Anti-Mycobacterial Evaluation of 7-Chloro-4-Aminoquinolines and Hologram Quantitative Structure–Activity Relationship (HQSAR) Modeling of Amino–Imino Tautomers

In an ongoing research program for the development of new anti-tuberculosis drugs, we synthesized three series (A, B, and C) of 7-chloro-4-aminoquinolines, which were evaluated in vitro against Mycobacterium tuberculosis (MTB). Now, we report the anti-MTB and cytotoxicity evaluations of a new series, D (D01–D21). Considering the active compounds of series A (A01–A13), B (B01–B13), C (C01–C07), and D (D01–D09), we compose a data set of 42 compounds and carried out hologram quantitative structure–activity relationship (HQSAR) analysis. The amino–imino tautomerism of the 4-aminoquinoline moiety was considered using both amino (I) and imino (II) forms as independent datasets. The best HQSAR model from each dataset was internally validated and both models showed significant statistical indexes. Tautomer I model: leave-one-out (LOO) cross-validated correlation coefficient (q2) = 0.80, squared correlation coefficient (r2) = 0.97, standard error (SE) = 0.12, cross-validated standard error (SEcv) = 0.32. Tautomer II model: q2 = 0.77, r2 = 0.98, SE = 0.10, SEcv = 0.35. Both models were externally validated by predicting the activity values of the corresponding test set, and the tautomer II model, which showed the best external prediction performance, was used to predict the biological activity responses of the compounds that were not evaluated in the anti-MTB trials due to poor solubility, pointing out D21 for further solubility studies to attempt to determine its actual biological activity.


Introduction
Tuberculosis (TB), a severe infectious bacterial disease caused by Mycobacterium tuberculosis (MTB), infected almost 9.6 million of people around the world and caused 1.5 million deaths in 2014 [1]. In the last five years, TB drug-resistant cases have increased dramatically; the number of multidrug-resistant (MDR) TB (i.e., resistance to at least both isoniazid and rifampicin) reported cases has almost doubled (250,000 to 440,000), while extensively drug-resistant (XDR) TB (i.e., resistance to any fluoroquinolone, and at least one of three second-line injectable drugs, i.e., capreomycin, kanamycin, and amikacin, in addition to MDR) was identified in 105 countries in 2015 [1].
Considering the alarming global situation of TB, as well as the high impact of MDR/XDR-TB on TB treatment, there is an urgent need to develop new anti-TB drugs. In this context, the quinoline nucleus, which occurs in natural products, besides being used in the design of bioactive compounds, is considered a "privileged structure" [2], showing a variety of biological activities, e.g., anticancer [3], antibacterial [4], antimalarial [5], antileishmanial [6], and anti-viral (e.g., anti-human immunodeficiency virus, HIV) [7].
In relation to anti-MTB activity, in December 2012, the United States Food and Drug Administration (U.S. FDA) approved a diarylquinoline derivative, namely bedaquiline (Sirturo®by Janssen/Johnson & Johnson), showing a new mechanism of action [8]. Bedaquiline has shown activity against MTB with minimum inhibitory concentration (MIC) values for the MTB wild-type H37Rv strain (MIC ≤ 0.1 µM) as well as MDR strains (MIC ≤ 0.5 µM) [9].

HQSAR (2D-QSAR) Modeling
In order to rationalize the structure-activity relationship (SAR) of the active derivatives from series D (D01-D09) and to guide future structural modifications, these compounds and 33 more from series A (A01-A13) [11], B (B01-B13) [10], and C (C01-C07) [12] were used to construct robust HQSAR models, considering both amino (I) and imino (II) tautomers as independent datasets (Table  2), since this potential amino-imino tautomerism could be important in the microenvironment of the target receptor. The 42 compounds divided into training set (

HQSAR (2D-QSAR) Modeling
In order to rationalize the structure-activity relationship (SAR) of the active derivatives from series D (D01-D09) and to guide future structural modifications, these compounds and 33 more from series A (A01-A13) [11], B (B01-B13) [10], and C (C01-C07) [12] were used to construct robust HQSAR models, considering both amino (I) and imino (II) tautomers as independent datasets (Table 2), since this potential amino-imino tautomerism could be important in the microenvironment of the target receptor. The 42 compounds divided into training set (32 compounds) and test set (10 compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values.
Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment Pharmaceuticals 2017, 10, 52 5 of 16 distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) cross-validated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SE cv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SE cv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SE cv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SE cv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SE cv = 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. Table 2. Chemical structures (tautomers I and II, R substituent) and biological activity values (pMIC) of the 7-chloro-4-aminoquinoline derivatives for series A-D. minimum inhibitory concentration) values.
Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes. compounds) present a regular distribution over the entire range of experimental pMIC (-Log of the minimum inhibitory concentration) values. Initially, the 12 default hologram lengths (HL) were tested, aiming to reduce the chance of bad collisions. Then, the default fragment size (FS) (4-7 atoms) was used, setting the maximum number of components (NC) to fifteen, which is smaller than half the number of compounds used in the training set (32 compounds). Finally, the HQSAR runs were performed varying the six types of fragment distinction (FD) parameter (i.e., A = atoms, B = bonds, C = connections, Ch = chirality, H = hydrogen atoms, DA = hydrogen bond donor/acceptor atoms), obtaining twelve different models ( Table 3).
The three most predictive HQSAR models from this first optimization step of the tautomer I dataset (models A/C, A/C/H, and C/DA, Table 3) showed comparable leave-one-out (LOO) crossvalidated r 2 (q 2 ) values. In the first model, the atoms (A) and connections (C) FD parameters are combined: A/C (q 2 = 0.80; cross-validated standard error, SEcv = 0.32). The second model combines the atoms (A), connections (C), and hydrogen atoms (H) FD parameters: A/C/H (q 2 = 0.79; SEcv = 0.33). In the third model, the connections (C) and donor/acceptor atoms (DA) FD parameters are combined: C/DA (q 2 = 0.77; SEcv = 0.39). However, the A/C model was selected because of the high q 2 value, the low SEcv value, and the lower NC and HL. For the tautomer II data set, the most predictive model obtained showed the same combination of FD parameters: atoms and connections (A/C, q 2 = 0.77; SEcv= 0.35).
Aiming to improve the A/C models previously described for both tautomers I and II datasets, eight new models were generated considering FS starting from 2 to 5 until reaching 9 to 12 atoms per fragment, varying each fragment by four units. According to Table 4, the variation of this parameter did not improve the q 2 value, which proved that the default FS (4-7 atoms) led to better statistical results compared to other fragment sizes.                             In order to analyze the robustness of the obtained models, the Y-randomization test was performed ten times. In this test, the dependent variable (i.e., Y column, the biological activity data) was randomized and the HQSAR method was carried out for each training set (I and II). According to Table 5, the 10 randomized biological datasets are not correlated with the order of the original dataset (simple correlation coefficient, R, values close to zero); among the 10 runs, only 4 generated predictive models with low q 2 and r 2 values, indicating that an acceptable model was obtained for the original data set (i.e., non-randomized). Table 3. Summary of hologram quantitative structure-activity relationship (HQSAR) statistical indexes for various fragment distinction (FD) parameters using the default fragment size (FS = 4-7 atoms) for both tautomers I (amino) and II (imino) data sets.    Table 4. Summary of hologram quantitative structure-activity relationship (HQSAR) statistical indexes for the influence of various fragment sizes (FS) using the fragment distinction (FD) parameters atoms and connections (A/C) for both tautomers I (amino) and II (imino) data sets. leave-one-out (LOO) cross-validated r 2 ; r 2 : squared correlation coefficient; SE cv : cross-validated standard error; SE: standard error; NC: optimal number of components; HL: hologram length. Table 5. Values of R, q 2 , and r 2 of ten HQSAR runs generated for each randomized dataset (tautomers I and II, N = 32) using the Y-randomization test. −0.06 n.p. n.p. n.p. n.p. 10 0.02 n.p. n.p. n.p. n.p. a R: simple correlation coefficient between the random and original orders of the experimental pMIC values. b q 2 : leave-one-out (LOO) cross-validated r 2 ; r 2 : squared correlation coefficient; n.p.: no predictive model. After the internal validation (i.e., leave-one-out cross-validation, LOO-cv) of the models derived from the I and II data sets, the ability of these models to predict the activity values of new compounds (i.e., the test set) was measured. Therefore, the external validation of the HQSAR models was performed by predicting the biological potencies of the test set compounds using the best model derived from each training set. The predictive ability of the HQSAR models is expressed with predictive r 2 (r 2 Pred ) values, which is similar to the cross-validated r 2 (q 2 ) and is defined by Equation (1) (see HQSAR Modeling in the Materials and Methods section).

FS
The experimental pMIC (pMIC Exp ) and calculated pMIC (pMIC Calc ) values and residuals (pMIC Exp -pMIC Calc , respectively) of the training and test sets compounds obtained by the best HQSAR model of each data set are reported in Table 6, and the comparison plots between the experimental and calculated potencies of both the training and test sets are shown in Figure 2.  The limitation of the predictive ability of the best models for the tautomers I and II datasets was determined by the presence of outliers, i.e., compounds of the training and test sets showing residual values exceeding twice the standard deviation (SD) of estimate of the model (tautomer I, 2*SD = 0.47; tautomer II, 2*SD = 0.38) [17,18,27].
According to this criterion, there are four outliers for the tautomer I model (A01, B02, C02, and D01) and only two outliers for the tautomer II model (A01 and D09), all of them belonging to the test set (compounds marked with an asterisk in Table 6), but not exceeding one log unit (Table 6 and Figure 2). Although the tautomer I model (amino form) shows better statistical parameters for internal prediction (training set), it owns poor reliability on the external prediction (test set). Therefore, only the HQSAR tautomer II model (imino form) was used for the analysis of the HQSAR contribution maps to the biological activity of this dataset, as well for the prediction of the biological activity of the A10-A21 derivatives.
The SAR of the HQSAR models can be graphically represented by the contribution maps, where each compound can be colored using a standard color code that discriminates the individual atomic and connection contributions to the biological activity response: yellow, blue-green, and green reflect positive contribution; orange, orange-red, and red reflect negative contribution; and white reflects The limitation of the predictive ability of the best models for the tautomers I and II datasets was determined by the presence of outliers, i.e., compounds of the training and test sets showing residual values exceeding twice the standard deviation (SD) of estimate of the model (tautomer I, 2 * SD = 0.47; tautomer II, 2 * SD = 0.38) [17,18,27].
According to this criterion, there are four outliers for the tautomer I model (A01, B02, C02, and D01) and only two outliers for the tautomer II model (A01 and D09), all of them belonging to the test set (compounds marked with an asterisk in Table 6), but not exceeding one log unit (Table 6 and Figure 2). Although the tautomer I model (amino form) shows better statistical parameters for internal prediction (training set), it owns poor reliability on the external prediction (test set). Therefore, only the HQSAR tautomer II model (imino form) was used for the analysis of the HQSAR contribution maps to the biological activity of this dataset, as well for the prediction of the biological activity of the A10-A21 derivatives.
The SAR of the HQSAR models can be graphically represented by the contribution maps, where each compound can be colored using a standard color code that discriminates the individual atomic and connection contributions to the biological activity response: yellow, blue-green, and green reflect positive contribution; orange, orange-red, and red reflect negative contribution; and white reflects neutral contribution. Therefore, in order to understand the SAR of the HQSAR tautomer II model, the contribution maps of the most (A08, B05, C07, and D06) and least (A09, B03, C03, and D03) potent anti-MTB derivatives from each series (A, B, C, and D) are presented in Figure 3. In the N-alkyl series (A01-A13), the contribution maps of A08 (most potent) and A09 (least potent) (Figure 3) show the presence of a long aliphatic chain in A08, as compared to a short chain in A09, which is related to a larger number of fragments increasing the potency. This effect may be related to a greater lipophilicity conferred by the longer aliphatic chain. Consequently, the increase in the number of methylene groups of the aliphatic chain has a positive contribution to the overall biological activity.
In the monosubstituted phenyl-hydrazone series (B01-B13), the contribution maps of B05 (most potent) and B03 (least potent) (Figure 3) show the substitution of chlorine (B03) by bromine (B05) in the para position of the phenyl ring, which is related to a larger number of fragments in phenylhydrazone subunit, increasing the activity. Again, this effect may also be related to a greater lipophilicity conferred by bromine. However, at the same time, this substitution almost changed the entire character of the quinoline ring, which has a neutral contribution in B05.
Analogous to series B, in the polysubstituted phenyl-hydrazone series (D01-D21), the contribution maps of D06 (most potent) and D03 (least potent) (Figure 3) show the substitution of the hydroxyl (D03) by methoxy (D06) group at both the meta and para positions of the phenyl ring, which is related to a larger number of fragments increasing the activity, especially in the phenyl ring and C5/C6 atoms of the quinoline ring. As in series B, this effect should be related to a greater lipophilicity conferred by the methoxy group. In fact, the bromine and methoxy group increases the lipophilicity of B05 and D06, respectively, resulting in increased anti-MTB activity (Table 2).
Finally, in the heteroaromatic series (C01-C07), the contribution maps of C03 (most potent) and C07 (least potent) (Figure 3) show the replacement of the 2-pyridinyl by 5-nitro-2-thienyl group, which is related to a larger number of fragments increasing the activity, especially in both the heteroaromatic and quinoline rings. Again, this effect may be related to a greater lipophilicity conferred by the 5-nitro-2-thienyl group. On the other hand, heteroaromatic nitro compounds, such In the N-alkyl series (A01-A13), the contribution maps of A08 (most potent) and A09 (least potent) ( Figure 3) show the presence of a long aliphatic chain in A08, as compared to a short chain in A09, which is related to a larger number of fragments increasing the potency. This effect may be related to a greater lipophilicity conferred by the longer aliphatic chain. Consequently, the increase in the number of methylene groups of the aliphatic chain has a positive contribution to the overall biological activity.
In the monosubstituted phenyl-hydrazone series (B01-B13), the contribution maps of B05 (most potent) and B03 (least potent) (Figure 3) show the substitution of chlorine (B03) by bromine (B05) in the para position of the phenyl ring, which is related to a larger number of fragments in phenyl-hydrazone subunit, increasing the activity. Again, this effect may also be related to a greater lipophilicity conferred by bromine. However, at the same time, this substitution almost changed the entire character of the quinoline ring, which has a neutral contribution in B05.
Analogous to series B, in the polysubstituted phenyl-hydrazone series (D01-D21), the contribution maps of D06 (most potent) and D03 (least potent) (Figure 3) show the substitution of the hydroxyl (D03) by methoxy (D06) group at both the meta and para positions of the phenyl ring, which is related to a larger number of fragments increasing the activity, especially in the phenyl ring and C5/C6 atoms of the quinoline ring. As in series B, this effect should be related to a greater lipophilicity conferred by the methoxy group. In fact, the bromine and methoxy group increases the lipophilicity of B05 and D06, respectively, resulting in increased anti-MTB activity (Table 2).
Finally, in the heteroaromatic series (C01-C07), the contribution maps of C03 (most potent) and C07 (least potent) (Figure 3) show the replacement of the 2-pyridinyl by 5-nitro-2-thienyl group, which is related to a larger number of fragments increasing the activity, especially in both the heteroaromatic and quinoline rings. Again, this effect may be related to a greater lipophilicity conferred by the 5-nitro-2-thienyl group. On the other hand, heteroaromatic nitro compounds, such as C03, which is the most active of the dataset, are known for producing radical species that cause damage to bacterial DNA [28].
As previously stated, there are only two outliers in the HQSAR tautomer II model (Table 6 and Figure 2), and for both compounds, the pMIC values were underestimated (residuals of A01 = 0.68 and D09 = 0.53). The HQSAR model II underestimated the predictive pMIC value of outlier A01, besides, its experimental pMIC value is similar to those of compounds A02-A05 (Figure 4). This behavior could be attributed to the fact that the exchange of the chlorine atom by bulky amines could influence the permeability of these compounds through the cell wall.  In relation to outlier D09, which have three different substituents attached to the phenyl ring, there is no similar compound in the training set, however, the HQSAR models were obtained with compounds showing substituents in the same positions found in this outlier, e.g. D03, B01, and B12 ( Figure 4). When comparing the contribution maps of these compounds, the presence of a hydroxyl substituent increases the number of unfavorable fragments, contributing to the low estimated value.
The HQSAR model II was used to predict the biological activities (pMICCalc) of compounds with poor solubility in the culture media (D10-D21), where compounds with identical substituents (D10-D15) showed close pMICCalc values in comparison with those used in the training set (Table 7).
In the case of derivatives D16-D21, the HQSAR model II showed pMICCalc values which were not reliable due to the lack of compounds in the training test with substituents attached to the phenyl ring showing comparable chemical characteristics. Therefore, thermodynamic and kinetic studies are needed to determine their dissolution rates, according to the crystalline forms, to allow assessing their activities values [23,24]. This study would be especially interesting for compound D21 since the presence of the nitro group in its structure resulted in increasing favorable fragments and the predicted activity value was greater than that of the most active compound in the data set.  In relation to outlier D09, which have three different substituents attached to the phenyl ring, there is no similar compound in the training set, however, the HQSAR models were obtained with compounds showing substituents in the same positions found in this outlier, e.g. D03, B01, and B12 ( Figure 4). When comparing the contribution maps of these compounds, the presence of a hydroxyl substituent increases the number of unfavorable fragments, contributing to the low estimated value.
The HQSAR model II was used to predict the biological activities (pMIC Calc ) of compounds with poor solubility in the culture media (D10-D21), where compounds with identical substituents (D10-D15) showed close pMIC Calc values in comparison with those used in the training set (Table 7).
In the case of derivatives D16-D21, the HQSAR model II showed pMIC Calc values which were not reliable due to the lack of compounds in the training test with substituents attached to the phenyl ring showing comparable chemical characteristics. Therefore, thermodynamic and kinetic studies are needed to determine their dissolution rates, according to the crystalline forms, to allow assessing their activities values [23,24]. This study would be especially interesting for compound D21 since the presence of the nitro group in its structure resulted in increasing favorable fragments and the predicted activity value was greater than that of the most active compound in the data set. poor solubility in the culture media (D10-D21), where compounds with identical substituents (D10-D15) showed close pMICCalc values in comparison with those used in the training set (Table 7). In the case of derivatives D16-D21, the HQSAR model II showed pMICCalc values which were not reliable due to the lack of compounds in the training test with substituents attached to the phenyl ring showing comparable chemical characteristics. Therefore, thermodynamic and kinetic studies are needed to determine their dissolution rates, according to the crystalline forms, to allow assessing their activities values [23,24]. This study would be especially interesting for compound D21 since the presence of the nitro group in its structure resulted in increasing favorable fragments and the predicted activity value was greater than that of the most active compound in the data set. poor solubility in the culture media (D10-D21), where compounds with identical substituents (D10-D15) showed close pMICCalc values in comparison with those used in the training set (Table 7). In the case of derivatives D16-D21, the HQSAR model II showed pMICCalc values which were not reliable due to the lack of compounds in the training test with substituents attached to the phenyl ring showing comparable chemical characteristics. Therefore, thermodynamic and kinetic studies are needed to determine their dissolution rates, according to the crystalline forms, to allow assessing their activities values [23,24]. This study would be especially interesting for compound D21 since the presence of the nitro group in its structure resulted in increasing favorable fragments and the predicted activity value was greater than that of the most active compound in the data set. D15) showed close pMICCalc values in comparison with those used in the training set (Table 7). In the case of derivatives D16-D21, the HQSAR model II showed pMICCalc values which were not reliable due to the lack of compounds in the training test with substituents attached to the phenyl ring showing comparable chemical characteristics. Therefore, thermodynamic and kinetic studies are needed to determine their dissolution rates, according to the crystalline forms, to allow assessing their activities values [23,24]. This study would be especially interesting for compound D21 since the presence of the nitro group in its structure resulted in increasing favorable fragments and the predicted activity value was greater than that of the most active compound in the data set. D15) showed close pMICCalc values in comparison with those used in the training set (Table 7). In the case of derivatives D16-D21, the HQSAR model II showed pMICCalc values which were not reliable due to the lack of compounds in the training test with substituents attached to the phenyl ring showing comparable chemical characteristics. Therefore, thermodynamic and kinetic studies are needed to determine their dissolution rates, according to the crystalline forms, to allow assessing their activities values [23,24]. This study would be especially interesting for compound D21 since the presence of the nitro group in its structure resulted in increasing favorable fragments and the predicted activity value was greater than that of the most active compound in the data set.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Cytotoxicity Activity (Colorimetric Cell Viability Assay)
The cellular viability for a macrophage cell line J774 (ATCC ® TIB-67™) was determined by Mosmann's MTT microculture assay [25,26]. Macrophage viability was evaluated in the presence and absence of test compounds (D01-D09). The cells were plated in flat bottom 96-well plates (2.5 X 10 6 cells/well/100 µ L) cultured for 24 h in a controlled atmosphere (CO2 5% at 37 °C), and non-adherent cells were washed by gentle flushing with Roswell Park Memorial Institute (RPMI) 1640 medium, 4

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Cytotoxicity Activity (Colorimetric Cell Viability Assay)
The cellular viability for a macrophage cell line J774 (ATCC ® TIB-67™) was determined by Mosmann's MTT microculture assay [25,26]. Macrophage viability was evaluated in the presence and absence of test compounds (D01-D09). The cells were plated in flat bottom 96-well plates (2.5 X 10 6 cells/well/100 µ L) cultured for 24 h in a controlled atmosphere (CO2 5% at 37 °C), and non-adherent cells were washed by gentle flushing with Roswell Park Memorial Institute (RPMI) 1640 medium, 4

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µ L of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µ L of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µ g/mL. Plates were covered and sealed with parafilm and incubated at 37 °C for five days. After this time, 25 µ L of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Anti-Mycobacterial (Anti-MTB) Activity
The anti-MTB activities were tested using the microplate Alamar blue assay (MABA) [21,22]. Briefly, 200 µL of sterile deionized water was added to all outer-perimeter wells of sterile 96-well Falcon-3072 plates (Becton Dickinson, Lincoln Park, NJ, USA) to minimize evaporation of the medium in the test wells during incubation. The 96 plates received 100µL of the Middlebrook 7H9 broth (Difco Laboratories, Detroit, MI, USA), containing cultures of the Mycobacterium tuberculosis wild-type H37Rv strain susceptible to streptomycin, isoniazid, rifampin, and ethambutol (S.I.R.E.) drugs (American Type Culture Collection, ATCC ® 27294™), and a serial dilution of compounds D01-D21 was made directly on the plate. The final drug concentration ranged from 0.01 to 100 µg/mL. Plates were covered and sealed with parafilm and incubated at 37 • C for five days. After this time, 25 µL of a freshly prepared 1:1 mixture of Alamar blue (AccuMed International, Westlake, OH, USA) reagent and 10% Tween-80 was added to the plate and incubated for 24 h. A blue color in the well was interpreted as no bacterial growth, and a pink color was scored as growth. The minimal inhibitory concentration (MIC) was defined as the lowest drug concentration, which prevented a color change from blue to pink.

Cytotoxicity Activity (Colorimetric Cell Viability Assay)
The cellular viability for a macrophage cell line J774 (ATCC ® TIB-67™) was determined by Mosmann's MTT microculture assay [25,26]. Macrophage viability was evaluated in the presence and absence of test compounds (D01-D09). The cells were plated in flat bottom 96-well plates (2.5 × 10 6 cells/well/100 µL) cultured for 24 h in a controlled atmosphere (CO 2 5% at 37 • C), and non-adherent cells were washed by gentle flushing with Roswell Park Memorial Institute (RPMI)