Mapping the distribution of Golgi enzymes involved in the construction of complex oligosaccharides

The construction of complex, bi-antennary, N-linked oligosaccharides involves the sequential action of enzymes located in different parts of the Golgi apparatus (for reviews, see Kornfeld and Kornfeld, 1985; Roth, 1991). α? mannosidase I continues the trimming of mannose residues that started in the endoplasmic reticulum (ER) leaving a penta-mannose core to which the first N-acetylglucosamine is added by β1,2 N-acetylglucosaminyltransferase I (NAGT I). α1,3-1,6 mannosidase II (Mann II) removes two more mannose residues permitting addition of the final N-acetylglucosamine by β1,2 N-acetylglu-cosaminyltransferase II. Each branch can then be elongated by the addition of galactose by β1,4 galactosyltransferase (GalT) and sialic acid by α2,6 sialyltransferase (SialylT). Fucose may also be added prior to or following the addition of sialic acid.

GalT was the first of these enzymes to be localised, first to the trans cisterna (Roth and Berger, 1982) and later to the trans-Golgi network (TGN) (Lucocq et al., 1989; Nilsson et al., 1993). SialylT was found to localise to the trans Golgi cisterna and the TGN (Roth et al., 1985) and to have the same distribution in most though not all cells (Roth et al., 1986). NAGT I was found in medial cisternae (Dunphy et al., 1985) as, more recently, was Mann II (Velasco et al., 1993). The location of these enzymes strongly supported the idea that proteins undergoing transport moved through the stack in a cis to trans direction, sampling each of the compartments in turn.

The fact that most of these enzymes were usually found in two adjacent cisternae was taken as evidence of cisternal duplication. This interpretation was supported by the observation that the number of cisternae in the Golgi stack can vary widely from tissue to tissue and from organism to organism (see Fawcett, 1981). To investigate the possibility of cisternal duplication, we examined, using immunogold electron microscopy, the distribution of NAGT I and GalT in HeLa cells. Each enzyme was found in two, adjacent cisternae but, contrary to expectation, both were present in the trans cisterna (Nilsson et al., 1993). In other words, the two enzymes had an overlapping distribution such that each cisterna contained a unique mixture of enzymes not a unique set. In order to generalise these observations and obtain further evidence against cisternal duplication and in favour of each cisterna having a unique composition, we have extended our observations to other Golgi enzymes. To do this we have generated a series of stable HeLa cell lines expressing either epitope-tagged enzymes or ones to which antibodies were available.

HeLa cell lines expressing either human NAGT I (Kumar et al., 1990) tagged with a myc-epitope or murine Mann II (Moremen and Robbins, 1991) have been described elsewhere (Nilsson et al., 1993, 1994).

Human SialylT (Grundmann et al., 1990) was tagged with a VSV-G epitope (underlined) (Kreis, 1986; Soldati and Perriard, 1991) using PCR (Saiki et al., 1988) to introduce the epitope immediately prior to the stop codon. Primers used were:

5′GTCGACGGATCCACCATGATTCACACCAACCTGAAG3′;and 5′GTCGACGGATCCTTACTTTCCCAGCCTGTTCATCTCTA-TATCGGTGTAAGGGCAGTGAATGGTCCGGAAGCC3′.

The PCR product was sequenced and subcloned into the BamHI site of pSRα (DNAX, Palo Alto, CA).

A full length cDNA encoding rat TGN38 (Luzio et al., 1990) was digested with HindIII and complementary oligonucleotides encoding a BamHI site were introduced immediately following the stop codon. The complementary oligonucleotides were:

5′AGCTTTGAG3′ and 5′GATCCTCAA3′?The coding region of TGN38 was then excised and subcloned into the BamHI site of pCMUIV (Nilsson et al., 1989).

HeLa cells were transfected with tagged SialylT in pSRα either alone or together with TGN38 in pCMUIV. Transfection and isolation of stable lines was carried out essentially as described previously (Nilsson et al., 1994). Table 1 summarises relevant properties of the HeLa cell lines used.

SialylT-HeLa and TGN38/SialylT-HeLa cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco) supplemented with 10% foetal calf serum, penicillin (100 μg/ml), streptomycin (100 μg/ml) and geneticin (500 μg/ml) (Gibco). Approximately 10⁹ cells were used to isolate Golgi membranes (Balch et al., 1984) which were purified at least 10-fold over homogenate assayed by GalT activity (Bretz and Stäubli, 1977). Protein concentration was determined using the BCA protein assay kit (Pierce Chemical Co, Rockford, IL). SDS-PAGE was carried out essentially as described by Blobel and Dobberstein (1975) and western blotting as described previously (Nilsson et al., 1993). SialylT was assayed as described by Dunphy et al. (1981) using asialo-transferrin as the substrate.

Cells were fixed either in 2% paraformaldehyde and 0.2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4, or in 0.5% glutaraldehyde in the same buffer containing 0.2 M sucrose and processed as described previously (Rabouille et al., 1993). Briefly, cells were embedded in 10% gelatine and small blocks were infused with 2.3 M sucrose and frozen in liquid nitrogen. Ultra-thin cryosections were cut on an Ultracut E microtome with FC4E cryo-attachment and transferred onto collodion-carbon coated, copper or nickel, 100-mesh grids. All antibodies and gold conjugates were diluted in 0.5% fish skin gelatine in PBS.

The following primary antibodies were used: the 9E.10 mouse monoclonal antibody which recognises the c-myc epitope (Evan et al., 1985) at the C-terminus of NAGT I; a rabbit polyclonal antibody recognising rat Mann II (Moremen et al., 1991); a rabbit polyclonal antibody (N₁₁) recognising human GalT (Watzele et al., 1991); the P5D4 mouse monoclonal antibody which recognises the VSV-G tag (Kreis, 1986; Soldati and Perriard, 1991) at the C-terminus of SialylT; and a rabbit polyclonal antibody recognising rat TGN38 (Luzio et al., 1990). Goat anti-mouse antibodies coupled to gold (Biocell Research Laboratories, Cardiff, UK) were used to detect the primary monoclonal antibodies whereas goat anti-rabbit antibodies coupled to gold (Biocell Research Laboratories, Cardiff, UK) or Protein A gold (from Dept of Cell Biology, Utrecht School of Medicine, Utrecht, the Netherlands) were used to detect primary polyclonal antibodies.

Two protocols were used for double-labelling experiments. When one of the primary antibodies was a monoclonal and the other a poly-clonal, they were mixed together for the initial incubation with the section. Each of the secondary antibodies was then added sequentially (Nilsson et al., 1993). When both antibodies were polyclonal, incubation with the first primary antibody was followed by goat anti-rabbit or Protein A coupled to one size of gold. Sections were then fixed for 15 minutes in 4% paraformaldehyde and the second primary antibody added followed by goat anti-rabbit or Protein A coupled to a different size of gold (Slot et al., 1991).

Grids were stained with 2% neutral uranyl acetate and embedded in 2% methyl cellulose containing 0.4% uranyl acetate as described by Tokuyasu (1980). Grids were examined at 80 kV using a Philips CM10 electron microscope. Pictures were taken at a magnification of 15.5 or 21 K.

The compartments of the Golgi apparatus were defined as described previously (Nilsson et al., 1993; Ponnambalam et al., 1994). Briefly, the trans or T cisterna is defined as the last continuous cisterna on the side of the Golgi stack that labels for GalT. Since the Golgi stack in HeLa cells typically contains three cisterna, the T-1 cisterna most likely corresponds to the medial cisterna and the T-2 to the cis cisterna. The T+1 compartment is the TGN and comprises a tubulo-reticular network closely apposed to the trans face of the trans cisterna. It differs from the CGN (T-3) in having clathrin-coated (Pearse and Robinson, 1990) in addition to COP-coated buds. Clathrin coats have a different morphology and thickness to COP coats (Orci et al., 1984, 1985; Oprins et al., 1993). Nevertheless, it was occasionally difficult to distinguish the TGN from the CGN so double-labelling for GalT and the test enzyme was used in preliminary experiments to define the polarity of the stack.

The relative distribution of gold particles over the TGN and each cisterna was estimated by counting the number of gold particles falling within the boundary of each structure. The boundary of a cisterna was defined as the cisternal membrane. The boundary of the TGN was defined as the interface between the outermost membranes of the tubulo-reticular network and the immediately adjacent amorphous cytoplasm and was drawn on each micrograph. On occasion this boundary would include profiles of budded vesicles which were included in the quantitation whilst other structures (e.g. vacuolar endosomes) were omitted.

The linear density of gold particles/μm membrane was estimated as described previously for the T, T-1 and T-2 cisternae (Nilsson et al., 1993). The boundary of the TGN was drawn on each micrograph and the surface density was estimated by the point-hit method (Weibel, 1979; Ponnambalam et al., 1994). The length of every portion of membrane within this boundary was estimated by the intersection method (Weibel, 1979; Nilsson et al., 1993). Since the ratio of surface density to length was found to be constant (0.062±0.02) between different cell lines, the membrane length in most experiments was calculated from the surface density and this ratio.

The grid had a 5 mm spacing and the micrograph a final magnification that varied from 50 to 100 K. The linear density was calculated by dividing the number of gold particles by the membrane length.

TGN38/SialylT-HeLa cells were grown to 70% confluency on cover slips and incubated for 2 hours in the presence of 10 μg/ml cycloheximide. Cells were fixed and permeabilised essentially as described by Louvard et al. (1982). Bound primary antibodies were visualised using secondary antibodies coupled either to Texas Red (Vector Lab-oratories, Inc., Burlingame, CA) or FITC (Dakopatts, Copenhagen). Cells were visualised using a Zeiss Axiophot Epifluorescence micro-scope and photographed directly using Ilford black and white film.

GalT was the only endogenous enzyme under study that could be readily detected in cryo-sections of HeLa cells using affinity-purified antibodies to the deglycosylated protein (Watzele et al., 1991). Detection of the other enzymes was made possible by transfecting the parental HeLa cell line with the PSRα plasmid containing the appropriate cDNA (see Materials and Methods) and selecting stable cell lines in the presence of geneticin. Clones were picked at random and immunofluorescence microscopy was used to select those clones expressing approximately equal amounts of protein in all cells as described by Nilsson et al. (1994). Expressed protein was detected using either specific polyclonal antibodies or monoclonal antibodies to an epitope tag engineered onto the C-terminus of the enzyme (Table 1). The structure and topology of the proteins under study is summarised in Fig. 1. Stable HeLa cell lines expressing NAGT I (Nilsson et al., 1993) and Mann II (Nilsson et al., 1994) have been characterised previously. Cells expressing SialylT either alone or together with TGN38 were characterised by western blotting. As shown in Fig. 2 (lanes 1 and 2), Golgi membranes from either cell line expressed a single protein of 54 kDa. This is higher than the 47 kDa reported for the protein from rat liver (Weinstein et al., 1987) and presumably reflects increased gly-cosylation in HeLa cells. The converse was true for TGN38 in the TGN38/SialylT-HeLa cells. A single protein of 58 kDa was expressed (Fig. 2, lane 3), lower than the 85-95 kDa reported for the heterogeneously sialylated protein from NRK cells (Luzio et al., 1990).

Immunofluorescence microscopy of these two cell lines also gave the expected pattern. SialylT was localised to a compact reticulum on one side of the nucleus in both SialylT-HeLa cells (Fig. 3A) and TGN38/SialylT-HeLa cells (data not shown). TGN38 was present in the same structure but also in punctate structures throughout the cell cytoplasm which likely represent peripheral endosomes (Ponnambalam et al., 1994) (Fig. 3B).

The expression levels relative to endogenous protein were estimated in one of two ways and are summarised in Table 1. For NAGT I and SialylT the activity of the enzyme was measured and compared to the expression level of the endogenous protein in the parental HeLa cell line. Mann II activity could not be estimated in the same way because there were too many contaminating activities in whole cell homogenates. The level was, therefore, estimated by immuno-gold microscopy and compared with the level of the endogenous protein in NRK cells. The same procedure was used for TGN38 which has no known activity that could be measured.

The level of over-expression varied widely (Table 1). At the lower end was TGN38 (2.5-fold), NAGT I (4-fold) and Mann II (6-fold). At the higher end was SialylT which was expressed 10-fold over endogenous levels in the TGN38/SialylT-HeLa cell line and 50-100-fold in the SialylT-HeLa cell line. Interestingly the difference in expression levels between these two cell lines had no effect on the distribution of the protein within the Golgi apparatus. As shown in Table 2, 68% of the SialylT was present in the TGN in the TGN38/SialylT-HeLa cell line compared with 70% in the SialylT-HeLa cell line. Similar results were also obtained when the Mann II-HeLa cell line was compared with another clone expressing the protein at a 3-fold lower level (2-fold over NRK cells) (data not shown).

The effect of expressed proteins on endogenous Golgi proteins was determined by measuring the distribution of endogenous GalT within the Golgi apparatus by immuno-gold microscopy. In the parental cell line, 33±14% of the GalT was present in the stack, the rest in the TGN. As shown in Table 2 this was not changed significantly by stable expression of any of the proteins. The level of GalT in the stack in the stable cell lines ranged from 28 to 34%. Furthermore, the level of GalT in each of the stable cell lines was not affected by expression of any of the other proteins since the total number of gold particles/Golgi apparatus only varied between 34 and 40 (Table 2).

An extensive series of experiments was carried out to establish the distribution of the four Golgi enzymes and TGN38. One of them, GalT was used as the reference marker for each of the others. We had earlier shown that the distribution of NAGT I overlapped that of GalT (Nilsson et al., 1993); Mann II was found to overlap the distribution of GalT in the exactly same way (Fig. 4A) showing, by inference, that it had the same distribution as NAGT I. In contrast, Sialyl T had exactly the same distribution as GalT (Fig. 4B).

To provide a more quantitative measure of the distribution, we performed single labelling of all the cell lines and applied a Stereological method described earlier (Nilsson et al., 1993) in which the trans-most, or T cisterna of the stack was defined as the last continuous cisterna on the side of the stack that labelled for GalT. This cisterna was used as the reference point for all other compartments. The T-1 and T-2 cisternae most likely represent the medial and cis cisternae, respectively. This is because there are typically three cisternae in the stack in HeLa cells (Nilsson et al., 1993). The T-3 compartment likely represents the CGN but was not readily identified in many cross sections and was, for the most part, unlabelled for any of the Golgi proteins under study. It was not considered further.

The T+1 compartment represents the TGN which is a pleomorphic structure comprising flattened, cisternal elements abutting the trans cisterna linked to extensive tubulo-reticular elements that emanate a considerable distance from the stack. This structure was quantitated as described in Materials and Methods.

Labelling for NAGT I was restricted almost exclusively to the Golgi stack with little label over the TGN or other membrane compartments (Fig. 5A). More than 65% of the labelling (Fig. 6A) was present over two adjacent cisternae on the trans side of the stack (Nilsson et al., 1993). Mann II was also present over two cisternae on one side of the stack (Fig. 5B) which was shown to be the trans side by double-labelling for GalT (Fig. 4A). More than 75% of the labelling was present over the medial and trans cisternae (Fig. 6A).

The linear density of labelling for both NAGT I and Mann II in the medial and trans cisternae was at least 2 times higher than in the cis cisterna and 5-7 times higher than in the TGN (Fig. 6B).

Labelling for both SialylT (Fig. 5C) and GalT was restricted to the trans cisterna and the TGN. There was little labelling over the medial or cis cisternae. About 20% of labelling for GalT and SialylT was present in the trans cisterna and about 70% in the TGN (Fig. 6A). The co-distribution of these two enzymes was confirmed by double-labelling SialylT-HeLa cells for both GalT and SialylT (Fig. 4B).

Quantitation showed that the linear density of both GalT and SialylT in the trans cisterna was about 4.5 times that in the medial cisterna and about 20 times that in the cis cisterna (Fig. 6B). The average linear density in the TGN was lower than that in the trans cisterna even though it contained about 70% of both enzymes. This is because the TGN has a much longer membrane length.

TGN38 was originally described as a marker for the TGN in NRK cells (Luzio et al., 1990) and, when expressed either transiently at low levels (Ponnambalam et al., 1994) or stably (Fig. 5D) in HeLa cells, it is also present almost exclusively in the TGN. As shown in Fig. 6A, fully 90% of TGN38 was present in the TGN in the TGN38/SialylT-HeLa cell line. The linear density in the TGN was 7 times higher than that in the trans cisterna and 30 times higher than in the medial cisterna. None could be detected in the cis cisterna (Fig. 6B).

The work described in this paper both confirms and extends our earlier observations on the overlapping distribution of NAGT I and GalT (Nilsson et al., 1993) to include both Mann II and SialylT. The four enzymes fell into two groups: NAGT I co-distributed with Man II whereas GalT co-distributed with SialylT. The overlap was in the trans cisterna which had all four enzymes.

As before, considerable care was taken to ensure that the stably expressed protein had the same distribution in HeLa cells as the endogenous protein and did not alter the distribution of other Golgi enzymes. First, the origin of the cDNAs was either human (NAGT I, SialylT) or a closely related mammal (Mann II - murine, TGN38 - rat). The sequence similarity between Golgi enzymes from different mammals is typically in excess of 90% (for review, see Kleene and Berger, 1993). Second, the epitope tag, when present, was placed at the C-terminus, as far away as possible from the membrane-spanning domain that contains the signal for retention (for references, see Machamer, 1993). Third, the level of expression was in all but one case less than or equal to 10 times the level of the endogenous protein. Since none of the Golgi enzymes constitute more than a few per cent of Golgi membrane, such an increase could reasonably be expected to have a minimal effect on the distribution of the protein. In fact, the one exception showed that considerable over-expression had no effect on the distribution. SialylT was expressed at 10 times the endogenous level in TGN38/SialylT-HeLa cells but at 50-100 times in SialylT-HeLa cells, yet the distribution of SialylT between the trans cisterna and the TGN was almost exactly the same in both cell lines (Table 1). Lastly, the distribution of GalT was checked in each of the stable cell lines. The results showed clearly that none of the stably expressed proteins affected the distribution of at least this one Golgi protein.

The co-distribution of NAGT I and Mann II is in agreement with earlier work. Relocation of NAGT I to the ER by attachment of an ER retrieval signal causes accumulation of Mann II in the ER, and relocation of Mann II has a similar effect on NAGT I pointing to a very specific association between these two enzymes (Nilsson et al., 1994). When rat liver Golgi stacks are extracted with Triton X-100, most of the NAGT I and Mann II remain in the Triton pellet whereas most of the markers from other parts of the Golgi are released into the supernatant (Slusarewicz et al., 1994). Even earlier work showed that NAGT I and Mann II co-fractionated on sucrose gradients as did GalT and SialylT, but at a lower density of sucrose (Dunphy and Rothman, 1983; Goldberg and Kornfeld, 1983). Interestingly, the two sets of peaks overlapped in agreement with the overlap of all four enzymes in the trans cisterna. At that time the aim was to show that Golgi enzymes were in separate compartments so the overlap was either ignored or put down to a limitation in the resolution of the technique.

The co-distribution of GalT and SialylT agrees with most but not all published work. Though early biochemical studies showed that both enzymes co-fractionated on sucrose gradients (Dunphy and Rothman, 1983), studies on recycling proteins suggested that they were in different compartments (Duncan and Kornfeld, 1988; Huang and Snider, 1993). This discrepancy might reflect the fact that CHO cells were used for these experiments and there is evidence in this cell line that GalT and SialylT are in different compartments (Chege and Pfeffer, 1991). It will be important to confirm this using the approach we have used for HeLa cells.

Earlier microscopic studies, both immuno-gold and immunofluorescence, also suggested that GalT and SialylT were present in different compartments (see Berger et al., 1993, for references) but this discrepancy was due, at least in part, to the use of antibodies that recognised oligosaccharides as well as the enzyme polypeptide chain. The most recent immunofluorescence data show that GalT and SialylT co-localise almost entirely at the immunofluorescence level (Berger et al., 1993) and the data presented in this paper, also using antibodies to the polypeptide chain of GalT, show exact co-localisation by high resolution immuno-gold microscopy.

The mechanism that generates the overlapping distribution is unclear but one possibility relates to the recent finding that selective re-distribution of NAGT I and Mann II from the Golgi to the ER resulted in the disappearance of the Golgi stack (Nilsson et al., 1994). This strongly suggests that these enzymes (and perhaps others in the same cisternae) are involved in maintaining the structure of the stack. A simple model would be for these enzymes to interact with each other across the intercisternal space which means that they would have to be present in adjacent cisternae. The overlapping distribution of Golgi enzymes would, therefore, reflect the stacking mechanism of Golgi cisternae. A simple functional consequence of this arrangement is that transported proteins would meet the same set of enzymes twice. If they failed to be modified the first time, there would be a second chance.

Though the majority of enzyme was present in two adjacent compartments there was often significant amounts in the flanking cisternae. One possibility is that these enzymes represent that portion of the protein being recycled. No retention mechanism is perfect and enzymes might be expected to leak into transport vesicles. These might then be recycled as has been shown for both soluble and membrane proteins of the ER (for review, see Nilsson and Warren, 1994). There are several Golgi proteins that appear to recycle though the retrieval signal has not been identified (Alcalde et al., 1994; Johnston et al., 1994). Another possibility is that overlapping enzymes interact weakly with each other. As an example, relocation of NAGT I to the ER not only caused re-location of Mann II but also some of the GalT (Nilsson et al., 1994). The presence of some GalT and SialylT in the medial cisterna might then be explained by such an interaction. One last possibility is that the distribution is the consequence of a retention mechanism based on the increasing thickness of the lipid bilayer through the Golgi stack (Bretscher and Munro, 1993). Golgi enzymes would move until the length of the spanning domain matched the thickness of the bilayer. The increment in thickness might, however, be sufficiently small to permit accumulation in several adjacent and flanking cisternae.

Continued mapping of the Golgi apparatus in HeLa cells has strengthened the idea that it is a precisely defined structure. The Golgi enzymes involved in the construction of complex, N-linked oligosaccharides are not restricted to single cisternae but, so far, they are restricted to two adjacent ones. The overlapping distribution may reflect the structural organisation of the stack. It is, however, clear that the distribution of Golgi enzymes does vary from tissue to tissue and from organism to organism (Roth et al., 1986; Velasco et al., 1993). This suggests that the mechanism governing the localisation of Golgi enzymes is subject to another layer of control. The most obvious would be post-translational modifications such as phosphorylation, and work is presently underway to test this possibility.

We are indebted to Dr Paul Luzio (Cambridge, UK) and Dr George Banting (Bristol, UK) for kindly providing us with the cDNA and antibodies for TGN38; Dr Thomas Kreis (Geneva, Switzerland) for the P5D4 hybridoma; Dr Kelley Moremen (Athens, Georgia) for Mann II polyclonal antibodies; and Dr Sean Munro (Cambridge, UK) for cDNA encoding SialylT. We also thank the oligosynthesis facility at Clare Hall for high quality oligonucleotides; the photography department for high quality reproductions; and Drs Tom Misteli, Vas Ponnambalam and Francis Barr for critical comments and helpful discussions.