Probing the Active Site of Human Aldose Reductase SITE-DIRECTED MUTAGENESIS OF AND

Structural models of human aldose reductase com- plexed with NADPH have revealed the apposition of C4 of the nicotinamide ring with tyrosine 48 anr, histidine 110, suggesting that either of these residues could function as the proton donor in the reaction mechanism. Tyrosine 48 is also part of a hydrogen-bonding network that includes lysine 77 and aspartate 43. In order to study the potential catalytic roles of these 4 residues, we evaluated the kinetic properties of mutants containing structurally conservative replacements at these sites. type. Tyr-48 act- ing the acid-base catalyst in aldose reductase and confirm the importance of Asp-43, Lys-77, and His- 110 to the

dose reductase is found, the action of aldose reductase has been linked to the development of certain long-term complications of diabetes mellitus such as cataract, retinopathy, nephropathy, and neuropathy (reviewed in Ref. 1). Administration of aldose reductase inhibitors has been shown to delay onset or substantially prevent these complications in experimental animals (1)(2)(3)(4). These results have sparked intense study of the enzyme since the treatment of diabetic patients with aldose reductase inhibitors may represent a therapeutic route for preventing the onset or progression of diabetic complications in patients with chronic hyperglycemia.
Aldose reductases from various human and animal tissues have been purified and extensively studied. The enzyme is a monomer with a calculated molecular mass of about 35,800 Da.
Amino acid and nucleotide sequence comparisons of aldose reductases from several mammalian species have revealed their evolutionary relatedness with a family of oxidoreductases and structurally related proteins including steroid dehydrogenases (5, 61, prostaglandin synthases (7), lens crystallins (8), and xenobiotic metabolizing enzymes (9). Although heterogeneity has been described for aldose reductases extracted from different tissue donors (10) and from diabetic as compared to nondiabetic tissues (111, cDNA cloning studies have shown that aldose reductase expressed in different human tissues is the product of a single gene (12).
Despite some controversy over the kinetic mechanism of aldose reductase, the recent steady state and pre-steady state kinetic studies by Kubiseski et al. (13) have established that, at pH 7.0 and with DL-glyceraldehyde as the aldehyde substrate, the enzyme follows a sequential ordered mechanism in which NADPH binds before the aldehyde substrate and NADP+ is released after the alcohol product. A conformational change in the enzyme occurs both upon NADPH binding and prior to release of NADP+. The latter isomerization is reported to be the rate-limiting step in the forward reaction (13, 14). The physical basis for this isomerization was recently identified in crystal structures of aldose reductase complexed with NADPH (15, 16), as compared to the structure lacking NADPH (17), to be a hinge motion of the loop of residues Gly-213 to Leu-227 which helps to hold the coenzyme in place. Two other members of the aldoketo oxidoreductase superfamily, aldehyde reductase (18)(19)(20) and 3a-hydroxysteroid dehydrogenase (211, have also been shown to obey a sequential ordered reaction mechanism similar to that of aldose reductase. Until recently, very little was known about the nature of the catalytic center or any other structural features that may be important for this enzyme's function. Despite numerous chemical modification studies to determine the potential involvement of arginine, histidine, cysteine, or lysine residues in the catalytic mechanism, no conclusive results have been obtained (22)(23)(24). The only exception to the above were the chemical modi-25687 fication studies of human psoas muscle aldose reductase, where the involvement of lysine 262 in NADPH binding was demonstrated by affinity labeling with pyridoxal phosphate and pyridoxal 5'-diphospho-5'-adenosine (25). With the advent of crystal structures for porcine (17) and human (15,16) aldose reductases, together with systems to overexpress cloned human aldose reductase in Escherichia coli (26), systematic and rational site-directed mutagenesis studies to probe specific features of the enzyme's structure and function are now possible.
The catalytic mechanism of aldose reductase is thought to be relatively simple, involving a stereospecific transfer of the pro-R hydride from C4 of the nicotinamide (27) to the substrate carbonyl carbon atom and protonation of the substrate carbonyl oxygen atom by an enzyme functional group. The identity of the proton donor in this mechanism could not be ascertained from the crystal structures, although Tyr-48 was considered the most likely candidate based on its proximity to the hydridedonating nicotinamide C4 atom of NADPH and its hydrogenbonding interaction with other residues likely to facilitate its function as a proton donor (15). The possibility that His-110 could be the proton donor was also considered, but since it is found in a predominantly hydrophobic area, an environment not considered conducive to the formation of a positively charged imidazolium group, it was deemed less likely (15). In the present study, we individually replaced "-48 and His-110 with Phe and Asn, respectively, by site-directed mutagenesis in anticipation that the resulting mutants (Y48F and H11ON)' would reveal which one of these amino acid residues is the true proton donor. We also constructed point mutants at two other locations in the enzyme, namely at Asp-43 (resulting in the D43N mutant) and Lys-77 (K77M mutant), because these 2 residues participate in a hydrogen-bonding network linking the nicotinamide ribose 2"hydroxyl group with the hydroxyl of Tyr-48.

MATERIALS AND METHODS
Site-directed Mutagenesis-Human placental aldose reductase cDNA (containing a silent mutation introduced to abolish an internal NcoI site at a position corresponding to codon 143) was excised from pBLUE-SCRIPT I1 SK(-) phagemid (Stratagene) by digestion with Hind111 and NcoI. Plasmid pHuALR2-1 was created by ligating the resulting 1354base pair fragment with pMON20,400, a derivative of the plasmid expression vector described previously (28). The resulting pHuALR2-1 plasmid differs from our previous aldose reductase expression construct (26) by the inclusion of a M13 origin of replication and provision of a gene confemng ampicillin rather than spectinomycin resistance. Following the procedure of Kunkel et al. (291, mutations were introduced into the wild type ( W T ) cDNA sequence using single-stranded template DNA derived from pHuALR2-1. Oligonucleotide primers used to construct mutants are shown in Table I. Mutagenic primers were synthesized and purified as described previously (26). Clones containing the intended mutation were identified by nucleotide sequence analysis across the mutation site. Integrity of the entire coding sequence of the mutants was then confirmed by the same sequencing procedure using a set of eight internal sequencing primers.
Expression in E. coli and Purification-Plasmids encoding WT and mutant aldose reductases were transfected into E. coli strain JMlOl (30), and their respective protein products were expressed and purified under conditions described previously (26), except that cultures were maintained in the presence of ampicillin (50 pg/ml) rather than spectinomycin. Briefly, the enzyme was extracted from host cells by osmotic shock and purified by ammonium sulfate fractionation (50-80%), chromatofocusing on PBE 94 (Pharmacia LKB Biotechnology Inc.), and hydroxylapatite chromatography (Bio-Rad). All purification buffers contained 1 m~ dithiothreitol. For the mutants with reduced enzymatic activity, chromatography column fractions containing mutant reductases were identified by prominent protein bands co-migrating with purified WT aldose reductase on SDS-polyacrylamide gels. The    amide gel electrophoresis (31) followed by staining with Coomassie Brilliant Blue R-250. Protein concentrations were estimated by the Bradford method (32) using bovine serum albumin (Stratagene) as a standard.
Determination ofDissociation Constants-The dissociation constants for various binary enzyme-coenzyme complexes were determined by measuring the quenching of the intrinsic protein fluorescence (excitation, 288 nm; emission, 342 n m ) using a Perkin-Elmer MPF-66 fluorescence spectrophotometer. Titrations were camed out by sequential addition of 10-pl aliquots of a concentrated coenzyme solution (0.03-0.3 m~) to a 1-cm path length quartz cuvette containing mutant or WT aldose reductase (0.25-1.5 m) in 100 m~ potassium phosphate buffer, pH 7.0 at 15 "C, to a final volume of 2 or 3 ml. Glycyl-tryptophan standard solution was used to correct for dilution and inner filter effects. All exogenous thiol was removed from the protein solution immediately prior to the experiment by rapid gel filtration using Econo-Pac lODG columns (Bio-Rad). The Cleland package of kinetics software (33) was used to fit the data to a rectangular hyperbola described by Equation l (34): where A F is the difference in fluorescence signal obsellred in the presence and absence of coenzyme (L). The maximal change in fluorescence, A F , , , and the dissociation constant, Kd, were the derived quantities.
Enzyme Assays-During purification, WT aldose reductase activity was monitored spectrophotometrically at 340 nm using 5 m~ m-glyceraldehyde and 0.15 m~ NADPH as substrate and coenzyme, respectively (35). Kinetic constants for the WT enzyme were determined in a triple buffer system consisting of 25 nm MES, 25 m~ potassium phosphate, and 90 n m Tris at pH 7.0,0.15 nm NADPH and variable amounts of the appropriate substrate. For the mutant enzymes, assays were performed in the same triple buffer system, pH 5.0-10.1. Enzyme concentrations in the assays were 0.03-2.8 mg/ml for the active site mutants and about 2 pg/ml for the WT enzyme. Apparent K , and kc,, values were determined by fitting initial rate data obtained from duplicate assays to a rectangular hyperbola as described previously (26). At least two separate batches of each enzyme were analyzed.

RESULTS
Expression and Purification-The four newly constructed aldose reductase genes that code for D43N, Y48F, K77M, and HllON were overexpressed in E. coli strain JM 101 shakerflask cultures and purified to virtual homogeneity following the same procedure as described previously for the WT enzyme (26). All four mutant enzymes eluted upon chromatofocusing at positions, relative to other bacterial proteins, similar to that of the WT enzyme. On hydroxylapatite column chromatography, the mutant and WT enzymes exhibited similar elution profiles, appearing as single, distinct, protein peaks. Although the mutant enzymes exhibited reduced or no catalytic activity, their purification was easily followed by the appearance of a pre- four mutant enzymes exhibited dramatically altered kinetic behavior as compared to that of the WT enzyme. When evaluated with variable amounts of benzaldehyde, D-glyceraldehyde, L-glyceraldehyde, and p-nitrobenzaldehyde as substrates a t pH values ranging from 5.4 to 10.1, the activity of the Y48F mutant was not detectable. Similarly, no activity was detectable in the reverse direction with glycerol or benzyl alcohol. This indicated a decrease in enzymatic activity exceeding 3 orders of magnitude relative to the WT enzyme. AB a consequence of this dramatic decrease in activity, it was not possible to determine the K, and k , , values for this mutant using our assay system. The other three mutants had significantly decreased, but measurable, enzymatic activities.
Kinetic analysis of the HllON mutant using Dglyceraldehyde as the variable s u b strate was carried out at pH 5.4 (pH optimum for this mutant,  efficiency (k,&,) for this s u b strate was approximately 1.1 x 1O8-fold lower than WT. Kinetic constants using p-nitrobenzaldehyde, benzaldehyde, and bglyceraldehyde as substrates could not be accurately determined because sufficiently high substrate concentrations could not be achieved in our asmy system. No activity was detected in the reverse direction with glycerol or benzyl alcohol. At pH 7.0, the specific activity of the K77M mutant (1.4 milliunitdmg and 2.6 milliunitdmg with D and bglyceraldehyde, respectively) is about 1,460t o 700-fold lower than that for the WT enzyme (specific activity of WT is 2,050 milliunite/mg and 1,800 milliunitdmg with D and ~glyceraldehyde, respectively; 1 unit of enzyme oxidizes 1 pmol NADPW min). Using our enzyme assay (see Waterials and Methods"), we could not achieve conditions under which saturating amounts of NADPH could be maintained during the entire   able with either benzyl alcohol or glycerol in the reverse direction. The apparent kinetic constants for reduction and oxidation of various substrates by the D43N mutant are compared to those of the WT enzyme in Table 11. The apparent K,,, values for the three substrates in the forward reaction were increased 10.5to 16.7-fold when compared to the WT enzyme. The apparent turnover numbers (kcat) for the various substrates were decreased by 7.6-to 15.3-fold when compared to the WT enzyme. Likewise, the apparent catalytic efficiencies (kcatfK,,,) for those substrates have dropped approximately 80-to 176-fold, compared to those of the WT enzyme. Similar changes were observed in the kinetic constants with NADPH as the variable substrate: K,,, increased &fold, kmt decreased over 15-fold, and the catalytic efficiency dropped 77-fold. In the reverse reaction with benzyl alcohol as the variable substrate, the kinetic constants have changed similarly: K,,, increased over 30-fold, k,, decreased 2.6-fold, and the catalytic efficiency dropped 78-fold.
Coenzyme Binding Affinities-The dissociation constants (Kd) for NADPH and NADP' with the various enzyme forms are shown in Table 111 was the most likely candidate based on sequence conservation among the members of the enzyme superfamily and an interaction (hydrogen bonding with Lys-77 and Asp-43) that is likely to depress its PK, (Fig. 2). The other amino acid residues that could potentially function as the proton donor are Cys-298 and His-110 since they, in addition to Tyr-48, are the only residues within a 5 to 6 b; radius from the nicotinamide C4 that have side chains with the ability to donate a proton. We have recently ruled out 2 I. Tarle and J. M. Petrash, manuscript in preparation.
Cys-298 as the proton donor since Ala (data not shown) and Ser (26) mutants at that position show robust reductase activity. Our kinetic data in the present paper are consistent with Tyr-48 being the proton donor in the catalytic mechanism. Replacement of Tyr-48 with Phe resulted in an essentially inactive mutant, with 25000-fold reduction in enzymatic activity as compared with that of the WT enzyme (sensitivity limit of assay system). On the other hand, replacement of His-110 with Asn results in a much smaller decrease in activity, as evidenced by a 14-fold decrease in kcat.
This result may be somewhat surprising in light of the expected acidity of Tyr and His side chains: Tyr hydroxyl groups normally have a pK, of about 10 while His imidazolium side chains have pK, values in the 6-7 range. These pK, values would implicate His as the likely proton donor at physiologic pH values. However, in light of the known crystal structure of aldose reductase, it seems reasonable that the acidity of Tyr-48 might be altered. Tyrosine 48 is part of a three-membered hydrogen-bonding network that also includes Lys-77 and Asp-43 (Fig. 2). Lysine 77 donates a hydrogen bond through its side chain amino group to Tyr-48 hydroxyl oxygen (3.19 b; distance between the N and 0 atoms). A similar interaction with 2 Lys residues at the active site of a Class C 0-lactamase from Citrobacter freundii (37) appears to lower the p K , of Tyr-150, which acts as a general acid-base in the catalytic mechanism of this 0-lactamase. The presence of adjacent positive charges has been shown to cause a substantial decrease in PK, values of tyrosine copolymers (38) and in phenol-containing crown ethers complexed with ammonium ions (39). The D43N mutant is informative because numerous kinetic parameters can be measured relatively accurately. There appears to be good agreement between a &fold increase in K,-(NADPH) and a 5.7-fold increase in Kd(NADpH) consistent with the position of Asp-43 near the 2'-hydroxyl of the nicotinamide ribose of NADPH. That mutation ofhp-43 affects both the K, and kcat of aldehyde and alcohol substrates in both catalytic directions supports its indirect involvement in substrate binding and catalysis. The chemical (hydride transfer) step is not rate-limiting in the catalytic mechanism of native aldose reductase (13,14). Accordingly, the question arises as to whether the 7.6-to 15.5-fold decrease in kcat for the D43N mutant in the forward reaction results from a decrease in the rate of the demonstrated slow step in the kinetic mechanism, i.e. the isomerization of E*.NADP+ + E.NADF'+ (13,14), or from a larger decrease in the rate of the hydride transfer step such that the chemical step now becomes rate-limiting. If the latter scenario is correct, such a significant drop in the rate of hydride transfer may mean that Asp-43 plays an important role in facilitating the interaction between Lys-77 and Tyr-48. The increase in Km(o-g1yceraldehyde) in either scenario can be explained by a change in the precise orientation of the catalytic residues in this mutant, resulting in somewhat lower affinity of the enzyme for substrate.
Substmte Modeling in the Active Site-There is now s a lcient information available to construct a reasonable model of how a substrate might interact with the active site of aldose reductase. From the crystal structures, we know the detailed geometry of the active site residues and the coenzyme (15, 16). From the results presented in this paper, we postulate that the proton donor in the forward reaction (reduction of aldehyde) is the phenolic hydroxyl group of Tyr-48. From previous biochemical results, we know which hydrogen is transferred from NADPH to the substrate (the C4 pro-R hydrogen, Ref. 27) and which face of the substrate carbonyl receives this hydrogen (the re face, Ref. 41).3 In summary, we know the location of the reducing hydride, the acidic proton, and grossly how the substrate carbonyl must be oriented between the two.
Both a b initio ( 4 2 4 ) and semiempirical (45, 46) theoretical calculations on the hydride transfer reactions of dihydronicotinamides provide additional constraints. The atoms referred to below are nitrogen 1 (N1) and carbon 4 ((24) of the nicotinamide ring of NADPH, the pro-R hydrogen at C4 (H4R), the substrate carbonyl carbon (C) and oxygen (01, and the phenolic oxygen of the proton donor Tyr-48 (OH). The calculations indicate that in the transition state: l) the nicotinamide ring puckers slightly toward the substrate, placing H4R in a pseudoaxial conformation; 2) the hydride transfers in a bent mode, such that the C4-H4R-C angle is about 150"; 3) the hydride approaches the backside of the carbonyl, such that the H4R-C-0 angle is about 110"; and 4) the C4-C distance is about 2.6 A. Point 3 is in accord with studies of crystal structures wherein nucleophiles are closely apposed to carbonyl groups (47). Given all of these constraints, the substrate D-glyceraldehyde was modeled into the aldose reductase active site. The coordinates for D-glyceraldehyde were generated with MACRO-MODEL. 4 The gross position of D-glyceraldehyde was first adjusted so that the aldehydic hydrogen pointed toward Trp-20, the glycol side chain [-CH(OH)CH20Hl pointed toward His-110, and the carbonyl oxygen atom pointed toward Tyr-48; this ensured that the re face of the carbonyl received the hydride from NADYH. Next, D-glyceraldehyde was translated to place C about 2.6 A from C4, simultaneously satisfying the angular constraints mentioned above. C was also constrained to lie in the plane formed by N1, C4, and H4R. D-Glyceraldehyde was then rotated about C, to position 0 about 2.7 from OH. This rotation also ensured that the glycol side chain did not approach His-110 too closely. Finally, the D-glyceraldehyde torsion angles were adjusted to favorable staggered conformations, which also allowed the glycol side chain to point out of the active site toward the bulk solvent.
Our model of D-glyceraldehyde in the aldose reductase ac- tive site is presented in Fig. 3. A number of salient features are evident. First, our model explains the carbonyl face specificity (re, Ref. 41) of aldose reductase. In order for the si face of the carbonyl to accept the hydride from NADPH, the glycol side chain and the aldehydic hydrogen of D-glyceraldehyde must be interchanged; this would cause the glycol side chain of D-glyceraldehyde to collide with the indole ring of Trp-20.
Similar reasoning explains why methyl ketones are much poorer substrates for aldose reductase than are aldehydes (48). Second, it is clear that the fit of D-glyceraldehyde in the active site is rather tight. This is consistent with the postulate that enzymes achieve rate acceleration by stabilizing the transition state of a reaction (49). Third, the side chain of the aldehyde can extend up out of the active site, without unfavorable interactions with other amino acid side chains in the active site. Other excellent substrates, such as p-nitrobenzaldehyde and isohydrocortisone (501, can be modeled into the active site without difficulty (results not shown). We also attempted to model D-glyceraldehyde into the active site, using Ne' of His-110 as the proton donor, rather than OH of Tyr-48. This alternate model (not shown) is substantially less satisfactory. In order to prevent the glycol side chain from colliding with the indole ring of Trp-20, C must be moved out of the N1-C4-H4R plane by 0.7 8. Larger substrates (e.g. p-nitrobenzaldehyde) would require a further unfavorable adjustment. Also, the angular constraints mentioned above are not as well satisfied by this alternate model.
Our findings may have relevance for understanding the mechanism of action of enzymes other than aldose reductase. Alignment of sequences of all known members of the aldo-keto oxidoreductase superfamily reveals that Asp-43, Lys-77, and His-110 are strictly conserved throughout the superfamily. 5rosine 48 is conserved in all but one case, that exception being the European bull frog pcrystallin (8). This strict conservation of the catalytic residues among the members of the oxidoreductase superfamily suggests that the catalytic mechanism of aldose reductase extends to the other members of the superfamily.