Article No. mb981931 J. Mol. Biol. (1998) 281, 485±499

A pH-dependent Stabilization of an Active Site LoopObserved from Low and High pH Crystal Structures ofMutant Monomeric Glycinamide RibonucleotideTransformylase at 1.8 to 1.9 AÊ

Ying Su1, Mason M. Yamashita1, Samantha E. Greasley1

Christine A. Mullen2, Jae Hoon Shim3, Patricia A. Jennings2

Steven J. Benkovic3 and Ian A. Wilson1*

1Department of MolecularBiology and The SkaggsInstitute for Chemical BiologyThe Scripps Research Institute10550 North Torrey PinesRoad, La Jolla, CA 92093-0359USA2Department of Chemistry andBiochemistry, University ofCalifornia, San Diego, La JollaCA 92037, USA3Department of ChemistryPennsylvania State UniversityUniversity Park, PA 16802USA

E-mail address of the correspondwilson@scripps.edu

Abbreviations used: GarTfase, glribonucleotide transformylase; 10fTtetrahydrofolate; b-Gar, b-glycinamf-b-Gar, formyl b-Gar; DDATHF, 5tetrahydrofolate; 5dTHF, 5-deaza-5tetrahydrofolate; GARS, glycinamidsynthetase; AIRS, aminoimidazolesynthetase; MR, molecular replacem5,8-dideazafolate; fDDF, formyl DDglycol.

0022±2836/98/330485±15 $30.00/0

A mutation in the dimer interface of Escherichia coli glycinamide ribonu-cleotide transformylase (GarTfase) disrupts the observed pH-dependentassociation of the wild-type enzyme, but has no observable effect on theenzyme activity. Here, we assess whether a pH effect on the enzyme'sconformation is suf®cient by itself to explain the pH-dependence of theGarTfase reaction. A pH-dependent conformational change is observedbetween two high-resolution crystal structures of the Glu70Ala mutantGarTfase at pH 3.5 (1.8 AÊ ) and 7.5 (1.9 AÊ ). Residues 110 to 131 in GarT-fase undergo a transformation from a disordered loop at pH 3.5, wherethe enzyme is inactive, to an ordered loop-helix structure at pH 7.5,where the enzyme is active. The ordering of this ¯exible loop-helix has adirect effect on catalytic residues in the active site, binding of the folatecofactor and shielding of the active site from solvent. A main-chain car-bonyl oxygen atom from Tyr115 in the ordered loop forms a hydrogenbond with His108, and thereby provides electronic and structural stabiliz-ation of this key active site residue. Kinetic data indicate that the pKa ofHis108 is in fact raised to 9.2. The loop movement can be correlated withelevation of the His pKa, but with further stabilization, probably fromAsp144, after the binding of folate cofactor. Leu118, also in the loop,becomes positioned near the p-amino benzoic acid binding site, providingadditional hydrophobic interactions with the cofactor 10-formyl tetrahy-drofolate. Thus, the pH-dependence of the enzyme activity appears toarise from local active site rearrangements and not from differences dueto monomer-dimer association.

# 1998 Academic Press

Keywords: purine biosynthesis; folate cofactors; loop ¯exibility; monomer-dimer association; enzyme mechanism

*Corresponding author

ing author:

ycinamideHF, 10-formylide ribonucleotide;,10-dideaza-5,6,7,8-,6,7,8-e ribonucleotide

ribonucleotideent; DDF,F; PEG, polyethylene

Introduction

Glycinamide ribonucleotide transformylase (EC2.1.2.2; GarTfase) catalyses a formyl transfer fromthe cofactor 10-formyl tetrahydrofolate (10fTHF)to the substrate b-glycinamide ribonucleotide(b-Gar) to form formyl-b-glycinamide ribonucleo-tide (f-b-Gar; Figure 1). GarTfase is the thirdenzyme in the de novo purine biosynthesis path-way and has become a target for mediating neo-plasia due to the identi®cation of competitiveinhibitors, such as 5,10-dideaza-5,6,7,8-tetrahydro-

# 1998 Academic Press

Figure 1. The reaction equation for GarTfase with tetrahedral transition state.

486 pH-dependent Stabilization of GarfTfase Active Site Loop

folate (DDATHF: Taylor et al., 1985) and AG2034(Boritaki et al., 1996). A novel series of selectiveGarTfase inhibitors has been developed based onstructure-based drug design (Jackson 1994, 1997)including 5-thia-2,6-diamino-4(3H)-oxopyrimidineinhibitors (Varney et al., 1997). GarTfase fromEscherichia coli and other prokaryotes are mono-functional proteins, whereas eukaryotic GarTfasesare incorporated within a covalent, trifunctionalenzyme complex that includes glycinamide ribo-nucleotide synthetase (GARS) and aminoimidazoleribonucleotide synthetase (AIRS). Rational designof inhibitors for GarTfase has employed X-raystructures of the E. coli enzyme, due to itssequence and mechanistic similarities to theeukaryotic enzyme (Inglese et al., 1990a; Kan et al.,1992).

E. coli GarTfase is a 212 residue enzyme (23,237Daltons) with a pH-dependent kinetic rate pro®le(Inglese et al., 1990b). The kinetic rate maximumoccurs between pH 8.0 and 8.5 with a nearly bell-shaped drop at the pH extremes. E. coli GarTfaseexhibits a pH-dependent monomer-dimer equili-brium with the monomer preferably populatedabove pH 7.4 and the dimer below pH 6.6 (Mullen& Jennings, 1996). These solution studies are con-sistent with crystal studies of GarTfase at variouspH values. Wild-type phosphate-bound GarTfase,

crystallized at pH 6.75 (Chen et al., 1992; Almassyet al., 1992), and the multisubstrate adductBW1476U89 complexed with GarTfase at pH 6.3(Klein et al., 1995) are dimeric, whereas the ternarycomplex of the substrate b-Gar with 5-deaza-5,6,7,8-tetrahydrofolate (5dTHF) at pH 7.5(Almassy et al., 1992) crystallizes as four indepen-dent monomers per asymmetric unit. The proposalof a monomer-dimer transition near physiologicpH as a mechanism to control the activity ofGarTfase and, thereby, the production of purinesin E. coli, was an intriguing speculation from theseobservations (Mullen & Jennings, 1996).

To address this hypothesis, a single point mutantGlu70Ala (E70A) of GarTfase was constructed todisrupt the pH-dependent dimerization. Dynamiclight-scattering and gel ®ltration results (Mullen &Jennings, 1998) con®rmed the monomeric nature ofthe mutant GarTfase from pH 6.5 to 8.2. The X-raystructures, reported here, also con®rm a lack ofdimer formation at pH 3.5 and at 7.5. The kineticrate pro®le of the mutant protein is nearly identicalwith the wild-type, with both enzymes being inac-tive at lower pH values, even though the proteinremains monomeric. Thus, the pH-dependence ofthe monomer-dimer equilibrium and enzymeactivity appear to be decoupled. Dimerization ofE. coli GarTfase cannot then account for the

pH-dependent Stabilization of GarfTfase Active Site Loop 487

observed decrease in enzyme activity at moreacidic pH values.

So what factors lead to enzyme activity at highpH and inactivity at lower pH? A pH-dependenttertiary structural change was proposed as anexplanation for the observed enzymatic results(Mullen & Jennings, 1998). To identify whether anyconformational change occurs between the inactivelow-pH and active high-pH forms, we have deter-mined the crystal structures of the mutant E70AGarTfase at pH 3.5 and at pH 7.5 at 1.8 and 1.9 AÊ

resolution, respectively. Since E70A GarTfasemaintains a monomeric structure over a widerange of pH values, the conclusions derived aboutthe pH-dependence of the binding site confor-mation and enzymatic mechanism should not beobscured by dimer interactions. These monomericstructures can then be compared to previouslyreported GarTfase crystal structures in order toextend the structural database for the design ofnovel inhibitors that can induce or stabilize inac-tive enzyme conformations at physiologic pHvalues, and, hence, provide an alternative mode ofenzymatic inhibition.

Results and Discussion

Overall description of low and high-pHmutant structures

The pH 3.5 E70A mutant GarTfase structure, inspace group P212121, was determined from cryo-cooled crystals (Table 1) at 1.8 AÊ resolution bymolecular replacement (MR) using the programAMoRe (Navaza, 1994). A monomer from thehigh-resolution (1.96 AÊ ) dimeric multi-substrateadduct inhibitor complex structure GarTfase-BW1476U89 was used as the search probe (Brook-haven PDB code 1GAR; Klein et al., 1995). ThepH 7.5 structure, in space group P6122, was sub-sequently determined by MR at 1.9 AÊ with data,

Table 1. Crystallographic parameters and data statistics

CrystalpH 3.5

CrystalpH 7.5

Space group P212121 P6122Unit cell dimensions (AÊ ) a � 45.8,

b � 47.8 a � b � 50.4c � 107.2 c � 275.5

No. molecules per asymmetric unit 1 1VM (AÊ 3Daÿ1) 2.6 2.2Resolution range (AÊ ) 50-1.81 50-1.90No. observations 87,331 67,156No. unique reflections 21,864 16,340Multiplicitya 4.0 4.1Completeness (%); (last shell) 97.0; 77.6b 92.8; 81.8c

I/s 26.3; 3.1b 22.6; 2.6c

Rsym (%)d 5.0 5.0Rsym for last resolution shell 28.7 33.1

a Number of observations/unique re¯ections.b Last shell 1.84 to 1.80 AÊ .c Last shell 1.97 to 1.90 AÊ .d Rsym � (�h�i (Ihi ÿ hIhi)/�h�i Ih) � 100, where hIhi is the

mean of the Ihi observations of re¯ection h.

also collected at ÿ183�C, with the pH 3.5 mutantGarTfase structure as the probe. The low and high-pH forms of E70A GarTfase have been re®ned at1.8 and 1.9 AÊ , respectively (Table 2). The crystallo-graphic R values are 20.1 and 21.5% and Rfree

values are 25.1 and 27.4% for the low and high-pHstructures respectively, with good geometric par-ameters (Table 2). Residues 111 to 131, and 210 to212 are not included in the ®nal model of thepH 3.5 structure due to lack of electron density cor-responding to those residues. The pH 7.5 structureincludes the entire molecule except for theC-terminal residues 210 to 212. Both structures con-tain a tightly bound phosphate ion in the activesite as observed in the previously determinedstructures of native dimeric wild-type GarTfase(Chen et al., 1992, Almassy et al., 1992).

The overall topology of monomeric E70AGarTfase at pH 3.5 and 7.5 is the same as the wild-type protein (Chen et al., 1992; Almassy et al., 1992;Klein et al., 1995). The N-terminal domain consistsof four parallel b-strands (b1 to b4) ¯anked onboth sides by a pair of a-helices (a1 and a2 on oneside, a3 and a4 on the other) that is connected tothe C-terminal domain by a ®fth parallel strand(b5) and two antiparallel strands (b6 and b7). Oneside of the C-terminal b-sheet is ¯anked by a longa-helix (a5) that leads to a short three-strandedb-meander (b8 to b10).

Glu70 is located in the middle of helix a3(Figure 2) and is involved in the dimeric inter-actions at the interface of the GarTfase dimer(Chen et al., 1992). A mutation at Glu70 to Ala70was not expected to cause any conformationalchanges in the helix, as Ala has the highest a-helix-forming propensity among all amino acids (Blaberet al., 1993; Myers et al., 1997). Indeed, both struc-tures show that the mutation E70A does not resultin any gross changes in this helix. In addition, themutation does not affect the secondary structure ofresidues in close contact with Glu70 in the wild-type dimer including helix a4 or the adjacent

Table 2. Re®nement statistics of the ®nal model

pH 3.5 pH 7.5

Resolution range (AÊ ) 50-1.8 50-1.9Crystallographic R-value (%) 20.1 21.5Free R-value (%) (10% of refl.) 25.1 27.4No. protein atoms 1439 1619No. water molecules 124 150R.m.s. deviation from:

Ideal bond lengths (AÊ ) 0.01 0.006Ideal bond angle (deg.) 1.4 1.4Ideal dihedral angles (deg.) 26.2 26.9Ideal improper angles (deg.) 0.8 0.8

Quality of Ramachandran plot:Residues in most favored region (%) 93.2 87.3Residues in additional allowed region (%) 6.8 11.0Residues in generously allowed region (%) 0.0 1.7Residues in disallowed region (%) 0.0 0.0

Crystallographic R-value � (�hjFo ÿ Fej/�hFo), where Fo and Fc

are the observed and calculated structure factors, respectively.Free R-value is computed as the R-value but with the test set ofre¯ections only.

Figure 2. Structure of a GarT-fase dimer at low pH (<6.9)and location of mutationE70A. The dimerization inter-face of GarTfase is formedthrough symmetrical pairingof the outer strand, b3, of thecentral seven-stranded b-sheetand helices a2 and a3. Thestructure shown was deter-mined by Klein et al. (1995),PDB code 1GAR. Residues 38to 78, in pink, form the dimerinterface. The side-chains ofresidues Glu70 and His73 con-tribute signi®cantly to thedimer association and are dis-

played in green and purple, respectively. A mutation at Glu70 to Ala70 would disrupt speci®c electrostatic inter-actions at the dimer interface but would be expected to maintain the overall secondary structure in helix a3.

488 pH-dependent Stabilization of GarfTfase Active Site Loop

strands b1 through b4. The r.m.s.d. for all back-bone atoms of helix a3 are 0.25 AÊ and 0.21 AÊ , forthe pH 3.5 and 7.5 structures, respectively, whencompared to the phosphate-bound GarTfasedimeric structure at pH 6.75 (Chen et al., 1992).

The comparison of the monomeric E70AGarTfase structures with phosphate-boundGarTfase at pH 6.75 (Chen et al., 1992) reveals theabsence of the previously observed dimer interfacefor both pH 3.5 and 7.5 structures. Wild-type phos-phate-bound GarTfase crystallizes as a dimer inC2221 at pH 6.75 (Chen et al., 1992) and the dimeri-zation is through interactions resulting from thesymmetrical pairing of the outer strand of the cen-tral seven-stranded b-sheet and helices a2 and a3(Figure 2). Mutant GarTfase crystallizes in P212121

at pH 3.5 and in P6122 at pH 7.5, each with onemolecule per asymmetric unit. The crystal contactsin both pH 3.5 and pH 7.5 crystals (Figure 3) revealthat the E70A GarTfase packs with different mol-ecular faces from the face used in the dimerizationof the wild-type GarTfase. Thus, mutant E70AGarTfase crystallizes as a monomer at pH 3.5 andpH 7.5.

Comparison of the low and high-pH structures

The primary structural differences between theinactive pH 3.5 and active pH 7.5 mutant GarTfaseare con®ned to three loops between strands b5 andb6 (residues 111 to 131), strands b6 and b7 (resi-dues 141 to 145) and a stretch of residues from 158to 165 (Figure 4(a)). The r.m.s. deviations betweenthe low and high-pH forms are 1.40 AÊ for allatoms (excluding residues 111 to 131, which aredisordered at pH 3.5) and 0.52 AÊ for a carbonatoms without the three loops of signi®cant differ-ence. The corresponding r.m.s. deviation for allatoms, excluding the three loops, is 1.08 AÊ .

The long loop from 111 to 131 is totally disor-dered in the pH 3.5 structure. Throughout there®nement, no electron density was traceable forany part of this loop, suggesting that it is too ¯ex-

ible or disordered to assume a single discrete con-formation. In contrast, the same loop in the pH 7.5structure is clearly visible in the electron densitymap, but is still less well-de®ned compared tomost of the residues in the N-terminal domain. Theaverage B-value for the loop is 59 AÊ 2, about twicethe average of 31 AÊ 2 for the entire protein (Table 3).The ternary complex of the substrate b-Gar with5dTHF (Almassy et al., 1992) has a similar ¯exible,but still well-de®ned, loop 111-131 conformation(Figure 4(b)). A possible implication for this loopconformational variation and ordering at high pHfor the mechanism of action of GarTfase is dis-cussed later.

Loop 141-145 is ¯exible in both the pH 3.5 and7.5 structures, as indicated by weak electron den-sity and higher B-values (Table 3). The relativeorientations of this loop vary considerably in allthe different GarTfase structures, as shown froman overlap of the various structures (Figure 4(c)and shown individually in Figure 4(a) and (b)).The 141 to 145 loops diverge substantially in thelow and high-pH structures, resulting in a 10.4 AÊ

displacement for the a-carbon position of the activesite residue, Asp144. It is uncertain if the displace-ment of Asp144 is signi®cant in vitro for the inacti-vation of GarTfase at low pH, since it is involvedin crystal packing interactions in both the high andlow-pH structures. The crystal contacts utilizedifferent molecular faces of GarTfase in the lowand high-pH mutant structures, as represented inFigure 3(a) and (b), respectively. Although thesolvent content is lower in the pH 3.5 versus 7.5structure, contacts between residues in the loop141-145 are similar based on a 4 AÊ cutoff distance.In the low-pH structure, Leu143 contacts Gly159and Asp160, and Asp144 contacts Gly159. Thehigh-pH structure also shares three intra-molecularcontacts but with a different molecular face;Leu143 contacts Pro113, Pro205 and Gln206. Crys-tal contacts have often been invoked as a perturb-ing force which either stabilizes or destabilizes theorder of the contact (Kossiakoff et al., 1992; Rader

pH-dependent Stabilization of GarfTfase Active Site Loop 489

& Agard, 1997; Young et al., 1994) and, therefore,may be responsible for the difference in the loopscontaining Asp144. However, it is likely that bind-ing of the folate cofactor is required for stabiliz-ation of this loop.

Finally, residues 158 to 166 form a connectingloop between strand b7 and the long helix a5, anddiffer signi®cantly between the pH 3.5 and 7.5structures (Figure 4(a)). In the pH 7.5 structure,helix a5 is unwound slightly at the amino end andthe connecting loop is displaced as a rigid body.Again, this region has higher-than-average thermalfactors in all the GarTfase structures (Table 3).Crystal contacts in the low-pH structure areGly159 with Leu143, and Asp144 and Asp160 withLeu143. The same 158-166 loop in the high-pHstructure has Phe157 contacting Gln15 and Asp18.As with the loop 141-145, loop positional variation(Figure 4(c)) may be a function of crystal contacts.

Figure 3(a) (legen

pH-induced local conformational changes of Tyrand Trp residues in GarTfase have been studied byabsorbance and ¯uorescence spectroscopies(Mullen & Jennings, 1996). In the E70A mutants,0.4 tyrosine residues per molecule are more sol-vent-accessible at pH 6.5 compared to pH 8.2.Apart from Tyr115, the ®ve remaining tyrosineresidues (Tyr67, Tyr78, Tyr100, Tyr177 and Tyr208)superimpose when comparing the pH 3.5 and 7.5structures, all with well-de®ned electron density inboth structures, except for Tyr208, which is locatednear the C terminus. Tyr115 is located in a totallydisordered loop in the pH 3.5 structure comparedto its location in a de®ned loop at pH 7.5. There-fore, at pH 3.5, Tyr115 adopts a disordered confor-mation and is more solvent-accessible, consistentwith the increase in tyrosine solvent-accessibilityobserved in solution (Mullen & Jennings, 1998).

d on page 490)

Figure 3. Stereo view of crystal contacts in (a) the pH 3.5 E70A GarTfase in P212121 and in (b) the pH 7.5 E70A inP6122. Regions 141±145 and 158±166 are in green and the faces normally used for dimerization are in cyan. Note thecomplete loss of the normal dimer interface in both crystal forms.

490 pH-dependent Stabilization of GarfTfase Active Site Loop

GarTfase exhibits a pH-independent maximum¯uorescence emission at 343 nm and a pH-depen-dent intensity (Mullen & Jennings, 1996). Fluor-escence quenching studies reveal that both wild-type and mutant GarTfases share identical quench-ing patterns as the pH decreases, suggesting thatthe quenching pattern is not correlated to dimeri-zation. Mullen & Jennings (1996) concluded thatone or both of the GarTfase tryptophan residues,Trp183 and/or Trp197, contributed to ¯uorescence

and that their solvent-accessibilities do not varywith pH. They suggest that the ¯uorescencequenching is actually due to the presence of anionizable histidine residue near one or both of thetryptophan residues (Mullen & Jennings, 1998).Fluorescence quenching of tryptophan by the pre-sence of a nearby histidine residue is commonlyfound in proteins as, for example, in barnase(Loewenthal et al., 1992), human erythrocyte glu-cose transporter (Chin et al., 1992) and murine

pH-dependent Stabilization of GarfTfase Active Site Loop 491

interleukin-6 (Matthews et al., 1997). A comparisonof the pH 3.5 and 7.5 mutant crystal structures con-®rms that there is no pH-dependent conformation-al change for either Trp183 or Trp197 and thatboth are solvent-exposed. Only Trp197 has a near-by histidine residue, His192, which is solvent-exposed and approximately 4.5 AÊ away. Therefore,the interaction between Trp197 and His192 is verylikely associated with the pH-dependence of the¯uorescence studies of GarTfase.

Figure 4(a) (legen

Comparison of the active site with otherGarTfase structures

All reported structures of wild-type GarTfase(Almassy et al., 1992; Chen et al., 1992; Klein et al.,1995) are dimeric, with the exception of the pH 7.6GarTfase-5dTHF-b-Gar ternary complex structure,which is monomeric (Almassy et al., 1992). Examin-ation of the pH-dependence of the steady-state kin-etic parameter kcat for the E70A GarTfase revealed

d on page 493)

492 pH-dependent Stabilization of GarfTfase Active Site Loop

a pro®le unchanged from the wild type enzyme(Figure 5). The dependency of kcat on pH follows abell-shaped curve, with pKa � 7.1 � 0.1 andpKa � 9.1 � 0.2 with kcat maximum at 71(�14) sÿ1.For the wild-type enzyme, a similar bell-shapedcurve was observed with pKa � 7.1 � 0.1 and

Figure 4(b) (leg

pKa � 9.2 � 0.2 with kcat maximum at 58(�11) sÿ1.At enzyme concentrations <1.0 nM, there was nokinetic evidence that aggregation in¯uences thepH-rate pro®le for the wild-type enzyme. Thesestructures share the same general fold as the E70Amutant GarTfase structures (Table 4). Residues 111

end opposite)

Figure 4. Structures of various GarTfases showing the location of ¯exiable loops. (a) Individual representations of thepH 7.5 and pH 3.5 E70A GarTfase; (b) the GarTfase-5dTHF-b-Gar complex and the GarTfase-BW1476U89 complex;(c) the location of the three ¯exible regions, residues 111 to 131, 141 to 145, and 158 to 166, are represented for var-ious structures of GarTfase (pH 7.5 structure in gray, pH 3.5 structure in red, GarTfase-BW1476U89 complex structurein cyan, GarTfase-5dTHF-b-Gar complex structure in pink, phosphate-bound structures by Chen et al. (1992) in blueand by Almassy et al. (1992) in green). Ca atoms of residues 1 to 110, 132 to 140, 146 to 157 and 167 to 209 are usedin the alignment. The Ca trace of the pH 7.5 E70A GarTfase is shown as a reference. Note the large differences of twohighly mobile loops, residues 141 to 145 and 158 to 166, one of which contains the catalytic residue Asp144. Thewide range of positions of Asp144 is likely due to various crystal contacts or lack of folate cofactor in the differentstructures.

pH-dependent Stabilization of GarfTfase Active Site Loop 493

to 131 are unde®ned in the ``unliganded'', butphosphate-bound, structures (Almassy et al., 1992;Chen et al., 1992) and only partially de®ned in the

Table 3. Ratio of average B-values (AÊ 2) for all atoms of ththe overall molecule

Residues E70A pH 3.5 E70A pH 7.5

111-131 ± 1.9141-145 1.4 1.5158±165 2.2 2.1moleculec (AÊ 2) 25 31

a Almassy et al. (1992).b Klein et al. (1995) for monomer 1 and monomer 2.c Average B-value (AÊ 2) for all protein atoms.

dimeric wild-type GarTfase-BW1476U89 (a pico-molar multisubstrate adduct inhibitor) structure(Klein et al., 1995; see Figure 6). The pH 7.6

e three ¯exible loops in GarTfase structures compared to

Structure1CDEa 1GARb 1GARb

1.4 ± ±1.8 1.4 1.51.9 1.6 1.5

25 36 33

Figure 5. pH rate pro®les for the E70A mutant andwild-type GarTfase are nearly identical withpKa � 7.1 � 0.1 and pKa � 9.1 � 0.2 with kcat maximumat 71(�14) sÿ1 for the mutant and pKa � 7.1 � 0.1 andpKa � 9.2 � 0.2 with kcat maximum at 58(�11) sÿ1 forthe wild-type enzyme.

Table 4. R.m.s.d. of Ca between 1CDE (Almassy et al.,1992) and the other GarTfase structures (AÊ )

1GRCa 1CDDbE70A

pH 3.5E70A

pH 7.5 1GARc 1GARc

0.62 0.60 0.66 0.60 0.48 0.44

The following residues were used for superpositions: 1 to 110,132 to 140, 146 to 157 and 167 to 209.

a Chen et al. (1992).b Almassy et al. (1992).c Klein et al. (1995), monomer 1 and monomer 2.

494 pH-dependent Stabilization of GarfTfase Active Site Loop

GarTfase-5dTHF-b-Gar ternary complex structure(Almassy et al., 1992) reveals an ordered loop (resi-dues 113 to 118) and helix (residues 119 to 128).Two additional ¯exible loops (residues 141 to 145and 158 to 166) are observed in all X-ray crystalstructures of GarTfase but their relative positionsvary extensively, as shown in Figure 4. These

Figure 6. Stereo view showing the proximity of the loopHis108 and Asp144. The binary complex structure of GarE70A GarTfase structure in pink. Residues 122 to 126 areand His108 are proposed to stabilize the oxyanion of theresidues in the ordered loop-helix that interact with the cathe binding site.

highly variable loops have greater than averageB-values when compared to the overall protein(Table 3) and are not involved in dimeric interfacecontacts. As no signi®cant change was noted in theregions around residues that form the dimer inter-face, we conclude that the monomer-dimer tran-sition does not substantially change the secondarystructure of GarTfase.

Protein ¯exibility is often directly linked to pro-tein function. Many proteins require this ¯exibilityin order to ful®l their biological functions. Forexample, in E. coli thymidylate synthase, the ¯exi-bility of the C terminus allows the protein to adopta ``closed'' and an ``open'' conformation in order tosequester reactants from bulk solvent and to pos-ition the substrate and cofactor for catalysis(Montfort et al., 1990). In savinase, a subtilisin-likeproteinase with broad speci®city, residues 99 to

helix (110±131) to key residues in the active site, Asn106,Tfase-BW1476U89 is represented in cyan and the pH 7.5

not ordered in the GarTfase-BW1476U89 structure. Asn106transition state (see Figure 7). Tyr115 and Leu118 are keytalytic residues and bound cofactor, and help to desolvate

Figure 7. Mechanism of action for GarTfase as modi®edfrom Klein et al. (1995). (1) b-Gar binds to the substratesite, placing its primary amine group in a position toattack the formyl carbon atom of 10-formyl-tetrahydro-folic acid. The formyl oxygen atom is hydrogen bondedto Asn106 and the positively charged imidazoliumgroup of His108, and thereby helps to withdraw elec-trons and activate the formyl carbon atom for nucleo-philic attack. (2) The reaction proceeds by nucleophilicattack by the primary amine group on the formyl car-bon atom, resulting in the tetrahedral transition inter-mediate formation. (3) Electron rearrangement andtransfer of a hydrogen atom from formyl-b-Gar to thecofactor N10 is mediated through the structural water ora bulk solvent molecule. (4) Products are formed andreleased. R1, glutamate-p-aminobenzoic acid; R2, pteri-dine ring; R3, phosphoribosyl glycinamide.

pH-dependent Stabilization of GarfTfase Active Site Loop 495

104 have an increased mobility at higher pH sothat the substrate channel widens and allows thesubstrate easier access to the active site (Langeet al., 1994). In other examples, rigidi®cation ofa ¯exible segment facilitates activation. Intriosephosphate isomerase, the rigidity of theactive site loops, due to subunit-subunit interactionand substrate binding, is important for optimal cat-alysis (Phillips et al., 1977; Borchert et al., 1995).

The conformation of loop-helix residues 111 to131 in the pH 7.5 E70A GarTfase structure is verysimilar to the structure in the ternary GarTfase-5dTHF-b-Gar structure (Figure 4(a) and (b)). ThepH 3.5 E70A GarTfase, pH 7.5 E70A GarTfase, andthe ternary structures may represent a continuumfrom a completely disordered, inactive state, to afully ordered loop-helix, active state. Our crystallo-graphic studies suggest that the ordering of theloop-helix (111 to 131) is pH-dependent and is themost signi®cant difference between the low-pH(6.8) wild-type dimer and low-pH (3.5) and high-pH (7.5) monomeric structures.

Is the ordering of this loop-helix critical for theactivity of GarTfase, perhaps for preparing theactive site for ef®cient catalysis? One consequenceof the loop-helix ordering is the formation of ahydrogen bond between the main-chain carbonylgroup of Tyr115 and the side-chain of His108,which could affect the position and pKa of thisessential active site residue (Klein et al., 1995;Warren et al., 1996). A second consequence is toprovide a hydrophobic lid to the formyl transfer,p-aminobenzoic acid and substrate binding sitesthat, in addition to increasing favorable hydro-phobic interactions, may further affect the pKa ofHis108 by desolvation of His108 and ordering ofthe Asp144 loop. Indeed, the solvent-exposed areaof His108 decreases to 30 AÊ 2 in the active pH 7.5E70A structure from 68 AÊ 2 in the pH 3.5 E70Astructure. The side-chain imidazole group assumesidentical positions in the active pH 7.5 E70Amutant and ternary GarTfase structures, whereasin the pH 3.5 E70A structure, as well as most low-pH GarTfase structures, a high degree of disorderor multiple conformations is seen for His108. In thecase of the multisubstrate adduct BW1476U89 com-plex with GarTfase, His108 assumes two distinctconformations, one towards and one away fromthe active site. These observations imply increasedconformational constraints on the active site due toloop-helix ordering. Leu118 is an additional resi-due found in this loop-helix region that is impor-tant for cofactor binding (Figure 6). Leu118 ispositioned to interact with the p-aminobenzoic acidmoiety of the folate cofactor. Therefore, rigidity ofthe loop-helix and its proximity to the active sitemay enhance ef®cient catalysis.

The implications of the loop-helix conformation-al transition to the mechanism of action andregulation of GarTfase can now be explored.A mechanism of action for GarTfase was proposedby Klein et al. (1995) and this is represented in amodi®ed form in Figure 7. A protonated imidazo-

lium group of His108 is positioned near the formylgroup of 10-fTHF and acts to withdraw electrons.The formyl carbon atom is now activated fornucleophilic attack by the primary amine of b-Gar.His108, along with Asn106, can then stabilize theoxyanion of the transition state analogue. Transferof a proton from b-Gar to the N10 of folate is pro-posed to be mediated though a structural water orbulk solvent molecule and is followed by productrelease from the active site. Site-directed mutagen-esis studies suggest that none of these three activesite residues (Asn106, His108 or Asp144) is absol-utely required for catalysis; a limited set of singlemutants can give rise to an active enzyme, albeit atsubstantially reduced activity. However, any com-bination of double or triple mutants is completelyinactive (Warren et al., 1996).

Figure 8. A revised mechanism for the GarTfase reac-tion. The loop-helix forms at physiologic pH, donatingthe main-chain carbonyl group of Tyr115 to ®x the pos-ition of the imidizolium side-chain of His108, which inturn stabilizes the oxyanion transition state intermediateshown in Figure 6. This Figure is shown in the sameorientation as Figure 6 with a 90� rotation about thehorizontal axis. R1 represents a pterin, R2 the p-amino-benzoate-glutamate, and R3 the remaining portion ofb-Gar. At low pH, the loop-helix disorder leads to disor-dering of His108 and a possible hydrogen transfer to theoxyanion could trap the reaction in a secondaryhydroxyl amine. Binding of cofactor appears to helporder the 141-145 loop and brings Asp144 in contactwith His108.

496 pH-dependent Stabilization of GarfTfase Active Site Loop

As a result of our present study, we suggest thatthe original proposal by Klein et al. (1995) need bemodi®ed only slightly to include stabilization ofthe positively charged imidazolium group by theconformational ordering of the ¯exible loop-helixregion (Figure 8). The loop-helix contributes amain-chain carbonyl oxygen atom from Tyr115that hydrogen bonds to the imidazolium group ofHis108 (C � O- - - -H-Nd distance of 2.6 AÊ ). Hence,this situation demands that His108 be protonatedif it is to act either as an acid or to stabilize theoxyanion transition state by direct hydrogen bond-ing through its NHe group. This main-chain hydro-gen bond has two effects: (1) it partially neutralizes

the charge of the imidazolium group, therebyincreasing its pKa and reducing its ability to donatea hydrogen atom to the oxyanion of the transitionstate; and (2) locks the imidazolium in a confor-mation that, together with Asn106, will stabilizethe charge of the oxyanion transition state. Thereaction can then proceed with hydrogen transfervia a structural water or bulk solvent molecule andproduct formation.

The decreased activity at lower pH can then beattributed to disorder of the ¯exible loop-helixwith concomitant change in position and pKa ofthe His108 imidazolium ring. Our proposalrequires the existence of a positively charged imi-dazolium group at physiologic pH values thatwould require elevation of its pKa from around 6to at least 8 (the pH where maximal catalytic rateoccurs). In fact, kinetic evidence suggest thatHis108 is the titratable group at the high pH endof the catalytic rate pro®le; therefore, its pKa mustbe raised to 9.2 (J.H.S. & S.J.B., unpublishedresults). Depending on the local environment, thepKa values of ionizable residues in proteins can besigni®cantly different from that of their isolatedstates in solution. For example, in bovine low mol-ecular mass protein tyrosine phosphatase, the pKa

values of His66 and His72 are shifted to 8.3 and9.2, respectively, by electrostatic interactions withthe carboxyl groups of glutamate residues(Tishmack et al., 1997). The pKa of His95 in triose-phosphate isomerase is shifted in the oppositedirection, from 6 to 4.5, by its location at theN-terminal end of an a-helix where it can be stabil-ized by the partial positive charge from the helixdipole (Lodi & Knowles, 1991).

The desolvation of His108 to a mainly hydro-phobic environment and acquisition of a singlehydrogen bond from the main-chain carbonylgroup of Tyr115, both resulting from closure of theloop-helix, is probably not suf®cient to account forthe entire pKa shift in His108 from 6 to 9.2. Wepropose that the additional stabilization is derivedfrom a buried salt-bridge from Asp144 to theHis108 imidazolium, as seen in the multisubstrateadduct inhibitor BW1476U89 complex (Klein et al.,1995) and the 5dTHF and b-Gar ternary complex(Almassy et al., 1992). In both low and high-pHmutant structures, the loop containing Asp144 islocated away from His108, probably as a resultof a lack of bound folate cofactor. Thus, thecofactor appears to be responsible for ordering ofthe 141-145 loop that brings Asp144 into proxi-mity with His108 (Figure 8). In theGarTfase-10-formyl-5,8,10-trideazatetrahydrafolate-b-Gar complex structure (S.E.G. & I.A.W., unpub-lished results), the backbone amide group ofGly117 also forms a hydrogen bond to the side-chain of Asp144 and presumably helps stabilizethe charged form of Asp144.

Thus, the ordering of the loop-helix has becomean intriguing player in the proposed mechanism ofthe GarTfase reaction. The importance of the loop-helix is consistent with the results of earlier

pH-dependent Stabilization of GarfTfase Active Site Loop 497

mutation work, where residues 118 through 122were deleted and replaced by a bridge of two ser-ine residues. This loop deletion was not capable ofcomplementing the TX680 auxotrophic cell lines,indicating that the mutant enzyme was inactive(Warren et al., 1996). These data also support a keyrole for the loop-helix ordering in the active sitestabilization and enzyme mechanism. The view ofthe loop-helix as relatively ¯exible and weaklyassociated with the active site of GarTfase expandsthe scope of the rational drug design from a merestatic model to a more realistic view of a multi-con-formational active site that might potentially leadto novel methods of enzyme inhibition.

Materials and Methods

Materials

The overexpressing mutant E70A GarTfase E. coli wasprepared as described by Mullen & Jennings (1998).Luria broth (LB) and agar were obtained from Life Tech-nologies, Gaithersburg, MD and all other common buf-fers and reagents were purchased from Sigma-AldrichCorporation, St. Louis, MO.

Protein preparation

Mutant E70A GarTfase E. coli was prepared asdescribed by Mullen & Jennings (1998). Cells weregrown in LB medium with 100 mg/ml ampicillin at 32�Cto an absorbance at 600 nm of 0.6. The protein wasinduced with 0.5 mM isopropyl-b,D-thiogalactopyrano-side (IPTG) and grown for an additional six hours. Cellswere harvested by centrifugation and resuspended in aminimal volume of 50 mM Tris-HCl (pH 7.3), 1 mMEDTA.

Protein purification

Cells were disrupted by sonication and the lysateclari®ed by centrifugation at 20,000 g for one hour. Pro-tein was precipitated from the supernatant by additionof 50% saturated ammonium sulfate and centrifugationfor 30 minutes. The resulting pellet was resuspended inbuffer A and loaded onto a 50-ml DEAE-anion-exchangecolumn (Whatman, Maidenstone, England) and elutedwith a 0 M to 1 M KCl gradient. Fractions with the pro-tein mutant GarTfase were detected by SDS-PAGE andpooled. The pooled fractions were loaded onto a 500-mlSuperdex 300 gel ®ltration column (Pharmacia BiotechInc., Piscataway, NJ) equilibrated with 50 mM Bis-Tris(pH 6.0), 1 mM EDTA. A pH of 6.0 was chosen to dimer-ize any wild-type GarTfase expressed by the E. coli (i.e.not expressed from the plasmid), thereby separating itfrom the mutant GarTfase (a monomer at pH 6.0) basedon molecular mass. Fractions were again pooled on thebasis of SDS-PAGE analysis, which revealed greater than99% purity. The protein focused to a smeared band witha pI from 5.5 to 6.5 on isoelectric focusing gels (identicalwith the pI pro®le of the wild-type GarTfase).

Kinetic measurements

Instead of the unstable natural cofactor, 10-fTHF, akinetically equivalent and more stable fDDF was used inkinetic measurements (Inglese et al., 1990a). The initial

velocity for the reaction of GAR transformylase withGAR and fDDF was determined by monitoring the pro-duction of DDF at 295 nm (�e � 18.9 mMÿ1 cmÿ1). Thereactions were done in a 1 ml cuvette thermosatically con-trolled on a Gilford 252 spectrophotometer, and the reac-tion was initiated with one of the substrates at anenzyme concentration of 0.5 to 1 nM in MTEN buffer(50 mM 2-(N-morpholino)ethanesulfonic acid, 25 mMethanolamine, 25 mM Tris, 100 mM sodium chloride).

Crystallization and data collection

Crystallization trials were performed using the sitting-drop vapor diffusion method (Ducruix & GiegeÂ, 1992)where equal volumes (1.8 ml) of a 20 mg/ml protein sol-ution and crystallization buffer were mixed and thedrops left to equilibrate at 22�C. A number of crystalswere obtained from an incomplete factorial screen devel-oped by Jancarik & Kim (1991). Crystals from two ofthese initial conditions were selected for data collection;one at low pH and one at high pH. The low-pH crystalsgrow from 15% (w/v) PEG 1500 at pH 3.5, while thehigh-pH crystal form grows from a solution of 2% (v/v)PEG 400, 2.0 M ammonium sulfate, 0.1 M sodium Hepes(pH 7.5).

Data for both crystal forms were collected on a Mar-Research imaging plate system on beam-line 7-1 at Stan-ford Synchrotron Radiation Laboratory. The data werecollected at a temperature of ÿ183�C by ¯ash-cooling(Petsko, 1975; Hope, 1988) the crystals in the nitrogengas stream. To prepare crystals for ¯ash-cooling, theywere mounted in a cryo-loop (Teng, 1990; HamptonResearch, Laguna Hills, CA) and transferred quicklythrough a drop containing glycerol at 15% and 30%(v/v) in crystallization buffer for the low and high-pHcrystals, respectively, before transferring to the goni-ometer head. The low-pH crystal form diffracted to1.8 AÊ resolution and data were collected on an 18 cmMAR plate at a distance of 130 mm with an oscillationangle of 2�. Due to a long unit cell edge in the high-pHcrystal form, data were collected on a 30 cm plate at adistance of 300 mm with the plate raised by 60 mm tocollect data to 1.9 AÊ resolution. The crystal was alignedto have the longest cell axis parallel with the f-rotationaxis with a 2� oscillation angle. Data were processedusing DENZO and SCALEPACK (Otwinowski, 1993).Data statistics are summarized in Table 1.

Structure solutions and refinement

The crystal structures were determined by molecularreplacement using the program AMoRe (Navaza, 1994)incorporated into the CCP4 Program Package (CCP4,1994). The probe model for the pH 3.5 crystal was amonomer from the dimeric multi-substrate adductinhibitor complex structure GarTfase-BW1476U89 (Brook-haven PDB reference code 1GAR; Klein et al., 1995). There®ned pH 3.5 crystal structure was then used as thesearch model for the crystal structure at pH 7.5. Thetranslational search ascertained that P6122 was the cor-rect enantiomorphic space group for the pH 7.5 crystal.The R-values after molecular replacement were 44.1 and48% (resolution 12 to 4 AÊ ) for the low and high-pHstructures, respectively. Electron density map calcu-lations and re®nement of molecular replacement sol-utions were carried out using X-PLOR (BruÈ nger, 1992a)with a cut-off of Fo > 2sF. The Rfree value (BruÈ nger, 1992b)was calculated at each stage of X-PLOR re®nement with a

498 pH-dependent Stabilization of GarfTfase Active Site Loop

sequestered set of data that comprised about 10% of theobserved data (2200 and 1600 re¯ections for the pH 3.5and 7.5 dataset, respectively). Model building was per-formed using the O program (Jones et al., 1991).

The ®rst calculated 2Fo ÿ Fc electron density mapsusing data between 10 and 3 AÊ after rigid-bodyre®nement of the MR solutions (10 to 4 AÊ ) showed thatextensive rebuilding was required in two regions, resi-dues 140 to 145 and 158 to 165, in both structures. TheR-values after rigid-body re®nement were 40.7 and43.8% (resolution 12 to 4 AÊ ) for the low and high-pHstructures, respectively. The region 111 to 131 was com-pletely disordered in the initial pH 3.5 map and through-out all subsequent stages of re®nement. This sameregion in the pH 7.5 crystal was better de®ned and couldbe clearly modeled. Several cycles of re®nement andmodel building were performed using data between 8 AÊ

and the highest resolution for each structure. Each cycleincluded model rebuilding, based on 2Fo ÿ Fc or Fo ÿ Fc

shake-omit maps, followed by positional re®nement orsimulated annealing re®nement and by individual B-value re®nement. An overall anisotropic data correctionof B11 � B22 � 13 AÊ 2, B33 � 19 AÊ 2 using X-PLOR wasapplied in the pH 7.5 structure re®nement. This correc-tion raised the overall B-value derived from the Wilsonplot from 15.0 AÊ 2 to 26.3 AÊ 2. Water molecules were mod-eled using the program SHELXL (Sheldrick, 1996), whichemploys the following criteria to assign water molecules:(1) the highest peaks in Fo ÿ Fc maps; (2) no bad con-tacts; and (3) creates at least one geometrically plausiblehydrogen bond to a hydrogen-bonding partner. Theidenti®ed water molecules were subjected to X-PLORre®nement. After each round of the re®nement, watermolecules with B-values greater than twice the averageB-value of the protein or without reasonable hydrogenbonds to the protein were discarded. All water moleculeswere assigned unit occupancy. A bulk solvent correctionwas applied to the low and high-pH structures towardsthe end of water molecule building using X-PLOR (Jiang& BruÈ nger, 1994). The re®nement results are summarizedin Table 2. The current R-values are 20.1 and 21.5% forthe low and high-pH structures, respectively, and theRfree values are 25.1 and 27.4%, with the differencesbetween R and Rfree values of 5.0 and 5.9% which is con-sistent with structures at this resolution and state ofre®nement (Kleywegt & BruÈ nger, 1996). The assignedPDB ID codes are 2gar and 3gar for the pH 3.5 andpH 7.5 E70A GarTfase, respectively.

Analysis and graphics

Ramachandran plot analyses were computed usingPROCHECK (Laskowski et al., 1993). Buried surfaceswere calculated using the program MS (Connolly, 1983)with a probe radius of 1.4 AÊ . InsightII (MSI, San Diego,CA) and Adobe Photoshop (Adobe Systems Incorpor-ated, San Jose, CA) were used for preparing the Figures.

Acknowledgments

We thank Mark S. Warren for his kind gift of themutant E70A GarTfase vector, the staff of Stanford Syn-chrotron Radiation Laboratory (SSRL) beam line 7-1 forhelpful support, and Dr Dale Boger for helpful discus-sions. This project has been supported by NIH grantsPO1-CA63536 (I.A.W. and S.J.B.) and GM54038 (P.A.J.),

NIH training grant T32 MH19185 (Y.S.) and T32CA09523 (C.A.M.), the Hellman Family Foundation(P.A.J.), the Lucille P. Markey Charitable Trust (C.A.M.),and the ARCS Foundation (C.A.M.).

References

Almassy, R. J., Janson, C. A., Kan, C.-C. & Hostomska,Z. (1992). Structures of apo and complexed Escheri-cia coli glycinamide ribonucleotide transformylase.Proc. Natl Acad. Sci. USA, 89, 6114±6118.

Blaber, M., Zhang, X. J. & Matthews, B. W. (1993). Struc-tural basis of amino acid alpha helix propensity.Science, 260, 1637±1640.

Borchert, T. V., Kishan, K. V., Zeelen, J. P., Schliebs, W.,Thanki, N., Abagyan, R., Jaenicke, R. & Wierenga,R. K. (1995). Three new crystal structures of pointmutation variants of monoTIM: conformational¯exibility of loop-1, loop-4 and loop-8. Structure, 3,669±679.

Boritaki, T. J., Burlett, C. A., Zhang, C., Howland, E. F.,Margosiak, S. A., Palmer, C. L., Romines, W. H. &Jackson, R. C. (1996). AG2034 a novel inhibitor ofglycinamide ribonucleotide formyltransferase. Inves-tigat. New Drugs, 14, 295±303.

BruÈ nger, A. T. (1992a). X-PLOR, Version 3.1, A System forX-Ray Crystallography and NMR, Yale UniversityPress, New Haven, CT.

BruÈ nger, A. T. (1992b). Free R value: a novel statisticalquantity for assessing the accuracy of crystal struc-tures. Nature, 355, 472±474.

CCP4, (1994). The CCP4 suite; programs for proteincrystallography. Acta Crystallog. sect. D, 50, 760±763.

Chen, P., Schulze-Gahmen, U., Stura, E. A., Inglese, J.,Johnson, D. L., Marolewski, A., Benkovic, S. J. &Wilson, I. A. (1992). Crystal structure of glycina-mide ribonucleotide transformylase from Escherichiacoli at 3.0 AÊ resolution. A target enzyme for che-motherapy. J. Mol. Biol. 227, 283±292.

Chin, J. J., Jhun, B. H. & Jung, C. Y. (1992). Structuralbasis of human erythrocyte glucose transporterfunction: pH effects on intrinsic ¯uorescence. Bio-chemistry, 31, 1945±1951.

Connolly, M. L. (1983). Analytical molecular surface cal-culation. J. Appl. Crystallog. 16, 548±558.

Ducruix, A. & GiegeÂ, R. (1992). Crystallization of NucleicAcids and Proteins: A Practical Approach, Oxford Uni-versity Press, New York.

Hope, M. (1988). Cryocrystallography of biologicalmacromolecules: a generally applicable method.Acta Crystallog. sect. B, 44, 22±26.

Inglese, J., Johnson, D. L., Shiau, A., Smith, J. M. &Benkovic, S. J. (1990a). Subcloning, characterization,and af®nity labeling of Escherichia coli glycinamideribonucleotide transformylase. Biochemistry, 29,1436±1443.

Inglese, J., Smith, J. M. & Benkovic, S. J. (1990b). Active-site mapping and site-speci®c mutagenesis of glyci-namide ribonucleotide transformylase from Escheri-chia coli. Biochemistry, 29, 6678±6687.

Jackson, R. C. (1994). Inhibitors of thymidylate synthaseand glycinamide ribonucleotide transformylase.Advan. Exp. Med. Biol. 370, 179±183.

Jackson, R. C. (1997). Contributions of protein structure-based drug design to cancer chemotherapy. Semin.Oncol. 24, 164±172.

pH-dependent Stabilization of GarfTfase Active Site Loop 499

Jancarik, J. & Kim, S. (1991). Sparse matrix sampling: ascreening method for crystallization of proteins.J. Appl. Crystallog. 24, 409±411.

Jiang, J. S. & BruÈ nger, A. T. (1994). Protein hydrationobserved by X-ray diffraction. Solvation propertiesof penicillopepsin and neuraminidase crystal struc-tures. J. Mol. Biol. 243, 100±115.

Jones, A. T., Zou, J. Y. & Cowan, S. W. (1991). Methodsfor building protein models in electron densitymaps and location of errors in these models. ActaCrystallog. sect. A, 47, 110±119.

Kan, C. C., Gehring, M. R., Nodes, B. R., Janson, C. A.,Almassy, R. J. & Hostomska, Z. (1992). Heter-ologous expression and puri®cation of activehuman phosphoribosylglycinamide formyltransfer-ase as a single domain. J. Protein Chem. 11, 467±473.

Klein, C., Chen, P., Arevalo, J. H., Stura, E. A.,Marolewski, A., Warren, M. S., Benkovic, S. J. &Wilson, I. A. (1995). Towards structure-based drugdesign: crystal structure of a multisubstrate adductcomplex of glycinamide ribonucleotide transformy-lase at 1.96 AÊ resolution. J. Mol. Biol. 249, 153±175.

Kleywegt, G. J. & BruÈ nger, A. T. (1996). Checking yourimagination: applications of the free R value. Struc-ture, 4, 897±904.

Kossiakoff, A. A., Randal, M., Guenot, J. & Eigenbrot, C.(1992). Variability of conformations at crystal con-tacts in BPTI represent true low-energy structures:correspondence among lattice packing and molecu-lar dynamics structures. Proteins: Struct. Funct.Genet. 14, 65±74.

Lange, G., Betzel, C., Branner, S. & Wilson, K. S. (1994).Crystallographic studies of savinase, a subtilisin-likeproteinase, at pH 10.5. Eur. J. Biochem. 224, 507±518.

Laskowski, R. A., MacArthur, M. W., Moss, D. S. &Thornton, J. M. (1993). PROCHECK: a program tocheck the stereochemical quality of protein struc-tures. J. Appl. Crystallog. 26, 283±291.

Lodi, P. J. & Knowles, J. R. (1991). Neutral imidazole isthe electrophile in the reaction catalyzed by triose-phosphate isomerase: structural origins and cataly-tic implications. Biochemistry, 30, 6948±6956.

Loewenthal, R., Sancho, J. & Fersht, A. R. (1992). Histi-dine-aromatic interactions in barnase. Elevation ofhistidine pKa and contribution to protein stability.J. Mol. Biol. 224, 759±770.

Matthews, J. M., Ward, L. D., Hammacher, A., Norton,R. S. & Simpson, R. J. (1997). Roles of histidine 31and tryptophan 34 in the structure, self-association,and folding of murine interleukin-6. Biochemistry,36, 6187±6196.

Montfort, W. R., Perry, K. M., Fauman, E. B., Finer-Moore, J. S., Maley, G. F., Hardy, L., Maley, F. &Stroud, R. M. (1990). Structure, multiple site bind-ing, and segmental accommodation in thymidylatesynthase on binding dUMP and an anti-folate (pub-lished erratum appears in Biochemistry (1990) 29,10864). Biochemistry, 29, 6964±6977.

Mullen, C. A. & Jennings, P. A. (1996). Glycinamideribonucleotide transformylase undergoes pH-depen-dent dimerization. J. Mol. Biol. 262, 746±755.

Mullen, C. A. & Jennings, P. A. (1998). A single mutata-tion disrupts the pH-dependent dimerization of gly-cinamide ribonucleotide transformylase. J. Mol. Biol.276, 819±827.

Myers, J. K., Pace, C. N. & Scholtz, J. M. (1997). A directcomparison of helix propensity in proteins and pep-tides. Proc. Natl Acad. Sci. USA, 94, 2833±2837.

Navaza, J. (1994). AMoRe: an automated package formolecular replacement. Acta Crystallog. sect. A, 50,157±163.

Otwinowski, Z. (1993). Data Collection and Processing,SERC Daresbury Laboratory, Warrington, UK.

Petsko, G. A. (1975). Protein crystallography at sub-zerotemperatures: cryo-protective mother liquors forprotein crystals. J. Mol. Biol. 96, 381±392.

Phillips, D. C., Rivers, P. S., Sternberg, M. J. E.,Thornton, J. M. & Wilson, I. A. (1977). An analysisof the three-dimensional structure of chicken triosephosphate isomerase. Biochem. Soc. Trans. 5, (B),642±646.

Rader, S. D. & Agard, D. A. (1997). Conformational sub-states in enzyme mechanism: the 120 K structure ofalpha-lytic protease at 1.5 AÊ resolution. Protein Sci.6, 1375±1386.

Sheldrick, G. M. (1996). SHELX-96, Program for CrystalStructure Re®nement, University of GoÈ ttingen.

Taylor, E. C., Harrington, P. J., Fletcher, S. R., Beardsley,G. P. & Moran, R. G. (1985). Synsthesis of theantileukemic agents 5,10-dideaza-aminopterin and5.10-dideaza-5,6,7,8-tetrahydroaminopterin. J. Med.Chem. 78, 914±921.3.

Teng, T. Y. (1990). Mounting of crystals for macromol-ecular crystallography in a free-standing thin ®lm.J. Appl. Crystallog. 23, 287±391.

Tishmack, P. A., Bashford, D., Harms, E. & Van Etten,R. L. (1997). Use of 1H NMR spectroscopy and com-puter simulations to analyze histidine pKa changesin a protein tyrosine phosphatase: experimental andtheoretical determination of electrostatic propertiesin a small protein. Biochemistry, 36, 11984±11994.

Varney, M. D., Palmer, C. L., Romines, W. H., 3rd,Boritzki, T., Margosiak, S. A., Almassy, R., Janson,C. A., Bartlett, C., Howland, E. J. & Ferre, R. (1997).Protein structure-based design, synthesis, and bio-logical evaluation of 5-thia-2,6-diamino-4(3H)-oxo-pyrimidines: potent inhibitors of glycinamideribonucleotide transformylase with potent cellgrowth inhibition. J. Med. Chem. 40, 2502±2534.

Warren, M. S., Marolewski, A. E. & Benkovic, S. J.(1996). A rapid screen of active site mutants in gly-cinamide ribonucleotide transformylase. Biochemis-try, 35, 8855±8862.

Young, A. C., Tilton, R. F. & Dewan, J. C. (1994). Ther-mal expansion of hen egg-white lysozyme. Com-parison of the 1.9 AÊ resolution structures of thetetragonal form of the enzyme at 100 K and 298 K.J. Mol. Biol. 235, 302±317.

Edited by D. Rees

(Received 19 February 1998; received in revised form 30 April 1998; accepted 5 May 1998)