Membrane Topology of the Mammalian CMP-Sialic Acid Transporter*

Nucleotide sugar transporters form a family of distantly related membrane proteins of the Golgi apparatus and the endoplasmic reticulum. The first transporter sequences have been identified within the last 2 years. However, information about the secondary and tertiary structure for these molecules has been limited to theoretical considerations. In the present study, an epitope-insertion approach was used to investigate the membrane topology of the CMP-sialic acid transporter. Immunofluorescence studies were carried out to analyze the orientation of the introduced epitopes in semipermeabilized cells. Both an amino-terminally introduced FLAG sequence and a carboxyl-terminal hemagglutinin tag were found to be oriented toward the cytosol. Results obtained with CMP-sialic acid transporter variants that contained the hemagglutinin epitope in potential intermembrane loop structures were in good correlation with the presence of 10 transmembrane regions. This building concept seems to be preserved also in other mammalian and nonmammalian nucleotide sugar transporters. Moreover, the functional analysis of the generated mutants demonstrated that insertions in or very close to membrane-spanning regions inactivate the transport process, whereas those in hydrophilic loop structures have no detectable effect on the activity. This study points the way toward understanding structure-function relationships of nucleotide sugar transporters.

Nucleotide sugar transporters form a family of structurally related multimembrane-spanning proteins of the Golgi apparatus and the endoplasmic reticulum (ER). 1 Their function resides in translocating activated sugars from the cytosol into the lumen of the ER and Golgi apparatus (1)(2)(3). Transporters therefore provide essential components of the glycosylation pathways in eukaryotic cells (for review, see Ref. 4). Recently, the first nucleotide sugar transporters have been identified at the molecular level. Complementation cloning in mutant cells lacking specific nucleotide sugar transport activities identified the mammalian transporters (Tr) for CMP-sialic acid (5,6), UDP-galactose (UDP-Gal) (7,8), and UDP-N-acetylglucosamine (UDP-GlcNAc) (9), the yeast transporters for UDP-Glc-NAc (10) and UDP-Gal (11), and the Leishmania GDP-mannose transporter (12,13). Additional putative nucleotide sugar transporter sequences have been identified using sequence homologies (8,10,13). Surprisingly high sequence homology has been found between the mammalian transporters for CMPsialic acid, UDP-Gal, and UDP-GlcNAc (8,9), whereas the conservation between transporters of identical specificity in different biological kingdoms can be low (9,10).
As mentioned above, cloning of transporters was achieved by complementation, and the cDNAs isolated were demonstrated to correct the mutant phenotype. Transport activity, however, has only been proven for the murine CMP-Sia-Tr, which could be functionally expressed in Saccharomyces cerevisiae (14). Because yeast cells lack sialic acids, this result clearly demonstrates that the cloned cDNA in fact encodes the CMP-Sia-Tr and not an accessory protein required in the transport process. Moreover, the canine UDP-GlcNAc-Tr, although only 22% identical to the yeast orthologue, has been isolated by expression cloning in the UDP-GlcNAc-Tr negative mutant of Kluyveromyces lactis. This result allows us to speculate that structural elements involved in specific substrate recognition are formed via the tertiary and/or quaternary organization of the transport proteins.
Limited information is available on the regulation of CMPsialic acid translocation, or on how variations in the translocation rates influence sialylation reactions in the Golgi lumen. It is well established that lumenal CMP stimulates CMP-sialic acid uptake by Golgi vesicles, indicating antiport of CMP and CMP-sialic acid (15). However, CMP-sialic acid is also translocated in the absence of CMP (14,15), and CMP and derivatives of CMP-sialic acid have been shown to compete with CMPsialic acid translocation if added onto the cytosolic side of the Golgi membrane (16). These reagents could therefore be used to block the sialylation of cell surfaces. Artificial reduction of cell surface sialylation via application of CMP-sialic acid derivatives has been demonstrated to reduce growth and metastasis of tumor cells (17). Moreover, because sialic acids provide recognition and receptor structures for viral and bacterial pathogens (for review, see Ref. 18), reversible inhibition of cell surface sialylation has been discussed as a perspective to protect healthy cells against these invasive organisms.
Understanding structure-function relationships of nucleotide sugar transporters requires knowledge of the three dimensional organization in the plane of the lipid bilayer. The first important step toward this aim is the determination of the membrane topology. Hydrophobicity analyses of the nucleotide sugar transporters cloned to date suggest between 6 and 10 transmembrane domains and the use of secondary structure prediction algorithms proposed models with eight transmembrane domains for both CMP-Sia-Tr and UDP-galactose transporter (6,7).
This study was undertaken to develop a detailed topological model of the CMP-Sia-Tr. An epitope-insertion approach was used to map membrane orientation. This approach, in contrast to other methods (e.g. methods using truncated proteins fused to reporter proteins), takes into account that even small changes in sequences surrounding transmembrane domains can alter the membrane topology (19) and that it is therefore important to confirm unaltered membrane orientation of the analyzed protein by the remaining activity. The results of this study strongly suggest a 10-transmembrane domain topology in which amino and carboxyl termini are oriented toward the cytosol.
Construction of Insertion and Deletion Mutants-Mouse CMP-Sia-Tr, with carboxyl-terminal HA tag and amino-terminal FLAG sequence, respectively, were generated as described previously (5,6). Using the FLAG-tagged construct as template, overlapping extension polymerase chain reaction (24) was used to generate two series of insertion mutants, the N constructs (N1-N15) and the HA constructs. For the production of N construct, primers were designed that introduced the sequence GGATCCAACGCTAGC at selected sites of the CMP-Sia-Tr (see Table I). This sequence, which encodes the pentapeptide GSNAS, introduces BamHI and NheI restriction sites and harbors a potential N-glycosylation site. The newly introduced restriction sites were then used to produce HA constructs by ligating a linker encoding the HA epitope (HA1-HA15). The linker was prepared by annealing the 5Јphosphorylated oligonucleotides HA sense (5Ј-GATCCTACCCTTATG-ACGTCCCCGATTACGCCAGCCTGG-3Ј) and HA antisense (5Ј-CTA-GCCAGGCTGGCGTAATCGGGGACGTCATAAGGGTAG-3Ј). In the final constructs, the peptide sequence GSYPYDVPDYASLAS was inserted after the amino acid residue indicated in Table I (the epitope of mAb 12CA5 is underlined). Construct HA16 was generated by overlap extension polymerase chain reaction, directly introducing the HA encoding sequence (see above) between nucleotides 270 and 271 of the CMP-Sia-Tr coding sequence. A carboxyl-terminal truncated version of construct HA12 was prepared by digestion with NheI, fill-in with Klenow enzyme, and religation, resulting in HA12STOP with a stop codon immediately downstream of the HA epitope. All constructs were verified by DNA sequencing.
Transient Transfections-Cells were seeded at 2.5 ϫ 10 5 cells per 35-mm cell culture dish or at 1.5 ϫ 10 6 per 10-cm dish. For immunofluorescence analysis, cells were seeded onto glass coverslips. Epitopetagged CMP-Sia-Tr cDNAs were transfected using LipofectAMINE (Life Technologies, Inc.) following the instructions of the manufacturer. Briefly, 1 g cDNA was mixed with 6 l (24 l) LipofectAMINE in 1 ml Opti-MEM medium (Life Technologies, Inc.) added to the cells that had been washed twice with Opti-MEM and incubated for 6 -8 h. Transfections were stopped by adding 2 volumes of Dulbecco's modified Eagle's medium/Ham's F-12 medium (supplemented with 10% fetal calf serum, 1 mM sodium pyruvate, 100 units/ml penicillin, and 100 g/ml streptomycin), and cells were analyzed 48 h later.
Indirect Immunofluorescence-Cells were washed twice with PBS, fixed in 4% paraformaldehyde for 15 min, again washed in PBS, and incubated for 20 min in 50 mM NH 4 Cl in PBS to neutralize residual paraformaldehyde. Thereafter, cells were permeabilized with 0.2% saponin, 0.1% BSA in PBS for 15 min and incubated with the respective primary antibody for 2 h at room temperature or overnight at 4°C. Antibodies used were anti-FLAG mAb M5 (5.4 g/ml), anti-HA mAb 12CA5 (2.6 g/ml), and rabbit anti-␣-mannosidase II antiserum (1: 2000) in 0.2% saponin, 0.1% BSA in PBS. After washing four times in 0.1% BSA/PBS, cells were incubated with anti-mouse Ig-DTAF (1:100) and anti-rabbit Ig-TRITC (1:100) for 1 h at room temperature. The incubation was stopped by washing three times in 0.1% BSA/PBS. After a final wash in PBS, slides were briefly rinsed in water and mounted in moviol. Samples were visualized under a Zeiss Axiophot Epifluorescence microscope.
To selectively permeabilize the plasma membrane of transfected CHO cells, we used low concentrations of digitonin (25). Cells were fixed in paraformaldehyde as described above and washed three times with PBS, and the plasma membrane was permeabilized by incubating the cells for 15 min at 4°C in 5 g/ml digitonin, 0.3 M sucrose, 0.1 M KCl, 2.5 mM MgCl 2 , 1 mM EDTA, and 10 mM Hepes, pH 6.9. Thereafter, cells were washed three times with PBS, and nonspecific binding sites were blocked with 1% BSA in PBS at room temperature for 30 min. Control cells were treated in the same way, but 0.1% saponin was added to the blocking solution to achieve complete permeabilization. For staining, cells were processed as described above, but saponin was omitted from solutions containing primary antibodies.
Structure Predictions-Putative transmembrane-spanning regions and helix ends of CMP-Sia-Tr, human and yeast UDP-Gal-Tr and a related putative nucleotide sugar transporter of Caenorhabditis elegans were estimated by using the programs PredictProtein (27) and interfacial hydrophobicity analysis (28). The murine CMP-Sia-Tr sequence was aligned to the human UDP-Gal-Tr, the Schizosaccharomyces pombe UDP-Gal-Tr and the related C. elegans protein ZK370.7 using CLUSTAL (29). Helix ends were estimated by assuming that helix ends in corresponding transmembrane domains of different transporters are identical, if the sequences surrounding the helix ends are highly conserved. When different results were obtained for different transporter sequences or with different algorithms, the most plausible distribution of charged and polar residues was fitted to the experimental results.

Amino and Carboxyl Termini of CMP-Sia-Tr Are
Oriented toward the Cytosol-The murine wild-type CMP-Sia-Tr, with either amino-terminal FLAG or carboxyl-terminal HA tag, was transiently expressed in the Lec2 clone 8G8 (5), and the orientation of the epitope tags was determined by immunofluorescence using digitonin-permeabilized cells. Low concentrations of digitonin selectively permeabilize the plasma membrane because of its higher concentration of cholesterol compared with intracellular membranes (30). Cells were simultaneously stained with M5 and polyclonal ␣-mannosidase II antibodies or with mAb 12CA5 and polyclonal ␣-mannosidase II antibodies. Because the ␣-mannosidase II antiserum recognizes the catalytic domain of the enzyme exclusively (31), this staining could be used to control the integrity of Golgi membranes after per-meabilization. As shown in Figs. 1A and 2A, both mAb 12CA5 and mAb M5 recognize their epitopes in digitonin-treated cells, whereas no ␣-mannosidase II staining was detectable (Figs. 1B and 2B). After saponin permeabilization, all antibodies reached their epitopes, but the intensity of the staining with mAbs 12CA5 and M5 was not increased (Figs. 1, A and C, and 2, A and C), demonstrating that HA and FLAG epitopes were fully accessible in semipermeabilized cells. These results clearly demonstrate that CMP-Sia-Tr is composed of an even number of transmembrane domains, with amino and carboxyl termini oriented toward the cytosol.
Evidence for 10 Membrane-spanning Domains in the CMP-Sia-Tr-The orientation of amino and carboxyl termini as evaluated in the first experiment was in agreement with the predicted model of the CMP-Sia-Tr (6). Therefore, we used this model to select putative intermembrane loops for the generation of insertion mutants (Fig. 3). In the first set of mutants, a short nucleotide sequence that encodes the pentapeptide GS-NAS was introduced at various positions of the amino-terminally FLAG-tagged murine CMP-Sia-Tr. The resulting mutants received the name N constructs, because the inserted pentapeptide contained the consensus motif for N-glycosylation. The foreign nucleotide sequence, in addition, introduced unique endonuclease restriction sites, which in a second step could be used to ligate a linker that encodes the HA epitope consisting of the amino acid sequence GSYPYDVPDYASLAS. The latter constructs received the name HA constructs. All insertion mutants are summarized in Table I.
To evaluate the membrane topology of the CMP-Sia-Tr, the HA constructs were transiently expressed in CHO 8G8 cells (5), and the orientation of the HA epitope was determined by immunofluorescence with the mAb 12CA5 in permeable (saponintreated) and semipermeable (digitonin-treated) cells. Simultaneous staining with the ␣-mannosidase II antiserum served as a control for the integrity of Golgi membranes in digitonintreated cells. The HA epitopes inserted at amino acid positions 38 (HA1), 109 (HA14), 165 (HA4), and 233 (HA6) could be detected after saponin but not after digitonin treatment of the cells (Fig. 4, 12CA5), indicating that the tags in these constructs are part of lumenal loops. In contrast, the HA epitopes introduced in mutants HA2, HA13, HA5, and HA7 (insertions were at amino acid positions 83, 137, 202, and 266, respectively) were detectable in both semipermeabilized and saponintreated cells, demonstrating cytosolic orientation. Together, these results identified the hydrophobic regions I, II, IIIa, IIIb, IV, V, and VI (see Fig. 3C) as membrane-spanning domains. Moreover, the cytosolic orientation of the HA epitope in construct HA2 and the lumenal orientation of the tag in construct HA14 clearly show that amino acids between these regions, despite of the relatively weak hydrophobic character, span the Golgi membrane. The co-localization of the transporter mutants with ␣-mannosidase II (Fig. 4, compare third and fourth columns) indicates that Golgi targeting was not affected.
The transporter variants HA9, HA3, HA15, HA11, HA8, and HA12 (see Figs. 3C and 8A) were undetectable in immunofluorescence studies with mAb 12CA5; however, expression and localization could be displayed with the anti-FLAG mAb M5 (examples are displayed in Fig. 5). In the case of HA9, only few transfected cells were detectable by immunofluorescence (see Fig. 5), and the protein was not visible in Western blot (see Fig.  7). All other variants were stained with mAb 12CA5 and migrated with an apparent molecular mass of 31 kDa (see Fig. 7). The inaccessibility of the HA tags HA9, HA3, HA15, HA11, HA8, and HA12 in immunofluorescence suggests that these epitopes are inserted into transmembrane domains, or, as in the case of construct HA3, they are in close proximity to the membrane. Co-localization with ␣-mannosidase II was found for all constructs, but HA9 was partially retained in the ER. Correct targeting of the transporter mutants is consistent with their correct folding.
Neither the HA tag of construct HA8 nor that of HA12 could be detected by immunofluorescence; therefore, it remained unclear whether the hydrophobic regions VII and VIII (see Fig.  3C) transverse the membrane, form integral parts of the membrane, or are tightly membrane associated. To resolve this problem, construct HA12STOP was generated by introducing a stop codon immediately after the HA sequence in construct HA12. Immunofluorescence was then used to search for the HA epitope in transiently transfected cells. As shown in Fig. 6C, the HA epitope was still not detectable, although the correctly targeted protein was stained with anti-FLAG mAb M5 (Fig. 6, A  and B), and a protein of the expected molecular mass was developed in Western blot with mAb 12CA5 (Fig. 6D). In contrast, the HA constructs obtained after insertion of the tag sequence after Ser-298, Gly-302, and Ser-309 led to inactive proteins that were not detectable in either immunofluorescence or Western blot (data not shown). These results argue against membrane association and promote the assumption that the hydrophobic regions VII and VIII are integral parts of the Golgi membrane.
Activity of Insertion Mutants-Even small sequence changes FIG. 1. Orientation of the carboxyl terminus of CMP-Sia-Tr. CHO cells on glass coverslips were transfected with HA-tagged mouse CMP-Sia-Tr. Two days after transfection, cells were fixed in paraformaldehyde and incubated in 5 g/ml digitonin to selectively permeabilize the plasma membrane (A and B) or in saponin (C and D). Cells were then subjected to indirect immunofluorescence using anti-HA mAb 12CA5 (A and C). Simultaneously, all cells were stained with ␣-mannosidase II antiserum (B and D) exclusively directed against the catalytic domain. Bound primary antibodies were visualized using anti-mouse Ig-DTAF and anti-rabbit Ig-TRITC conjugates. Inaccessibility of the ␣-mannosidase II antiserum to its antigen confirms integrity of the Golgi membrane in digitonin-treated cells (A and B). Bar, 20 m.
near membrane-spanning domains can alter the membrane topology (19). The model deduced from the epitope insertion study was therefore verified by testing activity of the mutants. Although loss of transport activity does not necessarily imply a changed topology, an active transporter strongly indicates intact membrane topology. In the activity tests, we took advantage of the fact that CHO wild-type cells in contrast to Lec2 mutants express polysialic acid (5), detectable with the mAb 735 (20). mAb 735 in different experimental systems detects polysialic acid with a much higher sensitivity than lectins (e.g. Maackia amurensis agglutinin) (5). 2 Constructs were transiently transfected into 8G8 cells, and the expression of mutant CMP-Sia-Tr was analyzed by Western blotting with the anti-FLAG mAb M5 and the anti-HA mAb 12CA5 (Fig. 7). Bands of approximately 31 kDa were detected in transfectants expressing mutant or wild-type CMP-Sia-Tr. A difference in the migration behavior that could be attributed to N-glycosylation, however, was not found in the N constructs ( Fig. 7B and data not shown). The inability of the pentapeptide sequence to serve as glycosylation site most probably reflects insufficient spacing between the potential N-glycosylation site

FIG. 2. Orientation of the amino terminus of CMP-Sia-Tr. CHO cells were grown on glass coverslips and transiently
transfected with the amino-terminally FLAG-tagged mouse CMP-Sia-Tr using LipofectAMINE. Two days after transfection, cells were fixed in paraformaldehyde and incubated in 5 g/ml digitonin to selectively permeabilize the plasma membrane (A and B) or in saponin (C and D). Thereafter cells were subjected to indirect immunofluorescence using the anti-FLAG mAb M5 (A and C). Simultaneously, all cells were stained with ␣-mannosidase II antiserum (B D). Bound primary antibodes were visualized using anti-mouse Ig-DTAF and anti-rabbit Ig-TRITC conjugates. Again, inaccessibility of the ␣-mannosidase II antiserum to its antigen revealed that the Golgi membrane was not permeabilized by digitonin treatment. Bar, 20 m.

FIG. 3. Position of HA epitope insertions analyzed in this study.
A, hydrophobicity plot of mouse CMP-Sia-Tr using hydrophobicity values as described (52) and a window size of nine amino acids. B, probability of transmembrane helices of CMP-Sia-Tr as proposed by the neural-network based program PredictProtein (27). C, hydrophobic segments (I-VIII) predicted to form membrane-spanning helices. Note that segments IIIa and IIIb were previously proposed to form a single transmembrane domain. The positions of introduced HA epitopes are indicated.  4. Orientation and localization of HA epitopes of HA-tagged CMP-Sia-Tr. CHO cells were transiently transfected with the HA-tagged CMP-Sia-Tr mutants (indicated at left) and simultaneously stained for HA epitopes using mAb 12CA5 and ␣-mannosidase II after digitonin or saponin permeabilization, as described in the legend to Fig. 1. Data obtained for constructs HA10 and HA16 were identical to those for HA2. Bar, 20 m. and the membrane. According to Nilsson and von Heijne (32), a minimal distance of 12-14 amino acids is required.
Transporter mutants for which the HA epitope was undetectable by immunofluorescence (HA9, HA3, HA11, HA15, HA8, and HA12) were unable to correct the phenotype of the PSA-negative CHO mutant 8G8 (Fig. 7A), and transfection of the corresponding N constructs led to the same results (Fig.  7B). On the other hand, only two of the HA epitopes inserted into hydrophilic loops, namely those in constructs HA14 and HA7, inactivated the transporter, as shown by the absence of polysialic acid after transfection into 8G8 cells (Fig. 7A). For the mutant pair N6/HA6, a gradual diminishing of the ability to complement the 8G8 mutant was observed with the size of the insert. Although N6 led to the same level of PSA expression as the wild-type construct, complementation of 8G8 cells with construct HA6 was less efficient. The level of PSA surface expression varied between different experiments and was strongly dependent on the transfection efficiency (data not shown), yet it never reached the level obtained with N6. Interestingly, constructs HA10 and HA16, which destroy the leucine zipper motif (amino acids 56 -77), produced fully active transporters.
The Membrane Topology of the CMP-Sia-Tr-The above experiments strongly suggest a 10 membrane-spanning domain topology for the CMP-Sia-Tr, but the boundaries of transmembrane helices remain undetermined. In order to propose a model of the CMP-Sia-Tr that can be used to determine further structural details, we used different algorithms (see under "Experimental Procedures") to predict possible helix ends. The methods were applied to the CMP-Sia-Tr, to the closely related UDP-Gal-Trs from human and yeast, and to the homologous protein from C. elegans. The tertiary structure predictions in combination with the experimental data discussed above allow the deduction of the CMP-Sia-Tr model shown in Fig. 8A. The predicted transmembrane domains were further analyzed by helical wheel projection. The projections shown in Fig. 8B demonstrate the presence of hydrophobic and hydrophilic faces in most transmembrane helices. Conserved amino acids are FIG. 5. Localization of HA-tagged CMP-Sia-Tr. HA-tagged CMP-Sia-Trs that were undetectable with mAb 12CA5 were transiently transfected into CHO cells. Two days after transfection, cells were saponin permeabilized and subjected to immunofluorescence analysis using anti-HA mAb 12CA5 or anti-FLAG mAb M5. All cells stained with mAb M5 were simultaneously stained with ␣-mannosidase II antiserum. Results similar to those for HA11 and HA3 were obtained for constructs HA15, HA8, and HA12 (data not shown). Bar, 20 m.

FIG. 6. Expression of a carboxyl-terminally truncated CMP-Sia-Tr.
A carboxyl-terminally truncated CMP-Sia-Tr with an HA tag at amino acid position 294, followed by a stop codon (construct HA12STOP), was expressed in CHO 8G8 cells and analyzed by immunofluorescence as described in Fig. 5 using anti-FLAG mAb M5 (A), ␣-mannosidase II antiserum (B), or anti-HA mAb 12CA5 (C). D, cell lysates of 8G8 cells expressing HA12STOP (lane 2) or transfected with the empty vector pcDNA3 (lane 1) were analyzed by Western blotting using mAb 12CA5. A specific band with the expected molecular mass of approximately 28 kDa was detectable (arrowhead). Bands visible in both lanes are due to nonspecific binding of the primary and secondary antibodies. thereby concentrated in the more hydrophilic faces. In fact, only 16% of the hydrophobic residues are conserved, whereas 33% of the nonhydrophobic amino acids are identical in the four transporter sequences. DISCUSSION Complementation cloning in Lec2 cells identified the first mammalian nucleotide sugar transporter (5). The highly hydrophobic protein was shown to be localized in the Golgi apparatus and to span the lipid bilayer multiple times (5,6). Additional nucleotide sugar transporters were cloned from different species (5,7,9,10,12). Depending on the algorithm used to predict structural features, and depending on the transporter analyzed, the number of predicted putative transmembrane domains varied considerably. Therefore, an epitope insertion study was performed in the present study to elucidate the membrane topology of the CMP-Sia-Tr. This experimental approach has clear advantages over other strategies used to define the arrangement of type III membrane proteins. The modifications introduced in the primary sequence are small. Functional activity can be retained in many mutants and indicates correct molecular organization of the protein. Using epitope insertion and indirect immunofluorescence studies, we provide strong evidence for 10 transmembrane passages (Fig.  8) in the CMP-Sia-Tr.
Insertions in general bear the risk of changing the topogenic activity of the adjacent membrane-spanning domains. Therefore, the activity of the mutated proteins was assayed and taken as an indicator for intact membrane organization. For this reason, we measured PSA expression in the transiently transfected Lec2 mutant 8G8.
Indirect immunofluorescence in transiently transfected 8G8 cells demonstrated that both the amino and carboxyl are oriented to the cytosolic side of the Golgi membrane. These data indicate an even number of transmembrane domains for the CMP-Sia-Tr. Moreover, the epitope localization of the constructs HA1, HA10, HA13, HA4, HA5, and HA6 confirmed the existence of TM1, TM2, TM5, TM6, and TM7. All constructs used in this part of the study were functionally active (see also Refs. 5 and 6). In contrast, HA tags introduced into HA14 and HA7 inactivated transport, but targeting of the proteins to the Golgi apparatus suggested correct folding. These constructs therefore confirmed TM3, TM4, and TM8.
We were unable to unequivocally demonstrate that TM9 and TM10 transverse the Golgi membrane, although the hydrophobic character of this region suggests the presence of integral membrane domains. HA tags inserted after the amino acid residues Ser-287 (HA8) and Gln-294 (HA12, HA12STOP) were found in Western blot but were undetectable by immunofluorescence, and further downstream insertion of the HA tag (after Ser-298, Gly-302, and Ser-309) resulted in unstable proteins that were not detectable neither in Western blot nor immunofluorescence (data not shown). If the region TM9-TM10 were not an integral part of the membrane but associated with the cytosolic side of the Golgi, we would have expected to see the cytosolically oriented HA tag of HA12STOP. The inaccessibility of the HA tag in the constructs (see Fig. 6) provides strong evidence that the hydrophobic regions VII and VIII (Fig.  3C) are integral parts of the membrane. Moreover, all structure prediction algorithms used so far suggest two membrane integral helices in this domains. Additional studies using different techniques (e.g. cysteine-modified proteins, as described in Ref. 33) are required to find out whether a part of this region is accessible from the lumenal site of the Golgi.
It will be interesting to determine whether all nucleotide sugar transporters exhibit a common membrane topology. The high similarity of the hydrophobicity plots and sequence comparisons of the different transporters support this possibility. Ten hydrophobic domains can be clearly distinguished in the GDP-mannose transporter of Leishmania donovani (12), are compatible with the sequences of the K. lactis UDP-GlcNAc-Tr, the related proteins from S. cerevisiae and C. elegans (10), and are in accordance with the primary sequence of a putative human nucleotide sugar transporter (8). In contrast, the S. pombe UDP-Gal-Tr, although showing a strong sequence homology to CMP-Sia-Tr, lacks the first transmembrane domain (11). A second exception from the 10-transmembrane model is a putative nucleotide sugar transporter (ZK370.7) identified in C. elegans. In this protein the region corresponding to the first two transmembrane domains in CMP-Sia-Tr is absent. So far, no functional data on the C. elegans transporter are available; however, a mutant of the hamster CMP-Sia-Tr has been identified in our laboratory that lacks the first two transmembrane domains and is functionally inactive (34). In contrast to the variability observed in the amino-terminal region, significant sequence similarity between the different transporters has been found in the carboxyl-terminal membrane regions (6,9,13). Taken together, these data suggest that the majority of nucleotide sugar transporters are composed of 10 transmembrane domains, although some may lack the first or the first and second membrane-spanning domain.
Some transporter proteins form dimers (35)(36)(37), and based on data obtained with the related adenosine 3Ј-phosphate 5Јphosphosulfate transporter (38), it has been proposed that nucleotide sugar transporters also function as dimers (39). The idea is further supported by the facts that (i) many Golgi resident proteins form dimers (40 -42), and (ii) some nucleotide sugar transporters contain a leucine zipper motif that could be involved in dimerization (4). Introduction of a pentapeptide sequence or a hemagglutinin epitope between the second and third (constructs N10 and HA10) or the third and fourth leucine residues of the leucine zipper motif (construct HA16) did not inactivate the CMP-Sia-Tr. These mutants were able to fully correct the Lec2 phenotype. In agreement with these data, the simultaneous change of the second and third leucine resi-due of the zipper to alanine did not interfere with Golgi targeting or transport activity. 2 Although these data cannot answer the question on the active transport unit, they clearly demonstrate that the leucine zipper does not play an essential role in the formation and/or stabilization of the active transporter.
With the exception of construct HA9, the insertion mutants were correctly targeted to the Golgi apparatus. In the case of insertions into lumenal or cytosolic loops, this observation fits well with the present view that mainly the transmembrane domains of Golgi resident membrane proteins determine the targeting process (43)(44)(45). The insertion of 15 amino acids into a membrane-spanning helix was, however, expected to perturb correct folding of the proteins. It was therefore interesting to see that some of these constructs were correctly targeted. In case of HA9, both ER and Golgi localization were observed, most probably reflecting a decreased stability of HA9 (compare the Western blot in Fig. 7A). From these results we conclude that not mistargeting, but perturbation of functional organization is responsible for the loss of CMP-sialic acid transport activity in insertion mutants. The second lumenal and forth cytosolic loop form essential parts of the molecule, because the constructs HA14, N14, HA7, and N7 are inactive. The functional importance of the fourth cytoplasmic loop is further supported by the high conservation between the mammalian transporters for CMP-Sia, UDP-Gal, and UDP-GlcNAc (9). Also involved in the architecture of the transport unit is the forth lumenal loop. Although insertion of a pentapeptide in N6 did not interfere with activity, the insertion of the larger HA tag (HA6) led to a drastic reduction of the PSA signal. In contrast, insertion of the pentapeptide or of the HA tag into the first (HA1) and third (HA4) lumenal and the first (HA10, HA16, and HA2), second (HA13), and third (HA5) cytosolic loop did not affect CMP-Sia-Tr activity.
Biochemical studies on nucleotide sugar transport showed that the corresponding nucleoside monophosphates, but not the free sugars, are efficient competitive inhibitors of transport activity (2,46). These data provide strong evidence that the specificity of translocation is mediated by the nucleotide moiety, suggesting a specific nucleotide binding site. In view of these results, it seems surprising that the recently isolated canine UDP-GlcNAc-Tr exhibits only 22% amino acid sequence identity to its orthologue isolated from K. lactis (10), but 40% identity to the murine CMP-Sia-Tr sequence. However, sialic acids appeared late in evolution, and in eukaryotes, they are restricted to chordates and echinoderms (47). Thus, the relatively high sequence homology of the mammalian nucleotide sugar transporters might reflect a common evolutionary origin.
Hydrophobic moment and substitution moment often point to opposite faces of helices of type III membrane proteins, suggesting that the less conserved hydrophobic faces interact with membrane lipids, whereas the hydrophilic faces interact with other transmembrane domains or with substrate molecules (48 -51). Helical wheel projections revealed an asymmetric distribution of conserved and hydrophobic amino acids for most transmembrane helices of the CMP-Sia-Tr (Fig. 8B). Therefore, it seems likely that the less conserved, hydrophobic faces are oriented toward the lipid phase of the membrane and that the more conserved hydrophilic faces interact with other transmembrane domains, or with the nucleotide sugar. TM4 exhibits the highest degree of conservation (50% identity with the UDP-Gal-Tr). The conserved amino acids are equally distributed and suggest that TM4 is less exposed to the lipid phase. With respect to the different substrate molecules of CMP-Sia-Tr and UDP-Gal-Tr, it is plausible to speculate that the highly conserved polar residues are involved in the interactions between adjacent membrane domains and not in substrate recognition. Nevertheless, cytidine and uridine are chemically similar, and the CMP-Sia-Tr is inhibited by both CMP and UMP (2). Conserved hydrophilic residues could therefore participate in nucleotide sugar recognition and binding. The generation of chimeric proteins and site directed mutations of conserved amino acid residues provide promising strategies to resolve these questions in the future. The model developed for the CMP-Sia-Tr in the present study will greatly facilitate these studies and provides a basis for investigating structure function relationships of nucleotide sugar transporters in general.