Results

3.1 Dbv29 is a FAD-containing dimer

To begin our investigations, we solved crystal structures of Dbv29 free and in complex with substrate. The free structure superimposes very well with the complexed form (r.m.s. deviation of . for C), suggesting that binding of ligand does not induce substantial conformational transitions. As shown for the bound enzyme (Figure 2a), the structure consists of two domains: the F domain, which binds the flavin cofactor, and the S domain, which recognizes substrates, confirming that Dbv29 belongs to the flavin-dependent p-cresol methylhydroxylase (PCMH) superfamily¹⁹. The crystal structure also offers clear evidence that the flavin cofactor is flavin adenine dinucleotide (FAD) and not FMN. We suspect that the previous suggestion of an FMN-containing enzyme was likely because of hydrolysis of FAD in solution. The major difference between the two structures is that apo Dbv29 appears to form a dimer on the basis of a large subunit-subunit interface ( , 6 ² per monomer). However, in the complex, an asymmetric unit contains four molecules but does not demonstrate substantial protein-protein interaction, suggesting that the bound form exists as a monomer. This is in contrast to three other members of the PCMH family, glucooligosaccharide oxidase (GOOX)¹⁷, aclacinomycin oxidoreductase (AknOx)¹⁶ and berberine bridge enzyme (BBE)²⁸, which were solved recently in monomeric forms.

3.2 Contributions to cofactor- and substrate-binding sites

As discussed above, our prior studies had highlighted unusual roles for four residues in controlling flavin binding and reactivity. The crystal structure now reveals a possible explanation for these functions (Figure 2). Though the isoalloxazine ring of flavin is covalently linked to the side chains of His91 and Cys151, it is closely connected to Tyr165 and Tyr473 via the substrate and a bridging water (Figure 2c). We revisited our

14

enzyme activity studies to obtain improved semi-kinetic data and observed that, as previously reported, single H91A and C151A mutants, as well as the H91A/C151A double mutant sequence, retained low levels of activity (11%, 23% and 5% activities, respectively, relative to wild type; Table 3). The mutated enzymes also retained their characteristic yellow color, in stark contrast to the corresponding mutants of GOOX, AknOx and BBE, which lost their color and activities. In contrast to our earlier work, however, the improved kinetic analysis demonstrated that Y165F and Y473F mutations did substantially affect enzyme activity, reducing function to 70% or 23% of the wild-type function, respectively; the Y165F/Y473F double mutant further led to loss of the yellow color and total loss of catalytic activity. Notably, the Y473E mutation similarly abolished activity; we suspect the lower pKa of this side chain or the alternate conformation adopted by the nonplanar ligand might explain this observation. We additionally examined the impact of mutations at the nearby positions Arg360, Ser364 and Thr366, as well as six other tyrosine residues to search for other contributing interactions. However, aside from the Y135F mutant protein, which could not be expressed, none of these mutations disrupted flavin binding (Table 3).

The substrate-binding site in the S domain is somewhat hidden by a long loop (residues 351–362) that crosses over the binding pocket. The electron density for the aglycone shows some discontinuity, whereas that for the N-acyl amino sugar moiety is apparent and clear. The sugar ring anchors against the si face of the isoalloxazine ring of FAD and exposes the C5 carbon close to the N5 nitrogen of FAD, while the C4 hydroxyl group forms a hydrogen bond with Asn427. The N-acyl group extends from the sugar ring into a hydrophobic lipid cavity (Figure 2b). We tested the importance of packing interactions in this lipid cavity by comparing the enzymatic turnover of two substrates with short (C4, 3) and long (C10, 6) alkane chains; the longer chain substrate

15

was preferred by approximately ten-fold, indicating contributions from but not a strict requirement for hydrophobic packing.

Beyond Tyr165 and Tyr473, which contact the C6 hydroxyl group, most residues that line the ligand-binding site are located in different loop regions. Mutational analysis of some of these residues showed that Ser364, Thr366, Tyr370 and Tyr470—though they did not impact flavin binding—did significantly alter enzyme function, with resultant activities of 4–21% of wild type. Additionally, Trp399 and Ile401 proved to be critical for function, as the mutated enzymes displayed either no activity or no more than 10%

of wild-type function (Table 3).

3.3 A diol intermediate leads to catalytic redirection

Although the structure of the complex was solved by cocrystallization with teicoplanin, the electron density is best fitted with neither teicoplanin (the substrate) nor oxo-teicoplanin (the product). Instead, the appearance of extra out-of-plane electron density on the C6 terminus of the amino sugar was best explained by the presence of two OH groups in sp³ configuration or a water-coordinated diol (Figure 2c). We had previously suggested a basic series of reactions needed to convert the C6 alcohol to a carboxylate in which a diol intermediate, or aldehyde equivalent, is used as the transformed substrate for the second half of the oxidation. The structural information obtained here now allows us to build on our prior analysis to propose a specific model for Dbv29 function (Figure 2d): in the first oxidation step, the C6-OH is deprotonated by the hydroxyl group of Tyr473, which in turn is activated by the hydroxyl group of Tyr165. These residues together drive the pro-R hydride transfer from the C6 carbon to the N5 of the isoalloxazine ring and yield the aldehyde intermediate corresponding to the diol found in the crystal structure. In the second oxidation, a second hydride transfer to FAD is likely facilitated by a proton relay network involving the tyrosine pair, the

16

substrate diol and the diol-coordinated water molecule. In each two-electron oxidation reaction, the reduction of FADH2 back to FAD converts one molecule of molecular oxygen to hydrogen peroxide.

Beyond the mere presence of this unexpected chemical group on the protein substrate, we also discovered that this moiety appeared to be directly exposed to solvent. To test this hypothesis, we conducted two sets of reactions using isotopically labeled reactants.

When the teicoplanin and Dbv29 were mixed in ¹⁸O-labeled water, a dominant mass of M+4 instead of M+2 was found, indicating both oxygens of the diol intermediate were labeled with ¹⁸O (Figure 3a). The simplest explanation for this result is that the aldehyde is accessible to water and that the aldehyde and diol are in equilibrium. In addition, when we reduced teicoplanin with NaBD4 in the presence and absence of Dbv29, MS analyses of the product recovered from the reaction with protein showed a mass increased by 1 mass unit (Figure 3b). Again, the simplest explanation for this result is that the NaBD4 was able to access the oxidized aldehyde in the reactive site and reduce it to a deuterated alcohol species.

The accessibility of the aldehyde further led us to explore the possibility of trapping and modifying this intermediate as a way to produce new antibiotic analogs. First, we tested the reductive amination of an aldehyde with an amine²⁷. Enzymatic reactions were carried out in the presence of teicoplanin, Dbv29, sodium cyanoborohydride (Na(CN)BH3) and either n-decylamine or n-hexylamine. Remarkably, the LC-MS traces (Figure 3c) indicated that although reaction with n-decylamine resulted in the expected formation of decylaminated teicoplanin (10), the reaction with hexylamine led to n-hexylamidated teicoplanin (11), both of which were confirmed by MS (Figure 3d,e).

These results can be explained by differential insertion of the amine into the active site.

Thus, the hexylamine would react with the aldehyde intermediate, and the resulting

17

gem-hydroxyl-alkylamine could remain in the active site to undergo the second oxidation reaction. In contrast, decylamine appears to be too bulky to enter the active site and so instead reacts with the aldehyde functional group to form an imine that remains exposed to solution or is even released into solution, where it is subsequently reduced. Thus, the reactive intermediates can be trapped at varying stages during the catalytic course.

To take advantage of this reactivity, we constructed a chemoenzymatic strategy to create additional analogs with variations in both the length and the structure of the lipid chains at both C2- and C6-positions of teicoplanin and oxo-teicoplanin (Figure 4). Our protocol used Dbv21, an enzyme catalyzing deacetylation of the GlcNac moiety^{10, 14}, and Dbv8¹⁵, an enzyme catalyzing the acylation of the glucosamine moiety, in combination with Dbv29 and various coenzyme A derivatives. When mixed in the presence of organic solvents (for example, 50% DMSO) and selective reducing agents (for example, 5 mM sodium cyanoborohydride), we were able to obtain a series of new analogs 12–31. The enzymes were quite tolerant to new functional groups, as demonstrated by the inclusion of amantidine and related compounds as well as alkyne and azide handles that can be further manipulated through ‘click’ chemistry for applications such as identification of a second mode of action for broader spectrum analogs^{29, 30}.

3.4 Analogs provide alternate antimicrobial protection

Having gained access to new chemical space with our modified teicoplanin analogs, we wanted to examine whether we had similarly reached new biological space;

accordingly, we determined the minimum inhibition concentrations (MICs) of several of the compounds against collections of vancomycin-resistant Enterococcus (VRE) strains (Table 1). We observed that inhibition is proportional to the length of acyl side chains in

18

compounds 2–7, with the carboxylate (9) showing no change in activity as compared to teicoplanin. The bi-lipid analogs 10 and 11, however, showed significantly enhanced bactericidal activities against all five tested strains of Enterococcus fecalis when compared to vancomycin and teicoplanin. Notably, the different functional groups in compounds 10 and 11 yielded opposite outcomes for the different strains: the shorter-chain analog 11 is more effective against antibiotic-sensitive strains (for example, ATCC 33186), whereas the longer-chain analog 10 is more effective against drug-resistant strains (for example, ATCC 51559 and 700221). Strikingly, analog 25, tailored with benzylamine (Figure 10), is equally effective against sensitive and resistant strains with dose potency one to two orders of magnitudes lower than vancomycin and teicoplanin and thus may complement compounds such as the A40926 derivative dalbavancin, which is active against Staphylococci (for example, MRSA) and Streptococci but weak against VRE^31-34. As a first proof of principle for in vivo efficacy, we tested analog 25 against mice infected with VRE (ATCC 51559) via intravenous injection and found that this molecule was more effective than either vancomycin or teicoplanin at reducing blood bacterial counts (Figure 5).

3.5 Dbv29 has a dynamic quaternary structure

We remained intrigued by the different quaternary forms of the protein observed in the apo and bound forms (Figure 6). To investigate this matter further, we biophysically characterized several constructs, including the wild-type protein, the double mutant that lacks covalent linkages to flavin (H91A/C151A), the nonfunctional double mutant (Y165F/Y473F) and an additional construct with alanine mutations at each of two residues located at the dimeric interface (R41A/N44A). The latter two constructs did not form dimers as monitored by gel filtration analysis or analytical ultracentrifuge experiments and showed some minor deviations from wild type by circular dichroism

19

(Figures 7, 8). Our earlier functional test had demonstrated that the Y165F/Y473F double mutant lacked catalytic activity, but we were also surprised to observe that the enzyme structure was significantly destabilized, as indicated by a decrease in melting temperature of almost 35 °C (Figure 9). Additionally, though Arg41 and Asn44 were not anticipated to participate in the enzyme mechanism directly, the catalytic efficiency of the double mutant is substantially reduced (59% of wild-type activity; Table 3).

Although the mechanistic importance of this is not fully understood, we speculate that the protein quaternary structure dynamics may be correlated to subtle structure changes when a ligand is loaded on or off the enzyme.

20