**1. Introduction**

Although the drug discovery process has historically relied on high-throughput screening (HTS) to identify hits to be developed into drug candidates, either on a given target (in most cases, an essential enzyme) or through whole-cell screenings, it still remains an experimentally laborious, expensive, and time-consuming process. Unlike traditional drug discovery, however, drug design is not necessarily based on the screening of large libraries, which is intrinsically a trial and error process, but builds on the available knowledge for a given biological target. A particular, well-known case of drug design is called structure-based drug design (SBDD), which uses the three-dimensional structural knowledge of the target to find or optimize molecules that can bind to the target with high affinity and selectivity. The potential of using structural information for discovering candidate drugs was apparent from the early days of structural biology [1], but it took several years to achieve the first successful examples, i.e., human immunodeficiency virus (HIV) protease inhibitors [2] and carbonic anhydrase inhibitors for the treatment of glaucoma [3].

A three-dimensional model of the target is, therefore, a prerequisite for SBDD. The structure can be obtained experimentally in different ways, in most cases by X-ray crystallography, although nuclear magnetic resonance (NMR) and, more recently, cryo-electron microscopy (cryo-EM) have also attracted attention [4], especially following the spectacular increase in the resolution capabilities of single-particle cryo-EM, which is commonly referred to as the resolution revolution [5,6]. For targets whose experimental structure is elusive, in silico structure prediction is also routinely performed and can be achieved, whenever suitable models are available, through homology modeling from a closely related homologous protein [7].

The identification of potential drug binding sites can be obvious when ligand-bound target structures are available. However, the capability to detect binding sites for substrates or modulators becomes highly relevant not only for targets for which no ligand-bound three-dimensional structure is available, but also to detect allosteric sites or protein–protein interaction surfaces that might be specifically targeted [8], and for these, dedicated libraries are now available [9]. Considering that binding site identification is not always straightforward, tools have been developed to infer "druggable" pockets from the identification of concave regions—that could accommodate drug-size molecules—by screening for appropriate binding properties, such as volume, hydrophobicity, hydrogen bonding, energy potential, solvent accessibility, and desolvation energy [10].

The search for potential ligands on a given target can then follow two different approaches, commonly referred to as ligand-based drug design (LBDD) or target-based drug design (TBDD). LBDD does not require, a priori, direct structural knowledge of the target, but relies on the identification of hits, usually congeneric compounds, with an established biological effect; hit-to-lead optimization can then proceed through the definition of appropriate chemical descriptors of the series, quantitative structure-activity relationship analysis (QSAR), and pharmacophore modeling [11], all followed by experimental validation. QSAR-based virtual screening approaches, alone or in combination with HTS screenings, can also be used to enlarge the panel of bioactive compounds against the target or pathway of interest, increasing the overall hit rate [12]. TBDD, in contrast, uses the physical-chemical constrains from the target three-dimensional structure, and possibly a well-defined binding pocket, to perform virtual screening of libraries, either of natural or synthetic compounds, usually applying appropriate filters like compliance to Lipinski's rules or QSAR models, or to design molecules de novo in a step-wise manner. A particular case of de novo molecule design is fragment-based drug design (FBDD), in which low-affinity target binding molecules are identified from appropriate libraries, using a variety of biophysical approaches, and then merged or linked together using the available three-dimensional information to achieve larger binders with improved properties [13,14].

It is common knowledge in the field that although SBDD approaches have proven to be successful for non-transmissible diseases and viral diseases, they have proven to be much less effective in the antibiotic discovery field [15,16]. Unfortunately, tuberculosis represents no exception to this trend [17], despite the considerable progress achieved during the last twenty years in the understanding of the pathogen molecular physiology, starting from the seminal publication of the *M. tuberculosis* genome sequence in 1998 [18], the extensive structural genomics campaigns during the following years [19–21], the advances in understanding host-pathogen interactions and the development of the disease [22,23], and notwithstanding the development of genetic and biochemical tools to allow the in vivo and in vitro validation of targets [24,25].

The purpose of this review was not to provide an exhaustive overview of the capabilities now offered by in silico approaches for antibiotic development against *M. tuberculosis*, already reviewed elsewhere [26,27], nor to make a survey of the current state of structural knowledge of the pathogen proteome and the experimental structures relevant for drug discovery, for which we point the reader to very recent, extensive work [28]. Rather, we focused here on three emblematic case studies of *M. tuberculosis* targets that attracted most efforts for anti-tuberculosis compound development by HTS campaigns, computer-aided, and structure-driven compound identification: the serine/threonine (Ser/Thr) kinases—protein kinase (Pkn)B and PknG—two amongst the most known, supposedly promising new targets offered by the post-genomic era, and the DNA gyrase, the 'old' but the well-proven target of fluoroquinolones.

#### **2. Protein Ser**/**Thr Kinases as Drug Targets**

Protein phosphorylation is a well-known, widespread mechanism for signal transduction and regulation of several biological functions. Protein kinases have long been known as major drug targets [29], especially for the treatment of cancer, where the deregulation of signaling mechanisms is a hallmark of the disease [30]. Protein kinases are not only pharmaceutical targets in cancer

chemotherapy but also for the treatment of parasitic infections, ranging from the ones caused by Trypanosomatids or *Leishmania* [31] to malaria [32]. Since the sequence of *M. tuberculosis* H37Rv genome was first reported in 1998 [18], the pathogen has been known to possess eleven genes coding for Hanks-type Ser/Thr kinases [33], named from *pknA* to *pknL* (but no *pknC*), providing one of the first challenges to the paradigm of prokaryotic cell signaling as being entirely driven by two-component systems. One gene coding for a transmembrane Ser/Thr phosphatase, *pstP*, was also identified as lying on the same cluster as *pknA* and *pknB*, forming a putative operon [34], in addition to two genes (*ptpA* and *ptpB*) coding for Tyr phosphatases [35]. Nine out of eleven Ser/Thr kinases (all but PknG and PknK) were predicted to be integral membrane proteins, all sharing the same topology with an N-terminal catalytic domain in which the archetypal Hanks motifs could be identified [33], a single transmembrane segment and a very variable extracellular, C-terminal domain, hypothesized to be acting as a signal sensing domain. The role of such C-terminal domains is, in most cases, still puzzling, although, in some kinases like PknB, it has been suggested to be involved in the kinase activation process, possibly by controlling the kinase oligomerization state as a function of the external signal [36]. Work from Tom Alber's group in Berkeley indeed demonstrated the role of the 'back-to-back' dimerization to promote an active kinase conformation, not only for PknB [37,38] but also for PknD [39], notwithstanding the observation that mutations in the dimerization interface do not abolish kinase activity, as reported by separate groups [37,40]. PknB was also the first *M. tuberculosis* kinase for which the crystal structure of the catalytic domain was described in 2003 [41,42]. Indeed, this first crystal structure confirmed the overall conservation of the bi-lobed protein kinase fold and the initial assumptions about the presence of the known structural features of eukaryotic protein kinases, as predicted by the detection of the Hanks motifs [42], thus underlining the common origin of eukaryotic and prokaryotic kinases. Following these first milestones and the genetic proof of the in vitro essentiality of *pknB* coming both from transposon mutagenesis [43] and targeted studies [44,45], PknB then attracted the attention of tuberculosis (TB) research community as an ideal target for structure-based drug design. The community's interest rise when considering the increasing evidence, over the years, of its role in the control of peptidoglycan synthesis and cell division (recently reviewed in [46]). The first description of the potential use of PknB inhibitors as anti-mycobacterial agents reported the compound H-7 (1-(5-isoquinolinesulfonyl)-2-methylpiperazine; Table 1), well before any bacterial protein kinase structure was available [47]. Once the PknB catalytic domain structure was solved, virtual screening carried out on a library of about 40,000 compounds for hits into the PknB adenosine triphosphate-binding (ATP-binding) pocket was performed. The screening led to identifying mitoxantrone, a chemotherapeutic agent known for its DNA intercalating properties, as a sub-micromolar PknB inhibitor [48], similar to staurosporine, K-252-a, and K-252-b (Table 1), who were identified as hits after testing a few commercially available eukaryotic kinase inhibitors [44]. The crystal structure of the PknB-mitoxantrone complex was the first showing the kinase catalytic domain in complex with a non-ATP analog, kinase inhibitor. Two crucial, hydrogen bonding interactions were evidenced between the mitoxantrone hydroxyl groups and main chain atoms from the PknB hinge region that connects the two kinase lobes, i.e., the carbonyl oxygen of Glu93 and the amino group of Val95 [48], opening the way to compound optimization (Figure 1A,B). Several other PknB inhibitors were proposed in the course of the following years, starting from hits developed either from known kinase inhibitors like staurosporine analogs [37], 2-aminopurine and its derivatives, including organometallic compounds [49]), or from hits obtained from HTS on public or proprietary libraries using GarA as the substrate [50–53], or phytocompounds [54] (Table 1). Most of the published work deals with compounds inhibiting the kinase in the micro and sub-micromolar range but with limited activity on mycobacteria. The exception to this trend is IMB-YH-8 (Table 1), a compound that, despite showing an IC50 (half maximal inhibitory concentration) on PknB in the 20 μM range, shows good selectivity for mycobacterial PknB and PknA and a MIC (minimal inhibitory concentration) in the sub-micromolar range [53]. Compounds highly active on both PknB and PknA have also been reported recently by others in the form of substituted quinazolines, and, for a compound derived

from this series (a pyrimidine analog), the crystal structures of the respective complexes with both kinases have been described [55] (Figure 1, Table 1). Both structures underline the common binding mode in the ATP pocket and the crucial interaction with the hinge region (Figure 1B). Despite the inhibition constant for both kinases falls in the nanomolar range, these dual-targeting inhibitors display promising but yet limited antibacterial effect with MIC on *M. tuberculosis* in the lower micromolar range [55]. A different PknB inhibitor for which a crystal structure in complex with the target is available is GSK690693 (Figure 1C, Table 1), also identified through a virtual screening approach on known kinase inhibitors [56]. This compound, member of the imidazopyridine aminofurazans class, also displays conserved features in its binding mode to the PknB hinge region (Figure 1B), and sub-micromolar affinity to the kinase, but no significant antimycobacterial activity on *Mycobacterium smegmatis* or *Mycobacterium bovis* BCG (bacillus Calmette-Guerin). However, the MIC is significantly lowered if the compound is associated with a sub-MIC50 concentration of meropenem, suggesting a synergistic action between PknB inhibitors and β-lactams [56].

**Figure 1.** Experimental structures of protein kinase B (PknB)-inhibitor complexes. (**A**) Electrostatic surface from the X-ray structure of PknB in complex with AMP-PCP (β,γ-methyleneadenosine 5- -triphosphate) in red (pdb: 1O6Y), superimposed with the coordinates of four other PknB complexes with, respectively, mitoxantrone in green (pdb: 2FUM), KT5720 in orange (pdb: 3F69), GSK690693 in cyan (pdb: 5U94), and 'compound 38' in pink (pdb: 6B2P). This compound has been reported as active on both PknA and PknB [55]. (**B**) Side and top view of the superimposed ligands with the hinge region from the pdb entry 6B2P. Kinase residues are indicated. (**C**) Ball-and-stick representation of the ligands kept in the same relative binding orientation as in (**B**) and colored by element.



**Table 1.** *Cont.*


**Table 1.** *Cont.*

ND: not determined. <sup>a</sup> measured on a PknB mutant (M145L, M155V) [37].

On the other hand, another *M. tuberculosis* kinase that attracted as much attention for its potential druggability is PknG, a soluble Ser/Thr kinase that was described, by Jean Pieters and coworkers, as a virulence factor secreted into the human macrophage, where it would inhibit the phagosome-lysosome fusion [57]. Most notably, in the same work, the authors showed that chemical inhibition of PknG by the compound AX20017 (a tetrahydrobenzothiophene identified by HTS and found to inhibit PknG with IC50 in the sub-micromolar range; Table 2) led to the accumulation of *M. tuberculosis* inside lysosomes [57]. Noteworthy, the 2.4 Å resolution crystal structure of an N-terminal truncated form of PknG in complex with AX20017, published later [58], shows a great similarity between the AX20017 binding to PknG and the binding of ATP and mitoxantrone to PknB. Not only the compound occupies the adenine binding pocket but also makes similar interactions with the main chain atoms of residues Glu233 and Val235 (Figure 1). It was only a few years later that further structural work allowed to elucidate the binding mode of ATP to PknG and suggested a regulatory role of the rubredoxin-like domain by a reversible occlusion of the active site entrance [59]. These initial findings paved the way to further structure-based lead optimization and undoubtedly generated excitement for the potential chemical targeting of this kinase, which would allow preventing the arrest of phagosome maturation, directing *M. tuberculosis* to lysosomes. Attempts to develop more potent compounds starting from AX20017 yielded sub-micromolar inhibitors, some of which had no activity in macrophage assays [50], and no other structure of a PknG-inhibitor complex has so far been reported. It is, however, worth noting that, more than ten years later, the role of PknG in arresting the phagosome maturation is still elusive. Although progress has been made in identifying the SecA2 system as responsible for the export of PknG outside *M. tuberculosis* [60,61], and interference by PknG on the host Rab7l1 signaling pathway has been reported [62], it is largely accepted that several mechanisms, and not a single virulence factor, contribute to the *M. tuberculosis* capability to escape the phagocytic route and survive into macrophages [22,63]. Moreover, an increasing amount of evidence has since validated PknG as a key signaling element in the control of central metabolism, as first shown in *Corynebacterium glutamicum* [64,65], then in mycobacteria [66,67], where genetic and metabolomic evidence point to a role of PknG in regulating the 2-oxoglutarate node according to nutrient availability [68]. Alternative roles of PknG in biofilm formation, adaptation to oxidative stress [69], and hypoxia [70] have also been proposed. Therefore, the observed effects of the lack of PknG on the phagosome-lysosome fusion, and the decreased viability of a *pknG*-deprived *M. tuberculosis* mutant, both in vitro and in the mouse model [71], may not necessarily indicate a direct interference of the kinase on the host signaling pathways but could be ascribed to metabolic alterations caused by the depletion of PknG [68,72]. During the last years, further search for PknG inhibitors has allowed identifying sclerotiorin, an azaphilone derivative isolated from *Penicillium* sp. ZJ27 (Table 2), a known inhibitor of other unrelated enzymes like lipoxygenase, that, despite showing an IC50 on PknG around 76 μM and no inhibition on *M. tuberculosis* growth in vitro, displays capability to partially reduce the growth of *Mycobacterium bovis* BCG inside macrophages, and to enhance the effect of rifampicin [73]. On the other hand, the recently described PknG inhibitor NU-6027 (2,6-diamino-4-cyclohexylmethoxy-5-nitrosopyrimidine; Table 2) was identified from phenotypic screening as having an MIC99 value of 1.56 μM on *M. bovis* BCG and shown to partially inhibit PknG and PknD autophosphorylation in vitro [74]. Another series of four compounds, all inhibiting PknG in the micromolar range, was also recently reported after the screening of an 80-compound, commercially available kinase inhibitor library (Table 2); three of them were found to promote the transfer of mycobacteria to lysosomes, and two to inhibit *M. bovis* BCG growth in macrophages [75]. None of the reported compounds, however, issued from a structure-based screening approach.


**Table 2.** *M. tuberculosis* PknG inhibitors.

μ

μ

μ

μ

ND: not determined.
