Kinetic properties of saquinavir-resistant mutants of human immunodeficiency virus type 1 protease and their implications in drug resistance in vivo

In order to study the basis of resistance of human immunodeficiency virus, type 1 (HIV-1), to HIV-1 protease inhibitor saquinavir, the catalytic and inhibition properties of the wild-type HIV-1 protease and three saquinavir resistant mutants, G48V, L90M, and G48V/L90M, were compared. The kinetic parameter kcat/Km was determined for these proteases using eight peptide substrates whose sequences were derived from the natural processing site sequences of HIV-1. The kcat/Km values were determined using conventional steady-state kinetics as well as initial velocities of mixed substrate cleavages under the condition where the substrate concentrations [S]o < Km. The independently determined kcat and Km values for some of the substrates confirmed the accuracy of the mixed-substrate method and also permitted the calculation in all cases of true rather than relative kcat/Km values. The Ki values were also determined. Using a previously described kinetic model [Tang, J., & Hartsuck, J. A. (1995) FEBS Lett. 367, 112-116], the relative processing activities of HIV-1 protease variants were estimated in the saquinavir concentration range of 0-10(-7) M. Although the protease activity of G48V, L90M, and G48V/L90M are only about 10, 7, and 3% of that of the wild-type HIV-1 protease in the absence of inhibitor, the resistance tendencies of the three mutants are clearly manifest by relatively less activity loss as inhibitor concentration becomes higher. Also, the ratios of the activities of the four protease species at certain saquinavir concentrations appear to correlate with the population ratios of the four protease species at different time points of clinical trials. This correlation suggests that the population ratio of the protease species is driven by in vivo saquinavir concentration, which appears to be in the range 10(-10)-10(-9) M during the clinical trials.     

Resistance to HIV protease inhibitors: a comparison of enzyme inhibition and antiviral potency

Resistance of HIV-1 to protease inhibitors has been associated with changes at residues Val82 and Ile84 of HIV-1 protease (HIV PR). Using both an enzyme assay with a peptide substrate and a cell-based infectivity assay, we examined the correlation between the inhibition constants for enzyme activity (Ki values) and viral replication (IC90 values) for 5 active site mutants and 19 protease inhibitors. Four of the five mutations studied (V82F, V82A, I84V, and V82F/I84V) had been identified as conferring resistance during in vitro selection using a protease inhibitor. The mutant protease genes were expressed in Escherichia coli for preparation of enzyme, and inserted into the HXB2 strain of HIV for test of antiviral activity. The inhibitors included saquinavir, indinavir, nelfinavir, 141W94, ritonavir (all in clinical use), and 14 cyclic ureas with a constant core structure and varying P2, P2' and P3, P3' groups. The single mutations V82F and I84V caused changes with various inhibitors ranging from 0.3- to 86-fold in Ki and from 0.1- to 11-fold in IC90. Much larger changes compared to wild type were observed for the double mutation V82F/I84V both for Ki (10-2000-fold) and for IC90 (0.7-377-fold). However, there were low correlations (r2 = 0.017-0.53) between the mutant/wild-type ratio of Ki values (enzyme resistance) and the mutant/wild-type ratio of viral IC90 values (antiviral resistance) for each of the HIV proteases and the viruses containing the identical enzyme. Assessing enzyme resistance by "vitality values", which adjust the Ki values with the catalytic efficiencies (kcat/Km), caused no significant improvement in the correlation with antiviral resistance. Therefore, our data suggest that measurements of enzyme inhibition with mutant proteases may be poorly predictive of the antiviral effect in resistant viruses even when mutations are restricted to the protease gene.     

Counteracting HIV-1 protease drug resistance: structural analysis of mutant proteases complexed with XV638 and SD146, cyclic urea amides with broad specificities

The long-term therapeutic benefit of HIV antiretroviral therapy is still threatened by drug-resistant variants. Mutations in the S1 subsite of the protease are the primary cause for the loss of sensitivity toward many HIV protease inhibitors, including our first-generation cyclic urea-based inhibitors DMP323 and DMP450. We now report the structures of the three active-site mutant proteases V82F, I84V, and V82F/I84V in complex with XV638 and SD146, two P2 analogues of DMP323 that are 8-fold more potent against the wild type and are able to inhibit a broad panel of drug-resistant variants [Jadhav, P. K., et al. (1997) J. Med. Chem. 40, 181-191]. The increased efficacy of XV638 and SD146 is due primarily to an increase in P2-S2 interactions: 30-40% more van der Waals contacts and two to four additional hydrogen bonds. Furthermore, because these new interactions do not perturb other subsites in the protease, it appears that the large complementary surface areas of their P2 substituents compensate for the loss of P1-S1 interactions and reduce the probability of selecting for drug-resistant variants.     

In vivo resistance to a human immunodeficiency virus type 1 proteinase inhibitor: mutations, kinetics, and frequencies

Resistance to saquinavir (Ro 31-8959), an inhibitor of human immunodeficiency virus type I proteinase, was studied in peripheral blood mononuclear cell-derived proviral DNA from patients undergoing prolonged treatment. A Leu90-->Met exchange was the predominant resistance mutation in vivo; Gly48-->Val or doubly mutant virus was rarely observed. After 8-12 months of treatment with saquinavir alone (600 mg, 3 times/day) or in combination with zidovudine (200 mg, 3 times/day), approximately 45% of all patients carried provirus with mutant proteinase; the incidence was lower (22%) in patients treated with a combination of saquinavir, zidovudine, and dideoxycytidine. There was a good relationship between genotypic analysis of saquinavir resistance and data from virus assays, confirming that Leu90-->Met and Gly48-->Val are the essential exchanges in the proteinase that determine loss of sensitivity to this inhibitor. Absence of genotypic resistance correlated with a sustained decrease in plasma viral RNA. There was a positive correlation between a Met90 mutation and some residues at natural polymorphic sites (positions 10, 36, 63, and 71).     

Mutations in the reduced folate carrier gene which confer dominant resistance to 5,10-dideazatetrahydrofolate

L1210/D3 mouse leukemia cells are resistant to 5, 10-dideazatetrahydrofolate due to expansion of cellular folate pools which block polyglutamation of the drug (Tse, A., and Moran, R. G. (1998) J. Biol. Chem. 273, 25944-25952). These cells were found to have two point mutations in the reduced folate carrier (RFC), resulting in a replacement of isoleucine 48 by phenylalanine and of tryptophan 105 by glycine. Each mutation contributes to the resistance phenotype. Genomic DNA from resistant cells contained both the wild-type and mutant alleles, but wild-type message was not detected. Folic acid was a much better substrate, and 5-formyltetrahydrofolate was a poorer substrate for transport in L1210/D3 cells relative to L1210 cells. Enhanced transport of folic acid was due to a marked, approximately 20-fold, decrease in the influx Km. Influx of methotrexate and 5,10-dideazatetrahydrofolate were minimally altered. Transfection of mutated rfc cDNA into RFC-null L1210/A cells produced the substrate specificity and 5, 10-dideazatetrahydrofolate resistance observed in the L1210/D3 line. Transfection of the mutant cDNA into wild-type cells also conferred resistance to 5,10-dideazatetrahydrofolate. We conclude that the I48F and W105G mutations in RFC caused resistance to 5, 10-dideazatetrahydrofolate, that the region of the RFC protein near these two positions defines the substrate-binding site, that the wild-type allele was silenced during the multistep development of resistance, and that this mutant phenotype represents a genetically dominant trait.     

Impaired fitness of human immunodeficiency virus type 1 variants with high-level resistance to protease inhibitors

One hope to maintain the benefits of antiviral therapy against the human immunodeficiency virus type 1 (HIV-1), despite the development of resistance, is the possibility that resistant variants will show decreased viral fitness. To study this possibility, HIV-1 variants showing high-level resistance (up to 1,500-fold) to the substrate analog protease inhibitors BILA 1906 BS and BILA 2185 BS have been characterized. Active-site mutations V32I and I84V/A were consistently observed in the protease of highly resistant viruses, along with up to six other mutations. In vitro studies with recombinant mutant proteases demonstrated that these mutations resulted in up to 10(4)-fold increases in the Ki values toward BILA 1906 BS and BILA 2185 BS and a concomitant 2,200-fold decrease in catalytic efficiency of the enzymes toward a synthetic substrate. When introduced into viral molecular clones, the protease mutations impaired polyprotein processing, consistent with a decrease in enzyme activity in virions. Despite these observations, however, most mutations had little effect on viral replication except when the active-site mutations V32I and I84V/A were coexpressed in the protease. The latter combinations not only conferred a significant growth reduction of viral clones on peripheral blood mononuclear cells but also caused the complete disappearance of mutated clones when cocultured with wild-type virus on T-cell lines. Furthermore, the double nucleotide mutation I84A rapidly reverted to I84V upon drug removal, confirming its impact on viral fitness. Therefore, high-level resistance to protease inhibitors can be associated with impaired viral fitness, suggesting that antiviral therapies with such inhibitors may maintain some clinical benefits.     

Kinetic characterization of human immunodeficiency virus type-1 protease-resistant variants

Passage of human immunodeficiency virus type-1 (HIV-1) in T-lymphocyte cell lines in the presence of increasing concentrations of the hydroxylethylamino sulfonamide inhibitor VX-478 or VB-11328 results in sequential accumulation of mutations in HIV-1 protease. We have characterized recombinant HIV-1 proteases that contain these mutations either individually (L10F, M46I, I47V, I50V) or in combination (the double mutant L10F/I50V and the triple mutant M46I/I47V/I50V). The catalytic properties and affinities for sulfonamide inhibitors and other classes of inhibitors were determined. For the I50V mutant, the efficiency (kcat/Km) of processing peptides designed to mimic cleavage junctions in the HIV-1 gag-pol polypeptide was decreased up to 25-fold. The triple mutant had a 2-fold higher processing efficiency than the I50V single mutant for peptide substrates with Phe/Pro and Tyr/Pro cleavage sites, suggesting that the M46I and I47V mutations are compensatory. The effects of mutation on processing efficiency were used in conjunction with the inhibition constant (Ki) to evaluate the advantage of the mutation for viral replication in the presence of drug. These analyses support the virological observation that the addition of M46I and I47V mutations on the I50V mutant background enables increased survival of the HIV-1 virus as it replicates in the presence of VX-478. Crystal structures and molecular models of the active site of the HIV-1 protease mutants suggest that changes in the active site can selectively affect the binding energy of inhibitors with little corresponding change in substrate binding.     

Characterization of human immunodeficiency virus type 1 mutants with decreased sensitivity to proteinase inhibitor Ro 31-8959

A human immunodeficiency virus type 1 (HIV-1) variant with highly reduced susceptibility to Ro 31-8959, an inhibitor of the viral proteinase, has been selected by repeated passage of wild-type virus in CEM cells in the presence of increasing concentrations of the inhibitor. Peptide sequences of the proteinase of selected virus were obtained from proviral DNA. Sequence comparison to wild-type (wt) proteinase demonstrated two amino acid substitutions in the resistant virus, a Gly to Val exchange at position 48 and a Leu to Met exchange at position 90. Furthermore, sequences of intermediate passage virus suggest contributions from positions 12, 36, 57, and 63 in early steps of resistance development. The selected virus showed a ca. 40-fold increase in 50% inhibitory concentration of Ro 31-8959. Growth kinetics of resistant virus were comparable to wild-type virus and the resistant genotype proved to be stable in the absence of inhibitor. Directed mutagenesis of the HIV-1 HXB2 proteinase at positions 48 and 90 suggested that each mutation alone led to a moderate decrease in sensitivity of the recombinant virus to proteinase inhibitor. However, a recombinant virus carrying both mutations in the proteinase gene showed a significant reduction in its sensitivity to Ro 31-8959 thus proving the importance of these exchanges for the resistance phenotype.     

Overexpression and characterization of the chromosomal aminoglycoside 6'-N-acetyltransferase from Enterococcus faecium

The chromosomal gene aac(6')-Ii, encoding an aminoglycoside 6'-N-acetyltransferase in Enterococcus faecium, renders this organism resistant to moderate levels of many aminoglycoside antibiotics. The ubiquitous presence of aac(6')-Ii in E. faecium complicates the selection of antibiotics for treatment of infections caused by this organism. In view of the importance of this enzyme, we have initiated studies to gain an understanding of its molecular mechanism of acetyl transfer. The AAC(6')-Ii enzyme was overexpressed in Escherichia coli and purified in a simple three-step procedure which yields 55 mg of pure dimeric protein per liter of cell culture. Steady-state kinetic analyses revealed a broad substrate specificity and demonstrated that acetylation occurs exclusively at position N-6'. k(cat)/Km values were on the order of 10(4) M(-1) s(-1), which is relatively low compared to other aminoglycoside-modifying enzymes. In addition, MIC values were positively correlated with k(cat), the rate when the enzyme is saturated with the aminoglycoside substrate, and not with k(cat)/Km, the rate at low aminoglycoside (sub-Km) concentrations. These results describe an enzyme which is not optimally evolved for aminoglycoside inactivation and suggest that this chromosomally encoded enzyme may have an alternate physiological function.     

Role of specific amino acid residues in T4 endonuclease V that alter nontarget DNA binding

Endonuclease V is a pyrimidine dimer-specific DNA glycosylase-apurinic (AP)1 lyase which, in vivo or at low salt concentrations in vitro, binds nontarget DNA through electrostatic interactions and remains associated with that DNA until all dimers have been recognized and incised. On the basis of the analyses of previous mutants that effect this processive nicking activity, and the recently published cocrystal structure of a catalytically deficient endonuclease V with pyrimidine dimer-containing DNA [Vassylyev, D. G., et al. (1995) Cell 83, 773-782], four site-directed mutations were created, the mutant enzymes expressed in repair-deficient Escherichia coli, and the enzymes purified to homogeneity. Steady-state kinetic analyses revealed that one of the mutants, Q15R, maintained an efficiency (k(cat)/Km) near that of the wild-type enzyme, while R117N and K86N had a 5-10-fold reduction in efficiency and K121N was reduced almost 100-fold. In addition, K121N and K86N exhibited a 3-5-fold increase in Km, respectively. All the mutants experienced mild to severe reduction in catalytic activity (k(cat)), with K121N being the most severely affected (35-fold reduction). Two of the mutants, K86N and K121N, showed dramatic effects in their ability to scan nontarget DNA and processively incise at pyrimidine dimers in UV-irradiated DNA. These enzymes (K86N and K121N) appeared to utilize a distributive, three-dimensional search mechanism even at low salt concentrations. Q15R and R117N displayed somewhat diminished processive nicking activities relative to that of the wild-type enzyme. These results, combined with previous analyses of other mutant enzymes and the cocrystal structure, provide a detailed architecture of endonuclease V-nontarget DNA interactions.     

Three-dimensional structure of a mutant HIV-1 protease displaying cross-resistance to all protease inhibitors in clinical trials

Analysis of mutational effects in the human immunodeficiency virus type-1 (HIV-1) provirus has revealed that as few as four amino acid side-chain substitutions in the HIV-1 protease (M46I/L63P/V82T/I84V) suffice to yield viral variants cross-resistant to a panel of protease inhibitors either in or being considered for clinical trials (Condra, J. H., Schleif, W. A., Blahy, O. M., Gadryelski, L. J., Graham, D. J., Quintero, J. C., Rhodes, A., Robbins, H. L., Roth, E., Shivaprakash, M., Titus, D., Yang, T., Teppler, H., Squires, K. E., Deutsch, P. J., and Emini, E. A. (1995) Nature 374, 569-571). As an initial effort toward elucidation of the molecular mechanism of drug resistance in AIDS therapies, the three-dimensional structure of the HIV-1 protease mutant containing the four substitutions has been determined to 2.4-A resolution with an R factor of 17.1%. The structure of its complex with MK639, a protease inhibitor of the hydroxyaminopentane amide class of peptidomimetics currently in Phase III clinical trials, has been resolved at 2.0 A with an R factor of 17.0%. These structures are compared with those of the wild-type enzyme and its complex with MK639 (Chen, Z., Li, Y., Chen, E., Hall, D. L., Darke, P. L., Culberson, C., Shafer, J., and Kuo, L. C. (1994) J. Biol. Chem. 269, 26344-26348). There is no gross structural alteration of the protease due to the site-specific mutations. The C alpha tracings of the two native structures are identical with a root-mean-square deviation of 0.5 A, and the four substituted side chains are clearly revealed in the electron density map. In the MK639-bound form, the V82T substitution introduces an unfavorable hydrophilic moiety for binding in the active site and the I84V substitution creates a cavity (unoccupied by water) that should lead to a decrease in van der Waals contacts with the inhibitor. These changes are consistent with the observed 70-fold increase in the Ki value (approximately 2.5 kcal/mol) for MK639 as a result of the mutations in the HIV-1 protease. The role of the M46I and L63P substitutions in drug resistance is not obvious from the crystallographic data, but they induce conformational perturbations (0.9-1.1 A) in the flap domain of the native enzyme and may affect the stability and/or activity of the enzyme unrelated directly to binding.     

Molecular basis for the coupling ion selectivity of F1F0 ATP synthases: probing the liganding groups for Na+ and Li+ in the c subunit of the ATP synthase from Propionigenium modestum

The conserved glutamate residue at position 65 of the Propionigenium modestum c subunit is directly involved in binding and translocation of Na+ across the membrane. The site-specific introduction of the cQ32I and cS66A substitutions in the putative vicinity to cE65 inhibited growth of the single-site mutants on succinate minimal agar, indicating that both amino acid residues are important for proper function of the oxidative phosphorylation system. This growth inhibition was abolished, however, if the cF84L/cL87V double mutation was additionally present in the P. modestum c subunit. The newly constructed Escherichia coli strain MPC848732I, harboring the cQ32I/cF84L/cL87V triple mutation, revealed a change in the coupling ion specificity from Na+ to H+. ATP hydrolysis by this enzyme was therefore not activated by NaCl, and ATP-driven H+ transport was not affected by this alkali salt. Both activities were influenced, however, by LiCl. These data demonstrate the loss of the Na+ binding site and retention of Li+ and H+ binding sites within this mutant ATPase. In the E. coli strain MPC848766A (cS66A/cF84L/cL87V), the specificity of the ATPase was further restricted to H+ as the exclusive coupling ion. Therefore, neither Na+ nor Li+ stimulated the ATPase activity, and no ATP-driven Li+ transport was observed. The ATPase of the E. coli mutant MPC32N (cQ32N) was activated by NaCl and LiCl. The mutant ATPase exhibited a 5-fold higher Km for NaCl but no change in the Km for LiCl in comparison to that of the parent strain. These results demonstrate that the binding of Na+ to the c subunit of P. modestum requires liganding groups provided by Q32, E65, and S66. For the coordination of Li+, two liganding partners, E65 and S66, are sufficient, and H+ translocation was mediated by E65 alone.     

A triple mutation in the a subunit of the Escherichia coli/Propionigenium modestum F1Fo ATPase hybrid causes a switch from Na+ stimulation to Na+ inhibition

Previously we have shown that the Na+-translocating Escherichia coli (F1-delta)/Propionigenium modestum (Fo+delta) hybrid ATPase acquires a Na+-independent phenotype by the c subunit double mutation F84L, L87V that is reflected by Na+-independent growth of the mutant strain MPC8487 on succinate [Kaim, G., and Dimroth, P. (1995) J. Mol. Biol. 253, 726-738]. Here we describe a new class of mutants that were obtained by random mutagenesis and screening for Na+-independent growth on succinate. All six mutants of the new class contained four mutations in the a subunit (S89P, K220R, V264E, I278N). Results from site-specific mutagenesis revealed that the substitutions K220R, V264E, and I278N were sufficient to create the new phenotype. The resulting E. coli mutant strain MPA762 could only grow in the absence but not in the presence of Na+ ions on succinate minimal medium. This effect of Na+ ions on growth correlated with a Na+-specific inhibition of the mutant ATPase. The Ki for NaCl was 1. 5 mM at pH 6.5, similar to the Km for NaCl in activating the parent hybrid ATPase at this pH. On the other hand, activation by Li+ ions was retained in the new mutant ATPase. In the absence of Na+ or Li+, the mutant enzyme had the same pH optimum at pH 6.5 and twice the specific activity as the parent hybrid ATPase. In accordance with the kinetic data, the reconstituted mutant ATPase catalyzed H+ or Li+ transport but no Na+ transport. These results show for the first time that the coupling ion selectivity of F1Fo ATPases is determined by structural elements not only of the c subunit but also of the a subunit.     

Subcellular localization, substrate specificity and crystallization of duodenase, a potential activator of enteropeptidase

Duodenase, a serine protease from bovine duodenum mucosa, was located in endoplasmic reticulum, the Golgi secretory granules of epithelial cells and ducts of Brunner's glands by the A-gold immunocytochemical method. Duodenase exhibits trypsin-like and chymotrypsin-like specificities with a preference for substrates having lysine at the P1 and proline at the P2 positions. The kinetic constants for the hydrolysis of 21 potential duodenase substrates are reported. The best substrates were found to be alpha-N-tosylglycylprolyllysine 4-nitroanilide (k[cat]/Km of 35000 M[-1] s[-1]), alpha-N-succinylthreonylprolyllysine 4-nitroanilide (k[cat]/Km of 18000 M[-1] s[-1]) and alpha-N-serylprolyllysine 4-nitroanilide (k[cat]/Km of 2600 m[-1] s[-1]), all of which contain the P1-P3 sequence of the enteropeptidase zymogen/activation site. On the basis of its catalytic properties and sites of localization, duodenase has been postulated to be an activator of the enteropeptidase precursor. A tetradecapeptide (LVTQEVSPKIVGGS) having the P9-P5'sequence of the cleavage site of zymogen activation of bovine proenteropeptidase was synthesized, and kinetic parameters of its hydrolysis by duodenase were determined (Km of 87 microM; k[cat] of 1.4 s[-1]; k[cat]/Km of 16000 M[-1] s[-1]). Crystals of duodenase frozen in a stream of liquid nitrogen diffracted synchrotron X-rays to 0.2-nm resolution.     

Crystal structures of complexes of a peptidic inhibitor with wild-type and two mutant HIV-1 proteases

Crystal structures of the protease of human immunodeficiency virus type 1 (HIV-1) and two mutant proteases, V82D and V82N, have been determined. In all three cases the enzyme forms a complex with the peptidic inhibitor U-89360E. All structures have been determined to 2.3 A resolution and have satisfactory agreement factors: 0.173 for wild type, 0.175 for V82D, and 0.182 for V82N. Comparison of the three crystal structures provides explanations which are consistent with the known kinetic properties of these mutant enzymes with the U-89360E inhibitor [Lin, Y., Lin, X., Hong, L., Foundling, S., Heinrikson, R. L., Thaisrivongs, S., Leelamanit, W., Raterman, D., Shah, M., Dunn, B.M., & Tang, J. (1995) Biochemistry 34, 1143-1152]. Unfavorable van der Waals interactions between the inhibitor and the mutated side chains at position 82 are consistent with diminished affinity for the inhibitor by the mutant enzymes. If a mutation is potentially resistant to an inhibitor, the mutant enzyme should not only have an increased Ki for the inhibitor but should also preserve considerable catalytic capability. The V82D mutant possesses these qualities. In the V82D crystal structure, a water molecule, which connects the protease flap to the inhibitor, is missing or of low occupancy. Absence of this bridge may be important in determining catalytic capability. Moreover, mutation at position 82 induces change in two polypeptide backbone regions, 35-41 and 67-68, which may be related to protease flap mobility.     

Mutational analysis of the three cysteines and active-site aspartic acid 103 of ketosteroid isomerase from Pseudomonas putida biotype B

In order to clarify the roles of three cysteines in ketosteroid isomerase (KSI) from Pseudomonas putida biotype B, each of the cysteine residues has been changed to a serine residue (C69S, C81S, and C97S) by site-directed mutagenesis. All cysteine mutations caused only a slight decrease in the k(cat) value, with no significant change of Km for the substrate. Even modification of the sulfhydryl group with 5,5'-dithiobis(2-nitrobenzoic acid) has almost no effect on enzyme activity. These results demonstrate that none of the cysteines in the KSI from P. putida is critical for catalytic activity, contrary to the previous identification of a cysteine in an active-site-directed photoinactivation study of KSI. Based on the three-dimensional structures of KSIs with and without dienolate intermediate analog equilenin, as determined by X-ray crystallography at high resolution, Asp-103 was found to be located within the range of the hydrogen bond to the equilenin. To assess the role of Asp-103 in catalysis, Asp-103 has been replaced with either asparagine (D103N) or alanine (D103A) by site-directed mutagenesis. For D103A mutant KSI there was a significant decrease in the k(cat) value: the k(cat) of the mutant was 85-fold lower than that of the wild-type enzyme; however, for the D103N mutant, which retained some hydrogen bonding capability, there was a minor decrease in the k(cat) value. These findings support the idea that aspartic acid 103 in the active site is an essential catalytic residue involved in catalysis by hydrogen bonding to the dienolate intermediate.