From Obscurity to Opportunity: LpxH Emerges as a Promising Antibiotic Target in the Battle Against Gram-Negative Pathogens

Abstract

The surging crisis of multidrug-resistant Gram-negative pathogens underscores the urgent need for antibiotics with novel mechanisms of action. A promising strategy is to target previously unexploited pathways, such as lipid A biosynthesis. Lipid A functions as the membrane anchor of lipopolysaccharide and constitutes the outer monolayer of the outer membrane of Gram-negative bacteria. LpxH, a Mn²⁺-dependent phosphoesterase, catalyzes the conversion of UDP-2,3-diacylglucosamine to lipid X, a key precursor in lipid A production. Disruption of this essential step compromises outer membrane integrity, leading to bacterial death, making LpxH an attractive antibiotic target. Since AstraZeneca’s discovery of the first small-molecule LpxH inhibitor a decade ago, research has progressed substantially. The development of non-radioactive LpxH activity assays has enabled rapid screening and characterization of inhibitors. Structural and biochemical studies have revealed the architecture of LpxH and dynamic properties of the bound inhibitors, informing structure- and dynamics-based inhibitor design. Notably, recent breakthroughs from academic institutions and pharmaceutical companies have produced LpxH inhibitors with potent antibacterial activity against wild-type Enterobacterales in both in vitro and in vivo models. This review describes the biological role of LpxH and its paralogs, highlights recent advances in assay development and structural analysis, and surveys the current landscape of LpxH-targeting compounds in preclinical development. These collective advances establish LpxH as a novel target in the battle against multidrug-resistant Gram-negative infections and highlight a promising therapeutic opportunity that could reinvigorate the antibiotic pipeline.

Graphical Abstract

INTRODUCTION

Pathogenic bacteria pose a serious threat to public health. In 2019 alone, an estimated 7.7 million deaths were associated with 33 bacterial pathogens.¹ While antibiotics have been critical in treating bacterial infections, reducing infant mortality, and extending human lifespans,^{2, 3} their widespread use has inadvertently driven antimicrobial resistance (AMR), diminishing their effectiveness and severely limiting treatment options.⁴ This escalating AMR crisis has dramatically increased global healthcare costs and is projected to worsen significantly in the coming decades.⁵ Of particular concern is the rise of multidrug-resistant Gram-negative bacterial pathogens, including the extended-spectrum β-lactamase (ESBL)-producing Enterobacterales and carbapenem-resistant Enterobacterales (CREs), presenting a substantial challenge to the efficacy of current antibiotics.

Compounding this crisis is a significant shortfall in pharmaceutical research and development programs for novel antibiotics, leading to a paucity of truly innovative compounds in the pipeline. Alarmingly, no new class of antibiotics targeting Gram-negative bacilli has been approved since the 1980s.³ In the meanwhile, widespread antibiotic resistance is steadily eroding the efficacy of treatments, a problem particularly acute for Gram-negative pathogens. Their inherently impermeable outer membrane confers natural resistance to many antibiotics, further complicating treatment. The 2024 WHO Bacterial Priority Pathogens List (WHO BPPL) highlights the urgency of this issue, categorizing Gram-negative bacteria resistant to last-resort antibiotics—including Acinetobacter baumannii and various pathogens within the order Enterobacterales—as critical priorities.⁶ This designation reflects their ability to transfer resistance genes, the severity of the infections they cause, and their substantial global burden. Therefore, there is a desperate need for new antibiotics that target unexploited pathways by commercial antibiotics to combat the growing menace of multidrug-resistant Gram-negative bacteria. In this context, the lipid A biosynthetic pathway has emerged as a particularly promising avenue for therapeutic intervention.

Discovery of LpxH in Gram-Negative Bacterial Lipid A Biosynthesis

First demonstrated by Hans Christian Gram in 1884,⁷ the distinction between Gram-negative and Gram-positive bacteria remains a cornerstone of modern microbiology. Gram’s staining method revealed a defining characteristic of Gram-negative bacteria: a protective outer membrane shielding the cell. This barrier, largely impermeable to dyes, detergents, and many antibiotics, consists of lipopolysaccharide (LPS), a critical component of the bacterial envelope. The core component of LPS is lipid A (Figure 1A), a hydrophobic anchor embedded in the outer leaflet of the outer membrane. Lipid A is also known as endotoxin, as it is the active component responsible for the septic shock caused by Gram-negative bacterial infections.⁸ The constitutive biosynthesis of lipid A is essential for bacterial viability and fitness and is a key determinant of pathogenicity.⁸ Bacteria with compromised lipid A biosynthesis are hypersensitive to antibiotics due to increased outer membrane permeability.⁹ Lipid A biosynthesis remains unexploited by any commercial antibacterial therapies and presents an untapped opportunity for development of novel antibiotics.

Figure 1. Lipid A biosynthesis and non-radioactive LpxH activity assays.

(A) The Raetz pathway of lipid A biosynthesis. (B) Accumulation of lipid X in psgA-/psgB1-deficient E. coli. Left panel shows lipids from WT strain, whereas the right panel shows lipids from the psgA-/psgB1-deficient strain. Accumulation of lipid X is indicated with a red circle. Adapted from the original image by Nishijima et al. (ref. 10). (C) LpxH activity can be detected either through phosphate release via an AaLpxE-coupled malachite green assay (purple path) or through detection of UMP using the Promega UMP/CMP-Glo™ Glycosyltransferase Assay (green path). Modified from the original image by Lee et al. (ref. 30).

The elucidation of lipid A biosynthesis was catalyzed by the serendipitous discovery of lipid X, an unusual lipid species accumulated in a phosphatidylglycerol-deficient Escherichia coli (E. coli) mutant strain (Figure 1B).¹⁰ The chemical structure of lipid X was identified as 2,3-diacylglucosamine 1-phosphate, and it was recognized as a direct precursor to lipid A.^{11, 12} This finding marked a major turning point, laying the groundwork for conceptualizing a dedicated lipid A biosynthetic pathway.

In 1985, Raetz and colleagues identified UDP-N-acetylglucosamine (UDP-GlcNAc) as the starting substrate for lipid A synthesis.¹³ Remarkably, by systematically tracing lipid A intermediates in E. coli extracts without any knowledge of the enzymes, they delineated a multistep pathway of lipid A production, today known as the Raetz pathway.¹³ These insights served as a conceptual blueprint for the eventual discovery of all enzymes involved in lipid A biosynthesis (Figure 1A).⁸ The enzyme responsible for the production of lipid X was identified nearly two decades after the original detection of lipid X. This enzyme, known as LpxH, catalyzes the hydrolysis of UDP-diacylglucosamine (UDP-DAGn) to lipid X and constitutes the fourth step in lipid A biosynthesis.¹⁴

LpxH, a Mn²⁺-dependent calcineurin-like phosphoesterase (CLP),¹⁵ is unique among lipid A biosynthetic enzymes: although the enzymatic conversion of UDP-2,3-diacylglucosamine (UDP-DAGn) to lipid X is universally conserved across Gram-negative bacteria, LpxH itself is restricted to β- and γ-proteobacteria. These classes encompass a large percentage of clinically relevant pathogens, including Enterobacteriaceae, Pseudomonas aeruginosa (P. aeruginosa), and Acinetobacter baumannii (A. baumannii). In other bacterial lineages, this critical step is catalyzed by functional paralogs of LpxH (Figure 1A). For instance, α-proteobacteria utilize LpxI—a structurally and mechanistically unrelated enzyme—to perform the same reaction.^{16, 17} In Chlamydiae, this function is carried out by LpxG, which shares minimal sequence similarity with LpxH¹⁸.

The Raetz pathway has garnered considerable interest within the field of antibiotic development due to its essential role in maintaining bacterial viability and fitness, and in protecting Gram-negative bacteria from many commercial antibiotics.^19–23 In E. coli, inhibition of any of the first six steps of lipid A biosynthesis is lethal.^{24, 25} However, A. baumannii isolates have been observed with null mutations of lpxA, lpxC, and lpxD^{26, 27} and to survive without lipooligosasccharides in their outer membrane.^26–28 In contrast, deletion of lpxH in A. baumannii was fatal, due to the toxic accumulation of UDP-DAGn, which compromises inner membrane integrity.²⁹ This dual-killing mechanism—blocking an essential pathway while simultaneously causing the buildup of detergent-like lipid A intermediates—significantly reduces the likelihood of off-target resistance development. These features position LpxH as a particularly attractive antibiotic target. Given the clinical significance of A. baumannii and P. aeruginosa and the widespread prevalence of multidrug resistance in these pathogens, the pursuit of LpxH inhibitors represents a strategic and timely therapeutic endeavor. Such agents would not only address a critical unmet medical need but also offer an opportunity to introduce antibiotics with a novel mode of action and a potentially lower risk of resistance development.

Development of Non-Radioactive LpxH Activity Assays

Initially, the ³²P-autoradiographic TLC assay was the only method available to characterize LpxH.¹⁴ This assay was highly sensitive, allowing the detection of product accumulation down to picomolar concentrations. However, this method was costly and inconvenient due to the short half-life of ³²P, making routine testing of LpxH inhibitors challenging. To overcome this limitation, the Zhou lab at Duke University developed a non-radioactive protocol by leveraging the unique ability of Aquafex aeolicus LpxE (AaLpxE) to dephosphorylate lipid X as its non-native substrate (Figure 1C).^{30, 31} The release of an inorganic phosphate from lipid X can be quantitatively measured using the malachite green assay.³² A comparison of these two methods revealed nearly identical specific activity values of LpxH, indicating that the AaLpxE-coupled assay is suitable for quantitative measurement of LpxH activity.³⁰ The development of this non-radioactive assay eliminated the bottleneck in rapid evaluation of LpxH inhibitors, and the assay was successfully employed to establish an initial pharmacophore model of LpxH inhibitors.³⁰ Alternatively, the by-product UMP can be converted to ATP and quantified by the luciferase reaction using the UMP/CMP-Glo™ Glycosyltransferase Assay kit from Promega. The practicality of these non-radioactive assays has facilitated the discovery of novel LpxH inhibitors.

Discovery of AZ1, the First Reported Inhibitor of LpxH

The first reported inhibitor of LpxH, a sulfonyl-piperazine based small molecule hereafter referred to as AZ1 (Table 1), was discovered by Nayar and colleagues at AstraZeneca a decade ago through a high-throughput phenotypic screening campaign.³³ The screening aimed at detecting the disruption of cell wall biosynthesis in E. coli with a deficient efflux pump (ΔtolC). Whole-genome sequencing of isolated spontaneous resistant strains revealed that all mutants had single amino-acid substitutions in lpxH, strongly suggesting LpxH as the molecular target. This conclusion was further supported by target validation experiments showing that overexpression of lpxH reduced AZ1’s antibacterial activity. Further examination of E. coli incubated with a sub-lethal concentration of AZ1 showed elongated cell morphology and a loss of membrane integrity.

Table 1.

The structure and activity of various LpxH inhibitors reported in literature.

Team	Compound	Chemical Structure	Activity	Reference
AstraZeneca (Nayar et al. 2015)	AZ1		Ki: 146 nM (KpLpxH) Ki: 53.4 nM (EcLpxH) MIC: 0.25 μg/mL (Ec ATCC 25922 ΔtolC)	33
Duke (Cho et al. 2020)	JH-LPH-33		Ki: 10.5 nM (KpLpxH) Ki: 17.6 nM (EcLpxH) MIC: 1.6 μg/mL (Kp 10031)	34
Duke (Kwak et al. 2023)	JH-LPH-50		Ki: 3.1 nM (KpLpxH); MIC: 3.3 μg/mL (Kp 10031)	36
Duke (Ennis et al. 2024)	JH-LPH-86		Ki: 34.4 nM (KpLpxH) MIC: 0.25 μg/mL (Kp 10031)	37
	JH-LPH-88		Ki: 1289 nM (KpLpxH) MIC: 64 μg/mL (Kp 10031)	37
	JH-LPH-89		Ki: 998 nM (KpLpxH) MIC: >64 μg/mL (Kp 10031)	37
	JH-LPH-90		Ki: 45.4 nM (KpLpxH) MIC: 2.0 μg/mL (Kp 10031)	37
	JH-LPH-92		Ki: 1.9 nM (KpLpxH) MIC: 0.08 μg/mL (Kp 10031)	37
	JH-LPH-93		Ki: 40.5 nM (KpLpxH) MIC: 1.3 μg/mL (Kp 10031)	37
	JH-LPH-106		Ki: 0.02 nM (KpLpxH) Ki: 0.02 nM (EcLpxH) MIC: 0.63 μg/mL (Ec 25922) MIC: 0.04 μg/mL (Kp 10031)	37
	JH-LPH-107		Ki: 0.05 nM (KpLpxH) Ki: 0.05 nM (EcLpxH) MIC: 0.31 μg/mL (Ec 25922) MIC: 0.04 μg/mL (Kp 10031)	37
Roche (WO2023/061617A1, 2023)	017		MIC: 0.30 μg/mL (Ec 25922) MIC: 0.45 μg/mL (Kp 43816)	46
	047		MIC: 0.35 μg/mL (Ec 25922) MIC: 0.30 μg/mL (Kp 43816)	46
	058		MIC: 1.1 μg/mL (Ec 25922) MIC: 1.1 μg/mL (Kp 43816)	46
	067		MIC: 0.30 μg/mL (Ec 25922) MIC: 0.57 μg/mL (Kp 43816)	46
Roche (WO2023/072794A1, 2023)	5		MIC: 8.08 μg/mL (Ec 25922) MIC: 16.15 μg/mL (Kp 43816)	45
	23		MIC: 30.45 μg/mL (Ec 25922) MIC: 60.9 μg/mL (Kp 43816)	45
	78		MIC: 0.22 μg/mL (Ec 25922) MIC: 1.76 μg/mL (Kp 43816)	45
	73		MIC: 0.44 μg/mL (Ec 25922) MIC: 0.44 μg/mL (Kp 43816)	45
	76		MIC: 0.51 μg/mL (Ec 25922) MIC: 1.03 μg/mL (Kp 43816)	45
	77		MIC: 0.44 μg/mL (Ec 25922) MIC: 0.44 μg/mL (Kp 43816)	45
	116		MIC: 0.47 μg/mL (Ec 25922) MIC: 0.94 μg/mL (Kp 43816)	45
	79		MIC: 14.22 μg/mL (Ec 25922) MIC: >28.43 μg/mL (Kp 43816)	45
	82		MIC: 6.98 μg/mL (Ec 25922) MIC: 6.98 μg/mL (Kp 43816)	45
Roche (WO2023/166103A1. 2023)	065		MIC: 0.26 μg/mL (Ec 25922) MIC: 0.52 μg/mL (Kp 43816)	44
Roche (WO2024/213610A1. 2024)	036		MIC: 0.14 μg/mL (Ec 25922) MIC: 0.14 μg/mL (Kp 43816)	43
Roche (WO2024/213625A1. 2024)	026		MIC: 0.085 μg/mL (Ec 25922) MIC: 0.085 μg/mL (Kp 43816)	47
Uppsala (Huseby, et al. 2024)	EBL-3647		Ki: 1.7 nM (EcLpxH) MIC: 2.0 μg/mL (Ec 25922) MIC: 0.5 μg/mL (Kp 13883) in vivo active against Kp and Ec	41
	EBL-3599		Ki: 2.6 nM (EcLpxH) MIC: 2.0 μg/mL (Ec 25922) MIC: 1.0 μg/mL (Kp 13883) in vivo active against Kp and Ec	41
	JEDI-852		Ki: 982 nM (EcLpxH) MIC: >64 μg/mL (Kp 13833) MIC: >64 μg/mL (Ec 25922)	41
	JEDI-1444		Ki: 1.96 nM (EcLpxH) MIC: 2 μg/mL (Ec 25922) MIC: 2 μg/mL (Kp 13833)	41
Uppsala (Benediktsdottir et al. 2024)	S1		MIC: 1 μg/mL (Ec 25922)	42
	S2		MIC: 64 μg/mL (Ec 25922)	42
	S3		MIC: 64 μg/mL (Ec 25922)	42
	S4		MIC: 4 μg/mL (Ec 25922)	42

Ec: Escherichia coli; Kp: Klebsiella pneumoniae; Literature reported IC₅₀ values are converted to K_i values following the mode of competitive inhibition: IC₅₀ = K_i × (1 + [S]/K_M) using K_M values of 68.1 μM for KpLpxH and 61.7 μM for EcLpxH, respectively.

While AZ1 demonstrated activity against efflux-deficient E. coli, it lacked efficacy against wild-type (WT) Gram-negative pathogens, highlighting the need for further optimization. Consequently, AstraZeneca deprioritized the development of LpxH inhibitors; instead, researchers at academic labs—most notably at Duke University^{30, 34–39} and Uppsala University^40–42—and scientists at Hoffmann–La Roche (Roche)^43–47 pushed forward. These efforts have culminated in the development of significantly more potent LpxH inhibitors with robust activity against WT strains of Enterobacteriaceae, marking a major breakthrough in the development of LpxH-targeting antibiotics.

Crystal Structure of AZ1 Bound to LpxH

The discovery of AZ1 sparked intense efforts to elucidate its binding mode with LpxH. Shortly after its identification, crystal structures of lipid X-bound LpxH from Haemophilus influenzae (H. influenzae),¹⁵ P. aeruginosa,⁴⁸ and Klebsiella pneumoniae (K. pneumoniae)³⁴ (Figure 2A) were reported. These structures confirmed LpxH as a member of the CLP family, owing to its conserved CLP fold and the presence of a catalytic dimanganese cluster in the active site. These structural observations are consistent with the critical role of manganese in LpxH activity.⁴⁹ Distinct features of LpxH were also revealed from these studies: LpxH harbors a unique insertion lid that caps the CLP fold and forms a hydrophobic chamber between them to accommodate the 2-N-acyl chain shared by both the substrate and product. ¹⁸O incorporation studies suggested that catalytic activity stems from a Mn²⁺-activated water molecule that hydrolyzes the α-phosphate of the pyrophosphate bond.¹⁴ Mapping AZ1-resistant mutations onto these structures supported a model that AZ1 binds within the substrate-binding chamber, although the precise molecular details remained unclear.

Figure 2. Structural and biochemical characterization of the interaction of KpLpxH with its inhibitor, AZ1, and optimization of LpxH inhibitors aided by ligand dynamics.

(A) Crystal structure of the KpLpxH/lipid X complex (PDB: 6PH9). (B) Crystal structure of the KpLpxH/AZ1 complex (PDB: 6PIB). (C) Overlay of AZ1 with the product lipid X. (D) 2D KpLpxH-AZ1 interaction map. (E) Chemical structures of AZ1 and the diazirine-based photoaffinity probe, JH-LPH-13. (F) Location of photocrosslinked residues in the structural model of E. coli LpxH confirms that the phenyl group of AZ1 is located away from the active site. Reproduced from the original images by Cho et al. (ref. 34). (G) ¹⁹F NMR of AZ1 in the KpLpxH-unbound (free) and -bound states. Two distinct signals of AZ1 in the KpLpxH-bound state indicate the presence of two ligand conformations. (H) Exchange of AZ1 in two different KpLpxH-bound conformations delineates an expanded ligand envelope that guides the design of di-substituted AZ1 analog, JH-LPH-33. (I) Crystal structure of JH-LPH-33 bound to KpLpxH (PDB: 6PJ3). (J) Overlay of JH-LPH-33 with lipid X, showing that the chloride substitution occupies the space of the terminal methyl group of the product lipid X. (K) Omit map of AZ1. (L) Omit map of JH-LPH-33. In panels (K, L), the purple meshes represent 2mFo-DFc omit maps of AZ1 and JH-LPH-33 in the KpLpxH-bound complexes contoured at 1σ. Modified from the original images by Cho et al. (ref. 34).

Bohl and colleagues used the H. influenzae LpxH structure to generate docking models of AZ1.⁵⁰ Of the two predicted binding poses, the one placing the trifluoromethyl-substituted phenyl group near the catalytic Mn²⁺ cluster was considered compelling. However, this model was not supported by the crystal structure of AZ1 bound to K. pneumoniae LpxH (KpLpxH) (PDB: 6PIB) reported later.³⁴ Such a discrepancy underscores the importance of obtaining experimental structural data in guiding lead optimization.

Obtaining the AZ1–LpxH complex structure was particularly challenging due to the poor solubility and weak binding of AZ1. After four years of persistent effort, success was achieved by including AZ1 throughout the purification process, an approach that was only made possible with the support of large-scale chemical synthesis of AZ1. Consistent with previous spontaneous resistance mutation data,³³ AZ1 binds in a narrow hydrophobic chamber neighboring the active site, which accommodates the 2-N-acyl chain of both UDP-DAGn and lipid X (Figure 2A,B).³⁴ The sulfonyl group of AZ1 forms a distinct “L-shaped” bend, mirroring that found in lipid X. From this bend, the N-acetyl indoline moiety points toward the active site, while the piperazine–trifluoromethyl phenyl group extends to the chamber’s far end (Figure 2C). Multiple van der Waals contacts and key π–π and π–cation interactions were observed, notably between the terminal trifluoromethyl phenyl ring and Phe141 and between the indoline ring and Arg80. Hydrogen bonding of AZ1 with residues Asn79, Arg157, and Trp46 further stabilize the complex (Figure 2D). The extensive overlap of AZ1 with the LpxH substrate in crystal structures suggests that this class of compounds functions as competitive inhibitors. This was confirmed experimentally using a more potent AZ1 derivative.³⁴

As noted above, crystallographic analysis revealed that AZ1 binds in a direction opposite to the docking pose suggested by Bohl et al.⁵⁰ For confirmation, a diazirine-based photoaffinity labeling probe of AZ1 (JH-LPH-13, Figure 2E) was synthesized.³⁴ This compound was shown to inhibit LpxH similarly to AZ1,³⁴ and mass spectrometry analysis of the photocrosslinked complex confirmed that labeled residues were clustered at the distal end of the hydrophobic chamber (Figure 2F), providing unequivocal support for the inhibitor orientation observed in the crystal structure.

The molecular visualization of the AZ1-LpxH complex paved the way for accelerated development of LpxH-targeting antibiotics. Researchers at Duke, Uppsala, and Roche built on each other’s work, reporting IC₅₀ values from assays with different substrate concentrations and MIC values against different bacterial strains. For consistent comparison, we have converted the reported IC₅₀ values to K_i values and, whenever possible, will focus on the MIC comparison using E. coli 25922, which is a clinical reference strain used by all three groups to evaluate the potency of LpxH inhibitors.

Development of AZ1 Derivatives Driven by Insights into Dynamic Ligand Binding

Despite the well-defined electron density of AZ1 in the KpLpxH–AZ1 crystal structure, which revealed a single ligand-binding conformation, the solution-phase ¹⁹F NMR spectrum of AZ1 bound to KpLpxH displayed two distinct ¹⁹F signals.³⁴ Analysis of these signals indicated that AZ1 adopts two bound conformations: a predominant one (84.5%) and a minor one (15.5%) (Figure 2G). Since free AZ1 exhibited only a single ¹⁹F signal, the second ¹⁹F signal of AZ1 observed in complex with KpLpxH was attributed to an alternate binding conformation of the trifluoromethyl phenyl group. This alternate conformation likely arises from rotation around the bond connecting the substituted phenyl ring to the piperazine moiety, resulting in a ring flip of the trifluoromethyl phenyl group (Figure 2H). The major conformation corresponds to the crystal structure, while the minor conformation positions the ring-flipped trifluoromethyl group deeper into the tunnel, occupying the cavity typically filled by the terminal methyl of the 2-N-linked acyl chain in lipid X.

Based on the assumption that AZ1 has two rotameric conformations when bound to LpxH, the Zhou and Hong groups at Duke University explored derivatives of AZ1 with substitutions at the meta-position of the trifluoromethyl phenyl ring (Figure 2H). Structure–activity relationship (SAR) analysis revealed increased potency with substitutions in the order of H < F < CH₃ < Cl, followed by a decline with Br and CF₃.^{34, 35} The crystal structure of KpLpxH in complex with JH-LPH-33 (PDB: 6PJ3; Figure 2I) confirmed the existence of the secondary binding pocket, which is now occupied by the meta-substituted chlorine atom (Figure 2J–L).³⁴

In vitro analysis revealed that JH-LPH-33 was vastly more potent than AZ1. JH-LPH-33 exhibited K_i values of 11 nM against KpLpxH and 17 nM against E. coli LpxH (EcLpxH), representing 4- to 8-fold improvements over AZ1 (K_i values of 53 nM and 146 nM, respectively; see Table 1). Importantly, while AZ1 failed to inhibit the growth of K. pneumoniae 10031 at 64 μg/mL, JH-LPH-33 demonstrated a MIC at 1.6 μg/mL. Notably, the strategy of enhancing compound potency through meta-substitution of the trifluoromethyl phenyl group was independently pursued by other groups. The Roche team employed the same chloro-substitution,^43–45 while the Uppsala team explored a cyano (CN) substitution.⁴¹ Although meta-substitution alone was not sufficient for activity against WT E. coli W3110, JH-LPH-33 was the first LpxH inhibitor to demonstrate antibiotic activity against a WT bacterial strain: K. pneumoniae 10031.

Expansion of Chemical Space — A New Generation of LpxH Inhibitors

New designs by Duke University.

To overcome limited aqueous solubility of compounds like AZ1 and JH-LPH-33, Duke University researchers developed LpxH inhibitors that incorporated trifluoromethyl-substituted heteroaromatic rings (e.g., pyridine, thiazole, and thiadiazole) to expand their chemical space.³⁷ Compounds incorporating a five-membered heteroaromatic ring, or a pyridine ring with its nitrogen meta or para to the piperazine, exhibited low inhibitory activity (Table 1). In contrast, those featuring a pyridine ring with the nitrogen positioned ortho to the piperazine, such as JH-LPH-86 and JH-LPH-90, demonstrated markedly enhanced potency (Figure 3A). These analogs significantly improved the K_i of AZ1 from 146 nM to 34 nM for JH-LPH-86 (a 4-fold improvement) and 45 nM for JH-LPH-90 (a 3-fold improvement). This systematic approach, informed by structure–activity relationships, led to the identification of compounds with significantly improved LpxH inhibitory activity, demonstrating a clear path from an initial hit to more optimized lead candidates.

Figure 3. Evolution of LpxH inhibitors from AZ1 to pyridine and pyrimidine derivatives and extension into the active site by the Duke team.

(A) Iterative design and optimization strategy by the Duke team in developing potent LpxH inhibitors. The figure highlights the chemical structures, corresponding K_i values against KpLpxH, and fold improvement for various pyridine and pyrimidine derivatives of AZ1. (B) Development of metal-chelating inhibitor JH-LPH-50 from JH-LPH-33. (C) Crystal structure of the KpLpxH/JH-LPH-50 complex (PDB: 7SS7). LpxH is shown in the cartoon representation; JH-LPH-50 is shown in the stick model; and the dimanganese cluster is shown in spheres. The purple mesh represents the 2mFo-DFc omit map of JH-LPH-50 at 0.8σ. (D) Overlay of JH-LPH-50 with lipid X in complexes with KpLpxH (PDB: 6PH9). (E) Metal coordination of JH-LPH-50. Adapted from the original images by Kwak et al. (ref. 36).

X-ray crystallography of JH-LPH-86 and JH-LPH-90 bound to KpLpxH revealed binding modes highly similar to those of AZ1 and JH-LPH-33,^{34, 37} offering no clear structural basis for their enhanced activity. To elucidate this, Duke researchers employed symmetry-adapted perturbation theory (SAPT)^{51, 52} to analyze non-covalent π–π stacking interactions between the Phe141 residue of LpxH and the aromatic rings of the inhibitors. Computational analysis indicated that the improved activity of JH-LPH-86 and JH-LPH-90 resulted from more favorable interaction energies between their pyridine rings and Phe141, relative to the phenyl ring in AZ1.

Surprisingly, the pyrimidine ring analog (JH-LPH-93) exhibited weaker inhibition than JH-LPH-86 (Figure 3A). Similar to the meta-chlorine enhancement observed when AZ1 was modified to JH-LPH-33, incorporating a chlorine atom into JH-LPH-86 (K_i = 34 nM) resulted in the more potent JH-LPH-92 (K_i = 1.9 nM). Notably, the pyridine ring was also utilized in Roche compounds, while the Uppsala group primarily built their scaffold around the pyrimidine ring (vide infra), likely due to its enhanced aqueous solubility.⁴¹

A key observation is that AZ1 and its close derivatives, such as JH-LPH-33 and JH-LPH-86, do not extend into the LpxH active site. Rather, they bind within a nearby hydrophobic chamber normally occupied by the 2-N-acyl chain of the LpxH substrate.^{34, 37} This has led to the hypothesis that extending AZ1 derivatives into the LpxH active site could significantly enhance their inhibitory potency. In pursuit of this strategy, the Duke team introduced metal-binding groups into the existing scaffold of JH-LPH-33. The resulting compounds, represented by JH-LPH-50,³⁶ demonstrated markedly improved enzymatic inhibition against KpLpxH (K_i = 3.1 nM) (Figure 3B). The crystal structure of the KpLpxH/JH-LPH-50 complex resolved at 1.73 Å resolution showed that the only new interaction between KpLpxH and JH-LPH-50 was the bidentate Mn²⁺ coordination by the hydroxamic acid moiety of JH-LPH-50 (Figure 3C–E). This observation suggested that the improved LpxH inhibition by JH-LPH-50 was due to its coordination to the dimanganese cluster within the active site. Similar to earlier inhibitors, JH-LPH-50 anchors its sulfonyl piperazine tail group in the 2-N-acyl chain chamber, while its acyl hydroxamate moiety extends into the active site to chelate the metal ions. (Figure 3C,D). The N-hydroxyl group bridges both manganese ions, whereas the carbonyl oxygen coordinates a single manganese ion, resulting in pseudo-octahedral geometry at the coordinating manganese and pseudo-square pyramidal geometry at the non-coordinating one (Figure 3E). Despite its improved enzymatic inhibition, JH-LPH-50 did not show enhanced in vitro MICs, likely due to impaired bacterial cell penetration by the flexible acyl hydroxamate group.

New designs by Hoffmann–La Roche.

While efforts by the team at Duke University focused on extension into the active site, teams at Roche explored diverse alternative modifications, documented in five patents filed between 2023 and 2024.^43–47 Roche’s LpxH inhibitors share a common scaffold with those reported by research groups at Duke University and Uppsala University (vide infra), but feature extensive structural modifications to broaden the SAR of AZ1 (Figure 4). Initial patents^{45, 46} focused on compounds with pyrrolidinone or oxazolidinone extensions and replaced the trifluoromethyl group in AZ1 and JH-LPH-33 with a difluoromethylbenzene group. The pyrrolidinone/oxazolidinone ring was extensively explored, with amino pyrrolidinone being the most common fragment (for example, compound 78 in Figure 4A). The introduction of bulky substituents, such as gem-dimethyl (for example, compound 5 in Figure 4B) or tetrahydropyran with triol (for example, compound 23 in Figure 4B), led to reduced antibacterial activity. The most potent derivatives in patent WO2023/061617A1 incorporated a primary amine into the extended N-acyl chain (for example, compound 017 in Figure 4C; compounds 017, 047, 058, and 067 in Table 1),⁴⁶ while those in WO2023/072794A1 featured a difluoromethylbenzene group (compounds 73, 77, and 78 in Table 1).⁴⁵ Several of these compounds showed notable antibacterial activity (MIC values of 0.2–0.5 μg/mL against K. pneumoniae and E. coli). A later patent⁴⁴ introduced a new structural framework, where benzamide or indoline amide is linked to piperazinyl sulfonyl groups and replaces the pyrrolidinone or oxazolidinone rings. For example, compound 065 (Figure 4D), which features a long primary amine chain, exhibited potent activity against E. coli 25922 (MIC value of 0.26 μg/mL). The presence of two primary amine chains may improve its ability to penetrate bacterial cell membranes in accordance with the “eNTRy rules” (vide infra).^{53, 54} Additionally, extensive replacement of the chlorine atom incorporated in JH-LPH-33 with various moieties revealed that only substitution with a methyl group maintained comparable activity (Figure 4E).^{44, 45} Functional groups that were smaller, bulkier, or more polar than methyl substitution resulted in reduced antibacterial activity, as indicated by increased MIC (for example, compounds 79 and 82 in Table 1).

Figure 4. The structures and MIC values of representative LpxH inhibitors developed by the Roche team.

The structures of representative LpxH inhibitors developed by Roche are shown in panels (A) to (G). Modifications on the common core are shown in blue and pink. The compound codes, patent information, and in vitro antibiotic activities (MIC, μg/mL) against K. pneumoniae (Kp) and E. coli (Ec) are shown with each lead compound.

In a 2024 patent, Roche described a set of substituted rigid tricyclic compounds.⁴³ These tricyclic compounds, which incorporate a dihydro-2H-1,4-oxazine ring, showed improved activity against E. coli 25922 and K. pneumoniae 43816, exemplified by compound 036 (Figure 4F) (MIC = 0.14 μg/mL against E. coli 25922 and K. pneumoniae 43816, respectively). This compound’s sole difference from compound 116 (in WO2023/072794A1) lies in its oxazine ring,⁴⁵ suggesting that a rigid tricyclic ring system could improve antibacterial activity. Notably, compound 036 achieved the lowest MIC values against both E. coli 25922 and K. pneumoniae 43816. In addition to extending the compounds to reach deeper into the active site, the Roche team significantly improved compound membrane permeability by incorporating primary amines onto the central scaffold, aggressively leveraging the eNTRy rule.^{53, 54} The eNTRy rules for Gram-negative cell permeability are a set of heuristics suggesting preferable traits among small molecules with high membrane permeability in E. coli and closely related bacteria, emphasizing the inclusion of ionizable nitrogen atoms (preferably primary amines), low three-dimensionality/high globularity, and high rigidity (a low number of rotatable bonds).⁵⁴ A series of difluoro-alkyl tails containing primary amine groups were attached to the pyridine ring of LpxH inhibitors (Figure 4G, compound 026).^{46, 47} The Roche team reported impressive MIC values (< 0.1 μg/mL) for compound 026 (0.085 μg/mL against both E. coli 25922 and K. pneumoniae 43816).

New designs by Uppsala University.

Researchers at Uppsala University also leveraged the eNTRy rules, combining the incorporation of primary amines with the development of a new scaffold.^{41, 42} The Uppsala team conducted a series of phenotypic screens of publicly available chemical libraries against efflux-deficient E. coli to identify compounds with antibacterial activity against Gram-negative bacteria. These efforts led the Uppsala team to identify JEDI-852 from the PubChem Library (PubChem AID 573). Despite showing no activity against WT E. coli, JEDI-852 demonstrated an MIC of 4–8 μg/mL against efflux pump-deficient E. coli (ΔtolC) and low cytotoxicity (IC₅₀ >32 μM) against HepG2 cells. These preliminary results demonstrated the early antibiotic potential of JEDI-852. Further investigation through whole genome sequencing of resistant strains identified mutations in lpxH, providing strong evidence that LpxH is the probable target of JEDI-852.

Although the MIC of JEDI-852 against efflux pump–deficient E. coli was promising, the poor antibacterial activity of JEDI-852 against WT Gram-negative bacteria warranted further optimization. Recognizing similarities in the structures of JEDI-852 and AZ1, the Uppsala team combined key elements of these two compounds, resulting in JEDI-1444. Unlike JEDI-852, JEDI-1444 showed good activity against efflux-proficient WT E. coli 25922 (MIC = 2 μg/mL) and K. pneumoniae 13833 (MIC = 2 μg/mL). A low frequency of resistance for JEDI-1444 was observed in studies with efflux-deficient E. coli. In 2022, several compounds derived from JEDI were patented,⁴⁰ with 15 inhibitors exhibiting MIC values ≤0.125 μg/mL against WT E. coli or K. pneumoniae (Figure 5).

Figure 5. The structures, MIC values, and binding modes of representative LpxH inhibitors developed by the Uppsala team.

(A) Key structural motifs incorporated into JEDI-1444 from AZ1 and JEDI-852 are highlighted in pink and blue, respectively. Unique structural features in EBL-3647 and EBL-3599 are highlighted in orange. The compound codes and in vitro antibiotic activities (MIC, μg/mL) against K. pneumoniae and E. coli are shown with each lead compound. (B) Binding mode of lipid X (green; PDB: 6PH9) in the KpLpxH active site. Only the receptor residues Asn79, Arg80 and His195, together with the two Mn²⁺ ions, are shown for clarity. Hydrogen bonds are depicted as yellow dashed lines. (C) Binding mode of JEDI-852 (salmon; PDB: 8QKA). (D) Binding mode of EBL-3647 (cyan; PDB: 8QK5).

Despite JEDI-1444’s promising activity against Gram-negative bacteria, it faced several limitations, such as low solubility, poor metabolic stability, and reduced antibiotic activity in the presence of serum. To address these limitations, the Uppsala team adopted a combination of hit expansion, SAR studies, and structure-based drug design (SBDD) aimed at improving ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties.⁴¹ The poor ADMET properties of JEDI-1444 were due to its relatively high lipophilicity and aromaticity. Therefore, the Uppsala team’s optimization strategy focused on preparing analogs with reduced lipophilicity (quantified as calculated logD or clogD).

Similar to Roche, the Uppsala team leveraged the eNTRy rule, attaching primary amine groups to their lead compounds to enhance compound accumulation in Gram-negative bacteria. These polar extensions mimic the solvent exposed acyl arm of lipid X. To reduce overall lipophilicity while maintaining target affinity with the lipophilic binding pocket of the trifluoromethyl substituted phenyl group, they also incorporated heterocycles akin to the pyridine inclusion employed by the Duke team. The resulting optimization of π–π stacking interactions between the heteroaromatic ring and Phe141 presumably counteracted the loss in target affinity with the lipophilic binding pocket. The combination of the polar extension and pyrimidine ring incorporation yielded EBL-3647 and EBL-3599, which had lipophilicity (clogD) values improved by 4 to 5 log units compared to JEDI-1444. EBL-3647 and EBL-3599 both demonstrated enhanced MIC activity in the presence of serum, over 5,000-fold greater thermodynamic solubility, and a more than 25-fold increase in stability in human- and mouse-liver microsomes. The enhanced stability is attributed to the reduced affinity of polar extension compounds for the lipophilic CYP450 enzyme, which mitigates CYP450-mediated oxidative N-demethylation.

Both EBL-3647 and EBL-3599 have MICs of 2 μg/mL against E. coli 25922 and 0.5–1 μg/mL against K. pneumoniae 13883 (Figure 5). These compounds also have measurable MICs against efflux-defective A. baumannii, P. aeruginosa, and P. mirabilis, with MICs ranging from 4–64 μg/mL). More importantly, these compounds are active in vivo. Doses of EBL-3647 or EBL-3599 at ≥50 mg/kg yielded efficacy similar to the positive control group of ciprofloxacin (13 mg/kg) against K. pneumoniae.⁴¹

The crystal structures of JEDI- and EBL-series compounds bound to EcLpxH and KpLpxH provided valuable insights into the key role played by the N-methyl-N-phenyl-methanesulfonamide moiety.⁴¹ The crystal structure of lipid X binding to KpLpxH shows hydrogen bonding between a phosphate oxygen and Arg80 as well as hydrogen bonding between the amide carbonyl oxygen and Asn79 (Figure 5B). LpxH inhibitors—including AZ1 and most others—incorporate a common amide bond occupying the same space as the lipid X amide and share the same carbonyl oxygen–Asn79 hydrogen bonding. The lipid X phosphate oxygen–Arg80 hydrogen bonding interaction, however, was not captured in LpxH inhibitors until JEDI-852. The N-methyl-N-phenyl-methanesulfonamide oxygens of JEDI-852 are hydrogen bond acceptors for both Asn79 and Arg80 (Figure 5C), replicating the native hydrogen bonding interaction. Later N-methylsulfonamide derivatives have the same N-methylsulfonamide oxygen Asn79/Arg80 interactions, as seen with EBL-3647 (Figure 5D). These biomimetic substitutions strongly improved compound interactions with the LpxH binding pocket.

The Uppsala team further explored the impact of their ortho-N-methyl-sulfonamide group by shifting the position the N-methyl-sulfonamide to the meta-position.⁴² While the meta-N-methyl-sulfonamide analog (S3) lost activity against WT E. coli compared to its ortho-substituted analog, removal of the N-methyl group (S4) restored activity against WT E. coli (MIC = 4 μg/mL). Because both S3 and S4 were equally active against the efflux-deficient strain, the efflux susceptibility of the meta-N-methyl compounds may be controlled by the N-methyl group.

Enhancing the potency of LpxH inhibitors through hybridization of key fragments from existing compounds.

The N-methyl-N-phenyl-methanesulfonamide group of Uppsala compounds appears to be a versatile moiety, as its incorporation into the Duke scaffold further improved the activity of JH-LPH-92.³⁷ When tested against E. coli strain 25922, Duke’s compounds (JH-LPH-106 and JH-LPH-107) exhibited 3–7 fold greater antibiotic activity than the Uppsala’s lead compounds (EBL-3647 and EBL-3599), with MIC values of 0.3–0.6 μg/mL (Figure 6). This enhanced potency suggests a synergistic effect between the structural features of Duke and Uppsala LpxH inhibitors. The promising MIC values further validate the strategy of fragment hybridization in optimizing LpxH inhibitors. Future modifications focusing on the central scaffold and peripheral structural moieties may yield more potent LpxH inhibitors with broad-spectrum activity.

Figure 6. Lead LpxH inhibitors by Duke University.

Incorporation of the ortho-N-methyl-N-phenyl-methanesulfonamide moiety (highlighted in blue) into the Duke scaffold results in the most potent LpxH inhibitors reported to date.

Mechanism of Antibiotic Activity and Spontaneous Resistance Mutations

The development of potent LpxH inhibitors has enabled deeper investigation into their mechanism of action. These compounds function as bactericidal antibiotics, reducing bacterial viability by >1,000-fold within 6 h in vitro (Figure 7A).³⁷ Remarkably, they remain effective against strains resistant to currently available antibiotics, ⁴¹ underscoring LpxH as a promising new target within a previously untapped biosynthetic pathway.

Figure 7. LpxH inhibitors exhibit bactericidal activity and a low frequency of spontaneous resistance.

(A) Time-kill kinetics of JH-LPH-107 against K. pneumoniae 10031. Error bars represent S.E.M. (n=3). Adapted from the original images by Ennis et al. (ref. 37). (B) Spontaneous resistance development of select LpxH inhibitors. ^a Spontaneous resistance was measured at 4× MIC. Values shown in parentheses were measured at 8× MIC. ^b Resistance for JH-LPH-107 was measured against ATCC 10031. Resistances for EBL-3599 and EBL-3647 were measured against ATCC 13883. ^c Resistance for JH-LPH-107 was measured after incubation for 48 h. ^d Resistances for EBL-3599 and EBL-3647 were measured after incubation for 24 h.

Spontaneous resistance mutation rates to LpxH inhibitors are exceptionally low, typically around 10^–9 when measured at 4× MIC against E. coli 25922 and K. pneumoniae 10031 or 13883 (Figure 7B).^{37, 41} Sequence analysis of resistant mutants revealed that nearly all mutations were localized to the lpxH gene.^{37, 41} The sole exception was a mutation in the ramR gene isolated from K. pneumoniae 13883,⁴¹ which regulates the overexpression of efflux pumps. Collectively, these findings support the view that inhibiting LpxH is an effective antibacterial strategy because bacteria cannot easily compensate for the loss of the LpxH function, likely due to the toxic accumulation of lipid intermediates.

FUTURE PERSPECTIVES

A decade after the initial report of the first LpxH inhibitor AZ1 in 2015,³³ the field has witnessed remarkable progress in developing LpxH-targeting antibiotics. AZ1 exhibited limited antibiotic activity, being effective only against efflux-deficient E. coli, but not against the WT strain. However, independent efforts by researchers at Duke, Uppsala, and Roche have led to the development of next-generation LpxH inhibitors. These new inhibitors are highly potent, boasting sub-nM K_i values, and exhibit antibiotic activity against WT Enterobacteriaceae both in vitro and in vivo.^{37, 41, 43–47} Furthermore, they demonstrate excellent safety profiles, showing minimal to no cytotoxicity in HEK293 and HepG2 cells, even at concentrations 100-fold higher than their MIC values.³⁷

Despite these advances, major challenges remain. Firstly, membrane permeability continues to restrict the antibacterial activity of LpxH inhibitors. While efforts based on the eNTRy rules^{53, 54} have shown promise for improved cell permeability, this strategy is restricted to E. coli and closely related bacterial species, such as K. pneumoniae. Membrane penetration of P. aeruginosa appears to be porin-independent and requires a different set of properties defined by the PASsagE (Pseudomonas Aeruginosa self-promoted entry) rule,^{55, 56} whereas A. baumannii presents another challenge, as its outer membrane contains LOS rather than LPS,^{57, 58} suggesting another permeability profile. Emphasis on one heuristic (i.e. eNTRy versus PASsagE) may incorporate design strategies or moieties that disfavor the other and limit broad spectrum permeability. For example, eNTRy emphasizes low three-dimensionality and rigidification, whereas PASsagE does not define these variables; efforts to promote these variables in compound design may disfavor variables under PASsagE. These differences underscore the need for continued investigation of delivery strategies.

Secondly, none of the LpxH inhibitors developed thus far have demonstrated antibiotic activity against WT P. aeruginosa or A. baumannii—two clinically important Gram-negative pathogens that both rely on LpxH. In addition to variations in membrane composition that affect compound permeability, this lack of activity may reflect key sequence or structural differences in LpxH that reduce inhibitor binding. Further optimization will therefore be required to generate effective compounds against these challenging pathogens and to expand the spectrum of LpxH-targeting antibiotics. Notably, LpxH is found in ~70% of Gram-negative bacteria, while the remaining species reply on LpxI or LpxG. Although these enzymes either belong to an entirely different structural family or show very low sequence similarity to LpxH, they act on similar substrates. Hence, it remains to be established whether current LpxH inhibitors could also target these paralogues.

Finally, while the Uppsala team reported in vivo efficacy for both EBL-3647 and EBL-3599 in the murine peritonitis model,^{37, 41} percentage animal survival was not reported. It is also crucial to establish in vivo activity of LpxH inhibitors in other clinically relevant disease models, such as urinary tract infection (UTI) and pneumococcal infection models. As LpxH inhibitors move toward preclinical and clinical studies, careful consideration of their ADMET and pharmacokinetic/pharmacodynamic (PK/PD) properties will also be essential. Investigating potential synergistic effects with existing antibiotics presents another promising research direction for these LpxH inhibitors.

Historically, preclinical antibiotic development targeting the Raetz pathway has largely focused on LpxC, the second enzyme in lipid A biosynthesis.^{20, 59–62} However, after a decade of intense research by both academic and pharmaceutical teams, inhibitors of LpxH, the fourth enzyme in the Raetz pathway, have now emerged as compelling candidates in the fight against multidrug-resistant Gram-negative infections. In comparison with LpxC inhibitors, developing LpxH-targeting antibiotics is uniquely advantageous. First, LpxH inhibitors fall into entirely different chemical classes from LpxC inhibitors, and hence they are unlikely to induce the same cardiovascular toxicity that plagued the development of LpxC inhibitors in clinical trials.⁶³ Second, LpxH, but not LpxC, is essential in A. baumannii,²⁹ suggesting that potent A. baumannii LpxH inhibitors could be highly effective antibiotics for this clinically important pathogen. Third, the dual toxicity caused by disruption of lipid A biosynthesis and accumulation of toxic lipid A intermediates suggests that LpxH inhibition cannot be overcome though compensation of other pathways (e.g., resistance to LpxC inhibitors caused by reduced fatty acid biosynthesis due to fabZ mutation^{64, 65}). These advantages highlight the benefits of developing novel antibiotics targeting LpxH. The continued development and application of drug design tools assisted by artificial intelligence (AI) represent an exciting opportunity for optimizing these inhibitors. Sustained progress—fueled by detailed structural analysis, ligand dynamics studies, and AI-assisted drug design—holds great promise for developing clinically effective LpxH-targeting antibiotics for treating multidrug-resistant Gram-negative bacterial infections.