Interactions between Transhydrogenase and Thio-nicotinamide Analogues of NAD(H) and NADP(H) Underline the Importance of Nucleotide Conformational Changes in Coupling to Proton Translocation*

Transhydrogenase couples the reduction of NADP+ by NADH to inward proton translocation across mitochondrial and bacterial membranes. The coupling reactions occur within the protein by long distance conformational changes. In intact transhydrogenase and in complexes formed from the isolated, nucleotide-binding components, thio-NADP(H) is a good analogue for NADP(H), but thio-NAD(H) is a poor analogue for NAD(H). Crystal structures of the nucleotide-binding components show that the twists of the 3-carbothiamide groups of thio-NADP+ and of thio-NAD+ (relative to the planes of the pyridine rings), which are defined by the dihedral, Xam, are altered relative to the twists of the 3-carboxamide groups of the physiological nucleotides. The finding that thio-NADP+ is a good substrate despite an increased Xam value shows that approach of the NADH prior to hydride transfer is not obstructed by the S atom in the analogue. That thio-NAD(H) is a poor substrate appears to be the result of failure in the conformational change that establishes the ground state for hydride transfer. This might be a consequence of restricted rotation of the 3-carbothiamide group during the conformational change.

Transhydrogenase is found in the inner membrane of animal mitochondria and in the cytoplasmic membrane of bacteria. The enzyme provides NADPH for biosynthesis and for reduction of glutathione, and in some mammalian tissues, it probably participates in the regulation of flux through the tricarboxylic acid cycle (1,2). Under most physiological conditions transhydrogenase is driven in the "forward" direction by the proton electrochemical gradient (⌬p) generated by respiratory (or photosynthetic) electron transport.
There is general agreement that coupling between the redox reaction and proton translocation is mediated by changes in protein conformation, although the character of these conformational changes is not known (reviewed in Refs. [3][4][5]. Coupling mechanisms that involve large conformational changes operating over considerable distances are emerging as a common feature in proteins that translocate solutes/ions across membranes, and the amenable properties of transhydrogenase make it an attractive model in the search for fundamental principles. The enzyme has three components. The dI component, which binds NAD ϩ and NADH, and the dIII component, which binds NADP ϩ and NADPH, are extrinsic proteins protruding from the membrane (on the matrix side in mitochondria and on the cytoplasmic side in bacteria), and dII spans the membrane. The enzyme is essentially a "dimer" of two dI-dII-dIII "monomers," although the polypeptide composition is variable among species. Crystal structures of Rhodospirillum rubrum dI (6,7), bovine dIII (8), human dIII (9,10), and R. rubrum dI 2 dIII 1 complex (11,12), and an NMR structure of R. rubrum dIII (13) have recently been published. Studies on the transient state kinetics of transhydrogenation reveal that the redox reaction between the two nucleotides is direct (14,15). Thus, the nicotinamide and dihydronicotinamide groups are brought into apposition to allow transfer of a hydride ion equivalent between the C-4 positions of the rings. The reaction is stereo-specific for the pro-R (A-side) of NAD(H) and the pro-S (B-side) of NADP(H) (16,17).
Recent interpretations of kinetic and structural work on transhydrogenase have focused on the importance of conformational changes in the nucleotides as well as in the protein (3). It may be possible to test these interpretations through experiments using nucleotide analogues. Thio-NAD(H) and thio-NADP(H), which have a 3-carbothiamide substituent in place of the 3-carboxamide of the pyridine/dihydropyridine ring, have been used extensively in the study of soluble "dehydrogenases" (18). The binding properties of the analogues can be different relative to those of the physiological nucleotides and the catalytic rate can be affected; both increases and decreases have been observed in different enzymes (19). An x-ray structure of dihydrofolate reductase (DHFR) 1 with bound thio-NADP ϩ showed how small distortions of the nucleotide conformation can lead to pronounced effects on catalysis (20).
Because the absorbance band of its reduced form is redshifted relative to that of the physiological substrate (and therefore has minimal spectral overlap with NADH), thio-NADP ϩ has often been used to monitor the activity of proton-translocating transhydrogenase (21), but few experiments have been carried out using thio-nicotinamide analogues with a view to elucidating mechanistic details of the enzyme. In this report we compare hydride transfer rates to thio-NAD ϩ and to thio-NADP ϩ with those to NAD ϩ and NADP ϩ , respectively, in both the intact R. rubrum transhydrogenase and in dI 2 dIII 1 complexes. It emerges that thio-NAD ϩ is a poor substrate in the dI site, but thio-NADP ϩ is a good substrate in the dIII site. To attempt to explain these differences, we have solved the crystal structures of human dIII in its thio-NADP ϩ form (for comparison with dIII.NADP ϩ ; Protein Data Base 1DJL) and of the R. rubrum dI 2 dIII 1 complex loaded with thio-NAD ϩ and NADP ϩ (for comparison with complex loaded with NAD ϩ and NADP ϩ ; Protein Data Base 1HZZ). The increased van der Waals' radius of the S atom in the carbothiamide group (1.9 Å compared with 1.4 Å of the O atom) and the increased length of the C ϭ S bond (1.65 Å compared with 1.25 for C ϭ O) have only quite subtle effects on the conformation of the bound nucleotides and the arrangement of the side chains of invariant amino acids at the binding site. We explain the results in terms of the structural changes at the catalytic center that are required to bring together the nicotinamide and dihydronicotinamide rings during hydride transfer, and we discuss the conclusions in the context of the suggestion that these structural changes are associated with proton translocation.

EXPERIMENTAL PROCEDURES
Recombinant dI and dIII (wild type and the E155W mutant) from R. rubrum transhydrogenase and human heart dIII were expressed in Escherichia coli and purified by column chromatography as described in Refs. 22-25. After supplementing with 25% glycerol, they were stored at Ϫ20°C. Thawed proteins were either used directly in experiments or were first concentrated in Vivaspin centrifugal filters (5-kDa cut-off for dIII and 10-kDa cut-off for dI). Protein concentrations (given with respect to subunits) were determined by the microtannin procedure (26). Complexes (dI 2 dIII 1 ) of dI and dIII are generated spontaneously (K d Ͻ 60 nM (27)) upon mixing the two components in solution.
The dIII proteins are normally isolated in their NADP ϩ -bound forms. Where required, the NADP ϩ was replaced by NADPH as described in Ref. 15. To replace with thio-NADP ϩ , human dIII and R. rubrum dIII were first washed in Vivaspin filters with 10 mM Tris-HCl, pH 8.0, 1 mM dithiothreitol, 4 M NADP ϩ . The protein (ϳ35 mg ml Ϫ1 ) was then incubated in 10 mM Tris-HCl, pH 8.0, 1 mM dithiothreitol, 10 mM thio-NADP ϩ at 4°C for 1 h, a period sufficient to permit release of all tightly bound NADP ϩ (24). In crystallization experiments and in measurements of the cyclic reaction (see below), this solution was used directly. In stopped flow experiments and in measurements of nucleotide release, the solution was washed again in 10 mM Tris-HCl, pH 8.0, 1 mM dithiothreitol, and 50 M thio-NADP ϩ .
Everted cytoplasmic membranes (chromatophores) were isolated from phototrophically grown cultures of wild-type R. rubrum strain S1 and from a transhydrogenase-overexpressing strain RTB2 (28) by French pressing the cells as described in Ref. 29. The bacteriochlorophyll concentration was determined using the in vivo extinction coefficient of 140 mM Ϫ1 cm Ϫ1 at 880 nm (30). Where indicated, the dI component was washed from the membranes by centrifugation in the absence of NADP(H) (22). Reconstitution with recombinant dI protein was achieved by simple mixing.
Assays of steady state transhydrogenation were performed at 25°C on a Perkin Elmer Lambda 16 double-beam spectrophotometer using extinction coefficients given in Refs. 21 and 31. The rate of the reverse reaction (compare Equation 1) with physiological nucleotides was determined in the presence of an NADPH-regenerating system comprising 6 g ml Ϫ1 NADP-linked isocitrate dehydrogenase (Sigma I2002) and 4.0 mM isocitrate. Absorbance changes in the transient state were recorded at 20°C using an Applied Photophysics DX17-MV in its absorbance mode; the mixing time of the instrument was 1.31 ms (15,31). Protein fluorescence was measured using a Spex FluoroMax in the time drive mode. Heteronuclear single quantum coherence experiments (32) on 15 N-labeled protein were performed on a Bruker AMX500 NMR spectrometer, essentially as described in Ref. 33.
Human dIII in its thio-NADP ϩ form was crystallized essentially as described for the protein in its NADP ϩ form (10). Diffraction data were collected at beamline ID14 -1 at the European Synchrotron Radiation Facility (Grenoble, France) using a MAR-CCD detector. The crystals were flash-cooled to 100 K in the cryostream immediately prior to data collection. A complete data set was collected to 2.4 Å. The R. rubrum dI 2 dIII 1 complex in its thio-NAD ϩ /NADP ϩ form was crystallized essentially as described previously for (dI.Q132N) 2 dIII 1 mutant protein in its NAD ϩ /NADP ϩ form (11,12). A complete data set was collected to 2.6 Å on an ADSC detector on beamline ID 14 -2 also at the European Synchrotron Radiation Facility. In both cases, the crystals were isomorphous with those loaded with the physiological substrates. The data were integrated and scaled using the programs MOSFLM (34) and SCALA (35). The wild-type structures of dIII with NADP ϩ bound and dI 2 dIII 1 with NAD ϩ /NADP ϩ bound were refined against the structure factor amplitudes of dIII with thio-NAD ϩ bound and dI 2 dIII 1 with NAD ϩ /NADP ϩ bound, respectively, using the program CNS (36,37). The refinement statistics are summarized in Table I. Simulated annealing omit maps, both f o Ϫ f c and 2f o Ϫ f c , calculated using CNS, confirmed the calculated positions of the atoms in the carbothiamide groups of the two structures. Further confirmation of the sulfur atom positions was obtained from test refinements using various combinations of the parameter files for the physiological and analogue nucleotides and the reflection files from the structures. The information for the sulfur atom positions was found to reside predominantly in the reflection files of the thio-NAD(P) files for both structures. The cut-off values for hydrogen bond determination were 2.35-3.2 Å. Ribbon diagrams were prepared using the programs MOLSCRIPT (38) and TURBO-FRODO (39). The human dIII.thio-NADP ϩ structure and R. rubrum dI 2 dIII 1 with bound thio-NAD ϩ and NADP ϩ appear as Protein Data Bank entries 1PT9 and 1PTJ, respectively. The values are percentages of amino acids in the "core," "allowed," "generously allowed," and "disallowed" regions, respectively, according to the definition given in Ref. 53. e 5% of data have been set aside for cross-validation calculations.
Amino acid residues in R. rubrum dI and dIII are numbered according to their position in the recombinant proteins, as in Ref. 11. Residues in the human dIII are numbered according to the sequence of the intact enzyme, as in Ref. 10. The dihedral angle X am describes the twist of the carboxamide (or carbothiamide) group relative to the plane of the pyridine ring of a nicotinamide nucleotide. It is defined by viewing atoms C-2-C-3-C-7-O-7 (or S7) along C-3 3 C-7; it is 0°when O-7 (or S7) is perfectly cis to C-2, positive for a clockwise rotation of the carboxamide (carbothiamide) from this value and negative for an anticlockwise rotation. X n is the dihedral that describes the twist of the nicotinamide ring relative to the ribose ring across the glycosidic bond; it is defined by the atoms O-4 -C-1-N-1-C-2 and is 0°when O-4 is cis to C-2. When X n is between 0 and 180°, the rings are said to be in a syn conformation, and when X n is between 0 and Ϫ180°they are anti. The dihedral angles were calculated using TURBO-FRODO. All of the nucleotides and nucleotide analogues were obtained from Sigma.

Thio-NAD ϩ Is a Poor Substrate in the Reverse and Cyclic
Reactions Catalyzed by R. rubrum Transhydrogenase-The steady state rates of NADPH oxidation by NAD ϩ , by thio-NAD ϩ , and by AcPdAD ϩ (all "reverse" transhydrogenation reactions) in R. rubrum strain RTB2 membranes at saturating nucleotide concentrations, using the assay buffer described in the legend of Fig. 1, were 8.5, 4.5, and 16.0 mol mol Ϫ1 bacteriochlorophyll min Ϫ1 , respectively. However, the rates of these reactions do not closely reflect events at the hydride transfer step because reverse transhydrogenation is at least partly limited by slow NADP ϩ release (40). "Cyclic" transhydrogenation (Scheme 1) more reliably indicates the rate of hydride transfer because it can proceed without NADP ϩ (or NADPH) leaving the enzyme (41). Fig. 1 shows that the maximum rate of cyclic reduction of thio-NAD ϩ by NADH plus NADPH (Scheme 1A) was only 5.5 mol mol Ϫ1 bacteriochlorophyll min Ϫ1 , whereas the maximum rate of cyclic AcPdAD ϩ reduction by NADH plus NADPH (Scheme 1B) under equivalent conditions was 130 mol mol Ϫ1 bacteriochlorophyll min Ϫ1 . The left-hand arms of these two cyclic reactions (that is, the reduction of bound NADP ϩ by NADH in Scheme 1, A and B) are identical. Therefore, the oxidation of bound NADPH by AcPdAD ϩ in intact transhydrogenase is much faster than that by thio-NAD ϩ (the right-hand arms). Despite the large difference in rate, the K m for thio-NAD ϩ is only about 2.5 times greater than that for AcPdAD ϩ in the respective reactions. We were unable to detect any oxidation of thio-NADH by NADP ϩ (a forward transhydrogenation) in R. rubrum membranes under either darkened or illuminated conditions. Mixtures of isolated, purified dI and dIII of R. rubrum transhydrogenase spontaneously form stable dI 2 dIII 1 complexes (11,22,27). The steady state rates of oxidation of NADPH by NAD ϩ , by thio-NAD ϩ , and by AcPdAD ϩ catalyzed by dI 2 dIII 1 complexes were all very similar (ϳ2 mol mol Ϫ1 dIII min Ϫ1 ) but, even more than in the intact enzyme, these reverse transhydrogenations are limited by the very low rate of product NADP ϩ release (23). Cyclic reduction of thio-NAD ϩ by NADH plus NADPH catalyzed by dI 2 dIII 1 complexes (140 mol mol Ϫ1 dIII min Ϫ1 ; Scheme 1A) was considerably slower than cyclic reduction of AcPdAD ϩ by NADH plus NADPH (typically 2000 -3000 mol mol Ϫ1 dIII min Ϫ1 (12, 23); Scheme 1B). Following the same arguments as above, this indicates that, as in the intact enzyme, thio-NAD ϩ is a very poor acceptor of hydride equivalents from NADPH.
Experiments in the stopped flow spectrophotometer provide complementary information. It was previously shown that mixing NADPH-loaded dI 2 dIII 1 complexes with AcPdAD ϩ leads to a rapid burst of hydride transfer preceding the steady state reaction (15); the burst arises because the binding of AcPdAD ϩ , hydride transfer, and release of AcPdADH are all very fast relative to the rate of NADP ϩ release. Subsequently, measurements of changes in Trp fluorescence revealed an equivalent rapid burst of reaction between NADPH-loaded dI 2 dIII 1 complexes and NAD ϩ (42,43). In the experiment shown in Fig. 2, NADPH-loaded dI 2 dIII 1 complexes were mixed in the stopped flow spectrophotometer with thio-NAD ϩ . A burst of reaction was observed but with a much slower rate than that observed with either NAD ϩ or AcPdAD ϩ as hydride acceptors. Approximately similar concentrations of AcPdAD ϩ , NAD ϩ , and thio-NAD ϩ were required to give the maximum rate constants for the respective reactions (k app was ϳ550 s Ϫ1 for NADPH 3 AcPdAD ϩ , ϳ600 s Ϫ1 for NADPH 3 NAD ϩ , and ϳ8 s Ϫ1 for NADPH 3 thio-NAD ϩ ). In a subsequent experiment, dIII loaded with NADPH from one syringe was mixed with dI plus thio-NAD ϩ (1.0 mM after mixing) from the other. Again the burst kinetics were observed and with a k app ϭ ϳ6 s Ϫ1 (data not shown), proving that the slow rate of reaction is not a result of slow binding of thio-NAD ϩ to dI. The analysis of the k app values in terms of their microscopic rate constants, for AcPdAD ϩ and NAD ϩ as hydride acceptors, was discussed previously (15).
The transient state kinetics of forward transhydrogenation on dI 2 dIII 1 complexes with AcPdADH/NADP ϩ and with NADH/NADP ϩ were described in earlier works (31,42,43). These reactions also take place as a rapid single-turnover burst, here the slow steady state rate resulting from limiting NADPH release. For AcPdADH/NADP ϩ , measured from the 375 nm absorbance change at saturating AcPdADH, k app ϭ ϳ90 s Ϫ1 , and for NADH/NADP ϩ , measured from a Trp fluorescence change under continuous flow conditions at saturating NADH, k app ϭ ϳ21000 s Ϫ1 . When the dI 2 dIII 1 complex loaded with NADP ϩ was mixed with thio-NADH in the stopped flow spectrophotometer, a single-turnover burst of reaction was observed, but it was considerably slower than with either AcP-dADH or NADH as hydride donor (k app ϭ ϳ0.7 s Ϫ1 at saturating thio-NADH).
Thio-NADP ϩ Is a Good Substrate in the Forward Reaction of R. rubrum Transhydrogenase-The use of thio-NADP ϩ as an analogue for NADP ϩ in the forward reaction (Equation 1) catalyzed by transhydrogenases from several sources is well documented (21). The steady state rate of the reaction catalyzed by wild-type R. rubrum chromatophores with saturating concentrations of nucleotides in dark conditions is typically 0.1 mol mol Ϫ1 bacteriochlorophyll min Ϫ1 . With saturating photosynthetic illumination, the rate increases in the order of 10-fold as a result of the increased proton electrochemical gradient. Be- cause the steady state rate of this reaction is probably limited, at least partly, by thio-NADPH release from the enzyme (compare the reverse reaction, above and see Ref. 40), we monitored variants of steady state cyclic transhydrogenation and transient state forward transhydrogenation to compare the rates of hydride transfer to NADP ϩ and to thio-NADP ϩ in dI 2 dIII 1 complexes.
First, Fig. 3 shows that the rate of cyclic reduction of AcP-dAD ϩ by NADH in the presence of bound thio-NADP(H) (Scheme 1C) was only slightly less than that in the presence of bound NADP(H) (Scheme 1B). This proves that neither onenzyme hydride transfer from NADH 3 thio-NADP ϩ nor onenzyme hydride transfer from thio-NADPH 3 AcPdAD ϩ is substantially slower than the respective reactions with NADP ϩ and NADPH.
Second, the transient state kinetics of thio-NADP ϩ reduction by NADH and by AcPdADH on dI 2 dIII 1 complexes were investigated in the stopped flow spectrophotometer. With both hydride donors a rapid burst of thio-NADPH formation preceded the very slow steady state reaction (Fig. 4). This points to a kinetic mechanism similar to that observed with other nucleotides: rapid binding of NADH (or AcPdADH), rapid hydride transfer, and slow release of thio-NADPH. At close to saturating concentrations of AcPdADH (100 -200 M) the dominant (and faster) kinetic component in the burst had a k app value of ϳ200 s Ϫ1 , which compares with a k app of ϳ90 s Ϫ1 for AcPdADH 3 NADP ϩ (31). Equivalently, the increase in the k app for the burst of thio-NADP ϩ reduction with the initial concentration of NADH was similar to that seen for the burst of NADP ϩ reduction (the latter measured from the change in dIII Trp fluorescence in the E155W mutant (43)). In neither set of experiments was there any indication of saturation by NADH SCHEME 1. The cyclic reaction of transhydrogenase. E represents an enzyme catalytic site at the interface between dI and dIII. The dIII nucleotidebinding site is shown to be permanently occupied by either NADP ϩ /NADPH (A and B) or thio-NADP ϩ /thio-NADPH (C). The two double-headed arrows in each panel show consecutive events at the catalytic site. For example, in A, at the left arrows, NADH binds (to dI) and reduces the (dIII-bound) NADP ϩ , and NAD ϩ then dissociates; at the right arrows, thio-NAD ϩ binds (to dI) and oxidizes the (dIIIbound) NADPH, and thio-NADH then dissociates. The experiments were carried out under similar conditions, but the dIII protein was pretreated to exchange its NADP ϩ for thio-NADP ϩ (see "Experimental Procedures"), and the NADP ϩ in the assay buffer was replaced with thio-NADP ϩ . The temperature was 25°C. The curves through the data are not meant to imply a known or modeled relationship.
up to the limit of resolution of the instrument (k app ϭ ϳ800 s Ϫ1 , reached at 100 -200 M nucleotide). The data show that the rates of hydride transfer from NADH to NADP ϩ and to thio-NADP ϩ are similar.
Rate of Release of Thio-NADP ϩ from the dIII Component of R. rubrum Transhydrogenase-Isolated dIII is locked in an "occluded state" resembling an intermediate in turnover of the intact enzyme (3). An important property of the occluded state is its slow rate of exchange of bound NADP(H) with nucleotide in the solvent. We have therefore investigated the rate of thio-NADP ϩ release from dIII. The E155W mutant of R. rubrum dIII has very similar kinetic and thermodynamic properties to wild-type dIII, but fluorescence from its unique Trp residue is sensitive to the redox state of bound nucleotide (24). Thus, the fluorescence emission from dIII.NADP ϩ is 25% higher than that from dIII.NADPH, and this fact can be used to determine the occupancy of the binding site. In the present experiments, it was observed that Trp fluorescence from dIII.thio-NADP ϩ is even lower than that from dIII.NADPH and that the Trp fluorescence of dIII.thio.NADPH is about 10% lower than that of dIII.thio-NADP ϩ (data not shown). The experiment illustrated by the upper trace in Fig. 5 was performed with dIII.E155W presaturated with thio-NADP ϩ . Following a short period of preincubation, during which the rates of thio-NADP ϩ release and rebinding were equalized, NADPH was added to the protein solution. The initial, prompt fluorescence decrease was due to inner filtering by the nucleotide. The subsequent slow increase in Trp fluorescence was the result of replacement of the bound thio-NADP ϩ by NADPH. Separately it was established that the NADPH used in the experiment was in excess, and under these conditions, the fluorescence increase gives the first order rate constant for thio-NADP ϩ release (see scheme and compare with Refs. 27 and 44). The calculated value (k off ϭ 0.028 s Ϫ1 ) is similar to that determined for NADP ϩ release (k off ϭ 0.022 s Ϫ1 ). In a complementary experiment (Fig. 5, lower  trace), excess thio-NADP ϩ was added to dIII.E155W in its NADP ϩ form. Following the rapid, initial, inner filtering effect, the decrease in fluorescence is attributed to the replacement of the physiological nucleotide by the analogue. The rate constant for the fluorescence decrease corresponds to that for NADP ϩ release; the calculated value (k off ϭ 0.030 s Ϫ1 ) is indeed similar to that determined following an NADPH pulse (27).
Perturbation in the NMR Spectrum of R. rubrum dIII by Thio-NADP ϩ and Thio-NADPH-In the HSQC experiment, the 1 H and 15 N spins of amide groups in a protein are correlated. Almost all the peaks in the 1 H, 15 N HSQC spectrum of R. rubrum dIII.NADP ϩ are now assigned (13). The HSQC spectrum of dIII.thio-NADP ϩ differed from that of dIII.NADP ϩ only in amides that are spatially close to the nicotinamide ring of the nucleotide (Table II), especially those in the "nicotinamide binding loop" between strand ␤2 and helix B. This indicates that the only changes in the protein structure caused by substituting the physiological nucleotide with the analogue are in the locality of the thio-nicotinamide ring.
Chemical shift changes are observed in the HSQC spectra of R. rubrum (33) and E. coli dIII (45) when bound NADP ϩ is replaced by NADPH. The changes, mapped onto the high resolution structures of the R. rubrum protein, reveal that magnetic perturbations (atomic displacements or charge redistributions or both (46)) not only take place in the vicinity of the nicotinamide ring but also extend into helix D/loop D and loop E of the protein, and they might reflect events associated with the gating mechanism of transhydrogenase (10). Chemical shift changes in the 1 H, 15 N HSQC spectrum of 15 N-labeled dIII, consequent upon the reduction of bound thio-NADP ϩ by addition of a low concentration of unlabelled R. rubrum dI protein plus NADH, are listed in Table III. Reduction was not complete (ϳ30%), but clearly, the amino acid residues whose amide groups are affected are the same as those affected by substitution of NADP ϩ by NADPH (33).
Crystal Structures of Isolated Transhydrogenase Components with Bound Thio-analogues of Nicotinamide Nucleotides-To try to understand why thio-NAD ϩ is a poor substrate for transhydrogenase but thio-NADP ϩ is a good substrate, we have determined high resolution structures of isolated components of the enzyme in different nucleotide-bound states. Human dIII in its thio-NADP ϩ form was crystallized under conditions similar to those described for the protein in its NADP ϩ form (9, 10), and its structure was solved by x-ray diffraction. The fold of dIII.thio-NADP ϩ is very similar to that of dIII.NADP ϩ (root mean square difference of C ␣ ϭ 0.3 Å). Briefly, the protein adopts a Rossmann fold; it is a six stranded parallel ␤ sheet flanked by helices, organized into two ␤␣␤␣␤ motifs. As in the classical Rossmann fold, the NADP ϩ /thio-NADP ϩ is bound in a crevice at the C-terminal end of the ␤ sheet, but the orientation of the nucleotide is reversed relative to that found in other structures; the adenine moiety is located over the second ␤␣␤␣␤ motif, and the nicotinamide mononucleotide is located over the first. The nicotinamide and thionicotinamide rings in the respective structures are bound by the loop between strand ␤2 and helix B (Fig. 6). A striking difference between the two structures is that, in dIII.NADP ϩ , the 3-carboxamide group is approximately coplanar with the pyridine ring, the oxygen atom being trans to C-2 (X am ϭ 179°), but in dIII.thio-NADP ϩ , the 3-carbothiamide is twisted relative to the pyridine plane (X am ϭ 140°). However, the hydrogen bond between N-7 of the thio-nicotinamide and the Ala 923 carbonyl group is preserved, and the pyridine ring of the thio-NADP ϩ is still maintained in a similar position in the binding loop to that of the NADP ϩ pyridine ring; contacts between the ring atoms and side chains of amino acid residues in the loop move only slightly. Importantly, the C-4 atom moves no more than 0.3 Å from the polypeptide backbone, and the si face of the ring is exposed to the solvent in the same way as the physiological nucleotide. It will be discussed below that during hydride transfer the si face of the NADP ϩ (thio-)nicotinamide at C-4 is presented to the pro-R hydrogen at the C-4 atom of the NADH dihydronicotinamide. The confinement of structural changes to the vicinity of the nicotinamide ring in the crystalline state is nicely consistent with the limited changes in the amide chemical shifts observed in solution experiments by NMR (Table II). We have also determined the x-ray structure of the complex formed from a mixture of isolated dI and dIII components of R. rubrum transhydrogenase (compare Ref. 11), but in crystals grown in the presence of a combination of thio-NAD ϩ and NADP ϩ , two oxidized nucleotides, ensure that hydride transfer does not take place during crystallization. Again the overall fold was not affected by the substitution of NAD ϩ by thio-NAD ϩ (root mean square difference of C ␣ ϭ 0.3 Å). The complexes are dI 2 dIII 1 heterotrimers. The two symmetrically organized dI polypeptides (A and B) are each composed of two domains (dI.1 and dI.2) that are separated by deep clefts and linked by two long helices. Both dI.1 and dI.2 comprise mostly parallel ␤-sheets flanked by helices and have the form and connectivity of the Rossmann fold. Only the cleft of dI(B) is associated with a dIII polypeptide. The fold of the dIII polypeptide and the conformation of its bound NADP ϩ are very similar to those seen in the structures of the isolated dIII.NADP ϩ of dIII upon reducing bound NADP ϩ and bound thio-NADP ϩ HSQC spectra of dIII⅐NADP ϩ , dIII⅐NADPH, dIII⅐thio-NADP ϩ , and dIII⅐thio-NADPH were recorded as described under "Experimental Procedures" and "Results." The chemical shift perturbation (see Table II) was measured for (NADP ϩ Ϫ NADPH) and for (thio-NADP ϩ Ϫ thio-NADPH) (the latter are described as tNADP ϩ and tNADPH in the table). Only residues with perturbations Ͼ50 Hz either for NADP ϩ Ϫ NADPH or tNADP ϩ Ϫ tNADPH are listed. Data for (NADP ϩ Ϫ NADPH) are similar to those presented previously (33) but include additional assignments in dIII⅐NADPH. NA, not assigned.  mammalian enzymes (see above). As observed in the R. rubrum dI 2 dIII 1 complex crystallized with NAD ϩ and NADP ϩ and discussed in Refs. 27, 33, and 42, there is good electron density for (thio-)NAD ϩ only in the A polypeptide of the new structure; the binding site is located at the C-terminal ends of the strands in the ␤-sheet of dI.2. The adenosine moieties of the NAD ϩ and thio-NAD ϩ bind in the same way. The nicotinamide and thionicotinamide rings occupy quite similar positions in the binding pocket, but there are differences of detail (Fig. 7). Most obviously, the dihedral angles, X am , signifying the twist of the 3-carboxamide group relative to the pyridine ring, and X n , the rotation of the pyridine ring relative to the nicotinamide ribose, are both altered in the dI(A)-bound thio-NAD ϩ . Note that the orientation of the 3-carboxamide group of NAD ϩ in our earlier structure is unusual with the oxygen atom approximately cis to C-2 of the pyridine ring (X am ϭ Ϫ47°; see "Discussion"). The S atom of the 3-carbothiamide group in the new structure is also cis to C-2 and (in contrast to the situation with thio-NADP ϩ in dIII; see above) is twisted toward the plane of the pyridine ring (X am ϭ Ϫ2°). The dihedral angle X n for thio-NAD ϩ (Ϫ124°) and that for NAD ϩ (Ϫ145°) both fall in the range that defines the anti conformation relative to the nicotinamide ribose group. The largest movement of the C-4 atom of the thio-nicotinamide ring relative to the polypeptide backbone is 0.5 Å.

DISCUSSION
It is instructive to compare the crystal structures of DHFR and dIII from transhydrogenase. In both DHFR.NADP ϩ and human dIII.NADP ϩ the carboxamide group of bound nucleotide is in the trans position, approximately coplanar with the pyridine ring (X am ϭ ϳ180°). However, when these two proteins are loaded with thio-NADP ϩ , the carbothiamide group of the analogue in both cases is twisted relative to the plane of the pyridine ring, although the twist has the opposite sense: X am is Ϫ160°in DHFR and ϩ140°in dIII. The increased twist of the carbothioamide group in DHFR was suggested to avoid unfavorable contact between the sulfur atom and the CH group at position 4 of the pyridine ring (20). It was noted to be less than in a model compound, 2-methyl-4(thiocarbamoyl)-pyridine (47). McTigue et al. (20) concluded that a larger rotation in DHFR is prohibited by the binding site geometry of the protein and that the unfavorable interactions between the CH group of pyridine ring and the S atom are not completely relieved. The oppositely directed twist of the carbothiamide group in dIII (Fig. 6) might similarly result from this kind of structural compromise; the tendency of the group to twist from a coplanar position to prevent unfavorable contact between the large S atom and C4H is limited by the constraints imposed by interactions between the nucleotide and the protein-binding pocket.
The changes in the nucleotide-binding site in dI(A) resulting from the substitution of NAD ϩ with thio-NAD ϩ in the R. rubrum dI 2 dIII 1 complex have a different character. They appear to be dominated by unfavorable interactions between the bulky carbothiamide S atom and the protein-binding pocket, especially the side chain of Gln 132 . The carbothiamide group is twisted away from Gln 132 into the plane of the pyridine ring, evidently incurring the penalty of a close contact between the S atom and C2H. The torsion angle of the glycosidic bond is distorted, the Gln 132 side chain is displaced, there is enforced close contact (2.8 Å) between the thioamide-S atom and the Arg 127 carbonyl, and the single hydrogen bond that links the carboxamide group of NAD ϩ with Ile 128 is broken (Fig. 7).
In the crystal structure of a ternary complex of DHFR, the rotated sulfur atom in the thio-NADP ϩ results in destabilization of the bound biopterin; the latter is moved away from the nicotinamide ring, and its temperature factors are increased relative to those in the NADP ϩ structure (20). This suggested an explanation for the finding that thio-NADPH is a very poor coenzyme for DHFR; in the formation of the transition state for hydride transfer, the mutual approach of the nicotinamide ring and the substrate pteridine ring are impeded by the sulfur atom of the thioamide. In contrast to DHFR, thio-NADP ϩ is a good substrate for intact transhydrogenase and its isolated components (1). The rapid rate of the cyclic reaction in R. rubrum dI 2 dIII 1 complexes and of the single turnover burst of thio-NADP ϩ reduction by NADH and by AcPdADH show that the hydride transfer step is functioning normally (2). The reduction of thio-NADP ϩ by NADH in suspensions of membranes was stimulated upon generation of a proton electrochemical gradient by photosynthetic electron transfer, showing that the reaction is well coupled to proton translocation (3). The rates of release of NADP ϩ and thio-NADP ϩ from isolated dIII, thought to be locked in the occluded state, were very similar (4). The chemical shift changes in isolated dIII accompanying the reduction of thio-NADP ϩ are very similar to those resulting from NADP ϩ reduction; the change in magnetization through helix D/loop D and loop E, thought to be related to a gating step in the enzyme, is not substantially affected. It appears that neither the increased atomic radius of the sulfur atom nor the increased twist of the carbothiamide group compromise the behavior of the nucleotide in transhydrogenation. Whether hydride transfer occurs entirely by way of an over-the-barrier mechanism or whether there is a contribution from quantum mechanical tunneling is not clear (15), but it is evident that the C-4 atoms of the dihydropyridine and pyridine ring systems (for example, of NADH and NADP ϩ ) must approach one another to facilitate the reaction, and we conclude that the bulky sulfur atom of the analogue despite its rotated position does not prevent this approach.
The explanation as to why, in contrast, thio-NAD(H) is a poor substrate for transhydrogenase (in both the intact enzyme and in dI 2 dIII 1 complexes) is more difficult to define, but driving force effects can probably be ruled out. There are differences in the standard redox potentials of the nucleotides used in this work, and these differences will lead to differences in the driving force at the hydride transfer step; in aqueous solution the E 0 Ј values of thio-NAD ϩ /thio-NADH, AcPdAD ϩ /AcPdADH, and NAD ϩ /NADH are Ϫ0.285, Ϫ0.247, and Ϫ0.320 volt, respectively (18) (the presence of a 2Ј-phosphate group on the adenosine ribose does not significantly affect E 0 Ј). However, pairs of experiments with the dI 2 dIII 1 complex under comparable, single-turnover conditions show that these differences do not account for the observation that thio-NAD ϩ is a poor hydride acceptor, and thio-NADH is a poor hydride donor. Thus, in reverse transhydrogenation, the driving force on the reaction NADPH 3 thio-NAD ϩ is greater than that on NADPH 3 NAD ϩ (calculated from the solution E 0 Ј values), but the rate of reaction is much slower. Equivalently, in forward transhydrogenation, the driving force on thio-NADH 3 NADP ϩ is greater than that on AcPdADH 3 NADP ϩ , but again the rate of the reaction is much slower.
To understand the behavior of nucleotide in the dI site of transhydrogenase, it has to be recognized that there are probably changes in protein and nucleotide conformation in this site that precede and follow the hydride transfer reaction (3,11,12). The conformation of the A polypeptide and its bound nucleotide in the dI 2 dIII 1 complex (Fig. 7) probably corresponds to that in an "open state" of the intact enzyme. This is the state in which nucleotide reactants bind and products dissociate during turnover, but it is important that hydride transfer is prevented in this state. After NADH and NADP ϩ binding (e.g. during forward transhydrogenation), the conformation is driven by protonation/deprotonation reactions associated with proton translocation into an occluded state in which hydride transfer does proceed. We suggest that the low reactivity of thio-NAD(H) results from an obstruction of the conformational changes occurring during interconversion of the open and occluded states. When NAD(H) from the A polypeptide of the dI 2 dIII 1 complex is modeled into the B polypeptide to anticipate the pretransition state for hydride transfer to NADP ϩ in dIII, the dihydro-and nicotinamide rings are in a "distal" conformation; the C-4 atoms of the nucleotides are too far apart to allow hydride transfer (11). A switch to a "proximal" nucleotide position is required to bring the C-4 atoms into apposition in the occluded state. The crystal structures of isolated dI show that there is considerable flexibility in the conformation of NAD(H) in its binding site (6,7) and that the distal-to-proximal switch can be achieved by rotating the nicotinamide mononucleotide moiety toward the NADP ϩ bound to dIII (3,11,12). The conformation of thio-NAD ϩ seen in the A-polypeptide of the new structure is also in the distal position and will have to be driven (by movements associated with changes in the width of the cleft between dI.1 and dI.2) into the proximal position to permit hydride transfer.
A specific suggestion as to why the conversion from the distal to the proximal position might be restricted with thio-NAD(H) can be deduced from an unusual feature in the conformation of NAD(H) in dI. In crystals of the lithium salt of NAD ϩ dihydrate (48) (and related nucleotides in the Cambridge Structural Database) the oxygen atom of the 3-carboxamide group is cis to C-2 of the nicotinamide ring (X am ϭ ϳ0°). A preference for the cis conformation of this group is also revealed in theoretical studies (49 -51). However, in the majority of protein crystal structures that have bound nicotinamide nucleotides, the oxygen is approximately trans to C-2 (X am ϭ ϳϮ180°); the conclusion is usually based on the pattern of hydrogen bond donors and acceptors to the carboxamide group because few structures have been determined at a high enough resolution to show it directly. Of fifty unique NAD(P)(H)-binding proteins (not including transhydrogenase dI) deposited most recently in the Protein Data Bank, only three had nucleotide in the cis conformation; of these, two had an hydrogen bond organization that did not exclude the possibility of a trans conformation and the resolution of the other was probably not good enough to discriminate. The nucleotide in the dI site of transhydrogenase is one of the few clear exceptions to the rule. Thus, the probable hydrogen bond between the 3-carboxamide group of NAD ϩ and the carbonyl of Ile128 in the R. rubrum dI 2 dIII 1 complex (11), and of NADH and the equivalent atom in the dI.NADH struc-ture (7), suggests the unusual cis conformation (the Ile carbonyl can only be a hydrogen bond acceptor and the carboxamide-NH 2 group would have to be the donor). Note that the 3-carboxamide of NAD ϩ in polypeptides A and D of isolated dI (6), was arbitrarily set in the commonly observed trans position, although the hydrogen bonds to the carboxylate of Asp 135 in the former do not discriminate against a cis conformation, and there are no hydrogen bonds in the latter to determine an orientation (the electron density in polypeptides B and C, on the other hand, is too weak to define the nicotinamide position and conformation). The finding that the carboxamide is probably in the cis conformation in at least some of the transhydrogenase structures is all the more remarkable in view of the results of ab initio molecular orbital calculations on transition state models of hydride transfer between nicotinamide and dihydronicotinamide (52). It was shown that X am can affect the rate of reaction, notably with a trans organization lowering the energy of the transition state. Although the modeling studies assume a symmetrical transition state structure that is unlikely in the enzyme, they would appear to indicate that the cis conformation of the 3-carboxamide group of NAD(H), seen in current structures of transhydrogenase, is relatively unfavorable for the redox chemistry. We therefore suggest that the C-3-C-7 bond rotates into a trans form during the distal to proximal conformational change. The calculated barrier to rotation is only a few kcal mol Ϫ1 for both nicotinamide and 1,4-dihydronicotinamide (50,51). The device could be important to help prevent hydride transfer until the appropriate configuration is reached, thus minimizing slippage in the coupling to proton translocation. The proposed swiveling of the C-3-C-7 bond might also help to facilitate, or to steer, the (dihydro-) nicotinamide ring between the hydrogen bonding organizations of the distal and the proximal positions. The hypothesis would explain why thio-NAD(H) is a poor substrate for transhydrogenase. Equivalently to NAD ϩ , the S atom of thio-NAD ϩ is cis to C-2 in the structure of the dI 2 dIII 1 complex (this is evident from the stronger electron density of the S atom rather than the hydrogen bond pattern). If movement from the distal to the proximal position requires rotation of the C-3-C-7 bond from cis to trans, then we can appreciate that this will be restricted with thio-NAD(H) because the barrier to rotation would be increased by the bulkier sulfur atom. The limited rate of formation of the proximal arrangement of nucleotides would lead to inhibition of hydride transfer.
The 3-carboxamide oxygen of NADP ϩ in dIII is trans to C-2 in the observed x-ray structures of isolated dIII (Fig. 6) and the dI 2 dIII 1 complex, and, according to the ab initio MO studies (52), this is the favored conformation for hydride transfer. However, conformational changes in this site are also expected during turnover, but on the basis that thio-NADP ϩ is a good substrate for transhydrogenase, these changes are not expected to involve equivalent rotations in the C-3-C-7 bond of the nucleotide.