Mechanism of inhibition of 3a,2013-hydroxysteroiddehydrogenase by a licorice-derived steroidal inhibitor

Debashis Ghoshl*, Mary Erman1, Zdzislaw Wawrzak 2t, William L Duax1

and Walter Pangborn11Hauptman-Woodward Medical Research Institute (formerly Medical Foundation of Buffalo, Inc.), 73 High Street, Buffalo,NY 14203-1196, USA and 2Technical University of Lodz, Institute of Physics, ul. Wolczanska, 219, 93-005 Lodz, Poland

Background: Bacterial 3,20p-hydroxysteroid dehy-drogenase (3a,201-HSD) reversibly oxidizes the 3a and203 hydroxyl groups of androstanes and pregnanes anduses nicotinamide adenine dinucleotide as a cofactor.3ot,201-HSD belongs to a family of short-chain dehydro-genases that has a highly conserved Tyr-X-X-X-Lyssequence. The family includes mammalian enzymesinvolved in hypertension, digestion, fertility and sperm-atogenesis. Several members of the enzyme family,including 3,203-HSD, are competitively inhibited byglycyrrhizic acid, a steroidal compound found in licorice,and its derivative, carbenoxolone, an anti-inflammatoryglucocorticoid.Results: The three-dimensional structure of theenzyme-carbenoxolone complex has been determined

and refined at 2.2 A resolution to a crystallographic R-factor of 19.4%. The hemisuccinate side chain of car-benoxolone makes a hydrogen bond with the hydroxylgroup of the conserved residue Tyr152 and occupies theposition of the nicotinamide ring of the cofactor. Theoccupancies of the inhibitor in four independent catalyticsites refine to 100%, 95%, 54% and 36%.Conclusions: The steroid binds at the catalytic site in amode much like the previously proposed mode ofbinding of the substrate cortisone. No bound cofactormolecules were found. The varying occupancy of steroidmolecules observed in the four catalytic sites is either dueto packing differences or indicates a cooperative effectamong the four sites. The observed binding accounts forthe inhibition of 3a,203-HSD.

Structure 15 October 1994, 2:973-980

Key words: carbenoxolone, crystallography, enzyme-inhibitor complex, steroid

Introduction3c,20{-hydroxysteroid dehydrogenase from Streptomyceshydrogenans (3ot,2003-HSD; EC 1.1.1.53) is a nicotin-amide adenine dinucleotide [NAD(H)]-linked short-chain dehydrogenase that reversibly oxidizes the3ao-hydroxyl and 20-hydroxyl groups of androstane andpregnane derivatives [1,2]. It is one of the first-enzymesof the short-chain superfamily to be studied by X-raycrystallographic techniques [3,4]. Another member ofthis superfamily whose structure is known is dihydro-pteridine reductase [5]. The three-dimensional structureof the enzyme-NAD complex of 3oa,2013-HSD revealedthe existence of the putative catalytic triad of conservedresidues, tyrosine, lysine and serine in the active site, andprovided the basis for modeling the steroid-proteininteraction and a hypothesis for the mechanism of theoxidoreductive reaction involving the conserved aminoacids [4]. Site-directed mutagenesis studies providedindependent evidence that the conserved tyrosine andlysine residues play major roles in the catalytic processesof short-chain dehydrogenases [6-9]. Structural results[3] also enabled us to confirm the previous suggestionthat there is only one catalytic site for activities at boththe C3 and C20 ends of steroidal substrates [10].Depending on their nature and the reaction conditions(such as the prevailing pH), substrates can orient them-selves with either the C3 or the C20 end towards thecatalytic residues.

In addition to 3a,2013-HSD, the analogous enzymes15-hydroxyprostaglandin dehydrogenase and l1 3-HSDhave been shown to be inhibited by glycyrrhizic acid (asteroidal compound derived from licorice) and by itsderivative, carbenoxolone, an anti-inflammatory gluco-corticoid [11-13]. In the case of 3t,20P-HSD,carbenoxolone exhibited competitive inhibition with50% inhibition (K) at about 1 ,uM. Single crystalssuitable for X-ray studies were obtained from theinhibited enzyme [11].

We report here the crystal structure at 2.2 A resolutionof 3at,203-HSD in complex with carbenoxolone, aglucocorticoid, bound at the catalytic site of theenzyme. The unbiased electron density map obtained forthe carbenoxolone molecule from the present studyenabled us to position it directly and unequivocally atthe catalytic site of the enzyme, in close proximity to thecatalytically important residues. The atomic model ofthe catalytic site reveals two hydrogen-bonding inter-actions of the inhibitor with the protein, in addition tovan der Waals contacts with the hydrophobic aminoacids. The most specific contact is a hydrogen bondbetween the hemisuccinate side chain at the C3 positionof the carbenoxolone and the Tyr152 hydroxyl group.The binding of the inhibitor is strikingly similar to theproposed model of binding for cortisone [4]; however,carbenoxolone binds with the C3 position inside the

© Current Biology Ltd ISSN 0969-2126

*Corresponding author. tPresent address: University of Pennsylvania, Department of Chemistry, 231 South 34th Street, Philadelphia,PA 19104-6323, USA.

973

974 Structure 1994, Vol 2 No 10

catalytic cleft, while cortisone is a substrate at the C20position. The hemisuccinate side chain at the C3position of carbenoxolone occupies the nicotinamideposition of the cofactor. NAD is therefore eitherprevented from binding or competitively displaced bythe inhibitor.

Results and discussionInteraction with catalytic residues and steroid conformationThe overall structure of a monomer of the tetrameric3a,20P-HSD is shown schematically in Fig. la. Itconsists of a a3 motif resulting in a seven-strandedparallel P-sheet, flanked by three parallel oa-helices on

Fig. 1. (a) The overall structure of amonomer of 3ct,2013-HSD in complexwith carbenoxolone (green), an anti-inflammatory glucocorticoid. The a-helices are shown as magenta coils,[g-strands as blue ribbons ending inarrows, and, non-secondary structureregions as cyan ropes. Also shown, inorange, are residues Tyr152, Lys156,and Ser139, which are believed to becatalytically important. (Figure gen-erated using SETOR [201.) (b) Chemicalstructures of the glucocorticoid inhibitorcarbenoxolone and the substratecortisone.

Inhibited complex of 3a,2013-HSD Ghosh et al. 975

each side. The asymmetric unit is a tetramer; thesubunits are related by three mutually perpendicular two-fold axes, described previously as P, Q and R axes [4].

Fig. 2a shows the refined molecule of carbenoxoloneat the active site and its unbiased electron densitymap. The initial model of the inhibitor was obtainedfrom studying a single crystal of carbenoxoloneOF Griffin, personal communication). The crystal struc-ture of carbenoxolone and its structure as a complexwith the enzyme are essentially the same; they differonly in the conformation of the hemisuccinate sidechain at the C3 position which is gauche in the complexas opposed to trans in the crystal structure. This changeis effected by a rotation about the C29-C30 bond. Fig.2b shows the calculated (3 Fobs-2Fcalc) electron densityfor the refined carbenoxolone molecule at the endof refinement.

In Fig. 2c the protein environment of the inhibitor inthe catalytic cleft is shown. The carbenoxolone moleculeis anchored at both ends of the molecule by hydrogenbonds. The hemisuccinate terminal oxygen forms ahydrogen bond (2.8 A) to Tyrl52, one of the two strictlyconserved residues of the catalytic triad. Such an inter-action of the conserved tyrosine side chain with thesubstrate keto-oxygen has been proposed to initiate thehydride-transfer reaction [4]. In this instance, thehydrogen bonding does not occur directly with theoxygen at the C3 position of the steroid, as postulatedfor enzyme substrates such as 17p-hydroxy-5a-androstan-3-one. The presence of the hemisuccinate sidechain at this position on carbenoxolone precludes such adirect contact but makes it a viable inhibitor byenhancing its solubility.

The structure of carbenoxolone at the C20 end of themolecule is significantly different from that of a substratesuch as cortisone or progesterone at the C20 position,while the C3 end of carbenoxolone more closely mimicsthe A-ring of 17-hydroxy-5a-androstan-3-one.However, modeling the stereospecific 3-keto to 3-hydroxyl reaction for 17p-hydroxy-5a-androstan-3-onesuggests that the substrate molecule is oriented with thea-face of the steroid towards Tyrl52, while the hydridefrom the cofactor approaches the C3 atom from the 3-face. Carbenoxolone, on the other hand, is bound withits -face towards Tyr152, such that the C3-0 groupresembles a 3P-hydroxyl group.

A second, weak hydrogen-bonding interaction (3.5 A)occurs between the 11-keto oxygen of carbenoxoloneand the hydroxyl on Ser91. This side chain assumes adifferent conformation in the inhibited complex fromthat observed in the holo-enzyme. Interestingly, car-benoxolone is also an inhibitor for human 11 -HSD[12], but has no effect on the activity of human170-HSD (S-X Lin, personal communication). Theother hydrogen bond that the steroid molecule makeswith the protein is a short interaction (2.5 A) between

the carboxylic oxygen 034 at the C21 end of themolecule, and the hydroxyl of the SerlOO side chain ofthe D-subunit of a symmetry-related molecule. Thisinteraction appears to be incidental since inhibition ofthe enzyme is achieved in a very dilute solution [11].

Examination of the packing environments of the ends ofthe inhibitor molecules that are farthest from thecatalytic triad reveals that they are similar in subunitsA and B. Thus, the atom 034 of the steroid in theB-subunit makes a similar hydrogen bond (2.5 A) withthe SerlOO hydroxyl of the C-subunit of the crystallo-graphically related tetramer. The environments of thisend of the steroid in subunits C and D, on the otherhand, are similar but different from those in subunits Aand B. The nearest protein atoms to the 034 atoms ofthe steroids in subunits C and D are LyslO06 side chainnitrogens from crystallographically related subunits Band A at distances of 5.0 A and 6.1 A, respectively. Inaddition, three water molecules could be built into themodel, one in the vicinity of the 034 atom in theC-subunit and two in the D-subunit. The hydrogen-bonding distances of these water molecules are in therange 2.8-3.4 A.

The Lys156 side chain, the other strictly conservedresidue that is catalytically active [14], shows nohydrogen-bonding interaction with the Tyr152 hydroxyl(the separation of Tyr152 OH and Lys156 N5 is 5.1 A,which is slightly greater than in the NAD-containingcomplex structure). A strong interaction between theseside chains has been proposed in the transition state ofthe enzyme-catalyzed reaction [4]. A water molecule hasbeen found that is tightly bound to the Lys156 sidechain. The third member of the so-called catalytic triad,Ser139, makes a stronger hydrogen bond (2.8 A) withthe Tyr152 hydroxyl in the inhibitor complex than in theNAD-containing complex structure (3.8 A). This inter-action may also be important for either hydride orproton transfer or both.

Other residues in the catalytic cleftThe catalytic cleft is mainly hydrophobic in nature.Hydrophobic/aromatic residues that surround the car-benoxolone molecule are Gly93, Met94, Ala140,Alal41, Leu146, Leu148, Thr149, Met184, Metl89,Tyr202, Trp243, and residues Val252 and Met253 fromthe B-subunit (in the active site of the A-subunit), whosea-carbon trace is visible below the steroid molecule inFig. 2c. Some of the residues that are in van der Waalscontact with the inhibitor molecule, such as Met184,Met189 and Thr193, may play a role in substrate recog-nition and selectivity. A dotted Connolly surface of theactive site and a stick model of carbenoxolone is shownin Fig. 3 to illustrate the complementarity of the bindingcavity and the inhibitor molecule. Nearly 75% of the280 A2 accessible surface area of the inhibitor is buriedin the catalytic cleft. Since the fused ring system is non-polar, a significant part of the binding energy is derivedfrom hydrophobic interactions.

976 Structure 1994, Vol 2 No 10

Fig. 2. (a) Unbiased electron densitymap for the carbenoxolone moleculebound at the active site of the A-subunitof 3a,2013-HSD. The density mapshown is a (3Fobs- 2 Fcalc) map contouredat or, prior to the inclusion of car-benoxolone in the model. The steroidmolecule shown is the final refinedmodel of carbenoxolone. Protein atomsin the surrounding density have beenremoved for clarity. (b) Calculated final(3Fobs-2 Fcalc) electron density map andthe final refined model for the car-benoxolone molecule. This map is alsocontoured at 1r. (c) The protein envi-ronment of the steroid molecule at thecatalytic cleft in the A-subunit and itshydrogen-bonding interactions. Proteinand steroid atoms are shown instandard colors. The a-carbon trace ofthe A-subunit is shown in red. Directlybelow, the trace of the carboxyl-terminus of the B-subunit (cyan) is alsovisible (in two discontinuous parts, onebelow text 93 and the other below textC20). Atoms from the crystallographi-cally related D-subunit are shown inblue. Hydrogen-bonding interactions(dashed white lines) are shown betweenthe hemisuccinate carboxyl oxygen andthe hydroxyl of Tyr152 (2.8 A), betweenthe 11-keto oxygen of the inhibitor andthe Ser91 hydroxyl (3.5 A), andbetween the carboxylic oxygen 034and the crystallographically relatedSer100 hydroxyl of the D-subunit (inblue). The side chain of Ser139 is alsohydrogen bonded (2.8 A) to the Tyr152hydroxyl. A few other key residues arealso shown. See text for further details.(Figure generated by CHAIN [16].)

Inhibited complex of 3a,2013-HSD Ghosh et al. 977

Fig. 3. A dotted Connolly surface (with1.35 A probe radius) of the catalyticcleft from the A-subunit with the boundcarbenoxolone molecule showing com-plementarity of the active site and theinhibitor molecule. The inhibitormolecule which has predominantlynon-polar atoms is buried with nearly75% of its surface within the cleft.(Figure generated by CHAIN [16].)

The subunit structure is almost identical in the steroid-bound form and the holo (NAD-containing)-form ofthe enzyme. When 253 pairs of a-carbon atoms fromtwo subunits were subjected to least-squares fitting, theroot mean square deviation (rmsd) for a-carbons was0.6 A. When the a-carbon atoms of all four subunits ofthe tetramers were similarly treated, the rmsd was also0.6 A. Therefore, no significant change has occurredeither in the individual subunits or in the quaternarystructure of the enzyme upon forming the complex. Thelargest deviations between the two backbones occur atthe two termini and at the semi-helical region betweenresidues 184 and 189, next to the catalytic site (Fig. 4).Flexibility in this partially helical loop regibn of thecatalytic cleft may be important for substrate recognitionand access to the active site. The electron density in thisloop region is somewhat diffuse, both in the complexedand uncomplexed structures, complicating interpretationof structural differences and their significance.

Occupation of nicotinamide site by hemisuccinate side chainAlthough NADH was present (at a concentration of1 mM) in the solution from which the 3,201-HSD

carbenoxolone complex crystals were grown, noevidence for the presence of NADH in the crystals wasfound in the electron density maps. A superposition ofthe catalytic clefts of the NAD-containing holo-enzymeand the carbenoxolone complex structure is shown inFig. 5. The hemisuccinate side chain present in theinhibitor molecule (green in Fig. 5), and absent fromsteroidal substrates, occupies the binding site of thenicotinamide ring of the cofactor molecule (yellow inFig. 5). The cofactor is either prevented from binding ordisplaced by the inhibitor molecule. Carbenoxoloneappears to play a dual role of inhibition, competing notonly with the binding of the substrate but also with thatof NAD(H).

Varied occupancies of steroid molecules in four active sitesA ribbon diagram of the tetrameric enzyme withinhibitor bound in each of the subunits is shown inFig. 6. Although the C3 end of the inhibitor is wellburied, the 034 end appears to protrude out of thecatalytic cavity. The strength of interaction at the 034end of the inhibitor with a residue in the neighboringtetramer in the crystal may determine the relative

Fig. 4. A stereoview showing superim-posed backbone atoms of NAD-con-taining (blue) and carbenoxolone-containing (red) complexes of 3a,20[3-HSD, near the catalytic site. The largestdeviation between two backbonesoccurs in the region between residues184 and 189. Some side chain confor-mations, such as those of Ser91 andSer139, are different in the two struc-tures, while the strictly conservedresidues Tyr152 and Lys156 assumenearly identical conformations.Carbenoxolone is shown in green.(Figure prepared using SETOR [20].)

978 Structure 1994, Vol 2 No 10

Fig. 5. A stereoview of superimposedbackbone atoms of NAD-containing(blue) and carbenoxolone-containing(red) complexes of 3a,203-HSD, nearthe catalytic site, as in Fig. 4. Thecofactor molecule is shown in yellowand carbenoxolone in green. Thehemisuccinate side chain of carbenox-olone overlaps with the nicotinamide-ring position of the cofactor. (Figureprepared using SETOR [20].)

Fig. 6. A ribbon diagram of the 3a,20P3-HSD tetramer in complex with car-benoxolone. Subunits are shown astraces of their backbone atoms indifferent colors. The inhibitor molecules,shown in contrasting colors, have theirC3 ends well-buried within the catalyticcavity, but have their 034 ends, whichare important in crystal packing interac-tions, exposed to solvent. The view isnearly parallel to the Q-axis two-fold(one of the symmetry elements of thenon-crystallographic 222 symmetrywhich relates the subunits of thetetramer). The P-axis is vertical and theR-axis is horizontal in this view. (Figureprepared using SETOR [201.)

half-lives in each of the four binding sites, and, hence,their relative occupancies. The occupancies of thesteroid molecules bound in the catalytic clefts ofsubunits B, C and D refined to 0.95, 0.54 and 0.36,respectively, while that for the A-subunit was heldconstant at 1.00. Visual inspection of the qualities ofunbiased electron densities for steroid molecules atthese sites agreed quite well with these values. Since theinhibition experiment was completed in a dilutesolution of the enzyme prior to the nucleation of thecrystal, progressively lower occupancies of the steroidmolecules cannot simply be attributed to an artifact ofcrystallization.

Subunits A and C (also B and D) are related by theQ-axis dimer interface, the most intimate of the threetypes of interfaces (P, Q and R) found in the tetramericenzyme, which takes the form of a four-helix bundle.Although having no direct contact with the boundsteroid, residues Leu165 to Val168 of the a-helix F ofthe C-subunit are in close proximity to the outer end of

the steroid-binding pocket of the A-subunit, andvice versa. Furthermore, residues Pro247 to Gln255 fromthe B-subunit are situated at the outer end of the steroidbinding cleft of the R-axis-related A-subunit, and viceversa. Examination of the crystal packing contactsreveals, as previously discussed, that the environments ofthe steroid-binding clefts are similar in subunits A andB, and also, in subunits C and D. The additional non-specific but rather strong hydrogen bonds in subunits Aand B may be important for the stability of the boundinhibitor. The absence of this hydrogen bond fromsubunits C and D may allow the inhibitor to slowly dis-sociate and diffuse out of the crystal lattice, resulting inreduced occupancies at these sites. This hypothesis issupported by the finding that the derived occupancies ofcarbenoxolone in subunits A and C are similar to thosein subunits B and D, respectively. If negative cooperativ-ity is not the explanation behind the reduced occupan-cies of the steroid at these sites, simple packingconsiderations and possible dissociation of the inhibitormay account for this observation.

Inhibited complex of 3a,20P-HSD Ghosh et al. 979

Biological implicationsBacterial 3,20-hydroxysteroid dehydrogenase(3a,20P-HSD) belongs to a superfamily of dehy-drogenases, classified as short-chain. A largenumber of prokaryotic and eukaryotic enzymes,such as Drosophila alcohol dehydrogenases, humanplacental 17-HSD, human renal 11P-HSD andhuman placental 15-hydroxyprostaglandin dehy-drogenase, have been identified as members ofthis family of enzymes. They are implicated in awide range of physiological processes includingalcohol metabolism, blood pressure control andproliferation of breast cancer. Common charac-teristics of this superfamily are a nicotinamideadenine dinucleotide binding fold and a strictlyconserved Tyr-X-X-X-Lys sequence of residues.The results of mutagenesis studies for somemembers of the family have confirmed theabsolute necessity of the conserved tyrosine andlysine residues for catalytic activity. X-raystructure analyses demonstrated the presence ofthese residues in the catalytic cavity. Models havealso been proposed for the binding of substratesin the active site and participation of theseconserved residues in the catalytic mechanism.

This paper presents the first direct evidence ofbinding of a substrate-like moiety in the catalyticsite of a member of this family of enzymes. Thestructure of the complex of 3a,201-HSD with theglucocorticoid molecule carbenoxolone (a deriva-tive of glycyrrhizic acid and the active ingredientin licorice), which is a competitive inhibitor ofthe enzyme, reveals that the steroid moleculebinds in the region previously identified as thecatalytic site, thereby preventing substratebinding. Carbenoxolone also competes effectivelywith the binding of the essential cofactor nicoti-namide adenine dinucleotide. The binding inter-actions include the formation of a specifichydrogen bond with the invariant Tyr152.Together with the site-directed mutagenesis studythat showed tyrosine to be essential for catalyticactivity [6-9], this result suggests that Tyr152could interact with the steroidal substrates in asimilar fashion. Analogous binding of carbenox-olone to human renal 111-HSD is reported to bea contributing factor in licorice-induced hyper-tension. Detailed information on the structure ofholo, apo, and inhibited forms of the short-chaindehydrogenase family of enzymes will be criticalto determining their mechanisms of action andthe design of selective and specific inhibitors.Such inhibitors could have direct therapeuticapplications as drugs with reduced side effects forthe treatment of short-chain dehydrogenase-related diseases, for example, inhibition of17[-HSD as a therapy for breast cancer.

Materials and methodsInhibition and crystallizationThe enzyme was inhibited and crystallized as described previ-ously [4]. All materials for the experiments, including car-benoxolone (31-hydroxy-11-oxoolean-12-en-30-oic acid3-hemisuccinate) and the crude enzyme, were purchased fromSigma Chemical Company. The crystals thus obtained sufferedseverely from twinning. Nearly 50 crystals from 15 prep-arations were examined before a crystal suitable for datacollection was obtained.

Data collectionThe crystal belonged to the space group P1 and the unit cellparameters are a=72.05 A, b=60.04 A, c=59.84 A, a=101.78 °,P=104.41°, y=96.31 °. A complete tetramer of 106 kDa was inthe unit cell. The data collection was performed on anR-AXIS IIc image plate area detector and processed with theR-AXIS program package (version 3.40 from Rigaku). Alldata were collected using one crystal. A total of 48 972 reflec-tions were collected having 34 405 unique data, between reso-lutions 51.08 A and 2.20 A. The overall Rmerge(F 2) was 11.4%.The data set was 71% complete overall to 2.20 A and 65%complete between 2.30 A and 2.20 A. The average F2/(F2value was 2.38 in the last shell.

Structure solution and refinementThe structure was solved by molecular replacement using theprogram package X-PLOR [15] and the structure of the holo-enzyme tetramer (without the NAD molecules) as the searchmodel. Rotation function calculations were performed withdata between 10.0-3.0 A resolution. Four peaks of similarmagnitudes were obtained, indicating four possible ways oforienting the tetramer. These four peaks had rotation functionvalues (in arbitrary units) between 9.3 and 8.5 while those forbackground peaks were <6.5 with a standard deviation of 0.6.The molecule was oriented in accordance with the peak ofgreatest magnitude and positioned with the center of mass at(0,0,0) of the unit cell.

The refinement of the tetramer was initiated with X-PLOR,at 10.0-3.0 A resolution. The starting crystallographic R-factorwas 34.7%. After one cycle of simulated annealing refinementstarting at 3000 K following the 'slow cool' protocol, the R-factor was 18.5%. At this point, the quality of the model wasexamined from the electron density maps using CHAIN, amodified version of FRODO [16] and also throughPROCHECK [17] outputs. Electron densities belonging tothe bound steroid molecules were identified. Refinement wasextended to include higher resolution data in steps of about2000-3000 additional reflections, through a series of positionalX-PLOR refinements. In the last step, simulated annealing at3000 K was performed again, using data in the resolutionrange 8.00-2.20 A. The R-factor at this stage was 21.7%.Loosely restrained non-crystallographic symmetry was usedamong four subunits throughout the refinement.

The entire molecule was carefully examined and problem areasrebuilt. A carbenoxolone molecule having the conformationobserved in a single crystal study was fitted into the densityobserved in each of the four substituted binding sites. Duringthe next cycle of refinement, four carbenoxolone moleculeswere included in the X-PLOR refinement. Only small shiftsin atomic positions of the inhibitors from the starting positionswere observed in refinement. The isotropic temperaturefactors of carbenoxolone atoms in the A-subunit were similar

980 Structure 1994, Vol 2 No 10

to those of the protein atoms in the immediate environment.In subunits B, C and D the electron density corresponding tothe inhibitor appeared weaker than that in subunit A. In orderto estimate the occupancy parameters for the steroid moleculesin the B, C and D subunits, their isotropic temperature factorswere fixed at the average temperature factor of the neigh-boring protein atoms, which were similar in all four subunits.In the next few cycles of positional refinement, the thermalparameters of all steroid atoms and the group occupancy ofthe carbenoxolone molecule in A-subunit were heldconstant, while the group occupancies of the inhibitors insubunits B, C and D were refined. The refinement of occu-pancies quickly converged. Subsequently, the isotropic tem-perature factors of carbenoxolone atoms were refined whileoccupancy parameters were kept fixed. The average B-factorsfor the carbenoxolone molecules in four subunits was 21 A2,while those for the protein atoms and water oxygens were20 A2 and 26 A2, respectively. The R-factor at the convergenceof the refinement was 20.3%.

A total of 89 well-defined solvent molecules (as wateroxygens) were included in the model at this time. Thesemolecules essentially were all first shell waters on the surface ofthe protein. A few of them were located at subunit interfaces.The final R-factor was 0.194. Table 1 provides a summary ofthe refinement statistics.

Quality of the modelAlthough Seri50 has well-defined electron density, it is in thedisallowed region of the Ramachandran plot [18] in all foursubunits, as it was in the holo structure [4]. Of 832 non-glycine and non-proline residues, 86.7% are in the mostfavored region of the plot. The electron density is weakest forresidues 187-200 for all subunits, as was observed in theholo-enzyme structure. Residues 1 and 212 could not belocated in the map, as in the holo-enzyme structure. Therandom positional error estimated from a Luzzati plot [19] ofthe final model is 0.25 A.

The coordinates for the complex structure are being depositedwith the Protein Data Bank, Brookhaven National Laboratory.

Acknowledgements: We wish to thank Dr Jane F Griffin for helpfuldiscussion and critical reading of the manuscript. This research isfunded by Grant No. DK26546 from the National Institutes ofHealth.

References1. White, I.H. & Jeffrey, J. (1972). Structural features of ring A and the

interaction of 20-oxosteroids with cortisone reductase. Eur. J.Biochem. 25, 409-414.

2. Edwards, C.A.F. & Orr, J.C. (1978). Comparison of the 3a- and 33-hydroxysteroid dehydrogenase activities of the cortisone reductaseof Streptomyces hydorgenans. Biochemistry 17, 4370-4376.

3. Ghosh, D., et al., & Orr, J.C. (1991). Three-dimensional structure ofholo 3a,2013-hydroxysteroid dehydrogenase: a member of the short-chain dehydrogenase family. Proc. Natl. Acad. Sci. USA 88,10064-10068.

4. Ghosh, D., Wawrzak, Z., Weeks, C.M., Duax, W.L. & Erman, M.(1994). The refined three-dimensional structure of holo 3,2013-hydroxysteroid dehydrogenase and possible roles of residuesconserved in short-chain dehydrogenases. Structure 2, 629-640.

5. Varughese, K.I., Skinner, M.M., Whiteley, J.M., Matthews, D.A. &Xuong, N.H. (1992). Crystal structure of rat liver dihydropteridinereductase. Proc. Natl. Acad. Sci. USA 89, 6080-6084.

6. Chen, Z., Jiang, J.C., Lin, Z.-G., Lee, W.R., Baker, M.E. & Chang,S.H. (1993). Site specific mutagenesis of Drosophila alcohol dehy-drogenase: evidence for involvement of tyrosine-152 and lysine-156in catalysis. Biochemistry 32, 3342-3346.

7. Albalat, R., Gonzalez-Duarte, R. & Atrian, S. (1992). Protein engi-neering of Drosophila alcohol dehydrogenase. The hydroxyl groupof Tyr152 is involved in the active site of the enzyme. FEBS Lett.308, 235-239.

8. Ensor, C.M. & Tai, H.-H. (1991). Site-directed mutagenesis of theconserved tyrosine 151 of human placental NAD+-dependent 15-hydroxy-prostaglandin dehydrogenase yields a catalytically inactiveenzyme. Biochem. Biophys. Res. Commun. 176, 840-845.

9. Obeid, J. & White, P.C. (1992). Tyr179 and Lys183 are essential forenzymatic activity of 11 [3-hydroxysteroid dehydrogenase. Biochem.Biophys. Res. Commun. 188, 222-227.

10. Sweet, F. & Samant, B.R. (1980). Bifunctional enzyme activity at thesame active site: study of 3 and 2013 by affinity alkylation of3a,201-hydroxysteroid dehydrogenase with 1 7-(bromoacetoxy)-steroids. Biochemistry 19, 978-986.

11. Ghosh, D., Erman, M., Pangborn, W., Duax, W.L. & Baker, M.E.(1992). Inhibition of Streptomyces hydrogenans 3,2013-hydroxy-steroid dehydrogenase by licorice-derived compounds and crystal-lization of an enzyme-cofactor-inhibitor complex. . SteroidBiochem. Mo/. Biol. 42, 849-853.

12. Peskar, B.M., Holland, A. & Peskar, B.A. (1976). Effect of carben-oxolone on prostaglandin synthesis and degradation. J. Pharm.Pharmacol. 28, 146-148.

13. Monder, C., Stewart, P.M., Lakshmi, V., Valentino, R., Burt, D. &Edwards, C.R.W. (1989). Licorice inhibits corticosteroid 111 3-hydroxysteroid dehydrogenase of rat kidney and liver: in vivo and invitro studies. Endocrinology 125, 349-354.

14. Persson, B., Krook, M. & Jornvall, H. (1991). Characteristics of short-chain alcohol dehydrogenases and related enzymes. Eur. J.Biochem. 200, 537-543.

15. Bronger, A.T. (1992). X-PLOR (Version 3.1) Manual. YaleUniversity, New Haven, CT.

16. Jones, T.A. (1978). A graphics model building and refinementsystem for macromolecules. J. App. Crystallogr. 11, 268-272.

17. Laskowski, R.A., MacArthur, M.W., Moss, D.S. & Thornton, J.M.(1993). PROCHECK: a program to check the stereochemical qualityof protein structures. J. App. Crystallogr. 26, 283-291.

18. Ramachandran, G.N. & Sasisakharan, V. (1968). Conformation ofpolypeptides and proteins. Adv. Protein Chem. 23, 283-437.

19. Luzzati, V. (1952). Traitment statistique des erreurs dans la deter-mination des structures cristallines. Acta Crystallogr. 5, 802-810.

20. Evans, S.V. (1993). SETOR: hardware lighted three-dimensionalsolid model representation of macromolecules. J. Mol. Graphics11,134-138.

Received: 25 ul 1994; revisions requested: 19 Aug 1994;revisions received: 26 Aug 1994. Accepted: 30 Aug 1994.

Table 1. Statistics of 3x,20[-HSD-carbonoxolone complex from thefinal cycle of X-PLOR refinement.

No. of protein atoms 7368Steroid atoms 164Water oxygens 89

Resolution range (A) 8.0-2.20No. of reflections used (F2> F2) 30996R-factor for data used 0.194Average B-factor (A2) 19.8Rms deviations

Bond lengths (A) 0.012Bond angles () 1.80Dihedral angles (°) 24.9Improper angles () 1.65

Ramachandran plot statisticsResidues in most favored region (%) 86.7Average rms deviation of main-chain atoms from 222 symmetry (A) 0.49