Biosynthesis of GlcNAc-rich N - and O -glycans in the Golgi apparatus does not require the nucleotide sugar transporter SLC35A3

Nucleotide sugar transporters, encoded by the SLC35 gene family, deliver nucleotide sugars throughout the cell for vari-ous glycosyltransferase-catalyzed glycosylation reactions. GlcNAc, in the form of UDP-GlcNAc, and galactose, as UDP-Gal, are delivered into the Golgi apparatus by SLC35A3 and SLC35A2 transporters, respectively. However, although the UDP-Gal transporting activity of SLC35A2 has been clearly demonstrated, UDP-GlcNAc delivery by SLC35A3 is not fully understood. Therefore, we analyzed a panel of CHO, HEK293T, and HepG2 cell lines including WT cells, SLC35A2 knockouts, SLC35A3 knockouts, and double-knockout cells. Cells lacking SLC35A2 displayed significant changes in N - and O -glycan synthesis. However, in SLC35A3-knockout CHO cells, only limited changes were observed; GlcNAc was still incorporated into N -glycans, but complex type N -glycan branching was impaired, although UDP-GlcNAc transport into Golgi vesicles was not decreased. In SLC35A3-knockout HEK293T

The glycan moiety is formed by glycosyltransferases, the majority of whose catalytic centers face the lumen of the endoplasmic reticulum (ER) and Golgi apparatus. Nucleotide sugar transporters (NSTs) encoded by the SLC35 (solute carrier 35) gene family provide glycosyltransferases with activated sugars (nucleotide sugars), which serve as substrates for glycosylation reactions (2). NSTs also transport the nucleotide monophosphate byproduct of the glycosyltransferase reaction back to the cytoplasm. The ability of NSTs to form dimers or higher order oligomers (2) allows for multiple and simultaneous interactions between several NSTs, as well as between NSTs and glycosyltransferases (3)(4)(5)(6)(7). Initially, NSTs were demonstrated to be specific for the translocation of a single nucleotide sugar (8), but later multisubstrate transporters were reported (9)(10)(11). Mutations in genes encoding mammalian NSTs are associated with severe defects in development and physiology, demonstrating their requirement for proper functioning of multicellular organisms (12)(13)(14)(15)(16).
To date, the crystal structures of Saccharomyces cerevisiae GDP-Man transporter (Vrg4) in substrate-free, GDP-Manbound, and GMP-bound forms (17,18), mouse CMP-sialic acid (CMP-Sia) transporter in complex with CMP and CMP-Sia (19), and also the Zea mays analog of CMP-Sia transporter in complex with CMP (20) have been determined. The overall structure of NSTs revealed dimeric organization, the presence of 10 transmembrane a-helices connected by short loops, exposition of both N and C termini at the cytoplasmic side of the Golgi membrane, separate binding sites for the nucleotide and sugar, and a requirement for short-chain lipids.
GlcNAc is the most common among several monosaccharides used for synthesis of glycoconjugates, including all types of glycans, not only for extension of the synthesized oligosaccharide structures but importantly for generation of oligosaccharide branches. Golgi UDP-GlcNAc transporter (SLC35A3 and its homologs) has been characterized, i.e. in yeasts (21,22), parasites (23), plants (24), and mammals (25)(26)(27). Established data suggested that SLC35A3 is the key, if not the only NST responsible for UDP-GlcNAc transport into the Golgi lumen. This assumption was made based mostly on (i) phenotypic correction of Kluyveromyces lactis mutant deficient in Golgi transport of UDP-GlcNAc using cDNA encoding SLC35A3 from Madin-Darby canine kidney cells (25), (ii) phenotypic complementation of K. lactis mutant with mouse SLC35A3 transporter (28), and (iii) UDP-GlcNAc transport using mouse SLC35A3 reconstituted in S. cerevisiae mutant and in artificial proteoliposomes (28). It is worth noting, however, that the above-mentioned approaches were based on a heterologous yeast expression system producing different types of N-glycans as compared with mammalian cells and on proteoliposomes using the protein overexpressed in S. cerevisiae cells. In contrast, our gene silencing analysis of the SLC35A3 gene did not result in significant changes in N-glycans (4). Chinese hamster ovary (CHO), Lec8 (CHO mutant cells lacking functional SLC35A2 gene), and HeLa cells with significantly decreased SLC35A3 activity exhibited only a decreased amount of highly branched tri-and tetra-antennary complex type N-glycans, whereas monoantennary and diantennary ones remained unchanged or even were accumulated.
Another important monosaccharide, galactose (Gal), is transported into the Golgi apparatus by the UDP-Gal transporter (SLC35A2) (29)(30)(31)(32)(33)(34)(35)(36). Importantly, we demonstrated that biological function of UDP-GlcNAc and UDP-Gal transporters in glycosylation of proteins may be coupled. This hypothesis was based on the observations that (i) overexpression of UDP-GlcNAc transporter in mutant cells defective in UDP-Gal transporter partially restores galactosylation of N-glycans (37), (ii) chimeric proteins composed of regions derived from both transporters restore galactosylation in cells defective in UDP-Gal transporter (38,39), and (iii) both overexpressed and native transporters form complexes detected not only by fluorescence lifetime imaging microscopy-FRET or in situ proximity ligation assay analyses but importantly also by coimmunoprecipitation assay, suggesting not only their close localization but also strong interaction (3,7,40).
This study presents novel data that broaden our knowledge of glycosylation process occurring in the Golgi apparatus. Taking into consideration that siRNA knockdown did not cause a complete inactivation of mRNA encoding SLC35A3 (4), as well as cell-specific glycosylation pattern, in this study we performed knockout of the SLC35A3 gene in three mammalian cell lines using CRISPR-Cas9 technology. Again, no significant changes in N-glycan synthesis were observed. Moreover, the knockout cells also continued to produce GlcNAc-containing O-glycans. Interestingly, we observed not only cell-specific glycosylation pattern but also differences in UDP-GlcNAc transporting activity. These findings suggest that SLC35A3 may not be the key transporter of UDP-GlcNAc and/or that different mechanisms of UDP-GlcNAc transport into the Golgi apparatus should be taken into consideration. We constructed knockout cells lacking not only the SLC35A3 transporter but also both SLC35A3 and SLC35A2 proteins. This attempt allowed us to follow changes occurring in Nand Oglycans in different cell lines resulting not only from a single deletion of the respective transporter but also from a combined deficiency of the two highly homologous and functionally related SLC35A subfamily members, which is important for broadening our knowledge on the functional glycosylation complexes and potential involvement of their members in the glycosylation process.

Inactivation of the SLC35A2 and SLC35A3 genes
Taking into account the phenomenon of cell-specific glycosylation, a panel of CHO, HEK293T, and HepG2 cell lines consisting of (i) WT cells, (ii) single-knockout cells lacking SLC35A2, (iii) single-knockout cells lacking SLC35A3, and (iv) doubleknockout cells lacking both SLC35A2 (A2KO) and SLC35A3 (A3KO) were analyzed. To obtain cell lines deficient in particular proteins, we employed CRISPR-Cas9 or CRISPR-Cas9 double nickase technology. The absence of SLC35A2 protein was confirmed by Western blotting analysis using highly specific anti-SLC35A2 antibody (Fig. 1). A similar approach was impossible in the case of SLC35A3 because of the lack of specific antibody. Therefore, to confirm that inactivation of the corresponding genes was effective, genomic DNA and RNA were isolated from WT cells and from several clones of the knockout cells. PCR and RT-PCR with subsequent DNA sequencing showed that the changes we detected do not allow for production of functional full-length or truncated protein, confirming that the CRISPR-Cas9 strategy successfully inactivated SLC35A3 gene in HepG2, HepG2 A2KO, HEK293T, HEK293T A2KO, CHO, and Lec8 cells (Figs. 2 and 3 and Table 1).

Analysis of N-glycans
To characterize oligosaccharides produced by selected cell lines, first fluorescently labeled N-glycans were separated, and their N-glycan profiles compared. In contrast to differences clearly visible in the case of SLC35A2-deficient cells, no significant changes in N-glycan profiles between the WT and SLC35A3-deficient cells were observed (Fig. S1). Then 2-aminobenzamide (2-AB) derivatives were analyzed in detail using MALDI-TOF MS. If SLC35A3 was the main Golgi UDP-GlcNAc transporter, no complex type structures could be formed in cells lacking its activity. However, in the case of all single-knockout SLC35A3 cells, every complex type N-glycan typical of the parental cells could be detected . However, subtle quantitative differences could be observed. In the case of CHO cells, four tetra-antennary structures were considerably decreased (Fig. 4, m/z of ;2660, 3026, 3390, and 3754). This was less evident in the case of HEK293T cells (Fig. 5). Interestingly, the cells lacking both SLC35A2 and SLC35A3 displayed much more profound defects compared with the single-knockout cells. First of all, in the case of the double CHO and HEK293T knockout cells, a complete disappearance of a tetra-antennary, fucosylated structure could be observed (Figs. 4 and 5, m/z of ;2011). In the case of CHO cells lacking both SLC35A2 and SLC35A3, a triantennary, nonfucosylated structure (m/z of ;1663) was also absent from the N-glycan profile. In addition, a triantennary, fucosylated structure (m/z of ;1802) was completely lost in the case of HEK293T doubleknockout cells, resulting in a unique profile completely devoid of complex type N-glycans with more than two antennae, whereas in the case of CHO cells, the relative amount of this structure was substantially reduced. In the case of HepG2 knockout cells, however, we did not observe major changes in N-glycan profiles of SLC35A3-deficient cells compared with the WT cells (Fig. 6). Importantly, in contrast to CHO and HEK293T cell lines, no significant changes could be observed in the case of SLC35A3 knockout constructed in SLC35A2deficient cells.
To analyze mature N-glycans in secreted proteins, we employed a reporter glycoprotein assay. For this purpose, we engineered a modified secreted alkaline phosphatase (SEAP) and established an efficient overexpression and purification protocol. 1 PNGase-released and 2-AB-labeled N-glycans were separated, and N-glycan profiles were qualitatively compared in all cell lines examined in this study. No significant changes were observed between the WT and SLC35A3 knockout cells (Fig. S2). This procedure allowed us also to confirm no significant changes in sialylated N-glycans. Next, we carried out a more detailed analysis using MALDI-TOF MS in positive ion mode. N-Glycans released from SEAP secreted by SLC35A3 knockout cells were mainly of complex type, and no significant changes in GlcNAc-containing structures were observed when compared with the parental cells (Figs. 7-9). In the case of double CHO knockout cells, a complete disappearance of triantennary structures (m/z of ;1663 and ;1809) could be observed  when compared with Lec8 cells. N-Glycans released from SEAP produced by double HepG2 knockout cells were comparable with structures derived from HepG2 SLC35A2 knockout cells. In the case of single-knockout cells deficient in SLC35A3 a triantennary, a fully galactosylated structure could be observed (m/z of ;2148). This glycan was completely absent from the corresponding profile obtained for the WT cells, which contained only a monogalactosylated triantennary structure (m/z of ;1824). Surprisingly, in HEK293T SLC35A2 and double-knockout cells, we observed a substantial degree of galactosylation of SEAP-derived N-glycans. Double-knockout cells contained more galactosylated glycans than single-knockout SLC35A2 cells (five versus two glycans). All N-glycan structures detected by MALDI-TOF MS are listed in Table S2.

Analysis of O-glycans
To further examine involvement of SLC35A3 in protein glycosylation, we employed the cellular O-glycome reporter/ amplification strategy (41) to identify O-glycans produced by the WT and SLC35A3-deficient cells. Oligosaccharides were isolated, and their permethylated derivatives were analyzed using MALDI-TOF MS. In addition to O-glycan changes clearly visible in the case of the cells lacking SLC35A2, SLC35A3 deficiency did not prevent GlcNAc incorporation into O-glycans (Figs. [10][11][12]. MALDI-TOF MS spectra obtained for O-glycans synthesized by the SLC35A3-deficient HEK293T and HepG2 cells showed that all GlcNAc-containing structures were still produced (Figs. 10 and 11). Even O-glycan with two incorporated GlcNAc moieties (m/z of ; 1260) could still be detected. In SLC35A3-deficient HEK293T cells, the relative amounts of the two nonsialylated GlcNAc-containing structures (m/z of 1044.2 and 1218.2) were substantially decreased. However, this was not visible in the case of their more complex, sialylated biosynthetic derivatives (m/z of ;1406, ;1580, and ;1767). CHO cells produce only core 1type structures (Fig. 12) and therefore are not suitable for studying this phenomenon. All O-glycan structures detected by MALDI-TOF MS are listed in Table S3.

Transport assay
In a classical NST activity assay enriched, intact Golgi vesicles are incubated with radiolabeled NSTs and the label within the vesicles is mesured (8). However, this approach has several drawbacks. First of all, it requires enormous amounts of cells to obtain enough Golgi vesicles. Second, isolated vesicles are prone to damage during handling and storage. In addition, there is a high risk of sample contamination by the ER membranes. For all these reasons we employed an alternative approach to measure UDP-GlcNAc Golgi uptake, experimentally designed based on the procedure used for UDP-Gal uptake (42). This strategy relies on incubation of semipermeabilized cells with a radiolabeled nucleotide sugar and a membrane-permeable synthetic acceptor followed by a measurement of a cell-bound radioactivity. UDP-Gal uptake is expected to be   (4,37). We indeed observed a negligible UDP-Gal uptake in single-knockout cells lacking SLC35A2, as well as in double-knockout cells lacking both SLC35A2 and SLC35A3 compared with the WT cells ( Fig. 13). Next, we performed a similar assay with UDP-GlcNAc with the difference that in this case sugar incorporation into endogenous acceptors instead of a synthetic one was attempted. Surprisingly, in SLC35A3-deficient CHO cells, UDP-GlcNAc uptake did not decrease but was even higher than in the WT cells, which is in accordance with our gene silencing studies (4). In the case of HEK293T cells, however, SLC35A3 deficiency was accompanied by a statistically significant decrease in UDP-GlcNAc uptake, which was partially restored by knocking out SLC35A2 gene. Nevertheless, it must be emphasized that Golgi vesicles derived from SLC35A3-deficient HEK293T cells imported UDP-GlcNAc at a rate exceeding 25% of the WT, whereas in the case of SLC35A2-deficient HEK293T cells, UDP-Gal transport was virtually negligible. We did not examine UDP-GlcNAc transport into Golgi vesicles derived from HepG2 cells, because in their case the N-glycan profile changed upon SLC35A3 deficiency to a much lesser extent compared with CHO and HEK293T cells.
A summary of all the observed phenotypic effects is presented in Table 2. Previously, we showed decreased UDP-Gal transporting activity in CHO SLC35A3 knockdown cells as compared with CHO WT cells, whereas in Lec8 SLC35A3 knockdown cells, UDP-Gal transporting activity was comparable with that determined for Lec8 cells but lower compared with CHO WT cells (4). Here, CHO cells lacking SLC35A3 exhibit transporting activity similar to CHO WT cells. These differences can be explained, at least in part, by different methods used to determine transporting activity in CHO cells. In the method used here, we can observe structures galactosylated with b1,4-Gal, whereas in the method used previously, all types of galactosylated structures were subjected to analysis. The data obtained here for HEK293T cells are very similar. However, observations made for UDP-GlcNAc transporting activity are dissimilar in both cell lines, suggesting a cell-specific effect of glycosylation.

Analysis of gene expression
Searching for other UDP-GlcNAc transporters, we determined relative gene expression of potential candidates. In our previous study (43), we demonstrated that the first potentially strong candidate, SLC35B4, should be excluded. We showed that this protein is not involved in UDP-GlcNAc delivery for glycan synthesis because of its ER localization and the lack of glycophenotypic changes after the gene knockout. Nevertheless, this protein was further examined, together with SLC35A5, SLC35D1, and SLC35D2, in regard to transcript production. Analysis of mRNA encoding all examined potential candidates showed no dramatic changes between the WT and knockout cells (Fig. 14). Although we noticed an increase in mRNA encoding SLC35A5 in HepG2 cells lacking SLC35A3 and, at the same time, decreases in mRNAs encoding SLC35A5, SLC35B4, SLC35D1, and SLC35D2 in CHO cells lacking SLC35A3, the differences were rather subtle.

Discussion
The goal of this study was to gain a better understanding of the role played in glycosylation by SLC35A3, which was consid-ered for the long time the main Golgi UDP-GlcNAc transporter in mammals. GlcNAc plays a crucial role in numerous posttranslational modifications of macromolecules including Nand O-glycosylation. Without the efficient delivery of UDP-GlcNAc into the Golgi lumen complex type N-glycans could not be synthesized. GlcNAc attachment is a prerequisite for galactosylation of N-glycans, which in turn enables their sialylation. Moreover, GlcNAc incorporation into N-glycans allows for their branching, i.e. formation of antennae. In addition, elongation of the a1,3-mannose arm of N-glycan with GlcNAc is considered a prerequisite for the core fucosylation. Multiantennary galactosylated, sialylated, and fucosylated structures SLC35A3 knockout does not affect GlcNAc-rich glycans are the fully mature forms of complex type N-glycans. These species are most capable of eliciting biological responses within the body. Our data obtained from an siRNA approach showed that the limited expression of the SLC35A3 gene causes a reduction in N-glycan branching but does not prevent GlcNAc from being incorporated into mono-and biantennary complex type structures (4). However, one cannot exclude the possibility that the residual expression of the SLC35A3 gene could have provided an intraluminal pool of UDP-GlcNAc sufficient for the biosynthesis of some complex type N-glycan species. We felt that there is a need for obtaining more convincing data on the contribution of SLC35A3 in N-glycosylation, and therefore we applied the CRISPR-Cas9 strategy to completely disable SLC35A3 expression. We also employed a larger repertoire of cell lines to avoid making conclusions from the effects that might turn out to be cell-specific. Moreover, for the first time we examined the contribution of SLC35A3 in O-glycosylation. Overall, this study has been greatly extended and refined when compared with the already published research.
Apart from the knockout cells deficient only in SLC35A3, in this work we also examined double-knockout cells in which both SLC35A2 and SLC35A3 genes were inactivated. We did it for several reasons. First of all, complex type N-glycans produced by the SLC35A2-deficient cells have simplified structures comparing with the WT cells because of the absence of galactose and sialic acid residues. In addition, because galactosylation and branching of complex type N-glycans are concurrent reactions (44), one might expect that turning off galactosylation would increase the subset of multiantennary structures and therefore provide a more suitable model for studying SLC35A3 function. Finally, we also demonstrated that SLC35A2 and SLC35A3 interact with each other and appear to be functionally related (3,45), which raises a possibility that SLC35A2 could partially compensate for SLC35A3 deficiency in the single-knockout cells. For all the above reasons, we assumed that the putative effects of the SLC35A3 deficiency may be more pronounced and easier to demonstrate in cells that have already been deprived of SLC35A2. To obtain CHO cells lacking both NSTs, we knocked out the SLC35A3 gene in a well-established Lec8 mutant cells (34,46). In the case of HepG2 and HEK293T cell lines, we sequentially inactivated both genes, because their SLC35A2-deficient variants were not available.
The specific decrease in the amount of tetra-antennary species in CHO and HEK293T cell lines suggests that in the absence of SLC35A3, the most compromised enzyme is Mgat5. Interestingly, we found that both SLC35A2 and SLC35A3 interact with mannoside N-acetylglucosaminyltransferases (Mgats) SLC35A3 knockout does not affect GlcNAc-rich glycans including Mgat5 as demonstrated by the fluorescence lifetime imaging microscopy-FRET and in situ proximity ligation assay analyses (5). Recently, several ternary complexes formed with the involvement of some Mgats, and the selected SLC35A subfamily members have been demonstrated (7). According to this study, Mgat5 was the only Mgat not being capable of interacting with other Mgats. We therefore hypothesize that SLC35A2 and SLC35A3 may play a crucial role in preventing displacement of Mgat5 from the putative multiprotein complex that possibly regulates N-glycosylation.
The relatively mild effect of SLC35A3 knockout on N-glycans produced by CHO and HEK293T cells was remarkably potentiated by the lack of SLC35A2. This appears rather surprising given the fact that GlcNAc addition is a prerequisite for Gal attachment. To our knowledge, it is not a common phenomenon that an early step of glycosylation is affected by a step occurring later within the pathway. This may suggest that in addition to transporting activity, SLC35A2 could play also a modulatory role in the multiprotein complex, because its deficiency impairs both galactosylation and GlcNAc attachment. Importantly, SLC35A2 and SLC35A3 are known to tightly interact with each other (3), and SLC35A3 overexpression in SLC35A2-deficient cells was shown to partially restore galactosylation of N-glycans (37). The current work supports the idea that these two proteins are functionally related and suggests their collaboration in the multiprotein complex performing the biosynthesis of complex type N-glycans. However, it has to be emphasized that in the case of the HepG2 cell line, knocking out SLC35A2 did not potentiate the outcome of SLC35A3 deficiency, so this phenomenon also appears to be cell-specific.
Profiles of SEAP-derived N-glycans differed from the ones obtained for the corresponding whole cell lysates in several aspects. The most novel and striking finding was the firm presence of Gal in N-glycans derived from SEAP secreted by the double and SLC35A2 single-knockout HEK293T cells. This may mean that SLC35A2 is not the sole UDP-Gal transporter, at least in this cell line. In fact, one of our previous studies dem-onstrated some residual galactosylation of N-glycans synthesized by the cells deficient in SLC35A2 activity (32). Alternatively, it could be concluded that cellular and secretory proteins are not evenly glycosylated by the knockout cells.
Another interesting finding is the increase of galactosylation of N-glycans in single SLC35A3 knockout HepG2 cells compared with the WT cells as well as double-knockout HEK293T cells compared with single-knockout SLC35A2 knockout cells. It appears that in these cell lines, the lack of SLC35A3 potentiates galactosylation of complex type N-glycans.
Apart from N-glycans, in mammalian cells GlcNAc is often incorporated into O-glycans. This monosaccharide can be found in core 2-, 3-, 4-, and 6-type O-linked structures. However, the contribution of SLC35A3 in O-glycan biosynthesis has SLC35A3 knockout does not affect GlcNAc-rich glycans never been explored. HepG2 and HEK293T cells produce a variety of core 2-type O-glycans. These species are synthesized with the involvement of the b-1,3-galactosyl-O-glycosyl-glycoprotein b-1,6-N-acetylglucosaminyltransferase (C2GnT). Its affinity for UDP-GlcNAc has been established at 1.9 mM (46) and more recently at 0.36 mM (47). Both values significantly exceed the K m value postulated for Mgat1 (0.04 mM) (48). One therefore might expect that C2GnT activity would become severely compromised by the limited UDP-GlcNAc availability in the Golgi lumen of SLC35A3-deficient cells, and the range of the produced structures would be narrowed down to core 1. The double-knockout cells are, however, expected to produce Tn antigen only, because galactose addition to GalNAc by the glycoprotein-GalNAc 3-b-galactosyltransferase 1 (C1GalT1) is a prerequisite for C2GnT action. To our surprise, O-glycans synthesized by the SLC35A3-deficient HepG2 and HEK293T cells contained all GlcNAc-containing structures, including the one possessing two incorporated GlcNAc moieties. This finding strongly argues against the idea of SLC35A3 being a general and unfocused UDP-GlcNAc transporter in the Golgi membrane of mammalian cells.
A knockout of the gene encoding a UDP-GlcNAc transporter should typically result in a significantly decreased uptake of this nucleotide sugar into the Golgi vesicles. However, in the case of CHO cells deficient in SLC35A3, UDP-GlcNAc transport was even slightly increased, whereas in the case of HEK293T cells, it was reduced to ;25%. This relatively high residual UDP-GlcNAc import into the Golgi apparatus in the latter cells is in line with the persistence of GlcNAc-containing Nand O-glycan structures and provides another argument against the concept of SLC35A3 being the sole UDP-GlcNAc Golgi transporter in mammals. CHO and HEK293T cell lines clearly responded differentially to SLC35A3 deficiency in terms of UDP-GlcNAc uptake. Perhaps these differences result from a generally lower transport rate in HEK293T comparing with CHO cells. Moreover, in the case of HEK239T cells, GlcNAc is incorporated both in N-and O-glycans, whereas in the case of CHO cells, GlcNAc is present only in N-glycans. In addition, it has to be emphasized that the overall uptake of both UDP-Gal and UDP-GlcNAc was 1 order of magnitude lower in HEK293T than in CHO cells. Moreover, in the WT CHO cell line, UDP-Gal transport was higher than UDP-GlcNAc transport, whereas in the case of the WT HEK293T cell line, this was completely reversed. All these discrepancies make the obtained results difficult to compare between the two cell line panels.
Although it has been reported so far that SLC35A3 is the main Golgi UDP-GlcNAc transporter in mammals, our findings do not support this idea. This is in sharp contrast to the other well-characterized, apparently monospecific NSTs related with N-glycosylation, i.e. SLC35A1, SLC35A2, and SLC35C1, for whom the postulated substrate specificity is fully supported by the phenotypes of the mutation-bearing cell lines and human individuals (Table S1). In our opinion, unlike GDPfucose, UDP-Gal, and CMP-Sia, the main supplier of UDP-GlcNAc for N-glycan biosynthesis has yet to be identified. Obviously, SLC35A3 activity is not required for O-glycosylation. Despite a complete inactivation of the SLC35A3 gene, we were not able to obtain the expected GlcNAc-deficient Nand O-glycophenotypes. There appears to be a puzzling discrepancy between the experimentally determined UDP-GlcNAc transporting activity of SLC35A3 and its native function. However, this is not the only case in which the substrate specificity of an NST cannot be easily translated into its biological role. For instance, although SLC35A2 is considered the sole UDP-

SLC35A3 knockout does not affect GlcNAc-rich glycans
Gal transporter in mammals, its deficiency does not completely abolish galactosylation of N-glycans (Ref. 32 and this work). The recently resolved three-dimensional structure of the CMP-Sia transporter from maize clearly demonstrates a trans-porting capability of this protein (20). However, it is widely accepted that plants do not incorporate sialic acid into glycoconjugates (49); therefore the physiological significance of such finding is obscure. Another example is the surprising lack of  SLC35A3 knockout does not affect GlcNAc-rich glycans glycophenotypic effects of the putative UDP-GlcNAc/UDPxylose transporter deficiency (43). The remaining members of the SLC35A subfamily, namely SLC35A4 and SLC35A5, also appear to be members of multiprotein complex with modulatory rather than transporting properties (40,50). For SLC35A5 a decrease in UDP-GlcNAc uptake has been demonstrated in cells deficient in this protein, although it was not reflected in the corresponding glycophenotype (50). Nevertheless, SLC35A5 may be considered as an accessory (or alternative) Golgi UDP-GlcNAc transporter. Gene expression analysis demonstrated in this study suggests that, at least at the mRNA level, no induction of SLC35A5 gene expression caused by SLC35A3 gene knockout could be observed. Another Golgi-resident NST suspected of UDP-GlcNAc delivery is SLC35D2 (51). We hypothesize that some members of the SLC35 family could have evolved toward novel biological functions while retaining transporting activity. Nevertheless, it should be realized that within cells many factors can affect transport activity and modulate substrate specificity of the putative NSTs. UDP-GlcNAc transporting activity has been assigned to SLC35A3 based on experimental approaches using conditions that poorly reflect the dynamics and complexity of the mammalian secretory pathway. We feel that the assignment of the substrate specificity to a putative NST should be made with the support of the findings resulting from a knockdown or a knockout of the corresponding gene in a cell line of interest followed by an extensive characterization of the resulting glycoconjugates.
In this work we developed and characterized single and double-knockout cell lines lacking SLC35A3 or both SLC35A2 and SLC35A3, respectively. The study involved three cell lines selected based on differences in their Nand O-glycan repertoires, e.g. the ability to produce GlcNAc-containing O-glycans, polyLacNAc extensions, and highly branched structures. The phenotypic effects of SLC35A3 depletion differed depending on the cell line used. Regardless of these differences, it can be boldly assumed that if SLC35A3 was the only UDP-GlcNAc transporter in mammals, its deficiency should disable all GlcNAc transferases including enzymes displaying high affinities toward UDP-GlcNAc. However, we clearly demonstrated that SLC35A3 deficiency does not prevent GlcNAc incorporation into Nand O-glycans. Moreover, branching of complex type N-glycans is affected to a different extent depending on the cell line used, and this phenomenon is strongly potentiated by the SLC35A2 deficiency. The lack of SLC35A3 appears to affect some specific steps of N-glycosylation, which suggests it could be a member of a catalytically more efficient multiprotein complex rather than function as a single transporter. This appears to be particularly supported by the phenotype of CHO cells lacking SLC35A3, which display perturbations in N-glycan branching despite showing no decrease in UDP-GlcNAc transport. This suggests that the role of SLC35A3 played in glycosylation is different from what has been expected so far. We believe this might be related with the complexes formed by SLC35A2, SLC35A3, and Mgats. Specific and very restricted defects in glycosylation suggest that glycoproteins produced as a result of a given step of glycosylation may not be properly committed to the next one, because consecutively performed reactions become uncoupled because of the altered composition of the corresponding multiprotein complexes. However, further studies are required to address this idea.
Our findings clearly indicate that there are some gaps in the understanding of glycosylation. First of all, the biological role of SLC35A3 has to be specified. This task might not be possible to fulfill until the architecture of glycosylation-related multiprotein complexes is fully deciphered. It may also require resolving the three-dimensional structure of SLC35A3. Second, the ways by which UDP-GlcNAc is delivered into the Golgi lumen of mammalian cells are to be clarified.
HepG2 and HEK293T WT cells were transfected with a pool of three plasmids (containing three different SLC35A3-specific gRNAs), according to the manufacturer's instructions (Santa Cruz Biotechnology). Preselection was performed in MEM (for HepG2 cells) or DMEM (for HEK293T cells) complete medium supplemented with 1 mg/ml puromycin (InvivoGen). After 3 weeks of preselection, clones overexpressing GFP and RFP (the second and third selection markers in the employed CRISPR-Cas9 system) were isolated. CHO and Lec8 cells were transfected with a mixture of double nickase plasmids according to the manufacturer's instructions (Santa Cruz Biotechnology) and selected according to the manufacturer recommendations. For further analyses, clones overexpressing GFP (the second Figure 14. Analysis of gene expression of selected nucleotide sugar transporters. Relative mRNA levels were determined in HepG2 (A), HEK293T (B), and CHO (C) WT and SLC35A3 knockout (A3KO) cells. GAPDH gene was used as a reference gene. *, p , 0.05; ***, p , 0.01; ****, p , 0.001; ns, difference not statistically significant. All samples were run in two independent experiments in triplicate for the target and reference genes.
SLC35A3 knockout does not affect GlcNAc-rich glycans selection marker in the employed CRISPR-Cas9-NIC system) were isolated.

Evaluation of the SLC35A3 and SLC35A2 knockout efficiency
To confirm the SLC35A3 gene knockout in HepG2 and HEK293T cells, genomic DNA was isolated, and PCR was performed. PCR products were separated in 1.6% (w/v) agarose gel, visualized with ethidium bromide, and sequenced as described previously (40). Confirmation of the SLC35A3 gene knockout in CHO and Lec8 cells was performed at the RNA (RT-PCR) level as reported previously (40,43) using primers listed in Table 3. To confirm the SLC35A2 gene knockout in all cell lines studied, we used the highly specific antibody against SLC35A2 protein (Merck) and Western blotting approach (4).

Western blotting
Cell lysates were subjected to SDS-PAGE using 10% polyacrylamide gels and transferred onto nitrocellulose (Whatman) as described previously (4). Briefly, after the transfer, the membrane was cut above the 52-kDa and below the 140-kDa prestained marker bands (Thermo Fisher Scientific). The upper part of the membrane was subjected to reaction with a rabbit anti-EEA1 antibody (1:2000; Santa Cruz), and the lower part of the membrane was subjected to reaction with a rabbit anti-SLC35A2 antibody (1:1000), both followed by anti-rabbit antibody conjugated with horseradish peroxidase (1:10,000; Sigma-Aldrich). Chemiluminescence reaction products were visualized as described previously (4).

Transport assay
The cells were harvested at 70-80% confluency and permeabilized using hypotonic conditions as described by others (52). UDP-Gal transport assay was performed as reported by Ng et al. (42). UDP-GlcNAc transport assay was carried out at 37°C using 0.75 mM UDP-3 H-GlcNAc (30 Ci/mmol; GlcNAc-6-3 H (N); PerkinElmer). Time and concentration of unlabeled UDP-GlcNAc were optimized for each cell line separately. Finally, for CHO cells the reaction was carried out for 10 min in the presence of 0.25 mM unlabeled UDP-GlcNAc (Sigma-Aldrich). For HEK293T the reaction was carried out for 40 min without unlabeled substrate. The reaction was terminated by addition of 10 volumes of cold transport buffer. As a negative control, the reaction was conducted in the presence of 0.1% Triton X-100, which disrupts Golgi integrity. After the transport assay, the cells were spun down (3 min, 2000 rpm), and the resulting cell pellets were sonicated in the presence of 70% ethanol. After two rounds of ethanol extraction, the pellets were dried and dissolved in 5% SDS, and the radioactivity was measured. For each cell line transport assays were performed in three independent biological replicates, using different cell preparations and performed on different days. For statistical analysis, a oneway analysis of variance test with a Bonferroni's post hoc test and ranking method was employed. All analyses were performed with GraphPad Prism (GraphPad Software, La Jolla, CA). Statistical significance was assigned to p values ,0.05.

Analysis of fluorescently labeled N-and O-glycans
Total cellular N-glycans were enzymatically released, isolated, fluorescently labeled with 2-AB, and separated as reported previously (37). Briefly, the cells were collected and lysed using the CelLytic M detergent solution (Sigma). Proteins from whole cell lysates were precipitated with cold acetone. Precipitates were dissolved in denaturation buffer and digested using peptide:Nglycosidase F (New England Biolabs). Released N-glycans were purified with Hypercarb Hypersep 200 mg SPE columns (Thermo Scientific) and labeled with 2-AB. Prior to HPLC analysis, the samples were purified on blotting filter paper (Whatman). Prior to MALDI-TOF MS analysis, 2-AB-labeled glycans were desialylated by using a2-3,6,8,9-neuraminidase A (New England Biolabs) and purified using Hypercarb Hypersep with 10-200-ml tips (Thermo Scientific). The last steps were also applied to purification of SEAP-derived N-glycans.
To analyze O-glycans, the procedure reported by Kudelka et al. (41) was adopted, as reported recently (50,51). Briefly, the cells were incubated with 50 mM peracetylated O-glycan precursor (Ac 3 GalNAcBn) in medium containing 5% serum and grown for 72 h. Then the medium was collected and centrifuged (1000 3 g), and the supernatant was subjected to glycan extraction procedure. Isolated N-glycans and permethylated O-glycans were subsequently analyzed using MALDI-TOF MS (4).
MALDI-TOF-MS experiments were performed on an Axima Performance TOF spectrometer (Shimadzu Biotech, Manchester, UK), equipped with a nitrogen laser (337 nm). The pulsed extraction ion source accelerated the ions to a kinetic energy of 20 keV. The data were obtained in a positive-ion linear mode. The calibration of the linear mode analysis was done using PEG in mass range up to 5000 Da. The accuracy of the product ion calibration is ;1.5 Da. The mass calibration was conducted based on the average masses. The samples were dissolved in 20% acetonitrile in water. As a matrix, 2,5-dihydroxybenzoic acid (20 mg/ml) dissolved in 20 mM sodium acetate in 20% acetonitrile in water was used. The sample and matrix were combined at a 1:1 ratio. The resulting solution (1 ml) was spotted on a 384-well MALDI-TOF plate, followed by evaporation of the solvent at ambient temperature without any assistance. Each mass spectrum was accumulated from at least 200 laser shots and processed by Biotech Launchpad version 2.9.1 program (Shimadzu).

Secreted glycoprotein reporter assay
To overexpress and purify reporter glycoprotein, we constructed genetically modified vector, enabling improved ex- SLC35A3 knockout does not affect GlcNAc-rich glycans pression and purification of SEAP. 1 Briefly, psiTEST plasmid (Invivogen) was used as a template to add His 6 at the C terminus of the SEAP. In addition, using a Q5 site-directed mutagenesis kit (New England Biolabs), a new N-glycosylation site at 278amino acid position was introduced. The cells were transiently transfected with the modified plasmid using FuGENE 6 transfection reagent (Promega) at 3:1 transfection reagent:DNA ratio and grown in respective serum-containing medium. Media containing SEAP-His 6 were collected 24 h after transfection by centrifugation (500 3 g, 5 min) and examined with QUANTI-Blue phosphatase reagent according to the manufacturer's protocol (Invivogen). Recombinant SEAP-His 6 was purified using nickelnitrilotriacetic acid magnetic agarose beads (Jena Bioscience) according to the modified protocol. 3 Purified SEAP-His 6 obtained after elution with 23 glycoprotein denaturing buffer (New England Biolabs) was subjected to peptide:N-glycosidase F (New England Biolabs) treatment. Released N-glycans were isolated, labeled, and separated as described previously (37).
Quantitative reverse transcriptase-PCR RNA was isolated from 1.2 3 10 6 cells using the ExtractMe total RNA kit (Blirt). Isolated RNA was treated with DNase I and purified using the Clean-Up RNA concentrator kit (A&A Biotechnology). RNA integrity was verified using a spectrophotometric method. Reverse transcription was carried out using 1 mg of RNA and the SensiFAST cDNA synthesis kit (Bioline). Real-time PCR was carried out using the RT HS-PCR mix SYBR A kit (A&A Biotechnology) and the LightCycler 96 System (Roche). The amplification reaction included initial denaturation at 95°C for 2 min, 40 cycles of denaturation at 95°C for 15 s, primer annealing at 60°C for 30 s, and extension at 72°C for 30 s. The melting curves were analyzed to monitor the quality of PCR products. The relative quantification of expression of respective genes was determined compared with the human or hamster GAPDH gene as a reference, using the DDC q method and LightCycler 96 System software (Roche). All samples were run in two independent experiments in triplicate for the target and reference genes. No-template controls were included on each reaction plate to check for contamination. All primers used in this study are listed in Table 4.

Data availability
The MS data has been deposited in the GlycoPost under ID GPST000080. Some unpublished data are available upon request from the corresponding author (E-mail address: mariusz.olczak@uwr.edu.pl).