J. Mol. Biol. (1997) 273, 740±751
Solution Structure of the Spectrin Repeat:
a Left-handed Antiparallel Triple-helical Coiled-coil
Jaime Pascual, Mark Pfuhl, Dirk Walther, Matti Saraste*
and Michael Nilges
European Molecular Biology
Laboratory, Meyerhofstr. 1
69012 Heidelberg, Germany
Cytoskeletal proteins belonging to the spectrin family have an elongated
structure composed of repetitive units. The three-dimensional solution
structure of the 16th repeat from chicken brain a-spectrin (R16) has been
determined by NMR spectroscopy and distance geometry-simulated
annealing calculations. We used a total of 1035 distance restraints, which
included 719 NOE-based values obtained by applying the ambiguous
restraints for iterative assignment (ARIA) method. In addition, we performed a direct re®nement against 1H-chemical shifts. The ®nal ensemble
Ê from the mean for the
of 20 structures shows an average RMSD of 1.52 A
backbone atoms, excluding loops and N and C termini. R16 is made up
of three antiparallel a-helices separated by two loops, and folds into a
left-handed coiled-coil.
The basic unit of spectrin is an antiparallel heterodimer composed of two
homologous chains, b and a. These assemble a tetramer via a mechanism
that relies on the completion of a single repeat by association of the partial repeats located at the C terminus of the b-chain (two helices) and at
the N terminus of the a-chain (one helix). This tetramer is the assemblage
able to cross-link actin ®laments. Model building by homology of the
``tetramerization'' repeat from human erythrocyte spectrin illuminates the
possible role of point mutations which cause hemolytic anemias.
# 1997 Academic Press Limited
*Corresponding author
Keywords: cell elasticity; membrane skeleton; hemolytic anemias;
heteronuclear NMR
Introduction
Spectrin (also called fodrin) is a common component of cytoskeletal structures associated with
the cell membrane in metazoan organisms (Shenk
& Steele, 1993). Electron microscopy studies of
spectrin samples reveal a ¯exible elongated molecule composed of two loosely intertwined strands
Current addresses: M. Pfuhl, Department of
Biochemistry and Molecular Biology, University College
London, WC1E 6BT, UK; D. Walther, Department of
Cellular and Molecular Pharmacology, University of
California, San Francisco, CA 94143-0450, USA.
Abbreviations used: R16, 16th repeat of chicken brain
a-spectrin; 2D and 3D, two and three-dimensional;
NOE, nuclear Overhauser effect; COSY, correlated
spectroscopy; TOCSY, total correlation spectroscopy;
TOWNY, TOCSY without NOESY; HSQC, heteronuclear
single quantum coherence; T1, longitudinal relaxation
time; T2, transverse relaxation time; RMSD, root-meansquare deviation; ARIA, ambiguous restraints for
iterative assignment.
0022±2836/97/430740±12 $25.00/0/mb971344
that appear to be tightly associated at both ends,
where the antiparallel or head-to-tail dimerization
occurs (Shotton et al., 1979). Each strand is made of
two homologous chains (b and a) which associate
into tetramers through a head-to-head interaction
(the C terminus of the b subunit with the N terminus of the a subunit) at the tetramerization site.
The tetramer exposes actin-binding sites at its ends
and is thus able to cross-link membrane-associated
actin ®laments (Bennett & Gilligan, 1993).
Although spectrin is present in most animal tissues, it has been most thoroughly studied in red
cells.
The human erythrocyte is characterized by a distinctive biconcave shape and remarkable elasticity.
These properties are largely determined by the
membrane skeleton, a ¯exible meshwork of proteins on the inner surface of the cell membrane
which is mainly formed by non-covalent interactions between spectrin, F-actin and integral
membrane proteins (Bennett & Gilligan, 1993).
These interactions result in a relatively uniform
# 1997 Academic Press Limited
741
Solution Structure of the Spectrin Repeat
two-dimensional network which is primarily organized in the shape of hexagons. Actin and the
associated proteins are located at the center and on
the six corners of the hexagon and spectrin lies on
the sides. This network is intimately coupled to the
membrane at a limited number of sites via two
molecular contacts between integral membrane
proteins and spectrin. These attachments are provided by ankyrin which interacts with the band 3
protein and spectrin, and by protein 4.1 which
interacts with glycophorin C and spectrin.
A widely held model for red cell membrane elasticity is based on the assumption that spectrin exhibits entropic-spring behavior (Svoboda et al., 1992).
Several other studies indicate that the elastic
deformability of the erythrocyte depends on
enthalpic events (Vertessy & Steck, 1989). According to both models, the membrane elasticity is
mainly due to the properties of spectrin. However,
the rigidity of the erythrocyte membrane also
depends on integral membrane proteins that span
the bilayer, mainly glycophorin C and band 3 and
their oligomerization states (Feng & MacDonald,
1996), indicating that all components of the membrane skeleton contribute to the elasticity of the
red cells.
Most of the spectrin sequence is composed of a
series of contiguous motifs called spectrin repeats
that characterize all members of the spectrin
family, namely spectrin, a-actinin, dystrophin, and
utrophin (Hartwig, 1994). Additional structural
features include an actin-binding domain made of
two calponin homology (CH) domains (Djinovic
et al., 1997), a plekstrin homology (PH) domain
(Macias et al., 1994), a Src homology 3 (SH3)
domain (Musacchio et al., 1992), and a calmodulinlike domain with four EF-hands (Trave et al., 1995).
The ability of the spectrin molecule to contract
and expand has been attributed to the modular
structure made of repeats, initially identi®ed by
partial peptide sequences (Speicher & Marchesi,
1984). The global molecular architecture was later
revealed by complete sequencing of cDNAs for the
a and b subunits from several species. The analysis
of the conformational phase of the repeat using
recombinant fragments of different lengths has
established the boundaries for properly folded
repeats (Winograd et al., 1991). Despite low
sequence homology, theoretical models have proposed that the repeat folds into a left-handed
coiled-coil made of three antiparallel a-helices (A,
B, and C) separated by two short loops (AB and
BC; Parry et al., 1992). However, the crystal structure of the dimeric 13th repeat of Drosophila a-spectrin (PDB ID: 2SPC) has showed an intermolecular
triple helical bundle where helices A and B from
one molecule pack against helix C0 of the other
molecule in the dimer and the proposed BC loop
appears in an a-helical conformation as part of a
continuous helix B-C(Yan et al., 1993). In the intact
spectrin molecule, both termini must point in
opposite directions. The termini in the crystal
structure are oriented in the same direction which
is incompatible with the global arrangement of
repeats inside spectrin. The conformation found in
the crystal is likely to be an artifact caused by the
dimerization process that this particular repeat suffers in solution at room temperature when it is
recombinantly produced as a GST-fusion (Ralston
et al., 1996). From the X-ray data (Yan et al., 1993),
a model for the monomer was proposed that
agrees with the secondary but not with the tertiary
structure proposed by modeling (Parry et al., 1992),
i.e. it proposes the existence of the BC loop but
suggests that the repeat is a helical bundle rather
than a coiled-coil.
The importance of the spectrin repeat to the elastic properties of the membrane skeleton and to the
structural and functional integrity of the normal
red cell is demonstrated by the fact that it is a target for mutations which cause hemolytic anemias,
such as some cases of hereditary elliptocytosis,
pyropoikilocytosis and spherocytosis characterized
by abnormally shaped erythrocytes (Davies & Lux,
1989; Winkelmann & Forget, 1993; Hassoun &
Palek, 1996). Mutations associated with these anemias occur in helical regions of the spectrin
repeats, and frequently disrupt the spectrin tetramer formation (Tse et al., 1990). Impairment of the
tetramer formation alters the membrane skeleton
structure and leads to reduced mechanical stability
of the membrane. Mutations involve nucleotide
substitutions, exon skipping, alternative splice-site
selection, short deletions and frameshifts (Hassoun
et al., 1996).
In earlier studies (MacDonald et al., 1994;
Pascual et al., 1996), we showed that the 16th
repeat from chicken brain a-spectrin (R16;
Wasenius et al., 1989) in solution folds into a monomeric helical structure composed of three helices
separated by two loops at room temperature
(298 K) and at the 1 mM concentration required for
NMR. Recently, it has been demonstrated that the
isolated repeat 1 from human erythrocyte a-spectrin (DeSilva et al., 1997) and repeat 2 from human
dystrophin (Calvert et al., 1996) are also monomeric. Here, we report the complete assignment of
the 15N, 1H and 13C chemical shifts of R16 and
determination of its three-dimensional structure in
solution, providing the ®rst experimental threedimensional structure of a spectrin repeat in native
conformation.
Results
Resonance assignment
The recombinantly expressed R16 domain consists of 110 residues spanning from Ala1763 to
Glu1872 of chicken brain a-spectrin (SwissProt ID:
SPCN_CHICK). Backbone and aromatic side-chain
assignment has already been reported (Pascual
et al., 1996). Con®rmation of the aromatic resonances was accomplished by analyzing a 2D 1H-13C
HSQC spectrum (Stonehouse et al., 1994) where the
13
C carrier frequency was placed at 126 ppm. Due
742
Solution Structure of the Spectrin Repeat
Table 1. Experimental restraints and structural statistics
for the best 20 structures
Experimental restraints
Total number of restraints
NOE restraints
unambiguous
intraresidue
sequential
medium-range
long-range
ambiguous
dihedral angle restraints
hydrogen bond restraints
Deviations from experimental restraints:
RMSD:
Ê)
of unambiguous NOE restraints (A
Ê)
of ambiguous NOE restraints A
of dihedral angle restraints ( )
Ê)
of hydrogen bond restraints (A
of 1H chemical shifts (ppm)
Structural statistics for the ensemble:
Quality indices:
overall G value (PROCHECK)
what-check score (WHATIF)
combined z-score (PROSA)
RMSD from the average structure:
all residues (10-107)
Ê)
backbone (N, Ca, C) (A
Ê)
heavy atoms (A
residues in a-helical conformation
Ê)
backbone (N, Ca, C) (A
Ê)
heavy atoms (A
1035
719
602
157
75
324
46
117
170
146
0.025
0.015
0.287
0.003
0.241
0.003
0.005
0.019
0.001
0.012
0.15
0.47
ÿ5.44
0.03
0.18
0.52
1.83
2.71
0.12
0.19
1.52
2.44
0.15
0.20
to the high a-helical content, the spectral dispersion
of the Ca and Ha chemical shifts was poor, precluding the extension of the assignment to the
side-chains. Taking advantage of the 1H-13C-13C-1H
correlation, we could obtain side-chain assignments, primarily from an analysis of the 3D
HCCH-COSY and TOCSY spectra. We used the
13 a
C and 1Ha chemical shifts as a starting point for
the validation of the previous sequential backbone
assignment. In order to help the assignment, information from the expected 13C chemical shift ranges
and 1H spin system topology was used as an initial
criterium. Further con®rmation was obtained by
checking the resonances in the 3D 13C HSQCTOWNY spectrum. A list containing the 15N, 1H,
and 13C chemical shifts assigned and referenced
according to Wishart et al. (1995) is available as
supplementary material.
Experimental restraints
Table 1 shows the experimental restraints used
to calculate the structure. Due to severe overlap,
we could manually identify only 355 unambiguous
distance restraints between protons. Application of
the ``ambiguous restraints for iterative assignment''
(ARIA) method (Nilges et al., 1997) increased the
number of NOE-based distance restraints. As
described in Materials and Methods, after eight
iterations a total of 719 distance restraints could be
obtained, where 602 were unambiguous and 117
ambiguous. The former contained 157 intraresidue,
75 sequential, 324 medium-range and 46 long-
Figure 1. Superposition of the Ca trace for the ®nal
ensemble of R16 structures (from His10 to Gln107). The
triangles point towards the C terminus of each of the
helices.
range restraints. The elongated shape of the molecule results in a lower number of long-range NOE
contacts in comparison to a globular protein of
similar size where the closer proximity of different
segments allows more interactions. The ambiguous
restraints contained 70 with two, 36 with three,
seven with four, three with ®ve and one with
seven assignment possibilities. Inside this group of
ambiguous restraints, two long-range restraints
were found, the ®rst with two assignment possibilities and the second with three.
Measurement of the 3JHN-Ha coupling constant
for R16 (Pascual et al., 1996) allowed the constraint
of 85 f angles to ÿ60(20) . Since the pattern of
sequential and medium-range NOEs and the consensus chemical shift index for these 85 residues
indicate a helical conformation, their c angles were
743
Solution Structure of the Spectrin Repeat
Figure 2. View of the backbone average structure of R16
(from His10 to Gln107). From the N to the C terminus,
helix A is colored green, the AB loop red, helix B yellow, the BC turn red, and helix C cyan. Some of the
side-chains of residues implicated in interhelical contacts
are shown in white and labeled according to the one
letter code for amino acid residues and in parenthesis
according to the helices nomenclature.
also constrained to ÿ40(20) . In order to de®ne
the helices better, we added 73 hydrogen bonds
canonical of a-helical conformation along the main
chain as two distance restraints (2.50 < dOiÊ
Ê ).
NHi ‡ 4 < 2.70 A
and 2.5 < dOi-Ni ‡ 4 < 3.70 A
These were assigned to the residues in the central
part of the helices. Proline residue 61 (the sequence
is shown in Figure 6) and residues around it were
left unrestrained.
Structure analysis
The termini are highly disordered, as indicated
by their 15N-T1, T2 and 15N-1H heteronuclear NOE
values (Pascual et al., 1996), and therefore a sensible superposition cannot be made with the full
sequence. Consequently, calculation of an average
structure and analysis of the ensemble was carried
out with structures spanning from His10 to
Gln107. Table 1 shows the structural statistics of
the ensemble. The analysis shows that the structures barely violate the empirical restraints and
only slightly deviate from ideal geometry having a
small force ®eld total energy. None of the strucÊ
tures showed NOE violations bigger than 0.5 A
nor dihedral angle violations bigger than 5 . The
RMSD value for the 1H chemical shifts is indicative
of a generally correct assignment. The average
RMSD from the mean structure for the backbone
atoms (N, Ca, and C) of residues in helical conforÊ , and 2.44 A
Ê for all heavy atoms.
mation is 1.52 A
Regarding the quality indices, the overall G value
(Laskowski et al., 1996) should be above ÿ0.5, the
``what-check'' score value (Vriend & Sander, 1993)
for a good structure is expected to be above ÿ0.5
and the combined z-score (Sippl, 1993) varies with
the sequence length and for a 100-residue protein
is expected to be ÿ8.0 2.0. As shown in Table 1,
these criteria are met by the R16 structure.
The ensemble is shown in Figure 1, and the average structure with key residues implicated in interhelical contacts in Figure 2. The Ramachandran
map (Figure 3) for all 20 structures shows that
more than 90% of the residues appear in the
allowed regions. Just a few loop residues in some
of the structures have f/c values in disallowed
regions. A plot showing the distribution of the
number of NOE-restraints per residue and backbone atoms RMSD from the average per residue is
shown in Figure 4a and b. Figure 4c shows the
relative solvent accessibility. This plot reveals that
the most buried residues match with the a and d
positions of the heptad pattern (McLachlan &
Stewart, 1975), shown in Figure 6, characteristic of
coiled-coils (Crick, 1953; Lupas, 1996, 1997). An
improvement in the ®t between the observed and
calculated 1H chemical shifts when comparing
ensembles calculated without and with 1H chemical shift re®nement can be observed in Figure. 5a
and b. The analysis of the ®nal ensemble by PROCHECK-NMR indicates that the overall structural
Ê rescharacteristics are equivalent to a typical 2.5 A
olution crystal structure.
Description of the R16 fold
Analysis of the secondary structure of the average structure using the program DSSP (Kabsch &
Sander, 1983) shows that R16 consists of three
a-helices separated by two loops. The ®rst helix, A,
spans from Gln11 (A6, according to the nomenclature of Yan et al. (1993); see Figure 6) to Ser32
(A27); the second, B, begins with Val42 (B1) and
ends with Asp76 (B35) accommodating the transPro61 (B20); the last one, C, extends from Lys81
(C1) to Gly106 (C26). Helices A and C have seven
turns and are smoothly bent throughout their
744
Solution Structure of the Spectrin Repeat
Figure 3. Ramachandran plot of the f (Phi) and (Psi) torsion angles for all 20 structures in the ensemble generated
using PROCHECK-NMR. Each square represents the f/c values for a residue from the structure of the ensemble
indicated by the number inside the square. Small triangles represent Gly residues. A, B, and L de®ne the most
favored regions; a, b, l, and p de®ne the allowed regions; a, b, l, and p de®ne the generously allowed regions.
Squares that appear on the latter region are topped by the residue name and number. Notice that all of them belong
to loop regions and none appears on the disallowed region systematically.
length. Helix B is the longest with ten turns. It has
a kink (y ˆ 21.8 ; Barlow & Thornton, 1988) which
is caused by Pro61. A loop, from Glu33 to Thr40
connects the ®rst two helices, whereas a tight turn
Ê ) from Asn77 to Gly80
(Cai to Cai ‡ 3 distanceˆ 7.0 A
makes the connection between helices B and C.
Most of the contacts at the interface between the
helices are made by hydrophobic side-chains,
which are highly conserved in the spectrin repeats
and occupy the positions a and d of the heptad pattern of coiled-coils. However, ionic interactions
between charged side-chains of variable residues
mainly at g and e positions also appear to stabilize
the structure. A total of 46 interhelical NOEs de®ne
several contact regions and allow the mapping of
the interior of the structure (Figure 2). At the top,
there are interactions between the side-chains of
Phe12 (A7), the variable Gly70 (B29) and Ile84
(C4); below, contacts appear between Met16 (A11),
Val66 (B25) and Phe91 (C11). In the middle, the
almost invariant Trp22 (A17) in the g position
interacts with the variable His59 (B18) and with
Trp95 (C15). Near the bottom, contacts involving
Val30 (A25), Leu52 (B11) and Ala102 (C22) can be
observed. Finally, an electrostatic rather than a
hydrophobic interaction occurs at the bottom
Solution Structure of the Spectrin Repeat
745
Figure 5. Plots of the difference between the average
observed and calculated 1H chemical shifts versus the
sequence number (from His10 to Gln107). In a, an
ensemble of ten structures coming from the last ARIA
iterative calculation without 1H chemical shift re®nement was used. In b, the ®nal 20 structures re®ned in
vacuum and against 1H chemical shifts were the input
for the calculation.
Figure 4. Plot showing the distribution of the number of
NOE restraints (a), backbone atoms RMSD from the
average (b) and the relative solvent accessible area (c)
per R16 residue (from His10 to Gln107).
between the variable Glu33 (A28), the conserved
Asp34 (A29) and two highly conserved positively
charged residues, namely Lys48 (B7) and Arg105
(C25) both in g positions.
Analysis of the tertiary structure of the average
structure of R16 was performed using the program
HELIX (Walther et al., 1996). Table 2 shows the
geometrical and packing parameters measured for
each pair of helices. Notice that the helices A/B
and B/C are antiparallel while helices A/C are
parallel. The column gives the crossing angle
value for each pair of helices. Depending on the
orientation of the helices in the pair (parallel or
antiparallel), the crossing angle varies by 180 .
Regarding the helical packing, the ratio column
presents the percentage of residues in close contact
(all column) that pack in ``knobs into holes'' manner (153 column). Measurement of these parameters shows that R16 folds into a left-handed
antiparallel triple helical coiled-coil with crossing
angles for the parallel (A/C) and antiparallel (A/B
and B/C) helices of 30.8 , ÿ144.7 and ÿ162.3 ,
Ê . On
respectively, and an average pitch of 178.6 A
average, 52% of the residues implied in R16 interhelical contacts show the classical ``knobs into
holes'' type of packing characteristic of coiled-coils.
A comparison of the same parameters with the
crystal structure of the Drosophila repeat (2SPC)
and with coil-Ser (Lovejoy et al., 1993; PDB ID:
1COS), a synthetic trimer composed of three
a-helices folded into an antiparallel coiled-coil, is
also shown in Table 2. Analysis of the R16 fold
against the PDB using the program DALI (Holm &
746
Solution Structure of the Spectrin Repeat
Figure 6. Alignment of spectrin repeats. The numbering corresponds to the fragment used in the solution structure.
The sequences aligned are the 16th repeat of chicken brain a-spectrin (a16) solved by NMR, the 13th repeat of Drosophila a-spectrin (a13) solved by X-ray, and the ``tetramerization'' repeat made of the partial repeats of human erythrocyte b/a-spectrin (b17/a0). The consensus line was obtained from a complete alignment of all repeats from human
erythrocyte spectrin. The heptad pattern as well as the helices nomenclature are shown. The multiple sequence alignment was made using the program CLUSTALW (Thompson et al., 1994).
Sander, 1993) reported as the highest score
(z ˆ 6.1) the match with the avian farnesyl diphosphate synthase (PDB ID: 1FPS). Both structures
Ê for 70% of the
superimpose with a RMSD of 2.8 A
a
C atoms of the repeat including fragments from
all three helices.
Discussion
The spectrin repeat folds into a
triple-helical coiled-coil
As previously reported (Pascual et al., 1996), R16
folds into a monomeric triple-helical structure. The
fold of R16 is composed of three long helices (A, B,
C) separated by a loop (AB) and a turn (BC). As
depicted in Figures 1 and 4b, both the AB loop and
the BC turn correspond to the more disordered
areas. However, the close correspondence between
the positional disorder and their mobile behavior
observed via the relaxation measurements suggests
that the disorder is not due to the scarcity of
restraints for those residues but re¯ects real mobility in solution. Moreover, the AB loop shows a
higher degree of disorder than the BC turn. This is
in agreement with the relaxation parameters of the
repeat where a shorter value for 15N-T1, longer for
T2 and a smaller 15N-1H heteronuclear NOE are
observed for the AB loop residues compared to
those of the BC turn (Pascual et al., 1996). Regarding the helices, a high proportion of the residues
implicated in interhelical contacts pack their sidechains following the well-de®ned ``knobs into
holes'' arrangement characteristic of a coiled-coil.
The crossing angle and the pitch values corrobo-
rate the coiling of the helices (Table 2). The transPro61 (f ÿ66.4, c ÿ29.6) is accommodated in the
middle of helix B by a kink. Other repeats also
show a proline residue at the same helical position
(B20; Pascual et al., 1997), indicating that a single
proline residue can be present inside a helix without preventing its folding (Barlow & Thornton,
1988).
Superposition of the backbone atoms of the helical residues between the average solution structure
(R16) and the monomeric model from the dimeric
Ê.
X-ray structure (2SPC) shows a RMSD of 2.1 A
These repeats have diverged early in evolution
(Pascual et al., 1997) and have only 22% identical
residues at equivalent positions (Figure 6). Of these
23 identical residues, 13 are highly conserved
among all spectrin repeats. Most of them are
hydrophobic residues that occupy the a and d heptad positions (A4, B1, B4, B8, B22, B32, C15, C18,
and C29) and show low dispersion of their w1
angle in the ensemble of NMR structures. However, these side-chains do not superimpose well
with those from the crystal model, showing some
of them to have distinct gauche/trans w1 rotamers.
Moreover, the aromatic cluster made by A17/B18/
C15 shows a different disposition of the rings
(Figure 7).
Contrasting their tertiary structure (Table 2),
different values are obtained for the crossing angle,
pitch and ratio of residues packed according to the
``knobs into holes'' arrangement. For comparison,
Table 2 includes the corresponding values for a
synthetic antiparallel triple-helical coiled-coil
(1COS) with almost identical helical length. Notice
that 1COS should provide the ideal values for an
Table 2. Helical geometry and packing parameters of R16, 2SPC, and 1COS
Molecule
R16
2SPC
1COS
a
Helix 1
A
A
B
A
A
B
A
A
B
Helices
Helix 2
B
C
C
B
C0 a
C0 a
B
C
C
( )
Geometry
Ê)
Pitch (A
ÿ144.7
30.8
ÿ162.3
ÿ157.0
12.5
ÿ160.8
ÿ157.7
32.7
ÿ159.3
This helix comes from the second molecule in the crystal dimer.
198.4
157.2
180.2
251.2
206.1
308.8
197.2
167.6
192.8
153
5
7
7
9
12
8
14
17
15
Packing
All
10
12
15
23
26
26
25
24
19
Ratio (%)
50
58
47
39
46
31
56
71
79
747
Solution Structure of the Spectrin Repeat
dues with opposite charge are not observed in the
sequence alignment of the repeats.
Molecular explanation for hemolytic anemias
Figure 7. Superposition between the X-ray monomeric
model (depicted in red) and the average solution structure (depicted in green) of the conserved hydrophobic
side-chains A17, B18, and C15 in the core of the repeat.
These side-chains show a low dispersion of their w1
angles in the ensemble of solution structures.
antiparallel triple-helical coiled-coil since it only
contains leucine residues at the a and d positions.
Since R16 shows a higher crossing angle, lower
pitch and higher ratio of residues packed in
``knobs into holes'' as compared to 2SPC, we conclude that the repeat folds into a coiled-coil in solution whereas the crystal structure of the repeat is
better described as a bundle. These discrepancies
could re¯ect the possibility that the crystal structure resembles more the ``tetramerization'' repeat
(a very special repeat constituted by intermolecular
interactions of the C-terminal helices A and B of
b-spectrin and the N-terminal helix C of a-spectrin)
rather than a typical internal repeat exempli®ed by
the NMR structure.
Antiparallel interactions between repeats
within the dimer
Since the functional unit of spectrin is a dimer
and that of dystrophin is a monomer, differences
in the pattern of conservation of residues between
repeats in both proteins could explain that differential property (Winder et al., 1995). There are only
two highly conserved and exposed (either in b, c or
f positions) residues within the helices of the spectrin repeats that are not conserved in the dystrophin repeats. These are Asp18 (A13) and Glu25
(A20) which are both in consecutive c heptad positions of the helix A (Figure 6). Conservation of a
residue or a physical property at a speci®c position
or area could indicate an important structural
and/or functional role. In this case, however, a
possible involvement of these charges in functions
like repeat-repeat interactions in the antiparallel
dimer is unclear since conserved and exposed resi-
Several point mutations in the repeats from
human erythrocyte spectrin have been related to a
subset of hemolytic anemias such as hereditary
elliptocytosis, pyropoikilocytosis and spherocytosis. Most of those mutations have been mapped
inside the ``tetramerization'' repeat. We have built
a model by homology of the tetramerization repeat
using the R16 structure as a template and employing the WHATIF program (Vriend, 1990) to interpret the puzzling fact that a single point mutation
leads to the disease.
This repeat is built by interactions between two
molecules, and it is therefore likely to be very sensitive to mutations that destabilize the interhelical
packing. One of the mutations appears in position
A17 where the highly conserved Trp is mutated to
Arg (Parquet et al., 1994). This conserved Trp has
been described as one of the key residues for the
folding of the repeat where a conservative
mutation to Phe already causes a signi®cant
decrease in its global stability (Pantazatos &
MacDonald, 1997). Apart from this, the model indicates that an Arg residue in this position would
cause a charge repulsion with the Arg in position
B21. This could prevent the proper folding of the
repeat. An interesting mutation is caused by the
replacement of Ala with Pro in position B10 (Tse
et al., 1990). The presence of a single Pro residue is
not rare in helix B of the spectrin repeats, especially
in position B20. In fact, the tetramerization repeat
contains a Pro residue in position B28. The presence of a second Pro in the same helix most likely
causes another kink that disrupts the proper packing. An additional peculiar feature of this repeat is
the presence of ®ve Arg residues in the helix C for
which mutations have been found (C7, C8, C14,
C21, and C25; Palek & Sahr, 1992; Winkelmann &
Forget, 1993). For instance, the substitution of Arg
with Leu in position C8 causes the loss of an interhelical salt bridge with the Glu at B26 according to
our model. The same is applicable to the Arg to
Ser mutation at C25 which may lead to the breakage of another highly conserved interhelical ionic
interaction with the Asp at A29 (Figure 8).
Elastic properties of spectrin
It has been proposed (Bloch & Pumplin, 1992)
that the residues connecting two sequential repeats
could act as a hinge region being largely responsible for the elastic properties of spectrin. However,
this region is rather short, in general comprising
only two residues which are rarely Gly or Pro.
Thus, it is possible that there is a continuous C-A
helix between the repeats, although the heptad pattern is lost between helix C of one repeat and helix
A of the next one. The likely residues implicated in
inter-repeat contacts are conserved, hydrophobic
748
Solution Structure of the Spectrin Repeat
Figure 8. Model of the arrangement
of the conserved side-chains A29,
B7, and C25 of the ``tetramerization'' repeat colored in green interacting via an intermolecular salt
bridge as well as a model for the
effect of the Arg to Ser (colored in
red) mutation in C25 that prevents
the tetramerization. The inset the
backbone of the model for the tetramerization repeat; helices A and
B from the b-chain are in white
and helix C from the a-chain is in
cyan. Notice the absence of the BC
loop. Each triangle points towards
the C terminus of its respective
helix.
and exposed like A2 (His7 in R16) that occupies
the heptad postition f, the ®rst (Tyr35) and the ®fth
(Leu39) residue in the AB loop and the ®fth
(Thr78) residue in the BC turn (Figure 6). Unfortunately, these residues show the highest disorder
both in the crystal and solution structures of single
repeats. Consequently, further insights into the
molecular mechanism of spectrin elasticity will
become available when a re®ned structure of a
double repeat has been determined, and the structure of the proposed hinge region and the interrepeat contacts emerge.
Materials and methods
NMR spectroscopy
The expression, puri®cation and 15N/13C labeling of
R16 was carried out according to Pascual et al. (1996).
The NMR sample contained 1 mM of uniformly
15
N/13C-labeled R16 dissolved in 90% H2O/10% 2H2O,
10 mM potassium phosphate (pH 6.0). The NMR spectra
were recorded at 308 K using Bruker DMX 500/600
MHz spectrometers. Water suppression was accomplished using WATERGATE (Piotto et al., 1992) or coherence selection by pulsed ®eld gradients combined with
sensitivity enhancement (Kay et al., 1992). The experiments acquired for the backbone and aromatic side
chains assignment have already been described (Pascual
et al., 1996). In order to assign the side-chains 3D 13C-edited HCCH-COSY (Ikura et al., 1991) and HCCH-TOCSY
(Bax et al., 1990) spectra were obtained, using 20 ms as
TOCSY mixing time, as well as a 3D 13C HSQC-TOWNY
modi®ed from a 13C HSQC-NOESY according to
Majumdar & Zuiderweg (1993) by introducing the
TOWNY scheme (Kadkhodaei et al., 1993) with a mixing
time of 60 ms. For structure calculation purpose, two 3D
NOESY-type spectra, namely 15N NOESY-HSQC
(Sklenar et al., 1993) and 13C HSQC-NOESY (Majumdar
& Zuiderweg, 1993), were acquired with mixing times of
70 ms. All 3D experiments were recorded in the phasesensitive mode using the States-TPPI method (Marion
et al., 1989). The NMR spectra were processed and analyzed on Silicon Graphics workstations using the programs AZARA (Boucher, 1994) and AURELIA (Neidig
et al., 1995). Linear prediction was applied to the
indirectly detected dimensions. The residual water signal
in some of the 3D spectra was suppressed by means of
the Karhunen-Loeve transformation (Mitschang et al.,
1991). The ®nal size of the 3D data matrices was
1024 512 256 real points.
Structure calculation
Structure re®nement was achieved employing the
ambiguous restraints for iterative assignment (ARIA)
methodology (Nilges et al., 1997). ARIA performs an
automatic assignment of NOESY spectra peaks based on
the chemical shifts list and its compatibility with the
interproton distances measured in the initial structures.
Merging of the manually assigned NOE restraints with
the automatically assigned NOE restraints is done in
order to calculate better de®ned structures that will be
used in the next iteration. The structure re®nement is
obtained by molecular dynamics using version 3.851 of
X-PLOR (BruÈnger, 1992) adapted to interpret ambiguous
distance restraints (Nilges, 1995) and applying speci®c
simulated annealing protocols as described by Nilges
et al. (1997). Assignment of distance restraints implying
methylene and isopropyl group resonances is done using
the ¯oating chirality approach (Weber et al., 1988;
Folmer et al., 1997).
In our case, three data sets were used. The ®rst one
consisted of 355 NOE-based unambiguous distance
restraints derived from manual assignment of different
NOESY spectra. The second and the third data sets were
obtained from the 3D 15N NOESY-HSQC and 13C
HSQC-NOESY spectra using the automatic peak-picking
routine of AURELIA. A manual cleaning for obvious
artifacts and solvent residual signal of the peak-picked
lists was performed. Automatic assignment of the picked
peaks was obtained using a frequency window tolerance
around their chemical shift values of 0.02 ppm in the
1
H acquisition dimension, 0.1 ppm in the 1H indirect
dimension and 0.2 ppm in the heteronuclear one. Conversion of peak volumes into distance restraints was
accomplished through the calibration procedure
described by Nilges et al. (1997). The initial structures
used for the ARIA re®nement were the seven lower in
energy from 500 structures calculated employing the distance geometry program DIANA (Guntert et al., 1991)
749
Solution Structure of the Spectrin Repeat
applying the REDAC strategy (Guntert & Wuthrich,
1991). They were obtained using 355 NOE-based unambiguous distance restraints manually assigned including
32 long-range NOEs; cross-peak intensities were classi®ed as strong, medium or weak using contour levels for
calibration, and the upper limits for the distance
Ê , 4.0 A
Ê and 5.0 A
Ê , respectively.
restraints were set to 3.0 A
Eight ARIA iterations were run modifying the assignment parameter (p) from 0.999 to 0.8, the violation
Ê to 0.0 A
Ê and the violation ratio
threshold (Tv) from 2.0 A
(Nv) from 0.5 to 0.75. In each iteration ten structures
were re®ned although just the seven structures with
lower energy were used for assignment purpose. At the
®nal iteration, a total of 719 distance restraints based on
NOEs were obtained, where 602 were unambiguous and
117 ambiguous.
Further re®nement to detect ionic interactions on the
surface of the molecule was accomplished by the simulation of non-bonded interactions in vacuum using all
719 NOE-derived distance restraints and an X-PLOR
energy function that included an electrostatic potential
term and a 1H-chemical shift pseudopotential
(Kuszewski et al., 1995). Starting with the best 40 structures from the initial distance geometry calculation, 40
new structures were calculated in vacuum. The 20 structures with lower energy were selected for structure analysis. All 20 structures were superimposed to wellde®ned regions (from His10 to Gln107) and an average
structure spanning those residues was calculated. A
quality analysis of the ®nal superimposed ensemble was
assessed using the programs PROCHECK-NMR
(Laskowski et al., 1996), WHATIF (Vriend & Sander,
1993) and PROSA (Sippl, 1993). The coordinates of the
ensemble of 20 structures (PDB ID: 1AJ3) plus a list with
the experimental NMR restraints (PDB ID: R1AJ3MR)
are available at the Brookhaven Protein Data Bank.
Helical geometry and packing analysis
The de®nitions and sign conventions of helical geometry and packing parameters were adapted from those of
Walther et al. (1996). The measurements were obtained
using the program HELIX (Walther et al., 1996) and performed for each pair of helices. Parameters studied
included the dihedral packing (or helix-crossing) angle
(
) associated with the line of closest approach between
the pair of helices; the pitch or helical supercoiling, estimated by assigning local lines of closest approach to
both ends of a given helix-helix interface and relating the
dihedral angle associated with the vector connecting the
middle points of the local lines to the length of the vector; and the automated identi®cation of the so-called
``knobs into holes'' (cell 153, according to Walther et al.
(1996)) type of helical side-chain packing distinctive of
coiled-coils (Crick, 1953).
Acknowledgments
We thank Y. Prigeant and D. Davoust for providing
measuring time at the NMR laboratory of the Universite
de Haute Normandie, Rouen (France). We also thank
Francisco Blanco for his constant help and support
during the project and constructive criticism of the
manuscript.
References
Barlow, D. J. & Thornton, J. M. (1988). Helix geometry
in proteins. J. Mol. Biol. 201, 601± 619.
Bax, A., Clore, M., Driscoll, P., Gronenborn, A. & Ikura,
M. (1990). Practical aspects of proton-carbon-proton
three-dimensional correlation spectroscopy of
13
C-labeled proteins. J. Magn. Reson. 87, 620± 628.
Bennett, V. & Gilligan, D. M. (1993). The spectrin-based
membrane skeleton and micron-scale organization
of the plasma membrane. Annu. Rev. Cell Biol. 9,
27±66.
Bloch, R. & Pumplin, D. (1992). A model of spectrin as a
concertina in the erythrocyte membrane skeleton.
Trends Cell Biol. 2, 186± 189.
Boucher, W. (1994). AZARA v1.0, Department of Biochemistry, University of Cambridge, UK.
BruÈnger, A. (1992). X-PLOR 3.1. A System for X-ray Crystallography and NMR, Yale University Press, New
Haven, CT.
Calvert, R., Kahana, E. & Gratzer, W. (1996). Stability of
the dystrophin rod domain fold: evidence for
nested repeating units. Biophys. J. 71, 1605± 1610.
Crick, F. (1953). The packing of a-helices: simple coiledcoils. Acta Crystallog. 6, 689± 697.
Davies, K. & Lux, S. (1989). Hereditary disorders of the
red cell membrane skeleton. Trends Genet. 5, 222 ±
227.
DeSilva, T., Harper, S., Kotula, L., Hensley, P., Curtis,
P., Otvos, L. & Speicher, D. (1997). Physical properties of a single-motif erythrocyte spectrin peptide: a
highly stable independently folding unit. Biochemistry, 36, 3991± 3997.
Djinovic, K., Banuelos, S. & Saraste, M. (1997). Crystal
structure of a calponin homology domain. Nature
Struct. Biol. 3, 175± 179.
Feng, S. & MacDonald, R. C. (1996). A tethered adhesive
particle model of two-dimensional elasticity and its
application to the erythrocyte membrane. Biophys. J.
70, 857± 867.
Folmer, R., Hilbers, C., Konings, R. & Nilges, M. (1997).
Floating stereospeci®c assignment revisited: application to an 18 kDa protein and comparison with J
coupling data. J. Biomol. NMR, 9, 245± 258.
Guntert, P. & Wuthrich, K. (1991). Improved ef®ciency
of protein structure calculations from NMR data
using the program DIANA with redundant dihedral angle constraints. J. Biomol. NMR, 1, 447± 456.
Guntert, P., Braun, W. & Wuthrich, K. (1991). Ef®cient
computation of three-dimensional protein structures
in solution from NMR data using the program
DIANA and the supporting programs CALIBA,
HABAS and GLOMSA. J. Mol. Biol. 217, 517± 530.
Hartwig., J. H. (1994). Actin binding proteins 1: spectrin
superfamily. Protein Pro®le, 1, 711± 762.
Hassoun, H. & Palek, J. (1996). Hereditary spherocytosis:
a review of the clinical and molecular aspects of the
disease. Blood Rev. 10, 129±147.
Hassoun, H., Vassiliadis, J. N., Murray, J., Yi, S. J.,
Hanspal, M., Johnson, C. A. & Palek, J. (1996). Hereditary spherocytosis with spectrin de®ciency due
to an unstable truncated beta spectrin. Blood, 87,
2538± 2545.
Holm, L. & Sander, C. (1993). Protein structure comparison by alignment of distance matrices. J. Mol. Biol.
233, 123± 138.
Ikura, M., Kay, L. & Bax, A. (1991). Improved threedimensional 1H-13C-1H correlation spectroscopy of a
750
Solution Structure of the Spectrin Repeat
13
C labeled protein using constant time evolution.
J. Biomol. NMR, 1, 299 ± 304.
Kabsch, W. & Sander, C. (1983). Dictionary of protein
secondary structure: pattern recognition of hydrogen-bonded and geometrical features. Biopolymers,
22, 2577± 2637.
Kadkhodaei, M., Hwang, T. L., Tang, J. & Shaka, A. J.
(1993). A simple windowless mixing sequence to
suppress cross relaxation in TOCSY experiments.
J. Magn. Reson. ser. A, 105, 104 ± 107.
Kay, L., Keifer, P. & Saarinen, T. (1992). Pure absorption
gradient enhanced heteronuclear single quantum
correlation spectroscopy with improved sensitivity.
J. Am. Chem. Soc. 114, 10663± 10665.
Kuszewski, J., Gronenborn, A. & Clore, M. (1995). The
impact of direct re®nement against proton chemical
shifts on protein structure determination by NMR.
J. Magn. Reson. ser. B, 107, 293± 297.
Laskowski, R., Rullman, J., MacArthur, M., Kaptein, R. &
Thornton, J. (1996). AQUA and PROCHECK-NMR:
programs for checking the quality of protein structures solved by NMR. J. Biomol. NMR, 8, 477 ± 486.
Lovejoy, B., Choe, S., Cascio, D., McRorie, D., DeGrado,
W. & Eisenberg, D. (1993). Crystal structure of a
synthetic triple-stranded a-helical bundle. Science,
259, 1288± 1293.
Lupas, A. (1996). Coiled coils: new structures and new
functions. Trends Biochem. Sci. 21, 375 ± 382.
Lupas, A. (1997). Predicting coiled-coil regions in
proteins. Curr. Opin. Struct. Biol. 7, 388± 393.
MacDonald, R. I., Musacchio, A., Holmgren, R. A. &
Saraste, M. (1994). Invariant tryptophan at a
shielded site promotes folding of the conformational
unit of spectrin. Proc. Natl Acad. Sci. USA, 91, 1299±
1303.
Macias, M. J., Musacchio, A., Ponstingl, H., Nilges, M.,
Saraste, M. & Oschkinat, H. (1994). Structure of the
pleckstrin homology domain from beta-spectrin.
Nature, 369, 675± 677.
Majumdar, A. & Zuiderweg, E. P. (1993). Improved
13
C-resolved HSQC-NOESY spectra in H2O, using
pulsed ®eld gradients. J. Magn. Reson. ser. B, 102,
242 ± 244.
Marion, D., Ikura, M., Tschudin, R. & Bax, A. (1989).
Rapid recording of 2D NMR spectra without phase
cycling. Application to the study of hydrogen
exchange in proteins. J. Magn. Reson. 85, 393 ± 397.
McLachlan, A. & Stewart, M. (1975). Tropomyosin
coiled-coil interactions: evidence for an unstaggered
structure. J. Mol. Biol. 98, 293 ± 304.
Mitschang, L., Cieslar, C., Holak, T. & Oschkinat, H.
(1991). Application of the Karhunen-Loeve transformation to the suppression of undesired resonances in three-dimensional NMR. J. Magn. Reson.
92, 208 ± 217.
Musacchio, A., Noble, M., Pauptit, R., Wirenga, R. &
Saraste, M. (1992). Crystal structure of a Src-homology 3 (SH3) domain. Nature, 359, 851± 854.
Neidig, K.-P., Geyer, M., Gorler, A., Antz, C., Saffrich,
R., Beneicke, W. & Kalbitzer, H. R. (1995). AURELIA, a program for computer-aided analysis of multidimensional NMR spectra. J. Biomol. NMR, 6, 255±
270.
Nilges, M. (1995). Calculation of protein structures with
ambiguous distance restraints. Automated assignment of ambiguous NOE crosspeaks and disulphide
connectivities. J. Mol. Biol. 245, 645±660.
Nilges, M., Macias, M., O'Donoghue, S. & Oschkinat, H.
(1997). Automated NOESY interpretation with
ambiguous distance restraints: the re®ned NMR solution structure of the pleckstrin homology domain
from b-spectrin. J. Mol. Biol. 269, 408± 422.
Palek, J. & Sahr, E. (1992). Mutations of the red blood
cell membrane proteins: from clinical evaluation to
detection of the underlying genetic defect. Blood, 80,
308± 330.
Pantazatos, D. & MacDonald, R. (1997). Site-directed
mutagenesis of either the highly conserved W22 or
the moderately conserved W95 to a large, hydrophobic residue reduces the thermodynamic stability
of a spectrin repeating unit. J. Biol. Chem. 272,
21052± 21059.
Parquet, N., Devaux, I., Boulanger, L., Galand, C.,
Boivin, P., Lecomte, M., Dhermy, D. & Garbarz, M.
(1994). Identi®cation of three novel spectrin aI/74
mutations in hereditary elliptocytosis: further support for a triple-stranded folding unit model of the
spectrin heterodimer contact site. Blood, 84, 303 ±
308.
Parry, D. A., Dixon, T. W. & Cohen, C. (1992). Analysis
of the three-alpha-helix motif in the spectrin superfamily of proteins. Biophys. J. 61, 858 ± 867.
Pascual, J., Pfuhl, M., Rivas, G., Pastore, A. & Saraste,
M. (1996). The spectrin repeat folds into a threehelix bundle in solution. FEBS Letters, 383, 201± 207.
Pascual, J., Castresana, J. & Saraste, M. (1997). Evolution
of the spectrin repeat. BioEssays, 19, 811 ±817.
Piotto, M., Saudek, V. & Sklenar, V. (1992). Gradient-tailored excitation for single-quantum NMR spectroscopy of aqueous solutions. J. Biomol. NMR, 2,
661± 665.
Ralston, G., Cronin, T. & Branton, D. (1996). Self-association of spectrin's repeating segments. Biochemistry,
35, 5257± 5263.
Shenk, M. & Steele, R. (1993). A molecular snapshot of
the metazoan `Eve'. Trends Biochem. Sci. 18, 459 ±
463.
Shotton, D., Burke, B. & Branton, D. (1979). The molecular structure of human erythrocyte spectrin. J. Mol.
Biol. 131, 303± 329.
Sippl, M. (1993). Recognition of errors in three-dimensional structures of proteins. Proteins: Struct. Funct.
Genet. 17, 355± 362.
Sklenar, V., Piotto, M., Leppik, R. & Saudek, V. (1993).
Gradient-tailored water suppression for 1H-15N
HSQC experiments optimized to retain full
sensitivity. J. Magn. Reson. ser.A, 102, 241± 245.
Speicher, D. W. & Marchesi, V. T. (1984). Erythrocyte
spectrin is comprised of many homologous triple
helical segments. Nature, 311, 177± 180.
Stonehouse, J., Shaw, G. L., Keeler, J. & Laue, E. D.
(1994). Minimizing sensitivity losses in gradientselected 15N-1H HSQC spectra of proteins. J. Magn.
Reson. ser.A, 107, 178± 184.
Svoboda, K., Schmidt, C. F., Branton, D. & Block, S. M.
(1992). Conformation and elasticity of isolated red
blood cell membrane skeleton. Biophys. J. 63, 784 ±
793.
Thompson, J. D., Higgins, D. G. & Gibson, T. J. (1994).
CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence
weighting, position-speci®c gap penalties and
weight matrix choice. Nucl. Acids Res. 22, 4673±
4680.
Trave, G., Lacombe, P. J., Pfuhl, M., Saraste, M. &
Pastore, A. (1995). Molecular mechanism of the calcium-induced conformational change in the spectrin
EF-hands. EMBO J. 14, 4922± 4931.
751
Solution Structure of the Spectrin Repeat
Tse, W. T., Lecomte, M. C., Costa, F. F., Garbarz, M.,
Feo, C., Boivin, P., Dhermy, D. & Forget, B. G.
(1990). Point mutation in the b-spectrin gene associated with aI/74 hereditary elliptocytosis. Implications for the mechanism of spectrin dimer selfassociation. J. Clin. Invest. 86, 909 ± 916.
Vertessy, B. & Steck, T. (1989). Elasticity of the human
red cell membrane skeleton. Biophys. J. 55, 255 ±262.
Vriend, G. (1990). A molecular modeling and drug
design program. J. Mol. Graphics, 8, 52 ± 56.
Vriend, G. & Sander, C. (1993). Quality control of protein models: directional atomic contact analysis.
J. Appl. Crystallog. 26, 47± 60.
Walther, D., Eisenhaber, F. & Argos, P. (1996). Principles
of helix-helix packing in proteins: the helical lattice
superposition model. J. Mol. Biol. 255, 536± 553.
Wasenius, V. M., Saraste, M., Salven, P., Eramaa, M.,
Holm, L. & Lehto, V. P. (1989). Primary structure of
the brain alpha-spectrin (published erratum appears
in J. Cell Biol. Mar 1989; 108(3): following 1175).
J. Cell Biol. 108, 79± 93.
Weber, P., Morrison, A. & Hare, D. (1988). Determining
stereo-speci®c 1H nuclear magnetic resonance
assignments from distance geometry calculations.
J. Mol. Biol. 204, 483 ±487.
Winder, S., Gibson, T. & Kendrick-Jones, J. (1995). Dystrophin and utrophin: the missing links!. FEBS
Letters, 369, 27±33.
Winkelmann, J. & Forget, B. (1993). Erythroid and nonerythroid spectrins. Blood, 81, 3173± 3185.
Winograd, E., Hume, D. & Branton, D. (1991). Phasing
the conformational unit of spectrin. Proc. Natl Acad.
Sci. USA, 88, 10788± 10791.
Wishart, D. S., Bigam, C. G., Yao, J., Abildgaard, F.,
Dyson, J., Old®eld, E., Markley, J. & Sykes, B. D.
(1995). 1H, 13C and 15N chemical shift referencing in
biomolecular NMR. J. Biomol. NMR, 6, 135 ± 140.
Yan, Y., Winograd, E., Viel, A., Cronin, T., Harrison,
S. C. & Branton, D. (1993). Crystal structure of the
repetitive segments of spectrin. Science, 262, 2027±
2030.
Edited by P. E. Wright
(Received 12 June 1997; received in revised form
4 August 1997; accepted 5 August 1997)
http://www.hbuk.co.uk/jmb
Supplementary material comprising one Table is
available from JMB Online.