Bioorganic & Medicinal Chemistry 22 (2014) 1568–1585 Contents lists available at ScienceDirect Bioorganic & Medicinal Chemistry journal homepage: www.elsevier .com/locate /bmc Antiprotozoan lead discovery by aligning dry and wet screening: Prediction, synthesis, and biological assay of novel quinoxalinones http://dx.doi.org/10.1016/j.bmc.2014.01.036 0968-0896/� 2014 Elsevier Ltd. All rights reserved. ⇑ Corresponding author. Tel./fax: +34 963543156. E-mail addresses: ymarrero77@yahoo.es, ymponce@gmail.com (Y. Marrero-Ponce), escario@farm.ucm.es (J.A. Escario), vjaran@iqm.csic.es (V.J. Arán). URL: http://www.uv.es/yoma/ (Y. Marrero-Ponce). Miriam A. Martins Alho a, Yovani Marrero-Ponce b,d,c,⇑, Stephen J. Barigye b, Alfredo Meneses-Marcel b, Yanetsy Machado Tugores b, Alina Montero-Torres b, Alicia Gómez-Barrio f, Juan J. Nogal f, Rory N. García-Sánchez f,h, María Celeste Vega f,g, Miriam Rolón f, Antonio R. Martínez-Fernández f, José A. Escario f, Facundo Pérez-Giménez d, Ramón Garcia-Domenech d, Norma Rivera i,j, Ricardo Mondragón i,j, Mónica Mondragón i,j, Froylán Ibarra-Velarde k, Atteneri Lopez-Arencibia l, Carmen Martín-Navarro l, Jacob Lorenzo-Morales l, Maria Gabriela Cabrera-Serra l, Jose Piñero l, Jan Tytgat e, Roberto Chicharro m, Vicente J. Arán n a CIHIDECAR (CONICET), Departamento de Química Orgánica, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, C1428EGA Buenos Aires, Argentina b Unit of Computer-Aided Molecular ‘Biosilico’ Discovery and Bioinformatic Research (CAMD-BIR Unit), Faculty of Chemistry-Pharmacy, Universidad Central ‘Marta Abreu’ de Las Villas, Santa Clara 54830, Villa Clara, Cuba c Environmental and Computational Chemistry Group, Facultad de Química Farmacéutica, Universidad de Cartagena, Cartagena de Indias, Bolivar, Colombia d Unidad de Investigación de Diseño de Fármacos y Conectividad Molecular, Departamento de Química Física, Facultad de Farmacia, Universitat de València, Spain e Laboratory of Toxicology, University of Leuven (KULeuven), Campus Gasthuisberg, O&N2, PO Box 922, Herestraat 49, 3000 Leuven, Belgium f Departamento de Parasitología, Facultad de Farmacia, Universidad Complutense, 28040 Madrid, Spain g Centro para el Desarrollo de la Investigación Científica (CEDIC), Fundación Moisés Bertoni/Díaz Gill Medicina Laboratorial, Pai Perez 1165, Asunción, Paraguay h Laboratorio de Investigación de Productos Naturales Antiparasitarios de la Amazonía, Universidad Nacional de la Amazonía Peruana, Pasaje Los Paujiles s/n, A A.H H Nuevo San Lorenzo, San Juan Bautista, Iquitos, Peru i Departamento de Bioquímica, Centro de Investigaciones y Estudios Avanzados del IPN, Av. Instituto Politécnico Nacional No 2508, Col. San Pedro Zacatenco, México DF 07360, Mexico j Departamento de Microbiología y Parasitología, Facultad de Medicina, UNAM, México DF 04510, Mexico k Department of Parasitology, Faculty of Veterinarian Medicinal and Zootecnic, UNAM, Mexico DF 04510, Mexico l Instituto Universitario de Enfermedades Tropicales y Salud Pública de Canarias, University of La Laguna, Avda. Astrofísico Fco. Sánchez, S/N, 38203 La Laguna, Tenerife, Canary Islands, Spain m Instituto de Química Orgánica General, CSIC, c/Juan de la Cierva 3, 28006 Madrid, Spain n Instituto de Química Médica, CSIC, c/Juan de la Cierva 3, 28006 Madrid, Spain a r t i c l e i n f o Article history: Received 29 May 2012 Revised 13 January 2014 Accepted 21 January 2014 Available online 31 January 2014 Keywords: In silico study TOMOCOMD–CARDD software Non-stochastic and stochastic linear indices Classification model Machine learning-based QSAR Antiprotozoan database In vitro assay Antimalarial Antitrypanosomal Antitoxoplasma Antitrichomonas Leishmanicide Cytotoxicity a b s t r a c t Protozoan parasites have been one of the most significant public health problems for centuries and sev- eral human infections caused by them have massive global impact. Most of the current drugs used to treat these illnesses have been used for decades and have many limitations such as the emergence of drug resistance, severe side-effects, low-to-medium drug efficacy, administration routes, cost, etc. These drugs have been largely neglected as models for drug development because they are majorly used in countries with limited resources and as a consequence with scarce marketing possibilities. Nowadays, there is a pressing need to identify and develop new drug-based antiprotozoan therapies. In an effort to overcome this problem, the main purpose of this study is to develop a QSARs-based ensemble classifier for antipro- tozoan drug-like entities from a heterogeneous compounds collection. Here, we use some of the TOMO- COMD-CARDD molecular descriptors and linear discriminant analysis (LDA) to derive individual linear classification functions in order to discriminate between antiprotozoan and non-antiprotozoan com- pounds as a way to enable the computational screening of virtual combinatorial datasets and/or drugs already approved. Firstly, we construct a wide-spectrum benchmark database comprising of 680 organic chemicals with great structural variability (254 of them antiprotozoan agents and 426 to drugs having other clinical uses). This series of compounds was processed by a k-means cluster analysis in order to design training and predicting sets. In total, seven discriminant functions were obtained, by using the whole set of atom-based linear indices. All the LDA-based QSAR models show accuracies above 85% in the training set and values of Matthews correlation coefficients (C) vary from 0.70 to 0.86. The external http://crossmark.crossref.org/dialog/?doi=10.1016/j.bmc.2014.01.036&domain=pdf http://dx.doi.org/10.1016/j.bmc.2014.01.036 mailto:ymarrero77@yahoo.es mailto:ymponce@gmail.com mailto:escario@farm.ucm.es mailto:vjaran@iqm.csic.es http://www.uv.es/yoma/ http://dx.doi.org/10.1016/j.bmc.2014.01.036 http://www.sciencedirect.com/science/journal/09680896 http://www.elsevier.com/locate/bmc M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1569 validation set shows rather-good global classifications of around 80% (92.05% for best equation). Later, we developed a multi-agent QSAR classification system, in which the individual QSAR outputs are the inputs of the aforementioned fusion approach. Finally, the fusion model was used for the identification of a novel generation of lead-like antiprotozoan compounds by using ligand-based virtual screening of ‘avail- able’ small molecules (with synthetic feasibility) in our ‘in-house’ library. A new molecular subsystem (quinoxalinones) was then theoretically selected as a promising lead series, and its derivatives subse- quently synthesized, structurally characterized, and experimentally assayed by using in vitro screening that took into consideration a battery of five parasite-based assays. The chemicals 11(12) and 16 are the most active (hits) against apicomplexa (sporozoa) and mastigophora (flagellata) subphylum parasites, respectively. Both compounds depicted good activity in every protozoan in vitro panel and they did not show unspecific cytotoxicity on the host cells. The described technical framework seems to be a promis- ing QSAR-classifier tool for the molecular discovery and development of novel classes of broad—antipro- tozoan—spectrum drugs, which may meet the dual challenges posed by drug-resistant parasites and the rapid progression of protozoan illnesses. � 2014 Elsevier Ltd. All rights reserved. 1. Introduction only 3 (and 20 additional in a second study), 19, and 40 know drugs Diseases caused by tropical parasites affect hundreds of mil- lions of people worldwide, mainly distributed in tropical and sub- tropical regions. In fact, parasitic diseases have been one of the most significant public health problems for centuries with note- worthy mortality and devastating social and economic conse- quences. Parasites belonging to phylum protozoa are the most important causal pathogens and cause several human infections with globally massive impact. For instance, malaria (Plasmodium spp.),1 leishmaniasis (Leishmania spp.),2 trypanosomiasis (Trypano- soma brucei [sleeping sickness]3 and Trypanosoma cruzi [Chagas disease]4) as well as giardiasis5/amebiasis6 (Giardia lamblia/Ent- amoeba histolytica) are among the main neglected parasitic dis- eases with great social impact. Trichomoniasis, one of the most common sexually transmitted diseases (with around 120 million vaginitis infections worldwide every year) caused by the flagellate protozoa Trichomonas vaginalis, is increasingly recognized as an important infection in women and men.7Other serious disease caused by a related apicomplexan parasite, Toxoplasma gondii, has gained increasing relevance in immunocompromised patients, such as patients with transplants, cancer, or AIDS, and in congeni- tally infected infants.8 Although most of the current anti-protozoan drugs are well known and broadly used in medical treatments, most of them are decades old and have many limitations, including the emergence of drug resistance, severe side-reactions (toxicity), low-to-medium efficacy, limitations in the routes of administration, price and other important inconveniences. These drawbacks of cur- rent antiprotozoan chemotherapy make the search for new drugs an urgent need. However, the development of such drugs has been largely neglected because they are intended for the treatment of pathologies that mainly affect poor people in regions of the world with limited resources and with scarce marketing possibilities, par- ticularly in today’s post-merger climate. Nevertheless, the search for antiprotozoan compounds is now on the desktop of medicinal chemists and great efforts to reinvig- orate the drug development pipeline for these diseases are being addressed by new consortia of scientists from the academy and industry, which are driven in large part by support from major phi- lanthropies.9 Recently, using whole-organism screening with com- pounds derived from libraries containing drugs already approved for human use (with other therapeutic use, but ‘off-label’ like anti- parasitic efficacy), a few hits were identified in diversity screening assays against T. brucei, Plasmodium falciparum and leishmania.10–13 In this ‘trial-and-error’ search for antiprotozoan drug-like com- pounds a lot of chemicals had to be experimentally screened (>15,000) and the efficacy of this process was very low, yielding with efficacy equal to or greater than that of the drugs used cur- rently against leishmania-, malaria- or trypanosoma-reference (control) compounds, respectively.10–13 In addition to the low effi- ciency of this type of drug discovery landscape, the usually expen- sive and time consuming approaches impose on us the necessity to develop alternative and more rational techniques in the classical— trial and error—screenings. In order to reduce costs, pharmaceutical companies have to find new technologies in the quest of new chemical entities (NCE), where an in silico ‘virtual’ world of data, analysis and computer- aided molecular design can be seen as an adequate alternative to the ‘real’ world of synthesis and screening of compounds in the laboratory. By such means, ‘the expensive commitment to actual synthesis and bioassay is made only after exploring the initial concepts with computational models and screens’. In silico screening is now incorporated in all areas of lead discovery; from target identifica- tion and library design, to hit analysis and compound profiling. This theoretical(dry)-to-experimental(wet) integration procedure will be used here in order to find predictive models that permit the ‘rational’ identification of new antiprotozoan drug-like compounds. 1.1. Background-review of TOMOCOMD–CARDD method in drug discovery for parasitic diseases: meeting the challenge Some of our research teams have previously reported several antimicrobial-chemoinformatic studies to drive the selection of novel chemicals as promising NCEs. In these studies, the TOMOCOMD–CARDD (acronym of TOpological MOlecular COMputational Design–Computer-Aided ‘Rational’ Drug Design) method14 and linear discriminant analysis (LDA), have been used in order to parameterize molecules in a database and for develop- ing classification functions, respectively. The LDA is one of most important and simple (supervised, linear and parametric) patter recognition technique that can be used to determine which vari- ables discriminate between two or more naturally occurring groups. The TOMOCOMD–CARDD approach is a novel scheme the rational—in silico—molecular design and Quantitative Structure Activity/Property Relationships (QSAR/QSPR).15–19 It calculates several new families of 2D,3D-Chiral (2.5) and 3D (geometric and topographic) non-stochastic and (simple and double) stochastic (as well as canonical forms) atom- and bond-based molecular descriptors (MDs), denominated quadratic, linear and bilinear indi- ces in analogy to the quadratic, linear and bilinear mathemati- cal.15–19 For instance, the TOMOCOMD–CARDD strategy has been used for the in silico screening of novel molecular subsystems hav- ing a desired activity against Trichomonas vaginalis.19–21 It was also 1570 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 successfully applied to the virtual (computational) screening of no- vel antihelmintic compounds, which were then synthesized and evaluated in vivo on Fasciola hepatica.22 Studies for the fast-track discovery of novel paramphistomicides,17 antimalarial,23,24 and antitrypanosomal/leishmania4,25,26 compounds were also con- ducted with this theoretical method. As result of previous studies developed by our group focused on the synthesis and activity of several families of heterocyclic beta- ines and salts, we have prepared many indazole,27–30 indole,31 cinnoline32 and quinoxaline33 derivatives, which have shown interesting properties as trichomonacidal,19–21,30 antichaga- sic,25,26,30 antimalarial24 and antineoplastic28–30 drugs. Nowadays, the effort for the search of novel antiprotozoan drugs has increased considerably. However, do effective broad spectrum antiparasitic agents exist? Therapeutics that are efficient against most of the parasitic species are interesting (and very important) because in regions of the world where these parasites are endemic, they indeed do overlap, and several infections can oc- cur at the same time and sometimes with similar symptoms. We initially have developed ‘general’ QSAR models (based on activity datasets comprising diverse compounds corresponding to a num- ber of mechanisms of action) to describe and predict the individ- ual-antiprotozoan activity.4,19,20,23–26,34 Nonetheless, by using this approach a different model must be used to predict specific antipar- asitic activities for a given set of chemicals for each of the antipro- tozoan species. For this reason, is important to develop a more general model, which includes all chemicals reported as active against any protozoan parasite. This strategy will allow us, to ob- tain general models with a broad application domain (antiprotozo- an space) and maybe, to discover drug-like agents with possible broad spectrum antiparasitic activity. In this report, we will explore the potential of TOMOCOMD– CARDDMDs to seek a QSARs-based ensemble classifier for antipro- tozoan drug-like compounds obtained from a heterogeneous series of compounds. In the first step, we selected a wide-spectrum data- base of antiprotozoan drugs, which include active compounds against all kinds of protozoan parasites with diverse action modes. Next, the aforementioned MDs were calculated for this large series of active/nonactive compounds and LDA was subsequently used to fit every individual classification function. The LDA was selected as statistical technique due to its broad use and simplicity. Later on, we developed a multi-agent QSAR classification system (ensemble classifier), in which the individual QSAR outputs are the inputs for the fusion approach. Finally, the fusion model was used for the identification of a novel generation of lead-like antiprotozoans by using ligand-based virtual screening (LBVS) of smallmolecules ‘available’ (with synthetic feasibility) in our ‘in-house’ library. A new molecular sub-system was then theoretically selected as promising lead series, which were subsequently synthesized, structurally characterized, and experimentally assayed. Here, we also describe the original synthesis and the spectroscopic charac- terization of 10 molecules (new quinoxalinones) that had not been previously reported. The in vitro screening carried out here was de- signed by taking into account a battery of assays that included the two most representative parasites of the protozoa subphylum: (1) mastigophora (flagellata) and (2) apicomplexa (sporozoa). These ‘parasite-based’ assays are suitable for describing a rather complete profile of antiprotozoan activities of these new chemicals. Recently, a study based on a multi-pathogen screening strategy, integrating activity and cytotoxicity data for the selection and prioritization of lead compounds, has been reported in the literature.35 While this approach presents a plausible advantage, particularly when exper- imental screening is performed on a structurally diverse data set, it is imperative that the analyzed data be obtained under homoge- nous experimental conditions. 2. Results and discussion 2.1. In silico studies Three different computational experiments were developed in this study. Firstly, we present the result obtained in the construc- tion of classification models and their assemblies by using a fusion approach (multiagent-system). Each individual model was evalu- ated based according to the Organization for Economic Coopera- tion and Development (OECD) principles.36 Later, we describe the selection of new leads by using LBVS as well as the preparation of these new chemicals for simple and efficient methods of synthe- sis. Finally, the biological characterization against four different species of protozoan parasites will be presented in order to close the lead discovery cycle (experimental corroboration). 2.2. Discussion on the classification-based general QSAR for the description of antiprotozoan activity The development of discriminant functions that allows the classification of organic-chemical drugs as active or inactive is the key step in the present approach for the discovery of new wide-spectrum antiprotozoan agents. It was therefore necessary to select a training data set of active and inactive compounds containing broad structural variability and action modes, as well as therapeutic uses. It is well-know that the general performance and extrapolation power of the learning methods decisively depends on the selection of compounds for the training series used to build the classifier mod- el. For this reason, and with the purpose of ensuring molecular and pharmacological diversity, we have selected a benchmark dataset composed by a great number of molecular entities, some of them re- ported as antiprotozoan37,38 and the rest with a series of other phar- macological uses.37,38 We consider a large database of 680 drugs having great structural variability; 254 of them are active (antipro- tozoan agents) and the others are non-antiprotozoan (426 com- pounds having other clinical uses, such as antivirals, sedative/ hypnotics, diuretics, anticonvulsivants, haemostatics, oral hypogly- cemics, antihypertensives, antihelminthics, anticancer compounds and so on). The classification of these compounds as ‘inactive’ (with- out antiprotozoan activity) does not guarantee that any of these compounds present antiparasitic activities not yet detected. Initially, two k-means cluster analyses (k-MCA) were performed for active and inactive series of chemicals, which permitted split- ting the dataset into training (learning) and predicting (test) series. All cases were processed by using k-MCA in order to design train- ing and predicting data series in a ‘‘rational’’ way. The main idea consists in carrying out a partition of either active or inactive series of chemicals in several statistically representative classes of chem- icals. Thence, one may select from the members of all these classes of training and predicting series. This procedure ensures that any chemical class (as determined by the clusters derived from k- MCA) will be represented in both series of compounds. Then, selec- tion of the training and prediction sets was performed by randomly selecting compounds belonging to each cluster. The training set was composed by 204 antiprotozoans and 300 inactives from a set of 680 chemicals (504, �75%). The resting group composed of 50 actives and 126 compounds with different biological activities was prepared as a test data set for validation of the models (all molecules in the four groups as a zip file). These 176 (�25%) drugs were never used in the development of the classification models. For these sets of compounds, two atom-based TOMOCOMD– CARDD MDs families (kth order non-stochastic [APfk ð�xÞ ] and stochastic [APsfk ð�xÞ] linear indices) were computed.16,22,26,34,39 These linear maps use a complete atomic properties (AP) scheme, M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1571 which characterizes a specific aspect of the atomic structure. All indices were calculated for H-atom explicit molecular graphs, that is, APf H k ð�xÞ and APsf H k ð�xÞ for non-stochastic and stochastic linear indices, respectively. Two local (L) atom-type indices for heteroat- oms (group = heteroatoms (E): E = S, N, O), not considering [APfkL ð�xEÞ] and considering [APfkL ð�xEÞ ] H-atoms in the molecule, were computed as well. All Classification-based QSAR equations were derived by using forward stepwise LDA and all the set of total and local atom-based linear computed indices are shown below: Class ¼ �3:84� 3:14 � 10�4Mf H 5 ð�xÞ þ 2:79 � 10�2Mf1ð�xÞ þ 4:19 � 10�3Mf2ð�xÞ þ 2:72 � 10�8Mf12ð�xÞ � 2:45 � 10�3Mf H 4Lð�xEÞ þ 4:23 � 10�6Mf H 10Lð�xEÞ � 2:40 � 10�8Mf14Lð�xEÞ ð1Þ Class ¼ �3:97� 2:32 � 10�5Pf H 8 ð�xÞ þ 6:23 � 10�3Pf5ð�xÞ � 1:87 � 10�4Pf9ð�xÞ þ 6:47 � 10�6Pf12ð�xÞ � 6:55 � 10�8Pf15ð�xÞ þ 2:37 � 10�6Pf H 11Lð�xEÞ � 1:46 � 10�8Pf15Lð�xEÞ ð2Þ Class ¼ �4:03� 1:34 � 10�9Vf H 14ð�xÞ þ 3:37 � 10�3Vf1ð�xÞ þ 8:23 � 10�9Vf13ð�xÞ � 2:47 � 10�3Vf H 4Lð�xEÞ þ 1:78 � 10�7Vf H 12Lð�xEÞ þ 1:84 � 10�2Vf2Lð�xEÞ � 4:12 � 10�9Vf15Lð�xEÞ ð3Þ Class ¼ �3:84� 1:36 � 10�4Kf H 8 ð�xÞ þ 3:42 � 10�5Kf H 9 ð�xÞ þ 0:27Kf0ð�xÞ � 6:76 � 10�3Kf3ð�xÞ � 6:96 � 10�2Kf H 2Lð�xEÞ þ 3:76 � 10�5Kf H 9Lð�xEÞ � 1:71 � 10�8Kf15Lð�xEÞ ð4Þ Class ¼ �4:06þ 2:8 � 10�8Mf12ð�xÞ � 4:53 � 10�8Pf15Lð�xEÞ þ 1:34 � 10�7Vf H 12Lð�xEÞ þ 9:23 � 10�3Vf2Lð�xEÞ � 1:36 � 10�5Kf H 8 ð�xÞ þ 0:14Kf0ð�xÞ � 6:35 � 10�2K f H 2Lð�xEÞ ð5Þ Class ¼ �3:13� 5:28 � 10�2Msf H 2 ð�xÞ þ 0:26Msf2ð�xÞ � 0:18Msf10ð�xÞ þ 0:10Msf H 1Lð�xEÞ � 5:46 � 10�2Msf1Lð�xEÞ � 0:20Msf H 2Lð�xEÞ þ 0:15Msf14Lð�xEÞ þ 3:73Msf3Lð�xH�EÞ ð6Þ Class ¼ �4:00þ 0:74Psf H 1 ð�xÞ � 0:72Psf H 2 ð�xÞ � 0:56Psf H 11ð�xÞ þ 0:87Psf4ð�xÞ � 2:12Psf H 1Lð�xEÞ þ 1:12Psf H 2Lð�xEÞ þ 1:31Psf H 3Lð�xEÞ ð7Þ Class ¼ �3:79þ 0:14Vsf H 0 ð�xÞ � 0:08Vsf H 2 ð�xÞ � 0:03Vsf H 7 ð�xÞ � 0:73Vsf H 3 ð�xÞ þ 1:94Vsf H 5Lð�xEÞ þ 0:16Vsf H 6Lð�xEÞ � 1:30Vsf H 7Lð�xEÞ � 0:07Vsf0Lð�xEÞ ð8Þ Class ¼ �4:27þ 3:25Ksf H 5 ð�xÞ � 3:23Ksf H 7 ð�xÞ þ 2:97Ksf H 1Lð�xEÞ þ 3:34Ksf H 6Lð�xEÞ � 1:62Ksf1Lð�xEÞ � 4:29Ksf6Lð�xEÞ � 13:81Ksf H 6Lð�xE�HÞ þ 60:01Ksf H 10Lð�xE�HÞ � 46:26Ksf H 12Lð�xE�HÞ ð9Þ Class ¼ �3:93� 9:25 � 10�2Msf2Lð�xEÞ � 0:98Psf H 1Lð�xEÞ þ 0:18Vsf H 6Lð�xEÞ þ 5:98 � 10�2Vsf0Lð�xEÞ þ 10:99Ksf H 10Lð�xE�HÞ � 11:23Ksf H 12Lð�xE�HÞ ð10Þ Class ¼ �3:97� 1:89 � 10�3Mf H 4Lð�xEÞ þ 2:59 � 10�2V f2Lð�xEÞ � 6:96 � 10�2K f H 2Lð�xEÞ þ 9:24 � 10�6K f H 9Lð�xEÞ � 0:72Psf H 1Lð�xEÞ þ 0:10Vsf H 7Lð�xEÞ ð11Þ In total eleven models were obtained, the first four Eqs. (1)–(4) developed with the non-stochastic bond-based linear indices and models 6–9 obtained with the stochastic MDs. The overall perfor- mances of all the obtained models are given in Table 1, together with the Wilks’ statistics (k), the square of the Mahalanobis dis- tances (D2), and the Fisher ratio (F). The selected models are statis- tically significant at p-level <0.001. Table 1 also shows the result obtained for the Eqs. 5 and 10 in both cases (non-stochastic and stochastic molecular fingerprints) resulting in a combination of all pairs of atom weights (atomic labels). In addition, the Eq. 11 was carried out by using the entire MDs set (mixing non-stochastic and stochastic linear indices) and was the best models in the learn- ing set (see Table 1). As can be observed in Table 1, the fitted models 5 and 10, result- ing of the combination of weighting schemes for the non-stochas- tic and stochastic atom-level linear indices, respectively, as well as the Eq. 11 (mixing non-stochastic and stochastic indices) exhibit the best results. These best two equations based on both individual set of linear indices (Eqs. 5 and 10) correctly classified 91.27% of the training set, and showed values of the Matthews correlation coefficients (C) of 0.82. However, Eq. 5 (non-stochastic linear indi- ces) showed more false positive rate than Eq. 10, fitted by using only stochastic MDs. Even then, the best result was obtained when all MD sets were used. The Eq. 11 showed 93.06% of global good classification and a C of 0.86. The most common parameters in medical statistics for all the models are depicted in the same Ta- ble 1. Classifications of every compound in the learning series are shown in Tables SI1 and SI2, respectively, of Supporting informa- tion (see structures as zip file). Likewise a plot of the DP% (see Sec- tion 4) for the entire training set by using the best model 11, is illustrated in Figure 1. Another crucial problem in chemometric and QSAR studies is the definition of the Applicability Domain (AD) of a classification or regression model. ‘Not even a robust, significant, and validated QSAR model can be expected to reliably predict the modeled property for the entire universe of chemicals. In fact, only the predictions for chemicals falling within this domain can be considered reliable and not extrapolations model’.40 The AD is a theoretical region in chem- ical space, defined by the model descriptors and modeled response, and thus by the nature of the chemicals in the training set, is rep- resented in each model by specific MDs. That is to say, the AD of the QSAR model is ‘the range within which it tolerates a new molecule’. Figure 2 shows the Williams plot for the AD of Eq. 11. As can be noted in Figure 2, almost all chemicals used lie within this area. Actually, some chemicals have leverage (h) values much higher than the threshold but show residuals within the limits, for exam- ple in the test set, Trypan red (h = 0.371) and Dithiophos (h = 0.156). These active and inactive compounds are outside the AD of this model and these chemicals can influence model parameters. Considering this fact, we must check the effect of withdrawal of these compounds on the model performance. When we studied the new parameters of the model after removal of these chemicals we detected no significant variations. Therefore, the influence of these compounds was not critical neither for model parameters nor its performance. Consequently, their removal was not justified. In addition, Sch 18545 (antiprotozoan with h of 0.113) and Sicc- amid (nonantiprotozoan with h of 0.109) had higher h in the train- ing set. However, these compounds presented residuals rather lower than the previous ones. Given that these chemicals are in 0 100 200 300 400 500 Chemicals -100.00 -80.00 -60.00 -40.00 -20.00 0.00 20.00 40.00 60.00 80.00 100.00 Δ P% False Positive False Negative Figure 1. Plot of the DP% from Eq. 11 for every compound in the training set. Compounds 1–204 and 205–504 are active and inactive, respectively. Table 1 Prediction performances and statistical parameters for LDA-based QSAR models in the training set Eqs. Atomic labelsa Matthews corr. coeff. Accuracy ‘Qtotal’ (%) Specificity (%) Sensitivity ‘hit rate’ (%) False ‘+’ rate (%) Landa Wilks D2 F Non-Stochastic linear indices 1 (M) 0.72 86.71 83.41 83.82 11.33 0.49 4.28 73.36 2 (P) 0.80 90.48 91.49 84.31 5.33 0.49 4.29 73.66 3 (V) 0.79 89.88 90.05 84.31 6.33 0.493 4.24 72.75 4 (K) 0.80 90.48 91.05 84.80 5.667 0.467 4.72 80.83 General model (combining all atomic labels) 5 (NS) 0.82 91.27 92.11 85.78 5.00 0.467 4.98 80.83 Stochastic linear indices 6 (M) 0.70 85.71 84.02 79.90 10.33 0.60 2.73 40.83 7 (P) 0.76 88.49 88.83 81.86 7.00 0.52 3.77 64.63 8 (V) 0.79 89.68 88.38 85.78 7.67 0.52 3.74 56.02 9 (K) 0.76 88.29 87.96 82.35 7.667 0.51 3.92 52.06 General model (combining all atomic labels) 10 (SS) 0.82 91.27 93.48 84.31 4.00 0.46 4.83 96.93 Mixing all MDs (non-stochastic and stochastic indices) 11 (NS–SS) 0.86 93.06 92.89 89.71 4.67 0.435 5.35 107.3 Bold values represents the Best models. a M: atomic mass, P: atomic polarizability, K: atomic Mulliken electronegativity, V: van der Waals atomic volume.51 NS, SS and NS–SS means non-stochastic MDs, stochastic MDs and whole set of MDs, respectively. 1572 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 the same experimental space as other 20 cases in the training set which slightly exceed the critical hat value (vertical line), they are slightly influential in the model development: the predictions for new compounds in this sense (for instance, included in an external test set, where there are 13 cases that slightly exceed the critical h⁄ value) can be considered as reliable as those of the training chemicals and the possible erroneous prediction could probably be attributed to wrong experimental data rather than to the molecular structure. Finally, two compounds Myralact (r = 3.09) and Tosulur sodium (r = 3.187), which are cases of training and test sets, depicted outlier behavior with standardized residuals greater than three standard deviation units. That is to say, both chemicals were wrongly predicted (>3r); these two compounds as well as the initial two chemicals (Trypan red and Dithiophos) are completely outside the AD of the model. Thus, there are only four compounds that are either a response outlier or a high leverage chemical. Therefore, the model can be used with high accuracy in this AD. In the next section we re-take this analysis in order to determine the reliability of prediction of selected mol- ecules as good candidates by virtual screening protocols. Statistical validation of models is another key feature in good QSAR practice regarding with the diagnosis of developed models. In this sense, a QSAR model should be associated with appropriate measures of goodness-of-fit and robustness (internal validation), as well as predictivity (external validation). The evaluation of perfor- mance of models by using external validation (one or more exter- nal test sets) is viewed as a superior alternative because the good behavior of models in internal experiments is a necessary but not sufficient condition for the model to have high predictive power. That is, the predictivity can be claimed only if the model is success- fully applied in the prediction of the external chemicals, which were not used in the model development. For this reason, in this 0,00 0,03 0,05 0,08 0,11 0,16 0,20 0,28 0,37 0,40 Leverage -4,00 -3,00 -2,19 -1,54 -1,00 -0,46 0,09 0,63 1,17 1,71 2,39 3,00 4,00 St d. R es id ua ls Training set Test set Figure 2. William plot of Eq. 11: outlier will be chemicals atpoints with standardized residuals greater than three standard deviation units; influential chemicals are points with high leverage values higher than the threshold or cut-off value h⁄ = 0.042. The training and test sets are represented by blues circles and red squares, respectively. M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1573 report we describe the external performance evaluation by using a prediction set of active and inactive compounds. The key parameters for statistical diagnostic of all obtained models are presented in Table 2. As can be observed, the predictive performance for LDA-based QSAR models in the test set was ade- quate. Here, the results show that the equations obtained with non-stochastic indices are better than models derived with sto- chastic MDs. In addition, the best LDA-based QSAR is the Eq. 11, with an accuracy of 92.05% versus 85.80% depicted by model 5. Fi- nally, the classification of every compound in prediction series is Table 2 Prediction performances for LDA-based QSAR models in the test set Eqs. Atomic labelsa Matthewscorr. coeff. Accuracy‘Qtotal ’ (%) Non-stochastic linear indices (NS) 1 (M) 0.66 85.23 2 (P) 0.61 83.52 3 (V) 0.61 84.09 4 (K) 0.66 84.09 General model (combining all atomic labels) 5 (NS) 0.67 85.80 Stochastic linear indices (S) 6 (M) 0.35 71.02 7 (P) 0.41 74.43 8 (V) 0.57 81.25 9 (K) 0.42 73.30 General model (combining all atomic labels) 10 (SS) 0.52 79.55 Mixing all MDs (non-stochastic and stochastic indices, NS–S) 11 (NS–SS) 0.81 92.05 Bold values represents the Best models. a M: atomic mass, P: atomic polarizability, K: atomic Mulliken electronegativity, V: van MDs and whole set of MDs, respectively. illustrated as Supporting information (Tables SI3 and SI4, see struc- tures as zip file). Likewise a plot of the DP% (see Section 4) for the entire test set by using the best models 11, is shown in Figure 3. 2.3. Drug (lead)-like discovery by virtual (in silico) screening and dry selection: to be or not to be The ligand-based methods are supported by the principle of similarity—similar compounds are assumed to produce similar effects—and serve for modeling the complex phenomena of Specificity (%) Sensitivity ‘hit rate’ (%) False ‘+’ rate (%) 70.69 82.00 13.49 68.42 78.00 14.29 71.15 74.00 11.90 66.67 88.00 17.46 71.19 84.00 13.49 49.23 64.00 26.19 54.24 64.00 21.43 63.93 78.00 17.46 52.17 72.00 26.19 62.07 72.00 17.46 83.33 90.00 7.14 der Waals atomic volume.51 NS, SS and NS–SS means non-stochastic MDs, stochastic 0 20 40 60 80 100 120 140 160 180 Chemicals -100.00 -80.00 -60.00 -40.00 -20.00 0.00 20.00 40.00 60.00 80.00 100.00 P% Δ False Positive False Negative Figure 3. Plot of the DP% from Eq. 11 for every compound in the test set. Compounds 1–50 and 51–176 are active and inactive, respectively. 1574 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 molecular recognition. Similarity-based methods are cornerstones of chemoinformatic and computer-aided pharmaceutical research. To this effect, LBVS has been used to identify novel active compounds in many biological applications. This indicates that ‘similarity’ methods should have substantial ‘selectivity’ in recog- nizing diverse active compounds. Current purposes to integrate chemoinformatics into ‘real-life’ applications, to step-ahead in drug discovery are of main importance nowadays. The algorithm described above, and the obtained good results prompted us to make in silico evaluations of all the chemicals con- tained in our ‘in-house’ collections of indazole, indazolols, indole, cinnoline, and quinoxaline derivatives (as well as other new re- lated chemicals and their derivatives), which have been recently obtained by our chemical synthesis team. On the basis of com- puter-aided predictions we selected potential antiprotoazoan leads (virtual hits). The following criteria were used for the hits selection: (1) compounds were selected as hits if the value of posterior prob- ability of possessing antiprotozoan activity exceeded 15% (DP >15%) with all LDA-based QSAR models (fusion approach or mul- ti-classification system), and (2) If, among the compounds de- signed (or those that would be obtained in our laboratory) by our chemical team, too many similar compounds satisfied criterion 1, then only several representative structures were selected. Here, we performed in silico mining of our library and some heterocyclic leads were identified (selected) as novel antiprotozo- an compounds by using the discriminant functions obtained through the TOMOCOMD–CARDD method and LDA data-mining technique as an ensemble classifier, CE. That is, here every individ- ual classifier (CI) is fused into the CE through a voting system, where the outputs of CI are used as inputs for CE, which will have a voting score for the query molecules M (for more detail see Sec- tion 4). To provide an intuitive picture, a flowchart to show how these CI are fused into the CE is given in Figure 4. One series of compounds (quinoxalinones derivatives) was se- lected as antiprotozoan lead-like compounds, showing good agree- ment between the in silico predictions and in vitro assays in several cell (parasite)-based tests (see more below). The values of DP% for this subset are depicted in Table 3. This result shows an experimental example of QSAR application for the development of drug discovery; besides, it could be effec- tive help for further design and optimization in this type of lead compounds as a way to improve the antiprotozoan activity, from the selection of hits, followed by the elucidation of the behavior in the pharmacological and toxicological assays. However, it is generally acknowledged that QSARs are valid only within the same domain for which they were developed. In fact, even if the models are developed on the same chemicals, the AD for new chemicals can differ from model to model, depend- ing on the specific MDs. One of the main aims of the present work was to develop a model for predicting antiprotozoan activity at early stages of drug discovery and development. Consequently, one may not pretend to extrapolate the use of these models to other classes of antiprotozoan activity as this would result in uncertain predictions in conditions different from those fixed to derive the model.41,42 Therefore, the chemical designed in these studies only were synthetized and in vitro evaluated after they were found to lie in the AD of obtained models. For instance, another William plot (Fig. 5) of Eq. 11 (with the training set and quinoxalinone series discovered as novel antiprotozoan leads) was carried out. As can be noted in Figure 5, all quinoxalinones used lie within this area, which ensures great reliability for the prediction of this kind of leads used in the virtual screening. This proves the good assessment for the classification of these quinoxalinones as novel antiprotozoan leads. Therefore, this model can be used with high accuracy for new compound predictions in this applicability domain.41,42 2.4. Chemistry result Owing to their direct involvement with the present paper, spe- cial mention deserves the study performed on the synthesis and biological activity of a series of 3-alkoxy-1-[5-(dialkylami- no)alkyl]-5-nitroindazoles,30 as well as previous work on the syn- thesis and reactivity of quinoxalinium salts prepared from substituted acetanilides through intramolecular quaternization reactions.33 Molecule list Imput (Total & Local MDs) New virtual Leads non-stochastic Indices (NS) vdW volume Atomic Mass ...... Ensemble classifier: multi-agent predictor/fusion approach (Fuse outputs by weighted voting) Final Output Individual outputs (1-4) polariz- ability (NS), (SS) MDs(NS+SS) MDs-APNS-AP Individual outputs (1-3) SS-AP stochastic Indices (SS) vdW volume Atomic Mass ...... Individual outputs (1-4) polariz- ability Chemometric techniques: Linear Discriminant Analysis Figure 4. Flowchart illustrating how the individual classifiers are fused into the ensemble classifier through a voting system. Here we show the fusion of the discriminant functions by using TOMOCOMD–CARDD MDs into a prediction engine. Table 3 Results of ligand-based in silico screening by using CI and CE Compound* Result by using whole set of CI CE classb DP%a Eq. 1 DP%a Eq. 2 DP%a Eq. 3 DP%a Eq. (4 DP%a Eq. 5 DP%a Eq. 6 DP%a Eq. 7 DP%a Eq. 8 DP%a Eq. 9 DP%a Eq. 10 DP%a Eq. 11 9 88.52 65.68 71.94 72.28 86.99 44.53 88.64 95.44 90.49 96.78 92.95 11 10 82.37 23.57 81.68 71.39 86.91 41.86 67.79 96.88 87.79 97.56 91.73 11 11 82.77 20.49 81.90 71.52 87.77 34.61 67.80 97.12 89.02 97.52 91.84 11 12 83.36 15.96 81.87 75.16 88.81 37.49 68.75 97.40 90.31 97.53 91.82 11 13 74.25 63.10 63.94 75.34 78.29 56.37 60.75 98.39 68.54 98.26 93.50 11 14 77.56 70.57 58.60 86.10 77.77 68.55 80.79 98.51 72.14 98.29 95.17 11 15 79.64 24.10 83.89 77.29 84.65 53.17 58.56 96.29 84.39 97.79 89.31 11 16 80.10 21.02 84.09 77.39 85.63 46.80 58.56 96.58 85.94 97.76 89.46 11 17 80.77 16.50 84.06 80.37 86.84 49.35 59.72 96.90 87.56 97.76 89.43 11 18 70.47 63.42 67.92 80.52 74.73 65.54 50.10 98.08 60.90 98.43 91.58 11 * The molecular structures of the compounds represented with codes (numbers) are shown in Scheme 1. a DP% = [P(Active) � P(Inactive)] � 100 of each compounds in this screening set (see Section 4). Classification of each compounds using every obtained CI models in the following order: Eqs. 1–11. Here, in order to consider every query molecule as active chemical we used DP% >15%, because with this cut-off we avoid the not classified example as well as the risk of false active can be less. b Classification of each compounds using the Ce (see Eqs. 13–17 in Section 4). M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1575 On this basis and taking into consideration the early in silico selection of the quinoxaline molecular sub-systemas promisorial antiprotozoan lead series, we decided to prepare (synthesis and spectroscopical characterization) and evaluate the biological efficacy of 7-nitroquinoxalin-2-ones 9–18, carrying at position 4 a 5-(dialkylamino)pentyl chain similar to that of the mentioned indazole derivatives, according to the synthetic pathway shown in the Scheme 1. Thus, treatment of substituted aniline 1 with bromoacetyl bromide yielded 2-bromoacetanilide 2, which cyclized easily to the spiro quinoxalinium bromide 5. This salt, as well as the corresponding 1-methyl analogue 6, could also be prepared by treat- ment of the previously prepared33 chlorides 3 and 4 with hydrobro- mic acid through a halogen exchange reaction. The piperidine ring of salts 5 and 6 was then cleaved in refluxing nitromethane to yield the corresponding 4-(5-bromopentyl)quinoxalinones7 and 8. Finally, treatment of compounds 7 and 8 with the required sec- ondary amines (dimethylamine, pyrrolidine, piperidine, homopi- peridine or 1,2,3,4-tetrahydroisoquinoline) yielded the final 4-[5-(dialkylamino)pentyl]-7-nitroquinoxalin-2-ones 9–18, which were isolated as the corresponding hydrobromides. The previously prepared33 chloro analogues of 7 and 8 were rather unreactive un- der the conditions used in this work (see Section 4) and were not appropriate for the preparation of the desired final compounds. The structures for all compounds have been established on the basis of their analytical and spectral data. The latter are similar to those of related 1-[5-(dialkylamino)alkyl]indazoles,30 quinoxalines and intermediates33 previously prepared by our research team. Thus, NMR spectra of 2-bromoacetanilide 2 show that this com- pound, like the corresponding chloro analogue,33 appears in CDCl3 solution as the Z-rotamer. On the other hand, owing to the rigidity of spiro bromides 5 and 6, NCH2 protons of piperidine rings are 0.000 0.027 0.055 0.087 0.200 0.276 0.400 Leverage -4.00 -3.00 -2.19 -1.54 -1.00 -0.46 0.09 0.63 1.17 1.71 2.39 3.00 4.00 St d. R es id ua ls Training set New Leads Figure 5. LDA models applicability domain for learning and new lead series. The training set is represented by blues circles and the new compounds are represented by red triangles. 1576 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 anisochronic and, according to their different coupling patterns, they can be distinguished as equatorial (He) and axial (Ha). Similar features were observed for the cyclic secondary amine-derived fi- nal products 10–13 and 15–18, according to their structure of ter- tiary ammonium bromides. The NCH2 protons of piperidine rings of compounds 11 and 16 can also be distinguished as Ha and He. Nevertheless, the assignment (equatorial or axial) of other protons of the piperidine rings and protons of pyrrolidine (10, 15), homopi- peridine (12, 17) and 1,2,3,4-tetrahydroisoquinoline (13, 18) deriv- atives is not easy; when separate signals are observed, they have been mentioned in the description of 1H NMR spectra as HA and HB. 2.5. In vitro screening and wet evaluation In the present section we describe the main results obtained in the experimental assays (wet evaluation) in five different proto- zoan-parasite tests for the new chemicals selected as lead series in our in silico experiment. Here, we developed a wet screening experiment taking into account a battery of tests, that include the two most representative types of the subphylum protozoa par- asites: (1) T. vaginalis, T. cruzi and Leishmania braziliensis, which be- long to mastigophora (flagellata) subphylum and also, (2) two different apicomplexa (sporozoa) parasites: P. falciparum and T. gondii. These parasite-based tests will permit to depict a rather complete profile of antiprotozoan activity of these new compounds. Firstly, we evaluate the designed compounds against T. vaginalis and T. cruzi. In the case of the latter, the epimastigote form was used in the in vitro experiment taking into consideration that this form is an obligate invertebrate (replicative form) intracellular stage. In addition, unspecific cytotoxicity to macrophages was tested for all compounds. The in vitro efficacy against T. vaginalis and T. cruzi (as well as unspecific cytotoxicity) is shown in Tables 4 and 5, respectively. The specific activity against T. cruzi and T. vaginalis are ex- pressed as percentages of anti-epimastigote activity and growth inhibition (cytostatic activity), respectively. The cytocidal activity (percentage of reduction with respect to the control) against T. vag- inalis is shown in brackets. Metronidazole and Nifurtimox were used as trichomonacidal and trypanocidal reference drugs, respec- tively. Unspecific cytotoxic activity to macrophages is expressed as cytotoxicity percentage. In general, all chemicals showed low unspecific cytotoxicity, ex- cept for compounds 13, 17, and 18 at 100 lg/mL. Most of the tested compounds, exhibited a trichomonacidals activity near to 100% (11–18, 14) at a concentration of 100 lg/mL. Only compound 10 and 9 were inactive at this concentration. However, only chemicals 15–17 showed cytocidal activity against T. vaginalis at 10 lg/mL after 24 h of contact. These derivatives showed rather good anti- protozoan action at this level (near 90%; percentage of reduction with respect to the control), but this effect does not appear at 48 h of contact. At this time, only at the first concentration of 100 lg/mL 11–18 were active. In the same form, most of the tested compounds also exhibited a trypanocidal activity of 80–100% (10–13 and 16) at 100 lg/mL. This activity was specific, since all of them, except for compound 13, showed cytotoxicity lower than the anti-epimastigote activity (see Table 5). However, the trypanocidal activity dramatically de- creases at the lowest dose. Only compound 16 retained 60% of activity at 10 lg/mL; at this concentration no cytotoxicity was shown for this compound. In addition, we tested this series of chemicals against L. braziliensis which is the strain that causes a mucocutaneous form N NO2N O R1 H N COCH2Br N O2N + Br _ NH2 N O2N N NO2N O R1 +Cl _ 1 2 5: R1 = H 6: R1 = CH3 i 3: R1 = H 4: R1 = CH3 ii iii N NO2N O R1 N R3 R2 N NO2N O R1 Br 7: R1 = H 8: R1 = CH3 R1 = H R2,R3 CH3,CH3 [CH2]4 [CH2]5 [CH2]6 R1 = CH3 o-CH2-C6H4-[CH2]2 9 10 11 12 13 14 15 16 17 18 5' 4'3' 2' 1' 3 2 18 6 5 x HBr iv v 9-18 2 3 5 8 7 2' 3' 4' Scheme 1. Reagents and conditions: (i) BrCH2COBr, acetone, rt, 30 min; (ii) CH3NO2, reflux, 25 min; (iii) 48% aq HBr, vacuum evaporation to dryness (3 times); (iv) CH3NO2, reflux (48 h for 7 and 24 h for 8), argon; (v) R2R3NH, dioxane, 100– 110 �C (autoclave) or reflux, 5–10 h. Table 4 Percentages of citostatic and/or citocidal activity [brackets] for the three concentra- tions assayed in vitro against Trichomonas vaginalis Compound* In vitro activityb (lg/mL) Obs.a %CA24 h [%C24 h] %CA48 h [%C48 h] 100 10 1 100 10 1 9 � 29.39 11.43 1.22 28.33 14.68 0 10 � 75.61 21.02 3.53 34.26 1.64 0 11 + [99.37] 20.94 0 [100] 5.74 0 12 + [100] 12.94 2.35 [100] 0 0 13 + [100] 83.76 3.53 [100] 44.06 0 14 + [100] 45.71 8.98 [100] 11.26 0 15 ++ [100] [89.25] 0 [100] 67.7 4.1 16 ++ [100] [92.63] 0 [100] 86.52 0 17 ++ [100] [91.61] 10.98 [100] 70.41 2.87 18 + [100] 70.98 4.71 [100] 23.28 4.1 Metronidazole +++ [100] [99.1] [98.0] [100] [100] [99.5] * The molecular structures of the compounds represented with codes (numbers) are shown in Scheme 1. a Observed (experimental activity) classification against T. vaginalis. b Pharmacological activity of each tested compound, which was added to the cultures at doses of 100, 10 and 1 lg/mL: %CA# = Cytostatic activity(24 or 48 h) and [%C#] = Cytocidal activity(% of reduction)(24 or 48 h). Metronidazole was used as positive control (concentrations for metronidazole were 2, 1 and 0.5 mg/mL, respectively). M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1577 of leishmaniasis in humans. This strain in very important because the actual clinical treatment failure, especially in patients with kala-azar, mucocutaneous leishmaniasis, and diffuse cutaneous leishmaniasis, is becoming a common problem in many areas of endemicity.43 Treatment of the parasites (metacyclic promastigotes) with VAM2 compounds resulted in a concentration-dependent inhibi- tion effect on the proliferation of the tested L. braziliensis strain in seven of them. The compounds 17 and 15 displayed highest activity at a concentration range of 5.30 ± 0.22 > IC50 > 5.43 ± 0.00 lg/mL. Among the other products, the compounds 14 and 16 showed a significantly activity, IC50 8.15 (±0.00) and 9.57 (±0.56) lg/mL, respectively. Among the rest of compounds evaluated, products 13, 12 and 11 displayed an appreciable activity among promastigotesas well [IC50 of 31.27 (±3.16), 39.28 (±0.22) and 64.56 (±2.73) lg/mL, correspondingly]. The rest of the compounds (9, 10 and 18) were inactive. Comparing the activity against the three species of parasites (as well as cell toxicity) of these ten compounds it is possible to con- clude that 15–17 are the best chemicals. Specifically, 16 was the most active compound in T. cruzi and T. vaginalis and also showed high activity in L. braziliensis. Therefore, this compound can be con- sidered as a hit against mastigophora subphylum parasites. From these experiments, some relevant conclusions can be made about structure-activity relationship of these compounds. For instance, the methyl group at N-1 (14–18) enhances the activity against both species of flagellate protozoan parasites. The 6-member ring substituted at N4 (11 and 16) is the best chemical functional group, in contrast to the open form of this ring which was lethal during bioactivity assays (9 and 14) as well as when aromatic fragments were used (13 and 18), which also raised the toxicity observed for this lead series (see last column in Table 5). In the second step, we evaluated the same compounds against Toxoplasma gondii, an important protozoon of human medical interest. Here, we initially tested the efficacy of these chemicals against the tachyzoite form of T. gondii (RH strain).44,45 Tachyzoites (1 � 106) were exposed to VAM2 compounds for four hours at room temperature in order to evaluate the viability of the para- sites. Five hundred tachyzoites were counted and the viability per- centage was evaluated by the trypan blue exclusion method by counting the number of live tachyzoites.44,45 Results are shown in in Table 6. The compounds 10–12 showed toxoplasmicidal effects at con- centrations of 500 lM and 1 mM. The VAM2-17 was active against the parasite at 1 mM concentration. The evaluation of the parasites by light microscopy (data not shown) demonstrated that these four compounds produced the extrusion of the tachyzoites cytoplasm contents. Damage of tachyzoites with VAM2-11 was more severe than that caused by the other three compounds. Assays were made by triplicate. Negative controls without drug treatment showed 96% viability during the assay. Compounds 13 and 18 were not evaluated because their insolubility in the cell culture MEM media. Under these experimental approaches, we showed that some of the tested compounds have toxoplasmicidal activity mainly in the tachyzoite form. The compound 11 had the most potent anti- toxoplasma activity at high concentrations. These results suggested that the compounds 10–12 of this series might be considered as possible candidates in the development of toxoplasmicidal chemo- therapy. More studies need to be done to evaluate the effect of the VAM2 chemicals on the structural, functional and virulent proper- ties of Toxoplasma gondii in vitro and in vivo in order to design new drugs against these reemerging parasitic zoonoses. These stud- ies are in progress and will be published in a forthcoming paper. In conclusion, compound 11, with the same function at N4 as 16, but with an H-atom in N1 was the most active compound. This result indicates that H-atom in the N1 is necessary for the anti-toxo- plasma activity in contrast to that obtained in flagellate parasites, where the methylation of this N-atom was desired. The differences found in the effects of the evaluated compounds in both protozoan Table 5 Antitrypanosomal activity and inespecific cytotoxicity at three different concentrations (100, 10 and 1 lg/mL) assayed in vitro against Tripanosoma cruzi and macrophagic cells, respectively Compound* Obs.a Concentration (lg/mL) % anti-epimastigotesb ± % SD % cytotoxicityc ± % SD NT 100 NT NT 9 10 1 100 83.54 ± 0.44 0 ± 0.55 10 + 10 5.35 ± 0.25 0 ± 2.19 1 4.38 ± 0.30 0 ± 2.14 100 82.4 ± 0.68 3.36 ± 1.47 11 + 10 17.68 ± 1.24 0 ± 1.51 1 1.78 ± 8.63 0 ± 1.97 100 97.73 ± 0.45 59.14 ± 1.77 12 + 10 23.84 ± 1.27 5.78 ± 0.58 1 8.35 ± 5.11 0 ± 1.07 100 87.83 ± 0.06 100 ± 0.15 13 + 10 56.77 ± 1.41 13.25 ± 0.46 1 12.49 ± 1.85 9.89 ± 1.21 NT 100 NT NT 14 10 1 100 6.36 ± 4.81 49.25 ± 0.4 15 � 10 2.51 ± 5.97 0 ± 2.26 1 0 ± 3.38 0 ± 1.25 100 79.12 ± 3.86 61.38 ± 0.53 16 + 10 60.68 ± 2.78 11.57 ± 2.01 1 7.93 ± 4.42 NT 100 65.46 ± 5.47 75.75 ± 0.9 17 � 10 15.38 ± 2.83 20.24 ± 1.2 1 0 ± 3.84 N 100 19.78 ± 5.94 99.44 ± 0.2 18 � 10 15.62 ± 5.06 24.44 ± 0.26 1 11.77 ± 4.35 NT 100 98.73 ± 0.5 25.9 ± 3.9 Nifurtimox + 10 90.0 ± 1.8 0.6 ± 3.9 1 75.5 ± 3.9 0.0 ± 2.1 * The molecular structures of the compounds represented with codes (numbers) are shown in Scheme 1. a Observed (experimental activity) classification against T. cruzi. Experimentally observed activity (compounds with % anti-epimastigote >70 at 100 (lg/mL) were con- sidered as active ones). b Anti-epimastigotes percentage and ±standard deviation (SD). c Inespecific cytotoxicity in macrophages cells and standard deviation (SD). NT means not tested. Reference drug and positive control: Nifurtimox. Table 6 In vitro efficacy against Toxoplasma gondii Tachyzoites Compound* Obs.a % tachyzoites parasitesb 1 mM 500 lM 200 lM 100 lM 9 � 73 93 96 95 10 + 0 0 85 91 11 + 0 0 68 92 12 + 0 0 82 88 13 NT NT NT NT NT 14 � 81 93 90 81 15 � 38 58 84 89 16 � 36 40 90 96 17 ± 0 71 77 81 18 NT NT NT NT NT DMSO � 85 93 92 91 * The molecular structures of the compounds represented with codes (numbers) are shown in Scheme 1. a Observed (experimental activity) against Toxoplasma gondii tachyzoites (RH strain). b Biochemical studies of percentages of parasites (tachyzoites) for every chemi- cals evaluated in the range of 1 mM, 500 lM, 200 lM, 100 lM. DMSO: dimethyl sulfoxide. 1578 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 subphylums might reside in the intrinsic biological properties of the mentioned parasites. Finally, these compounds were assayed in two different tests for antimalarial screening. The first technique used was an enzymatic in vitro assay, the so-called: ferriprotoporphyrin IX biocrystalliza- tion inhibition test (FBIT).46 During their digestion of host cell haemoglobin, intraerythrocytic malaria parasites produce large amounts of toxic ferriprotoporphyrin IX (FP). The inhibition of biomineralization of FP to b-hematin by some antimalarial compounds such as chloroquine underlies their action mode and in this sense, it can be used to give criteria about the antimalarial properties of such compounds.24,46 The global results for the se- lected chemicals in this enzymatic in vitro model are depicted in Table 7. From ten compounds, only 3 cases (13, 17 and 18) showed activity at IC50 values lower than 2.0 lg/mL in the biomineraliza- tion microassay. The remaining seven, resulted inactive. In this as- say, all compounds resulted less active than chloroquine (see Table 7). In terms of activity, the assayed compounds can be or- dered as follows: 18 > 13 > 17. However, these chemicals had unspecific cytotoxicity at 100 lg/mL. Afterwards, a cell-based approach was also used to evaluate the in vitro effectivity of the designed series. This second in vitro cell- based assay was carried out by using a radioisotopic microtest in Plasmodium falciparum (strain 3D7).47 Here, every compound was evaluated against cultured intraerythrocytic asexual forms of the human malaria parasite P. falciparum. The uptake of [G-3H]hypo- xanthine by parasitized erythrocytes in microtiter plates was used as an indicator of drug activity. As can be seen in Table 7, the com- pound 18 was also active in this wet evaluation, while chemicals 13 and 17 were inactive. However, compound 12 showed rather high activity in this cell assay. This compound had low cytotoxicity and was also active against T. gondii, therefore this chemical core and SAR result (H atom at N-1 and a 6-membered ring at N-4) can be considered as an important starting point for the design of novel Table 7 In vitro antimalarial activity as a function of ferriprotoporphyrin IX biocrystallization inhibition test and radioisotopic microtest in strain 3D7 of Plasmodium falciparum Compound* Obs.a Ferriprotoporphyrin IX biocrystallization inh. test Radioisotopic microtest in strain 3D7 of Plasmodium falciparum IC50 b (mg/mL) IC50 c (mg/mL) 9 VAM2-9 � >2 >10 10 VAM2-10 � >2 >10 11 VAM2-11 � >2 >10 12 VAM2-12 + >2 5.72 13 VAM2-13 + 1.53 >10 14 VAM2-14 � >2 >10 15 VAM2-15 � >2 >10 16 VAM2-16 � >2 >10 17 VAM2-17 + 1.95 >10 18 VAM2-18 ++ 0.95 6.47 Chloroquine ++ 0.04 0.04 Bold values represents the Compound number. * The molecular structures of the compounds represented with codes (numbers) are shown in Scheme 1. a Observed (experimental activity) as a function of two different in vitro assays. b IC50 values calculated from the percentage of inhibition obtained in ferriprotoporphyrin IX biocrystallization inhibition test (IC50 >2 lg/mL were considered as inactives). c IC50 values calculated from the percentage of inhibition obtained in radioisotopic microtest in strain 3D7 of Plasmodium falciparum (IC50 >10 lg/mL were considered as inactives). Chloroquine was used as antimalarial reference drug in both assays. M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1579 anticomplexan drugs. In this sense, new refining algorithms are needed for optimizing the pharmacological, toxicological and physicochemical properties. Additionally, all chemicals were screened for activity against Acanthamoeba (Sarcodina subphylum) and none of the compounds evaluated was active against Acanthamoeba castellanii. All biologi- cal details and their results are show as Supporting information (see page 24). In summary, these results can be considered as a promising starting point for the future design and refinement of novel com- pounds with high anti-protozoan activity and low toxicity, although compounds 15–17 (lead series for anti-mastigophora subphylum) and 10–12 (lead series for anti-apicomplexan subphy- lum) were active at higher doses than their respective reference drugs. Analyzing all these in vitro results, it is clear to see that fur- ther refinement algorithms are needed to identify the ways in which the activity and ADME-Tox of the present chemical core can be optimized. Therefore, these chemicals, mainly 11(12) and 16, can be taken as hits, which are amenable for further chemistry optimization in order to derive products with appropriate combi- nation of potency, pharmacokinetic properties, toxicity etc., as well as with good activity in animal models. 3. Conclusion The integration (aligning) of dry and wet screening for diverse compounds libraries is an essential step in the quest and design of antiprotozoan lead compounds. The results of our in silico pre- diction and posterior in vitro screening by using a battery of para- sites-cell assays are encouraging and show that such kind of methodological approaches can be successful. Within this one set of an in house library, we have identified 10 novel chemicals not yet reported (virtual hits) as antiprotozoan leads. All novel quinox- alinones were then synthetized by using simple and efficient prep- arations methods. The spectroscopical (structural) characterization was also presented in this report. Finally, the biological evaluation showed that most of the tested compounds, exhibited an adequate antiprotozoan activity against the four different types of parasites (T. vaginalis, T. cruzi, L. braziliensis, T. gondii and P. falciparum). In general, all chemicals showed low unspecific cytotoxicity, except for compounds 13, 17, and 18 at 100 lg/mL. However, the most ac- tive compounds, 11(12) and 16, do not present cytotoxicity in mac- rophages at any level. These chemicals showed preliminary evidence of efficacy and selectivity for broad antiprotozoan activity with potential for scaffold optimization. 3.1. Future perspective The development of a new drug is a lengthy and complex pro- cess. The identification of an appropriate lead molecule is the most critical component of this phase. To this effect, here we have shown how the combination of validated QSAR-modeling and LBVS, could be successfully used as innovative technologies, to ensure high expected hit rates in the discovery of new bioac- tive compounds. In future outlooks, these models which relate the chemical structure with a specific endpoint, could be pro- grammed into expert systems helping in exhaustive search of bioactive molecules within huge chemical libraries. That is to say, the preliminary identification of novel antiprotozoan leads in this work is promising and strongly supports the LBVS of addi- tional compounds libraries; chemicals with diverse scaffolds must be considered in the search of new anti-parasitic com- pounds. In fact, the ensembleclassifier presented here will be used to identify new antiprotozoan drugs from well-known drug databases already approved for human use with potential ‘off-la- bel’ antiparasitic application. The logic of this approach is that hits from such screens are low-hanging fruit that will require less development before they are able to enter in clinical trials as antiparasitic drugs. Some work in this direction is now under progress. The action mode of the novel quinoxalinones described in this study is a question that has not been addressed. While this is be- yond the scope of this report, it is extremely relevant, and we are currently following up on the top leads. A recent study on the ac- tion mechanisms of 7-nitroquinoxalin-2-ones as a meaning of evaluating their effectiveness against T. cruzi suggests as possible action mode the inhibition of the enzyme trypanothione reduc- tase.48In the same spirit, we intend to explore the ADMETox prop- erties of the screened antiprotozoan leads, as they will illuminate future studies on theiroptimization, but first explored using the- oretical models. In this sense, our research group is working in the application of new 3D MDs and data mining techniques to these problems. We also intend to concentrate our efforts on the use of more sophisticated statistical techniques with the TOMOCOMD–CARRD MDs in order to describe the activity of or- ganic compounds against important pharmacological targets of antiprotozoan drugs. Another direction to explore in future stud- ies is the multi-optimization (approach) in order to characterize the biological response of one target chemical versus multitarget chemicals, for instance: different species, different molecular tar- gets, and so on. 1580 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 4. Experimental section 4.1. Computational strategies 4.1.1. Data set and classification strategy A benchmark dataset usually consists of a learning (or training) dataset and an independent testing dataset. The learning dataset is one of the important components for a statistical predictor because it is used for training the predictor’s ‘engine’, whereas the testing dataset is used for examining the predictor’s accuracy via an exter- nal test. The benchmark dataset was composed by 680 drugs (see all structures as a zip file) having great structural variability; 254 of them are active (antiprotozoan agents) and 426 inactive com- pounds (drugs having other clinical uses).37,38 4.1.2. Representation of molecules samples Several kinds of representations are generally used in this re- gard, all well-known molecular descriptor (MDs) or molecular indi- ces. These parameters are numbers that characterize a specific aspect of the molecule structure. The so-called topological (and topo-chemical) indices are among the most useful MDs known nowadays. These theoretical indices are numbers that describe the structural information of molecules through graph theoretical invariants and can be considered as structure-explicit descriptors. In the present report, a novel 2D TOMOCOMD–CARDD MDs family, namely atom, atom-type, and total linear indices were used in order to codify the molecular structure of every molecule in the dataset. These MDs are based on the calculation of linear maps (lin- ear form) in Rn in canonical basis sets.16,22,26,34,39 The computation of the non-stochastic and stochastic linear indices is develop by using the kth ‘nonstochastic and stochastic graph–theoretical elec- tronic-density matrices’ Mkand Sk, correspondingly, as matrices of the mathematical forms.16,22,26,34,39 These matricial operators are graph-theoretical electronic-structure models, like the ‘extended Hückel MO model’. The M1 matrix considers all valence-bond elec- trons (- and p-networks) in one step, and their power k (k = 0, 1, 2, 3,...) can be considered as an interacting-electronic chemical-net- work in the kth step. The present approach is based on a simple model for the intramolecular (stochastic) movement of all outer- shell electrons. The theoretical scaffold of these atom-based MDs and their use to represent small-to-medium size organic chemi- cals, as well as QSAR and drug design studies has been explained in detail elsewhere.16,22,26,34,39 4.1.3. Computational methods: TOMOCOMD–CARDD approach The TOMOCOMD is an interactive program for molecular design and bioinformatics research, developed upon the base of a user-friendly philosophy.14 In this report, we only used the CARDD subprogram. All MDs [total and local (both atom and atom-type)] non-stochastic and stochastic linear indices were calculated in this software. 4.1.4. Chemometric studies The statistical software package STATISTICA was used to devel- op the k-MCA.49 The number of members in each cluster and the standard deviation of the variables in the cluster (kept as low as possible) were taken into account, to have an acceptable statistical quality of data partitions into the clusters. The values of the stan- dard deviation between and within clusters, the respective Fisher ratio and their p level of significance, were also examined. The LDA was also carried out with the STATISTICA software.49 The considered tolerance parameter (proportion of variance that is unique to the respective variable) was the default value for min- imum acceptable tolerance, which is 0.01. A forward-stepwise search procedure was fixed as the strategy for variable selection. The principle of parsimony (Occam’s razor) was taken into account as a strategy for model selection. The quality of the models was determined by examining Wilks’ k parameter (U statistic), the square Mahalanobis distance (D2), the Fisher ratio (F), and the cor- responding p level [p(F)] as well as the percentage of good classi- fication (accuracy) in the training and test sets. The classification of cases was performed by means of the posterior classification probabilities. By using the models, a compound can then be classified as active, if DP% >0, being DP% = [P(Active) � P(Inac- tive)] > 100, or as inactive otherwise. The P(Active) and P(Inactive) are the probabilities with which the equations classify a com- pound as active or inactive, respectively. Performing the assess- ment of the obtained models, the sensibility, the specificity (also known as ‘hit rate’), the false positive rate (also known as ‘false alarm rate’), and Matthews correlation coefficient (C), were calcu- lated; and checked in the training and test sets.50 Finally, the leverage approach41was used to evaluate the AD of the QSAR mod- els. Through of this method it is possible to verify whether a new chemical will lie within the structural model domain. The leverage h of a compound measures its influence on the model. That is, the leverage used as a quantitative measure of the model AD is suitable for evaluating the degree of extrapolation, which represents a sort of compound ‘distance’ from the model experimental space. Pre- diction should be considered unreliable for compounds of high leverage values (h > h⁄). A leverage greater than the warning lever- age h⁄ means that the compound predicted response can be extrapolated from the model, and therefore, the predicted value must be used with great care. Only predicted data for chemicals belonging to the chemical domain of the training set should be proposed. 4.1.5. Prediction algorithms and ensemble classifier (multi-agent predictor or fusion approach) Here, we used nonstochastic and stochastic linear indices to de- velop classification-based QSAR models in order to classify mole- cules as antiprotozoan or inactive compounds. These MDs have a few parameters that can be ‘modified’ in the calculation process. The number of these uncertain parameters depends on the atom- labels (AP scheme) used for the prediction engine. It would be much more tedious and time-consuming to determine the optimal values for AP [AP:51 atomic mass (AP = M), atomic polarizability (AP = P), atomic Mulliken electronegativity (AP = K) and van der Waals atomic volume (AP = V)]. In addition, the number of uncer- tain parameters also depends on which MDs sets are used to rep- resent the chemical samples. For instance, here every model can be fitted by two kinds of MD sets: (1) non-stochastic MDs (NS), (2) stochastic MDs (SS). To solve the problem, we use a [2AP + 1NS + 1SS + 1(NS + SS)]-dimensional fusion approach (11 models in total), similar to that done in protein research. First, the basic individual classifiers to be generally expressed asCI(NS � AP, SS � AP, NS, SS, NS + SS)andthe predicted classifica- tion results for a query molecule M by each of the individual clas- sifiers can be formulated by, CIðNS� AP; SS� AP;NS; SS;NSþ SSÞ›M ¼ CNS�AP;SS�AP;NS;SS;NSþSSðMÞ 2 S ð12Þ where, the symbol › is an action operator meaning using CI(NS � AP, SS � AP, NS, SS, NS + SS) to classify M, S representing the union of the two subsets defined (active or inactive). Therefore, the final pre- dicted result should be determined by a fusion approach through the following voting mechanism. Now let us introduce an ensemble classifier CE, which is formed by fusing all set of the basic individual classifiers CI(NS � AP, SS � AP, NS, SS, NS + SS) and can be formu- lated as follows: M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1581 CE ¼ C1ðM;NSÞ8C2ðK;NSÞ8C3ðP;NSÞ8C4ðV;NSÞ8C5ðall AP;NSÞ::: 8C6ðM; SSÞ8C7ðK; SSÞ8C8ðP; SSÞ8C9ðV; SSÞ8C10ðall AP; SSÞ::: 8C11ðall AP;NSþ SSÞ ð13Þ where the symbol " denotes the fusing operator. Then, the voting score for the query molecules M belonging to the cth class is given by, pc ¼ X4 AP¼1 X2 MDs¼1 wAP;MDsDðAP;MDs;ScÞ þ X3 MDs¼1 wall�AP;MDsD ðall� AP;MDs;ScÞ; ðc ¼ 1;�1Þ ð14Þ where, Sc = 1 is for antiprotozoans and Sc = �1 for non-antiprotozo- ans, wAP,MDs and wall-AP,MDs are the weight factors and were set at 1 for simplicity. The delta functions in Eq. 14 is given by, DðAP;MDs;ScÞ ¼ 1 if CAP;MDsðMÞ 2 Sc 0 otherwise � ð15Þ Dðall� AP;MDs;ScÞ ¼ 1 if CP;MDsðMÞ 2 Sc 0 otherwise � ð16Þ thus the query Molecule M is predicted belonging to the class (c) or subset Sc for which the score of Eq. 14 is the highest; that is, l ¼ arg max c pcf g ðc ¼ 1;�1Þ ð17Þ where, l is the argument of c that maximizes pc. If there is a tie, then the final predicted result will be randomly assigned (or is take as unclassified) to one of the corresponding subsets although this kind of tie case rarely happens and was actually not observed in the current study. 4.2. Chemistry 4.2.1. Instrumental data Mps were determined in a Stuart Scientific melting point appa- ratus SMP3. The mps of quinoxalinium salts 5 and 6, as well as those of some of the final products (hydrobromides 9–18) are not very well defined; these compounds decompose on heating and the observed mps are frequently heating-rate dependent and previous softening is usual. 1H (300 or 400 MHz) and 13C (75 or 100 MHz) NMR spectra were recorded on Varian Unity 300 or Var- ian Inova 400 spectrometers. The chemical shifts are reported in ppm from TMS (d scale) but were measured against the solvent sig- nal. The J values are given in Hz. The assignments have been per- formed by means of different standard 1D and 2D correlation experiments (NOE, COSY, HMQC and HMBC). The numbering used in the description of NMR spectra of spiro compounds 5 and 6, and 4-substituted quinoxalinones 7–18 is shown in Scheme 1; double primed numbers refer to the cyclic secondary amine rings of final compounds 9–18. The electron impact (EI) and electrospray (ES) mass spectra were obtained at 70 eV on a Hewlett Packard 5973 MSD spectrometer or on a Hewlett Packard 1100 MSD spectrome- ter, respectively. DC-Alufolien silica gel 60 PF254 (Merck, layer thickness 0.2 mm) was used for TLC. Microanalyses were per- formed by the department of Analysis, Center of Organic Chemistry ‘Manuel Lora Tamayo’ CSIC, Madrid, Spain. 4.2.2. Procedure for the preparation of all chemicals 4.2.2.1. 2-Bromo-50-nitro-20-piperidinoacetanilide (2) Bromoacetyl bromide (9.08 g, 45 mmol) was dropped (ca. 5 min) into a solution of 5-nitro-2-piperidinoaniline (1)27 (8.85 g, 40 mmol) in acetone (150 mL). After 15 min, an additional amount of bromoacetyl bromide (ca. 1 mL) was dropped and the mixture stirred for 15 min. The obtained suspension (2 � HBr) was poured into water (1 L), and the mixture stirred for 30 min. The solid in sus- pension, collected by filtration, washed with water (4 � 100 mL) and air-dried was shown to be bromoacetanilide 2 (13.28 g, 97% yield). This compound, crystallized from ethanol, melts partially and resolidifies at 123–125 �C (decomposition to spiro salt 5, TLC), showing a further mp at 186–190 �C (corresponding to that of salt 5, see below); 1H NMR (CDCl3): d 9.40 (s, 1H, NH), 9.20 (d, J = 2.7 Hz, 1H, 60-H), 7.97 (dd, J = 8.8, 2.7 Hz,1H, 40-H), 7.21 (d, J = 8.8 Hz, 1H, 30-H), 4.09 (s, 2H, 2-H), 2.87 (m, 4H, 200-,600-H), 1.81 (m, 4H, 300-,500-H), 1.62 (m, 2H, 400-H); 13C NMR (CDCl3): d 163.46 (C-1), 148.95 (C-20), 144.16 (C-50), 132.65 (C-10), 120.31, 120.01 (C-30, -40), 114.47 (C-60), 53.32 (C-200, -600), 29.59 (C-2), 26.35 (C-300, -500), 23.75 (C-400); MS (EI): m/z (%) 343 (12) ([M+2]+), 341 (12) (M+), 262 (85), 220 (100), 203 (35), 192 (13), 174 (25), 164 (16), 145 (10), 118 (19). Anal. Calcd for C13H16BrN3O3 (342.19): C 45.63; H 4.71; N 12.28. Found: C 45.70; H 4.67; N 12.12. 4.2.2.2. 6-Nitro-3-oxo-1,2,3,4-tetrahydroquinoxaline-1-spiro-10- piperidinium bromide (5). (a) From bromoacetanilide 2: A solution of anilide 2 (0.68 g, 2.0 mmol) in nitromethane (10 mL) was refluxed for 25 min. After cooling, the insoluble bromide 5 (0.59 g, 87% yield) was collected by filtration, washed with acetone (3 � 10 mL) and air-dried. (b) From tetrahydroquinoxaline-1-spiro- 10-piperidinium chloride 3: Chloride 3 (prepared33 by cyclization of 2-chloro analogue of 2) (7.44 g, 25 mmol) was dissolved in 48% aq hydrobromic acid and evaporated to dryness. This process was repeated twice and, after addition of acetone (100 mL), the insoluble salt 5 (8.47 g, 99% yield) was collected by filtration, washed with acetone (3 � 40 mL) and air-dried. Mp 187–192 �C (decomp.) (water); 1H NMR (DMSO-d6): d 11.88 (s, 1H, 4-H), 8.30 (d, J = 9.0 Hz, 1H, 8-H), 8.13 (dd, J = 9.0, 2.6 Hz,1H, 7-H), 8.01 (d, J = 2.6 Hz, 1H, 5-H), 4.89 (s, 2H, 2-H), 4.12 (m, Jgem = (�)12.0 Hz, Ja,a = 9.6 Hz, 2H, 20-,60-Ha), 3.84(br d, Jgem = (�)12.0 Hz, 2H, 20-,60- He), 2.18 (m, 2H) and 1.98-1.50 (m, 4H) (30-,40-,50-H); 13C NMR (DMSO-d6): d 160.74 (C-3), 148.59 (C-6), 134.98, 133.47 (C-4a, -8a), 122.84 (C-8), 118.38 (C-7), 112.89 (C-5), 61.72 (C-20, -60), 55.05 (C-2), 19.92 (C-40), 19.29 (C-30, -50); MS (ES+): m/z (%) 523 (20) ([2(M�Br)�1]+), 262 (100) ([M�Br]+); MS (EI) of salt 5 is iden- tical to that of bromoalkyl derivative 7 arising from its thermal decomposition. Anal. Calcd for C13H16BrN3O3 (342.19): C 45.63; H 4.71; N 12.28. Found: C 45.50; H 4.87; N 12.52. 4.2.2.3. 4-Methyl-6-nitro-3-oxo-1,2,3,4-tetrahydroquinoxaline- 1-spiro-10-piperidinium bromide (6). Spiro chloride 4 (prepared33 from 2,20-dichloro-N-methyl-50-nitroacetanilide and piperidine or by cyclization of 2-chloro-N-methyl analogue of 2) (7.79 g, 25 mmol), treated with 48% aq hydrobromic acid as de- scribed for the preparation of salt 5, afforded the title bromide 6 (7.93 g, 89% yield). Mp 159–162 �C (decomp.) (ethanol); 1H NMR (DMSO-d6): d 8.34 (d, J = 9.0 Hz, 1H, 8-H), 8.23 (dd, J = 9.0, 2.4 Hz,1H, 7-H), 8.18 (d, J = 2.4 Hz, 1H, 5-H), 4.97 (s, 2H, 2-H), 4.14 (m, Jgem = (�)12.1 Hz, Ja,a = 9.8 Hz, 2H, 20-,60-Ha), 3.90 (br d, Jgem = (�)12.1 Hz, 2H, 20-,60-He), 3.45 (s, 3H, 4-CH3), 2.17 (m, 2H) and 1.93–1.51 (m, 4H) (30-,40-,50-H); 13C NMR (DMSO-d6): d 160.14 (C-3), 148.93 (C-6), 136.39, 135.34 (C-4a, -8a), 122.62 (C-8), 118.87 (C-7), 112.82 (C-5), 61.48 (C-20, -60), 55.18 (C-2), 29.69 (4-CH3), 19.93 (C-40), 19.32 (C-30, -50); MS (ES+): m/z (%) 633 (5) ([2 M�Br+2]+), 631 (5) ([2 M�Br]+), 276 (100) ([M�Br]+); MS (EI) of salt 6 is identical to that of bromoalkyl derivative 8 arising from its thermal decomposition. Anal. Calcd for C14H18BrN3O3 (356.22): C 47.20; H 5.09; N 11.80. Found: C 47.48; H 4.87; N 11.62. 4.2.2.4. 4-(5-Bromopentyl)-7-nitro-3,4-dihydro-1H-quinoxalin- 2-one (7). A suspension of bromide 5 (6.84 g, 20 mmol) in nitromethane (50 mL) was refluxed for 48 h under argon atmo- sphere. After cooling, the solid in suspension, collected by filtration, 1582 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 washed with nitromethane (2 � 10 mL) and air-dried, was shown to be the bromopentyl derivative 7 (6.43 g, 94% yield). Similar re- sults were obtained starting from bromoacetanilide 2, following the same procedure but without isolation of the intermediate salt 5. Mp 192–195 �C (decomp.) (nitromethane).1H NMR (DMSO-d6): d 10.78 (s, 1H, 1-H), 7.77 (dd, J = 9.3, 2.7 Hz,1H, 6-H), 7.59 (d, J = 2.7 Hz, 1H, 8-H), 6.77 (d, J = 9.3 Hz, 1H, 5-H), 4.04 (s, 2H, 3-H), 3.53 (t, J = 6.7 Hz, 2H, 50-H), 3.34 (t, J = 7.3 Hz, 2H, 10-H), 1.84 (m, 2H, 40-H), 1.59 (m, 2H, 20-H), 1.42 (m, 2H, 30-H); 13C NMR (DMSO- d6): d 163.82 (C-2), 140.20 (C-4a), 136.46 (C-7), 125.57 (C-8a), 120.58 (C-6), 109.66 (C-8), 109.17 (C-5), 51.00 (C-3), 48.96 (C-10), 35.00 (C-50), 31.99 (C-240), 24.92 (C-30), 23.64 (C-20); MS (EI): m/z (%) 343 (24) ([M+2]+), 341 (24) (M+), 262 (17), 206 (100), 178 (45), 160 (10), 132 (23), 118 (8). Anal. Calcd for C13H16BrN3O3 (342.19): C 45.63; H 4.71; N 12.28. Found: C 45.55; H 4.61; N 12.52. 4.2.2.5. 4-(5-Bromopentyl)-1-methyl-7-nitro-3,4-dihydro-1H- quinoxalin-2-one (8). A suspension of bromide 6 (7.12 g, 20 mmol) in nitromethane (50 mL) was refluxed for 24 h under ar- gon. The solvent was then evaporated to dryness and the residue triturated with ethanol (20 mL); the insoluble material was collected by filtration, washed with cold ethanol (2 � 10 mL) and air-dried yielding compound 8 (5.91 g, 83% yield). Mp 118–120 �C (ethanol). 1H NMR (DMSO-d6): d 7.88 (dd, J = 9.3, 2.4 Hz,1H, 6-H), 7.69 (d, J = 2.4 Hz, 1H, 8-H), 6.85 (d, J = 9.3 Hz, 1H, 5-H), 4.12 (s, 2H, 3-H), 3.53 (t, J = 6.7 Hz, 2H, 50-H), 3.37 (t, J = 7.6 Hz, 2H, 10-H), 3.32 (s, 3H, 1-CH3), 1.84 (m, 2H, 40-H), 1.58 (m, 2H, 20-H), 1.43 (m, 2H, 30-H; 13C NMR (DMSO-d6): d 163.27 (C-2), 141.71 (C-4a), 136.85 (C-7), 127.64 (C-8a), 120.79 (C-6), 109.65, 109.51 (C-5, -8), 51.02 (C-3), 49.01 (C-10), 34.94 (C-50), 31.92 (C-40), 28.25 (1-CH3), 24.87 (C-30), 23.57 (C-20); MS (EI): m/z (%) 357 (33) ([M+2]+), 355 (33) (M+), 276 (17), 220 (100), 192 (72), 160 (7), 146 (29), 131 (12), 104 (5). Anal. Calcd for C14H18BrN3O3 (356.22): C 47.20; H 5.09; N 11.80. Found: C 47.48; H 5.19; N 11.62. 4.2.2.6. Preparation of 4-[5-(dialkylamino)pentyl]quinoxalin-2- ones hydrobromides 9–18. For dimethylamino derivatives 9 and 14, the corresponding bromide (7 or 8) (3 mmol) and dimeth- ylamine (7.5 mmol; 1.34 mL of a 5.6 M solution in ethanol) in 1,4- dioxane (100 mL) was heated in an autoclave at 100–110 �C until the starting bromide was consumed (ca. 6 h). For cyclic secondary amines derivatives 10–13 and 15–18, a mixture of the correspond- ing bromide (7 or 8) (3 mmol) and the required amine (7.5 mmol) in 1,4-dioxane (100 mL) was refluxed until the starting bromide was consumed (5–10 h). After eventual separation (filtration or decantation), some tars appeared when using dimethylamine or pyrrolidine, dioxane was evaporated to dryness and ethanol (10 mL) and 48% aq hydrobromic acid (0.5 mL) were added. The mixture was stirred for 2 h and the precipitated hydrobromide col- lected by filtration, washed with ethanol (2 � 5 mL) and air-dried (83–98% yield). 4.2.2.7. 4-[5-(Dimethylamino)pentyl]-7-nitro-3,4-dihydro-1H- quinoxalin-2-one hydrobromide (9). Yield: 0.98 g (84%); mp 204–207 �C (methanol); 1H NMR (DMSO-d6): d 10.81 (s, 1H, 1-H), 9.44 (br s, 1H, 50-NH+), 7.77 (dd, J = 9.3, 2.7 Hz,1H, 6-H), 7.60 (d, J = 2.7 Hz, 1H, 8-H), 6.80 (d, J = 9.3 Hz, 1H, 5-H), 4.06 (s, 2H, 3-H), 3.35 (t, J = 7.5 Hz, 2H, 10-H), 3.03 (m, 2H, 50-H), 2.75 [s, 6H, N(CH3)2], 1.61 (m, 4H, 20-,40-H), 1.33 (m, 2H, 30-H); 13C NMR (DMSO-d6): d 163.86 (C-2), 140.27 (C-4a), 136.43 (C-7), 125.59 (C-8a), 120.60 (C-6), 109.66, 109.30 (C-5, -8), 56.34 (C-50), 51.08 (C-3), 48.84 (C-10), 42.09 [N(CH3)2], 24.03, 23.44, 23.09 (C-20, -30, -40). MS (ES+): m/z (%) 695 (12) ([2 M�Br+2]+), 693 (12) ([2 M�Br]+), 308 (20) ([M�Br+1]+), 307 (100) ([M�Br]+).Anal. Calcd for C15H23BrN4O3 (387.27): C 46.52; H 5.99; N 14.47. Found: C 46.50; H 5.77; N 14.21. 4.2.2.8. 7-Nitro-4-(5-pyrrolidinopentyl)-3,4-dihydro-1H-qui- noxalin-2-one hydrobromide (10). Yield: 1.22 g (98%); mp 233–235 �C (decomp.) (water); 1H NMR (DMSO-d6): d 10.81 (s, 1H, 1-H), 9.60 (br s, 1H, 100-H), 7.77 (dd, J = 9.3, 2.7 Hz,1H, 6-H), 7.60 (d, J = 2.7 Hz, 1H, 8-H), 6.80 (d, J = 9.3 Hz, 1H, 5-H), 4.06 (s, 2H, 3-H), 3.49 (br s, 2H, 200-,500-HA), 3.35 (t, J = 7.4 Hz, 2H, 10-H), 3.10 (m, 2H, 50-H), 2.97 (br s, 2H, 200-,500-HB), 1.95 (br s, 2H) and 1.87 (br s, 2H) (300-,400-H), 1.63 (m, 4H, 20-,40-H), 1.33 (m, 2H, 30-H). 13C NMR (DMSO-d6): d 163.79 (C-2), 140.22 (C-4a), 136.38 (C-7), 125.54 (C-8a), 120.59 (C-6), 109.62, 109.29 (C-5, -8), 53.58 (C-50), 52.94 (C-200, -500), 51.05 (C-3), 48.85 (C-10), 24.85, 23.96, 23.23 (C-20, -30, -40), 22.61 (C-300, -400).MS (ES+): m/z (%) 747 (14) ([2 M�Br+2]+), 745 (13) ([2 M�Br]+), 334 (23) ([M�Br+1]+), 333 (100) ([M�Br]+). Anal. Calcd for C17H25BrN4O3 (413.31): C 49.40; H 6.10; N 13.56. Found: C 49.50; H 6.37; N 13.72. 4.2.2.9. 7-Nitro-4-(5-piperidinopentyl)-3,4-dihydro-1H-quinox- alin-2-one hydrobromide (11). Yield: 1.24 g (97%); mp 246– 248 �C (decomp.) (methanol); 1H NMR (DMSO-d6): d 10.81 (s, 1H, 1-H), 9.08 (br s, 1H, 100-H), 7.78 (dd, J = 9.0, 2.7 Hz,1H, 6-H), 7.60 (d, J = 2.7 Hz, 1H, 8-H), 6.80 (d, J = 9.0 Hz, 1H, 5-H), 4.06 (s, 2H, 3- H), 3.37 (m, 4H, 10-H, 200-,600-He), 3.00 (m, 2H, 50-H), 2.83 (m, 2H, 200-,600-Ha), 1.67 (m, 9H, 20-,40-,300-,500-H, 400-HA), 1.33 (m, 3H, 30-H, 400-HB); 13C NMR (DMSO-d6): d163.85 (C-2), 140.26 (C-4a), 136.43 (C-7), 125.59 (C-8a), 120.59 (C-6), 109.66, 109.28 (C-5, -8), 55.59 (C-50), 51.95 (C-200, -600), 51.06 (C-3), 48.81 (C-10), 24.02, 23.31, 22.93 (C-20, -30, -40), 22.45 (C-300, -500), 21.34 (C-400). MS (ES+): m/z (%) 775 (8) ([2 M�Br+2]+), 773 (8) ([2 M�Br]+), 348 (25) ([M�Br+1]+), 347 (100) ([M�Br]+). Anal. Calcd for C18H27BrN4O3 (427.34): C 50.59; H 6.37; N 13.11. Found: C 50.50; H 6.47; N 13.32. 4.2.2.10. 4-(5-Azepanylpentyl)-7-nitro-3,4-dihydro-1H-quinox- alin-2-one hydrobromide (12). Yield: 1.28 g (97%); mp 235–237 �C (decomp.) (methanol); 1H NMR (DMSO-d6): d 10.82 (s, 1H, 1-H), 9.13 (br s, 1H, 100-H), 7.79 (dd, J = 9.0, 2.7 Hz,1H, 6- H), 7.61 (d, J = 2.7 Hz, 1H, 8-H), 6.80 (d, J = 9.0 Hz, 1H, 5-H), 4.06 (s, 2H, 3-H), 3.35 (m, 4H, 10-H, 200-,700-HA), 3.06 (m, 4H, 50-H, 200- ,700-HB), 1.90–1.50 (m, 12H, 20-,40-,300-,400-,500-,600-H), 1.33 (m, 2H, 30-H); 13C NMR (DMSO-d6): d 163.83 (C-2), 140.26 (C-4a), 136.42 (C-7), 125.58 (C-8a), 120.58 (C-6), 109.64, 109.27 (C-5, -8), 56.06 (C-50), 53.53 (C-200, -700), 51.06 (C-3), 48.82 (C-10), 25.94 (C-300, -600), 24.04, 23.32, 23.29 (C-20, -30, -40), 22.86 (C-400, -500). MS (ES+): m/z (%) 803 (18) ([2 M�Br+2]+), 801 (17) ([2 M�Br]+), 362 (24) ([M�Br+1]+), 361 (100) ([M�Br]+). Anal. Calcd for C19H29BrN4O3 (441.36): C 51.70; H 6.62; N 12.69. Found: C 51.98; H 6.67; N 12.62. 4.2.2.11. 7-Nitro-4-[5-(1,2,3,4-tetrahydroisoquinolin-2-yl)pen- tyl]-3,4-dihydro-1H-quinoxalin-2-one hydrobromide (13) Yield: 1.34 g (94%); mp 196–198 �C (decomp.) (0.5 M aq HBr); 1H NMR (DMSO-d6): d 10.83 (s, 1H, 1-H), 9.73 (br s, 1H, 200-H), 7.79 (dd, J = 9.3, 2.7 Hz,1H, 6-H), 7.61 (d, J = 2.7 Hz, 1H, 8-H), 7.32-7.16 (m, 4H, 500-,600-,700-,800-H), 6.82 (d, J = 9.3 Hz, 1H, 5-H), 4.55 [br d, J = (�)15.3 Hz, 100-HA], 4.29 [br dd, J = (�)15.3, 8.3 Hz, 100-HB], 4.07 (s, 2H, 3-H), 3.70 (m, 1H, 300-HA), 3.35 (t, J = 7.2 Hz, 2H, 10-H), 3.30–2.95 (m, 5H, 50-,400-H, 300-HB), 1.79 (m, 2H, 40-H), 1.63 (m, 2H, 20-H), 1.38 (m, 2H, 30-H). 13C NMR (DMSO-d6): d 163.73 (C-2), 140.19 (C-4a), 136.44 (C-7), 131.25, 128.48, 126.59, 126.55 (C-500, -600, -700, -800), 128.31, 127.64 (C-400a, -800a), 125.55 (C-8a), 120.48 (C-6), 109.62 (C-8), 109.25 (C-5),54.90 (C-50),51.71 (C-100), 51.05 (C-3), 48.81 (C-10), 48.76 (C-300), 24.78 (C-400), 24.00 (C-20), 23.21 (C-30), 23.03 (C-40). MS (ES+): m/z (%) 871 (6) ([2 M�Br+2]+), 869 (6) ([2 M�Br]+), 396 (28) ([M�Br+1]+), 395 (100) ([M�Br]+). Anal. Calcd for C22H27BrN4O3 (475.38): C 55.58; H 5.72; N 11.79. Found: C 55.50; H 5.67; N 11.52. M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 1583 4.2.2.12. 4-[5-(Dimethylamino)pentyl]-1-methyl-7-nitro-3,4- dihydro-1H-quinoxalin-2-one hydrobromide (14). Yield: 1.00 g (83%); mp 182–184 �C (decomp.) (ethanol); 1H NMR (DMSO-d6): d 9.38 (br s, 1H, 50-NH+), 7.88 (dd, J = 9.3, 2.4 Hz,1H, 6-H), 7.70 (d, J = 2.4 Hz, 1H, 8-H), 6.88 (d, J = 9.3 Hz, 1H, 5-H), 4.13 (s, 2H, 3-H), 3.38 (t, J = 7.1 Hz, 2H, 10-H), 3.32 (s, 3H, 1-CH3), 3.03 (m, 2H, 50-H), 2.74 [s, 6H, N(CH3)2], 1.63 (m, 4H, 20-,40-H), 1.34 (m, 2H, 30-H); 13C NMR (DMSO-d6): d 163.35 (C-2), 141.81 (C-4a), 136.88 (C-7), 127.70 (C-8a), 120.85 (C-6), 109.78, 109.61 (C-5, -8), 56.35 (C-50), 51.13 (C-3), 48.91 (C-10), 42.10 [N(CH3)2], 28.30 (1-CH3), 23.99, 23.42, 23.09 (C-20, -30, -40); MS (ES+): m/z (%) 723 (8) ([2 M�Br+2]+), 721 (8) ([2 M�Br]+), 322 (20) ([M�Br+1]+), 321 (100) ([M�Br]+). Anal. Calcd for C16H25BrN4O3 (401.30): C 47.89; H 6.28; N 13.96. Found: C 47.64; H 6.47; N 13.92. 4.2.2.13. 1-Methyl-7-nitro-4-(5-pyrrolidinopentyl)-3,4-dihydro- 1H-quinoxalin-2-one hydrobromide (15). Yield: 1.13 g (88%); mp 206–208 �C (decomp.) (ethanol).;1H NMR (DMSO-d6): d 9.46 (br s, 1H, 100-H), 7.90 (dd, J = 9.0, 2.4 Hz,1H, 6-H), 7.72 (d, J = 2.4 Hz, 1H, 8-H), 6.87 (d, J = 9.0 Hz, 1H, 5-H), 4.14 (s, 2H, 3-H), 3.50 (br s, 2H, 200-,500-HA), 3.38 (t, J = 7.3 Hz, 2H, 10-H), 3.33 (s, 3H, 1-CH3), 3.10 (m, 2H, 50-H), 2.96 (br s, 2H, 200-,500-HB), 1.97 (br s, 2H) and 1.83 (br s, 2H) (300-,400-H), 1.61 (m, 4H, 20-,40-H), 1.35 (m, 2H, 30-H).); 13C NMR (DMSO-d6): d 163.34 (C-2), 141.79 (C-4a), 136.84 (C-7), 127.67 (C-8a), 120.87 (C-6), 109.76, 109.63 (C-5, -8), 53.62 (C-50), 53.01 (C-200, -500), 51.12 (C-3), 48.94 (C-10), 28.30 (1-CH3), 24.88, 23.96, 23.24 (C-20, -30, -40), 22.58 (C-300, -400); MS (ES+): m/z (%) 775 (15) ([2 M�Br+2]+), 773 (15) ([2 M�Br]+), 348 (24) ([M�Br+1]+), 347 (100) ([M�Br]+). Anal. Calcd for C18H27BrN4O3 (427.34): C 50.59; H 6.37; N 13.11. Found: C 50.33; H 6.61; N 13.33. 4.2.2.14. 1-Methyl-7-nitro-4-(5-piperidinopentyl)-3,4-dihydro- 1H-quinoxalin-2-one hydrobromide (16). Yield: 1.24 g (94%); mp 214–216 �C (decomp.) (ethanol); 1H NMR (DMSO-d6): d 9.24 (br s, 1H, 100-H), 7.88 (dd, J = 9.3, 2.4 Hz,1H, 6-H), 7.69 (d, J = 2.4 Hz, 1H, 8-H), 6.88 (d, J = 9.3 Hz, 1H, 5-H), 4.13 (s, 2H, 3-H), 3.37 (m, 4H, 10-H, 200-,600-He), 3.31 (s, 3H, 1-CH3), 3.00 (m, 2H, 50-H), 2.85 (m, 2H, 200-,600-Ha), 1.90-1.50 (m, 9H, 20-,40-,300-,500-H, 400-HA), 1.33 (m, 3H, 30-H, 400-HB); 13C NMR (DMSO-d6): d 163.34 (C-2), 141.79 (C-4a), 136.84 (C-7), 127.68 (C-8a), 120.87 (C-6), 109.76, 109.62 (C-5, -8), 55.58 (C-50), 51.93 (C-200, -600), 51.12 (C-3), 48.91 (C-10), 28.30 (1-CH3), 23.96, 23.31, 22.89 (C-20, -30, -40), 22.43 (C-300, -500), 21.34 (C-400); MS (ES+): m/z (%) 803 (15) ([2 M�Br+2]+), 801 (13) ([2 M�Br]+), 362 (24) ([M�Br+1]+), 361 (100) ([M-Br]+). Anal. Calcd for C19H29BrN4O3 (441.36): C 51.70; H 6.62; N 12.69. Found: C 51.57; H 6.67; N 12.45. 4.2.2.15. 4-(5-Azepanylpentyl)-1-methyl-7-nitro-3,4-dihydro- 1H-quinoxalin-2-one hydrobromide (17). Yield: 1.24 g (91%); mp 223–225 �C (decomp.) (ethanol); 1H NMR (DMSO-d6): d 9.11 (br s, 1H, 100-H), 7.90 (dd, J = 9.0, 2.6 Hz,1H, 6-H), 7.72 (d, J = 2.6 Hz, 1H, 8-H), 6.87 (d, J = 9.0 Hz, 1H, 5-H), 4.13 (s, 2H, 3-H), 3.39 (m, 4H, 10-H, 200-,700-HA), 3.33 (s, 3H, 1-CH3), 3.06 (m, 4H, 50- H, 200-,700-HB), 1.90-1.50 (m, 12H, 20-,40-,300-,400-,500-,600-H), 1.33 (m, 2H, 30-H); 13C NMR (DMSO-d6): d 163.35 (C-2), 140.81 (C-4a), 136.87 (C-7), 127.70 (C-8a), 120.87 (C-6), 109.77, 109.62 (C-5, -8), 56.07 (C-50), 53.57 (C-200, -700), 51.13 (C-3), 48.92 (C-10), 28.30 (1-CH3), 25.93 (C-300, -600), 24.01, 23.31 (2C) (C-20, -30, -40), 22.89 (C-400, -500). MS (ES+): m/z (%) 831 (13) ([2 M�Br+2]+), 829 (12) ([2 M�Br]+), 376 (28) ([M�Br+1]+), 375 (100) ([M�Br]+). Anal. Calcd for C20H31BrN4O3 (455.39): C 52.75; H 6.86; N 12.30. Found: C 52.49; H 6.59; N 12.51. 4.2.2.16. 1-Methyl-7-nitro-4-[5-(1,2,3,4-tetrahydroisoquinolin- 2-yl)pentyl]-3,4-dihydro-1H-quinoxalin-2-one hydrobromide (18). Yield: 1.29 g (88%); mp 184–187 �C (decomp.) (metha- nol); 1H NMR (DMSO-d6): d 9.95 (br s, 1H, 200-H), 7.89 (dd, J = 9.1, 2.4 Hz,1H, 6-H), 7.70 (d, J = 2.4 Hz, 1H, 8-H), 7.32-7.14 (m, 4H, 500- ,600-,700-,800-H), 6.90 (d, J = 9.1 Hz, 1H, 5-H), 4.54 (br s, 100-HA), 4.34 (br s, 100-HB), 4.15 (s, 2H, 3-H), 3.71 (m, 1H, 300-HA), 3.41 (t, J = 7.1 Hz, 2H, 10-H), 3.32 (s, 3H, 1-CH3), 3.30-2.95 (m, 5H, 50-, 400-H, 300-HB), 1.83 (m, 2H, 40-H), 1.63 (m, 2H, 20-H), 1.40 (m, 2H, 30-H). 13C NMR (DMSO-d6): d 163.35 (C-2), 141.80 (C-4a), 136.86 (C-7), 131.31 (CH), 128.58 (CH), 128.41 (Cipso), 127.74 (Cipso), 127.68 (Cipso), 126.68 (CH), 126.63 (CH) (C-8a, -400a, -500, -600, -700, -800, -800a), 120.88 (C-6),109.76 (C-8), 109.64 (C-5), 54.95 (C-50), 51.81 (C-100), 51.16 (C-3), 48.93 (C-10), 48.85 (C-300), 28.31 (1-CH3), 24.87 (C-400), 24.00 (C-20), 23.26 (C-30), 23.13 (C-40). MS (ES+): m/z (%) 897 (7) ([2 M�Br+2]+), 895 (7) ([2 M�Br]+), 410 (29) ([M�Br+1]+), 409 (100) ([M�Br]+). Anal. Calcd for C23H29BrN4O3 (489.41): C 56.45; H 5.97; N 11.45. Found: C 56.57; H 6.21; N 11.69. 4.3. Wet evaluation: pharmacological assays 4.3.1. Determination of in vitro trichomonacidal activity The biological activity was assayed on Trichomonas vaginalis JH31A #4 Ref. No. 30326 (ATCC, MD, USA) in modified Diamond medium supplemented with equine serum and grown at 37 �C (5% CO2). The compounds were added to the cultures at several con- centrations (100, 10, and 1 lg/mL) after 6 h of seeding (0 h). Viable protozoa were assessed at 24 and 48 h after incubation at 37 �C by using the Neubauer chamber. Metronidazole (Sigma-Aldrich SA, Spain) was used as a reference drug at concentrations of 2, 1, 0.5 lg/mL. Cytocidal and cytostatic activities were determined by calculation of percentages of cytocidal (%C) and cytostatic activities (%CA), in relation to controls as previously reported.52 4.3.2. T. cruzi epimastigote susceptibility assay For this in vitro test,25,53the CL strain parasites (clone CL-B5) stably transfected with the Escherichia coli b-galactosidase gene (LacZ) were used. The epimastigotes were grown at 28 �C in liver infusion tryptose broth (LIT) with 10% foetal bovine serum (FBS), penicillin and streptomycin and harvested during the exponential growth phase. The screening assay was performed in 96-well microplates (Sarstedt, Sarstedt, Inc.) with cultures that had not reached the stationary phase. Briefly, epimastigotes form, CL strain, was seeded at concentration of 1 � 105 per milliliter in 200 lL media. The plates were then incubated at 28 �C for 72 h with var- ious concentrations of the drugs (100, 10 and 1 lg/mL), at which time 50 lL of CPRG solution was added to give a final concentra- tion of 200 lM. The plates were incubated at 37 �C for 6 hrs and the absorbances read at 595 nm. Each concentration was tested in triplicate and in order to avoid drawback, medium, negative and drug controls were used in each test. The anti-epimastigote percentage (%AE) was calculated as follows: %AE = [(AE � AEB)/ (AC � ACB)] � 100, where AE = absorbance of experimental group; AEB = blank of compounds; AC = Absorbance of control group; ACB = blank of culture medium. Stock solutions of the compounds to be assayed were prepared in DMSO, with the final concentration in a water/DMSO mixture never exceeding 0.2% of the latter sol- vent.25,53 Nifurtimox was used as reference drug. 4.3.3. In vitro cytotoxicity on macrophage cells Murine J774 macrophages were grown in plastic 25 lL flasks in (RPMI)-1640 medium (Sigma) supplemented with 20% heat inacti- vated (30 min, 56 �C) foetal calf serum (FCS) and 100 IU penicillin/ mL + 100 lg/mL streptomycin, in a humidified 5% CO2/95% air atmosphere at 37 �C and subpassaged once a week. The J774 mac- rophages were seeded (70,000 cells/well) in 96-well flat-bottom microplates (Nunc) with 200 lL of medium. The cells were allowed to attach for 24 h at 37 �C and then exposed to the compounds (dissolved in DMSO, maximal final concentration of solvent was 0.2%) for another 24 h. Afterwards, the cells were washed with 1584 M. A. Martins Alho et al. / Bioorg. Med. Chem. 22 (2014) 1568–1585 PBS and incubated (37 �C) with 3-(4,5-dimethylthiazol-2-yl)-2,5- diphenyltetrazolium bromide (MTT) 0.4 mg/mL for 60 min. The MTT solution was removed and the cells solubilized in DMSO (10 lL). The extent of reduction of MTT to formazan within cells was quantified by measurement of OD595.54 Each concentration was assayed three times and six cell growth controls were used in each test. The assays were performed in duplicate. Nifurtimox cytotox- icty was also determined. Cytotoxic percentages (%C) were deter- mined as follows: %C = [1 � (ODp � ODpm)/(ODc � ODm)] � 100, where ODp represents the mean OD595 value recorded for wells with macrophages containing different doses of product; ODpm represents the mean OD595 value recorded for different concen- trations of product in medium; ODc represents the mean OD595 value recorded for wells with macrophages and no product (growth controls), and ODm represents the mean OD595 value re- corded for medium/control wells. The 50% cytotoxic dose (CD50) was defined as the concentration of drug that decreases OD595 up to 50% of that in control cultures.25 4.3.4. L. braziliensis promastigotes susceptibility test The tested chemicals were solubilized in DMSO (Sigma) to pre- pare a working solution of 10 mg/mL. Later on it was diluted in RPMI 1640 medium to the final highest concentration of DMSO on 1.5%, which was not toxic to the parasites. In this study we used L. braziliensis (MHOM/PE/95/LQ2), which was isolated in the province of La Convención, Cuzco, Perú. Cul- tures were handled as previously described.55 Promastigotes were adapted for culture in RPMI 1640 liquid medium (Gibco-BRL) sup- plemented with 20% heat inactivated fetal bovine serum, vitamins and amino acids, at 22 �C. Logarithm phase cultures of promastig- otes were used for experimental purposes. The inhibition of promastigotes growth in vitro was assessed by using a quantitative colorimetric assay with the oxidation–reduc- tion indicator Alamar Blue� Assay.56Briefly, promastigotes were serially diluted in 200 lL RPMI 1640 medium without phenol red and supplemented with 20% heat-inactivated fetal bovine serum in 96-well plates. To these wells were added parasites (106/well), and the drug concentration to be tested. After addition of 10% of Alamar Blue�, the plates were incubated at 22 �C. After 72 h, the plates were analyzed on a Microplate Reader Model 680 (Biorad, Hercules, CA) by using a test wavelength of 570 nm and a reference wavelength of 630 nm. The 50% inhibitory concentrations (IC50) were calculated by linear regression analysis with 95% confidence limits. All experiments were performed three times each in dupli- cate, and the mean values were also calculated. A paired two-tailed t-test was used for analysis of the data. Values of p <0.05 were con- sidered significant. 4.3.5. In vitro efficacy studies with Toxoplasma gondii tachyzoites The efficacy of chemicals was tested against tachyzoites form of Toxoplasma gondii.44,45 Tachyzoites (1 � 106)weresettled in epen- dorf microtubes (500 lL, Axygen Scientific), and exposed to com- pounds 9–18 for four hours at room temperature in order to evaluate the viability of the parasites. One hundred and fifty tach- yzoites were counted and the viability percentage was taken with trypan blue exclusion method by counting the number of living tachyzoites. All chemicals were first dissolved in dimethyl sulfoxide [DMSO, sigma, 99.5% (GC)], and then diluted in BME (basal medium eagle) Sigma-Aldrich. The compounds were assayed in the range of 1 mM, 500 lM, 200 lM, 100 lM. The final concentration of DMSO did not exceed 0.2% which caused no damage to t