Glycosylation Can Influence Topogenesis of Membrane Proteins and Reveals Dynamic Reorientation of Nascent Polypeptides within the Translocon

All constructs were made by site-directed mutagenesis using PCR with Vent polymerase (New England Biolabs). The final constructs for in vivo expression were subcloned into the vector pECE (Ellis et al. 1986) and verified by sequencing. To construct chimeric proteins with conflicting signals, the coding sequence for the 98 NH₂-terminal residues of influenza hemagglutinin (strain A/Jap/305/57 [H2N2]) (Gething and Sambrook 1981) was fused via a linker sequence (Gly-Ser) in front of codon 4 of the cDNA of the asialoglycoprotein (ASGP) receptor subunit H1 (Spiess et al. 1985). The codon for methionine in position 95 was mutated to one for isoleucine to eliminate potential internal translation initiation. The resulting sequence encoded a chimeric protein with two ER targeting signals separated by a spacer of 128 residues. By PCR with this sequence as a template, 5′ truncations were prepared consisting of a 5′ Bgl II site, the COOH-terminal 100, 80, 60, 40, or 20 residues of the spacer sequence, and the signal and the COOH-terminal remainder of H1. In front of these sequences, the cDNAs of the signal sequences of hemagglutinin (H), human prepro-vasopressin-neurophysin II (V), bovine preprolactin (P), and the ASGP receptor H1 (with the wild-type hydrophilic sequence, A, or with a 30 residue deletion, ΔA) (Beltzer et al. 1991), as shown in Fig. 1 C, were ligated by BamH I/Bgl II (Gly-Ser) to produce five series of plasmids. The constructs (schematically shown in Fig. 1 A) were named to indicate the origin of the first signal, the length of the hydrophilic spacer sequence, and the origin of the second signal (e.g., V40A). The series without the second signal (H20− etc.) was generated by deleting the entire coding region of the H1 signal from the premade constructs harboring two conflicting signals (H20A etc.) by PCR.

To generate potential N-glycosylation sites in the spacer sequence, the codons for IleThrLeu (ITL), MTM, or GS were inserted into the spacer sequence after the codon for Asn-13 of the H1 sequence of the spacer sequence between the signals. To destroy the glycosylation site NITL, the codon for Thr was mutated to one for Asn. To test a different spacer sequence with a natural glycosylation site (NFT), the residues 26–59 of V40A were replaced by residues 61–100 of H1, resulting in V45(NTF)A. The glycosylation site was mutated to QTF in V45(QTF)A.

To generate the truncated constructs H55, H75, H95, and H115, the 5′ sequences of constructs H40(NIT)A, H60(NIT)A, H80(NIT)A, and H100(NIT)A encoding the hemagglutinin signal and the spacer sequences were fused via two methionine codons and a Kpn I site encoding Gly-Thr to the 14 COOH-terminal residues of H1, which are recognized by the antibody anti-H1C. Constructs V55, V75, V95, and V115 were constructed accordingly.

Cell culture reagents were from Life Technologies, Inc. COS-1 cells were grown in modified MEM supplemented with 10% FCS, 2 mM l-glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin at 37°C with 7.5% CO₂. Transient transfection was performed with lipofectin (Life Technologies, Inc.) according to the manufacturer's instructions in 6-well clusters, and the cells were processed the second day after transfection. For in vivo labeling, transfected cells were incubated for 40 min in methionine-free medium, labeled for 30 min at 37°C with 100 μCi/ml [³⁵S]methionine in starvation medium, transferred to 4°C, washed twice with PBS, lysed, and immunoprecipitated using a rabbit antiserum directed against a synthetic peptide corresponding to residues 277–287 near the COOH terminus of the ASGP receptor H1 (anti-H1C). The immune complexes were isolated with protein A–Sepharose (Amersham Pharmacia Biotech) and analyzed by SDS-PAGE and fluorography. For deglycosylation, the immune complexes were released from protein A–Sepharose by boiling in 50 mM Na citrate, pH 6, 1% SDS, and incubated with 1 mU endo-β-d-N-acetyl glucosaminidase H (endo H) for 5 h at 37°C, before gel electrophoresis. Quantitation was performed using a PhosphorImager (Molecular Dynamics, Inc.). For constructs with the vasopressin signal, the results were corrected for the loss of a methionine within the signal peptide upon signal cleavage.

Alkaline extraction was performed as described previously (Gilmore and Blobel 1985; Wessels et al. 1991). In brief, cells were homogenized and incubated in Hepes buffer, pH 11.5, for 15 min on ice. One half of each sample was centrifuged through a sucrose cushion, and the membrane pellet and the supernatant were immunoprecipitated separately and compared with the immunoprecipitate of the untreated second half of the sample (total). In vitro translation was performed with standard procedures using rabbit reticulocyte lysate and dog pancreas microsomes (Wessels et al. 1991). Protease protection assay was performed as described by Wahlberg and Spiess 1997, except that the scraped cells were homogenized in a cell cracker (from EMBL) as described by Leitinger et al. 1995.

To characterize the in vivo topogenesis of membrane proteins and to test the linear insertion model, we constructed a series of chimeric polypeptides containing an NH₂-terminal cleaved signal sequence and an internal type II signal-anchor sequence. An NH₂-terminal portion of influenza virus hemagglutinin with its cleavable signal sequence was fused to the NH₂ terminus of subunit H1 of the ASGP receptor, a type II membrane protein. By successive truncation of the hydrophilic sequence separating the two hydrophobic signals, a series of constructs with spacers of ∼20, 40, 60, 80, and 100 residues was created (H20A, H40A, etc.) (Fig. 1 A). In the wild-type context, each of the two signals efficiently directs translocation of its COOH-terminal end across the membrane. According to the linear insertion model, the NH₂-terminal hemagglutinin signal will target the polypeptide to the ER membrane and induce translocation of the spacer sequence. The internal H1 signal will then enter the translocon as a stop-transfer sequence and halt further translocation, leaving the COOH terminus of the polypeptide, including the only two potential sites for N-glycosylation, in the cytosol (Fig. 1 B, left). If, in contrast, the topogenic information of the internal signal was dominant, the COOH-terminal portion of H1 would be translocated and glycosylated, whereas the NH₂-terminal signal would be forced to insert with the opposite N_exo/C_cyt orientation or would fail to insert at all (Fig. 1 B, right).

The constructs were transiently expressed in COS-1 cells, labeled for 30 min with [³⁵S]methionine, immunoprecipitated, and analyzed by SDS-gel electrophoresis and fluorography (Fig. 2 A). Consistent with the linear insertion model, the constructs H100A and H80A yielded single products which were not glycosylated, as revealed by their resistance to deglycosylation by endo H. However, the constructs with shorter spacers between the signals, H60A, H40A, and H20A, produced an increasing fraction of an additional species of lower electrophoretic mobility. The shift in mobility was due to glycosylation, as shown by endo H digestion. Both the glycosylated and the unglycosylated products were stably integrated into the ER membrane, since they were recovered in the membrane pellet after alkaline extraction (Fig. 2 B). Under the same conditions, a control protein lacking a membrane anchor (H20−), was completely extracted into the soluble fraction. The results demonstrate that the internal signal, when separated by <80 residues from the NH₂-terminal signal, did not always function as a stop-transfer sequence, but was able to direct translocation of its COOH terminus across the membrane and to anchor the protein in a N_cyt/C_exo orientation.

Whereas the sequence context of the internal H1 signal is the same in all constructs, that of the hemagglutinin signal is always different. As a result, the flanking charges, which are known to influence the topological preference of signal sequences, are not constant. The charge difference between the 15 COOH-terminal and the 15 NH₂-terminal flanking residues, Δ(C–N), as defined by Hartmann et al. 1989, is −0.5 for the hemagglutinin signal in its wild-type context. Δ(C–N) is more negative in H40A (−1.5), H60A (−1), and H80A (−1). Thus, the observed insertion pattern cannot be explained simply by an effect of the flanking charges on the orientation of the NH₂-terminal signal. Only in H20A, the construct with the shortest spacer, in which the flanking segments of the two signals largely overlap, is Δ(C–N) positive for the hemagglutinin signal (+2) and might favor translocation of the NH₂ terminus. To test experimentally the insertion behavior of the hemagglutinin signal in the different chimeras without the potential interference by a second signal, a series of constructs was made lacking the hydrophobic part of the internal signal (H20−, H40−, etc.). In all cases, the COOH-terminal sequence was completely translocated and glycosylated (Fig. 2 C), demonstrating that different flanking sequences did not affect the functionality of the hemagglutinin signal as a secretory signal. It required the competition by the second signal to cause insertion of H20A, H40A, and H60A with a translocated COOH terminus.

To determine the contribution of the NH₂-terminal signal on topogenesis, the hemagglutinin signal in the H#A constructs was replaced by the cleavable signals from preprolactin (P#A) or the vasopressin precursor protein (V#A), or by the signal-anchor sequence of the ASGP receptor H1, either with (A#A) or without the complete 40-residue NH₂-terminal, hydrophilic domain (ΔA#A) (Fig. 1 C). The constructs were expressed in COS cells and analyzed by gel electrophoresis and fluorography (Fig. 3). The fraction of glycosylated products corresponding to the topology with a translocated COOH terminus was quantified (Fig. 4). The constructs with the preprolactin signal yielded fewer COOH-terminally translocated proteins than those with the hemagglutinin signal: a maximum of 35% for P20A and none for P60A. Since the presequence of preprolactin is relatively long, signal cleavage could be observed for P20A. The unglycosylated, cleaved products displayed a slightly higher electrophoretic mobility than the glycosylated, uncleaved products (indicated by an asterisk in Fig. 3, panel P) after endo H digestion, consistent with their expected topologies (Fig. 1 B).

The constructs with the signal of vasopressin generated more glycosylated products than the other series: ∼90% for V20A and ∼50% with a spacer of 60 residues (V60A). However, with spacers of 80 or 100 residues, no glycosylated products were generated for any of the constructs. Based on these results, the strength of the different secretory signals, i.e., the ability to dominate the insertion process, can be ranked P > H > V.

In the construct series A#A and ΔA#A, the first copy of the internal signal-anchor of the ASGP receptor H1 within the polypeptide appeared to dictate the topology, since glycosylated products were only generated with the shortest spacer length. In A20A and ΔA20A, the positive flanking charges at the NH₂ terminus of the second signal are also part of the COOH-terminal flanking region of the first, and are thus likely to weaken the first signal. Deletion of most of the 40-amino acid hydrophilic NH₂ terminus in ΔA20A increased the fraction of glycosylated products from ∼55 to ∼90%. Thus, the larger NH₂ terminus of A20A inhibits the topology with a translocated COOH terminus.

Two mechanisms could explain the observed behavior of conflicting signals in the insertion process: competition between the signals for the recruitment of SRP in the cytosol (Fig. 5 A), and/or competition for the preferred topology within the translocon (Fig. 5 B). According to the first model, the kinetics of SRP binding to the first signal are slow enough that the second signal may emerge from the ribosome before SRP was recruited to the first. The two signals then compete directly for SRP binding. If the first signal binds SRP, it will initiate translocation of the spacer sequence. If the second signal does, it will induce translocation of the COOH terminus, whereas the first signal will remain in the cytosol or will subsequently insert into the membrane with an N_exo/C_cyt orientation. This model predicts, in agreement with our results, that the ability of the second signal to compete for SRP decreases with the length of the spacer, and depends on the relative affinities of the two signals for SRP.

This model requires that at the time when the second signal emerges from the ribosome, there are still polypeptides that have not been targeted to the ER. To test this condition, we generated the constructs consisting of the signal of vasopressin or hemagglutinin, followed by 55, 75, 95, or 115 residues, including a diagnostic glycosylation site (Fig. 6 A) (V55, H55, etc.). Since at least 35 residues of a nascent polypeptide are hidden within the ribosome and inaccessible to SRP, translation will reach the stop codon and trigger disassembly of the ribosome when the first 20, 40, 60, or 80 residues following the signal sequence have emerged from the ribosome. This corresponds to a moment in the translation of the constructs V20A, V40A, V60A, and V80A (or of the corresponding constructs H#A) when the second hydrophobic sequence is just beginning to emerge from the ribosome (Fig. 6 A). When the stop codon is reached, only polypeptides that have already been targeted to the ER membrane will produce glycosylated products; all others will be released into the cytosol and remain unglycosylated. In vitro translation in reticulocyte lysate with dog pancreas microsomes added either during or after translation confirmed that these constructs could not be posttranslationally translocated (data not shown).

As shown in Fig. 6 B, even the shortest construct with a vasopressin signal, V55, was >90% glycosylated (lanes 1 and 2), indicating that the binding of SRP to the vasopressin signal and its targeting to the ER membrane was fast. Only few polypeptides of the constructs V#A thus had the opportunity to be targeted by the second signal. In contrast, the polypeptides with a hemagglutinin signal produced significant fractions of unglycosylated products: ∼40% of H55, ∼25% of H75, and still ∼10% of H95. Therefore, SRP recruitment is considerably slower than with a vasopressin signal. This suggests that the second signal in the H#A constructs might be able to compete for SRP binding. However, the numbers of untargeted polypeptides cannot account for the observed populations of H20A, H40A, and H60A with a translocated COOH terminus (Fig. 4). Most importantly, a larger fraction of the V#A than of the H#A constructs inserted with a translocated COOH terminus, even though targeting by the vasopressin signal is faster. Therefore, the different topologies are more likely the result of competition between the signals within the translocon (Fig. 5 B) than of the kinetics of SRP binding and targeting by the first signal.

According to the second model (Fig. 5 B), the spacer sequence of all polypeptides is exposed to the ER lumen, at least transiently. To probe the lumenal exposure of the spacer, a diagnostic N-glycosylation site was introduced. In the construct V40(NIT)A, a potential glycosylation site NIT was generated by insertion of three additional codons at asparagine 13 of the spacer segment. As shown in Fig. 7 A, lanes 3 and 4, the site was efficiently modified, since all polypeptides were now glycosylated either once or twice, demonstrating that they had translocated either the spacer sequence (one glycosylation) or the COOH-terminal domain (two glycosylations). However, in comparison to V40A (Fig. 7 A, lanes 1 and 2), the fraction of twice glycosylated proteins dropped significantly. In a protease protection assay, it was confirmed that the efficiency of glycosylation in the translocated COOH terminus was not reduced (data not shown). Therefore, the ratio between the two topologies was significantly altered by addition of the glycosylation site in favor of polypeptides with a translocated spacer sequence. The fraction of products with a translocated COOH terminus decreased from ∼75 to ∼30% (Fig. 7 B). A similar effect was seen for the longer spacer of 60 residues, where insertion of the glycosylation site reduced this fraction from ∼50% in V60A to <20% in V60(NIT)A (Fig. 7 A, lanes 9–12, and Fig. 7 B). This showed that topogenesis is affected by glycosylation, a lumenal modification, which clearly takes place after targeting.

To exclude the possibility that the mutation exerted its effect by altering the conformational properties of the spacer sequence rather than by glycosylation itself, the sequence NIT was mutated to NIN, which is not modified by oligosaccharyl transferase (V40[NIN]A; Fig. 7 A, lanes 5 and 6, and Fig. 7 B): the effect on the topology was largely reverted. A different consensus sequence, NMT, which was also efficiently glycosylated, showed the same increase of polypeptides with translocated spacer sequences as NIT (Fig. 7 B). The sequence NGS, in contrast, did not affect topology, but also produced hardly any polypeptides with a single glycan (Fig. 7 A, lanes 7 and 8, and Fig. 7 B). This confirms that a consensus sequence N-X-S/T is not sufficient for the topogenic effect, but that efficient glycosylation is required. (Interestingly, NGS was reported to be the most efficient N-X-S consensus sequence for N-glycosylation in an in vitro study [Shakin-Eshleman et al. 1996]. Apparently, the sequence context dramatically affects the relative efficiencies of potential glycosylation sites). The effect of glycosylation on topology was also reproduced with an entirely different spacer sequence of 45 amino acids corresponding to a segment of the exoplasmic portion of H1 with a natural glycosylation site NFT. Mutation to QFT caused the fraction of COOH-terminally translocated polypeptides to increase from ∼30% (V45[NFT]A) to ∼50% (V45[QFT]A) (Fig. 7 B). As expected, the topogenic influence of glycosylation was also observed with hemagglutinin as the NH₂-terminal signal sequence (H40[NIT]A vs. H40A) (Fig. 7 B). Glycosylation affects topogenesis and thus takes place before the topology of the polypeptides is determined, most likely by trapping the spacer sequence on the lumenal side, preventing its return to the cytosolic side of the membrane (as illustrated in Fig. 8).

The insertion of multispanning membrane proteins is a complex process that is still poorly understood. To test how successive topogenic determinants in a polypeptide chain are decoded by the targeting and translocation machinery of the mammalian ER, we expressed a series of chimeric proteins containing two conflicting signal sequences in COS cells. Only when the two signals are sufficiently separated from each other can the results be explained by a simple linear insertion process, in which the most NH₂-terminal signal sequence determines its own orientation as well as that of a subsequent transmembrane domain. The required spacer length depends on the characteristics (the strength) of the two signals. The weakest NH₂-terminal signal tested, that of vasopressin, completely forced the internal signal-anchor of the ASGP receptor H1 into a stop-transfer orientation with a spacer of 80 residues or longer, whereas the H1 signal-anchor itself could do so with a spacer of only 40 amino acids. With shorter spacers, mixed topologies were observed, indicating that topogenic information in the second signal of the polypeptide codetermined the insertion process. Hence, insertion did not occur in a strictly linear manner from the NH₂ to the COOH terminus.

The situation is thus comparable to that found in bacteria by Coleman et al. 1985 using proteins with two copies of the prolipoprotein signal separated by either 13 or 27 residues. The spacer was predominantly translocated with the longer spacer, and the COOH terminus with the shorter one. However, a point mutation that shifted the first signal from a cotranslational to a posttranslational mode of action caused COOH-terminal translocation also with the longer spacer, suggesting that targeting and insertion was now directed by the second signal. Similarly, if in our system the binding kinetics of SRP to the NH₂-terminal signal are sufficiently slow, SRP binding and targeting by the second signal might explain our results. Experiments by Johnsson et al. (1994) provided evidence that in yeast many signal sequences are surprisingly slow in targeting to the ER membrane. The signal sequences tested were of different efficiencies, allowing for between ∼100 (invertase) and ∼300 residues (carboxypeptidase Y and Ste6) to be exposed to the cytosol for 50% of the polypeptides. However, most of the weak signals, such as that of caboxypeptidase Y, were later shown to function independently of SRP (Ng et al. 1996), and even invertase secretion was only slightly affected in an SRP-defective strain (Rothe and Lehle 1998). The question of SRP-dependent targeting kinetics in mammalian cells, where only few proteins are known to be targeted independently of SRP (Zimmermann et al. 1990; Kutay et al. 1995), was still open.

In this study, we analyzed the kinetics of SRP-dependent targeting in COS cells using the hemagglutinin and vasopressin signals followed by reporter polypeptides of increasing length. Targeting was clearly faster (relative to translation) than in all the examples tested in yeast. 60% of the chains with a hemagglutinin signal and >90% with a vasopressin signal had been targeted by the time 55 residues past the signal had been translated. The relative rate of targeting might reflect that pro-vasopressin–neurophysin II is a relatively short protein of 145 residues in comparison to hemagglutinin with 547 amino acids, and thus needs to be targeted more rapidly. From the sequence, it is not clear what makes the vasopressin signal more efficient for SRP recruitment. For our constructs with conflicting signal sequences, the targeting rates of the hemagglutinin and vasopressin signals do not provide an explanation for the observed topologies, since there is no correlation with COOH-terminal translocation. Although a fraction of some of the proteins, e.g., with a hemagglutinin signal and with the shortest spacer, may be targeted by the second signal, the topogenic influence of the second signal appears to be largely independent of how the protein was brought to the translocon.

Glycosylation sites introduced into the spacer segment between the signal sequences provided a surprising insight into the dynamics of the orientation process in the translocon. N-glycosylation significantly affected the equilibrium of topologies in favor of that with a translocated spacer sequence and a cytosolic COOH terminus. This implies that with or without a glycosylation site, the spacer segment destined for cytosolic localization is transiently exposed to the ER lumen. With the glycosylation site, the spacer is modified by oligosaccharyl transferase when it appears in the lumen and is trapped there (Fig. 8). The results indicate that the polypeptides dynamically reorient within the translocon, exploring the possible topologies favored by either of the two signals. Hydrophilic sequences of up to 60 residues between transmembrane domains move in and out of the ER before the topology is decided. A likely mechanism for this glycosylation effect is steric hindrance of the modified segment to return to the cytosolic side. It has been previously observed in vitro that a glycosylated segment of a model membrane protein was finally exposed to the cytosol (Lipp et al. 1989). We did not recover any threefold glycosylated products indicative of a lumenal COOH terminus and a glycosylated spacer, suggesting either that in our system glycosylated sequences cannot flip back to the cytosolic side or, if they can, that they are rapidly deglycosylated in the cytosol (Suzuki et al. 1998).

For the insertion process of natural multispanning membrane proteins, our results illustrate that successive membrane-spanning sequences insert in a coordinated manner, if sufficiently close within the polypeptide. What determines the topogenic strength of a signal is not obvious from the few examples analyzed. It does not simply correlate with the charge difference, since Δ(C–N) of the constructs with a spacer of 40 residues, for example, are −6.5 (P40A), −1.5 (H40A), and −5 (V40A), listed in the order of decreasing topogenic strength. Other criteria shown to influence the orientation of individual signals, such as hydrophobicity (Wahlberg and Spiess 1997) and the size of the NH₂-terminal hydrophilic segment (Fig. 3 and Fig. 4A vs. ΔA20A; Denzer et al. 1995), are likely to contribute as well. In addition, longer and more hydrophobic signals, like that of H1 and, to a lesser extent, that of preprolactin, may exit the translocation pore more rapidly (Martoglio et al. 1995; Mothes et al. 1997) and might thereby limit the influence of the second signal.

Spacer segments of 80 or more residues may be sterically unable to reorient themselves within the translocon. An additional mechanism that could limit an effect of the second topogenic sequence is the binding of lumenal chaperones, particularly BiP. The yeast homologue of BiP, Kar2p, has been shown to promote posttranslational translocation, but probably acts also in cotranslational transport (Brodsky et al. 1995; Panzner et al. 1995; Pilon et al. 1998; Zimmermann 1998; Matlack et al. 1999). It is likely that spacer sequences that are long enough to engage with lumenal chaperones will be anchored by this interaction, thus determining the topology.

In summary, topologenesis of membrane proteins appears to be a dynamic process in which topogenic information of closely spaced signal and transmembrane sequences throughout the polypeptide is integrated. Enzymes associated with the translocation machinery and acting on the protein to be inserted, such as oligosaccharyl transferase and potentially signal peptidase and chaperones, can affect the process and thereby contribute to the efficient and uniform insertion of natural membrane proteins.

We thank Nicole Beuret for technical assistance, and Dr. N. Kralli and our colleagues in the laboratory for critical reading of the manuscript.

This work was supported by Grant 31-43483.95 from the Swiss National Science Foundation.