CardiotoxinV4II fromNaja mossambica mossambica

J. MOE. Biol. (1999) 214, 281-297

Cardiotoxin

Vf from Naja mossambicci mossambica The Refined Crystal Structure

B. Rees, A. Bilwes, J. P. Samama and D. Moras Institut

de Biologic Mole’culaire et Cellulaire du Centre National Recherche ScientiIque, 15 rue Rene’ Descartes F-67084 Strasbourg Cedex, France (Received

5 January

1990; accepted 13 March

de la

1990)

The crystal structure of cardiotoxin Vi’ from Naja mossambica mossawrbicawas refined to 25 A resolution. Fifty ordered solvent sites were localized and included in the refinement. The final R factor is 0.197 (n/(2sin@ <5 A; F>3o). The three-dimensional structure is characterized by two fl-sheets. Of particular interest is the two-stranded b-sheet in the N-terminal region. This shows a large right-handed twist and, though strongly connected to the core of the molecule, and in particular to the C-terminal end, protrudes out of the bulk of the molecule. The segment of four amino acid residues connecting the two strands of this sheet is particularly exposed. It contains an invariant proline residue that has probably an important structural role, and is completely hydrophobic. Two other conserved hydrophobic zones were identified; the largest extends over the second and third loops, on one side only of the molecule. All side-chains of invariant hydrophobic character (except proline residues) belong to one of these three zones. Also discussedare the dimeric assembly and the rather loose packing in the crystal. The three-dimensional structure is compared with that of short and long a-neurotoxins. Comparison with two-dimensional nuclear magnetic resonance results on the 68% homologous cardiotoxin CTXIIb shows an excellent overall agreement. A few differences are probably genuine.

1. Introduction

1.4 A (1 A=61 nm) (Tsernoglou & Petsko, 1977; Walkinshaw et al., 1980; Love & Stroud, 1986; Smith et al., 1988). The only X-ray structure of a cardiotoxin published so far is that of Vy from Naja mossambicu mossambica, determined at an intermediate resolution, 3 A (Rees et al., 1987). This lack of structural information is probably due to the difficulty usually encountered in crystallizing the cardiotoxins. A better knowledge of the threedimensional structure should be important in the understanding of the still obscure molecular mechanisms involved. Solution studies may provide valuable information, sometimes complementary to the crystal structure, even if the level of accuracy is lower. The most detailed results are given by two-dimensional n.m.r.t A complete n.m.r. study of the cardiotoxin CTXIIb from N. m. mossambicahas been published

Cardiotoxins are one of the components of the venom of certain species of snakes (cobras), where they coexist with other lethal components, mainly a-neurotoxins (for reviews, see Harvey, 1985; Dufton & Hider, 1988). Cardiotoxins and neurotoxins are closely related in their structure but act in a different way. The target of neurotoxins is well identified as the nicotinic acetylcholine receptor at the post-synaptic level of the neuro-muscular junction and, even if a detailed mechanism of action is not known, it is clear that the function of neurotoxins is to block the receptor and thus prevent binding of acetylcholine. By contrast, very little is known about the mechanisms of action of cardiotoxins, or even about their targets, except that they are localized on muscular and nervous cell membranes, which cardiotoxins depolarize. Also, the part of the molecule interacting directly with the membrane, the “toxic site”, is still controversial. 7 Abbreviations used:n.m.r., nuclear magnetic The information available on the molecular strucresonance;n.c., non-crystallographic; r.m.s., root-meantures parallels the biochemical knowledge. The square;m.l.r., multiple isomorphousreplacement; crystal structures of several neurotoxins have been s.i.r., single isomorphous replacement. NOE, nuclear determined, some of them at a resolution as high as Overhauser enhancement. 281

0022-2836/90/130281-17

$03.00/O

0

1990 Academic

Press Limited

282

R. Rees et al. 2-fold non-crystallographic

Table 1

the molecular

Molecular and crystallographic characteristics Name Species Molecular mass (Da)

Cardiotoxin Vi’ or toxin III Naja mossambica mossambica 6715

Number of amino acid residues Number of disulphide bridges

60 4

Crystallographic system and space group Unit cell dimensions (A) Number of molecules/ asymmetric unit Solvent for crystallization

Hexagonal a=b=73.9,

(n.c.) symmetry within was confirmed

P6, c=590

2. Structure

Refinement

The main steps of the refinement, procedure are indicated in Table 2. Step 1 started with a model constructed on a Richard’s box. The m.i.r. phases were combined with density modification information (see below). The construction was limited to 1 half of the dimer. The 2nd half was generated by application of the n.c. symmetry operation on the coordinates of the 1st monomer. The R factor in the 5 to 10 A resolution range was 0.493. In this step, the refinement was performed with Hendrickson’s rigid-body program ROTLSQ, modified here for the case of 2 rigid monomers.

potassium

phosphate buffer 1.07 (kO.01) 61

(Steinmetz et al., 1988). From the numerous geometric constraints derived, a three-dimensional molecular reconstitution was obtained that is in good overall agreement with the X-ray structure of cardiotoxin Vi’. A more detailed comparison will be made here. The essential molecular and crystallographic characteristics of cardiotoxin Vir are reported in Table 1. Of interest are the rather large solvent

(a) Restrainedrefinement All other steps were followed by a PROLSQ refinement. The refined parameters were the atomic co-ordinates and the temperature factors (or, in the last steps, the atomic occupancies of ordered water molecules). The “observations” were the geometry parameters of an ideal protein as well as the X-ray structure amplitudes. In an early stage, the existence of n.c. symmetry was also imposed, but not the location of the symmetry axis or the rotation angle. The n.r. symmetry restraints were progressively

content of the crystals and the existence of two molecules in the asymmetric unit. This, and the fact that the self-rotation function shows strong peaks for 180” rotations implies the probable existence of a

Table 2 Summary of crystal structure re$ne,ment

SteD

1 2 3

4 5 6 7 8 9 10 11

Model

building

Rigid-body displacement of each monomer FRODO reconstruction after density-modification Density-modification with new envelope and optimized n.c. symmetry axis “Minimum bias” map Omit map Omit map and 3P,, - 2F,,, map 3F,,, - 2F,,, map 3F,,. - 2F,,, map Omit map and 2F0b,-F,,1

map

GROMOS

refinement

2F,,, - Fcaland Fobs- pEaI

by the

and crystal structure determination have been described (Rees et al., 1987). Here, we are mainly concerned with the refinement of the structure.

&Methyl-2,4-pentanediol

Density (g cme3) Solvent content of the crystals (%, v/v)

which

structure determination. The steps of crystallization, preparation of heavyatom derivatives, X-ray diffraction data collection

2 solution,

dimer,

Resolution range in refinement (4

Number of reflections

H fart,or

5510

705

0477

4-10

1428

0413

3--7

3121

3-10 3-10 3-5 3--5

3308 3308 2611 2611

0360 0321 0.297 0283

3-5 2.8-5 2.55

2611 3371 4002

0265 0269 0.277

25-5 2.5-5

4002 4002

0.247 0.197

map; 48 water molecules and 2 phosphate groups

Except for the 1st step, model building was performed on a graphics picture system Evans & Sutherland PS330 or PS390, using the program FRODO (Jones, 1982), and each step ended with a PROLSQ restrained least-squares refinement (Hendrickson & Konnert, 1980). The resolution range, the number of observed reflections used and the final agreement factor of this refinement are given.

Cardiotoxin

V’i from

283

N. mossambica mossambica

relaxed, and completely discarded in the last steps of the refinement process. In each refinement, the weights u: applied to all “observations”, and particularly to the X-ray structure factors, were chosen so that the average residuals WA’ (where A is the difference between the observed and the calculated value) would be of the order of 1 for each type of observation. Because of the large solvent content of the crystals, the program PROLSQ was modified to enable a correction for disordered solvent. The effect of this correction is discussed in Appendix I. (b) Density modz&ation In steps 1 to 3 (Table 2), the electron density maps were calculated with phases obtained after an iterative density-modification procedure. Density-modification included solvent levelling, lower and upper limit restriction of the electron density, and imposition of n.c. 2-fold symmetry within a molecular envelope of the dimer. After each cycle, the phases calculated from the modified density were combined with the initial m.i.r. phases (Bricogne, 1976). After convergence, a few more cycles were performed, during which uncombined calculated phases were used. The orientation of the n.c. symmetry axis was initially determined from the self-rotation function and its position from that of the heavy-atoms Pt and W. Refinement was attempted later through maximization of the correlation coefficient between the m.i.r. map and its n.c. symmetric, within the molecular envelope. As seen in Table 3, these different determinations are in excellent agreement. The molecular envelope is defined on a 3-dimensional grid and was first obtained by visual inspection of a lowresolution map (5 A). To achieve this, it was found useful to compute a correlated density, defined in Appendix IT. This is close to the usual density wherever there is a strong correlation at points related by the n.c. symmetry, i.e. mainly within the envelope, but tends to vanish outside. The molecular envelope was redetermined in later stages from the current molecular model in the following way: (1) take all grid points inside spheres of suitable radius centred on selected atoms (usually C” atoms, possibly also some extra dummy atoms in regions where the model is badly defined and where one would wish, for safety, to have a more extended envelope); (2) sort the grid point co-ordinates and eliminate redundant points;

(3) eliminate regions that overlap with neighbouring molecules (this is best done by checking the redundancy of envelope grid points after reducing them to the asymmetric unit); (4) to avoid cutting some side-chains, and also to avoid possible holes inside the envelope, it is possible to extend the envelope by 1 or 2 grid points in all directions where there is no overlap with neighbouring molecules; (5) as a final check, visualize the envelope together with the molecular model on a graphics display. If necessary, add more spheres in some particular regions and restart steps (1) to (3). The routines to perform the above operations are part of a density modification package called RMOL, the core of which originates from J. E. Johnson. (c) Unbiased

maps

From step 4 on (Table 2), the phases calculated from the model were used in the calculation of the density map. We tried to reduce the resulting bias toward the current model in several ways. (1) The combination with m.i.r. or density-modification phases (Rice, 1981) was not very successful, probably because of the poor quality of the observed phases. (2) The “minimum bias” method (Z. Otwinowski, unpublished results), used in step 4, also combines observed and calculated phase information. The observed phases are normally m.i.r. phases, but in our case better results were obtained starting with the phases after density-modification. The resulting density map was less ambiguous in some places than previously computed maps. However, the map showed less detailed features. probably due to a downweighting of the data of highest resolution. (3) The easiest way of reducing bias is to replace the Fobs coefficients in the Fourier transformation by Fobs + @ohs -I”,,,). This was done with n=2 (3F,,,,-2F,,, synthesis) in earlier stages and with n= 1 (2Fobs-Fca, synthesis) toward the end of the refinement. (4) Omit maps were calculated either by omitting a few residues from the structure factor calculation, or in the systematic way described by Bhat & Cohen (1984). In most cases the result was almost identical with a

3F,,,,-2Fca,

or

a

2F,,,,- Foal map.

However.

Table 3 Non-crystallographic

Weg.) Self rotation function$ From heavy atoms positions Optimized on m.i.r. map Final model11

90 996 91.9 91.5

in

a few

cases, there was a significant improvement. An example is the conformation of residues 1617: the peptide plane linking these residues was found turned by 180” in one molecule of the dimer compared to the other, so that one molecule presented at this place a /?-turn of type II and the 2nd molecule a turn of type I. The ZF,,,- Fcai map

symmetry

4W-s.) 52.5 53.3 52.2 52.5

axis

zli (4

4 (4

s 0.5 0.6

96 -01 o-2

Error

(A)?

66 0.7

The orientation and position of the axis A may be defined by the 4 numbers C#I,$, 2, and d,. C$ and J, are the spherical co-ordinates of A (J/ is the angle with Oz, C$ the angle of the projection on the ry plane with Ox). H is the point of A nearest to the hexagonal axis Oz, 2, the height of H above the q plane, d, its distance to Oz (taken positive in the A x c direction). 7 r.m.s. displacement of the n.c. symmetric atoms when the axis is changed from the position defined in this line to that in the final model. $ Resolution range 6 to 15 A. Integration radius 25 A. 5 The origin is chosen arbitrarily on the 6, axis. 11 Axis of the best superposition of the Cp atoms of the 2 molecules, excepting residues 28 to 33.

284

B. Rees et al.

seemed to confirm this surprising difference, but the omit map (strand 14 to 18 omitted) was in better agreement with a model having similar type II conformations in phe 2 molecules. (d) Phase extension, at higher resolution There was no phase information available from isomorphous replacement beyond 3 A resolution, but the data from the native protein extended to 2.5 A. Between 3 and >5 A, the resolution was gradually extended. This was achieved through several PROLSQ refinements. After convergence of each refinement, reflections of higher resolution were included, by steps of @l A. Manual reconstruction was done at 2.8 A (step 8 of Table 2) and at 25 A (step 9). (e) Refinement with molecular

dynamics

In step 10, the refinement was continued with program GROMOS (Fujinaga et al., 1989). The energy, which includes an “X-ray term” %J(F,,,-Ir’,,,)‘, was first minimized, essentially to regularize the molecular geometry. A total of 500 steps of O-002 ps each of molecular dynamics were then performed. The initial spectrum of atomic velocities was that of a Maxwellian distribution at 300 K. The force field was derived from the same potential energy as used in the energy minimization step, including the X-ray term. This term was weighted by w=fs -*(F) with a(P)=300, a value close to that used in the PROLSQ calculations. With this weight, the true energy aa well as the X-ray term decreased at a similar rate during the refinement procedure. All observed reflections between 2-5 and 10 A resolution were included in the calculation. The resulting molecular model showed a few anomalous features like distorted tyrosine rings. The normal geometry was easily re-established in a few PROLSQ cycles, with practically no increase in the R factor. It is quite noteworthy that, starting from a molecular model that was considered almost completely refined, the complete process of step 10 resulted in a reduction of the R factor by 3 %. (f) Solvent molecules

listed. Those at the correct distance from an electronegative 0 or N atom, or from an already identified solvent molecule, were included in further PROLSQ refinement as possible water molecules. Of course, the map was also examined visually. This process was repeated several times. GROMOS refinements, with inclusion of water molecules, were also attempted, but they did not improve the agreement between the model and the experimental data significantly. Finally, 2 zones of rather large residual density, approximately symmetric around the n.c. symmetry axis, were tentatively reinterpreted as phosphate groups, after it had been recognized that an interpretation as clusters of water molecules brought the centre of these molecules too close to each other. The B factor of the water molecules was fixed at 20 A2 and their occupancy allowed to vary freely during the last PROLSQ refinement cycles. The lowest occupancy for the water molecules included was 0.38. The occupancy of the phosphate groups was kept fixed, at @5 and @6, respectively, and their B factors submitted to a restrained refinement. (g) Initial

phases: m.i.r. or s.i.r.? The effect of den&y mod@cation

During the early soaking experiments, it had been relatively easy to obtain a platinum derivative, but then a great deal of effort had to be devoted to the search of other heavy-atom derivatives. Only tungstate gave another usable derivative. Additionally, 2 mixed derivatives (platinum + tungstate) were obtained, however. without any new heavy-atom sites. The tungstate derivative had a much weaker phasing power than the platinum derivative. However, the “best” m.i.r. phases were taken as starting value. After completion of the refinement, the m.i.r. phases and the platinum s.i.r. phases were compared with the phases calculated from the final model, considered as the “true” phases. Table 4 shows that the m.i.r. phases are worse than the Cr. phases. Upon density modification (columns 4 and 5 of Table 4), the quality of the 2 sets of phms is less different, but the phases derived from s.i.r. are still 3” better on average than those derived from m.i.r.

At the end of the refinement, an Fobs-FCal difference density map was computed and its largest peaks were

The same Table addresses another question: what was the phase improvement due to the density modification

Table 4 Phasesfrom

m.i.r.

and s.i.r.,

before and after density-modijcation mix.

Figure of merit at 3 A n.c. symmetry correlation$ Correlation with “true” density Correlation with initial m.i.r. or six. density Average phase errorj Phase difference with initial m.i.r./s.i.r. phases

six.

m.i.r.

s.i.r.

Modified?

Modifiedt

@45 0.68 0.26

037 068 0.29

67”

62” -~

079 0.87 043 062 58” 35”

0.84 088 @46 0.57 55” 38”

The phases, and the corresponding electron density, are compared with the “true” values (calculated from the final model). Resolution range, 3 to 20 A. 7 Phases combined from m.i.r. or 8.i.r. and density modification (including n.c. symmetrization). $ Correlation coefficient between the electron density and ita n.c. symmetry-equivalent within the envelope of the dimer. The correlation coefficient between densities p, and p2 is < p1p2 > /( I” I’*) (assuming = =O). < > represents the average over a volume that is usually the asymmetric unit, and the molecular envelope in the case of n.c. symmetry correlation. 0 Weighted by FoF,, times the figure of merit.

Cardiotoxin

Vii from N. mossambica

procedure? Let us remember that density modification was essential to solve the crystal structure, the m.i.r. or s.i.r. maps being uninterpretable. It is surprising that density-modification improved the phases by less than 10” on the average. This confirms the well-known fact that the average phase error should be less than about 60” for the map to be interpretable. Note, however, that this improvement of a few degrees of the phases corresponds to almost doubling the correlation coefficient with the correct density. The refined co-ordinates are deposited with the Brookhaven Protein Data Bank.

3. Results The final R factor for all reflections with Fobs> 3a O-197 in the resolution range 2.5 to 5 A (4002 reflections). It was 6218 in the resolution range 2.5 to 20 A (4800 reflections, local scaling). The corresponding figures for all measured reflections were 0222 (4960 reflections between 2.5 and 5 A, theoretically observable: 5610) and 6237 (5761 reflections between 2.5 and 20 A, theoretically observable: 6421). All restrained geometric parameters were close to their ideal value, with an average r.m.s. deviation always smaller than the assumed standard deviation for each type of “observation”. The r.m.s. deviation of all bond lengths from their target value was 0.015 A, it was 2.4” for the valence angles and 95” for the torsion angle o around the peptide bond. was

(a) Molecular structure Cardiotoxin is a flat molecule with overall dimensions 32 A x 24 A x 15 A (Fig. 1). It is formed essentially of B-strands that interact within two antiparallel P-sheets: a three-stranded sheet, made of residues 20 to 26, 32 to 39 and 48 to 54, and a two-stranded sheet with residues 1 to 5 and 10 to 14. The hydrogen bond network of each P-sheet is

Figure 1. Cardiotoxin

Vi1 A stereoscopic

mossambica

285

continued by at least one additional bond: NH(46). . . OC(49) in the first case, NH(58). . . OC(2) in the second (Fig. 2). The polypeptide strands are held together strongly by four disulphide bridges close to each other in the “bottom’? region of the molecule (in the orientation shown in Fig. l), so that three loops protrude in the opposite direction. These will be called loop I, loop II and loop III. The small /3-sheet is part of loop I, while the three-stranded sheet extends over loop II and one strand of loop III. Three of the disulphide bridges mark the end of the p-sheets. Two of them, 3-21 and 14-38, connect the two p-sheets. The function of the fourth, 54-59, is probably to make the C-terminal loop rigid. Details on the conformation of the four disulphide bridges are given in Table 5; each has a different conformation. Two of them are left-handed, with x3 close to -9O”, and the two others are right-handed, with x3 close to +90”. x1 is close to -6O”, the most probable value a priori (Richardson, 1981), in five cysteine residues, to 180” in one and to the infrequent value of +60” in the two cysteine residues 54 and 59 forming the bridge within the C-terminal loop. This bridge is very peculiar, as it is completely symmetric, and just the mirror-image of the most commonly encountered conformation x1 XX’, z -6W; x2z:x;xx3 x -90” (Richardson, 1981). The shortest main-chain NH. . . 0 distances are reported in Table 6. The hydrogen bond pattern between the two first strands of the large p-sheet is altered locally by a /?-bulge at residue 24. This is of the so-called classic type (Richardson, 1981), with the three side-chains Met24, Va134 and Lys35 pointing in the same direction (seeFig. 3). Figure 2 gives also the values of the local twist angle in the /?-sheets.This is the angle between C?- 1 Cai+l and c’q_,c”i+,) on two consecutive strands of a /I-sheet (Salemme & Weatherford, 1981). It is

view of 1 of the 2 molecules of the asymmetric

unit of the crystal.

286

B. Rees et al.

Figure 2. The main-chain hydrogen bond network. A NH. .O interaction is reported wherever at) least, I of t,hr 2 corresponding N 0 distances determined in the 2 molecules of the asymmetric unit is smaller than 3.5 A (see Table 6). The 2 interactions between Cys53 and Cys53’ connect the 2 monomers of the asymmet)rir unit, thus forming a 6 stranded j-sheet. The Figure also indicates the twist of the 2 /?-sheets. The local intrrchain twist angle between C~~,C~+ 1 and CT- icy, 1 on 2 consecutive strands of a p-sheet (Salemme & Weatherford. 198 I ) is printed between the middle residues i and i’. The angles are given in degrees and are an average between the 2 independent molecules (r.m.8. difference between the 2 molecules, 2.6”).

Table 6 Main-chain

Table 5 The disulphide s-s

Xl

x2

NH

bridges

x3

x;

x;

d(C”. P)

-69 -63

-48 -53

-104 -106

-35 -34

-74 -78

55 c5.7

14-38

-37 -46

-76 -67

-64 -79

-115 -111

-55 -57

5.7 6.0

42-53

-165 -163

-1-72 +64

+87 +s7

f113 f112

-58 -47

60 5.9

+93 +99

+91 +104

f59 f59

69 6.4

3-21

5459

+58 f57

+91 +92

The 5 characteristic torsion angles (deg.), and the distance between the C” atoms (A) are given. x1 and x’, are the torsion angles around the 2 c”-CB bonds, xz and x’* around the U-S bonds, x3 around S-S’. For each disulphide bridge, the 2 lines contain the values refined for the 2 crystallographically independent molecules.

OC

A. Two-stranded 14... 1 3...12 12... 3 5...10 B. Three-stranded 20 39 39 .20 22 .37 37 .22 24 35 35 .24 34.. .24 26. 32 32 .26 21.. .54

Mol.

1

/,I sheet 2.8 2.8

2.6

hydrogen

Mol. 2.9 2.9 2.6 3.1

3.0 /I sheet 3.0 3.3 2.9 2.8 3.3 3.1 :+2 34 “-9 2.9 34 34 ?7 2.6 23 3. I 3.4 3.6 2.9 2%

2 NH

bonds O(

54...21 23 5“I 5%...23 25 .50 50 25 27 .4H

Mol.

-

1

Mol. 2

32 :91 32 3.1 3.1 2.x

C”. Other interaetims 58. 2 30 3-t 8... 5 IO... ‘i 3.3 18. .I5 3.7 46 .49 2.7 59 56 3.0 I). /ntermotrcutar 53 (53’

-.>.(J

The NH 0 distances (A) determined in each of thr 2 molecules of the asymmetric unit are reported, wherever at least 1 of the 2 values is smaller than 35. The shortest NH. 0 distances b&ween the 2 molecules are also given.

Cardiotoxin

V’f, from

N. mossambica

mossambica

287

R

b

Figure 3. The final model, superimposed on the Fourier transform of (a#‘,,,- F,,,)exp(i$,,,). (a) Part of the 3-stranded p-sheet, with the /?-bulge at residue 24. Map contour level, 1.9 times the r.m.s. density. fb) A zone of contact between the 2 molecules of the dimer. The strand from Cys53 to Ser55 of the 2nd molecule occupies the upper right corner. Several short contacts with the 1st molecule can be seen, in particular the main-chain interactions between the 2 Cys53 residues. The interactions involving a bridging water molecule are indicated by broken lines. The side-chain of Asn4.5 is connected t,o a 2nd water molecule seen in the background. Notice on the left side the disulphide bridge 42-53. To make the solvent molecules apparent, the map is contoured here at a lower level, 1.4 r.m.s.

experimentally better defined than the torsion angle 6 = OiCiCi+20i+2 often used to characterize the twist of B-strands (Chou et al., 1982). The right-handed twist is indicated by the minus sign of all the twist angles. The spread of the individual values is rather large, which means that both sheets are not very regular, but it is clear that the two-stranded sheet is much more twisted than the three-stranded sheet, with an average value of 29” for the latter versus 40” for the former. The C-terminal end of the molecule is well structured. It forms a loop held by the disulphide 54-59,

with a b-turn of type II (Richardson, 1981) between residues 56 and 59 (see Table 7). Another type-II p-turn connects residues 15 and 18. The third residue of this loop is the characteristic glycine. A /?-turn of type I ends loop III, but this is somewhat distorted, with a O(46). HN(49) distance as large as 3.6 and 3-7 A in the two molecules of the dimer, while the NH.. . 0 distance between the same residues is significantly shorter (Fig. 2 and Table 6). This distortion could be due to the interactions between the two molecules, discussed below, which involve, for example, the

B. Rees et al.

288

Table 7 Conformation Molecule

A. Loop I

B. Loop II

c. Loop III

The torsion

of turns

1

Molecule

2

4

CL

9

(1,

Leu6 Ile7 Pro8 Ile9

-48 -96 -60 -66

-54 110 -8 -15

-41 -88 -76 -93

-64 113 25 -1

Glu16 Gly17

-96 102

139 -11

-65 86

145 2

Ala27 Ser28 Lys29 Lys30 Met31

-73 -57 58 -99 84

-16 91 -53 - 13 165

-60 -45 108 - 143 -11

-21 72 -95 115 77

-76 158

- 103 -128

-44 141

-45 -16

-64 -120

-29 -18

Asn40 Val41

-69 -120

Ala47 Leu48

-62 -97

Asp57 Arg58

-42 49

129 41

angles C$ and $ of the residues

forming

side-chain of Asn45. There may also be competition between a 0(46) . . .N(49) and a O(46). NS(Lys50) interaction: the latter distance is quite short: 3.2 A and 3.3 A, in the two molecules, respectively. The conformation of loop I is somewhat unusual. Four amino acid residues connect the two strands of the small p-sheet. Their torsion angles 4 and Ic/ are given in Table 7. Three of the four residues, Leu6, Pro8 and Ile9, are in a right-handed helical conformation. Pro8 and Ile9 may even be considered as a fragment of a 3rc helix, which is known to form a tight turn close to a type-I j-turn (Richardson, 1981). Only Ile7 is in a /?-conformation. Its positive II/ angle, probably due to the steric hindrance of Pros, results in an elongation of the polypeptide chain.

-28 63 turns

are given

106 35

Type

of turn

j-Turn

type

II

/&Turn

type

I

P-Turn

type

II

for the 2 independent

molecules.

The electron density is quite clear and the chain tracing unambiguous for loops I and TIT. This is not so for the end loop II, which is probably affected by some structural disorder. There are significant differences in this region between the two molecules of the dimer, in the orientation of the side-chains and even in the conformation of the polypeptide chain (seeTable 7). Very likely, what we seein this region is an average between conformations occurring in the two molecules with a different frequency. A number of side-chains were found at short distances from other side-chain or main-chain atoms indicating probable interactions, principally hydrogen bonds or salt bridges. The most significant interactions are those observed independently in the two molecules of the asymmetric unit. They are reported in Table 8.

Table 8 Hydrogen bonds or salt bridges involving Mol. N(Lys2) N(Asn4) O(Asn4). N’(Asn4) Od(Asn4) O(Leu6) O(Pro8) W(Arg36) N(Gly37) Oa(Asn45) O(Ser46)

Od(Asp57) Od(Asn60) N6(Asn60) O(Asn60) N”(Arg58) N”(Arg36) Y(Arg58) O’(Asn60) N&(AsnGO) N(Tyr61) N’;(Lys50)

‘25 2.7 32 34 3.1 3.1 2.7 3.3 35 29 3.2

I

side-chains Mol.

2

2.8 2.9 31 2.6 28 33 3.0 2.7 3.4 2.9 3.3

Distances are given for pairs of 0. N or S atoms, at least one of them from a side-chain, and separated by at least 2 residues. Only distances ~3.5 A in both independent molecules are reported.

(b) The dimer The two molecules of the asymmetric unit were refined independently (at least in the later stages of the refinement). They possessan essentially identical structure. Only a few external loops show significant differences in the polypeptide chain configuration. Figure 4 shows the two molecules after least-squares superposition of the two polypeptide chains. The overall r.m.s. distance between superimposed main-chain atoms is 0.40 A (0.32 A when residues 28 to 32 of loop II are omitted). Figure 5 shows the average (r.m.s.) distance between corresponding main-chain atoms for each residue after this superposition. The largest differences are seen in the tip of loop II, and are probably related to partial structural disorder. Smaller differences are

Cardiotoxin

Figure 4. Comparison of the 2 molecules of the asymmetric atoms of the 2 polypeptide chains.

observed around residues 9 (loop I), 49 (loop II), 18 and 39 (the 2 “bottom” loops). The differences are of course

larger

for the side-chains:

while

the r.m.s.

distance after superposition is 0.36 A for the 60 C” atoms, it is 1.10 A for the 65 Cy atoms and 1.86 A for the 51 Cd atoms. The two molecules build a fairly strong dimeric association. Two short CO . . NH interactions continue the hydrogen-bond network of the threestranded /l-sheet of each molecule, so that a sixstranded antiparallel pleated sheet extends over the

10

20

Figure 5. Comparison

30

289

V’f, from N. mossambica mossambica

N”CPKNSAL.“KWCCSTDRCN 40 SO

00

of the 2 molecules of the asymmetric unit. (a) The refined B values (in AZ, averaged over the 4 main-chain atoms). (b) The r.m.s. distance (A) between the main-chain atoms of homologous residues, after least-squares superposition.

unit, after least-squares

dimer.

In spite

superposition

of the small

of the N, C”, C and 0

number

of main-chain

hydrogen bonds connecting the two middle strands, the complete B-sheet exhibits the characteristic right-handed twist, with a remarkable continuity of the interchain twist (Fig. 2): C’(36) C’(38) C”(21) C”(23)= -21”, C”(21) C”(23) C”(52) C”(54)= -23”, C”(52) C’(54) Ca(52’) C”(54’)= -21”. Other interactions involve side-chains. The strongest are listed in Table 9. (c) Side-chain

distribution

Cardiotoxin is characterized by continuous patches of hydrophobic residues at the surface of the molecule. The extremity of loop I (5 consecutive residues, from Leu6 to AlalO, followed by Tyrll) is completely hydrophobic, and flanked by a number of basic residues, (lysine and arginine). Furthermore, these five residues are either invariant (Leu6 and Pro8) or mutable only into other hydrophobic side-chains, in all 41 sensu strictu cardiotoxins sequencedso far (this excludes the so-called homologous cardiotoxins, which have more than 60 residues, and show larger differences: see Dufton & Hider, 1988). The second hydrophobic zone covers a large part of loops II and III, and is continuous over the crystallographic dimer. The side-chains are mostly on one side, the “convex” side of the molecule or of the dimer, opposite the C terminus: Va132, Va134, Tyr22, Met24, Leu26, Ala47, Leu48, Val 49 and Tyr51 all point in the same direction, thus filling a flat hydrophobic surface, which is a continuation of the hydrophobic zone of loop I. Note that the two tyrosine residues 22 and 51 are at the border of this surface, as Tyrll marks the limit of the first hydrophobic zone. Met25, Met31, Pro33 and Va152 are the

290

B. Bees et al.

.--

I

ALA

Figure (proline

6. Distribution excepted)

either

of side-chains strictly

of invariant

invariant

hydrophobic

or mutable

into

other

character. hydrophobic

The

Figure shows all hydrophobic amino acids only.

(e) Heavy-atoms

position

Not surprisingly, both platinum sites in the crvstals soaked with PtCl:-are close to rnethioninc residues; one is at a distance of 2.2 A from Sd(Met25) of one molecule of the dimer, the other at’ 2.1 a from the equivalent atom of the second molecule. The sulphur atoms of the two Met31 residues are at 5.1 and 6.0 A, respectively, but a conformational change of t’his side-chain would easily bring them within bonding distance. It is therefore possible that in the derivat’ive each Pt, atom is sandwiched between the t’wo methionine residues 25 and

(d) Solvent structure The positions and occupancies of 48 ordered water molecules were refined, and two clusters of positive density were tentatively assigned to phosphate groups. The strongest interactions with mainchain and side-chain atoms are shown in Table 10. The phosphate groups occupy similar positions in the two molecules of the dimer, each being close to the amine group of Lysl2 and to the hydroxyl

31.

Table 9 interactions

between the 2 molecules Mol.

1

Mol.

2

of the asymrmetric


Mol. 2.

Mol.

N(Cys53)

O(Cys53)

2.9

31

N(Ser5.5) OY(Ser55) OY(Ser55) OY(Ser55)

N6(Asn45) O(Pro43) HN(Asn4.5) Od(Asn45)

322 3.6 27 2.7

34 3c.2 3.3 3.3

N(Thr56) OY(Thr56) OY(Thr56)

N’(Asn45) Nd(Asn45) N<(LysSO)

3.5 3.0 3.1

3.4 2.7 29

Distances

are given

in A

residues

group of Tyr22. A number of significant water moecules were also found in similar locations in the two molecules. In some cases, a water molecule or a small cluster of water molecules bridge two amino acid residues. This is particularly frequent when one residue belongs to the C-terminal region. Also noteworthy is the bifurcated bridge bet,ween the carbonyl groups of residues 43 and 51 and the amide group of Asn45 in one molecule, and the hydroxyl group of Ser55 in the other molecule of the dimer. This is observed twice, in n.c. symmetric locations (Table 10).

only hydrophobic residues of this region that point to the other side of the molecule, but the latter residues are variable, with the exception (probably due to structural reasons) of Pro33, while the former without a single exception are always hydrophobic, and t,hus define an extended invariantly hydrophobic surface. This is quite apparent in Figure 6, where these semi-invariant hydrophobic residues are highlighted. Finally, a third zone is formed by Leul, Leu20, Va141 and Ile39; all lie in the same plane and surround the molecule below the disulphide bridges. Again, the first three residues have their side-chain on the convex side. All four are hydrophobic in all cardiotoxins (except in one case for Leul).

Main

ILL

ALA

I

Cardiotoxin

Table 10 Ordered solvent molecules A. Molecule

1 NH,(l) Lys2 Asn4

~~

NH(7) Lysl2 ~~

WI

CO(14)

w2 w3 w3

Tyrll Arg58

Pl

Tyr22

NH’(38) LyslH Asnl9

~~

w5

Asn40

W8

w9

WlO I

CO(31)

I

w15 w19’ I w13

Wll I W12

Ser28

- CO(55)

COO(60)

Lys23 I

CO(59)

LA,,,

W16

CO(28) I W17

w19 I NH(30)

NH(31)

Lys29

W18

Lys30

(‘O(42)

w20

Asp57

w25

B. Molecule2 Lys2

W26

\M’27 Lys5

Asp57

W28

Tyr22

w32

Glu16

1433 Ser28 ~~~ W34

(1. Molecule

CO(29)~~- w35 /I W36 w37

Lys30

Asn40

w41

~-

1 -molecule

/I w22

W40

291

Vif, from N. mossambica mossambica

~ W42

2 (‘) bridging

CO(43) I CO(51)--~ w21 ~ SIX55 I W23 ~ Asn45 CO(43’) I CO(51’)-- W24 ~ Ser55 I Asn45’

Pl, P2 and Wi are solvent molecules, Pl and P2 probably phosphate groups, the Wi are water molecules. The lines represent distances shorter than 35 A. When the solvent molecule interacts with side-chain atoms, the name of the amino acid is given.

Side-chain conformational changes between native and derivative crystals are probable for the tungstate derivative; no protein atom is at a bonding distance from any of the two W sites. Among the nearest neighbours is Nr(Lys5) (66 and 44 A, respectively). Again, a moderate conformational change of the side-chains, without alteration of the polypeptide backbone, would allow this amine group to make a salt bridge with WO,. Tungstate is a chemical substitute of phosphate, so that this interaction could be a model of the interaction of Lys5 with a phospholipid phosphate.

(f) Comparison

with post-synaptic

neurotoxins

Four X-ray structures of a-neurotoxins are available: neurotoxin B by Tsernoglou & Petsko (1977, in the Brookhaven Protein Data Bank under 1NXB); erabutoxin B by Smith et al. (1988, entry SEBX); a-cobratoxin, by Walkinshaw et al. (1980, entry 1CTX); a-bungarotoxin, by Love & Stroud (1986, entry BABX). The first two are in fact the structure of the same short neurotoxin (62 amino acid residues), determined at high resolution (1.4 A). They are extremely similar, and we will consider only the most recently refined one. The other two compounds are long neurotoxins: 71 amino acid residues in cobratoxin and 74 in bungarotoxin. Their structure was determined at a lower resolution (2.8 and 2.5 A, respectively). Comparison of the primary sequences showed that cardiotoxins and short neurotoxins possess eight identically situated cysteine residues, which are likely to form the same four disulphide bridges. Extensive three-dimensional similarities could therefore be expected for all toxins of the two families. The long neurotoxins have a fifth disulphide bridge, inserted at the extremity of loop II. Additional amino acid residues are inserted at the C-terminal end. X-ray diffraction showed that the four remaining disulphide bridges are at the same location as in short neurotoxins. All neurotoxins have exclusively a b-structure with at least one b-sheet. Figure 7 is the result of a least-squares superposition of the common part of this p-sheet (residues 21 to 24, 35 to 38 and 50 to 54 in cardiotoxin). The degree of conformational identity between cardiotoxin and neurotoxins for this domain can be measured as the r.m.s. distance of the superimposed C” atoms: 0.3 A for erabutoxin, 67 A for cobratoxin and 1.2 A for bungarotoxin. The distances between the C” atoms of all homologous residues are shown in the Figure. The cardiotoxin-bungarotoxin superposition (not shown) gives results similar to the cardiotoxin-cobratoxin superposition, but with larger

distances.

It is clear from Figure 7 that superposition of the central p-sheet brings the largest part of the molecules in fair coincidence. This is particularly so for the N-terminal and C-terminal ends. Larger differences are seen for all loops other than the C terminus, with maximum values at loop II.

B. Rees et al.

292

Erabu

10 20 RICFNHQPQTTKTCSPGESS@&J~SDFRGTII *

:NLCYKhJMLl

20-

LSKKB JVPVKRGC 30 -

20 DIl ‘SKDCPNGH Tl-

5 4 i

60

u

I NV

‘KNSALVKYVCCSTDRCI\

1

40

50

60

10

V-n

40

50

60

WCDFRGH :RVDmAATCPTVKTGVISTDNCN

IIIlId1,d CPE

Cardio

50 GCPTVKPGIKLSCCESEVCX

(a 1

Cobra 6

40 [m

4

Cardio IRCF

30

:NLCYKMMI

,ASKKhW

20-

30 (b)

‘PVKRGC

-

I NVCPP lNSA LVKYVCCSTDRCN

40

50

60

Figure 7. Comparison of the tertiary structures of (a) cardiotoxin and erabutoxin 6 and (b) of caardiotoxin and a-cobratoxin. Distances are shown (in A) between homologous C” atoms. after a least-squares superposition of part of the 3-stranded /?-sheet common to the 3 toxins considered. The aligned sequences are given below and above each diagram. Insertions are indicated by a star. The superimposed C” atoms are underlined (or overlined) in the sequence. The sequence alignment follows that described by Endo & Tamiya (1987) for the neurotoxins. The cardiotoxin alignment was discussed by Rees et al. (1987).

The long neurotoxin is generally less superimposable with cardiotoxin than is the short neurotoxin. An interesting exception is the peptide strand 39-43 (“bottom” of loop III): C” distances between cardiotoxin and cobratoxin are small, which indicates similar conformations, in contrast with erabutoxin, where two residues are deleted in this region and distances of the remaining C” atoms are large. Figure 8 shows the structural changes in a more explicit way. Starting from the superposition of the central /?-cores, other molecular domains may be brought into the best possible coincidence by an additional least-squares superposition. The additional rotations necessary to perform this are represented in the Figure. Strictly speaking, a general displacement may be described as a helicoidal motion, i.e. a rotation followed by a translation along the rotation axis, but in the present case the translation is always very small (less than @5 A). The Figure shows clearly the coincidence of the N-terminal and the C-terminal domains in cardio-

toxin and erabutoxin when t,he central B-cores are superimposed. The superposition of the C-terminal domains (cardiotoxin residues 54 to 60). can be improved only slightly by a rot’ation of 5” (not shown) and the r.m.s. C” distance is then 0.3 A. Cardiotoxin and erabutoxin are both flat and slightly concave molecules, but, the concave side is not the same; it is on the C terminus side in cardiotoxin, on the opposite side in erabutoxin. Figure 8 shows that all three main loops contribute to this change in concavity. The most spectacular change is of course that of loop II. which is turned hy almost 90”. When the rotations shown in the Figure are applied, there is a very good coincidence of the loop I B-sheet (r.m.s. C” distance, 0.4 ,h) and a fail coincidence of loop III (1.1 A). The conformat’ion of loop II is different, and the r.m.s. distance after superposition is larger (1.7 8). Also shown is the rotation that superimposes the polypeptide strand. made of residues 15 to 20 in cardiotoxin, that links the two-stranded and the three-stranded /?-sheets.

Cardiotoxin

V’i from N. mossambica

mossambica

293

Figure 8. Comparison of the tertiary structures of (a) cardiotoxin and erabutoxin b and (b) of cardiotoxin and a-cobratoxin. The thick line is the C” backbone of cardiotoxin, the thin line that of neurotoxin. The common part of the central j-sheet is superimposed by least-squares, as in Fig. 5. The Figure shows the additional rotation (axis and signed angle) necessary to bring into coincidence other parts of the molecule: the 2stranded b-sheet (residues 1 to 5 and 10 to 14 in cardiotoxin), loop II (residues 26 to 32), loop III (residues 42 to 53) and, for the comparison with erabutoxin only, the link between the 2 b-sheets.

The position of this strand is different in the two toxins but its conformation is very similar (the r.m.s. distance after superposition is 0.7 A). Considering now the long neurotoxin, an interesting observation is that the operation that brings loop III into coincidence with cardiotoxin is practically the same as for erabutoxin; the same rotation angle of 27” and similar position of the axis, in spite of the deletion of two residues at the bottom of loop III in erabutoxin. In contrast, the B-sheet of

loop I is displaced in a quite different way in the two neurotoxins; the rotation is small for erabutoxin, larger and in the opposite sense for cobratoxin. (g) Comparison with two-dimensional n.m.r. A complete assignment of the two-dimensional ‘H n.m.r. spectra of cardiotoxin CTXIIb has been carried out and extensive structural constraints

294

B. Rees et al.

derived (Steinmetz et al., 1988). This toxin has 680/b sequence homology with cardiotoxin Vi’. The threedimensional crystal structure of cardiotoxin Vi’ and the structure in aqueous solution of cardiotoxin CTXIIb, as determined by n.m.r., are extremely similar. The agreement includes a number of structural features, such as the existence and extension of the two /?-sheets and the b-turns 46-49 and 56 59. The torsion angles also agree with the indications given by the n.m.r. coupling constants. For example, the bounding values ascribed to three cysteine x1 angles, -20” to - 100” for Cys3, 140” to 220” for Cys42 and 20” to loo” for Cys54, are consistent with t,he X-ray results reported in Table 5. There are, however, several interesting discrepancies, which may be due either to a conformational change in solution or, more likely, to differences in the amino acid sequence; the suggested polyprolinelike conformation of the extremity of loop I could be due t’o the succession of two proline residues at positions 8 and 9. Interestingly, in spite of this difference between t’he two toxins, the n.m.r. distance-geometry calculations indicate that the two-stranded P-sheet is strongly twisted, as observed also in the present X-ray study. In loop IT, the three hydrogen bonds after the p-bulge are not detected by n.m.r., and at the top of the loop the torsion angle 4 at residue 29 is in the interval - 160” to -80” for CTXZZB. in contrast to the positive X-ray values given in Ta’ble 7. Rut in this region the two toxins have a completely different sequence. There is also a small difference in loop III, where a NOE is observed for a O(46) . . N(49) interaction, as expected for a normal b-turn, rather t,han N(46). . O(49) (Table 6). Note, however, that the torsion angles obtained from the distance-geometry calculations with n.m.r. con straint’s (42= -61°. $2= -21”. 43= -91”, $3= -26”. for the “best” solution). match well with t,he X-ray values (Table 7). A larger change seems to occur in the loop connecting the two p-strands (residues 14 to 20); no p-turn 17-l 8 is observed in (‘TXllb (no slowly exchanging proton), and the NOE Ha(18))H”(57) indicating proximity with t’he C-terminal region is at’ complete variance wit#h the X-ray dist(ance of 12.6 A between the respective P/ atoms.

4. Discussion The refined structure of the cardiotoxin molecule is characterized by a well-structured core constituted by the two b-sheets and the C-terminal loop. consolidated by the four disulphide bridges. This core contrasts with less well ordered loops, especially the protruding and rather floppy loop II. This is shown clearly by the variable quality of the electron density maps in the different regions of the molecule, by the magnitude of the temperature factors. and by the more or less exact matching of the two crystallographically independent molecules (Figs 4 and 5).

The C-terminal loop (residues 54 to 60) is in sharp contrast to the other loops; its structure is rein forced by the disulphide bridge 54459 and a @turn. A number of interactions secure it to the rest of the core: Table 8 shows quite clearly that this part of the molecule. and particularly the last residue. Asn60, is tightly bonded to the N-terminal end. Among the C-terminallN-terminal interactions is the NH(58). O(2) hydrogen bond, which may be considered as an extension of the twostranded ,&sheet of loop T. The disulphide bridge 42253 and the interactions of the terminal residue Asn60 with Arg36 cont’ribute to the fastening of the C-terminal loop to t’he central part of the molecule. Ordered solvent molecules add even more to the stabilit)y. The small valur of the thermal motion compared to the rest of the molecule is certainly a consequence of t’his (Fig. 5). Arg36 is invariant, in short neurotoxins as well as in cardiotoxins. Indeed, the irrt,eraction between the guanidyl group of this arginine base and the (l-terminal carboxyl group was also shown in erabutoxin R (Kimball et al., 1979: Tsernoglou B Petsko, 1977). It) is in agreement with t,he unusual pK, value of this carboxyl found by n.m.r. for botjh cardiotoxin Vi’ and a short neurot,oxin; the strong tendency for deprotonation is indicat,ivc of the vicinity of a positive charge (I,aut,erwein of nl.. 1978). The live polypeptide fragments connecting t~hc p-strands are less well ordered and are chara&erized by larger B factors and larger differences between the two molecules of the dimer. This is principal]) so for the large and protruding loop II. which is completely immersed in the solvent. This lack of defined stru&ure correlates with t,hr proba.bIc absence of any functional role (Gatincau et al.. unpublished results), in contrast, to nrurotoxins. where it, is an essential part of the toxic site (Endo & Ta,miya, 1987). Quite different is loop I, which plays a,n important role in the toxic action of cardiotoxin (Dufourcq et al.. 1982; Gatineau et al.. unpublished results). Although a number of interactions connect it to the rest, of the molecule, and more part’icularly to t)he C-terminal end. it) is tjopologically quite distinct. It is characterized by a strongly twisted /?-sheet (Fig. 2). The link between the two strands of this sheet is formed by four residues. the third of which is proline. invariant in all cardiotoxinn sequenced so far. The effect of this proline residue could be to disrupt, t#he otherwise helical conformation of the link and thus cause its elongation, while maintaining a relat,ively rigid structure. This brings the extremities of the B-strands wide apart. and may cause the observed t)wist Another striking feature is the number of intrractions between the two molecules of the asyrn metric unit. which lead t,o formation of a sixstranded ant,iparallel b-sheet. Hydrophobic sidechain interactions. hydrogen bonds between sidechain atoms and bridging water molecules st.rengthen the dimeric association. Whether this assocation is relevant to the toxic activity remains an

Cardiotoxin

V; from N. mossambica

open question. A very similar situation has been observed in the crystal structure of a-bungarotoxin (Love & Stroud, 1986). The comparison between cardiotoxin and the known structures of post-synaptic neurotoxins leads to the same distinction between a stable core and more variable and flexible external loops. The core is made of the central /I sheet and the C-terminal loop, and a few residues at the N terminus. It has the same conformation in cardiotoxin and in the short neurotoxin, and can be very well superimposed, while additional rotations are needed to superimpose the external loops. Essentially the same conclusions are reached when the comparison is made with long neurotoxins, although larger differences are observed in the core, probably due to the additional strand at the C terminus. An interesting observation is that the relative positions of loops II and III are different in cardiotoxin and in neurotoxins, but are the same in the short and long neurotoxins compared. This could be illustrative of a structural invariance of the relative position of those loops in the post-synaptic neurotoxins. Such an invariance could be a biological necessity if the site of interaction with the acetylcholine receptor extends over the two loops II and III, as was wellestablished for several short and long neurotoxins (Kimball et al., 1979; Menez et al., 1982; Inagaki et aE., 1981; Endo & Tamiya, 1987).

x

mossambica

295

Cardiotoxin is characterized by its basicity and its hydrophobicity. Of the 60 amino acid residues, 12 are basic (10 lysine and 2 arginine) and only two are acidic (1 glutamate and 1 aspartate). In many cases, basic and hydrophobic residues are neighbours in the spatial arrangement. This and the peculiar distribution of hydrophobic zones is certainly related with the toxic activity, even if the mechanisms involved, and even the target, are still controversial. In one way or another, the hydrophobic clusters of cardiotoxin certainly interact with the hydrophobic core of the phospholipid membrane, while the basic side-chains form bridges with external negative parts. Here, the contrast with neurotoxins is striking; while the molecular architecture is very similar, the distribution of hydrophobic residues is completely different. For example, in erabutoxin, loop I is essentially hydrophilic. One is rather reminded of scorpion neurotoxins, which, like the cardiotoxins, are membrane toxins and where similar hydrophobic patches exist, in spite of a completely different folding (Fontecilla-Camps et al., 1982). The solvent content of the crystals of cardiotoxin is unusually large for this protein size. This implies an unusual packing of the molecules, which is probably the main reason for the difficulties encountered in growing crystals. This can now be understood by a study of intermolecular contacts. Figure 9 shows

x

Figure 9. Crystal packing. The Figure shows, in the hexagonal unit cell, a dimer (a, a’) forming one asymmetric unit, and its closest neighbours. The contacts are essentially of van der Waals’ type, between hydrophobic side-chains. Each molecule is represented by the c” backbone with the hydrophobic side-chains.

296

B. Rees et al.

the crystal packing. The shortest contacts are observed between one molecular dimer (a, a’ in the Figure) and its two neighbours b and b’ in the 6, helix of the hexagonal packing around the z axis. Almost all those contacts are between hydrophobic residues of the central hydrophobic patch (see above); namely, Met24, Leu26, Va134 and Va149. Leu6 of loop I of each monomer is also involved, in contact with a neighbouring Leu48. The only hydraphilic interactions are given by Lys35; in one monomer it is in close contact (2.9 A) with Ser49, in the other it may interact weakly with Tyr51 (at 3.8 A).

A

smaller

number

of

contacts

is

Correction

I

for Disordered

0.3

? 1

seen

between two different helices. (a with c and a’ with c’ in Fig. 9.) They involve mostly the hydrophobic residues of loop I. Obviously, the crystal packing is dominated by the necessity, for energetic reasons, to hide the exposed hydrophobic side-chains from hydrophilic parts of other molecules or from the solvent. A closer packing would result only in a marginal gain in energy, more than compensated by the loss due to more exposed hydrophobic side-chains. It, would be interesting to know if similar, even if only temporary, molecular associations exist in solution. and their bearing on the mode of action of the cardiotoxins in the venom.

Appendix

0.6

0.2

0.1 20

IO

5

4

3

i 2-5

X(2Sl”8d)

Figure 10. Correction for disordered solvent.

Thr

Figure reports the R factor versus the resolution, after restrained refinements in the resolution range 2.5 to 20 A. This total range wasdivided into 20 domains equidistant in sin20/1’. All reflections with F > &r(F) were included in the calculation of the R factor. (a2(F) = crf + (2F)‘. where ~7~ is the standard deviation from counting statistics.) Dotted lines, the refinement included a correction for disordered solvent. Continuous lines. no solvent correction. overall H factor. Broken lines. no solvent correction. local scaling (scale factor determintd separately for each domain of resolution).

Solvent

A large solvent content of the crystals has a big systematic effect on t,he diffracted intensities, resulting in large variations of the scale factor between Fobsand uncorrected Fca, values, as a function of resolution. The effect of disordered solvent can be approximated by subtracting from the calculated amplitude diffracted by each atom that of the corresponding volume of uniformly distributed solvent. For simplicity, an identical spherical volume is considered for all atoms of the same type (M. S. Lehmann, unpublished results). A common temperature factor is applied to all the solvent volumes, at least equal to the largest temperature factors of the protein atoms. The restrained refinement program PROLSQ was modified to include this contribution of the disordered solvent to the structure factors and to their derivatives. This correction results in a change of a factor of about, 2 in the average scale factor (10.6 instead of 5.6) and an increase of 19 A2 of the apparent temperature factor. Figure 10 shows the variation of the agreement factor R with the resolution, with and without solvent correction, after several cycles of PROLSQ in the 2.5 to 20 A resolution range. Since, in the absence of correction, the scale factor is very dependent on resolution, a local scaling in

small resolution ranges was performed before the calculation of the R factor. In the solvent correction case, this makes no significant difference in the R factor.

It is clear from this Figure that, even with this local scaling, the observed data agree het.ter wit.h the solvent-corrected calculation at low resolution. At intermediate resolution: however, from 7 A to about 4 A, the reverse is true. which seemsto indicate that this solvent model is probably too crude when the resolution is increased. At even higher resolution, the effect>of the correction is just thab of a change in scale and H factor, so that the t,wo models become equivalent. For these reasons. the latest stages of refinement were done without solvent correction, but discarding all low-resolution reflections (low resolution limit, 5 A). Note that, while large on the temperat,ure factor. the effect of the solvent correction on the atomic coordinates is small; after the two PROLSQ refimments mentioned above, the r.m.s. positional difference was 0.09 A for all atoms, and only 0.035 A for the main-chain atoms alone.

Appendix II Correlated Density If one wants to make use of n.c. symmetry in a density-modification approach, it is necessary to define a molecular envelope, i.e. the limits of a three-dimensional region within which the n.c. symmetry applies. The automatic mask determination methods used for solvent levelling cannot be

Cardiotoxin

V’f, from N. mossambica mossambica

used here, since such methods can define the limits between protein and solvent, but not between adjacent proteins. The problem is usually solved by visual inspection of a low-resolution map. Sometimes the density is first averaged about the n.c. axis, since this enhances the region of interest; if pt is the true density, the expected value of the average is pt within the envelope, but close to zero outside, because of the lack of correlation. The concept of correlation can be used in a more systematic way. If p is considered as a statistical variable of mean 0, an estimate of the correlation coefficient between the densities at the n points r=ri,. . . r, related by the n.c. symmetry is: Covariance y(r) = Variance

= ~&+jp(rib(ri)/(n2 -4 xP(ri)21n

=-&(g!$ -1). In the present case, n = 2 and y(r) = 2Arl Wd/ (P(ri)2 +p(r2)2). The correlated density y(r)p(r) has the following advantage over the simple average of densities; for a given value of p(rl), the correlated density is maximum when p(r2) =p(rl), and smaller if p(r2) is either smaller or larger than p(rl). The average density is larger the larger p(r2), even if the matching with p(rl) is poor. An averaged correlated density map may make the molecular boundaries even more clearly apparent.

References Bhat, T. N. & Cohen, G. H. (1984). J. Appl. Crystallogr. 17, 244-248. Bricogne, G. (1976). Acta CrystaZZogr. sect. A, 32, 832-847. Chou, K. C., Pottle, M., Nemethy, G., Ueda, Y. & Scheraga, H. A. (1982). J. Mol. BioZ. 162, 89-112.

Edited

297

Dufourcq, J., Faucon, J. F., Bernard, E., Pezolet, M., Tessier, M., Bougis, P., Van Rietschoten J., Delori, P. $ Rochat, H. (1982). Toxicon, 20, 165-174. Dufton, H. J. & Hider, R. C. (1988). Pharmacol. Ther. 36, 140. Endo, T. & Tamiya, N. (1987). Pharmacol. Ther. 34, 403451. Fontecilla-Camps, J. C., Almassy, R. J., Suddath, F. L. & Bugg, C. E. (1982). Toxicon, 20, l-7. Fujinaga, M., Gross, P. t Van Gunsteren, W. F. (1989). J. AppZ. CrydaZZogr. 22, l-8. Harvey, A. L. (1985). J. ToxicoZ.-Toxin Rev. 4, 41-69. Hendrickson, W. A. & Konnert, J. H. (1980). In Cry8takgraphy (Diamond, R., Computing in Ramaseshan, S. & Venkatesan, K., eda), pp. 13.01-13.23, Indian Inst. Sci., Bangalore. Inagaki, F., Clayden, N. J., Tamiya. N. & Williams, J. P. (1981). Eur. J. B&hem. 120, 313-322. Jones, T. A. (1982). In ComputationaZ CrystaZZography, (Sayre, D., ed.), pp. 3033317, Oxford University Press, Oxford. Kimball, M. R., Sato, A., Richardson, J. S., Rosen, L. S. & Low, B. W. (1979). Biochem. Biophys. Re8. Commun. 88, 950-959. Lauterwein, J., Lazdunski, M. 6 Wiithrich, K. (1978). Eur. J. B&hem. 92, 361-371. Love, R. A. & Stroud, R. M. (1986). Protein Eng. 1,37-46. Menez, A., Boulain, J. C., Faure, G., Couderc, J., Liacopoulos, P., Tamiya, N. & Fromageot, P. (1982). Toxicon, 20, 95103. Rees, B., Samama, J. P., Thierry, J. C., Gilibert, M., Fischer, J., Schweitz, H. & Lazdunski, M. (1987). Proc. Nat. Acad. Sci., U.S.A. 84, 3132-3136. Rice, D. W. (1981). Acta CrystaZZcgr. sect. A, 37, 491-500. Richardson, J. S. (1981). Advan. Protein Chem. 34, 167-339. Salemme, F. R. k Weatherford, D. W. (1981). J. Mol. BioZ. 146, 101-117. Smith, J. L., Corfield, P. W. R., Hendrickson, W. A. & Low, B. W. (1988). Acta Crystallogr. sect. A, 44, 357-368. Steinmetz, W. E., Bougis, P. E., Rochat, H., Redwine, 0. D., Braun, W. & Wiithrich, K. (1988). Eur. J. Biochem. 172, 101-116. Tsernoglou, D. & Petsko, G. A. (1977). Proc. Nat. Acad. Sci., U.S.A. 74, 971-974. Walkinshaw, M. D., Saenger, W. & Maelike, A. (1980). Proc. Nat. Acad. Sci., U.S.A. 77, 2400-2404.

by R. Huber.