Design, synthesis and biological evaluation of novel isoniazid derivatives with potent antitubercular activity

*Highlights (for review)

ACCEPTED MANUSCRIPT

Highlights



Robust and highly predictive QSAR models were developed

17 isoniazid (INH) derivatives were designed and 13 actually synthesized

Five newly synthesized INH derivatives showed MICs against H37Rv lower

than INH



One compound showed a six fold fall in MIC against the katG (S315T) mutated

strain

Compounds were shown not to be cytotoxic

*Revised Manuscript

Click here to view linked References

ACCEPTED MANUSCRIPT

DESIGN, SYNTHESIS AND BIOLOGICAL

EVALUATION OF NOVEL ISONIAZID

DERIVATIVES WITH POTENT

ANTITUBERCULAR ACTIVITY

a,*

a

a,b

a,c

Filomena Martins, , Susana Santos , Cristina Ventura , Ruben Elvas- Leitão , Lídia

a,1

a,2

a,b

a,3

a

Santos , Susana Vitorino , Marina Reis , Vanessa Miranda , Henrique F. Correia ,

d

d,4

d,e

f

João Aires-de-Sousa , Vasyl Kovalishyn , Diogo A.R.S. Latino , Jorge Ramos ,

Miguel Viveiros

f

a

Centro de Química e Bioquímica (CQB), Departamento de Química e Bioquímica,

Faculdade de Ciências, Universidade de Lisboa, Ed. C8, Campo Grande 1749-016

Lisboa, Portugal

b

Instituto Superior de Educação e Ciências, Alameda das Linhas de Torres, 179, 1750-

1

42 Lisboa, Portugal

c

Área Departamental de Engenharia Química, Instituto Superior de Engenharia de

Lisboa (ISEL), Instituto Politécnico de Lisboa, R. Conselheiro Emídio Navarro, 1959-

0

07 Lisboa, Portugal

d

REQUIMTE and CQFB, Departamento de Química, Faculdade de Ciências e

Tecnologia, Universidade Nova de Lisboa, Monte de Caparica, 2829-516 Caparica,

Portugal

e

Centro de Ciências Moleculares e Materiais (CCMM), Departamento de Química e

Bioquímica, Faculdade de Ciências, Universidade de Lisboa, Ed. C8, Campo Grande,

1

749-016 Lisboa, Portugal

f

Grupo de Micobactérias, Unidade de Microbiologia Médica, Instituto de Higiene e

Medicina Tropical, Universidade Nova de Lisboa, Rua da Junqueira, 100, 1349-008

Lisboa, Portugal

Abstract The disturbing emergence of multidrug-resistant strains of Mycobacterium tuberculosis (Mtb)

has been driving the scientific community to urgently search for new and efficient antitubercular drugs.

Despite the various drugs currently under evaluation, isoniazid is still the key and most effective

component in all multi-therapeutic regimens recommended by the WHO. This paper describes the QSAR-

oriented design, synthesis and in vitro antitubercular activity of several potent isoniazid derivatives

(

isonicotinoyl hydrazones and isonicotinoyl hydrazides) against H37Rv and two resistant Mtb strains.

QSAR studies entailed RFs and ASNNs classification models, as well as MLR models. Strict validation

procedures were used to guarantee the models‟ robustness and predictive ability. Lipophilicity was shown

not to be relevant to explain the activity of these derivatives, whereas shorter N-N distances and lengthy

substituents lead to more active compounds. Compounds 1, 2, 4, 5 and 6, showed measured activities

*Corresponding author: Tel: +351 21 7500870; Fax: +351 21 7500088. E-mail address:

filomena.martins@fc.ul.pt (F. Martins).

1

Present address: CENIMAT – Centro de Investigação em Materiais, Faculdade de Ciências e Tecnologia,

Universidade Nova de Lisboa, Monte da Caparica, 2829 - 516 Caparica, Portugal.

2

Present address: Department of Surgery and Cancer, Imperial College London, Charing Cross Campus,

Fulham Palace Road, London, W6 8RF, United Kingdom.

3

Present address: Instituto de Tecnologia Química e Tecnológica (ITQB), Universidade Nova de Lisboa,

Av. da República, 2780-157, Oeiras, Portugal.

4

Present address: Institute of Bioorganic Chemistry & Petroleum Chemistry, National Ukrainian

Academy of Sciences, Kyiv-94, 02660, Murmanskaya, 1, Ukraine.

1

ACCEPTED MANUSCRIPT

against H37Rv higher than INH (i.e., MIC  0.28 M), while compound 9 exhibited a six fold decrease in

MIC against the katG (S315T) mutated strain, by comparison with INH, (i.e., 6.9 vs. 43.8 M). All

compounds were ineffective against H37Rv_I_N_H(ΔkatG), a strain with a full deletion of the katG gene, thus

corroborating the importance of KatG in the activation of INH-based compounds. The most potent

compounds were also shown not to be cytotoxic up to a concentration 500 times higher than MIC.

Keywords Antitubercular Activity; Isoniazid Derivatives; Mycobacterium tuberculosis;

Resistance; QSARs; Synthesis

2

ACCEPTED MANUSCRIPT

1

. Introduction

With nearly one-third of the global population infected with the Mycobacterium

tuberculosis bacilli (Mtb), tuberculosis (TB) is still a major cause of mortality and

morbidity. The World Health Organization (WHO) estimated for 2012 8.6 million new

cases and 1.3 million deaths due to TB, of which 320.000 among HIV-positive [1]. It

was also reported that about 3.6% of the new patients and 20% of the previously treated

ones had multidrug-resistant TB (MDR-TB), which is caused by Mtb strains resistant to

the most effective first-line drugs, isoniazid (INH) and rifampicin (RIF). Furthermore,

as by September 2013, 92 countries had reported at least one case of extensively drug-

resistant TB (XDR-TB), a form of TB resistant to INH and RIF, as well as to any

fluoroquinolone and to any of the second-line injectable drugs (amikacin, kanamycin or

capreomycin). Recent WHO updates claim that approximately 9.6% of all MDR-TB

cases are XDR-TB [2]. To make the scenario even more serious, a new designation,

“total drug resistance” tuberculosis (TDR-TB), although not yet recognized by the

WHO, has been used in the literature to describe an allegedly new (or deadlier) form of

TB, resistant to all first- and second-line drugs [3-5]. These resistant strains are

emerging as a major public health problem, threatening the success of the DOTS

program (Directly-Observed Treatment Short-course), the WHO program launched in

1

994 for monitoring TB detection, treatment and drug supply [6, 7].

Besides the need to make the treatment of drug-sensitive TB shorter, simpler, safer and

more accessible, also treatment of MDR- and XDR-TB must be greatly improved to

become more efficient, less toxic, better tolerated and less expensive, as well as

compatible with antiretroviral therapies due to the increasing number of patients co-

infected with HIV/AIDS [8-11]. Current drugs against TB are mostly inadequate to

address this fight against tuberculosis.

After decades of stagnation in research for new antitubercular drugs, a new interest from

pharmaceutical companies and consortia, a more consistent support from governmental

institutions and some effective funding from international non-governmental

organizations, have led to the appearance of several classes of molecules which trigger

different biological pathways such as cell wall synthesis, nucleic acid synthesis, protein

synthesis or membrane energy production and which are now under various stages of

_

___________

Abbreviations: AE, average error; AAE, absolute average error; ASNNs, associative neural networks;

BCG, Bacillus Calmette-Guérin; CCC, concordance correlation coefficient; ClogP, calculated logarithm

(

base 10) of the partition coefficient; DMSO d , dimethylsulfoxide deuterated; DOTS, directly-observed

6

treatment short-course; enoyl-ACP, enoyl-acyl carrier protein; EtOAc, ethylacetate; EtOH, ethanol; GU,

growth unit; IC , concentration necessary to reduce cell growth to 50%; IC , concentration necessary to

5

0

99

inhibit 99% of bacterial activity; INH, isoniazid; InhA, enoyl-ACP reductase in Mtb; kasA, gene encoding



-ketoacyl carrier protein synthase; katG, gene encoding KatG; KatG, Mtb catalase peroxidase; LMO,

leave-many-out; LOO, leave-one out; MDR-TB, multidrug-resistant tuberculosis; MeOH, methanol;

MIC, minimum inhibitory concentration; MLR, multiple linear regressions; Mtb, Mycobacterium

tuberculosis; MTT, (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide; ndh-2, gene coding

for Mtb type II NADH:menaquinone oxidoreductase; NIAID, National Institute of Allergy and Infectious

Diseases; NN, neural networks; oxyR–ahpC, oxidative regulator-alkyl hydroperoxidase; OECD

Organization for Economic Co-operation and Development; OOB, out-of-bag; PLS, partial least squares;

2

Q , cross-validated correlation coefficient; R , determination coefficient; RFs, random forests; RIF,

rifampicin; RMSE, root mean squared error; SD, standard deviation; SI, selectivity index; SL, significance

level; SVM, support vector machines; TB, tuberculosis; TDR-TB, total drug resistance tuberculosis;

WHO, World Health Organization; XDR-TB, extensively drug-resistant tuberculosis.

3

ACCEPTED MANUSCRIPT

clinical trials [11-13].

Despite the accelerated approval granted by FDA to bedaquiline, the first drug to seek a

specific indication for (adult) MDR-TB in the USA and also the first drug with a new

mechanism of action (MOA) for treating TB in over 40 years [14-16], the fact remains

that isoniazid is still the most effective drug against wild type Mtb and is therefore the

treatment of choice for TB.

The clinical usefulness of isoniazid (isonicotinic acid hydrazide) was discovered in the

1

950‟s when its specific high activity against Mtb was first noticed, higher than any

other compound used at the time [17, 18]. INH is indeed particularly active against

Mycobacterium tuberculosis with a minimum inhibitory concentration (MIC) of 0.05

g/mL [18].

It is thought that INH enters Mtb through passive diffusion through the cell-wall and

that it is active only against dividing bacteria [19, 20]. The first study regarding the

mechanism of action of INH was published in 1970 by Winder and Collins [21],

connecting INH with mycolic acid biosynthesis. In order for INH to be effective against

Mtb it needs to be activated by the multifunctional catalase-peroxidase enzyme KatG

into a range of activated species, such as an isonicotinoyl radical, that can acylate

numerous compounds [22, 23]. The generally accepted mechanism of action of INH

postulates that the isonicotinoyl radical binds to the nicotinamide adenine dinucleotide

+

(

NAD ) and the resulting adduct inhibits the enoyl-ACP reductase InhA, a NADH-

dependent enoyl-acyl carrier protein (enoyl-ACP) reductase of the fatty acid synthase

type II system (FASII). This inhibition causes accumulation of long -chain fatty acids,

inhibition of mycolic acid biosynthesis and, ultimately, cell death [18, 24].

The main mechanism of resistance to isoniazid resides in the presence of mutations in

its activator, KatG, product of the katG gene [25], whereas mutations in the inhA gene

represent the second most common cause of resistance. Altogether, mutations in these

two genes are responsible for approximately 75% of all cases of Mtb resistance to

isoniazid in the clinical setting [26]. Resistance to INH has also been associated with

mutations in several other genes (e.g., ndh, kasA and oxyR–ahpC intergenic region)

[

18], but their direct association with resistance is still not clear. It has also been

demonstrated very recently that drug efflux contributes likewise to the overall

resistance, particularly in the case of acquired isoniazid resistance [27, 28], much in the

same way as active-efflux plays a role in the failure of cancer chemotherapy [29].

As mentioned before, isoniazid remains a key component in all multiple drug treatment

regimens recommended by the WHO albeit resistant isolates are rapidly generated

during monotherapy or inappropriate treatment [30]. Similarly to some other candidates

in the pipeline that were chemically tailored from drugs to which Mtb was already

resistant, departing from isoniazid might also bring new insights into the development

of new antitubercular agents. Hence, improvement of INH by introducing chemical

modifications in its core structure in order to enhance the biological response against

Mtb and/or circumvent resistance phenomena continues to be an interesting scientific

challenge.

To face this challenge, in vitro, in vivo or in silico methods are being used in early

stages of drug development to avoid possible failures, especially those related with drug

metabolism, pharmacokinetic profiles and toxicity issues [31, 32]. In particular,

quantitative structure–activity relationships (QSAR), is an in silico methodology

4

ACCEPTED MANUSCRIPT

extensively used in Medicinal Chemistry to find relationships between molecular

properties of chemical compounds and measured biological activities [32-34].

With the purpose of establishing robust and predictive MLR- and NN-based QSAR

models to assist us in the design of new active compounds, we have developed a

systematic analysis of a large data set of potentially antitubercular agents [35-37]. In

this work we describe the details of the QSAR-oriented design, synthesis and in vitro

antibacterial activity evaluation of 13 compounds derived from the INH core structure

(Figure 1) which were assessed against the wt Mtb H37Rv strain and two mutated

strains, one carrying solely a katG S315T mutation (the most frequent mutation related

with INH resistance), and the other resulting from a full deletion of the katG gene

(katG) [28].

Figure 1

5

ACCEPTED MANUSCRIPT

2

. Methods

2

.1 Data Sets

Different blocks of data, based on two independent approaches, were used to build anti-

tubercular activity models in order to design novel potentially active isoniazid

derivatives, namely i- “RFs and ASNNs data sets”, used to develop classification

models that were integrated into a consensus model to predict the propensity of a

compound to be active [36, 38]. These classification models were applied to a “virtual

data set” of compounds assembled using the INH molecule as a common scaffold; ii-

“MLRs data sets”, used to develop regression models for the design of another set of

potentially active compounds. Additionally, these MLR models were also used for a

quantitative evaluation of some selected compounds proposed as active by the

classification models.

2

.1.1 RFs and ASNNs Data Sets

Four different data sets were used to establish classification models with Associative

Neural Networks (ASNNs) and Random Forests (RFs) methods. All compounds were

divided into two classes, according to their MIC values, i.e., their minimum inhibitory

concentrations: active (MIC ≤ 5 µM) and inactive (MIC > 5 µM).

Data set i consisted of 173 hydrazide derivatives with known MIC values retrieved

from the literature, 136 compounds tested against the BCG Mtb strain [39] and 37

against the H37Rv [40-43]. Data set ii consisted of 372 INH derivatives obtained from

two sources: 259 compounds from the NIAID and GVK database s [44], evaluated

against H37Rv, and 113 retrieved from several other sources [39-43, 45-47], of which

5

3 were tested against BCG and 60 against H37Rv. Data set iii corresponded to the 113

INH derivatives subset of data set ii whereas data set iv corresponded to the remaining

59 INH derivatives. An additional external evaluation set composed of 45 INH

2

derivatives tested against the H37RV strain, data set v, was also selected from the

literature [48-50]. All compounds were stored in SMILES and MDL SDF formats.

RFs and ASNNs data sets are given in Table S1 in Supporting Information.

2

.1.2 Virtual Data Set

A set of 430 drug-like molecules was generated with an in-house developed program

starting from 10 skeletons comprising derivatives of two INH families, isonicotinoyl

hydrazides and isonicotinoyl hydrazones, and 43 common substitution patterns (see

Table S2, in Supporting Information). This set was then screened by the developed

classification models.

2

.1.3 MLR Data Sets

For building the MLR models, biological data were retrieved from literature for 45

compounds tested against the BCG strain [39], and 37 compounds tested against the

H37Rv strain [40-43], in a total of 82 isoniazid derivatives. Although data refer to two

different strains, no systematic deviation among the various sets of MIC values that

could be attributed to data inconsistency was ever noticed in the setting up of the

6

ACCEPTED MANUSCRIPT

models either in this work or in our previous investigations involving INH derivatives

[

35-37].

Additionally, a set of 13 compounds from the Novartis TB public access database, with

MIC values measured against the H37Rv strain, were used to make a further external

validation of the established models [51].

2

.2 Molecular Descriptors

.2.1 Molecular Descriptors for RFs and ASNNs Models

A set of 3224 theoretical molecular descriptors divided in 22 types was calculated using

the DRAGON program [52] for all mentioned RFs and ASNNs data sets. Constant and

near-constant descriptors were deleted.

A set of 1270 descriptors, for the above mentioned data sets, was also calculated with

the ADRIANA.Code program [53]. For this purpose, the 3D structures were generated

with the embedded CORINA program.

The type of descriptors used in each case is presented in Table S1 of Supporting

Information.

2

.2.2 Molecular Descriptors for MLR Models

Twenty-eight molecular descriptors were calculated for each of the 82 isoniazid derived

compounds: 5 Abraham‟s descriptors calculated with the Absolv program [54-56], and

8

energetic, 9 geometrical, 3 structural (Verloop‟s parameters), and 2 electronic

+

parameters calculated with the Molecular Modeling Pro Plus (MMP ) software [57].

Additionally, ClogP, taken as a measure of lipophilicity, was calculated using

ChemDraw Ultra, 11.0.1 [58]. For descriptors dependent on 3D structures, the geometry

of each compound was previously optimized using the MM2 method, a molecular

mechanics method incorporated in the software. Partial charges were calculated using

the Del Re method and dipole moments by a combination of PEOE and Huckel/4

methods also included in the referenced software. Molecular structures were drawn with

ChemDraw Ultra, and saved as SMILES strings and are available as Supporting

Information in Table S3. Descriptor values for all 82 compounds, as well as the

corresponding MIC values from literature, are also given as Supporting Information in

Table S4.

2

.3 Classification and Regression Techniques

.3.1 Random Forests

A Random Forest [59], RF, is an ensemble of unpruned classification trees which is

created using bootstrap samples of the training set and random subsets of descriptors to

define the best split at each node. The final prediction for an object from a RF is

obtained by majority voting of the individual trees. The bootstrapping technique is used

to internally assess the model‟s performance from the prediction errors for the objects

left out in the bootstrap procedure (out-of-bag estimation, OOB). RF can also be used

for descriptor selection. The RF algorithm estimates the importance of a descriptor by

determining how much the prediction error increases when data for that descriptor is

7

ACCEPTED MANUSCRIPT

permuted keeping all others unchanged. RFs were grown with R program v. 2.10.0 and

2

.11.0 [60] using the Random Forest library. Forest size and number of descriptors

selected at each split were optimized, being the number of trees set to 1000 and the

number of descriptors tried at each split set to 30.

2

.3.2 Associative Neural Networks

Associative neural networks (ASNNs) [61] integrate an ensemble of feed-forward

neural networks (FFNNs) with a memory of experimental data. The ensemble consists

of independently trained FFNNs, which contribute to a single prediction. The final

prediction for an object from an ASNN is obtained from (a) the outputs produced by the

ensemble of individual NNs and (b) the most similar cases in the memory (here, the

training set). Further details on the application of this technique can be found in

Supporting Information, Section 1.

2

.3.3 Multiple Linear Regressions

In the context of this work, a multiple linear regression (MLR) methodology was used

to establish a mathematical relationship between the biological activity of the target

compounds (Y) and a set of molecular descriptors (X ) that numerically encode the

i

features of the compounds‟ chemical structure and which can be expressed in general

terms as follows:

N

(1)

Ya  a X 

0



i

i1

0 i

where  stands for the regression residuals and a and a are the regression coefficients

obtained by minimizing the sum of the squared residuals. The order of magnitude of the

regression coefficients reflects the degree of influence of the molecular descriptors,

whereas their sign can be interpreted in terms of positive or negative contributions to

the activity. A correct use of a MLR-QSAR approach presupposes that the descriptors

are not intercorrelated to avoid redundancy in descriptor information [62] and that the

number of compounds for model building is at least four times the number of

descriptors used, to avoid overfitting and therefore chance correlations [62, 63]. The

success of a MLR-QSAR approach in the development of interpretative and predictive

model equations requires representative, homogeneous and good quality da ta, a well-

defined domain of applicability [64-66], and strict validation procedures [63, 67, 68].

2

.4 Selection of Descriptors for ASNNs and RF Models

The selection of descriptors to develop RF and ASNN classification models was

performed in two steps: (a) using the ranking of descriptors established by RFs and (b)

using pruning methods implemented in the ASNN program. A pre-selection of the 300

to 400 most important descriptors obtained from RF models was performed and used in

the next step where the ASNN pruning methods were applied. Details on the ASNNs

pruning methods are given in Supporting Information, Section 2.

8

ACCEPTED MANUSCRIPT

The influence of the number of selected descriptors on the ASNN model quality was

checked on the basis of the leave-one-out (LOO) results for the training sets (please

refer to Supporting Information, Section 1). Pre-selecting less than 300–400 descriptors

by RF decreased the ASNN models predictive ability. With the final sets of descriptors

selected by ASNNs pruning methods for each one of the data sets, ASNNs and RFs

were used to obtain the final models.

2

.5 Validation Procedures

.5.1 Validation of RFs and ASNN Models

Both ASNNs and RF models were first internally validated on the basis of the training

set by their intrinsic validation procedures, namely out-of-bag (OOB) and leave-one-out

method, respectively (see section 2.3). Furthermore, models were submitted to an

external validation using an independent test set (data set v).

The traditional way of summarizing “lack of fit” in QSAR models is to use the root

mean squared error (RMSE) and the mean absolute error (MAE) between predicted and

experimental activities. For ASNNs, the RMSE is calculated by the expression

N

k

2

Y_i_jO_i_j



(2)





i=1 j=1

RMSE =

N k

where Yij and Oij are the predicted and measured activity j for molecule i, respectively,

N is the number of compounds, and k is the number of output neurons.

To assess the classification ability and to separately monitor the classification

performance of the two classes (active and non-active), sensitivity (Sn), specificity (Sp),

and overall accuracy (Ac) were calculated as follows:

(3)

(4)

(5)

Sn = TP / (TP + FN)

Sp = TN / (TN + FP)

Ac = (TP + TN) / (TP + FN + TN + FP)

where TP, FP, TN and FN are true positives, false positives, true negatives and false

negatives, respectively. In general, the overall accuracy, Ac, is always used to measure

the predictive power of classification models, as the number of active and inactive

compounds is approximately the same in all data sets.

2

.5.2 Validation Procedures Used on MLR Models

For MLR validation purposes, data was split into a training set and an independent test

set with similar variability both in the dependent variable, i.e., the biological activity,

and in the chemical/structural multidimensional space covered by the descriptors.

9

ACCEPTED MANUSCRIPT

Training/test set splitting took also into consideration the fact that test set compounds

must exhibit structural similarity with their training set counterpa rts to hamper

predictions by extrapolation which are more likely to be unreliable [62, 66].The training

set was used to derive the models, whereas the test set (and, for that matter, also the

Novartis external data set) was used to evaluate the models‟ predictive ability.

The best MLR models were found by a forward stepwise procedure starting with

equations containing a single descriptor and adding up terms, one at a time. All

combinations of the 28 molecular descriptors were tested and descriptors were retained

or disregarded in each step according to rigorous statistical criteria (see below).

Also, the intercorrelation among descriptors used in each regression was always

2

checked (for descriptors to be considered non-redundant, r between any two descriptors

2

should be smaller than 0.5 and R for a given descriptor against a linear combination of

all others, should be lower than 0.8) [69].

Additionally, models were examined for anomalous observations, the so-called outliers.

These allegedly suspicious values were fully assessed using two criteria: the

conventional criterion of |Ycalc - Yexp| > 2 SD, where SD stands for the standard deviation

of the fit, and a more refined measure, the Cook‟s distance [70, 71].

The goodness-of-fit of the best models was also evaluated by checking their linearity

through residual analysis and normal probability plots. The models‟ robustness,

(

evaluated with training set compounds) was assessed by traditional statistical criteria,

2

namely, the determination coefficient, R , the standard deviation of the fit, SD, the F

statistics, and the significance level (SL) of each adjusted parameter (parameters were

kept in the regressions only if SL > 95%).

Internal validation was further gauged by the leave-many-out (LMO) cross validation

2

correlation coefficient, Q , given by:

training

2

ˆ

^Yi ^^Y

i



i1





2

(6)

Q  1 

training

2

Y  Y







i

i1

where ꢁꢂꢃꢄꢅꢆꢃꢇꢃare the measured, predicted and averaged (over the whole data set)

ꢀ

2

values of the dependent variable, respectively. A Q LM O (a verage) > 0.6 is usually taken

as a good measure of the robustness of the model [35, 63, 72]. Root mean squared errors

(RMSE) were also calculated.

The predictive power of the MLR models was evaluated through several demanding and

complementary external validation criteria, using precautionary threshold values, i.e .,

values significantly higher than those normally applied in current QSAR studies [73-

7

Q

7

6]. The “measures” of predictive ability calculated in this work were the exter nal

2

ext(F1) statistics [62, 74], the r

m

metrics (and also the corresponding r )[73, 76,

7] and the concordance correlation coefficient, CCC [74, 75]. Details of these metrics

are given in Supporting Information, Section 3. Additionally, scatter plots of predicted

vs. experimental data were also obtained, as recent studies have recommended the visual

1

0

ACCEPTED MANUSCRIPT

inspection of these plots as important complementary indicators of model predictivity

[

75].

The predictive capacity of each model was further evaluated by computing various

measures of fit between predicted and experimental values, namely, the average error

(AE), the absolute average error (AAE) and the already referred RMSE which should all

be as low as possible. Finally, models had still to comply with the following criteria:

2

0

R  R

2

R > 0.6;

 0.1 ; 0.85 < m < 1.15

2

R

2

where R and R have the same meaning as be fore, and m is the slope of the regression

0

between predicted and experimental values [63, 67].

A Y-randomization was also performed to eliminate the possibility of chance correlation

in the best MLR models. A randomization experiment was carried out by randomly

shuffling the Y-column values (log (1/MIC)) twenty five times while keeping the

descriptor matrix unchanged. Each randomized model was subsequently used to make

2

predictions for the test set, being the corresponding R and RMSE calculated each time.

2

Both an increase in RMSE and a substantial decrease in R vis-à-vis the non-randomized

model support the robustness and statistical reliability of the original model. To help

2

judging if the difference between the original R and the average R for the randomized

2

models, R_r, was significant, a further statistical parameter, the so-called Todeschini

c

2

R_pparameter, was also computed [78]:

c

2

p

2

r

(7)

R  R (R  R )

c

2

QSAR models that present R_pvalues higher that 0.5 are considered robust and reliable

and not merely a result of chance correlation.

A further important aspect to take into consideration in this type of analysis is the

applicability domain (AD) of the obtained QSAR models, which is essential to

guarantee that they make reliable predictions [62, 64, 65, 79]. To assess the AD in the

developed MLR models, two independent filtering methods were considered: i- the

range of individual descriptors, and ii- the leverage approach. In the first case, the

descriptors‟ range for training set compounds was calculated and the minimum and

maximum values of the descriptors were used to define their interval of validity. Test

set (and designed) compounds descriptors were then compared with the corresponding

average values in the training set to ensure that predictions were performed for

compounds whose molecular descriptors fell within the established limits. In the second

case , the leverage value, h, of each compound was determined as described elsewhere

[

64, 79, 80], thus providing a measure of the distance of the compound from the

centroid of the training set. A “warning leverage”, h*, is generally fixed at 3p/N, where

N is the number of compounds in the training set and p the number of model variables

plus one [64, 79, 80]. This method allows a graphical assessment of the standardized

residuals, SR, as a function of the leverage values (Williams plot) thus permitting a

“better” visualization of the AD and the detection of influential compounds, i.e.,

1

ACCEPTED MANUSCRIPT

compounds structurally distant from training set compounds (h > h*) and of response

outliers (SR >  3 SD units) [79, 80]. In this type of plot the AD corresponds to the

squared area defined by the range  3 SR and the threshold leverage value h* [64, 65,

8

0].

3

. Results and Discussion

3

.1 Chemistry

Synthesis of all compounds was successfully achieved using general organic synthetic

pathways and their chemical structure (please refer to Figure 1) was elucidated by

1

13

extensive spectroscopic analysis ( H and C NMR, COSY, DEPT, HMQC, HMBC,

NOESY and MS).

Isonicotinoyl hydrazones 1, 2, 5-8 were prepared by functionalization of INH at N-2

using standard methods. Typically, treatment of INH with the appropriate ketone or

aldehyde in ethanol at room temperature led to the corresponding Schiff base, which

was generally isolated as a crystalline solid after column chromatography. For

compounds 1 and 2 a 1:0.5 INH/ aldehyde ratio assured the reaction of just one of the

aldehyde functions. Compounds 3 and 4 were obtained by reduction of the parent

compounds 1 and 2 with NaBH in ethanol, and further recrystallization.

4

The stereochemistry of the double bond in hydrazones 1-4 and 7-8 was assigned as

1

synperiplanar E, based on H NMR experiments. It is well known that acylhydrazones

may exist as E/Z geometric isomers about the C=N double bonds and as

synperiplanar/antiperiplanar amide conformers (Figure 2). However, it is reported in

literature that hydrazones derived from aldehydes and substituted hydrazides are present

in solution mainly in the E form [81-83]. GC and HPLC analysis of our compounds

confirmed that, in all cases, only one isomer was present. NOESY experiments were

conducted in DMSO d , and revealed a well-defined cross peak between the CONH

6

proton and the iminic proton, only possible in the E geometry of the double bond.

Another cross-peak between the CONH proton and the H-2/ H-6 protons disclosed a

synperiplanar CO–NH conformation. The existence of an equilibrium in solution of the

major synperiplanar and the minor antiperiplanar CO–NH conformation was also

1

o

observed. H NMR experiments at several temperatures (20, 40, 70 and 90 C) were run

in order to confirm that the double signs with very small intensity observed in the

spectrum, were actually due to the minor antiperiplanar conformer. The coalescence of

o

the signals was complete at 90 C, being reversible when temperature was returned to

2

0 °C, thus confirming that the conversion of the two conformers is the only process of

1

isomerization. The presence of conformers can be obtained only by H NMR

experiments since the interconversion is too fast to be detected by HPLC [81].

The configuration of C=N double bond in the Schiff bases was confirmed by X-ray

diffraction crystallography. As an example, the ORTEP diagram of compound 7 is

presented in Figure 3 showing that the asymmetric unit consists of one Schiff base

molecule and two water molecules. The structure of the compound reveals quasi co-

planarity of the whole molecular skeleton, with a dihedral angle of 2.34° between the

benzene and the pyridine rings. The E configuration of the C(7)=N(3) double bond, is

evidenced by the C(6)-N(2)-N(3)-C(7) torsion angle of -178.54(10). The O(1) atom

and the hydrazone N(3) atom are cis with respect to C(6)-N(2) which confirms the

amide synperiplanar arrangement.

1

2

ACCEPTED MANUSCRIPT

In the crystal structure, the water molecules link the molecules into a three-dimensional

network by the intermolecular H(2N)···O(1W) and O(2)···H(4W) hydrogen bonds

(Figure 3).

Figure 2

Figure 3

Isonicotinoyl hydrazides 9, 10 and 11 were synthesized using standard literature

methods; compound 13 was obtained according to a specific procedure reported in the

literature [84]. Compound 12 was obtained by treatment of the corresponding

hydrazone (compound 6) with NaBH CN, followed by repeated column

3

chromatography, in order to separate the two diastereoisomers formed in the reduction.

To the best of our knowledge, compounds 1 , 3, 4, 6-8, 10 and 12 are herein reported for

the first time. Compounds 5 [85] and 13 [84], although already reported in the literature

were spectroscopically characterized for the first time in this work. Compound 9 [86-

8

8] has also been previously synthesized but, along with compound 13, has no reported

MIC values. Likewise, compound 2 [89], used as a precursor of compound 4, was

described and characterized by others, but as far as we know has never been tested

against Mtb. Compound 11 has a reported MIC value against the BCG [39] strain of 15



g/ mL. Finally, compound 5 has MIC values reported against Mtb H37Rv strain (0.05

g/ mL) and against a INH resistant strain (> 0.75g/ mL) [85].

3

.2 Information Retrieved from QSAR Models

.2.1 RFs and Neural Networks Models

ASNN Results. In the first stage, ASNN models were developed using the total set of

descriptors. The initial number of descriptors was then submitted to an additional

reduction procedure in which descriptors with constant values were removed. Then the

importance of the descriptors for the anti-TB activity was evaluated by Random Forests

and ASNN gradual pruning methods (see Supporting Information, Section 2). New

models were developed with the final selection of descriptors. Results are summarized

in Table 1. The total accuracy Ac for the training sets ranged from 87.2 to 96.5%. The

4

5 compounds in the external test set v were predicted with an accuracy of 66.6-77.8 %.

RF Results. The results of this method for all data sets were statistically similar to the

ASNN results. However, the predictive ability for the test sets was in some cases

(models RF1 and RF3) slightly higher than that for ASNNs models (Table 1). The

observed Sn and Sp values indicate that the models are balanced.

Table 1

To visualize the regions of the chemical space covered by the different training sets and

the overlap of these regions with the compounds of the virtual set, a principal

component analysis was performed. The top 2 principal components were calculated

using the Weka 3.7.3 software [90] with the full set of 3059 initial descriptors, common

to all data sets, for the training sets i, iii, iv (training set ii consists in training set iii and

iv), the test set and the virtual set. The algorithm to calculate principal components does

not take into account any information concerning the activity of the compounds. The

visualization of the chemical space based on the projection of the 2 top principal

components is also another approach to the applicability domain of the models. In this

1

3

ACCEPTED MANUSCRIPT

case it shows that the test set is in the chemical space of the different training sets, and

most of the compounds of the virtual set are also in the same chemical space of the

training sets (Figure 4). The virtual set is concentrated in the center of the chemical

space, a region with high density of compounds particularly from training set iv. An

additional representation is included as Supporting Information (Figure S1) with active

and inactive compounds of training and test sets labeled differently. This representation

shows that the region in the chemical space with higher density of active compounds is

near the center of the chemical space. Moreover, this region also shows a high density

of compounds of the virtual set.

Figure 4

3

.2.2 MLR Models

Table 2 shows the results for the best found MLR models. The 82 initial compounds

were divided in two sets: a training set of 62 compounds and a test set of 20 compounds

(~ a 75 % : 25 % splitting, well within recommended limits). A close inspection of the

training set matrix immediately pinpointed four outliers: compounds 10, 21, 23 and 39

in Table S3. In fact, these compounds show very high MIC values (between 149.32 and

6

00 g/mL) i.e., a very low activity when compared with the remaining 58 compounds

which show an average MIC value of 8.35 g/mL, and were, hence, promptly removed.

Models were built with this new training set and 24 new outliers were sequentially

detected according to the two criteria mentioned in section 2.5 (the identified outliers

were compounds 1, 2, 9, 11-12, 14, 15, 18, 24, 30-31, 33, 35-36, 40, 51, 62-67, 70 and

7

9 in Table S3, i.e., 9 hydrazones and 19 hydrazides). These outliers were also

eliminated, one-by-one, from the training set, and the models refitted after each removal

until we reached the final models for a training set comprising 34 compounds.

Table 2

The figures of merit for all six models, shown in Table 2, fulfill the statistical criteria

set above for both internal and external validation, in particular in the case of the two

best models, models A3 and B3, which conform to the most demanding thresholds and

are able to predict log (1/MIC) for this set of compounds with an average SD of 0.47 log

units.

To remove any possibility of attributing the quality of the statistics of these models to a

chance correlation between the response variable and the descriptors, a Y-randomization

was performed as described in section 2.5 and results are presented in Table 3. As can

be observed, there is a significant decrease in the quality of the randomized models

when compared to the original non-randomized ones, and therefore there seems to be no

chance correlation, as corroborated by the value of ^c

2

p

, quite above the 0.5 threshold

R

value.

Table 3

A close analysis of Table 2 shows that the biological behavior of the studied INH

derivatives against Mtb does not depend on their physicochemical, energetic or

electronic characteristics. Descriptors falling in these categories were found not to be

important to model log (1/MIC) values. Also, the lipophilicity characteristics measured

by Clog P, do not appear as relevant to explain the antitubercular activity of these

1

4

ACCEPTED MANUSCRIPT

derivatives. A similar result had already been reported by us in a previous study [35]

involving the more general class of hydrazide derivatives.

Conversely, steric and geometrical features emerge as very effective in this modeling

process. In fact, models A and B both include the steric Verloop‟s parameters L, B

and B , where L is a measure of the substituent‟s length, B is essentially a measure of

the size of the first atom in the substituent (largely a steric e ffect) and B is an attempt to

define the effective volume of the whole substituent [91, 92]. Concurrently, model A

involves an extra geometrical parameter, d , which refers to the interatomic distance

between the two exocyclic nitrogen atoms, whereas model B contains an additional

parameter, d , which corresponds to the 2D distance between the pyridinic nitrogen and

1

,

5

1

5

3

k

the terminal nitrogen of the hydrazide functionality (see Figure 5 ) [39].

Figure 5

Both models A and B in their complete form, i.e., models A3 and B3, explain virtually

the same variability in the response variable (80.1 % and 79.3 %, respectively) since d₃

and d are highly correlated, although no significant collinearity was observed between

k

2

any pair of parameters in the same equation (r was always below 0.59) or between one

parameter against all others (R was always lower than 0.68), as can be seen by the

2

intercorrelation matrices (Table 4). However, the predictive power of model A3 is

slightly superior to that of model B3, as suggested by the set of measures used to assess

the quality of the predictions for the same test set (Table 2). CCC for model B3 is also

marginally below the suggested threshold value of 0.85. The scatter plot for model A3

represented in Figure 6 shows that there are no systematic deviations from the ideal line

(represented as a dashed line in the figure); in other words, there are no local, scale or

local plus scale shifts in the data. This type of plot allows an easy visual inspection of

the quality of the model, thus precluding the possibility of having good statistical

validation criteria but a “bad” predictive model [75]. Additionally, as already

mentioned, we have tested the predictive power of our models towards a totally external

data set of compounds collected from the Novartis database. These compounds were

selected using the INH core structure as a first filter, a similarity level of 40 % and a

MIC threshold value of less than 400 g/mL. A sub-set of 13 compounds from those

returned from the database was chosen so that the variability of both the dependent and

independent variables fell into the same range as that of compounds in the training and

test sets of the best models. Model A3 did not perform as well as model B3 and so in

Table 2 we present the results retrieved only from model B3 for the Novartis data set.

As can be seen from the metrics displayed and also from Figure 7, model B3 shows a

good predictive ability for the Novartis data set which is a further important criterion to

assess its quality.

Table 4

Figure 6

Figure 7

The six MLR models displayed in Table 2 also perform well in terms of applicability

domain. In fact, training and test set compounds of all models fall within  3 SR,

signaling the absence of any outlier (see Table S5, in Supporting Information).

Moreover, there are also no significant influential compounds in the training sets since

when h > h* (models A1, A2 and B2) the distance to h* is quite small [64, 79]. Figure

8

shows as an example the Williams plot for model A3 (see also Table S5). Three

1

5

ACCEPTED MANUSCRIPT

compounds from the test set (compounds 7, 13 and 49) have leverage values slightly

higher than the cut-off value, h*, being somewhat deviated from the structural centroid

of the training set thus falling slightly outside the model AD. However, their residuals

are close to zero and so they are accurately predicted by the model. Data points such as

these are sometimes called “good high leverage points or “good influential points”[66].

Figure 8

Looking at the descriptors for the two best models, A3 and B3, one can see that higher

values of interatomic distances (d and d ) lead to a decrease in the activity and the

3

K

same effect is observed when the volume of the substituent, measured by B , increases.

5

On the contrary, the molecular specificities encoded by B and L enhance the activity of

1

these INH derivatives against Mtb.

The observation that the chain length of the substituent promotes the activity of these

compounds is indeed remarkable. Actually, the favorable effect on the activity of INH

derivatives caused by an increase in the chain length of the substituent has already been

reported in the literature [41]. The authors suggested that the effect was due to an

enhanced ability of the larger compounds to penetrate the cell membrane. However, as

already referred in our previous studies [35], and again in this work, the lipophilicity

character of the INH derivatives, as measured by ClogP, is not found to be critical to

explain the biological behavior of these compounds against Mtb. In fact, for the 82 INH

derivatives that constitute the object of our study, the findings did not reveal any

significant correlation between Clog P and the biological activity expressed in terms of

log (1/MIC). It should also be noted that the intercorrelation between ClogP and L is

very low (0.248). These observations might thus imply at least one of two hypotheses:

(i) the penetration process through the cell membrane is indeed not determinant for the

antitubercular activity of these compounds; or (ii) the measure of lipophilicity by an

isotropic lipophilicity parameter such as the n-octanol–water partition coefficient is not

an adequate model to mimic these specific drug-membrane interactions [35]. It is

known that there are potential antitubercular compounds with either strong or weak

lipophilic/hydrophilic characteristics [88] and the importance of the

lipophilicity/hydrophilicity balance on the activity of antitubercular agents has been

recognized by various authors [44, 93-95]. However, the relationship between

lipophilicity and activity is not always evident, making it difficult to define a unique and

consensual optimal range for the lipophilicity of a good antitubercular drug candidate.

3

.3 QSAR-Oriented Design and Compound Selection

The obtained classification models from RFs and ASNNs methods were used to

automatically screen a data set of 430 virtual compounds, potentially active against Mtb,

and with distinct substitution patterns - see Table S2. These compounds belong to two

broad INH derived families, the isonicotinoyl hydrazide family and the isonicotinoyl

hydrazone family, and were classified as active/inactive by the eight classification

models. A first ranking of these compounds was achieved taking into account their

average probability to be active, determined by the probability measure of the RFs.

After this initial ranking, two additional pruning operations were performed: i) virtual

compounds classified as inactives by a given classification model were deleted; and ii)

virtual compounds with an average probability of being active lower than 0.8 were also

deleted. As a consequence, the final number of virtual compounds suggested for

1

6

ACCEPTED MANUSCRIPT

synthesis was reduced to 118 (see section 2.1 on Data sets). A further refinement of the

virtual screening using the web interface OSIRIS [96], allowed an evaluation of these

compounds in terms of predicted drug relevant properties such as toxicity

(mutagenicity, tumorigenicity, irritation and reproductive effectiveness), drug-likeness,

drug score, ClogP, solubility and synthetic accessibility. Based on this information, a

subset of 44 compounds was proposed for synthesis. From these, 13 were chosen to be

synthesized due their synthetic feasibility, but were first also evaluated in terms of their

predicted activity by the MLR models referred to in section 3.2.2 (estimated MIC values

had to be  5 M for a compound to proceed for synthesis).

From the best found MLR models, 17 compounds with suitable molecular

characteristics were evaluated, including the 13 proposed by the classification models.

Compounds were designed so that they would lead to high activity values as predicted

by the majority of the models, i.e., so that their descriptors B and L were higher than

1

the average values in the training set, and B and d (or d ) were lower than those

5

3

k

values. This balance was, however, difficult to attain and therefore we also had to take

into consideration the relative weights of the descriptors to explain the antitubercular

activity. Table 5 presents the descriptors values computed for the 17 compounds, as

well as maximum and minimum values for training and test set descriptors and mean

values for training set descriptors. Values in bold correspond to descriptors that fulfill

the specified requisites (see 3.2.2). Based on the 6 MLR models presented in Table 2,

MIC values (expressed in M) were subsequently estimated for each of these

compounds and results are shown in Table 6. Using the same activity threshold as

above, 15 INH derivatives with MIC  5 M, as predicted by at least four MLR models,

which concomitantly passed the referred OSIRIS filters, were selected as potential

antitubercular agents (in bold in Table 6). From these, 10 compounds (including 6

suggested by the consensus approach) were chosen to be synthesized due to their greater

synthetic workability, namely compounds 1 to 10, in bold and underlined in Table 6.

However, before progressing with the synthesis, we verified the compliance of the

designed compounds with the applicability domain of the derived models by

introducing them in the Williams plot represented in Figure 8 for model A3. All the

synthesized compounds lie well within  3 SD and the great majority of them are

structurally close to training set compounds as their leverage values are lower than the

warning leverage value, being therefore within the AD of the models. Obvious

exceptions are compounds 5 and 9 which, despite being reliably predicted (low

standardized residuals), only display h < h* for two of the models (see Table S5), and

hence are farther from the training set structural centroid. Yet, this procedure reinforced

the decision to go on with the synthesis, since compounds 5 and 9 fall into the category

of “good high leverage points” according to [66] (see section 3.2.2). The Williams plot

for model B3 is presented in Figure S2 as Supporting Information. Similar trends are

observed for all the represented sets, being one of the Novartis compounds structurally

more distant from the model centroid than all the others, even if it is the most-well

predicted compound from the set.

Although compounds 12 and 13, shown in Tables 5 and 6, have only been synthesized

for comparative purposes, it is easily seen from Figure 8 (and Figure S2 in SI) and

Table S5 that they have h >> h*, and are therefore structurally remote from the model

centroid, besides having relatively high SD, close to the lower limit of the AD. This

observation is consistent with their behavior in terms of descriptors‟ range and MIC

values (see Tables 5 and 6).

1

7

ACCEPTED MANUSCRIPT

Table 5

Table 6

From the 10 selected compounds only two belong to the isonicotinoyl hydrazide family

(compounds 9 and 10) the remaining being isonicotinoyl hydrazones. In terms of the

considered molecular descriptors (Table 5), these two compounds differ from all others

especially in the interatomic distance, d , and in the length of the substituent, L.

3

On the other hand, these 10 compounds have very distinct lipophilicity characteristics,

with experimental logP_o_/_wranging from 0.220 (compound 5) to 3.700 (compound 8)

[

88, 97]. This large spectrum of lipophilicity values in compounds with high predicted

activity corroborates the aforementioned statement that lipophilicity is not per se a good

criterion to judge the suitability of a new compound as an antitubercular agent.

In order to further extend the variability of logP, we have also synthesized compound

1

3, a high hydrophilic compound just as INH (logP_o_/_w(INH) = -0.851[88]) even if its

descriptors do not follow, in general, the tendency expected to promote activity (please

refer to Table 5), as can be confirmed by the high estimated MIC values in Table 6.

Furthermore, to be able to compare the effect upon the activity of having a hydrazone or

the corresponding hydrazide analogue, we have likewise synthesized compound 12, the

reduced counterpart of compound 6. Again, estimated MIC values for compound 12 are

much higher than the established threshold value of 5 M, although its logP_o_/_wvalue

lies within the limits of lipophilicity defined by the other compounds.

3

.4 Biological Activity

The synthesized compounds were tested in vitro against three Mtb strains: the H37Rv

T

ATCC27294 strain, a Mtb clinical strain with a katG S315T mutation and the INH

T

resistant strain H37Rv ATCC27294 (H37RvINH) with a full deletion in the katG gene,

as referred in section 5.2.

Results are shown in Table 7 where INH, the reference compound, is introduced for

comparative purposes.

Table 7

3

.4.1 Activities against the wt strain

The ten designed compounds show experimental MIC values (expressed in M units)

against the wild type strain of Mtb very close to values predicted by all MLR models

which give predictions for these compounds < 5 M (column 3, Table 7) – i.e., by a

consensus prediction [36, 38] – and by the two best MLR models, models A3 and B3

(column 5, Table 7). The mean average deviations (AE) are, respectively, 0.25 and 0.36.

Within experimental uncertainty, compounds are more active than predicted in all cases,

with the exception of compound 10.

A very relevant result is that five out of the 10 compounds (compounds 1, 2, 4, 5 and 6)

for which the majority of descriptors clearly contribute to enhance the activity (see

Table 5), show a measured activity against H37Rv higher than that of INH, measured

1

8

ACCEPTED MANUSCRIPT

with the same protocol, BACTEC 960, in the same laboratory (please refer to section

5

.2) - Figure 9. These findings clearly refute the arguments of some authors who claim

that INH analogous backbones never achieve the (in vitro) efficiency of INH of which

they would be nothing else than mere precursors, thus postulating that INH is not a lead

compound [94]. If the action of these derivatives would be to simply produce,

completely and readily, INH, as referred [94], a higher activity than INH would never

be observed.

Figure 9

Compared to the previous compounds, compound 3, rather unexpectedly based solely

on its descriptor values, is approximately twice less active than INH. Additionally,

compounds 7 to 9 have a very unfavorable B (taken as the effective volume of the

5

whole substituent) which seems to cancel out the positive effects of the other

descriptors. Compound 10, on the other hand, has a consensus predicted activity of 1.44



M but is the less active compound of this set. These observations cannot be

rationalized in terms of simple electronic effects, which in any case did not appear as

relevant descriptors in any of the best models. The behavior of these compounds is most

probably related with the steric characteristics of the R substituents and their ability or

incapability to promote the formation of the isonicotinoyl radic al post ulated to be

formed in the generally accepted mechanism of action of INH [18, 24, 98, 99]. In fact, it

is reported in the literature that access to the heme active site of KatG poses steric

constraints and therefore any factor influencing the stereochemistry of these compounds

should be relevant [100, 101]. Some authors [102] suggest that the formation of the INH

+

analogues adducts with NAD requires acyl radicals sufficiently long-lived, and that, for

aromatic hydrazides, the nature of substituents should play an important role in the

stabilization of radical intermediates involved in the overall process of INH activation.

A comparison between the hydrazone compound 6 and the corresponding hydrazide

compound 12, shows that the hydrazone is more active than its analogue (0.27 M vs.

0

.44 M), as already perceived by some authors [39] and contested by others [94],

based on the assumption that MIC measurements in the two situations were not

standardized, which is clearly not the case here. The higher activity of hydrazones is

possibly due to a higher efficacy in the formation, in loco, of the prodrug INH, as a

result of hydrazone hydrolysis, and the subsequent production of the acyl radical, as a

result of INH activation by KatG.

The biological behavior of the isonicotinoyl hydrazide compounds 12 and 13, is very

similar, with reasonably low experimental MIC values (0.44 and 0.42 M, respectively)

in spite of their (not surprisingly) high estimated MIC values (41.61 and 59.24 M,

respectively). These compounds show, however, very different lipophilic characteristics

as given by their logP_o_c_t_/_w(Table 5), which, again, further supports the inference that

there is no clear relationship between lipophilicity and in vitro activity.

To summarize, the information in Tables 5 and 6 shows that N-N distances (d and d )

3

k

and the length of the substituent (L) are the properties with the most impact in the

activity of these compounds. Shorter distances and lengthy substituents appear to favor

the activity by allegedly facilitating the production of the isonicotinoyl radical.

3

.4.2 Activities against resistant strains

1

9

ACCEPTED MANUSCRIPT

Given the results obtained for the Mtb wt strain, we carried out the testing of these

compounds against INH mono-resistant strains whose mechanism of resistance is

related only to the diminished effectiveness of KatG. Values in Table 7 reveal that

seven of the synthesized compounds are more effective than INH against the mutated

S315T resistant strain, i.e., compounds 3 to 9, with compounds 7 to 9 showing a

significant decrease in MIC - Table 7 and Figure 10. Among these seven INH

derivatives, six are isonicotinoyl hydrazones, being compound 9 the only isonicotinoyl

hydrazide. However, it is precisely this compound that presents the most significant

difference in activity showing a six fold increase in activity by comparison with INH.

Figure 10

It has been reported in the literature that katG (S315T) confers resistance to INH

through subtle changes in the INH binding site [103], but without losing its ability to

bind INH [98, 104]. Moreover, it has been described that the addition of a methyl group

in the mutated katG actually constricts the accessibility to the heme active site by

closing down the dimensions of the narrowest part of the access channel from 6 Å in the

wt katG to 4.7 Å in the S315T mutated katG [100-102]. This steric effect is taken as

interfering with the interaction between katG and INH, leading some authors to suggest

that INH binding within the heme pocket is most probably a pre-requisite for INH

activation [101]. Furthermore, these steric effects on INH binding allegedly have an

impact upon INH oxidation, thus decreasing the conversion of INH to a biologically

active form [105, 106]. However, the effect of any perturbation on INH binding upon

drug activation needs to be further analyzed [102].

Preliminary docking studies of INH with both wt and mutated KatG carried out by us

have shown that the S315T mutation subtly narrows the accessibility channel to the

catalytic centre of KatG but the in silico binding constants of INH to both strains turned

out to be similar within the method uncertainty. Although the exact binding site of INH

to the CP-KatG wt enzyme (and for that matter, also to the KatG (S315T)) is still

controversial [102] and needs further investigation, recent studies [100, 103, 107]

indicate that it might be close to the KatG heme pocket, at about 12 Å from the heme

iron for both enzymes. Therefore, any factor affecting the stereochemistry of a

compound might be important to rationalize its activity, especially in the mutated strain.

These evidences make the performance of compound 9 in the mutated strain even more

peculiar because its molecular structure, with a long alkyl chain, does anticipate large

steric constrictions. To have a measure of the importance of the alkyl group upon the

antitubercular activity, we have synthesized the acetyl analogue, compound 11, (N'-

acetylisonicotinohydrazide), which showed a very low activity both in the wt strain and

in the mutated strain. Hence, the C10 chain of compound 9 seems indeed crucial to

enhance its mycobacterial activity. Some authors have also reported improved activities

for other long chain hydrazides but there seems to be no consi stenc y between the length

of the chain and the corresponding MIC value [41, 87, 108]. Regarding the

effectiveness of compound 9, our hypothesis is that the active site in the mutated strain

might be located in a more peripheral area of the KatG enzyme and/or that the

compound-enzyme interaction may be more efficient.

Since we do not expect that a substitution of a serine by a threonine at amino acid

residue 315 shall introduce very significant modifications in the active site of KatG,

molecular dynamics studies to model the behavior of both compounds, i.e., INH and

compound 9, vis-à-vis the wild type and the mutated strain, as well experimental

2

0

ACCEPTED MANUSCRIPT

determinations of binding constants in these four situations are underway and will

hopefully give us further insights into the real molecular picture of these drug-target

interactions.

Results presented in this work seem to suggest that redesign of the INH molecule in

order to improve drug binding may indeed be a feasible approach to overcome

resistance in katG (S315T), the most frequent and important cause of INH resistance

worldwide [109]. In support of this hypothesis is the total ineffectiveness of the new

INH derivatives against the H37RvINH (ΔkatG) that suffered a full deletion of the katG

gene after prolonged exposure to INH (Table 7) which confirms the importance of the

activation role of KatG upon INH and its derivatives.

New INH derivatives redesigned to take advantage of the diminished but still effective

activation of INH by the mutated katG, would thus allow the maintenance in the

therapeutic regimen of one of the two most active and effective anti-TB drug known so

far, i.e., isoniazid [28, 110].

3

.5 Cytotoxicity assays

In parallel with the antitubercular activity evaluation, all compounds with MIC values

against the H37Rv strain < 0.5 M were screened towards African green monkey

kidney cells (VERO cells) using the MTT assay, in order to determine their in vitro

cytotoxicity (IC ) and to establish the selectivity index (SI). Results showed that these

5

0

compounds were non-toxic to the host cells in the concentration range studied (1MIC-

00MIC, with MIC in g/mL) – Table 8. Isoniazid was also analyzed and had a IC₅₀

value > 200. The SI was determined as the ratio of the measured IC to the measured

5

0

IC (H37Rv Mtb MIC value) and was in all cases > 400 (for INH, SI > 500). SI is

9

commonly used to estimate the therapeutic window of a dr ug and to identify drug

candidates for further studies. According to literature [49, 111, 112] candidates for new

drugs must have a SI 10, with MIC values lower than 6.25 g/mL and a low

cytotoxicity, as is indeed the case for all tested compounds (compounds 1-6 and 9).

Table 8

4

. Conclusions

The use of a rational design approach based on well-validated classification and

regression QSAR models has allowed a successful synthesis of several potent

antitubercular isoniazid derivatives.

Compounds 1, 2, 4, 5 and 6 showed an activity against H37Rv higher (MIC  0.28 M)

than the one exhibited by the reference compound, INH (MIC = 0.29 M). It is

postulated that the behavior of these compounds is probably dependent on

stereochemical aspects associated with the nature of the substituents and on their ability

to promote the formation (and stabilization) of the isonicotinoyl radical in the overall

process of activation by KatG. As suggested in our previous works, lipophilicity was

again not identified as an important factor to describe the activity of these compounds,

which was now further confirmed by the absence of correlation between measured

logP_o_/_wand MIC values.

2

1

ACCEPTED MANUSCRIPT

On the other hand, compounds 3 to 9 showed to be more effective than INH against a

Mtb clinical strain carrying only a katG S315T mutation. In particular, compound 9,

with an alkyl chain in C10, showed a six fold increase in activity against this resistant

strain. This finding seems to question the recently proposed relationship between the

increased resistance of katG (S315T) to INH and a putative larger steric constraint in the

access channel to the heme active site in the mutated strain. Experimental and in silico

studies to evaluate the interaction of compound 9 with the biological target in both

strains are now being carried out and will be reported in due course.

Finally, the ineffectiveness of all the new INH derivatives against the INH high-level

resistant strain H37Rv_I_N_H(ΔkatG) with a full deletion of the katG gene, indubitably

attests the crucial role of KatG in the activation process of these compounds.

In summary, our results clearly demonstrate that using a properly validated QSAR

methodology it is possible to design more active compounds departing from the INH

core structure and overcome the decreased susceptibility of the katG mutated strains

towards INH.

2

ACCEPTED MANUSCRIPT

5

. Experimental Section

.1 Chemistry

.1.1 General

All solvents and reagents were obtained from commercial suppliers and were used

without further purification. The reactions were monitored by thin layer

chromatography performed on silica gel 60 F254 aluminum sheets and visualized with

UV light (254 nm). Flash column chromatographies were performed on silica-gel 60,

4

0-63 m. Melting points were taken in open capillary tubes using an apparatus Stuart®

SMP 30 or in a Reichert Thermovar Microscope and are uncorrected. IR spectra were

obtained using a Mattson Satellite FTIR and only the diagnostic absorption bands were

-1

1

13

reported in cm . H NMR (400 MHz) and C NMR spectra (100.6 MHz) were

,

recorded on a Bruker AMX spectrometer; chemical shifts were expressed as δ values

and referenced to the residual solvent peak (DMSO-d ,  = 2.50); coupling constants

6

H

were reported in units of Hertz (Hz). HRMS were recorded in a mass spectrometer

Apex Ultra FTICR-MS. The purity of all compounds ( 98 %) was assessed by GC-MS

analysis using a ThermoQuest 8000 series chromatograph coupled with a MS Trio

1

000-Fisions [DB-1 coated capillary column (5m x 0.53 mm x 2.65m) using N as

2

o o o

carrier gas; the GC oven was programmed at 120 C (2 min), heated to 200 C at 10 C

-1

o

-1

min , held 2min at 200 C, further heated to 280 C at 20 C min and held at final

temperature for 30 min] or a Surveyor Plus HPLC system (Thermo Finnigan) equipped

with a Surveyor Autosampler Plus with a 100 L loop, a quaternary pump Surveyor LC

Pump Plus, and a diode-array detector Surveyor PDA Plus Merck [column

Lichrospher100 RP-18 column(250 × 4 mm, particle size 5 m), gradient elution (40

%

acetonitrile to 100% 0.1 % TFA over 20 min and return to the initial conditions

within 13 min)].

5

.1.2 General procedure for the synthesis of hydrazones 1 and 2

Isoniazid (1 equiv.) was added in portions over a period of three hours to a solution of

the dialdehyde (1 equiv.) in absolute ethanol. The mixture was stirred at room

temperature for 24 h, the resulting precipitate was filtered off and the mother liquor

concentrated to afford a crude product which was further purified by flash

chromatography and recrystallization.

5

.1.2.1 N'-(E)-(3-methanoylbenzylidene)isonicotinohydrazide (compound 1)

White crystals (flash column chromatography with EtOAc/MeOH 93:7 and

o

recrystalization from ethanol/water 1:1), yield: 41 % (1.027 g), mp 210.1-211.5 C. IR

-

1

(

KBr):  (cm ) = 3300 (NH amide), 3045-2854 (Ar C-H), 1676 (CO amide and

1

aldehyde), 1550 (CN imine). H NMR (DMSO d ): δ = 12.39 (s, 1H, NH), 10.01 (s,

6

1

7

H, CHO), 8.81 (AA‟ part of a AA‟XX‟system, 2H, J = 4.6, 1.4 Hz; H-2, H-6), 8.56(s,

H, CH=N), 8.29 (s, 1H, H-2‟), 8.07 (d, 1H, J = 7.8 Hz), 8.00 (d, 1H, J = 7.6, H-4‟),

.85 (XX‟ part of a AA‟XX‟system, 2H, J = 4.6, 1.5 Hz, H-3, H-5), 7.72 (t, 1H, J = 7.7

1

3

Hz, H-5‟). C NMR (DMSO d ): δ = 193.0 (CHO) 161.8 (NHCO), 150.3 (C-2, C-6),

6

1

47.6 (CHN), 140.3 (C-4), 136.6 (C-3‟), 135.0 (C-1‟), 132.9 (C-6‟), 130.8 (C-4‟), 129.8

2

3

ACCEPTED MANUSCRIPT

+

(

C-5‟), 127.8 (C-2‟), 121.5(C-3, C-5). HRMS: m/z 254.09231 [M+H] (calcd. for

C H N O +H 254.09240).

14

11

3

2

5

.1.2.2 N'-(E)-(4-methanoylbenzylidene)isonicotinohydrazide (compound 2)

o

White crystals, mp 222.1-223.3 C [lit. 219-220 ºC] [89]. HRMS: m/z 254.09214

+

[

M+H] (calcd. for C H O N +H 254.09240).

14 11 2 3

5

.1.3 General procedure for the synthesis of hydrazones 3 and 4

1

equiv of NaBH was added, portion wise, to a solution of compounds 1 or 2 dissolved

4

in 96 % ethanol. After stirring the reaction mixture for 2 h at room temperature, water

was added and the resulting precipitate was filtered off to afford a crude product which

was further purified by recrystallization.

5

.1.3.1 N'-(E)-3-(hydroxymethyl)benzylidene)isonicotinohydrazide (compound 3)

o

White crystals (recrystallization from water), yield: 43 % (0.108 g), mp 230.4-232.3 C.

-1

IR (KBr):  (cm ) = 3373 (OH), 3202 (NH amide), 3034-2897 (Ar C-H), 1671 (CO

1

amide), 1568 (CN imine). H NMR (DMSO d ): δ = 12.08 (s, 1H, NH), 8.80 (AA‟ part

6

of a AA‟XX‟system, 2H, J = 5.6 Hz, H-2, H-6), 8.47 (s, 1H, CHN), 7.84 (XX‟ part of a

AA‟XX‟system, 2H, J = 5.8 Hz, H-3, H-5), 7.76 (s, 1H, H-2‟), 7.60 (d, 1H, J = 7.1 Hz,

1

3

H-6‟), 7.41 (m, 1H, H-5‟), 4.56 (d, 2H, J = 5,3 Hz CH OH). C NMR (DMSO d ):): δ =

2

6

1

61.6 (NHCO), 150.3 (C-2, C-6), 149.1 (CHN), 143.3 (C-3‟), 140.4 (C-4), 133.8 (C-1‟),

28.6 (C-4‟), 126.4 (C-5‟), 126.0 (C-6‟), 124.7 (C-2‟), 121.5 (C-3, C-5), 62.5 (CH OH).

2

+

HRESIMS: m/z 256.10806 [M+H] (calcd. for C H N O +H 256.10805).

1

4

13

3

2

5

.1.3.2 N'-(E)-4-(hydroxymethyl)benzylidene)isonicotinohydrazide (compound 4)

White crystals (recrystallization from EtOH/H O 1:1), yield: 29 % (0.119 g), mp 220.1-

2

o

-1

2

1

21.3 C. IR (KBr):  (cm ) = 3399 (OH), 3300 (NH amide), 3045-2854 (Ar C-H),

667 (CO amide), 1567 (CN imine). H NMR (CD OD): δ = 8.77 (AA‟ part of a

1

3

AA‟XX‟system, 2H, J = 4.6, 1.5 Hz, H-2, H-6), 8.38 (s, 1H, CHN), 7.91 (XX‟ part of a

AA‟XX‟system, 2H, J = 4.4, 1.6 Hz, H-3, H-5), 7.84 (AA‟ part of a AA‟XX‟system,

2

5

H, J = 8.2 Hz, H-2‟, H-6‟), 7.46 (XX‟‟ part of a AA‟XX‟system, 2H, J= 8.0, H-3‟, H-

1

3

‟), 4.67 (s, 2H, CH OH). C NMR (CD OD): δ = 163.2 (NHCO), 150.3 (CHN), 149.7

2 3

(

5

C-2, C-6), 144.6 (C-4‟), 141.1 (C-4), 132.8 (C-1‟), 127.7 (C-2‟, C-6‟), 126.8 (C-3‟, C-

+

‟), 121.8 (C-3, C-5), 63.3 (CH OH). HRESIMS: m/z 256.10805 [M+H] (calcd. for

2

C H N O +H 256.10805).

14

13

3

2

5

.1.4 General procedure for the synthesis of hydrazones 5-8

A solution of the aldehyde or ketone (1 equiv.) in absolute ethanol was added dropwise

to a solution of isoniazid (1 equiv.) in water. The mixture was stirred at room

temperature for 24 h and the resulting solid collected and washed with cold water. Flash

2

4

ACCEPTED MANUSCRIPT

column chromatography over silica-gel and/or recrystallization afforded pure

compounds.

5

.1.4.1 N'-cyclopentylideneisonicotinohydrazide (compound 5)

o

White crystals (recrystallization from water), yield: 63 % (1.877 g), mp 182.3-183.2 C.

-1

IR (KBr):  (cm ) = 3200 (NH amide), 3084-3013 (Ar C-H), 2976-2816 (aliph. C-H),

1

8

658, (CO amide), 1645 (CN imine). H NMR (DMSO d ): δ = 10.58 (s, 1H, NH),

6

.73 (AA‟ part of a AA‟XX‟system, 2H, J = 5.6 Hz, H-2, H-6), 7.72 (XX‟ part of a

AA‟XX‟ system, 2H, J = 5.6Hz, H-3, H-5), 2.44 (t, 2H, J = 6.8Hz, H-5‟), 2.41 (t, 2H, J

1

3

=

6.8 Hz, H-2‟), 1.72 (m, 4H, H-3‟, H-4‟). C NMR (DMSO d ): δ = 172.5 (C-1‟),

6

1

5

62.2 (NHCO), 150.5 (C-2; C-6), 141.7 (C-4), 122.2 (C-3; C-5), 33.6 (C-2‟), 29.4 (C-

‟), 24.8 (C-3‟*), 24.7 (C-4‟*). HRESIMS: m/z 204.11308 [M+H] (calcd. for

+

C H N O+H 204.11314).

11

13

3

5

.1.4.2 N'-(4-phenylcyclohexylidene)isonicotinohydrazide (compound 6)

White crystals (recrystallization from ethanol/water 1:2), yield: 74 % (3.252 g), mp

o

-1

1

(

78.3-180.3 C (dec.). IR (KBr):  (cm ) = 3443 (NH amide), 2935 (Ar C-H), 1665,

1

CO amide), 1636 (CN imine). H NMR (DMSO d ): δ = 11.0 (s, 1H, NH), 8.74

AA‟ part of a AA‟XX‟system, 2H, J = 4.6; 1.2 Hz, H-2, H-6), 7.76 (XX‟ part of a

6

AA‟XX‟ system, 2H, J = 4.6, J = 1.2 Hz, H-3, H-5), 7.29 (m, 4H, H-2‟‟, H-3‟‟, H-5‟‟,

H-6‟‟), 7.19 (m, 1H, H-4‟‟), 3.04 (d, 1H, J = 13.7 Hz, H-6 ‟), 2.89 (t, 1H, J = 12.08 Hz,

eq

H-4‟), 2.56 (d, 1H, J = 13.6 Hz, H-2 ‟), 2.46 (ddd, 1H, J = 13.6, 13.6, 4.86 Hz, H-2 ‟),

eq

ax

2

.09 (ddd, 1H, J = 13.7, 13.7, 5.24 Hz, H-6 ‟), 1.96 (dt, 1H, J = 12.8 Hz, H-5 ‟), 1.71

ax eq

(

dddd, 1H, J = 12.7, 12.7, 4.24 Hz, H-3 ‟), 1.61 (dddd, 1H, J = 12.8, 12.8, 3.6 Hz, H-

ax

13

5_a_x‟), 2.03 (m, 1H, H-3 ‟). C NMR (DMSO d ): δ = 166.7 (C-1‟), 161.8 (NHCO),

eq

6

1

5

2

50.1 (C-2; C-6), 145.7 (C-1‟‟), 141.1 (C-4), 128.4* (C-2„‟, C-6‟‟), 126.7* (C-3„‟, C-

‟‟), 126.1 (C-4‟‟), 121.7 (C-3; C-5), 42.4 (C-4‟), 34.8 (C-2‟), 33.9 (C-3‟), 33.0 (C-5‟),

+

7.3 (C-6‟). HRESIMS: m/z 294.15969 [M+H] (calcd. for C H N O+H 294.16009).

1

8

19

3

5

.1.4.3 Methyl 4-[(E)-(isonicotinoylhydrazone)methyl] benzoate (compound 7)

White crystals (recrystallization from acetone), yield: 59 % (1.808 g,), mp 216.2- 217.1

o

-1

C. IR (KBr):  (cm ) = 3262 (NH amide), 3000-3126 (Ar C-H), 2956- 23848 (aliph.

1

C-H), 1716 (CO ester), 1665 (CO amide), 1604 (CN imine). H NMR (DMSO d ):

6

δ = 12.26 (s, 1H, NH), 8.81 (AA‟ part of a AA‟XX‟system, 2H, J = 4.4, 1.5 Hz, H-2‟,

H-6‟), 8.52 (s, 1H, CHN), 8.05 (AA‟ part of a AA‟BB‟system, 2H, J =8.3 Hz, H-2, H-

6

), 7.90 (BB‟‟ part of a AA‟BB‟ system, 2H, J = 8.3 Hz, H-3, H-5), 7.84 (XX‟ part of a

1

3

AA‟XX‟system, 2H, J = 4.4, 1.5 Hz, H-3‟, H-5‟), 3.88 (s, 3H, OCH₃). C NMR

(

DMSO d ): δ = 166.3 (CH OCO), 162.3 (NHCO), 150.9 (C-2‟; C-6‟), 148.0 (CHN),

6 3

1

40.7 (C-4‟), 138.9 (C-4), 131.2 (C-1), 130.2 (C-2, C-6), 127.9 (C-3, C-5), 122,0 (C-3‟,

+

C-5‟), 52.8 (OCH ). HRESIMS: m/z 284.10687 [M+H] (calcd. for C H O N +H

3

15 14

3

2

84.10297).

5

.1.4.4 N'-(E)-(4-phenoxybenzylidene)isonicotinohydrazide (compound 8)

2

5

ACCEPTED MANUSCRIPT

White crystals (flash column chromatography with EtOAc and recrystallization from

o

-1

ethanol/water), yield: 77 % (1.552 g), mp 175.8– 176.8 C. IR (KBr):  (cm ) = 3421

(

1

NH amide), 3103-3038 (Ar C-H), 1659, (CO amide),1607 (CN imine). H NMR

DMSO d ): δ = 12.65 (s, 1H, NH), 8.79 (AA‟ part of a AA‟XX‟system, 2H, J = 5.9 Hz,

(

6

H-2, H-6), 8.45 (s, 1H, CHN), 7.83 (XX‟ part of a AA‟XX‟system, 2H, J = 5.9 Hz, H-3,

H-5), 7.77 (AA‟ part of a AA‟XX‟system, 2H, J = 8.7 Hz, H-2‟, H-6‟), 7.45 (AA‟ part

of a AA‟BB‟C system, 2H, J = 7.9 Hz, H-3‟‟, H-5‟‟), 7.21 (C part of a AA‟BB‟C

system, 1H, J = 7.4 Hz, H-4‟‟), 7.10 (BB‟ part of a AA‟BB‟C system, 2H, J = 7.7 Hz,

1

3

H-2‟‟), 7.07 (XX‟ part of a AA‟XX‟system, 2H, J = 8.6 Hz, H-3‟, H-5‟). C NMR

(

DMSO d ): δ = 162.0 (NHCO), 159.3 (C-4‟), 156.1 (C-1‟‟), 150.8 (C-2, C-6), 148.9

6

(CHN), 141.0 (C-4), 130.7 (C-3‟‟, C-5‟‟), 129.7 (C-2‟, C-6‟), 129.4 (C-1‟), 124.7 (C-

4

3

‟‟), 122.0 (C-3, C-5), 120.0 (C-2‟‟, C-6‟‟), 118.7 (C-3‟, C-5‟). HRESIMS: m/z

+

18.12909 [M+H] (calcd. for C H O N +H 318.12370).

1

9

16

2

3

5

.1.5 N'-decanoylisonicotinohydrazide (compound 9)

A solution of isoniazid (4 g, 29.2 mmol) and N-methylmorpholine (3.2 mL, 29.1 mmol)

in acetonitrile (65 mL) was stirred and heated to reflux. Then, a solution of decanoyl

chloride (6.0 mL, 28.9 mmol) in acetonitrile (15 mL) was added dropwise, and the

reaction mixture kept at reflux for 2 h. After cooling to room temperature, the

precipitate was filtered off and washed with water.

Several recrystallizations from ethanol/water 2:8 and n-hexane/ethanol 10:2.5 afforded

compound 9.

o

White crystals, yield: 35 % (2.977 g), mp 118.4-119.2 C [lit. 110-112 ºC] [87].

+

HRESIMS: m/z 292.20190 [M+H] (calcd. for C H N O +H 292.20195).

1

6

25

3

2

5

.1.6 N’-(2-methylheptanoyl)isonicotinohydrazide (compound 10)

2

-methylheptanoic acid (10 mL, 61 mmol) was refluxed with 35.8 mL (487 mmol) of

thionyl chloride for two hours. The excess of thionyl chloride was removed under

reduced pressure and the crude 2-methylheptanoyl chloride was used in the next step

without further purification. A solution of isoniazid (8.36 g, 60.9 mmol) in acetonitrile

(136 mL) was stirred and heated to reflux. Then, a solution of 2-methylheptanoyl

chloride (8.96 g, 60.31 mmol) in acetonitrile (30 mL) was added dropwise, and the

reaction mixture kept at reflux for 1 h and left overnight at room temperature. The

reaction mixture was neutralized with a 25 % solution of ammonium hydroxide and the

formed precipitate washed with water and filtered off.

o

-1

White solid, yield: 30.6 % (4.866 g), mp 122.2-123.5 C. IR (KBr):  (cm ) = 3194 (N-

1

H amide), 3027 (ArC-H), 2954-2930 (aliph. C-H), 1602 (CO amide). H NMR

(

DMSO d ): δ =10.79(s, 1H, pyCONH), 9.57 (s, 1H, NHCOCH ), 8.66 (d, 2H, J = 5.0

6

2

Hz, H-2, H-6), 7.63 (d, 2H, J = 5.7 Hz, H-3, H-5), 2.47 (ddd, 1H, J = 13.8; 6.8 Hz, H-

2

3

‟), 1.65 (m, 1H, H-3 ‟), 1.42 (m, 1H, H-3 ‟), 1.25 (s, 6H, H-4‟, H-5‟, H-6‟), 1.17 (d,

H, J = 6.80 Hz, H-8‟), 0.86 (t, 3H, J = 5.9 Hz, H-7‟); C NMR (DMSO d ): δ = 175.1

6

a b

1

3

(

3

C-1‟), 162.6 (pyCONH), 150.4 (C-2, C-6), 138.3 (C-4), 121.1 (C-3, C-5), 39.1 (C-2´),

4.0(C-3‟), 31.7(C-5‟), 27.0 (C-4‟), 22.5 (C-6‟), 17,61(C-8‟), 14.0 (C-7‟). HRESIMS:

m/z 264.107748 [M+H]+(calcd. for C14H21N3O2+H 264.107653).

2

6

ACCEPTED MANUSCRIPT

5

.1.7 N'-acetylisonicotinohydrazide (compound 11)

Acetic anhydride (1.8 mL, 19.0 mmol) was added dropwise to a solution of isoniazid (2

g, 14.6 mmol) in water (20 mL). The reaction mixture was stirred at room temperature

for 30 min and the resulting solid filtered off, washed with dichloromethane and

recrystallized from water.

o

White crystals, yield: 53.5 % (1.399 g), mp 163.0-163.9 C, lit. [162.0-163.0 C] [113].

+.

+

EIMS (probe) 70 eV: m/z 179 [M] , 137 [C H N O] ,106 [C H NO] , 78 [C H N] , 51

6

7

3

6

4

5

4

+

[

C H ] .

4 3

5

.1.8 N’-(4-phenylcyclohexyl)isonicotinohydrazide (compound 12)

A solution of methanolic 5M HCl was added, dropwise, to a solution of compound 6

1.004 g, 3.4 mmol) and NaBH CN (143 mg, 2.3 mmol) in 50 mL of methanol until

(

3

pH3 was reached. The solution was stirred for 3 h, and methanol was removed under

reduced pressure. The residue was flash chromatographed over silica-gel

(

acetone/CH Cl 3:7). Fractions containing the reduced compound were further

2 2

chromatographed using hexane/acetone 6:4 as eluent.

o

-1

White crystals, yield: 10 % (0.107 g), mp 114-115 C (dec.). IR (KBr):  (cm ) = 3296

(

NH amide), 2922 (Ar C-H), 1674, (CO amide). 1H NMR (CDCl ): δ = 8.73 (AA‟ part

3

of a AA‟XX‟system, 2H, J = 4.6; 1.4 Hz, H-2, H-6), 7.65 (XX‟ part of a AA‟XX‟

system, 2H, J = 4.6, J = 1.2 Hz, H-3, H-5), 7.30* (m, 2H, H-3‟‟, H-5‟‟), 7.28* (m, 2H,

H-2‟‟, H-6‟‟), 7.20 (tt, 1H, J = 7.1; 1.5 Hz H-4‟‟), 3.36 (m, 1H, H-1‟), 2.60 (m, 1H, H-

1

3

4

‟), 1.94** (m, 4H, H-2‟, H-6‟), 1.69 (m, 4H, H-3‟, H-5‟). C NMR (CDCl ): δ = 165.6

3

(NHCO), 150.6 (C-2; C-6), 147.1 (C-1‟‟), 140.2 (C-4), 128.4* (C-2„‟, C-6‟‟), 126.9*

(

2

C-3„‟, C-5‟‟), 125.9 (C-4‟‟), 120.9 (C-3; C-5), 53.9 (C-1‟), 43.7 (C-4‟), 29.1** (C-3‟),

+

8.0** (C-2‟). HRESIMS: m/z 296.17583 [M+H] (calcd. for C H N O+H

18

21

3

96.175739).

5

.1.9 2-hydrazinoisonicotinohydrazide (compound 13)

Compound 13 was prepared according to a procedure reported in literature [84]. Yellow

o

crystals (recrystallization from water), yield: 25 % (0.270 g), mp 215.8-216.9 C, [lit.

-

1

2

(

1

(

16.5-217.0 ºC]. IR (KBr):  (cm ) = 3312-3123 (NH amide and amine), 3020-2920

1

Ar C-H), 1653 (CO amide). H NMR (DMSO d ): δ = 9.84 (s, 1H, NHNH ), 8.03 (dd,

H, J =5.2, 0.4 Hz, H-6), 7.64 (s, 1H, NHNH ), 7.08 (s, 1H, H-3), 6.85 (dd, 1H, J =5.2,

6

2

1

3

.6 Hz, H-5). C NMR (DMSO d ): δ = 164.8 (NHCO), 162.4 (C-2), 147.8 (C-6) 141.4

6

+

C-4), 109.3 (C-5), 104.2 (C-3). HRESIMS: m/z 168.08797[M+H] (calcd. for

C H N O+H 168.08799).

6

9

5

.1.10 Crystallographic structure determination

Single crystal X-ray analysis of a suitable quality crystal of compound 7 was performed

with a Bruker AXS-KAPPA APEX II diffractometer using graphite-monochromated

Mo-K radiation ( = 0.71069 Å) and operating at 50 kV and 30 mA. A single crystal

was mounted loop with protective oil and placed under a flow of nitrogen gas at 150 K.

All measurements were conducted on the Crystallography Laboratory at Instituto

2

7

Article Doi

Design, synthesis and biological evaluation of novel isoniazid derivatives with potent antitubercular activity

DOI: 10.1016/j.ejmech.2014.04.077

Source and publish data:

Authors:

Article abstract of DOI:10.1016/j.ejmech.2014.04.077

Full text of DOI:10.1016/j.ejmech.2014.04.077

Products guided by the article

R&D Labs maybe for 1616082-21-3

Relevant to this article

Hot Product