Femtomolar transition state analogue inhibitors of 5'-methylthioadenosine/S-adenosylhomocysteine nucleosidase from Escherichia coli.

Escherichia coli 5'-methylthioadenosine/S-adenosyl-homocysteine nucleosidase (MTAN) hydrolyzes its substrates to form adenine and 5-methylthioribose (MTR) or S-ribosylhomocysteine (SRH). 5'-Methylthioadenosine (MTA) is a by-product of polyamine synthesis and SRH is a precursor to the biosynthesis of one or more quorum sensing autoinducer molecules. MTAN is therefore involved in quorum sensing, recycling MTA from the polyamine pathway via adenine phosphoribosyltransferase and recycling MTR to methionine. Hydrolysis of MTA by E. coli MTAN involves a highly dissociative transition state with ribooxacarbenium ion character. Iminoribitol mimics of MTA at the transition state of MTAN were synthesized and tested as inhibitors. 5'-Methylthio-Immucillin-A (MT-ImmA) is a slow-onset tight-binding inhibitor giving a dissociation constant (K(i)(*)) of 77 pm. Substitution of the methylthio group with a p-Cl-phenylthio group gives a more powerful inhibitor with a dissociation constant of 2 pm. DADMe-Immucillins are better inhibitors of E. coli MTAN, since they are more closely related to the highly dissociative nature of the transition state. MT-DADMe-Immucillin-A binds with a K(i)(*) value of 2 pm. Replacing the 5'-methyl group with other hydrophobic groups gave 17 transition state analogue inhibitors with dissociation constants from 10(-12) to 10(-14) m. The most powerful inhibitor was 5'-p-Cl-phenylthio-DADMe-Immucillin-A (pClPhT-DADMe-ImmA) with a K(i)(*) value of 47 fm (47 x 10(-15) m). These are among the most powerful non-covalent inhibitors reported for any enzyme, binding 9-91 million times tighter than the MTA and SAH substrates, respectively. The inhibitory potential of these transition state analogue inhibitors supports a transition state structure closely resembling a fully dissociated ribooxacarbenium ion. Powerful inhibitors of MTAN are candidates to disrupt key bacterial pathways including methylation, polyamine synthesis, methionine salvage, and quorum sensing. The accompanying article reports crystal structures of MTAN with these analogues.

bond is partially cleaved at the transition state without significant bond order to the attacking water nucleophile (15)(16)(17). The Immucillin transition state analogues designed for these enzymes were modified to incorporate chemical features of MTA and were the starting point for our inhibitor design program. Kinetic isotope effect analysis for MTAN has established the transition state structure to be more closely related to a highly dissociated ribooxacarbenium ion. 2 Highly dissociated transition states for N-ribosyltransferases are closely related to the human and malarial purine nucleoside phosphorylases and to ricin, which have all been shown to form highly dissociated S N 1-like transition states (18 -20). The second-generation DADMe-Immucillins are closer mimics of highly dissociated N-ribosyltransferase transition states (20,21), and we modified the DADMe-Immucillins to incorporate features of MTA for the second-generation MTAN inhibitors (Fig. 2). Transition state structures derived from kinetic isotope effect analysis have proven to provide robust atomic blueprints for the design of transition state analogues (15,(21)(22)(23). In contrast, protein structural similarity has not been a strong predictor of transition state structure. For example, the catalytic sites of human and bovine purine nucleoside phosphorylases are identical within the error of x-ray diffraction experiments, yet the transition states differ considerably (20). Thymidine phosphorylase, another ribosyltransferase, was predicted from structural and computational analysis to have a dissociative transition state, but kinetic isotope effect analysis established a nearclassical S N 2 transition state (24,25). Recent analyses have implicated dynamic motion as an important component of transition state formation. The only dependable way to understand enzymatic transition states is with methods such as kinetic isotope effect analysis that are independent of dynamics (21).
Transition state analogue inhibitors are designed from the hypothesis that chemically stable analogues that mimic geometric and molecular electrostatic features of the transition state will be bound to the enzyme tighter than the substrate by the factor equal to the catalytic rate of acceleration imposed by the enzyme (22,23). Transition state analogues with nearly irreversible binding can hypothetically be designed for enzymes with typical catalytic rate enhancements (10 10 to 10 15 ). Nucleoside hydrolases enhance reaction rates by ϳ10 12 . With K m values of 0.43 and 4.3 M for MTA and SAH, respectively (26 -28), perfectly designed transition state analogues for MTAN should have dissociation constants of ϳ10 Ϫ19 to 10 Ϫ18 M. However, it is impossible to design perfect transition state analogues, since the actual enzymatic transition state has nonequilibrium bond lengths and charges that cannot be accurately copied in chemically stable molecules. Features of the transition state that distinguish it from substrate are shown in blue. 5Ј-Thio substituted Immucillin-A molecules mimic an early transition state where the bond between the ribosyl and adenine groups retain partial bond order. Features important for transition state recognition are shown in blue (B). 5Ј-Thio-substituted DADMe-Immucillin-A molecules mimic a late transition state where the ribosyl cation is fully dissociated from the adenine leaving group, a distance of ϳ3 Å (B). In fully dissociated ribosyl transition states, the site of carbocation formation is at C1Ј, and the N1Ј mimics this geometry. The methylene bridge positions the leaving group at an appropriate distance from the ribooxacarbenium ion site; 9-deazaadenine provides a carbon-carbon bridge for chemical stability and to increase the pK a at N7. Features important for transition state recognition are shown in blue.
FIG. 1. Pathways connecting polyamine synthesis, adenine salvage, methionine salvage, and tetrahydrofuran synthesis in bacteria. The tetrahydrofurans are precursors for the synthesis of autoinducer-2 molecules (AI-2), while N-acyl-substituted homoserine lactones are autoinducer-1 molecules (AI-1). Both are signaling molecules for quorum sensing. MTAN is encoded by the pfs gene in E. coli and catalyzes two reactions in this cycle. The diagram is adapted from the summary provided in Ref. 9. SAM, S-adenosylmethionine; RH, ribosyl-homocysteine; THF, tetrahydrofolate.
We synthesized putative transition state analogues for MTAN that mimic properties of both partially and fully dissociated S N 1 transition state structures. Efficient transition state analogue inhibitors include ribooxacarbenium ion (carbocation) mimics and a 9-deazaadenine group to mimic the elevated pK a for N7 of MTA at the transition state (Fig. 2). The same approach and some of the same transition state analogues have also been shown to be powerful inhibitors of human MTA phosphorylase (29,30). Transition state mimics of the Immucillin family include an iminoribitol moiety as a mimic of the ribooxacarbenium ion. A stable C-C ribosidic bond replaces the N-ribosidic bond of substrates and an elevated pK a at the N7 position is accomplished by the use of 9-deazaadenine. Iminoribitol mimics resemble early transition states where significant N-ribosidic bond order remains at the transition state. To mimic a fully dissociated ribooxacarbenium transition state, the DADMe-Immucillin family was synthesized. These analogues include the pyrrolidine moiety as the ribooxacarbenium ion mimic and a methylene bridge between the ribooxacarbenium ion mimic and the 9-deazaadenine to provide a geometric approximation of the N-ribosidic bond distance for a fully dissociated transition state. Chemical stability of the pyrrolidine nitrogen is achieved by three neighboring methylene groups, generating analogues resembling 2Ј-deoxyribosyl nucleosides (Fig. 2). Dissociation constants of the Immucillin analogues extend to the picomolar range whereas the DADMe-Immucillin inhibitors extend to the femtomolar range. Competitive inhibitors with femtomolar dissociation constants are rare in enzymology, making these compounds among the most powerful noncovalent enzyme inhibitors yet reported.

MATERIALS AND METHODS
Enzyme Preparation-The fragment containing full-length E. coli MTAN was obtained by EcoRI/NotI digestion of p5Xmtan (28) and cloned into a pPROEX Hta expression vector. The vector was transformed into E. coli strain TOP10FЈ. The transformed cells were cultured in LB medium containing 100 g/ml ampicillin at 37°C, and MTAN was expressed by induction with 1 mM isopropyl ␤-thiogalactopyranoside. This construct expressed a protein with six histidines at the N terminus and was purified by Ni-NTA chromatography (28,31). The histidine tag was removed by chymotrypsin digestion following purification. The digested protein has a linker region of ten amino acid residues at its N terminus. The purified protein was analyzed for purity by SDS-PAGE gels stained with Coomassie Blue and was stored at 15 mg/ml at Ϫ70°C following flash freezing in dry ice-acetone.
MT-DADMe-3-deaza-ImmA [49] was synthesized via a Mannich reaction with formaldehyde, 3-hydroxy-4-thiomethylpyrrolidine and N-Ac-3,9-dideazaadenine, followed by deprotection of the acyl group. The N-Ac-3,9-dideazaadenine moiety was prepared from 2-hydroxy-4methyl-3-nitropyridine. Treatment of this hydroxypyridine with POCl 3 provided the chloride, which was displaced with ammonia to afford 2-amino-4-methyl-3-nitropyridine. The amino group was acylated before reaction with Brederick's reagent, which provided the enamine. This enamine was then reductively cyclized with zinc dust in acetic acid to give N-Ac-3,9-dideazaadenine. Me-Sulfoxide-ImmA [44], as a ϳ1:1 mixture of diastereomers, was prepared by treating MT-ImmA [12] with 30% hydrogen peroxide. In a similar fashion, hydrogen peroxide treatment of MT-DADMe-ImmA , MT-Pz-ImmA [7], and formycin A [20] were made from sample weight. Concentration of MT-ImmH [19] was determined with the extinction coefficient of 9-deazahypoxanthine of 9.54 mM Ϫ1 cm Ϫ1 at 261 nm (36). The reactions for measuring the inhibition constants for Immucillins were initiated by adding the enzyme (1-5 nM) to the reaction mixture, typically containing 200 M MTA, 100 mM HEPES, pH 7.5, 50 mM KCl and varying amount of inhibitor concentration in 1-ml reaction volume at 25°C. For tight binding DADMe-Immucillins having affinities in the low picomolar to femtomolar range, MTA concentrations of 2-3 mM were required to observe appropriate reaction rates. The coupled assay with xanthine oxidase was used with high concentrations of substrate. Reaction mixtures for the xanthine oxidase-coupled assay contained 100 mM HEPES, 50 mM KCl, 2-3 mM MTA, 0.5 unit/ml xanthine oxidase, and varying amount of inhibitor concentrations in a reaction volume of 1 ml at 25°C. The reactions were initiated by adding 1-5 nM MTAN. Controls having no inhibitor and no enzyme were included in all experiments. The K i values for inhibitors were obtained by fitting the initial rate and inhibitor concentration to the following expression of competitive inhibition where VЈ o is the initial rate in the presence of inhibitor, and V o is the initial rate in the absence of inhibitor, [I] is the inhibitor concentration, and [S] is the substrate concentration. This expression is valid only under the condition where the inhibitor concentration is 10 times greater than the enzyme concentration. However, when the inhibitor concentration was only a few fold greater than the enzyme concentration, the effective inhibitor concentration was obtained by the expression, where IЈ is the effective inhibitor concentration, VЈ o and V o are the initial rate in the presence and absence of inhibitor, and E t is the total enzyme concentration. Most tight binding transition state analogue inhibitors displayed a second linear reaction rate following slow-onset tight binding of the complex, indicating that they had reached thermodynamic equilibrium with the enzyme. Examples are given in the slow-onset inhibition curves of Figs. 3-5. The equilibrium dissociation constant (K i *) was obtained by fitting the rates to the following equation of competitive inhibition, [21]. Data analysis for the tightest binding DADMe-Immucillin-A derivative. The reaction rate and slow-onset inhibition was monitored by the conversion of MTA to 2,8-dihydroxyadenine in a coupled reaction with xanthine oxidase (upper panel). The coupled assay permits use of high substrate concentration (2.0 mM, 4,650 ϫ K m for MTA) to compete against these powerful inhibitors. Values of K i (middle panel) and K i * (lower panel) were obtained from the initial (0 -5 min) and the final rates (25-30 min). Similar experiments were used to measure inhibition by slow-onset tight binding inhibitors shown in Figs. 6 and 7. Control reactions demonstrated that the inhibitors did not inhibit the xanthine oxidase reaction. with correction to [I] as described above. 3 Comparison of inhibition constants throughout this work are values representing true thermodynamic equilibrium (dissociation constants) between enzyme and inhibitor. In cases where both K i and K i * are reported, K i * represents this dissociation constant. In cases where only K i is reported, K i is the dissociation constant. Thus, it is correct to compare K i and K i * values in some cases.

MTAN Transition State Analogue Inhibitor Design-E. coli
MTAN hydrolyzes the N-ribosidic bond of its substrates with highly developed ribooxacarbenium-ion character at the transition state. 2 Other N-ribosyl transferases including purine nucleoside phosphorylases from human and malarial sources, ricin A-chain and human MTA phosphorylase are also characterized by ribooxacarbenium ion character at their transition states (18,20,29). Transition states for purine N-ribosyltrans- 3 Slow-onset, tight binding inhibition is common with transition state analogues (43). In the first phase, inhibitor binds reversibly to form the EI complex, defined by the dissociation constant K i . In the second step, EI undergoes conformational tightening (EI 3 EI*) where the inhibitor is bound more tightly. The equilibrium between free enzyme and the EI* complex is defined by K i *, the equilibrium dissociation constant .   FIG. 4. Inhibition of E. coli MTAN by pChPhT-ImmA [1]. An example of data analysis for a slow-onset, tight-binding Immucillin derivative. The rate traces in the upper panel are from the direct assay that monitored the difference spectra between MTA and adenine. Initial rates of inhibition were used to evaluate K i from fits to the equation for competitive inhibition as indicated in the methods (middle panel). Following slow-onset inhibition, a second set of slopes is observed (80 -100 min), and these are fitted to the equation for competitive inhibition to evaluate K i *, the equilibrium dissociation constant (lower panel). Details of the analysis and equations are given in the methods. Similar methods were used to measure other inhibitors shown in Figs. 6 and 7. [41]. An example of data analysis for a slow-onset, tight binding 5Јthio-substituted DADMe-Immucillin derivative. The reaction rate and slow-onset inhibition were monitored by the conversion of MTA to 2,8-dihydroxyadenine in a coupled reaction with xanthine oxidase (upper panel). The coupled assay permits use of high substrate concentration (2.0 mM, 4,650 ϫ K m for MTA) to compete against these powerful inhibitors. Values of K i (middle panel) and K i * (lower panel) were obtained from the initial (0 -5 min) and the final rates (25-30 min). Similar experiments were used to measure inhibition by slow-onset tight binding inhibitors shown in Fig. 8. ferases fall into two groups, those with early transition states that have ribooxacarbenium ion character at the transition state but partial bond order remaining to the purine leaving group and those with fully dissociated ribooxacarbenium ions at the transition state. The transition state for E. coli MTAN is similar to N-ribosyltransferases with fully dissociated transition states (Fig. 2). Immucillins are known to be powerful inhibitors for the first class of N-ribosyltransferases, while DADMe-Immucillins are powerful inhibitors for enzymes where the transition states are fully dissociated carbocations (21). The inhibition constants described below for MTAN with Immucillins and DADMe-Immucillins support a transition state structure with a fully developed carbocationic ribosyl group and are consistent with transition state analysis based on kinetic isotope effects. 2 Immucillin Inhibitor Design for MTAN-Immucillins are transition state analogue inhibitors that incorporate the features of an early ribooxacarbenium-ion transition state (Fig.  2B). The N4Ј-imino group of Immucillins has a pK a of 6.9 and is known to be cationic at the catalytic sites of N-ribosyltransferases (37). Replacing N9 with carbon makes the glycosidic bond chemically stable and alters the conjugative pattern to increase the pK a of N7 to Ͼ10, resulting in its protonation. 4 N-Ribosyltransferases are proposed to gain purine leaving group activation by hydrogen bond formation or protonation of N7 from the enzyme as the pK a of N7 increases with elongation of the N-ribosidic bond. In MTAN, Asp 197 interacts with N7 and the elevated pK a in the purine ring permits formation of a favorable H-bond (see Refs. 38 and 39 and the accompanying article (40)).

FIG. 5. Inhibition of E. coli MTAN by cHeptylT-DADMe-ImmA
MT-ImmA [12] was designed to mimic an early transition state with MTA as substrate. It is a powerful inhibitor exhibiting slow-onset tight-binding inhibition with a K i value of 130 pM and the K i * equilibrium constant of 77 pM (Fig. 6). The K m value for MTA is 0.43 M to give a K m /K i * value of 5,600 for MT-ImmA [12]. 5 The increase in apparent affinity of MT-ImmA [12] can be partially attributed to a H-bond between the N7 and Asp 197 . An additional H-bond between the N8 in MT-Pz-ImmA [7] and Ser76 O␥ (38) improves its K m /K i * value to 16,500.
The pocket in MTAN that binds the 5Ј-methylthio group is hydrophobic and this part of the catalytic site is composed of flexible amino acid side chains. Substrate specificity demonstrates that the region has sufficient conformational flexibility to accept both methylthio and homocysteine groups, although the K m for SAH [39] is 10-fold greater than that for MTA [17] (4.3 versus 0.43 M). The 5Ј-methylthio site can accommodate both polar and non-polar groups as indicated by the binding of SAH [39] and 5Ј-substituted substrates such as 5Ј-(p-nitrophenylthio)adenosine (26). Incorporating other hydrophobic and halogenated aliphatic and aromatic groups at the 5Ј-thio position of 5Ј-thio-ImmA increased the binding affinity to give dissociation constants of 6 to 12 pM in pClPhT-ImmA [1], pTolT-ImmA [2], mTolT-ImmA [3], and BnT-ImmA [4]. Aromatic and small nonpolar, aliphatic residues that surround the 5Ј-methylthio binding site include Met 9 , Ile 50 , Val 102 , Phe 105 , Tyr 107 , Pro 113 , and Phe 207 , and these are proposed to interact with the 5Ј-hydrophobic substituents. Among the 5Ј-thio-substituted Immucillin-A family, pClPhT-ImmA [1] with K i * of 2 pM is the tightest binding inhibitor with K m /K i * of 214,800 with respect to MTA [17] and 2,148,000 relative to SAH [39]. The slow-onset inhibition pattern is readily observed in the inhibition pattern (Fig. 4).
Design of bacteria-specific inhibitors requires discrimination against human enzymes catalyzing similar reactions. In humans there is no MTAN and 5Ј-methylthioadenosine phosphorylase (MTAP) is the only enzyme capable of metabolizing MTA. Comparisons between MTAP and MTAN reveal that MTAP has a more restricted 5Ј-alkylthio binding site, thus accounting for the weaker binding of BnT-ImmA [4] to MTAP than to MTAN with dissociation constants of 26,000 and 12 pM, respectively (29,30,41). This provides a specificity factor of 2,200 for the bacterial enzyme and provides potential for targeting bacterial MTAN without inhibiting human MTAP.
DADMe-Immucillin Inhibitor Design for MTAN-The DADMe-Immucillins (Fig. 7) were originally designed to match the geometry and molecular electrostatic features of fully dissociated ribooxacarbenium ion transition states such as those found in human purine nucleoside phosphorylase (20). MTAN from E. coli also has a fully dissociated transition state 2 and it would be anticipated that the DADMe-Immucillin analogues would bind better than the Immucillins. MT-DADMe-ImmA [32] has K i and K i * values of 48 and 2 pM, respectively, with a K m /K i * of 215,000 with respect to MTA and 2,150,000 with respect to SAH. The 5Ј-methythio binding pocket readily accommodates larger alkyl groups including ethyl, propyl, and butyl. The affinity increases with the increase in length of the 5Ј-alkyl chain. EtT-DADMe-ImmA [29], PrT-DADMe-ImmA [26], and BuT-DADMe-ImmA [22] bind with K i * values of 950, 580, and 296 fM, respectively. BuT-DADMe-ImmA [22] is the second most powerful inhibitor of the DADMe-Immucillins. It binds 14,500,000-fold more tightly than SAH [39].
Comparing Immucillins and DADMe-Immucillins with the same 5Ј-thio substituents indicates the thermodynamic benefit of increasing the distance between 9-deazaadenine and the ribooxacarbenium mimics. pClPhT-ImmA [1] and pClPhT-DADMe-ImmA [21] gave dissociation constants of 2 pM and 47 fM, a factor of 43 tighter binding for the DADMe inhibitor. Similar values are seen in comparing pFPhT-ImmA [6] and pFPhT-DADMe-ImmA [25] where the DADMe derivative binds 36-fold more tightly. Likewise, PhT-DADMe-ImmA [30] binds 16-fold tighter than PhT-ImmA [10] (Figs. 6 and 7). Finally, the MT, mClPhT, and EtT groups achieve an extra 26-, 27-, and 28-fold affinity, respectively, from being in the DADMe context. BnT-ImmA [4] and BnT-DADMe-ImmA [24] have K i * values of 12 pM and 460 fM, respectively, a factor of 26. The consistent affinity ratio in comparing Immucillins and DADMe-Immucillins indicate that the methylene bridge and DADMe feature improves binding by ϳ2 kcal/mol. Changes in the purine base can also be interrogated for the effects on binding affinity. Thus, BnT-Pz-DADMe-ImmA [23] binding is equivalent to that for BnT-DADMe-ImmA [24]. The 3-fold improved binding of MT-Pz-ImmA [7] relative to MT-ImmA [12] is consistent with a favorable hydrogen bond between N8 and Ser 76 at the active site or may reflect a more favorable H-bond at N7 due to its pK a (9.6 for N7 of [7]). Incorporating polar groups, like chloro and para, to the thiol significantly improves the binding affinity of DADMe-Immucillins as was also seen for the Immucillins. pClPhT-DADMe-ImmA [21] shows slow-onset inhibition with K i and K i * values of 2.6 pM and 47 fM and binds 40 times more tightly than PhT-DADMe-ImmA [30] (Fig. 7). It is the tightest binding inhibitor of the DADMe-Immucillins against E. coli MTAN with K m /K i * values of 9.1 and 91 million relative to MTA [17] and SAH [39], respectively. Binding affinity 91 million times tighter than substrates is unprecedented for noncovalent enzymatic inhibitors. This corresponds to over 12 kcal/mol favorable binding energy relative to SAH. The improved binding of DADMe-Immucillins indicates that they are closer mimics of the transition state by 2 kcal/mol than are the Immucillins. The N1Ј pyrrolidine nitrogen in DADMe-Immucillins has a pK a near 9.0 (42), making this group more cationic than the N4Јimino group in Immucillins (pK a of 6.9). The accompanying article (40) describes the structure of E. coli MTAN with MT-DADMe-ImmA [32] and demonstrates that the N1Ј cation of MT-DADMe-ImmA [32] provides a favorable electrostatic/Hbond interaction between N1Ј and the nucleophilic water at a distance of 2.7 Å.
Although the N3 of MT-DADMe-ImmA [32] does not directly interact with any active site residues, the N3 to C3 substitution in MT-DADMe-3-deaza-ImmA [49] caused a 32,500-fold decrease in binding affinity. A loss of ϳ6 kcal/mol in binding energy is proposed to reflect the altered bond conjugation in the purine ring with different pK a values and decreased H-bond interactions at N1, N6, and N7.
Summary and Conclusions-This report adds to the growing demonstration that a detailed understanding of transition state chemistry is readily applied to the design of transition state analogues. Seventeen transition state analogue inhibitors with dissociation constants of 10 Ϫ12 to 10 Ϫ14 M are described for MTAN from E. coli. Transition state analogue design is based on the ribooxacarbenium ion character of the MTAN transition state. With dissociation constants to 47 femtomolar, the 5Јthio-substituted DADMe-ImmucillinA analogues are among the most powerful noncovalent enzymatic inhibitors. The best of these, pClPhT-DADMe-ImmA [21], exhibits a K m /K i * value of 91 million relative to the substrate S-adenosylhomocysteine. In addition to the 17 analogues with dissociation constants from 47 fM to 2 pM, a second group of 21 transition state analogue inhibitors for E. coli MTAN are described with dissociation constants from 2 pM to 1 nM. These inhibitors are potential antibiotics to interfere with the metabolic pathways involved in methylation, polyamine biosynthesis, methionine recycling, and quorum sensing pathways.