Next generation Glucose-1-phosphate thymidylyltransferase (RmlA) inhibitors: An extended SAR study to direct future design

Graphical abstract


Introduction
The continued global emergence of multi-drug resistant bacteria is a major health concern. The time taken for resistance to new drugs to arise is often rapid and the pace of antibiotic discovery has slowed since the golden era of the 1940-60s. 1 The development of novel antimicrobials that avoid the existing mechanisms of resistance and target new pathways is recognised as a high priority for research.
The outer membrane (OM) protects gram-negative bacteria from antibiotic attack and is essential for survival. 2 The OM is composed of lipopolysaccharides (LPS) that in many, but not all, bacteria contain Lrhamnose, a C6 sugar unit. For example, in P. aeruginosa L-rhamnose is a component of the LPS and deletion of one of the genes responsible for its biosynthesis results in a bacterium that has much lower virulence in a mouse model. 3 In M. tuberculosis the arabinogalactan unit in the cell wall is linked to the peptidoglycan by a disaccharide phosphodiester linker that has a L-rhamnose component (a decaprenyl-diphospho-N-acetylglucosamine rhamnosyl molecule 4 ). The enzymes involved in the biosynthesis of L-rhamnose are therefore potential anti-tuberculosis drug targets. [5][6] The L-rhamnose biosynthetic pathway involves four enzymes, RmlA-RmlD, which catalyse the conversion of glucose-1-phosphate (Glu-1-P) to the L-rhamnose precursor deoxythymidine diphosphate-Lrhamnose (dTDP-L-rhamnose 3 , Scheme 1). Since this biosynthetic pathway is not found in eukaryotes, these enzymes are attractive targets for the development of novel selective antibiotics. Small molecule inhibitors of RmlA 7-9 and RmlC [10][11] have already been reported.
RmlA, a glucose-1-phosphate thymidylyltransferase, is the first enzyme in the pathway and catalyses the condensation of Glu-1-P with deoxythymidine triphosphate (dTTP) to give dTDP-D-glucose (Scheme 1). 3,12 The inhibition of RmlA by dTDP-L-rhamnose (the final product of the four step reaction sequence) has been reported 3,13 and it would therefore appear that, in bacteria, RmlA is the point of control for flux through the biosynthetic pathway. RmlA exists as a dimer of dimers and is functional as a tetramer. As a consequence of its structure, the active sites cluster at the dimer-dimer interface and the allosteric (or regulatory) sites cluster at the monomer-monomer interface within each dimer. 3,14 We have previously reported a series of potent novel small molecule allosteric inhibitors of P. aeruginosa RmlA 7 (for example Compound 8a in Figure 1A, compound numbering taken from the original report 7 ). In the previous work 7 , examples of our in vitro RmlA inhibitors were tested for their ability to inhibit the growth of M.tuberculosis (H37Rv) in which RmlA has been shown to be essential. 15 Even though sequence alignment of RmlA from P.aeruginosa and M.tuberculosis showed the two proteins are highly conserved, the selected compounds demonstrated only weak activity against M. tuberculosis bacteria (MIC 100 values＞25 μg/mL). For example, the potent in vitro P.aeruginosa RmlA inhibitor 8a (IC 50 = 0.073 ± 0.001 μM 7 ) was shown to have only a weak effect on M. tuberculosis (MIC 100 = 100 μg/mL). Apart from off-target effects and minor sequence differences in the RmlA homologues from the two bacteria, another possible reason for the poor effect on live bacteria is the inability of the tested analogues to penetrate the bacterial cells. The cell envelope of mycobacteria, comprised of polysaccharides and lipids, functions as a natural shield that is effective at blocking the entry of small molecules into the protoplasm. [16][17] Re-examination of the RmlA-8a complex [PDB 4ASJ] revealed a hydrophobic pocket that was only partly occupied by the N 1 -substituent ( Figures 1B and 1C). In an initial attempt to explore further the impact of structural changes on the binding of compound 8a, we chose to vary the R 1 substituent ( Figure 1A). However, the main focus of this report builds on the observation that the C6-NH 2 group of 8a points out of the allosteric binding pocket based on our analysis of the RmlA-8a complex ( Figure 1C). It was decided to assess whether substitution of one of the NH bonds in the C6-NH 2 group with an extended alkyl chain (represented by R 2 in Figure 1A) could be tolerated by P. aeruginosa RmlA as this could provide a vector out of the allosteric pocket to an open space whilst retaining the in vitro inhibitory activity of the current series of Scheme 1. L-Rhamnose biosynthetic pathway involving 4 enzymes which catalyse the conversion of Glu-1-P to dTDP-L-rhamnose. Chemical structure and biological activity of the previously optimized inhibitor 8a. [7] IC 50 against the Pseudomonas aeruginosa RmlA protein, MIC 100 against Mycobacterium tuberculosis. The aims of this work were to modify the N 1 -and C6-NH 2 positions. B. A representation of 8a bound in the allosteric site of RmlA based on our previous Xray crystallographic analysis of the RmlA-8a complex [PDB 4ASJ]. Residues that make up the N 1substituent sub-pocket are highlighted. C. Schematic representation of pocket interactions between 8a and the enzyme showing that the C6-NH 2 in 8a has the tendency to point out of the allosteric pocket into solution.
analogues. If successful this could provide the foundation for the development of a new series of RmlA inhibitors. For example, the attachment of a bacterial cell wall permeabilizer could be achieved via the newly incorporated linker unit at the C6-NH 2 position.

Results and discussion
Our previous studies focused on structure activity relationship (SAR) analyses involving the sulfonamide and N 5a -alkyl substituents in 8a ( Figure 1A). 18 However, no attempts to optimize the substituent at the N 1 -position were made. This current study started with examination of the reported structure of the RmlA-8a complex [PDB 4ASJ] and revealed that the N 1 -substituent pocket was formed from both main-and sidechain atoms of residues Leu45, His119, Glu120, Ile256, Arg259 and Gln260 ( Figure 1B). Visual inspection showed that this binding pocket was not ideally filled by the unsubstituted N 1 -benzyl group in 8a and therefore it was proposed that alternative N 1 -substituents could improve target binding. A pilot SAR study was performed to explore this hypothesis (see linked Data in Brief report for a detailed discussion). In summary, it was concluded that the key driver in increasing the potency was the presence of a substituent at the 4-position of the N 1 -benzyl group. It was found that a para-bromobenzyl substituent (compound 1a in Table 1) was optimal (see linked Data in Brief report and PDB codes 5FUH, 5FYE, 5FU0, 5FTS, 5FTV, 5FU8). Consequently, all R 2 -modified analogues were prepared in the N 1 -p-bromobenzyl series with one exception (compound 1f, Scheme 2 and Table 1).
The X-ray crystallographic analysis of the RmlA-1a complex ( Figure 2, PDB 5FTV) confirmed that in the p-bromobenzyl series, as well as for 8a, substitution at the C6-NH 2 position should enable positioning of a modifiable functional group in proximity to the mouth of the allosteric site ( Figure S1A). If this could be achieved, not only would the only remaining position available for modification in this inhibitor series have been explored, but future efforts to prepare analogues with enhanced bacterial cell wall permeability would also be facilitated (Figures S1B and S1C). Preliminary molecular modelling studies predicted that the C6-NH 2 modified analogue 1b (Table 1 and Scheme 2 for structure) binds in the allosteric site of the enzyme in a similar confirmation to the parent analogue 1a (Figure S2). In addition, the extended C6-aminoalkyl chain was predicted to reach out towards the mouth of the allosteric pocket, as designed. Analogues 1b and 1c with n-propyl and ethyl-containing linkers were therefore synthesised (Scheme 2).
It was decided to incorporate the extended C6-aminoalkyl chains of 1b and 1c early in the reaction sequence (Scheme 2). Selective N 1alkylation of the starting material 6-chlorouracil (2) with 4-bromobenzyl chloride under basic conditions enabled isolation of the N 1benzylated product to give 3a. [19][20] 6-Aminouracils 4b and 4c were then synthesized by reaction of 3a with the corresponding amines 5b (2 × CH 2 ) and 5c (3 × CH 2 ) in moderate yield. The remaining stepsbromination (to give 6b and 6c), addition of methylamine (to give 7b and 7c) and sulfonamide formation were based on our previous report 7 (see linked Data in Brief report for additional examples of this reaction sequence) and enabled the successful conversion of 4b and 4c to the N-Boc protected versions (8b and 8c) of the final compounds. Subsequent Boc deprotection of 8b and 8c using TFA gave 1b and 1c respectively as the TFA salts (Scheme 2). The introduction of the ethyl C6 aminoalkyl chain in 1c led to a complete loss of activity 21 against P. aeruginosa RmlA, whereas incorporation of the n-propyl linker in 1b retained activity (IC 50 of 0.86 µM, Table 1, entries 2 and 3, see Figures S3 and S4 legends for a discussion on the lack of activity of 1c). X-ray crystallographic analysis of the complex of RmlA with 1b [PDB 6TQG] showed that 1b was bound in the allosteric site of RmlA as expected. Compared with the C6-NH 2 unsubstituted analogue 1a (Table 1, entry 1), most of the ligand-protein interactions were retained in the RmlA-1b complex ( Figure S5), however, some differences were observed. For example, whereas the C6-NH 2 group in 1a showed hydrogen bonding to the protein backbone (Gly115 and His119) through the interaction with two different molecules of water, 1b retained the interaction with Gly115 but lost the watermediated hydrogen bond to His119 (as expected, Figure S3). Consistent with the docking studies, the extended aminoalkyl chain in 1b pointed out of the allosteric pocket and the distance between the nitrogen of the newly introduced terminal methylamine in 1b to the Cterminal Tyr293 residue was 3.8 Å. The terminal NH in 1b interacted with a network of water molecules ultimately linking to the C-terminal Tyr293 ( Figure S3). Based on the initial success with 1b being a sub-micromolar inhibitor of P. aeruginosa RmlA, it was decided to extend the length of the linker unit from n-propyl in 1b to n-butyl in 1d and n-pentyl in 1e (Scheme 2) in an attempt to position the terminus of the C6-aminoalkyl chain outside the allosteric pocket. In the case of 1d and 1e, a primary amine was incorporated at the end of the chain (see Figure S5 legend for more discussions). There was a risk that the more extended and flexible alkyl chains in 1d and 1e may undergo hydrophobic collapse. Therefore a heteroaromatic ring was incorporated into the linker unit in an attempt to minimse the chances of this occurring. The triazole-containing compound 1f was therefore synthesized (Scheme 2). In the case of 1f, the para-Br in the N 1 -benzyl moiety was also removed to provide additional room for the triazole group in 1f to adjust its position in the allosteric site. The synthesis of the additional analogues 1d and 1e was achieved in an analogous manner to the synthesis of 1b and 1c (Scheme 2). The synthesis of 1f required incorporation of an azide functional group at the terminus of the C6-linker unit through formation of 4f (n = 3, R 2 = N 3 , Scheme 2). The azide group was compatible with the subsequent steps enabling 4f to be successfully converted to 8f. The copper-catalysed azide-alkyne click (CuAAC) reaction of 8f with propargylamine gave 1f although the unoptimized yields over the final two steps in the sequence were low (Scheme 2). If analogue 1f was found to retain activity against P.aeruginosa RmlA, future work should enable the relatively easy incorporation of a bacterial cell wall permeabilizer using this CuAAC approach.
The increased length of the linker chain in analogues 1d and 1e compared to 1b led to around a 2.5-fold increase in potency with 1d and 1e having IC 50 values of 0.303 ± 0.026 μM and 0.316 ± 0.023 μM respectively ( Table 1, entries 4 and 5 vs. entry 2). The structure of the RmlA-1d complex [PDB 6 T38] confirmed that instead of interacting with any protein residues, the terminal amine of the C6-aminoalkyl chain in 1d was positioned out of the pocket, approximately equidistant between His119 and Tyr293 ( Figures 3A and 3B).
Whilst there was a notable decrease in the potency of the triazolecontaining analogue 1f in the RmlA inhibition assay compared to 1c, 1d and 1e ( Table 1, entry 6 vs. entries 2, 4 and 5), 1f was still able to inhibit P.aeruginosa RmlA. The structure of the RmlA-1f complex [PDB 6 T37] revealed that the triazole moiety of 1f stacks between the imidazole ring of His119 and its own benzyl group in the N 1 position forming an unusual sandwich structure in which the extended C6 chain in 1f is stabilized ( Figure 3C). The terminal NH 2 in 1f does not appear to interact with any protein residues and extends out of the pocket ( Figure 3D and Figure S6). Superposition of the structures of the RmlA-8a with RmlA-1f complexes ( Figure S7) revealed that the introduction of the triazole moiety in 1f forces the repositioning of its N 1 -benzyl group. Compared to the situation with 8a, the N 1 -benzyl group in 1f is positioned much closer to Arg 259 and Glu 255 which form the backbone of the hydrophobic pocket. This may be one factor in explaining the observed drop in potency associated with 1f.
The physicochemical properties (Table S1) of this series of new compounds in terms of Ligand Efficiency (LE, ranging from 0.21 to 0.37) and Lipophilic Ligand Efficiency (LLE, ranging from 3.9 to 5.2) showed high drug-likeness 22 , while the CLogP values (ranging from 1.14 − 2.77) are relatively higher than those of therapeutic antibacterial agents, which mostly cluster near 0. 23 To address the possible low permeability of these compounds (due to their relative high lipophilicity), a "trojan horse strategy" 24 could be considered. Due to the poor solubility of Fe 3+ salts, most microorganisms cannot use them directly. A siderophore is an iron chelator secreted by bacteria. Having chelated Fe 3+ , the siderophore is recognised by a specific outer membrane receptor and then transported to the bacterial cytoplasm. In this way the bacteria can take in iron as an essential nutrient required to survive. 25 Similarly, synthesised siderophore-antibiotic conjugates [26][27][28] can be recognised by bacteria and transported into the cell. As soon as the conjugates are transported into the cell, the antibiotic, if the siderophore-antibiotic linkage is designed correctly, can be released to kill the bacteria. Many studies have shown that synthetic siderophore-drug conjugates can act as novel antimicrobial agents and can help treat disease caused by antibiotic resistant bacteria. The terminal amine of 1d and 1f could provide a possible site to attach a siderophore, laying the foundation to prepare bacterial cell wall penetrating RmlA inhibitors. In addition, the presence of the more basic amine functionalities in these novel compounds may impact on activity against both gram positive and gram negative bacteria (the eNTRy rules 29,30 ).

Synthesis of analogues
All intermediates and final compounds were prepared according to the protocols supplied in the Supporting Information and Data in Brief. Examples of the synthesis of 1d (intermediates and final compound) and the synthesis of 1b, 1c, 1e and 1f (final compounds) are shown below.

Protein Crystallization, co-Crystallization and soaking
Crystals were grown by the sitting drop vapour diffusion method as previously described. 7 Drops contained 1 μL of protein (10 mg mL -1 mixed with 1 μL precipitant (4-12% PEG 6000, 0.1-0.15 M MES pH 6.0, 0.05-0.1 M MgCl 2 , 0.1-0.15 M NaBr, 1% β-mercaptoethanol). Crystals grew overnight to dimensions of 0.2 × 0.2 × 0.1 mm. Complexes of RmlA with inhibitor were prepared by soaking or co-crystallization. For soaking, solid compound was added to drops containing crystals and allowed to incubate for between 2 and 24 h prior to data collection. For co-crystallization, solid compound was incubated with protein in solution for 1 h prior to setting up sitting drops.

Data collection
Data were collected at the Diamond Light Source synchrotron or inhouse using a Rigaku MicroMax 007HFM x-ray generator. Data were processed with iMOSFLM 33 or XIA2 34 incorporating XDS 35 . Each structure was solved using MOLREP 36 with 4ASJ 7 as the search model with the inhibitor removed. REFMAC 37 was used to refine the models with model building in COOT 38 and ligands built with PRODRG 39 . Structural figures were prepared using Pymol 40 and CCP4MG 41 .

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.