Comprehensive physicochemical, pharmacokinetic and activity profiling of anti-TB agents

Abstract

Objectives

The discovery and development of TB drugs has met limited success, with two new drugs approved over the last 40 years. Part of the difficulty resides in the lack of well-established in vitro or in vivo targets of potency and physicochemical and pharmacokinetic parameters. In an attempt to benchmark and compare such properties for anti-TB agents, we have experimentally determined and compiled these parameters for 36 anti-TB compounds, using standardized and centralized assays, thus ensuring direct comparability across drugs and drug classes.

Methods

Potency parameters included growth inhibition, cidal activity against growing and non-growing bacteria and activity against intracellular mycobacteria. Pharmacokinetic parameters included basic physicochemical properties, solubility, permeability and metabolic stability. We then attempted to establish correlations between physicochemical, in vitro and in vivo pharmacokinetic and pharmacodynamic indices to tentatively inform future drug discovery efforts.

Results

Two-thirds of the compounds tested showed bactericidal and intramacrophage activity. Most compounds exhibited favourable solubility, permeability and metabolic stability in standard in vitro pharmacokinetic assays. An analysis of human pharmacokinetic parameters revealed associations between lipophilicity and volume of distribution, clearance, plasma protein binding and oral bioavailability. Not surprisingly, most compounds with favourable pharmacokinetic properties complied with Lipinski's rule of five.

Conclusions

However, most attempts to detect in vitro–in vivo correlations were unsuccessful, emphasizing the challenges of anti-TB drug discovery. The objective of this work is to provide a reference dataset for the TB drug discovery community with a focus on comparative in vitro potency and pharmacokinetics.

Introduction

One-third of the world's population is infected with TB, of whom 5%–10% may develop active TB in their lifetime.¹Mycobacterium tuberculosis (Mtb) is the causative agent of TB, a disease that causes 1.4 million deaths each year. In 2012, around 8.6 million people developed TB (including ∼400 000 MDR TB cases).² The treatment of TB is becoming challenging due to the emergence of MDR TB and XDR TB, and increasing numbers of TB–HIV coinfections. Recent reports of totally drug-resistant³ TB cases that are resistant to all anti-TB drugs further emphasize the need for novel drugs that are active against all forms of TB.

It is widely recognized that, apart from in vitro potency on target, physicochemical and pharmacokinetic (PK) properties need to be considered early in drug discovery. Following studies in the late 1990s indicating that poor PK and toxicity were important causes of costly late-stage failures in drug development, in vitro assays of absorption, distribution, metabolism and elimination (ADME), as well as in vitro toxicity assays, were integrated into the early stages of drug discovery. Early in vitro ADME and toxicity assessment resulted in a dramatic reduction of attrition associated with poor drug exposure and adverse events.⁴ Physicochemical parameters affect absorption, oral bioavailability and other ADME parameters. Experimental and computational analyses of very large sets of small-molecule drugs resulted in Lipinski's rule of five, whereby poor absorption or permeation is more likely if a compound exhibits one or more of the following properties: molecular weight (MW) >500, calculated logP or octanol:water partition coefficient (cLogP) >5, hydrogen bond donors (HBD) >5 or hydrogen bond acceptors (HBA) >10.⁵ Veber's rule added that compounds with <10 rotatable bonds and a polar surface area (PSA) <140 Ų are more likely to have good oral bioavailability.⁶ Further, Gleeson's analysis revealed that almost all ADME parameters deteriorate with increasing MW, logP or both, thus recommending MW <400 and cLogP <4 to improve ADME parameters.⁷ In addition, Egan et al. showed that cLogP and PSA could be used as a tool (the Egan Egg model) for predicting good passive gut absorption.⁸

In early drug discovery, emphasis is most often placed on in vitro potency, while desirable physicochemical and PK parameters are considered as secondary attributes in the decision-making process. To be efficacious, anti-infectives have to (i) effectively penetrate the microbe and inhibit an essential cellular process and (ii) reach the site of infection—lung tissue and lesions for pulmonary TB. Hence, the chemical space occupied by antibacterials⁹ and antimycobacterials¹⁰ is distinct from that of drugs targeting non-infectious diseases.

Recently, Franzblau et al.¹¹ carried out a comprehensive analysis of various microbiological and animal study protocols used by different TB research groups to characterize compounds. A guide for medicinal chemists to optimize compounds against TB has also been discussed.^12,13 Further, the physicochemical, biopharmaceutical and clinical PK–pharmacodynamic (PD) properties of selected anti-TB compounds and novel agents under development have been reviewed.^14–20 These reviews do not include comparative analyses and correlation between in silico, in vitro and in vivo properties. Here we have generated in vitro potency and in vitro PK for a set of 36 compounds using standardized procedures. In silico physicochemical and in vivo pharmacological parameters have also been compiled. We have used computational approaches to seek relationships between in silico, in vitro and human in vivo parameters to guide early drug discovery efforts against TB. Not surprisingly, lipophilicity emerged as a major factor driving the in vitro and in vivo PK properties.

Materials and methods

Bacterial strains, culture media and chemicals

Mtb H37Rv (ATCC #27294) was maintained in Middlebrook 7H9 broth supplemented with 0.05% Tween 80 and 10% ADS (5% albumin, 2% dextrose and 0.81% sodium chloride) supplement. Alternatively, Dubos broth (Difco) supplemented with 0.05% (v/v) Tween 80, and 10% Dubos medium albumin (Difco) was used for the Wayne hypoxia model.^21,22 For nutrient starvation (the Loebel model), Mtb cells grown in 7H9 medium were washed and resuspended using PBS supplemented with 0.25% Tween 80 and rolled in a 1 L roller bottle for 14 days at 37°C as described elsewhere.²³ All the standard compounds were obtained from commercial sources except for bedaquiline (TMC-207), PA-824, delamanid (OPC-67683) and PNU-100480, which were synthesized at Novartis. Stock solutions of the compounds were prepared using 90% DMSO. The 36 compounds used in the study (Table 1) were isoniazid, rifampicin, pyrazinamide, ethambutol, streptomycin, kanamycin, amikacin, p-aminosalicylic acid, cycloserine, ethionamide, rifabutin, rifapentine, moxifloxacin, levofloxacin, gatifloxacin, ciprofloxacin, ofloxacin, sparfloxacin, capreomycin, thioacetazone, linezolid, prothionamide, clarithromycin, amoxicillin, clavulanate, meropenem, clofazimine, metronidazole, thioridazine, mefloquine, vancomycin, valnemulin, PNU-100480, PA-824, bedaquiline and delamanid.

Table 1.

Anti-TB agents and their properties

No.	Compound	SM/NP	Route	Activity	Chemical class
1	isoniazida	SM	oral	TB	pyridines
2	rifampicin	NP	oral	TB/Gm+	rifamycins
3	pyrazinamidea	SM	oral	TB	pyrazines
4	ethambutol	SM	oral	TB	ethylenediamine
5	streptomycin	NP	injectable	TB/Gm+/Gm−	aminoglycosides
6	kanamycin	NP	injectable	Gm+/Gm−	aminoglycosides
7	amikacin	NP	injectable	Gm+/Gm−	aminoglycosides
8	p-aminosalicylic acida	SM	oral	TB	salicylates
9	cycloserine	NP	oral	TB	—
10	ethionamidea	SM	oral	TB	pyridines (thioamides)
11	rifabutin	NP	oral	TB	rifamycins
12	rifapentine	NP	oral	TB	rifamycins
13	moxifloxacin	SM	oral	Gm+/Gm−	fluoroquinolones
14	levofloxacin	SM	oral	Gm+/Gm−	fluoroquinolones
15	gatifloxacin	SM	oral	Gm+/Gm−	fluoroquinolones
16	ciprofloxacin	SM	oral	Gm+/Gm−	fluoroquinolones
17	ofloxacin	SM	oral	Gm+/Gm−	fluoroquinolones
18	sparfloxacin	SM	oral	Gm+/Gm−	fluoroquinolones
19	capreomycin	NP	injectable	TB	aminoglycosides
20	thioacetazonea	SM	oral	TB	—
21	linezolid	SM	oral	Gm+	oxazolidinones
22	prothionamidea	SM	oral	TB	thioamides
23	clarithromycin	NP	oral	Gm+	macrolides
24	amoxicillin	NP	oral	Gm+/Gm−	penicillins
25	clavulanate	NP	oral	Gm+/Gm−	β-lactams
26	meropenem	NP	injectable	Gm+/Gm−	carbapenem
27	clofazimine	NP	oral	leprosy/TB	riminophenazine
28	metronidazolea	SM	oral	anaerobic	nitroimidazole
29	thioridazine	SM	oral	antipsychotic	phenothiazine
30	mefloquine	SM	oral	antimalarial	isoquinolines
31	vancomycin	NP	injectable	Gm+	polypeptides
32	valnemulin	NP	oral	Gm−	pleuromutilins
33	PNU-100480a	SM	oral	TB	oxazolidinone
34	PA-824a	SM	oral	TB	nitroimidazole
35	bedaquiline	SM	oral	TB	diarylquinoline
36	delamanida	SM	oral	TB	nitroimidazole

SM, synthetic molecule; NP, natural product (including compounds of natural origin and semi-synthetic derivatives); Gm+, Gram-positive bacteria; Gm−, Gram-negative bacteria.

^aPro-drug.

Physicochemical parameters

Data on MW, cLogP, PSA (which measures the surface sum of the overall polar atoms), rotatable bonds, HBD and HBA were generated using proprietary software based on standard algorithms (Medchem FOCUS).

MIC₅₀ determination

MIC₅₀ was determined using the laboratory strain Mtb H37Rv by turbidimetric assay as previously described,^24,25 with slight modifications. Briefly, compounds dissolved in 90% DMSO were 2-fold serially diluted in duplicates and spotted by mosquito HTS liquid handler (TTP LabTech, Hertfordshire, UK) in to 384-well clear plates, resulting in 10 dilutions of each compound. A volume of 50 μL of Mtb culture (final OD₆₀₀ of 0.02) was added to each well, and the plates were incubated at 37°C for 5 days. OD₆₀₀ values were recorded using a SpectraMax M2 spectrophotometer and MIC₅₀ curves were plotted using Graph Pad Prism 5 software.

Bactericidal assays [MBC₉₀, Wayne cidal concentration (WCC₉₀) and Loebel cidal concentration (LCC₉₀) determination]

Compounds dissolved in 90% DMSO were 2-fold serially diluted in duplicates and spotted by mosquito HTS liquid handler (TPP LabTech) in to 96-well plates, resulting in 10 dilutions of each compound. Depending on the assay, we added 200 μL of either aerobically growing Mtb or 14-day-old Loebel nutrient-starved non-growing Mtb cells to each well and the plates were incubated at 37°C under aerobic conditions for 5 days for the determination of MBC and LCC, respectively.^21,23,24 Rifampicin (8.22 mg/L) and isoniazid (13.7 mg/L) were used as positive and negative controls, respectively, during LCC determination. For WCC determination, 200 μL of Mtb H37Rv that had been subjected to growth in the Wayne hypoxia model^21,22 for 20 days was added to 96-well plates containing serially diluted compounds and the plates were incubated under hypoxic conditions for 5 days at 37°C. Metronidazole (68.5 mg/L) and isoniazid (4.3 mg/L) were used as positive and negative controls, respectively, during WCC determination. Viability was determined by plating on 7H11 agar plates and by cfu determination. The concentration at which 90% of the initial inoculum was killed after 5 days of incubation under aerobic, nutrient-rich conditions was defined as the MBC₉₀. Similarly, a 90% reduction in the number of viable bacteria under aerobic, nutrient-starved and hypoxic, nutrient-rich conditions was defined as the LCC₉₀ and WCC₉₀, respectively_.

Macrophage assay

All compounds were profiled for their activity against Mtb-infected mouse macrophage cells (RAW264.7) at the Institut Pasteur, South Korea, using standard procedures.²⁶

Cytotoxicity

Cytotoxicity was tested in HepG2 (ATCC #HB-8065) and BHK-21 (ATCC #CCL-10) cell lines in 96-well microplates as described elsewhere.²⁵ Relative fluorescence units were measured at OD₄₅₀ using a Tecan Safire² and CC₅₀ curves were plotted using Graph Pad Prism 5 software.

In vitro PK studies

In vitro PK parameters such as solubility, permeability [parallel artificial membrane permeability assay (PAMPA) and Caco-2] and metabolic stability in mouse and human liver microsomes were measured at Cyprotex using standard protocols as described at www.cyprotex.com/admepk/.^27,28 Thirty-two compounds were profiled in various assays. Thermodynamic solubility was determined in PBS buffer at pH 7.4 using HPLC-UV. For solubility (mg/mL), the mean values from two independent replicates are given. Mean apparent permeability (P_app with units being 10⁻⁶ cm/s) values from four replicates are given for PAMPA permeability. Caco-2 permeability was provided as mean P_app (10⁻⁶ cm/s) (i) from apical to basolateral (A–B) and (ii) from basolateral to apical (B–A), as well as the ratio between the two values, from two independent replicates. In vitro metabolic stability in both mouse and human liver microsomes was determined and their intrinsic clearance (μL/min/mg protein) values are given. PAMPA, Caco-2 and metabolic stability sample analysis was carried out using optimized LC-MS/MS methods.

Mouse in vivo PK and efficacy studies

All procedures involving mice were reviewed and approved by the institutional animal care and use committee of the Novartis Institute for Tropical Diseases. Animal studies were carried out in accordance with the Guide for the Care and Use of Laboratory Animals. All animals were kept under specific pathogen-free conditions and fed water and chow ad libitum, and all efforts were made to minimize suffering or discomfort.

In vivo PK data for ethambutol, cycloserine, ethionamide, mefloquine, clofazimine, valnemulin and PA-824 were obtained in CD-1 female mice. For isoniazid, rifampicin, pyrazinamide, moxifloxacin, linezolid and bedaquiline, PK parameters were compiled from the literature (Table S1, available as Supplementary data at JAC Online). The compounds were formulated either in 0.5% carboxy methyl cellulose or 10% hydroxypropyl-β-d-cyclodextrin suspension and 10% lecithin in water for PA-824²⁹ and administered orally at doses ranging from 10 to 200 mg/kg. Blood samples were collected by retro-orbital bleeding at various timepoints (0.08 to 24 h) post-dose. All procedures during the PK experiments were performed under isoflurane anaesthesia.³⁰ Groups of three mice were used for each timepoint. Blood was centrifuged at 13 000 rpm for 7 min at 4°C and plasma was harvested and stored at –20°C until analysis. Plasma samples were analysed using optimized LC-MS/MS conditions and PK parameters were calculated using WinNonlin software version 5.1.

In vivo acute mouse efficacy studies were carried out as described earlier.^11,25 Briefly, animals were infected intranasally with 10³ Mtb H37Rv bacilli. One week post-infection, animals were treated for 1 month with the appropriate dose. At 2 and 4 weeks post-dosing, the lungs were aseptically removed and homogenized and the bacterial loads were estimated by plating serial dilutions on 7H11 agar plates. Mean log cfu reductions (mean ± SD) were calculated from five animals at each timepoint and compared with untreated controls.

Statistical analysis

Correlations were tested using TIBCO Spotfire version 4.0.2 (http://spotfire.tibco.com). Statistical analysis was carried out using Prism software (GraphPad Prism, version 5.02, San Diego, CA, USA; www.graphpad.com). A Spearman's correlation analysis was performed to determine the relationship between various parameters.

Results and discussion

To benchmark and compare the physicochemical, potency and PK properties of the available anti-TB agents, we experimentally determined and compiled these parameters for 36 anti-TB compounds using standardized and centralized assays, thus ensuring direct comparability across drugs and drug classes. The compound set included first- and second-line anti-TB agents, clinical candidates as well as drugs that exhibited in vitro activity against Mtb (Table 1). Of the 36 compounds selected, 15 were natural products (compounds of natural origin and semi-synthetic derivatives) and the remaining 21 were fully synthetic molecules. The majority of these drugs are administered orally, with a few injectables. Half of the compounds were initially developed as either Gram-positive or Gram-negative antibacterials. These were introduced as part of combination regimens for the treatment of MDR TB due to the emergence of resistance to first-line anti-TB drugs (Table 1 and Figure S1).

Physicochemical properties

Physicochemical parameters and Lipinski rule violations are presented in Table 2. Most compounds follow Lipinski's rule of five and Veber's rule with MW <500 (n = 25), cLogP <5 (n = 30), HBD <5 (n = 29) and HBA <10 (n = 27), and with PSA <120 Ų (n = 26) and rotatable bonds <10 (n = 34), which is predictive of good oral bioavailability (Table 2).^5,6 The majority of the compounds display acceptable drug-like properties with cLogP <5, the exception being natural product derivatives such as aminoglycosides and rifamycins (Figure 1a–e). All compounds were subjected to the Egan Egg analysis (Figure 2).⁸ Twenty-four compounds were within the inner ellipsoid and are therefore predicted to exhibit good passive absorption through the gastrointestinal tract, in agreement with acceptable oral bioavailability (Table S2). Compounds falling outside the ellipsoid are natural products (Table 1). Interestingly, both rifampicin and rifapentine have good oral bioavailability in humans, suggesting that they may be actively transported.⁸ Amoxicillin, located on the borderline of the ellipsoid, is known to be absorbed by active transport processes.³¹

Table 2.

Physicochemical parameters

Compound	MW	cLogP	HBD	HBA	Lipinski violations	PSA (Å2)	RB
Isoniazid	137.14	−0.67	2	4	0	68.01	2
Rifampicin	822.96	3.71	6	16	3	220.15	5
Pyrazinamide	123.12	−0.68	1	4	0	68.87	1
Ethambutol	204.31	0.12	4	4	0	64.52	9
Streptomycin	581.58	−4.26	14	19	3	331.43	11
Kanamycin	484.51	−5.17	11	15	2	282.61	6
Amikacin	586.60	−6.36	13	18	3	326.15	10
p-Aminosalicylic acid	153.14	1.06	3	4	0	83.55	1
Cycloserine	102.09	−1.19	2	4	0	64.35	0
Ethionamide	166.25	1.73	1	2	0	38.91	2
Rifabutin	847.03	4.73	5	15	2	205.55	5
Rifapentine	877.05	5.09	6	16	4	220.15	6
Moxifloxacin	401.44	−0.08	2	7	0	86.27	4
Levofloxacin	361.38	−0.51	1	7	0	77.48	2
Gatifloxacin	374.42	3.40	1	6	0	74.24	4
Ciprofloxacin	331.35	−0.73	2	6	0	77.04	3
Ofloxacin	361.38	−0.51	1	7	0	77.48	2
Sparfloxacin	392.41	−0.61	3	7	0	103.06	3
Capreomycin	668.72	−7.06	15	22	4	375.92	10
Thioacetazone	236.30	0.88	3	5	0	79.51	4
Linezolid	337.35	0.17	1	7	0	71.11	4
Prothionamide	180.27	2.26	1	2	0	38.91	3
Clarithromycin	747.97	2.37	4	14	2	182.91	8
Amoxicillin	365.41	−1.87	4	8	0	132.96	4
Clavulanate	199.16	−1.07	2	6	0	87.07	2
Meropenem	383.47	−3.28	3	8	0	110.18	5
Clofazimine	473.41	7.70	1	4	1	44.68	4
Metronidazole	171.16	−0.46	1	6	0	86.34	3
Thioridazine	370.58	6.00	0	2	1	6.48	4
Mefloquine	378.32	3.67	2	3	0	45.15	4
Vancomycin	1449.29	−1.14	19	33	3	530.49
Valnemulin	564.83	5.57	3	7	2	118.72	10
PNU-100480	353.42	1.00	1	6	0	61.88	4
PA-824	359.26	2.79	0	8	0	93.8	6
Bedaquiline	555.52	7.25	1	4	2	45.59	8
Delamanid	534.50	5.25	0	10	2	106.27	9
Anti-TB compounds, n = 36	428.31 (102–1449)	0.86 (−7.1 to 7.7)	3.78 (0–19)	8.59 (2–33)		127.71 (6.5–530.5)	0–11
Antibacterials (only Gram-positive activity), n = 33	813 (290–1880)	2.1 (−5.1 to 6.4)	7.1 (1–29)	16.3 (2–46)		243 (29–764)
Antibacterials (only Gram-negative activity), n = 113	414 (138–1203)	−0.1 (−5.0 to 3.3)	5.1 (1–23)	9.4 (41–29)		165 (64–491)

cLogP, calculated logP or octanol:water partition coefficient; Lipinski violations, number of Lipinski rule violations (referring to Lipinski's rule of five); RB, rotatable bond.

Mean values and ranges (in parentheses) are provided for the summary parameters. Gram-positive and Gram-negative antibacterial physicochemical parameters were obtained from O'Shea and Moser.⁹

Figure 1.

Physicochemical properties of anti-TB compounds and their relationship with cLogP. (a) MW. (b) PSA. (c) Rotatable bonds. (d) HBD. (e) HBA. The horizontal and vertical lines are the cut-offs from Lipinski's rule of five (a, d and e) and Veber's rule (b and c). The majority of the compounds lie within Lipinski's space (lower left quadrant). The compounds are categorized as synthetic molecules (open squares) and natural products (filled squares) based on their origin (Table 1).

Figure 2.

Egan Egg analysis of 36 anti-TB compounds. The numbering of compounds is the same as in Table 1. Compounds within the ellipsoid are known to have good passive absorption and those outside have poor absorption. Compounds within the red and blue dashed circles are injectables and rifamycins, respectively.

Biological profiling of anti-TB compounds

Mtb in human lung lesions exists in growing and non-growing states. Further, bacilli can be found extracellularly and intracellularly. Therefore, a battery of in vitro potency assays is employed to characterize the antimycobacterial properties of substances under various conditions, in addition to standard determinations of growth inhibition and cidal activity. While these data are for the most part available in the literature, we sought to generate a centralized dataset using standardized assays in a single laboratory, thus eliminating interlaboratory variability and generating values that can be reliably compared across compound classes. The results are presented in Table 3. Compounds displayed a wide range of MIC₅₀ values, from 0.01 to 11 mg/L (Table 3). Pyrazinamide, metronidazole, meropenem and amoxicillin were found to be inactive (MIC₅₀ >30 mg/L). Pyrazinamide, known to be inactive in standard neutral pH medium and to require an acidic pH to exert its growth inhibitory activity, showed an MIC₅₀ of 25 mg/L at pH 5.5. Although metronidazole was inactive against growing Mtb, it was active against hypoxic non-growing Mtb with a WCC₉₀ of 8.56 mg/L (Table 3).²² Meropenem in combination with clavulanate has shown good inhibitory activity both in vitro and in vivo^32,33 but when tested individually neither drug showed any in vitro inhibitory activity against growing or non-growing Mtb (Table 3). The MIC₅₀ data presented in the current study are comparable to the data obtained by Franzblau et al.¹¹ using similar media. Two-thirds of the compounds tested showed cidal activity against replicating Mtb. For cidal compounds, the MBC₉₀ values were in general 2- to 4-fold greater than the MIC₅₀ values. Some compounds, such as p-aminosalicylic acid, linezolid, thioacetazone, vancomycin and PNU-100480, showed low MIC values but had poor cidal activity. For bedaquiline, only weak cidal activity was observed with the standard incubation time. The discrepancy between this and published data is likely to be due to the longer incubation time required to detect a full cidal activity of bedaquiline.³⁴ Approximately two-thirds of the compounds were active against intracellular Mtb at a concentration of 5 mg/L or lower (Table 3).

Table 3.

In vitro potency and cytotoxicity

Compounds	MIC50 (mg/L)	MBC90 (mg/L)	Macrophage IC90 (mg/L)	WCC90 (mg/L)	LCC90 (mg/L)	CC50 (mg/L)a
Isoniazid	0.04	0.04–0.09	0.04	>13.71	>13.71	>2.74
Rifampicin	0.02	0.06	0.26	0.82	8.23	>16.46
Pyrazinamide	>30	>9.85	>2.46	>12.31	>12.31	>2.46
Ethambutol	0.60	0.51–1.02	>4.09	>20.43	>20.43	>4.09
Streptomycin	0.09	0.37	0.73	5.82	>58.16	>11.63
Kanamycin	0.76	2.42	2.42	9.69	>48.45	>9.69
Amikacin	0.22	0.37	11.73	5.87	>58.66	>11.73
p-Aminosalicylic acid	0.09	1.53	>3.06	>15.31	>15.31	>3.06
Cycloserine	2.14	8.17	>2.04	>10.21	>10.21	>2.04
Ethionamide	0.34	0.83–1.66	3.33	>16.63	>16.63	>3.33
Rifabutin	0.01	0.03	<0.03	0.42	8.47	>16.94
Rifapentine	0.02	0.07	<0.03	0.44	8.77	>17.54
Moxifloxacin	0.11	0.12–0.25	1.00	4.01	>40.14	>8.03
Levofloxacin	0.30	0.45–0.90	>7.23	18.07	>36.14	>7.23
Gatifloxacin	0.07	0.24	0.47	18.72	>37.44	>7.49
Ciprofloxacin	0.31	0.83	—	16.57	>33.14	—
Ofloxacin	0.56	0.11–0.23	3.61	>36.14	>36.14	>7.23
Sparfloxacin	0.08	0.12–0.25	0.98	3.92	>39.24	>7.85
Capreomycin	0.67	>0.84	3.34	6.69	>66.87	>13.37
Thioacetazone	0.05	>18.90	>4.73	>23.63	—	—
Linezolid	0.51	3.37	6.75	>33.74	>33.74	>6.75
Prothionamide	0.16	0.90	0.90	>18.03	>18.03	>3.61
Clarithromycin	—	—	7.48	>74.80	—	—
Amoxicillin	>30	>29.23	>7.31	>36.54	>36.54	>7.31
Clavulanate	>15.93	>15.93	>3.98	>19.92	>19.92	>3.98
Meropenem	>30.68	>30.68	>7.67	38.35	>38.35	>7.67
Clofazimine	0.12	18.94	9.47	23.67	>47.34	>9.47
Metronidazole	>136.93	>136.93	>3.42	8.56	>17.12	>3.42
Thioridazine	5.21	14.82	7.41	7.41–18.53	18.53	>7.41
Mefloquine	3.78	9.46	>7.57	9.46	9.46	4.20/2.80
Vancomycin	11.01	>115.94	28.99	>144.93	—	>28.99
Valnemulin	0.68	2.82	11.30	56.48	—	—
PNU-100480	0.18	1.77	—	—	—	—
PA-824	0.3	0.23	0.45	7.19	—	>7.19
Bedaquiline	0.18	5.56	0.35	2.78	>11.11	>11.11
Delamanid	0.02	—	—	—	—	—

—, Not tested.

^aCytotoxicity was measured against HepG2 (human hepatocellular carcinoma) and BHK-21 (baby hamster kidney) cell lines. The highest concentration tested was 20 μM, with all compounds having a CC₅₀ >20 μM against both cell lines, reported here in mg/L.

Two models of non-growing Mtb were employed: the Wayne model (where the dormant state is induced by oxygen starvation) and the Loebel model (where Mtb is subjected to nutrient starvation). In general, non-growing Mtb was not affected by most compounds (i.e. the concentrations that killed 90% of the initial inoculum were greater than the highest concentrations tested, which ranged between 10 and 145 mg/L) (Table 3). Rifampicin, rifabutin and rifapentine were the most active compounds against the ‘Wayne’ and ‘Loebel’ bacilli. Aminoglycosides, fluoroquinolones, capreomycin, clofazimine, metronidazole, thioridazine and mefloquine showed moderate activity against hypoxic non-growing Mtb cells. Bedaquiline showed good WCC₉₀ activity, followed by PA-824. Most of the WCC₉₀ and LCC₉₀ data obtained in the present study are comparable to published results. A cytotoxicity (CC₅₀) measurement was carried out using two common cell lines, namely BHK-21 (kidney cell line) and HepG2 (liver cell line). Most compounds tested were non-cytotoxic up to a maximum concentration tested (2.5–29 mg/L), with CC₅₀ to MIC₅₀ ratios of >7–800. Mefloquine was the only exception, with CC₅₀ values of 4.2 and 2.8 mg/L and an MIC₅₀ of 3.78 mg/L (Table 3). It is important to note here that many anti-TB agents are pro-drugs that are primarily activated inside Mtb. Cytotoxicity assays in human cell lines do not capture the potential toxicity of these active metabolites if bioconversion does not take place within host cells.

In vitro PK

In vitro solubility, permeability and liver microsomal metabolic stability are key predictors of oral bioavailability. These properties were determined using standardized assays and are summarized in Table 4. Ninety percent of the compounds showed moderate-to-high solubility, except clofazimine, PA-824 and bedaquiline, which were poorly soluble. For predicting permeability, PAMPA is employed as a simple in vitro model of passive diffusion through an artificial membrane. The Caco-2 cell layer permeability assay is widely used as a more predictive in vitro model of absorption through the intestinal epithelium. While PAMPA only measures passive diffusion, the Caco-2 assay integrates active uptake, efflux and permeation via the paracellular route. Most of the compounds (62%) exhibited low PAMPA permeability while 90% of the compounds showed moderate to high Caco-2 permeability and no or low active efflux except for rifampicin, thioacetazone and clarithromycin. Exceptions were streptomycin and clofazimine displaying low Caco-2 permeability. More than 90% of the compounds displayed low-to-medium intrinsic clearance in both mouse and human liver microsomes. There was a good correlation between the two species, except for ethionamide and prothionamide, with high and medium clearance in mouse and human microsomes, respectively (Table 4).

Table 4.

In vitro PK properties and in vivo PK–PD parameters

Compound	Solubility (mg/mL)	PAMPA (Papp 10−6 cm/s)	Caco-2 (A-B) (Papp 10−6 cm/s)	Caco-2 (B-A) (Papp 10−6 cm/s)	Efflux (B–A)/(A–B)	Mouse CLint (μL/min/mg)	Human CLint (μL/min/mg)	fC  max/MIC	fAUC/MIC	fT>MIC (%)
Isoniazid	4.93	ND	ND	ND	—	21.7	13.9	408.9	756.9	62
Rifampicin	1.79	1.6	1.52	18.4	12.1	3.04	2.84	89.9	1187.5	100
Pyrazinamide	2.72	1.73	70.4	37.1	0.5	low	low	8.6	17.9	15
Ethambutol	0.79	ND	ND	ND	—	low	low	8.1	24.7	25
Streptomycin	2.04	ND	0.78	0.57	0.7	9.76	low	—	—	—
Kanamycin	ND	6.32	ND	ND	—	14.6	2.97	—	—	—
Amikacin	2.94	ND	ND	ND	—	32.8	16.8	—	—	—
p-Aminosalicylic acid	0.97	ND	ND	ND	—	low	41.9	—	—	—
Cycloserine	2.94	ND	ND	ND	—	ND	ND	100.1	98.0	20
Ethionamide	0.70	14.6	57.3	29.6	0.5	210	77.1	1.2	0.4	0.6
Rifabutin	0.21	63.2	12.4	17.4	1.4	16.2	27.4	—	—	—
Rifapentine	0.11	19.3	7.68	15	1.95	9.1	11.2	—	—	—
Moxifloxacin	2.76	0.22	17.2	29.1	1.7	low	0.78	16.5	85.7	70
Levofloxacin	3.1	0.02	8.23	17.3	2.1	low	4.32	—	—	—
Gatifloxacin	2.8	0.06	8.65	19.4	2.2	0.94	4.01	—	—	—
Ciprofloxacin	—	—	—	—	—	—	—	—	—	—
Ofloxacin	2.11	0.02	9.25	20.1	2.2	2.77	5.8	—	—	—
Sparfloxacin	0.99	0.59	28.8	31.2	1.1	0.14	2.63	—	—	—
Capreomycin	ND	0.09	ND	ND		8.24	0.80	—	—	—
Thioacetazone	0.16	ND	3.87	57.2	14.8	5.78	4.75	—	—	—
Linezolid	1.97	0.05	18.1	31.2	1.7	2.98	0.89	87.9	99.3	19
Prothionamide	0.26	52	58.5	27.8	0.5	264	56	—	—	—
Clarithromycin	0.57	19.5	1.31	54.3	41.4	7.82	41.1	—	—	—
Amoxicillin	2.22	ND	ND	ND	—	low	6.48	—	—	—
Clavulanate	>5	0.46	37.5	21.4	0.6	3.8	1.46	—	—	—
Meropenem	>5	ND	ND	ND	—	ND	ND	—	—	—
Clofazimine	<0.005	2.58	0.17	0.18	1.0	low	low	0.02	0.6	0
Metronidazole	1.34	0.59	39.8	27.6	0.7	3.98	low	—	—	—
Thioridazine	2.57	91	3.8	5.0	1.3	107	44.6	—	—	—
Mefloquine	—	—	—	—	—	—	—	0.1	1.6	0
Vancomycin	—	—	—	—	—	—	—	—	—	—
Valnemulin	>5	ND	ND	ND	—	low	0.62	0.3	1.5	0
PNU–100480	0.45	92.8	5.54	5.9	1.1	0.91	1.41	—	—	—
PA-824	0.03	16.8	47.6	30.2	0.6	1.3	5.78	6.0	101.2	52
Bedaquiline	<0.005	ND	ND	ND	—	low	2.33	0.03	0.5	0
Delamanid	—	—	—	—	—	—	—	—	—	—
Lowa	10% (3)	62% (13)	10% (2)			90% (27)	93.3% (28)
Moderatea	33% (10)	—	19% (4)			3.3% (1)	6.7% (2)
Higha	47% (14)	38% (8)	71% (15)			6.7% (2)	—
Very higha	10% (3)	—	—			—	—

—, Not tested; ND, not detected or data not available due to analytical/technical issues.

Numbers in parentheses indicate the number of compounds in each group.

^aCompounds were grouped into various categories based on the following cut-offs. Solubility: low, <0.05 mg/mL; moderate, 0.05–1 mg/mL; high, 1–5 mg/mL; and very high, >5 mg/mL. PAMPA: low, <10 P_app 10⁻⁶ cm/s; and high, >10 P_app 10⁻⁶ cm/s. Caco-2: low, <1 P_app 10⁻⁶ cm/s; moderate, 1–5 P_app 10⁻⁶ cm/s; and high, >5 P_app 10⁻⁶ cm/s. CL_int (intrinsic clearance): low, <50 μL/min/mg of protein; moderate, 50–150 μL/min/mg of protein; and high >150 μL/min/mg of protein.

Correlations between in silico, in vitro and in vivo parameters

As expected, some level of correlation was observed between in vitro PK and physicochemical properties. Generally, compounds with MWs <400 and cLogP of <4 showed better solubility and Caco-2 permeability (Figure 3a–d). This is in agreement with Gleeson's findings. Most of the outliers were natural products either belonging to the rifamycin class or aminoglycoside injectables. There was a significant negative correlation between solubility and cLogP (r_s = −0.5824, P = 0.0007) (Figure 3b). Further, examination of clinical PK parameters retrieved from the literature (Supplementary Data) revealed a moderate-to-high oral bioavailability for compounds that follow Lipinski's rule of five and Veber's rule (Figure S2A–E),^5,6 as well as compounds with a solubility of >1 mg/mL and Caco-2 of >5 × 10⁻⁶ cm/s (Supplementary Data). In addition, a significant positive correlation was observed between (i) cLogP and volume of distribution, an indicator of drug distribution into tissues (Figure 4a), (ii) cLogP and systemic clearance of the unbound fraction (Figure 4b) and (iii) cLogP and plasma protein binding (Figure 4c). A significant negative correlation was observed for cLogP and oral bioavailability (Figure 4d). As expected, lipophilicity emerged as an important parameter influencing in vitro and in vivo PK parameters. Mtb possesses a notoriously impermeable cell wall. Thus, we tested the hypothesis that lipophilicity and permeability could be a determinant of in vitro potency. Across all drug classes included in this study, no significant correlation was observed between antimycobacterial activity and cLogP, PAMPA or Caco-2 permeability, indicating that permeability and/or lipophilicity is not a predictor of activity inside the mycobacterium, at least not across this large set of compound classes. Examination of potential correlations within subgroups of compounds (natural products, synthetic molecules, injectables and oral drugs) did not reveal any apparent association either. Within a given series, however, a positive correlation is often detected between cLogP and potency, as recently published for indolcarboxamide and tetrahydropyrazolopyrimidine carboxamide series.^35,36

Figure 3.

In silico–in vitro correlations. (a) Solubility versus MW. (b) Solubility versus cLogP. (c) Caco-2 versus MW. (d) Caco-2 versus cLogP. The vertical lines indicate cut-off values as defined in Gleeson's analysis of MW and cLogP. The horizontal lines delineate the boundaries between the low, moderate and high categories for solubility (a and b) and Caco-2 permeability (c and d). The compounds are categorized as synthetic molecules (open squares) and natural products (filled squares) based on their origin (Table 1).

Figure 4.

Correlation between cLogP and volume of distribution (a), unbound clearance (CL_u), (b) plasma protein binding (c) and oral bioavailability (d). A significant positive correlation was observed between cLogP and volume of distribution, cLogP and CL_u (ratio of clearance and unbound fraction⁴⁴), and cLogP and plasma protein binding. A significant negative correlation was observed between cLogP and oral bioavailability. The compounds are categorized as synthetic molecules (open squares) and natural products (filled squares) based on their origin (Table 1). r_s is the Spearman's rank correlation coefficient and P is the level of significance.

Correlations between in vivo PK and in vivo efficacy

To examine the PK–PD relationships of anti-TB compounds, mouse in vivo PK and efficacy data were either retrieved from the literature or generated as part of this study (Supplementary Data). Doses were selected based on the efficacious exposure observed in humans. Based on the MIC₅₀ and PK parameters, plasma PK–PD indices (fC_max/MIC, fAUC/MIC and fT_>MIC) were calculated and corrected for plasma protein binding to reflect the free drug fraction (indicated by ‘f’) (Table 4)_. Overall, unbound C_max values were above the MIC at some point during the dosing interval, except for compounds with high plasma protein binding such as mefloquine, clofazimine and bedaquiline (Table 4 and Supplementary Data). Except for PA-824—for which efficacy is driven by the fraction of the dosing interval during which free plasma concentration exceeds the MIC or fT_>MIC—the efficacy of all the compounds tested was concentration dependent, i.e. either fAUC/MIC and/or C_max/MIC best correlated cfu reduction in mouse lungs.^37–41 However, no clear trend was observed between PK–PD indices and mean log cfu reduction across compounds or compound classes. This is an important observation and clearly shows that simple PK–PD rules of thumb cannot be used to guide the discovery and development of novel TB drugs.¹³ This feature is specific to TB drugs and is in stark contrast to many other antibacterials for which plasma PK–PD indices usually fall within relatively narrow and reasonably predictive windows.⁴² The complexity of the pathology of TB, where the pathogen resides in remote lesion compartments, is likely to contribute to the lack of correlation between plasma PK and efficacy.⁴³

Conclusions

Drug discovery aims to deliver a candidate that shows efficacy, exposure and tolerability in a relevant animal model and in man. In silico and in vitro assays are employed in lead finding and optimization, attempting to predict these in vivo properties. In the present study, we have comprehensively determined and compiled physicochemical parameters, in vitro PK properties and potency values for 36 antimycobacterials, most of which are in clinical use against TB. Attempts to detect in vitro–in vivo correlations between physicochemical, PK and efficacy parameters met with limited success, once again highlighting the challenges of anti-TB drug discovery. We hope that this standardized dataset represents a useful reference for the TB drug discovery community.

Funding

This work was funded and supported by the Novartis Institute for Tropical Diseases (NITD), Singapore.

Transparency declarations

None to declare.

Author contributions

S. B. L. and S. P. S. R. analysed in silico properties. T. B. H. and S. P. S. R. performed/analysed mycobacterial cell-based assays. S. B. L. and S. P. S. R. analysed in vitro PK data. S. B. L. and V. D. analysed in vivo PK data. U. H. M. and S. P. S. R. performed/analysed mouse efficacy studies. S. B. L., P. C. H., U. H. M., V. D., T. D. and S. P. S. R. supervised and directed the work. S. B. L. and S. P. S. R. conceptualized and carried out correlation studies. S. B. L., T. D., V. D. and S. P. S. R. wrote the manuscript. All authors discussed the results and commented on the manuscript.