Targeting dormant tuberculosis bacilli: Results for molecules with a novel pyrimidone scaffold

Article abstract of DOI:10.1111/cbdd.12373

Our inability to completely control TB has been due in part to the presence of dormant mycobacteria. This also renders drug regimens ineffective and is the prime cause of the appearance of drug-resistant strains. In continuation of our efforts to develop novel antitubercular agents that especially target dormant mycobacteria, a set of 55 new compounds belonging to the pyrimidone class were designed on the basis of CoMFA and CoMSIA studies, and these were synthesized and subsequently tested against both the dormant and virulent BCG strain of M. tuberculosis. Some novel compounds have been identified which selectively inhibit the dormant tuberculosis bacilli with significantly low IC₅₀ values. This study reports the second molecule after TMC-207, having the ability to inhibit tuberculosis bacilli exclusively in its dormant phase. The synthesis was accomplished by a modified multicomponent Biginelli reaction. A classification model was generated using the binary QSAR approach - recursive partitioning (RP) to identify structural characteristics related to the activity. Physicochemical, structural, topological, connectivity indices, and E-state key descriptors were used for generation of the decision tree. The decision tree could provide insights into structure-activity relationships that will guide the design of more potent inhibitors. This paper reports the second molecule after TMC-207, with the ability to inhibit tuberculosis bacilli in its dormant phase. The paper reports molecules with a novel Pyrimidone Scaffold, the synthesis of which was accomplished with a modified multi-component Biginelli reaction. A classification model was generated using recursive partitioning (RP) technique to identify structural characteristics of the molecules with their varying activities.

Full text of DOI:10.1111/cbdd.12373

Received Date : 12-Nov-2013
Revised Date : 31-May-2014
Accepted Date : 04-Jun-2014
Article type : Research Article
Targeting Dormant Tuberculosis Bacilli: Results for molecules with a novel Pyrimidone Scaffold
Rohit R. Joshia, Avinash Barchhaa, Vijay M. Khedkara, Raghuvir R. S. Pissurlenkara, Sampa Sarkarb,
b
c
c
d
a
†
Dhiman Sarkar , Rohini R. Joshi , Ramesh A. Joshi , Anamik K. Shah and Evans C. Coutinho
aDepartment of Pharmaceutical Chemistry, Bombay College of Pharmacy, Kalina, Santacruz (E), Mumbai
400 098.
bCombi Chem Bio Resource Centre, National Chemical Laboratory, Pune 411 008.
cOrganic Chemistry Division, National Chemical Laboratory, Pune 411 008.
dDepartment of Chemistry, Saurashtra University, Rajkot 360 005.
†
Corresponding Author
Tel. No.: +91-22-26670905
Fax. No.: +91-22-26670816
Email: evans@bcpindia.org
This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as an
'Accepted Article', doi: 10.1111/cbdd.12373
This article is protected by copyright. All rights reserved.

Keywords
Antitubercular activity; dormant tuberculosis bacilli; pyrimidones; recursive partitioning
Abstract
Our inability to completely control TB has been due in part to the presence of dormant mycobacteria.
This also renders drug regimens ineffective and is the prime cause of the appearance of drug resistant
strains. In continuation of our efforts to develop novel antitubercular agents that especially target
dormant mycobacteria, a set of fifty-five new compounds belonging to the pyrimidone class were
designed on the basis of CoMFA and CoMSIA studies, these were synthesized and subsequently tested
against both the dormant and virulent BCG strain of M. tuberculosis. Some novel compounds have been
identified which selectively inhibit the dormant tuberculosis bacilli with significantly low IC50 values.
This paper reports the second molecule after TMC-207, having the ability to inhibit tuberculosis bacilli
exclusively in its dormant phase. The synthesis was accomplished by a modified multi-component
Biginelli reaction. A classification model was generated using the binary QSAR approach - recursive
partitioning (RP) to identify structural characteristics related to the activity. Physicochemical, structural,
topological, connectivity indices and E-state key descriptors were used for generation of the decision tree.
The decision tree could provide insights into structure-activity relationships that will guide the design of
more potent inhibitors.
Introduction
Mycobacterium tuberculosis (MTB) infects about 32% of the world's population. M. tuberculosis has two
characteristic features that render it the deadliest disease: its high virulence and its ability to enter into a
dormant state with subsequent reactivation. Approximately ⅓ of the world population is infected with
This article is protected by copyright. All rights reserved.

latent TB caused by the dormant, non-replicating form of the mycobacterium(1). Although the
physiology of latent MTB is still not clear, a hypoxic condition has been found to trigger the dormant
state of mycobacterium. Normally, a small population of bacilli enters into the dormant stage in the
granuloma formed by immune cells and fibroblasts. Dormant bacilli are adapted to anaerobic conditions
and maintain their ability to resume growth and aggravate the disease by deterioration of the immune
system. This unique property also renders tolerance to conventional antimycobacterial drugs. It has been
now realized that the necessity of a minimum 6 months treatment for TB is due to the difficulty in
eradicating non-replicating dormant mycobacterium tuberculosis. Therefore there is a pressing need for
development of new drugs which are effective against MTB in both the active and dormant states.
The first molecule reported to possess selective activity against dormant MTB was metronidazole
which undergoes reduction of the nitro group to a reactive species in a hypoxic environment, that causes
DNA damage and subsequent cell death(2). Nitrofurans have also shown bactericidal activity against
active as well as dormant mycobacterium. However, this activity was observed at very high
concentrations of metronidazole or nitrofurans, rendering limited clinical value to the compounds. More
than 350 products of natural origin have been reported as anti-mycobacterial agents from 2003
onwards(3). However, most of these efforts were focused on inhibiting the growth of the actively
growing mycobacterium.
The current situation clearly shows that TB is far from under control and demonstrates the need
for a re-evaluation of our approach to treating TB. Drug development for tuberculosis and other
neglected diseases has been at a virtual standstill for decades, but increased awareness and advocacy in
recent years have led to new initiatives in TB drug development which includes a search for new drug
targets and new drugs. Recognizing these facts, we have initiated a program which includes design,
synthesis and screening of some new compounds that target dormant mycobacteria. Efforts to optimize
This article is protected by copyright. All rights reserved.

the structure and biological activity have resulted in several classes of compounds with potential
antitubercular activity (4-7). We had reported the discovery of ring-substituted quinolines as a promising
new structural class of anti-tuberculosis agents, which emerged from a broad structure-based screening
approach of new chemical entities against various pathogens including mycobacteria (8).
Currently we are focusing our efforts on developing an entirely new class of antitubercular agents
possibly acting on completely novel targets with a mechanism of action different from those of the
existing drugs. This effort has proven to be highly rewarding and has resulted in the identification of a
new class of tetrahydropyrimidines with potent antimycobacterial activity against both drug-sensitive and
dormant strains of mycobacteria with significantly low IC50 values. We report here the second molecule
after TMC-207, which has the ability to inhibit dormant tuberculosis bacilli.
Pyridines(4), pyrimidines and pyrimidones(5) have been identified with antitubercular activity
and bear a potential to be developed as novel structural classes of antitubercular agents (Figure 1). A
class of pyridines known as N-alkyl-1,2-dihydro-2-thioxo-3-pyridine carbothioamides (Figure 1, I) have
shown good antitubercular activity(9). N-pyridinylsalicylamides (Figure 1, II) which are related to the
pyridine class of compounds have also shown activity against MTB. The ramifications of various
substituents on the activity for this class of compounds were earlier studied by 3D-QSAR methods like
CoMFA and CoMSIA(5). Within the pyrimidine group, 2,4-diaminopyrimidines (Figure 1, III) have been
identified with IC50 of 5.8 nM and a safety index > 600(10). Thus, this structural class holds great hope
for development of new and effective antitubercular agents. Various 1,4-dihydropyridines (Figure 1, IV)
have been synthesized and evaluated as antitubercular drugs on the assumption that these compounds
would act as prodrugs, which after penetration into the cell would be converted into the 3,5-dicarboxylate
anion by enzymatic hydrolysis. The most effective analog has shown more than 97% inhibition on
mycobacteria at a concentration of 2.50 μg/ml. This series was examined in detail by QSAR analysis (11,
12). In order to further explore this structural class, we have modified the 1,4-dihydropyridines as
This article is protected by copyright. All rights reserved.

tetrahydropyrimidines (Figure 1, V),which can also be viewed as aza analogs of pyridines, with the hope
of developing a new class of antitubercular agents.
This paper describes the synthesis and biological evaluation of a series of substituted pyrimidone
derivatives as potential antitubercular agents especially against dormant mycobacteria. Further, to
classify these pyrimidone derivatives according to their activities, a recursive partitioning (RP) model was
developed. The 'Recursive Partitioning' (RP) method is a specific case of the 'binary QSAR' approach
that is capable of classifying class analog activity data by considering appropriate descriptors recursively.
Methods and Materials
Chemistry
The synthesis of N-phenyl-6-methyl-2-oxo-4-phenyl-1,2,3,4-tetrahydropyrimidine-5-carboxylates
(Scheme 1) was accomplished with a multi-component Biginelli reaction. Acid catalyzed one pot
cyclocondensation of various acetoacetanilides with the appropriate aromatic aldehyde and urea/thiourea
afforded the substituted tetrahydropyrimidines. Various acetoacetanilides were synthesized from
substituted anilines and ethyl/butyl acetoacetate. All molecules listed in Table 1 contain at least one
chiral centre in the pyrimidone ring with an equal preference for formation of both stereoisomers under
the conditions of synthesis and were not separated.
Melting points were determined using a BUCHI-5300 melting point apparatus and are
uncorrected. IR spectra were recorded on a PERKIN ELMER spectrophotometer (nujol mull method)
and only characteristic peaks are reported in cm-1. 1H NMR spectra were recorded in DMSO-d6, with
TMS as internal standard using a Bruker AC-300F NMR Spectrometer (300 MHz). Chemical shifts (δ)
are reported in parts per million (ppm) of the applied field. Reactions were monitored over silica gel-G
(Merck) TLC plates and spots were visualized by iodine vapor or by ultraviolet light (254 nm). Spectral
This article is protected by copyright. All rights reserved.

data (IR and 1H-NMR) of all the synthesized compounds were found to be in agreement with the proposed
structures.
General procedure for the synthesis of N-phenyl-6-methyl-2-oxo-4-phenyl-1,2,3,4-tetrahydropyrimidine-
5-carboxylates
An equimolar mixture (1:1) of substituted acetoacetanilide (3) (1 mol) and appropriate aldehyde (4) (1
mol) with urea or thiourea (1.5 mol) in absolute ethanol was heated under reflux for 4-8 hours in the
presence of a catalytic amount of conc. HCl /pTSA (para-toluene sulfonic acid). The reaction was
monitored with TLC. The precipitate (5) was filtered, dried and crystallized from the appropriate solvent.
Details about the characterization of these derivatives are provided in the supplementary material.
Antitubercular activity
Determination of growth and NR activity
Growth of M. bovis BCG in microplate format of dormancy model was measured by reading the
absorbance at 620nm as well as by determining the CFU/ml of the culture at different time intervals. The
lowest concentration of the drug yielding a differential absorbance (A620) of approximately zero or less
was defined as MIC. Nitro reductase (NR) activity of M. bovis BCG in microplate dormancy model was
measured by following the Griess reaction method described earlier (13). Briefly, 1% solution of
sulphanilic acid (in 20% HCl) and 0.1% of napthyl ethylene diamine hydrochloride solution (in distilled
water) was added to a whole cell culture in 1:1:1 ratio and the absorbance at 540 nm was read after 15
min of incubation.
This article is protected by copyright. All rights reserved.

Computational Studies
All computational studies were carried out on a Linux workstation with the CentOS Enterprise Linux 4.9
operating platform. The computations were executed with the molecular modeling software – Sybyl v7.1
from Tripos Inc. St. Louis, MO, USA (14) and Cerius2 v4.11 from Accelrys Inc, USA (15).
Recursive partitioning (RP) (16-20) is a simple, yet powerful statistical technique that seeks to decode the
elusive relationships in complex data sets involving thresholds, interactions and nonlinearities. These
factors impede an analysis that is based on assumptions of linearity such as multiple linear regression,
principal component analysis or partial least squares. RP can be used for the division of a dataset into
groups of higher and lower responses according to their appropriate descriptors. Further the decision tree
derived from the RP analysis can be used to qualitatively predict activities or activity classes in structure–
activity relationship analysis. It uses the best chemical features to split the large data set into smaller and
more homogeneous subsets. At each non-terminal node in the decision tree, molecules are split into two
groups by a particular descriptor. Each of the two subgroups is split further by another descriptor or
variable. To identify the best split for a particular node, the algorithm considers all possible binary splits
for each descriptor and selects the optimal one by some criterion. This recursive partitioning process is
terminated when further splitting is impossible or a threshold value of the termination rule is reached.
The tree display of the SAR, without considering complex statistical analysis, is clear, easy to interpret
and often suggestive. It is inherently fast when compared with other grouping techniques, such as
clustering and the 2D nature of this approach make it especially good for large datasets that are difficult to
sieve into usable divisions of classification.
The Cerius2 molecular modeling package was used to generate the 2D descriptors (Table 2). RP
analysis was carried out using the CSAR recursive partitioning method and the decision tree was
generated from the results of RP. The antitubercular activity of compounds was described in terms of
their IC50 values. It is necessary to classify the activity values into two classes, active (1) and less active
This article is protected by copyright. All rights reserved.

(0). Descriptors with zero variance as well as descriptors containing 95% of zero values were eliminated.
The correlation matrices were built for the descriptors and some descriptors were eliminated on the basis
of the correlation threshold of R = 0.5. The pruned set of descriptors was used as the independent
variables (X) while the biological activity formed the dependent variable (Y) for all analyses. The
CARTTM (classification and regression trees) method in Cerius2 was applied to generate the decision
tree model. The activity classes were weighted equally and the splits were scored using Twoing rule
scoring function. The pruning factor was varied between 0-3. Various values were used for maximum
tree depth (layers < 10) while the default values were set for maximum number of generic splits (30) and
the number of knots per variable (20). The internal validation for the RP model was performed using
cross-validation with the number of cross-validation groups set to 5.
Results and Discussion
Biological assay/anti-tubercular activity
As stated earlier all the synthesized compounds were obtained as a racemic mixture and were used as
such for anti-tubercular testing. These compounds were evaluated for their antitubercular activity in
accordance with protocols developed by Sarkar et al (21). They were also tested against dormant
mycobacteria by the nitrate reductase model (21). IC50 values were determined from the dose-response
curves and are provided in Table 1. Compound 14 (IC50 = 14.3 µg/ml) was found to be most active
against normally growing mycobacteria. Interestingly, compounds 13 and 14 showed good (IC50 = 6.3
µg/ml) activity against dormant tuberculosis bacteria by the nitrate reductase model.
A close appraisal of the molecules shows that R1 groups which are electron withdrawing and
positioned at C3 and C4 (5-8) decrease the activity of the compounds against aerobic bacteria (Table 1).
Similarly, with reference to R2, groups which are electron withdrawing and positioned at C3 and C4 (9
This article is protected by copyright. All rights reserved.

and 12) also decrease the antitubercular activity. However, interestingly when the groups R1 and R2 are
of electron withdrawing nature and simultaneously present at position 3 in their respective rings (13 and
14), the antimycobacterial activity increases and one of them i.e. compound 13 displays exclusive activity
against bacteria under anaerobic conditions. One probable explanation for the differential activity could
be that the intracellular environment of M.tuberculosis is different under aerobic and anaerobic condition.
It is possible that compound 13 is metabolized faster compared to compound 14 under aerobic conditions
consequently leading to early loss of activity. On the contrary compound 14 is relatively stable in this
environment and is therefore able to retain the anti-tubercular activity. Such a metabolic degradation may
not exist in the anaerobic environment and therefore no such variation in activity is observed.
Further, compounds with both R1 and R2 as electron withdrawing and with the former (R1)
stationed at positions 2 and 3 and the latter (R2) resident at position 3 on their respective rings, exhibit
activity against both aerobic as well as anaerobic bacteria. Compounds with R2 as an aliphatic substituent
for e.g. cyclohexyl group (20 and 35) also show medium activity but only when R1 at positions 2, 3 and 4
is electron withdrawing by nature.
Recursive Partitioning (RP) Analysis
Two independent classification models were developed to classify the pyrimidone derivatives into their
own activity class using the calculated descriptors (Table 2). For deriving the classification model, a
dataset of 20 molecules was selected from the complete set of synthesized molecules on the basis of
chemical and biological diversity using similarity search techniques viz. D-optimal design, Tanimoto
similarity coefficient and the Euclidian distance matrix criteria defined in Cerius2. The selection was
carried out such that the final set of molecules had structural diversity as well as a maximal distribution of
biological activities.
This article is protected by copyright. All rights reserved.

The first classification model is based on the activity exhibited by the molecules against the
dormant MTB while the second model is derived based on activity against virulent MTB. The RP models
were developed by varying parameters discussed in the experimental section so as to improve the
prediction for compounds correctly in their respective classes. This is defined by three terms- a
percentage of the total number of compounds observed to be in each class (Class % Observed Correct);
percent of the ratio of total number of compounds correctly classified to the number of compounds
predicted in the class (Overall % Predicted Correct) and the enrichment factor which for a specific class
is the ratio of the (Overall % Predicted Correct) to the original percentage of compounds belonging to
that class.
RP model for the activity against dormant MTB
Recursive partitioning splits the dataset into smaller, homogeneous subsets by deriving a binary
decision tree in which descriptor values are used as decision points. The dataset was divided into two
classes, inactives (0) and actives (1) where class 1 contains 15 inactive compounds having IC50 values
greater than 100μg/ml while the class 2 contains 5 active compounds with IC50 values less than 100μg/ml
against dormant MTB. The statistical results of the RP model and activity predictions are summarized in
tables 3 and 4 respectively.
In class 1, all 15 'inactives' compounds were correctly classified as class 1. Also for the class 2
i.e. the class of actives, all five molecules active against dormant bacteria were correctly predicted as
active, showing 100% class prediction. The number of true positives among the predictions in each
activity class is listed as the term 'Overall % Predicted Correct'. It is also noteworthy that for both
classes ('actives' and 'inactives') 'Overall % Predicted Correct' was observed to be 100%. The enrichment
factor (1.33 for class 1 and 4.00 for class 2 respectively) also suggest that the final RP model is
statistically significant and can ably classify any new set of molecules.
This article is protected by copyright. All rights reserved.

A 6-leaf recursive partitioning decision tree was obtained for the pyrimidone analogs based on the
2D descriptors (Figure 2). The model contains 6 terminal and 5 non-terminal nodes where a true
response to any given split follows the branch to the downside while a false response follow the branch to
the upside in a decision tree.
The first primary split is observed on the radius of gyration which is a spatial descriptor related to
the size of the molecule. The terminal node 1 contains two molecules (16 and 39) with radius of gyration
greater than 4.36 which is shown in class 2 (active) while the remaining 18 molecules having the values
above 4.36 formed the non-terminal node. These molecules were further split on the basis of molecular
volume (Vm) which is a 3D spatial descriptor that defines the molecular volume inside the contact
surface. It is calculated as a function of conformation and is related to the binding and transport
properties of a molecule. The terminal node 2 which is the active node contains one molecule (42) with
molecular volume greater than 339.13 and has been correctly assigned as active by the RP model.
The third split was observed on molecular surface area which is again 3D spatial descriptor that
describes the van der Waals area of a molecule. It determines the extent to which a molecule exposes
itself to the external environment. It is also related to the binding, solubility and transport properties of a
molecule. The terminal node 3 contains ten molecules (12, 20, 43, 44, 46, 48, 51, 54, 55 and 35) with
area greater than 394.54 which are correctly classified as in class 1 (inactive) while the remaining 7
compounds with area greater than 394.54 are placed on the non-terminal node. These were further split
on the basis of density which is a 3D spatial descriptor calculated as the ratio of molecular weight to its
volume. It reflects the types of atoms and how tightly they are packed in a molecule and therefore it can
be related to transport and melt behavior of a molecule. The terminal node 6 which is the inactive node
contains four molecules (7, 22, 26 and 34) with density less than 1.23 and have been correctly classified
as class 1 i.e. inactives. Remaining 3 molecules in the dataset formed the non-terminal node which were
This article is protected by copyright. All rights reserved.

then split on the basis of Principal moment of inertia at a cut off value of 1383.28. Terminal node 4
which is the active node contains two correctly classified active molecules (13 and 14) with PMI greater
than 1383.28 while the terminal node 5 was assigned molecule 49 with PMI less than the cut-off value
and has been correctly classified as inactive by the classification model.
To avoid over fitting and to improve generalization of the classification model, this model was
internally validated using the technique of 5-fold cross-validation. A set of 5 molecules are extracted
each time from the whole dataset and a new model is generated using the reduced dataset which is then
employed to predict the activity of the excluded molecules. The procedure is repeated iteratively until all
molecules have been omitted and predicted once, so that the statistics can be derived from a comparison
of the predicted data with the observed data. Approximately four-fifths of the compounds in the dataset
formed the training set that was used to derive the models and the remaining compounds shaped the test
set. The statistical results of cross-validation are summarized in table 5. An acceptable percent
classification was achieved for the training set and the predictivity of the test set was also good. This
reliability of the predictions for the molecules in the test set suggests that the model can be used with
confidence for predictions of unknown but related molecules. The key descriptors determined in the trial
set were consistent with those from the total set. The parameters used to build the RP model suggest that
the model is well able to classify and predict the activity class of new candidates.
RP model for the activity against virulent MTB
The same dataset of 20 molecules identified in the previous study was again used to build the
classification model to identify the features governing activity against virulent tuberculosis bacteria. The
dataset was divided into two classes, inactives (0) and actives (1). Class 1 contains 8 inactive compounds
with the IC50 values greater than 100μg/ml while class 2 contains 12 active compounds with IC50 values
This article is protected by copyright. All rights reserved.

less than 100μg/ml against the virulent tuberculosis bacteria. The statistical results of the RP model and
activity predictions from it are summarized in tables 6 and 7 respectively.
Analysis of the decision tree shows that in class 1, i.e. the class of 'inactives', 8 out of 8
compounds are correctly classified as belonging to class 1, while in class 2 which is the class of 'actives',
12 out of 12 compounds are correctly classified as well. It is noteworthy that for both classes ('actives'
and 'inactives') the 'Overall % Predicted Correct' is observed to be 100%. The enrichment factor (2.50
for class 1 and 1.67 for class 2 respectively) suggest that the final RP model is statistically significant and
can be confidently used for classifying a new library of compounds.
Figure 3 displays the optimized 6-leaf recursive partitioning decision tree generated using the 2D
descriptors. The decision tree consists of six terminal and five non-terminal nodes. The class 1 i.e. the
class of inactives has been plotted in red while Class 2 which is the class of actives has been plotted with
green color in the decision tree. The terminal nodes 2, 4 and 5 are associated with class 2 while the
terminal nodes 1, 3 and 6 are linked to class 1. The descriptors- dipole moment, principle moment of
inertia (PMI), AlogP98, radius of gyration and sum of atomic polarizability (Apol) form the decision
points in the decision tree (Table 7).
The first primary split is observed on the dipole moment which is an electronic descriptor
reflecting the strength and orientation behavior of a molecule in an electrostatic field. Dipole moments
have been found to be correlated to long range ligand–receptor recognition and receptor binding. The
dataset of 20 molecules was split into two branches- those molecules with dipole moment less than 4.42
follows a branch to the downside while those with dipole moment above the cut-off follows a branch to
the upside. Ten molecules (48, 26, 22, 14, 16, 20, 46, 54, 34 and 35) with dipole moment greater than
4.42 are further split on the basis of AlogP98 at a cut off mark of 4.05. LogP reflects the hydrophobic
character of a molecule. In this atom-based approach, each atom of the molecule is assigned to a
This article is protected by copyright. All rights reserved.

particular class with additive contributions to the total value of logP. Terminal node 1 contains one
molecule (46) with AlogP98 greater than 4.05 which is correctly classified as inactive by the RP model
while the terminal node 2 which is the active node contains nine correctly classified molecules (14, 16,
20, 22, 26, 34, 35, 48 and 54) having logP less than 4.05.
Switching to the first primary split, ten molecules (51, 49, 13, 43, 44, 42, 12, 39, 7 and 55) with
dipole moment values less than 4.42 follows a branch downside. These molecules were further split
based on principal moment of inertia which is a spatial descriptor. The terminal node 6 contains 5
molecules with PMI less than 1519.07 (49, 13, 44, 12 and 7) correctly classified as inactives while the
remaining five molecules (51, 43, 42, 39 and 55) with PMI above the cut off form the non-terminal node.
These molecules are again split on the basis of radius of gyration, another spatial descriptor related to the
size of the molecule. Molecules with radius of gyration less than 4.14 form the terminal node 5 (42 and
55) which are shown in class 2 while those having the value above 4.14 form the non-terminal node.
These molecules are then split on the basis of sum of atomic polarizability (Apol) with a cut off value of
15390.03. The terminal node 3 contains two molecules (43 and 51) with Apol greater than 15390.03
correctly classified as inactives while molecule 39 with Apol less than 15390.03 forms the terminal node
4 which is the active node.
A 5-fold cross-validation was performed to gauge the stability and the statistical significance of
the RP model. The statistical results for the same are summarized in Table 8. An acceptable
classification percentage observed for the classes signifies the stability of this model for external
predictability. Also the key descriptors identified in trial set are consistent with those observed in the
whole set. The statistical parameters indicate that this RP model is valid for classifying and predicting the
activity class of new candidate molecules that target tubercular bacteria.
This article is protected by copyright. All rights reserved.

A close appraisal of both the classification models reveal that features contributing towards
antitubercular activity against dormant MTB are radius of gyration molecular volume, molecular surface
area, density and principal moment of inertia while the activity against virulent MTB is shown to be
governed by dipole moment, principle moment of inertia, AlogP98, radius of gyration and sum of atomic
polarizability. This difference in contributing factors definitely provides some guidelines to channelize
the activity against dormant or virulent strains of mycobacterium.
Conclusions
The primary goal of this endeavor was to explore pyrimidones as a novel structural class targeting
dormant mycobacteria. The molecules were designed based on 3D-QSAR results and were synthesized
by a modification of the Biginelli reaction. The compounds display promising antitubercular activity
especially on dormant TB bacilli. Interestingly, some of the compounds show activity against both active
as well as dormant bacilli while some compounds show activity only against dormant bacilli e.g.
compound 13. This is only the second compound reported so far to show activity specifically against
dormant TB bacilli. Thus the hypothesis we initially proposed has been validated and the extensive
QSAR analysis on this nucleus will surely lead to a more potent candidate for targeting dormant bacilli.
RP helped to develop a discriminative model for analyzing the structure-activity relationships for these
pyrimidone derivatives. The classification using 2D descriptors has considerable discriminative power in
spite of high degree of structural similarity in the library. Also the two independent RP models derived
for these molecules could identify features essential for activity against virulent and dormant tubercular
bacteria. Cytotoxicity studies and elucidation of the mechanism of action of these molecules and
determination of the site of action is a priority in our future efforts. In conclusion, considering the clinical
importance of dormant tubercular bacilli, development of molecules active against dormant TB bacilli
This article is protected by copyright. All rights reserved.

might decrease the amount of time required to cure TB to almost half of the 6-24 months required
currently and could improve therapy against TB thus limiting its spread across the world.
Acknowledgements
The computational facilities were jointly provided by the All India Council of Technical
Education through grant (F. No. 8022/RID/NPROJ/RPS-5/2003-04), the Department of Science and
Technology through their FIST program (SR/FST/LSI-163/2003), Department of Biotechnology
(BT/PR11810/BRB/10/690/2009) and the Council of Scientific and Industrial Research (01/23/99/10-
EMR-II).
REFERENCES
1.
Wayne LG (1994) Dormancy of Mycobacterium tuberculosis and latency of disease. Eur J Clin
Microbiol Infect Dis;13: 908-14.
2.
Papadopoulou MV, Bloomer WD, McNeil MR (2007) NLCQ-1 and NLCQ-2, two new agents
with activity against dormant Mycobacterium tuberculosis. Int J Antimicrob Agents;29: 724-7.
3.
Copp BR, Pearce AN (2007) Natural product growth inhibitors of Mycobacterium tuberculosis.
Nat Prod Rep;24: 278-97.
4.
Manvar AT, Pissurlenkar RR, Virsodia VR, Upadhyay KD, Manvar DR, Mishra AK, et al. (2009)
Synthesis, in vitro antitubercular activity and 3D-QSAR study of 1,4-dihydropyridines. Mol Divers;285-
305.
This article is protected by copyright. All rights reserved.

5.
Virsodia V, Pissurlenkar RR, Manvar D, Dholakia C, Adlakha P, Shah A, et al. (2008) Synthesis,
screening for antitubercular activity and 3D-QSAR studies of substituted N-phenyl-6-methyl-2-oxo-4-
phenyl-1,2,3,4-tetrahydro-pyrimidine-5-carboxami des. Eur J Med Chem;43: 2103-15.
6.
Virsodia V, Manvar A, Upadhyay K, Loriya R, Karia D, Jaggi M, et al. (2009) Synthesis of 1-
(2,6-dichlorophenyl)-3-methylene-1,3-dihydro-indol-2-one derivatives and in vitro anticancer evaluation
against SW620 colon cancer cell line. Eur J Med Chem;44: 1355-62.
7.
Manvar A, Malde A, Verma J, Virsodia V, Mishra A, Upadhyay K, et al. (2008) Synthesis, anti-
tubercular activity and 3D-QSAR study of coumarin-4-acetic acid benzylidene hydrazides. Eur J Med
Chem;43: 2395-403.
8.
Nayyar A, Malde A, Jain R, Coutinho E (2006) 3D-QSAR study of ring-substituted quinoline
class of anti-tuberculosis agents. Bioorg Med Chem;14: 847-56.
9.
Ubiali D, Pagani G, Pregnolato M, Piersimoni C, Pedraz Munoz JL, Rodriguez Gascon A, et al.
(2002) New N-alkyl-1,2-dihydro-2-thioxo-3-pyridinecarbothioamides as antituberculous agents with
improved pharmacokinetics. Bioorg Med Chem Lett;12: 2541-4.
10.
Rosowsky A, Forsch RA, Queener SF (2002) Inhibition of Pneumocystis carinii, Toxoplasma
gondii, and Mycobacterium avium dihydrofolate reductases by 2,4-diamino-5-[2-methoxy-5-(omega-
carboxyalkyloxy)benzyl]pyrimidines: marked improvement in potency relative to trimethoprim and
species selectivity relative to piritrexim. J Med Chem;45: 233-41.
11.
Desai B, Sureja D, Naliapara Y, Shah A, Saxena AK (2001) Synthesis and QSAR studies of 4-
substituted phenyl-2,6-dimethyl-3, 5-bis-N-(substituted phenyl)carbamoyl-1,4-dihydropyridines as
potential antitubercular agents. Bioorg Med Chem;9: 1993-8.
This article is protected by copyright. All rights reserved.

12.
Kharkar PS, Desai B, Gaveria H, Varu B, Loriya R, Naliapara Y, et al. (2002) Three-dimensional
quantitative structure-activity relationship of 1,4-dihydropyridines as antitubercular agents. J Med
Chem;45: 4858-67.
13.
14.
15.
16.
Nicholas DJD, Nason A (1957) Determination of nitrate and nitrite. Methods Enzymol;3: 981-4.
Sybyl. (2005) USA: Tripos Associates Inc.
Accelrys (1998) Cerius2. San Diego, CA, USA.
Strobl C, Malley J, Tutz G (2009) An introduction to recursive partitioning: rationale, application,
and characteristics of classification and regression trees, bagging, and random forests. Psychol
Methods;14: 323-48.
17.
Choi SY, Shin JH, Ryu CK, Nam KY, No KT, Park Choo HY (2006) The development of 3D-
QSAR study and recursive partitioning of heterocyclic quinone derivatives with antifungal activity.
Bioorg Med Chem;14: 1608-17.
18.
Young SS, Hawkins DM (2004) Using recursive partitioning analysis to evaluate compound
selection methods. Methods Mol Biol;275: 317-34.
19.
Rusinko A, 3rd, Young SS, Drewry DH, Gerritz SW (2002) Optimization of focused chemical
libraries using recursive partitioning. Comb Chem High Throughput Screen;5: 125-33.
20.
Blower P, Fligner M, Verducci J, Bjoraker J (2002) On combining recursive partitioning and
simulated annealing to detect groups of biologically active compounds. J Chem Inf Comput Sci;42: 393-
404.
21.
Khan A, Sarkar D (2008) A simple whole cell based high throughput screening protocol using
Mycobacterium bovis BCG for inhibitors against dormant and active tubercle bacilli. J Microbiol
Methods;73: 62-8.
This article is protected by copyright. All rights reserved.

Table 1.
Structures and activities of the molecules synthesized and tested against aerobic and
dormant (NR) strains of M. bovis.
Sr. No.
R1
H
R2
H
Activity (IC50 µg/ml)
> 100 µg/ml
5
6
H
3-Cl
> 100 µg/ml
7
H
4-NO2
2,3-diCl
H
>100 µg/ml
8
H
> 100 µg/ml
9
2,4-diCl
2,4-diCl
2,4-diCl
2,4-diCl
3-Cl
> 100 µg/ml
10
11
12
13
3-Cl
>100 µg/ml
2,3-diCl
4-NO2
3-Cl
> 100 µg/ml
>100 µg/ml
> 100 µg/ml (aerobic stage)
6.3 µg/ml (N.R. activity)
This article is protected by copyright. All rights reserved.

14
3-Cl
2,3-diCl
14.3 µg/ml (aerobic stage)
6.3 µg/ml (N.R. activity)
> 100 µg/ml
15
16
3-Cl
3-Cl
3-NO2
4-NO2
54.95 µg/ml (aerobic stage)
6.67 µg/ml (N.R. activity)
> 100 µg/ml
17
18
19
20
4-Cl
2,3-diOCH3
H
H
3-Cl
> 100µg/ml
cyclohexyl
cyclohexyl
> 100 µg/ml
2,4-diCl
60.37 µg/ml (aerobic stage)
> 100 µg/ml (N.R. activity)
> 100 µg/ml
21
22
23
24
25
26
27
28
29
30
3-NO2
3-Cl
H
2-Cl
10.29 µg/ml (aerobic stage)
> 100 µg/ml
3-NO2
3-Cl
2,3-diCl
4-OCH3
4-OCH3
3-Cl
> 100 µg/ml
3-NO2
3-NO2
3-Cl
> 100 µg/ml
10.52 µg/ml (aerobic stage)
> 100 µg/ml
2-CH3
3-NO2
3-Cl
morpholinyl
morpholinyl
> 100 µg/ml
> 100 µg/ml
> 100 µg/ml
This article is protected by copyright. All rights reserved.

31
32
33
> 100 µg/ml
> 100 µg/ml
> 100 µg/ml
34
35
3-Cl
3-Cl
2-pyridinyl
cyclohexyl
12.5 µg/ml (aerobic stage)
27.17 µg/ml (aerobic stage)
This article is protected by copyright. All rights reserved.

Table 1. (contd.)
Structures and activities of the molecules synthesized and tested against aerobic
and dormant (NR) strains of M. bovis.
Sr. No.
36
R1
H
H
H
H
R2
Activity (IC50 µg/ml)
> 100
3-Cl
2,3-diCl
3-NO2
4-NO2
> 100
37
>100
38
80.69 (aerobic stage)
7.41 (N.R. activity)
>100
39
2,4-diCl
2,4-diCl
2,4-diCl
3-Cl
40
41
42
2,3-diCl
4-NO2
> 100
25.1 (aerobic stage)
6.97 (N.R. activity)
> 100
3-Cl
3-Cl
3-Cl
43
44
2,3-diCl
> 100
This article is protected by copyright. All rights reserved.

H
cyclohexyl
cyclohexyl
H
> 100
>100
45
2,4-diCl
3-Cl
46
> 100
47
3-NO2
3-Cl
2-Cl
18.96 (aerobic stage)
> 100
48
2-Cl
49
3-NO2
3-NO2
3-Cl
2-CH3
> 100
50
2,3-diCl
4-OCH3
4-OCH3
3-Cl-4-F
3-Cl-4-F
2-CH3
> 100
51
> 100
52
3-NO2
3-NO2
3-Cl
> 100
53
3.45 (aerobic stage)
4.13 (aerobic stage)
> 100
54
55
3-Cl
56
3-Cl
morpholinyl
morpholinyl
morpholinyl
> 100
57
58
3-Cl
> 100
3-NO2
> 100
59
Rifampicin
0.04 (aerobic stage) 0.26
(N.R activity)
Isoniazid
0.04 (aerobic stage) 0.25
(N.R activity)
Streptomycin
0.17 (aerobic stage)
0.50 (N.R activity)
0.16 (aerobic stage)
0.3 (N.R activity)
0.5 (aerobic stage)
0.30 (N.R activity)
Ethambutol
Pyrazinamide
This article is protected by copyright. All rights reserved.

Table 2.
Descriptors used to develop the classification model.
X-Descriptors
Description
Structural descriptor
Kappa indices
the number of H-bond donor and acceptor
the shape of molecule
E_state_keys
the electrotopological interaction for each atom
Charge, Apol , HOMO, LUMO, Dipole moment.
Electronic descriptors
Topological descriptors
Wiener index (W), Zagreb index (Zagreb), Hosoya index (Z), Kier & Hall
molecular connectivity index (χ)
Spatial descriptors
Rad of Gyration, Jurs descriptors, Molecular volume, Area , Density,
Principal moment of inertia
Table 3.
Statistical results of recursive partitioning for the model built for activity against dormant
mycobacteria.
Class
No. of
molecules
Activity % of molecules
Class % as
Obs Correct
100
Overall % as
Pred Correct
100
Enrichment
Ratio
1
2
15
5
0
1
75
25
1.33
4.00
100
100
Table 4. Activity prediction in terms of binary codes by recursive partitioning for activity against
dormant mycobacteria.
Molecule Biological
RP
Radius of Molecular
Molecular Density
Surface
Principal
moment of
inertia
ID
activity in
binary
code
Predicted
activity in
binary code
Gyration
Volume
(Vm)
Area
This article is protected by copyright. All rights reserved.

7
0
0
1
1
1
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
1
1
1
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
4.35
4.06
4.13
4.13
4.38
4.04
4.11
4.21
4.07
4.08
4.36
4.08
4.14
3.88
4.06
4.20
4.13
4.23
4.26
4.14
302.08
329.24
306.26
319.37
315.66
324.25
306.33
315.32
287.13
310.00
311.69
339.34
316.09
329.76
333.87
325.00
316.05
338.91
329.68
320.67
369.24
413.14
381.56
393.77
395.32
419.31
374.71
393.55
353.20
407.93
383.10
429.73
400.06
417.22
432.70
407.36
390.70
420.44
411.73
404.45
1.17
1.28
1.23
1.29
1.23
1.18
1.23
1.23
1.19
1.12
1.18
1.29
1.24
1.29
1.19
1.24
1.24
1.29
1.28
1.28
1459.97
1459.26
1495.92
1683.92
1709.37
1161.28
1276.94
1542.22
1083.97
1078.15
1584.96
1610.45
1594.63
1396.60
1264.83
1419.66
1369.96
1815.23
1882.99
1781.55
12
13
14
16
20
22
26
34
35
39
42
43
44
46
48
49
51
54
55
This article is protected by copyright. All rights reserved.

Table 5.
Statistical results of recursive partitioning (cross validation)(dormant)
Class
No. of
molecules
Activity % of molecules
Class % as
Obs Correct
66.67
Overall % as
Pred Correct
71.43
Enrichment
Ratio
1
2
15
5
0
1
75
25
1.95
1.97
50.00
66.67
Table 6.
Statistical results of recursive partitioning for the model built for activity against virulent
mycobacteria
Class
No. of
molecules
Activity % of molecules
Class % as
Obs Correct
100
Overall % as
Pred Correct
100
Enrichment
Ratio
1
2
8
0
1
40
60
2.50
1.67
12
100
100
Table 7. Activity prediction in terms of binary codes by recursive partitioning for the model built using
activity against virulent mycobacteria.
Radius
of
Biological
Molecule activity in
RP Predicted
activity in
binary code
Dipole-
mag
Apol
PMI
AlogP98
ID
binary
code
Gyration
4.35
7
0
0
0
1
0
0
0
1
13961.24
17040.80
15965.72
17505.50
3.25
2.89
3.46
4.43
1459.97
1459.26
1495.92
1683.92
1.89
3.22
3.32
3.99
12
13
14
4.06
4.13
4.13
This article is protected by copyright. All rights reserved.

16
20
22
26
34
35
39
42
43
44
46
48
49
51
54
55
1
1
1
1
1
1
1
1
0
0
0
1
0
0
1
1
1
1
1
1
1
1
1
1
0
0
0
1
0
0
1
1
15501.02
15262.60
15965.72
15501.02
14166.42
13722.82
15279.04
18358.60
17283.52
18823.30
16580.40
16818.82
17283.52
18358.60
16931.20
17395.90
5.10
6.69
4.57
5.30
4.87
6.01
3.42
3.06
3.40
4.25
6.58
5.43
4.41
4.25
5.80
3.45
4.38
4.04
4.11
4.21
4.07
4.08
4.36
4.08
4.14
3.88
4.06
4.20
4.13
4.23
4.26
4.14
1709.37
1161.28
1276.94
1542.22
1083.97
1078.15
1584.96
1610.45
1594.63
1396.60
1264.83
1419.66
1369.96
1815.23
1882.99
1781.55
2.55
3.60
3.32
2.55
2.05
2.94
2.79
4.12
4.22
4.89
4.50
3.45
4.22
4.12
3.66
4.43
Table 8.
Statistical results of recursive partitioning (cross validation) (virulent)
Class
No. of
molecules
Activity % of molecules
Class % as
Obs Correct
50.00
Overall % as
Pred Correct
54.44
Enrichment
Ratio
1
2
8
0
1
40
60
1.11
1.06
12
58.33
63.64
This article is protected by copyright. All rights reserved.

Figure 1.
Chemical structures or cores known to have antitubercular activity
Figure 2
Recursive Partitioning tree generated with moderate pruning; score splits using Twoing
Rule for prediction of antitubercular activity (dormant) classes. Those marked 1:1, 2:1,
3:0, 4:1, 5:0 and 6:0 correspond to terminal nodes 1-6 and each terminal node
corresponds to the value of 1 (active) or 0 (inactive).
This article is protected by copyright. All rights reserved.

Figure 3
Recursive Partitioning tree generated with moderate pruning; score splits using Twoing
Rule for prediction of antitubercular activity (virulent) classes. Those marked 1:0, 2:1, 3:0, 4:1, 5:1 and
6:0 correspond to terminal nodes 1-6 and each terminal node corresponds to the value of 1 (active) or 0
(inactive).
This article is protected by copyright. All rights reserved.

Scheme 1.
The modified Biginelli reaction for synthesis of substituted tetrahydropyrimidone
derivatives. (i) toluene; (ii) urea or thiourea in presence of conc. HCl / pTSA,
absolute ethanol, 4-8 hours reflux.
O
O
H
HN
O
O
O
NH2
+
(i)
R1
(4)
R2
R2
+
O
(3)
(1)
(2)
(ii)
R1
O
R2
N
H
NH
(5)
N
H
O
This article is protected by copyright. All rights reserved.