Target selection is the first step in our drug discovery pipeline. A systematic analysis of literature on gene essentiality is performed in order to select essential drug-targets. These studies include various molecular genetic approaches, such as gene trapping, homologous recombination, transposon insertion mutagenesis (Van Opijnen and Camilli, 2013; Chao et al., 2016), and more recently developed CRISPR/CAS9 gene editing system (Wang et al., 2015; Evers et al., 2016; Morgens et al., 2016; Senturk et al., 2017; Bartha et al., 2018; You et al., 2020), to identify essential genes by first incorporating inactivating mutations in them and then determining the viability of the cells or organism carrying these mutations. The next step is to look for the similarity of these essential genes to those of the human genome. Targets similar to a gene in the human genome or human gut microbiome are not pursued in order to avoid cross-reactivity of the developed antibiotic and maintain the health of the microbiome. The protein products of these genes are considered for selecting a target. This is followed by the ligandability/druggability analysis by detecting and evaluating the ligand binding sites using modelling programs, such as the α-shape based approach (Edelsbrunner and Mücke, 1994) in MOE (MOLECULAR OPERATING ENVIRONMENT, 2016 Integrated Computer-Aided Molecular Design Platform), SiteMap (Halgren, 2009), LIGSITE (Huang and Schroeder, 2006), CASTp (Dundas et al., 2006) and more recently by generating fragment HOTSPOT maps (Radoux et al., 2016), taking advantage of the available 3-dimensional structure for potential target proteins. Cryptic binding sites can be detected by molecular dynamics simulations (Kuzmanic et al., 2020) and the cosolvent mapping method (Yang and Wang, 2010). In cases where the 3-dimensional structures are not available, they can be modelled using software such as Modeller (Šali and Blundell, 1993; Eswar et al., 2006) and more recently developed deep-learning algorithms, such as AlphaFold (Senior et al., 2020), AlphaFold2 (Jumper et al., 2021) and RoseTTAFold (Baek et al., 2021), and the resulting structure can be then analysed for ligandability. Other criteria while selecting a target include preference for non-membrane proteins and availability of well-established biochemical/functional assays in order to achieve a speedy drug-discovery process.

Once the target protein is selected, the corresponding gene is cloned into an expression vector using molecular biology techniques such as PCR (Mullis, 1990; Garibyan and Avashia, 2013; Hoseini and Sauer, 2015), restriction digestion (Roberts, 2005) and ligation (Cohen et al., 1973) resulting in an expression plasmid. The target protein is often overexpressed in Escherichia coli using this expression plasmid. The overexpressed protein is purified in large quantity using protein chemistry techniques such as affinity chromatography (Rodriguez et al., 2020), ion-exchange chromatography (Walls and Walker, 2017) and size exclusion chromatography (Burgess, 2018; Held and Kilz, 2021).

Designing/selecting a generic fragment library for a fragment-screening campaign in general involves considerations such as diversity, solubility, molecular weight (MW), cLogP, polar surface area (PSA), Fsp3 (Lovering et al., 2009; Wei et al., 2020), natural product-likeness and the number of the fragments (Mahmood and Ramachandraiah, 2021). On the other hand, there are certain criteria that are important to consider while assembling/choosing a fragment library with high chemical and structural diversity to be used for screening against a selected target. These involve ligand and protein structure-based approaches of library design/selection (Schuffenhauer et al., 2005). The ligand-based approach involves the knowledge of binding mode of known ligands for the target and the fragments that can undergo similar interactions with the target are selected. Importantly, such interactions are often conserved over the whole target family and hence a fragment identified using this approach can be used to target other members of the family. An example of such approach include selection of metal chelators in a fragment screening campaign against gelatinase (Wang et al., 2002). Commercial fragment libraries dedicated for matrix metalloprotease (MMPs) such as Chelator Fragment Libraries (CFL 1.1), comprising a range of metal chelating moieties (Agrawal et al., 2008) and protease inhibitor-enriched library (Baell and Holloway, 2010), have been deliberately chosen to target a matrix metalloproteinase, LasB (Garner et al., 2012; Kany et al., 2018a). Another example of target-focussed libraries are halogen-enriched fragment libraries (HEFLibs), comprising chemical probes that identify halogen bonds as the main feature of binding mode, and originally constructed to find chemical moieties that stabilise p53 mutants (Heidrich et al., 2019). Structure-based approaches for design/selection of the fragment library may involve in silico screening as described for DNA gyrase ligands (Boehm et al., 2000). The hit rates in fragment screening with such target-focused fragment libraries are much higher than those with generic fragment libraries.

Biophysical, biochemical, structural biology and computational methods are used to screen a library of fragments (∼1,000) against the target. The most popular biophysical method used for screening is differential scanning fluorimetry (DSF), often known as thermal shift assay, a technique that measures the denaturation temperature (melting temperature, Tm) of the protein (Senisterra et al., 2012) and allows the detection of compounds that increase the Tm of a target protein on binding by promoting protein stability (Niesen et al., 2007). Surface plasmon resonance (SPR), a label-free technique, is another biophysical technique that is used for direct screening of fragment libraries (Navratilova and Hopkins, 2010). Fragment-screening by SPR is advantageous owing to its cost-effectiveness, the possibility of high-throughput mode and the requirement of a very small amount of protein (Neumann et al., 2007; Chavanieu and Pugnière, 2016). NMR spectroscopy can also be used to perform initial screening, possibly using cocktails of fragments to accelerate the procedure (Leach, 2006).

Microscale thermophoresis (MST), another well-established biophysical technique to quantify any kind of biomolecular interaction (Jerabek-Willemsen et al., 2014; Mueller et al., 2017), is based on thermophoretic mobility, the directed motions of biomolecules and macromolecular complexes in solution in a temperature gradient; these strongly depend on molecular properties such as size, charge, hydration shell or conformation. MST enables the identification of compounds whose binding to a target changes the thermophoretic mobility of the target (Asmari et al., 2018), including even the weak binders, such as fragments (Linke et al., 2016). It has been demonstrated that MST is amenable for implementation into high-throughput screening cascades (Rainard et al., 2018) and has potential to maximize the efficacy of fragment screening campaigns owing to a high degree of automation in the technique generating quantitative data for affinity ranking in a rapid and precise manner (Rainard et al., 2018). Biolayer interferometry (BLI), especially useful for targeting protein–protein interactions (Wartchow et al., 2011), and nanoelectrospray ionization mass spectrometry (ES-MS) (Maple et al., 2012) are other methods that have recently been used for fragment screening and ranking.

Crystal-based screening of a fragment library has also been developed (Patel et al., 2014), and this is routinely practiced at the XChem facility of Diamond Light Source, United Kingdom as a highly streamlined process, allowing more than 1,000 fragments to be screened in less than a week (Douangamath et al., 2020. https://www.diamond.ac.uk/Instruments/Mx/Fragment-Screening.html). For this, many crystals of the target protein are produced and the fragments are soaked in these crystals using the acoustic droplet ejection technique (Collins et al., 2017). High-throughput X-ray diffraction data collection (36 h of unattended beamtime) is carried out on the resulting crystals at the dedicated beamline, IO4-1, of the Diamond Light Source. The processing of X-ray data is performed automatically while the data are being collected, using the dedicated data processing softwares, such as xia2 (Winter 2010) among others. The structure solution and map analysis is performed by the XChemExplorer package (Krojer et al., 2017) including modules such as DIMPLE (Difference Map PipeLinE) for initial refinement and automated difference map calculation to allow for quick assessment of X-ray data to see if a ligand has bound to the structure; AceDRG (Long et al., 2017) for generation of ligand restraints; REFMAC (Murshudov et al., 2011) for refinement; PanDDA (Pearce et al., 2017) for the streamline density interpretation and hit detection; a number of tools from PHENIX (Adams et al., 2010) for structure validation; and Coot (Emsley et al., 2010) for automated model building. The advantage of crystal-based screening is that it can detect fragments that have very weak binding affinities to the target protein.

To cut the cost of experiments and as a pre-screen, virtual docking of the fragment library on the three-dimensional structure of target protein, known as virtual screening (VS), is often performed in order to obtain initial hits (Bielska et al., 2011; Singh et al., 2018; Yamaotsu and Hirono, 2018; Gimeno et al., 2019). However, it is of note that these hits are hypothetical and warrant experimental validation (Zhu et al., 2013). VS is advantageous in cases where a very large library comprising millions of scaffolds has to be screened, where it can reduce the number of molecules needed to be screened experimentally by other biophysical/biochemical methods. VS campaign can be supplemented by machine and deep learning principles, especially when dealing with large data sets (Melville et al., 2009). Machine learning approaches in ligand-based VS can address complex compound classification problems and help predict new active molecules (Lavecchia, 2015). Deep learning in the field has emerged in recent years and it has been demonstrated to have utility beyond bioactivity prediction to other drug discovery problems such as de novo molecular design, synthesis prediction and biological image analysis (Chen et al., 2018).

Hits from initial screening are confirmed by ligand-based NMR spectroscopy (Hajduk et al., 1997; Mayer and Meyer, 1999; Dalvit et al., 2001). Real-time fragment-binding affinity and kinetics can be determined using SPR (Yadav et al., 2012; Capelli et al., 2020) and thermodynamics of fragment binding is determined using isothermal titration calorimetry (ITC) (Srivastava and Yadav, 2019). The 3D structures of protein-fragment complexes are determined using X-ray crystallography to instruct the process of fragment elaboration. Functional biochemical assays are performed to measure the inhibition activity of the fragments.

Guided by the 3D structure of protein-fragment complexes the fragments are elaborated using growing (Hoffer et al., 2018), merging (Nikiforov et al., 2016; Miyake et al., 2019) or linking (Bancet et al., 2020), whichever suits in order to generate compounds with higher affinity and drug-like properties. In-silico molecular modelling is a remarkably beneficial tool as, together with X-ray and cryo-electron microscopy, it provides a means to develop a suitable linker to attach to the low-affinity molecules (Scoffin and Slater, 2015; de Souza Neto et al., 2020). We do this with the help of our chemistry collaborators. The elaborated compounds are then tested by ITC for their affinities to target and functional assays for their inhibition activity against the target. An iterative cycle of fragment growing/merging/linking, followed by biophysical, biochemical and structural analyses while maintaining high ligand efficiency is established. The final compounds, often achieving nanomolar affinities and having potent in vitro inhibition, are then used for cell-based assays to check their minimal inhibitory concentrations (MICs) (Kowalska-Krochmal and Dudek-Wicher, 2021). We do this with the help of our cell biologist collaborators.

Small “fragments” of drug-like molecules may bind to hotspots on a defined protein target. The initial screening of a library of fragments (∼1,000) often results in several fragments (∼20–50) binding to a target. Hit validation should lead to 5–10 fragments having significant affinity to the target. Fragments can be chemically “grown” or “linked” to deliver compounds with nanomolar binding affinities to the target and altering its activity. Such small-molecule inhibitors developed against essential protein targets can be developed into lead compounds. Further chemical modification of the chemical structure of the lead compound can lead to improvement in potency, selectivity and pharmacokinetic parameters. The pharmacologically active moiety thus obtained may have poor drug likeness which can usually be further modified chemically to result in a more drug-like compound for testing biologically or clinically. Such a compound should be able to compromise the growth and survival of the pathogen inside the host organism.

The presence of an outer membrane in Gram-negative bacteria (GNB) protects the bacterial cell not only from the harsh environment but also antibiotics (Koch, 2017). Lipopolysaccharide (LPS), the major component of the outer leaflet of this membrane (Raetz and Whitfield, 2002), are virulence factors essential in many clinically important GNBs, such as P. aeruginosa, where it plays important roles in the structural integrity of the bacteria and its defence against the host; hence the enzymes of the LPS biosynthesis pathway are attractive drug targets (Cryz et al., 1984; King et al., 2009). Although there are currently no antibiotics targeting LPS biosynthesis, compounds inhibiting this biochemical pathway can lead to the development of new antibiotics with novel mechanism of action (Jackman et al., 2000; Joo, 2015). Lipid A, a glucosamine disaccharide that is connected to multiple fatty acid chains of various lengths, is the minimal component of LPS required for cellular viability in most GNBs (Anderson et al., 1993; Rotella, 1997; Raetz and Whitfield, 2002). Moreover, lipid A is the antigenic determinant of LPS that triggers septic shock. The enzymes of the LPS biosynthesis pathway, including lipid A biosynthesis, are attractive drug targets for therapeutic interventions. An in silico study suggests that, in P. aeruginosa, a total of thirteen enzymes are involved in this LPS biosynthesis, of which seven enzymes matched with the list of candidate essential genes obtained by transposon mutagenesis study (Perumal et al., 2007). These enzymes have no human homologues and hence they can serve as potential drug targets.

LpxA, LpxC and LpxD are the first three enzymes in lipid. A biosynthesis pathway. Of these, LpxC has been extensively targeted by antibiotic discovery leading to the development of many small molecule inhibitors with antibacterial properties (Kalinin and Holl, 2017). In contrast, LpxA and LpxD have remained largely unexplored for the development of small molecule inhibitors and are targeted mostly by antibacterial peptides (Williams et al., 2006; Jenkins and Dotson, 2012; Jenkins et al., 2014; Dangkulwanich et al., 2019). LpxA and LpxD are amenable to dual-targeting inhibitors due to their structural similarity and the advantage of such inhibitors include increased potency and reduced likelihood of resistance against the inhibitors (Silver, 2007). FBDD against LpxA and LpxD to develop dual-targeting inhibitor was performed by Kyle G. Kroeck and co-workers (126). An initial fragment screening by virtual docking of a subset of the ZINC fragment library of John J. Irwin and Brian K. Shoichet (Irwin and Shoichet, 2005), followed by SPR characterization of binding fragments and X-ray crystallography analysis of enzyme-fragment complexes to verify binding and to reveal binding modes for lead optimization have identified novel small-molecule scaffolds that can serve as starting point for future inhibitor discovery (Kroeck et al., 2019). Several of these novel ligands, which have shown dual-binding activity (Table 2), have the potential to result in small molecule inhibitors targeting LpxA and LpxD simultaneously. For instance compound 1 in its complex with LpxA and LpxD adopted similar binding poses in the two acyl-chain binding pockets revealing both the structural similarities and differences that can guide the development of future dual-targeting inhibitors. Interestingly, these structures also highlight additional binding hot spots shared by the two enzymes that can be exploited for further lead optimization.

The bacterial enzyme UDP-3-O-acyl-N-acetylglucosamine deacetylase (LpxC) is an attractive target for the development of novel therapeutic agents (Erwin, 2016; Chen et al., 2019). It is essential to most Gram-negative bacteria including P. aeruginosa (Raetz and Whitfield, 2002, 2002) and catalyzes the removal of an N-acetyl group from UDP-3-O-acyl-N-acetylglucosamine (which constitutes the core of Lipid A), the first committed step in biosynthesis of the LPS, which is an essential component of the bacterial cell wall. Several series of inhibitors have been developed against P. aeruginosa LpxC (PaLpxC) employing non-FBDD approach (Erwin, 2016; Kalinin and Holl, 2017; Chen et al., 2019) exploiting the following features of molecules at the binding site: 1) a zinc-chelating motif, 2) a polar group occupying the UDP binding pocket, 3) a linker that is mainly hydrophobic but sometimes makes at least one hydrogen bond interaction, 4) an extended hydrophobic moiety that binds in the tunnel, and 5) often addition of groups out into the solvent at the end of the tunnel to modulate compound properties. The most advanced compounds carry a hydroxamate moiety that coordinates the zinc ion at the core of the enzyme. One of these compounds, ACHN-975 (Cohen et al., 2019; Lpxc et al., 2019), a hydroxamate-based histone deacetylase inhibitor (HDACI) entered clinical trial and was approved by FDA for its use in oncology applications but was discontinued beyond oncology (Lpxc et al., 2019) due to its off-target side effects associated with hydroxamate group (Shen and Kozikowski, 2016).

Therefore, an FBDD campaign, to identify LpxC inhibitors with a non-hydroxamate metal-coordinating group, was started (Yamada et al., 2020). 1,152 compounds from the Vernalis Research fragment library (Baurin et al., 2004; Chen and Hubbard, 2009) were screened against PaLpxC by ligand-observed NMR (STD, water-LOGSY, and CPMG), using cocktails of 6 fragments (Hubbard et al., 2007; Hubbard and Murray, 2011) followed by singleton NMR competition assays to look for the competitive binding of the fragments with respect to binding of the tool compounds 6 and 7, (hydroxamate group containing inhibitors reported previously) to LpxC. The 28 fragments showing competitive binding were further assessed in the Fluorescence Polarization (FP) binding assay followed by Fluorescamine-Based Functional Activity Assay. Some of which such as 8 showed clear inhibition with an IC₅₀ of 41.9 μM. Two zinc-chelating fragments, a glycine fragment 8 and an imidazole fragment 9, both of which were competitive with 6, were selected for further investigation. Both of these were shown to stabilize LpxC in a TSA experiment further confirming the binding. The crystal structure of PaLpxC-fragment 8 complex guided the initial medicinal chemistry and iterative cycle of protein-ligand co-crystallization, confirming binding poses, structure-based modeling predicting the binding poses and guiding the possibility of enhancing potency and chemical modification. This led to the development of a compound 10 with many fold increase in the activity (functional IC₅₀ 0.00603 μM) as compared to initial fragment 8 (functional IC₅₀ 41.9 μM). However, this compound showed only weak antibacterial activity even in the presence of phenylalanine-arginine β-naphthylamide (PAβN) (an efflux pump inhibitor) (MIC of 8 μg/ml against P. aeruginosa and 64 μg/ml against P. aeruginosa ATCC27853), so this series was not pursued any further. Similar medicinal chemistry on imidazole fragment 9 resulted in an advanced lead compound 2-(1 Hydroxyethyl) imidazole 11 which exhibited low nanomolar inhibition of PaLpxC (functional IC₅₀ of 20 nM) and a minimal inhibitory concentration (MIC) of 4 μg/ml against P. aeruginosa (Figure 2). Further optimization and in vivo efficacy measurement are under consideration for this compound and will be described in a future publication. The lead compounds of both the series exhibited significant selectivity towards zinc metalloenzyme, LpxC. It was demonstrated that maintaining the zinc-chelation motif as it was in the fragments, engineering aliphatic linkers of appropriate length extending the fragments in the hydrophobic tunnel of LpxC and appending various hydrophobic groups at the para position of the benzyl successfully increased the potencies of compounds with retained ligand efficiencies.

The fatty acid synthesis type II (FAS II) system of bacteria has been identified as an attractive target for therapeutic interventions and several antibiotics targeting this pathway are known, such as triclosan, isoniazid and thiolactomycin (Campbell and Cronan, 2001; Zhang et al., 2004; Heath and Rock, 2006; Parsons and Rock, 2011; Pan et al., 2012). FabG is a NADPH-dependent 3-oxoacyl-acyl carrier protein (ACP) reductase that plays a key role in the FAS II system of pathogenic microorganisms and has been identified as an attractive drug target. It catalyses the reduction of 3-oxoacyl-ACP to 3-D-hydroxyacyl-ACP intermediates during the elongation cycle of fatty acid biosynthesis (Rock and Jackowski, 2002; Goodman and McFadden, 2008; Chan and Vogel, 2010). FabG qualifies to be a promising drug target due to its essentiality, high conservation across bacteria and presence of only a single isoform in most of the bacterial species (Zhang et al., 2004). Inhibitors of P. aeruginosa FabG (PaFabG) identified include largely natural product extracts that pose significant drug development challenges and hence none of them have reached the clinic (Zhang and Rock, 2004; Tasdemir et al., 2006; Wickramasinghe et al., 2006; Sohn et al., 2008; Zhang et al., 2008). As part of the AEROPATH project, Cyprian D. Cukier and coworkers have established the essentiality of fabG gene in P. aeruginosa using gene knockout procedure and following a FBDD approach they then developed a series of novel PaFabG inhibitors with IC₅₀ values in nanomolar to low micromolar range and ligand efficiencies in the range of 0.37–0.53 (Cukier et al., 2013). Although the compounds show no phenotypic response in the Gram-negative P. aeruginosa either due to poor penetration of the compounds through the Gram-negative cell wall or to rapid efflux of the compounds, the diverse chemotypes of these inhibitors presents a number of options for optimization to increase intracellular concentrations. Structural investigation of 16 PaFabG-inhibitor complexes by X-ray crystallography in this study reveals that inhibitors bind at a novel allosteric site (cryptic binding site) located at the dimer-dimer interfaces of PaFabG (Figure 3) and this binding induces the conformational changes that propagate to the active site and results in the disturbance of the catalytic triads (residues S141, Y154, and K158) with loss in the binding affinity of cosubstrate NADH, thus inhibiting the enzyme. Kinetic analysis of inhibition suggested a noncompetitive mode of inhibition with respect to NADH.

LasB (or pseudolysin, Pseudomonas elastase B) is the most abundant extracellular collagenase protease secreted by P. aeruginosa with a hydrolysis activity against a broad spectrum of substrate proteins from the host, causing damage to host tissues, disruption of the host immune response and promoting inflammation associated with P. aeruginosa virulence and disease pathology (Wretlind and Pavlovskis, 1983; Saint-Criq et al., 2018; Galdino et al., 2019). Since LasB is extracellular the direct inhibitors are not required to cross the difficult-to-penetrate P. aeruginosa cell membrane and the fact that the target belongs to a class of validated drug targets (metalloprotease), against which there are already clinacally useful drugs, makes LasB an attractive antivirulence target for therapeutic intervention (Galdino et al., 2019). Synthesis of small-molecule inhibitors of P. aeruginosa LasB (PaLasB) had started more than 40 years ago and since then eighteen inhibitors have been developed (Everett and Davies, 2021). The FBDD approach to develop inhibitors against PaLasB started in 2012 by screening a library of 96 metal chelating fragments (CFL-1.1) (a library designed specifcally with fragments which can coordinate with the metal ion in the active site of metalloproteins) against PaLasB and the initial hits were followed up by medicinal chemistry optimization program (Agrawal et al., 2008; Garner et al., 2012). This led to the development of a thiopyridone 13 (Garner et al., 2012) with an IC₅₀ of 2.73 μM for PaLasB but was promiscuous with regard to other mettaloproteases, and the tropolone 14 (Fullagar et al., 2013) exhibiting similar potency (IC₅₀ of 1 μM for LasB) but had better selectivity over human metalloproteases, such as matrix metalloproteases (MMPs) and carbonic anhydrase II (Figure 4A). They were the first potent non-peptidic small molecule antagonists discovered (Cathcart et al., 2011) and the first targetted compounds exhibiting antiswarming activity (Garner et al., 2012).

With the aim of expanding the chemical space of PaLasB inhibitors, a functional screening based on FRET-based in vitro assay (Nishino and Powers, 1980) was performed against PaLasB using 330 fragments (Maybridge Fragment Library) and a protease inhibitor-enriched library (Baell and Holloway, 2010) comprising of 1,192 low molecular weight compounds (Kany et al., 2018a). This study led to the development of a thiobenzamide (N-(3,4-dichlorophenyl)-2-sulfanylacetamide) 15, LasB inhibitor with thiol warhead, with an IC₅₀ of 6.6 μM (Figure 4A). This was selective with regard to human MMPs. The X-ray structure of PaLasB-15 complex revealed that binding of 15 to LasB does not necessarily lead to closure of the binding site (Kany et al., 2018a), unlike that described for thermolysin-like proteases (Adekoya et al., 2015), enzymes with high structural similarity to LasB. Moreover, two molecules of 15 were observed in the structure (Figure 4B) and the binding of the second molecule was supposed to be supported by the binding of the first molecule. However, the Hill coefficient of 1 in the in vitro assay for 15 suggested that only one binding event was necessary for full inhibition of PaLasB. Neverthless, this structure with two binding sites for 15 paved the way for the development of novel LasB inhibitors targeting the open conformation of the enzyme. Attempts to merge these two molecules into one N-benzylamide derivative did not provide the desired activity. Growing the non-zinc-chelating molecule of 15 in the PaLasB-15 complex led to the development of N-arylsuccinimide 16 with an IC₅₀ of 4 μM, which showed, 1) significant selectivity for bacterial LasB over human MMPs and three other off-targets and 2) no signs of cytotoxicity in human cell lines (Konstantinović et al., 2020). In a another subsequent study, the zinc-chelating molecule of 15 in the PaLasB-15 complex was modified by changing the zinc-chelating moiety leading to a hydroxamate 17 (Kany et al., 2018b) with an IC₅₀ of 17 μM with moderate cytotoxic effects towards mamalian cell lines. Since only one molecule of 17 was observed in the crystal structure of PaLasB-17 complex (Figure 4C), unlike that of 15 in PaLasB-15 complex, and 17 was less susceptible to oxidation in air, the authors preferred 17 over 15 for further exploration. Moreover, 17 could undergo the characteristic hinge-bending motion resulting in clusure of LasB binding site unlike open conformation as observed in PaLasB-15 complex. Overall, given their modest activities, the authors suggest that the molecules such as 15, 16 and 17 should be considered as starting points for further chemistry rather than as potential drug development candidates which could pave the way for the rational development of selective protease inhibitors as potential new antibiotics.

Exploiting the alternative binding modes of 15 to guide efficient fragment growing resulted in a series of compounds with better activities (Kaya et al., 2021). Compound 18 thus developed had an IC₅₀ of 1.2 μM (Figure 4A) and maintains the selectivity as 15. PaLasB-18 complex structure was determined by x-ray crystallography. Here again, the binding of 18 to PaLasB leads to closure of the binding site. This structure provided a deeper understanding of the possible interactions in the surrounding unoccupied space and complemented with docking analysis it paved the way for further optimization. A focused, substrate-inspired structure-based optimization of 18 (substitution of methyl group at the para position on the aromatic core of N-arylacetamide) resulted in compound 19 (Figure 4A) with fourteen-fold boost in activity (IC₅₀ 0.48 μM) compared to 15 (173).

In vivo efficacy of these compounds was investigated using an insect model, Galleria mellonella and PA14, a virulent stain of P. aeruginosa. Since Galleria mellonella and mice show similar virulence patterns when infected with mutant PA14 starins (PA14, virulent strain of P. aeruginosa) this insect is likely to be a good model system to perform in vivo studies (Jander et al., 2000). Administering the anti-LasB compound 15 thus generated by FBDD campaign has shown significant increase in the survival of Galleria mellonella larvae infected with PA14. For instance, injection of 2.5 nM of 15 in the P14 infected larvae increased the survival of the larve from 43 to 73% after 65 h (Kany et al., 2018a). Compound 19 with better IC₅₀ showed enhanced in vivo efficacy compared to 15, thereby accelerating the translational path (Kaya et al., 2021). Antivirulent agents targeting quorum sensing of P. aeruginosa have previously been shown to enhance the survival of Galleria mellonella larvae infected with PA14 (Lu et al., 2014; Thomann et al., 2016). These compounds attain this efficacy by inhibiting the swarming and biofilm formationa rather than being bacteriocidal. For instance, presence of compound 17 reduces the formation of biofilm and release of extracellular DNA by P. aeruginosa. The advantage of such inhibitors is that they are capable of disrupting several important bacterial resistance mechanisms and hence open novel avenues to combat multidrug resistant strains of P. aeruginosa.

PqsD is a key enzyme in the biosynthesis of signal molecules 2-heptyl-4-hydroxyquinoline (HHQ) and Pseudomonas quinolone signal (PQS) that are involved in the regulation of virulence factor (pyocyanine, elastase B, lectin A, rhamnolipids, and hydrogen cyanide) production and biofilm formation in P. aeruginosa (Van Delden and Iglewski, 1998; Diggle et al., 2003; Déziel et al., 2005; Yang et al., 2009). It has been shown that a mutant P. aeruginosa having a transposon insertion in the pqsA gene (deficient in HHQ and PQS production) forms less biofilm than the wild type (Gallagher et al., 2002; Müsken et al., 2010). Using a ligand-based approach the first class of PqsD inhibitors were identified which repressed HHQ and PQS production and biofilm formation in P. aeruginosa, validating PqsD as an attractive anti-biofilm target for the development of novel anti-infectives (Storz et al., 2012). Elisabeth Weidel and co-workers performed an SPR-based fragment screening against P. aeruginosa PqsD (PaPqsD) using a library of 500 fragments (Maybridge) with high structural diversity covering large chemical space in order to identify new scaffold for drug discovery (Koch, 2017). This screen resulted in identification of three fragments exhibiting a moderate inhibition of PaPqsD (Table 3). Two compounds showed strong inhibition of PaPqsD and hence these may be the starting point for future investigations.

Oxidative protein folding is essential for assembly and function of many secreted and membrane proteins, and DSB (disulfide bond) proteins that catalyze the disulfide bond formation are essential for virulence of many Gram-negative bacteria and hence these proteins are targets for novel antibacterial drugs (Heras et al., 2009). DsbA, an enzyme of disulfide oxidoreductase family, and its membrane-bound partner DsbB together catalyze the oxidative folding of disulfide bond containing proteins, many of which are virulence factors including secreted toxins and cell surface components, such as adhesins and pili. DsbA1 in P. aeruginosa (PaDsbA1) plays a pivotal role in the oxidative folding of virulence factors qualifying it as an attractive target for the development of new anti-virulence antimicrobials (Braun et al., 2001; Urban et al., 2001; Ha et al., 2003). In a FBDD approach, PaDsbA1 was screened against a library of 1,137 fragments (Bradley et al., 2013) using ligand-detected STD NMR, which identified small molecules that bind selectively to PaDsbA1 over E. coli DsbA (EcDsbA) suggesting the feasibility of species-specific development of narrow-spectrum inhibitor of PaDsbA1 (Mohanty et al., 2017). Structural characterization of the complex of PaDsbA1 with fragment 23 (highest affinity fragment with K_D of 0.9, ligand efficiency 0.27 and lipophilic ligand efficiency of 0.26) using both X-ray crystallography and HADDOCK model revealed that the fragment is positioned at an interface between the thioredoxin (TRX) and helical domains of PaDsbA1 on the non-catalytic face (Figure 5) of the enzyme. Unfortunately, fragment 1 had no inhibitory effect on the enzymatic activity of PaDsbA1 as shown in an in vitro model-peptide folding assay. Nevertheless, these findings represent a starting point for the development of high affinity allosteric inhibitors of PaDsbA1.