Fragment-Based Discovery of 6-Azaindazoles As Inhibitors of Bacterial DNA Ligase

Abstract

Herein we describe the application of fragment-based drug design to bacterial DNA ligase. X-ray crystallography was used to guide structure-based optimization of a fragment-screening hit to give novel, nanomolar, AMP-competitive inhibitors. The lead compound 13 showed antibacterial activity across a range of pathogens. Data to demonstrate mode of action was provided using a strain of S. aureus, engineered to overexpress DNA ligase.

Bacterial DNA ligase (LigA) is an NAD⁺-dependent enzyme, which is essential for DNA replication and has attracted interest as a novel target for antibacterial therapy.¹ LigA is responsible for ligating two strands of DNA via the formation of a phosphodiester bond between the 3′-hydroxyl end of one oligonucleotide and the 5′-phosphate end of another.^2,3 This process involves a three-step mechanism. Initially, reaction between NAD⁺ and an active-site lysine leads to an adenylated form of the protein. Enzyme-bound AMP is then transferred to the 5′-phosphate end of a nicked DNA strand. Finally, attack on the AMP-DNA bond by the 3′-hydroxyl of a second strand of DNA seals the phosphate backbone and releases AMP. LigA has been shown to be essential for viability in all Gram-positive and Gram-negative organisms tested to date.^4,5 It is highly conserved across bacterial species and is phylogenetically quite distinct from its human, ATP-dependent, counterpart. This provides encouragement that inhibitors of LigA may exhibit both broad-spectrum antibacterial activity and selectivity over human isozymes.⁶

Widespread bacterial resistance to current classes of approved antibiotics has led to an unmet medical need for compounds with novel modes of action that are not compromised by pre-existing resistance mechanisms.⁷ As a clinically unexploited target, inhibitors of LigA would fall into this category, and several classes of compounds have been reported in the literature.⁸⁻¹¹ In most cases, high throughput screening (HTS) hits have provided chemical starting points, which were then optimized, often using structure-based drug design. Notable examples, shown to exhibit target-mediated antibacterial activity, include the adenosine analogue 1 and naphthyridine 2 (Figure 1).^8,10

Figure 1.

Inhibitors of DNA ligase described in the literature.^8,10

Although in cases such as LigA molecular targeted HTS has delivered chemical leads, success rates from antibacterial HTS are typically lower than for targets from other therapeutic areas.¹² In this letter, we describe an alternative approach to the discovery of LigA inhibitors using fragment-based drug design (FBDD).¹³ FBDD has become established as an approach for the generation of chemical leads for drug targets and has been employed in a variety of therapeutic areas, including antibacterials.^14,15 In this approach specialized detection methods are used to identify small chemical compounds, fragments (MW ≤ 16 heavy atoms) that bind to the drug target, and structural biology is usually employed to establish their binding mode and to facilitate their optimization. A key feature of fragment-derived leads is that they are, on average, significantly smaller and less lipophilic than historical leads derived from HTS.¹⁶ Given that successful antibiotics are typically associated with lower lipophilicity than other drug classes,^17,18 FBDD seems well suited to this therapeutic area. To our knowledge, this is the first published example of a fragment-derived LigA inhibitor.

An essential prerequisite to a successful fragment screening cascade for LigA was the identification of a stable, deadenylated, soakable crystal form of the protein. Constructs from multiple pathogens, including E. faecalis, H. influenzae, S. aureus, and E. coli, were prepared and subject to deadenylation by incubation with nicotinamide mononucleotide (NMN). LigA constructs showing reproducible and sustained levels of complete deadenylation (as determined by mass spectrometry) were progressesd to large scale crystallization trials. Of these, LigA from S. aureus provided the most stable soakable crystal form of the enzyme, and this was selected as the preferred system for X-ray crystallographic screening. Approximately 1500 compounds from our fragment library were then screened using a combination of high throughput LigA X-ray crystallography, ligand observed NMR (via water LOGSY), and a thermal shift (T_m) assay. Hits were then followed up using isothermal titration calorimetry (ITC) to determine binding affinities and ligand efficiencies (LE).¹⁹

The starting point for chemical optimization was provided by the fragment-based screening hit 3 [Figure 2, K_d (ITC) = 38 μM, ligand efficiency, LE = 0.50]. The X-ray crystallographic structure of LigA (S. aureus) in complex with 3 revealed that the fragment binds in the AMP pocket and forms hydrogen bonds with the side chains of Lys283 and Glu110 (Figure 3a). An additional, water-mediated, hydrogen bond exists between the triazole nitrogen and the protein backbone NH of Ile113. Further interaction is provided by the side chain of Tyr219, which partially stacks with the pyrazine ring.

Figure 2.

Fragment optimization and growth of 3 toward 13. Data presented as K_d (ITC) or IC₅₀ (biochemical assay) using DNA ligase from S. aureus.

Figure 3.

(a) X-ray crystal structure of fragment 3 bound to LigA (S. aureus) showing key hydrogen bonds (purple dotted lines), a key water molecule (red sphere), and partially resected Connoly surface. (gray). Growth vectors toward the hydrophobic pocket, ribose binding region, and the protein backbone are shown by the red arrows. (b) X-ray crystal structure of 12 bound to LigA (S. aureus).

Pyrazine 3 offered an attractive starting point for structure-based optimization for three reasons. First, 3 was one of the most ligand efficient hits observed for LigA. Second, the 2-chloro group partly fills the hydrophobic pocket (Figure 3a), a region of the active site which is occluded in the human form of the enzyme, providing an opportunity to obtain selectivity.²⁰ Third, replacing the pyrazine with a pyridine would provide a growth vector to explore the ribose-binding region, via the pyridyl 4-position, provided this replacement was tolerated.

The pyridine analogue 4 [K_d (ITC) = 10 μM, IC₅₀ = 17 μM, LE = 0.54] (Figure 2) was prepared, which, gratifyingly, showed improved activity over the fragment hit 3. One concern was the potentially reactive chlorine at position 2. To mitigate this risk, a number of alternative groups were explored at this position. Deletion of the 2-substituent (compound 5) or introduction of an electron donating substituent (e.g., OMe analogue 6) were both detrimental to activity. 2-Alkyl substituents (e.g., cyclobutyl 8), although appearing to fill the hydrophobic region more effectively than chlorine, surprisingly offered little advantage. These findings suggested that the electronics of the 2-substituent may be more important than sterics and that increasing the electron withdrawing nature of the alkyl group could be advantageous. The trifluoromethyl analogue 10 (IC₅₀ = 16 μM, LE = 0.44) was synthesized, which proved equipotent with 4, albeit with a modest loss in ligand efficiency.

As described above, one of the interactions formed by these compounds is a water-mediated hydrogen bond between the triazole 4-position nitrogen and the backbone NH of Ile 113 (shown in Figure 3a). Our next strategy was to expand the triazole ring, in order to fill the pocket more effectively, displace this bridging water molecule and form a direct hydrogen bond with Ile 113.

To this end, the 6-azaindazole analogue 11 (IC₅₀ = 0.22 μM, LE = 0.48) was prepared, which proved to be ∼70-fold more potent than 10. An X-ray structure of 11 could not be obtained, which we ascribed to the low solubility (solubility = 12 μg/mL)²¹ of this compound. Next, attention was turned toward exploring substitution at the pyridine 4-position in order to access the ribose-binding pocket. The more soluble ethanolamine derivative 12 (solubility = 113 μg/mL) was prepared, which was comparable in potency to 11 and enabled an X-ray structure of the protein–ligand complex to be obtained. The structure of 12 complexed with DNA ligase clearly shows N6 of the 6-azaindazole making the intended hydrogen-bond to the backbone NH of Ile113 (Figure 3b). The other interactions are conserved. An inspection of the ribose-binding region indicates a hydrogen-bond between the backbone carbonyl of Leu80 and the 4-aminopyridine substituent. However, the electron density around the last two atoms of the ethanolamine chain was ambiguous, and therefore, the exact position of the OH group is uncertain. Although the ethanolamine motif did not improve potency in the biochemical assay, it did result in 12 being the first analogue to show signs of antibacterial activity, particularly against S. aureus [MIC S.aureus (Oxford) = 8 μg/mL] (Table 1).

Table 1. MIC Data for Compounds 1, 12, and 13.

	MIC (μg/mL)
compd	1a	12	13
Gram-Positive Pathogens
S. aureus Oxford	8	8	4
S. pneumoniae 1629	2	64	2
S. pyogenes 1307006P	8	128	8
Gram-Negative Pathogens
H. influenzae H128	16	128	16
E. coli 7623	>128	>128	>128
K. pneumonia 1161486	>128	>128	>128
P. aeruginosa PAO(MV)	>128	>128	>128
A. baumannii BM4454	>128	>128	>128
Gram-Negative Efflux Knockout Strainsb
H. Influenzae H128 Acr B-	2	8	1
E. coli 7623 TolC-	8	128	16
K. pneumonia 1161486a TolC-	8	>128	32
P. aeruginosa PAO322	64	>128	32
A. baumannii BM4652 (efflux mutant)	NT	>128	32
cytotox IC50 (μg/mL)	NT	16	>36

^aCompound described in ref (8).

^bEngineered bacterial strains lacking key efflux transporter proteins (e.g., AcrB, TolC).

Further analysis of the binding mode led us to the hypothesis that compound 12 almost certainly binds in an unfavorable conformation.²² This is due to both steric (hydrogens) and electrostatic (heterocyclic nitrogen lone pairs) clashes between the pyridine and azaindazole rings (Figure 4). This is not the case in the more favored (unbound) conformation. Replacing the pyridine with a pyrimidine would remove the steric hindrance in the bound conformation and also create a clash (between nitrogen lone pairs) in the unbound conformation. Overall, this would contribute to stabilization of the bound state and therefore, hopefully, an increase in potency. Consequently, pyrimidine 13 [K_d (ITC) = 0.025 μM, LE = 0.45] (Figure 2) was prepared, which proved to be 15-fold more potent (compared to 12).

Figure 4.

Illustration showing the alternative conformations of 12 (ethanolamine side chain removed for clarity).²² Electrostatic and steric clashes are shown in red, and positive interactions are shown in black.

Compound 13 exhibited antibacterial activities (Table 1), with MIC values of 8 μg/mL or less against wild-type Gram-positive pathogens S. aureus, S. pneumoniae, and S. pyogenes. Moderate activities (16–32 μg/mL) were also seen against efflux mutants of Gram-negative pathogens such as E. coli, P. aeruginosa, and A. baumannii. The lack of activity against most wild-type Gram-negative strains suggested that compound 13 was subject to active efflux out of the cell. This may also be compounded with inadequate cell permeability. General cytotoxicity against a mouse lung lymphoma cell line was not observed for compound 13 up to the highest concentration tested (100 μM). MIC data for literature compound 1 is shown for comparison.⁸

Antibacterial activity was determined to be caused by inhibition of LigA using a strain of S. aureus engineered to overexpress DNA ligase (Table 2). MICs of compound 13 were found to correlate with expression levels of DNA ligase.

Table 2. MIC for 13 in S. aureus Parent and LigA Overexpression Strains^a.

	MIC (μg/mL) vs S. aureus RN4220
	pYH4		pYH4-YerG (DNA ligase)
compd	–	+	–	+
13	8	8	32/16	128

^apYH4 = vector control. pYH4-YerG = vector + DNA gene. +/– = with or without inducer (0.1 μg/mL anhydrotetracycline).

This result is consistent with a DNA-ligase mediated mode of antibacterial action in this species. In the Gram-negative pathogen E. coli, the mode of action was also tracked to LigA inhibition by showing cross-resistance with LigA target mutants (Table 3).²³

Table 3. MIC for 13 in E. coli Efflux Mutant Parent and LigA Target Mutant Strains.

	MIC (μg/mL)
E. coli	compd 13
TOP10 ΔTolC	8
TOP10 ΔTolC LigA G180E	128/64
TOP10 ΔTolC LigA R150S	128/64
TOP10 ΔTolC LigA R518H	64
E. coli LigA IC50 (uM)	0.097

In summary, we have identified 13 (K_d = 25 nM, LE = 0.45, LLE_AT = 0.45²⁴) as an inhibitor of bacterial DNA ligase. Starting from pyrazine fragment 3 (K_d = 38 μM, LE = 0.50, LLE_AT = 0.56), X-ray crystallographic data was used to establish key determinants for affinity and to guide structure based design. In particular, a strategy of establishing additional hydrogen bonds to the protein backbone and stabilizing the enzyme-bound conformation led to over a 1000-fold increase in activity, relative to starting point 3. Good LE and lipophilic ligand efficiency (LLE_AT) were maintained during this process. Compound 13 demonstrated single-digit MICs across a range of Gram-positive pathogens, which, in the case of S. aureus, was shown to be target mediated. The 6-azaindaole scaffold provides a novel, nonpurine, chemotype to the LigA field, and to our knowledge, 13 is the first published example of a fragment-derived LigA inhibitor. Moreover, this work provides further validation of FBDD as an effective, complementary approach in the field of antibacterials.

Supporting Information Available

Assay conditions (including error limits and data for reference compounds), microbiology methods, biophysical methods, and synthetic procedure/characterization of compounds 4–13. This material is available free of charge via the Internet at http://pubs.acs.org.

The authors declare no competing financial interest.