Engineering Yarrowia lipolytica to Produce Glycoproteins Homogeneously Modified with the Universal Man3GlcNAc2 N-Glycan Core

Yarrowia lipolytica is a dimorphic yeast that efficiently secretes various heterologous proteins and is classified as “generally recognized as safe.” Therefore, it is an attractive protein production host. However, yeasts modify glycoproteins with non-human high mannose-type N-glycans. These structures reduce the protein half-life in vivo and can be immunogenic in man. Here, we describe how we genetically engineered N-glycan biosynthesis in Yarrowia lipolytica so that it produces Man3GlcNAc2 structures on its glycoproteins. We obtained unprecedented levels of homogeneity of this glycanstructure. This is the ideal starting point for building human-like sugars. Disruption of the ALG3 gene resulted in modification of proteins mainly with Man5GlcNAc2 and GlcMan5GlcNAc2 glycans, and to a lesser extent with Glc2Man5GlcNAc2 glycans. To avoid underoccupancy of glycosylation sites, we concomitantly overexpressed ALG6. We also explored several approaches to remove the terminal glucose residues, which hamper further humanization of N-glycosylation; overexpression of the heterodimeric Apergillus niger glucosidase II proved to be the most effective approach. Finally, we overexpressed an α-1,2-mannosidase to obtain Man3GlcNAc2 structures, which are substrates for the synthesis of complex-type glycans. The final Yarrowia lipolytica strain produces proteins glycosylated with the trimannosyl core N-glycan (Man3GlcNAc2), which is the common core of all complex-type N-glycans. All these glycans can be constructed on the obtained trimannosyl N-glycan using either in vivo or in vitro modification with the appropriate glycosyltransferases. The results demonstrate the high potential of Yarrowia lipolytica to be developed as an efficient expression system for the production of glycoproteins with humanized glycans.


Introduction
There is increasing demand for efficient expression systems for the economical production of biopharmaceuticals. The properties of recombinant biopharmaceutical proteins can be fine-tuned by manipulating the glycan structures attached to them. However, versatile production methods for producing specific glycoforms are few and involve mostly laborious in vivo pathway engineering.
To rapidly generate different glycoforms of a particular biopharmaceutical for functional studies and subsequent production, it would be valuable to have a microbial expression system that produces N-glycoproteins homogenously modified with the Man 3 GlcNAc 2 N-glycan core. This core is common to all mammalian N-glycan structures, and any complex type N-glycan can be built in vitro on this core using the appropriate glycosyltranferases and sugar-nucleotide donors. However, no convenient expression system producing this Man 3 GlcNAc 2 core is currently available. Our objective was to engineer the yeast Yarrowia lipolytica for this purpose.
Yeasts combine the ease of genetic manipulation and up-scaling of microbial cultures with the ability to secrete and modify proteins with the major eukaryotic post-translational modifications. Saccharomyces cerevisiae and the methylotrophic yeasts Pichia pastoris and Hansenula polymorpha are the most frequently used yeast hosts for recombinant protein production, but there is growing interest in the dimorphic yeast Yarrowia lipolytica. This yeast can grow to high cell density on long-chain fatty acids. The promoters of acyl-CoA oxidase (POX) genes are strongly induced on this carbon source and are therefore used to drive heterologous gene expression. Moreover, Y. lipolytica has long been used for the production of lipases for the agro-food industry and is therefore classified as GRAS (generally regarded as safe).
To generate a Y. lipolytica strain producing Man 3 GlcNAc 2 on its glycoproteins, we engineered the ER-localized components of the N-glycosylation pathway. At the cytoplasmic side of the ER membrane, N-glycosylation starts with the synthesis of a dolichol linked glycan precursor ( Figure 1A). The intermediate Man 5 -GlcNAc 2 -PP-Dol structure flips to the luminal side of the ER, where it is further elongated, first by the a-1,3-mannosyltransferase Alg3p, and then by other mannosyltransferases until Man 9 GlcNAc 2 is formed. This dolichol linked sugar is then glucosylated by the a-1,3-glucosyltransferase Alg6p, after which two more glucoses are added. The resultant glycan (Glc 3 Man 9 GlcNAc 2 ) is transferred to the nascent polypeptide chain ( Figure 1A) [1].
In a process of quality control for protein folding [2], all glucose residues are trimmed sequentially. The first two glucose molecules are removed rapidly by the consecutive action of glucosidase I and II, whereas the last a-1,3-linked glucose residue is removed more slowly by glucosidase II (GII). Monoglucosylated proteins are recognized by calnexin and/or calreticulin. These ER chaperones aid the folding of the glycoprotein and do not reassociate with the glycoprotein once the last glucose residue is removed by GII. If the glycoprotein does not fold properly, it is glucosylated again by the UDP-glucose:glycoprotein glucosyltransferase, after which it again binds calnexin and/or calreticulin and reenters the folding cycle. When the glycoprotein is correctly folded and the sugars are trimmed to Man 8 GlcNAc 2 by ER mannosidase I, the protein proceeds along the secretory pathway. In the Golgi apparatus of yeasts, the Man 8 GlcNAc 2 N-glycans are further extended by the addition of mannose and phospho-mannose residues. This elongation is initiated by the a-1,6-mannosyltransferase Och1p [3,4]. In contrast, higher eukaryotes first trim the glycans to Man 5 GlcNAc 2 by Golgi mannosidases I and then further modify them to complex type glycans [5][6][7].
Several methods can be envisioned to engineer yeast for the production of homogeneous, universal glycan 'scaffolds' on which different types of eukaryotic N-glycans can be built [8]. One approach is to engineer only Golgi-localized processes so that the more essential ER-localized steps of the N-glycosylation pathway are not affected. This has been successfully implemented in P. pastoris [9,10]. Another approach is to interfere with the ER steps of the pathway. This is particularly attractive at the ALG3 step: disruption of ALG3 is expected to lead to the glycosylation of proteins with Man 5 GlcNAc 2 N-glycans, which should be easy to trim to Man 3 GlcNAc 2 with an a-1,2-mannosidase ( Figure 1B). Man 3 GlcNAc 2 is the common core of all types of eukaryotic Nglycans and provides an ideal scaffold for in vitro or in vivo synthesis of different glycoforms.
However, at least in S. cerevisiae [11,12], the situation is complicated because Man 5 GlcNAc 2 -PP-Dol is glucosylated by Alg6p less efficiently than Man 9 GlcNAc 2 -PP-Dol. Glucosylation of the N-glycan precursor is important for its efficient transfer to nascent proteins by oligosaccharyltransferase, and reduced glucosylation diminishes this transfer. Previous studies have not addressed this shortcoming of this otherwise attractive engineering approach. Here, we report that the glucose residues on glycoproteins produced in alg3 strains are not removed efficiently by Yarrowia GII, and we describe the engineering strategy we used to solve this problem ( Figure 1B). Through this integrated 'systems engineering' approach, we succeeded in creating a glyco-engineered Y. lipolytica strain that produces glycoproteins homogeneously modified with the trimannosyl core N-glycan (Man 3 -GlcNAc 2 ).

ALG3 Gene Knock-out
In order to alter Y. lipolytica to produce heterologous proteins glycosylated with Man 3 GlcNAc 2 , we interfered with biosynthesis of the core N-glycan ( Figure 1B, step1). Elimination of Alg3p a-1,3-mannosyltransferase prevents the addition of an a-1,3-mannose to the a-1,6-arm of the ER Man 5 -PP-Dol structure. Knock-out of ALG3 should lead to accumulation of its substrate, Man 5 GlcNAc 2 [13,14].
To disrupt the Y. lipolytica ALG3 gene, we constructed a plasmid that includes parts of the promoter and terminator of ALG3 and has a URA3 selection marker cassette (pYLalg3PUT). The NotI and PacI sites were used to linearize the vector in order to remove the E. coli related DNA elements before transformation of wild type (WT) Y. lipolytica MTLY60 (Table 1). Double homologous recombination at the promoter and terminator sites replaced ALG3 with the URA3 selectable marker, which resulted in the alg3::URA3 mutant strain YLA3 (Table 1). To study the effect of this mutation, we analyzed the N-glycan profile of proteins that completely traverse the yeast's secretory system, i.e. cell wall mannoproteins. Whereas the wild type mannoproteins contained mainly Man 8 GlcNAc 2 and Man 9 GlcNAc 2 N-glycans ( Figure 2, panel C), the alg3 mutants proteins had three glycan structures ( Figure 2, panel D). As expected, one of these structures ran at about the same position in electrophoresis as the Man 5 GlcNAc 2 sugar structure of RNaseB, but two others ran at positions corresponding to one and two extra monosaccharide units ( Figure 2, panel D). This was the case for all transformants that were confirmed by PCR on gDNA to be alg3 knock-outs.
To further elucidate the structures of the two additional Nglycans, we performed exoglycosidase digests with a-1,2-mannosidase, Jack Bean (JB) a-mannosidase and purified rat liver GII and analysed the products using capillary electrophoresis. The peak that had reached the same position as Man 5 GlcNAc 2 of the RNaseB marker shifted two glucose units after a-1,2-mannosidase treatment ( Figure 2, panel E) and four glucose units after broadspecificity a-mannosidase (JB) digestion ( Figure 2, panel F). This fits with the dolichol-linked Man 5 GlcNAc 2 structure, as expected. The additional two glycans are not affected by a-1,2-mannosidase digestion. Also, both peaks shifted only one glucose unit upon amannosidase (JB) digestion. However, both glycans were sensitive to GII digestion and were converted to Man 5 GlcNAc 2 ( Figure 2 Compensation for Underoccupancy of the N-glycan Sites by Overexpressing ALG6 The alg3 mutation in S. cerevisiae causes underoccupancy of Nglycosylation sites [12,13,[15][16][17]. Efficient transfer of the dolichol linked N-glycan precursor to a protein by the oligosaccharyltransferase complex (OST) requires the triglucosyl glycotope on the dolichol-linked precursor [1,18]. The first glucosyltransferase, Alg6p, can glucosylate the Man 5 GlcNAc 2 -PP-Dol structure in alg3 S. cerevisiae [12], but with low efficiency. This results in underglucosylation of the dolichol linked precursor, poor transfer by OST, and reduced occupancy of N-glycosylation sites. Anticipating this problem, we incorporated an Alg6p constitutive overexpression cassette in the alg3 knock-out vector ( Figure 1B, step2). The resultant vector (pYLalg3PUT-ALG6) was transformed into WT Y. lipolytica MTLY60, yielding strain YLA3-A6 (Table 1). Upon DSA-FACE analysis of the N-glycans derived from mannoproteins, all transformants in which alg3 knock-out was confirmed by PCR exhibited a change in glycosylation pattern. The proportion of glucosylated Man 5 GlcNAc 2 increased substantially ( Figure 3, panel E) compared to the alg3 mutant without Alg6p overexpression (Figure 3, panel D). This indicates that Alg6p activity was indeed augmented and clearly shows that the endogenous Y. lipolytica GII activity was insufficient to deglucosylate its suboptimal Glc 1-2 Man 5 GlcNAc 2 substrates.
To evaluate the underoccupancy of N-glycosylation sites in our different strains, we examined the N-glycosylation of overexpressed Y. lipolytica lipase 2 (LIP2), which has two glycosylation sites [19,20]. We analyzed the pattern of secreted proteins before and after Ndeglycosylation with PNGaseF. For the wild type strain, a single LIP2 band with a smear of hyper-N-glycosylation is observed ( Figure 4 It should be noted that this phenotype was observed in cells in mid-log phase of growth, and that it was much less pronounced in stationary-phase cells (data not shown). The difference is probably due to the considerably slower flux of proteins through the N-glycosylation pathway in stationary phase.
Interestingly, no hyperglycosylation of LIP2 was seen in the alg3 and alg3ALG6 strains, which means that our strategy need not involve knocking out any Golgi mannosyltransferases to obtain homogeneous glycosylation, contrary to previous approaches [9,10].
Consequently, we solved the underglycosylation problem of the alg3 mutant by overexpressing Alg6p, but this was at the expense of further augmenting the fraction of undesired glucosylated Man 5 GlcNAc 2 derivatives.

Removal of Capping Glucoses
In strains in which alg3 is disrupted, the N-glycans are capped by GII-hydrolyzable glucose residues. This type of capping is more pronounced when the ALG6 gene is overexpressed. Since the presence of these glucose residues prevents conversion of Man 5 GlcNAc 2 to Man 3 GlcNAc 2 by an introduced a-1,2-mannosidase ( Figure 1B, step 4), our next objective was to eliminate those glucose residues by further in vivo engineering.
Removal of capping glucose residues: mutanase and T.
brucei GII. We examined the possibility of using the mutanase of Trichoderma harzianum to remove the capping glucose residues on the Man 5 GlcNAc 2 glycans. Both unwanted glucose residues are a-1,3-linked to the rest of the sugar, and this mutanase has a-1,3glucosidase activity. A dilution series of the Novozyme 234 mutanase preparation was added to the oligosaccharides derived from the YLA3-A6 strain (Man 5 GlcNAc 2 , GlcMan 5 GlcNAc 2 and Glc 2 Man 5 GlcNAc 2 ). The DSA-FACE profile ( Figure 5B, panel G) shows that Glc 2 Man 5 GlcNAc 2 was effectively hydrolyzed to GlcMan 5 GlcNAc 2 . However, GlcMan 5 GlcNAc 2 was not deglucosylated further. It should be noted that Man 5 GlcNAc 2 was also trimmed, most probably by a contaminating mannosidase in the crude enzyme mixture. Since complete deglucosylation could not be obtained with this mutanase, we abandoned this approach. The approach used to engineer the yeast specific pathway. A: Standard N-glycosylation pathway in the ER. The early steps in N-glycosylation start with the synthesis of a dolichol-linked Man 5 GlcNAc 2 glycan precursor that flips to the ER lumen, where it is further elongated with mannoses starting with the activity of Alg3p mannosyltransferase. The resulting dolichol-linked Man 9 GlcNAc 2 precursor is then also glucosylated starting with the activity of Alg6p glucosyltransferase. When complete, the Glc 3 Man 9 GlcNAc 2 glycan is transferred en bloc to the nascent polypeptide chain. These glycans are then subjected to a protein folding quality control process involving de-glucosylation by glucosidases I and II (GI, GII) and re-glucosylation glucosyltransferase. B: The engineering strategies used to obtain a Y. lipolytica strain that produces glycoproteins homogeneously modified with the trimannosyl core N-glycan (Man 3 GlcNAc 2 ). First, ALG3 was knocked out (1), then Alg6p was overexpressed (2), then GII was overexpressed (3), and finally a-1,2-mannosidase was overexpressed (4). Conforming to the representation proposed by the Consortium for Functional Glycomics Nomenclature Committee, the green and blue spheres represent mannose (Man) and glucose (Glc), respectively, and blue squares represent Nacetylglucosamine residues (GlcNAc). C: Man 3 GlcNAc 2 -glycans can be further modified to any complex-type N-glycan structure using a combination of glycosyl-transferases, either in vitro or in vivo. doi:10.1371/journal.pone.0039976.g001 As an alternative strategy, we overexpressed the T. brucei GII asubunit. T. brucei uses a dual N-glycosylation system that can transfer both Man 9 GlcNAc 2 and Man 5 GlcNAc 2 to proteins ( Figure 5A) [21]. Furthermore, unlike organisms that exclusively transfer Glc 3 Man 9 GlcNAc 2 , the GII enzyme in T. brucei uses GlcMan 5 GlcNAc 2 as a preferred substrate [22]. Therefore, we tested whether the T. brucei enzyme can deglucosylate these structures in our engineered strains. We transformed the YLA3-A6 strain with pYLHmAXTbGIIa, which resulted in a YLTBGIIA strain (Table 1) and analyzed its cell wall mannoprotein glycans. No deglucosylation was observed ( Figure 5B, panel D). As GII is heterodimeric [23], we considered the possibility that the asubunit of T. brucei GII cannot dimerize with the b-subunit of Y. lipolytica GII and would thus not be retained in the endoplasmic reticulum. So we introduced an HDEL ER-retrieval tag at the Cterminus of the a-subunit of T. brucei GII. Moreover, we expressed the T. brucei enzyme once with its own signal peptide and once with the Y. lipolytica LIP2 signal peptide in the YLA3-A6 strain (yielding strains YLTBGIIAHDEL and YLTBpreGIIAHDEL, respectively) ( Table 1). N-glycan analysis of the clones overexpressing the HDEL-tagged a-subunit showed reduced abundance of the mono-glucosylated Man 5 GlcNAc 2 peak ( Figure 5B, panel E and F), whereas the di-glucosylated Man 5 GlcNAc 2 structure was not hydrolyzed. Evidently, this latter structure is not a substrate for the T. brucei GII. Consequently, this engineering approach also did not solve our problem, so we abandoned it.
Removal of capping glucoses by overexpression of the endogenous GII. To eliminate mono-and di-glucosylated Man 5 GlcNAc 2 structures in vivo, the YLA3-A6 strain was genetically engineered to overexpress the Y. lipolytica GII. This enzyme is a heterodimer consisting of two subunits, of which the asubunit is catalytically active [23] and contains a GH31 family domain [24]. We started by overexpressing the a-subunit in our YLA3-A6 strain. Glucosylation of the various glycans in the resultant strain, YLYLGIIA, was not reduced (Figure 3 It is believed that the b-subunit, which contains an HDEL tag, serves primarily to retain the a-subunit in the ER [23,[25][26][27]. Therefore, first we tried mimicking the b-subunit's function by adding an HDEL tag to the C-terminus of the a-subunit of the Y. lipolytica GII. This way, the tag would serve to retrieve the enzyme from the Golgi apparatus to the ER via COPI vesicles and thereby help to maintain the enzyme at its site of action. Again, aglucose removal was not improved in any of the transformation clones of the resultant YLYLGIIAHDEL strain (Figure 3, panel G).
Several studies have indicated the necessity of the b-subunit of the GII complex for maturation, solubility, stability and enzymatic activity on natural substrates [25][26][27][28][29]. Overexpression of the a subunit of Y. lipolytica GII alone was not sufficient to reduce the unwanted glucosylation on the Man 5 GlcNAc 2 glycan. Therefore we simultaneously overexpressed the b-subunit in two strains that overexpress the Y. lipolytica GII a-subunit with or without HDEL tag and we tested both the hp4d and the TEF promoter. We retained the clone with the best glycan profile, i.e. the one that removed a-glucose most efficiently. The best result was obtained in a strain that overexpressed the Y. lipolytica GII a-subunit with the HDEL tag, with a slightly improved effect when the Y. lipolytica GII b-subunit was expressed from the TEF promoter compared to the hp4d promoter. Therefore, we created a strain that overexpressed both the Y. lipolytica GII a and b-subunit driven by the TEF promoter. The strain was named YLYLGIIAB (Figure 3, panel H). However, though overexpression of both aand bsubunits of Y. lipolytica GII significantly reduced the proportion of glucosylated Man 5 GlcNAc 2 , it was still insufficiently effective for homogeneous glycoprotein production.

Removal of capping glucoses by overexpression of the A.
niger GII. Kainz and colleagues [30] recently reported that knockout of ALGC, the ALG3 homologue in the filamentous fungus A. niger, leads to the synthesis of Man 3-6 GlcNAc 2 glycans. In vitro digestion of these glycans with a-1,2-mannosidase gave almost exclusively Man 3 GlcNAc 2 [30]. Hence, no glucosylated glycan structures were detected when the ALG3 gene was disrupted in A. niger. Therefore, we assumed that the GII of A. niger can cope better with the alterations in N-glycan substrate structures caused by inactivation of the ER-mannosyltransferase Alg3p. Indeed, overexpression of the HDEL-tagged a-subunit of A. niger GII alone in our Y. lipolytica alg3 strain overexpressing ALG6, i.e. YLA3-A6, resulted in trimming of the glucosylated Man 5 GlcNAc 2 forms in the newly made YLANGIIA strain (Figure 3, panel I). No differences were seen between the strains that overexpressed the asubunit of A. niger GII under control of the TEF or under control of the hp4d promoter (data not shown). We subsequently overexpressed the b-subunit of A. niger GII in the YLANGIIA strain that overexpressed the HDEL-tagged a-subunit of A. niger GII, also under control of the TEF promoter. The resultant strain was named YLANGIIAB. Analysis of the glycan structures on glycoproteins produced by this strain showed very efficient conversion of glucosylated to non-glucosylated Man 5 GlcNAc 2 glycan structures (Figure 3, panel J), which represented about 80% of the total cell wall mannoprotein N-glycan pool.

Discussion
Y. lipolytica has emerged as a suitable system for heterologous protein expression [33]. With the increasing importance of yeasts  as an alternative host for recombinant protein production, it has become important to glyco-engineer yeasts for production of humanized glycans for therapeutic purposes. We aimed to engineer the Yarrowia ER glycosylation pathway for the production of the Man 3 GlcNAc 2 core N-glycan structure, which can be converted to any desired mammalian N-glycan using Golgi glycosyltransferases ( Figure 1C).
Upon disruption of the ALG3 gene in Y. lipolytica, we observed the expected Man 5 GlcNAc 2 (dolichol-linked type) as well as two additional glycan structures: GlcMan 5 GlcNAc 2 and Glc 2 Man 5 GlcNAc 2 . Both glucose residues could be removed in vitro by purified rat liver GII. It has also been reported that Nglycosylation sites of secretory proteins are underoccupied in alg3 mutants [12,13,[15][16][17]. Various studies have shown that the glucose residues on the lipid-linked oligosaccharide facilitate the transfer of the oligosaccharide to protein [1,18]. Nonglucosylated or partially glucosylated oligosaccharides can be transferred to protein, but with a reduced efficiency. In alg3 mutants of baker's yeast, the resulting Man 5 GlcNAc 2 lipid-linked glycan is not glucosylated efficiently [12]. Apparently, the 69 branch of the oligosaccharide is a major structural determinant in the specificity and activity of the Alg6p, dolichol-P-Glc:Man9GlcNAc2-PP-Dol glucosyl transferase, which is the first glucosyltransferase in the ER [16,34]. We anticipated this problem and avoided it by constitutively overexpressing the Y. lipolytica ALG6 gene. Indeed, overexpression of ALG6 largely remedied the defect in Nglycosylation site occupancy in the lipase secreted by the alg3 mutant. However, this complemented strain secreted proteins with more Man 5 GlcNAc 2 glucosylation, most likely because of the transfer of a larger fraction of nonglucosylated Man 5 GlcNAc 2 to proteins.
Remarkably and beneficially, Y. lipolytica Golgi glycosyltransferases does not seem to further modify the glycans upon disruption of the ALG3 gene. This was also reflected in the increased homogeneity of secreted LIP2 lipase on SDS-PAGE gels. Most likely, YlOch1p does not recognize the ER-type Man 5 GlcNAc 2 or its glucosylated derivates.
In contrast, N-glycans released from an alg3och1 mutant strain of P. pastoris contain the expected Hex 5 GlcNAc 2 structure, as well as large quantities of glycans of higher molecular weight ranging from Hex 6 GlcNAc 2 to Hex 12 GlcNAc 2 [35]. Upon treatment with a-1,2-mannosidase, the Man 5 GlcNAc 2 was converted to Man 3 -GlcNAc 2 , which is consistent with the alg3 Man5 structure. The other glycans, however, were mostly resistant to treatment with broad-specificity a-mannosidase. Amongst these, only Hex 6 Glc-NAc 2 was shown to contain glucose, which is consistent with a GlcMan 5 GlcNAc 2 structure [35]. The presence of larger structures implies the existence of P. pastoris Golgi glycosyltransferases capable of acting on these substantially truncated substrates. This is clearly different from the situation in Yarrowia. In a S. cerevisiae alg3sec18 mutant, a substantial proportion of the glycan chains on the model protein invertase were the mono-, diand triglucosylated Man 5 GlcNAc 2 structures [12,16,36].
In contrast, in the plant Arabidopsis thaliana, an alg3cgl mutant yielded Man 3-4 GlcNAc 2 glycans, which led to the hypothesis that an aberrant Man 5 GlcNAc 2 structure, once it is transferred to a protein, is trimmed by the Golgi a-1,2-mannosidase [37]. Similarly, analysis of whole cell extracts from the filamentous fungus A. niger algC knock-out (the ALG3 homologue) revealed the presence of Man 3-6 GlcNAc 2 N-glycans [30]. Moreover, proteins secreted by an alg3 mutant of the yeast Hansenula polymorpha contain almost no glucosylated glycans [38]: model glycoproteins contain predominantly Man 5 GlcNAc 2 . The less abundant Hex 6-8 GlcNAc 2 structures can be almost completely converted to Man 3 GlcNAc 2 by in vitro digests with a-1,2and a-1,6-mannosidases. Deletion of the endogenous OCH1 gene encoding the initiating a-1,6mannosyltransferase decreases the overall abundance of Hex 6-8 GlcNAc 2 structures and only a minor fraction of Hex 6 GlcNAc 2 remains. This Hex 6 GlcNAc 2 glycan quite likely contains a capping glucose residue [38].
The presence of glucose residues on the alg3 Man 5 GlcNAc 2 glycans implies either the existence of an endogenous glucosyltransferase or, more likely, insufficient activity of ER-resident GII, which normally cleaves both a1,3-linked glucose residues successively from Glc 2 Man 9 GlcNAc 2 . GII's substrate specificity includes the 69 pentamannosyl branch of its glucose-containing oligosaccharide substrates. Its activity seems to decrease with reduction of the number of mannoses on the 69 branch of the N-glycan substrate. Mammalian GII activity was several times higher with Glc 1-2 Man 9 GlcNAc 2 as substrate than with Glc 1-2 Man 7 GlcNAc 2 . Moreover, oligosaccharides lacking the four outermost mannose residues on the 69 branch were very poor substrates [39]. Similar results were obtained by other investigators [40][41][42][43]. More recently, it was found that the rate of GII-mediated trimming is specifically dependent on the presence of the a-1,2-linked mannose on the C-arm [44]. The b-subunit of GII contains a mannose-6-phosphate-homology (MRH) domain that recognizes carbohydrates and contributes to substrate recognition [45]. Sequence alignments indicated that all residues involved in mannose binding in the MRH domain are conserved in GII b, except for those that interact with the phosphate group. Indeed, there is evidence that the GII b-subunit plays a key role in enhancing the specific activity of the heterodimeric GII enzyme towards natural N-glycan substrates [28,29,[46][47][48][49].
From all the above observations, it can be concluded that, GII of Y. lipolytica is much more specific for its natural substrate than, for example, the GII of A. thaliana or A. niger. Here, we used this broader substrate specificity of A. niger GII to reduce the glucosylation of our YLA3-A6 strain.
The feasibility of our integrated system's engineering approach illustrates the current level of understanding of the N-glycosylation pathway's intricacies. We anticipate that this strain will find use in the structure-function analysis of N-glycan modifications in many settings, such as in the fine-tuning of biopharmaceutical protein Nglycans to particular therapeutic goals.

Strains, Reagents and Culture Conditions
Escherichia coli strains MC1061, TOP10, and DH5a were used for the amplification of recombinant plasmid DNA. T. brucei GII and mutanase tested as engineering approach. (A) The dual N-glycosylation system in T. brucei. Both Man 9 GlcNAc 2 and Man 5 GlcNAc 2 can be transferred to proteins. Next, these proteins are reglucosylated and deglucosylated in the folding cycle by glucosyltransferase and GII, respectively. (B) DSA-FACE analysis of reference N-glycans and N-glycans derived from strains engineered with T. brucei GII or treated with mutanase. A, Oligomaltose reference. B, N-glycans from RNaseB reference. C, N-glycans from the alg3 mutant strain overexpressing Alg6p. D-F, N-glycan from the alg3 mutant strain overexpressing Alg6p and engineered in different ways: D, engineered with T. brucei GII; E, engineered with T. brucei GII with HDEL tag; F, engineered with T. brucei GII with HDEL tag and pre-lip2 signal. G, N-glycans derived from the alg3 mutant strain overexpressing Alg6p treated with mutanase. doi:10.1371/journal.pone.0039976.g005 Yarrowia lipolytica MTLY60 (Table 1) [50] was used as parent strain. All yeast strains were cultured at 28uC. They were grown on YPD (20 g/L dextrose, 20 g/L bacto-peptone and 10 g/L yeast extract) or MM (1.7 g/L yeast nitrogen base (YNB) without amino acids and ammonium sulfate, 10 g/L glucose, 5 g/L NH 4 Cl, 50 mM K + /Na + phosphate buffer pH 6.8, and 7.7 g/L Complex Serum-free Medium (CSM)); for selection of Ura + and Leu + transformants, 7.7 g/L CSM -ura or CSM -leu was added instead of CSM.

Standard Genetic Techniques
For transformation of Y. lipolytica, competent cells were prepared as described [51]. Briefly, cells were pretreated with lithium acetate and incubated with the DNA to be transformed together with salmon sperm carrier DNA. PEG 4000 was added, and after a heat shock at 42uC, cells are plated on selective plates.
Genomic DNA was isolated using the MasterPure TM Yeast DNA Purification Kit according to the instructions of the manufacturer (Epicenter Biotechnologies). PCR amplification was performed in a volume of 50 mL containing 20 mM Tris-HCl pH 8.4, 50 mM KCl, different concentrations of MgCl 2 , 0.4 mM of dNTPs, 50 ng of template DNA, 50 pmol of primers, and 2.5 units of either Taq or Pfu DNA polymerase. Cycling conditions were as follows: denaturation at 94uC for 10 min followed by hot start at 80uC and 30 cycles of 94uC for 45 s, suitable annealing temperature for 45 s, and extension at 72uC for 1 min per kbp, followed by 10 min of final extension at 72uC.
DNA fragments in PCR reactions and those recovered from gels were purified using NucleoSpin extract II (Macherey-Nagel).

Vector Construction
Knocking out the ALG3 gene. We used a knock-out strategy that makes use of the Cre-lox recombination system, which facilitates efficient marker rescue [52]. The genomic region upstream of the ALG3 ORF (GenBank Accession No: XM_503488; Genolevures: YALI0E3190g) was amplified from genomic DNA of Y. lipolytica MTLY60 by PCR with primers ALG3Pfw and ALG3Prv (Table 2) using Taq polymerase (Invitrogen, Carlsbad, CA, USA). The overhanging A was removed with T4 DNA polymerase (Fermentas, Burlington, Ontario, Canada). The genomic region downstream of the ALG3 ORF was amplified from genomic DNA of Y. lipolytica MTLY60 by PCR with primers ALG3Tfw and ALG3Trv (Table 2) using Pfu DNA polymerase (Fermentas). The presence of overlapping primer sequences containing I-SceI restriction sites allowed the linking of the fragments by PCR with primers ALG3Pfw and ALG3Trv using Taq polymerase. This co-amplicon was then subcloned in pCR-2.1-TOPO-TA (Invitrogen, Carlsbad, CA, USA) and sequenced. It was then cloned between the NotI and PacI sites in a derivative of pBluescriptIISK (Stratagene, Cedar Creek, Texas, USA) to yield pBLUYLalg3PT. Next, the URA3 selection marker flanked by lox sites originating from pKS-LPR-URA3 [52] (a gift from J.M. Nicaud, INRA) was inserted in the introduced I-SceI site between upstream and downstream regions, yielding pYlalg3PUT. Similarly, pYlalg3PLT was constructed by exchanging the URA3 cassette in pYlalg3PUT with the LEU2 selection marker from pKS-LPR-LEU2 [52] by means of I-SceI digestion.
Cloning the ALG6 gene. The ORF (1725 bp) of ALG6 together with the 415-bp downstream region (GenBank Accession No: XM_502922; Genolevures: YALI0D17028g) was cloned from genomic DNA of Y. lipolytica MTLY60 by PCR with primers ALG6fw and ALG6rv (Table 2) using Pfu DNA polymerase. The amplified fragment was cloned in pCR-Blunt-II-TOPO (Invitrogen, Carlsbad, CA, USA) and sequenced. Next, it was cloned between the BamHI and AvrII sites of pYLHmA (pINA1291) [53], which contains the hp4d promoter [54] and the LIP2 terminator. It was then subcloned in the intermediate vector pBLUYLalg3PT in the unique ClaI and HindIII restriction sites present in the downstream region of ALG3. The URA3 selection marker flanked by lox sites, which was obtained from pKS-LPR-URA3, was inserted in the introduced I-SceI site between promoter and terminator fragments of the ALG3 gene. The resultant plasmid was named pYlalg3PUT-ALG6. Similarly, pYlalg3PLT-ALG6 was made by exchanging the URA3 cassette in pYlalg3PUT-ALG6 with the LEU2 selection marker from pKS-LPR-LEU2 be means of I-SceI digestion.
Cloning the GII alpha-subunit of Y. lipolytica with and without HDEL tag. The ORF (2766 bp) of the Y. lipolytica GII a-subunit gene (GenBank Accession No: XM_500574) was amplified from genomic DNA of Y. lipolytica MTLY60 by PCR with primers YlGlucIIafw and YlGlucIIarv ( Table 2) using Pfu DNA polymerase. The PCR fragment was cloned in pCR-Blunt-II-TOPO (Invitrogen, Carlsbad, CA, USA) and confirmed by Sanger sequencing. Next, it was cloned (BglII/BamHI and AvrII) under control of the hp4d promoter in pYLHmAX (pYLHmA carrying the URA3 selection marker) yielding pYLHmAXYlGIIa. To add the HDEL coding sequence to the ORFof GII a-subunt of Y. lipolytica, a PCR was performed on the obtained plasmid pYLHmAXYlGIIa with primers YlGlucIIafw and YlGlucIIaH-DELrv (Table 2), and the amplified fragment was cloned as described above for the version without HDEL tag.
Cloning the GII alpha-subunit of Trypanosoma brucei with and without HDEL tag. The ORF (2421 bp) of the GII a-subunit gene was amplified from genomic DNA of T. brucei (GenBank Accession No: AJ865333; a gift from Stijn Roge, Institute of Tropical Medicine, Antwerp) by PCR with primers TbGlucIIafw and TbGlucIIarv (Table 2) using Pfu DNA polymerase. The amplified fragment was cloned in pCR-Blunt-II-TOPO (Invitrogen, Carlsbad, CA, USA) and confirmed by sequencing. Next, it was subcloned BamHI-AvrII in pYLHmAX, which contains the hp4d promoter and the URA3 marker, yielding pYLHmAXTbGIIa. To add an HDEL tag to the T. brucei GII asubunit, PCR was performed on the obtained plasmid with primers TbGlucIIafw and TbGlucIIaHDELrv (Table 2), and the amplified fragment was cloned in the same way as without HDEL tag.
Cloning the GII beta-subunit of Y. lipolytica. The ORF (1288 bp) of the GII b-subunit gene was cloned from genomic DNA of Y. lipolytica MTLY60 (GenBank Accession No: XM_500467; Genolevures: YALI0B03652g) by PCR with primers YlGlucIIbfw and YlGlucIIbrv (Table 2) and Pfu DNA polymerase. Two other vectors (pYLHL and pYLTL) carrying the LEU2 selection marker were constructed for protein expression controlled by the hp4d or TEF promoter, respectively. Next, the ORF of Y. lipolytica GII b-subunit was cloned BamHI-AvrII in these vectors, yielding pYLHLYlGIIb and pYLTLYlGIIb.
Cloning the GII alpha-subunit of Aspergillus niger. cDNA for a fusion of the ORF of the a-subunit of A. niger GII and an HDEL tag, flanked by SnaBI and AvrII, was synthesized by Geneart AG (Regensburg, Germany). The sequence was codon-optimized for expression in Y. lipolytica. First, two intermediate vectors were constructed, pYLTUXL2pre and pYLHUXL2pre, by introducing the pre sequence of LIP2 in pYLHmAX and pYLTmAX. The latter was derived from pYLHmAX by replacing the hp4d promoter by the TEF promoter. The introduction of the pre sequence of LIP2 was performed by annealing two primers (Table 2) and cloning them BamHI-AvrII in pYLHmAX and pYLTmAX. The above-mentioned cDNA of the glucosidase a-subunit of A. niger flanked by SnaBI and AvrII was cloned in the corresponding restriction sites of pYLTUXL2pre and pYLHUXL2pre after SacII digestion + T4 DNA polymerase blunting and AvrII digestion. The resultant plasmids (pYLTUXL2preAnGlucIIa and pYLHUXL2-preAnGlucIIa, respectively) were confirmed by sequencing.
Cloning the GII beta-subunit of A. niger. The coding sequence for the b-subunit of A. niger GII flanked by Eco47III and AvrII restriction sites was synthesized by Geneart AG (Regensburg, Germany) as cDNA codon-optimized for expression in Y. lipolytica. Two intermediate vectors (pYLTLL2pre and pYLHLL2pre) were constructed by introducing the pre sequence of LIP2 in pYLTL and pYLHL, respectively, as described above. The resultant plasmids were named pYLTLL2pre and pYLHLL2pre. The above-mentioned synthesized cDNA was then cloned in the Eco47III and AvrII sites of these vectors by using SacII digestion + T4 DNA polymerase blunting and AvrII digestion. The resultant plasmids (pYLTLL2preAnGlucIIb and pYLHLL2preAnGlucIIb, respectively) were confirmed by sequencing.
Cloning the Trichoderma reesei a-1,2-mannosidase with HDEL tag. We used an expression plasmid derived from pYLTUXL2preManHDEL [32] by digestion with I-SceI followed by replacement of the URA3 selection marker with the hygromycin selection marker (obtained from pKS-LPR-HYG, a gift from J.M. Nicaud, INRA) [52]. The resultant plasmid, pYLTHygL2pre-ManHDEL, contains the T. reesei a-1,2-mannosidase coding sequence codon-optimized for Y. lipolytica, under control of a TEF promoter, preceded by the Y. lipolytica LIP2 pre signal sequence, and C-terminally tagged with an HDEL retrieval sequence.
Selection marker rescue. In all plasmids, the selection marker cassette is flanked by loxP and loxR sites to facilitate marker rescue by transient overexpression of the Cre recombinase. For overexpression of Cre recombinase, we used pRRQ2 (a gift from J.M. Nicaud, INRA) [52], which expresses the enzyme under control of the hp4d promoter and carries the LEU2 resistance gene.

Preparation of Mannoproteins, N-glycan Analysis and Exoglycosidase Digests
Yeast strains were inoculated and grown overnight in 10 mL of standard YPD medium in 50 mL Falcon tubes rotating at 250 rpm in a 28uC incubator. The cells were then pelleted at 4000 rpm in a cooled Eppendorf 5810R centrifuge. The supernatants were removed, and the cells were first washed with 2 mL of 0.9% NaCl solution followed by two washes with 2 mL of water and subsequently resuspended in 1.5 mL of 0.02 M sodium citrate pH 7 in an Eppendorf tube. After autoclaving for 90 min at 121uC, they were vortexed and the cellular debris was spun down. Then the supernatants were collected and the mannoproteins were precipitated overnight with four volumes of methanol at 4uC on a rotating wheel. After centrifugation, the pellets were allowed to dry and then dissolved in 50 mL of water.
The whole 50 mL of the cell wall protein solution was used to prepare N-glycans labeled with 8-aminopyrene-1.3.6-trisulphonic acid (APTS) according to a published method [55]. Then, fluorophore-assisted carbohydrate electrophoresis (FACE) was performed with an ABI 3130 DNA sequencer.
For the exoglycosidase digests, one tenth of the prepared APTSlabeled N-glycans was used. Exoglycosidase treatment of APTSlabeled glycans with Jack bean a-mannosidase (20 mU/digest, Sigma Biochemicals, Bornem, Belgium) or a-1,2-mannosidase (0.33 mg/digest, made in house) was performed overnight at 37uC in 50 mM ammonium acetate (pH 5.0). GII treatment of APTSlabeled glycans was performed with a purified rat liver mixture of alpha and beta (5 mU/mL, a gift from Dr. Terry Butters, Glycobiology Institute, Department of Biochemistry, Oxford, UK) [56]. Equal volumes of enzyme (in 80 mM triethylamine buffer, pH 7, containing 0.15 M NaCl and 10% glycerol) and sample were incubated together at 37uC overnight. The samples were then vacuum dried, resuspended in 10 mL of water, and analyzed on the ABI 3130 DNA sequencer. PNGaseF Treatment of Glycoproteins Proteins in the Yarrowia culture medium were precipitated with two volumes of ice-cold acetone. After incubation on ice for 20 min and centrifugation at 14,000 rpm for 5 min, the supernatant was removed and the protein pellet was resuspended in 100 mL of 50 mM Tris-HCl, pH 8. SDS and b-mercaptoethanol were added to a final concentration of 0.5% and 1%, respectively. Samples were incubated for 5 min at 100uC, after which G7 buffer (106 buffer, New England Biolabs), NP-40 (final concentration of 1%), complete protease inhibitor (Roche) and in-house produced PNGaseF (15 IUBMB milliunits) were added. After overnight incubation at 37uC, proteins were precipitated by the deoxycholate/trichloroacetic acid (DOC/TCA) procedure, resuspended in 26 Laemmli buffer, and analyzed by SDS-PAGE.

In vitro Digestion with Trichoderma Harzianum Mutanase
T. harzianum mutanase Novozyme 234, L1412 was obtained from Sigma-Aldrich Corporation, Spruce St., St. Louis, MO, USA. A stock solution of the enzyme (10 g/L) was prepared by dissolving 40 mg in 4 mL of 5 mM NH 4 Ac pH5 buffer. Five serial five-fold dilutions were made, and the final dilution (0.2 mL) was used to treat 0.5 mL of APTS-labeled N-glycans in a total volume of 10 mL buffered to a final concentration of 50 mM NH 4 Ac pH5. This reaction mixture was incubated overnight at 37uC and analyzed on an ABI 3130 DNA sequencer after desalting on a Sephadex G10 column [55].