The Cytoplasmic, Transmembrane, and Stem Regions of Glycosyltransferases Specify Their in Vivo Functional Sublocalization and Stability in the Golgi*

We provide evidence for the presence of targeting signals in the cytoplasmic, transmembrane, and stem (CTS) regions of Golgi glycosyltransferases that mediate sorting of their intracellular catalytic activity into different functional subcompartmental areas of the Golgi. We have constructed chimeras of human α1,3-fucosyltransferase VI (FT6) by replacement of its CTS region with those of late and early acting Golgi glycosyltransferases and have stably coexpressed these constructs in BHK-21 cells together with the secretory reporter glycoprotein human β-trace protein. The sialyl Lewis X:Lewis X ratios detected in β-trace protein indicate that the CTS regions of the early acting GlcNAc-transferases I (GnT-I) and III (GnT-III) specify backward targeting of the FT6 catalytic domain, whereas the CTS region of the late acting human α1,3-fucosyltransferase VII (FT7) causes forward targeting of the FT6in vivo activity in the biosynthetic glycosylation pathway. The analysis of the in vivo functional activity of nine different CTS chimeras toward β-trace protein allowed for a mapping of the CTS donor glycosyltransferases within the Golgi/trans-Golgi network: GnT-I < (ST6Gal I, ST3Gal III) < GnT-III < ST8Sia IV < GalT-I < (FT3, FT6) < ST3Gal IV < FT7. The sensitivity or resistance of the donor glycosyltransferases toward intracellular proteolysis is transferred to the chimeric enzymes together with their CTS regions. Apparently, there are at least three different signals contained in the CTS regions of glycosyltransferases mediating: first, their Golgi retention; second, their targeting to specific in vivofunctional areas; and third, their susceptibility toward intracellular proteolysis as a tool for the regulation of the intracellular turnover.

We provide evidence for the presence of targeting signals in the cytoplasmic, transmembrane, and stem (CTS) regions of Golgi glycosyltransferases that mediate sorting of their intracellular catalytic activity into different functional subcompartmental areas of the Golgi. We have constructed chimeras of human ␣1,3-fucosyltransferase VI (FT6) by replacement of its CTS region with those of late and early acting Golgi glycosyltransferases and have stably coexpressed these constructs in BHK-21 cells together with the secretory reporter glycoprotein human ␤-trace protein. The sialyl Lewis X:Lewis X ratios detected in ␤-trace protein indicate that the CTS regions of the early acting GlcNAc-transferases I (GnT-I) and III (GnT-III) specify backward targeting of the FT6 catalytic domain, whereas the CTS region of the late acting human ␣1,3-fucosyltransferase VII (FT7) causes forward targeting of the FT6 in vivo activity in the biosynthetic glycosylation pathway. The analysis of the in vivo functional activity of nine different CTS chimeras toward ␤-trace protein allowed for a mapping of the CTS donor glycosyltransferases within the Golgi/trans-Golgi network: GnT-I < (ST6Gal I, ST3Gal III) < GnT-III < ST8Sia IV < GalT-I < (FT3, FT6) < ST3Gal IV < FT7. The sensitivity or resistance of the donor glycosyltransferases toward intracellular proteolysis is transferred to the chimeric enzymes together with their CTS regions. Apparently, there are at least three different signals contained in the CTS regions of glycosyltransferases mediating: first, their Golgi retention; second, their targeting to specific in vivo functional areas; and third, their susceptibility toward intracellular proteolysis as a tool for the regulation of the intracellular turnover.
The assembly of protein-or lipid-linked oligosaccharides is mediated by the reactions of a series of glycosidases and glycosyltransferases that localize in the subcompartments of the secretory pathway of mammalian cells (1). According to the current consensus, the enzymes should be arranged in a sequential manner within the Golgi stacks. The control mechanisms that underlie the distribution of glycosyltransferases into different Golgi subcompartments are not understood. Some key enzymes like ␣-mannosidase II and GnT-I 1 have been localized in the medial Golgi and trans-Golgi, whereas several terminal glycosyltransferases (GalT-I, ST6Gal I, ST3Gal III, FT5, FT6) have been localized in the trans-Golgi/ TGN (2)(3)(4)(5)(6)(7). Evidence is provided in the literature that most glycosyltransferases show an overlapping distribution into more than only one morphological defined subcompartment (e.g. GnT-I has been localized to the medial and trans-Golgi, and GalT-I and ST6Gal I have been localized to the trans-Golgi as well as the TGN (2,4)). The transmembrane region as well as the flanking domains of type II Golgi resident glycosyltransferases have been identified to maintain Golgi retention (8 -10). The bilayer thickness model for Golgi retention of glycosyltransferases (11,12) postulates that the length of the transmembrane region of transferases mediates Golgi retention. A second hypothesis (3,13,14) proposes a disulfide-linked homo-/ hetero-oligomerization of the enzymes to function as a Golgi retention signal by preventing the large complexes from being delivered to secretory vesicles and ongoing transport to the plasma membrane. Neither model provides sufficient information about the mechanisms that control the in vivo functional organization of the different members of the glycosyltransferase families and how their sequential arrangement within different subcompartments might be accomplished. Immunochemical localization techniques lack the sensitivity to resolve in detail the distribution of, for example, the many late acting Golgi glycosyltransferases in the functional network of the trans-Golgi/TGN (10). A further complication with immunodetection of enzymes in defined subcompartments results from the migration of the newly synthesized membrane-bound glycosyltransferases from the endoplasmic reticulum through the compartments of the secretory pathway until they arrive at their final destination in individual functional Golgi stacks. It has been shown that glycosyltransferases themselves undergo complex-type N-glycosylation including terminal sialylation (15)(16)(17). In addition, there are several reports describing different levels of intracellular proteolytically cleaved forms of certain glycosyltransferases that might change under different physiological conditions in different cells (18 -23). A major problem of immunohistochemical methods is that they do not provide any information concerning the in vivo functional activity of the detected enzyme species. Assuming a defined in vivo acceptor substrate specificity for glycosyltransferases, their sequential action and possibly also their sequential distribution along the secretory pathway should be reflected in the final oligosaccharide structure of the biosynthetic products of a cell (e.g. in a secretory glycoprotein). Therefore, the structural characterization of a reporter glycoprotein expressed at a constant level from cells transfected with new glycosyltransferase genes should allow to identify the position of the newly introduced enzyme within the biosynthetic reaction sequence of the host cell.
We have recently analyzed in detail the in vivo biosynthetic activity of the human ␣1,3/4-fucosyltransferases III-VII (FT3-FT7) in BHK-21 cells (17) by stable coexpression of each individual enzyme together with human ␤-TP, which is decorated exclusively with diantennary complex-type N-glycans (17, 24 -26). We found that each human ␣1,3/4-fucosyltransferase is characterized in vivo by the synthesis of an individual ratio of sLe x :Le x , with FT7 forming exclusively sLe x and FT4 preponderantly (90%) Le x , whereas FT6 expression results in a 1.1:1 mixture of sLe x and Le x motifs in the oligosaccharides of the coexpressed reporter glycoprotein ␤-TP. The in vitro specificity data of the enzymes clearly support the exclusive sLe x -forming specificity of the FT7 catalytic domain and the Le x -forming specificity of the FT4 catalytic domain. Consequently, in order to get access to its Gal(␤134)GlcNAc-R substrate, FT4 should be localized in a cellular subcompartment before ␣2,3-sialylation occurs, since, according to all data available, the human ␣2,3-sialyltransferase ST3Gal III does not transfer NeuAc to Le x (compare Fig. 1). Likewise, FT7 action strictly depends on the proper supply with ␣2,3-sialylated acceptors. FT7 should either colocalize with ST3Gal III in the same functional area, or, more preferably, should reside in a later subcompartment in order to get access to high acceptor substrate concentrations. The human FT6 catalytic domain recognizes both, in vivo and in vitro, the sialylated or unsialylated acceptor motifs with a high efficiency (17,27); therefore, this enzyme should either colocalize with ST3Gal III, resulting in competition for the common Gal(␤134)GlcNAc-R substrate, or should have a broader subcompartmental distribution. We have also shown that a variant of FT6 constructed by replacement of its CTS region with the signal peptide sequence of human interleukin-2 is efficiently secreted from cells but does not show in vivo functional activity when expressed at a total activity level comparable with wt-FT6 cells (17). This result as well as reports by other groups suggesting the transmembrane and flanking regions of several glycosyltransferases (ST6Gal I, GalT-I, GnT-I, and ␣1,2-fucosyltransferase) as playing an important role in their Golgi retention (12, 28 -40) prompted us to investigate the properties of glycosyltransferase CTS regions in the in vivo functional targeting of the human FT6 catalytic domain to different biosynthetically active Golgi subcompartments. If localizing to early Golgi compartments, the FT6 cat-alytic domain would be expected to encounter low levels of sialylated N-glycans and preferentially would transfer Fuc to Gal(␤134)GlcNAc-R, resulting in increased Le x synthesis detectable in the secreted product. Likewise, its targeting to a later compartment would be expected to lead to preferential formation of sLe x motifs by the enzyme from the availability of already ␣2,3-sialylated precursor substrates. In the present report, we have fused CTS regions of different donor glycosyltransferases to the N terminus of the FT6 catalytic domain and have stably expressed the constructs in BHK-21 cells together with human ␤-TP, resulting in expression levels comparable with those in wt-FT6 cells. The characterization of the oligosaccharides attached to the secreted reporter glycoprotein should also allow the identification of any possible interference with the integrity of the cellular glycosylation pathway that could have resulted from the genetic engineering procedure.
Molecular Cloning of Glycosyltransferases-The cDNAs encoding the human ␣1,3/4-fucosyltransferases FT3, FT6, and FT7 as well as the human ST6Gal I were those from previous publications (17,25). Plasmids containing the cDNAs of GnT-I (41) and GnT-III (42) were kindly provided by Professor Geyer (University of Giessen, Giessen, Germany) and Professor Taniguchi (Osaka University Medical School, Osaka, Japan), respectively. A human GalT-I cDNA (43) was cloned from reverse-transcribed HL-60 mRNA as described (17) by using the Expand High Fidelity PCR system (Roche Molecular Biochemicals) and the primers (upper/lower) 5Ј-AAG ATG AGG CTT CGG GAG CCG CTC/5Ј-CTA GCT CGG TGT CCC GAT GTC CAC (35 cycles of denaturation, 15 s, 94°C, annealing, 20 s, 45°C, and extension, 120 s, 72°C). The cDNAs of three BHK-21 cell sialyltransferases were cloned by PCR using primers homologous to human ST3Gal III (44), human ST3Gal IV (45), and Chinese hamster ovary cell ST8Sia IV (46). BHK-21 cell mRNA isolation and cDNA synthesis was essentially as described above for HL-60 cells. The PCR was performed as above using the following primer pairs (upper/lower): 5Ј-AG ATG GGA CTC TTG GTA TTT GT/5Ј-TCA GAT GCC ACT GCT TAG ATC AGT GAT (ST3Gal III), 5Ј-AAC ATG GTC AGC AAG TCC CGC T/5Ј-GGT CAG AAG GAC GTG AGG TTC (ST3Gal IV), and 5Ј-TTA TAC CAA GAG AAG GTG CC/5Ј-GAT CCT TCA ATA TGT GCT TTA TT (ST8Sia IV) and 35 cycles with 15 s at 94°C, 20 s at 45°C/52°C/50°C, respectively, and 120 s at 72°C. The homology of the BHK-21 cDNA sequences to the human sequences was found to be 91% for ST3Gal III, 87% for ST3Gal IV, and 91% for ST8Sia IV. The PCR products were cloned into the vector pCR3.1 (Invitrogen) according to the manufacturer's instructions and were subsequently used as templates for the generation of CTS mutants of human FT6 as detailed below.
Construction of Chimeric FT6 Mutants-Chimeric constructs of FT6 were generated by fusing the CTS regions of nine different mammalian Golgi glycosyltransferases (51-126 aa) to the N terminus of the FT6 catalytic domain (compare Fig. 2 and Table I  The in vivo specificity of human FT3-FT7 toward N-glycoproteins has been reported previously (17). According to current concepts, ␣2,3-sialylation by ST3Gal III does not occur on already ␣1,3-fucosylated N-acetyllactosamine antennae. *, It is noteworthy that human FT3 has a high preference for type I motifs and therefore is considered a Lewis A/sialyl Lewis A enzyme (16,17,27).
Coexpression of FT6 Chimeras and ␤-TP in BHK-21 Cells-BHK-21 cells stably expressing human ␤-TP and CTS variants of FT6 were generated with the calcium phosphate precipitation method and selected with G418-sulfate as described (17). In most cases, cell lines showing a similar ␤-TP expression levels in Western blots and similar FT6 in vitro activity in cellular extracts compared with the well characterized wt-FT6 cell line (17) were used for the further characterization of the reporter glycoprotein. For this, the cells were cultivated for 2-3 weeks in Dulbecco's modified Eagle's medium containing alternatingly 0 or 2% fetal bovine serum, respectively, and about 0.5 mg of recombinant human ␤-TP was purified from 500 -1000 ml of culture supernatants by immunoaffinity chromatography as described (17).

SDS-Polyacrylamide Gel Electrophoresis and Western Blot
Analysis-SDS-polyacrylamide gel electrophoresis was performed according to Laemmli (47) using 12.5 and 3% acrylamide in the resolution and stacking gels, respectively. For Western blot analysis, proteins were transferred to Immobilon NC membranes (Millipore Corp.) by using a TransBlot™ SD transfer cell (Bio-Rad). The membrane was blocked for 1 h with Tris-buffered saline containing 10% horse serum and 3% bovine serum albumin and was incubated overnight with rabbit anti-␤-TP antiserum in blocking buffer at a 1:1000 dilution. The second antibody, goat anti-rabbit immunoglobulin coupled to horseradish peroxidase, was used at a 1:500 dilution. The blots were developed with Tris-buffered saline containing 0.5 g/liter 4-chloro-1-naphtol solubilized in methanol and 0.2% H 2 O 2 . Immunodetection of secreted FT6 variants was performed essentially as described for ␤-TP using a rabbit antiserum raised against the human FT6 peptide R 125 RQGQRWIWFSM-ESPSHCWQLK following immunoaffinity purification on the peptide antigen coupled to Affi-Gel 15 (Bio-Rad).
Characterization of Endoproteolytically Cleaved FT6 Chimeras-Cell culture supernatants of stable cell lines expressing CTS variants of FT6 were analyzed for secreted forms of the enzymes by an in vitro ␣1,3fucosyltransferase assay as detailed previously for wt-FT6 cells (17). The secreted chimeric enzymes were partially purified by affinity chromatography on a GDP-Fractogel column and characterized by Western blotting. The N terminus of the secreted form of wt-FT6 was determined by gas phase sequencing of the protein following transfer onto an Immobilon-P (Millipore) membrane.
Isolation and Characterization of the N-Glycans of the Reporter Glycoprotein ␤-TP-Purified ␤-TP was reduced, carboxamidomethylated, and digested with trypsin, and the free reducing N-glycans were obtained from reverse phase high pressure liquid chromatography-purified glycopeptides by peptide-N 4 -(N-acetyl-␤-D-glucosaminyl)asparagine amidase F digestion as detailed earlier (24). The released oligosaccharide material was subsequently analyzed by HPAE-PAD using conditions identical to those that were applied for the characterization of ␤-TP glycans from wt-FT6 cells (17). Identification and quantitation of the ␤-TP N-glycans was achieved by comparison with the elution profile of ␤-TP oligosaccharides from wt-FT6 cells, since this material contained all possible ␣1,3-fucosylated diantennary structures as was revealed in our previous investigation by using methylation analysis and mass spectrometry of individual oligosaccharide fractions (17). Similarly to wt-FT6 cells, more than 90% of the oligosaccharides from CTS variant cells were of the diantennary N-acetyllactosamine type, differing only in their content and the distribution of ␣2,3-linked NeuAc and ␣1,3-linked Fuc residues.

Construction and Expression of CTS Variants Containing the Human FT6 Catalytic Domain
The design of human FT6 CTS-variants is based on our previous finding that the catalytic domain of FT6 contains the in vivo specificity to transfer Fuc in ␣1,3-linkage to GlcNAc in both sialylated and unsialylated type II N-acetyllactosamine oligosaccharide chains of coexpressed human ␤-TP with a similar efficiency (17). The transmembrane domain as well as flanking regions of FT6 are required for its in vivo functional activity, since an engineered truncated form of the enzyme lacking the first 51 aa, when fused to the interleukin-2 signal peptide, is efficiently secreted from BHK-21 cells and does not fucosylate any oligosaccharide of coexpressed ␤-TP when expressed at an intracellular enzyme activity level comparable with wt-FT6 cells (17). As shown in Fig. 2 and Table I, CTSregions from donor glycosyltransferases of different length were fused to the N terminus of the human FT6 catalytic domain. The precise location of the catalytic domains of many glycosyltransferases is unknown (10). In analogy to the data reported for truncated forms of the highly homologous human FT3 (48), the catalytic domain of human FT6 was defined by the shortest C-terminal part that showed no loss of enzyme activity in vitro and contained the minimum sequence starting from Pro-62 ( . . . PLILLWTW . . . ), with the exception of ft7-FT6 and ft3-FT6, where the FT6 sequence starts with Trp-67 or Ile-92, respectively. Thus, the new CTS regions introduced into the chimeric FT6 variants contained the entire cytoplasmic and transmembrane domain of the corresponding donor glycosyltransferase as well as the stem region or parts thereof replacing 51-91 aa of the N terminus of wt-FT6. In the case of st6-FT6 and galt-FT6, the stem region part included also the proteolytic cleavage sites that have been identified for rat ST6Gal I (19) and human GalT-I (20).
Stable BHK-21 cell lines coexpressing wt-FT6 or chimeric FT6 variants together with constant levels of human ␤-TP were generated essentially as described previously for fucosyltransferase-expressing cells (17). An almost identical ␤-TP expression of all cell lines (0.5-1 g/10 6 confluent cells/48 h) was verified by Western blotting analysis of supernatants and by comparing the signal intensity with that of different concentrations of a known purified standard ␤-TP preparation from recombinant BHK-21 cells (25), using a monoclonal antibody raised against human ␤-TP (26, 49). About 0.5 mg of the reporter glycoprotein ␤-TP was quantitatively recovered from 500 -1000 ml of culture supernatant from each cell line by quantitative immunoaffinity purification (25,26), and detailed N-glycan mapping of each ␤-TP preparation was performed as described (17,25).

In Vitro FT6 Activity Levels in Stably Transfected BHK-21 Cells and Culture Supernatants
For the chimeric gnt1-, st6-, gnt3-, pst-, galt-, ft3-and ft7-FT6 constructs, stably transfected cells were obtained with intracellular enzyme activity very similar to wt-FT6 cells (Table II) using the type II Gal(␤134)GlcNAc-O-(CH 2 ) 8 -COOCH 3 acceptor substrate along with GDP-Fuc (17). For several chimeras, considerable amounts of FT6 activity were released into the culture supernatant (Table II). Intracellular proteolytic cleavage of fucosyltransferases has been reported to occur in both, recombinant host cells (6,16,17) as well as in human tumor cell lines and tissues (23). The proteolytic cleavage product from wt-FT6 cells was purified from the medium by affinity chromatography and was subjected to N-terminal sequencing. Three N termini were identified indicating cleavage at the C-terminal end of the putative transmembrane domain of FT6, with the largest resulting sequence being H 2 NY 33 -LRVSQDDPTVYPNGS . . . , and two further sequences lacking 1 or 3 additional amino acids, respectively. The secreted products of several chimeric enzymes were analyzed by Western blotting (Fig. 3), using FT6 peptide-specific antibodies. From the comparison of the intensity of the immunoreactive protein bands and the corresponding FT6 activity measured in cell culture supernatants, we conclude that the specific ␣1,3-fucosyltransferase activity of the CTS variants is very similar to that of secreted wt-FT6. The sizes of the soluble forms of gnt1-FT6, st3-FT6, and pst-FT6 indicate that cleavage must have occurred close to the C-terminal end of the putative transmembrane domains similarly as shown above for wt-FT6. Similar cleavage at the C-terminal end of the transmembrane domain has also been reported for human ␤1,4-GalNAc-transferase (50), whereas cleavage of rat ST6Gal I and human GalT-I was found at positions 37 and 34 of their putative luminal domains, respectively (19,20). The apparent mass of secreted st6-FT6 is about 2 kDa smaller than the value calculated for a full-length luminal domain, which might be due to cleavage in a region homologous to rat ST6Gal I. Secreted galt-FT6, however, is about 6 -8 kDa larger than the calculated mass of its luminal domain; most likely, this is due to N-glycosylation in the stem region and some additional posttranslational modifications oc-curring in the trans-Golgi/TGN as has been suggested in the case of a cell surface-targeted bovine GalT-I variant (51). Interestingly, no cleavage of the ft7-FT6 chimera was detected, which is in agreement with our previous observation that human FT7 is also resistant to intracellular proteolysis when stably expressed from BHK-21 cells (17). Similarly, no FT6 activity was released into the supernatants of stable cells transfected with the gnt3-FT6 construct. Taken together, our data suggest that the proteolytic cleavage characteristics of CTS donor glycosyltransferases have been transferred to the chimeric enzymes together with the CTS regions.

CTS Chimeras of FT6 Exhibit Altered in Vivo Functional Activity toward Unsialylated and Sialylated N-Acetyllactosamine Motifs
Functional Activity of FT6 Variants Containing the CTS Regions of GnT-I, GnT-III, and GalT-I-In the biosynthetic pathway of N-linked oligosaccharides, enzymes like GnT-I, GnT-III, and GalT-I must or should act before FT6. Therefore, we investigated if the CTS-regions of these glycosyltransferases would indeed lead to a targeting of the FT6 catalytic domain to functionally earlier Golgi compartments. We have previously shown that all asialo, mono, and disialo forms of diantennary complex N-acetyllactosamine-type glycans with or without ␣1,3-Fuc attached to ␤-TP can be resolved and quantitated by HPAE-PAD (17). As depicted in Table III, the total N-glycans of ␤-TP secreted by wt-FT6 cells consist of 17% asialo, 42% monosialo, and 42% disialo diantennary structures containing either one (11%) or two (41%) ␣1,3-linked Fuc (compare also HPAE-PAD elution profiles in Fig. 4), resulting in a ratio of 1.1:1 of the sLe x and Le x motifs on the oligosacharide antennae (Fig. 5). By contrast, the pattern of oligosaccharides found in ␤-TP secreted by gnt1-FT6 cells showed a large decrease of ␣1,3-fucosylated di-and monosialylated structures (Table III and Fig. 4) and also a lower overall peripheral fucosylation of only 20% of total N-linked oligosaccharides compared with 52% in the case of wt-FT6 cells. Consequently, ␤-TP N-glycans from gnt1-FT6 cells have a significantly lower sLe x : Le x ratio of 1:4.3 (Fig. 5). Since the cellular FT6 activity associated with gnt1-FT6 cells is even higher than observed with wt-FT6 cells (Table II) when measured with the Gal (␤134)-GlcNAc-R substrate, we conclude that the FT6 expression level does not account for the reduced fucosylation of ␤-TP in gnt1-FT6 cells. In fact, it appears that the CTS region of human GnT-I causes the FT6 catalytic domain to reside in an intracellular biosynthetic compartment where it is largely excluded from access to its Gal(␤134)GlcNAc-R substrate and, even more pronounced, to the sialylated form thereof. This view is strongly supported by the detection of almost no sLe x in the disialo oligosaccharide fraction (see Table III). This differs sig-  nificantly from the situation observed for the in vivo fucosylation specificity of wt-FT6, where 35% of the total diantennary ␤-TP N-glycans contain the sLe x motif (Table III). These data are best explained by the in vivo localization of the gnt1-FT6 chimera in an early Golgi subcompartment, presumably the same compartment where also GnT-I resides (2)(3)(4)12).
Further support for a different intracellular functional targeting of the FT6 catalytic domain by means of newly transferred CTS regions was obtained from the glycosylation analysis of ␤-TP secreted by gnt3-FT6 cells. GnT-III has not been immunolocalized so far, but from the current understanding of the N-glycosylation pathway, GnT-III should act soon after GnT-I has initiated complex-type N-glycan biosynthesis. We found that gnt3-FT6 cells secrete ␤-TP with a higher degree of fucosylation (43%) compared with gnt1-FT6 cells and also larger amounts of ␣1,3-fucosylated mono-and disialo glycans, resulting in an increased sLe x :Le x ratio of 1:2.6 (compare Figs. 4 and 5). Finally, ␤-TP oligosaccharides from galt-FT6 cells show a slightly decreased overall fucosylation (44%) and a slightly decreased sLe x :Le x ratio (1:1.5) when compared with wt-FT6 cells. In summary, our data lead us to conclude that the fusion of the CTS region from early acting glycosyltransferases to the catalytic domain of the trans-Golgi/TGN enzyme FT6 results in the targeting of the chimeras into early in vivo functional compartments that are presumably equivalent to those occupied by the wt-CTS donor glycosyltransferases. The sequential decrease in sLe x antennae of ␤-TP oligosaccharides from wt-FT6 over galt-FT6 and gnt3-FT6 to gnt1-FT6 suggests a retention of the FT6 catalytic domain in functionally earlier compartments of the biosynthetic glycosylation pathway caused by the properties of the donor CTS regions.
Functional Activity of FT6 Variants Containing CTS Regions of Terminal Glycosyltransferases-In a second set of experiments, we tested the possibility that a donor CTS region can also be used in the forward targeting of the in vivo functional activity of the FT6 catalytic domain. However, there are no data available from the literature that distinguish glycosyltransferase localization within different subcompartments of the TGN. According to common view, all terminal glycosyltransferases should localize in the trans-Golgi/TGN, as has been shown by immunolocalization for ST3Gal III (7, 52), ST6Gal I (4, 29, 53-55), FT5 (5), and FT6 (6). We have compared the targeting properties of the CTS regions of FT3 and FT7, since our previous results have shown that in vivo FT3 synthesizes preponderantly and FT7 exclusively sLe x structures. As has been already mentioned in the Introduction, in order to get access to favorable high amounts of its ␣2,3-sialylated acceptor substrate, FT7 should preferably localize in a  Fig. 4 and Table III).

TABLE III
In vivo fucosylation of ␤-TP N-glycans by stably transfected FT6 chimeras The reporter glycoprotein ␤-TP was isolated from the culture supernatants of each cell line, and the diantennary oligosaccharides (Ͼ90% of total) were quantitated by HPAE-PAD analysis. 0 Fuc, 1 Fuc, and 2 Fuc represent a content of zero, one or two ␣1,3-linked Fuc, respectively, in addition to proximal, ␣1,6-linked Fuc that was invariantly found in all structures.  FT6 and chimeric FT6 variants. A, wt-FT6; B-E, chimeric FT6 variants containing CTS regions from early and late acting glycosyltransferases. The symbols indicate proximally fucosylated, diantennary oligosaccharide structures modified with ␣2,3-linked NeuAc (S) and/or ␣1,3-linked Fuc (Lex). All peaks were identified also by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry as described previously (17). functional subcompartment later than ST3Gal III. However, in view of the strict substrate specificity of FT7 for ␣2,3-sialylated glycans in vitro (56) and in vivo (17), a spatial separation of the catalytic activity of ST3Gal III and FT7 is not a prerequisite for the FT7 bioactivity in vivo. The fucosylation characteristics of ␤-TP N-glycans resulting from stable expression of ft7-FT6 in BHK-21 cells are included in Table III and Fig. 4. These data support our hypothesis that wt-FT7 should localize later in the glycosylation pathway than wt-FT6, since the ␤-TP oligosaccharides from ft7-FT6 cells show a significant increase in sLe x motifs over Le x motifs (1.8:1) when compared with those from wt-FT6 cells expressing similar intracellular FT6 activity (Fig.  5). However, a complete conversion of the in vivo catalytic activity of FT6 into an FT7 specificity is obviously not achieved. The possibility cannot be excluded that the detected Le x motifs have at least partially been generated by a FT6 in vivo activity during transport of the chimera to the final biosynthetic functional compartment specified by the FT7 CTS region. This is corroborated by the results of stable cell lines expressing different amounts of intracellular FT6 activity (Table IV). A ft7-FT6 cell line expressing only 35% ␣1,3-fucosylated ␤-TP due to reduced intracellular FT6 activity was found to contain significantly increased amounts of sLe x motifs. Interestingly, ft7-FT6 cells synthesize almost equal amounts of ␣1,3-monofucosylated (10%) and ␣1,3-difucosylated (13%) disialo oligosaccharides (Table III) very similar as detected from wt-FT7 cells (11% in both cases (17)). Thus, the FT7 CTS-region appears to confer forward targeting of the FT6 catalytic domain and, therefore, different from that of the wt-FT6 CTS-region, has the properties to concentrate the in vivo enzyme activity later in the biosynthetic subcompartment, where it encounters higher concentrations of the NeuAc(␣233)Gal(␤134)GlcNAc-R acceptor substrate.

FIG. 4. HPAE-PAD analysis of native N-linked oligosaccharides from ␤-TP coexpressed in BHK-21 cells together with wt-
In contrast to ft7-FT6 cells, the fucosylation pattern of ␤-TP glycans from ft3-FT6 cells is almost identical to that obtained from wt-FT6 cells and shows only a slightly reduced overall fucosylation (Tables II and III and Fig. 5). This suggests an identical or at least a considerably overlapping in vivo functional localization of FT3 and FT6. From this result, it can be concluded that the preferential synthesis of sialylated Le x structures by FT3 in vivo is an intrinsic property of its catalytic domain and not due to a localization in compartments later than FT6. It should be emphasized that FT3 in fact is a Lewis A enzyme (16) and that the low in vivo activity of this enzyme with type II acceptors is probably only due to overexpression of FT3 in the recombinant cell line (17).
Mapping of the in Vivo Functional Targeting Properties of Sialyltransferase CTS Regions-In the biosynthesis of the Lewis-type carbohydrate structures, the different members of the sialyl-and fucosyltransferases are considered to play an important regulatory role. Therefore, it is of interest to understand the regulation and subcompartmental functional distribution of these enzymes involved in the assembly of terminal oligosaccharides motifs in glycoproteins. We analyzed FT6 variants containing the CTS regions of ST3Gal III, ST3Gal IV, ST6Gal I, or ST8Sia IV (compare Table I) in order to map the targeting properties of their donor CTS regions compared with that of wt-FT6. The fucosylation characteristics of ␤-TP coexpressed in our st3-FT6(I) and st3-FT6(II) cell lines (Tables III  and IV) indicates that st3-FT6 is localized earlier than wt-FT6. Although the intracellular activities of both cell lines differ from those of the cells listed in Table II, it can be expected that average expression of st3-FT6 would lead to a sLe x :Le x ratio intermediate between the ratios of 1:3 and 1:4.7 that were found for the low and high expressing cell lines, respectively. In the case of st4-FT6 cells, the in vitro activity measured in cellular extracts was only 1 ⁄30 compared with wt-FT6 cells (see Table IV). Remarkably, however, 19% of total ␤-TP N-glycans from st4-FT6 cells had acquired ␣1,3-linked Fuc, which is even higher than the value detected in st3-FT6(I) cells (14%). In contrast, st4-FT6 cells showed an ␤-TP oligosaccharide pattern significantly different from st3-FT6(I) cells (Table III and Fig.  6), resulting in a sLe x :Le x ratio of 2.3:1. This observation led us to conclude that ST3Gal IV localizes in a later subcompartmental area than ST3Gal III. Irrespective of the low intracellular FT6 activity measured in st4-FT6 cells, we would tentatively assign st4-FT6 a subcompartmental localization later than wt-FT6.
The detection of sLe x and Le x antennae in a ratio of 1:4.4 in ␤-TP N-glycans from st6-FT6 cells suggests that st6-FT6 is localized earlier than wt-FT6 and presumably even earlier than st3-FT6. This result corroborates the expected colocalization of ST6Gal I and ST3Gal III suggested previously (25). When compared with st6-FT6 cells, pst-FT6 cells showed similar FT6 activity and degree of ␣1,3-fucosylation of total ␤-TP glycans (Table II). The ratio of sLe x :Le x , however, was found to be significantly increased to 1:2 (compare Table III). Based on this result, one would localize ST8Sia IV to a later functional subcompartment than ST3Gal III, but earlier than wt-FT6 can act on its substrate.
Proposed Sequence of the in Vivo Functional Localization of Glycosyltransferases-From the in vivo functional activity of the FT6 CTS variants with N-linked oligosaccharides of coexpressed ␤-TP, we propose an intracellular sublocalization of the corresponding wt-CTS-donor glycosyltransferases according to a sequential arrangement in the following order: GnT-I Ͻ The data published for the immunolocalization of GnT-I and GalT-I (medial and trans-Golgi) as well as the terminal sialyl-and fucosyltransferases (trans-Golgi/TGN) also suggest their sequential but overlapping distribution along the biosynthetic glycosylation pathway (4). However, to our knowledge, a differential distribution of sialyl-or fucosyltransferases within the late trans-Golgi/TGN has not been identified. Our results show that the CTS regions of mammalian glycosyltransferases mediate backward or forward targeting of the catalytic domain of FT6 into different in vivo functional subcompartments of BHK cells. The results from the present work are compatible with the view that the CTS regions of the hybrid glycosyltransferases contain the signals that localize the FT6 catalytic domain to the proper biosynthetic functional compartments of the Golgi, where also the CTS-donor glycosyltransferases localize.

DISCUSSION
Different secretory N-glycoproteins synthesized by a given cell presumably must pass the same members of the oligosaccharide processing machinery en route through the Golgi but may acquire a different polypeptide-specific modification that is detectable in the final product. For example, human erythropoietin is decorated with preponderantly tetraantennary oligosaccharides with 1-3 N-acetyllactosamine repeats, and human antithrombin III is modified by a mixture of di-and triantennary glycans and some tetraantennary chains, whereas human ␤-TP contains only highly sialylated diantennary N-acetyllactosamine-type structures when expressed from BHK-21 or Chinese hamster ovary cells (17,25,57,58). This indicates that in a given host cell the final oligosaccharide structures on a glycoprotein are largely determined by the polypeptide itself, and thus, the glycosylation of a product reflects the accessibility of the catalytic domains of the Golgi membrane-anchored glycosyltransferases to the glycosylation domain of the protein (59). In the experimental approach used here, the secretory reporter glycoprotein ␤-TP produced from cells transfected with the different CTS variants encounters the same catalytic domain of FT6. Therefore, it is conceivable that the different ␣1,3-fucosylation and ␣2,3-sialylation characteristics observed for ␤-TP N-glycans from the variant cells is brought about by the anchoring of the FT6 catalytic domain in functionally defined subcompartmental Golgi areas, which is specifically mediated by each of the donor CTS regions fused to the enzyme. According to the general concept of a sequential organization of the intracellular glycosylation pathway in eukaryotic cells, glycosyltransferases that act early in protein-linked oligosaccharide biosynthesis are located in the medial/ trans-Golgi (e.g. GnT-I), whereas the late glycosyltransferases that are involved in the terminal decoration of oligosaccharide motifs (e.g. ␣2,3/6-sialyltransferases and ␣1,3/4-fucosyltransferases) are localizing later within the trans-Golgi/TGN, respectively. This has been experimentally verified by immunolocalization of some of the enzymes in the pertinent subcompartments of cells (28,29,31,38,53,60). However, immunochemical techniques such as high resolution laser scanning immunofluorescence microscopy and immunolabeling of cryosections of cells have also revealed overlapping localization of transferases, e.g. GnT-I in the medial Golgi and trans-Golgi, GalT-I, ST3Gal III, ST6Gal I, and FT6 in the trans-Golgi and TGN (2, 4, 6, 7). It should be emphasized that immunolocalization techniques cannot be expected to detect subtle differences in the subcompartmental arrangement of the vast majority of the glycosyltransferases that reside within the trans-Golgi/TGN. Most importantly, these methods do not provide any information about the in vivo functional activities of the detected enzyme molecules (8 -10).
All recent studies concerning the specific Golgi retention and thus prevention of the forward movement to the plasma membrane of Golgi glycosyltransferases (ST6Gal I, GalT-I, and GnT-I) indicate that this is mediated by signals contained in the three N-terminal polypeptide domains of the transferases, the cytoplasmic, transmembrane, and stem regions (9, 10, 29, 31-33, 38, 40). Fukuda and co-workers (40) reported a decrease of core 2 O-glycan synthesis in Chinese hamster ovary cells transfected with the core 2 ␤1,6-GlcNAc-transferase where the CTS region was replaced by amino acids 1-70 of human ST6Gal I when compared with cells transfected with the wildtype core 2 GlcNAc-transferase. Since the wild-type enzyme appears to be a cis-Golgi enzyme, these authors concluded that the ST6Gal I CTS region localized the chimera into the trans-Golgi. We have previously shown that a mutant form of the human FT6 containing the human interleukin-2 signal peptide sequence in place of its CTS-region is efficiently secreted but does not show in vivo functional activity toward N-glycans of cosecreted human ␤-TP. Only when an 80-fold higher expression level of the truncated FT6 variant is achieved, small amounts of peripheral fucosylation is detectable in secreted ␤-TP (17). However, in contrast to full-length FT6-expressing cells, the majority of the diantennary oligosaccharides from the cells expressing the truncated secreted FT6 acquire only a single ␣1,3-linked Fuc. This result then supports the view of the importance of the CTS region not only for retention in the Golgi but also for a proper in vivo function of glycosyltransferases. We have previously shown that in vivo human FT7 synthesizes only sLe x structures, whereas FT4 forms preponderantly the Le x motif, and cells transfected with FT6 synthesize ␤-TP with a roughly 1:1 mixture of sLe x and Le x structures. The increasing amount of Le x glycans formed is paralleled by a concomitant decrease in the overall sialylation of ␤-TP, since the BHK cell endogenous ST3Gal III does not act on Le x structures (17). Therefore, the substrate specificity of wt-FT6 toward both neutral and ␣2,3-sialylated oligosaccharides appears to be an intrinsic property of its catalytic domain, since an enzyme form lacking the CTS region also acts on both substrates in vitro (17,27). From our results obtained with the gnt1-FT6 variant, we conclude that the GnT-I CTS region prevents the forward movement of the enzyme to the late Golgi subcompartment, where wt-FT6 exerts its biosynthetic activity and, therefore, gnt1-FT6 has access to only limited amounts of the Gal(␤134)GlcNAc-R substrate. This result is corroborated by our observation that, with a gnt1-FT6 cell line expressing 1 ⁄10 of the wt-FT6 activity, no peripheral fucosylation of ␤-TP oligosaccharides was observed. This in vivo activity for gnt1-FT6 then is in agreement with the subcompartmental localization reported for GnT-I and GalT-I. Two-thirds of GnT-I has been immunolocalized in the medial Golgi of HeLa cells, and 1 ⁄3 of the enzyme co-immunolocalizes in the trans-Golgi together with 50% of the cellular GalT-I in HeLa cells (2,4). In context with discussions of the kin recognition hypothesis (3), human GnT-1 has been shown to oligomerize with ␣-mannosidase II by interactions mediated by their stalk regions (38), which retain the two proteins in the same two Golgi compartments. Since our gnt1-FT6 variant comprises the human GnT-I stalk region (aa 1-102), our in vivo specificity data could in fact be explained by such an interaction of this chimera with cellular ␣-mannosidase II. However, similar interactions have not been reported for any other pair of Golgi membrane enzymes (10).
The ft7-FT6 variant, like wt-FT7 described earlier (17), was found to be completely resistant toward intracellular proteolysis as is the case for the gnt3-FT6 chimera, whereas large amounts of FT6 activity were measured in the medium of most of the other stably transfected cell lines including those where a sensitivity to intracellular endoproteolysis has already been described for the wild-type enzymes (ST6Gal I, GalT-I, and FT3). From the detection of intracellular proteolysis of st3-FT6 and pst-FT6, we conclude that the two wild-type sialyltransferases are also susceptible to cleavage, although no data have been reported so far. Apparently, there are at least three different signals contained in the CTS regions of glycosyltransferases mediating, first, their Golgi retention (8,12,29,31); second, as proposed from our results, their targeting to specific in vivo functional areas; and third, susceptibility of the enzymes toward intracellular proteolysis, which might constitute a tool for regulation of the intracellular switch-off of glycosyltransferases. We propose that the in vivo functional distribution of glycosyltransferases can be mapped by the stable expression of CTS variants and the analysis of their biosynthetic products. The results obtained for the ft7-FT6 variant indicate a forward targeting of this chimera into a functional Golgi compartment later than wt-FT6, where the ft7-FT6 catalytic domain has access to higher amounts of already ␣2,3-sialylated precursor N-glycans. In the case of wt-FT6, the earlier ␣1,3fucosylation of ␤-TP asialo oligosaccharides by the FT6 catalytic domain leads to decreased overall ␣2,3-sialylation of the reporter glycoprotein (17). In contrast, ␤-TP from ft7-FT6 cells used here exhibits a higher overall degree of sialylation, similar to the value detected in ␤-TP secreted from wt-BHK-21 cells that do not coexpress fucosyltransferase (17,25).
In the biosynthesis of fucosylated ligands for the selectin family of carbohydrate receptors, the different cellular sialyland fucosyltransferase activities play a crucial role, since for example ␣1,3-fucosylation of Gal(␤134)GlcNAc-R motifs prevents subsequent sialylation by ST3Gal III, and ␣2,6-sialylated oligosaccharides are not a substrate for ␣1,3/4-fucosyltransferases (17). The expression of different sialyl-and fucosyltransferases in the same cell could lead to a competition for common acceptor substrates; therefore, it was of interest to identify possible different targeting signal properties of sialyltransferase CTS regions compared with fucosyltransferases. Our results obtained for the stably expressed st6-FT6 would point to an earlier in vivo functional localization of this variant compared with wt-FT6. The high proportion of Le x already detectable in the st3-FT6(I) product also points to a retention in an earlier compartment of this chimera when compared with wt-FT6 and indicates that the high amounts of Le x structures obtained with the st3-FT6(II) cell line are not only attributable to the bioactivity of enzyme molecules en route to their final Golgi localization. However, by comparison of the functional activity of FT6 CTS variants expressed at different enzyme levels, it is clear that in principal increasing amounts of the enzyme lead to increased synthesis of Le x oligosaccharides, most likely due to an increased concentration of active enzyme already during its transport. Our results obtained in the present investigation perfectly agree with the suggested functional colocalization and in vivo competition of the two enzymes for their common substrate Gal(␤134)GlcNAc-R (25).
The st4-FT6 chimera differed significantly in its sLe x :Le x ratio (2.3:1) of ␤-TP oligosaccharides from st3-FT6(I) cells (1:3) that expressed a similar low intracellular FT6 activity. From this result, we would allocate ST3Gal IV to an in vivo functional Golgi compartment significantly later than ST3Gal III, most probably intermediate between wt-FT6 and ft7-FT6. In the literature, ST3Gal IV has been discussed in the sialylation of Le x motifs, although no final evidence for this hypothesis has been provided so far (61)(62)(63). If in fact ST3Gal IV would be involved in the in vivo sialylation of Le x , this enzyme could explain the low amounts of sLe x that were detected in ␤-TP oligosaccharides from gnt1-FT6 cells in the present study or with the expression of human FT4 reported in our previous work (17). The homologous enzyme is present in BHK-21 cells, since we have cloned the corresponding cDNA from this source (see "Experimental Procedures"). The pst-FT6 variant produced a sLe x :Le x ratio intermediate between st3-FT6/st6-FT6 and wt-FT6. It is important to note that the length of transmembrane region (13 aa) assigned to polysialyltransferase (ST8Sia IV, PST) is significantly shorter than the same region of all other transferases used in this study (17-21 aa). PST is involved in the biosynthesis of polysialylated (NeuAc-(␣238)) n NeuAc(␣233) Gal(␤134)GlcNAc-R structures, which are considered as important signals in tissue developmental processes (64). One would expect that the enzyme functionally localizes together with the cellular ST3Gal III in the same or in a later functional Golgi area, since PST requires ␣2,3-sialylated substrates. The in vivo data resulting from the expression of the pst-FT6 variant support such a hypothesis and clearly indicate a compartmental localization earlier than FT6 and FT7.
In summary, the overall ␣1,3-fucosylation of the reporter glycoprotein ␤-TP and its sLe x :Le x ratio allow for a mapping of the functional localization of the chimeras either before or after wt-FT6, and we hypothesize that this sequential arrangement represents also the sequential distribution of the CTS donor glycosyltransferases in the following order: GnT-I Ͻ (ST6Gal I, ST3Gal III) Ͻ GnT-III Ͻ ST8Sia IV Ͻ GalT-I Ͻ (FT3, FT6) Ͻ ST3Gal IV Ͻ FT7. Apparently, the position allocated to GalT-I within this scheme does not fit to the current model of complextype N-glycan biosynthesis, since GalT-I should provide the substrate for the sialyltransferases and therefore should reside in an earlier Golgi area. However, it could be questioned if the GalT-I used as the CTS donor transferase is the relevant enzyme involved the early biosynthesis of type II N-acetyllactosamine chains, since recently five new human ␤1,4-GalT genes have been identified (65). Recent publications suggest that GalT-I is mainly involved in the biosynthesis of poly-Nacetyllactosamines on N-or O-glycans (66 -69), which is in perfect agreement with our results.
It is noteworthy that the secreted ␤-TP from all transfected cell lines contained exclusively oligosaccharides with intact N-acetyllactosamine antennae. We did not observe truncated oligosaccharides in any of the ␤-TP preparations isolated from the supernatant of the different CTS variant cells. It appears that the partition of the new galt-and gnt1-FT6 variants into the pre-existing gradient of the endogenous glycosyltransferases in the compartments does not interfere with the regular oligosaccharide modification pathway of the cells. This is in agreement with the view that it would be difficult to saturate the mechanism underlying Golgi retention of transferases (10). We have obtained some evidence that the in vivo functional activity of FT6 can be modulated by the overexpression of the enzyme, since in a single cell clone expressing 3 times higher intracellular FT6 activity than was obtained routinely, 75% of the ␤-TP oligosaccharides were modified with peripheral Fuc, 2 ⁄3 of which contained two Le x motifs. This observation can easily be explained by the early modification of Gal (␤134)-GlcNAc-R chains encountered by the high level of fucosyltransferase molecules exhibiting functional activity already during their transport through early compartments. Consequently, the early fucosylation of acceptor substrate leads to reduced amounts available for subsequent sialylation by ST3Gal III and also to substrate depletion for sLe x formation by FT6 molecules that might have been gated into later subcompartmental areas in the high expression cell line. The detection of any such late residing enzyme subpopulation that might have resulted from a hypothetical saturation of the subcompartmental retention mechanism, of course, is prevented by the in vivo activity approach used in our investigation. However, our data obtained with chimeras expressing similar intracellular FT6 activities clearly indicate the presence of signals contained in the CTS regions that can cause a forward (ft7) or a backward (gnt1, st3, st6, gnt3) targeting of enzymes operating in the late biosynthetic glycosylation pathway and thus provide a regulatory means for the spatial separation of enzyme activities competing for the same substrates in the same compartments.
The model for Golgi retention of glycosyltransferases (11, 12) suggests a lipid-mediated sorting mechanism to be responsible for preventing Golgi membrane proteins to be transported to the plasma membrane, since the length of the transmembrane domain of 17-22 aa of most Golgi proteins is about 5 aa shorter than those of the plasma membrane (9, 10). The cholesterol concentration gradient formed throughout the secretory compartments would result in a lipid bilayer with increasing diameter across the Golgi, where the Golgi-resident proteins partition into different lipid/glycolipid microdomains when compared with plasma membrane proteins (12). However, in view of our mapping results, and considering the different transmembrane domains of PST (13 aa), ST6Gal I (17 aa), and GnT-I (21 aa), it is difficult to understand how this model can be applied to all Golgi enzymes. Also, computer modeling of the different glycosyltransferase transmembrane domains does not lead to satisfying results 2 ; therefore, it is conceivable that the flanking domains also contribute to the partition of glycosyltransferases into lipid microdomains and that the enzymes are in fact targeted to nonequivalent in vivo functional areas of the late glycosylation compartment by the signals contained in the CTS region. The significance of the entire CTS region of glycosyltransferases for their in vivo function is also emphasized by our observation that this polypeptide moiety must contain signals mediating resistance (ft7, gnt3) or susceptibility to intracellular cleavage (gnt1, st3, st6, galt, pst, ft3, ft6) by yet unidentified endoprotease(s). The physiological significance of this phenomenon may be considered as playing a role for the rapid elimination of glycosyltransferase activities when, for example, new enzyme genes become activated during differentiation or in developmental processes.