Eur. J. Biochem. 251, 8212829 (1998)
 FEBS 1998
Stop-transfer function of pseudo-random amino acid segments
during translocation across prokaryotic and eukaryotic membranes
Annika SÄÄF, Erik WALLIN and Gunnar VON HEIJNE
Department of Biochemistry, Stockholm University, Stockholm, Sweden
(Received 3 October/27 November 1997) 2 EJB 97 1408/3
We have measured the efficiency of stop-transfer function for a set of pseudo-random, 18-residue
amino acid segments, both in Escherichia coli and in mammalian microsomes. In general, stop-transfer
function correlates well with the mean hydrophobicity of the segment, though exceptions exist. Kinetic
studies suggest that polar segments are rapidly translocated through the E. coli inner membrane and that
strongly hydrophobic segments become permanently anchored, while sequences with an intermediate
mean hydrophobicity become partly trapped in a transmembrane disposition for a considerable time before
being released to the periplasm or degraded.
Keywords : membrane protein; protein translocation ; stop transfer sequence; endoplasmic reticulum.
Stop-transfer sequences in integral membrane proteins generally consist of long stretches of hydrophobic residues that interrupt membrane translocation and provide a permanent anchoring in the lipid bilayer [1]. Mutations that weaken the hydrophobic character of such sequences often result in defective stoptransfer function [2210], and theoretical methods based on the
average hydrophobicity measured over a 15220-residue window have proved quite effective in localizing stop-transfer sequences [11, 12].
Even though the basic idea that overall hydrophobicity is the
major determinant of stop-transfer function is thus quite well
established, it is not clear whether the transition, as a function
of hydrophobicity or some other sequence-related parameter,
from a functional to a non-functional stop-transfer sequence is
gradual or abrupt. This lack of understanding of how sequences
near the threshold for stop-transfer function behave is also reflected in the difficulty of developing prediction methods that
cleanly discriminate between stop-transfer sequences and protein segments that can be translocated. Finally, it is unclear how
stop-transfer sequences are recognized by the translocon, and
how they eventually move out of the translocon into the surrounding lipid.
To address these kinds of problems, we have developed experimental systems that allow us to measure the stop-transfer
efficiency of defined sequences both in Escherichia coli and, in
parallel, in a eukaryotic in vitro system. In this initial study, a
collection of 18-residue, pseudo-random segments spanning a
wide range of hydrophobicities has been analyzed, following an
approach previously used to analyze the sequence requirements
for functional signal peptides [13, 14] and mitochondrial
targeting peptides [15, 16]. As expected, there is a good but not
Correspondence to G. von Heijne, Dept. of Biochemistry, Stockholm
University, S-106 91 Stockholm, Sweden
Fax: 146 8 15 36 79.
E-mail: [email protected]
URL : http://www.biokemi.su.se/
Abbreviations. ATA, aurintricarboxylic acid; CCCP carbonylcyanide
m-chlorophenylhydrazone; PSBT, P. shermanii biotin transcarboxylase;
PhMeSO2 F, phenylmethylsulfonyl fluoride; BCCP, biotin carboxyl carrier protein.
perfect correlation between stop-transfer efficiency and overall
hydrophobicity, and between stop-transfer function in the prokaryotic and eukaryotic systems. A number of segments of intermediate hydrophobicity are found to be partly translocated and
partly trapped spanning the inner membrane. Detailed analysis
of one such segment suggests that it delays translocation to the
periplasm in a fraction of the molecules, while the remaining
molecules are slowly degraded.
MATERIALS AND METHODS
Enzymes and chemicals. Unless otherwise stated, all enzymes were from Promega. T7 DNA polymerase was from Pharmacia. [35S]Met and streptavidin-horseradish peroxidase were
from Amersham. Proteinase K was from Gibco BRL. Tosylphenylalanylchloromethane-treated trypsin was from Serva GmBH.
Ribonucleotides, deoxyribonucleotides, dideoxyribonucleotides,
carbonyl cyanide m-chlorophenylhydrazone (CCCP), and the
cap analog m7G(5′)ppp(5′)G were from Pharmacia. Aurintricarboxylic acid (ATA) was from Sigma Chemical Co. Plasmid
pGEM1, transcription buffer and rabbit reticulocyte lysate was
from Promega. The glycosylation acceptor peptide N-benzoylAsn-Leu-Thr-N-methylamide and the non-acceptor peptide Nbenzoyl-Asn-Leu-(allo)Thr-N-methylamide were synthesized
according to [17]. Oligonucleotides were from Kebo Lab.
Strains and plasmids. Experiments were performed in E.
coli strain MC1061 [∆lacX74, araD139,∆(ara, leu)7697, galU,
galK, hsr, hsm, strA] [18] and CC118 [∆(ara-leu)7697 ∆lacX74
∆phoA20 galE galK thi rpsE rpoB argE(am) recA1] [19]. The
biotinylatable Propionibacterium shermanii biotin transcarboxylase (PSBT) domain was obtained from plasmid pCY66 [20]
All constructs were expressed in E. coli from the pING1 plasmid
[21] by induction with arabinose and in vitro from plasmid
pGEM1 using SP6 RNA polymerase.
DNA techniques. Site-specific mutagenesis was performed
according to the method of Kunkel [22], as modified by Geisselsoder et al. [23]. All mutants were confirmed by DNA sequencing of M13 or plasmid DNA using T7 DNA polymerase.
Pseudo-random oligonucleotides of the sequence ATTCGTCTT-
822
Sääf et al. (Eur. J. Biochem. 251)
TCCGAGACTAGT(A0.30C0.15G0.25T0.30 /A0.25C0.20G0.15T0.40 /A0.00 C0.95G0.05T0.00)18GGATCCGTGACGCACCGC were amplified
by PCR using primers complementary to their unique ends and
were cloned into a modified lep gene between a SpeI site introduced by site-directed mutagenesis in codons 226 and 227 and
a BglII site introduced in codons 231 and 232. The gene used
in the microsome experiments also carried an Asn214→Gln mutation (removing the potential glycosylation site in wild-type
Lep) and mutations converting residues 96298 and residues
2582260 to, respectively, Asn-Ser-Thr and Asn-Ala-Thr (introducing two new potential glycosylation sites). Fusions to the
phoA gene were made by introducing a KpnI site in lep codon
253, and fusing the KpnI2SmaI fragment from a previously constructed lep fusion [24] carrying a phoA gene lacking the 5′
segment coding for the signal sequence behind the new KpnI
site in lep. Similarly, the 78-residue biotinylatable domain from
the 1.3S subunit of PSBT was amplified by PCR from plasmid
pCY66 [20] and was fused to lep at the KpnI site in codon 253.
For cloning into and expression from the pGEM1 plasmid,
an XbaI2SmaI fragment carrying a lep gene with the above
modifications (i.e., the SpeI, BglII and KpnI sites, glycosylation
sites in residues 96298 and 2582260, and an Asn214→Gln mutation) was cloned behind the SP6 promoter in a previously constructed pGEM1 derivative [25] containing a lep gene that had
been modified in its 5′ upstream region by the introduction of
an XbaI site and by changing the context 5′ to the initiator ATG
codon to a ‘Kozak consensus’ sequence [26]. Thus, the 5′ region
of the gene was modified to : ...ATAACCCTCTAGAGCCACCATGGCGAAT... (XbaI site and initiator codon underlined). In
a second step, SpeI2KpnI fragments from pING1 plasmids carrying lep-phoA fusion genes with pseudo-random inserts were
ligated between the SpeI and KpnI sites, thus creating intact lep
genes with pseudo-random inserts and with potential glycosylation sites in positions 96 and 258 that could be expressed in
vitro.
Protease-protection assay of stop-transfer function in E.
coli. E. coli strain MC1061 transformed with the pING1 vector
[21] carrying the relevant constructs under control of the arabinose promoter was grown at 37°C in M9 minimal medium supplemented with 100 mg/ml ampicillin, 0.5 % fructose, 100 mg/
ml thiamin, and all amino acids (50 mg/ml each) except methionine. An overnight culture was diluted 1:25 in fresh medium,
shaken for 3.5 h at 37°C, induced with arabinose (0.2%) for
5 min, and labeled with [35S]methionine (75 µCi/ml). After
1 min, non-radioactive methionine was added (500 mg/ml) and
incubation was continued for various times after which it was
stopped by chilling on ice. Cells were centrifuged at 20 800 g for
2 min, resuspended in ice-cold buffer (40% mass/vol. sucrose,
33 mM Tris, pH 8.0) and incubated with lysozyme (5 mg/ml)
and 1 mM EDTA for 15 min on ice. Aliquots of the cell suspension were incubated 10 min on ice and 5 min at room temperature, either with no additions, with the addition of 25 mg/ml
proteinase K, or with 25 mg/ml proteinase K and 0.2 % Chaps
[27]. After addition of phenylmethylsulfonyl fluoride (PhMeSO2F), samples were acid precipitated with trichloroacetic acid
(10% final conc.), resuspended in 10 mM Tris/2 % SDS, immunoprecipitated with antisera to PhoA, washed and analyzed by
SDS/PAGE. Gels were scanned in a FUJIX Bas 1000 phosphoimager and analyzed using the MacBAS software (version 2.1).
To block the function of SecA [28] or dissipate the proton
motive force across the inner membrane, sodium azide or CCCP
were added to final concentrations of 2 mM and 50 µM, respectively, 1 min before pulse labeling.
PhoA activity assay. Alkaline phosphatase activity measurements were performed in strain CC118 transformed with the ap-
Fig. 1. Topology of Lep (left) and of Lep(SpeI/BglII)−PhoA-derived
clones with pseudo-random stop-transfer sequence inserts (hatched
segment). When the insert has stop-transfer function (middle), a topology with three transmembrane segments and an inactive PhoA moiety
in the cytoplasm results. For inserts lacking stop-transfer function (right),
the PhoA moiety is translocated to the periplasm where it folds into
an enzymatically active, protease-resistant species. Proteolysis of intact
spheroplasts thus gives rise to a protected fragment encompassing the
stop-transfer sequence plus the PhoA domain in the former case, and
to a slightly smaller protease-resistant PhoA moiety in the latter. After
disruption of the spheroplasts with detergent, only the folded, periplasmic form of PhoA is protease resistant.
propriate pING1-derived plasmids (in the presence of 0.2 % arabinose) as described [29].
Biotinylation assay. Biotinylation of the PSBT domain in
Lep2PSBT fusion proteins was assayed in strain MC1061
grown to mid-log phase in Luria-Bertani medium by induction
of the fusion protein with arabinose (see above) for 1 h, followed
by preparation of spheroplasts, treatment with 10 µg/ml trypsin
for 20 min on ice and addition of PhMeSO 2F to a final concentration of 1.4 mg/ml. After SDS/PAGE, proteins were transferred to nitrocellulose, blotted with either streptavidin2horseradish peroxidase according to the manufacturers recommendations or with a Lep antiserum that had been affinity purified
[30], and visualized by the ECL procedure (Amersham). Band
intensities were quantified by laser densitometry using a Molecular Dynamics Personal Densitometer.
In vitro transcription and translation in reticulocyte lysate. pGEM1 plasmids carrying the relevant constructs were
used for in vitro transcription. Synthesis of RNA by SP6 RNA
polymerase and translation in reticulocyte lysate in the presence
of dog pancreas microsomes was performed as described [31].
Translocation of polypeptides to the lumenal side of the microsomes was assayed both by prevention of N-linked glycosylation
through competitive inhibition by addition of a glycosylation
acceptor tripeptide but not by a non-acceptor tripeptide, and by
proteinase K treatment of either intact or detergent-solubilized
microsomes as described [32]. Kinetics of glycosylation was
measured as described in [33], and the results were quantitated
by phosphoimager analysis.
Hydrophobicity analysis. Hydrophobicity analysis was carried out using the Engleman-Steitz hydrophobicity scale [34]
and the TOPPRED algorithm [35, 36] with a full window size
of 21 residues and a core widow size of 15 residues.
RESULTS
Assays for stop-transfer efficiency in E. coli. As a model protein for measuring stop-transfer efficiency, we chose the well-
Sääf et al. (Eur. J. Biochem. 251)
characterized E. coli inner membrane protein leader peptidase
(Lep). Lep has two transmembrane segments (H1, H2) near the
N-terminus, a short cytoplasmic loop (P1), and a large, periplasmic C-terminal domain (P2), Fig. 1. Translocation of the P2 domain requires an active Sec machinery [37] and depends on the
proton-motive force (pmf) across the membrane [38].
For these studies, two unique restriction sites (SpeI and
BglII) were introduced by site-directed mutagenesis into the
middle of the P2 domain (in codons 2262227 and 2312232,
respectively), and pseudo-random oligonucleotides encoding a
stretch of 18 predominantly hydrophobic amino acids were
cloned between these sites.
To allow an easy measure of the stop-transfer efficiency of
the pseudo-random segments in E. coli, the alkaline phosphatase
(phoA) gene lacking the portion encoding the signal peptide was
fused at a newly introduced KpnI site in codon 253 of Lep(SpeI/
BglII). Since two critical disulfide bonds are necessary for PhoA
activity [39], PhoA will only be active when located in the oxidizing environment of the periplasm but not in the cytoplasm
[40, 41]. Moreover, active, properly folded PhoA is highly protease-resistant, whereas inactive forms of the protein are readily
degraded by proteases [42, 43]. The stop-transfer efficiency of
any stretch of residues cloned between the SpeI and BglII sites
in Lep(SpeI/BglII)/PhoA should thus be reflected both in the
enzymatic activity of the fusion protein, and in the level of protease resistance of the PhoA moiety in detergent-permeabilized
cells. Finally, protease treatment of intact spheroplasts should
result in the appearance of the protease-resistant PhoA core in
clones with no stop-transfer function, but should give rise to a
slightly larger protected fragment including both the stoptransfer sequence and the entire PhoA domain in clones with
stop-transfer activity.
Design of pseudo-random oligonucleotides. Three criteria
were considered in the design of the pseudo-random oligonucleotides. First, we wanted to avoid stop codons as much as possible. Second, we wanted to produce a collection of sequences
spanning the critical hydrophobicity interval that statistical
analyses of natural stop-transfer sequences have suggested as a
threshold for stop-transfer function [36]. Third, we wanted a set
of sequences with a reasonably balanced overall amino acid
composition. A simple target function that includes these three
factors was used:
Uo
20
S 5 f stop 1 4
i5 1
U
fi hi 2 H0 1
20
o (f 2 0.05)
2
i
i5 1
where fstop is the frequency of stop-codons, fi is the frequency of
amino acid type i, hi is the hydrophobicity of amino acid type i,
H0 is the target mean hydrophobicity, and 0.05 is the target frequency for all amino acid types. For the hydrophobicity calculation, we used the Engelman-Steitz scale [34] with H0 5 0.6. Solutions with f stop . 0.0053 (corresponding to more than 10% of
all possible 20-residue segments having a stop-codon) were discarded. All possible combinations of base frequencies (in steps
of 5 %) in the three codon positions were generated, and evaluated with the target function. Based on the minimal value of S
found, we chose to synthesize a pseudo-random collection of
oligonucleotides of the following composition :
ATTCGTCTTTCCGAGACTAGT(A0.30C0.15G0.25T0.30/
A0.25C0.20G0.15T0.40/
A0.00C0.95G0.05T0.00)18GGATCCGTGACGCACCGC
Note that the 5′ and 3′ ends are non-degenerate and encode,
respectively SpeI and BamHI restriction sites (underlined) used
for cloning. These unique flanking sequences were used in a first
823
round of PCR amplification, and the resulting double-stranded
molecules were digested with SpeI and BamHI (compatible with
BglII) and cloned into the Lep(SpeI/BglII)/PhoA vector. 30 different clones were selected at random and used in the subsequent
analysis. For these clones, the overall base composition in
the three randomized codon positions was found to be
A0.31C0.15G0.15T0.39 /A0.17C0.23G0.11T0.48 /A0.00C0.98G0.02T0.00 ; i.e., reasonably close to the design target.
Stop-transfer activity correlates with overall hydrophobicity.
In an initial screen of the 30 clones, PhoA activities were measured and plotted as a function of the mean hydrophobicity of
the pseudo-random segments as calculated by the TOPPRED
algorithm [35, 36], Table 1 and Fig. 2. Two classes of sequences
could be identified: those with high PhoA activity (similar to
the activities reported previously for PhoA fusions to the P2
domain of wild type Lep [44]; see clone no. 1113 in Table 1)
and low mean hydrophobicity, and those with relatively low
PhoA activity and high mean hydrophobicity. However, within
each set of sequences, there was no clear correlation between
PhoA activity and mean hydrophobicity measured by a number
of different hydrophobicity scales (data not shown).
Somewhat surprisingly, even the most hydrophobic pseudorandom segments gave clones with PhoA activities clearly above
background. To study this further, three controls were included :
the original Lep(SpeI/BglII)/PhoA fusion without an insert
(clone no. 1113), a fusion with 18 leucine residues cloned between the SpeI and BglII sites (18L), and a similar fusion but
with 5 arginines immediately following the 18 leucines (18L5R). As expected, the first fusion had a high activity (Table 1),
and the 18L-5R construct had essentially no activity. However,
the 18L construct again had a low but non-zero activity. Two
possible explanations are either that C-terminal positively
charged residues are needed in addition to the hydrophobic segment for full stop-transfer function [45], or that a hydrophobic
segment placed in the P2 domain can act also as a signal sequence and compete with the H2 segment (which is not very
hydrophobic and is followed by a positively charged arginine)
for initiation of translocation of downstream segments of the
nascent chain. We favor the latter possibility, since flanking positively charged residues have previously been found to increase
the stop-transfer efficiency only of marginally hydrophobic segments [3] and since translocation of the P2 domain is known to
be efficiently blocked by C-terminal positively charged residues
[46]. A similar situation with competition for initiation of
translocation between an inefficient signal peptide and a downstream stop-transfer sequence has recently been described [2].
We conclude that there is a reasonably good correlation between the mean hydrophobicitiy of the pseudo-random segment
and the PhoA activity, but that the high background activity of
the clones with more hydrophobic segments impairs the sensitivity of the assay.
Functional stop-transfer sequences detected by a proteaseprotection assay. As a more direct assay for stop-transfer function, we carried out protease-protection experiments. As noted
above, protease treatment of spheroplasts is expected to give rise
to a PhoA-sized fragment in clones without stop-transfer function, and to a slightly larger stop-transfer (ST) fragment when
the pseudo-random sequence functions as a stop-transfer sequence. Protease treatment of detergent-permeabilized cells
should again leave a PhoA-sized fragment in clones without (or
with partial) stop-transfer function and should completely digest
the fusion protein in clones with full stop-transfer activity where
the PhoA moiety is cytoplasmic.
824
Sääf et al. (Eur. J. Biochem. 251)
Table 1. Amino acid sequence, alkaline phosphatase activity, and mean hydrophobicity (kHyfl) of constructs. The 18-residue inserts are
underlined.
Clone
Sequence
Activity
khyfl
13
20
22
24
26
27
29
30
34
36
38
60
62
66
67
68
69
71
76
77
79
82
84
86
89
90
91
92
93
94
1113
18L
18L-5
GIRLSETSLAPYSFITFFNSIVILTSRSVTHRIL
GIRLSETSIIYPSNCHFISHFNSFTPRSVTHRIL
GIRLSETSYFICPVSSSASIGHAFISRSVTHRIL
GIRLSETSSTNFYINKSLNDLIRNIPRSVTHRIL
GIRLSETSIIYPSNCHFISHFNSFTPRSVTHRIL
GIRLSETSFPNFFPFGDSSFINFFSPRSVTHRIL
GIRLSETSDFSPSFVSSLYIFFMCFSRSVTHRIL
GIRLSETSNYVATFYIFDFSFFFVYVRSVTHRIL
GIRLSETSFTAHIPGLCFFPSFGFATRSVTHRIL
GIRLSETSPPNSTIINIDFSGHLIYSRSVTHRIL
GIRLSETSACPVISYRIFTVNYIPISRSVTHRIL
GIRLSETSITFIFANCSHLISCIIPSRSVTHRIL
GIRLSETSVFIVSVGSFITSVLFIVIRSVTHRIL
GIRLSETSFCNVYLYVRLIRISSFNIRSVTHRIL
GIRLSETSYNLYIYAFFASSLVHGFIRSVTHRIL
GIRLSETSTTFFAIFVIFIFLTLVCFRSVTHRIL
GIRLSETSAPNLFCFLVFIFICHFIDRSVTHRIL
GIRLSETSFIIDTRSCIAYFNYSAVNRSVTHRIL
GIRLSETSLTIFAFLFVNFLFVLTCNRSVTHRIL
GIRLSETSFFDIFFFSSNSITFVFDLRSVTHRIL
GIRLSETSTIIICIVFLSCFPLFLCGRSVTHRIL
GIRLSETSGNIFCSVSINVCFFTLVPRSVTHRIL
GIRLSETSVFYNICFFITSLLLIDIFRSVTHRIL
GIRLSETSLYISSDFVIYTYSTIHILRSVTHRIL
GIRLSETSINFNITFPIIYSILSFHIRSVTHRIL
GIRLSETSIYYRAFFNNFFLSYCVSSRSVTHRIL
GIRLSETSHLGICVVCFIVAFYPFRGRSVTHRIL
GIRLSETSFLIINLYPVHTFPIFCTSRSVTHRIL
GIRLSETSVFYIYVTVSTSFPYSIITRSVTHRIL
GIRLSETSAVGDFFNVSSTFINVFYTRSVTHRIL
Lep(SpeI/BglII)
GIRLSETSLLLLLLLLLLLLLLLLLLRSVTHRIL
GIRLSETSLLLLLLLLLLLLLLLLLLRSRRRRRL
369
390
375
604
330
384
124
207
531
391
354
431
92
302
475
218
241
460
230
388
117
343
174
363
383
330
195
171
477
472
597
116
32
1.42
0.29
1.19
21.30
0.29
0.50
1.49
1.13
1.51
20.14
0.22
1.32
2.10
20.51
1.07
2.59
1.73
20.35
2.08
0.95
2.27
1.38
1.44
0.54
1.09
0.06
1.55
1.14
1.30
0.52
Fig. 2. Alkaline phosphatase activity correlates with the mean hydrophobicity of the pseudo-random insert. The mean hydrophobicity was
calculated over a 21-residue, trapezoid window using the EnglemanSteitz hydrophobicity scale (see Materials and Methods). The number of
each clone is shown (see Table 1). Clones indicated by an open square
have stop-transfer function in dog pancreas microsomes; those indicated
by a gray square do not (see Fig. 6B).
A number of clones were tested in the protease assay. A
clone with a low mean hydrophobicity (no. 24) produced only a
PhoA-size protected fragment (Fig. 3 A, lanes 426), and one
with high mean hydrophobicity (18L) gave rise to the expected,
slightly larger protected species (ST-fragment, Fig. 3A, lanes
2.51
2.51
123). A pulse/chase analysis carried out for the 18L-5R clone
demonstrated that the stop-transfer segment was stably integrated in the membrane for at least 60 min (Fig. 3 B, lanes 5,
6) but that the periplasmic domain preceding the stop-transfersegment was slowly degraded by endogenous proteases as seen
by the disappearance of the full-length molecules at the 60 min
chase point (Fig. 3 B, lane 4).
Protease-protection experiments were carried out for a selection of the 30 clones. A clear correlation between the relative
amount of ST-fragments seen after a 1 min labeling period and
the mean hydrophobicity was apparent, suggesting a loosely defined threshold for stop-transfer function around a mean hydrophobicity of 121.5 (Fig. 3 C). In the transition region, a number
of clones with intermediate mean hydrophobicities gave rise to
a significant amount of ST-fragment, suggesting that they were
not entirely devoid of stop-transfer function.
Marginally hydrophobic segments have partial stop-transfer
function. The observation of the diagnostic protected ST-fragment in clones with intermediate mean hydrophobicity prompted
us to further characterize their behavior in a pulse-chase assay.
The ST-fragment was stable during an 8 min chase in clone 18L
with high mean hydrophobicity (Fig. 4A, compare with the result for clone 18L-5R above), whereas no ST-fragment was observed in a clone (no. 24) with a low mean hydrophobicity even
at the earliest chase time. Notably, in a clone (no. 67) with intermediate mean hydrophobicity, the ST-fragment gradually disappeared with a half-time of approximately 3 min.
Sääf et al. (Eur. J. Biochem. 251)
Fig. 3. Stop-transfer function correlates with the appearance of a
diagnostic protease-protected ST-fragment. (A) Cells were labeled
with [35S]Met for 1 min, converted to spheroplasts, treated with proteinase K (PK) either in the absence or presence of Chaps, and immunoprecipitated with a polyclonal PhoA antiserum. Protease treatment of clone
18L gives rise to a prominent protected fragment (ST-fragment, lane 2),
somewhat larger in size than mature PhoA, that is completely digested
in the presence of detergent (lane 3), whereas a PhoA-sized fragment is
seen in clone no. 24 both in the absence and presence of detergent (lane
5). (B) Pulse/chase analysis of clone 18L-5R. Cells were labeled as in
(A), chased in the presence of non-radioactive Met for either 4 min or
60 min, and then processed as in (A). (C) Fraction of ST-fragment observed after a 1-min labeling period versus mean hydrophobicity for the
different clones. Clones indicated by an open square have stop-transfer
function in dog pancreas microsomes; those indicated by a gray square
do not. FL, full-length fusion protein ; ST, ST-fragment ; PhoA, PhoAsized fragment.
As seen in Fig. 4B, a number of clones with intermediate
hydrophobicity gave rise to varying levels of a somewhat unstable ST-fragment, whereas this fragment was not seen in
clones with a low mean hydrophobicity (no. 24, no. 66) or in a
Lepwildtype2PhoA fusion (no. 1113; data not shown).
In principle, the gradual disappearance of the ST-fragment
could be due either to translocation or to endogenous proteolysis. Unfortunately, the relatively slow folding of PhoA in the
825
Fig. 4. Marginally hydrophobic segments have partial stop-transfer
function. (A) Pulse/chase analysis of stop-transfer function in clones
18L, no. 67 and no. 24. Cells were labeled with [35S]Met for 1 min, nonradioactive methionine was added, and incubation was continued for the
indicated times. Aliquots were chilled on ice, converted to spheroplasts
and treated with proteinase K as in Fig. 3. (B) Percentage of ST-fragment
relative to the amount of (FL1PhoA1ST) fragments as a function of
time for the indicated clones.
periplasm [47] takes place on a similar time-scale as the disappearance of the ST-fragment in clones such as no. 67 with marginally hydrophobic ST segments both at 37°C and at 21°C
(data not shown), making it impossible to cleanly correlate the
disappearance of the ST-fragment and the appearance of a stably
folded PhoA-domain in the periplasm. Simple pulse/chase experiments on PhoA fusion proteins thus could not distinguish
between translocation and degradation.
Slow translocation and degradation of a clone with a marginally hydrophobic segment. Biotinylation has been used as
an alternative to pulse/chase/proteolysis assays for studying the
rate of translocation of periplasmic domains of inner membrane
proteins [20]. In this approach, a biotinylatable domain is fused
to the periplasmic domain in question, and its degree of biotinylation (catalyzed by the cytoplasmic enzyme biotin ligase) is
taken as a measure of the time spent in the cytoplasm before
translocation. To this end, the PhoA moiety in clones no. 1113,
24, 67 and 18L-5R was replaced by a 78-residue, biotinylatable
domain from the 1.3S subunit of P. shermanii biotin transcarboxylase (PSBT), and cells were induced for production of the
fusion protein for 1 h before conversion to spheroplasts, digestion with trypsin, SDS/PAGE, transfer to nitrocellulose, and
826
Sääf et al. (Eur. J. Biochem. 251)
Fig. 5. A marginally hydrophobic segments reduces the rate of
translocation and increases degradation. Clones with the C-terminal
PhoA domain replaced by a biotinylatable PSBT domain were expressed
and steady-state levels of protein reactive to Lep antiserum and streptavidin-horseradish peroxidase were measured by western blotting and ECL.
Lane 1, wild type Lep (no PSBT fusion) ; lane 2, clone no. 1113 ; lane
3, clone no. 24 ; lane 4, clone no. 67 ; lane 5, clone 18L-5R. (A) Spheroplasts probed with Lep antiserum. (B) Spheroplasts probed with streptavidin-horseradish peroxidase. (C) Trypsin-treated spheroplasts probed
with streptavidin-horseradish peroxidase. (D) Trypsin-treated, Triton X1002solubilized spheroplasts probed with streptavidin-horseradish peroxidase. Full-length molecules are indicated by FL, protease-resistant,
biotinylated degradation products are indicated by *, and the cytoplasmic, biotinylated biotin carboxyl carrier protein (17 kDa) and its protease-resistant core domain (6 kDa) are indicated by BCCP and **, respectively. The integrity of the spheroplasts can be judged by the protease-resistance of the cytoplasmic BCCP protein (C). (E) Fraction of biotinylated full-length molecules calculated as the quotient between the
intensity of the streptavidin- and Lep-reactive full-length bands (A and
B) normalized to the value for clone 18L-5R which was set to 1.0. Mean
values from two independent experiments are shown
ECL detection with streptavidin2horseradish peroxidase or Lep
antiserum.
As seen in Fig. 5A, the full-length fusion proteins reacted
with the Lep antiserum, although clones nos 67 and 18L-5R
gave a considerably weaker signal than did wild type Lep and
clones nos 1113 and 24. When probed with streptavidin
(Fig. 5 B), the full-length proteins gave signals of similar
strengths; in addition, clone 18L-5R gave rise to a strong signal
in the 10215-kDa region (similar results were obtained when
trichloroacetic acid-precipitated whole cells were analyzed, data
not shown). While the full-length proteins were largely degraded
by trypsin treatment of the spheroplasts (70290% degradation
relative to the biotinylated cytoplasmic biotin carboxyl carrier
protein BCCP, compare Fig. 5B and C), the biotinylated fragment in clone 18L-5R was resistant. After solubilization in detergent, all Lep2PSBT fusion proteins were degraded by trypsin, and only a protease-resistant fragment of BCCP remained
(Fig. 5 D; see [20]).
The results for construct 18L-5R are consistent with the topology determined for the 18L-5R/PhoA fusion, i.e., the C-terminal PSBT domain remains in the cytoplasm and is efficiently
biotinylated. Apparently, the full-length form is partly degraded
by endogenous proteases to C-terminal, biotinylated fragments
of a size consistent with cleavage near the ST-segment; as noted
above (Fig. 3 B), the 18L-5R-PhoA fusion is also slowly degraded in vivo to the C-terminal ST-fragment.
Clones nos 1113 and 24 which lack hydrophobic ST-segments are inefficiently biotinylated (strong signal with Lep anti-
body, weak with streptavidin) and are sensitive to trypsin in
spheroplasts, suggesting that the C-terminal PSBT domain is
translocated slowly enough to allow a small fraction to become
biotinylated. This fraction can be estimated to around 5% if we
assume that the full-length form of the 18L-5R2PSBT fusion is
fully biotinylated and that the signal obtained with the Lep antiserum reflects the total amount of full-length protein (Fig. 5 E).
Clone no. 67, finally, gives a weak signal both with Lep antiserum and streptavidin, and is fully degraded by trypsin. It is
thus translocated to the periplasm but accumulates to significantly lower steady-state levels than all the other clones tested
(22% compared with clone no. 1113 based on the relative intensities of the Lep signals in Fig. 5A corrected by the levels of the
cytoplasmic BCCP protein in Fig. 5 B). Almost half of the fulllength molecules are biotinylated (Fig. 5 E) suggesting that
translocation of the PSBT domain is markedly slower than in
clones nos 1113 and 24. It thus appears that the marginally hydrophobic ST-segment in clone no. 67 significantly slows down
the rate of translocation compared to clones with a more polar
ST-segment, but that a large fraction of the molecules are degraded rather than translocated. Few if any of the molecules in
clone no. 67 remain in a stable transmembrane disposition, as
seen both by the disappearance of the ST-fragment in Fig. 4 and
by the lack of a biotinylated, protease-resistant fragment under
the steady-state conditions reported in Fig. 5 B ; possibly, the
marginally hydrophobic character of the pseudo-random segment in this clone renders the protein unstable in the lipid bilayer.
It thus appears that the segment of intermediate mean hydrophobicity in clone no. 67 remains transmembrane for a significant period of time before being translocated to the periplasm
(<20% of the molecules) or degraded (<80% of the molecules),
as shown by the appearance of an apparent translocation intermediate of a well-defined size (Fig. 4A), by a large increase in
the fraction of biotinylated full-length molecules (Fig. 5 E), and
by a strong reduction in the steady-state level Lep-reactive protein (Fig. 5 A).
Correlation between stop-transfer activity in E. coli and in
dog pancreas microsomes. To be able to directly compare stoptransfer function in E. coli and in the mammalian endoplasmic
reticulum (ER), some of the constructs analyzed above were
cloned into the pGEM1 eukaryotic expression vector and proteins were translated in a reticulocyte lysate in the absence and
presence of dog pancreas microsomes. In these experiments,
PhoA fusions were not used, but the entire P2 domain from Lep
with a potential glycosylation-acceptor site in position 258 was
retained since N-linked glycosylation is a reliable marker for
translocation of the P2 domain to the lumenal side of the microsomal membrane [32]. In addition, a second glycosylation-acceptor site was introduced in position 96 located between H2
and the insertion site for the pseudo-random sequences (Fig. 6).
Thus, inserts devoid of stop-transfer function should be glycosylated on both sites, whereas those with stop-transfer function
should only be glycosylated on Asn96.
As shown in Fig. 6, clone no. 93 which has an insert of low
mean hydrophobicity was efficiently glycosylated on both sites
(Fig. 6, lane 2), and a fragment corresponding to the doubly glycosylated H2-P2 domain was protected from degradation upon
proteinase K treatment (Fig. 6, lane 5). When the reaction was
carried out in the presence of a competitive peptide inhibitor of
glycosylation, both doubly, singly, and non-glycosylated molecules were visible (Fig. 6, lane 3). In contrast, clone no. 62 was
glycosylated only on one site and was not protected from proteinase K digestion, as expected if the hydrophobic insert has
stop-transfer function.
Sääf et al. (Eur. J. Biochem. 251)
827
Fig. 6. Pseudo-random segments have similar stop-transfer activities in E. coli and in dog pancreas microsomes. On the left, clone no. 62 is
glycosylated once (the pseudo-random insert has stop-transfer function), whereas clone no. 93 is glycosylated twice (the pseudo-random insert lacks
stop-transfer function). Proteins were expressed in vitro from plasmid pGEM1 either in the absence or presence of rough microsomes (RM). An
acceptor peptide (AP, a competitive inhibitor of N-linked glycosylation), a related non-acceptor peptide (NAP), proteinase K (PK), and Triton X-100
(T-X) were included as indicated. Note that the shift in molecular mass in the presence of microsomes (lane 2) is smaller for clone no. 62 than
for no. 93, indicative of single and double glycosylation, respectively. The lumenal location of the P2 domain in clone no. 93 is shown by the
protease-protected fragment remaining after proteinase K treatment of the intact microsomes (lane 5) ; no such fragment is seen for clone no. 62.
We have previously shown that the glycosylated, protease-protected H2-P2 fragment in microsome-inserted wild-type Lep has the same mobility
on SDS/PAGE as non-glycosylated, full-length Lep [32]; the somewhat higher mobility of the protected fragment in lane 5 is consistent with the
presence of the extra glycosylation site. On the right, expression of the original Lep construct without an inserted, pseudo-random segment (no. 1113)
and clones nos 34, 66, 67, 68 and 92 in the absence and presence of rough microsomes. Non-glycosylated products are indicated by a black dot,
singly glycosylated products by a white dot, and doubly glycosylated products by two white dots. The Lep construct used is schematically shown.
Two unique potential acceptor sites for N-linked glycosylation at Asn96 and Asn258 provide topological markers, since glycosylation is only
possible in lumenally exposed domains. Note that the intact Lep protein, not the Lep/PhoA fusion, is used.
Fig. 7. Kinetic analysis of the formation of singly and doubly glycosylated molecules during translocation across the microsomal membrane. Protein synthesis was initiated at time zero by addition of reticulocyte lysate to the mRNA/microsome mix (Fig. 6). Further chain initiation was blocked after 1.5 min by addition of the inhibitor ATA. At the
indicated times, aliquots were removed and Triton X-100 was added to
solubilize the microsomes and prevent further glycosylation. Translation
was continued up to a total time of 1 h, and the amounts of singly and
doubly glycosylated molecules were then determined by SDS/PAGE and
phosphoimager analysis. Results (mean of two experiments) for clone
no. 66 are shown by circles and for clone no. 67 by squares. Open symbols, singly glycosylated molecules ; filled symbols, doubly glycosylated
molecules.
Similar experiments were carried out for a number of other
clones, Fig. 6; singly and doubly glycosylated molecules could
easily be distinguished on the basis of the difference in molecular mass between the non-glycosylated and glycosylated forms
(<2 kDa/added glycan). The results are summarized in Figs 2
and 3C, where singly glycosylated clones are indicated by white
squares and doubly glycosylated ones by gray squares. Only in
one case (no. 92) was the behavior found to be significantly different in E. coli and in microsomes: while this sequence of
rather low mean hydrophobicity gives rise to a fairly stable STfragment (Fig. 4 B), it lacks detectable stop-transfer function in
the microsomal system.
To further characterize the kinetics of translocation of clones
such as no. 67, which gives rise to a slowly disappearing STfragment in E. coli, the appearance of singly and doubly glycosylated molecules was followed using an approach first introduced by Rothman and Lodish [33, 48]. 1.5 min after initiation
of translation in the presence of microsomes, the inhibitor ATA
was added to block further chain initiation. The timing of glycosylation was then determined by removing aliquots from the reaction mixture at different times, immediately adding the detergent Triton X-100 to these aliquots to solubilize the microsomes, and continuing translation up to a total time of 1 h. Control experiments showed that no glycosylation took place after
detergent solubilization of the microsomes, whereas translation
proceeded unhindered to completion (data not shown). As
shown in Fig. 7, the kinetics of appearance of both singly and
doubly glycosylated molecules were indistinguishable for clones
nos 66 and 67, suggesting that the slow translocation of the marginally hydrophobic segment in clone no. 67 is peculiar to E.
coli. A possible caveat is that translation is so slow in the in
vitro system that it may mask a temporary slowing down of
translocation; this can only be discounted by in vivo studies. In
any event, there is no suggestion from the present data that
short-lived transmembrane intermediates can be generated in the
microsomal system, in contrast to E. coli.
828
Sääf et al. (Eur. J. Biochem. 251)
DISCUSSION
We have generated a collection of pseudo-random, 18-residue amino acid segments, and have tested their stop-transfer
function, both in vivo in E. coli and in a eukaryotic in vitro
system supplemented with ER-derived microsomes. In general,
we find a good correlation between stop-transfer function and
mean hydrophobicity in both systems, and we have not found
any strongly hydrophobic sequences lacking stop-transfer function or any strongly hydrophilic ones with stop-transfer activity.
The apparent threshold between sequences with and without
measurable stop-transfer activity is approximately 1.5 as calculated by the TOPPRED algorithm (Fig. 3 C). This value is somewhat lower than the threshold defined in a recent study where
the stop-transfer activity of 21-residue sequences composed only
of Ala and Leu residues was measured in E. coli [2]; using the
same algorithm and hydrophobicity scale as above, one obtains
a threshold of approximately 1.8 in this case. We thus conclude
that overall hydrophobicity is the major determinant of stoptransfer function both in E. coli and in the ER, and that the
detailed amino acid sequence of the stop-transfer sequence,
while not without effect, is only of secondary importance.
Strictly speaking, since the translocation of the periplasmic loop
preceding the randomized segments is SecA- and ∆µH1-dependent as assayed by its sensitivity to treatment with azide and
CCCP (data not shown), this conclusion is valid only for membrane proteins that use the Sec pathway for membrane insertion
[49]. Beyond hydrophobicity, it is known that both N-terminally
and C-terminally flanking segments can influence stop-transfer
activity [3, 50] and thus also influence the exact value of the
hydrophobicity threshold.
In spite of the good correlation between overall hydrophobicity and stop-transfer function, the stop-transfer activity of sequences near the apparent hydrophobicity threshold cannot be
accurately predicted using any of the standard hydrophobicity
scales, suggesting that more specific features than overall hydrophobicity may play a role for such sequences. The most obvious
case is clone no. 92, which has a rather low calculated hydrophobicity, yet gives rise to a nearly stable ST-fragment. The
presence of two central prolines in this sequence, (Table 1), may
suggest that prolines are less unfavorable for stop-transfer function in E. coli than suggested by the hydrophobicity analysis.
Unfortunately, our present collection of sequences is too small
to allow a reliable statistical analysis of residue-specific or position-specific effects. The identification of such second-order corrections to predictions based on mean hydrophobicity may be
one way to further improve current prediction algorithms.
One interesting result of this study is the unexpected finding
that certain marginally hydrophobic sequences appear to remain
in a transmembrane disposition for many minutes prior to their
disappearance from the inner membrane. Using biotinylatable
fusion proteins to characterize the rate of translocation, it was
found that the C-terminal domain of clone no. 67 (which has
a marginally hydrophobic segment) appears to be translocated
significantly more slowly than the corresponding regions in
clones that lack hydrophobic segments in this domain (Fig. 5).
In addition, the low steady-state level of clone no. 67 suggests
that the transmembrane intermediate is also sensitive to endogenous proteases.
A possible scenario is that marginally hydrophobic sequences pause in the translocation channel but are not hydrophobic enough to be able to move laterally into the lipid bilayer to
become bona fide transmembrane anchors. Instead, either the
continued action of the Sec machinery will eventually overcome
the translocation block, or the protein will be degraded. Another
possibility is that marginally hydrophobic sequences move out
of the translocon into the lipid bilayer but that their weaker interaction with the lipids can make them more susceptible to degradation. Both models are broadly in agreement with recent
results using chemical cross-linking to follow the progression of
a transmembrane segment from its initial location within the ER
or E. coli translocons to its final location the lipid bilayer [512
53]. In vitro studies using inverted E. coli inner membrane vesicles have also shown that certain short hydrophobic segments in
the pro-OmpA protein pause in the translocon during translocation [54].
In any case, our results suggest that stop-transfer sequences
may be proof-read twice: first, by their ability to pause in the
translocon, and, second, by their ability to become stably anchored in the lipid bilayer rather than translocated or degraded;
essentially the same conclusion was reached recently on the basis of an in vitro study of the translocation of a marginally hydrophobic segment engineered into pro-OmpA [55]. A possible
advantage of such two-step recognition is that it would leave
segments of intermediate hydrophobicity (e.g., hydrophobic segments containing one or a few charged residues) more time to
find their way into the bilayer or to interact with other hydrophobic segments in the nascent protein, and thus allow for a wider
range of sequence variation in stop-transfer segments.
Finally, we note that the criteria for functional stop-transfer
sequences appear to be largely conserved between prokaryotes
and eukaryotes, and that most sequences behave similarly in the
two systems, (Fig. 6). However, we have come across one, marginally hydrophobic, sequence (no. 92) that has stop-transfer
function in E. coli but not in microsomes, pointing to the existence of subtle differences in the way the two translocons interact with nascent polypeptides in transit.
We are indebted to Elin Grahn and Dr Paul Whitley for expert assistance during the initial stages of the project. Plasmid pCY66 was a gift
from Dr Jon Beckwith, Harvard Medical School. This work was supported by grants from the Swedish Natural Sciences Research Council
(NFR), the Swedish Technical Sciences Research Council (TFR), the
Swedish Cancer Foundation, and the Göran Gustafsson Foundation to
G.v. H.
REFERENCES
1. von Heijne, G. (1997) Mol. Microbiol. 24, 2492253.
2. Chen, H. F. & Kendall, D. A. (1995) J. Biol. Chem. 270, 14 1152
14 122.
3. Kuroiwa, T., Sakaguchi, M., Mihara, K. & Omura, T. (1991) J. Biol.
Chem. 266, 925129255.
4. Davis, N. G. & Model, P. (1985) Cell 41, 6072614.
5. Davis, N. G., Boeke, J. D. & Model, P. (1985) J. Mol. Biol. 181,
1112121.
6. Zerial, M., Huylebroeck, D. & Garoff, H. (1987) Cell 48, 1472155.
7. Adams, G. A. & Rose, J. K. (1985) Cell 41, 100721015.
8. Adams, G. A. & Rose, J. K. (1985) Mol. Cell. Biol. 5, 144221448.
9. Nilsson, I., Gafvelin, G. & von Heijne, G. (1993) FEBS Lett. 318,
7210.
10. Seligman, L. & Manoil, C. (1994) J. Biol. Chem. 269, 19 8882
19 896.
11. von Heijne, G. (1994) Annu. Rev. Biophys. Biomol. Struct. 23, 1672
192.
12. Fasman, G. D. & Gilbert, W. A. (1990) Trends Biochem. Sci. 15,
89292.
13. Kaiser, C. A., Preuss, D., Grisafi, P. & Botstein, D. (1987) Science
235, 3122317.
14. Kaiser, C. A. & Botstein, D. (1990) Mol. Cell. Biol. 10, 316323173.
15. Baker, A. & Schatz, G. (1987) Proc. Natl Acad. Sci. USA 84, 31172
3121.
16. Lemire, B. D., Fankhauser, C., Baker, A. & Schatz, G. (1989) J.
Biol. Chem. 264, 20 206220 215.
Sääf et al. (Eur. J. Biochem. 251)
17. Erickson, B. W. & Merrifield, R. B. (1976) in The Proteins (Hill,
R. L. & Neurath, H., eds) pp. 2552527, Academic Press, London.
18. Dalbey, R. E. & Wickner, W. (1986) J. Biol. Chem. 261, 13 8442
13 849.
19. Lee, E. & Manoil, C. (1994) J. Biol. Chem. 269, 28 822228 828.
20. Jander, G., Cronan, J. E. & Beckwith, J. (1996) J. Bacteriol. 178,
304923058.
21. Johnston, S., Lee, J. H. & Ray, D. S. (1985) Gene (Amst.) 34, 1372
145.
22. Kunkel, T. A. (1985) Proc. Natl Acad. Sci. USA 82, 4882492.
23. Geisselsoder, J., Witney, F. & Yuckenberg, P. (1987) Biotechniques
5, 7862791.
24. Whitley, P., Nilsson, I. & von Heijne, G. (1994) Nat. Struct. Biol.
1, 8582862.
25. Johansson, M., Nilsson, I. & von Heijne, G. (1993) Mol. Gen. Genet.
239, 2512256.
26. Kozak, M. (1989) Mol. Cell. Biol. 9, 507325080.
27. Rusch, S. L., Chen, H. F., Izard, J. W. & Kendall, D. A. (1994) J.
Cell. Biochem. 55, 2092217.
28. Oliver, D. B., Cabelli, R. J., Dolan, K. M. & Jarosik, G. P. (1990)
Proc. Natl Acad. Sci. USA 87, 822728231.
29. Manoil, C. (1991) Methods Cell. Biol. 34, 61275.
30. Harlow, E. & Lane, D. (1988) Antibodies 2 a laboratory manual,
Cold Spring Harbor Laboratory Press, Cold Spring Harbor NY.
31. Liljeström, P. & Garoff, H. (1991) J. Virol. 65, 1472154.
32. Nilsson, I. & von Heijne, G. (1993) J. Biol. Chem. 268, 579825801.
33. Garoff, H., Huylebroeck, D., Robinson, A., Tillman, U. & Liljeström, P. (1990) J. Cell. Biol. 111, 8672876.
34. Engelman, D. M., Steitz, T. A. & Goldman, A. (1986) Annu. Rev.
Biophys. Biophys. Chem. 15, 3212353.
35. Claros, M. G. & von Heijne, G. (1994) Comput. Appl. Biosci. 10,
6852686.
829
36. von Heijne, G. (1992) J. Mol. Biol. 225, 4872494.
37. Wolfe, P. B., Rice, M. & Wickner, W. (1985) J. Biol. Chem. 260,
183621841.
38. Wolfe, P. B. & Wickner, W. (1984) Cell 36, 106721072.
39. Derman, A. I. & Beckwith, J. (1991) J. Bacteriol. 173, 771927722.
40. Prinz, W. A. & Beckwith, J. (1994) J. Bacteriol. 176, 641026413.
41. Manoil, C. & Beckwith, J. (1986) Science 233, 140321408.
42. Akiyama, Y. & Ito, K. (1989) J. Biol. Chem. 264, 4372442.
43. Chou, M. M. & Kendall, D. A. (1990) J. Biol. Chem. 265, 28732
2880.
44. San Millan, J. L., Boyd, D., Dalbey, R., Wickner, W. & Beckwith,
J. (1989) J. Bacteriol. 171, 553625541.
45. Boyd, D. & Beckwith, J. (1989) Proc. Natl Acad. Sci. USA 86,
944629450.
46. Andersson, H. & von Heijne, G. (1991) Proc. Natl Acad. Sci. USA
88, 975129754.
47. Akiyama, Y. & Ito, K. (1993) J. Biol. Chem. 268, 814628150.
48. Rothman, J. & Lodish, H. (1977) Nature 269, 7752780.
49. de Gier, J.-W. L., Valent, Q., von Heijne, G. & Luirink, J. (1997)
FEBS Lett. 408, 124.
50. Yost, C. S., Lopez, C. D., Prusiner, S. B., Myers, R. M. & Lingappa,
V. R. (1990) Nature 343, 6692672.
51. Do, H., Falcone, D., Lin, J., Andrews, D. W. & Johnson, A. E.
(1996) Cell 85, 3692378.
52. Mothes, W., Heinrich, S., Graf, R., Nilsson, I., von Heijne, G., Brunner, J. & Rapoport, T. (1997) Cell 89, 5232533.
53. Sato, K., Mori, H., Yoshida, M., Tagaya, M. & Mizushima, S. (1997)
J. Biol. Chem. 272, 20 082220 087.
54. Sato, K., Mori, H., Yoshida, M., Tagaya, M. & Mizushima, S. (1997)
J. Biol. Chem. 272, 588025886.
55. Duong, F. & Wickner, W. (1998) EMBO J., in the press.