Structure of the cytoplasmic domain of p23 in solution. Implications for the formation of
COPI vesicles
Marcus Weidler†, Constanze Reinhard*, Felix Wieland*, and Paul Rösch†)
*
Biochemie-Zentrum Heidelberg (BZH), Ruprecht-Karls-Universität Heidelberg
†
Institut für Struktur und Chemie der Biopolymere, Universität Bayreuth
§
) Corresponding author:
Prof. Dr. Paul Rösch
Institut für Struktur und Chemie der Biopolymere
Universität Bayreuth
Universitätsstr. 30
D-95440 Bayreuth
Tel.:
+49-921-553540
Fax.: +49-921-553544
E-mail:
paul.roesch@uni-bayreuth.de
1
ABSTRACT
Coatomer, the coat protein complex of COPI vesicles, is involved in the budding of these
vesicles, but the underlying mechanism is unknown. Here, we present the structure of peptides
analogous to the cytoplasmic domain of p23 as determined by two-dimensional nuclear
magnetic resonance spectroscopy. These peptides are able to interact with coatomer and to
induce a conformational change of it, leading to the polymerization of the complex. An
improved strategy for structure calculation revealed that the peptides form -helices and adopt
a tetrameric state. Based on these results we propose the first model for the binding of
coatomer by p23 and the thereby induced conformational change of coatomer which results in
its polymerization, and thus drives formation of the bud on the Golgi membrane during
biogenesis of a COPI vesicle.
2
COPI-coated vesicles mediate protein transport in the early secretory pathway[Rothman, 1994
#1; Rothman, 1996 #2; Schekman, 1996 #4]. The COPI coat consists of a small G protein,
ADP-ribosylation factor 1 (ARF1) [Serafini, 1991 #6; Taylor, 1992 #7] and coatomer, a
hetero-oligomeric protein complex of seven subunits (- COPs, for coat proteins) [Waters,
1991 #8; Stenbeck, 1993 #9; Harter, 1995 #10]. COPI bud formation is initiated by membrane
recruitment of ARF1, which in its GTP-bound form [Donaldson, 1992 #11; Palmer, 1993
#12], together with members of the p24 family, provides membrane binding sites for coatomer
[Stamnes, 1995 #13; Nickel, 1997 #14]. Subsequent coat assembly leads to membrane
deformation and the morphological appearance of a bud [Orci, 1993 #15]. Recently, it was
shown that upon interaction with peptides analogous to the cytoplasmic domain of p23, a
member of the p24 family [Sohn, 1996 #16], coatomer undergoes a conformational change,
which induces the polymerization of the complex [XXX Paper von Constanze XXX]. p23, a
type I transmembrane protein, is highly enriched in COPI vesicles and is present in a ratio to
coatomer of approximately 4:1 [Sohn, 1996 #16]. The cytoplasmic domain of p23
(YLRRFFKAKKLIE) is structurally similar to a classical dilysine motif (KKXX) for retrieval
to the endoplasmic reticulum (ER) [Nilsson, 1989 #17; Jackson, 1993 #19; Jackson, 1990
#18; Cosson, 1994 #20], and binds coatomer with the same efficiency as the KKXX motif
[Sohn, 1996 #16]. p23 binding, however, depends on its phenylalanine residues as well as its
lysine residues.
Interestingly, a dimeric form of the above mentioned peptide which is disulfid-bridged via
additional cysteines at the N-terminus is far more efficient in inducing the conformational
change of coatomer leading to the formation of a bud [XXX Paper von Constanze XXX]:
3
_1 5 9 13
p23w
t-m _YLRRFFKAKKLIE
CYLRRFFKAKKLIE
S
S
CYLRRFFKAKKLIE
XXX (Eventuell entfallen lassen) The essential components needed for the biogenesis of
p23w
t-d
additional coated vesicles are defined as well. Clathrin coated [Takei, 1998 #21] and COPIIcoated vesicles [Matsuoka, 1998 #22] have been generated in vitro from defined constituents.
Most recently, a conformational change was shown upon binding of GTP to Dynamin
[Sweitzer, 1998 #23], and this structural change was correlated to the process of pinching off
a coated vesicular intermediate. However, two questions remain: (i) how are coat proteins
normally bound and (ii) how does this association lead to a mechanical deformation of a
membrane. XXX
In view of the central role played by the cytoplasmic domain of p23 in the formation of COPIcoated vesicles, the determination of the three-dimensional structure of peptides analogous to
this domain is a problem of considerable biochemical interest.
We decided to determine the structure of these peptides in the presence of the crowding
reagent, 2,2,2-trifluoroethanol, in order to approximate the conditions on the surface of a
membrane. Our studies show that the small cytoplasmic tail domains of p23 adopt a tetrameric
state, and offer a model as to how the amino acid residues essential for binding coatomer are
positioned within the tetramer on the surface of the membrane
The strikingly higher efficiency of p23wt-d to aggregate coatomer might imply that the
dimeric peptide adopts a distinct and defined structure in solution. Therefore, a complete
structural analysis of p23wt-m and p23wt-d by NMR techniques was performed in aqueous
buffer solution with addition of various amounts of trifluoroethanol (TFE) [Luo, 1997 #45].
4
MATERIALS AND METHODS
Synthetic Peptides. p23wt-m and p23wt-d (sequences are shown above) were obtained as a
commercial product (Deutsches Krebsforschungszentrum, Heidelberg). Dimers were formed
by disulfide bridges linking the peptides via N-terminally introduced cysteine residues. For
this purpose, newly synthesized monomeric peptides were oxidized in 20 % dimethyl
sulfoxide in water for 48 h. Subsequently the dimers were isolated by high pressure liquid
chromatography.
Nuclear Magnetic Resonance Spectroscopy of p23wt-m and p23wt-d. The solution
structures of p23wt-m and p23wt-d were obtained from 2D 1H NMR data (4.0 mM peptide,
pH 3.6 in 9:1 H2O:D2O, 8:2 H2O:d2-TFE, 7:3 H2O:d2-TFE, or 6:4 H2O:d2-TFE, 100 mM
potassium phosphate buffer, 50 mM NaCl, 280 K. Complete sequence-specific assignments of
backbone and side-chain protons were obtained by total correlation spectroscopy (TOCSY)
with 80 ms mixing time [Braunschweiler, 1983 #27; Griesinger, 1988 #28], correlation
spectroscopy (COSY) [Jeener, 1979 #29], and NOE spectroscopy (NOESY) (mixing times
150 ms and 300 ms) [States, 1982 #30]. All spectra were acquired on a Bruker DRX600
spectrometer using standard pulse sequences [Wüthrich, 1986 #31]. Saturation of the water
signal was accomplished by continuous coherent irradiation prior to the first excitation pulse
and during the mixing time of the NOESY experiments. 4096  512 data points were
collected with a spectral width of 6024 Hz in both dimensions. Base-line correction up to 6th
order was performed for all two-dimensional spectra along the F2 as well as the F1 dimension.
A sinebell-squared filter with a phase shift of /2, /4 or /8 prior to Fourier transformation
was used. Zero-filling resulted in a data size of 4096  1024 data points in the frequency
domain. In addition to the standard Bruker spectrometer control software, the NDEE 2.0
5
software package (Software Symbiose, Inc., Bayreuth, Germany) was used for data
processing. Chemical shift values are reported relative to 2,2-dimethyl-2-silapentane
sulfonate.
For structure calculations, only NOEs visible in the spectra recorded in the presence of 40 %
TFE with 150 ms mixing time were taken into account. Identical calculations combining
information obtained in the presence of 20 % TFE resulted in virtually identical structures for
p23wt-d. The structure of p23wt-m, however, could not be determined at TFE concentrations
lower than 40 %.
An estimate of secondary structure element can be obtained from the proton chemical shifts
[Wishart, 1995 #52; Wishart, 1995 #53; Wishart, 1992 #54; Wishart, 1991 #55]. CH
resonances shifted to high field relative to the corresponding random coil values [Wishart,
1995 #52] indicate local -helical structure, whereas downfield shifted resonances are typical
for local -sheet conformation. Elements of regular secondary structure are assumed to be
present if a deviation from the random coil value of more than 0.1 ppm is observed. To get a
more reliable picture it is suggested that only resonances with the same sense chemical shift
deviation for a stretch of more than 3 sequential residues be taken into account [Wishart, 1995
#52].
Restrained And Unrestrained Molecular Dynamics calculations.
Simulated annealing calculations [Nilges, 1988 #36] were performed on DEC Alpha
workstations with X-PLOR V3.840 [Brünger, 1993 #37]. The 3JN coupling constants were
obtained from cross-peak measurements in a COSY spectrum. The NOE intensities were
classified as strong, medium, and weak and assumed to correspond to proton-proton distances
of 1.8 - 2.7 Å, 1.8 - 4.0 Å, and 1.8 - 5.5 Å, respectively. Stereospecific assignments for the
methyl, methylene, and aromatic protons have not been performed, and appropriate
corrections were added for constraints including pseudoatoms[Clore, 1987 #32; Wagner, 1987
6
#33; Qi, 1994 #34]. Torsion angles were derived from 3JN coupling constants [Pardi, 1984
#35]. An interval of 30° around the experimental value was allowed. Frequency degenerated
cross-peaks were incorporated into the structure calculations as ‘ambiguous’ in order to
extract as much structural information as possible from the NOESY spectrum [Brünger, 1993
#37]. Subsequently, the proton-proton distances in the calculated structures were determined
using the program ‘BackCalc_db 2.0’ (Software Symbiose, Inc., Bayreuth, Germany) and
compared with the combinations of distances possible for each frequency degenerated
NOESY cross-peak. If only one of the possible distance combinations was fulfilled in more
than 50% of the calculated structures, the distance information was used in further structure
calculations. This procedure was repeated several times.
Elements of regular secondary structure were deduced by chemical shift analysis (cf. above)
and the inspection of the NOE pattern. In addition, the structures were checked for the
existence of secondary structural elements by MOLMOL 2.5.1 [Koradi, 1996 #56].
For the unrestrained molecular dynamics simulation (MD) a water/TFE box consisting of
3392 water and 768 TFE molecules with a spatial extension of 6.2 x 6.2 x 6.2 nm was
generated [Jorgensen, 1983 #39; Sticht, 1994 #38]. The structure with the lowest energy was
chosen as the starting structure for the MD calculations. The further calculation strategy was
described earlier [Jorgensen, 1983 #39; Sticht, 1994 #38; Powell, 1977 #40; Verlet, 1967 #41;
van Gunsteren, 1977 #42].
Size Exclusion Chromatography of Monomeric and Dimeric p23wt. Size exclusion
chromatography was performed on a SMART System (Pharmacia Biotech) using the
Superdex Peptide PC 3.2/30. The column was equilibrated with buffer 1 at a flow rate of 40
µl/min at 8°C. Peptides were dissolved in buffer 1 at a concentration of 1 mg/ml (Fig. 7A, B)
or 0.1 mg/ml (Fig. 7D) and injected as indicated in the figure. Chromatography of the peptides
was performed at a flow rate of 20 µl/min at 8°C and fractions of 40 µl were collected.
7
8
RESULTS
Sequence-specific assignments of the spin systems identified in the COSY and TOCSY
spectra could be performed with standard techniques [Wüthrich, 1986 #31] as the spectra
were well resolved (Tab. 5-2, Tab. 5-5, Dipl. Arbeit). Chemical shift data of the CH
resonances was analyzed according to the chemical shift index strategy [Wishart, 1995 #53;
Wishart, 1992 #54], yielding an estimate of elements of regular secondary structure (Fig.
Wishart-Plot). This preliminary estimate was confirmed and refined by analysis of NOESY
crosspeak patterns: According to sequential short-range dNN(i,i+1) NOEs, medium-range
d(i,i+3), d(i,i+3), and d(i,i+4) NOEs (Fig. 5A, B), both, p23wt-m and p23wt-d form helices from L2 to I12, fringing at Y1 and E13.
For p23wt-m, 180 intraresidual, 179 interresidual NOE connectivities and 10 dihedral angle
constraints derived from coupling constants (Tab. 2) were used in an SA protocol with
subsequent refinement to generate 100 structures from an elongated starting conformation.
The 30 structures with lowest total energies were superimposed, and the local root mean
square deviation (rmsd) of the backbone atoms were calculated. The four residues involved in
coatomer binding, F5, F6, and K9, K10 [Sohn, 1996 #16], are located at the same side of the
helix (Fig. 6B, C). Remarkably, this helix clearly shows amphipathic character, its
hydrophobic face composed of residues Y1, F5, A8, and I12. Amphipathicity is increased by a
salt bridge between K9 and E13, stable in molecular dynamics simulations (Fig. 5-26, Dipl.
Arbeit). The root-mean-square deviation (RMSD) for the backbone heavy atoms of the 10
structures with lowest internal energies of p23wt-m is 0.17 Å (Fig. 6A). A break down of the
NOEs of both p23wt-m and p23wt-d into short, medium and long range is listed in table 1, as
are average violations of the distance and dihedral restraints in the final ensemble of
9
conformers, deviations of the structures from ideal geometries and rms deviations from the
average coordinates.
The structure of p23wt-d, calculated under the assumption of a monomeric state of the
dimerized peptide (data not shown), could not satisfactorily explain 15 long-range NOEs
between I12 and Y1 and I12 and F5 (at lower TFE concentrations, 20 and 15 % (v/v),
additional NOEs between F6 and E13 could be observed). This implies oligomerization of
p23wt-d at least to a p23wt-d dimer, analogous to a tetrameric form of the p23 cytoplasmic
domain. In order to analyze this dimerization of p23wt-d biochemically, size exclusion
chromatography was performed at pH 7.4 without the addition of TFE and under the
conditions used for NMR spectroscopy. p23wt-m and p23wt-d reduced by dithiothreitol elute
in a single peak (Fig. 7A). p23wt-d, however, gives rise to two peaks (Fig. 7B): Peak 2 elutes
with the same retention time as p23wt-m (Fig. 7A), indicating very similar Stokes’ radii of the
two species, whereas peak 1, judged from its retention time, represents a p23wt-d dimer.
Accordingly, rechromatography of the fraction corresponding to peak 1 in Fig. 7B results in
an elution profile similar to the p23wt-d sample applied in Fig. 7B. However, a different ratio
of peak sizes is observed with peak 2 enlarged at the expense of peak 1 (Fig. 7C), indicating
an equilibrium of the two states of the peptide. This equilibrium depends on the concentration
of the peptide (Fig. 7D), revealing its bimolecular nature. Thus, peak 1 is caused by the
dimeric, and peak 2 by the monomeric state of p23wt-d. Size exclusion chromatography
clearly confirms a dimeric state of p23wt-d even in the absence of TFE (Experiments carried
out under the conditions used for NMR spectroscopy yielded identical results).
These results clearly show that p23wt-d is dimerized in solution. Possible arrangements of the
two peptides of a dimer are sketched in fig. XXX. Comparing the experimental NMR spectra
with the spectra expected for the four possible structures of a dimer, only one structure is
plausible (fig. XXXD). In detail: Structure A can not explain the 15 observed NOEs between
10
I12 (near COOH-terminus) and Y1 or F5 (near NH2-terminus). The same is true for model B.
While model C fulfills the above mentioned distance restraints, it does not reflect the
symmetry detected by NMR spectroscopy. In the NMR spectra the four amino acids with the
same sequence position give rise only to one spin system, i.e. the magnetic environment of the
four residues must be identical (Fig. Ausschnitt aus TOCSY-Spectrum). Only an arrangement
of the two p23wt-d with a XXX symmetry depicted in Fig. XXD meets all experimental
constraints: (i) dimeric; (ii) helical conformation; (iii) proximity of NH2- and COOHterminus; (iv) an identical magnetic environment for the four helices of a p23wt-d dimer.
In the Clean-TOCSY spectrum of p23wt-d dimer three different spin systems, corresponding
to three different conformations, for each of the COOH-terminal residues I12 and E13 are
discernible. In both cases, however, only one of the spin systems shows interresidual NOEs
(data not shown), which were then used in the structure determination.
No indication for an oligomerization state of the dimerized peptide higher than a dimer
corresponding to four -helices could be observed.
The overall structure of the p23wt-d dimer is bipartite (Fig. 6C). The -helices are well
defined, and a hydrophobic core is formed through interactions of the side chains of residues
F5, A8, L11, and I12 of the two dimers. A halfcircle is formed by residues K9 and K10 of two
antiparallel helices (Fig. 6C). The corresponding diphenylalanine motifs of the two
antiparallel -helices are separated by this "K-halfcircle" (Fig. 6C). This quaternary structure
"FKF motif" is present twice in the tetrameric arrangement. Thus, p23wt-d dimer comprises a
tetramer of the cytoplasmic domain of p23. Likewise, the stoichiometry found for both p23
and coatomer in isolated COPI vesicles and in coatomer precipitated by p23 wt-d was
approximately 4:1 XXX Paper von Constanze XXX. Taken together, from this data we
propose that this tetramer is the species active in coatomer binding in vivo (Fig. 8).
11
The stability of the calculated structure was probed by and MD simulation over 200 ps and in
Fig. XXX (Abb. 5-35 Dipl. Arbeit) the radius of gyration, a measure for the compactness of a
molecule, as a function of the simulation time is shown. After an rapid increase which is
caused by the equilibration of the system, the radius of gyration decreases slightly from 12.0
to 11.8 Å.
12
DISCUSSION
The structure of p23 that triggers the recently observed conformational change of coatomer
[XXX Paper von Constanze XXX] is a tetramer of four equivalent -helical domains, as
judged from nuclear magnetic resonance spectroscopy and molecular dynamics simulations. A
tetrameric form of this domain is confirmed by size exclusion chromatography and
determination of its stoichiometry in the precipitated coatomer complex [XXX Paper von
ConstanzeXXX]. The structure arising from tetramerization of the p23-tail domain represents
a symmetrical molecule of two equivalent half's, two "FKF"-motifs (Fig. 6 C). We figure that
one such motif is directed towards the surface of the Golgi membrane and may well be
stabilized by interactions with membrane lipid headgroups, and the other motif protrudes from
the membrane and serves as a "plug" for coatomer.
In vitro precipitation of coatomer was recently observed in the presence of some bivalent
aminoglycoside antibiotics [Hudson, 1997 #46], and interpreted in terms of aggregation of
coatomer by crosslinking. In light of the data described here these aminoglycosides may well
fit into the binding site for the cytoplasmic domain of p23 that resides in the -subunit of the
complex [Harter, in press #44] and induce its conformational change.
It is of note that a Wbp1p-peptide with a characteristic ER-retrieval motif does bind coatomer
but, in contrast to the cytoplasmic domain of p23, does not trigger a conformational change of
the complex. Thus, the two classes of domains have distinct and different functions:
interaction of the Wbp1p-type of peptide might serve the sorting of cargo to be retrieved to the
ER into retrograde COPI vesicles. p23, however, may represent part of the machinery of a
13
COPI vesicle. Not only does it bind to coatomer during the budding reaction, but it also
triggers a conformational change that leads to polymerization of the complex (Fig. 8).
The stoichiometry described here in the precipitate between p23 peptide and coatomer and the
solution structure of the dimeric peptide strongly favour a tertramerized state of p23 when it
binds to coatomer in vivo. Similarly, other members of the p24 family known to bind
coatomer and bearing a diphenylalanine and dilysine motif might interact in a tetrameric state
with coatomer. Thus receptor induced polymerization of coatomer might represent a general
mechanism for COPI vesicle formation. However, at present it is not known in which
stoichiometries the various p24 family members interact with each other [Dominguez, 1998
#47] or whether they are able also to form homooligomers. In any case, various oligomers
would allow different classes of COPI vesicle to exist, specified by the different p24 members
they contain. According to this model, oligomerization of p23 and its relatives must be
regulated in vivo. Such regulation might occur via interaction of p24 members with cargo, if
p24 members serve as cargo receptors [Schimmoller, 1995 #48; Fiedler, 1996 #49; Kuehn,
1998 #50]. Another candidate to regulate oligomerization of p24 members is ARF 1, the
recruitment of which to Golgi membranes precedes binding of coatomer. However, such a
role of ARF 1 in oligomerization of p24 members waits to be elucidated.
As coatomer is used in many cycles, an important question concerns a reversibility of the
conformational change of complex. Since ARF 1 is a component of COPI vesicle and known
to be involved in the uncoating process by hydrolysis of its bound GTP [Tanigawa, 1993 #51],
this energy providing step might well help to reverse the conformational change of coatomer
and thus dissociate the coat.
14
In general, receptor induced polymerization of coat proteins during their recruitment would
strongly increase the efficiency of the coating process, and with the resulting geometry ruled
by the nature of the conformationally changed proteins, may well be the driving force to shape
a membrane.
15
FIGURE LEGENDS
Fig. 1. Precipitation of coatomer with synthetic peptides corresponding to the COOH-termini
of p23wt, p23AS and Wbp1p. (A) Peptides used in this study. p23wt represents the
cytoplasmic domain of p23 which binds coatomer depending on its dilysine and
diphenylalanine motifs, whereas p23AS lacks these residues and therefore does not bind
coatomer. Wbp1p represents the cytoplasmic domain of a subunit of the yeast Noligosaccharyl transferase complex bearing a characteristic KKXX ER-retrieval motif known
to bind coatomer. (B, C) Precipitation of coatomer (0.09 µM) with monomeric peptides (B)
and dimeric peptides (C). Only p23wt (filled squares) precipitates the complex, whereas
mutated p23 peptide (p23AS, filled circles) and Wbp1p (filled triangles) have no effect. The
error bars indicate standard errors (n=3).
Fig. 2. Limited proteolysis of coatomer. Proteolysis of coatomer (0.09 µM) with thermolysin
(0.008 µM) in the presence of dimerized peptides (50 µM), p23wt-d (A), p23AS-d (B) and
Wbp1p-d (C). Western blot analysis of proteolytic fragments of the -COP subunit reveals a
better susceptibility of this subunit to the protease in the presence of p23wt-d (A, lanes 2, 3) as
compared to p23AS-d (B, lanes 2, 3) and Wbp1p-d (C, lanes 2, 3).
Fig. 3. Limited proteolysis of COPI vesicles. (A) Suspension of purified COPI vesicles
containing  0.009 µM coatomer treated with thermolysin (0.009 µM) in the presence of
p23AS-d (50 µM) as a substrate for the protease. Western blot analysis of proteolytic
fragments of -COP reveals a prominent fragment of about 59 kD (lane 2) which is highly
susceptible to the protease (lane 3). (B) Coatomer (0.009 µM) incubated with p23wt-d (50
16
µM) and treated with thermolysin under the same conditions as used for the COPI vesicles
yields a result comparable to that of COPI vesicles (lanes 2, 3). In contrast, partial digestion in
the presence of p23AS-d (50 µM) yields a different result: the 59 kD fragment is more stable
(C, lanes 2, 3).
Fig. 4. Stoichiometry between coatomer and p23wt-d. 125I-labelled p23wt-d (filled squares) or
p23AS-d (filled circle) were incubated with increasing concentrations of coatomer . The
amounts of peptide in the precipitates were determined by counting the
125
I radioactivity. The
error bars indicate standard errors (n=4).
Fig. 5. NOEs indicative of secondary structures. The width of the lines represents the relative
strength of the NOEs; a gray line indicates that the NOE could not be identified because of
spectral overlap. (A) p23wt-m, (B) p23wt-d.
Fig. 6. (A) Superimposition of the 10 structures with lowest internal energies of p23wt-m
obtained from simulated-annealing molecular dynamics using NOE-derived an dihedral
restraints. The heavy atoms of all structures are shown. (B) Ribbons depiction of p23wt-m
showing the secondary structure, diphenylalanine (green) and the dilysine (blue) motif. Top:
view perpendicular to the helix axis; bottom: parallel view. A similar representation from the
structure of p23wt-d dimer, representing the tetramer of the cytoplasmic domain of p23, is
shown in (C).
Fig. 7. Dimer formation of p23wt-d confirmed by size exclusion chromatography. (A)
Monomeric p23wt (p23wt-m: 1 mg/ml, inject 20 µl). (B) Dimeric p23wt (p23wt-d: 1 mg/ml,
inject 20 µl). Peak 1 represents the dimerized form of p23wt-d, peak 2 the monomeric form of
17
p23wt-d. (C) Rechromatography of peak 1 of (B) (40 µl). The appearance of both peak 1 and
2 demonstrates an equilibrium between monomeric and dimeric states of p23wt-d. (D)
Dimeric p23wt (p23wt-d: 0.1 mg/ml, inject 40 µl). The concentration dependence of the
equilibrium of peak 1 and 2 indicates a bimolecular event. Thus, peak 1 is caused by the
dimeric and peak 2 by the monomeric state of the peptide.
Fig. 8. Hypothetical model for bud formation. According to this model, interaction of
coatomer with a tetramer of p23 induces a conformational change of the complex. This leads
to its polymerization on the surface of a membrane resulting in the formation
18
REFERENCES
Acknowledgements
We thank Dr. Suzanne Pfeffer, Dr. Kai Simons, and Dr. Graham Warren, for critically reading
the manuscript. This work was supported by grants of the Deutsche Forschungsgemeinschaft
(SFB 352, to C. H., J.B. H., F. W.), the Human Frontier Science Program (to F. W.) and the
Fonds der Chemischen Industrie (to F. W., P. R.).
19