Structure determination of the glycosomal triosephosphate isomerase from Trypanosoma brucei brucei at 2.4 Å resolution

.I. Mol. Riol. (1987) 198. 109-121

Structure Determination of the Glycosomal Triosephosphate Isomerase from Trypanosoma brucei brucei at 2-4 A Resolution Rik K. Wierenga, Kor H. Kalk and Wim G. J. Ho1 Laboratory of Chemical Physics, University of Groningen Nqenborgh 16, 9747 AG Groningen, The Netherlands (Received 1 April

1987, and in revised form

7 July

1987)

The three-dimensional crystal structure of the enzyme triosephosphate isomerase from the unicellular tropical blood parasite Trypanosowm brucei brucei has been determined at 2.4 A resolution. This triosephosphate isomerase is sequestered in the glycosome, a unique trypanosomal microbody of vital importance for the energy-generating machinery of the trypanosome. The crystals contain one dimer per asymmetric unit. The structure could be solved by the method of molecular replacement, using the refined co-ordinates of chicken triosephosphate isomerase as a search model. The positions and individual isotropic temperature factors of the 3792 atoms of the complete dimer have been refined by the Hendrickson & Konnert restrained refinement procedure. While tight restraints have been maintained on the bonded distances, the R-factor has dropped to 23.2% for 12317 reflections between 6 A and 2.4A. A total of O-6 mg of enzyme was used for establishing the correct crystallization conditions and solving the three-dimensional structure. Although the sequences of trypanosomal and chicken triosephosphate isomerase are identical at only 52% of the 247 common positions, the overall folds are very similar. The architecture of the active sites is virtually the same with 85% of the side-chains being identical. On the other hand, the residues involved in the dimer contacts are the same at only 55% of the positions. Nevertheless, the position of the local 2-fold axis in the chicken and glycosomal dimers is similar. A remarkable feature of glycosomal triosephosphate isomerase is its high overall positive charge. This extra charge is concentrated in four clusters of positively charged side-chains on the surface of the dimer, quite far away from the active site. These clusters may be involved in the mechanism of import of this triosephosphate isomerase into the glycosome.

To date nine sequences (eukaryotic as well as prokaryotic) have been determined. The alignment of the trypanosomal (Swinkels et aE., 1986) and the chicken (Straus & Gilbert, 1985) sequences is shown in Figure 2. Throughout this paper we will use the trypanosomal TPIase (gTPIase) sequence numbering. The crystal structures of chicken TPIase (Banner et al., 1975; Alber et al., 1981) and yeast TPIase (Alber et al., 1981) have been described. The chicken structure in particular is well characterized. Each subunit is composed of alternating b-strands and cc-helices. The eight parallel b-strands form an

1. Introduction Triosephosphate isomerase (EC 5.3.1.1; TPIaset), is a dimeric enzyme, consisting of two polypeptide chains of approximately 250 residues each. The reaction catalyzed by this enzyme is shown in Figure 1. Only intact dimers are catalytically fully active (Waley, 1973), but co-operative interactions between the subunits have not been observed. Cofactors are not required. TPIase is a well-characterized enzyme, in terms of both its structural and its catalytic properties.

inner

t Abbreviations used: TPIase, triosephosphate isomerase; gTPIase, glyrosomal triosephosphate isomerase from Trypanosoma brucei brucei; F,, observed structure amplitude; F,, CI,, calculated structure amplitude and calculated phase, respectively; r.m.s., root mean square; 1 .A = 0.1 nm. 0022-2836/87/210109-13

$03.00/O

cylinder.

This

P-barrel

is shielded

from

the

solvent by eight helices, running antiparallel to the strands of the inner barrel (Fig. 3). The active site resides at the C-terminal end of the p-barrel. Its properties are influenced: (1) by a number of catalytic residues; (2) by two short 109

0 1987 Academic Press Limited

110

R. K. Wierenga et al Dihydroxyocetone -phosphate E

chicken TPIase, Fig. 2). For the subsequent steps a histidine (His95; His95 in chicken TPTase) and a lysine (Lysl3; Lys13 in chicken TPlase) are

D - glycemldehyde 3 phosphote

TPIose (no -cofactors)

important,

P d-0 ! ‘coo DHAP

GAP

Figure 1. The tea&on

cntalyzed

probably for proton transfer from 0-I to

O-2 (by His95) and stabilization of the negative charge of the intermediat,es (by 1~~~13). The positions of these residues in the folded protein are shown schematically in Figure 3. The free-energy profile of t.he reaction. deduced from the rates of the individual steps, has been determined by Knowles $ Albery (1977). The analysis showed that the rate-limiting steps in vivo are the on/off steps, which themselves are diffusion limited. It has therefore been suggested that TPTase is a “perfect” enzyme, which has arrived at the end of its evoiutionary development as a catalyst, in the sense that any further acceleration of the catalytic steps does not have an effect on the rate of the overall reaction in vivo (Knowles 62 Albery. 1977). Recentlv a TPIase mutant has been characterized in w&h the active site Glul67 is changed into Asp (Straus et al.. 1985). Owing to this mutat,ion. the active-site architecture is changed in such a way that t’he catalytic step has become the ratr-limiting step. The change in position of the carboxylicx acid moiety in the active site could be as small as 1 A; nevertheless. t.he catalytic rate is reduced by a fact’or of 1000 (Raines et nl., 1986). Alagona et aL6. (1986) have carried out some energy calculations to

by triosephosphate

isomerase.

helices, a phosphate-binding helix and an active site helix (Fig. 3); and (3) by the position of two loops, a “flexible loop”, which apparently closes over the active site after substrate bindmg, and a welldefined “interface loop”, which protrudes from the other subunit into the active site (Alber et al., 1981; and Fig. 3). The reaction mechanism and its energetics have been studied in great detail. A cis-enediolate (enediol) intermediate has been demonstrated by Rose (1962). It is formed from dihydroxyacetone phosphate after abstraction of the l-pro-R proton by t,he active site glutamic acid (Glul67: Glul65 in p1 QPlAAAlll@ KFFVCDllY@ L Ill)

TSINHD

vqcvvA5

( 31)

A Y t 5 A 0

TEVVCGA

gTPIas@

(56)

SHPKF

Ch,ck.nTPIase

( 56)

D A

gTPIOlC Chicken

TPIase

L VI

AAq

GVAA9

K I

IAIAKS.GAflGE"S

LPLLXD

NC"KVPKGAFTGEIS

PAlklXD nO.4

_84 (86)

gTPIare clvtlr.n

TPIosr

I 86)

YlVtG

FGVN

YYILG

IGAA

w @S

E R R:?'i'i?G : .,......: @SfRRHVFGfSD

t T "

El"ADKYAll"A ELlGqX"AHAL.4

$35 gTPIor* Ch,ck.n

TPIore

(119)

SGF

,tV,AClGE

TLqERESGRTA

(119)

E GL

GVIACIGf

KLOEREAGITE

L KV"IAV@

PVYAIGTCKVATP

KVVLAV@

PVYAIGTGKTATP

na7

L gTPlos*

(198)

IGADVRGE

Ch,ckmTPIose

(1%)

Y 5 D AV

AQ 5

LRILVGG

SV"G

1 R I I V G G

SVlG

J-L

w

~Tr,VLI

(2251

GNCKELASP

1223i

-

gTP1-M

1226)

R 0

VNGFL

"GGASLKP

E‘"Ol1XA

Ch,ck."TPJoa*

1224)

" 0

VOGFL

VGGASLYP

EFYOI

gTPIor*

(2491

T 4

ch,ck.nTPlose

(247)

K H

lid81 INA

12461

Figure 2. The sequences of trypanosomal (gTPTaue) and chicken (TPTase) enzymes. The numbers in parentheses are the residue numbers at the beginning and at the end of each line. The dots f ) indicate deletions. Only the residues in thp 8 /?-strands and 8 a-helices, which together form the characteristic TPIase-barrel, are specifically labeled fil to fiS and stl snd a8. respectively. The active site residues, Cysl3, Hi&5 and Glul67. are encircled. The unique gTPTase sequence near the active site, AlalOO-TyrlOl-TyrlOl, is shown within a box of dotted lines. The hot spot 1 sequence is within broken lines and the hot spot 2 peptide is within continuous lines. The sequences of trypanosomal and chicken enzymes are 5204, identical.

Glycosomal

Triosephosphate

lsomerase

111

Structure

acbve site cavity

flexible loop

active site helix

phosphatebmdmghehx

interface loop

Triosephosphate (subunit

isomerose I)

Figure 3. A schematic drawing of subunit 1 of the gTPTase dimer. The approximate positions of the active-site residues Lysl3, His95 and Glu167 are shown. The black dots, emphasizing the interface region, indicate the 27 residues with atoms that are closer than 3.5 i% to atoms of the other subunit (see also Table 3). The interface loop (72-79) points towards the active site of subunit 2. The flexible loop (169-179) closes over the active site after substrate binding (Alber el ccl., 1981). The 8 p-strands and 8 cc-helices forming the core of the TPIase barrel are labeled as /?l t,o 88 and al to ~8. respectively. The active-site helix is formed by residues 95 to 103. The phosphate binding helix is formed by residues 234 to 239. (This picture is a modification of the artistic drawing made by ,J. Richardson (1981).)

try

to

simulate the effects of this mutation, that even a crude model of the active site leads to reasonable energy parameters, at least for one of the steps in the reaction path. Nevertheless, a quantitative understanding of the energetics of showing

the entire

catalytic

reaction

path

has not yet been

reached. Our studies will extend the database, from which further progress can be made to a full understanding of the properties of TPTase. At the same time, our structure determination project aimed at a better

is also part of a broad understanding of the

enzymes of the trypanosomal glycolytic system. Trypanosomes are protozoan hemoflagellates that (*ause a number of diseases of great medical and economic importance (Molyneux & Ashford, 1983). In tropical Africa they cause sleeping sickness in man and nagana in cattle, affecting the physical and economic aspects of life in a region as large as Europe. En t)he bloodstream form these parasites are dependent entirely on glycolysis for energy production, and hence for survival. Therefore, the glyco!ytic pathway has been suggested as a promlsmg target for drugs against trypanosomiasis (Opperdoes, 1985; Several 1984). Wang, biochemical peculiarities of the trypanosomal glycolytic pathway enhance the probability that selective inhibitors can be found. For example, the first seven enzymes

of glycolysis

are sequestered

in

a unique microbody-like organelle, the glycosome (Opperdoes et aE., 1984). It has been established that these enzymes are synthesized outside the glycosome and are translocated without any processing (Hart et aZ., 1987).

As it is advantageous, for rational drug design. to know the three-dimensional structure of target enzymes (Goodford, 1984; Hol, 1986), we initiated the crystallization of these glycolytic enzymes (Wierenga et al., 1984; Read et al., 1987). This structure determination of gTPIase is the first, structure of a parasitic enzyme to emerge from these investigations. With this structure we can try to understand the kinetic properties of gTPIase (Lambeir et al., 1987). It may also shed more light on the postulated mechanism for the import, of the glycolytic enzymes into the glycosome (Wierenga et al., 1987). In addition, it may be the start of the rational drug design cycle (Hol, 1986). From the point of view of rational drug design, not, only the active-site

architecture

but the complete

structure

is important. In the case of trypanosomal TPTase. those parts of the structure that differ from human TPIase are crucial. Any compound that binds specifically to gTPIase might be modified via a cyclic learning process in such a way that it will eventually interfere with the proper functioning of gTPIase. Such compounds might either be inhibitors of the catalytic process, interfere with the import into the glycosomes, or just destabilize the folded dimer; or they might combine some of these properties. 2. Materials

and Methods

(a) Puri$cation, crystallization, data collection and data processing Purified glycosomal TPIase was kindly supplied by Dr Opperdoes (ICP, Brussels) and collaborators. It is

R. K. Wierenga et al.

112

obtained via a purification procedure in which all 9 glycosomal enzymes are purified simultaneously, as described by Misset et al. (1986). Although gTPIase is obtained in the smallest amounts of all enzymes by this procedure, it appeared that the 0.6 mg available was sufficient for finding the correct crystallization conditions, growing 6 crystals with minimal dimensions of 0.2 mm and solving the structure as described in this paper. The gTPIase crystals grow in a 60% saturated ammonium sulfate solution in a 0.2 iu-3-(N-morpholino)-propane sulfonic acid (Mops) buffer (pH 7.0), containing also 1 mM-EDTA, 1 mM-dithiothreitol and 1 mM-sodium azide (Wierenga et al., 1984). The space group is P2,2,2, with cell dimensions a = 113.2 A, b = 97.77 A, c = 46.50 A. The crystals are well-ordered, diffracting to at least 1.8 A resolution. There is 1 dimer per asymmetric unit. Three crystals were used for data collection by the oscillation method, using an .Elliott GX20 rotating anode as the X-ray source. The crystal-to-film distance was 75 mm. Altogether, the data were recorded on 56 film packs, each consisting of 2 films. These films were digitized by a Scandig microdensitometer with a 100 x 100 pm* raster. Subsequently, these digitized films were processed and merged into 1 dataset by programs of the Groningen BIOMOL crystallographic software package. For the final dataset, 32,158 structure factors of fully recorded reflections were scaled and averaged into 1 dataset, consisting of 13,795 unique reflections The overall Rmcrgewas 5.8%, where: R merge=

1 2 I I Gkll-I hkl

p hkl., I/( hk, 1 Nj I lTihklI )

j=l

The resolution varies from 23 A to 2.37 A. The data are 8O”/b complete to 2.8 A, and 70% complete to 2.5 A.

= 240 .o tg 200 'hE 160

I

I

I

I

I

I

I

III

IO 20 30 40 50 60 70 00 90 PC”)

Figure 4. Cross-rotation functions (7 to 3 A data) of chicken T)ersua glycosomal TPIase. Two complete crossrotat,ion functions are compared: (1) FC is calculated with all atoms of the chicken TPIase dimer (-O-O-); (2) F, is calculated with the atoms of the polyalanine skeleton of the chicken TPIase dimer (-•-•-). The highest peak of a p-section (vertical axis) is plotted as fun&ion of /? (horizontal axis). /? is the Eulerian angle, as used by the fast rotation function program (Crowther, 1972) and specified in Materials and Methods. These 2 peaks of the cross-rotation function are related by crystallographic symmetry to, respectively. a = 35”, b = 125”, y = 30” and CI= 267,5”, /3= 30”, y = 280”. which are related to each other by the 2-fold symmetry operation observed in the self-rotation function. Applying the transformation specified by the second set of angles resulted in a correct orientation of the chicken TPIase dimer in the trypanosomal enzyme unit cell. In this new orientation the local 2-fold axis has the same direction as observed in the self-rotation function (cp = 155”, I(/ = 90”) of gTPIase.

(b) The self- and cross-rotation function The self-rotation function has been calculated with the Crowther (1972) fast-rotation function program. In this program the orientation variables are the Eulerian angles CI, /? and y. The definitions of c(, p and y in this program are: t( is a rotation about the z-axis, fi is a rotation about the new y-axis and y subsequently specifies a rotation about the final z-axis. The calculations have been done with data between 7 A and 3 A resolution, using a radius of integration of 17 A. The origin-peak is 50 (arbitrary units); the next highest peak (19) is observed at c1= O”, /l = 130”, y = 180”, which corresponds in a polar system (Rossmann k Blow, 1962) to a 2-fold axis, K = 180”, with cp = 155”, $ = 90”. The third highest peak is 16. The sequence of chicken TPIase is 520/b identical with trypanosomal TPIase. Therefore. the chicken enzyme structure is a good search model for finding the orientation and position of gTPIase in its unit cell. For the structure-factor calculations the co-ordinates of a refined chicken gTPIase dimer, kindly made available by Drs D. C. Phillips and P. Artymiuk, were used. The dimer was placed in a sufficiently large orthogonal unit cell (a = 75 A, b = 100 A, c = 70 A) with symmetry Pl. The cross-rotation function (Crowther, 1972) was calculated with data between 7 A and 3 A. using a radius of integrat’ion of 17 A. The highest peaks are at a = 145”, fi = 55”, y = 210” (peak

height,

176 arbitrary

units)

(c) The translation function Knowing the correct orientation permits the calculation of a translation function (Crowther & Blow, 1967). Three translation functions, corresponding to the 3 crystallographic 2-fold screw axes, were calculated with data between 6 A and 4 A. Since the spacegroup is P2,2,2,, the maxima in the translation function are expected in sections z = l/2, y = l/2 and z = l/2. The highest peaks were very prominent (Fig. 5) and indeed did occur in the expected sections. As can be seen from the data in Table 1, the 3 highest peaks of the 3 relevant sections are related by translations of half a unit cell. allowing for a straightforward solution of the translation function. (d) Rigid body refinement Subsequently, the position of the chicken TPIase dimer in the gTPIase unit cell was marginally changed on the

Table 1 Maxima

and

M = 87.5”, p = 30”, y = 280” (peak height, 174). The next highest peak, at c( = 117.5, fl = 85”: y = 230”, is much lower: with a peak height of 162.5 arbitrary units (as cross-rotation function shown in Fig. 4). Another calculated with a polyalanine chicken enzyme model, confirms the importance of the 2 highest peaks (Fig. 4).

in the translation Fractional

Value of

Value of the

co-ordinates of the highest peak

the highest peakt

next highest

0.283

99

0.32 :

99 99

54 70 73

s S&ion z = 4 ’ Section y = 4 0.230

Section 2 = 4 0.80 t Arbitrary

function

units.

Ok f

0.52

pkt

Glycosomal Triosephosphate Isomerase Structure

113

Section y = $ I-0

0

O-9

-0

0.8

-

O

0 0

0

0

0

0

o-7 0.6 .Y ::

0

-

a

,a

0

b

0

o-5-

:,

0

0

0.4 0.3 0.2

o

-

0

0

0

0 0

0.1

0

I@ O

@

-

d

0

0-I

0

0.2

, 0.3

I 0.4

0

0 I 0.5

I 0.6

I 0.7

Q 0.8

00 0.9

IO

x-axis

Figure 5. The translation function (Crowther & Blow, 1967); section y = l/2, 6 to 4 A data. F, is calculated from the complete chicken TPIase dimer, oriented properly according to the cross-rotation function. The highest value in this section is 99 (arbitrary units). Contour lines have been drawn at levels 39, 59, 79 and 99. basis of an R-factor search using data between 10 A and 6 A resolution (Fig. 6). The R-factor was reduced from 43.5% to 42% by shifting the molecule 0.5A along the x-axis; small rotations of the complete dimer around z (2”) and y (1”) further reduced the R-factor to 41.1% (10 to 6 A data): for the 10 to 5 A data the R-factor is 41.9%. The overall r.m.s. displacement is 08 A. Finally, the subunits of the dimer were allowed to move independently in a CORELS refinement run (Sussman, 1985) with data between 10 A and 5 A, reducing the R-factor from 41.9% to 40.1 yc (10 to 5 A data). The overall r.m.s. displacements due to the CORELS refinement, are 0.7 A for subunit 1 and 0.5 A for subunit 2. At least 2 observations further strengthened our confidence in the found solution. (1) The TPIase dimers are nicely packed in the gTPIase unit cell (data not shown). There are no clashes of overlapping atoms and reasonable packing contacts are present. (2) Side-chains of the correct size could be seen in a Sim-weighted m(2lF,J--lF,)exp(ia,) map, for which F, and a, had been calculated from a pruned chicken TPIase model in which all side-chain atoms beyond CB had been removed (Fig. 7(a) and (b)). The figure of merit was calculated according to Sim (1960). Throughout’ these investigations the Fourier maps,

(e) Model building in a 3 A electron density map The co-ordinates of the chicken TPIase model as obtained after the rigid body refinements (see section (d) above) were used as the starting point for building a model of the gTPIase structure into a Sim-weighted m(3 IF,, I- 2 I F, I)exp(ia,) map, calculated with data between 7 A and 3 A resolution. F, and a, were calculated from a st,ructure that was obtained from the rigid-body refinement model (see section (d), above), after the 3 following steps. (1) Atoms occurring in chicken TPIase, but not in gTPIase were removed from the model. (2) This co-ordinate set was subjected to a few cycles of Fast Fourier Transform refinement, (Agarwal, 1978) (r.m.s. displacement with 0.3 A) intermittent regularization by the procedure of Dodson et al. (1976). (3) Residues near insertions/deletions were removed. Starting at the pu’ terminus, the complete subunit 1 of chicken TPIase was fitted as well as possible into the electron density map, using the interactive computergraphics program GUIDE (Brandenburg et al.. 1981). In most regions this first gTPIase model could be built in a quite straightforward manner. Whenever sequence differences occurred. then the chicken enzyme side-chain was replaced by the gTPIase side-chain. Also deletions and insertions were appropriately accounted for at this stage. Subunit 2 was generated by superimposing the fitted structure of subunit 1 on top of subunit 2 of the chicken model. This first complete gTPIase model was subsequently improved, as described in the next sections. (f) Refinement of the model at 24 A resolution

42 1 40 ’

which are on an arbitrary scale, have been calculated with the XRAY System (1976) on a Cyber 170/760 and inspected on an Evans & Sutherland Picture System 2, using the computer graphics program GUIDE (Brandenburg et al., 1981).

I -6

I -4

I 0

I -2 Shifts

I 2

I 4

I 1 6

(8)

Figure 6. The rigid-body R-factor search; 10 to 6 A data. F, is calculated from the chicken TPIase dimer, oriented and positioned according to the rotation and translation function F, is from gTPIase. The shifts are along the z-axis (O), the y-axis (0) and the z-axis (a).

The model was refined by the restrained least-squares procedure of Hendrickson & Konnert (Hendrickson, 1985). The program used was the Purdue Cyber 205 version, kindly made available by Michael Rossmann and co-workers, and adapted to the Cyber 205 in Amsterdam by Anne Volbeda. van der Waals’ restraints and torsion restraints were not applied. The existing noncrystallographic symmetry was constrained to the situation as observed after the rigid body refinements.

114

H. K. Wierenga

et al.

(a)

( b)

i

I 7

(d)

Fig. 7.

Glycosomal Triosephosphate Isomerase Structure For the first cycles, data between 6 A and 3.5 A were used; the resolution of the data was gradually increased to 2.8 A at cycle 5. After 7 cycles, individual temperature factors of the 1896 atoms were also allowed to vary, resulting, after 19 cycles, in an R-factor of 31 O/Ofor the 9128 reflections between 6A and 2.88. The r.m.s. displacement for all atoms due to this refinement is 0.5 A. The 2.X A model was analyzed for a number of features, such as van der Waals’ contacts, q-$ angles and individual B-factors. Furthermore, the average eleitron density per residue in a 2.8 A Sim-weighted m(21E’~1-IFEl)exp(icc,) electron density map (F, and Q, from the 2.8 A model) was determined. Subsequently. the residues with bad geometry. as well as the residues occurring in weak electron density, were removed from the refined 2.8 A model. Altogether, 30 residues (472 at,oms) were removed from each subunit, consisting originally of 250 residues (1896 a.toms). With this partial model a new Sim-weighted m(2 1F, I- 1FL l)exp(iab) electron density map was calculated with 1F: 1 and a: the structure amplitude and phase of the partial model. respectively. With this map, contoured at a level equal to 0.8 times the r.m.s. electron density. all residues of subunit, 1 were rebuilt. Only in a few regions (residues l-3. 30-36. 151-159, 176179, 220-222, 250) was it impossible to decide where to place the atoms of the residues. Rebuilding resulted in a 0.8 A r.m.s. displacement for all atoms of subunit 1. The subunit 2 coordinates were obtained through the non-crystallographic: symmetry operations. as established after the rigid body refinement. by CORELS. (g) Rejnement

115

Table 2 ReJinement statistics r.m.s. deviat,ions

Target

from ideal values

valuet

Distance restraints (A) Bond distance Angle distance Planar l-4 distance Plane restraints (A) Chiral centre restraints (A3) Non-bonded contact restraint.s (A) Single torsion Multiple torsion

Possible hydrogen bond Torsion angle restraints (“) Planes Staggered Orthonormal Individual, isotropic thermal factor restraints (A2) Main-chain bond Main-chain angle Side-chain bond Side-chain angle Number

0.017 0.051

0430 0~0.50

04)5l

0.060

0.007 0.140

0.020 0.150

0.266 0.384 0.333

0.500 0.500 0~500

1.:3 26.I) 41.2

34 154 “04

15.5 18.1 16.2 17.7

304 304 304 30.0

of protein atoms; 3792; number of solvent, atoms, 0; 2.4 A; number of reflections, 12,317; R-factor,

resolution, 6 to 23.2%:

TThese target values are the input estimated standard deviations that determine the relative weights of the corresponding restraints.

of the model at 2.4 A resolution

At this stage all the 3792 atoms of the dimer were allowed to move independently, without, any restraints arising from local symmetry. Solvent molecules were not yet included. van der Waals’ restraints and torsion restraints were applied, but’ only weakly. The resolution was increased from 2-8 A at cycle 1 to 2.4 A at cycle 7. At. first only an overall B-factor was refined, but’ from cycle 9 onwards individual temperature factors were allowed to vary. The start values of these B-factors were the individual H-factors as obtained from the 2.8 A refinement: however. t,he individual temperature factors of the atoms of the uncertain regions, as established in the previous section. were set arbitrarily to 60 A2. Eventually the R-factor dropped from 36% (cycle 1) to 23.:!~,(cycle 27) for 12.317 reflections between 6A and 2.4 -4. Table 2 summarizes the important statistics at this stage of the refinement. The geometry of the model is good, except for some non-bonded distances. Large fluctuations do occur in the individual B-factors, which vary from 2 A2 to approximately 90 A2. These, however, have been very weakly restrained; therefore, poor parts of the model will contribute less to the calculated structure fartors and phases.

Figure 7. Electron density maps near Cys14-AsnlS-Glyl6

3. Results (a) The current m.odel procedure molecular replacement The (Rossmann, 1972) smoothly led us to the correct orientation and position of the chicken TPIase model in the gTPIase unit cell. The current electron density map (Fig. 7(c) and (d)) is of good quality. Nevertheless, some parts of the polypeptide chain are in regions of low electron density, as indicated in Figure 8, in which the average electron density for each residue, calculated at t’he positions of the N, c” and C main-chain atoms is plotted for subunits 1 and 2. Especially in the regions 31-37, 151-169 and 220-223, the main-chain electron density near the atom positions is low. Tn these regions. the current model is not well-defined. These parts of the structure tend to have high R-factors (Fig. 8). Also the cp-~+G angles of these residues regions of the often map in the unallowed Ramachandran plot (Figs 8 and 9).

and Arg99-AlalOO. The model co-ordinates are from the model. (a) and (b) m(2jFJ --IF& exp (ia,). Th ese Sim-weighted maps (contoured at 0.6 times the r.m.s. value of the map density) have been calculated using F, and a, (7 to 3.2 A) from a pruned chicken TPIase model (after the rigid body refinements), before any model building had been done. All the side-chain atoms beyond CB had been removed from this model, nvertheless there is clearly density present for the side-chains of Asn15 and Arg99. (c) and (d) m(2(F,(--]F,I) exp(iu,). Th ese Sim-weighted electron density maps (contoured at 0.7 times the r.m.s. value of the map density) have been calculated from the current model using data between 6 A and 2.5 A. The positions of the side-chain atoms of Asn15 and Ag99, as well as the carbonyl oxygen main-chain atoms are well-defined in this map.

current

116

R. K. Wierenga

60

c I.1

cl

et al.

II

IO

20

30

40

50

60

70

60

SO

100

110

I20

Residue

I30

I40

160

160

170

I60

190

200

210

220

230

240

260

140

I50

160

170

160

190

200

210

220

230

240

2

numbw

160

60

-20

L 0

IO

20

30

40

50

60

70

60

SO

100

110

120

Residue

130 number

(b) Figure 8. Main-chain electron density value per residue, calculated as the average at’ the positions of the N, C” and C atoms. The electron density value associated with each atom is an average value, calculated from the 8 grid-points nearest to that particular atom. Residues for which the individual temperature factor of either N, C” or C is higher than 60 A2 are marked (V) on a line at level zero. Residues with unallowed rp-$ angles are marked (A) on a line at level 40. Protein regions with low electron density are often associated with high B-factors and unallowed cp-$ angles. (a) Subunit 1; (b) subunit 2.

(b) The overall structure The overall folding of the two gTPIase polypeptide chains of the complete dimer is shown in Figure 10. The dimer is approximately 80 A long and 50 A wide. The structure is very similar to the chicken TPIase structure. After superposition of the C” atoms of the chicken enzyme subunit 1 on top of the gTPIase subunit 1, there is a 0.6 A r.m.s. difference between the 238 common C” atoms. When subunits 1 of gTPIase and chicken TPIase

are optimally superimposed, the two subunits 2 overlap with an r.m.s. difference of 1.7 L%for the 238 common C” atoms, indicating a possible difference in relative position of the two subunits in the two dimers. However, at this stage it is too early to decide if the association of the two subunits is significantly different in the TPIase dimer compared with the chicken dimer. Nevertheless, a substantial number of amino acid replacements do occur at the interface between the two subunits, as will be discussed below.

Glycosomal Triosephosphate Ismerase Structure

117

180

60

-60

-180 -180

-120

-60

0

120

60

180

Phi (a) 180

60

-60

. . -180

-120

-60

0

60

120

180

Phi (b)

Figure 9. The Ramachandran plot. The cp-$ angles of the non-glycine residues are indicated by (a), whereas glycine cp-tJ angles are given by (a). The areas within the continuous lines define the fully allowed conformations for z(N, C”, C) = 110”. The broken lines show the regions obtained by relaxing the van der Waals’ contact constraints as well as by allowing T to increase to 115” (Ramachandran & Sasisekharan, 1968). (a) Subunit 1; (b) subunit 2. The 238 C” atoms of subunit 1 (gTPIase) and which were used for the subunit 2 (gTPIase), previous comparisons, superimpose with a r.m.s. difference of 0.4 d, indicating excellent agreement with the local 2-fold symmetry. Therefore, significant structural differences between subunit 1 and subunit 2 of gTPIase cannot be given with any certainty.

(c) The dimer interface From studies established that

with only

rabbit TPIase it has been the complete dimer is fully

active (Waley, 1973). This can be understood from the structure, because residues of the loop 72 to 79 of subunit 2 extend into the active site of subunit 1. In particular, Thr75 (labeled as Thr375 in Fig. 12) points towards the active site. Surprisingly a oneresidue deletion occurs in this loop, just, before residue Gly72 (Fig. 2). As can be seen from Table 3, substantial sequence differences are observed for the interface residues. Altogether in gTPIase only 15 out of 27 interface residues are identical (55%) with the chicken residues, which is equal to the overall percentage of identical residues for the

R. K. Wierenga

118

et al.

Figure 10. The complete dimer (similar view as in Fig. 3). The structure of the upper subunit (subunit 1) is visualized by drawing lines between the C” atoms of the residues, while the folding pattern of subunit 2 is indicated by the positions of the 4 main chain atoms, X’, C”, C and 0; of each residue. The conformations of the active site residues Lys13, His95 and Glul67 of subunit 1 are exnlicitlv shown. The structure of the unique gTPIase peptide AlalOO-TyrlOlTyrlO2 is shown by thick lines. TPIase and chicken enzyme sequences. Much higher conservation of sequence identity within this set of interface residues is observed when comparing, for example, the human versus chicken (78% identity) or the yeast versus chicken sequence (70% identity). The consequences of these intriguing

Table 3 Interface

residues Accessible

Accessible

Residue Amino acid surface (A2) surface (A’) number in Amino acid in chicken in gTPIase in gTPIase in gTPIase monomert gTPIase TPIase dimert

Asn (:I> Ser Gin

Lys Met As.11 Gig Asp Lys

65 114 105 39 51 110

44 0 55 25 22 53

44 45 46 47 48

Thr Phr Val His Leu

Pro SW Ile T,W LPU

2x 70 117 1x

I 8 1 0

i

0

65

Gin

Ghl

46

1

72 73 74 75 76 77 78 79

Uly Ala Phe Thr my Glu Val Ser

Cily

51

Ala

86

Phe Thr Gly GlU Ile Ser

128 152 67 75 30 4

28 50 38 0 7 6 1 0

82 85 86

Ile Asp Phe

Met Asp Ile

87 134 110

8 56 21

97 98 101 102

GIU Arg TY~ Tyr

GlU *rg Val Phr

51 66 175 147

23 8 142 69

13 14 15 16 17 18

Lys CYS

Residues that have atoms closer than 3.5 A to atoms of the other subunit are included in this Table. t The accessibility has been calculated with the DSSP package (Kabsch & Sander. 1983).

interface differences on the properties of the gTPIase dimer are difficult to predict. The interface of gTPIase seems to be somewhat more hydrophilic than that of the chicken enzyme (Table 3), due to substitutions in gTPIase such as Thr44 (instead of Pro), His47 (Tyr), TyrlOl (Val) and Tyr102 (Phe). At this interface a potential salt bridge, unique to gTPIase, is formed between a histidine and an aspartic acid residue. In the current model the distance between NE2 (His47: subunit 1) and ODl (Asp%; subunit 2) is 3.9 a and the distance from ODl (Asp%: subunit 1) to NE2 (His47; subunit 2) is 4.0 .&. This unique gTPIase His-Asp interaction might be the cause of the different kinetic properties of gTPIase at high pH (pH > 6.4) and low salt conditions, i.e. ionic strength below 0.1 (Lambeir et al., 1987). Under these conditions the K, of gTPIase is significantly higher, whereas V,,, only slightly increases in comparison with the values at ionic strength greater than 0.1. These properties are pH-dependent with a pK of 6.4. Above a pH of about 7 His47 becomes deprotonated, therefore the salt bridge is no longer formed and a less-stable dimer is obtained. Apparently this stabilizing interaction is only essential at low ionic strength, which may be due to the fact that under such conditions less electrostatic screening occurs; consequently the repulsion between the two subunits, with a charge of +6 each. will increase, which will reduce the dimer stability. (d) The “hot spots” The overall charge of the gTPIase subunit, which is +6, is much higher than the overall charges of the other TPIase sequences (Swinkels et al., 1986), where it varies from 0 (rabbit) to -4 (yeast); the overall charge of chicken enzyme = - 1. The extra positive charge of gTPIase is due to the fact that the number of negatively charged residues is significantly less compared with the &her TPIase sequences.

Glycosomal Triosephosphate Isomerase Structure

119

Figure 11. The 4 hot spots (similar view as in Fig. 3, but slightly rotated around a vertical axis). The positions of hot spot 1 (HS-1) and hot spot’ 2 (HS-2) on each subunit are indicated by dotted van der Waals’ surfaces of Lys153 and Lys155 (H&l), and Lys217 and Arg220 (HS-2). The positions of the side-chains of the active-site residues of subunit 1, Lysl3, His95 and Glu167 are explicitly shown. The 4 hot spots are at the extreme edges of the dimer. The distance between HS-1 (subunit 1) and HS-2 (subunit 2) is 39 A. The distances between any other set of 2 hot spots is greater than 45 A. These distances are calculated between the average positions of the CG atoms of the 2 dotted side-chains of each hot spot. On the surface of gTPIase two unique clusters of positive charges (hot spots) occur: Lys152-Lys153Leu154-Lys155-Lys156 (hot spot 1) and Lys217Asn218-Ala219-Arg220 (hot spot 2) Wierenga et al., 1987). The charge of these two peptides is +6, whereas in chicken enzyme the homologous peptides have a charge of + 1. Therefore the extra positive charge of gTPIase compared with other TPTases is largely concentrated in these two hot spots. The other glycosomal enzymes, except for phosphoglucose isomerase, also have an excess of positive charge when compared with their cytosolic counterparts as can be deduced from their high p1 values (Misset et al., 1986). A function for these extra positive charges has not been definitely established, but, it has been found also that on the surface of other glycosomal enzymes unique positively charged clusters (hot spots) occur

(Wierenga et al., 1987). Subsequently, it has been postulated that these hot spots are essential for the of the glycosomal enzymes import into the glycosome (Wierenga et al., 1987). The exact conformations of the hot spot peptides of gTPIase are uncertain at this stage of the refinement, because the electron density of the corresponding regions of the Fourier map is rather low, as is shown in Figure 8. Further model building and refinement may clarify these parts of the map but, alternatively, these residues might have intrinsic flexibility, which in some way could be related to their putative function for import into the glycosomes. The position of these hot spot peptides on the surface of the dimer is shown in Figure 11. Hot spot 1 is at the end of helix a5, while hot spot, 2 is near the beginning of helix a7. As can be seen from Figure 11 they are at the extreme edges of the

Figure 12. The geometry of the active site of subunit 1 of trypanosomal (thin lines, plus labels) and chicken TPIase (thick lines) (similar view to Fig. 3). The 238 common C” atoms of subunit 1 of gTPIase and chicken enzyme were used for this superposition (r.m.s. = 0.6 A). The position of all atoms of a residue are only drawn for a few important amino acids. otherwise only the main-chain atoms are shown. Important catalytic functions have been ascribed to Lysl3, His95 and Glu167. The substrate molecule is believed to be bound between the active-site helix (95-102) and the phosphate binding helix (234-239) (Alber et al., 1981). Thr174 is in the middle of the flexible loop. The position of the interface loop (72-79) of subunit 2 is also shown (bottom left). it starts at Gly72 (labeled as Gly372); Thr75 (labeled as Thr375) of this loop points into the active site of subunit 1.

120

R. K. Wierenga et al.

molecule. The distance between hot spot 1 in subunit 1 and hot spot 2 in subunit 1 is 39 A; the distance between any other set of two hot spots in the dimer is greater than 45 A (Fig. 11). (e) The active site The active site residues are much more conserved than the interface residues. Out of 40 residues having atoms within 10 A from NE2 of His95 34 ( = 85%) are identical in gTPIase and chicken enzyme. In Figure 12 it can be seen that the three active site residues of subunit 1, Lysl3, His95 and Glu167, take up very similar positions in the gTPIase and chicken structures. The side-chain of Thr75 (labeled as Thr375 in Fig. 12) of the interface loop of subunit 2 also points into the active site of subunit 1 (Fig. 12). This interface loop is shorter in gTPIase (Fig. 2), which explains the non-exact superposition near Gly72 (labeled as Gly372 in Fig. 12). At a somewhat greater distance from the cavity, sequence differences cause active-site structural differences that are relevant to our investigations. (1) The chicken TPIase sequence HislOO-VallOl-Phe102 in the second turn of the active site helix (between #?5 and ~5; Fig. 3) is replaced by the sequence AlalOO-TyrlOl-Tyr102 in gTPIase (Fig. 10). The distance between the hydroxyl oxygen atom of TyrlOl and NE2 of His95, taken as a reference atom in the active site, is 10.0 A. The Ala-Tyr-Tyr sequence is unique for gTPIase; in human TPIase as well as in other vertebrate TPIases the sequence His-Val-Phe is observed. From the point of view of rational drug design this is an important observation, because if a compound could be found that binds specifically to this gTPIase peptide, it might be modified in such a way that, after binding, it will selectively inhibit the catalytic process in the enzyme. (2) At a distance of approximately 22 A from NE2 of His95 charged hot spot 2 peptide the positively (subunit 1) occurs. The other clusters of positive charges are at about 23 A (hot spot 1, subunit 2), about 31 A (hot spot 1, subunit 2) and about 50 A (hot spot 2, subunit 2) away from His95. Molecules with high affinity for such a positively charged peptide can also be the starting point for selective inhibitors of gTPIase.

Table 4 Net charge distribution r (AH CL.5 5-10 10-15 15-20 20-25

around NE2 (His95)

gTPIase 0 +1 -2 -1 +10

Chicken

TPIase 0 +1 -2 -4 +1

For these calculations the residues Glu and Asp were given a charge of - 1, whereas the charges of Lys and Arg were set at + 1. Distances were calculated between NE2 (His95) and the CG atoms of these charged residues. t r defines the inner and outer radius of a spherical shell around NE2 (His95).

An analysis of the overall charges within a sphere with a radius of 25 A centered at NE2 of the active site residue His95 shows that in gTPIase this net charge is +8, whereas in chicken enzyme it is -4. Interestingly, as is indicated in Table 4, these “sphere charges” are identical up to 15 A away from His95. However, beyond 15 A the difference is +3 in the shell between 15 A and 20 A, and becomes as high as + 9 between 20 A and 25 A. Apparently the charged residues are distributed over the surface of gTPIase in such a way that large net charge differences do not occur near the active site. This probably explains why the kinetic properties of gTPIase and homologous TPIases are so similar, at least at ionic strengths greater than 0.1 (Lambeir et al., 1987), in spite of the fact that the negatively charged substrate molecule interacts with enzymes that differ greatly in overall charge.

4. Discussion and Conclusions The polypeptide chain of gTPIase is folded into the regular TPIase barrel structure, as expected from the 52% identity in amino acid sequence when comparing gTPIase with chicken enzyme. The residues at the active site are much more conserved. This agrees with the observation that the kinetic parameters of gTPIase and chicken enzyme are very similar at least at ionic strengths above 0.1 (Lambeir et al., 1987). Apparently the seven extra positive charges present in the gTPIase subunit compared with chicken TPIase do not affect the kinetic properties. This can be compared with the pH dependency of a subtilisin mutant in which an Asp-to-Ser mutation was introduced. In this case a change of just one surface charge, at a distance of 14 to 15 A from the active site of subtilisin did have a significant effect on the pK of the active-site histidine (Thomas et al., 1985), although the difference was only 0.3 of a pH unit at an ionic strength of 0.1. In contrast to the highly conserved active site, a much larger variation in the residues at the intersubunit contact area is observed. These differences could affect the stability of the dimer. It is not known if the dissociation constant of the gTPIase dimer deviates significantly from those of other TPIases, but the higher thermolabilit’y of gTPIase at low and high pH (Lambeir et al.: 1987) could be due to a larger dissociation constant. Such a reasoning is in line with arguments put forward by Daar et al. (1986) to explain the higher thermolability of a mutant of human TPIase. In this mutant Glu104 is changed into an Asp. Glu104 forms a salt bridge with Arg98, which interacts with Glu77 of the other subunit. The higher thermolability of this mutant proves that very subtle interactions are important for the overall stability of the folded protein. One can envisage that such an interference with stabilizing interactions, in this example caused by mutating a side-chain, might also be triggered by a compound that binds specifically to an interface side-chain.

Glycosomal Triosephosphate Isomerase Structure From the point of view of rational drug design such an approach may be important, as it offers an alternative approach to interfering selectively with the functioning of gTPIase, while leaving human TPIase unaffected. From the drug design point of view it is unfortunate that the active sites of gTPIase and human enzyme are so similar. Nevertheless, somewhat further away from the catalytic center, two unique gTPIase surface structures have been pointed out. (1) The Ala-Tyr-Tyr sequence of residues 100 to 102 in the active site helix. (2) The Lys-Asn-Ala-Arg sequence of hot spot 2. These unique features may be potential starting points for the development of selective inhibitors of the catalytic activity of gTPIase. Further refinement, especially against higherresolution data, should establish very accurately the atomic structure of gTPIase together with its solvent surrounding molecules. Subsequently, modeling studies and crystallographic binding experiments can guide us, in a cyclic process, towards the discovery of compounds with a high affinity for gTPIase, which inhibit the proper functioning of this enzyme in the sleeping-sickness parasite, while leaving human TPIase unaffected. Such studies are being initiated. It is a pleasure to thank Drs F. Opperdoes, 0. Misset and A. Lambeir (ICP, Brussels) for supplying us with the purified glycosomal TPIase; Drs P. Borst, K. Osinga $ B. Swinkels (NKI, Amsterdam) for making the gTPIase sequence available prior to publication; Drs D. C. Phillips and P. Artymiuk (Oxford) for giving us refined chicken TPIase co-ordinates; and Drs B. Dijkstra, R. Read, H. Schreuder and A. Volbeda for various discussions and assistance with the use of the BIOMOL programs of the Groningen protein crystallography group. Award of computer time on the Cyber 205 in Amsterdam by the “Werkgroep Supercomputers” is gratefully acknowledged. This research was supported by the UNDP/World Bank/WHO Special Programme for Research and Training in Tropical Diseases; by the Dutch Foundation of Chemical Research (SON) with financial aid from the Dutch Organization for the Advancement of Pure Research (ZWO), and by a special grant from the University of Groningen.

References Agarwal, R. C. (1978). Acta Cryatallogr. sect. A, 34, 791-809. Alagona, G., Ghio, C. & Kollman, P. A. (1986). J. Mol. Biol. 191, 23-27. Alber, T., Banner, D. W., Bloomer, A. C., Petsko, G. A., Phillips, D. C., Rivers, P. S. & Wilson, I. A. (1981). Phil. Trans. Roy. Sot. Ser. B, 293, 159-171. Banner, B. W., Bloomer, A. C., Petsko, G. A., Phillips, D. C., Pogson, C. I., Wilson, I.A., Corran, P. H., Furth, A. J., Milman, J. D., Offord, R. E., Priddle, J. D. & Waley, S. G. (1975). Nature (London), 255, 609-614. Brandenburg, N. P., Dempsey, S., Dijkstra, B. W., Lijk, L. J. & HOI, W. G. J. (1981). J. A&. @JStdtOgT. 14, 274-279. Crowther, R. A. (1972). In The Molecular Replacement Method (Rossmann, M. G., ed.), pp. 173-178, Gordon and Breach, New York. Edited

121

Crowther, R. A. & Blow, D. M. (1967). Acta Prystallogr. 23, 544-548. Daar, I. A., Artymiuk, P. J., Phillips, D. C. & Maquat, L. E. (1986). PTOC. Nat. Acad. Sci., V.S.A. 83, 7903-7907. Dodson, E. J., Tsaacs, N. W. & Rollett, J. S. (1976). Acta Crystabgr. sect. A, 32, 311-315. Goodford, P. J. (1984). J. Med. Chem. 27, 557.--564. Hart, D. T., Baudhuin, P., Opperdoes, F. R. & de Duve, C. (1987). EMBO J. 6, 1403-1411. Hendrickson, W.A. (1985). In Methods in Enzymology (Wyckoff, H. W., Hirs, C. H. W. & Timasheff, S. N., eds), vol. 11.5,pp. 252-270, Academic Press, New York. Hol, W. G. J. (1986). Angewundte Chemie, 25, 767-778. Kabsch, W. & Sander, C. (1983). Biopolymers, 22, 2577-2637. Knowles, J. R. & Albery, W. J. (1977). Act. (‘hem. Res. 10, 105-111. Lambeir, A.-M., Opperdoes, F. R. & Wierenga, R. K. (1987). EUT. J. B&hem. In the press. Misset, O., Bos, 0. J. M. & Opperdoes, F. R. (1986). EUT. J. Biochem. 157, 441-453. Molyneux, D. H. & Ashford, R. W. (1983). The Biology of Trypanosomes and Leishmania, Parasites of Man and Domestic Animals, Taylor & Francis, London. Opperdoes, F. R. (1985). BT. Med. Bull. 41, 130-136. Opperdoes, F. R., Baudhuin, P., Coppens, I., De Roe, C., Edwards, S. W., Weijers, P. J. & Misset, 0. (1984). J. Cell Biol. 98, 1178-1184. Raines, R. T., Sutton, E. L., Straus, D. R., Gilbert, W. & Knowles, J. R. (1986). Biochemislry, 25, 7142-7154. Ramachandran, G. N. & Sasisekharan, V. (1968). Advan. Protein Chem. 23, 283438. Read, R. J., Wierenga, R. K., Groendijk, H., Lambeir, A.-M., Opperdoes, F. R. & Hol. W. G. J. (1987). J. Mol. Biol. 194, 573-575. Richardson, J. S. (1981). Advan. Protein Chem. 34, 167-339. Rose, I. A. (1962). Brookhaven Symp. Biol. 15, 293-309. Rossmann, M. G. (1972). Editor of The MoZecuZar Replacement Method, Gordon & Breach, New York. Rossmann, M. G. & Blow, D. M. (1962). Acta (~TySkdlOgT. 15, 24-31. Sim, G. A. (1960). Acta Crystallogr. 13, 511-512. Straus, D. & Gilbert, W. (1985). PTOC. Nut. Acad. Sci., U.S.A. 82, 2014-2018. Straus, D., Raines, R., Kawashima, E., Knowles, J. R. & Gilbert, W. (1985). PTOC.Nat. Acad. Sci., l1.S.A. 82, 2272-2276. Swinkels, B. W., Gibson, W. C., Osinga, K. A.: Kramer, R., Veeneman, G. H., Van Boom, J. H. & Borst, P. (1986). EMBO J. 5, 1291-1298. Sussman, J. L. (1985). In Methods in Enzymology (Wyckoff, H. W., Hirs, C. H. W. C Timasheff, S. N., eds), vol. 115, pp. 271-303, Academic Press, New York, Thomas, P. G., Russell, A. J. & Fersht, A. R. (1985). Nature (L&on), 318, 375-376. Waley, S. G. (1973). Biochem. J. 135, 165-172. Wang, C. C. (1984). J. Med. Chem. 27, l-9. Wierenga, R. K., Hol, W. G. J., Misset, 0. & Opperdoes, F. R. (1984). J. Mo2. Biol. 178, 487490. Wierenga, R. K., Swinkels, B. W., Michels, P. A. M., Osinga, K., Misset, O., Van Beeumen, J., Gibson, W. C., Postma, J. P. M., Borst, P., Opperdoes, F. R. & Hol, W. G. J. (1987). EMBO J. 6, 215-221. XRAY System (1976). Dutch version of the XRAY system. Techn. Rep. TR-192. Computer Science Center, University of Maryland, College Park, Maryland, U.S.A.

by R. Huber