae, although high efficiency trimming greater than 50% in vivo has yet to be determined, by engineering a fungal mannosidase from A. saitoi into the endoplasmic reticulum (BR). The shortcomings of this approach are two-fold: (1) it is not clear whether the 30 MansGlcNAc 2 structures formed are in fact formed in vivo (rather than having been secreted and further modified by mannosidases outside the cell); and (2) it is not clear whether any Man 5 GlcNAc 2 structures formed, if in fact formed in vivo, 14 are the correct isoform to be a productive substrate for subsequent N-glycan modification by GlcNAc transferase I (Maras et al., 1997, Eur. J. Biochemn. 249, 701-707). [00321 With the objective of providing a more human-.ike glycoprotein derived 5 from a fungal host, U.S. Patent No. 5,834,251 discloses a method for producing a hybrid glycoprotein derived from Trichoderma reseei. A hybrid N-glycan has only mannose residues on the Manal-6 arm of the core mannose structure and one or two complex antennae on the Manal-3 arm. While this structure has utility, the method has the disadvantage that numerous enzymatic steps must be performed in' 10 vitro, which is costly and time-consuming. Isolated enzymes are expensive to prepare and need costly substrates (e.g. UDP-GlcNAc). The method also does not allow for the production of complex glycans on a desired protein. Intracellular Mannosidase Activity Involved in N-glycan Trimming 15 [00331 Alpha-1,2-mannosidase activity is required for the trimming of Man8GlcNAc 2 to form Man 5 GlcNAc 2 , which is a major intermediate for complex N-glycan formation in mammals. Previous work has shown that truncated murine, fungal and human a-1,2-mannosidase can be expressed in the methylotropic yeast P.pastoris and display Man8GlcNAc 2 to Man 5 GIcNAc 2 trimming activity (Lal et 20 al., Glycobiology 1998 Oct;8(10):981-95; Tremblay et al., Glycobiology 1998 Jun;8(6):585-95, Callewaert et al., 2001). However, to date, no reports exist that show the high level in vivo trimming of Man8GlcNAc 2 to MansGlcNAc 2 on a secreted glycoprotein from P.pastoris. [00341 While it is useful to engineer strains that are able to produce 25 Man 5 GlcNAc 2 as the primary N-glycan structure, any attempt to further modify these high mannose precursor structures to more closely resemble human glycans requires additional in vivo or in vitro steps. Methods to further humanize glycans from fungal and yeast sources in vitro are described in U.S. Pat. No. 5,834,251 (supra). As discussed above, however, if MansGlcNAc 2 is to be further humanized 30 in vivo, one has to ensure that the generated Man 5 GlcNAc 2 structures are, in fact, generated intracellularly and not the product of mannosidase activity in the medium. Complex N-glycan formation in yeast or fungi will require high levels of 15 MansGlcNAc 2 to be generated within the cell because only intracellular MansGlcNAc 2 glycans can be further processed to hybrid and complex N-glycans in vivo. In addition, one has to demonstrate that the majority of Man 5 GlcNAc 2 structures generated are in fact a substrate for GnTI and thus allow the formation 5 of hybrid and complex N-glycans. [0035] Moreover, the mere presence of an a-1,2-mannosidase in the cell does not, by itself, ensure proper intracellular trimming of MansGlcNAc 2 to Man 5 GlcNAc 2 . (See, e.g., Contreras et al. WO 02/00856 A2, in which an HDEL (SEQ ID NO: 5) tagged mannosidase of T. reesei is localized primarily in the ER 10 and co-expressed with an influenza haemagglutinin (HA) reporter protein on which virtually no Man 5 GlcNAc 2 could be detected. See also Chiba et al., 1998 (supra), in which a chimeric a-1,2-mannosidase/Ochlp transmrnembrane domain fusion localized in the ER, early Golgi and cytosol of S.cerevisiae, had no mannosidase trimming activity). Accordingly, mere localization of a mannosidase in the ER or 15 Golgi is insufficient to ensure activity of the respective enzyme in that targeted organelle. (See also, Martinet et al. (1998), supra, showing that a-1,2 mannosidase from T. reesei, while localizing intracellularly, increased rather than decreased the extent of mannosylation). To date, there is no report that demonstrates the intracellular localization of an active heterologous a-l,2 20 mannosidase in either yeast or fungi using a transmembrane localization sequence. [0036] Accordingly, the need exists for methods to produce glycoproteins characterized by a high intracellular Man5GlcNAc 2 content which can be further processed into human-like glycoprotein structures in non-human eukaryotic host cells, and particularly in yeast and filamentous fungi. 25 SUMMARY OF THE INVENTION [0037] Host cells and cell lines having genetically modified glycosylation pathways that allow them to carry out a sequence of enzymatic reactions which mimic the processing of glycoproteins in mammals, especially in humans, have 30 been developed. Recombinant proteins expressed in these engineered hosts yield glycoproteins more similar, if not substantially identical, to their mammalian, e.g., human counterparts. Host cells of the invention, e.g., lower eukaryotic micro 16 organisms and other non-human, eukaryotic host cells grown in culture, are modified to produce N-glycans such as MansGlcNAc 2 or other structures produced along human glycosylation pathways. This is achieved using a combination of engineering and/or selection of strains which: do not express certain enzymes 5 which create the undesirable structures characteristic of the fungal glycoproteins; which express heterologous enzymes selected either to have optimal activity under the conditions present in the host cell where activity is to be achieved; or combinations thereof; wherein the genetically engineered eukaryote expresses at least one heterologous enzyme activity required to produce a "human-like" 10 glycoprotein. Host cells of the invention may be modified further by heterologous expression of one or more activities such as glycosyltransferases, sugar transporters and mannosidases, to become strains for the production of mammalian, e.g., human therapeutic glycoproteins. [0038] The present invention thus provides a glycoprotein production method 15 using (1) a lower eukaryotic host such as a unicellular or filamentous fungus, or (2) any non-human eukaryotic organism that has a different glycosylation pattern from humans, to modify the glycosylation composition and structures of the proteins made in a host organism ("host cell") so that they resemble more closely carbohydrate structures found in mammalian, e.g., human proteins. The process 20 allows one to obtain an engineered host cell which can be used to express and target any desirable gene(s), e.g.,one involved in glycosylation, by methods that are well-established in the scientific literature and generally known to the artisan in the field of protein expression. Host cells with modified oligosaccharides are created or selected. N-glycans made in the engineered host cells have a 25 Man 5 GlcNAc 2 core structure which may then be modified further by heterologous expression of one or more enzymes, e.g., glycosyltransferases, glycosidases, sugar transporters and mannosidases, to yield human-like glycoproteins. For the production of therapeutic proteins, this method may be adapted to engineer cell lines in which any desired glycosylation structure may be obtained. 30 10039] Accordingly, in one embodiment, the invention provides a method for producing a human-like glycoprotein in a non-human eukaryotic host cell. The host cell of the invention is selected or engineered to be depleted in 1,6-mannosyl 17 transferase activities which would otherwise add mannose residues onto the N glycan on a glycoprotein. One or more enzymes (enzymatic activities) are introduced into the host cell which enable the production of a MansGlcNAc 2 carbohydrate structure at a high yield, e.g., at least 30 mole percent. In a more 5 preferred embodiment, at least 10% of the MansGlcNAc 2 produced within the host cell is a productive substrate for GnTI and thus for further glycosylation reactions in vivo and/or in vitro that produce a finished N-glycan that is similar or identical to that formed in mammals, especially humans. [0040] In another embodiment, a nucleic acid molecule encoding one or more 10 enzymes for production of a MansGlcNAc 2 carbohydrate structure is introduced into a host cell selected or engineered to be depleted in 1, 6 -mannosyltransferase activities. In one preferred embodiment, at least one enzyme introduced into the host cell is selected to have optimal activity at the pH of the subcellular location where the carbohydrate structure is produced. In another preferred embodiment, at 15 least one enzyme is targeted to a host subcellular organelle where the enzyme will have optimal activity, e.g., by means of a chimeric protein comprising a cellular targeting signal peptide not normally associated with the enzyme. [0041] The invention further provides isolated nucleic acid molecules and vectors comprising such molecules which encode an initiating al,6-mannosyl 20 transferase activity isolated from P.pastoris or from Klactis. These nucleic acid molecules comprise sequences that are homologous to the OCHI gene in S.cerevisiae. These and homologous sequences are useful for constructing host cells which will not hypermannosylate the N-glycan of a glycoprotein. [0042] In another embodiment, the host cell is engineered to express a 25 heterologous glycosidase, e.g., by introducing into the host one or more nucleic acid molecules encoding the glycosidase. Preferably, a nucleic acid molecule encodes one or more mannosidase activities involved in the production of MansGlcNAc 2 from ManBGlcNAc 2 or MangGlcNAc 2 . In a preferred embodiment, at least one of the encoded mannosidase activities has a pH optimum within 1.4 pH 30 units of the average pH optimum of other representative enzymes in the organelle in which the mannosidase activity is localized, or has optimal activity at a pH of between about 5.1 and about 8.0, preferably between about 5.5 and about 7.5. 18 Preferably, the heterologous enzyme is targeted to the endoplasmic reticulum, the Golgi apparatus or the transport vesicles between ER and Golgi of the host organism, where it trims N-glycans such as MangGlcNAc 2 to yield high levels of Man 5 GlcNAc 2 . In one embodiment, the enzyme is targeted by forming a fusion 5 protein between a catalytic domain of the enzyme and a cellular targeting signal peptide, e.g., by the in-frame ligation of a DNA fragment encoding a cellular targeting signal peptide with a DNA fragment encoding a glycosylation enzyme or catalytically active fragment thereof. [00431 In yet another embodiment, the glycosylation pathway of a host is 10 modified to express a sugar nucleotide transporter. In a preferred embodiment, a nucleotide diphosphatase enzyme is also expressed. The transporter and diphosphatase improve the efficiency of engineered glycosylation steps, by providing the appropriate substrates for the glycosylation enzymes in the appropriate compartments, reducing competitive product inhibition, and promoting 15 the removal of nucleoside diphosphates. 100441 The present invention also provides a combinatorial nucleic acid library useful for making fusion constructs which can target a desired protein or polypeptide fragment, e.g., an enzyme involved in glycosylation or a catalytic domain thereof, to a selected subcellular region of a host cell. In one preferred 20 embodiment, the combinatorial nucleic acid library comprises (a) nucleic acid sequences encoding different cellular targeting signal peptides and (b) nucleic acid sequences encoding different polypeptides to be targeted. Nucleic acid sequences of or derived from (a) and (b) are ligated together to produce fusion constructs, at least one of which encodes a functional protein domain (e.g., a catalytic domain of 25 an enzyme) ligated in-frame to a heterologous cellular targeting signal peptide, i.e., one which it normally does not associate with. [00451 The invention also provides a method for modifying the glycosylation pathway of a host cell (e.g., any eukaryotic host cell, including a human host cell) using enzymes involved in modifying N-glycans including glycosidases and 30 glycosyltransferases; by transforming the host cell with a nucleic acid (e.g., a combinatorial) library of the invention to produce a genetically mixed cell population expressing at least one and preferably two or more distinct chimeric 19 glycosylation enzymes having a catalytic domain ligated in-frame to a cellular targeting signal peptide which it normally does not associate with. A host cell having a desired glycosylation phenotype may optionally be selected from the population. Host cells modified using the library and associated methods of the 5 invention are useful, e.g., for producing glycoproteins having a glycosylation pattern similar or identical to those produced in mammals, especially humans. [0046] In another aspect, the combinatorial library of the present invention enables production of one or a combination of catalytically active glycosylation enzymes, which successfully localize to intracellular compartments in which they 10 function efficiently in the glycosylation/secretory pathway. Preferred enzymes convert Mans(a-1,2-Man) 3 gGlcNAc 2 to MansGlcNAc 2 at high efficiency in vivo. In addition, the invention provides eukaryotic host strains, and in particular, yeasts, fungal, insect, plant, plant cells, algae and insect cell hosts, capable of producing glycoprotein intermediates or products with MansGlcNAc 2 and/or 15 GlcNAcMansGlcNAc 2 as the predominant N-glycan. [0047] The present invention also provides recombinant molecules derived from a combinatorial nucleic acid library; vectors, including expression vectors, comprising such recombinant molecules; proteins encoded by the recombinant molecules and vectors; host cells transformed with the recombinant molecules or 20 vectors; glycoproteins produced from such transformed hosts; and methods for producing, in vivo, glycoprotein intermdiates or products with predominantly Man 5 GlcNAc 2 or GlcNAcMansGlcNAc 2 N-glycans covalently attached to appropriate glycosylation sites using the combinatorial library. [0048] Further aspects of this invention include methods, compositions and kits 25 for diagnostic and therapeutic uses in which presence or absence of MansGlcNAc 2 and/or GlcNAcMansGlcNAc 2 on a glycoprotein may be detected. 20 [0048a] A definition of the specific embodiment of the invention claimed herein follows. [0048b] In a broad format, the invention provides a lower eukaryotic host cell that does not display a 1,6 mannosyltransferase activity with respect to the N-glycan on a glycoprotein, said host cell further comprising a nucleic acid encoding a fusion protein comprising the catalytic domain of an a-1,2-mannosidase fused at the N terminus to an SECl2, VANI, or MNNI0 cellular targeting signal peptide for production of a Man 5 GlcNAc 2 carbohydrate structure, wherein Man 5 GlcNAc 2 as a productive substrate for GnTI in vivo is produced within the host cell at a yield of at least 30 mole percent. [Text continues on page 21.1 20a [00511 Figure 2 depicts construction of a combinatorial DNA library of fusion constructs. Fig. 2A diagrams the insertion of a targeting peptide fragment into pCR2.1-TOPO (Invitrogen, Carlsbad, CA). Fig. 2B shows the generated targeting peptide sub-library having restriction sites NotI - AscI. Fig. 2C diagrams the 5 insertion of a catalytic domain region into pJN347, a modified pUC19 vector. Fig. 2D shows the generated catalytic domain sub-library having restriction sites NotI, Asc and Paci. Fig. 2E depicts one particular fusion construct generated from the targeting peptide sub-library and the catalytic domain sub-library. [00521 Figure 3 (SEQ ID NOS: 45-46 respectively, in order of appearance) 10 illustrates the Mrmusculus a-1,2-mannosidase IA open reading frame. The sequences of the PCR' primers used to generate N-terminal truncations are underlined. [0053] Figures 4A - 4F illustrates engineering of vectors with multiple auxotrophic markers and genetic integration of target proteins in the P.pastoris 15 OCHJ locus. [0054] Figures 5A - SE show MALDI-TOF analysis demonstrating production of kringle 3 domain of human plasminogen (K3) glycoproteins having MansGlcNAc 2 as the predominant N-glycan structure in P. pastoris. Fig. SA depicts the standard MansGlcNAc 2 [a] glycan (Glyko, Novato, CA) and 20 MansGlcNAc 2 + Na [b]. Fig. 5B shows PNGase - released glycans from K3 wild type. The N-glycans shown are as follows: Man 9 GlcNAc 2 [d]; ManjoGcNAc 2 [e]; Man, 1 GlcNAc 2 [fj; Mani 2 GlcNAc 2 [g]. Fig. 5C depicts the ochi deletion resulting in the production of MansGIcNAc 2 [c] as the predominant N-glycan. Figs. 5D and 5E show the production of MansGlcNAc 2 [b] after in vivo trimming of 25 ManBGlcNAc 2 with a chimeric a-1,2-mannosidase. The predominant N-glycan is indicated by a peak with a mass (m/z) of 1253 consistent with its identification as MansGlcNAc 2 [b]. [0055] Figures 6A - 6F show MALDI-TOF analysis demonstrating production of IFN-0 glycoproteins having MansGlcNAc 2 as the predominant N-glycan 30 structure in P. pastoris. Fig. 6A shows the standard MansGlcNAc 2 (a] and Man 5 GlcNAc 2 + Na (b] as the standard (Glyko, Novato, CA). Fig. 6B shows PNGase - released glycans from IFN- # wildtype. Fig. 6C depicts the ochi knock 21 out producing MansGlcNAc 2 [c]; Man9GlcNAc 2 [d]; ManioGlcNAc 2 [e]; Man, ,GIcNAc 2 [fl; Man1 2 GlcNAc 2 [g]; and no production of MansGlcNAc 2 [b]. Fig. 6D shows relatively small amount of MansGlcNAc 2 [b] among other intermediate N-glycans MansGIcNAc 2 [c] to Mani 2 GlcNAc 2 [g]. Fig. 6E shows a 5 significant amount of MansGlcNAc 2 [b] relative to the other glycans Map4GlcNAc 2 [c] and Man 9 GIcNAc 2 [d] produced by pGC5 (Saccharomyces MNS1(m)/mouse mannosidase IB A99). Fig. 6F shows predominant production of Man 5 GlcNAc 2 [b] on the secreted glycoprotein IFN- by pFB8 (Saccharomyces SEC12 (m)/mouse mannosidase IA Al 87). The N-glycan is indicated by a peak 10 with a mass (m/z) of 1254 consistent with its identification as MansGlcNAc 2 [b]. [0056] Figure 7 shows a high performance liquid chromatogram for: (A) MangGlcNAc 2 standard labeled with 2-AB (negative control); (B) supernatant of medium P.pastoris, dochl transformed with pFB8 mannosidase, which demonstrates a lack of extracellular mannosidase activity in the supernatant; and 15 (C) Man 9 GlcNAc 2 standard labeled with 2-AB after exposure to T.reesei mannosidase (positive control). [0057] Figure 8 shows a high performance liquid chromatogram for: (A) ManGlcNAc 2 standard labeled with 2-AB (negative control); (B) supernatant of medium Ppastoris, dochl transformed with pGC5 mannosidase, which 20 demonstrates a lack of extracellular mannosidase activity in the supernatant; and (C) Man 9 GlcNAc 2 standard labeled with 2-AB after exposure to Treesei mannosidase (positive control). [00581 Figure 9 shows a high performance liquid chromatogram for: (A) MangGlcNAc 2 standard labeled with 2-AB (negative control); (B) supernatant of 25 medium P.pastoris, dochl transformed with pBC 18-5 mannosidase, which demonstrates lack of extracellular mannosidase activity in the supernatant; and (C) supernatant of medium Ppastoris, dochl transformed with pDD28-3, which demonstrates activity in the supernatant (positive control). [00591 Figures 1OA - 10B demonstrate the activity of an UDP-GlcNAc 30 transporter in the production of GlcNAcMansGlcNAc 2 in P. pastoris. Fig. IA depicts a P.pastoris strain (YSH-3) with a human GnTI but without the UDP GlcNAc transporter resulting in some production of GlcNAcMansGlcNAc 2 [b] but 22 a predominant production of MansGlcNAc 2 [a]. Fig 10B depicts the addition of UDP-GlcNAc transporter from Klactis in a strain (PBP-3) with the human GnTI, which resulted in the predominant production of GlcNAcMan 5 GlcNAc 2 [b]. The single prominent peak of mass (m/z) at 1457 is consistent with its identification as 5 GlcNAcMansGlcNAc 2 [b] as shown in Fig. 1OB. [0060] Figure 11 shows a pH optimum of a heterologous mannosidase enzyme encoded by pBB27-2 (Saccharomyces MN10 (s)/C. elegans mannosidase IB A3 1) expressed in P. pastoris. [0061] Figures 12A - 12C show MALDI-TOF analysis ofN-glycans released 10 from a cell free extract of Klactis. Fig. 12A shows the N-glycans released from wild-type cells, which includes high-mannose type N-glycans. Fig. 12B shows the N-glycans released from och1 mnn1 deleted cells, revealing a distinct peak of mass (m/z) at 1908 consistent with its identification as Man 9 GlcNAc 2 [d]. Fig. 12C shows the N-glycans released from och1 mnnl deleted cells after in vitro ca-1,2 15 mannosidase digest corresponding to a peak consistent with MansGIcNAc 2 . [00621 Figure 13 represents T-DNA cassettes with catalytic domain(s) of glycosylation enzymes fused in-frame to different leader sequences. The ends of ,the T-DNA are marked by the right (rb) and left borders (lb). Various promoters and terminators may also be used. The plant selectable marker can also be varied. 20 The right and left borders are required only for agrobacterium-mediated transformation and not for particle bombardment or electroporation. DETAILED DESCRIPTION OF THE INVENTION [0063] Unless otherwise defined herein, scientific and technical terms used in 25 connection with the present invention shall have the meanings that are commonly understood by those of ordinary skill in the art. Further, unless otherwise required by context, singular terms shall include pluralities and plural terms shall include the singular. The methods and techniques of the present invention are generally performed according to conventional methods well known in the art. Generally, 30 nomenclatures used in connection with, and techniques of biochemistry, enzymology, molecular and cellular biology, microbiology, genetics and protein 23 and nucleic acid chemistry and hybridization described herein are those well known and commonly used in the art. [0064] The methods and techniques of the present invention are generally performed according to conventional methods well-known in the art and as 5 described in various general and more specific references that are cited and discussed throughout the present specification unless otherwise indicated. See, e.g., Sambrook et al. Molecular Cloning: A Laboratory Manual, 2d ed., Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y. (1989); Ausubel et al., Current Protocols in Molecular Biology, Greene Publishing Associates (1992, and 10 Supplements to 2002); Harlow and Lane Antibodies: A Laboratory Manual Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y. (1990); Introduction to Glycobiology, Maureen E. Taylor, Kurt Drickamer, Oxford Univ. Press (2003); Worthington Enzyme Manual, Worthington Biochemical Corp. Freehold, NJ; Handbook of Biochemistry: Section A Proteins Vol 11976 CRC Press; Handbook 15 of Biochemistry: Section A Proteins Vol 111976 CRC Press; Essentials of Glycobiology, Cold Spring Harbor Laboratory Press (1999). The nomenclatures used in connection with, and the laboratory procedures and techniques of, molecular and cellular biology, protein biochemistry, enzymology and medicinal and pharmaceutical chemistry described herein are those well known and 20 commonly used in the art. [0065] All publications, patents and other references mentioned herein are incorporated by reference. [0066] The following terms, unless otherwise indicated, shall be understood to have the following meanings: 25 [0067] As used herein, the term "N-glycan" refers to an N-linked oligosaccharide, e.g., one that is attached by an asparagine-N-acetylglucosamine linkage to an asparagine residue of a polypeptide. N-glycans have a common pentasaccharide core of Man 3 GlcNAc 2 ("Man" refers to mannose; "Glc" refers to glucose; and "NAc" refers to N-acetyl; GlcNAc refers to N-acetylglucosamine). 30 The term " trimannose core" used with respect to the N-glycan also refers to the structure Man 3 GlcNAc 2 ("Man 3 "). N-glycans differ with respect to the number of branches (antennae) comprising peripheral sugars (e.g., fucose and sialic acid) that 24 are added to the Man 3 core structure. N-glycans are classified according to their branched constituents (e.g., high mannose, complex or hybrid). [00681 A "high mannose" type N-glycan has five or more mannose residues. A "complex" type N-glycan typically has at least one GlcNAc attached to the 1,3 5 mannose arm and at least one GlcNAc attached to the 1,6 mannose arm of the trimannose core. Complex N-glycans may also have galactose ("Gal") residues that are optionally modified with sialic acid or derivatives ("NeuAc", where "Neu" refers to neuraminic acid and "Ac" refers to acetyl). A complex N-glycan typically has at least one branch that terminates in an oligosaccharide such as, for example: 10 NeuNAc-; NeuAca2-6GalNAcal-; NeuAca2-3Galbl-3GalNAcal-; NeuAca2 3/6Galbl-4GlcNAcbl-; GlcNAcal-4Galbl-(mucins only); Fucal -2Galb 1-(blood group H). Sulfate esters can occur on galactose, GaINAc, and GIcNAc residues, and phosphate esters can occur on mannose residues. NeuAc (Neu: neuraminic acid; Ac:acetyl) can be O-acetylated or replaced by NeuGI (N-glycolylneuraminic 15 acid). Complex N-glycans may also have intrachain substitutions comprising "bisecting" GlcNAc and core fucose ("Fuc"). A "hybrid" N-glycan has at least one GlcNAc on the terminal of the 1,3 mannose arm of the trimannose core and zero or more mannoses on the 1,6 mannose arm of the trimannose core. [0069] The term "predominant" or "predominantly" used with respect to the 20 production of N-glycans refers to a structure which represents the major peak detected by matrix assisted laser desorption ionization time of flight mass spectrometry (MALDI-TOF) analysis. (0070] Abbreviations used herein are of common usage in the art, see, e.g., abbreviations of sugars, above. Other common abbreviations include "PNGase", 25 which refers to peptide N-glycosidase F (EC 3.2.2.18); "GlcNAc Tr' or "GnT," which refers to N-acetylglucosaminyl Transferase enzymes; "NANA" refers to N acetylneuraminic acid. [0071] As used herein, a "humanized glycoprotein" or a "human-like glycoprotein" refers alternatively to a protein having attached thereto N-glycans 30 having less than four mannose residues, and synthetic glycoprotein intermediates (which are also useful and can be manipulated further in vitro or in vivo) having at least five mannose residues. Preferably, glycoproteins produced according to the 25 invention contain at least 30 mole %, preferably at least 40 mole % and more preferably 50 - 100 mole % of the MansGlcNAc 2 intermediate, at least transiently. This may be achieved, e.g., by engineering a host cell of the invention to express a "better", i.e., a more efficient glycosylation enzyme. For example, a mannosidase 5 is selected such that it will have optimal activity under the conditions present at the site in the host cell where proteins are glycosylated and is introduced into the host cell preferably by targeting the enzyme to a host cell organelle where activity is desired. [0072] The term "enzyme", when used herein in connection with altering host 10 cell glycosylation, refers to a molecule having at least one enzymatic activity, and includes full-length enzymes, catalytically active fragments, chimerics, complexes, and the like. A "catalytically active fragment" of an enzyme refers to a polypeptide having a detectable level of functional (enzymatic) activity. [00731 A lower eukaryotic host cell, when used herein in connection with 15 glycosylation profiles, refers to any eukaryotic cell which ordinarily produces high -mannose containing N-glycans, and thus is meant to include some animal or plant cells and most typical lower eukaryotic cells, including uni- and multicellular fungal and algal cells. [0074] As used herein, the term "secretion pathway" refers to the assembly line 20 of various glycosylation enzymes to which a lipid-linked oligosaccharide precursor and an N-glycan substrate are sequentially exposed, following the molecular flow of a nascent polypeptide chain from the cytoplasm to the endoplasmic reticulum (ER) and the compartments of the Golgi apparatus. Enzymes are said to be localized along this pathway. An enzyme X that acts on a lipid-linked glycan or an 25 N-glycan before enzyme Y is said to be or to act "upstream" to enzyme Y; similarly, enzyme Y is or acts "downstream" from enzyme X [0075] The term "targeting peptide" as used herein refers to nucleotide or amino acid sequences encoding a cellular targeting signal peptide which mediates the localization (or retention) of an associated sequence to sub-cellular locations, e.g., 30 organelles. 10076] The term "polynucleotide" or "nucleic acid molecule" refers to a polymeric form of nucleotides of at least 10 bases in length. The term includes 26 DNA molecules (e.g., cDNA or genomic or synthetic DNA) and RNA molecules (e.g., mRNA or synthetic RNA), as well as analogs of DNA or RNA containing non-natural nucleotide analogs, non-native internucleoside bonds, or both. The nucleic acid can be in any topological conformation. For instance, the nucleic acid 5 can be single-stranded, double-stranded, triple-stranded, quadruplexed, partially double-stranded, branched, hairpinned, circular, or in a padlocked conformation. The term includes single and double stranded forms of DNA. A nucleic acid molecule of this invention may include both sense and antisense strands of RNA, cDNA, genomic DNA, and synthetic forms and mixed polymers of the above. 10 They may be modified chemically or biochemically or may contain non-natural or derivatized nucleotide bases, as will be readily appreciated by those of skill in the art. Such modifications include, for example, labels, methylation, substitution of one or more of the naturally occurring nucleotides with an analog, intemucleotide modifications such as uncharged linkages (e.g., methyl phosphonates, 15 phosphotriesters, phosphoramidates, carbamates, etc.), charged linkages (e.g., phosphorothioates, phosphorodithioates, etc.), pendent moieties (e.g., polypeptides), intercalators (e.g., acridine, psoralen, etc.), chelators, alkylators, and modified linkages (e.g., alpha anomeric nucleic acids, etc.) Also included are synthetic molecules that mimic polynucleotides in their ability to bind to a 20 designated sequence via hydrogen bonding and other chemical interactions. Such molecules are known in the art and include, for example, those in which peptide linkages substitute for phosphate linkages in the backbone of the molecule. [0077] Unless otherwise indicated, a "nucleic acid comprising SEQ ID NO:X" refers to a nucleic acid, at least a portion of which has either (i) the sequence of 25 SEQ ID NO:X, or (ii) a sequence complementary to SEQ ID NO:X. The choice between the two is dictated by the context. For instance, if the nucleic acid is used as a probe, the choice between the two is dictated by the requirement that the probe be complementary to the desired target. [0078] An "isolated" or "substantially pure" nucleic acid or polynucleotide (e.g., 30 an RNA, DNA or a mixed polymer) is one which is substantially separated from other cellular components that naturally accompany the native polynucleotide in its natural host cell, e.g., ribosomes, polymerases, and genomic sequences with which it is naturally associated. The term embraces a nucleic acid or polynucleotide that (1) has been removed from its naturally occurring environment, (2) is not associated with all or a portion of a polynucleotide in which the "isolated polynucleotide" is found in nature, (3) is operatively linked to a polynucleotide 5 which it is not linked to in nature, or (4) does not occur in nature. The term "isolated" or "substantially pure" also can be used in reference to recombinant or cloned DNA isolates, chemically synthesized polynucleotide analogs, or polynucleotide analogs that are biologically synthesized by heterologous systems. [0079] However, "isolated" does not necessarily require that the nucleic acid or 10 polynucleotide so described has itself been physically removed from its native environment. For instance, an endogenous nucleic acid sequence in the genome of an organism is deemed "isolated" herein if a heterologous sequence (i.e., a sequence that is not naturally adjacent to this endogenous nucleic acid sequence) is placed adjacent to the endogenous nucleic acid sequence, such that the expression 15 of this endogenous nucleic acid sequence is altered. By way of example, a non native promoter sequence can be substituted (e.g., by homologous recombination) for the native promoter of a gene in the genome of a human cell, such that this gene has an altered expression pattern. This gene would now become "isolated" because it is separated from at least some of the sequences that naturally flank it. 20 [0080] A nucleic acid is also considered "isolated" if it contains any modifications that do not naturally occur to the corresponding nucleic acid in a genome. For instance, an endogenous coding sequence is considered "isolated" if it contains an insertion, deletion or a point mutation introduced artificially, e.g., by human intervention. An "isolated nucleic acid" also includes a nucleic acid 25 integrated into a host cell chromosome at a heterologous site, a nucleic acid construct present as an episome. Moreover, an "isolated nucleic acid" can be substantially free of other cellular material, or substantially free of culture medium when produced by recombinant techniques, or substantially free of chemical precursors or other chemicals when chemically synthesized. 30 [0081] As used herein, the phrase "degenerate variant" of a reference nucleic acid sequence encompasses nucleic acid sequences that can be translated, 28 according to the standard genetic code, to provide an amino acid sequence identical to that translated from the reference nucleic acid sequence. [00821 The term "percent sequence identity" or "identical" in the context of nucleic acid sequences refers to the residues in the two sequences which are the 5 same when aligned for maximum correspondence. The length of sequence identity comparison may be over a stretch of at least about nine nucleotides, usually at least about 20 nucleotides, more usually at least about 24 nucleotides, typically at least about 28 nucleotides, more typically at least about 32 nucleotides, and preferably at least about 36 or more nucleotides. There are a number of different algorithms 10 known in the art which can be used to measure nucleotide sequence identity. For instance, polynucleotide sequences can be compared using FASTA, Gap or Bestfit, which are programs in Wisconsin Package Version 10.0, Genetics Computer Group (GCG), Madison, Wisconsin. FASTA provides alignments and percent sequence identity of the regions of the best overlap between the query and search 15 sequences (Pearson, 1990, herein incorporated by reference). For instance, percent sequence identity between nucleic acid sequences can be determined using FASTA with its default parameters (a word size of 6 and the NOPAM factor for the scoring matrix) or using Gap with its default parameters as provided in GCG Version 6.1, herein incorporated by reference. 20 [0083] The term "substantial homology" or "substantial similarity," when referring to a nucleic acid or fragment thereof, indicates that, when optimally aligned with appropriate nucleotide insertions or deletions with another nucleic acid (or its complementary strand), there is nucleotide sequence identity in at least about 50%, more preferably 60% of the nucleotide bases, usually at least about 25 70%, more usually at least about 80%, preferably at least about 90%, and more preferably at least about 95%, 96%, 97%, 98% or 99% of the nucleotide bases, as measured by any well-known algorithm of sequence identity, such as FASTA, BLAST or Gap, as discussed above. [0084] Alternatively, substantial homology or similarity exists when a nucleic 30 acid or fragment thereof hybridizes to another nucleic acid, to a strand of another nucleic acid, or to the complementary strand thereof, under stringent hybridization conditions. "Stringent hybridization conditions" and "stringent wash conditions" 29 in the context of nucleic acid hybridization experiments depend upon a number of different physical parameters. Nucleic acid hybridization will be affected by such conditions as salt concentration, temperature, solvents, the base composition of the hybridizing species, length of the complementary regions, and the number of 5 nucleotide base mismatches between the hybridizing nucleic acids, as will be readily appreciated by those skilled in the art. One having ordinary skill in the art knows how to vary these parameters to achieve a particular stringency of hybridization. [0085] In general, "stringent hybridization" is performed at about 25*C below the 10 thermal melting point (T,) for the specific DNA hybrid under a particular set of conditions. "Stringent washing" is performed at temperatures about 5*C lower than the Tm for the specific DNA hybrid under a particular set of conditions. The Tm is the temperature at which 50% of the target sequence hybridizes to a perfectly matched probe. See Sambrook et al., supra, page 9.51, hereby incorporated by 15 reference. For purposes herein, "high stringency conditions" are defined for solution phase hybridization as aqueous hybridization (i.e., free of formamide) in 6X SSC (where 20X SSC contains 3.0 M NaCI and 0.3 M sodium citrate), 1% SDS at 65'C for 8-12 hours, followed by two washes in 0.2X SSC, 0.1% SDS at 65'C for 20 minutes. It will be appreciated by the skilled artisan that hybridization at 20 65 0 C will occur at different rates depending on a number of factors including the length and percent identity of the sequences which are hybridizing. [0086] The term "mutated" when applied to nucleic acid sequences means that nucleotides in a nucleic acid sequence may be inserted, deleted or changed compared to a reference nucleic acid sequence. A single alteration may be made at 25 a locus (a point mutation) or multiple nucleotides may be inserted, deleted or changed at a single locus. In addition, one or more alterations may be made at any number of loci within a nucleic acid sequence. A nucleic acid sequence may be mutated by any method known in the art including but not limited to mutagenesis techniques such as "error-prone PCR" (a process for performing PCR under 30 conditions where the copying fidelity of the DNA polymerase is low, such that a high rate of point mutations is obtained along the entire length of the PCR product. See, e.g., Leung, D. W., et al., Technique, 1, pp. I1-15 (1989) and Caldwell, R. C. 30 & Joyce G. F., PCR Methods Applic., 2, pp. 28-33 (1992)); and "oligonucleotide directed mutagenesis" (a process which enables the generation of site-specific mutations in any cloned DNA segment of interest. See, e.g., Reidhaar-Olson, J. F. & Sauer, R. T., et al., Science, 241, pp. 53-57 (1988)). 5 [0087] The term "vector" as used herein is intended to refer to a nucleic acid molecule capable of transporting another nucleic acid to which it has been linked. One type of vector is a "plasmid", which refers to a circular double stranded DNA loop into which additional DNA segments may be ligated. Other vectors include cosmids, bacterial artificial chromosomes (BAC) and yeast artificial chromosomes 10 (YAC). Another type of vector is a viral vector, wherein additional DNA segments may be ligated into the viral genome (discussed in more detail below). Certain vectors are capable of autonomous replication in a host cell into which they are introduced (e.g., vectors having an origin of replication which functions in the host cell). Other vectors can be integrated into the genome of a host cell upon 15 introduction into the host cell, and are thereby replicated along with the host genome. Moreover, certain preferred vectors are capable of directing the expression of genes to which they are operatively linked. Such vectors are referred to herein as "recombinant expression vectors" (or simply, "expression vectors"). [0088] "Operatively linked" expression control sequences refers to a linkage in 20 which the expression control sequence is contiguous with the gene of interest to control the gene of interest, as well as expression control sequences that act in trans or at a distance to control the gene of interest. [0089] The term "expression control sequence" as used herein refers to polynucleotide sequences which are necessary to affect the expression of coding 25 sequences to which they are operatively linked. Expression control sequences are sequences which control the transcription, post-transcriptional events and translation of nucleic acid sequences. Expression control sequences include appropriate transcription initiation, termination, promoter and enhancer sequences; efficient RNA processing signals such as splicing and polyadenylation signals; 30 sequences that stabilize cytoplasmic mRNA; sequences that enhance translation efficiency (e.g., ribosome binding sites); sequences that enhance protein stability; and when desired, sequences that enhance protein secretion. The nature of such 31 control sequences differs depending upon the host organism; in prokaryotes, such control sequences generally include promoter, ribosomal binding site, and transcription termination sequence. The term "control sequences" is intended to include, at a minimum, all components whose presence is essential for expression, 5 and can also include additional components whose presence is advantageous, for example, leader sequences and fusion partner sequences. 10090] The term "recombinant host cell" (or simply "host cell"), as used herein, is intended to refer to a cell into which a nucleic acid such as a recombinant vector has been introduced. It should be understood that such terms are intended to refer 10 not only to the particular subject cell but to the progeny of such a cell. Because certain modifications may occur in succeeding generations due to either mutation or environmental influences, such progeny may not, in fact, be identical to the parent cell, but are still included within the scope of the term "host cell" as used herein. A recombinant host cell may be an isolated cell or cell line grown in 15 culture or may be a cell which resides in a living tissue or organism. [00911 The term "peptide" as used herein refers to a short polypeptide, e.g., one that is typically less than about 50 amino acids long and more typically less than about 30 amino acids long. The term as used herein encompasses analogs and mimetics that mimic structural and thus biological function. 20 [00921 The term "polypeptide" as used herein encompasses both naturally occurring and non-naturally-occurring proteins, and fragments, mutants, derivatives and analogs thereof. A polypeptide may be monomeric or polymeric. Further, a polypeptide may comprise a number of different domains each of which has one or more distinct activities. 25 [0093] The term "isolated protein" or "isolated polypeptide" is a protein or polypeptide that by virtue of its origin or source of derivation (1) is not associated with naturally associated components that accompany it in its native state, (2) when it exists in a purity not found in nature, where purity can be adjudged with respect to the presence of other cellular material (e.g., is free of other proteins from 30 the same species) (3) is expressed by a cell from a different species, or (4) does not occur in nature (e.g., it is a fragment of a polypeptide found in nature or it includes amino acid analogs or derivatives not found in nature or linkages other than 32 standard peptide bonds). Thus, a polypeptide that is chemically synthesized or synthesized in a cellular system different from the cell from which it naturally originates will be "isolated" from its naturally associated components. A polypeptide or protein may also be rendered substantially free of naturally 5 associated components by isolation, using protein purification techniques well known in the art. As thus defined, "isolated" does not necessarily require that the protein, polypeptide, peptide or oligopeptide so described has been physically removed from its native environment. [0094] The term "polypeptide fragment" as used herein refers to a polypeptide 10 that has an amino-terminal and/or carboxy-terminal deletion compared to a full length polypeptide. In a preferred embodiment, the polypeptide fragment is a contiguous sequence in which the amino acid sequence of the fragment is identical to the corresponding positions in the naturally-occurring sequence. Fragments typically are at least 5, 6, 7, 8, 9 or 10 amino acids long, preferably at least 12, 14, 15 16 or 18 amino acids long, more preferably at least 20 amino acids long, more preferably at least 25, 30, 35, 40 or 45, amino acids, even more preferably at least 50 or 60 amino acids long, and even more preferably at least 70 amino acids long. [0095] A "modified derivative" refers to polypeptides or fragments thereof that are substantially homologous in primary structural sequence but which include, 20 e.g., in vivo or in vitro chemical and biochemical modifications or which incorporate amino acids that are not found in the native polypeptide. Such modifications include, for example, acetylation, carboxylation, phosphorylation, glycosylation, ubiquitination, labeling, e.g., with radionuclides, and various enzymatic modifications, as will be readily appreciated by those well skilled in the 25 art. A variety of methods for labeling polypeptides and of substituents or labels useful for such purposes are well-known in the art, and include radioactive isotopes such as 125, 2 P, 3 1S, and 'H, ligands which bind to labeled antiligands (e.g., antibodies), fluorophores, chemiluminescent agents, enzymes, and antiligands which can serve as specific binding pair members for a labeled ligand. The choice 30 of label depends on the sensitivity required, ease of conjugation with the primer, stability requirements, and available instrumentation. Methods for labeling 33 polypeptides are well-known in the art. See Ausubel et al., 1992, hereby incorporated by reference. [00961 A "polypeptide mutant" or "mutein" refers to a polypeptide whose sequence contains an insertion, duplication, deletion, rearrangement or substitution 5 of one or more amino acids compared to the amino acid sequence of a native or wild type protein. A mutein may have one or more amino acid point substitutions, in which a single amino acid at a position has been changed to another amino acid, one or more insertions and/or deletions, in which one or more amino acids are inserted or deleted, respectively, in the sequence of the naturally-occurring protein, 10 and/or truncations of the amino acid sequence at either or both the amino or carboxy termini. A mutein may have the same but preferably has a different biological activity compared to the naturally-occurring protein. For instance, a mutein may have an increased or decreased neuron or NgR binding activity. In a preferred embodiment of the present invention, a MAG derivative that is a mutein 15 (e.g., in MAG Ig-like domain 5) has decreased neuronal growth inhibitory activity compared to endogenous or soluble wild-type MAG. [0097] A mutein has at least 70% overall sequence homology to its wild-type counterpart. Even more preferred are muteins having 80%, 85% or 90% overall sequence homology to the wild-type protein. In an even more preferred 20 embodiment, a mutein exhibits 95% sequence identity, even more preferably 97%, even more preferably 98% and even more preferably 99% overall sequence identity. Sequence homology may be measured by any common sequence analysis algorithm, such as Gap or Bestfit. [0098] Preferred amino acid substitutions are those which: (1) reduce 25 susceptibility to proteolysis, (2) reduce susceptibility to oxidation, (3) alter binding affinity for forming protein complexes, (4) alter binding affinity or enzymatic activity, and (5) confer or modify other physicochemical or functional properties of such analogs. [00991 As used herein, the twenty conventional amino acids and their 30 abbreviations follow conventional usage. See Immunology -A Synthesis ( 2 "d Edition, E.S. Golub and D.R. Gren, Eds., Sinauer Associates, Sunderland, Mass. (1991)), which is incorporated herein by reference. Stereoisomers (e.g., D-amino 34 acids) of the twenty conventional amino acids, unnatural amino acids such as a-, a-disubstituted amino acids, N-alkyl amino acids, and other unconventional amino acids may also be suitable components for polypeptides of the present invention. Examples of unconventional amino acids include: 4 -hydroxyproline, 5 y-carboxyglutamate, E-N,N,N-trimethyllysine, e-N-acetyllysine, 0-phosphoserine, N-acetylserine, N-formylmethionine, 3-methylhistidine, 5-hydroxylysine, s-N-methylarginine, and other similar amino acids and imino acids (e.g., 4-hydroxyproline). In the polypeptide notation used herein, the left-hand direction is the amino terminal direction and the right hand direction is the carboxy-terminal 10 direction, in accordance with standard usage and convention. [01001 A protein has "homology" or is "homologous" to a second protein if the nucleic acid sequence that encodes the protein has a similar sequence to the nucleic acid sequence that encodes the second protein. Alternatively, a protein has homology to a second protein if the two proteins have "similar" amino acid 15 sequences. (Thus, the term "homologous proteins" is defined to mean that the two proteins have similar amino acid sequences). In a preferred embodiment, a homologous protein is one that exhibits 60% sequence homology to the wild type protein, more preferred is 70% sequence homology. Even more preferred are homologous proteins that exhibit 80%, 85% or 90% sequence homology to the 20 wild type protein. In a yet more preferred embodiment, a homologous protein exhibits 95%, 97%, 98% or 99% sequence identity. As used herein, homology between two regions of amino acid sequence (especially with respect to predicted structural similarities) is interpreted as implying similarity in function. [0101] When "homologous" is used in reference to proteins or peptides, it is 25 recognized that residue positions that are not identical often differ by conservative amino acid substitutions. A "conservative amino acid substitution" is one in which an amino acid residue is substituted by another amino acid residue having a side chain (R group) with similar chemical properties (e.g., charge or hydrophobicity). In general, a conservative amino acid substitution will not substantially change the 30 functional properties of a protein. In cases where two or more amino acid sequences differ from each other by conservative substitutions, the percent sequence identity or degree of homology may be adjusted upwards to correct for 35 the conservative nature of the substitution. Means for making this adjustment are well known to those of skill in the art (see, e.g., Pearson et al., 1994, herein incorporated by reference). [0102] The following six groups each contain amino acids that are conservative 5 substitutions for one another: 1) Serine (S), Threonine (T); 2) Aspartic Acid (D), Glutarnic Acid (B); 3) Asparagine (N), Glutamine (Q); 4) Arginine (R), Lysine (K); 5) Isoleucine (1), Leucine (L), Methionine (M), Alanine (A), Valine (V), and 6) Phenylalanine (F), Tyrosine (Y), Tryptophan (W). [0103] Sequence homology for polypeptides, which is also referred to as percent 10 sequence identity, is typically measured using sequence analysis software. See, e.g., the Sequence Analysis Software Package of the Genetics Computer Group (GCG), University of Wisconsin Biotechnology Center, 910 University Avenue, Madison, Wisconsin 53705. Protein analysis software matches similar sequences using measure of homology assigned to various substitutions, deletions and other 15 modifications, including conservative amino acid substitutions. For instance, GCG contains programs such as "Gap" and "Bestfit" which can be used with default parameters to determine sequence homology or sequence identity between closely related polypeptides, such as homologous polypeptides from different species of organisms or between a wild type protein and a mutein thereof See, e.g., GCG 20 Version 6.1. [0104] A preferred algorithm when comparing a inhibitory molecule sequence to a database containing a large number of sequences from different organisms is the computer program BLAST (Altschul, S.F. et al. (1990) J Mol. Biol. 215:403-410; Gish and States (1993) Nature Genet. 3:266-272; Madden, T.L. et al. (1996) Meth. 25 Enzymnol. 266:131-141; Altschul, S.F. et al. (1997) Nucleic Acids Res.25:3389 3402; Zhang, J. and Madden, T.L. (1997) Genome Res. 7:649-656), especially blastp or tblastn (Altschul et al., 1997). Preferred parameters for BLASTp are: Expectation value: 10 (default); Filter: seg (default); Cost to open a gap: 11 (default); Cost to extend a gap: 1 (default); Max. alignments: 100 (default); Word 30 size: 11 (default); No. of descriptions: 100 (default); Penalty Matrix: BLOWSUM62. 36 [01051 The length of polypeptide sequences compared for homology will generally be at least about 16 amino acid residues, usually at least about 20 residues, more usually at least about 24 residues, typically at least about 28 residues, and preferably more than about 35 residues. When searching a database 5 containing sequences from a large number of different organisms, it is preferable to compare amino acid sequences. Database searching using amino acid sequences can be measured by algorithms other than blastp known in the art. For instance, polypeptide sequences can be compared using FASTA, a program in GCG Version 6.1. FASTA provides alignments and percent sequence identity of the regions of 10 the best overlap between the query and search sequences (Pearson, 1990, herein incorporated by reference). For example, percent sequence identity between amino acid sequences can be determined using FASTA with its default parameters (a word size of 2 and the PAM250 scoring matrix), as provided in GCG Version 6.1, herein incorporated by reference. 15 [01061 The term "fusion protein" refers to a polypeptide comprising a polypeptide or fragment coupled to heterologous amino acid sequences. Fusion proteins are useful because they can be constructed to contain two or more desired functional elements from two or more different proteins. A fusion protein comprises at least 10 contiguous amino acids from a polypeptide of interest, more 20 preferably at least 20 or 30 amino acids, even more preferably at least 40, 50 or 60 amino acids, yet more preferably at least 75, 100 or 125 amino acids. Fusion proteins can be produced recombinantly by constructing a nucleic acid sequence which encodes the polypeptide or a fragment thereof in-frame with a nucleic acid sequence encoding a different protein or peptide and then expressing the fusion 25 protein. Alternatively, a fusion protein can be produced chemically by crosslinking the polypeptide or a fragment thereof to another protein. [0107] The term "region" as used herein refers to a physically contiguous portion of the primary structure of a biomolecule. In the case of proteins, a region is defined by a contiguous portion of the amino acid sequence of that protein. 30 [01081 The term "domain" as used herein refers to a structure of a biomolecule that contributes to a known or suspected function of the biomolecule. Domains may be co-extensive with regions or portions thereof; domains may also include 37 distinct, non-contiguous regions of a biomolecule. Examples of protein domains include, but are not limited to, an Ig domain, an extracellular domain, a transmembrane domain, and a cytoplasmic domain. [0109] As used herein, the tern "molecule" means any compound, including, but 5 not limited to, a small molecule, peptide, protein, sugar, nucleotide, nucleic acid, lipid, etc., and such a compound can be natural or synthetic. [0110] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Exemplary methods and materials are described 10 below, although methods and materials similar or equivalent to those described herein can also be used in the practice of the present invention and will be apparent to those of skill in the art. All publications and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control. The materials, methods, and 15 examples are illustrative only and not intended to be limiting. [01111 Throughout this specification and claims, the word "comprise" or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated integer or group of integers but not the exclusion of any other integer or group of integers. 20 Methods For Producing Host Cells Having MansGlcNAc 2 Modified Oligosaccharides For The Generation of Human-like N-Glycans [0112] The invention provides a method for producing a glycoprotein having 25 human-like glycosylation in a non-human eukaryotic host cell. As described in more detail below, a eukaryotic host cell that does not naturally express, or which is engineered not to express, one or more enzymes involved in production of high mannose structures is selected as a starting host cell. Such a selected host cell is engineered to express one or more enzymes or other factors required to produce 30 human-like glycoproteins. A desired host strain can be engineered one enzyme or more than one enzyme at a time. In addition, a nucleic acid molecule encoding one or more enzymes or activities may be used to engineer a host strain of the invention. Preferably, a library of nucleic acid molecules encoding potentially 38 useful enzymes (e.g., chimeric enzymes comprising a catalytically active enzyme fragment ligated in-frame to a heterologous subcellular targeting sequence) is created (e.g., by ligation of sub-libraries comprising enzymatic fragments and subcellular targeting sequences), and a strain having one or more enzymes with 5 optimal activities or producing the most "human-like" glycoproteins may be selected by transforming target host cells with one or more members of the library. [0113] In particular, the methods described herein enable one to obtain, in vivo, MansGIcNAc 2 structures in high yield, at least transiently, for the purpose of further modifying it to yield complex N-glycans. A successful scheme to obtain 10 suitable MansGlcNAc 2 structures in appropriate yields in a host cell, such as a lower eukaryotic organism, generally involves two parallel approaches: (1) reducing high mannose structures made by endogenous mannosyltransferse activities, if any, and (2) removing 1,2- a- mannose by mannosidases to yield high levels of suitable MansGlcNAc 2 structures which may be further reacted inside the 15 host cell to form complex, human-like glycoforms. [01141 Accordingly, a first step involves the selection or creation of a eukaryotic host cell, e.g., a lower eukaryote, capable of producing a specific precursor structure of MansGlcNAc 2 that is able to accept in vivo GlcNAc by the action of a GlcNAc transferase I ("GnTI"). In one embodiment, the method involves making 20 or using a non-human eukaryotic host cell depleted in a 1,6 mannosyltransferase activity with respect to the N-glycan on a glycoprotein. Preferably, the host cell is depleted in an initiating 1,6 mannosyltransferase activity (see below). Such a host cell will lack one or more enzymes involved in the production of high mannose structures which are undesirable for producing human-like glycoproteins. 25 [0115] One or more enzyme activities are then introduced into such a host cell to produce N-glycans within the host cell characterized by having at least 30 mol % of the MansGlcNAc 2 (Mans") carbohydrate structures. MansGlcNAc 2 structures are necessary for complex N-glycan formation: MansGlcNAc 2 must be formed in vivo in a high yield (e.g., in excess of 30%), at least transiently, as subsequent 30 mammalian- and human-like glycosylation reactions require MansGlcNAc 2 or a derivative thereof. 39 [0116] This step also requires the formation of a particular isomeric structure of MansGlcNAc 2 within the cell at a high yield. While MansGlcNAc 2 structures are necessary for complex N-glycan formation, their presence is by no means sufficient. That is because MansGlcNAc 2 may occur in different isomeric forms, 5 which may or may not serve as a substrate for GlcNAc transferase I. As most glycosylation reactions are not complete, a particular glycosylated protein generally contains a range of different carbohydrate structures (i.e. glycoforms) on its surface. Thus, the mere presence of trace amounts (i.e., less than 5%) of a particular structure like MansGlcNAc 2 is of little practical relevance for producing 10 mammalian- or human-like glycoproteins. It is the formation of a GlcNAc transferase I-accepting MansGlcNAc 2 intermediate (Fig. 1B) in high yield (i.e., above 30%), which is required. The formation of this intermediate is necessary to enable subsequent in vivo synthesis of complex N-glycans on glycosylated proteins of interest (target proteins). 15 [0117] Accordingly, some or all of the MansGlcNAc 2 produced by the selected host cell must be a productive substrate for enzyme activities along a mammalian glycosylation pathway, e.g., can serve as a substrate for a GlcNAc transferase I activity in vivo, thereby forming the human-like N-glycan intermediate GlcNAcMansGlcNAc 2 in the host cell. In a preferred embodiment, at least 10%, 20 more preferably at least 30% and most preferably 50% or more of the MansGlcNAc 2 intermediate produced in the host cell of the invention is a productive substrate for GnTI in vivo. It is understood that if, for example, GlcNAcMansGlcNAc 2 is produced at 10% and MansGlcNAc 2 is produced at 25% on a target protein, that the total amount of transiently produced MansGlcNAc2 is 25 35% because GlcNAcMansGlcNAc2 is a product of MansGlcNAc 2 . [01181 One of ordinary skill in the art can select host cells from nature, e.g., existing fungi or other lower eukaryotes that produce significant levels of MansGlcNAc 2 in vivo. As yet, however, no lower eukaryote has been shown to provide such structures in vivo in excess of 1.8% of the total N-glycans (see e.g. 30 Maras et al., 1997). Alternatively, such host cells may be genetically engineered to produce the MansGlcNAc 2 structure in vivo. Methods such as those described in U.S. Patent No. 5,595,900 may be used to identify the absence or presence of 40 particular glycosyltransferases, mannosidases and sugar nucleotide transporters in a target host cell or organism of interest. Inactivation of Undesirable Host Cell Glycosylation Enzymes 5 [0119] The methods of the invention are directed to making host cells which produce glycoproteins having altered, and preferably human-like, N-glycan structures. In a preferred embodiment the methods are directed to making host cells in which oligosaccharide precursors are enriched in MansGlcNAc 2 . Preferably, a eukaryotic host cell is used that does not express one or more 10 enzymes involved in the production of high mannose structures. Such a host cell may be found in nature or may be engineered, e.g., starting with or derived from one of many such mutants already described in yeasts. Thus, depending on the selected host cell, one or a number of genes that encode enzymes known to be characteristic of non-human glycosylation reactions will have to be deleted. Such 15 genes and their corresponding proteins have been extensively characterized in a number of lower eukaryotes (e.g., S.cerevisiae, T. reesei, A. nidulans etc.), thereby providing a list of known glycosyltransferases in lower eukaryotes, their activities and their respective genetic sequence. These genes are likely to be selected from the group of mannosyltransferases e.g. 1,3 mannosyltransferases (e.g. MNN1 in 20 S.cerevisiae) (Graham, 1991), 1,2 mannosyltransferases (e.g. KTR/KE family from S.cerevisiae), 1,6 mannosyltransferases (OCHI from S.cerevisiae), mannosylphosphate transferases and their regulators (MNN4 and MNN6 from S.cerevisiae) and additional enzymes that are involved in aberrant, i.e. non human, glycosylation reactions. Many of these genes have in fact been deleted 25 individually giving rise to viable phenotypes with altered glycosylation profiles. Examples are shown in Table 1. [0120] Preferred lower eukaryotic host cells of the invention, as described herein to exemplify the required manipulation steps, are hypermannosylation-minus (ochl) mutants ofPichia pastoris or Klactis. Like other lower eukaryotes, 30 P.pastoris processes Man 9 GlcNAc 2 structures in the ER with an -1,2 mannosidase to yield ManaGlcNAc 2 (Fig. 1A). Through the action of several mannosyltransferases, this structure is then converted to hypermannosylated 41 structures (ManaGlcNAc 2 ), also known as mannans. In addition, it has been found that P.pastoris is able to add non-terminal phosphate groups, through the action of mannosylphosphate transferases, to the carbohydrate structure. This differs from the reactions performed in mammalian cells, which involve the 5 removal rather than addition of mannose sugars. It is of particular importance to eliminate the ability of the eukaryotic host cell, e.g., fungus, to hypermannosylate an existing MansGlcNAc 2 structure. This can be achieved by either selecting for a host cell that does not hypermannosylate or by genetically engineering such a cell. [01211 Genes that are involved in the hypermannosylation process have been 10 identified, e.g., in Pichia pastors, and by creating mutations in these genes, one can reduce the production of "undesirable" glycoforms. Such genes can be identified by homology to existing mannosyltransferases or their regulators (e.g., OCH1, MNN4, MNN6, MNN]) found in other lower eukaryotes such as C. albicans, Pichia angusta or S. cerevisiae or by mutagenizing the host strain and 15 selecting for a glycosylation phenotype with reduced mannosylation. Based on homologies amongst known mannosyltransferases and mannosylphosphate transferases, one may either design PCR primers (SEQ ID NOS: 7,8, 47 and 4 left to right, respectively, in order of appearance) (examples of which are shown in Table 2), or use genes or gene fragments encoding such enzymes as probes to 20 identify homologs in DNA libraries of the target or a related organism. Alternatively, one may identify a functional homolog having mannosyltransferase activity by its ability to complement particular glycosylation phenotypes in related organisms. 25 Table 2. PCR Primers PCRrimA Target Gene(s) in Homoloes ATGGCGAAGGCAGA TTAGTCCTTCCAAC 1,6- 0 iae, TGGCAGT TTCCTTC manosyltransfe Pchia albicans TAYTGGMGNGTNGA GCRTCNCCCANCK 1,2 KTR--- faily, RCYNGAYATHAA YTCRTA manosyltrnferases S.cerevisiae Legend- M= A or C, R A or G, W A or T, S = C or G Y =-C or T, K = G or T, V = A or C or G, H CA or C or T, D - A or G or T, B =C or 0 or T, N G4 or A or T or c. 30 42 [01221 To obtain the gene or genes encoding 1, 6 -mannosyltransferase activity in P. pastoris, for example, one would carry out the following steps: OCH1 mutants of S.cerevisiae are temperature sensitive and are slow growers at elevated temperatures. One can thus identify functional homologs of OCH1 in Ppastoris 5 by complementing an OCHJ mutant of S.cerevisiae with a P.pastoris DNA or cDNA library. Mutants of S.cerevisiae are available, e.g., from Stanford University and are commercially available from ResGen, an Invitrogen Corp. (Carlsbad, CA). Mutants that display a normal growth phenotype at elevated temperature, after having been transformed with a P.pastoris DNA library, are 10 likely to carry an OCH1 homolog of P. pastoris. Such a library can be created by partially digesting chromosomal DNA of P.pastoris with a suitable restriction enzyme and, after inactivating the restriction enzyme, ligating the digested DNA into a suitable vector, which has been digested with a compatible restriction enzyme. 15 [0123] Suitable vectors include, e.g., pRS314, a low copy (CEN6/ARS4) plasmid based on pBluescript containing the. Trp1 marker (Sikorski, R. S., and Hieter, P.,1989, Genetics 122, pg 19-27) and pFL44S, a high copy (2g) plasmid based on a modified pUC19 containing the URA3 marker (Bonneaud, N., et al., 1991, Yeast 7, pg. 609-615). Such vectors are commonly used by academic researchers and 20 similar vectors are available from a number of different vendors (e.g., Invitrogen (Carlsbad, CA); Pharmacia (Piscataway, NJ); New England Biolabs (Beverly, MA)). Further examples include pYES/GS, 2si origin of replication based yeast expression plasmid from Invitrogen, or Yep24 cloning vehicle from New England Biolabs. 25 [01241 After ligation of the chromosomal DNA and the vector, one may transform the DNA library into a strain of S.cerevisiae with a specific mutation and select for the correction of the corresponding phenotype. After sub-cloning and sequencing the DNA fragment that is able to restore the wild-type phenotype, one may use this fragment to eliminate the activity of the gene product encoded by 30 OCHI in P.pastoris using in vivo mutagenesis and/or recombination techniques well-known to those skilled in the art. 43 [0125] Alternatively, if the entire genomic sequence of a particular host cell, e.g., fungus, of interest is known, one may identify such genes simply by searching publicly available DNA databases, which are available from several sources, such as NCBI, Swissprot. For example, by searching a given genomic sequence or 5 database with sequences from a known 1,6 mannosyltransferase gene (e.g., OCH1 from Scerevisiae), one can identify genes of high homology in such a host cell genome which may (but do not necessarily) encode proteins that have 1,6 mannosyltransferase activity. Nucleic acid sequence homology alone is not enough to prove, however, that one has identified and isolated a homolog encoding 10 an enzyme having the same activity. To date, for example, no data exist to show that an OCHJ deletion in P.pastoris eliminates the crucial initiating 1,6 mannosyltransferase activity. (Martinet et al. Biotech. Letters 20(12) (Dec. 1998): 1171-1177; Contreras et al. WO 02/00856 A2). Thus, no data prove that the P.pastoris OCHJ gene homolog actually encodes that function. That 15 demonstration is provided for the first time herein. [0126] Homologs to several S.cerevisiae mannosyltransferases have been identified in P.pastoris using these approaches. Homologous genes often have similar functions to genes involved in the mannosylation of proteins in S.cerevisiae and thus their deletion may be used to manipulate the glycosylation pattern in 20 P.pastoris or, by analogy, in any other host cell, e.g., fungus, plant, insect or animal cells, with similar glycosylation pathways. [01271 The creation of gene knock-outs, once a given target gene sequence has been determined, is a well-established technique in the art and can be carried out by one of ordinary skill in the art (see, e.g., R. Rothstein, (1991) Methods in 25 Enzymology, vol. 194, p. 281). The choice of a host organism maybe influenced by the availability of good transformation and gene disruption techniques. [0128] If several mannosyltransferases are to be knocked out, the method developed by Alani and Kleckner, (Genetics 116:541-545 (1987)), for example, enables the repeated use of a selectable marker, e.g., the URA3 marker in yeast, to 30 sequentially eliminate all undesirable endogenous mannosyltransferase activity. This technique has been refined by others but basically involves the use of two repeated DNA sequences, flanking a counter selectable marker. For example: 44 URA3 may be used as a marker to ensure the selection of a transformants that have integrated a construct. By flanking the URA3 marker with direct repeats one may first select for transformants that have integrated the construct and have thus disrupted the target gene. After isolation of the transformants, and their 5 characterization, one may counter select in a second round for those that are resistant to 5-fluoroorotic acid (5-FOA). Colonies that are able to survive on plates containing 5-FOA have lost the URA3 marker again through a crossover event involving the repeats mentioned earlier. This approach thus allows for the repeated use of the same marker and facilitates the disruption of multiple genes without 10 requiring additional markers. Similar techniques for sequential elimination of genes adapted for use in another eukaryotic host cells with other selectable and counter-selectable markers may also be used. [0129] Eliminating specific mannosyltransferases, such as 1,6 mannosyltransferase (OCH1) or mannosylphosphate transferases (MNN6, or genes 15 complementing lbd mutants) or regulators (MNN4) in P.pastoris enables one to create engineered strains of this organism which synthesize primarily Man8GlcNAc 2 and which can be used to further modify the glycosylation pattern to resemble more complex glycoform structures, e.g., those produced in mammalian, e.g., human cells. A preferred embodiment of this method utilizes DNA sequences 20 encoding biochemical glycosylation activities to eliminate similar or identical biochemical functions in P. pastoris to modify the glycosylation structure of glycoproteins produced in the genetically altered P.pastoris strain. [0130] Methods used to engineer the glycosylation pathway in yeasts as exemplified herein can be used in filamentous fungi to produce a preferred 25 substrate for subsequent modification. Strategies for modifying glycosylation pathways in A.niger and other filamentous fungi, for example, can be developed using protocols analogous to those described herein for engineering strains to produce human-like glycoproteins in yeast. Undesired gene activities involved in 1,2 mannosyltransferase activity, e.g., KTR/KRE homologs, are modified or 30 eliminated. A filamentous fungus, such as Aspergillus, is a preferred host because it lacks the 1,6 mannosyltransferase activity and as such, one would not expect a hypermannosylating gene activity, e.g. OCH1, in this host. By contrast, other 45 desired activities (e.g., a-1,2-mannosidase, UDP-GlcNAc transporter, glycosyltransferase (GnT), galactosyltransferase (GaIT) and sialyltransferase (ST)) involved in glycosylation are introduced into the host using the targeting methods of the invention. 5 Engineering or Selecting Hosts Having Diminished Initiating ot-1,6 Mannosyltransferase Activity [0131] In a preferred embodiment, the method of the invention involves making or using a host cell which is diminished or depleted in the activity of an initiating 10 ca-1, 6 -mannosyltransferase, i.e., an initiation specific enzyme that initiates outer chain mannosylation on the a-1,3 arm of the Man 3 GlcNAc 2 core structure. In S.cerevisiae, this enzyme is encoded by the OCH gene. Disruption of the OCHI gene in S.cerevisiae results in a phenotype in which N-linked sugars completely lack the poly-mannose outer chain. Previous approaches for obtaining 15 mammalian-type glycosylation in fungal strains have required inactivation of OCH1 (see, e.g., Chiba, 1998). Disruption of the initiating a-1,6 mannosyltransferase activity in a host cell of the invention may be optional, however (depending on the selected host cell), as the Ochlp enzyme requires an intact Man 8 GlcNAc 2 for efficient mannose outer chain initiation. Thus, host cells 20 selected or produced according to this invention which accumulate oligosaccharides having seven or fewer mannose residues may produce hypoglycosylated N-glycans that will likely be poor substrates for Ochip (see, e.g., Nakayama, 1997). [0132] The OCH1 gene was cloned from P.pastoris (Example 1) and K.lactis 25 (Example 16), as described. The nucleic acid and amino acid sequences of the OCH1 gene from Klactis are set forth in SEQ ID NOS: 41 and 42. Using gene specific primers, a construct was made from each clone to delete the OCH gene from the genome of P.pastoris and Klactis (Examples I and 16, respectively). Host cells depleted in initiating a-1, 6 -mannosyltransferase activity and engineered 30 to produce N-glycans having a MansGlcNAc 2 carbohydrate structure were thereby obtained (see, e.g., Figs. 5 and 6; Examples 11 and 16). [0133] Thus, in another embodiment,.the invention provides an isolated nucleic acid molecule having a nucleic acid sequence comprising or consisting of at least 46 forty-five, preferably at least 50, more preferably at least 60 and most preferably 75 or more nucleotide residues of the Kilactis OCH1 gene (SEQ ID NO: 41), and homologs, variants and derivatives thereof The invention also provides nucleic acid molecules that hybridize under stringent conditions to the above-described 5 nucleic acid molecules. Similarly, isolated polypeptides (including muteins, allelic variants, fragments, derivatives, and analogs) encoded by the nucleic acid molecules of the invention are provided. Also provided are vectors, including expression vectors, which comprise the above nucleic acid molecules of the invention, as described further herein. Similarly, host cells transformed with the 10 nucleic acid molecules or vectors of the invention are provided. Host Cells of the Invention [0134] A preferred host cell of the invention is a lower eukaryotic cell, e.g., yeast, a unicellular and multicellular or filamentous fungus. However, a wide 15 variety of host cells are envisioned as being useful in the methods of the invention. Plant cells or insect cells, for instance, may be engineered to express a human-like glycoprotein according to the invention (Examples 17 and 18). Likewise, a variety of non-human, mammalian host cells may be altered to express more human-like or otherwise altered glycoproteins using the methods of the invention. 20 As'one of skill in the art will appreciate, any eukaryotic host cell (including a human cell) may be used in conjunction with a library of the invention to express one or more chimeric proteins which is targeted to a subcellular location, e.g., organelle, in the host cell where the activity of the protein is modified, and preferably is enhanced. Such a protein is preferably - but need not necessarily be 25 - an enzyme involved in protein glycosylation, as exemplified herein. It is envisioned that any protein coding sequence may be targeted and selected for modified activity in a eukaryotic host cell using the methods described herein. [0135] Lower eukaryotes that are able to produce glycoproteins having the attached N-glycan MansGlcNAc 2 are particularly useful because (a) lacking a high 30 degree of mannosylation (e.g. greater than 8 mannoses per N-glycan, or especially 30-40 mannoses), they show reduced immunogenicity in humans; and (b) the N-glycan is a substrate for further glycosylation reactions to form an even more 47 human-like glycoform, e.g., by the action of GIcNAc transferase I (Fig. IB; #l1,2 GnTI) to form GlcNAcMan 5 GlcNAc 2 . A yield is obtained of greater than 30 mole %, more preferably a yield of 50-100 mole %, glycoproteins with N-glycans having a Man 5 GlcNAc 2 structure. In a preferred embodiment, more than 50% of 5 the Man 5 GlcNAc 2 structure is shown to be a substrate for a GnTI activity and can serve as such a substrate in vivo. [0136] Preferred lower eukaryotes of the invention include but are not limited to: Pichia pastoris, Pichiafinlandica, Pichia trehalophila, Pichia koclanae, Pichia membranaefaciens, Pichia opuntiae, Pichia thermotolerans, Pichia salictaria, 10 Pichia guercuum, Pichia pjperi, Pichia stiptis, Pichia methanolica, Pichia sp., Saccharomyces cerevisiae, Saccharomyces sp., Hansenula polymorpha, Kluyveromyces sp., Kluyveromyces lactis, Candida albicans, Aspergillus nidulans, Aspergillus niger, Aspergillus oryzae, Trichoderma reseei, Chrysosporiun lucknowense, Fusariun sp. Fusarium gramineum, Fusarium venenatum and 15 Neurospora crassa. [0137] In each above embodiment, the method is directed to making a host cell in which the oligosaccharide precursors are enriched in MansGlcNAc 2 . These structures are desirable because they may then be processed by treatment in vitro, for example, using the method of Maras and Contreras, U.S. Patent No. 5,834,251. 20 In a preferred embodiment, however, precursors enriched in MansGIcNAc 2 are processed by at least one further glycosylation reaction in vivo - with glycosidases (e.g., a-mannosidases) and glycosyltransferases (e.g., GnTI) - to produce human like N-glycans. Oligosaccharide precursors enriched in Man 5 GlcNAc 2 , for example, are preferably processed to those having GlcNAcManxGlcNAc 2 core 25 structures, wherein X is 3, 4 or 5, and is preferably 3. N-glycans having a GlcNAcManxGlcNAc 2 core structure where X is greater than 3 may be converted to GIcNAcMan3GlcNAc 2 , e.g., by treatment with an ce-1,3 and/or G-1,6 mannosidase activity, where applicable. Additional processing of GlcNAcMan 3 GlcNAc 2 by treatment with glycosyltransferases (e.g., GnTII) 30 produces GIcNAc 2 Man3GlcNAc 2 core structures which may then be modified, as desired, e.g., by ex vivo treatment or by heterologous expression in the host cell of 48 additional glycosylation enzymes, including glycosyltransferases, sugar transporters and mannosidases (see below), to become human-like N-glycans. [0138] Preferred human-like glycoproteins which may be produced according to the invention include those which comprise N-glycans having seven or fewer, or 5 three or fewer, mannose residues; and which comprise one or more sugars selected from the group consisting of galactose, GlcNAc, sialic acid, and fucose. Formation of complex N-glycans [0139] Formation of complex N-glycan synthesis is a sequential process by 10 which specific sugar residues are-removed and attached to the core oligosaccharide structure. In higher eukaryotes, this is achieved by having the substrate sequentially exposed to various processing enzymes. These enzymes carry out specific reactions depending on their particular location within the entire processing cascade. This "assembly line" consists of ER, early, medial and late 15 Golgi, and the trans Golgi network all with their specific processing environment. To re-create the processing of human glycoproteins in the Golgi and ER of lower eukaryotes, numerous enzymes (e.g. glycosyltransferases, glycosidases, phosphatases and transporters) have to be expressed and specifically targeted to these organelles, and preferably, in a location so that they function most efficiently 20 in relation to their environment as well as to other enzymes in the pathway. [0140] Because one goal of the methods described herein is to achieve a robust protein production strain that is able to perform well in an industrial fermentation process, the integration of multiple genes into the host cell chromosome involves careful planning. As described above, one or more genes which encode enzymes 25 known to be characteristic of non-human glycosylation reactions are preferably deleted. The engineered cell strain is transformed with a range of different genes encoding desired activities, and these genes are transformed in a stable fashion, thereby ensuring that the desired activity is maintained throughout the fermentation process. 30 10141] Any combination of the following enzyme activities may be engineered singly or multiply into the host using methods of the invention: sialyltransferases, mannosidases, fucosyltransferases, galactosyltransferases, GlcNAc transferases, 49 ER and Golgi specific transporters (e.g. syn- and antiport transporters for UDP galactose and other precursors), other enzymes involved in the processing of oligosaccharides, and enzymes involved in the synthesis of activated oligosaccharide precursors such as UDP-galactose and CMP-N-acetylneuraminic 5 acid. Preferably, enzyme activities are introduced on one or more nucleic acid molecules (see also below). Nucleic acid molecules may be introduced singly or multiply, e.g., in the context of a nucleic acid library such as a combinatorial library of the invention. It is to be understood, however, that single or multiple enzymatic activities may be introduced into a host cell in any fashion, including 10 but not limited to protein delivery methods and/or by use of one or more nucleic acid molecules without necessarily using a nucleic acid library or combinatorial library of the invention. Expression OfGlycsyltransferases To Produce Complex N-a lycans: 15 [0142] With DNA sequence information, the skilled artisan can clone DNA molecules encoding GnT activities (e.g., Examples 3 and 4). Using standard techniques well-known to those of skill in the art, nucleic acid molecules encoding GnTI, IL, III, IV or V (or encoding catalytically active fragments thereof) may be inserted into appropriate expression vectors under the transcriptional control of 20 promoters and other expression control sequences capable of driving scription in a selected host cell of the invention, e.g., a fungal host such as Pichia sp., Kluyveromyces sp. and Aspergillus sp., as described herein, such that one or more of these mammalian OnT enzymes may be actively expressed in a host cell of choice for production of a human-like complex glycoprotein (e.g., Examples 15, 25 17 and 18). [01431 . Several individual glycosyltransferases have been cloned and expressed in S.cerevisiae (GaiT, GnTI), Aspergillus nidulans (GnTI) and other fungi, without however demonstrating the desired outcome of "humanization" on the glycosylation pattern of the organisms (Yoshida, 1995; Schwientek, 1995; Kalsner 30 1995). It was speculated that the carbohydrate structure required to accept sugars by the action of such glycosyltransferases was not present in sufficient amounts, which most likely contributed to the lack of complex N-glycan formation. 50 [0144] A preferred method of the invention provides the functional expression of a GnT, such as GnTI, in the early or medial Golgi apparatus as well as ensuring a sufficient supply of UDP-GlcNAc (e.g., by expression of a UDP-GlcNAc transporter; see below). 5 Methods for Providing Sugar Nucleotide Precursors to the G oli Apparatus: [0145] For a glycosyltransferase to function satisfactorily in the Golgi, the enzyme requires a sufficient concentration of an appropriate nucleotide sugar, which is the high-energy donor of the sugar moiety added to a nascent 10 glycoprotein. In humans,.the full range of nucleotide sugar precursors (e.g. UDP N-acetylglucosamine, UDP-N-acetylgalactosamine, CMP-N-acetylneuraminic acid, UDP-galactose, etc.) are generally synthesized in the cytosol and transported into the Golgi, where they are attached to the core oligosaccharide by glycosyltransferases. 15 [01461 To replicate this process in non-human host cells such as lower eukaryotes, sugar nucleoside specific transporters have to be expressed in the Golgi to ensure adequate levels of nucleoside sugar precursors (Sommers, 1981; Sommers, 1982; Perez, 1987). Nucleotide sugars may be provided to the appropriate compartments, e.g., by expressing in the host microorganism an 20 exogenous gene encoding a sugar nucleotide transporter. The choice of transporter enzyme is influenced by the nature of the exogenous glycosyltransferase being used. For example, a GlcNAc transferase may require a UDP-GlcNAc transporter, a fucosyltransferase may require a GDP-fucose transporter, a galactosyltransferase may require a UDP-galactose transporter, and a sialyltransferase may require a 25 CMP-sialic acid transporter. [0147] The added transporter protein conveys a nucleotide sugar from the cytosol into the Golgi apparatus, where the nucleotide sugar may be reacted by the glycosyltransferase, e.g. to elongate an N-glycan. The reaction liberates a nucleoside diphosphate or monophosphate, e.g. UDP, GDP, or CMP. Nucleoside 30 monophosphates can be directly exported from the Golgi in exchange for nucleoside triphosphate sugars by an antiport mechanism. Accumulation of a nucleoside diphosphate, however, inhibits the further activity of a 51 glycosyltransferase. As this reaction appears to be important for efficient glycosylation, it is frequently desirable to provide an expressed copy of a gene encoding a nucleotide diphosphatase. The diphosphatase (specific for UDP or GDP as appropriate) hydrolyzes the diphosphonucleoside to yield a nucleoside 5 monosphosphate and inorganic phosphate. [0148] Suitable transporter enzymes, which are typically of mammalian origin, are described below. Such enzymes may be engineered into a selected host cell using the methods of the invention (see also Examples 7-10). [01491 In another example, a 2,3- or a 2

,

6 -sialyltransferase caps galactose 10 residues with sialic-acid in the trans-Golgi and TGN of humans leading to a mature form of the glycoprotein (Fig. IB). To reengineer this processing step into a metabolically engineered yeast or fungus will require (1) a 2,3- or a 2,6 sialyltransferase activity and (2) a sufficient supply of CMP-N-acetyl neuraminic acid, in the late Golgi of yeast (Example 6). To obtain sufficient a 2,3 15 sialyltransferase activity in the late Golgi, for example, the catalytic domain of a known sialyltransferase (e.g. from humans) has to be directed to the late Golgi in fungi (see above). Likewise, transporters have to be engineered to allow the transport of CMP-N-acetyl neuraminic acid into the late Golgi. There is currently no indication that fungi synthesize or can even transport sufficient amounts of 20 CMP-N-acetyl neuraminic acid into the Golgi. Consequently, to ensure the adequate supply of substrate for the corresponding glycosyltransferases, one has to metabolically engineer the production of CMP-sialic acid into the fingus. UDP-N-acetylglucosamine 25 [0150] The cDNA of human UDP-N-acetylglucosamine transporter, which was recognized through a homology search in the expressed sequence tags database (dbEST), has been cloned (Ishida, 1999 J. Biochen. 126(1): 68-77). The mammalian Golgi membrane transporter for TJDP-N-acetylglucosamine was cloned by phenotypic correction with cDNA from canine kidney cells (MDCK) of 30 a recently characterized Iluyveromyces lactis mutant deficient in Golgi transport of the above nucleotide sugar (Guillen, 1998). Results demonstrate that the mammalian Golgi UDP-GlcNAc transporter gene has all of the necessary 52 information for the protein to be expressed and targeted functionally to the Golgi apparatus of yeast and that two proteins with very different amino acid sequences may transport the same solute within the same Golgi membrane (Guillen, 1998). [0151] Accordingly, one may incorporate the expression of a UDP-GlcNAc 5 transporter in a host cell by means of a nucleic acid construct which may contain, for example: (1) a region by which the transformed construct is maintained in the cell (e.g. origin of replication or a region that mediates chromosomal integration), (2) a marker gene that allows for the selection of cells that have been transformed, including counterselectable and recyclable markers such as ura3 or T-urfl3 10 (Soderholm, 2001) or other well characterized selection-markers (e.g., his4, bla, Sh ble etc.), (3) a gene or fragment thereof encoding a functional UDP-GlcNAc transporter (e.g. from K lactis, (Abeijon, (1996) Proc. Natl. Acad. Sci. US.A. 93:5963-5968), or from H.sapiens (Ishida, 1996), and (4) a promoter activating the expression of the above mentioned localization/catalytic domain fusion construct 15 library. GDP-Fucose [0152] The rat liver Golgi membrane GDP-fucose transporter has been identified and purified by Puglielli, L. and C. B. Hirschberg (Puglielli, 1999 J Bio. Chem. 20 274(50):35596-35600). The corresponding gene has not been identified, however, N-terminal sequencing can be used for the design of oligonucleotide probes specific for the corresponding gene. These oligonucleotides can be used as probes to clone the gene encoding for GDP-fucose transporter. 25 UDP-Galactose [0153] Two heterologous genes, gmal2(+) encoding alpha 1,2 galactosyltransferase (alpha 1,2 GalT) from Schizosaccharomycespombe and (hUGT2) encoding human UDP-galactose (UDP-Gal) transporter, have been functionally expressed in S.cerevisiae to examine the intracellular conditions 30 required for galactosylation. Correlation between protein galactosylation and UDP-galactose transport activity indicated that an exogenous supply of UDP-Gal transporter, rather than alpha 1,2 GaIT played a key rolefor efficient 53 galactosylation in S.cerevisiae (Kainuma, 1999 Glycobiology 9(2): 133-141). Likewise, an UDP-galactose transporter from S. pombe was cloned (Aoki, 1999 J.Biochem. 126(5): 940-950; Segawa, 1999 Febs Letters 451(3): 295-298). 5 CMP-N-acetylneuraminic acid (CMP-Sialic acid). [01541 Human CMP-sialic acid transporter (hCST) has been cloned and expressed in Lec 8 CHO cells (Aoki, 1999; Eckhardt, 1997). The functional expression of the murine CMP-sialic acid transporter was achieved in Saccharomyces cerevisiae (Berninsone, 1997). Sialic acid has been found in some 10 fungi, however it is not clear whether the chosen host system will be able to supply sufficient levels of CMP-Sialic acid. Sialic acid can be either supplied in the medium or alternatively fungal pathways involved in sialic acid synthesis can also be integrated into the host genome. 15 Expression of Diphosphatases: [0155] When sugars are transferred onto a glycoprotein- either a nucleoside diphosphate or monophosphate is released from the sugar nucleotide precursors. While monophosphates can be directly exported in exchange for nucleoside triphosphate sugars by an antiport mechanism, diphosphonucleosides (e.g. GDP) 20 have to be cleaved by phosphatases (e.g. GDPase) to yield nucleoside monophosphates and inorganic phosphate prior to being exported. This reaction appears to be important for efficient glycosylation, as GDPase from S.cerevisiae has been found to be necessary for mannosylation. However, the enzyme only has 10% of the activity towards UDP (Berninsone, 1994). Lower eukaryotes often do 25 not have UDP-speciflc diphosphatase activity in the Golgi as they do not utilize UDP-sugar precursors for glycoprotein synthesis in the Golgi. Schizosaccharomycespombe, a yeast which adds galactose residues to cell wall polysaccharides (from UDP-galactose), was found to have specific JDPase activity, further suggesting the requirement for such an enzyme (Berninsone, 30 1994). UDP is known to be a potent inhibitor of glycosyltransferases and the removal of this glycosylation side product is important to prevent glycosyltransferase inhibition in the lumen of the Golgi (Khatara et al. 1974). 54 Methods For Altering N-Glycans in a Host By Expressing A Targeted Enzymatic Activity From a Nucleic Acid Molecule 5 [0156] The present invention further provides a method for producing a human like glycoprotein in a non-human host cell comprising the step of introducing into the cell one or more nucleic acid molecules which encode an enzyme or enzymes for production of the MansGlcNAc 2 carbohydrate structure. In one preferred embodiment, a nucleic acid molecule encoding one or more manmosidase activities 10 involved in the production of MansGlcNAc 2 from Man8GlcNAc 2 or Man 9 GlcNAc 2 is introduced into the host. The invention additionally relates to methods for making altered glycoproteins in a host cell comprising the step of introducing into the host cell a nucleic acid molecule which encodes one or more glycosylation enzymes or activities. Preferred enzyme activities are selected from the group 15 consisting of UDP-GlcNAc transferase, UDP-galactosyltransferase,

GDP

fucosyltransferase, CMP-sialyltransferase, LJDP-GlcNAc transporter, UDP galactose transporter, GDP-fucose transporter, CMP-sialic acid transporter, and nucleotide diphosphatases. In a particularly preferred embodiment, the host is selected or engineered to express two or more enzymatic activities in which the 20 product of one activity increases substrate levels of another activity, e.g., a glycosyltransferase and a corresponding sugar transporter, e.g., GnTI and UDP GlcNAc transporter activities. In another preferred embodiment, the host is selected or engineered to expresses an activity to remove products which may inhibit subsequent glycosylation reactions, e.g. a UJDP- or GDP-specific 25 diphosphatase activity. [0157] Preferred methods of the invention involve expressing one or more enzymatic activities from a nucleic acid molecule in a host cell and comprise the step of targeting at least one enzymatic activity to a desired subcellular location (e.g., an organelle) by forming a fusion protein comprising a catalytic domain of 30 the enzyme and a cellular targeting signal peptide, e.g., a heterologous signal peptide which is not normally ligated to or associated with the catalytic domain. The fusion protein is encoded by at least one genetic construct ("fusion construct") comprising a nucleic acid fragment encoding a cellular targeting signal peptide 55 ligated in the same translational reading frame ("in-frame") to a nucleic acid fragment encoding an enzyme (e.g., glycosylation enzyme), or catalytically active fragment thereof. [0158] The targeting signal peptide component of the fusion construct or protein 5 is preferably derived from a member of the group consisting of: membrane-bound proteins of the ER or Golgi, retrieval signals, Type II membrane proteins, Type I membrane proteins, membrane spanning nucleotide sugar transporters, mannosidases, sialyltransferases, glucosidases, mannosyltransferases and phosphomannosyltransferases. 10 [0159] The catalytic domain component of the fusion construct or protein is preferably derived from a glycosidase, mannosidase or a glycosyltransferase activity derived from a member of the group consisting of GnTI, GnTII, GnTII, GnTIV, GnTV, GnTVL GaIT, Fucosyltransferase and Sialyltransferase. The catalytic domain preferably has a pH optimum within 1.4 pH units of the average 15 pH optimum of other representative enzymes in the organelle in which the enzyme is localized, or has optimal activity at a pH between 5.1 and 8.0. In a preferred embodiment, the catalytic domain encodes a mannosidase selected from the group consisting of C. elegans mannosidase IA, C. elegans mannosidise I3, D. melanogaster mannosidase IA, H sapiens mannosidase IB, P. citrinum 20 mannosidase I, mouse mannosidase IA, mouse mannosidase IB, A. nidulans mannosidase IA, A. nidulans mannosidase IB, A. nidulans mannosidase IC, mouse mannosidase II, C. elegans mannosidase II, H. sapiens mannosidase II, and mannosidase IIL 25 Selecting a Glycosylation Enzyme: p1H Optima and Subcellular Localization [0160] In one embodiment of the invention, a human-like glycoprotein is made efficiently in a non-human eukaryotic host cell by introducing into a subcellular compartment of the cell a glycosylation enzyme selected to have a pH optimum similar to the pH optima of other enzymes in the targeted subcellular compartment. 30 For example, most enzymes that are active in the ER and Golgi apparatus of S.cerevisiae have pH optima that are between about 6.5 and 7.5 (see Table 3). Because the glycosylation of proteins is a highly evolved and efficient processthe 56 internal pH of the ER and the Golgi is likely also in the range of about 6-8. All previous approaches to reduce mannosylation by the action of recombinant mannosidases in fungal hosts, however, have introduced enzymes that have a pH optimum of around pH 5.0 (Martinet et al., 1998, and Chiba et al., 1998). At pH 5 7.0, the in vitro determined activity of those mannosidases is reduced to less than 10%, which is likely insufficient activity at their point of use, namely, the ER and early Golgi, for the efficient in vivo production of MansGlcNAc 2 on N-glycans. [0161] Accordingly, a preferred embodiment of this invention targets a selected glycosylation enzyme (or catalytic domain thereof), e.g., an c-mannosidase, to a 10 subcellular location in the host cell (e.g., an organelle) where the pH optimum of the enzyme or domain is within 1.4 pH units of the average pH optimum of other representative marker enzymes localized in the same organelle(s). The pH optimum of the enzyme to be targeted to a specific organelle should be matched with the pH optimum of other enzymes found in the same organelle to maximize 15 the activity per unit enzyme obtained. Table 3 summarizes the activity of mannosidases from various sources and their respective pH optima. Table 4 summarizes their typical subcellular locations. Table 3. Mannosidases and their pH optimum. 57 Source Enzyme pH Reference optimum Aspergillus saitoi a-1,2-mannosidase 5.0 Ichishima et al., 1999 Biochem. J. 339(Pt 3):589-597 Trichoderma reesei a-1,2-mannosidase 5.0 Maras et al., 2000 J. Biotechnol. 77(2-3):255 263 Penicillium a-D-1,2- 5.0 Yoshida et al., 1993 citrinum mannosidase Biochem. J. 290(Pt 2):349-354 C.elegans c-1,2-mamosidase 5.5 see Fig. 11 Aspergillus a- 1,2-mannosidase 6.0 Eades and Hintz, 2000 nidulans Homo sapiens a-1,2-mannosidase 6.0 IA(Golgi) Homo sapiens IB a-1,2-mannosidase 6.0 (Golgi) Lepidopteran Type I a-1,2-Man 6 - 6.0 Ren et al., 1995 insect cells mannosidase Biochem. 34(8):2489 2495 Homo sapiens a-D-mannosidase 6.0 Chandrasekaran et al., 1984 Cancer Res. 44(9):4059-68 Xanthomonas c-1,2,3-mannosidase 6.0 U.S. Pat. No. 6,300,113 manihotis Mouse IB (Golgi) cc-1,2-mannosidase 6.5 Schneikert and Herscovics, 1994 Glycobiology. 4(4):445 50 Bacillus sp. a-D-1,2- 7.0 Maruyama et al., 1994 (secreted) mannosidase Carbohydrate Res. 251:89-98 [0162] In a preferred embodiment, a particular enzyme or catalytic domain is targeted to a subcellular location in the host cell by means of a chimeric fusion construct encoding a protein comprising a cellular targeting signal peptide not 5 normally associated with the enzymatic domain. Preferably, an enzyme or domain is targeted to the ER, the early, medial or late G6gi of the trans Golgi apparatus of the host cell. [01631 In a more preferred embodiment, the targeted glycosylation enzyme is a mannosidase, glycosyltransferase or a glycosidase. In an especially preferred 10 embodiment, mannosidase activity is targeted to the ER or cis Golgi, where the 58 early reactions of glycosylation occur. While this method is useful for producing a human-like glycoprotein in a non-human host cell, it will be appreciated that the method is also useful more generally for modifying carbohydrate profiles of a glycoprotein in any eukaryotic host cell, including human host cells. 5 10164] Targeting sequences which mediate retention of proteins in certain organelles of the host cell secretory pathway are well-known and described in the scientific literature and public databases, as discussed in more detail below with respect to libraries for selection of targeting sequences and targeted enzymes. Such subcellular targeting sequences may be used alone or in combination to target 10 a selected glycosylation enzyme (or catalytic domain thereof) to a particular subcellular location in a host cell, i.e., especially to one where the enzyme will have enhanced or optimal activity based on pH optima or the presence of other stimulatory factors. [0165] When one attempts to trim high mannose structures to yield 15 MansGlcNAc 2 in the ER or the Golgi apparatus of a host cell such as S.cerevisiae, for example, one may choose any enzyme or combination of enzymes that (1) has a sufficiently close pH optimum (i.e. between pH 5.2 and pH 7.8), and (2) is known to generate, alone or in concert, the specific isomeric Man 5 GlcNAc 2 structure required to accept subsequent addition of GlcNAc by GnTI. Any enzyme or 20 combination of enzymes that is shown to generate a structure that can be converted to GlcNAcMansGlcNAc 2 by GnTI in vitro would constitute an appropriate choice. This knowledge may be obtained from the scientific literature or experimentally. [0166] For example, one may determine whether a potential mannosidase can convert MansGlcNAc 2 -2AB ( 2 -aminobenzamide) to MansGlcNAc 2 -AB and then 25 verify that the obtained MansGlcNAc 2 -2AB structure can serve a substrate for GnTI and UDP-GlcNAc to give GlcNAcMansGlcNAc 2 in vitro. Mannosidase IA from a human or murine source, for example, would be an appropriate choice (see, e.g., Example 11). Examples described herein utilize 2 -aminobenzamide labeled N-linked oligomannose followed by HPLC analysis to make this determination. 30 Table 4. Cellular location and pH optima of various glycosylation-related enzymes of S.cerevislae. 59 Gene Activity Location PH optimum Reference(s) KTR1 a- 1,2 Golgi 7.0 Romero et al., mannosyltransferase 1997 Biochem. J. 321(Pt 2):289 295 MNSJ a- 1,2- mannosidase ER 6.5 CWH41 glucosidase I ER 6.8 - mannosyltransferase Golgi 7-8 Lehele and Tanner, 1974 Biochim. Biophys. Acta 350(1):225-235 KRE2 a- 1,2 Golgi 6.5-9.0 Romero et al., mannosyltransferase 1997 [0167] Accordingly, a glycosylation enzyme such as an a-1,2-mannosidase enzyme used according to the invention has an optimal activity at a pH of between 5.1 and 8.0. In a preferred embodiment, the enzyme has an optimal activity at a 5 pH of between 5.5 and 7.5. The C. elegans mannosidase enzyme, for example, works well in the methods of the invention and has an apparent pH optimum of about 5.5). Preferred mannosidases include those listed in Table 3 having appropriate pH optima, e.g. Aspergillus nidulans, Homo sapiens IA (Golgi), Homo sapiens IB (Golgi), Lepidopteran insect cells (IPLB-SF21AE), Homo sapiens, 10 mouse lB (Golgi), Xanthomonas manihols, Drosophila melanogaster and C. elegans. [0168] The experiment which illustrates the pH optimum for an a-1,2 mannosidase enzyme is described in Example 14. A chimeric fusion protein BB27-2 (Saccharomyces MNN1O (s)/C. elegans mannosidase IB A3 1), which leaks 15 into the medium was subjected to various pH ranges to determine the optimal activity of the enzyme. The results of the experiment show that the a-1,2 mannosidase has an optimal pH of about 5.5 for its function (Fig. 11). [0169] In a preferred embodiment, a single cloned mannosidase gene is expressed in the host organism. However, in some cases it may be desirable to 20 express several different mannosidase genes, or several copies of one particular gene, in order to achieve adequate production of MansGlcNAc 2 . In cases where multiple genes are used, the encoded mannosidases preferably all have pH optima 60 within the preferred range of about 5.1 to about 8.0, or especially between about 5.5 and about 7.5. Preferred mannosidase activities include a-1,2-mannosidases derived from mouse, human, Lepidoptera, Aspergillus nidulans, or Bacillus sp., C. elegans, D. nelanogaster, P. citrinum, X laevis or A. nidulans. 5 In Vivo Alteration of Host Cell Glycosylation Using a Combinatorial DNA Library [0170] Certain methods of the invention are preferably (but need not necessarily 10 be) carried out using one or more nucleic acid libraries. An exemplary feature of a combinatorial nucleic acid library of the invention is that it comprises sequences encoding cellular targeting signal peptides and sequences encoding proteins to be targeted (e.g., enzymes or catalytic domains thereof, including but not limited to those which mediate glycosylation). 15 [0171] In one embodiment, a combinatorial nucleic acid library comprises: (a) at least two nucleic acid sequences encoding different cellular targeting signal peptides; and (b) at least one nucleic acid sequence encoding a polypeptide to be targeted. In another embodiment, a combinatorial nucleic acid library comprises: (a) at least one nucleic acid sequence encoding a cellular targeting signal peptide; 20 and (b) at least two nucleic acid sequences encoding a polypeptide to be targeted into a host cell. As described further below, a nucleic acid sequence derived from (a) and a nucleic acid sequence derived from (b) are ligated to produce one or more fusion constructs encoding a cellular targeting signal peptide functionally linked to a polypeptide domain of interest. One example of a functional linkage is when the 25 cellular targeting signal peptide is ligated to the polypeptide domain of interest in the same translational reading frame ("in-frame"). [0172] In a preferred embodiment, a combinatorial DNA library expresses one or more fusion proteins comprising cellular targeting signal peptides ligated in-frame to catalytic enzyme domains. The encoded fusion protein preferably comprises a 30 catalytic domain of an enzyme involved in mammalian- or human-like modification of N-glycans. In a more preferred embodiment, the catalytic domain is derived from an enzyme selected from the group consisting of mannosidases, glycosyltransferases and other glycosidases which is ligated in-frame to one or more targeting signal peptides. The enzyme domain may be exogenous and/or endogenous to the host cell. A particularly preferred signal peptide is one normally associated with a protein that undergoes ER to Golgi transport. [0173] The combinatorial DNA library of the present invention may be used for 5 producing and localizing in vivo enzymes involved in mammalian- or human-like N-glycan modification. The fusion constructs of the combinatorial DNA library are engineered so that the encoded enzymes are localized in the ER, Golgi or the trans-Golgi network of the host cell where they are involved in producing particular N-glycans on a glycoprotein of interest. Localization of N-glycan 10 modifying enzymes of the present invention is achieved through an anchoring mechanism or through protein-protein interaction where the localization peptide constructed from the combinatorial DNA library localizes to a desired organelle of the secretory pathway such as the ER, Golgi or the trans Golgi network. [01741 An example of a useful N-glycan, which is produced efficiently and in 15 sufficient quantities for further modification by human-like (complex) glycosylation reactions is MansGlcNAc 2 . A sufficient amount of MansGlcNAc 2 is needed on a glycoprotein of interest for further human-like processing in vivo (e.g., more than 30 mole %). The MansGlcNAc 2 intermediate may be used as a substrate for further N-glycan modification to produce GlcNAcMansGlcNAc 2 (Fig. IB; see 20 above). Accordingly, the combinatorial DNA library of the present invention may be used to produce enzymes which subsequently produce GlcNAcManGlcNAC 2 , or other desired complex N-glycans, in a useful quantity. [0175] A further aspect of the fusion constructs produced using the combinatorial DNA library of the present invention is that they enable sufficient and often near 25 complete intracellular N-glycan trimming activity in the engineered host cell. Preferred fusion constructs produced by the combinatorial DNA library of the invention encode a glycosylation enzyme, e.g., a mannosidase, which is effectively localized to an intracellular host cell compartment and thereby exhibits very little and preferably no extracellular activity. The preferred fusion constructs of the 30 present invention that encode a mannosidase enzyme are shown to localize where the N-glycans are modified, namely, the ER and the Golgi. The fusion enzymes of the present invention are targeted to such particular organelles in the secretory 62 pathway where they localize and act upon N-glycans such as ManaGlcNAc 2 to produce MansGlcNAc 2 on a glycoprotein of interest. [0176] Enzymes produced by the combinatorial DNA library of the present invention can modify N-glycans on a glycoprotein of interest as shown for K3 or 5 IFN-p proteins expressed in P.pastoris, as shown in Fig. 5 and Fig. 6, respectively (see also Examples 2 and 11). It is, however, appreciated that other types of glycoproteins, without limitation, including erythropoietin, cytokines such as interferon-a, interferon-p, interferon-y, interferon-o, and granulocyte-CSF, coagulation factors such as factor VIII, factor IX, and human protein C, soluble 10 IgB receptor a-chain, IgG, IgG fragments, IgM, interleukins, urokinase, chymase, and urea trypsin inhibitor, IGF-binding protein, epidermal growth factor, growth hormone-releasing factor, annexin V fusion protein, angiostatin, vascular endothelial growth factor-2, myeloid progenitor inhibitory factor-1, osteoprotegerin, a-1 antitrypsin, DNase II, a- feto proteins, AAT, rhTBP-1 15 (onercept, aka TNF Binding protein 1), TACI-Ig (transmembrane activator and calcium modulator and cyclophilin ligand interactor), FSH (follicle stimulating hormone), GM-CSF, GLP-l w/ and w/o FC (glucagon like protein 1) IL-I receptor agonist, sTNFr (enbrel, aka soluble TNF receptor Fc fusion) ATI, rhThrombin, glucocerebrosidase and CTLA4-Ig (Cytotoxic T Lymphocyte associated Antigen 4 20 - Ig) may be glycosylated in this way. Constructing a Combinatorial DNA Library of Fusion Constructs: [01771 A combinatorial DNA library of fusion constructs features one or more cellular targeting signal peptides ("targeting peptides") generally derived from N 25 terminal domains of native proteins (e.g., by making C-terminal deletions). Some targeting peptides, however, are derived from the C-terminus of native proteins (e.g. SEC12). Membrane-bound proteins of the ER or the Golgi are preferably used as a source for targeting peptide sequences. These proteins have sequences encoding a cytosolic tail (ct), a transmembrane domain (tmd) and a stem region 30 (sr) which are varied in length. These regions are recognizable by protein sequence alignments and comparisons with known homologs and/or other localized proteins (e.g., comparing hydrophobicity plots). 63 [0178] The targeting peptides are indicated herein as short (s), medium (in) and long (1) relative to the parts of a type II membrane. The targeting peptide sequence indicated as short (s) corresponds to the transmembrane domain (tmd) of the membrane-bound protein. The targeting peptide sequence indicated as long (I) 5 corresponds to the length of the transmembrane domain (tmd) and the sten region (sr). The targeting peptide sequence indicated as medium (in) corresponds to the transmembrane domain (tmd) and approximately half the length of the stem region (sr). The catalytic domain regions are indicated herein by the number of nucleotide deletion with respect to its wild-type glycosylation enzyme. 10 Sub-libraries [0179] In some cases a combinatorial nucleic acid library of the invention may be assembled directly from existing or wild-type genes. In a preferred embodiment, the DNA library is assembled from the fusion of two or more sub 15 libraries. By the in-frame ligation of the sub-libraries, it is possible to create a large number of novel genetic constructs encoding useful targeted protein domains such as those which have glycosylation activities. Catalytic Domain Sub-Libraries Encoding Glycosylation Activities 20 [0180] One useful sub-library includes DNA sequences encoding enzymes such as glycosidases (e.g., mannosidases), glycosyltransferases (e.g., fucosyl transferases, galactosyltransferases, glucosyltransferases), GIcNAc transferases and sialyltransferases. Catalytic domains may be selected from the host to be engineered, as well as from other related or unrelated organisms. Mammalian, 25 plant, insect, reptile, algal or fungal enzymes are all useful and should be chosen to represent a broad spectrum of biochemical properties with respect to temperature and pH optima. En a preferred embodiment, genes are tnmcated to give fragments some of which encode the catalytic domains of the enzymes. By removing endogenous targeting sequences, the enzymes may then be redirected and 30 expressed in other cellular loci. [0181] The choice of such catalytic domains may be guided by the knowledge of the particular environment in which the catalytic domain is subsequently to be 64 active. For example, if a particular glycosylation enzyme is to be active in the late -Golgi, and all known enzymes of the host organism in the late Golgi have a certain pH optimum, or the late Golgi is known to have a particular pH, then a catalytic domain is chosen which exhibits adequate, and preferably maximum, activity at 5 that pH, as discussed above. Targeting Peptide Sequence Sub-Libraries [0182] Another useful sub-library includes nucleic acid sequences encoding targeting signal peptides that result in localization of a protein to a particular 10 location within the ER, Golgi, or trans Golgi network. These targeting peptides may be selected from the host organism to be engineered as well as from other related or unrelated organisms. Generally such sequences fall into three categories: (1) N-terminal sequences encoding a cytosolic tail (ct), a transmembrane domain (tmd) and part or all of a stem region (sr), which together 15 or individually anchor proteins to the inner (lumenal) membrane of the Golgi; (2) retrieval signals which are generally found at the C-terminus such as the HDEL (SEQ ID NO: 5) or KDEL (SEQ ID NO: 6) tetrapeptide; and (3) membrane spanning regions from various proteins, e.g., nucleotide sugar transporters, which are known to localize in the Golgi. 20 [0183] In the first case, where the targeting peptide consists of various elements (ct, tmnd and sr), the library is designed such that the ct, the tmd and various parts of the stem region are represented. Accordingly, a preferred embodiment of the sub-library of targeting peptide sequences includes ct, tmd, and/or sr sequences from membrane-bound proteins of the ER or Golgi. In some cases it may be 25 desirable to provide the sub-library with varying lengths of sr sequence. This may be accomplished by PCR using primers that bind to the 5' end of the DNA encoding the cytosolic region and employing a series of opposing primers that bind to various parts of the stem region. [01841 Still other useful sources of targeting peptide sequences include retrieval 30 signal peptides, e.g. the tetrapeptides HDEL (SEQ ID NO: 5) or KDEL (SEQ ID NO: 6), which are typically found at the C-terminus of proteins that are transported retrograde into the ER or Golgi. Still other sources of targeting peptide sequences 65 include (a) type II membrane proteins, (b) the enzymes listed in Table 3, (c) membrane spanning nucleotide sugar transporters that are localized in the Golgi, and (d) sequences referenced in Table 5. (The HDEL signal in column 1, cell 8 is shown in SEQ ID NO: 5). 5 Table 5. Sources of useful compartmental targeting se fences Gene or Sequence Organism Function Location of Gene SequenceProduct MNSI A.nidulans a-1,2-mannosidase ER MNSI A.niger a-1,2-mannosidase ER MNSI S.cerevisiae a-1,2-mannosidase ER GLSI S.cerevisiae glucosidase ER GLSI A.niger glucosidase ER GLSI A.nidulans glucosidase ER HDEL Universal in at C-terminus retrieval signal R SEC12 S.cerevisiae COPH vesicle protein ER/Golgi SEC12 A.niger COPH vesicle protein ER/Golgi OCH1 S.cerevisiae 1, 6 -mannosyltransferase Golgi (cis) OCHI P.pastoris 1, 6 -mannosyltransferase Golgi (cis) MNN9 S.cerevisiae 1, 6 -mannosyltransferase complex MAN9 A.niger undetermined Golgi VAN] S.cerevisiae undetermined Golgi VAN1 A.niger undetermined Golgi AATP S.cerevisiae undetermined Golgi HOCI S.cerevisiae undetermined Golgi MNN0 S.cerevisiae undetermined Golgi MNN10 A.niger undetermined Golgi MNN11 S.cerevisiae undetermined Golgi (cis) MNN11 A.niger undetermined Golgi (cis) MNTJ S.cerevisiae 1, 2 -mannosyltransferase Golgi (cis, medial 66 KTR1 P.pastoris undetermined Golgi (medial) KRE2 P.pastoris undetermined Golgi (medial) KTR3 P.pastoris undetermined Golgi (medial) MNN2 S.cerevisiae 1, 2 -mannosyltransferase Golgi (medial) KTR1 S.cerevisiae undetermined Golgi (medial) KTR2 S.cerevisiae undetermined Golgi (medial) M N S.cerevisiae 1, 3 -mannosyltransferase Golgi (trans) MNN6 S.cerevisiae Phosphomannosyltransfer Go1 (ta) _____________ase 2,6 ST H sapiens 2

,

6 -sialyltransferase trans Golgi network UDP-Gal T S. pombe UDP-Gal transporter Golgi [0185] In any case, it is highly preferred that targeting peptide sequences are selected which are appropriate for the particular enzymatic activity or activities to function optimally within the sequence of desired glycosylation reactions. For 5 example, in developing a modified microorganism capable of terminal sialylation of nascent N-glycans, a process which occurs in the late Golgi in humans, it is desirable to utilize a sub-library of targeting peptide sequences derived from late Golgi proteins. Similarly, the trimming of MangGlcNAc 2 by an a-1,2-mannosidase to give MansGlcNAc 2 is an early step in complex N-glycan formation in humans 10 (Fig. IB). It is therefore desirable to have this reaction occur in the ER or early Golgi of an engineered host microorganism. A sub-library encoding ER and early Golgi retention signals is used. [0186] A series of fusion protein constructs (i.e., a combinatorial DNA library) is then constructed by functionally linking one or a series of targeting peptide 15 sequences to one or a series of sequences encoding catalytic domains. In a preferred embodiment, this is accomplished by the in-frame ligation of a sub library comprising DNA encoding targeting peptide sequences (above) with a sub library comprising DNA encoding glycosylation enzymes or catalytically active fragments thereof (see below). 20 [0187] The resulting library comprises synthetic genes encoding targeting peptide sequence-containing fusion proteins. In some cases it is desirable to 67 provide a targeting peptide sequence at the N-terminus of a fusion protein, or in other cases at the C-terminus. In some cases, targeting peptide sequences may be inserted within the open reading frame of an enzyme, provided the protein structure of individual folded domains is not disrupted. Each type of fusion protein 5 is constructed (in a step-wise directed or semi-random fashion) and optimal constructs may be selected upon transformation of host cells and characterization of glycosylation patterns in transformed cells using methods of the invention. Generating Additional Sequence Diversity 10 [0188] The method of this embodiment is most effective when a nucleic acid, e.g., a DNA library transformed into the host contains a large diversity of sequences, thereby increasing the probability that at least one transformant will exhibit the desired phenotype. Single amino acid mutations, for example, may drastically alter the activity of glycoprotein processing enzymes (Romero et al., 15 2000). Accordingly, prior to transformation, a DNA library or a constituent sub library may be subjected to one or more techniques to generate additional sequence diversity. For example, one or more rounds of gene shuffling, error prone PCR, in vitro mutagenesis or other methods for generating sequence diversity, may be performed to obtain a larger diversity of sequences within the pool of fusion 20 constructs. Expression Control Sequences 101891 In addition to the open reading frame sequences described above, it is generally preferable to provide each library construct with expression control 25 sequences, such as promoters, transcription terminators, enhancers, ribosome binding sites, and other functional sequences as may be necessary to ensure effective transcription and translation of the fusion proteins upon transformation of fusion constructs into the host organism. [0190] Suitable vector components, e.g., selectable markers, expression control 30 sequences (e.g., promoter, enhancers, terminators and the like) and, optionally, sequences required for autonomous replication in a host cell, are selected as a function of which particular host cell is chosen. Selection criteria for suitable 68 vector components for use in a particular mammalian or a lower eukaryotic host cell are routine. Preferred lower eukaryotic host cells of the invention include Pichia pastoris, Pichiafnlandica, Pichia trehalophila, Pichia koclamae, Pichia membranaefaciens, Pichia opuntiae, Pichia thennotolerans, Pichia salictaria, 5 Pichia guercuum, Pichia piyperi, Pichia stiptis, Pichia methanolica, Pichia sp., Saccharomyces cerevisiae, Saccharomyces sp., Hansenulapolymorpha, Kluyveromyces sp., Kluyveronyces lactis, Candida albicans, Aspergillus nidulans, Aspergillus niger, Aspergillus oryzae, Trichoderma reesei, Chrysosporium luckatowense, Fusarium sp. Fusarium gramineum, Fusarium venenatum and 10 Neurospora crassa. Where the host is Pichia pastoris, suitable promoters include, for example, the AOX1, AOX2, GAPDH and P40 promoters. Selectable Markers [0191] It is also preferable to provide each construct with at least one selectable 15 marker, such as a gene to impart drug resistance or to complement a host metabolic lesion. The presence of the marker is useful in the subsequeht-selection of transformants; for example, in yeast the URA3, IS4, SUC2, G418, BLA, or SH BLE genes may be used. A multitude of selectable markers are known and available for use in yeast, fungi, plant, insect, mammalian and other eukaryotic 20 host cells. Transformation [0192] The nucleic acid library is then transformed into the host organism. In yeast, any convenient method of DNA transfer may be used, such as 25 electroporation, the lithium chloride method, or the spheroplast method. In filamentous fungi and plant cells, conventional methods include particle bombardment, electroporation and agrobacterium mediated transformation. To produce a stable strain suitable for high-density culture (e.g., fermentation in yeast), it is desirable to integrate the DNA library constructs into the host 30 chromosome. In a preferred embodiment, integration occurs via homologous recombination, using techniques well-known in the art. For example, DNA library elements are provided with flanking sequences homologous to sequences of the 69 host organism. In this manner, integration occurs at a defined site in the host genome, without disruption of desirable or essential genes. [0193] In an especially preferred embodiment, library DNA is integrated into the site of an undesired gene in a host chromosome, effecting the disruption or deletion 5 of the gene. For example, integration into the sites of the OCH1, fAN1, or MNN4 genes allows the expression of the desired library DNA while preventing the expression of enzymes involved in yeast hypermannosylation of glycoproteins. In other embodiments, library DNA may be introduced into the host via a nucleic acid molecule, plasmid, vector (e.g., viral or retroviral vector), chromosome, and may 10 be introduced as an autonomous nucleic acid molecule or by homologous or random integration into the host genome. In any case, it is generally desirable to include with each library DNA construct at least one selectable marker gene to allow ready selection of host organisms that have been stably transformed. Recyclable marker genes such as ura3, which can be selected for or against, are 15 especially suitable. Screening and Selection Processes [01941 After transformation of the host strain with the DNA library, transformants displaying a desired glycosylation phenotype are selected. Selection 20 may be performed in a single step or by a series of phenotypic enrichment and/or depletion steps using any of a variety of assays or detection methods. Phenotypic characterization may be carried out manually or using automated high-throughput screening equipment. Commonly, a host microorganism displays protein N glycans on the cell surface, where various glycoproteins are localized. 25 [0195] One may screen for those cells that have the highest concentration of terminal GIcNAc on the cell surface, for example, or for those cells which secrete the protein with the highest terminal GlcNAc content. Such a screen may be based on a visual method, like a staining procedure, the ability to bind specific terminal GlcNAc binding antibodies or lectins conjugated to a marker (such lectins are 30 available from E.Y. Laboratories Inc., San Mateo, CA), the reduced ability of specific lectins to bind to terminal mannose residues, the ability to incorporate a radioactively labeled sugar in vitro, altered binding to dyes or charged surfaces, or 70 may be accomplished by using a Fluorescence Assisted Cell Sorting (FACS) device in conjunction with a fluorophore labeled lectin or antibody (Guillen, 1998). [0196] Accordingly, intact cells may be screened for a desired glycosylation phenotype by exposing the cells to a lectin or antibody that binds specifically to the 5 desired N-glycan. A wide variety of oligosaccharide-specific lectins are available commercially (e.g., from BY Laboratories, San Mateo, CA). Alternatively, antibodies to specific human or animal N-glycans are available commercially or may be produced using standard techniques. An appropriate lectin or antibody may be conjugated to a reporter molecule, such as a chromophore, fluorophore, 10 radioisotope, or an enzyme having a chromogenic substrate (Guillen et al., 1998. Proc. Nat!. Acad. Sci. UISA 95(14): 7888-7892)). [0197] Screening may then be performed using analytical methods such as spectrophotometry, fluorimetry, fluorescence activated cell sorting, or scintillation counting. In other cases, it may be necessary to analyze isolated glycoproteins or 15 N-glycans from transformed cells. Protein isolation may be carried out by techniques known in the art. In a preferred embodiment, a reporter protein is secreted into the medium and purified by affinity chromatography (e.g. Ni-affinity or glutathione -S-transferase affinity chromatography). In cases where an isolated N-glycan is preferred, an enzyme such as endo-p-N-acetylglucosaminidase 20 (Genzyme Co., Boston, MA; New England Biolabs, Beverly, MA) may be used to cleave the N-glycans from glycoproteins. Isolated proteins or N-glycans may then be analyzed by liquid chromatography (e.g. HPLC), mass spectroscopy, or other suitable means. U.S. Patent No. 5,595,900 teaches several methods by which cells with desired extracellular carbohydrate structures may be identified. In a preferred 25 embodiment, MALDI-TOF mass spectrometry is used to analyze the cleaved N glycans. [0198] Prior to selection of a desired transformant, it may be desirable to deplete the transformed population of cells having undesired phenotypes. For example, when the method is used to engineer a functional mannosidase activity into cells, 30 the desired transformants will have lower levels of mannose in cellular glycoprotein. Exposing the transformed population to a lethal radioisotope of mannose in the medium depletes the population of transformatshaving the 71 undesired phenotype, i.e. high levels of incorporated manmose (Huffaker TC and Robbins PW., Proc Natl Acad Sci U SA. 1983 Dec;80(24):7466-70). Alternatively, a cytotoxic lectin or antibody, directed against an undesirable N-glycan, may be used to deplete a transformed population of undesired 5 phenotypes (e.g., Stanley P and Siminovitch L. Somatic Cell Genet 1977 Jul;3(4):391-405). U.S. Patent No. 5,595,900 teaches several methods by which cells with a desired extracellular carbohydrate structures may be identified. Repeatedly carrying out this strategy allows for the sequential engineering of more and more complex glycans in lower eukaryotes. 10 [0199] To detect host cells having on their surface a high degree of the human like N-glycan intermediate GIcNAcMan 3 GlcNAc 2 , for example, one may select for transformants that allow for the most efficient transfer of GlcNAc by GlcNAc Transferase from UDP-GlcNAc in an in vitro cell assay. This screen may be carried out by growing cells harboring the transformed library under selective 15 pressure on an agar plate and transferring individual colonies into a 96-well microtiter plate.. After growing the cells, the cells are centrifuged, the cells resuspended in buffer, and after addition of UDP-GlcNAc and GnT I, the release of UDP is determined either by HPLC or an enzyme linked assay for UDP. Alternatively, one may use radioactively labeled UDP-GlcNAc and GnT H1, wash 20 the cells and then look for the release of radioactive GlcNAc by N actylglucosaminidase. All this may be carried manually or automated through the use of high throughput screening equipment. Transformants that release more UDP, in the first assay, or more radioactively labeled GlcNAc in the second assay, are expected to have a higher degree of GlcNAcMan 3 GlcNAc 2 on their surface and 25 thus constitute the desired phenotype. Similar assays may be adapted to look at the N-glycans on secreted proteins as well. [0200] Alternatively, one may use any other suitable screen such as a lectin binding assay that is able to reveal altered glycosylation patterns on the surface of transformed cells. In this case the reduced binding of lectins specific to terminal 30 mannoses may be a suitable selection tool. Galantus nivalis lectin binds specifically to terminal (x-1,3 mannose, which is expected to be reduced if sufficient mannosidase H activity is present in the Golgi. One may also enrich for 72 desired transformants by carrying out a chromatographic separation step that allows for the removal of cells containing a high terminal mannose content. This separation step would be carried out with a lectin column that specifically binds cells with a high terminal mannose content (e.g., Galantus nivalis lectin bound to 5 agarose, Sigma, St.Louis, MO) over those that have a low terminal mannose content. 10201] In addition, one may directly create such fusion protein constructs, as additional information on the localization of active carbohydrate modifying enzymes in different lower eukaryotic hosts becomes available in the scientific 10 literature. For example, it is known that human p 1,4-GalTr can be fused to the membrane domain of MNT, a mannosyltransferase from S.cerevisiae, and localized to the Golgi apparatus while retaining its catalytic activity (Schwientek et al., 1995). If S.cerevisiae or a related organism is the host to be engineered one may directly incorporate such findings into the overall strategy to obtain complex N 15 glycans from such a host. Several such gene fragments in P.pastoris have been identified that are related to glycosyltransferases in S.cerevisiae and thus could be used for that purpose. Alteration of Host Cell Glycosylation Using 20 Fusion Constructs From Combinatorial Libraries [0202] The construction of a preferred combinatorial DNA library is illustrated schematically in FIg. 2 and described in Example 11. The fusion construct may be operably linked to a multitude of vectors, such as expression vectors well-known in 25 the arl. A wide variety of such fusion constructs were assembled using representative activities as shown in Table 6. Combinations of targeting peptide/catalytic domains may be assembled for use in targeting mannosidase, glycosyltransferase and glycosidase activities in the ER, Golgi and the trans Golgi network according to the invention. Surprisingly, the same catalytic domain may 30 have no effect to a very profound effect on N-glycosylation patterns, depending on the type of targeting peptide used (see, e.g., Table 7, Example 11). Mannosidase Fusion Constructs 73 [0203] A representative example of a mannosidase fusion construct derived from a combinatorial DNA library of the invention is pFB8, which a truncated Saccharomyces SEC12(m) targeting peptide (988-1296 nucleotides of SEC12 from SwissProt P11655) ligated in-frame to a 187 N-terminal amino acid deletion of a 5 mouse a-mannosidase IA (Genbank AN 6678787). The nomenclature used herein, thus, refers to the targeting peptide/catalytic domain region of a glycosylation enzyme as Saccharomyces SEC12 (m)/mouse mannosidase IA A187. The encoded fusion protein localizes in the ER by means of the SEC12 targeting peptide sequence while retaining its mannosidase catalytic domain activity and is capable 10 of producing in vivo N-glycans having a MansGlcNAc 2 structure (Example 11; Fig. 6F, Fig. 7B). [02041 The fusion construct pGC5, Saccharomyces MNSJ(m)/mouse mannosidase IB A99, is another example of a fusion construct having intracellular mannosidase trimming activity (Example 11; Fig. 5D, Fig. 8B). Fusion construct 15 pBC18-5 (Saccharomyces VANJ(s)/C. elegans mannosidase IB A80) is yet another example of an efficient fusion construct capable of producing N-glycans having a MansGlcNAc 2 structure in vivo. By creating a combinatorial DNA library of these and other such mannosidase fusion constructs according to the invention, a skilled artisan may distinguish and select those constructs having optimal intracellular 20 trimming activity from those having relatively low or no activity. Methods using combinatorial DNA libraries of the invention are advantageous because only a select few mannosidase fusion constructs may produce a particularly desired N glycan in vivo. [0205] In addition, mannosidase trimming activity may be specific to a particular 25 protein of interest. Thus, it is to be further understood that not all targeting peptide/mannosidase catalytic domain fusion constructs may function equally well to produce the proper glycosylation on a glycoprotein of interest. Accordingly, a protein of interest may be introduced into a host cell transfected with a combinatorial DNA library to identify one or more fusion constructs which express 30 a mannosidase activity optimal for the protein of interest. One skilled in the art will be able to produce and select optimal fusion construct(s) using the combinatorial DNA library approach described herein. 74 10206] It is apparent, moreover, that other such fusion constructs exhibiting localized active mannosidase catalytic domains (or more generally, domains of any enzyme) may be made using techniques such as those exemplified in Example 11 and described herein. It will be a matter of routine experimentation for one skilled 5 in the art to make and use the combinatorial DNA library of the present invention to optimize, for example, MansGlcNAc 2 production from a library of fusion constructs in a particular expression vector introduced into a particular host cell. Glycosyltransferase Fusion Constructs 10 [02071 Similarly, a glycosyltransferase combinatorial DNA library was made using the methods of the invention. A combinatorial DNA library of sequences derived from glycosyltransferase I (GnTI) activities were assembled with targeting peptides and screened for efficient production in a lower eukaryotic host cell of a GlcNAcMansGlcNAc2 N-glycan structure on a marker glycoprotein. A fusion 15 construct shown to produce GlcNAcMansGlcNAc 2 (pPB104), Saccharomyces MNN9(s)/human GnTI A38 was identified (Example 15). A wide variety of such GnTI fusion constructs were assembled (Example 15, Table 10). Other combinations of targeting peptide/GnTI catalytic domains can readily be assembled by making a combinatorial DNA library. It is also apparent to one 20 skilled in the art that other such fusion constructs exhibiting glycosyltransferase activity may be made as demonstrated in Example 15. It will be a matter of routine experimentation for one skilled in the art to use the combinatorial DNA library method described herein to optimize GlcNAcMansGlcNAc 2 production using a selected fusion construct in a particular expression vector and host cell line. 25 [02081 As stated above for mannosidase fusion constructs, not all targeting peptide/GnTI catalytic domain fusion constructs will function equally well to produce the proper glycosylation on a glycoprotein of interest as described herein. However, one skilled in the art will be able to produce and select optimal fusion construct(s) using a DNA library approach as described herein. Example 15 30 illustrates a preferred embodiment of a combinatorial DNA library comprising targeting peptides and GnTI catalytic domain fusion constructs involved in producing glycoproteins with predominantly GlcNAcMansGlcNAc 2 structure. 75 Using Multiple Fusion Constructs to Alter Host Cell Glycosylation [02091 In another example of using the methods and libraries of the invention to alter host cell glycosylation, a P.pastoris strain with an OCHI deletion that 5 expresses a reporter protein (K3) was transformed with multiple fusion constructs isolated from combinatorial libraries of the invention to convert high mannose N glycans to human-like N-glycans (Example 15). First, the mannosidase fusion construct pFB8 (Saccharomyces SEC12 (m)/mouse mannosidase IA A 187) was transformed into a P.pastoris strain lacking 1,6 initiating mannosyltransferases 10 activity (i.e. ochl del1fiw;. Example 1). Second, pPB103 comprising a Klactis MNN2-2 gene (Genbank AN AF106080) encoding an UDP-GlcNAc transporter was constructed to increase further production of GlcNAcMansGcNAc 2 . The addition of the UDP-GlcNAc transporter increased production of GlcNAcMansGlcNAc 2 significantly in the P.pastoris strain as illustrated in Fig. 15 10B. Third, pPB104 comprising Saccharomyces MNN9 (s)/human GnTI A38 was introduced into the strain. This P.pastoris strain is referred to as "PBP-3." [0210] It is understood by one skilled in the art that host cells such as the above described yeast strains can be sequentially transformed and/or co-transformed with one or more expression vectors. It is also understood that the order of 20 transformation is not particularly relevant in producing the glycoprotein of interest. The skilled artisan recognizes the routine modifications of the procedures disclosed herein may provide improved results in the production of the glycoprotein of interest. [0211] The importance of using a particular targeting peptide sequence with a 25 particular catalytic domain sequence becomes readily apparent from the experiments described herein. The combinatorial DNA library provides a tool for constructing enzyme fusions that are involved in modifying N-glycans on a glycoprotein of interest, which is especially useful in producing human-like glycoproteins. (Any enzyme fusion, however, may be selected using libraries and 30 methods of the invention.) Desired transformants expressing appropriately targeted, active c-1, 2 -mannosidase produce K3 with N-glycans of the structure MansGlcNAc 2 as shown in Figs. 5D and 5E. This confers a reduced molecular 76 mass to the cleaved glycan compared to the K3 of the parent OCH1 deletion strain, as was detected by MALDI-TOF mass spectrometry in Fig. SC. [02121 Similarly, the same approach was used to produce another secreted glycoprotein: FN-f comprising predominantly MansGlcNAc 2 . The 5 MansGlcNAc 2 was removed by PNGase digestion (Papac et al. 1998 Glycobiology 8, 445-454) and subjected to MALDI-TOF as shown in Fig. 6A - 6F. A single prominent peak at 1254 (m/z) confirms MansGlcNA 2 production on IFN- # in Figs. 6E (pGC5) (Saccharomyces AMNS1(m)/mouse mannosidase IB A99) and 6F (pFB8) (Saccharomyces SEC12 (m)/mouse mannosidase IA Al 87). Furthermore; 10 in the P.pastoris strain PBP-3 comprising pFB8 (Saccharomyces SEC12 (m)/mouse mannosidase IA A187), pPB104 (Saccharomyces MNN9 (s)/human GnTI A38) and pPB103 (KlactisAMNN2-2 gene), the hybrid N-glycan GlcNAcMansGlcNAC 2 [b] was detected by MALDI-TOF (Fig. 10). [0213] After identifying transformants with a high degree of mannose trimming, 15 additional experiments were performed to confirm that mannosidase (trimming) activity occurred in vivo and was notpredominantly the result of extracellular activity in the growth medium (Example 13; Figs. 7-9)., Host Cells 20 [02141 Although the present invention is exemplified using a P.pastoris host organism, it is understood by those skilled in the art that other eukaryotic host cells, including other species of yeast and fungal hosts, may be altered as described herein to produce human-like glycoproteins. The techniques described herein for identification and disruption of undesirable host cell glycosylation genes, e.g. 25 OCHJ, is understood to be applicable for these and/or other homologous or functionally related genes in other eukaryotic host cells such as other yeast and fungal strains. As described in Example 16, ochi mnnl genes were deleted from Klactis to engineer a host cell leading to N-glycans that are completely converted to MansGlcNAc 2 by 1, 2 -mannosidase (Fig. 12C). 30 [0215] The AMNN gene was cloned from Klactis as described in Example 16. The nucleic acid and deduced amino acid sequences of the K lactis MNNI gene are shown in SEQ ID NOS: 43 and 44, respectively. Using gene-specific primers, a 77 construct was made to delete the MNN1 gene from the genome of Klactis (Example 16). Host cells depleted in och1 and mnn1 activities produce N-glycans having a Man 9 GlcNAc 2 carbohydrate structure (see, e.g., Fig. 10). Such host cells may be engineered further using, e.g., methods and libraries of the invention, to 5 produce mammalian- or human-like glycoproteins. [0216] Thus, in another embodiment, the invention provides an isolated nucleic acid molecule having a nucleic acid sequence comprising or consisting of at least forty-five, preferably at least 50, more preferably at least 60 and most preferably 75 or more nucleotide residues of the Klactis MNN gene (SEQ ID NO: 43), and 10 homologs, variants and derivatives thereof. The invention also provides nucleic acid molecules that hybridize under stringent conditions to the above-described nucleic acid molecules. Similarly, isolated polypeptides (including muteins, allelic variants, fragments, derivatives, and analogs) encoded by the nucleic acid molecules of the invention are provided. In addition, also provided are vectors, 15 including expression vectors, which comprise a nucleic acid molecule of the invention, as described further herein. Siiilarly host cells transformed with the nucleic acid molecules or vectors of the invention are provided. [0217] Another aspect of the present invention thus relates to a non-human eukaryotic host strain expressing glycoproteins comprising modified N-glycans 20 that resemble those made by human-cells. Performing the methods of the invention in species other than yeast and fungal cells is thus contemplated and encompassed by this invention. It is contemplated that a combinatorial nucleic acid library of the present invention may be used to select constructs that modify the glycosylation pathway in any eukaryotic host cell system. For example, the 25 combinatorial libraries of the invention may also be used in plants, algae and insects, and in other eukaryotic host cells, including mammalian and human cells, to localize proteins, including glycosylation enzymes or catalytic domains thereof, in a desired location along a host cell secretory pathway. Preferably, glycosylation enzymes or catalytic domains and the like are targeted to a subcellular location 30 along the host cell secretory pathway where they are capable of functioning, and preferably, where they are designed or selected to function most efficiently. 78 [0218] As described in Examples 17 and 18, plant and insect cells may be engineered to alter the glycosylation of expressed proteins using the combinatorial library and methods of the invention. Furthermore, glycosylation in mammalian cells, including human cells, may also be modified using the combinatorial library 5 and methods of the invention. It may be possible, for example, to optimize a particular enzymatic activity or to otherwise modify the relative proportions of various N-glycans made in a mammalian host cell using the combinatorial library and methods of the invention. [0219] Examples of modifications to glycosylation which can be affected using a 10 method according to this embodiment of the invention are: (1) engineering a eukaryotic host cell to trim mannose residues from Man 8 GlcNAc 2 to yield a MansGlcNAc 2 N-glycan; (2) engineering eukaryotic host cell to add an N-acetylglucosamine (GlcNAc) residue to MansGlcNAc 2 by action of GlcNAc transferase I; (3) engineering a eukaryotic host cell to functionally express an 15 enzyme such as an N-acetylglucosaminyl Transferase (GnTI, GnTII, GnTIII, GnTIV, GnTV, GnTVI), mannosidasel I, fucosyltransferase (FT), galactosyl tranferase (GaIT) or a sialyltransferase (ST). [0220] By repeating the method, increasingly complex glycosylation pathways can be engineered into a target host, such as a lower eukaryotic microorganism. In 20 one preferred embodiment, the host organism is transformed two or more times with DNA libraries including sequences encoding glycosylation activities. Selection of desired phenotypes may be performed after each round of transformation or alternatively after several transformations have occurred. Complex glycosylation pathways can be rapidly engineered in this manner. 25 Sequential Glycosylation Reactions In a preferred embodiment, such targeting peptide/catalytic domain libraries are designed to incorporate existing information on the sequential nature of glycosylation reactions in higher eukaryotes. Reactions known to occur early in 30 the course of glycoprotein processing require the targeting of enzymes that catalyze such reactions to an early part of the Golgi or the ER. For example, the trimming of MansGlcNAc2 to MansGIcNAc 2 by mannosidases is an early step in 79 complex N-glycan formation. Because protein processing is initiated in the ER and then proceeds through the early, medial and late Golgi, it is desirable to have this reaction occur in the ER or early Golgi. WhIen designing a library for mannosidase I localization, for example, one thus attempts to match ER and early 5 Golgi targeting signals with the catalytic domain of mannosidase I. Integration Sites [02211 As one ultimate goal of this genetic engineering effort is a robust protein production strain that is able to perform well in an industrial fermentation process, 10 the integration of multiple genes into the host (e.g., fungal) chromosome preferably involves careful planning. The engineered strain may likely have to be transformed with a range of different genes, and these genes will have to be transformed in a stable fashion to ensure that the desired activity is maintained throughout the fermentation process. As described herein, any combination of 15 various desired enzyme activities maybe engineered into the fungal protein expression host; e.g., al ylt fe-es mannosidases, fucosyltransferases galactosyltransferases, glucosyltransferases, GlcNAc transferases, ER and Golgi specific transporters (e.g. syn and antiport transporters for UDP-galactose and other precursors), other enzymes involved in the processing of oligosaccharides 20 and enzymes involved in the synthesis of activated oligosaccharide precursors such as UDP-galactose, CMP-N-acetylneuraminic acid. Examples of preferred methods for modifying glycosylation in a lower eukaryotic host cell, such as Pichia pastors, are shown in Table 6. (The HDEL and KDEL signal peptides in the second row of the third column are shown in SEQ ID) NOS: 5 and 6, respectively). 25 80 Table 6. Some preferred embodiments for modifying glycosylation in a lower eukaroytic microorganism Desired Suitable Suitable Sources of Suitable Suitable Structure Catalytic Localization Gene Transporters Activities Sequences Deletions and/or Phosphatases MansGIcNAc 2 a-1,2- MnsI (N-terminus, OCHI none mannosidase S.cerevisiae) MNN4 (murine, Ochi (N-terminus, fNN6 human, S.cerevisiae, Bacillus sp., P.pastoris) A.nidulans) Ktrl Mnn9 Mntl (S.cerevisiae) KDEL, HDEL (C-terminus) GlcNAcMan 5 GIcNAc2 GlcNAc Ochi (N-terminus, OCHI UDP-GlcNAc Transferase I, S.cerevisiae, AMNN4 transporter (human, P.pastoris) MNN6 (human, urine, murine, rat KTRI (N-terminus) K.lactis) etc.) UDPase Mnnl (N-terminus, (human) S.cerevisiae) MntI (N-terminus, S.cerevisiae) GDPase (N-terminus, S.cerevisiae) G1cNAcMan 3 GIcNAc 2 mannosidase KtrI OCHI UDP-GlcNAc II Mni (N-terminus, MNN4 transporter S.cerevisiae) MNN6 (human, murine, Mntl(N-terminus, Klactis) S.cerevislae) UDPase Kre2/Mntl (human) (S.cerevisiae) Kre2 (P.pastoris) Ktr1 (S.cerevislae) Ktrl (P.pastons) Mn1I (S.cerevisiae) GIcNAca. GIcNAc Mum (N-terminus, OCH1 UDP-GIcNAc 4)Man 3 GlcNAc 2 Transferase S.cerevisiae) MNN4 transoter IL lIL IV, V MntI (N-terminus, MNN6 (human, murine, (human, S.cerevisiae) Klactis) murine) Kre2/Mnt1 UDPase (S.cerevisiae) (human) Kre2 (P.pastoris) Ktrl (S.cerevislae) Ktrl (P.pastoris) Manl (S.cerevisiae) 81 Ga1I.

4 )G1cNAc(, j3.-,4- Mnnl (N-terminus, OCH1 UDP-Galactose Man3GlcNAc 2 Galactosyl S.cerevisiae) MNN4 transporter transferase Mntl(N-terminus, MNN6 (human, (human) S.cerevisiae) S.pombe) Kre2/Mntl (S.cerevisiae) Kre2 (P.pastorls) Ktrl (S.cerevisiae) Ktrl (P.pastoris) Mnnl (S. cerevisiae) NANAr 4 )- a-2,6- KTR1 OCH1 Gak.

4 )GIcNAc( 2 Sialyltransfer MNNl (N-terminus, MNN4 transporter Man 3 GIcNAC 2 ase (human) S.cerevisiae) MNN6 (human) a-2,3- MNTl (N-terminus, Sialyltransfer S.cerevistae) ase Kre2/Mntl (S.cerevisiae) Kre2 (P.pastoris) Ktrl (S.cerevisiae) Ktrl (P.pastoris) MNNI (S.cerevisiae) [0222] As any strategy to engineer the formation of complex N-glycans into a host cell such as a lower eukaryote involves both the elimination as well as the addition of particular glycosyltransferase activities, a comprehensive scheme will 5 attempt to coordinate both requirements. Genes that encode enzymes that are undesirable serve as potential integration sites for genes that are desirable. For example, 1,6 mannosyltransferase activity is a hallmark of glycosylation in many known lower eukaryotes. The gene encoding alpha-1,6 mannosyltransferase (OCHJ) has been cloned from S.cerevisiae and mutations in the gene give rise to a 10 viable phenotype with reduced mannosylation. The gene locus encoding alpha-1,6 mannosyltransferase activity therefore is a prime target for the integration of genes encoding glycosyltransferase activity. In a similar manner, one can choose a range of other chromosomal integration sites that, based on a gene disruption event in that locus, are expected to: (1) improve the cells ability to glycosylate in a more 15 human-like fashion, (2) improve the cells ability to secrete proteins, (3) reduce proteolysis of foreign proteins and (4) improve other characteristics of the process that facilitate purification or the fermentation process itself. 82 Target Glycoproteins [02231 The methods described herein are useful for producing glycoproteins, especially glycoproteins used therapeutically in humans. Glycoproteins having specific glycoforms may be especially useful, for example, in the targeting of 5 therapeutic proteins. For example, mannose-6-phosphate has been shown to direct proteins to the lysosome, which may be essential for the proper function of several enzymes related to lysosomal storage disorders such as Giaucher's, Hunter's, Hurler's, Scheie's, Fabry's and Tay-Sachs disease, to mention just a few. Likewise, the addition of one or more sialic acid residues to a glycan side chain , 10 may increase the lifetime of a therapeutic glycoprotein in vivo after administration. Accordingly, host cells (e.g., lower eukaryotic or mammalian) may be genetically engineered to increase the extent of terminal sialic acid in glycoproteins expressed in the cells. Alternatively, sialic acid may be conjugated to the protein of interest in vitro prior to administration using a sialic acid transferase and an appropriate 15 substrate. Changes in growth medium composition may be employed in addition to the expression of enzyme activities involved in human-like glycosylation to produce glycoproteins more closely resembling human forms (S. Weikert, et al., Nature Biotechnology, 1999, 17, 1116-1121; Werner, Noe, et al 1998 Arzneimittelforschung 48(8):870-880; Weikert, Papac et al., 1999; Andersen and 20 Goochee 1994 Cur. Opin. Biotechnol. 5: 546-549; Yang and Butler 2000 Biotechnol.Bioengin. 68(4): 370-380). Specific glycan modifications to monoclonal antibodies (e.g. the addition of a bisecting GlcNAc) have been shown to improve antibody dependent cell cytotoxicity (Umana P., et al. 1999), which may be desirable for the production of antibodies or other therapeutic proteins. 25 [0224] Therapeutic proteins are typically administered by injection, orally, pulmonary, or other means. Examples of suitable target glycoproteins which may be produced according to the invention include, without limitation: erythropoietin, cytokines such as interferon-a, interferon-P, interferon-y, interferon-(o, and granulocyte-CSF, coagulation factors such as factor VIII, factor IX, and human 30 protein C, soluble IgE receptor a-chain, IgG, IgG fragments, IgM, interleukins, urokinase, chymase, and urea Irypsin inhibitor, IGF-binding protein, epidermal growth factor, growth hormone-releasing factor, annexin V fusion protein, 83 angiostatin, vascular endothelial growth factor-2, myeloid progenitor inhibitory factor-1, osteoprotegerin, a-I antitrypsin, DNase II, a- feto proteins, AAT, rhTBP 1 (onercept, aka TNF Binding protein 1), TACI-Ig (transmembrane activator and calcium modulator and cyclophilin ligand interactor), FSH (follicle stimulating 5 hormone), GM-CSF, GLP- 1 w/ and w/o FC (glucagon like protein 1) IL-I receptor agonist, sTNFr (enbrel, aka soluble TNF receptor Fc fusion) ATIEI, rhThrombin, glucocerebrosidase and CTLA4-Ig (Cytotoxic T Lymphocyte associated Antigen 4 - Ig) [0225 The following are examples which illustrate the compositions and 10 methods of this invention. These examples should not be construed as limiting: the examples are included for the purposes of illustration only. EXAMPLE 1 Cloning and Disruption of the OCH1 gene in P.pastoria 15 Generation of an OCHI mutant of P.pastoris: A 1215 bp ORF of the P.pastoris OCH1 gene encoding a putative a-1,6 mannosyltransferase was amplified from P.pastoris genomic DNA (strain X-33, Invitrogen, Carlsbad, CA) using the oligonucleotides 5' ATGGCGAAGGCAGATGGCAGT-3' (SEQ ID NO:7) and 5' 20 TIAGTCCTTCCAACTTCCTfC-3' (SEQ ID NO:8) which were designed based on the P.pastoris OCHJ sequence (Japanese Patent Application Publication No. 8 336387). Subsequently, 2685 bp upstream and 1175 bp downstream of the ORF of the OCHI gene were amplified from a P.pastoris genomic DNA library (Boehm, T. et al. Yeast 1999 May; 15(7):563-72) using the internal oligonucleotides 5' 25 ACTGCCATCTGCCTTCGCCAT-3' (SEQ ID NO:9) in the OCH gene, and 5' GTAATACGACTCACTATAGGGC-3' T7 (SEQ ID NO: 10) and 5' AATTAACCCTCACrAAAGGG-3' T3 (SEQ ID NO:11) oligonucleotides in the backbone of the library bearing plasmid lambda ZAP II (Stratagene, La Jolla, CA). The resulting 5075 bp fragment was cloned into the pCR2. 1 -TOPO vector 30 (Invitrogen, Carlsbad, CA) and designated pBK9. 10226] After assembling a gene knockout construct that substituted the OCH1 reading frame with a hIS4 resistance gene, P.pastoris was transformed and 84 colonies were screened for temperature sensitivity at 37*C. OCH1 mutants of S.cerevisiae are temperature sensitive and are slow growers at elevated temperatures. One can thus identify functional homologs of OCHI in P.pastoris by complementing an OCH1 mutant of S.cerevisiae with a P.pastoris DNA or 5 cDNA library. About 20 temperature sensitive strains were further subjected to a colony PCR screen to identify colonies with a deleted ochI gene. Several och I deletions were obtained. [02271 The linearized pBK9.1, which has 2.1 kb upstream sequence and 1.5 kb down stream sequence of OCH gene cassette carrying Pichia HIS4 gene, was 10 transformed into P.pastoris BK1 [GS 115 (his4 Invitrogen Corp., San Diego, CA) carrying the human IFN-P gene in the AOX locus] to knock out the wild-type OCHI gene. The initial screening of transformants was performed using histidine drop-out medium followed by replica plating to select the temperature sensitive colonies. Twenty out of two hundred histidine-positive colonies showed a 15 temperature sensitive phenotype at 37*C. To exclude random integration of pBK9.1 into the Pichia genome, the 20 temperature-sensitive isolates were subjected to colony PCR using primers specific to the upstream sequence of the integration site and to HIS4 ORF. Two out of twenty colonies were och1 defective and further analyzed using a Southern blot and a Western blot indicating the 20 functional ochi disruption by the ochl knock-out construct. Genomic DNA were digested using two separate restriction enzymes BgtIf and Clalto confirm the ochl knock-out and to confirm integration at the open reading frame. The Western Blot showed ochl mutants lacking a discrete band produced in the GS 115 wild type at 46.2 kDa. 25 EXAMPLE 2 Engineering of P.pastoris with cc-1,2-Mannosidase to Produce MansGlcNAcz-Containing IFN-# Precursors 30 [02281 An a-1,2-mannosidase is required for the trimming of MansGlcNAc 2 to yield MansGlcNAc 2 , an essential intermediate for complex N-glycan formation. While the production of a MansGlcNAc 2 precursor is essential, it is not necessarily sufficient for the production of hybrid and complex glycans because the specific 85 isomer of MansGlcNAc 2 may or may not be a substrate for GnTI. An ochi mutant of P.pastoris is engineered to express secreted human interferon-p under the control of an aox promoter. A DNA library is constructed by the in-frame ligation of the catalytic domain of human mannosidase lB (an c-1, 2 -mannosidase) with a 5 sub-library including sequences encoding early Golgi and ER localization peptides. The DNA library is then transformed into the host organism, resulting in a genetically mixed population wherein individual transformants each express interferon-p as well as a synthetic mannosidase gene from the library. Individual transformant colonies are cultured and the production of interferon is induced by 10 addition of methanol. Under these conditions, over 90% of the secreted protein is glycosylated interferon-p. [02291 Supernatants are purified to remove salts and low-molecular weight contaminants by Cis silica reversed-phase chromatography. Desired transformants expressing appropriately targeted, active a-1,2-mannosidase produce interferon-p 15 including N-glycans of the structure MansGlcNAc 2 , which has a reduced molecular mass compared to the interferon-p of the parent strain.-The purified interferon-p is analyzed by MALDI-TOF mass spectroscopy and colonies expressing the desired form of interferon-p are identified. 20 EXAMPLE 3 Generation of an ochl Mutant Strain Expressing an c-1,2-Mannosidase, GnTI and GnTI for Production of a Human-Like Glycoprotein. [0230] The 1215 bp open reading frame of the P.pastoris OCH1 gene as well as. 2685 bp upstream and 1175 bp downstream was amplified by PCR (see also WO 25 02/00879), cloned into the pCR2. 1 -TOPO vector (Invitrogen) and designated pBK9. To create an ochi knockout strain containing multiple auxotrophic markers, 100 pLg of pJN329, a plasmid containing an ochl::URA3 mutant allele flanked with Sf1I restriction sites was digested with SfiI and used to transform P.pasoris strain JC308 (Cereghino et al. Gene 263 (2001) 159-169) by 30 electroporation. Following incubation on defined medium lacking uracil for 10 days at room temperature, 1000 colonies were picked and re-streaked. URA clones that were unable to grow at 37*C, but grew at room temperature, were 86 subjected to colony PCR to test for the correct integration of the ochl::URA3 mutant allele. One clone that exhibited the expected PCR pattern was designated YJN153. The Kringle 3 domain of human plasminogen (K3) was used as a model protein. A NeoR marked plasmid containing the K3 gene was transformed into 5 strain YJN153 and a resulting strain, expressing K3, was named BK64-1. [0231] Plasmid pPB103, containing the Kluyveromyces lactis MNN2-2 gene which encodes a Golgi UDP-N-acetylglucosamine transporter was constructed by cloning a blunt BgII-HindIII fragment from vector pDL02 (Abeijon et al. (1996) Proc. Nati. Acad. Sci. US.A. 93:5963-5968) into BgllI and BamHI digested and 10 blunt ended pBLADE-SX containing the P.pastoris ADEI gene (Cereghino et al. (2001) Gene 263:159-169). This plasmid was linearized with EcoNI and transformed into strain BK64-1 by electroporation and one strain confirmed to contain the MNN2-2 by PCR analysis was named PBP 1. [0232] A library of mannosidase constructs was generated, comprising in-frame 15 fusions of the leader domains of several type I or type II membrane proteins from S.cerevisiae and P.pastoris fused-with the catalytic domains of several a-1,2 mannosidase genes from human, mouse, fly, worm and yeast sources (see, e.g., W002/00879, incorporated herein by reference). This library was created in a P.pastoris HIS4 integration vector and screened by linearizing with Sall, 20 transforming by electroporation into strain PBP1, and analyzing the glycans released from the K3 reporter protein. One active construct chosen was a chimera of the 988-1296 nucleotides (C-terminus) of the yeast SEC12 gene fused with a N terminal deletion of the mouse a-1,2-mannosidase IA gene (Fig. 3), which was missing the 187 nucleotides. A P.pastoris strain expressing this construct was 25 named PBP2. [0233] A library of GnTI constructs was generated, comprising in-frame fusions of the same leader library with the catalytic domains of GnTI genes from human, worm, frog and fly sources (WO 02/00879). This library was created in a P.pastoris ARG4 integration vector and screened by linearizing with AatI, 30 transforming by electroporation into strain PBP2, and analyzing the glycans released from K3. One active construct chosen was a chimera of the first 120 bp of the S.cerevisiae MNN9 gene fused to a deletion of the human GnTI gene, which 87 was missing the first 154 bp. A P.pastoris strain expressing this construct was named PBP3. [02341 A library of GnTII constructs was generated, which comprised in-frame fusions of the leader library with the catalytic domains of GnTII genes from human 5 and rat sources (WO 02/00879). This library was created in a P.pastoris integration vector containing the NSTR gene conferring resistance to the drug nourseothricin, The library plasmids were linearized with EcoRI, transformed into strain RDP27 by electroporation, and the resulting strains were screened by analysis of the released glycans from purified K3. 10 Materials for the Following Reactions [0235] MOPS, sodium cacodylate, manganese chloride, UDP-galactose and CMP-N-acetylneuraminic acid were from Sigma. Trifluoroacetic acid (TFA) was from Sigma/Aldrich, Saint Louis, MO. Recombinant rat o 2

,

6 -sialyltransferase 15 from Spodopterafrugiperda and Pl, 4 -galactosyltransferase from bovine milk were from Calbiochem (San Diego;-CA). Protein N:glycosidase F, mannosidases, and oligosaccharides were from Glyko (San Rafael, CA). DEAE ToyoPearl resin was from TosoHaas. Metal chelating "HisBind" resin was from Novagen (Madison, WI). 96-well lysate-clearing plates were from Promega (Madison, WI). Protein 20 binding 96-well plates were from Millipore (Bedford, MA). Salts and buffering agents were from Sigma (St. Louis, MO). MALDI matrices were from Aldrich (Milwaukee, WI). Protein Purification 25 [0236] Kringle 3 was purified using a 96-well format on a Beckman BioMek 2000 sample-handling robot (Beckman/Coulter Ranch Cucamonga, CA). Kringle 3 was purified from expression media using a C-terminal hexa-histidine tag. The robotic purification is an adaptation of the protocol provided by Novagen for their HisBind resin. Briefly, a 150uL (AL) settled volume of resin is poured into the 30 wells of a 96-well lysate-binding plate, washed with 3 volumes of water and charged with 5 volumes of 50mM NiSO4 and washed with 3 volumes of binding buffer (5mM imidazole, 0.5M NaCl, 20mM Tris-HCL pH7.9). The protein 88 expression media is diluted 3:2, media/PBS (60mM P04, 16mM KCl, 822mM NaCl pH7.4) and loaded onto the columns. After draining, the columns are washed with 10 volumes of binding buffer and 6 volumes of wash buffer (30mM imidazole, 0.5M NaCl, 20mM Tris-HC1 pH7.9) and the protein is eluted with 6 5 volumes of elution buffer (IM imidazole, 0.5M NaCl, 20mM Tris-HCI pH7.9). The eluted glycoproteins are evaporated to dryness by lyophilyzation. Release of N-linked Glycans [0237] The glycans are released and separated from the glycoproteins by a 10 modification of a previously reported method (Papac, et al. A. J. S. (1998) Glycobiology 8, 445-454). The wells of a 96-well MultiScreen IP (Immobilon-P membrane) plate (Millipore) are wetted with 1O0uL of methanol, washed with 3X150uL of water and 5OuL of RCM buffer (8M urea, 360mM Tris, 3.2mM EDTA pH8.6), draining with gentle vacuum after each addition. The dried protein 15 samples are dissolved in 30uL of RCM buffer and transferred to the wells containing 1OuL of RCM buffer. The wells are drained and washed twice with RCM buffer. The proteins are reduced by addition of 60uL of 0.1M DTT in RCM buffer for lhr at 37 0 C. The wells are washed three times with 300uL of water and carboxymethylated by addition of 60uL of 0.1M iodoacetic acid for 30min in the 20 dark at room temperature. The wells are again washed three times with water and the membranes blocked by the addition of 1OOuL of 1% PVP 360 in water for lhr at room temperature. The wells are drained and washed three times with 300uL of water and deglycosylated by the addition of 3OuL of 10mM NH 4

HCO

3 pH 8.3 containing one milliunit of N-glycanase (Glyko). After 16 hours at 37 0 C, the 25 solution containing the glycans was removed by centrifugation and evaporated to dryness. Matrix Assisted Laser Desorption Ionization Time of Flight Mass Spectrometry 30 [0238] Molecular weights of the glycans were determined using a Voyager DE PRO linear MALDI-TOF (Applied Biosciences) mass spectrometer using delayed extraction. The dried glycans from each well. were dissolved in 15uL of water and 0.5uL spotted on stainless steel sample plates and mixed with 0.5uL of S-DHB matrix (9mg/mL of dihydroxybenzoic acid, lmg/mL of 5-methoxysalicilic acid in 1:1 water/acetonitrile 0.1% TFA) and allowed to dry. [02391 Ions were generated by irradiation with a pulsed nitrogen laser (337nm) 5 with a 4ns pulse time. The instrument was operated in the delayed extraction mode with a 125ns delay and an accelerating voltage of 20kV. The grid voltage was 93.00%, guide wire voltage was 0.10%, the internal pressure was less than 5 X 10 7 torr, and the low mass gate was 875Da. Spectra were generated from the sum of 100-200 laser pulses and acquired with a 2 GHz digitizer. Man 5 GlcNAc 2 10 oligosaccharide was used as an external molecular weight standard. All spectra were generated with the instrument in the positive ion mode. The estimated mass accuracy of the spectra was 0.5%. EXAMPLE 4 15 Engineering a Strain to Produce Galactosyltransferase Galactosyltransferase Reaction [0240] Approximately 2mg of protein (r-K3:hPg [PBP6-5]) was purified by nickel-affinity chromatography, extensively dialyzed against 0.1% TFA, and 20 lyophilized to dryness. The protein was redissolved in 150pL of 50mM MOPS, 20mM MnCl2, pH7.4. After addition of 32.5pg (533nmol) of UDP-galactose and 4mU of 0 1,4-galactosyltransferase, the sample was incubated at 37'C for 18 hours. The samples were then dialyzed against 0.1% TFA for analysis by MALDI-TOF mass spectrometry. 25 [0241] The spectrum of the protein reacted with galactosyltransferase showed an increase in mass consistent with the addition of two galactose moieties when compared with the spectrum of a similar protein sample incubated without enzyme. Protein samples were next reduced, carboxymethylated and deglycosylated with PNGase F. The recovered N-glycans were analyzed by MALDI-TOF mass 30 spectrometry. The mass of the predominant glycan from the galactosyltransferase reacted protein was greater than that of the control glycan by a mass consistent with the addition of two galactose moieties (325.4 Da).

EXAMPLE 5 5 Engineering a Strain to Express functional and active Mannosidase II [0242] To generate a human-like glycoform, a microorganism is engineered to express a mannosidase H enzyme which removes the two remaining terminal mannoses from the structure GIcNAcMansGlcNAc 2 (see Fig. IB). A DNA library 10 including sequences encoding cis and medial Golgi localization signals is fused in frame to a library encoding mannosidase H catalytic domains. The host organism is a strain, e.g. a yeast, that is deficient in hypermannosylation (e.g. an och1 mutant) and provides N-glycans having the structure GlcNAcMansGlcNAc 2 in the Golgi and/or ER. After transformation, organisms having the desired 15 glycosylation phenotype are selected. An in vitro assay is used in one method. The desired structure GlcNAcMan3GlcNAc 2 (but not the undesired GlcNAcMan 5 GlcNAc 2 ) is a substrate for the enzyme GlcNAc Transferase H (see Fig. IB). Accordingly, single colonies may be assayed using this enzyme in vitro in the presence of the substrate, UDP-GlcNAc. The release of UDP is determined 20 either by HPLC or an enzymatic assay for UDP. Alternatively, radioactively labeled UDP-GlcNAc or MALDI-TOF may be used. [0243] The foregoing in vitro assays are conveniently performed on individual colonies using high-throughput screening equipment. Alternatively a lectin binding assay is used. In this case the reduced binding of lectins specific for terminal 25 mannoses allows the selection of transformants having the desired phenotype. For example, Galantus nivalis lectin binds specifically to terminal a-1,3-mannose, the concentration of which is reduced in the presence of operatively expressed mannosidase H activity. In one suitable method, G. nivalis lectin attached to a solid agarose support (available from Sigma Chemical, St. Louis, MO) is used to 30 deplete the transformed population of cells having high levels of terminal a-1,3 mannose. EXAMPLE 6 Engineering a Strain to Express Sialyltransferase 91 [0244] The enzymes a2,3-sialyltransferase and c.2, 6 -sialyltransferase add terminal sialic acid to galactose residues in nascent human N-glycans, leading to mature glycoproteins (see "a 2,3 ST; c2,6 ST" in Fig. IB). In human cells, the reactions occur in the trans Golgi or TGN. Accordingly, a DNA library is 5 constructed by the in-frame fusion of sequences encoding sialyltransferase catalytic domains with sequences encoding trans Golgi or TGN localization signals (Malissard et al. Biochem Biophys Res Commun 2000 Jan 7;267(l):169-73; Borsig et al. Biochem Biophys Res Commun 1995 May 5;210(1):14-20). The host organism is a strain, e.g. a yeast, that is deficient in hypermannosylation (e.g., an 10 och1 mutant), which provides N-glycans having terminal galactose residues in the late Golgi or TGN, and provides a sufficient concentration of CMP-sialic acid in the late Golgi or TGN. Following transformation, transformants having the desired phenotype are selected, e.g., using a fluorescent antibody specific for N-glycans having a terminal sialic acid. In addition, the strains are engineered to produce the 15 CMP-NANA precursors. Sialyltransferase Reaction [0245] After resuspending the (galactosyltransferase reacted) (Example 4) proteins in lOpL of 50mM sodium cacodylate buffer pH6.0, 30Opg (488nmol) of 20 CMP-N-acetylneuraminic acid (CMP-NANA) dissolved in 15pL of the same buffer, and 5pL (2mJ) of recombinant a-2,6 sialyltransferase were added. After incubation at 370C for 15 hours, an additional 200pg of CMP-NANA and ImU of sialyltransferase were added. The protein samples were incubated for an additional 8 hours and then dialyzed and analyzed by MALDI-TOF-MS as above. The 25 spectrum of the glycoprotein reacted with sialyltransferase showed an increase in mass when compared with that of the starting material (the protein after galactosyltransferase reaction). The N-glycans were released and analyzed as above. The increase in mass of the two ion-adducts of the predominant glycan was consistent with the addition of two sialic acid residues (580 and 583Da). 30 EXAMPLE 7 Engineering a strain to Express UDP-GIcNAc Transporter 92 [02461 The cDNA of human Golgi UDP-GlcNAc transporter has been cloned by Ishida and coworkers. (Ishida, N., et al. 1999 J. Biochem. 126(1): 68-77). Guillen and coworkers have cloned the canine kidney Golgi UDP-GIcNAc transporter by phenotypic correction of a Kluyveromyces lactis mutant deficient in Golgi UDP 5 GlcNAc transport. (Guillen, E., et al. 1998). Thus a mammalian Golgi UDP. GlcNAc transporter gene has all of the necessary information for the protein to be expressed and targeted functionally to the Golgi apparatus of yeast. These or other cloned transporter genes may be engineered into a host organism to provide UDP GIcNAc substrates for efficient GnT reactions in the Golgi and/or ER of the host. 10 Fig. 1 OB demonstrates the effect of a strain expressing a K lactis UDP-GlcNAc transporter. In comparison to Fig. 10A, which lacks a UDP-GlcNAc transporter, the effect of adding a UDP-GlcNAc transporter shows a dramatic increase in the production of GlcNAcMansGlcNAc 2 . 15 EXAMPLE 8 Engineering a strain to Express GDP-Fucose Transporter [0247] The rat liver Golgi membrane GDP-fucose transporter has been identified and purified by Puglielli, L. and C. B. Hirschberg 1999 J. Biol. Chem. 20 274(50):35596-35600. The corresponding gene can be identified using standard techniques, such as N-terminal sequencing and Southern blotting using a degenerate DNA probe. The intact gene is then expressed in a host microorganism that also expresses a fucosyltransferase. 25 EXAMPLE 9 Engineering a strain to Express UDP-Galactose Transporter [0248] Human UDP-galactose (UDP-Gal) transporter has been cloned and shown to be active in S.cerevisiae. (Kainuma, M., et al. 1999 Glycobiology 9(2): 133 30 141). A second human UDP-galactose transporter (hUGT1) has been cloned and functionally expressed in Chinese Hamster Ovary Cells. Aoki, K., et al. 1999 J.Biochem. 126(5): 940-950. Likewise, Segawa and coworkers have cloned a UDP-galactose transporter from Schizosaccharohnycespombe (Segawa, H., et al. 1999 Febs Letters 451(3): 295-298). These or other sequences encoding UDP 35 galactose transporter activities may be introduced into a host cell directly or may 93 be used as a component of a sub-library of the invention to engineer a strain having increased UDP-galactose transporter activity. EXAMPLE 10 Engineering a Strain to Express CMP-SIalic Acid Transporter 5 [0249] Human CMP-sialic acid transporter (hCST) has been cloned and expressed in Lec 8 CHO cells by Aoki and coworkers (1999). Molecular cloning of the hamster CMP-sialic acid transporter has also been achieved (Eckhardt and Gerardy Schahn 1997 Eur. J Biochem. 248(1): 187-192). The functional 10 expression of the murine CMP-sialic acid transporter was achieved in Saccharomyces cerevisiae by Berninsone, P., et al. 1997 J. Biol.Chem.272 (19):12616-12619. These or other sequences encoding CMP-sialic acid transporter activities may be introduced into a host cell directly or may be used as a component of a sub-library of the invention to engineer a strain having increased 15 CMIP-sialic acid transporter activity. EXAMPLE 11 Engineering of P.pastoris to Produce MansGlcNA 2 as the Predominant N Glycan Structure Using a Combinatorial DNA Library. 20 [02501 An ochi mutant of P.pastoris (see Examples 1 and 3) was engineered to express and secrete proteins such as the kringle 3 domain of human plasminogen (K3) under the control of the inducible AOXlpromoter. The Kringle 3 domain of human plasminogen (K3) was used as a model protein. A DNA fragment encoding the K3 was amplified using Pfu turbo polymerase (Strategene, La Jolla, CA) and .* 25 cloned into EcoRJ and XbaI sites of pPICZaA (Invitrogen, Carlsbad, CA), resulting in a C-terminal 6- His tag. In order to improve the N-linked glycosylation efficiency of K3 (Hayes et al. 1975 J. Arch. Biochem. Biophys. 171, 651-655), Pro46 was replaced with Ser4 using site-directed mutagenesis. The resulting plasmid was designated pBK64. The correct sequence of the PCR construct was 30 confirmed by DNA sequencing. [0251] A combinatorial DNA library was constructed by the in-frame ligation of murine ox-1,2-mannosidase EB (Genbank AN 6678787) and IA (Genbank AN 6754619) catalytic domains with a sub-library including sequences encoding Cop 94 II vesicle, ER, and early Golgi localization peptides according to Table 6. The combined DNA library was used to generate individual fusion constructs, which were then transformed into the K3 expressing host organism, resulting in a genetically mixed population wherein individual transformants each express K3 as 5 well as a localization signal/mannosidase fusion gene from the library. Individual transformants were cultured and the production of K3 was induced by transfer to a methanol containing medium. Under these conditions, after 24 hours of induction, over 90% of the protein in the medium was K3. The K3 reporter protein was purified from the supernatant to remove salts and low-molecular weight 10 contaminants by Ni-affinity chromatography. Following affmity purification, the protein was desalted by size exclusion chromatography on a Sephadex GlO resin (Sigma, St. Louis, MO) and either directly subjected to MALDI-TOF analysis described below or the N-glycans were removed by PNGase digestion as described below (Release of N-glycans) and subjected to MALDI-TOF analysis Miele et al. 15 1997 Biotechnol. AppL. Biochen. 25, 151-157. [0252] Following this approach, i diverse set of transformants were obtained; some showed no modification of the N-glycans compared to the ochl knockout strain; and others showed a high degree of mannose trimming (Fig. 5D, SE). Desired transfornants expressing appropriately targeted, active a-1,2-mannosidase 20 produced K3 with N-glycans of the structure MansGlcNAc 2 . This confers a reduced molecular mass to the glycoprotein compared to the K3 of the parent ochi deletion strain, a difference which was readily detected by MALDI-TOF mass spectrometry (Fig. 5). Table 7 indicates the relative MansGlcNAc 2 production levels. 25 Table 7. A representative combinatorial DNA library of localization sequences/catalytic domains exhibiting relative levels of MansGlcNAc 2 production. Targeting peptide sequences MNS1(s) MNS1(m) MNS1() SEC12(s) SEC12(m) Mouse mannosidase FB4 FB5 FB6 FB7 FB8 .1AA187 ++ + -+ Mouse mannosidase GB4 GB5 GB6 GB7 OB8 1B Ass ++ + + 4 + 95 Mouse mannosidase GC4 GC5 GC6

-

J GC7- GC8 1B A99 - +++ ++ + Mouse mannosidase GD4 GD5 GD6 GD7 GD8 1Bb70 - - A1- + + Table 8. Another combinatorial DNA library of localization sequences/catalytic domains exhibiting relative levels of MansGcNAc 2 production. 5 Targeting pep tide sequences VAN1(s) VAN1(m) VAN1(l) MNNIO(s) MNN10(m) MNN0(IO) C elegans BC18-5 BC19 BC20 BC27 BC28 BC29 mannosidase 1B +++++ H++ ++-+++ +4-++ ASO C. elegans BB18 BB19 BB20 BB18 BB19 BB20 e mannosidase 1B +++++ +-+ +M+ *- + A31 [0253] Targeting peptides were selected from MNS I (SwissProt P32906) in S.cerevisiae (long, medium and short) (see supra Nucleic Acid Libraries; 10 Combinatorial DNA Library of Fusion Constructs) and SEC12 (SwissProt P11655) in S.cerevisiae (988-1140 nucleotides: short) and (988-1296: medium). Although majority of the targeting peptide sequences were N-terminal deletions, some targeting peptide sequences, such as SEC12 were C-terminal deletions. Catalytic domains used in this experiment were selected from mouse mannosidase 1A with a 15 187 amino acid N-terminal deletion; and mouse mannosidase lB with a 58, 99 and 170 amino acid deletion. The number of (+)s, as used herein, indicates the relative levels of Man 5 GIcNA 2 production. The notation (-) indicates no apparent production of Man 5 GlcNA2. The notation (+) indicates less than 10% production of MansGIcNA 2 . The notation (++) indicates about 10-20% production of 20 Man 5 GlcNA 2 . The notation with (+++) indicates about 20-40% production of MansGIcNA 2 . The notation with (++4+) indicates about 50% production of MansGIcNA 2 . The notation with (+++++) indicates greater than 50% production ofMansGlcNA 2 . 96 Table 9 shows relative amount of MansGlcNAc 2 on secreted K3. Six hundred and eight (608) different strains of P.pastoris, AochI were generated by transforming them with a single construct of a combinatorial genetic library that was generated by fusing nineteen (19) ax-1,2 mannosidase catalytic domains to 5 thirty-two (32) fungal ER, and cis-Golgi leaders. Table 9 Amount of MansGlcNAc 2 on secreted K3 Number of constructs (%) (% of total glycans) N.D.* 19 (3.1) 0r-10% 341(56.1) 10-20% 50 (8.2) 20-40& 75(12.3) 40-60% 72 (11.8) More than 60% 51 (8.4) I Total 608 (100) 10 * Several fusion constructs were not tested because the corresponding plasmids could not be propagated in E.coli prior to transformation into P.pastoris. t Clones with the highest degree of MansGlcNAc 2 trimming (30/51) were further analyzed for mannosidase activity in the supernatant of the medium. The majority (28/30) displayed detectable mannosidase activity in the supernatant (e.g. Fig. 4B). 15 Only two constructs displayed high MansGlcNAc 2 levels, while lacking mannosidase activity in the medium (e.g. Fig. 4C). [02541 Table 7 shows two constructs pFB8 and pGC5, among others, displaying MansGIcNA 2 . Table 8 shows a more preferred construct, pBC18-5, a S.cerevisiae 20 VAN1(s) targeting peptide sequence (from SwissProt 23642) ligated in-frame to a C. elegans mannosidase IB (Genbank AN CAA98114) 80 amino acid N-terminal deletion (Saccharomyces Van1(s)/ C.elegans mannosidase IB A80). This fusion construct also produces a predominant Man 5 GIcNA 2 structure, as shown in Fig. SE. This construct was shown to produce greater than 50% MansGIcNA 2 25 (+4- ). Generation of a combinatorial localization/mannosidase library: 97 [02551 Generating a combinatorial DNA library of a-1, 2 -mannosidase catalytic domains fused to targeting peptides required the amplification of mannosidase domains with varying lengths of N-terminal deletions from a number of organisms. To approach this goal, the full length open reading frames (ORFs) of a-1,2 5 mannosidases were PCR amplified from either cDNA or genomic DNA obtained from the following sources: Homo sapiens, Mus musculus, Drosophila melanogaster, Caenorhabditis elegans, Aspergillus nidulans and Penicillium citrinum. In each case, DNA was incubated in the presence of oligonucleotide primers specific for the desired mannosidase sequence in addition to reagents 10 required to perform the PCR reaction. For example, to amplify the ORF of the M. musculus o-1,2-mannosidase IA, the 5'-primer ATGCCCGTGGGGGGCCTGTTGCCGCTCTTCAGTAGC (SEQ ID NO:12) and the 3'-primer TCATTTCTCTTTGCCATCAATT'CCTTCTTCTGTTCACGG (SEQ ID NO:13) were incubated in the presence of Pfu DNA polymerase 15 (Stratagene, La Jolla, CA) and amplified under the conditions recommended by Stratagene using the cycling parameters: 940C for 1min (1 cycle); 94*C for 30 sec, 68*C for 30 sec, 72"C for 3min (30 cycles). Following amplification the DNA sequence encoding the ORF was incubated at 720C for 5 min with 1U Taq DNA polymerase (Promega, Madison, WI) prior to ligation into pCR2.1-TOPO 20 (Invitrogen, Carlsbad, CA) and transformed into TOP10 chemically competent E. coli, as recommended by Invitrogen. The cloned PCR product was confirmed by ABI sequencing using primers specific for the mannosidase ORF. [0256] To generate the desired N-terminal truncations of each mannosidase, the complete ORF of each mannosidase was used as the template in a subsequent 25 round of PCR reactions wherein the annealing position of the 5'-primer was specific to the 5'-terminus of the desired truncation and the 3'-primer remained specific for the original 3'-terminus of the ORF. To facilitate subcloning of the truncated mannosidase fragment into the yeast expression vector, pIN347 (Fig. 2C) Ascl and PacIrestriction sites were engineered onto each truncation product, 30 at the 5'- and 3'-termini respectively. The number and position of the N-terminal truncations generated for each mannosidase ORF depended on the position of the transmembrane (TM) region in relation to the catalytic domain (CD). For instance, 98 if the stem region located between the TM and CD was less than 150bp, then only one truncation for that protein was generated. If, however, the stem region was longer than 150bp then either one or two more truncations were generated depending on the length of the stem region. 5 [0257] An example of how truncations for the M. musculus mannosidase IA (Genbank AN 6678787) were generated is described herein, with a similar approach being used for the other mannosidases. Fig. 3 illustrates the ORF of the M musculus a-1,2-mannosidase IA with the predicted transmembrane and catalytic domains being highlighted in bold. Based on this structure, three 5' 10 primers were designed (annealing positions underlined in Fig. 3) to generate the A65-, A105- and A187-N-terminal deletions. Using the A65 N-terminal deletion as an example the 5'-primer used was 5' GGCGCGCCGA CTCCAAGCTGCTCAGCGGGGTCCTGTTCCAC-3' (SEQ ID NO:14) (with the AscI restriction site highlighted in bold) in conjunction 15 with the 3'-primer 5' CCTTAATTAATCATTTCTCT

GCCATCAATTTCCTTCTTCTGTTCACGG

3' (SEQ ID NO:15) (with the Pacrestriction site highlighted in bold). Both of these primers were used to amplify a 1561 bp fragment under the conditions outlined above for amplifying the full length M musculus mannosidase IA ORF. 20 Furthermore, like the product obtained for the full length ORF, the truncated product was also incubated with Taq DNA polymerase, ligated into pCR2. I -TOPO (Invitrogen, Carlsbad, CA), transformed into TOP10 and ABI sequenced. After having amplified and confirmed the sequence of the truncated mannosidase fragment, the resulting plasmid, pCR2. 1 -A65mMannIA, was digested with Asct 25 and PacIin New England Biolabs buffer #4 (Beverly, MA) for 16h at 370C. In parallel, the pJN347 (Fig. 2C) was digested with the same enzymes and incubated as described above. Post-digestion, both the pJN347 (Fig. 2C) back-bone and the truncated catalytic domain were gel extracted and ligated using the Quick Ligation Kit (New England Biolabs, Beverly, MA), as recommended by the manufacturers, 30 and transformed into chemically competent DH5a cells (Invitrogen, Carlsbad, CA). Colony PCR was used to confirm the generation of the pJN347-mouse Mannosidase IAA6S construct, 99 [0258] Having generated a library of truncated -1, 2 -mannosidase catalytic domains in the yeast expression vector pJN347 (Fig. 2C) the remaining step in generating the targeting peptide/catalytic domain library was to clone in-frame the targeting peptide sequences (Fig. 2). Both the pJN 3 4 7-mannosidase constructs 5 (Fig. 2D) and the pCR2. 1TOPO-targeting peptide constructs (Fig. 2B) such as were incubated overnight at 37*C in New England Biolabs buffer #4 in the presence of the restriction enzymes NotI and AscI. Following digestion, both the pJN347-mannosidase back-bone and the targeting peptide regions were gel extracted and ligated using the Quick Ligation Kit (New England Biolabs, Beverly, 10 MA), as recommended by the manufacturers, and transfonned into chemically competent DH5a cells (Invitrogen, Carlsbad, CA). Subsequently, the pJN347 targeting peptide/mannosidase constructs were ABI sequenced to confirm that the generated fusions were in-frame. The estimated size of the final targeting peptide/alpha-1,2-mannosidase library contains over 1300 constructs generated by 15 the approach described above. Fig. 2 illustrates construction of the combinatorial DNA library.

Engineering a P.pastors OCH1 knock-out strain with multiple auxotrophic markers. 20 [0259] The first step in plasmid construction involved creating a set of universal plasmids containing DNA regions of the KX1 gene ofP.pastoris (Boehm et al. Yeast 1999 May, 15(7):563-72) as space holders for the 5' and 3' regions of the genes to be knocked out. The plasmids also contained the S.cerevisiae Ura-blaster (Alani et al., Genetics 116, 541-545. 1987) as a space holder for the auxotrophic 25 markers, and an expression cassette with a multiple cloning site for insertion of a foreign gene. A 0.9-kb fragment of the P.pastoris KEX1-5' region was amplified by PCR using primers GGCGAGCTCGGCCTACCCGGCCAAGGCTGAGATCATTGTCCAGC'TCA GA (SEQ ID NO:16) and 30 GCCCACGTCGACG ATCCGmACATCGATTGGAGAGGCTGACACC GCTACTA (SEQ ID NO: 17) and P.pastorls genomic DNA as a template and cloned into the Sac, Sallsites of pUC19 (New England Biolabs, Beverly, MA). 100 The resulting plasmid was cut with BanHI and Sail, and a 0.8-kb fragment of the KX1-3' region that had been amplified using primers CGGGATCCACTAGTATTTAAATCATATGTGCGAGTGTACAACTCTTCCC ACATGG (SEQ ID NO: 18) and 5 GGACGCGTCGACGGCCTACCCGGCCGTACGAGGAATTTCTCGG ATGACTCTTTTC (SEQ ID NO: 19) was cloned into the open sites creating pJN262. This plasmid was cut with BamHI and the 3.8-kb BamHI, Bgill fragment of pNKY51 (Alani et al. 1987) was inserted in both possible orientations resulting in plasmids pJN263 (Fig. 4A) and pJN284 (Fig. 4B). 10 [0260] An expression cassette was created with NotI and Pac as cloning sites. The GAPDH promoter of P.pastoris was amplified using primers CGGGATCCCTCGAGAGATCTITTGTAGAAATGTCTTGGTGCCT (SEQ ID NO:20) and GGACATGCATGCACTAGTGCGGCCGCCACGTGATAGTTGTTCA 15 ATTGATTGAAATAGGGACAA (SEQ ID NO:21) and plasmid pGAPZ-A (Invitrogen) as template and cloned into the BamH, Sphl sites of pUC19 (New England Biolabs, Beverly, MA) (Fig. 4B). The resulting plasmid was cut with SpeI and SphI and the CYC 1 transcriptional terminator region ("TT") that had been amplified using primers 20 CCTTGCTAGCTTAATTAACCGCGGCACGTCCGACGGCGGCCCA CGGGTCCCA (SEQ ID NO:22) and GGACATGCATGCGGATCCCTTAAGACCGGCACGCAA AAAGCCTTCGAGCGTCCC (SEQ ID NO:23) and plasmid pPICZ-A (Invitrogen) as a template was cloned into the open sites creating pJN261 (Fig. 25 4B). [0261] A knockout plasmid for the P.pastoris OCHI gene was created by digesting pJN263 with Sal and Spel and a 2.9-kb DNA fragment of the OCH1-5' region, which had been amplified using the primers GAACCACGTCGACGGCCATTGCG CAAAACCTI CCTATT 30 CAAACACAAGGCATTGC (SEQ ID NO:24) and CTCCAATACTAGTCGAAGATTATCTTCTACGGTGCCTGGACTC (SEQ ID NO:25) and P.pastoris genomic DNA as a template, was cloned into the open sites 101 (Fig. 4C). The resulting plasmid was cut with EcoRJ and PmeI and a 1.0-kb DNA fragment of the OCHI-3' region that had been generated using the primers TGGAAGGTTTAAACAAAGCTAGAGTAAAATAGATATAGCGAG ATTAGAGAATG (SEQ ID NO:26) and 5 AAGAATTCG TGGAAGGCCTTGTACCTTGATGTAGTTCCCGTT TTCATC (SEQ ID NO:27) was inserted to generate pJN298 (Fig. 4C). To allow for the possibility to simultaneously use the plasmid to introduce a new gene, the BamHJ expression cassette of pJN261 (Fig. 4B) was cloned into the unique BamHI site of pJN298 (Fig. 4C) to create pJN299 (Fig. 4E). 10 [0262] The P.pastoris Ura3-blaster cassette was constructed using a similar strategy as described in Lu. P., et al. 1998 (Cloning and disruption of the p isopropylmalate dehydrogenase gene (Leu2) of Pichia stipidis with URA3 and recovery of the double auxotroph. Apple. Microbiol. Biotechnol. 49, 141-146.) A 2.0-kb PstI, Spel fragment of P.pastoris URA3 was inserted into the PstI, XbaI 15 sites of pUC19 (New England Biolabs, Beverly, MA) to create pJN306 (Fig. 4D). Then a 0.7-kb Sac, PvuI DNA fragment of the lacZ open reading frame was cloned into the Sac!, Smalsites to-yield pJN308 (Fig. 4D). Following digestion of pJN308 (Fig. 4D) with PstI, and treatment with T4 DNA polymerase, the SacI Pvuf fragment from lacZ that had been blunt-ended with T4 DNA polymerase was 20 inserted generating pJN315 (Fig. 4D). The lacZ/URA3 cassette was released by digestion with Sac! and SphI, blunt ended with T4 DNA polymerase and cloned into the backbone of pJN299 that had been digested with Pme! and Aflit and blunt ended with T4 DNA polymerase. The resulting plasmid was named pJN329 (Fig. 4E). 25 [0263] A HIS4 marked expression plasmid was created by cutting pJN261 (Fig. 4F) with EcoICRI (Fig. 4F). A 2.7kb fragment of the Pichia pastoris IfS4 gene that had been amplified using the primers GCCCAAGCCGCCTTAAGGGATCTCCTGATGACTGACTCACTGATAATA AAAATACGG (SEQ ID NO:28) and 30 GGGCGCGTATTTAATACTAGTGGATCTATCGAATCTAAATGTAAGTTA AAATCTCTAA (SEQ ID NO:29) cut with NgoMTV and Swaland then blunt ended using T4 DNA polymerase, was then ligated into the open site. This 102 plasmid was named pJN337 (Fig. 4F). To construct a plasmid with a multiple cloning site suitable for fusion library construction, pJN337 was cut with NotI and Pad and the two oligonucleotides GGCCGCCTGCAGATTTAAATGAATTCGGCGCGCCTTAAT (SEQ ID 5 NO:30) and TAAGGCGCGCCGAATTCATTTAAATCTGCGGGC (SEQ ID NO:3 1), that had been annealed in vitro were ligated into the open sites, creating pJN347 (Fig. 4F). [0264] To create an och1 knockout strain containing multiple auxotrophic markers, 100 pg of pJN329 was digested with Sf1 and used to transform P.pastoris 10 strain JC308 (Cereghino et al. Gene 263 (2001) 159-169) by electroporation. Following transformation, the URA dropout plates were incubated at room temperature for 10 days. One thousand (1000) colonies were picked and restreaked. All 1000 clones were then streaked onto 2 sets of URA dropout plates. One set was incubated at room temperature, whereas the second set was incubated 15 at 37"C. The clones that were unable to grow at 37*C, but grew at room temperature, were subjected to colony PCR to test for the correct OCHI knockout. One clone that showed the expected PCR signal (about 4.5 kb) was designated YJN153. Example 12 20 Characterization of the Combinatorial DNA Library [0265] Positive transformants screened by colony PCR confirming integration of the mannosidase construct into the P.pastoris genome were subsequently grown at room temperature in 50ml BMGY buffered methanol-complex medium consisting of 1% yeast extract, 2% peptone, 100 mM potassium phosphate buffer, pH 6.0, 25 1.34% yeast nitrogen base, 4 X 10-'% biotin, and 1% glycerol as a growth medium) until ODc 0 oom 2-6 at which point they were washed with 1Oml BMMY (buffered methanol-complex medium consisting of 1% yeast extract, 2% peptone, 100 mM potassium phosphate buffer, pH 6.0,1.34% yeast nitrogen base, 4 X 10 5% biotin, and 1.5% methanol as a growth medium) media prior to induction of the 30 reporter protein for 24 hours at room temperature in 5ml BMMY. Consequently, the reporter protein was isolated and analyzed by mass spectrophotometry and HPLC to characterize its glycan structure. Using the targeting peptides in Table 6, 103 mannosidase catalytic domains localized to either the ER or the Golgi showed significant level of trimming of a glycan predominantly containing Man8GIcNAc 2 to a glycan predominantly containing Man 5 GlcNAc 2 . This is evident when the glycan structure of the reporter glycoprotein is compared between that of 5 P.pastoris ochi knock-out in Figs. 5C, 6C and the same strain transformed with M musculus mannosidase constructs as shown in Figs. 5D, 5E, 6D - 6F. Figs. 5 and 6 show expression of constructs generated from the combinatorial DNA library which show significant mannosidase activity in P.pastoris. Expression of pGC5 (Saccharomyces MNS1(m)/mouse mannosidase IB A99) (Fig. 5D, 6E) 10 produced a protein which has approximately 30% of all glycans-timmed to MansGlcNAc 2 , while expression of pFB8 (Saccharomyces SEC12(m)/mouse mannosidase IA A187) (Fig. 6F) produced approximately 50% Man 5 GlcNAc 2 and expression of pBC18-5 (Saccharomyces VAN1(s)/ C. elegans mannosidase LB A80) (Fig. 5E) produced 70% MansGlcNAc 2 . 15 Release of N-glycans (0266] The glycans were released and separated from the glycoproteins by a modification of a previously reported method (Papac et al. 1998 Glycobiology 8, 445-454). After the proteins were reduced and carboxymethylated and the 20 membranes blocked, the wells were washed three time with water. The protein was deglycosylated by the addition of 30 pl of 10 mM NH4HCO3 pH 8.3 containing one milliunit of N-glycanase (Glyko, Novato, CA). After 16 hr at 37 0 C, the solution containing the glycans was removed by centrifugation and evaporated to dryness. 25 Matrix Assisted Laser Desorption Ionization Time of Flight Mass Spectrometry [02671 After the N-glycans were released by PNGase digestion, they were analyzed by Matrix Assisted Laser Desorption Ionization Time of Flight Mass 30 Spectrometry. Molecular weights of the glycans were determined using a Voyager DE PRO linear MALDI-TOF (Applied Biosciences) mass spectrometer using delayed extraction. The dried glycans from each well were dissolved in 15 g1 of 104 water and 0.5 pl was spotted on stainless steel sample plates and mixed with 0.5 pl of S-DHB matrix (9 mg/ml of dihydroxybenzoic acid, 1 mg/ml of 5 methoxysalicilic acid in 1:1 water/acetonitrile 0.1% TFA) and allowed to dry. Ions were generated by irradiation with a pulsed nitrogen laser (337 nm) with a 4 ns 5 pulse time. The instrument was operated in the delayed extraction mode with a 125 ns delay and an accelerating voltage of 20 kV. The grid voltage was 93.00%, guide wire voltage was 0.1%, the internal pressure was less than 5 X 10-7 torr, and the low mass gate was 875 Da. Spectra were generated from the sum of 100-200 laser pulses and acquired with a 500 MHz digitizer. MansGlcNAc 2 oligosaccharide was 10 used as an external molecular weight standard. All spectra.were generated with the instrument in the positive ion mode. Example 13: Trimming in vivo by alpha-1,2-mannosidase [02681 To ensure that the novel engineered strains of Example 11 in fact 15 produced the desired MansGlcNAc 2 structure in vivo, cell supernatants were tested for mannosidase activity (see Figs. 7 - 9). For each construct/host strain described below, HPLC was performed at 30*C with a 4.0mm x 250 mm column of Altech (Avondale, PA, USA) Econosil-NH 2 resin (5pm) at a flow rate of 1.0 mI/win for 40 min. In Figs. 7 and 8, degradation of the standard Man 9 GlcNAc 2 [b] was 20 shown to occur resulting in a peak which correlates to Man8GlcNAc 2 . In Fig. 7, the MangGlcNAc 2 [b] standard eluted at 24.61 min and Man5GlcNAc 2 [a] eluted at 18.59 min. In Fig. 8, MangGlcNAc 2 eluted at 21.31 min and Man 5 GlcNAc 2 at 15.67 min. In Fig. 9, the standard MansGlcNAc 2 [b] was shown to elute at 20.88 min. 25 [0269] P.pastoris cells comprising plasmid pFB8 (Saccharomyces SEC12 (m)/mouse mannosidase IA A 187) were grown at 30*C in BMGY to an OD600 of about 10. Cells were harvested by centrifugation and transferred to BMMY to induce the production of K3 (kringle 3 from human plasminogen) under control of an AOX promoter. After 24 hours of induction, cells were removed by 30 centrifugation to yield an essentially clear supernatant. An aliquot of the supernatant was removed for mannosidase assays and the remainder was used for the recovery of secreted soluble K3. A single purification step using CM 105 sepharose chromatography and an elution gradient of 25mM NaAc, pH5.0 to 25mM NaAc, pH5.0, IM NaCl, resulted in a 95% pure K3 eluting between 300 500mM NaCl. N-glycan analysis of the K3 derived glycans is shown in Fig. 6F. The earlier removed aliquot of the supernatant was further tested for the presence 5 of secreted mannosidase activity. A commercially available standard of 2 aminobenzamide-labeled N-linked-type oligomannose 9 (Man9-2-AB) (Glyko, Novato, CA) was added to: BMMY (Fig. 7A), the supernatant from the above aliquot (Fig. 7B), and BMMY containing lOng of 75mU/mL of a-1,2-mnannosidase from Trichoderma reesei (obtained from Contreras et al., WO 02/00856 A2) (Fig. 10 7C). After incubation for 24 hours at room tempeature,.samples were analyzed by amino silica HPLC to determine the extent of mannosidase trimming. [0270] P.pastoris cells comprising plasmid pGC5 (Saccharomyces MNS1I(m)/mouse mannosidase [B A99) were similarly grown and assayed. Cells were grown at room temperature in BMGY to an OD600 of about 10. Cells were 15 harvested by centrifugation and transferred to BMMY to induce the production of K3 under control of an A OXl promoter. After 24 hours of induction, cells were removed by centrifugation to yield an essentially clear supernatant. An aliquot of the supernatant was removed for mannosidase assays and the remainder was used for the recovery of secreted soluble K3. A single purification step using CM 20 sepharose chromatography and an elution gradient of 25mM NaAc, pH5.0 to 25mM NaAc, pH5.0, IM NaCl, resulted in a 95% pure K3 eluting between 300 500mM NaCl. N-glycan analysis of the K3 derived glycans is shown in Fig. 5D. The earlier removed aliquot of the supernatant was further tested for the presence of secreted mannosidase activity as shown in Fig. 8B. A commercially available 25 standard of Man9-2-AB (Glyko, Novato, CA) were added to: BMMY (Fig. 8A), supernatant from the above aliquot (Fig. 8B), and BMMY containing lOng of 75mU/mL of a-1,2-mannosidase from Trichoderma reesei (obtained from Contreras et al., WO 02/00856 A2) (Fig. 8C). After incubation for 24 hours at room temperature, samples were analyzed by amino silica HPLC to determine the 30 extent of mannosidase trimming. [0271] Man9-2-AB was used as a substrate and it is evident that after 24 hours of incubation, mannosidase activity was virtually absent in the supernatant of the 106 pFB8 (Saccharomyces SEC12 (m)/mouse mannosidase IA A187) strain digest (Fig. 7B) and pGC5 (Saccharoinyces MNS1(m)/mouse mannosidase IB A99) strain digest (Fig. 8B) whereas the positive control (purified o'-1, 2 -mannosidase from T. reesei obtained from Contreras) leads to complete conversion of MangGlcNAc 2 to 5 MansGlcNAc 2 under the same conditions, as shown in Fig. 7C and 8C. This is conclusive data showing in vivo mannosidase trimming in P.pastoris pGC5 strain; and pFB8 strain, which is distinctly different from what has been reported to date (Contreras et al., WO 02/00856 A2). [02721 Fig. 9 further substantiates localization and activity of the mannosidase 10 enzyme. P.pastoris comprising pBC18-5 (Saccharomyces VANi(s)/ C. elegans mannosidase IB A80) was grown at room temperature in BMGY to an OD600 of about 10. Cells were harvested by centrifugation and transferred to BMMY to induce the production of K3 under control of an AOX1 promoter. After 24 hours of induction, cells were removed by centrifugation to yield an essentially clear 15 supernatant. An aliquot of the supernatant was removed for mannosidase assays and the remainder was used for the recovery of secreted soluble K3. A single purification step using CM-sepharose chromatography and an elution gradient 25mM NaAc, pH5.0 to 25mM NaAc, pH5.0, 1M NaCl, resulted in a 95% pure K3 eluting between 300-500mM NaCl. N-glycan analysis of the K3 derived glycans is 20 shown in Fig. 5E. The earlier removed aliquot of the supernatant was further tested for the presence of secreted mannosidase activity as shown in Fig. 9B. A commercially available standard of Man8-2-AB (Glyko, Novato, CA) was added to: BMMY (Fig. 9A), supernatant from the above aliquot pBC18-5 (Saccharomyces VANJ(s)/ C. elegans mannosidase IB A80) (Fig. 9B), and BMMY 25 containing media from a different fusion construct pDD28-3 (Saccharomyces MN1O(m) (from SwissProt 50108)/H. sapiens mannosidase IB A99) (Fig. 9C). After incubation for 24 hours at room temperature, samples were analyzed by amino silica HPLC to determine the extent of mannosidase trimming. Fig. 9B demonstrates intracellular mannosidase activity in comparison to a fusion construct 30 pDD28-3 (Saccharomyces MIN1O(m) H. sapiens mannosidase IB A99) exhibiting a negative result (Fig. 9C). 107 Example 14: pH Optimum Assay of Engineered ae,2-mannosidase [02731 P.pastoris cells comprising plasmid pBB27-2 (Saccharomyces MNN]0 5 (s) (from SwissProt 50108)/C. elegans mannosidase IB A3 1) were grown at room temperature in BMGY to an OD600 of about 17. About 80pL of these cells were inoculated into 600 L BMGY and were grown overnight. Subsequently, cells were harvested by centrifugation and transferred to BMNY to induce the production of K3 (kringle 3 from human plasminogen) under control of an AOX1 10 promoter. After 24 hours of induction, cellswmere removed by centrifugation to yield an essentially clear supernatant (pH 6.43). The supernatant was removed for mannosidase pH optimum assays. Fluorescence-labeled ManaGlcNAc 2 (0.5 pg) was added to 20tL of supernatant adjusted to various pH (Fig. 11) and incubated for 8 hours at room temperature. Following incubation the sample was analyzed by 15 HPLC using an Econosil NH2 4.6 X 250 mm, 5 micron bead, amino-bound silica column (Altech, Avondale, PA). The flow rate was 1.0 ml/min for 40 min and the column was maintained to 30*C. After eluting isocratically (68% A:32% B) for 3 min, a linear solvent gradient (68% A:32% B to 40% A:60% B) was employed over 27 min to elute the glycans (18). Solvent A (acetonitrile) and solvent B 20 (ammonium formate, 50 mM, pH 4.5. The column was equilibrated with solvent (68% A:32% B) for 20 min between runs. Example 15: Engineering of P.pastoris to Produce N-glycans with the Structure GlcNAcMansGlcNAc 2 25 [02741 GlcNAc Transferase I activity is required for the maturation of complex and hybrid N-glycans (U.S. Pat. No. 5,834,251). MansGlcNAc 2 may only be trimmed by mannosidase H, a necessary step in the formation of human glycoforms, after the addition of N-acetylglucosamine to the terminal a-1,3 mannose residue of the trimannose stem by GlcNAc Transferase I (Schachter, 30 1991 Glycobiology 1(5):453-461). Accordingly, a combinatorial DNA library was prepared including DNA fragments encoding suitably targeted catalytic domains of GlcNAc Transferase I genes from C. elegans and Homo sapiens; and localization 108 sequences from GLS, MNS, SEC, MNN9, VAN1, ANPJ, HOC], MNNJO, MNNI, MNT1, KTR1, KTR2, MNN2, MNN5, YURI, AMNN, and MNN6 from S.cerevisiae and P.pastoris putative a-1,2-mannosyltransferases based on the homology from Scerevisiae: D2, D9 and J3, which are KTR homologs. Table 10 includes but 5 does not limit targeting peptide sequences such as SEC and OCH1, from P.pastoris and Klactis GnTI, (See Table 6 and Table 10) Table 10. A representative combinatorial library of targeting peptide sequences/ catalytic domain for UDP-N-Acetylglucosaminyl Transferase I (GnTI) 10 -"Tageting peptide OCHI(s) OCH(m) OCBWI) MNN9(s) MNN9(m) W Human, GnTI, A38 PB105 PB106 PB107 PB104 N/A - Human, GnT A86 NB12 NB13 NB14 NB15 NB C.elegans, GnTI, A88 OA12 OA13 OA14 OA15 OA16 C.elegans, GnTI, A35 PA12 PA13 PA14 PA15 PA16 C.elegans, GnTI, A63 PB12 PB13 PB14 PB15 PB16 oXJeas, GnTI, A33 QA12 QA13 QA14 QA15 QA16 Xeavs, GuTI, A103 QB12 QB13 QB14 QB15 QB 16 [0275] Targeting peptide sequences were selected from OCHTin P.pastoris (long, medium and short) (see Example 11) and MNN9 (SwissProt P39107) in 15 S.cerevisiae short, and medium. Catalytic domains were selected from human GnTI with a 38 and 86 amino acid N-terminal deletion, C. elegans (gly-12) GnTI with a 35 and 63 amino acid deletion as well as C. elegans (gly-14) GnTI with a 88 amino acid N-terminal deletion and X leavis GnTI with a 33 and 103 amino acid N-terminal deletion, respectively. 20 [0276] A portion of the gene encoding human N-acetylglucosaminyl Transferase I (MGATI, Accession# NM002406), lacking the first 154 bp, was amplified by PCR using oligonucleotides 5'-TGGCAGGCGCGCCTCAGTCAGCGCTCTCG 3' (SEQ ID NO:32) and 5'-AGGTTAATTA AGTGCTAATTCCAGCTAGO-3' (SEQ ID NO:33) and vector pHG4.5 (ATCC# 79003) as template. The resulting 25 PCR product was cloned into pCR2.1 -TOPO and the correct sequence was confirmed. Following digestion with Asct and PacI the truncated GnTI was inserted into plasmid pJN346 to create pNA. After digestion of pJN271 with NotI 109 and Asc, the 120 bp insert was ligated into pNA to generate an in-frame fusion of the MNN9 transmembrane domain with the GnTI, creating pNA15. [0277] The host organism is a strain of P.pastoris that is deficient in hypermannosylation (e.g. an och1 mutant), provides the substrate UDP-GIcNAc in 5 the Golgi and/or ER (i.e. contains a functional UDP-GlcNAc transporter), and provides N-glycans of the structure MansGlcNAc2 in the Golgi and/or ER (e.g. P.pastoris pFB8 (Saccharomyces SEC12 (m)/mouse mannosidase IA A 187) from above). First, P.pastoris pFB8 was transformed with pPB103 containing the Kluyveromyces lactis MNN2-2 gene (Genbank AN AF106080) (encoding UDP 10 GlcNAc transporter) cloned into BaMIWfand BglI site of pBLADE-SX plasmid (Cereghino et al. Gene 263 (2001) 159-169). Then the aforementioned combinatorial DNA library.encoding a combination of exogenous or endogenous GnTL/localization genes was transformed and colonies were selected and analyzed for the presence of the GnTI construct by colony PCR. Our transformation and 15 integration efficiency was generally above 80% and PCR screening can be omitted once robust transformation parameters have been established. Protein Purification [0278] K3 was purified from the medium by Ni-affinity chromatography 20 utilizing a 96-well format on a Beckman BioMek 2000 laboratory robot The robotic purification is an adaptation of the protocol provided by Novagen for their HisBind resin. Another screening method may be performed using a specific terminal GlcNAc binding antibody, or a lectin such as the GSII lectin from Grffonia simplipfcolia, which binds terminal GlcNAc (BY Laboratories, San 25 Mateo,.CA). These screens can be automated by using lectins or antibodies that have been modified with fluorescent labels such as FITC or analyzed by MALDI TOF. [0279] Secreted K3 can be purified by Ni-affinity chromatography, quantified and equal amounts of protein can be bound to a high protein binding 96-well plate. 30 After blocking with BSA, plates can be probed with a GSII-FACS lectin and screened for maximum fluorescent response. A preferred method of detecting the above glycosylated proteins involves the screening by MALDI-TOP mass 110 spectrometry following the affinity-purification of secreted K3 from the supernatant of 96-well cultured transformants. Transformed colonies were picked and grown to an OD600 of 10 in a 2ml, 96-well plate in BMGY at 30 0 C. Cells were harvested by centrifugation, washed in BMMY and resuspended in 250ul of 5 BMMY. Following 24 hours of induction, cells were removed by centrifugation, the supernatant was recovered and K3 was purified from the supernatant by Ni affinity chromatography. The N-glycans were released and analyzed by MALDI TOF delayed extraction mass spectrometry as described herein. [0280] In summary, the methods of the invention yield strains of P.pastoris that 10 produce GlcNAcMansGlcNAc 2 in high yield, as shown in Fig. 10B. At least 60% of the N-glycans are GlcNAcMansGlcNAc 2 . To date, no report exists that describes the formation of GlcNAcMan 5 GlcNAc2 on secreted soluble glycoproteins in any yeast. Results presented herein show that addition of the UDP-GIcNAc transporter along with OnTI activity produces a predominant 15 G1cNAcMansGIcNAC 2 structure, which is confirmed by the peak at 1457 (m/z) (Fig. 10B). Construction of strain PBP-3: [02811 The P.pastoris strain expressing K3, (Aochi, arg-, ade-, his-) was transformed successively with the following vectors. First, pFB8 (Saccharomyces 20 SEC12 (m)/mouse mannosidase IA A187) was transformed in the P.pastoris strain by electroporation. Second, pPB103 containing Kluyveromyces lactis MNN2-2 gene (Genbank AN AF106080) (encoding UDP-GlcNAc transporter) cloned into pBLADB-SX plasmid (Ceregliino et al. Gene 263 (2001) 159-169) digested with BamHI and BglII enzymes was transformed in the P.pastoris strain. Third, 25 pPB104 containing Saccharomyces MN9(s)/human GnTI A38 encoding gene cloned as Not-PacI fragment into pJN336 was transformed into the P.pastoris strain. Example 16: Engineering XLiacis Cells to Produce N-glycans with the 30 Structure MansGlcNAc 2 Identification and Disruption of the Kgactis OCH1 ene 111 [02821 The OCH1 gene of the budding yeast S.cerevisiae encodes a 1,6 mannosyltransferase that is responsible for the first Golgi localized mannose addition to the Man8GlcNAc 2 N-glycan structure on secreted proteins (Nakayama et al, JBiol Chem.; 268(35):26338-45 (Dec 15, 1993)). This mannose transfer is 5 generally recognized as the key initial step in the fungal specific polymannosylation of N-glycan structures (Nakanishi-Shindo et al, 1993; Nakayama et al, 1992; Morin-Ganet et al, traffic 1(l):56-68. (Jan 2000)). Deletion of this gene in S. cerevisiae results in a significantly shorter N-glycan structure that does not include this typical polymannosylation or a growth defect at 10 elevated temperatures (Najcayamaet al, EMBOJ.;11(7):2511-9 (Jul 1992)). [02831 The Och1p sequence from S.cerevisiae was aligned with known homologs from Candida albicans (Genbank accession # AAL49987), and P.pastoris (B.K. Choi et al. in prep) along with the Hoc] proteins of Scerevisiae (Neiman et al, Genetics, 145(3):637-45 (Mar 1997) and Klactis (PENDANT EST 15 database) which are related but distinct mannosyltransferases. Regions of high homology that were in common among OchIp homologs but distinct from the Hocip homologs were used to design pairs of degenerate primers that were directed against genomic DNA from the K.lactis strain MG1/2 (Bianchi et al, Current Genetics 12, 185-192 (1987)). PCR amplification with primers RCD33 20 (CCAGAAGAATTCAATTYTGYCARTGG) (SEQ ID NO:34) and RCD34 (CAGTGAAAATACCTGGNCCNGTCCA) (SEQ ID NO:35) resulted in a 302 bp product that was cloned and sequenced and the predicted translation was shown to have a high degree of homology to Och1 proteins (>55% to S.cerevisiae Och1p). [0284] The 302 bp PCR product was used to probe a Southern blot of genomic 25 DNA from Klactis strain (MG1/2) with high stringency (Sambrook et al, 1989). Hybridization was observed in a pattern consistent with a single gene indicating that this 302 bp segment corresponds to a portion of the Klactis genome and Klactis (KiOCHI) contains a single copy of the gene. To clone the entire KLOCHlgene, the Southern blot was used to map the genomic locus. Accordingly, 30 a 5.2 kb BamHIPstI fragment was cloned by digesting genomic DNA and ligating those fragments in the range of 5.2 kb into pUC19 (New England Biolabs, Beverly, MA) to create a Klactis subgenomic library. This subgenomic library was 112 transformed into E. coli and several hundred clones were tested by colony PCR using RCD 33/34. The 5.2 kb clone containing the predicted KIOCHJ gene was sequenced and an open reading frame of 1362 bp encoding a predicted protein that is 46.5% identical to the S.cerevisiae OCHI gene. The 5.2 kb sequence was used 5 to make primers for construction of an ochl::KAMe deletion allele using a PCR overlap method (Davidson et al, Microbiology. 148(Pt 8):2607-15. Aug 2002). This deletion allele was transformed into two Klactis strains and G418 resistant colonies selected. These colonies were screened by both PCR and for temperature sensitivity to obtain a strain deleted for the OCH1 ORF. The results of the 10 experiment show strains which reveal a mutant PCR pattern, which were characterized by analysis of growth at various temperatures and N-glycan carbohydrate analysis of secreted and cell wall proteins following PNGase digestion. The ochi mutation conferred a temperature sensitivity which allowed strains to grow at 30*C but not at 35*C. Fig. 12A shows a MALDI-TOF analysis 15 of a wild type K lactis strain producing N-glycans of MansGlcNAc 2 [c] and higher. Identification, Cloning, and Disruption of the Klactis MNN gene [0285] S.cerevisiae MNAN is the structural gene for the Golgi a-1,3 mannosyltransferase. The product of MNN1 is a 762-amino acid type H1 membrane 20 protein (Yip et al., Proc Natl Acad Sci USA. 91(7):2723-7. (1994)). Both N linked and O-linked oligosaccharides isolated from mnn1 mutants lack a-1,3 mannose linkages (Raschke et al., JBiol Chem., 248(13):4660-6. (Jul 10, 1973). [0286] The Mnnlp sequence from S.cerevisiae was used to search the Klactis translated genomic sequences (PEDANT). One 405 bp DNA sequence encoding a 25 putative protein fragment of significant similarity to MnnIp was identified. An internal segment of this sequence was subsequently PCR amplified with primers KMN1 (TGCCATCTTIrAGGTCCAGGCCCGTTC) (SEQ ID NO:36) and KMN2 (GATCCCACGACGCATCGTATITCTrTC), (SEQ ID NO:37) and used to probe a Southern blot of genomic DNA from Klactis strain (MG1/2). Based on 30 the Southem hybridization data a 4.2 Kb BamWI-Pstlfragment was cloned by generating a size-selected library as described herein. A single clone containing the Klactis MNN gene was identified by whole colony PCR using primers KMN1 113 (SEQ ID NO:36) and KMN2 (SEQ ID NO:37) and sequenced. Within this clone a 2241 bp ORF was identified encoding a predicted protein that was 34% identical to the S. cerevisiae MNN1 gene. Primers were designed for construction of a mnnl::N t/ deletion allele using the PCR overlap method (Davidson et al. 2002). 5 [02871 This disruption allele was transformed into a strain of Klactis by electroporation and Noursethoicin resistant transformants were selected and PCR amplified for homologous insertion of the disruption allele. Strains that reveal a mutant PCR pattern may be subjected to N-glycan carbohydrate analysis of a known reporter gene. 10 [0288] Fig. 12B.depicts the N-glycans from the K.lactis och1 mnn1 deletion strain observed following PNGase digestion the MALDI-TOF as described herein. The predominant peak at 1908 (m/z) indicated as [d] is consistent with the mass of Man 9 GlcNAc 2 . 15 Example 17: Engineering Plant Cells To Express GlcNAc Transferases or Galactosyltransferases [0289] GlcNAc transferase IV is required for the addition of $ 1,4 GlcNAc to the a-1,6 mannose residue and the a-1,3 mannose residues in complex N-glycans in 20 humans. So far GIcNAc transferase IV has not been detected in or isolated from plants. A transgenic plant that is capable of adding human-like N-glycans must therefore be engineered to express GlcNAc transferase IV. Thus, the plant host cell or transgenic plant must also localize an expressed GlcNAc transferase IV to the correct intracellular compartment in the host so that the enzyme can add the 25 0 1,4 GIcNAc to the appropriate mannose residues. [0290] There is some evidence that glycosyltransferases from mammals and plants have similar targeting signals. For example, a full-length rat a-2,6 sialyltransferase has been shown to correctly localize to the trans Golgi network in transgenic arabidopsis though not necessarily active (Wee B et al. Plant Cell 1998 30 Oct;10(10):1759-68). A fusion construct having fifty-two N-terminal amino acids from a- 2 ,6-sialyltransferase fused to a green fluorescent reporter protein (GFP) was also shown to correctly localize to the plant Golgi (Boevink et aL Plant J 1998 Aug;15(3):441-7). Two mammalian proteins - TGN30 and furin - and AtELP, an 114 arabidopsis integral membrane protein (Sanderfoot et al. Proc Natl Acad Sci USA 1998 Aug 18;95(17):9920-5), which localize to the trans Golgi network, each contain a tyrosine tetrapeptide motif which targets them to the Golgi, probably by a recycling mechanism via the plasma membrane. Although mammals and plants 5 appear to share some common mechanisms related to protein targeting, exogenous glycosylases may nonetheless not target correctly in a plant cell, however, localization does not necessarily equal enzyme activity. It therefore becomes essential to devise means to correctly target in a plant cell these enzymes and/or other enzymes which participate in forming complex, human-like N-glycans. 10 102911 Glycosylation enzymes are integral membrane proteins which reside in the endoplasmic reticulum and Golgi apparatus. The targeting and localization signals are normally contained in the cytoplasmic and/ or transmembrane domains and in some cases are contained in some lumenal amino acids. For example, fifty two amino acids that make up the transmembrane domain, nine cytoplasmic amino 15 acids and twenty-six lumenal amino acids of a-2,6-sialyltransferase are required to target GFP to the trans Golgi network (Boevink et af. PlantJ 1998 Aug;15(3):441 7). [0292] Thus, a library of sequences encoding cellular targeting signal peptides comprising of either just the cytoplasmic and transmembrane domains or the 20 cytoplasmic, transmembrane and lumenal domains of endoplasmic reticulum and Golgi specific proteins is generated, as described in Example 11. The targeting peptide sequences maybe chosen from ER and Golgi-resident plant, yeast or animal proteins. A glycosylation related protein, e.g., an enzyme (or catalytic domain thereof) such as a glycosylase or integral membrane enzyme can be fused 25 in-frame to the library of targeting peptide sequences and introduced into plants (Fig. 13). Plant targeting peptide sequences may be most efficient in localizing the chimeric enzymes to the ER and Golgi, although targeting peptide sequences from fungi and mammals may also be effective. For example, the N-terminal 77 amino acids from tobacco N-acetylglucosaminyl Transferase I have been shown to 30 correctly target a reporter protein to the Golgi (Essl D. et al., FEBS Lett 1999 Jun 18;453(1-2):169-73). In one embodiment, one or more N-terminal fragments comprising these 77 amino acids (or subsets of these amino acids) is fused to one 115 or more fragments comprising a catalytic domain of GIcNAc transferase IV. At least one resulting fusion protein correctly localizes a functional GlcNAc transferase IV to the Golgi apparatus in a plant cell, as evidenced by monitoring the glycosylation state of a reporter glycoprotein resident or introduced into the 5 plant host cell using techniques described herein. [0293] Another plant enzyme shown to localize to the Golgi is Arabidopsis GIcNAc transferase II (Strasser R et al., GlycoconjJ 1999 Dec;16(12):787-91). Thus, in another embodiment, one or more different fragments of the arabidopsis GlcNAc transferase II targeting peptide are fused to a GlcNAc transferase IV 10 catalytic domain and fusion constructs produced and tested as described above. The plant specific 0 1,2-xylosyltransferase from Arabidopsis thaliana is another protein that localizes to the Golgi and its localization and retention in the Golgi is dependent on its cytoplasmic and transmembrane sequences (Dirnberger et al., Plant Mol Biol 2002 Sep;50(2):273-81). Thus, in another embodiment, one or 15 more fragments comprising the cytoplasmic and transmembrane sequences of pI,2-xylosyltransferase are fused to one or more fragments comprising a GIcNAc transferase IV catalytic domain and resulting fusion constructs are transformed into plant cells and tested for their ability to produce a human-like N-glycan and to otherwise modulate glycosylation in the plant host cell. 20 [0294] Because GlcNAc transferase IV or Galactosyltransferase from one organism may function more efficiently in a specific plant host than one from another organism, fragments comprising GlcNAc transferase IVs (or catalytic domains) from various eukaryotic organisms are fused in-frame to the library of endoplasmic reticulum (ER) and Golgi targeting peptide sequences and are then 25 introduced into plants. The use of a library of nucleic acids encoding enzyme domains isolated or derived from different species increases the chances of efficient glycosylation - in addition to correct localization and glycosylation by GlcNAc transferase IV. [0295] The methods and combinatorial nucleic acid libraries of the invention 30 may be used to introduce and localize, sequentially or en masse, multiple enzymes required to glycosylate proteins in a plant cell with human-like N-glycans. As different plant species may require different growth conditions, protocols for 116 transformation may vary depending on the species being transformed (Potrykus, "Gene transfer methods for plants and cell cultures." Ciba Found Symp 1990;154:198-208; discussion 208-12). The commonly used methods for generating transgenic plants include Agrobacterium mediated transformation, 5 particle bombardment (Sanford, J.C. et al, Biolistic plant transformation. Physiol. Plant. 1990, 79: 206-209) and electroporation. Agrobacterium method [0296] The catalytic domains of GlcNAc transferase IVs are fused in-frame to 10 multiple different targeting peptide sequences known to target proteins to the ER and Golgi in plants. Each of these fusion constructs is introduced under the control of the ubiquitously expressed promoters like the 35S CaMV, ubiquitin or actin promoters, tissue specific promoters or inducible promoters. A plant specific terminator region is also used. This cassette (Promoter.:targeting peptide-GlcNAc 15 transferase IV::terminator) is cloned into a vector suitable for Agrobacterium mediated transformation (Fig. 13). The vector also contains a selectable marker that allows one to select for transformed plants. The common selectable markers used include those resulting in kanamycin, hygromycin and basta resistance. The construct is introduced into Agrobacterium via well-established transformation 20 methods, which are available in the art. An Agrobacterium library of Golgi targeted GlcNAc transferase IVs is thereby generated. [0297] Embryonic and meristematic tissue may be transformed and can regenerate transgenic plants. To transform tissue, tissue explants (these could be plumules and radicals from germinated seeds) are first soaked and coated with an 25 Agrobacterium innoculum. They are then cultured on plates containing the innoculum to form an undifferentiated mass of cells termed the callus. Transformed plant cells are selected for by adding to the medium the relevant kanamycin, hygromycin or basta (depending on the selectable marker used on the construct). The transformed plant cells can either be grown in culture and remain 30 undifferentiated or they are treated with shoot regenerating and shoot elongation medium. Explants that differentiate are transferred onto rooting medium to generate transgenic plants. Some plants like Arabidopsis can be transformed by 117 dipping flowers into an Agrobacterium solution. Seeds from the transformed plants are germinated on plates containing the relevant herbicide or antibiotic selection. Transgenic plants are those that grow on the selection media. The transgenic plants are then screened for those with properly glycosylated proteins 5 (i.e., those which have complex, human-like N-glycans) by isolating glycoproteins from plant extracts and analyzing glycoprotein patterns as described elsewhere herein, e.g., by using a specific antibody or lectin. Although the Agrobacterium method is economical and simple, it is limited to certain species of plants. Accordingly, plants that cannot be transformed using Agrobacterium can be 10 transformed by ballistics or electroporation. Particle bombardment method and electroporation [02981 Compared to Agrobacterium mediated transformation, these methods have a greater tendency to insert multiple copies of the transgene into the genome. 15 This could result in gene silencing and cosuppression. However, unlike Agrobacterium mediated transformation, these methods are not species limited and are therefore useful when an Agrobacterium method cannot be employed to generate transgenic plants. In the particle bombardment method, cultured plant cells are bombarded with very small tungsten or gold particle that have been coated 20 with DNA (Promoter::targeting peptide-GLcNAc transferase IV terminator::selectable marker) (Fig. 13) (rb and lb not required) while in the electroporation method, plant cells in a DNA (Promoter::targeting peptide-GIcNAc transferase W-terminator::selectable marker) solution are treated with an electric pulse that perforates the cell, allowing it to take up DNA. The cells are then 25 cultured and allowed to recover. Stable transformants are selected for by culturing and regenerating plants on appropriate selection medium. Engineering Soybeans to express GIcNAc transferase IV using a soybean cotyledonary node Agrobacterium mediated transformation system 30 [0299] An Agrobacterium library of Golgi-targeted GlcNAc transferase IV is generated as described above. Soybean explants are transformed with the library using a protocol described by Hinchee et al (Bio/Technology 1988. 6:915). A 118 reporter protein is expressed with a His tag, purified and then analyzed. Transgenic plants are assayed for proteins with the x-1,6 inannose and the a-1,3 mannose residues using, e.g., mass spectroscopy. 5 Engineering pea to express GIcNAc transferase IV using particle bombardment [0300] A GlcNAc transferase 1V plasmid library is coated onto tungsten or gold particles and used as microprojectiles to bombard calli derived from pea embryonic tissue as described (Molnar et al., Symposium on Recent Advances in Plant Biotechnology, September 4-11, 1999, Stara Lesna, Slovak Republic). A reporter 10 protein is expressed with a His tag, purified and then analyzed. Transgenic plants are assayed for proteins with the a-1,6 mannose and the cc-1,3 mannose residues using, e.g., MALDI. Engineering plants to express GlcNAc transferase I 15 [0301] GlcNAc transferase I is involved in the addition of GlcNAc to the terminal a-1,3 mannose residue to form Man 5 GIcNAc 2 , an essential step in the maturation of complex N-glycans. Although GIcNAc transferase I has been isolated from plants and appears to have the same function as its mammalian homolog, it may not be the most efficient enzyme for glycosylation of mammalian 20 or exogenous proteins and may not be found in every plant species. As the addition of GIcNAc to the terminal a-1,3 mannose residue is a controlling step in the mammalian glycosylation pathway, it is advantageous to have transgenic plants that can carry out this step efficiently. To create transgenic plants that express GlcNAc transferase I that can function efficiently to promote the formation of 25 complex N-glycans, a library of GlcNAc transferase I isolated or derived from various organisms is fused in-frame to multiple plant Golgi targeting peptide sequences according to the methods described herein. The combinatorial library thus created is introduced into a plant cell or organism as described above for GlcNAc transferase IV. 30 Engineering maize to express GIcNAc transferase I using particle bombardment 119 [0302] Transgenic maize can be obtained using a protocol similar to the one used to generate peas that express GlcNAc transferase IV. Here the GlcNAc transferase I plasmid library is coated onto tungsten or gold particles and used to bombard call derived from maize embryonic tissue, e.g., using a protocol specific for the 5 generation of transgenic maize (Gordon-Kamm WJ et al., Plant Cell 1990 Jul;2(7):603-618)). Transgenic plants are assayed for proteins having GlcNAc on the terminal a-1,3 mannose residue, e.g., using specific antibodies or by assaying reduced binding of the N-glycans to certain lectins or by using MALDI-TOF. [0303] Other useful references for using plant host cells according to the 10 invention include: Christou P. Plant Mol Biol 1997 Sep;35(1-2):197-203; Chowrira GM et al. MolBiotechnol 1995 Feb;3(1):17-23; Dirnberger et al., Plant Mol Biol2002 Sep;50(2):273-81; Frame BR et al. Plant Physiol 2002 May,129(1):13-22; Gomord V et al. Biochimie 1999 Jun;81(6):607-18; Laursen CM et al. Plant Mol Biol 1994 Jan;24(l):51-61; Orci L et al. J Cell Biol 2000 Sep 15 18; 150(6):1263-70; Newell CA. Mol Biotechnol 2000 Sep; 16(1):53-65; Pawlowski Wpet al. MolBiotechnol 1996 Aug;6(l):17-30; Schroeder HE et al. Plant Physiol 1993 Mat;101(3):751-757; Sorokin, AP et al. Plant Sci. 2000 Jul 28;156(2):227 233; Strasser R et al. GlycoconjJ1999 Dec;16(12):787-91; and Tomes DT et al. Plant Mol Biol 1990 Feb;14(2):261-8. 20 Engineering plant cells to produce 1.

4 -malactosyltransferases [03041 P1, 4 -galactosyltransferase is an important human glycosyltransferase that is absent in plants. Lerouge P etal. Plant Mol Biol 1998 Sep;38(1-2):31-48. In mammals, P 1, 4 -galactosyltransferase is localized in the Golgi and is responsible 25 for the transfer of galactose residues to the terminal N-acetylglucosamine of the core Man 3 GlcNAc 2 of complex N-glycans. In plants, the Man 3 GIcNAc 2 core contains 0 1,2-xylose and al,3-fucose residues and lacks the pl,4-galactose. The xylose and fucose modifications are implicated in allergies and act as antigenic epitopes and are therefore not desirable modifications of therapeutic proteins. 30 [0305] The galactose modifications carried out by Pl, 4 -galactosyltransferase can be important for the proper functioning of the therapeutic proteins. In mammals, 1, 4 -galactosyltransferase acts after N-acetylglucosaminyltransferase I and N 120 acetylglucosaminyltransferase II and has been shown to initiate branching of the complex N-glycan. Lerouge P et al. Plant Mol Biol 1998 Sep;38(1-2):31-48. Palacpac N et al. Proc NatlAcad Sci USA 1999 Apr 13;96(8):4692-7. In tobacco cells, expression of human p 1, 4 -galactosyltransferase has been shown to result in 5 galactosylated N-glycans with reduced fucose and xylose modifications. Bakker H et al. Proc Natl Acad Sci U SA 2001 Feb 27;98(5):2899-904 Fujiyama Ket al. Biochem Biophys Res Commun 2001 Nov 30;289(2):553-7. Palacpac N et al. Proc Natl Acad Sci U S A 1999 Apr 13;96(8):4692-7. In these studies, a 1.2 kb fragment of human P 1, 4 -galactosyltransferase was cloned downstream of the cauliflower 10 mosaic virus promoter (35SCaMV), introduced into the binary vector pGA482, and finally into tobacco cells. Palacpac N et al. Proc Natl Acad Sci USA 1999 Apr 13;96(8):4692-7. [0306] Tobacco cells were transformed using the agrobacterium method described by Rempel et al. (Rempel, H.C.et al. 1995. Transgenic Res. 4(3):199 15 207.) Transformation of tobacco cells has also been described (An, G 1985. Plant Physiol. 79:568-570). Expression of P 1, 4 -galactosyltransferase under the 35SCaMV resulted in ubiquitous expression of the gene in tobacco cells. Tobacco cells expressing human 0 1

,

4 -galactosyltransferase showed the presence of galactosylated N-glycans. (Palacpac N et.aLProc NatlAcadSci USA 1999 Apr 20 13;96(8):4692-7). Bakker et al. showed that crossing tobacco plants expressing human $1, 4 -galactosyltransferase with plants expressing the heavy and light chain of a mouse antibody resulted in plants in which the antibody showed 30% galactosylation (Bakker H et al. Proc Natl Acad Sci USA 2001 Feb 27;98(5):2899-904). 25 [0307] A combinatorial DNA library can be constructed to obtain a p1,4 galactosyltransferase line for the addition of galactose residues. The combinatorial DNA library can effectively produce lines which are more efficient in the addition of galactose residues. Once such a line is made it can be easily crossed to lines expressing other glycosylation enzymes and to those expressing therapeutic 30 proteins to produce therapeutic proteins with human-like glycosylation. The final line can then be grown as plants and harvested to extract proteins or can be cultured as plant cells in suspension cultures to produce proteins in bioreactors. By 121 expressing the therapeutic proteins using the library of signal peptides, it is possible to retain the therapeutic protein within the cells or have them secreted into the medium. Tobacco cells expressing p1, 4 -galactosyltransferse secrete galactosylated N-glycans (Ryo Misaki et al. Glycobiology 2002 Dec 17;10: 1093). 5 While horseradish peroxidase isozyme C expressed in tobacco plants expressing P 1,4-galactosyltransferase contained xylose and fucose modifications, no xylose or fucose modification could be detected in horseradish peroxidase isozyme C expressed in tobacco cells expressing P 1, 4 -galactosyltransferase (GT6 cells). (Fujiyama Ket al. Biochem Biophys Res Commun 2001 Nov 30;289(2):553-7). 10 This indicates that it may be advantageous to express therapeutic proteins in cell lines instead of whole plants. Engineering plants to produce sialyltransferase [0308] In mammals, sialyltransferase is a trans golgi enzyme that adds terminal 15 sialic acid residues to glycosylated polypeptides. Thus far, terminal sialic acid residues-have not been detected in plants (Wee E et al. Plani Cell 1998 Oct;10(10):1759-68). Wee et al. expressed the rat a-2,6-sialyltransferase in transgenic arabidopsis and showed that the enzyme properly localized to the golgi and was functional. Wee et al. demonstrated that membranes isolated from 20 transgenic arabidopsis, when incubated with CMP-1-sialic acid and asialofetuin acceptor, resulted in the addition of sialic acid residues while membrane isolated from wild-type arabidopsis did not. While expressing the rat a-2,6 sialyltransferase in arabidopsis resulted in a functional enzyme that was able to incorporate sialic acid residues, fusing the mammalian enzymes oc-2,3 25 sialyltransferase and a- 2 ,6-sialyltransferase to a variety of transit peptides using the library approach of the present invention described earlier can result in more efficient sialylation in other plant species. Wee E et al had to isolate membranes and incubate them with CMP- 3 H-sialic acid and asialofetuin acceptor since arabidopsis does not have CMP-sialic acid or its transporter. In order to overcome 30 this additional step and obtain sialic acid addition in the plant, CMP-sialic acid biosynthetic pathway and the CMP-sialic acid transporter can be co-expressed in transgenic plants expressing cx-2,3-sialyltransferase and a- 2 ,6-sialyltransferuse. As 122 an alternative the CMP-sialic acid transporter can be co-expressed a-2,3 sialyltransferase and a-2,6-sialyltransferase in plant cells grown in suspension culture, and CMP-sialic acid or other precursors of CMP-sialic acid supplied in the medium. 5 Expressing a- 2

-

3 -sialvltransferase and a- 2

,

6 -sialyltransferase in lemna (0309] As described in the U.S. Pat. No. 6,040,498, lemna (duckweed) can be transformed using both agrobacterium and ballistic methods. Using protocols described in the patent, lemna will be transformed with a library of golgi targeted a- 2 ,3-sialyltransferase and/or a- 2 ,6-sialyltransferase and a library of mammalian 10 CMP-sialic acid transporters. Transgenic plants can be assayed for proteins with terminal sialic acid residues. Expressing a- 2

.

3 -sialvltransferase and a- 2

.

6 -sialltransferase in tobacco cells [0310] Alpha-2,3-sialyltransferase and/or a- 2

,

6 -sialyltransferase and/or a library of mammalian CMP-sialic acid transporters can also be introduced into tobacco 15 cells grown in suspension culture as described for 0 l, 4 -galactosyltransferases. CMP.sialic acid can be added to the medium. Both the cells and the culture medium (secreted proteins) can be assayed for proteins with terminal sialic acid residues residues. 20 Example 18: Engineering insect cells to produce glycosyltransferases [03111 Insect cells provide another mechanism for producing glycoproteins but the resulting glycoproteins are not complex human-like glycoproteins. Marz et al. 1995 Glycoproteins, 29:543-563; Jarvis 1997 The Baculoviruses 389-431. It is another feature of the present invention to provide enzymes in insect cells, which 25 are targeted to the organelles in the secretory pathway. In a preferred embodiment, enzymes such as glycosyltransferases, galactosyltransferases and sialyltransferases are targeted to the ER, Golgi or the trans Golgi network in lepidopteran insect cells (Sf9). Expression of mammalian #1, 4 -galactosyltransferase has been shown in Wf cells. Hollister et al. Glycobiology. 1998 8(5):473-480. These enzymes are 30 targeted by means of a chimeric protein comprising a cellular targeting signal peptide not normally associated with the enzyme. The chimeric proteins are made by constructing a nucleic acid library comprising targeting sequences as described 123 herein and the glycosylation enzymes. Baculovirus expression in insect cells is commonly used for stable transformation for adding mammalian glycosyltransferases in insect cells. Hollister et al. Glycobiology. 2001 11(1):1-9. 5 Table 11: DNA and Protein Sequence Resources 1. European Bioinformatics Institute (EBI) is a centre for research and services in bioinformatics: http://www.ebi.ac.uk/ 2. Swissprot database: http://www.expasy.ch/spr 3. List of known glycosyltransferases and their origin. 10 4. human cDNA, Kumar et al (1990) Proc. Natl. Acad. Sci. USA 87:9948 9952 5. human gene, Hull et al (1991) Biochem. Biophys. Res. Commun. 176:608 615 6. mouse cDNA, Kumar et al (1992) Glycobiology 2:383-393 15 7. mouse gene, Pownall et al (1992) Genomics 12:699-704 8. murine gene (5' flanking, non-coding), Yang et al (1994) Glycobiology 5:703-712 9. rabbit cDNA, Sarkar et al (1991) Proc. Natl. Acad. Sci. USA 88:234-238 10. rat cDNA, Fukada et al (1994) Biosci.Biotechnol.Biochem. 58:200-201 20 1.2 (GnTII) EC 2.4.1.143 11. human gene, Tan et al (1995) Eur. J. Biochem. 231:317-328 12. rat cDNA, D'Agostaro et al (1995) J. Biol. Chem. 270:15211-15221 13. P1.4 (GnTII) EC 2.4.1.144 14. human cDNA, Ihara et al (1993) J. Biochem.113:692-698 25 15. murine gene, Bhaumik et al (1995) Gene 164:295-300 16. rat cDNA, Nishikawa et al (1992) J. Biol. Chem. 267:18199-18204 §01.4 (GnTIV ) EC 2.4.1.145 17. human cDNA, Yoshida et al (1998) Glycoconjugate Journal 15:1115-1123 18. bovine cDNA, Minowa et al., European Patent EP 0 905 232 30 P1,6 (GnT V) EC 2.4.1.155 19. human cDNA, Saito et al (1994) Biochem. Biophys. Res. Commun. 198:318-327 20. rat cDNA, Shoreibah et al (1993) J. Biol. Chem. 268:15381-15385 p1,4 Galactosyltransferase, EC 2.4.1.90 (LacNAc synthetase) EC 2.4.1.22 35 (lactose synthetase) 21. bovine cDNA, D'Agostaro et al (1989) Eur. J. Biochem. 183:211-217 22. bovine cDNA (partial), Narimatsu et al (1986) Proc. Natl. Acad. Sci. USA 83:4720-4724 23. bovine cDNA (partial), Masibay & Qasba (1989) Proc. Natl. Acad. Sci. 40 USA 86:5733-5377 24. bovine cDNA (5' end), Russo et al (1990) 1. Biol. Chem. 265:3324 25. chicken cDNA (partial), Ghosh et al (1992) Biochem. Biophys. Res. Commun. 1215-1222 26. human cDNA, Masri et al (1988) Biochem. Biophys. Res. Commun. 45 157:657-663 124 27. human cDNA, (HeLa cells) Watzele & Berger (1990) Nucl. Acids Res. 18:7174 28. human cDNA, (partial) Uejima et al (1992) Cancer Res. 52:6158-6163 29. human cDNA, (carcinoma) Appert et al (1986) Biochem. Biophys. Res. 5 Commun. 139:163-168 30. human gene, Mengle-Gaw et al (1991) Biochem. Biophys. Res. Commun. 176:1269-1276 31. murine cDNA, Nakazawa et al (1988) J. Biochem. 104:165-168 32. murine cDNA, Shaper et al (1988) J. Biol. Chem. 263:10420-10428 10 33. murine cDNA (novel), Uehara & Muramatsu unpublished 34. murine gene, Hollis et al (1989) Biochem. Biophys. Res. Commun. 162:1069-1075 35. rat protein (partial), Bendiak et al (1993) Eur. J. Biochem. 216:405-417 2.3-Sialyltransferase, (ST3Gal II (N-linked) (Gal-1.3/4-GlcNAc) BC 2.4.99.6 15 36. human cDNA, Kitagawa & Paulson (1993) Biochem. Biophys. Res. Commun. 194:375-382 37. rat cDNA, Wen et al (1992) 3. Biol. Chem. 267:21011-21019 2,6-Sialyltransferase, (ST6Gal 1) EC 2.4.99.1 38. chicken, Kurosawa et al (1994) Eur. J. Biochem 219:375-381 20 39. human cDNA (partial), Lance et al (1989) Biochem. Biophys. Res. Commun. 164:225- 232 40. human cDNA, Grnmdmann et al (1990) Nucl. Acids Res. 18:667 41. human cDNA, Zettimeisi et al (1992) Patent EP0475354-A/3 42. human cDNA, Stamenkovic et al (1990) J. Exp. Med. 172:641-643 (CD75) 25 43. human cDNA, Bast et al (1992) J. Cell Biol. 116:423-435 44. human gene (partial), Wang et al (1993) 3. Biol. Chem. 268:4355-4361 45. human gene (5' flank), Aasheim et al (1993) Eur. J. Biochem. 213:467-475 46. human gene (promoter), Aas-Eng et al (1995) Biochim. Biophys. Acta 1261:166-169 30 47. mouse cDNA, Hamamoto et al (1993) Bioorg. Med. Chem. 1:141-145 48. rat cDNA, Weinstein et al (1987) J. Biol. Chem. 262:17735-17743 49. rat cDNA (transcript fragments), Wang et al (1991) Glycobiology 1:25-31, Wang et al (1990) J. Biol. Chem. 265:17849-17853 50. rat cDNA (5' end), O'Hanlon et al (1989) J. Biol. Chem. 264:17389-17394; 35 Wang et al (1991) Glycobiology 1:25-31 51. rat gene (promoter), Svensson et al (1990) J. Biol. Chem. 265:20863-20688 52. rat mRNA (fragments), Wen et al (1992) J. Biol. Chem. 267:2512-2518 [0312] Additional methods and reagents which can be used in the methods for 40 modifying the glycosylation are described in the literature, such as U.S. Patent No. 5,955,422, U.S. Patent No. 4,775,622, U.S. Patent No. 6,017,743, U.S. Patent No. 4,925,796, U.S. Patent No. 5,766,910, U.S. Patent No. 5,834,251, U.S. Patent No. 5,910,570, U.S. Patent No. 5,849,904, U.S. Patent No. 5,955,347, U.S. Patent No. 5,962,294, U.S. Patent No. 5,135,854, U.S. Patent No. 4,935,349, U.S. Patent No. 45 5,707,828, and U.S. Patent No. 5,047,335. Appropriate yeast expression systems 125 can be obtained from sources such as the American Type Culture Collection, Rockville, MD. Vectors are commercially available from a variety of sources. 103131 SEQUENCE LISTINGS 5 SEQ ID NO:1-6 can be found in U.S. Pat. Application No. 09/892,591 SEQ ID NO:7 Primer: regions of high homology within 1,6 mannosyltransferases 5 '-atggcgaaggcagatggcagt-3' 10 SEQ ID NO:8 Primer: regions of high homology within 1,6 mannosyltransferases 5'-ttagtccttccaacttcctto-3' 15 SEQ ID NO:9 internal primer: 5'-actgccatctgccttcgccat-3' SEQ ID NO:10 internal primer: 5'-GTAATACGACTCACTATAGGGC-3' T7 20 SEQ ID NO:11 Intemalprimer: 5'-AATTAACCCTCACTAAAGGG-3' T3 SEQ ID NO:12 25 Primer: atgccogtgg ggggcctgtt gccgctcttc agtagc SEQ ID NO:13 Primer: tcatttctct ttgccatcaa tttccttctt ctgttcacgg 30 SEQ ID NO:14 Primer: ggcgcgccga ctcctccaag ctgctcagcg gggtcctgtt ccac SEQ ID NO:15 Primer: ccttaattaa tcatttctct ttgccatcaa tttccttctt ctgttcacgg 35 SEQ ID NO:16 Primer ggcgagctccctacccgccaaggctga gatcatttgt ccagcttcaga SEQ ID NO:17 40 Primer gcccacgteg acgaatccat ttaaacatcg attggagagg ctgacaccgc tacta SEQ ID NO:18 Primer: cgg9LOO2tagt-att tcatatgtg cgagtgtaca actcttccca catgg 45 SEQ ID NO:19 Primer: ggacgc ac cctacc c cc tg aggaatttct cggatgacto ttttc 126 SEQ D NO:20 Primer: cgjzatrcct cgagagatct tttgtaga aatgtcttgg tgcct 5 SEQIDNO:21 Primer: ggacatca 2ccatc~cca~tatagttgt tcaattgatt gaaataggga caa SEQ ID NO:22 Primer: ccttctc tatac ccacgtc cgacggcggc ccacgggtcc ca 10 SEQ ID NO:23 Primer: ggacatgcat ggggatcect taa aggggg cagcttgcaa attaaagcct tcgagcgtcc c SEQ ID NO:24 15 Primer gaaca~acZL2gRcatt Zc ccaaaa ccttttcc tattcaaaca caaggcattg c SEQ ID NO:25 Primer: ctccatact acgaagat tatcttctac ggtgcctgga ctc 20 SEQ ID NO:26 Primer tggaaggtt aaaaagc It agagtaaaa tagatatagc gagattagag aatg SEQ ID NQ:27 Primier. aagaattegg ctggaaggcc ttgtaccttg atgtagttoc cgttttcatc 25 SEQ ID NO:28 Primer gcccaaggc~gccttaaggg atctcctgat gactgactca ctgataataa aaatacgg SEQ ID NO:29 30 Primer gggcgcgta ttatacta giggatetat cgaatctaaa tgtaagttaa aatctctaa SEQ ID NO:30 Primer: gacciactgg agafttaaat gaatcg cgc~ccttaat 35 SEQ ]D NO:31 Primer Magpcgcc egaatteatt taaatet~ca ggc SEQ ID NO:32 Primer 5'-tggcaggcgcgccagtcagcgctctcg-3 40 SEQ ID NO:33 Primer. 5'-aggttaatta agtgetaattccagctagg-3' SEQ ID NO:34 45 primer for Kiactis OCHI gene: ccagaagaat tcaattytgy cartgg SEQ ID NO:35 primer for Kiactis OCHJ gene: cagtgaaaat acctggnccn gtcca 127 SEQ ID NO:36 primer for Klactis MNNJ gene: tgccatcttt taggtccagg cccgttc 5 SEQ ID NO:37 primer for Klactis MANJ gene: gatcccacga cgcatcgtat ttctttc SEQ ID NO:38 DNA sequence of the 302 bp segment of the putative KIOCH1 gene: 10 gccttcagtgaaaatacctggcccggtecagttcataatateggtaccatetgtatttttggeggttttettttgttgatgttt gtaatttttgttgaacttctttttatccctcatgttgacattataatcatctgcaatgtettttaatacttcago atcatctaaaggaatgctgcttttaacatttgccacgctctccaatgttgttgcggtgatatttgtgatcaattegcgcaataa tggatggccagattttgattgtattgtccacgaaaattgaattctctggaagggc 15 SEQ ID NO:39 Translation of putative KiOCH1 gene (excluding primers): TIQSKSGHPLLRELITNITATTLESVANVKSSIPLDDAEVLKDIADDYNVNM RDKKKFNKNYKHQQKKTAKNTDGTDIMN 20 SEQ ID NO:40 DNA sequence of the 405 bp segment of the putative KLMNN1 gene: cccagcgtgccatacgtatttgccgccgtttgaaatatcaatattcatgatggttgtaaggegttttttatcattogogat 25 ataatatgccatettttaggto-caggcccgttctc-ttagetatetttggtgtetgtgetaccgtgatatggtacct attctttcagtctaatctgaagatggcagatttgnannnggtagcaacttc~aaggtatctttcacaagaaccgtcgttat cagaacttatgtcaaatgtgaagatcaagcetattgaagaaac-cccggtttcgccattggagttgattccagatatogaaa tatcgactagaaagaaatacgatgcgtcgtgggatetgttgttrcgtggtagaaaatataaatogttcaacgattatgat 30 SEQ ID NO:41 DNA sequence of the Klacti OCH1 gene: atggggttaccaaagatttcaagaagaacgaggtac-attattgtcattgtgotgatactgtacttattgttttctgtgcaatg 35 gaatactgcgaaagtgaatcaccatttctataacagcattggcacggtgettcccagtacagotcgogtggatcacttga acttgaaaaae-ttggacttagc-aggtacgagcaataacggtgatcatttgatggatctacgagttcaattggetagtcaat tccctacgattctcgagtacccatccenannaggtatggcagacctggaagattgatoccagttcaaagtcacaggtt tctccatttcaaaatgccagaatgattggaaac atttcagtgcatccgaggaaccgccatatcaataccaattaatcaca gatgatcaagtaccacttetagagcagotatatggtggggteccacaagtgataaaggettttgaatecttgccatt 40 ccaattaacgcttcgtctatctagaggtgaaattaatggatacgttcccatta caactaagtgaetaactgtttatgaeagtacggatagaagogattgggca cacantacacgcasaacttgagaegtgcaatgtaaagcagattcttagagatgetgaagtattaa 45 aagacattgcagatgattataatgtcaacatgagggi tnannngaagttcascannnatuacaaacatcaacaaaagaa aaccgcannaatacagatggtaccgatattatgaactggactggto~caggtttttcagatgttatttccagtatetta ataacgttatc~cagaagaatgatgacattatnittegataatcttaatgttatcaacaaacatggatccaaacatgata caactatgagattctataaagacattgttanantamcaaacgcaaccctcatgttetggggpttcttttcattgatga 128 5 SEQ ID NO:42 Translation of putative Kiactis CI gene: HLNUKNLDLAGTSNNGDHDLRVQLASQFPYDSRVPVQWKD 10 PSSKSQVSSISKCQNDWFSASEEPPYQYQLTDDQILLEQLYCG<-PQ RDKKFNKNYKHQQKKTAKNTDGThI4rGPGIISDIQYTNNVQK 15 NDIINNNqHSHT~~~VNQDPLWFSM SEQ ID NO:43 20 DNA sequence of the Kilactis MNNJ gene: atagtgagc ttacgtccaaattc-tttagccgcctcctgtttt gttttcactaag~ ttcttc-gcattagtgaatg mga 25 tagttttccaacgctttaactttcattagtagcttgaaae 25 cgtcrctgafgtc-gtcu~gatgagatcaggctgacgttc gtgaaattmctcagtagttctcaagggttttegataaattac agatgcataatgagtattgtfaeagaaagcagaatagttaa 30 pggtcatgtgeaaggttctttcagpeetagtgctgtcg ugtagcagtattt~aatc~gtaacataatgtacttaacac 30 aaggagcagctgtagatatgacggtc agcmnatct ctaggttgraggattcgatcacagpt actgagtcctacct ttcagaaattacaggtccattacgtaagtggaeU taggat 35 ttatc~cgggatfcttgga-tcpagtatcaacctgag ctgagtttccgtcg mtgattctgtcgoagcaggacagac 35 gctfgtctgggtagaagttcttcatacaacacatataaacg gaattacggcgtatacatcctrgtggtattttgtggagggc cagtgactat ggattgggttaatatgtta~ttacattgaat 40 taaaaaagcacetcfattgtgtatcatcaggtcattcgacc ctcf~aagctgmtga~gaaaaggtctgtgacrtgacaatct 40 aagatft~ tarttacagtta-tgaacagaggagtac gg aaagtaaagtctgatgtacaacatgaatgtatgctacanor actagptgtcgam~attgacaagcg~ cgtgggcaggt 45 gtgtttgttcgcaccatpettttgtcatmfgttcccaaa 129 gotgaaaccgtatacttataacacggtggtaactaaaggtgatttgatecgttteggagacgaggaaatcgaaagtatet ccaagattaataagatctggaatgatgctattattcagacggagcttaa 5 SEQ ID NO:44 Translation of putative K lactis MNN1 gene: MMVVRRFLSASRYNMPSFRSRPVLLAIFGVCATVlWYLFFFQSNLKMADL KKVATSRYLSQEPSLSELMSNVKIKPIEETPVSPLELIPDIEISTRKKYDASW 10 DLLFRGRKYKSFNDYDLHTKCEFYFQNLYNLNEDWTNNIRTFTFINDVD TSTKIDALKDSDGVQLVDEKAIRLYKRTHNVALATERLRLYDKCFVNSPG SNPLKMDHLFRSNKKSKTTALDDEVTGNRDTFTKTKKTSFLSDMDTSSFQ KYDQWDFEMFPMIPYFEHFTNVMPIFTGSNGGEPLPQGKFPVLDPKS GELLRVETFRYDKSKSLWKNWNDMSSASGKRGIILAAGDGQVDQCIRLIA 15 TLRAQGNALPIQIH QLNESVKLLSEAAKSTEFSSGRAQSLWLVNVGP TLESSIMNFGRFKNKWLSVIFNTFEEFFITDAISYMADYFNFKYKST GTLFFKDRSLAIGTBQKCGPLFETLEPRILEMYYFNTLPMINGDYVBQQCM GMLTPEEKVYKRFFEVGHQHNLESGLLAINKNEHIMGLVTATVLNIAPKV GGCGWGDKEFFWLLLVAGQRYSIYDIDASAIGVPQQKQSIANGDEFDEY 20 RICSLQVAHTSYDGHLLWINGGSQYCKKPBTFBGDWTNIKELRESYSDDKE KALKAYSDTVKVEAAIVPDSRSNGWGRD)DQRCKGYFWCGKFTSKLKPYT YNTVVTKGDIRFGDEEIESISKIKIWNDADIPDGA Definitions [0313a] The term "comprise" and variants of the term such as "comprises" or 25 "comprising" are used herein to denote the inclusion of a stated integer or stated integers but not to exclude any other integer or any other integers, unless in the context or usage an exclusive interpretation of the term is required. [0313b] Any reference to publications cited in this specification is not an admission that the disclosures constitute common general knowledge in Australia. 130 [03141 REFERENCES Aebi, M., J. Gassenhuber, et al. (1996). "Cloning and characterization of the ALG3 gene of Saccharomyces cerevisiae." Glycobiology 6(4): 439-444. Altmann, F., E. Staudacher, et al. (1999). "Insect cells as hosts for the expression 5 of recombinant glycoproteins." Glycoconiugate Journal 16(2): 109-123. Andersen, D. C. and C. F. Goochee (1994). "The effect of cell-culture conditions on the oligosaccharide structures of secreted glycoproteins." Current Opinion in Biotechnology 5: 546-549. Bardor, M., L. Faye, et al. (1999). "Analysis of the N-glycosylation of recombinant 10 glycoproteins produced in transgenic plants." Trends in Plant Science 4(9): 376-380. Bretthauer, R. K. and F. J. Castellino (1999). "Glycosylation of Pichia pastoris derived proteins." Biotechnology and Applied Biochemistry 30: 193-200. Burda, P. and M. Aebi (1999). "The dolichol pathway of N-linked glycosylation." 15 Biochimica Et Biophysica Acta-Gensritiibiects 1426(2): 239-257. Chiba, Y., M. Suzuki, et al. (1998). "Production of human compatible high mannose-type (Man(5)GIcNAc(2)) sugar chains in Saccharomyces cerevisiae." Journal of Biological Chemistry 273(41): 26298-26304. Cole, E. S., E. Higgins, et al. (1994). "Glycosylation Patterns of Human Proteins 20 Expressed in Transgenic Goat Milk" Journal of Cellular Biochemistry. 265-265. Davies et al. Biotechnol Bioeng. 2001 Aug 20;74(4):288-294. (Expression of GnTHI in a Recombinant Anti-CD20 CHO Production Cell Line: Expression of Antibodies with Altered Glycoforms Leads to an Increase in 25 ADCC Through Higher Affinity for FcgRIII). Dente, L., U. Ruther, et al. (1988). "Expression of Human Alpha-i-Acid Glycoprotein Genes in Cultured-Cells and in Transgenic Mice." Genes & Development 2(2): 259-266. Huffaker, T. C. and P. W. Robbins (1983). "Yeast Mutants Deficient in Protein 30 Glycosylation." Proceedings of the National Academy of Sciences of the United States of America-Biological Sciences 80(24): 7466-7470. Jarvis, D. L., Z. S. Kawar, et al. (1998). "Engineering N-glycosylation pathways in the baculovirus-insect cell system." Current Opinion in Biotechnology 9(5): 528-533. 35 Kimura, T., N. Kitamoto, et al. (1997). "A novel yeast gene, RHKl, is involved in the synthesis of the cell wall receptor for the HM-1 killer toxin that inhibits beta-1,3-glucan synthesis." Molecular & General Genetics 254(2): 139 147. Kimura, T., T. Komiyama, et al. (1999). "N-glycosylation is involved in the 40 sensitivity of Saccharomyces cerevisiae to HM-1 killer toxin secreted from Hansenula mrakii IFO 0895." Applied Microbiology and Biotechnology 51(2): 176-184. Malissard, M., S. Zeng, et al. (2000). "Expression of functional soluble forms of human beta-1,4- galactosyltransferase I, alpha-2,6-sialyltransferase, and 45 alpha-1,3-fucosyltransferase VI in the methylotrophic yeast Pichia pastoris." Biochemical and Biophysical Research Communications 267(1): 169-173. 131 Maras, M. and R. Contreras (1994). Methods of Modifying Carbohydrate Moieties. United States, Alko Group Ltd., Helsinki, Finland. Maras, M., A. De Bruyn, et al. (1999). "In vivo synthesis of complex N-glycans by expression of human N-acetylglucosaminyltransferase I in the filamentous 5 fungus Trichoderma reesei." Febs Letters 452(3): 365-370. Maras, M., X. Saelens, et al. (1997). "In vitro conversion of the carbohydrate moiety of fungal glycoproteins to mammalian-type oligosaccharides Evidence for N-acetylglucosaminyltransferase-I-accepting glycans from Trichoderma reesei." European Journal of Biochemistry 249(3): 701-707. 10 Martinet, W., M. Maras, et al. (1998). "Modification of the protein glycosylation pathway in the methylotrophic yeast Pichia pastoris." Biotechnology Letters 20(12): 1171-1177. McGarvey, P. B., J. Hammond, et al. (1995). "Expression of the Rabies Virus Glycoprotein in Transgenic Tomatoes." Bio-Technology 13(13): 1484 15 1487. Moens, S. and J. Vanderleyden (1997). "Glycoproteins in prokaryotes." Archives of Microbiology 168(3): 169-175. Nakanishishindo, Y., K. Nakayama, et al. (1993). "Structure of the N-Linked Oligosaccharides That Show the Complete Loss of Alpha-1,6-Polymannose 20 Outer Chain From Ochl, Ochl Mnnl, and Och1 Mnnl Alg3 Mutants of Saccharomyces- Cerevisiae." Journal of Biological Chemistry 268(35): 26338-26345. Raju, T. S., J. B. Briggs, et al. (2000). "Species-specific variation in glycosylation of IgG: evidence for the species-specific sialylation and branch-specific 25 galactosylation' and importance for engineering recombinant glycoprotein therapeutics." Glycobiology 10(5): 477-486. Sharma, C. B., R. Knauer, et al. (2001). "Biosynthesis of lipid-linked oligosaccharides in yeast: the ALG3 gene encodes the DoI-P-Man: Man(5)GlcNAc(2)-PP-DoI mannosyltransferase." Biological Chemistry 30 382(2): 321-328. Staub, J. M., B. Garcia, et al. (2000). "High-yield production of a human therapeutic protein in tobacco chloroplasts." Nature Biotechnology 18(3): 333-338. - Takeuchi, M. (1997). "Trial for molecular breeding of yeast for the production of 35 glycoprotein therapeutics." Trends in Glycoscience and Glycotechnology 9: S29-S35. Umana et al., Nat Biotechnol. 1999a Feb (17)176-180. (Engineered glycoforms of an antineuroblastoma IgG1 with optimized antibodydependent cellular 40 cytotoxic activity) Umana et al., Biotechnol Bioeng. 1999b Dec 5;65(5):542-549. (Regulated Overexpression of glycosyltransferase). Verostek, M. F., P. H. Atkinson, et al. (1993). "Glycoprotein-Biosynthesis in the Alg3 Saccharomyces-Cerevisiae Mutant .1. Role of Glucose in the Initial 45 Glycosylation of Invertase in the Endoplasmic-Reticulum." Lournal of Biological Chemistry 268(16): 12095-12103. Verostek, M. F., P. H. Atkinson, et al. (1993). "Glycoprotein-Biosynthesis in the Alg3 Saccharomyces-Cerevisiae Mutant .2. Structure of Novel Man6 132 10glcnac2 Processing Intermediates On Secreted Invertase." Journal of Biological Chemistry 268(16): 12104-12115. Weikert, S., D. Papac, et al. (1999). "Engineering Chinese hamster ovary cells to maximize sialic acid content of recombinant glycoproteins." Nature 5 Biotechnology 17(11): 1116-1121. Werner, R. G., W. Noe, et al. (1998). "Appropriate mammalian expression systems for biopharmaceuticals." Arzneimittel-Forschung-Drug Research 48(8): 870-880. Yang, M. and M. Butler (2000). "Effects of ammonia on CHO cell growth, 10 erythropoietin production, and glycosylation." Biotechnology and Bioenineering 68(4): 370-380.Zufferey, R., R. Knauer, et al. (1995). "Stt3, a Highly Conserved Protein Required..for Yeast Oligosaccharyl Transferase-Activity in-Vivo." EMBO Journal 14(20): 4949-4960. 133 Other embodiments of the invention as described herein are defined in the following paragraphs: 1. A method for producing a human-like glycoprotein in a non-human eukaryotic host cell that does not display a 1,6 mannosyltransferase activity with respect to the N-glycan on a glycoprotein, the method comprising the step of introducing into the host cell one or more enzymes for production of a Man 5 GlcNAc 2 carbohydrate structure, wherein Man 5 GlcNAc 2 is produced within the host cell at a yield of at least 30 mole percent. 2. The method of claim 1, wherein at least 10 percent ofthe Man 5 GlcNAc 2 produced within the host cell is a productive substrate for GnTI in vivo. 3. The method of claim 1, wherein at least one of the enzymes is selected to have optimal activity at the pH of the location in the host cell where the carbohydrate structure is produced. 4. The method of claim 2, wherein at least one of the enzymes is selected to have a pH optimum within about 1.4 pH units of the average pH optimum of other representative enzymes in the organelle in which the enzyme is localized. 5. The method of claim 1, wherein at least one of the enzymes is targeted to a subcellular location in the host cell where the enzyme will have optimal activity. 6. The method of claim 4, wherein the enzyme is targeted by means of a chimeric protein comprising a cellular targeting signal peptide not normally associated with the enzyme. 7. The method of claim 1, wherein at least one introduced enzyme is targeted to the endoplasmic reticulum, the early, medial, late Golgi or the trans Golgi network of the host cell. 8. The method of claim 1, wherein at least one of the enzymes is selected from the group consisting of mannosidases, glycosyltransferases and glycosidases. 9. The method of claim 6, wherein the enzyme is a mannosidase predominantly localized in the Golgi apparatus or the endoplasmic reticulum. 10. The method of claim 1, wherein the glycoprotein comprises N-glycans of which greater than 30 mole percent comprise six or fewer mannose residues. 11. The method of claim 1 , wherein the glycoprotein comprises N-glycans of which greater than 30 mole percent comprise three or fewer mannose residues. 12. The method of claim 1, wherein the glycoprotein comprises one or more sugars selected from the group consisting of GIcNAc, galactose, sialic acid, and fucose. 13. The method of claim 1, wherein the glycoprotein comprises at least one oligosaccharide branch comprising the structure NeuNAc-Gal-GIcNAc-Man. 14. The method of claim 1, wherein the host is selected from the group consisting of plant, algae, insect, fungi, yeast cells. 15. The method of claim 1, wherein the host is a lower eukaryotic cell. 1 'll 16. The method of claim 15, wherein the host cell is selected from the group consisting of Pichia pastoris, Pichiafinlandica, Pichia trehalophila, Pichia koclamae, Pichia membranaefaciens, Pichia opuntiae, Pichia thermotolerans, Pichia salictaria, Pichia guercuum, Pichiapijperi, Pichia stiptis, Pichia methanolica, Pichia sp., Saccharomyces cerevisiae, Saccharomyces sp., Hansenula polymorpha, Kluyveromyces sp., Kluyveromyces lactis, Candida albicans, Aspergillus nidulans, Aspergillus niger, Aspergillus oryzae, Trichoderma reesei, Chrysosporium lucknowense, Fusarium sp., Fusarium gramineum, Fusarium venenatum and Neurospora crassa. 17. The method of claim 1, wherein the host is deficient in the activity of one or more enzymes selected from the group consisting of mannosyltransferases and phosphomannosyltransferases. 18. The method of claim 17, wherein the host does not express an enzyme selected from the group consisting of 1,6 mannosyltransferase; 1,3 mannosyltransferase; and 1,2 mannosyltransferase. 19. The method of claim 1, wherein the host is an ochl mutant of P. pastoris. 20. The method of claim 1, wherein the host expresses GnTI and UDP-GlcNAc transporter activities. 21. The method of claim 1, wherein the host expresses a UDP- or GDP-specific diphosphatase activity. 22. The method of claim 1, further comprising the step of isolating the glycoprotein from the host. 23. The method of claim 22, further comprising the step of subjecting the isolated glycoprotein to at least one further glycosylation reaction in vitro, subsequent to its isolation from the host. 24. The method of claim 1, wherein the step of introducing into the host cell one or more enzymes for production of the Man 5 GlcNAc 2 carbohydrate structure comprises a nucleic acid molecule. 25. The method of claim 1, further comprising the step of introducing into the host a nucleic acid molecule encoding one or more mannosidase activities involved in the production of Man 5 GlcNAc 2 from MansGlcNAc 2 or Man 9 GlcNAc 2 . 26. The method of claim 25, wherein at least one of the encoded mannosidase activities has a pH optimum within about 1.4 pH units of the average pH optimum of other representative enzymes in the organelle in which the mannosidase activity is localized, or has optimal activity at a pH of between about 5.1 and about 8.0. 1'M 27. The method of claim 26, wherein the mannosidase has optimal activity at a pH of between about 5.5 and about 7.5. 28. The method of claim 26, wherein the mannosidase activity is an a-1,2-mannosidase derived from mouse, human, Lepidoptera, A spergillus nidulans, or Bacillus sp., C.elegans, D. melanogaster, P. citrinum or X laevis. 29. The method of claim 24, wherein at least one enzyme is localized by forming a fusion protein comprising a catalytic domain of the enzyme and a cellular targeting signal peptide. 30. The method of claim 29, wherein the fusion protein is encoded by at least one genetic construct formed by the in-frame ligation of a DNA fragment encoding a cellular targeting signal peptide with a DNA fragment encoding a catalytic domain having enzymatic activity. 31. The method of claim 30, wherein the encoded targeting signal peptide is derived from a member of the group consisting of: membrane-bound proteins of the ER or Golgi, retrieval signals, Type II membrane proteins, Type I membrane proteins, membrane spanning nucleotide sugar transporters, mannosidases, sialyltransferases, glucosidases, mannosyltransferases and phospho-mannosyltransferases. 32. The method of claim 24, wherein the catalytic domain encodes a glycosidase, mannosidase or a glycosyltransferase activity derived from a member of the group consisting of GnTI, GnTII, GnTIll, GnTIV, GnT V, GnT VI, GaIT, Fucosyltransferase and Sialyltransferase, and wherein the catalytic domain has a pH optimum within 1.4 pH units of the average pH optimum of other representative enzymes in the organelle in which the enzyme is localized, or has optimal activity at a pH between 5.1 and 8.0. 33. The method of claim 32, wherein the catalytic domain encodes a mannosidase selected from the group consisting of C. elegans mannosidase IA, C. elegans mannosidase IB, D. melanogaster mannosidase IA, H. sapiens mannosidase IB, P. citrinum mannosidase I, mouse mannosidase IB, mouse mannosidase IB, A. nidulans mannosidase IA, A. nidulans mannosidase IB, A. nidulans mannosidase IC, mouse mannosidase II, C. elegans mannosidase II, H. sapiens mannosidase II, and mannosidase III. 34. The method of claim 24, wherein the nucleic acid molecule encodes one or more enzymes selected from the group consisting of UDP-GlcNAc transferase, UDP-galactosyltransferase, GDP-fucosyltransferase, CMP-sialyltransferase, UDP-GlcNAc transporter, UDP-galactose transporter, GDP-fucose transporter, CMP-sialic acid transporter, and nucleotide diphosphatases. 35. The method of claim 24, wherein the host expresses GnTI and UDP-GlcNAc transporter activities.

36. The method of claim 24, wherein the host expresses a UDP- or GDP-specific diphosphatase activity. 37. A nucleic acid library comprising at least two different genetic constructs, wherein at least one genetic construct comprises a nucleic acid fragment encoding a glycosylation enzyme ligated in-frame with a nucleic acid fragment encoding a cellular targeting signal peptide which it is not normally associated with. 38. A DNA library of fusion constructs comprising: (a) at least two nucleotide sequences encoding a cellular targeting signal peptide and at least one nucleotide sequence encoding a catalytic domain region selected from the group consisting of mannosidases, glycosyltransferases and glycosidases; or (b) at least one nucleotide sequence encoding a cellular targeting signal peptide and at least two nucleotide sequences encoding a catalytic domain region selected from the group consisting of mannosidases, glycosyltransferases and glycosidases; wherein at least one nucleotide sequence encoding a catalytic domain region is ligated in frame to a nucleotide sequence encoding a cellular targeting signal peptide. 39. The DNA library of claim 37 comprising at least one nucleic acid comprising a naturally occurring sequence encoding a glycosylation enzyme. 40. The DNA library of claim 37 comprising at least one nucleic acid sequence previously subjected to a technique selected from the list: gene shuffling, in vitro mutagenesis, and error prone polymerase chain reaction. 41. The DNA library of claim 37, wherein the glycosyltransferase is selected from the group consisting of: mannosyltransferases, GleNAc transferases, phospho-GlcNAc transferases, galactosyltransferases, sialyltransferases and fucosyltransferases. 42. The DNA library of claim 37, wherein at least one nucleotide sequence encoding a catalytic domain region is derived from mouse, human, C.elegans, D. melanogaster, P. citrinum, X laevis, Bacillus sp.or A.nidulans. 43. The DNA library of claim 37, wherein the mannosidase catalytic domain is selected from the group consisting of: C. elegans mannosidase IA, C. elegans mannosidase IB, D. melanogaster mannosidase IA, H. sapiens mannosidase IB, P. citrinum mannosidase I, mouse mannosidase IA, mouse mannosidase IB, A. nidulans mannosidase IA, A. nidulans mannosidase IB, A. nidulans mannosidase IC, mouse mannosidase II, C. elegans mannosidase II, H sapiens mannosidase II, and mannosidase III. 44. The DNA library of claim 37, wherein the nucleic acid fragment encoding a cellular targeting signal peptide is selected from the group consisting of: 1 '1' membrane-bound proteins of the ER or Golgi, retrieval signals, Type II membrane proteins, Type I membrane proteins, membrane spanning nucleotide sugar transporters, mannosidases, sialyltransferases, glucosidases, mannosyltransferases and phosphomannosyltransferases. 45. The DNA library of claim 37, wherein the nucleic acid fragment encoding a cellular targeting peptide is selected from the group consisting of: Saccharomyces GLS1, Saccharomyces MNS1, Saccharomyces SEC 12, Pichia SEC, Pichia OCH 1, Saccharomyces MNN9, Saccharomyces VAN 1, Saccharomyces ANP 1, Saccharomyces HOC], Saccharomyces MNN 10, Saccharomyces MNN 11, Saccharomyces MNT1, Pichia D2, Pichia D9, Pichia J3, Saccharomyces KTR1, Saccharomyces KTR2, Kluyveromyces GnTI, Saccharomyces MNN2, Saccharomyces MNN5, Saccharomyces YURI, Saccharomyces MNN1, and Saccharomyces MNN6. 46. A vector comprising a fusion construct derived from a DNA library of any one of claims 37-45 operably linked to an expression control sequence, wherein said cellular targeting signal peptide is targeted to the ER, Golgi or trans-Golgi network. 47. The vector of claim 46 which, upon expression in a host cell, encodes a mannosidase activity involved in producing Man 5 GlcNAc 2 in vivo. 48. The vector of claim 46 which, upon expression in a host cell, encodes a glycosyltransferase activity involved in producing GlcNAcMan 5 GIcNAc 2 in vivo. 49. A eukaryotic host cell comprising at least one vector of claim 46. 50. The host cell of claim 49, selected from the group consisting of unicellular and multicellular fungi. 51. The host cell of claim 49, selected from the group consisting of Pichia pastoris, Pichia finlandica, Pichia trehalophila, Pichia koclamae, Pichia membranaefaciens, Pichia opuntiae, Pichia thermotolerans, Pichia salictaria, Pichia guercuum, Pichia pijperi, Pichia stiptis, Pichia methanolica, Pichia sp., Saccharomyces cerevisiae, Saccharomyces sp., Hansenula polymorpha, Kluyveromyces sp., Kluyveromyces lactis, Candida albicans, Aspergillus nidulans, Aspergillus niger, Aspergillus oryzae, Trichoderma reesei, Chrysosporium lucknowense, Fusarium sp. Fusarium gramineum, Fusarium venenatum and Neurospora crassa. 52. A method for producing a human-like glycoprotein in a non-human cell comprising the step of culturing a eukaryotic host cell comprising at least one vector of claim 46. 53. The method of claim 52, wherein the host is a unicellular or multicellular fungal cell. 54. The method of claim 52, wherein the host cell is selected from the group consisting of Pichia pastoris, Pichiafinlandica, Pichia trehalophila, Pichia koclamae, Pichia membranaefaciens, Pichia opuntiae, Pichia thermotolerans, Pichia salictaria, Pichia guercuum, 11)0 Pichia pijperi, Pichia stiptis, Pichia methanolica, Pichia sp., Saccharomyces cerevisiae, Saccharomyces sp., Hansenula polymorpha, Kluyveromyces sp., Kluyveromyces lactis Candida albicans, Aspergillus nidulans, Aspergillus niger, Aspergillus oryzae, Trichoderma reesei, Chrysosporium lucknowense, Fusarium sp. Fusarium gramineurn, Fusarium venenatum and Neurospora crassa. 55. A method for producing a human-like glycoprotein in a non-human host cell comprising the step of transforming the host cell with a DNA library of any one of claims 37-45 to produce a genetically mixed cell population expressing at least one glycosylation enzyme derived from the library. 56. The method of claim 55, further comprising the step of selecting from the mixed cell population a cell producing a desired human-like glycosylation phenotype. 57. The method of claim 56, wherein the selection comprises the step of analyzing a glycosylated protein or isolated N-glycan by one or more methods selected from the group consisting of: (a) mass spectroscopy; (b) MALDI-TOF; (c) liquid chromatography; (d) characterizing cells using a fluorescence activated cell sorter, spectrophotometer, fluorimeter, or scintillation counter; (e) exposing host cells to a lectin or antibody having a specific affinity for a desired oligosaccharide moiety; and (f) exposing cells to a cytotoxic or radioactive molecule selected from the group consisting of sugars, antibodies and lectins. 58. The method of claim 55, wherein the DNA fragment encoding the catalytic domain has an activity selected from the group consisting of mannosidase, UDP-GlcNAc transferase, UDP galactosyltransferase, and CMP-sialyltransferase activity and wherein the cellular targeting signal peptide localizes the enzyme predominantly in a host cell organelle selected from the group consisting of endoplasmic reticulum, cis Golgi, medial Golgi, and trans Golgi. 59. The method of claim 55, wherein said host cell further comprises a target glycoprotein of interest on which Man 5 GlcNAc 2 is produced in vivo. 60. The method of claim 59, wherein the Man 5 GlcNAc 2 produced in vivo is the predominant N-glycan on the target glycoprotein. 61. The method of claim 55, wherein said host cell further comprises a target glycoprotein of interest on which GlcNAcMan 5 GlcNAc2 is produced in vivo. 62. The method of claim 61, wherein the GlcNAcMan 5 GlcNAc2 produced in vivo is the predominant N-glycan on the target glycoprotein. 63. A host cell produced by the method of claim 1, 24 or 55. 64. A human-like glycoprotein produced by the method of claim 1, 24 or 55.

65. A method for altering the glycosylation pattern of a eukaryotic cell comprising the step of transforming the host cell with a DNA library of any one of claims 37-45 to produce a genetically mixed cell population expressing at least one glycosylation enzyme derived from the library. 66. An isolated nucleic acid molecule comprising or consisting of nucleic acid sequences selected from the group consisting of: (a) at least forty-five (45) contiguous nucleotide residues of SEQ ID NO:41; (b) homologs, variants and derivatives of (a); and (c) nucleic acid sequences that hybridize under stringent conditions to (a) but excluding sequences which encode the S. cerevisiae OCHI gene. 67. An isolated polypeptide comprising the amino acid sequence of SEQ ID NO:42. 68. An isolated nucleic acid molecule of claim 66 which encodes an OCHI activity upon expression in a host cell. 69. An isolated nucleic acid molecule of claim 66 which encodes a K.lactis OCHI gene. 70. An isolated nucleic acid molecule comprising or consisting of nucleic acid sequences selected from the group consisting of: (a) at least forty-five (45) contiguous nucleotide residues of SEQ ID NO:43; (b) homologs, variants and derivatives of (a); and (c) nucleic acid sequences that hybridize under stringent conditions to (a) but excluding sequences which encode the S. cerevisiae MNN] gene. 71. An isolated polypeptide comprising the amino acid sequence of SEQ ID NO:44. 72. An isolated nucleic acid molecule of claim 70 which encodes an MNN1 gene. 73. An isolated nucleic acid molecule of claim 70, wherein said sequence encodes a K.lactis MNNI gene. 74. A host cell comprising a disruption or mutation of SEQ ID NO:41 which is characterized by having a reduced expression level of SEQ ID NO:41 compared to a host cell without said disruption or mutation. 75. A host cell comprising a disruption or mutation of SEQ ID NO:43 which is characterized by having a reduced expression level of SEQ ID NO:43 compared to a host cell without said disruption or mutation. 76. A method of modifying plant glycosylation comprising introducing into a host at least one nucleotide sequence encoding a catalytic domain region ligated in-frame to a nucleotide sequence encoding a cellular targeting signal peptide. Still further embodiments are within the scope of the following claims. 140

Claims (14)

1. 2. The host cell of claim 1, wherein the host cell is capable of producing N-glycans of which greater than 30 mole percent comprise six or fewer mannose residues.
2. 3. The host cell of claim I or claim 2, wherein the host cell is capable of producing N-glycans of which greater than 30 mole percent comprise three or fewer mannose residues.
3. 4. The host cell of any one of claims I to 3, wherein the host cell is capable of producing N-glycans comprising one or more sugars selected from the group consisting of GlcNAc, galactose, sialic acid, and fucose.
4. 5. The host cell of any one of claims I to 4, wherein the host cell is capable of producing N-glycans with at least one oligosaccharide branch comprising the structure NeuNAcGalGlcNAcMan.
5. 6. The host cell of any one of claims I to 5, wherein the host cell is selected from the group consisting of Pichiapastoris, Pichiafinlandica, Pichia trehalophila, Pichia koclamae, Pichia membranaefaciens Pichia opuntiae Pichia thermotolerans Pichia salictaria, Pichia guercuum, Pichia pijperi, Pichia stiptis, Pichia methanolica, Pichia sp., Saccharonyces cerevisiae, Saccharomyces sp., Hansenula polymorpha, Kluyveromyces sp., Kluyveromyces lactis, Candida albicans, Aspergillus nidulans, Aspergillus niger, Aspergillus oryzae, Trichoderma reesei, Chrysosporium lucknowense, Fusarium sp., Fusarium gramineum, Fusarium venenatum and Neurospora crassa.
6. 7. The host cell of claim 6, wherein the host is deficient in the activity of one or more enzymes selected from the group consisting of mannosyltransferases and phosphomannosyltransferases.
7. 8. The host cell of claim 7, wherein the host does not express an enzyme selected from the group consisting of 1,6 mannosyltransferase; 1,3 mannosyltransferase; and 1,2 mannosyltransferase.
8. 9. The host cell of claim 8, wherein the host is an ochl mutant of P. pastoris.
9. 10. The host cell of any one of claims I to 9, wherein the host cell further expresses GnTI and UDP GlcNAc transporter activities.
10. 11. The host cell of any one of claims I to 10, wherein the a-1,2-mannosidase is derived from mouse, human, Lepidoptera, Aspergillus nidulans, or Bacillus sp., C.elegans, D.melanogaster, P.citrinum or X laevis.
11. 12. The host cell of any one of claims 1 to 11, wherein the host cell further includes a nucleic acid encoding a catalytic domain that encodes a glycosidase, mannosidase or a glycosyltransferase activity derived from a member of the group consisting of GnTI, GnTII, GnTIII, GnTIV, GnT V, GnT VI, GaIT, Fucosyltransferase and Sialyltransferase, and wherein the catalytic domain has a pH optimum within 1.4 1 A 1 pH units of the average pH optimum of other representative enzymes in the organelle in which the enzyme is localized, or has optimal activity at a pH between 5.1 and 8.0.
12. 13. The host cell of claim 12, wherein the mannosidase is selected from the group consisting of C. elegans mannosidase IA, C. elegans mannosidase IB, D. melanogaster mannosidase IA, H. sapiens mannosidase IB, P. citrinum mannosidase I, mouse mannosidase IA, mouse mannosidase IB, A. nidulans mannosidase IA, A. nidulans mannosidase IB, A. nidulans mannosidase IC, mouse mannosidase II, C. elegans mannosidase II, H. sapiens mannosidase II, and mannosidase III.
13. 14. The host cell of claim 12 or claim 13, wherein the nucleic acid molecule encodes one or more enzymes selected from the group consisting of UDP-GIcNAc transferase, UDP-galactosyltransferase, GDP-fucosyltransferase, CMP-sialyltransferase, UDP-GlcNAc transporter, UDP-galactose transporter, GDP-fucose transporter, CMP-sialic acid transporter, and nucleotide diphosphatases.
14. 15. The host cell of claim 12, wherein the host cell further expresses GnTI and UDP-GlcNAc transporter activities. Date: 21 September 2012 1 A')