Dual Inhibition of Mycobacterium tuberculosis and the Host TGFBR1 by an Anilinoquinazoline

Abstract

Tuberculosis (TB) control is complicated by the emergence of drug resistance. Promising strategies to prevent drug resistance are the targeting of nonreplicating, drug-tolerant bacterial populations and targeting of the host, but inhibitors and targets for either are still rare. In a cell-based screen of ATP-competitive inhibitors, we identified compounds with in vitro activity against replicating Mycobacterium tuberculosis (Mtb), and an anilinoquinazoline (AQA) that also had potent activity against nonreplicating and persistent Mtb. AQA was originally developed to inhibit human transforming growth factor receptor 1 (TGFBR1), a host kinase that is predicted to have host-adverse effects during Mtb infection. The structure–activity relationship of this dually active compound identified the pyridyl-6-methyl group as being required for potent Mtb inhibition but a liability for P450 metabolism. Pyrrolopyrimidine (43) emerged as the optimal compound that balanced micromolar inhibition of nonreplicating Mtb and TGFBR1 while also demonstrating improved metabolic stability and pharmacokinetic profiles.

Graphical Abstract

INTRODUCTION

Tuberculosis (TB) is a leading cause of death from infectious disease¹ and its treatment is frustrated by rampant drug resistance. More than 5% of TB cases are due to multidrug-resistant Mycobacterium tuberculosis (Mtb) strains, which are exceedingly difficult to treat.¹ Despite the recent introduction of three new anti-TB drugs,² Mtb has developed resistance to all drugs used in the clinic and antibiotic resistance to new antibiotics appears to be all but inevitable without targeting drug resistance mechanisms themselves. In addition to drug resistance, drug tolerance impedes TB treatment and is a major contributor to the failure of current drugs.³ Drug-tolerant bacteria are isogenic with drug susceptible bacteria but become drug tolerant through reduced metabolic activity associated with slowed or no replication.⁴ Development of TB drugs that are active against nonreplicating bacteria has led to dramatic reductions in treatment times in the past. For example, rifampicin reduced the standard treatment time from 18 to 9 months, and pyrazinamide further reduced the treatment time from 9 to 6 months. An extreme type of drug tolerance is found in persisters and is increasingly recognized as a driver of drug resistance and treatment failure.⁵ Persisters are a small subpopulation in a bacterial population that are characterized by slow growth and metabolism.⁶ They are drug tolerant even in the absence of genetic mutation and contribute to the long treatment times that are required for the cure; 6 months for drug-susceptible and 2 years and longer for drug-resistant TB.⁵ Only a few drug targets and inhibitors of nonreplicating or persistent Mtb have been identified to date, and identification of new molecular targets is a high priority for TB drug development.⁷ The persistent state of Escherichia coli and Mtb is accompanied and may be caused by low ATP levels, which could specifically sensitize ATPases to inhibition by ATP binding site-directed compounds.^8,9

Besides the direct targeting of tolerant and persistent bacteria, another strategy to combat drug resistance is host-directed therapy (HDT) since the bacterium cannot genetically develop resistance by altering the host drug target.¹⁰ HDT can have the dual benefit of contributing to direct clearance of infection and restricting destructive inflammation, a major cause of morbidity in TB.¹¹ However, despite intense interest, the number of HDT targets remains small. Transforming growth factor (TGF) β is a pleiotropic cytokine with immunosuppressive and anti-inflammatory properties. It inhibits lymphocyte function, including T cell proliferation and cytokine production, including IFNγ.¹² At sites of active Mtb infection in both murine and human TB, TGFβ is produced in excess in response to cell wall lipoarabinomannan, supporting the notion that induction of TGFβ during TB infection is beneficial to Mtb.^13,14 In clinical samples from patients with active TB, neutralization of TGFβ improves T-cell responses in vitro, but the extent of its role in vivo during TB infection is unclear.^15,16 A recent murine study linked TGFβ signaling to spatially localized immunosuppression in Mtb-infected granulomas. Blockade of TGFβ signaling restored CD4 T-cell function and reduced bacterial burden, suggesting TGFβ signaling as a new HDT target.¹⁷ TGFβ signals through a receptor serine/threonine kinase heterocomplex, the TGFBR1 and 2, both of which are required for functional signaling.¹⁸ The TGFBR complex has been extensively targeted in other therapeutic contexts, notably in oncology and fibrosis.^19,20 Initial clinical trials have shown safety and efficacy of inhibiting TGFβ signaling, and several drugs have reached phase III clinical trials,²¹ suggesting that it might be a tractable HDT target in TB disease as well.

To identify novel compounds with reduced propensity to elicit drug resistance, we sought to identify compounds active against nonreplicating and persistent Mtb by repurposing a collection of ATP-competitive small molecule inhibitors. We identified anilinoquinazoline (AQA) as an ATP analogue with activity against replicating and potent activity against nonreplicating and persistent, drug-tolerant Mtb. AQA demonstrated a minimal inhibitory concentration (MIC) against nonreplicating Mtb comparable to the first-line drug rifampicin. Since AQA is also a potent inhibitor of the kinase TGFBR1, albeit with low bioavailability,²² we sought to develop an improved pharmacological tool to explore the potential of dual targeting of Mtb and a potentially immunosuppressive host target within a single agent. We determined the structure–activity relationship for dual-targeting across each ring of the AQA core and identified an analogue that combined dual targeting activity with improved pharmacokinetic properties.

RESULTS

Mtb Whole-Cell Screen of ATP-Competitive Inhibitors.

To test the activity of ATP-competitive inhibitors against Mtb and to identify compounds also active against nonreplicating and persistent Mtb, we initially screened a library of ATP-competitive inhibitors against replicating Mtb in a whole-cell screen, followed by a secondary screen of hits against nonreplicating Mtb and lastly testing against drug-tolerant persisters. We initially hoped to identify inhibitors that were active in all three conditions, as those would arguably be the most desirable. We chose the open resource published kinase inhibitor set (PKIS) 1 and 2^23,24 for this screen: the PKIS 1 and 2 are diverse collections of drug-like ATP-competitive inhibitors that have been extensively curated for their activity against the human kinome; the PKIS chemotypes inhibit 150 (PKIS 1) and 250 (PKIS 2) human kinases, respectively. All PKIS compounds have undergone some level of medicinal chemistry optimization for favorable pharmacokinetic properties and are considered “drug-like”.

To test for anti-Mtb activity in the PKIS 1 and 2, we initially carried out a primary whole-cell screen against an Mtb H37Rv reporter strain expressing GFP from a strong 16S rRNA promoter²⁵ (Mtb-GFP), which allows for facile detection of growth inhibition by measuring fluorescence. Bacteria were seeded in a 96-well plate at a density of OD₆₀₀ of 0.005. Compounds were screened at a single concentration of 20 μM. Bacterial growth was then followed by measuring GFP fluorescence over 7 days. Each plate contained controls for growth of the inoculum without drug (vehicle only, 100% Mtb growth) and control drug isoniazid (INH) at the MIC (3.6 μM; 0.5 μg/mL). We calculated the % inhibition compared to that of the culture grown with vehicle (1% DMSO) only.

We defined initial hits as compounds with >90% of reduction of Mtb-GFP growth on day 7, resulting in a total of 15 hits; 2 hits from PKIS 1 and 13 hits from PKIS 2 (Figure 1). An additional 20 compounds inhibited Mtb between 60 and 90% but were not further evaluated. We next sought to validate the results of the fluorescence-based screen and rule out potential assay-specific artifacts by using a spot growth assay. We plated 5 μL of Mtb at an OD₆₀₀ of 0.05 with 20 μM compound on solid 7H10 medium, incubated for 3 weeks, and inspected for visible growth. All compounds that showed inhibition in the Mtb-GFP assay also showed inhibition in the spot assay.

Figure 1.

Overview of the Mtb whole-cell screen. A total of 879 compounds from the PKIS 1 and 2 collections were screened against an Mtb H37Rv strain constitutively expressing GFP. Percent inhibition was calculated relative to vehicle only on each plate. We defined hits as compounds showing more than 90% inhibition of GFP fluorescence compared to vehicle only. Twenty-six compounds showed above 50% increased fluorescence compared to vector only and were collapsed onto the x-axis. Compounds from PKIS 1 and 2 are separated by the vertical line.

MIC Determination of Hits.

We further characterized all hits with >90% inhibition at 20 μM from the primary screen by determining the MIC. We used the Mtb-GFP strain and measured growth inhibition in the presence of compounds along a 2-fold dilution series. Of the 15 hits with >90% of inhibition, seven did not show robust activity below 20 μM and were not evaluated further. The MIC₅₀ and MIC₉₀ values and chemical structures for the other eight Mtb inhibitors are shown in Figure 2, and their dose response curves are shown in Figure S1. To further confirm results, we determined MIC visually using the spot assay. The MICs obtained in the spot assay agreed with those from the fluorescence assay within the 2-fold range inherent in MIC assays. The eight Mtb inhibitors fall into seven different chemotypes: aminothiazoles, 2-aryl-4-anilinoquinazolines, 4,5-diarylimidazoles, 4-anilinoquinazolines, 4-anilino-7-azaquinazolines, 4-anilinoquinolines, and biarylamides (Figure 2).

Figure 2.

Structures, chemotypes, and MICs of hits. Structures of hit compounds with >90% inhibition of growth at 20 μM and their MIC₅₀, MIC₉₀, and MIC_90NR, the MIC₉₀ under hypoxic, nonreplicating (NR) conditions. The unit for all MICs is μM. ND: not determined.

Activity of Hits against Nonreplicating Mtb.

No or slow replication generally decreases the susceptibility of Mtb to drugs. Hypoxia is one of the best understood stress conditions causing nonreplication and is considered a reliable proxy for physiologic nonreplicating states also encountered in vivo.²⁶ To test hits from our primary screen for activity against nonreplicating Mtb in hypoxic culture, we added compounds as above and transferred the cultures to hypoxic conditions. Cultures were incubated in hypoxic conditions for 7 days, and oxygen depletion was monitored using a resazurin-based anaerobic indicator strip. We then measured viability of the culture by spot assay, as the GFP strain gave unreliable fluorescence readings in hypoxia and the spot assay had faithfully replicated MICs in normoxia. Two hits from the primary screen had better activity against nonreplicating Mtb as expressed by the MIC₉₀ in nonreplicating conditions (MIC_90NR) (Figure 2): AQA and GW578342X were 20- and 2-fold more potent against nonreplicating bacteria with MIC_90NR of 1 and 6 μM, respectively. The 4-anilinoquinazoline GW566221B also had potent activity against replicating and nonreplicating Mtb (MIC₉₀: 3 μM, MIC_90NR: 7 μM). AQA, which had a moderate MIC against replicating Mtb showed the most potent activity against nonreplicating bacteria and was of comparable potency to rifampicin (MIC_NR: 0.4 μM). Thus, we chose AQA for further characterization.

Activity of Hits against Drug-Tolerant Persisters.

Persisters are defined as a distinct bacterial subpopulation characterized by their survival of drug treatment that kills most of the bacterial population.^27,28 We next tested the activity of AQA specifically against persisters. To obtain persisters against which to test the activity of AQA, we treated an Mtb-GFP culture with rifampicin (RIF) and isoniazid (INH) at 100× the MIC. As expected, this treatment rapidly killed >99.9% of bacteria but a small, highly drug-tolerant persister population remained and readily regrew upon removal of drugs (Figure 3A). We next added AQA at 2 and 20 μM to the RIF/INH treatment. Addition of AQA resulted in further reduction of viable bacteria, indicative of persister killing (Figure 3A). AQA at 2 μM showed full activity against persisters and was comparable to its activity at 20 μM. To determine if this effect was persister-specific killing rather than the effect of just adding a third anti-TB drug, we next tested ethambutol (EMB), a clinical TB drug that is not known to kill persisters. Addition of EMB at the MIC did not increase the level of killing by RIF and INH further, indicating that AQA was indeed killing drug-tolerant persisters (Figure 3B). Two replicate experiments showed the same qualitative effect of AQA on persisters (Figure S2).

Figure 3.

Activity of AQA against persisters. (A) Colony forming units after treatment of Mtb with 100× the MIC of INH and RIF in the absence and presence of 2xMIC_90NR of AQA. (B) Unlike AQA, ethambutol at 1xMIC in addition to INH and RIF does not further reduce cfu. One of three representative experiments shown, see Figure S2 for replicate experiments.

AQA was initially developed as a TGFBR1 (ALK5) kinase inhibitor for the treatment of fibrosis.²² It is a highly selective inhibitor with nanomolar activity against TGFBR1, and it has shown minimal toxicity against 15 mammalian cell lines, except for moderate toxicity against the Cas1 glioblastoma cell line. AQA showed no acute toxicity when dosed at 5 mg/kg p.o. in rats, although its bioavailability was poor.²⁹ A resynthesized sample of AQA demonstrated the same MICs to those of the original material sourced from the PKIS.

Strategy for Optimization of AQA.

The dual activity of AQA against Mtb and host TGFBR1 dictated that the development of a bioavailable pharmacological probe required testing for both activities. The design of new dual acting Mtb inhibitors was guided initially by the cocrystal structure of AQA with TGFBR1 (PDB: 3HMM,²² Figure 4). Within the ATP-binding pocket, the 4-pyridine nitrogen forms a key hinge binding interaction with His283, while nitrogen atoms of the quinazoline and 2-pyridyl group stabilize a network of hydrogen bonds with the polar side chains of Glu45, Lys32, and Tyr49 and the backbone NH of Asp151 through a bound water molecule (Figure 4). The benzene ring of the quinazoline points toward the solvent exposed region. The structure suggests that three of the nitrogens in AQA are important for kinase inhibition but indicates that there may be room for modification around the quinazoline core and the pyridine rings.

Figure 4.

Key H-bond interactions of AQA with the ATP-binding pocket of TGFBR1 from analysis of the crystal structure PDB: 3HMM.

Since the bacterial target of AQA was unknown, three independent SAR studies were designed to define the chemical requirement for Mtb inhibition in each of the ring systems (A–C, Figure 5A). By determining the structural requirements for antibacterial activity independently, overlap with the SAR for kinase inhibition might allow for the optimization of dual targeting activity. In parallel to these SAR studies, we also sought to improve the pharmacokinetic properties of AQA. Although AQA was reported to have modest oral bioavailability in rats,²² it showed rapid clearance with a short half-life and low plasma levels in mice (vide infra). The optimal pharmacological probe would emerge from the convergence of the SAR for Mtb inhibition, TGFBR1 inhibition, and metabolic stability (Figure 5B).

Figure 5.

(A) Proposed modification of the AQA chemotype at rings A–C to develop SAR for Mtb inhibition, TGFBR1 inhibition, and metabolic stability. (B) The optimal pharmacological tool would lie in the overlap of the three SARs.

Synthesis of AQA Analogues.

AQA analogues were synthesized by either a four-step route from 2-aminobenzamide I (Scheme 1) or a two-step route from 2,4-dichloroquinolines/quinazolines III (Scheme 2). Selection of the preferred route was based on commercial availability and the cost of starting materials. In the four-step route (Scheme 1), the quinazoline ring was constructed from commercially available 2-aminobenzamide I by amide formation with an aryl carboxylic acid, followed by cyclization to form the quinazoline ring and subsequent POCl₃-mediated chlorination to yield the key 4-chloro-2-arylquinazoline II. Buchwald coupling of intermediate II with 4-aminoheterocyles or 4-aminoaromatics provided AQA analogues 1–4 and 7–24. The ether analogues 5 and 6 were synthesized by the S_NAr reaction of a phenol with intermediate II. In the two-step route, the S_NAr reaction of quinolone/quinazoline intermediate III with 4-aminopyridine under strongly basic conditions provided 2-chloro-N-(pyridin-4-yl)quinoline/quinazolones IV. Palladium-mediated Suzuki or Stille couplings were used to prepare the final compounds, 25–44, depending on the accessibility of the arylboronic acid or arylstannane. The isoquinoline 45 was synthesized using the two-step route from 1,3-dichloroisoquinoline and the corresponding quinoline 46 was prepared using the previously described method.²²

Scheme 1. Four-Step Synthesis of AQA Analogues^a.

^aReagents and conditions: (i) Ar¹–CO₂H, HOBT, EDC, Et₃N; (ii) NaOH, EtOH; (iii) POCl₃, toluene; (iv) Ar²–NH₂, Bu^tONa, Pd₂(dba)₃, BINAP. (v) Ar²–OH, DIPEA, PrⁱOH.

Scheme 2. Two-Step Synthesis of AQA Analogues^a.

^aReagents and conditions: (i) 4-aminopyridine, NaH, DMF; (ii) Ar¹–B(OH)₂, Pd(dppf)Cl₂, K₂CO₃; (iii) Ar¹–SnBu₃, Pd(PPh₃)₄.

Structure Activity for Mtb and TGFBR1 Inhibition.

The SAR study of AQA started with the ring A 4-anilino group. A range of aryl- and heteroaryl-bearing aniline, phenol, and tertiary amine derivatives 1–24 (Table 1) were synthesized. Compounds were tested in normoxic and hypoxic conditions that induce nonreplication, and antibacterial activity was scored by relative potency into four bins (Table 1), which is an established method of analyzing MIC data from antibacterial assays.³⁰ In this round of assays, AQA had an MIC₅₀ < 0.3 μM and an MIC₉₀ between 0.3 and 1.3 μM against nonreplicating Mtb. Analogues that matched AQA for potency were scored 4 (MIC₅₀) and 3 (MIC₉₀), respectively (Table 1). Several of the analogues matched the antibacterial activity of AQA, including the ether linked benzothiazoles 5 and 6, the 2-methylthiazole 7, the methylated benzotriazoles 15–17, and the benzothiophene 22. The 2-methylpyridine 1, benzothiazoles 2 and 3, tertiary amine 4, indazoles 11–13, benzofuran 20, and the substituted aromatics 21, 23, and 24 matched the potency of AQA at MIC₅₀ but were less potent at MIC₉₀, while the benzoxazole 8, N-methylimidazoles 9 and 10, benzotriazole 14, indole 18, and quinoline 19 had reduced bactericidal activity. The overall structure–activity of AQA Ring A showed that a wide range of aniline replacements retained phenotypic activity against nonreplicating Mtb.

Table 1.

Ring A Structure Activity for Mtb and TGFBR1

^aMtb MIC (μM) <0.3++++, 0.3–1.3+++, 1.3–5.0++, 5.0–10+, >10–.

^bTGFBR1 DiscoverX binding assay. ^cTGFBR1 enzymatic assay at 10 mM ATP.

In parallel, the TGFBR1 binding K_D of each analogue was determined using the DiscoverX KINOMEscan assay³¹ except for analogues 1 and 11 which were assayed in a radiometric enzyme inhibition assay at 1 μM concentration. Apart from the parent AQA which had potent TGFBR1 activity (K_D = 9.5 nM), only 3-thiomethylaniline 24 (K_D = 0.34 μM) demonstrated TGFBR1 binding activity below 1 μM. Whereas, indazole 11 showed >50% inhibition of TGFBR1 enzyme activity at 1 μM and the 2-methylpyridine 1 was only slightly less active at 45% inhibition. The only other analogues retaining micromolar TGFBR1 binding activity were the benzothiazole 2, benzoxazole 8, N-methylbenzimidazole 9, indazoles 11 and 12, indole 18, quinoline 19, and benzodioxolane 21. The remaining analogues were inactive at 10 μM in the TGFBR1 binding assay likely due to the loss of the critical H-bond interactions of these heterocycles within the ATP-binding pocket (Figure 4). In summary, the SAR of ring A revealed that while most of the AQA analogues showed bactericidal activity only those that contained potential kinase hinge-binding groups retained any TGFBR1 activity.

To explore SAR of the ring B at the quinazoline 2-position, AQA analogues containing five- and six-membered aryl and heteroaryl units were synthesized. In this round of Mtb assays, AQA had an MIC₅₀ and MIC₉₀ < 1.0 μM (Table 2). The 6-chloro, 6-fluoro, 6-trifluoromethyl, 6-methoxy, and 6-protio derivatives 25–29 were designed to probe the role of the 6-methyl substituent of the 2-pyridyl group in AQA. In the Mtb assay, the 6-chloro analogue 25 retained potent antibacterial activity, while the 6-fluoro 26, 6-trifluoromethyl 27, and 6-protio 29 analogues had reduced activity. The 6-methoxy analogue 28 had greatly reduced activity. The next series of analogues 30–36 were designed to probe the role of the 2-pyridyl nitrogen of AQA. Moving the nitrogen to the 3-position resulted in analogue 30 that was inactive in the Mtb assay. The pyrazines 31 and 32 as well as the pyrimidine 33 had a greatly reduced Mtb activity. The aryl analogues 34–36 that lacked the pyridyl nitrogen were inactive. Three analogues with methyl substituted five-membered heterocycles 37–39 were also inactive. These results demonstrated that 6-methyl-2-pyridyl was the optimal ring B group for antibacterial activity, with little tolerance for variation of either the 6-methyl group or the pyridyl nitrogen.

Table 2.

Ring B Structure Activity for Mtb and TGFBR1

^aMtb MIC (μM) <1.0++++, 1.0–2.5+++, 2.5–5.0++, 5.0–10+, >10–.

^bTGFBR1 enzymatic assay at 10 mM ATP ± 5%.

The ring B analogues 25–29 were tested for inhibition of TGFBR1 enzyme activity at two concentrations: 1 and 10 μM. AQA showed 100% enzyme inhibition at both concentrations. The other 2-pyridyl analogues 25–29 maintained robust TGFBR1 inhibition, while the 3-pyridyl analogue 30 had only weak inhibition. The modified pyrazine, pyrimidine, and aryl analogues 31–36 also had weaker TGFBR1 enzyme inhibition. Among the five-membered heterocycle analogues, the N-methylpyrazole 38 and 4-methylthiazole 39 showed good TGFBR1 inhibition. Overall, the SAR study determined that the original 6-methyl substituted 2-pyridyl group of AQA was both an essential and optimal ring B for dual Mtb and TGFBR1 inhibition, with little room for modification.

The next series of analogs focused on modification of the ring C through the introduction of substitution on either the 6- or 7-position or replacement of the benzene ring of AQA in analogues 40–43. In this third round of Mtb assays, AQA had an MIC₅₀ and MIC₉₀ < 1.0 μM (Table 3). Gratifyingly, the 6-methoxy analogue 40 retained both potent bactericidal activity and TGFBR1 enzyme inhibition at 1 and 10 μM. The 6-hydroxy 41 and 7-methoxy 42 analogues were less active in the Mtb assay but retained TGFBR1 inhibition. Notably, replacement of ring C with N-methylpyrrole gave analogue 43 that retained potent Mtb activity and still demonstrated micromolar TGFBR1 activity. The pyrimidine 44, in which ring C was removed, retained partial Mtb and TGFBR1 activity. These results revealed that the benzene ring portion of ring C was tolerant of a wide range of changes suggesting that it was not critical for engagement of the bacterial target. More dramatic effects on bactericidal activity were seen with the isoquinoline 45 and quinoline 46 derivatives, where one of the two nitrogens was removed from the quinazoline core. Their weaker activity in the Mtb assay showed that both nitrogens were required for optimal bactericidal activity. Analogues 25, 40, and 43 matched the activity of AQA in the Mtb assay while retaining the inhibition of TGFBR1 enzymatic activity. These three analogues were further tested for in-cell target engagement of TGFBR1 by NanoBRET where they demonstrated activity at concentrations matching their enzyme inhibition (Figure S4).

Table 3.

Ring C Structure Activity for Mtb and TGFBR1

^aMtb MIC (μM) <1.0++++, 1.0–2.5+++, 2.5–5.0++, 5.0–10+, >10–.

^bTGFBR1 enzymatic assay at 10 mM ATP ± 5%.

Metabolic Stability of AQA Analogues.

Incubation of AQA with mouse liver microsomes and NADPH resulted in only 14% of the parent remaining after 30 min (Table 4). The rapid metabolism of AQA suggested that it was a substrate for cytochrome P450s. The XenoSite P450 metabolism predictor³² was used to identify potential metabolic hotspots. XenoSite predicted that the ring B pyridyl 6-methyl group and ring A 4-pyridyl nitrogen were likely to be the primary sites for oxidative P450 metabolism (Figure 6).

Table 4.

Mouse Liver Microsome Stability

compound	microsome stability (%)a
AQA	14
7	18
11	9
12	9
15	28
16	30
17	15
21	30
22	8
25	18
26	18
27	29
29	20
40	23
41	20
42	12
43	39

^a% of compound remaining after 30 min of incubation with mouse liver microsomes in the presence of NADPH. All compounds were stable in the absence of NADPH.

Figure 6.

Predicted metabolic hot spots determined by XenoSite.

To confirm these predictions, we screened a range of the AQA analogues to establish a structure–metabolism relationship in mouse liver microsomes. Among the ring A analogues, benzothiazole 7, indazole 11, N-methylindazole 12, N-methylbenzotriazole 17, and benzothiophene 22 had very low metabolic stability, like the parent AQA (Table 5). In contrast, the benzodioxolane 21 and the isomeric N-methylbenzotriazoles 15 and 16 showed a 2-fold improved metabolic stability at 30 min incubation. The analogues with 6-chloro, 6-fluoro, 6-trifluoromethyl, and no substitution (25–27 and 29) on the ring B 2-pyridyl group also showed comparable, and 27 better metabolic stability compared to the parent AQA, which was consistent with the XenoSite prediction of a metabolic hot spot on the 6-methyl group. On ring C, the 6-methoxy and 6-hydroxy substituted quinazoline analogues (40 and 41) showed improved metabolic stability. The greater metabolic stability of 40 compared to that of 42 indicated that modification of the 6-position on the quinazoline ring had a larger effect on metabolic stability compared to the 7-position, which was in line with the XenoSite predictions. Gratifyingly, the 7-methyl-pyrrolopyrimidine 43 showed 3-fold improved metabolic stability compared to the parent AQA. In summary, modification of the predicted sites of oxidative P450 metabolism led to the identification of analogues with 2–3 fold improved metabolic stability. However, no single analogue was identified that retained Mtb and TGFBR1 inhibition and was highly resistant to metabolism by mouse liver microsomes. Three analogues, the benzodioxolane 21, 2-methyltriazolyl 16, and pyrrolo-pyrimidine 43, were selected for a snapshot pharmacokinetic study to determine if their greater metabolic stability in liver microsomes would translate to improved in vivo properties compared to AQA. Each analogue was administered orally at a dose of 10 mg/kg to CD1 mice and the plasma levels determined over 5 h by LC/MS/MS (Table 5). The parent AQA demonstrated a short half-life and low plasma concentration, as expected due to its low metabolic stability. Benzodioxolane 21 and N-methyltriazole 16 displayed lower clearance with half-lives >3 h. The pyrrolopyrimidine 43 demonstrated the highest drug exposure with C_max = 280 ng/mL and AUC_last = 640 h*ng/mL even though its half-life was slightly lower at ∼2 h. Furthermore, 43, which was active at 1 μM on TGFBR1 (Figure S4), was active on only an additional 7/192 human kinases at 10 μM in the K192 NanoBRET panel,³³ which was a comparable to the kinome selectivity profile of AQA (Table S3).

Table 5.

Snapshot Pharmacokinetic Profile AQA and Analogues 16, 21, and 43

Parameter	Unit	AQA	16	21	43
T1/2	h	1.8	4.6	3.4	1.9
Cmax	ng/mL	60	150	75	280
AUClast	h*ng/mL	75	440	250	640

The pyrrolopyrimidine 43 was further evaluated in a full pharmacokinetic study at a dose of 10 mg/kg in CD1 mouse by oral, intravenous, and intraperitoneal routes of administration with plasma levels monitored over 8 h (Figure 7). Pyrrolopyrimidine 43 showed a high volume of distribution of 4.65 L/kg with relative bioavailability of 25 and 65% by the oral and intraperitoneal routes, respectively, demonstrating its potential for use as a pharmacological tool in vivo.

Figure 7.

Mouse pharmacokinetic profile of indoloquinazoline 43.

DISCUSSION AND CONCLUSIONS

Drug resistance is increasingly slowing progress toward TB control. The emergence of drug resistance against all clinically used drugs suggests that strategies that specifically disable drug resistance mechanisms are needed. Among those, the killing of nonreplicating, drug-tolerant bacteria is increasingly recognized as important. However, the few drugs that kill nonreplicating Mtb such as rifampicin and bedaquiline only do so at much higher MICs than are required for killing replicating Mtb. HDT is also thought to reduce the emergence of bacterial resistance and is increasingly considered as a promising treatment avenue. For both approaches, molecular targets and chemical leads are still limited. Here, we discovered a molecule, AQA, with activity against replicating, nonreplicating, and extremely drug-tolerant Mtb persisters in addition to a host-directed activity. AQA is a useful probe to better understand the biology underlying nonreplicating Mtb and Mtb persisters. It also provides a unique lead to test the potential of dual host and pathogen targeting within a single molecule. To study the potential of dual targeting to overcome the development of drug resistance, the optimal pharmacological probe would have bioactivity against Mtb and TGFBR1 within a similar dose range combined with pharmacokinetic properties that support in vivo studies. Identification of a single agent that satisfied the three structure-activities was challenging. Having defined the structural requirements for bioactivity in the AQA chemotype, we identified multiple analogues with dual activity against nonreplicating Mtb and TGFBR1. Unfortunately, while the SAR studies of the AQA chemotype demonstrated the critical role of the pyridyl 6-methyl group in ring B for optimal bioactivity at Mtb and TGFBR1, the structure-metabolism study showed that the same methyl group was a liability for metabolic stability. Notably, the benzene ring of quinazoline ring C was relatively tolerant of modification and conferred some modest increase in metabolic stability. The pyrrolopyrimidine 43 emerged as the most promising AQA analogue. While it was not the most potent dual inhibitor, it balanced all three requirements by retaining low-micromolar activity against nonreplicating Mtb and micromolar inhibition of TGFBR1. Importantly, pyrrolopyrimidine 43 had a superior pharmacokinetic profile for in vivo studies in mice after intraperitoneal dosing. The results of pharmacological studies to test the dual targeting potential will be reported in due course. These structure–activity relationship studies also provide useful tool compounds to parse the contributions of AQA’s host- or pathogen-directed effect. For example, compound 39 lost activity against Mtb, while compound 7 lost activity against TGFBR1 without affecting the other activity.

The wide range of physiologic states in a bacterial population results in an equally wide range of drug susceptibilities that complicate and prolong the treatment of bacterial infections. Perhaps the most extreme example of this phenomenon is persisters, a subpopulation of phenotypic variants with an altered metabolic state that is defined by reversible but often extreme antibiotic tolerance. Although they only make up between 0.1 and 0.001% of a bacterial population, persisters are a stepping stone toward genetic resistance and prolong the treatment times required for full clearance of an infection. In this way, the small persister population has a large negative impact on treatment. The molecular basis for persistence is still poorly understood and as a result, molecular targets in persisters are also lacking, especially in Mtb. In recent years, some compounds with activity against bacterial persisters have been identified; however, their discovery is still largely serendipitous, and systematic efforts to identify persister drugs have been hampered by the small frequency of persisters in a bacterial population. The altered metabolic state of persisters further complicates their detection by typical assays used in drug screening, as the inherent slow-growth of persisters escapes most growth-based assays. Here, we chose a whole cell-based screening format that, by stepwise screening against replicating, nonreplicating, and last persister cells, downselected compounds with likely persister activity and identified the compound AQA from a relatively small initial library.

Whole-cell screens do not reveal the molecular target of a compound, and target identification can be challenging. The molecular efficacy target(s) of AQA also remain to be identified. We were not able to generate Mtb mutants resistant to AQA. While this would be a desired feature of an antibiotic, it also precludes one of the most straightforward ways to target identification. The use of a targeted chemical library, however, inherently provides some information about the likely molecular target: the chemical focus of the PKIS libraries on ATP-competitive inhibitors means that any target is likely a kinase, an ATPase, or another ATP-binding protein, a useful starting point for target identification and allows for prioritizing candidates from ‘omics approaches.

An unusual aspect of AQA is that its host activity, for which it was originally developed, could be advantageous rather than an impediment. A recent study showed that TGFBR is activated in TB and its activation suppresses a timely and effective T-cell response to Mtb in the lung. A corollary of this finding is that inhibition of TGFBR in the context of TB might be beneficial to the host and restore CD4 T-cell survival and function.¹⁷ The possibility of a dual activity of AQA—one against the host, one against Mtb—that both kill Mtb including persisters, needs further testing but makes AQA a highly interesting compound with potentially multiple beneficial anti-Mtb activities in a single molecule.

EXPERIMENTAL SECTION

Screening the PKIS (1) and (2) Libraries.

We obtained 359 and 520 compounds from PKIS 1 and PKIS 2 collections. The total of 879 compounds was screened in a whole-cell screen using Mtb H37Rv strain expressing GFP from a strong constitutive promoter (Rv::pHIGH22²⁵). The screen was carried out in 96-well plates in 200 μL total volume per well and Mtb at a final OD₆₀₀ of 0.005. Compounds were added at 20 μM in a final DMSO concentration of 1%. The outer wells of each plate were filled with medium only to avoid positional and evaporation effects. Each plate contained the following controls: isoniazid at 1xMIC, vector only (1% DMSO, the 100% growth control). Plates were incubated in the presence of compound for 7 days and GFP fluorescence measured. Blank values from medium only wells were subtracted from all measurements and % inhibition calculated relative to the vehicle only control. Compounds were considered hits if they reduced GFP signal by 90% or more compared to the vector only control. In parallel to the GFP assay, 5 μL of each sample was spotted onto 7H11 solid agar and incubated for ∼3 weeks. Plates were visually inspected for growth and those showing no growth were cross-referenced with the GFP data (and the other way around). For screening of hits under hypoxic conditions, plates were set up as above except for using 0.1 OD₆₀₀ of Mtb-GFP and immediately incubated in a GasPak Pro system after drug addition to rapidly deplete oxygen. Oxygen depletion was monitored by indicator strips. Cells were then further incubated for 1–5 days under aerobic conditions and CFU measured.

MIC Determination.

Several hits from the primary screen were further tested to determine the MIC. Serial 2-fold compound dilutions starting at 20 μM were added to Mtb as described above. Fluorescence was measured after 7 days and the cultures spotted onto 7H11 agar as previously described. The fluorescence readings were transformed into % inhibition relative to no inhibitor and fitted by nonlinear regression to determine the MIC₅₀ and MIC₉₀. Compounds were tested in triplicate.

For the determination of MICs of AQA analogues, we switched to a luminescence-based reporter strain with a larger dynamic range than the GFP strain. Frozen stock of Mtb H37Rv::luxCDAB was grown by dilution into Middlebrook 7H9 broth supplemented with 0.5% glycerol, 10% OADC, and 0.05% Tween80 and grown at 37 °C for at least two doublings. The culture was diluted to an OD₆₀₀ of 0.05. Test compounds in DMSO were placed in a white 96-well flat-bottom assay plate and serially diluted with 0.2% DMSO final concentration. Mtb was added to each well for a final volume of 250 μL/well, with some wells filled with a 10 or 1% dilution of the Mtb inoculum to serve as MIC₉₀ and MIC₉₉ reference controls. Each plate contained at least two replicates of each test condition. 125 μL of the final drug/Mtb mixture from each well was moved into a new assay plate to create an identical copy of each plate. One plate was used to determine the MICs under normoxic, one under hypoxic conditions. For normoxia, plates were placed in a gas-permeable plastic bag and incubated at 37 °C for 5 days. Luminescence was then quantified using a PHERAstar plate reader. For hypoxia-reaeration assays, plates were sealed into airtight FoodSaver bags with a BD GasPak to remove oxygen and an oxygen indicator strip. Sealed bags were incubated at 37 °C for 7 days and oxygen depletion monitored. After 7 days, the plates were removed from hypoxia, transferred to gas-permeable bags, grown for an additional 5 days at 37 °C, and luminescence quantified. The average luminescence values from the no-drug samples were used as the baseline, and the average values in the 10 and 1% diluted cultures were used as the cutoffs for assessing MIC₉₀ and MIC₉₉.

Persistence Assay.

Wild-type H37Rv culture was grown in 7H9 medium supplemented with OADC to an OD₆₀₀ of 1 and diluted 1:1 with drug in 7H9 to a final volume of 150 μL in a 96-well v-bottom plate. To select for persisters, INH and RIF were added at 100× MIC. Additional drugs (AQA, ETB) were added at 1xMIC. Plates were incubated at 37 °C without shaking for 14 days. Culture was removed at days 7 and 14 for cfu plating. After 14 days, cultures were washed twice with 150 μL 7H9 medium and then resuspended in fresh 7H9 medium. The cultures were incubated at 37 °C without shaking for another 28 days to allow for recovery. At days 21 and 28 after removing the drug, cultures were plated for cfu.

Purity Statement.

The purity of compounds submitted for biological screening was determined to be ≥95% as measured by HPLC.

Additional experimental information on biological assays and chemistry can be found in the Supporting Information.

ABBREVIATIONS USED

AQA

anilinoquinazoline

cfu

colony forming unit

EMB

ethambutol

HDT

host-directed therapy

IFNγ

interferon gamma

INH

isoniazid

MIC

minimal inhibitory concentration

OD

optical density

Mtb

Mycobacterium tuberculosis

PKIS

published kinase inhibitor set

RIF

rifampicin

SAR

structure–activity relationship

TB

tuberculosis

TGF

transforming growth factor

TGFBR1

transforming growth factor beta receptor 1