THE JOURNAL OF BIOLOGICAL CHEMISTRY Val. 264, No. 30, Issue of October 25, pp. 17681-17690, 1989 C 1989 by T h e American Societ) for Biochemistry and Molecular Biology. Inc. Printed in U. S. A.

Refined Atolnic Model of Glutamine Synthetase at 3.5 A Resolution* (Received for publication, June 5, 1989)

Mason M.. YamashitaS;, Robert J. Almassyg, Cheryl A. Jansong, Duilio CascioS;, and David EisenbergSll From the iMolecular Biology Institute and Department of Chemistry and Biochemistry, Uniuersity of California, Los Angeld:?, California 900%

An atomic model of 43,692 non-hydrogen atoms has been determined for the 12-subunit enzyme glutamine synthetase from Salmonella typhimurium, by methods of x-ray diffraction including restrained least-squares atomic gefinement against 65,223 unique reflections. At 3.5 A resolution the crystallographic R-factor (on 2~ data) is 25.8%. A.s reported earlier for the unrefined structure, the 12 subunits are arranged in two layers of six; at the interface of pairs of subunits within each layer, cylindrical active sites are formed by six anti- parallel @ strands c!ontributed by one subunit and two strands by the neighboring subunit. This interpreta- tion of the electron density map has now been sup- ported by comparison with glutamine synthetase from Escherichia coli by the Fourier difference method. Each active site cylinder holds two Mn2+ ions, with each ion having as ligands three protein side chains and two water molecules (one water shared by both metals), as well as, a histidyl side chain just beyond liganding distance. The protein ligands to Mn2+ 469 are Glu-131, Glu-212, and Glu-220; those to Mn2+ 470 are Glu-129, His-2169, and Glu-357. The two layers of subunits are held together largely by the apolar COOH terminus, a helical thong, which inserts into a hydro- phobic pocket formed by two neighboring subunits on the opposite ring. Also between layers, there is a hy- drogen-bonded sheet interaction, as there is between subunits within a :ring, but hydrophobic interactions account for most of the intersubunit stability. The cen- tral loop, which extends into the central aqueous chan- nel, is subject to attack by at least five enzymes and is discussed as an enzyme “passive site.”

Glutamine synthetase from Salmonella typhimurium is a large (relative molecular mass M, = 12 X 51,628) enzyme that serves as the central. element in the regulation of cellular nitrogen metabolism (Ginsburg, 1972; Ginsburg and Stadt- man, 1973; Stadtman and Ginsburg, 1974; Reitzer and Ma- gasanik, 1987). Electron micrographs of the closely related E. coli glutamine synthetase revealed 12 polypeptide chains ar- ranged in two rings o f six monomers with 622 point group symmetry (Valentine et al., 1968; Frey et al., 1975). Glutamine

* This work was supported by National Institutes of Health Grants GM 31299 and GM 51628 for general support and United States Public Health Service National Research Service Award Traineeship GM07184-14 (to M. Y.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “aduertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Pines Rd., La Jolla, CA 92037. Present address: Agouron Pharmaceuticals, Inc., 11025 N. Torrey

ll To whom correspondence should be addressed.

synthetase catalyzes the condensation of ammonia and glu- tamate with the aid of ATP to form glutamine (Levintow and Meister, 1954; Alberty, 1968). Glutamine in turn is the source of nitrogen in the biosynthesis of many metabolites, of which at least eight act as feedback inhibitors of glutamine synthe- tase. The sensitivity of glutamine synthetase to most of these inhibitors is accentuated by adenylylation of Tyr-397 (Sha- piro et al., 1967; Stadtman et al., 1968a, 1968b; Holzer et al., 1967; Holzer, 1969; Shapiro and Stadtman, 1970).

In earlier papers, we have reported the isolation and crys- tallization of glutamine synthetase (Janson et al., 1984), the inference of the amino acid sequence from the DNA sequence of the gene (Janson et al., 1986), and the determination of an unrefined atomic model with a crystallographic R-factor of 51.1%, by methods of protein crystallography (Almassy et al., 1986). Here we report the crystallographic refinement of this model. Whereas our preliminary report emphasized the struc- ture of the dodecamer (Almassy et al., 1986), this paper discusses the geometry of the active size and interatomic forces within and between monomers.

MATERIALS AND METHODS

Crystals and Initial Phase Determination-Completely unadenyl- ylated glutamine synthetase from S. typhimurium forms crystals of space group C2 and unit-cell dimensions a = 235.5 A, b = 134.5 A, c = 200.1 A, and p = 102.8” with one dodecameric molecule in the asymmetric unit (Janson et al., 1984). Data were collected with a multiwire area x-ray diffractometer (Hamlin, 1985) and phase deter- mination was by multiple isomorphous replacement coupled with real-space averaging and density modification, as described by Al- massy et al. (1986). The 3.5 A dataset contains 65,233 unique reflec- tions of a theoretical total of 76,889. For purposes of refinement, the data were scaled using a single scale factorbin, where a bin consists of reflections in a 5” scan range. After rejection of 393 outlying symmetry-related observations the merging R-factor from the inten- sities of the 175,640 observed reflections was 5.5%.

Atomic Refinement-The initial model was that described by Al- massy et al. (1986) built into 12-fold averaged electron density. The quality of this electron density is indicated by the low R-factor (9.7%) between the structure factors of its Fourier transform and the ob- served structure factors. This R-factor is significantly below the value (41%) expected for 12-fold averaging of random electron density (Rees, 1983).

The restrained least-squares refinement program of Hendrickson and Konnert (19811, as implemented by Rees et al. (1983), was used with derivatives (Jack and Levitt, 1978) calculated from a 12-fold averaged difference map. The 12-fold averaging was calculated with a local implementation of the Bricogne molecular replacement algo- rithm (Bricogne, 1974; Rossmann and Blow, 1963). Solvent was flattened in each cycle of refinement by replacing the density outside a molecular envelope by density equal to its mean value (Schevitz et aL, 1981; Wang, 19861.’ All programs have been implemented on a DEC VAX 8800 a t UCLA. The calculated shifts in coordinates and B-factors for one monomer were applied to all monomers, preserving the 622 non-crystallographic symmetry. Overlapping contacts he-

’ R. J. Almassy, manuscript in preparation.

17681

17682 Glutamine Synthetase tween monomers were removed during the four atomic model rebuilds described below. Each monomer has 3641 atoms X 3 positions, 468 residues X 1 B-factors, and two Mn'+ ions and 3 water molecules X 1 B-factor. The number of observations exceeds these 11,396 parame- ters of refinement by a factor of 5.7.

A total of 266 cycles of atomic refinement of the glutamine synthe- tase model were carried out, as summarized in Fig. 1. As explained below, after the initial 258 cycles, the model was restored to that of cycle 252 and eight further cycles were carried out with different B- factors; the cycle numbers for these last eight cycles are designated by primes. Drops in the R-factor a t cycles 70, 137, 210, 230, and 245 were associated with loosened restraints on bond lengths and angles and later tightened. Complete rebuilding of the glutamine synthetase model to improve its fit to the current electron density occurred a t cycles 130, 205, 235, and 252 using F,-F, and 2F,-F, maps calculated with model phases. A list of bad contacts between monomers was generated before each rebuild and care was taken to alleviate the overlapping contacts by rebuilding the model. By the fourth rebuild there were no bad contacts. The interactive graphics program FRODO (Jones, 1978) was used with an Evans and Sutherland PS330 and Digital microVAX I1 computer to adjust the model. During the final stage of refinement (cycles 205-260 and 252'-260') all (Y helices were restrained to maintain reasonable $/+ angles. In addition, three water molecules were placed into electron density around the two Mn2+ ions.

The locations of residues in the active site were examined using residue omit maps (Bhat and Cohen, 1984). All residues within 7 A of both Mn2+ ions, were deleted and a F,-F, map calculated. This density map contains minimal bias from the model near the active site, and it confirms the model in this vicinity. Special attention was paid to the side chains of Tyr-179 and Phe-180 which were shifted significantly by refinement. Before accepting the new positions, we examined several omit maps. One was produced after deleting from the model side chains of residues 174-183 and refining the model until convergence. All such maps fit better to the final model than the initial. Also Ser-266, a ligand to one Mn2+ ion in the initial model, was replaced by Glu-131.

The crystallographic R-factor decreased from 50.1 to 25.8% a t 3.5 A resolution over the course of the refinement. The final model deviates from ideal bond lengths by 0.027 A and from ideal bond

angles by 4.2". Table I summarizes the target inverse square root of the weights used during the refinement and the final r.m.s. deviation from ideal parameter values.

Refinement of Temperature (B)-Factors-Two B-factors/residue were refined starting a t cycle 205. The actual procedure was to

TABLE I Summary of parameters in the least-squares refinement

Distance 1-2 is between two atoms sharing a bond. Distance 1-3 is between two atoms at ends of a shared angle. Distance 1-4 is between two atoms a t ends of a shared dihedral.

Target a

R-factor Distances

1-2 0.020 1-3 0.020 1-4 0.040

Planes 0.020

Chiral volumes 0.150

Non-bonded contacts 1-4 0.5 Others 0.5

Angles Prespecified 12.0 Planar 3.0 Staggered 15.0 Orthonormal 20.0

Temperature factor Main chain 1-2 1.0 Main chain 1-3 2.0 Side chain 1-2 1.0 Side chain 1-3 2.0

- atomlc rebulld -----loosen geometry

I

9 0.4 Lz

1 i

I I :

Final value

0.258

0.027 0.040 0.072 0.025

0.382

0.295 0.413

15.5 4.3

30.6 35.2

0.315 0.522 0.684 1.180

0.2 I I I I I I J 0 50 100 I50 200 250

Cycle

FIG. 1. Progress of the atomic refinement, as monitored by the R-factor, Z(lF.1 - lFcl)/2 IF.1, where F. and F, are* the observed structure factors and structure factors calculated from the model. For all cycles, 10-3.5 A resolution data were used. At the start of refinement, the R-factor dropped dramatically from its initial value of 51% with little effect on the bond lengths and angles, but in cycles 50-70, refinement essentially stopped. To encourage further refinement, restraints on the bond lengths and angles were loosened a t cycle 70 and gradually reapplied starting at cycle 80, leading to a modest decrease in R-factor. This procedure of loosening and tightening of geometry restraints was repeated a t cycles 137, 210, 230, and 245 and accounts for the gradual increases in R-factor between cycles 150 and 200 and between cycles 215 and 225. A summary of deviation of bond lengths and bond angles from ideal values is in Table I. Complete rebuilding of the glutamine synthetase model to improve its fit to the current electron density occurred a t cycles 130, 205, 235, and 252. The large jump in the R- factor at cycle 235 was due to restoration of geometry to near ideality for a-helices and p strands. The R-factor dropped after a single cycle of refinement. After cycle 258, the validity of the refinement was checked by restoring the model to cycle 252, using a constant B-factor of 25 A'. The R-factor increased to 27.6% (dashed line). The B- factors were then re-refined as described in the text, with cycle numbers denoted by primes. The final R-factor after eight cycles was 25.8%.

Glutamine Synthetase 17683

calculate B-factor shifts for each atom of the residue and to then average a B factor for side chain and main chain before the next cycle. Also all B-factors were constrained to remain between 2 A’ and 50 A>. At cycle 258, examination of the B-factors revealed a highly biomodal distribution, in which the upper or lower constraint limits were met by a third of the values. This distribution raised questions about the validity of refinement, and in particular whether coordinate errors might be masked by erroneous B-factors.

To check the validity of refinement, the model was restored to cycle 252 (dashed line) and a constant B-factor of 25 A* was applied to all residues. This value, redetermined from a Wilson plot, is larger than the average value used in earlier cycles (10 A’). With application of the new single B-fa.ctor to the cycle 252 model, the R-factor increased to only 27.6 from 24.6%. This small increase suggests that the decline in R-factor from 50% for the initial model was from coordinate refinement, not the added degrees of freedom of residue B-factors. From cycle 2!j2’ to 260’, one B-factor/residue (along with coordinates) was refined with the R-factor dropped from 27.6 to 25.8%. The B-factors no longer showed a bimodal distribution, and seemed physically reasonable in that the low B-factors are generally correlated to a-helices and high B-factors to loops.

Atomic Solvation Energy-The estimates of atomic solvation en- ergy described under the “Discussion” were made according to the method of Eisenberg-and McLachlan (1986), except that the following values (in cal/mol-A*) were used for atomic solvation parameters: for carbon-containing groups, +18; for N- and 0-containing groups, -9; for 0--containing groups, -37; for N+-containing groups, -38; and for S-containing groups, -5.

Film Data for Difference Fourier Maps with Escherichia coli-X- ray diffraction film data recorded by the oscillation method on an Elliott GX-6 rotating anode from crystals of E. coli glutamine syn- thetase were used for l:!-fold averaged difference Fourier maps with data from S. typhirnuriurn glutamine synthetase. The E. coli gluta- mine synthetase crystals (Heidner et al., 1978) are isomorphous with the S. typhimurium glutamine synthetase crystals (Janson et al,, 1984). The data from E:. coli glutamine synthetase crystals to 6.4 A resolution were recorded on 50 films and consist of 12,400 reflections. The R-factor on intensities for symmetry related reflections is 7.7%. Films were scanned with the Kabsch spotfinder (Kabsch, 1977) and the data scaled with the Weissman Fourier-Bessel scaling procedure (Weissman, 1977).

Coordinates-The relined atomic parameters will be deposited in the Protein Data Bank, Chemistry Department, Brookhaven Na- tional Laboratory, Upton, NY, 11973.

RESULTS

Reliability of the Refinement-The resolution of 3.5 A is about the lower limit for which it is possible to trace the path of a polypeptide chain (e.g. Watson et al., 1970 for tosyl- elastase; Mavridis and Tulinsky, 1976 for 2-keto-3-deoxy-6- phosphogluconic aldolase; Suck et al., 1978 for Southern bean mosaic virus) and to refine an atomic model (e.g. Mercer et al., 1976, human sk.eletal muscle D-glyceraldehyde-3-phos- phate dehydrogenase). For this reason we sought indicators that refinement improved the model and that the protein fold is correct. Here we cite two quantitative measures of the improvement of the model by refinement and three qualitative indicators. The first quantitative measure is that during the course of refinement the phases calculated from the atomic model gradually approached the phases from the Fourier- inverted, 12-fold ave:raged electron density map (Almassy et al., 1986). This is decisive, because the 12-fold averaged elec- tron density map is (of high clarity. The r.m.s. difference in phase between the initial model and the map was 89.9”, nearly uncorrelated to the 12-fold averaged phases. At cycle 260’, the difference had been reduced to 34.8”, indicating a signifi- cant improvement in the model. Also indicating improvement was the increased height of identifiable features in the elec- tron density. For example, in the electron density computed from the phases of the initial model (cycle l), the Mn2+ 470 ion had a peak height of 4.0 on an arbitrary scale. With phases from the final model cycle 260’, the same feature had a peak

height of 129.0 on that same arbitrary scale. In the 12-fold averaged map, the Mn2+ 470 ion has height 134.0.

These two quantitative indicators reflect the improvement of the electron density during refinement. This improvement is evident from qualitative observations, three of which are the following: during the course of refinement it became evident that the unusual metal ligand Ser-266 of the initial model cycle 1 is not in fact a ligand of Mn2+ 469, but that the side chain of Glu-131 is in contact with the metal. Second, during the course of refinement three density peaks, now represented by water molecules, emerged adjacent to the two metal ions in reasonable ligand positions. Third, the refined model fit better into the 12-fold averaged electron density, than did the unrefined model. These three changes illustrate the improvement of the model that occurred during refine- ment.

Reliability of the Atomic Model-The atomic model fits the electron density well and reveals numerous features that are in accord with other types of experimental data, as discussed briefly in this section. Among these are 1) the nature and geometry of the metal ligands, 2) the reactivity of the NH2 and COOH termini, 3) the active site location and residues, and 4) the amino acid replacements compared with E. coli glutamine synthetase that are consistent with the electron density. All these points tend to confirm the overall structure presented above.

The glutamine synthetase metal ligands (glutamate and histidine; see below) are ligands found commonly in coordi- nation to metals. The stronger metal affinity of the nl-binding site relative to the n2 site is consistent with the net charge of the ligand sets. The nl divalent cation site has three Glu ligands (net charge of -3), whereas n2 has two Glu and one His ligand (about -1 net charge). Hence nl might be expected to bind cations more tightly. We note, however, that such electrostatic arguments do not always hold for protein-metal interactions. Also when nl binds Mn2+, a proton is released (Hunt and Ginsburg, 1981), which is consistent with an acidic ligand.

Our model has an exposed NH2 terminus and a buried COOH terminus; a result which matches experiments. The exposed NH2-terminal helix is visible in Fig. 2B at the very top and bottom of the glutamine synthetase dodecamer. Find- ings of DiIanni and Villafranca’ confirm the exposed position of the NH2 terminus because it is readily modified by reaction with pyridoxal 5’-phosphate. In contrast the buried COOH terminal helical thong is extremely resistant to carboxypep- tidase attack, which takes place only in the presence of sodium dodecyl sulfate (Ginsburg, 1972).

Amino acid residues found in the active site are compatible with binding and catalysis, and more decisively, the location of the active site between two subunits is supported by the earlier work of Maurizi and Ginsburg (1982) on the stability of glutamine synthetase complexed with the transition-state analog, methionine sulfoximine. These authors found that glutamine synthetase is resistant to dissociation into subunits by high concentrations of guanidine hydrochloride once it has been complexed with methionine sulfoximine. The intersub- unit stability provided by methionine sulfoximine can be rationalized by our model: the transition-state analog binds between subunits of the same ring. Still further evidence for the model comes from the work of Maurizi and Ginsburg (1982). They found that glutamine synthetase samples, treated first with substoichiometric amounts of methionine sulfoximine and then treated with EDTA and 5,5’-di- thiobis(nitrobenz0ic acid) yielded oligomers with 4, 6, 8, or 10

C. DiIanni and J. J. Villafranca, personal communication.

17684 Glutamine Synthetase

FIG. 2. The quaternary structure of glutamine synthetase, shown as line segments connecting the 468 se- quential a-carbon atoms for each of the six subunits of the top layer (panel A ) and for the six nearer sub- units of the two layers (panel B) . Each active site is indicated by a pair of spherical Mn2+ ions. In panel A , six cen- tral loops protrude into the central aqueous channel. The maximum dimen- sions of the moolecule including side chains are 103 A along the 6-fold axis and 143 A perpendicular to the 6-fold axis.

subunits, but not 2 subunits. This observation is consistent with glutamine synthetase up-down dimers that are strongly bound (by the helical thong and p loops) and which are bound to other dimers by the addition of methionine sulfoximine (at the active site between subunits of the same layer). Within the active site, the methionine sulfoximine molecule is held by amino acid residues of appropriate charge, as will be discussed el~ewhere.~

A stringent test of the model can be provided by examining the x-ray derived 12-fold averaged difference electron density of glutamine synthetase from S. typhimurium to that of E. coli. The test consists of matching peaks in the difference

3 M . M. Yamashita, R. J. Almassy, C. Gribskov, C. A. Janson, D. Cascio, and D. Eisenberg, in preparation.

density map with known differences in the sequences. From a first examination of the published sequences (Colombo and Villafranca, 1986; Janson e t al., 1986), it appears that there are 13 residue differences in the sequences of S. typhimurium and E. coli. However, we now believe there are only 10 differences and only one of these is a difference of more than two heavy ( i e . non-hydrogen) atoms. As explained in the following, this one substantial difference in sequence corre- sponds in our model to the largest peak in the difference map. This finding directly supports our model, but the test is clouded somewhat by the poor quality of the E. coli x-ray data.

The largest apparent difference in sequence is at residue 447, where the residue is Arg in E. coli and Pro in S. typhi- murium. However, the S. typhimurium map obtained from

Glutamine Synthetase 17685

c

FIG. 3. Ribbon representation of glutamine synthetase, looking into the active site, with the 6-fold axis roughly vertical. The COOH-terminal helical thong is at the bottom. Residue numbers are given at the start and end of segments of secondary structure. For clarity, the N-domain has been translated to the left; its junction to the COOH domain at residue 103 is shown by -. The eight p strands that form the active site are shown in a light shade. This diagram is based on a program by A. M. Lesk and K. D. Hardman (1982).

heavy atom and non-crystallographic-symmetry-averaged phases clearly showed electron density for a well-ordered Arg residue at this position (Almassy et al., 1986). Then subse- quently, this region of the Gln A gene of S. typhimurium was resequenced and was found to have an Arg at this position? Also, regions of the E. coli Gln A were resequenced, showing that residues 107 and 108 are the same as the S. typhimurium.5 This leaves the following 10 differences: 1) Ala-420 in S. typhimurium is a Glu in E. coli (a difference of four heavy atoms), 2) Pro-391 in S. typhimurium is an Ala in E. coli (a difference of two heavy atoms), 3) six other residues with a difference of one heavy atom, and 4) two other residues with no net difference in heavy atoms. Thus, the major difference is a t residue 420.

It is possible to compare the electron densities by the difference Fourier method because the two glutamine synthe- tase molecules crystallize in the same space group with vir- tually identical unit cell dimensions. We made this compari- son, using the present unadenylated S. typhimurium data set

' A. Kutsu, personal communication. J. J. Villafranca, personal communication.

and a 7 A resolution low state of adenylylation E. coli data set, Eo.

A 12-fold averaged difference map of fair quality was ob- tained from the Eo minus S. {yphimurium data. The largest difference peak (7.4 a) is 3.2 A from the C/3 atom of Ala-420 in S. typhimurium and represents the four atom difference 0.f replacing Ala with Glu. The next highest peak (5.1~) is 1.4 A from the C p atom of Ala-142 in S. typhimurium and repre- sents the one atom difference of replacing Ala with Ser. The next largest peak is -4.90, with many more peaks observed at just below this level. These smaller peaks could not be interpreted in terms of the differences in the protein sequence. Despite the noise of the map and its limited resolution, it lends direct support to our glutamine synthetase model.

Quaternury, Tertiary, and Secondary Structure-The 12 identical protein chains are arranged in two layers of six (Almassy et al., 1986) as shown in Fig. 2. Within one layer, pairs of subunits meet around the 6-fold axis at the active sites, marked by the pairs of Mn2+ ions in Fig. 2; These active sites are in cylindrical channels, roughly 10 A in $ameter that run paralldl to the 6-fold axis along the entire 50 A height of the subunit. The intersubunit contacts are mainly in three

17686 Glutamine Synthetase

A N-domain

B I"""""""""""""- 1

A

I"""""""" I

I I I

-1 I I I I I

5 6 4 1

I

From N Domain I

U .._.... B

"_"""""""""""" I I I """"""""

I I I

I

224 !i U i2 1

I

F"""

I I I I I 227

24: T I I I I I I I I I I I

1 1 401 407 414 424

I I

i G

._"" I I I I I I I I I I I I 1 I I 1 I I I I I

coo -

FIG. 4. Secondary structure in glutamine synthetase. a-Helices appear as rectangles. ,9 strands are shown by zig-zags. Panel A, the N domain. Panel B, the C domain. Panel C , a cartoon of the glutamine synthetase dodecamer, giving the notation for neighboring subunits. These identifiers appear on dotted lines at the sides of

Glutamine Synthetase 17687

regions, details of which are given below: the main contact of subunits within a layer is a p sheet interaction that is above and inside of the metal ions. By "above" we mean at larger absolute values of z, the coordinate along the 6-fold axis; by "inside" we mean closer to the 6-fold axis. The other two main contacts act between the upper and lower layers. These are: 1) the COOH terminus of each polypeptide chain forms a helical "thong" that inserts into the subunit below or above, and 2) a small four-stranded p sheet formed by the /3 loop of the subunit above and the (3 loop of the subunit below.

The tertiary fold of a single subunit is shown in Fig. 3, the view being into the active site. The subunit is formed by two folding domains: the smaller NHp-terminal domain (residues 1-102), which is linked covalently to the larger COOH-ter- minal domain (residues 103-468). The NH, domain (Fig. U), consisting of two a-helices and six strands of antiparallel @ sheet, sits mainly at the top of the molecule. The exception is the downward extension of the two-stranded /3 sheet whose strands are connected by the "Trp 57 loop." These two strands complete the cylindrical active site of the neighboring subunit. (This neighbor is the one clockwise to the subunit, when the molecule is viewed down the 6-fold axis. That is, the Trp-57 loop of subunit F of Fig. 4C completes the active site of subunit A).

The COOH-terminal domain (Fig. 4B) contains 13 helices and 9 strands of antiparallel (3 sheet. Two helices starting at Glu-402 and Ala-157 fit the density better as 310 rather than a-helices. The six /3 strands at the right-central region of Fig. 3 form the curved partial barrel structure which surrounds the Mn'+ ions of the active site. Supporting the partial barrel is a network of a-helices generally running perpendicular to the direction of the /3 strands.

Shown at the bottom of the C domain in Fig. 3 is the COOH-terminal helical thong, which inserts into the subunit of the opposite layer. J.ust above the thong are the two strands of the p loop (residues 137-154). These pair with two other /3 strands from the subunit below to form a small four-stranded sheet. Just to the right of this /3 loop is the central loop (residues 156-188) which protrudes into the central channel of the GS dodecamer. In the dodecamer, the view of Fig. 3 into the active site would be obscured partly by the NH2 domain of a neighboring subunit, completing the cylindrical barrel. This NH2 domain is from the subunit that is counter- clockwise to this subunit, when viewed down the 6-fold axis.

Metal ligands-Our crystals of glutamine synthetase were grown in the presence of Mn2+, and two strong spherical peaks of electron density, each with four protein ligands, are found in the active site cylinder. They are included in the glutamine synthetase model as Mn2' 469 and Mn2+ 470. Ion Mn2+ 469 probably corresponds to ion nl, previously identified as par- ticipating in the binding of glutamate (Villafranca et al., 1976a, 197613; Hunt and Ginsburg, 1980). This ion is more tightly bound than nL, which probably corresponds to our Mn2+ 470. Ion n2 is associated with ATP binding (Hunt et al., 1975). Other divalent cations are also known to bind to glutamine synthetase and to affect its kinetic properties, including M$+, Cap', .and Cd2+.

In our model, each R4n2+ ion is coordinated by three protein side chains and two water molecules, as listed in Table 11, in

TABLE I1 Ligand-metal distances in the active site

Distance/A

Mn2+ 469-Mn2'470 5.8 Mn2+ 469-Glu-212 OE2 1.9

Glu-131 OE2 2.2 Glu-220 OE2 2.3 Wat 471 0 2.6 Glu-220 OEl 2.7 Wat 472 0 3.3 Glu-212 OEl 3.3 Glu-131 OEl 3.5

Tyr-179 OEH" 5.0 His-210 NE2" 5.1

Mn2+ 470-Glu-129 OE2 1.6

His-269 ND1 2.1 Wat 473 0 3.0 Wat 472 0 3.0

Glu-357 OE2 2.0

Glu-357 OEl 3.3 Glu-129 OEl 3.3

His-271 NE2" 3.9 a Nearby protein atoms, not ligands.

a pattern that appears roughly octahcdral. The distance be- tween Mn2+ 469 and Mn2+ 470 is 5.8 A which corresponds to the spectroscopically determined metal-metal distance (Vil- lafranca et al., 1977). All six of these protein ligands are side chains of residues within the active site p sheet of the C domain (see Fig. 4B). The geometry of the ligands is shown in Fig. 5. The proiein side chains can be identified with little ambiguity at 3.5 A resolution, but the metal-ligand distances are uncertain by at least 0.4 A, and some of the glutamyl residues may contact the metal with both +oxygen atoms. Protein ligands to Mn2+ 469 are 3 glutamyl residues.*Also a nitrogen of the imidazole ring of His-210 is only 5.0 A away from Mn2+ 469, beyond the normal liganding distance. Two water molecules have been placed 2.6 and 3.3 A from Mn2+ 469 in regions of strong electron density. Protein ligands to Mn2+ 470 are 2 glutamyl residues and His-269. Again a histidyl tide chain (His-271) is just beyond liganding distance at 4.0 A. Two water molecules complete the coordination of Mn2+ 470, with one water being shared between MnZ+ 469 and Mn2+ 470. Their distances from Mn2+ 470 are both 3.0 A. Earlier Villafranca et al. (1976a) had concluded from NMR measure- ments that each Mn2+ ion is coordinated by two water mole- cules.

Helical Thong-The helical thong, represented in Fig. 6, is the third major intersubunit contact and like the (3 loops, it binds together subunits from the two layers. The COOH- terminal helix (residues 458-468) crosses from one layer of six subunits to the other and inserts into a pocket which is lined with apolar residues. The pocket for the thong from subunit A of Fig. 4C is created by residues from two other subunits (subunits G and L of Fig. 4C).

The helical thong and hydrophobic pocket consist of 77 residues of which 37 are apolar. Most of these apolar side

protein segments to indicate regions of intersubunit contacts. For example, in panel A, residues 30-34 have a hydrogen-bonded p interaction with His-209 in subunit B. In the C domain, the six strands of f l sheet forming the active site are numbered. They contain 6 residues that are metal ligands, indicated by ovals around the residue name for 1i::ands to the inner metal (Mn" 469) and by squares for ligands to the outer metal (Mn2+ 470). Notice the loop i.n the central lower region of panel B , having hydrogen bond interactions with subunit G. Also notice the COOH-terminal helical thong having int,eractions with subunits G and L. Segments of secondary structure were determined with the aid of CHARMM (Brooks et al., 1983). Residues were included within (Y or f l segments if either H-bonding distances or @/9 angles were within reasonable ranges.

17688 Glutamine Synthetase

A

+

i;: Jy9TYR

B

+

-U

+

SJ

FIG. 5. Stereo view for metal ions, ligands, and water molecules in the active site cylinder. The view is along the z axis from the bottom of a monomer looking up into the active site with the central channel to the left. Panel A shows metal ions as spheres and water molecules as crosses. The inner metal (MnZ' 469) is at the bottom of panel A , with ligands Glu-131, Glu-212, and Glu-220. The imidazole side chain of His-210 and the oxygen of Tyr-179 are somewhat too far (5 A) to be considered a metal ligand. The outer metal (MnZ+ 470) has ligands Glu-129, His-269, and Glu-357. The imidazole side chain of His-271 is somewhat too far (3.9 A) to be considered a metal ligand. The central water molecule is a ligand to both metals. Panel B shows a residue omit map where the metal ions, water molecules, and liganding residues have been deleted. The difference electron density (4u) is superimposed on the atomic model. Difference density is associated with the two Mn2+ ions, ligands, and one water molecule, to the upper left. Density for the other two water molecules appears at the 1.50 contour level.

Glutamine Synthetase 17689

FIG. 6. Stereo view of the COOH-terminal helical thong to illustrate the interaction of protein chains. Thlt view is the same as that of Fig. 3, from the central channel with 6-fold axis vertical. The thong of subunit A (A456-A468) enters from the front and top and extends downward (shown with heavy lines). To the right of the helical thong are the p loops of subunit A (A139-Al50) and subunit G (G139-Gl50) shown in medium lines. To the left of the thong is an a-helix from subunit G (G445-G457). To the bottom right a small section of the central 1.00~ of subunit L (L168-Ll71) completes the hydrophobic pocket.

chains are involved in direct contacts between the thong and pocket. Of these residues, 16% are proline and 10% are valine. In contrast, for the protein as a whole, the abundance for proline and valine residues are 6 and 7%, respectively. The unusual abundance of proline and valine at this contact may enhance its rigidity and cohesion.

DISCUSSION

Glutamine Synthettzse Enzyme Passive Site-Glutamine synthetase, as all enzymes, has an active site at which covalent chemistry is catalyzed. Unlike other enzymes, glutamine syn- thetase also seems to have a “passive site.” This is a local region of the molecule at which covalent chemical reactions are carried out on the enzyme. This passive site in glutamine synthetase is the central loop (residues 156-188) which can be seen in Fig. 2A as the segment of backbone that extends into the central aqueous cavity. This segment of polypeptide has been found to be susceptible to proteolysis by four secreted proteases, whereas no glutamine synthetase sites away from the central loop are attacked under the same mild conditions by these enzymes (Dautry-Versat et al., 1979; Lei et al., 1979; Monroe et al., 1985). The proteases are the V8 protease from Staphylococcus aureus which cleaves after Glu-165, trypsin which cleaves after Lys-169 and Lys-176, chymotrypsin which cleaves after Tyr-164 and after Tyr-179, and subtilisin which cleaves at an unknown site. Another type of covalent modi- fication of the central loop has been found recently by Moss et al. (1988). They discovered that Arg-172 is subject to ADP- ribosylation. All of these covalent modifications inhibit the enzyme. It should be noted that two other covalent modifi- cations to glutamine Yynthetase are known that do not take place at the passive site. These are adenylylation of Tyr-397 (Ginsburg, 1972), on the relatively mobile extended chain on the outside of the molecule, and proteolytic cleavage of oxi- dized glutamine synth.etase (Roseman and Levine, 1987) also at an exterior site.

What chemical or physical characteristics of the central

loop render it a passive site? Although the central loop is at the glutamine synthetase surface, it is not ea$ily accessible. I t extends into the central cavity, some 37 A below the top surface of the molecule, a position where it might not be expected to collide readily with enzymes. In a computer graph- ics experiment, we find there is just enough space in the central cavity for a single trypsin molecule to fit. But most of the rest of the glutamine synthetase surface is more accessible to diffusive collisions with enzymes. Perhaps it is the convex shape of the central loop that makes it a passive site. I t may be that this shape somehow renders amino acid side chains more accessible than they are at other sites on the glutamine synthetase surface. Alternatively, the central cavity may tem- porarily “trap” small enzymes while they vibrate and rotate, enhancing the chances of a productive collision.

Intersubunit Contacts-Each glutamine synthetase subunit (e.g. subunit A) contacts four neighboring subunits, subunits B, F, G, and L in Fig. 4C. Of these, the major contacts are with subunits B, F, and G. All three of these contacts involve hydrogen bonding and hydrophobic forces. The model sug- gests there are four intersubunit hydrogen bonds, and it may be possible to estimate the contribution to the free energy of stabilization by multiplying the number of bonds by an aver- age bond energy (in the range, -0.5 to 3.0 kcal/mol-of-hydro- gen bond, Fersht et al., 1985). This suggests that intersubunit hydrogen bonds stabilize the dodecamer by -2 to -12 kcal/ mol-of-monomer. Is it possible to estimate even roughly the free energy of stabilization of the hydrophobic forces?

A crude estimate of the free energy of stabilization provided by the hydrophobic interaction at the subunit interfaces is given by the method of Eisenberg and McLachlan (1986). In this method, the area of each type of atom is measured computationally in the dodecamer and then in a single mon- omer (assuming the structure is unchanged). The change in area is multiplied by an empirically derived atomic solvation parameter, and the contributions for all atoms are summed. This calculation suggests that the resulting “solvation free

17690 Glutamine Synthetase

energy change” for subunit A is -43 kcal/mol-of-subunit, providing more stabilization than hydrogen bond interaction. Of this energy, only 2 x -9 = -18 kcal/mol-of-subunit comes from contacts of subunit A with subunits B and F of the same ring, and only -1.4 kcal/mol-of-subunit comes from contact with subunit L diagonally related. In contrast, -25 kcal/mol- of-subunit comes from contact with subunit G, in the opposed subunit in other layer. Of this free energy, -18 kcal/mol-of- subunit comes from the apolar contacts with the helical thong, and about -7 kcal/mol-of-subunit from apolar contacts in- volving the 0 loops. We emphasize that these estimates are very crude, because, among other factors: 1) the areas of atoms depend somewhat on the accuracy of the model, 2) the computed free energy is based on the assumption that the structure of the subunit is unchanged when the intersubunit complex is formed, and 3) no estimate is included for the change in conformational entropy of the protein in forming the hydrophobic interaction. Despite these large uncertain- ties, the estimates suggest that the helical thong interacts strongly with the subunit in the opposite ring by hydrophobic interaction, and that intersubunit stabilization depends more on hydrophobic interactions than hydrogen bonds.

The strength of such inter-ring bonds compared to intra- ring bonds has been confirmed by electron microscopic studies of glutamine synthetase (Maurizi and Ginsburg, 1982; Has- cherneyer et al., 1982). The micrographs show partially disas- sembled glutamine synthetase molecules in which pairs of up- down subunits have been removed from glutamine synthetase dodecamers. This experiment is in accord with our estimate of solvation energy which suggests that the up-down forces (involving apolar helical thongs) provide much intersubunit stability.

Acknowledgments-We thank D. C. Rees and A. Ginsburg for helpful discussions, M. Wesson for computations of accessible surface areas, E. Goldsmith, R. Fenna, M. Harel, and J. Kobayashi for data collection and processing of x-ray films for E. coli glutamine synthe- tase, and M. C. Rayner for work on graphics.

REFERENCES Alberty, R. A. (1968) J. Bid. Chem. 2 4 3 , 1337-1343 Almassy, R. J., Janson, C. A., Hamlin, R., Xuong, N-H., and Eisen-

Bhat, T. N., and Cohen, G. H. (1984) J. Appl. Crystallogr. 17, 244-

Bricogne, G. (1974) Acta Crystallogr. A 3 0 , 395-405 Brooks, B. R., Bruccoleri, R. E., Olafson, B. D., States, D. J., Swa-

minathan, S., and Karplus, M. (1983) J. Comp. Chem. 4, 187-217 Colombo, G., and Villafranca, J. J. (1986) J. Bid. Chem. 261,10587-

10591 Dautry-Versat, A., Cohen, G. N., and Stadtman, E. R. (1979) J. Biol.

Chem. 254,3124-3128 Eisenberg, D., and McLachlan, A. D. (1986) Nature 3 1 9 , 199-203 Fersht, A. R., Shi, J-P, Knill-Jones, J., Lowe, D. M., Wilkinson, A.

J., Blow, D. M., Brick, P., Carter, P., Waye, M. M. Y., and Winter, G. (1985) Nature 3 1 4 , 235-238

Frey, T. G., Eisenberg, D., and Eiserling, F. A. (1975) Proc. Natl. Acad. Sci. U. S. A. 7 2 , 3402-3406

Ginsburg, A. (1972) Adu. Protein Chem. 26, 1-79 Ginsburg, A., and Stadtman, E. R. (1973) in The Enzymes of Gluta-

mine Metabolism (Prusiner, S., and Stadtman, E. R., eds) pp. 9-43, Academic Press, New York

berg, D. (1986) Nature 3 2 3 , 304-309

248

Hamlin, R. (1985) Methods Enzymol 1 1 4 , 416-452 Haschemeyer, R. H., Wall, J . S., Hainfield, J., and Maurizi, M. R.

(1982) J. Biol. Chem. 257, 7252-7253 Hendrickson, W. A., and Konnert, J . H. (1981) in Biomolecular

Structure, Function, Conformation and Euolution (Srinivasan, R., ed, Vol. 1, pp. 43-57, Pergamon, Oxford.

Heidner, E. G., Prey, T. G., Held, J., Weissman, L. J., Fenna, R. E., Lei, M., Harel, M., Kabsch, H., Sweet, R. M., and Eisenberg, D. (1978) J. Mol. Biol. 122, 163-173

Holzer, H., Mecke, D., Wulff, K., Liess, K., and Heilmeyer, L., Jr. (1967) Adu. Enzyme Regulation 5, 211-225

Holzer, H. (1969) Adu. Enzymol. 3 2 , 297-326 Hunt, J. B., Smyrniotis, P. Z., Ginsburg, A., and Stadtman, E. R.

Hunt, J. B., and Ginsburg, A. (1980) J. Biol. Chem. 255,590-594 Hunt, J. B., and Ginsburg, A. (1981) Biochemistry 2 0 , 2226-2233 Jack, A., and Levitt, M. (1978) Acta Crystallogr. A34,931-935 Janson, C. A., Almassy, R. J., Westbrook, E. M., and Eisenberg, D.

(1984) Arch. Biochem. Biophys. 2 2 8 , 512-518 Janson, C. A., Kayne, P. S., Almassy, R. J., Grunstein, M., and

Eisenberg, D. (1986) Gene (Amst.) 46, 297-300 Jones, T. A. (1978) J. App. Crystallogr. 11, 268-272 Kabsch, W. (1977) J. App. Crystallogr. 122, 426-429 Lei, M., Aebi, U., Heidner, E. G., and Eisenberg, D. (1979) J. Biol.

Lesk, A. M., and Hardmann, K. D. (1982) Science 1 6 , 539-540 Levintow, L., and Meister, A. (1954) J. Biol. Chem. 2 0 9 , 265-280 Maurizi, M. R., and Ginsburg, A. (1982) J. Bid. Chem. 2 5 9 , 7246-

Mavridis, I. M., and Tulinsky, A. (1976) Biochemistry 15,4410-4417 Mercer, W. D., Winn, S. I., and Watson, H. C. (1976) J . Mol. Biol.

Monroe, D. M., Noyes, C. M., Griffith, M. J., Lundblad, R. L., and

Moss, J., Stanley, S. J., and Levine, R. L. (1988) FASEB J. 2 , 1045

Rees, D. C. (1983) Acta Crystallozr. A39,916-920

(1975) Arch. Biochem. Biophys. 166, 102-124

Chem. 254,3129-3134

7251

104,277-283

Kingdon, H. S. (1985) Curr. Top. Cell. Regul. 2 7 , 361-372

(abstr.)

Rees, D. C., Lewis, M., and Lipscomb, W. N. (1983) J. Mol. Biol. 168,367-387

Reitzer, L. J., and Magasanik, B. (1987) in Escherichia coli and Salmonella typhimurium (Neidhart, F. C., ed ), Vol. 1, 302, Ameri- can Society for Microbiology, Washington, D. C.

Roseman, J. E., and Levine, R. L. (1987) J. Bid. Chem. 262, 2101- 2110

Rossmann, M. G., and Blow, D. M. (1963) Acta Crystallogr. 16, 39- 45

Schevitz, R. W., Podjarny, A. D., Zwick, M., Hughes, J. J., and Sigler, P. B. (1981) Acta Crystallogr. A37,669-677

Shapiro, B. M., Kingdon, H. S., and Stadtman, E. R. (1967) Proc. Natl. Acad. Sci. U. S. A. 58, 642-649

Shapiro, B. M., and Stadtman, E. R. (1970) Annu. Reu. Microbiol. 24,501-524

Stadtman, E. R., Shapiro, B. M., Ginsburg, A., Kingdon, H. S., and Denton, M. D. (1968a) Brookhauen Symp. Biol. 2 1 , 379-396

Stadtman, E. R., Shapiro, B. M., Kingdon, H. S., Woolfolk, C. A,, and Hubbard, J . S. (1968b) Adu. Enzyme Regul. 6 , 257-289

Stadtman, E. R., and Ginsburg, A. (1974) in The Enzymes (Boyer, P., ed), 10, 755-807

Suck, D., Rayment, I., Johnson, J. E., and Rossmann, M. G. (1978) Virology 8 5 , 187-197

Valentine, R. C., Shapiro, B. M., and Stadtman, E. R. (1968) Bio- chemistry 7,2143-2152

Villafranca, J. J., Ash, D. E., and Wedler, F. C. (1976a) Biochemistry 15,536-543

Villafranca, J. J., Ash, D. E., and Wedler, F. C. (1976b) Biochemistry 15,544-553

Villafranca, J. J., Balakrishnan, M. S., and Wedler, F. C. (1977) Biochem. Biophys. Res. Commun. 75,464-471

Wang, B-C. (1986) Methods Enzymol. 115,90-112 Watson, H. C., Shotton, D. M., Cox, J. M., and Muirhead, H. (1970)

Weissman, L. (1977) Ph.D. dissertation, University of California, Los Nature 225,806-811

Angeles