key: cord-104231-fi8pskod
authors: nan
title: The TGN38 glycoprotein contains two non-overlapping signals that mediate localization to the trans-Golgi network
date: 1994-04-02
journal: J Cell Biol
DOI: nan
sha: 
doc_id: 104231
cord_uid: fi8pskod

The membrane-spanning and cytoplasmic domains of CD4 and CD8 were replaced by those of TGN38. After transient expression in HeLa cells, the location of the hybrid proteins was determined using immunofluorescence and quantitative immuno-electron microscopy, FACS analysis and metabolic labeling. The membrane-spanning domain was found to contain a signal that localized hybrid proteins to the TGN. This was in addition to the signal previously identified in the cytoplasmic domain (Bos, K., C. Wraight, and K. Stanley. 1993. EMBO (Eur. Mol. Biol. Organ) J. 12:2219-2228. Humphrey, J. S., P. J. Peters, L. C. Yuan, and J. S. Bonifacino. 1993. J. Cell Biol. 120:1123-1135. Wong, S. H., and W. Hong. 1993. J. Biol. Chem. 268:22853-22862). The different properties of these two signals suggest that each operates by a different mechanism.

motif results in the retrieval of soluble ER proteins lost from this compartment (Munro and Pelham, 1987) . This sequence is recognized by a receptor which returns the protein to the ER (Semenza et al., 1990; Lewis and Pelham, 1992) . Some membrane proteins contain a double-lysine motif in the cytoplasmic tail which ensures retrieval to the ER (Jackson et al., 1993) .

Golgi retention signals were first identified in a viral protein (Swift and Machamer, 1991) and in Golgi glycosylation enzymes (Nilsson et al., 1991; Munro, 1991; Aoki et al., 1992; Burke et al., 1992; Colley et al., 1992; Tang et al., 1992; Teasdale et al., 1992; Wong et al., 1992) . In all cases, the retention signal lies in the membrane-spanning domain. More recently, membrane-spanning domains have been implicated in the localization of two proteins to the nuclear envelope (Wozniak and Blobel, 1992; Smith and Blobel, 1993) .

So far, none of the proteins studied have been shown to contain both a retention and a retrieval signal though there are indications that more than one signal is present in some of them. When the carboxy-terminal -KDEL signal is removed from the soluble BiP protein, it is secreted very slowly (Munro and Pelham, 1987) . When the double-lysine motif is removed from glucuronosyltransferase, an integral membrane protein, little if any escapes from the ER (Jackson et al., 1993) . Both these cases suggests that a retention signal is present in addition to the retrieval signal.

To determine whether a protein can contain both signals, we focused on TGN38, a resident of the trans-Golgi network (Luzio et al., 1990) . TGN38 has a tyrosine-based motif in the cytoplasmic domain that acts as a retrieval signal, return-ing the protein from the cell surface to the TGN (Bos et al., 1993; Humphrey et al., 1993; Wong and Hong, 1993 ). Here we have looked for the second signal and have found it in the membrane-spanning domain.

Chemicals and reagents were obtained from the sources described by Nilsson et al. (1993a Nilsson et al. ( , 1994 . Restriction enzymes, T4 DNA ligase, and polynucleotide kinase for recombinant DNA procedures were purchased from GIBCO BRL (Gaithershurg, MD), New England Biolahs (Bethesda, MD), or Pharmacia Corp. (Uppsala, Sweden) . Sequencing reagents and T7 polymerase were purchased from USB Biochemicals. Cell culture media, fetal calf serum, amino acids, and penicillin/streptomycin were from GIBCO BRL. 

The TGN38 eDNA (Lnzio et al., 1990) was mutagenized by PCR such that a BamHI site was introduced just after the stop codon. The HindIII-BamHI fragment containing the 5' untranslated region and the complete coding sequence was cloned between the same restriction sites in the pCMUIV plasmid (Nilsson et al., 1989) . The TGN38 eDNA in the mammalian expression vector pUEX (Luzio et al., 1990 ) was used as a template for PCR amplification under the conditions described by Salki et al. (1988) .

Oligonucleotides used in PCR reactions were generally synthesized with BamHI flanking sequences to facilitate easy cloning into plasmid vectors for sequencing. Oligonucleotides were used at I mM in PCR reactions. PCR products were ethanol precipitated, digested with BamHI, and cloned into the pBS plasmid (Stratagene Corp., La Jolla, CA) for sequencing. DNA sequencing was carried out using the dideoxy chain termination method and T7 polymerase (Tabor and Richardson, 1987) . Plasmid derivafives were digested with the appropriate enzymes to generate the correct 5' and 3' ends for cloning and construction of CD8 and CD4 eDNA chimeras.

CD8 hybrid proteins were constructed such that the first 153 amino acids of the extracellular domain of c~CD8 were preserved. This is sufficient for CD8 glycosylation and secretion (~y et al., 1992) , and should also not affect the ability of CD8 hybrid proteins to form homodimers (Leahy et al., 1992; Boursier et al., 1993) . PCR-generated DNA fragments were cloned into the pCMUIV expression vector (Nilsson et al., 1989) using 3 DNA fragments in a llgation. To construct the CD8-MC and CD8-MAC chimeric cDNAs the llgation mix contained (1) the HindHI-ApaLI DNA fragment that encoded most of the lumenal domain of the CD8 molecule, (2) an ApaLI-BamHI DNA fragment that encoded the membrane-spanning and cytoplasmic domain sequences of TGN38, and (3) the pCMUIV vector as a HindIlI-BamHI DNA fragment.

To construct the CDS-M membrane-spanning chimeric eDNA the ligation contained (1) the HindllI-ApaLI DNA fragment that encoded most of the lumenal domain of the CD8 molecule, (2) an ApaLI-SalI DNA fragment that contained the membrane-spanning domain of TGN38, and (3) the pCMUIV plasmid vector as a HindIII-SalI DNA fragment which also contalned the cytoplasmic domain of the CD8 molecule.

To construct the CD8-C and CD8-AC chimeric cDNAs the ligation contained (1) a SalI-BamHI DNA fragment that contained the cytoplasmic domain of TGN38, and (2) the pCMUIV vector as a BamI-II-SalI DNA fragment which also contained the lumenal and membrane-spanning domains of the CD8 molecule.

Cloning of the CD4 eDNA into the pCMUIV plasmid has been described by Nilsson et al. (1989) and we used this CD4 plasmid initially. However, there were no unique restriction sites in the CD4 eDNA at equivalent positions which would have allowed us to insert replacement membranespanning and cytoplasmic TGN38 sequences. We thus inserted unique restriction sites within the CD4 eDNA using PCR mutngenesis. Deen et al. (1988) have shown that a CD4 molecule terminating after residue 371 in the extracellular domain is efficiently secreted. As the threonine at 369 is followed by a proline residue, this would make it unlikely that there are any structural features after thr-369 necessary for correct folding and transport of CD4. Codon 367 (Maddon et al., 1985) in the CD4 eDNA (which encodes a serine residue) was mutated from TCC to TCA. This silent mutation created a single HincII site within the CD4 eDNA. Similarly, codons 400 and 401 (encoding arginine residues) were mutated from CGA AGG to CGT CGA. These silent mutations created a unique SalI site in the CD4 eDNA. There were now unique restriction sites on either side of the membranespanning region in the eDNA. PCR was carried out on the TGN38 eDNA using ollgonucleotides similar to those described above. Instead of the ApaLI site used in CD8 chimera constructions, 5' oligonucleotides that precede the TGN38 membrane-spanning domain were synthesized with a HincH site. A HincH-SalI or HincH-BamHI DNA fragment containing TGN38 sequences could be fused to CD4 by ligation as described previously. We thus ligated either a HindlII-HincII or HindIII-SalI DNA fragment (encoding the lumenal domain of the CD4 molecule) from pCMUIV-CD4, an appropriate PCR generated DNA fragment that encoded TGN38 sequences, and an appropriate pCMUIV vector. Ligations were transformed into E. coil strain JM101 using standard procedures (Sambrook et al., 1989 ) and clones were analyzed by restriction digestion followed by DNA sequencing. Large-scale preparations of the pCMUIV plasmids (for transfections) were made by alkaline lysis followed by CsCl-ethidium bromide purification (Sambrook et al., 1989) .

Monolayer HeLa cells (ATCC CCL185) were grown in DMEM supplemented with 10% FCS, non-essential amino acids, penicillin/streptomycin, and 2 mM glutamine. Cells were grown to 20-30% conttuency before transfection by the calcium phosphate method essentially as described by Nilsson et al. (1989) . For immunottuorescence, cells were seeded out into 30-ram, 6-weil plates (Becton Dickinson, Lincoln Park, NJ) containing sterile coverslips. After one day, cells were transfected with 2 ~g of plasmid DNA (CsCI-EtBr purified) and 3/tg of pUC19 plasmid DNA (used as a carrier to increase transfection efficiency). Cells were fixed and processed for immunottuorescence 60-72 h later (Nilsson et al., 1989; Jackson et al., 1990) . Transfected cells for electron microscopy and immunoprecipitation were prepared as described by Nilsson et al. (1989) in 100-mm tissue culture dishes and processed 60-72 h after DNA was added.

Cells were processed for immunofluorescence microscopy as described previously (Warren et al., 1984; Nilsson et al., 1991) . The rabbit antiserum to TGN38 has been described by Luzio et al. (1990) and was diluted (1:500) in 0.2% fish skin gelatin/PBS before use. GaIT was detected using either a rabbit polyclonal antiserum (Roth and Berger, 1982; Nilsson et al., 1993a ) at a dilution of 1:200 or the monoclonal antibody GT-2 (Berger et al., 1986) . Culture supernatants from the QS4120 (Healey et al., 1990) and OKT8 (ATCC CRL 8014) mouse bybridomas were used to detect CD4 and CDS, respectively. The secondary antibodies were horse anti=mouse conjngated to Texas red (Vector Labs, Burlingame, CA) and sheep anti-rabbit conjugated to FITC (Dakopatts, Copenhagen) and both were used at a dilution of 1:100. P-,hodamine-conjngated transferrin (Molecular Probes, Eugene, OR) was dissolved in PBS and used at a final concentration of 20 ~tg/ml; ceils were incubated in fresh media containing the ligand for 30 rain at 37°C before washing with PBS and fixation. Slides were examined on a Zeiss Axiophot fluorescence microscope using a 63)< objective oil immersion lens and a 10)< eyepiece.

Transfected cells were fixed in 2 % paraformaldehyde/0.2 % glutaraldehyde in 0.1 M phosphate buffer pH 7.4 for 3 h and processed for cryo-immunoelectron microscopy essentially as described by Rabouille et al. (1993) . Cell pellets embedded in gelatin were cut into blocks and infused with 2.3 M sucrose overnight. The blocks were frozen in liquid nitrogen, ultrathin cryosections were cut and incubated for 30 rain with primary antibodies. Polyclonal rabbit antiserum to TGN38 (Lnzio et al., 1990 ) was used at a dilution of 1:100 and an affinity-purified rabbit antibody to CD4 (Dr. Mark Marsh, MRC Centre, University College London) at a dilution of 1:100. Both were diluted into 0.5% fish skin golatin/PBS and the bound antibodies were detected using protein A-coupled to 10-nm gold particles used at a dilution of 1:50 (from Cell Biology Department, Ultrecht School of Medicine, Utrecht, Netherlands). The monoclonal antibody 14 to CD8 (Haynes, 1986) was used neat as a culture supernatant and bound antibodies were detected using a secondary goat anti-mouse antibody coupled to 10-rim gold particles at a dilution of 1:20 (Nilsson et al., 1994) . Immunolabeling was followed by staining with uranyl acetate and embedding in methyl cellulose as described by Tokuyasu (1980) . Quantitation of immunolabeling was carried out by first identifying a labeled area of the cell where profiles of the Crolgi stack and TGN were readily visible. The TGN was defined morphologically as described in the Results. Labeled cell sections were placed in one of two categories. Cell sections in which no eodosomes were labeled consistently contained fewer than 50-60 gold particles whereas cell sections containing labeled endosomes contained more than 50-60 gold particles. In each experiment, 200-1,000 gold particles were counted depending on the type of construct used for transfection (see Fig. 6 legend). The background observed in each experiment was extremely low and therefore ignored.

The relative distribution of immunolabeling was performed by counting the gold particles over the membranes (Golgi stack, TGN, endosornes, and plasma membrane) of selected cells. The linear density of the gold particles at the plasma membrane was estimated by the point hit method. Briefly, pictures were printed at 38,750× or 52,000× final magnification and overlaid with a grid containing a point-to-point spacing of 2 crn (plasma membrane) or 0.5 cm (TGN) (d). The number of intersections (I) of the membrane with this grid were counted. The total number of gold particles over the total membrane was counted and the linear density was then calculated using the formula:-Gold particles/(I X d/magnification).

Transfected cells were grown in 30-ram wells and processed for FACS analysis exactly as described by Nilsson et al. (1989) . Native and hybrid CD8 proteins were detected using the culture supernatant from the OKT8 hybridoma. Approximately 5 × 105 cells were incubated with the primary antibodies on ice for 20 rain followed by gentle washing. The secondary antibody was FITC-conjugated sheep anti-mouse (Vector Labs., Burlingame, CA) used at a dilution of 1:100. After washing, cells were fixed in 3% paraformaldehyde/PBS for 20 rain on ice. Immediately before FACS analysis, cells were washed three times with calcium/magnesium-free PBS and resuspended in 0.5 ml of this solution. A Becton-Dickinson FACScan 440 machine was used to analyze the samples. Quantitation of FAnS data was performed by calculating two parameters: the amount of positive cells expressing each protein and the arithmetic mean intensity of fluorescence of these positive cells (~:Ni.Ii2/~Nili). The mean intensity was multiplied by the number of positive cells which gave the total fluorescence intensity (arbitrary units) detected for the positive cell population. The total fluorescence intensity of the CD8 positive cells was represented as 100% and the total fluorescence intensity of positively gated cells expressing other CD8 hybrid proteins were expressed as a percentage of total CD8.

Immunopreeipitation was carried out essentially as described by Jackson et al. (1989) . Transfected cells in 100-mm dishes were labeled with 0.2 rnCi [35S]methionine/cysteine for 20 rain and either lysed in detergent immediately or chased with medium containing excess cold methionine and cysteine. After 15 rain on ice the cells were scraped off and lysates centrifuged at medium (12,000 g, 10 min at 4°C) and high speed (400000 g, 30 rain at 40C) to remove particulate material. The supernatants were precleared by incubation with mouse IgG bound to protein A-Sepharose (Pharmacia Corp., Uppsala, Sweden) and native and hybrid proteins were immunoprecipitated using QS4120 (anti-CD4) or OKT8 (anti-CDS) antibody bound to protein A-Sepharose. Immunoprecipitates were then washed sequentially twice with lysis buffer, once with lysis buffer containing 0.1% SDS, once with lysis buffer containing 0.5 M NaCI, twice with lysis buffer, and finally once with low ionic strength buffer (20 rnM Tris pH 7.5, 0.1% TX-100). After the final wash, immunoprecipitates were divided equally and resuspended in 15 #1 of neuraminidase digestion buffer (300 mM NaC1, 100 ram sodium acetate pH 5.5, 14 mM calcium chloride and I mM PMSF) containing either 25 mU neuraminidase (Type VIII; Sigma) or buffer (mock treatment). Samples were incubated at 37"C for 4 h before analysis by SDS-PAGE and fluorography.

We have used both CD8 and CIM as the reporter molecules.

These glycoproteins are members of the Ig superfamily and are normally expressed on the surface of T-lymphocytes Maddon et al., 1985) . The ~CD8 polypeptide consists of 214 amino acids (Littman et ai., 1985) and CD4 of 435 residues (Maddon et al., 1985) . The amino acid sequences of the membrane-spanning and cytoplasmic domains of CD8 and CD4 are shown in Fig. 1 A. Rat TGN38 is a type I glycoprotein with a predicted molecular weight of 38 kD but glycosylation increases this to 85-95 kD (Luzio et al., 1990) . Signal peptide cleavage produces a mature protein of 340 residues. The large lumenal domain (residues 1-287) is followed by a single membrane-spanning domain (residues 288-306) and a short cytoplasmic domain (residues 307-340; see Fig. I A) .

Chimeric cDNAs were constructed by replacing domains in CD4 and CD8 with those in TGN38 at identical amino acids in the three proteins (see Fig. 1 A) . This meant that hybrid proteins containing the membrane-spanning domain of TGN38 also included some of the flanking amino acids. Flanking sequences have been shown to influence substantiaily the ability of membrane-spanning domains to localize proteins to the Golgi apparatus (Nilsson et al., 1991; Munro, 1991; Colley et al., 1992) .

Rat TGN38 eDNA (Luzio et al., 1990) was cloned into the pCMUIV expression vector (see Materials and Methods) and the plasmid introduced into cells by calcium phosphate transfection. At 72 h after transfection, immunofluorescence microscopy showed that TGN38 was localized to a juxtanuclear reticulum (Fig. 2 A) very similar to that stained by antibodies to GaIT (Fig. 2 B) , a resident of the trans-Golgi cisterna (Roth and Berger, 1982) and the TGN (Lucocq et al., 1989; Taatjes et ai., 1992; Nilsson et al., 1993a) . Higher levels of TGN38 led to an accumulation of the protein in punctate structures in the cytoplasm (Fig. 2 A) and at the cell surface (data not shown; see later). These data are consistent with the localization of endogenous TGN38 in NRK cells and expressed protein in transfected CV-1 cells (Luzio et al., 1990) .

The membrane-spanning (M) and cytoplasmic (C) domains of TGN38 were grafted on to the lumenal domain of CD8 (see Fig. 1 B) . CD8 was mostly present on the cell surface though a small amount was present in small punctate structures, especially at high levels of expression (Fig. 2 C) . The hybrid protein (CD8-MC), however, was restricted to a compact juxtanuclear reticulum (Fig. 2 D) that was very similar to that stained by antibodies to GaIT (data not shown). The membrane-spanning and cytoplasmic domains of TGN38 were also grafted on to the lumenal domain of CD4. Transiently expressed CD4 was mostly localized to the plasma membrane (Fig. 2 E) , but the CD4-MC hybrid protein was again localized to a juxtanuclear compartment (Fig.  2 F) , very similar to that stained by antibodies to GaIT (data not shown). There was also faint staining of the nuclear envelope, suggesting some accumulation of both hybrid proteins within the ER (arrows in Fig. 2 , D and F).

To show that the membrane-spanning and cytoplasmic domains of TGN38 contained distinct localization signals, each of these domains in both CIM and CD8 was replaced by that domains of CD8, CD4, and TGN38. The membrane-spanning domain of CIM is that defined by Maddon et al. (1985) although it could inelude an additional two amino acids (-CV-) on the cytoplasmic side of the membrane. Amino acid residues at which domains were replaced in the different hybrid proteins are numbered. (B and C) Line sketches show CDS, CIM, and hybrid proteins containing the membrane-spanning and cytoplasmic domains of TGN38 (shaded and boxed). Hybrid proteins are defined by suffixes which denote the domains of CD8 or CD4 that were replaced by equivalent domains of TGN38. M = membranespanning domain and flanking residues; C = cytoplasmic domain; and AC = truncated cytoplasmic domain lacking 21 residues from the carboxyterminus.

in TGN38 (Fig. 1, B and C). The TGN38 membranespanning domain alone localized the hybrid proteins to the Golgi apparatus. Both CD8-M (Fig. 3 A) and CD4-M (Fig.  3 E) gave a discrete juxtanuclear staining pattern very similar to that for GAIT, especially at low levels of expression (Fig. 3 , B and F,, respectively). At higher levels of expression, staining was more diffuse suggesting accumulation in vesicles and at the cell surface (data not shown). This pattern was more frequently observed for CD4-M than for CDS-M.

The cytoplasmic domain of TGN38 contained, as expected, the other Golgi localization signal (Bos et al., 1993; Humphrey et al., 1993; Wong and Hong, 1993) . Both CD8-C (Fig. 3 C) and CIM-C (Fig. 3 G) were detected in a juxtanuelear reticulum very similar to that containing GaIT (Fig. 3, D and H, respectively) . At all levels of expression staining of small punctate structures (Fig. 3, C and G) was also observed and these were shown to be endosomes by immunoelectron microscopy (see below).

Replacing the membrane-spanning domain of either CIM or CD8 with that of TGN38 could have activated or corrupted a localization signal in the cytoplasmic domain of the hybrid proteins. CD8 has a tyrosine residue in the cytoplasmic domain but it is the penultimate amino acid that would not norreally be expected to operate as a retrieval signal (Fig. 1 A) . CD4 contains a signal for endocytosis (Shin et al., 1990 (Shin et al., , 1991 but this does not normally direct the protein to the TGN. Nevertheless, to eliminate the possibility of activating or corrupting a cytoplasmic signal, we exploited the fact that the localization signal in the cytoplasmic domain of TGN38 has been mapped to the sequence, SDYQRL (Bos et al., 1993; Humphrey et al., 1993; Wong and Hong, 1993) . Comparison of the hybrid protein CD8-MC (Fig. 4 A) with CD8-MAC (Fig. 4 B) , which lacked this sequence, showed that it had no significant effect on the localization of the protein to the Golgi apparatus. Both were localized to a juxtanuclear reticulum very similar to that stained by antibodies to GaIT (data not shown). Expression of the CD4-MAC hybrid protein gave similar results (Fig. 4 E) .

To show that we had in fact eliminated the localization signal in the cytoplasmic domain of TGN38, we expressed a CD8 or CD4 hybrid protein containing the truncated cytoplasmic domain from TGN38 (CD8-AC or CD4-~C; Fig. 1  B) . Unlike CD8-C (see Fig. 3 C) , the CD8-AC hybrid protein was not in the Golgi apparatus but exhibited a cell surface staining pattern and additional accumulation in intracel- * The two categories of transfected cells, low "expressers" (<) and high "expressers ~ (>), were defined as described in Materials and Methods. The total number of gold particles counted was 1695 CrGN38), 1036 (CD4-M), and 1302 (CD4-C). * The linear density at the plasma membrane and in the TGN was estimated for the expressed proteins as described in Materials and Methods. The number of gold particles counted as the plasma membrane was 300 (TGN38), 70 (CIM-M), and 60 (CD4-C). Note that for TGN38 and CIM-M, gold labeling at the plasma membrane correlated with labeling of endosomes. lular punctate structures (Fig. 4 C) . Some of these were shown to be early endosomes (arrowheads in Fig. 4 , C and D) by preincubating the cells with rhodamine-conjugated transferrin for 30 min before fixation and processing. CD4-AC was found to be expressed largely on the cell surface of transfected cells (Fig. 4 F) .

To determine the precise location of the hybrid proteins, cryosections of transfected cells were labeled with antibodies which were detected by appropriate gold conjugates. The TGN is defined morphologically as a tubulo-reticular network, part of which is closely apposed to the trans side of the Golgi stack. Budding and vesicle profiles coated by clathrin are often observed (Orci et al., 1984 (Orci et al., , 1985 Griffiths and Simons, 1986) . Transiently expressed TGN38 was found in the TGN, the Golgi stack, endosomes, and the plasma membrane (Fig. 5 A) . The distribution of TGN38 depended on the level of expression. At low levels (<60 gold particles/cell section) TGN38 was found exclusively in the Golgi apparatus (Table I) with most of the protein (85 + 7%) being found in the TGN (Fig. 6) . None was detected in either endosomes or on the plasma membrane (Table I) . At higher levels of expression (>60 gold particles/cell section) the percentage of labeling over the Golgi apparatus fell (to 60 5:7 %) and rose over endosomes (to 29 5:7 %) and the plasma membrane (to 11 + 7%) ( Table I) . Even at high levels of expression the linear density of labeling over the TGN was four times that over the plasma membrane showing that the protein was still concentrated in the TGN. Hybrid proteins containing both localization signals gave patterns of labeling almost indistinguishable from TGN38 both morphologically and quantitatively. Both CD8-MC (Fig. 5 B) and CD4-MC (Fig. 5 D) were present only in the Golgi apparatus at all levels of expression and more than 80 % of the label was present in the TGN (Fig. 6) .

Hybrid proteins containing only the membrane-spanning domain also gave results very similar to those for TGN38. CD4-M was present only in the Golgi apparatus at low levels of expression (<50 gold particles/cell section; Table I ) with 80 5:8 % being present in the TGN (Fig. 6) . At higher levels of expression (>50 gold particles/cell section) the percentage of total labeling over the Golgi apparatus fell (to 61 5: 8%) and rose over endosomes (to 35 + 8%) and plasma membrane (to 4 + 3%) ( Table I) . This distribution was not affected by pretreatment of the cells for 2 h with cycloheximide (Fig. 5 F) . The linear density of labeling over the TGN was 11 times that over the plasma membrane, even higher than that for TGN38 (Table I) . The other hybrid protein, CDS-M, was also present exclusively in the Golgi apparatus (Fig. 5 C) and most (66 5: 8%) was present in the TGN (Fig.  6) . However, more of the protein was present in the Golgi stack than for other hybrid proteins. We also examined cells expressing CD4-M after incubation in cycloheximide for 7 h and found no significant change in the pattern of labeling (data not shown).

The CD4 hybrid protein containing the cytoplasmic domain (CD4-C) was also localized using immunoelectron microscopy. CD8-C could not be detected by immuno-electron microscopy for reasons that are still unclear. More than 84 % of the CD4-C in the Golgi apparatus was present in the TGN (Fig. 6) The Journal of Cell Biology, Volume 125, 1994 5 E). However, CD4-C differed from other hybrid proteins and TGN38 in three respects. First, more than 35 % of the total protein was present in endosomal structures at low levels of expression and this increased to nearly 60 % at high levels. Second, the type of endosomal structure labeled also changed. At low levels, all of the endosomal structures were multivesicular bodies which are late endosomal compartments (Geuze et al., 1988; Hopkins et al., 1990) . At high levels of expression, labeling was also present in tubular endosomes. Because of the difficulties of distinguishing between tubular endosomes and the TGN (Geuze et al., 1988; Klumperman et al., 1993) , definitive identification of these structures will have to await further analysis using stably transfected cell lines. Third and last, the plasma membrane was labeled at both high and low levels of expression ( Table  I ). The linear density was the same (0.08 + 0.05 gold particles/#m) suggesting that CD4-C was as efficiently removed from the cell surface and delivered to endosomes. The protein was still concentrated in the TGN since the linear density was 9-24 times higher than that in the plasma membrane ( Table I) .

As a final control, CD4 was also expressed transiently in HeLa cells. All of the protein was found in endosomes and on the plasma membrane after treatment with cycloheximide and in agreement with studies carried out by Pelchen-Matthews et al. (1991) . Little if any was found in the TGN (data not shown). This shows that accumulation of the hybrid proteins in the TGN was not a consequence of the CD4 ectodomain.

FACS analysis was used to determine the amount of CD8 and CD8 hybrid proteins on the cell surface. Fixed, nonpermeabilized cells transfected with the appropriate eDNA were stained with the OKT8 antibody to CD8 and visualized using a secondary antibody conjugated to FITC (Fig. 7) . Mock-transfected cells were used as the negative control and the gate was set such that 1% of these cells scored positive.

The profile of CD8 expression was characteristic of that in a population of transiently transfected cells (Nilsson et al., 1989; Jackson et al., 1990) and 58% scored positive (Fig.  7 A, Table 1I ). When both localization signals were present (CDS-MC) only 6% of the cells showed surface staining (Fig. 7 B, Table ] I) and this rose to 19% when the cytoplasmic signal was deleted (CD8-MAC) (Fig. 7 D, Table 1I ) suggesting that the membrane-spanning signal alone could retain the hybrid protein. This was confirmed by expression of CD8-M. Only 13% of the cells scored positive (Fig. 7 C, Table II).

The cytoplasmic signal alone could also retain the hybrid protein. Only . Surface expression of CD8 and CD8 hybrid proteins analyzed using FACS analysis. HeLa cells were fixed and stained with monoclonal antibody to CD8 and visualized using a secondary antibody conjugated to FITC. The gate on the x-axis (fluorescence units) was set using mock-transfected cells as a negative control. Cells scoring to the right of this point were regarded as positive for cell surface fluorescence (see Table I1 ). * The gate was set using moek-transfeeted cells and 1% were positive. 5,000 cells were analyzed in each experiment. * The total fluorescence intensity was calculated by integrating the area under the positive cells and expressing the results as a percentage of the value for CD8.

( Fig. 7 E, Table II ). Removal of the tyrosine-based motif in this cytoplasmic signal destroyed its ability to retain hybrid proteins. More than 67 % of cells expressing CD8-AC scored positive (Fig. 7 F,, Table II ), slightly higher than the number expressing CD8 alone (58%, Fig. 7 A, Table II ). The total fluorescence of positive cells for each protein was also calculated by integration of the results presented in Fig. 7 . These calculations, presented in Table II , are in close agreement with the percentage of cells scoring positive and provide further evidence that both the membrane-spanning and cytoplasmic signals can, independently, retain hybrid proteins inside the cell.

These data were also supported by immunofluorescence microscopy on non-permeabilized cells (data not shown). There was no surface staining of cells expressing CD8-MC or CD8-C. A few of the transfected cells expressing CD8-M showed low levels of cell surface staining. Cells expressing CD8-AC showed high levels of cell surface staining, in agreement with the FACS analysis.

CD4 contains two N-linked glycosylation sites (Maddon et al., 1985) one of which is completely processed to the complex bi-antennary sialylated structure. Terminal sialic acid residues are added by ot2,6-sialyltransferase which is located in the trans-Golgi cisterna and the TGN (Roth et al., 1985) . Neuraminidase cleaves off these sialic acid residues producing a small mobility shift on SDS-PAGE (Nilsson et al., 1989; Jackson et al., 1990 ) that can be used to show that proteins have reached this part of the Golgi apparatus. Cells expressing CIM or CD4 hybrid proteins were labeled with [35S]methionine/cysteine for 20 min (see Materials and Methods) and either lysed immediately or chased with excess cold methionine and cysteine for 2 h. Immunoprecipitates were treated with neuraminidase or mocktreated and analyzed by SDS-PAGE (Fig. 8) . Newly synthesized CIM was insensitive to neuraminidase (cf. lanes I and 2) but completely sensitive after a 2-h chase (cf. lanes 3 and 4). Hybrid CIM proteins containing either the membranespanning (CIM-M) or cytoplasmic signal (CIM-C) were also completely siaiylated after a 2-h chase period (lanes 9-16). 3, 5, 7, 9, 11, 13, and 15) or digested with neuraminidase (lanes 2, 4, 6, 8, 10, 12, 14, and 16 ) and analyzed by SDS-PAGE and fluorography. The molecular size of 14C-radioactive markers are indicated on the left hand side of the panels.

However, CD4-MC, which contains both signals, was incompletely sialylated after the 2-h chase (lanes 7and 8) suggesting that transport to the trans part of the Golgi apparatus was slower. The reason for this is unclear but was not the consequence of using CD4 as the reporter molecule. CD8-MC was also incompletely sialylated (data not shown) in contrast to hybrid CD8 proteins containing either TGN38 signal (Fig. 9) .

Metabolic labeling of CD8 hybrid proteins could not be used to study arrival in the trans cisterna and the TGN because the bound oligosaccharides are all O-linked . Sialylation of O-linked oligosaccharides is thought to occur in the medial-Golgi cisterna (Locker et al., 1992) . However, pulse-chase analysis combined with neuraminidase digestion showed that CD8-M and CD8-C hybrid proteins had kinetic properties similar to those for native CD8 (Fig. 9) . In all cases the half-time for acquisition of terminal sialic acid residues was between 25-30 min. Taken further with the results for the CD4 hybrid proteins, these results show that the presence of either signal had little, if any, effect on synthesis, transport, and processing of the reporter molecules.

TGN38 was originally identified as a resident of the TGN (Luzio et al., 1990 ) but more recent work has shown that it cycles between the TGN and the cell surface (Jones et al., 1993; Reaves et al., 1993) . Movement to the cell surface occurs via exocytic vesicles (Jones et al., 1993) and is followed by retrieval which requires a signal present in the cytoplasmic domain that has been localized to the sequence SDYQRL (Bos et al., 1993; Humphrey et ai., 1993; Wong and Hong, 1993) . Retrieved TGN38 first moves to the endosomes and then to the TGN (Bos et al., 1993; Humphrey et al., 1993) . Using plasma membrane glycoproteins as reporter molecules, we have now shown that there is an additional signal for localizing the protein to the TGN. This signal is located in the membrane-spanning domain.

Four independent lines of evidence support this conclusion. First, immunofluorescence microscopy showed that CD4-M and CD8-M hybrid proteins localized to a juxtanuclear reticulum that could also be stained by antibodies to GAIT, a marker for the trans cisterna and TGN (Roth and Berger, 1982; Taatjes et al., 1992; Nilsson et al., 1993a) . This staining pattern was not affected by pretreamaent with 10 #g/ml cycloheximide for up to 7 h (data not shown) showing that these hybrid proteins were not simply moving slowly along the exocytic pathway. In addition, it is unlikely that a cryptic localization signal in the cytoplasmic domain of either CD4 or CD8 was activated as a result of generating hybrid proteins. Both CD8-MAC and CD4-MAC hybrid proteins which lack most of the cytoplasmic domain (including the cytoplasmic signal for TGN38) were also localized to the Golgi apparatus.

Second, quantitative immunoelectron microscopy showed that CIM-M had almost the same distribution as TGN38 at both high and low levels of expression. At low levels of expression, all of the CD4-M hybrid protein and TGN38 was present in the Golgi apparatus, 80% being present within the TGN. At higher levels of expression, CIM-M and TGN38 were also present in endosomes and on the cell surface, again to a similar extent. Even at these higher levels of expression, both proteins were still concentrated in the TGN as compared to the plasma membrane. For TGN38, the linear density in the TGN was four times that in the plasma membrane; for CD4-M it was even higher, being 12 times higher in the TGN than in the plasma membrane. The distribution of CD8-M was also similar to that for TGN38 being exclusively in the Golgi apparatus at low levels of expression. There was, however, more of this hybrid protein in the Golgi stack for reasons that are still unclear.

Third, FACS analysis showed that the membrane-spanning domain reduced the percentage of positive cells from 58 % Figure 9 . Metabolic labeling of CD8 and CD8 hybrid proteins. Transfected HeLa ceils were labeled with [35S]methionine/cysteine for 20 min and chased for0, 15, 30, 45, 75, or 120 min. Immunoprecipitates were mock-digested (-) or digested (+) with neurarninidase and analyzed by SDS-PAGE and fluorography.

(for CD8) to 13% (for CD8-M), a 4.5-fold decrease. This effect was not the consequence of activating a cryptic signal in the cytoplasmic domain of CD8 since removal of most of this domain (including the TGN38 cytoplasmic signal) still reduced the percentage of positive ceils to 19% (for CD8-MAC), a threefold decrease.

Fourth and finally, metabolic labeling studies showed that all molecules of CD4-M received sialic acid, showing that they had all reached the trans cisterna/TGN. The present studies are at variance with previous studies that identified only the cytoplasmic domain of TGN38 as a localiTation signal. Removal of most of the domain led to the appearance of the truncated protein on the cell surface (Luzio et al., 1990; Bos et al., 1993) . It should be noted that the earher study was carried out before any evidence existed for retention by membrane-spanning domains, a phenomenon first described for a viral protein (Swift and Machamer, 1991) and Golgi enzymes (Nilsson et al., 1991; Munro, 1991) . The expectation was that the signal would be found in the cytoplasmic domain and this appeared to be borne out by the appearance of protein on the cell surface. The likely reason for this, however, was that the levels of expressed protein were so high that they saturated the retention mechanism that operates on the membrane-spanning domain. Even cells expressing the wild-type protein (TGN38), containing both retention and retrieval signals, showed aberrant localization. After 24 h transfection, 50% of the cells had either fallen off the coverslip or showed cell surface staining (Luzio et al., 1990) . The more recent study by Bos et al. (1993) overcame this problem by using a similar expression vector (pSVL1) at a low transfection efficiency (~0.15%) to ensure more constant levels of expression in each cell. Unfortunately, their only assay was immunofluorescence microscopy and it was impossible to determine how much of the protein was still present in the Golgi apparatus. These problems did not arise in our system since the expressed levels of hybrid protein were lower and the distributions were quantitated in several independent ways. Two different reporter molecules were also used. However, it is conceivable that other small differences in the systems used are responsible for the apparent discrepancies. Wong and Hong (1993) , for example, found that the serine residue in the SDYQRL motif was important for Golgi localization, while Humphrey et al. (1993) found that an identical mutation in their system had no effect. Further studies using stable cell lines will be needed to resolve these differences.

Though the cytoplasmic and membrane-spanning signals can act independently to localize hybrid proteins to the TGN, it is likely that they operate in different ways. The cytoplasmic domain is a clear example of a retrieval signal. This acts after a protein has left the compartment in which it norreally functions and returns them from a later compartment. All of the retrieval signals so far characterized have been located in the cytoplasmic domains of proteins and many are based on tyrosine as the critical amino acid set in a particular motif (for review see Collawn et al., 1990; Pearse and Robinson, 1990; Kornfeld, 1992; Matter et ai., 1992) . The critical motif in the cytoplasmic domain of TGN38 is SDYQRL (Bos et al., 1993; Humphrey et al., 1993; Wong and Hong, 1993) and it operates once the protein reaches the cell surface. TGN38 is then returned to the TGN via endosomes.

The membrane-spanning domain is a typical Golgi retention signal. These act to keep proteins in the Golgi and, so far, all Golgi proteins (Machamer and Swift, 1991) and enzymes (Nilsson et al., 1991; Munro, 1991; Aoki et al., 1992; Burke et al., 1992; Colley et al., 1992; Tang et ai., 1992; Teasdale et al., 1992; Wong et al., 1992) have been found to be retained by the membrane-spanning domain. The mechanism of retention is still unclear though two possibilities have been proposed. TGN38 molecules could interact with their neighbors through the spanning domains forming kin oligomers so large that they can no longer enter vesicles budding from the TGN (Nilsson et al., 1994) . Alternatively, the length of the membrane-spanning domain might prevent forward movement of the protein past the TGN (Bretscher and Munro, 1993) . The latter seems less likely since the length of the membrane-spanning domain (19 amino acids) is longer than the estimated average for Golgi proteins (15 amino acids) but further work will be needed to determine whether either of these mechanisms, or a completely different one, are responsible for retention in the TGN.

Our results suggest that the membrane-spanning domain is an effective retention signal since alone it can confer localization properties on CD4 that make it virtually indistinguishable from TGN38. The predicted effect of its absence would be an increased flux of protein through the TGN to the cell surface. In other words, its absence should put a greatly increased load on to the retrieval system. As fast as the protein is retrieved and returned to the TGN, it would escape to the plasma membrane and need to be retrieved again. This would explain why CD4-C, which has the retrieval but not the retention signal, is present in both endosomes and at the plasma membrane even at low levels of expression. The rate-limiting step in retrieval is not endocytosis because the linear density of CD4-C at the plasma membrane was the same at both low and high levels of expression. The limiting step must be between endosomes and the TGN since CD4-C accumulates in endosomes as the expression level increases.

The only result that is a little puzzling is the presence of CD4-M in endosomes at high expression levels. This hybrid protein would be expected to leak very slowly to the cell surface, but, in the absence of a retrieval signal, it should accumulate there. One possibility is that the endocytic signal in the CD4 cytoplasmic domain (Shin et al., 1991 ) triggers movement to endosomes; another possibility is that the protein in the endosome is destined for degradation in lysosomes. This latter possibility is more likely for CDS-M which lacks an endocytic signal yet is found in small amounts in what appear to be endosomes by immunofluorescence microscopy.

It is likely that other proteins on the exocytic pathway have both retention and retrieval signals. The sum of the two processes acting on these signals would determine the amount of protein found in a particular compartment. For resident proteins which should spend as much of their time as possible in a particular compartment one would expect retention rather than retrieval to be the dominant process. For recycling proteins, such as receptors, one would expect the opposite so that substantial amounts of protein would be moving between two compartments. It is not yet clear whether TGN38 is best considered as a resident or a recycling protein (Luzio and Banting, 1993; Stanley and Howell, 1993) . That will have to await discovery of its function.

Golgi retention of a trans-Golgi membrane protein, galactosyltransferase, requires cysteine and histidiue residues within the membrane-anchoring domain

Monoclonal antibodies to soluble, human milk galactosyltransferase 0actose synthase A protein)

TGN38 is maintained in the trans-Goigi network by a tyrosine-containing motif in the cytoplasmic domain

Evidence for an extended of the T-cell co-receptor CDga as deduced from the hydrodynamic properties of soluble forms of the extracellular region

Cholesterol and the Golgi apparatus

The transmembrane and flanking sequences of fll,2-N-acetylglucosaminyltransferase I specify medial-Goigi localization

Transferrin receptor internalization sequence YXRF implicates a tight turn as the structural recognition motif for endocytosis

The signal anchor and stem regions of the fl-galactoside ot2,6-sialyltransferase may each act to localize the enzyme to the Golgi apparatus

Tubulovesicular processes emerge from tmas-Golgi cisteroae, extend along microtubules, and interlink adjacent trans-Golgi elements into a reticulum

Soluble form of CD4 ('I"4) protein inhibits AIDS virus infection

Protein folding in the cell

Intracellular site of aalaloglycoprotein receptor-ligand uncoupling: doublelabel immunoelectron microscopy during receptor mediated endocytosis

Sorting of mannose 6-phosphate receptors and lysosomal membrane proteins in endocytic vesicles

The trans Golgi network: sorting at the exit site of the Golgi complex

The dynamic nature of the Golgi complex

Summary of T cell studies performed during the second international workshop and conference on buman leukocyte differentiation antigens

Novel anti-CD4 monoclonal antibodies separate human immunodeficiency virus infection and fusion of CD4 + cells from virus binding

Movement of internalized ligand-receptor complexes along a continuous endosomal reticuinm

Localization of TGN38 to the trans-Golo network: involvement of a cytoplasmic tyrosine-contalning sequence

Identification of a consensus motif for retention of transmembrane proteins in the endoplasmic reticulum

Retrieval of transmembraue proteins to the endoplasmic-reticulum

A cytosolic complex of p62 and rab6 associates with TGN38/41 and is involved in budding of exocytic vesicles from the trans-Golgi network

Differences in the endosomal distributions of the two mannose-6-phosphate receptors

Structure and function of the mannose 6-phosphate/insulinlike growth factor II receptors

Cry.stal structure of a soluble form of the human T cell coreceptor CD8 at 2.6 A resolution

Ligand-induced redistribution of a human KDEL receptor from the Golgi complex to the endoplasmic reticulum

The isolation and sequence of the gene encoding T8: a molecule defining functional classes of T lymphccytes

O-glycosylation of the coronavirus M protein. Differential localization of sialyltransferases in N-and O-linked glycosylation

Mitotic Golgi fragments in HeLe cells and their role in the re, assembly pathway

Eukaryote membrane traffic-retrieval and retention to achieve organelle residence

Identification, sequencing and expression of an integral membrane protein of the trans-Golgi network (TGN38)

Golgi retention signals: do membranes hold the key?

The isolation and nucleotide sequence of a cDNA encoding the T cell surface protein T4: a new member of the immunoglobulin gene family

Basolateral sorting of LDL receptor in MDCK cells: the cytoplasmic domain contains two tyrosinedependent targeting determinants

The Golgi complex: in vitro veritas?

Sequences within and adjacent to the transmembrane segment of ot-2,6-sialyltransferase specify Golgi retention

A C-terminal signal prevents secretion of luminal ER proteins

Short cytoplasmic sequences serve as retention signals for transmembrane proteins in the endoplasmic reticulum

The membrane spanning domain of fll,4-galactosyltransferase specifies trans Golgi localization

Kin recognition between medial Golgi enzymes in HeLa cells

Overlapping distribution of two glycosyltransferases in the Golgi apparatus of HeLa cells

Kin recognition: a model for the retention of Golgi enzymes. FEBS (Fed. Fur

Perrelet. 1984. A clathrin-coated, Golgi-related compartment of the insulin secreting cell accumulates proinsulin in the presence of monensin

Clathi~-immunoreactive sites in the Golgi apparatus are concentrated at the trans pole in polypeptide hormone-secreting cells

Clathrin, adaptors, and sorting

Differential endoeytosis of CD4 in lymphoid and non-lymphoid cells

Control of protein exit from the endoplasmic reticulum

Segregation of MHC Class II molecules from MHC Class I molecules in the Golgi complex for transport to lysosomal compartments

The differential degradation of two cytosolic proteins as a tool to monitor autophagy in hepatocytes by immnnocytocbemistry

Thrze-dimensional electron microscopy: structure of the Golgi apparatus

TGN38/41 recycles between the cell surface and the TGN: brefeldin A affects its rate of return of the TGN

Immunocytochemical localization of galactosyltransferase in HeLa cells: codistribution with thiamine pyrophosphatase in trans-Golgi cisternae

Demonstration of an extensive trans-tubular network continuous with the Golgi apparatus stack that may function in glycosylation

Molecular dissection of the secretory pathway

Primer-directed enzymatic amplification of cDNA with a thermostable DNA polymerase

Molecular cloning: a laboratory manual

ERD2, a yeast gene required for the receptor-mediated retrieval of luminal ER proteins from the secretory pathway

Structural features of the cytoplasmic region of CD4 required for internalization

Phosphorylationdependent down-modulation of CD4 requires a specific structure within the cytoplasmic domain of CD4

The first membrane spanning region of the lamin B receptor is sufficient for sorting to the inner nuclear membrane

TGN38/41: a molecule on the move

A Golgi retention signal in a membrane spanning domain of coronaviros E1 protein

Immunocytochemical localization of/~1,4 galactosyltransferase in epithelial cells from bovine tissues using monoclonal antibodies

DNA sequence analysis with a modified bacteriophage T7 DNA polymerase

The transmembrane domain of N-glucosaminyltransferase I contains a Goigi retention signal. 3'

The signal for Golgi retention of bovine/31,4-galactosyltransferase is in the transmembrane doromn

Irmnnnochemistry on ultrathin frozen sections

Recycling of transferrin receptors in A431 cells is inhibited during mitosis

The SXYQRL sequence in the cytoplasmic domain of TGN38 plays a major role in trans-Golgi network localization

The 17-residue membrane domain of/~-galactoside tx2,6-sialyltransferase is sufficient for Golgi retention

The single transmembrane segment of gp210 is sufficient for sorting to the pore membrane domain of the nuclear envelope

We thank Dr. Mark Marsh for helpful advice and antibodies, Dr. Eric Berger, Dr. Peter Beverly, and Dr, Nancy Hogg for antibodies, and Dr. Per Peterson for the CD4 cDNA. We also thank Rose Watson, Felicia Hunte, and the Imperial Cancer Research Fund (ICRF) FACS Unit for invaluable technical assistance. We thank the ICRF Oligonucleotide Synthesis Facility for high quality oligonuclootides and the Photography Department for their help. We thank Dr. Kathryn Howell, Dr. Wanjin Hong, and Dr. Keith Stanley for preprints of papers before publication and helpful advice. We also thank members of the Warren lab for advice and critical reading of the manuscript.Received for publication 4 October 1993 and in revised form 7 December 1993.