Potential antitumoral 3,4-dihydropyrimidin-2-(1H)-ones: synthesis, in vitro biological evaluation and QSAR studies

Abstract

The search for novel anticancer agents with higher selectivity and lower toxicity remains a priority. This work aimed to design more potent and selective anticancer molecules among the class of 3,4-dihydropyrimidin-2-(1H)-ones. Thus, a series of molecules was synthesized through the Biginelli reaction and their in vitro antiproliferative activity was evaluated in different human cell lines. Then, a quantitative structure–activity relationship (QSAR) analysis was performed using Bayesian regularized artificial neural networks to model the relationships between in silico molecular descriptors and the observed antiproliferative activity of molecules across the tested cell lines. Interestingly, among the compounds prepared, the molecules containing chloro atoms in their structure demonstrated a relevant potency and a selective antiproliferative activity against a novel hepatic cancer cell line (HepaRG) without exhibiting noticeable cytotoxicity in normal dermal cells (NHDF). However, in prostatic (LNCaP), colon (Caco-2) and breast (T47D and MCF-7) cancer cell lines generally the compounds did not exhibit relevant cytoxicity. A statistically valid QSAR model was obtained (internal validation Q² = 0.663, RMSE_CV = 0.071, 10-fold cross-validation procedure, and external validation R_pred² = 0.740, RMSE = 0.077), which allowed the analysis of the involved relationships between molecular descriptors and the reliable prediction of the antiproliferative activity for hypothetical related compounds in the studied cell lines. Moreover, flow cytometry analysis showed that in HepaRG and MCF-7 cell lines, compound 16 did not decrease cell viability but, interestingly, led to an accumulation of cells in the G₀/G₁ phase of the cell cycle. Therefore, chlorinated 3,4-dihydropyrimidin-2-(1H)-ones may be considered promising compounds for further optimization as new antitumor agents.

---

Introduction

A multicomponent reaction (MCR) is generally defined as the process in which three or more reactants are combined in a single step to form a product that incorporates the structural characteristics of each reagent and can allow the synthesis of small drug-like molecules with several degrees of structural diversity.^1,2 MRCs can represent a form of combinatorial chemistry and diversity-oriented synthesis, and their interest has increased in the development of modern synthetic methodology in drug discovery research. MCRs offer simplicity, economical advantages and can allow the synthesis of a wide amount of molecules, which generally are purer than the final products from multi-step reactions.^3,4

One of the most famous MCR is the Biginelli reaction, which produces an interesting class of nitrogen heterocycles known as 3,4-dihydropyrimidin-2-(1H)-ones (DHPMs). The original reaction was firstly reported by Pietro Biginelli in 1893, who obtained DHPMs by refluxing a mixture of an aldehyde, a β-ketoester, and urea in the presence of an acid catalyst.⁵ Due to its relevance, over the years several approaches to perform this transformation were described, involving namely the use of microwaves, sonication,⁶ ionic liquids,⁷ as well as different catalysts.⁸

The functionalized DHPMs possess a wide spectrum of biological and pharmacological activities, which adds it even more interest to both synthetic and medicinal chemists. In this context, this class of heterocycles revealed to have, for example, anticancer,⁹ anti-malarial,¹⁰ antifungal and antibacterial,^11,12 anti-inflammatory,¹³ antioxidant¹⁴ and anti-thyroid¹⁵ properties. Additionally, they showed to be calcium channel modulators,¹⁶ HIV replication inhibitors,¹⁷ and melanin concentrating hormone receptor inhibitors.¹⁸

Regarding the anticancer activity, different molecules synthesized through Biginelli reaction have already been studied. Within these, some Biginelli adducts were reported as potent growth inhibitors of several cancer cell lines, such as ovarian, renal, non-small lung, prostatic, colon and glioma.¹⁹ Moreover, additional DHPMs have been evaluated for antitumoral effects in a mdr1-gene transfected mouse lymphoma cell line²⁰ and a series of 3,4-dihydropyrimidin-2-(1H)-thiones was also evaluated against several cancer cell lines²¹ with promising results.

Nowadays, the rational design of molecules with improved activity/selectivity can be assisted by quantitative structure–activity relationship (QSAR) studies.^22,23 QSAR studies can be performed by using different approaches, namely by means of Bayesian regularized artificial neural networks (BRANNs). BRANN models employ Bayesian inference to optimize the weights and regularization constant values by determining the posterior probability distribution of weights and related properties from a prior probability distribution. This enables the development of robust models, in which overfitting, architecture dependence and non-existent or redundant relationships are overcome.²⁴

Taking into account the interesting results of some DHPM as anticancer candidates, this work aimed the development of more potent and selective potential antitumoral molecules among the class of DHPMs. Specifically, it involved the synthesis of several DHPMs, with focus on halogenated molecules, some of which were not found in the literature, by means of the Biginelli reaction. The synthesis was followed by evaluation of the compounds effect on the in vitro proliferation of different human cancer cell lines, and analysis of the relationships between in silico calculated molecular descriptors and bioactivity by QSAR modelling. Finally, a cell cycle distribution assay was carried out to clarify the mechanism of cytotoxicity.

Experimental section

General information

The reagents urea, benzaldehyde, p-tolualdehyde, p-nitrobenzaldehyde, 2,4-dichlorobenzaldehyde and 2,3-difluorobenzaldehyde were purchased from Acros Organics (New Jersey, USA), ethyl acetoacetate, methyl acetoacetate, acetylacetone and anisaldehyde were purchased from Merck (Hohenbrunn, Germany) and 2,3-dichlorobenzaldehyde, furaldehyde, bismuth(III) nitrate pentahydrate, 5-fluorouracil (5-FU) and dimethyl sulfoxide (DMSO) were purchased from Sigma-Aldrich (St. Louis, MO, USA). The ethanol 99.9% was purchased from Manuel Vieira & C^a (Torres Novas, Portugal) and deuterated DMSO (DMSO-d₆) was purchased from Armar Chemicals (Leipzig, Germany). Infrared (IR) spectra were collected on a Thermoscientific Nicolet iS10: smart iTR, equipped with a diamond ATR crystal. For ATR data acquisition, a sample of the solid compound was placed onto the crystal and the spectrum was recorded. An air spectrum was used as a reference in absorbance calculations. The sample spectra were collected at room temperature in the 4000–400 cm⁻¹ range by averaging 16 scans at a spectral resolution of 2 cm⁻¹. Nuclear magnetic resonance (NMR) spectra (¹H-NMR and ¹³C-NMR) were acquired on a Bruker Avance 400 MHz spectrometer and were processed with the software TOPSPIN 3.1 (Bruker, Fitchburg, WI, USA). DMSO-d₆ was used as solvent. Chemical shifts are reported in parts per million (ppm) relative to deuterated solvent as an internal standard. Coupling constants (J values) are reported in hertz (Hz) and splitting multiplicities are described as s = singlet; brs = broad singlet; d = doublet; dd = double doublet; t = triplet; q = quartet; dq = double quartet and m = multiplet. High resolution mass spectrometry (ESI-HRMS) were performed by the microanalysis service on a QSTAR XL instrument (Salamanca, Spain).

Synthesis and structural characterization

To a mixture of an aldehyde (1 mmol), a β-ketoester/acetylacetone (1 mmol) and urea (1.3 mmol) was added bismuth(III) nitrate pentahydrate (0.1 mmol, 10 mol%) and the reaction was heated with stirring at 70 °C in a preheated oil bath for the appropriate time (Table 1).²⁵ The reaction was considered completed when solidified and, after being cooled to room temperature, was poured onto cold water and stirred from 20–30 min. The solid separated was filtered under suction, washed with ice-cold water, dried and then recrystallized from ethanol 99.9% to afford the pure product. All the products were characterized by IR, ¹H- and ¹³C-NMR. All compounds were determined to be >95% pure by ¹H-NMR. High resolution mass spectrum (HRMS) was also obtained for the new compounds.

Table 1 Bi(NO₃)₃·5H₂O-catalyzed synthesis of 3,4-dihydropyrimidin-2-(1H)-ones under solvent-free conditions at 70 °C^a

Compound	R	R′	Time (min)	Yield^b^,^c (%)
a Reaction conditions: aldehyde (1 mmol), β-ketoester/acetylacetone (1 mmol), urea (1.3 mmol), Bi(NO3)3·5H2O (10 mol%) at 70 °C.b Yield of isolated products after purification.c All products were characterized by ^1H- and ^13C-NMR, IR spectra and compared with available data in the literature.
1	C6H5	OCH2CH3	7	81
2	C6H5	OCH3	14	80
3	C6H5	CH3	6	93
4	4-(CH3)C6H4	OCH2CH3	12	74
5	4-(CH3)C6H4	OCH3	9	72
6	4-(CH3)C6H4	CH3	7	60
7	4-(NO2)C6H4	OCH2CH3	5	67
8	4-(NO2)C6H4	OCH3	8	61
9	4-(NO2)C6H4	CH3	5	71
10	4-(OCH3)C6H4	OCH2CH3	7	79
11	4-(OCH3)C6H4	OCH3	7	81
12	4-(OCH3)C6H4	CH3	30	64
13	2,3-(Cl)2C6H3	OCH2CH3	8	32
14	2,3-(Cl)2C6H3	CH3	8	45
15	2,4-(Cl)2C6H3	OCH2CH3	5	46
16	2,4-(Cl)2C6H3	OCH3	6	51
17	2,4-(Cl)2C6H3	CH3	5	50
18	2,3-(F)2C6H3	OCH2CH3	5	37
19	2,3-(F)2C6H3	OCH3	8	58
20	2,3-(F)2C6H3	CH3	7	52
21	2-Furyl	OCH2CH3	13	81
22	2-Furyl	OCH3	12	74
23	2-Furyl	CH3	18	76

---

Ethyl 6-methyl-2-oxo-4-phenyl-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 1). Yield: 81%, IR (ν_max/cm⁻¹): 3233, 3111, 2978, 1697, 1642, 1217, 1086; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.09 (t, 3H, J = 7.05 Hz, OCH₂CH₃), 2.24 (s, 3H, CH₃), 3.98 (q, 2H, J = 7.05 Hz, OCH₂), 5.14 (d, 1H, J = 3.04 Hz, CH), 7.20–7.36 (m, 5H, ArH), 7.73 (brs, 1H, NH), 9.18 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 14.06, 17.77, 53.95, 59.17, 99.25, 126.23, 127.25, 128.38, 144.86, 148.35, 152.12, 165.33.

Methyl 6-methyl-2-oxo-4-phenyl-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 2). Yield: 80%, IR (ν_max/cm⁻¹): 3328, 3213, 3103, 1692, 1664, 1412, 1236, 1092; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.25 (s, 3H, CH₃), 3.53 (s, 3H, CH₃OCO), 5.14 (d, 1H, J = 3.35 Hz, CH), 7.21–7.35 (m, 5H, ArH), 7.75 (brs, 1H, NH), 9.22 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.83, 50.78, 53.80, 99.00, 126.17, 127.29, 128.45, 144.67, 148.67, 152.18, 165.84.

5-Acetyl-6-methyl-4-phenyl-3,4-dihydropyrimidin-2(1H)-one (compound 3). Yield: 93%, IR (ν_max/cm⁻¹): 3329, 3253, 1700, 1673, 1597, 1234; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.10 (s, 3H, CH₃), 2.29 (s, 3H, CH₃CO), 5.26 (d, 1H, J = 3.39 Hz, CH), 7.21–7.36 (m, 5H, ArH), 7.82 (brs, 1H, NH), 9.18 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 18.90, 30.31, 53.80, 109.58, 126.42, 127.33, 128.51, 144.24, 148.12, 152.12, 194.25.

Ethyl 6-methyl-2-oxo-4-p-tolyl-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 4). Yield: 74%, IR (ν_max/cm⁻¹): 3238, 3112, 2981, 1700, 1645, 1217, 1086; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.10 (t, 3H, J = 7.05 Hz, OCH₂CH₃), 2.24 (s, 3H, CH₃), 2.26 (s, 3H, CH₃), 3.98 (q, 2H, J = 7.04 Hz, OCH₂), 5.10 (d, 1H, J = 3.23 Hz, CH), 7.12 (s, 4H, ArH), 7.68 (brs, 1H, NH), 9.15 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 14.09, 17.75, 20.63, 53.61, 59.14, 99.40, 126.13, 128.87, 136.35, 141.95, 148.14, 152.16, 165.35.

Methyl 6-methyl-2-oxo-4-p-tolyl-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 5). Yield: 72%, IR (ν_max/cm⁻¹): 3331, 3212, 3105, 2955, 1694, 1665, 1238, 1090; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.24 (s, 3H, CH₃), 2.26 (s, 3H, CH₃), 3.52 (s, 3H, CH₃OCO), 5.10 (d, 1H, J = 3.17 Hz, CH), 7.11 (s, 4H, ArH), 7.70 (brs, 1H, NH), 9.18 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.81, 20.64, 50.76, 53.48, 99.12, 126.08, 128.95, 136.40, 141.76, 148.47, 152.17, 165.85.

5-Acetyl-6-methyl-4-p-tolyl-3,4-dihydropyrimidin-2(1H)-one (compound 6). Yield: 60%, IR (ν_max/cm⁻¹): 3285, 3118, 2918, 1697, 1615, 1234; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.08 (s, 3H, CH₃), 2.26 (s, 3H, CH₃), 2.27 (s, 3H, CH₃CO), 5.21 (d, 1H, J = 3.22 Hz, CH), 7.12 (s, 4H, ArH), 7.77 (brs, 1H, NH), 9.14 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 18.85, 20.64, 30.21, 53.56, 109.54, 126.35, 129.02, 136.49, 141.30, 147.91, 152.11, 194.58.

Ethyl 6-methyl-4-(4-nitrophenyl)-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 7). Yield: 67%, IR (ν_max/cm⁻¹): 3228, 3113, 2976, 1698, 1640, 1517, 1210, 1084; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.09 (t, 3H, J = 7.09 Hz, OCH₂CH₃), 2.26 (s, 3H, CH₃), 3.98 (q, 2H, J = 7.10 Hz, OCH₂), 5.27 (d, 1H, J = 3.15 Hz, CH), 7.50 (d, 2H, J = 8.71 Hz, ArH), 7.88 (brs, 1H, NH), 8.22 (d, 2H, J = 8.70 Hz, ArH), 9.35 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 14.05, 17.86, 53.68, 59.38, 98.17, 123.83, 127.65, 146.72, 149.39, 151.73, 151.99, 165.05.

Methyl 6-methyl-4-(4-nitrophenyl)-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 8). Yield: 61%, IR (ν_max/cm⁻¹): 3364, 3221, 3114, 2950, 1689, 1638, 1514, 1225, 1093; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.27 (s, 3H, CH₃), 3.54 (s, 3H, CH₃OCO), 5.27 (d, 1H, J = 3.27 Hz, CH), 7.51 (d, 2H, J = 8.78 Hz, ArH), 7.90 (brs, 1H, NH), 8.21 (d, 2H, J = 8.78 Hz, ArH), 9.37 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.93, 50.92, 53.53, 97.97, 123.88, 127.61, 146.75, 149.64, 151.78, 165.57.

5-Acetyl-6-methyl-4-(4-nitrophenyl)-3,4-dihydropyrimidin-2(1H)-one (compound 9). Yield: 71%, IR (ν_max/cm⁻¹): 3247, 3110, 2945, 1672, 1606, 1514, 1234; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.18 (s, 3H, CH₃), 2.31 (s, 3H, CH₃CO), 5.38 (d, 1H, J = 3.42 Hz, CH), 7.50 (d, 2H, J = 8.64 Hz, ArH), 7.98 (brs, 1H, NH), 8.20 (d, 2H, J = 8.64 Hz, ArH), 9.33 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 19.18, 30.70, 53.20, 109.54, 123.88, 127.75, 146.75, 149.17, 151.60, 152.06, 194.07.

Ethyl 4-(4-methoxyphenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 10). Yield: 79%, IR (ν_max/cm⁻¹): 3235, 3104, 2956, 1703, 1647, 1218, 1085; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.10 (t, 3H, J = 7.12 Hz, OCH₂CH₃), 2.24 (s, 3H, CH₃), 3.72 (s, 3H, OCH₃), 3.98 (q, 2H, J = 7.12 Hz, OCH₂), 5.09 (d, 1H, J = 3.20 Hz, CH), 6.87 (d, 2H, J = 8.67 Hz, ArH), 7.14 (d, 2H, J = 8.67 Hz, ArH), 7.66 (brs, 1H, NH), 9.14 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 14.10, 17.74, 53.32, 55.04, 59.14, 99.55, 113.69, 127.38, 137.05, 148.00, 152.14, 158.43, 165.36.

Methyl 4-(4-methoxyphenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 11). Yield: 81%, IR (ν_max/cm⁻¹): 3240, 3108, 2955, 1710, 1680, 1651, 1236, 1095; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.24 (s, 3H, CH₃), 3.52 (s, 3H, CH₃OCO), 3.72 (s, 3H, OCH₃), 5.09 (d, 1H, J = 3.25 Hz, CH), 6.87 (d, 2H, J = 8.63 Hz, ArH), 7.14 (d, 2H, J = 8.63 Hz, ArH), 7.68 (brs, 1H, NH), 9.17 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.80, 50.76, 53.18, 55.05, 99.28, 113.76, 127.33, 136.85, 148.34, 152.16, 158.46, 165.86.

5-Acetyl-4-(4-methoxyphenyl)-6-methyl-3,4-dihydropyrimidin-2(1H)-one (compound 12). Yield: 64%, IR (ν_max/cm⁻¹): 3298, 3218, 1694, 1609, 1511, 1248, 1175; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.07 (s, 3H, CH₃), 2.27 (s, 3H, CH₃CO), 3.72 (s, 3H, OCH₃), 5.20 (d, 1H, J = 3.18 Hz, CH), 6.88 (d, 2H, J = 8.56 Hz, ArH), 7.16 (d, 2H, J = 8.56 Hz, ArH), 7.75 (brs, 1H, NH), 9.14 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 18.83, 30.16, 53.31, 55.06, 109.60, 113.85, 127.63, 136.37, 147.79, 152.08, 158.50, 194.36.

Ethyl 4-(2,3-dichlorophenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 13). Yield: 32%, IR (ν_max/cm⁻¹): 3305, 3235, 3100, 2968, 1697, 1638, 1558, 1215, 1085; ¹H-NMR (400 MHz, DMSO-d₆) δ: 0.97 (t, 3H, J = 7.08 Hz, OCH₂CH₃), 2.30 (s, 3H, CH₃), 3.89 (q, 2H, J = 7.08 Hz, OCH₂), 5.68 (dd, 1H, J = 2.67 Hz, CH), 7.29 (dd, 1H, J₁ = 7.81 Hz, J₂ = 1.58 Hz ArH), 7.35 (t, 1H, J = 7.81 Hz, ArH), 7.54 (dd, 1H, J₁ = 7.81 Hz, J₂ = 1.57 Hz, ArH), 7.78 (brs, 1H, NH), 9.32 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 13.86, 17.68, 52.37, 59.11, 97.63, 127.31, 128.64, 129.48, 129.85, 131.74, 144.42, 149.59, 151.11, 164.85.

5-Acetyl-4-(2,3-dichlorophenyl)-6-methyl-3,4-dihydropyrimidin-2(1H)-one (compound 14). Yield: 45%, IR (ν_max/cm⁻¹): 3230, 3108, 2961, 1636, 1520, 1236; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.09 (s, 3H, CH₃), 2.35 (s, 3H, CH₃CO), 5.70 (d, 1H, J = 3.23 Hz, CH), 7.23 (dd, 1H, J₁ = 7.83 Hz, J₂ = 1.49 Hz ArH), 7.34 (t, 1H, J = 7.83 Hz, ArH), 7.55 (dd, 1H, J₁ = 7.83 Hz, J₂ = 1.49 Hz, ArH), 7.81 (brs, 1H, NH), 9.32 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 18.96, 30.34, 52.33, 108.63, 126.94, 128.69, 129.65, 129.99, 131.96, 143.53, 149.11, 151.39, 193.82. HRMS (ESI-TOF): m/z [M⁺ + Na] calcd for C₁₃H₁₂N₂O₂Cl₂Na: 321.0176; found 321.0168.

Ethyl 4-(2,4-dichlorophenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 15). Yield: 46%, IR (ν_max/cm⁻¹): 3356, 3217, 3098, 2972, 1693, 1638, 1560, 1225, 1092; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.00 (t, 3H, J = 7.03 Hz, OCH₂CH₃), 2.29 (s, 3H, CH₃), 3.90 (q, 2H, J = 7.03 Hz, OCH₂), 5.59 (d, 1H, J = 2.83 Hz, CH), 7.32 (d, 1H, J = 8.34 Hz, ArH), 7.41 (dd, 1H, J₁ = 8.34 Hz, J₂ = 2.11 Hz ArH), 7.56 (d, 1H, J = 2.11 Hz, ArH), 7.74 (brs, 1H, NH), 9.31 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 13.92, 17.69, 51.17, 59.13, 97.47, 127.97, 128.68, 130.27, 132.54, 132.65, 140.95, 149.57, 151.13, 164.83.

Methyl 4-(2,4-dichlorophenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 16). Yield: 51%, IR (ν_max/cm⁻¹): 3360, 3219, 3096, 2944, 1695, 1641, 1227, 1099; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.29 (s, 3H, CH₃), 3.46 (s, 3H, CH₃OCO), 5.58 (d, 1H, J = 2.59 Hz, CH), 7.31 (d, 1H, J = 8.41 Hz, ArH), 7.41 (dd, 1H, J₁ = 8.42 Hz, J₂ = 1.72 Hz, ArH), 7.56 (d, 1H, J = 1.72 Hz, ArH), 7.75 (brs, 1H, NH), 9.34 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.77, 50.77, 51.10, 97.34, 127.99, 128.80, 130.18, 132.59, 132.63, 140.81, 149.70, 151.20, 165.35.

5-Acetyl-4-(2,4-dichlorophenyl)-6-methyl-3,4-dihydropyrimidin-2(1H)-one (compound 17). Yield: 50%, IR (ν_max/cm⁻¹): 3332, 3100, 2976, 1637, 1559, 1226, 1040; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.08 (s, 3H, CH₃), 2.33 (s, 3H, CH₃CO), 5.62 (d, 1H, J = 2.91 Hz, CH), 7.26 (d, 1H, J = 8.39 Hz, ArH), 7.40 (dd, 1H, J₁ = 8.38 Hz, J₂ = 1.87 Hz, ArH), 7.59 (d, 1H, J = 1.87 Hz, ArH), 7.78 (brs, 1H, NH), 9.30 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 18.96, 30.30, 51.17, 108.39, 128.02, 128.96, 129.94, 132.75, 132.83, 140.12, 149.03, 151.40, 193.83.

Ethyl 4-(2,3-difluorophenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 18). Yield: 37%, IR (ν_max/cm⁻¹): 3228, 3106, 2982, 1694, 1651, 1480, 1232, 1101; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.04 (t, 3H, J = 7.13 Hz, OCH₂CH₃), 2.28 (s, 3H, CH₃), 3.94 (dq, 2H J₁ = 7.12 Hz, J₂ = 2.38 Hz, OCH₂), 5.50 (d, 1H, J = 2.83 Hz, CH), 7.08–7.14 (m, 1H, ArH), 7.16–7.22 (m, 1H, ArH), 7.28–7.37 (m, 1H, ArH), 7.77 (brs, 1H, NH), 9.33 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 13.83, 17.72, 48.32, 59.15, 97.08, 116.10, 116.27, 123.88, 124.90, 134.44, 134.55, 149.31, 151.31, 164.85. HRMS (ESI-TOF): m/z [M⁺ + H] calcd for C₁₄H₁₅N₂O₃F₂: 297.1052; found 297.1045.

Methyl 4-(2,3-difluorophenyl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 19). Yield: 58%, IR (ν_max/cm⁻¹): 3373, 3219, 3095, 2956, 1692, 1644, 1482, 1228, 1096; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.27 (s, 3H, CH₃), 3.48 (s, 3H, CH₃OCO), 5.47 (d, 1H, J = 2.84 Hz, CH), 7.06–7.11 (m, 1H, ArH), 7.14–7.20 (m, 1H, ArH), 7.28–7.36 (m, 1H, ArH), 7.78 (brs, 1H, NH), 9.35 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.84, 48.35, 50.81, 96.92, 116.21, 116.39, 123.75, 124.93, 134.18, 134.29, 149.52, 151.41, 165.40. HRMS (ESI-TOF): m/z [M⁺ + H] calcd for C₁₃H₁₃N₂O₃F₂: 283.0896; found 283.0889.

5-Acetyl-4-(2,3-difluorophenyl)-6-methyl-3,4-dihydropyrimidin-2(1H)-one (compound 20). Yield: 52%, IR (ν_max/cm⁻¹): 3305, 3218, 3119, 1693, 1607, 1487, 1240; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.13 (s, 3H, CH₃), 2.31 (s, 3H, CH₃CO), 5.55 (d, 1H, J = 3.08 Hz, CH), 7.03–7.08 (m, 1H, ArH), 7.12–7.196 (m, 1H, ArH), 7.27–7.36 (m, 1H, ArH), 7.84 (brs, 1H, NH), 9.30 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 19.02, 30.42, 48.30, 108.14, 116.25, 116.41, 123.60, 124.95, 133.80, 133.91, 148.82, 151.53, 193.72. HRMS (ESI-TOF): m/z [M⁺ + H] calcd for C₁₃H₁₃N₂O₂F₂: 267.0947; found 267.0940.

Ethyl 4-(furan-2-yl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 21). Yield: 81%, IR (ν_max/cm⁻¹): 3351, 3233, 3115, 2976, 1694, 1644, 1230, 1098; ¹H-NMR (400 MHz, DMSO-d₆) δ: 1.14 (t, 3H, J = 7.19 Hz, OCH₂CH₃), 2.23 (s, 3H, CH₃), 4.03 (dq, 2H J₁ = 7.20 Hz, J₂ = 1.92 Hz, OCH₂), 5.20 (d, 1H, J = 3.34 Hz, CH), 6.09 (d, 1H, J = 3.14 Hz, ArH), 6.34–6.37 (m, 1H, ArH), 7.55 (brs, 1H, ArH), 7.75 (brs, 1H, NH), 9.24 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 14.15, 17.74, 47.73, 59.22, 96.75, 105.28, 110.34, 142.15, 149.36, 152.41, 155.93, 165.01.

Methyl 4-(furan-2-yl)-6-methyl-2-oxo-1,2,3,4-tetrahydropyrimidine-5-carboxylate (compound 22). Yield: 74%, IR (ν_max/cm⁻¹): 3313, 2954, 1672, 1637, 1432, 1237, 1087; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.24 (s, 3H, CH₃), 3.57 (s, 3H, CH₃OCO), 5.20 (d, 1H, J = 3.39 Hz, CH), 6.35 (d, 1H, J = 3.13 Hz, ArH), 6.34–6.36 (m, 1H, ArH), 7.56 (brs, 1H, ArH), 7.78 (brs, 1H, NH), 9.27 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 17.77, 47.63, 50.88, 96.55, 105.30, 110.37, 142.21, 149.67, 152.39, 155.83, 165.49.

5-Acetyl-4-(furan-2-yl)-6-methyl-3,4-dihydropyrimidin-2(1H)-one (compound 23). Yield: 76%, IR (ν_max/cm⁻¹): 3277, 3151, 2954, 1675, 1595, 1235; ¹H-NMR (400 MHz, DMSO-d₆) δ: 2.16 (s, 3H, CH₃), 2.24 (s, 3H, CH₃CO), 5.31 (d, 1H, J = 3.33 Hz, CH), 6.12 (d, 1H, J = 3.17 Hz, ArH), 6.34–6.37 (m, 1H, ArH), 7.56 (brs, 1H, ArH), 7.84 (brs, 1H, NH), 9.23 (brs, 1H, NH); ¹³C-NMR (100 MHz, DMSO-d₆) δ: 18.88, 29.95, 47.84, 105.61, 107.22, 110.34, 142.36, 148.80, 152.44, 155.87, 193.79.

In vitro studies

Cell culture. MCF-7, LNCaP, NHDF, T47D and Caco-2 cell lines were obtained from American Type Culture Collection (ATCC; Manassas, VA, USA) and HepaRG cell line was obtained from Life Technologies – Invitrogen™ (through Alfagene, Portugal). They were maintained in 75 cm² culture flasks at 37 °C in a humidified air incubator with 5% CO₂. The high-glucose Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS; Sigma-Aldrich, St Louis, MO, USA), and 1% antibiotic/antimycotic (10 000 units per mL penicillin G, 100 mg mL⁻¹ streptomycin and 25 μg mL⁻¹ amphotericin B) (Ab; Sigma-Aldrich, St Louis, MO, USA) was used to culture MCF-7 cells. Caco-2 cells were cultured in high glucose DMEM supplemented with 10% FBS and 1% of the antibiotic mixture of 10 000 units per mL penicillin G and 100 mg mL⁻¹ of streptomycin (sp; Sigma-Aldrich, St Louis, MO, USA). LNCaP and T47D cells were cultured in RPMI 1640 medium with 10% FBS and 1% sp. NHDF cells have grown in RPMI 1640 medium supplemented with 10% FBS, 2 mM L-glutamine, 10 mM HEPES, 1 mM sodium pyruvate and 1% Ab. Finally, HepaRG cells were seeded in Williams' E medium supplemented with 10% FBS, 1% sp, 5 μg mL⁻¹ insulin, and 5 × 10⁻⁵ M hydrocortisone hemisuccinate (Sigma-Aldrich, St Louis, MO, USA). For all cell lines, the medium was renewed every 2–3 days until cells reach approximately 90–95% confluence. Then, they were detached by gentle trypsinization (trypsin–EDTA; Sigma-Aldrich, St Louis, MO, USA) and, before the experiments, viable cells were counted by the trypan-blue exclusion assay and suitably diluted in the adequate complete culture medium.

Preparation of compounds solutions. All compounds were dissolved in DMSO in a concentration of 10 mM and stored at 4–8 °C. From this stock solution, the various working solutions of the compounds in different concentrations were prepared by adequate dilutions in the complete culture medium before each experiment. The maximum DMSO concentration in the studies was 1% and previous experiments revealed that this solvent level has no significant effects in cell proliferation (data not shown).

MTT assay. The in vitro antiproliferative effects were evaluated by the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT; Sigma-Aldrich, St Louis, MO, USA) assay. After reaching confluence, cells were trypsinized and counted using a hemocytometer and by trypan-blue exclusion of dead cells. Then, 100 μL of cell suspension per well with an initial density of 2 × 10⁴ cells per mL was seeded in 96-well culture plates and left to adhere for 48 h. After adherence, the medium was replaced by the several solutions of the compounds in study (30 μM for preliminary studies and 0.01, 0.1, 1, 10, 50 and 100 μM for concentration–response studies) in the appropriate culture medium for approximately 72 h. Untreated cells were used as the negative control. Each experiment was performed in quadruplicate and independently repeated. Then, the medium was removed, 100 μL of phosphate buffer saline (NaCl 137 mM, KCl 2.7 mM, Na₂HPO₄ 10 mM and KH₂PO₄ 1.8 mM, pH 7.4) were used to wash the cells and then 100 μL of the MTT solution (5 mg mL⁻¹), prepared in the appropriate serum-free medium, was added to each well, followed by incubation for approximately 4 h at 37 °C. Then, the MTT containing medium was removed and the formazan crystals were dissolved in DMSO. The absorbance was measured at 570 nm using a microplate reader Bio-rad Xmark spectrophotometer. After background subtraction, cell proliferation values were expressed as percentage relatively to the absorbance determined in negative control cells.

Cell viability. The analysis of cell viability was performed by flow cytometry after staining dead cells with propidium iodide (PI) (solution of PI 1 mg mL⁻¹ in 0.1% of azide and water, Sigma Aldrich, St Louis, MO, USA). Briefly, 3 mL of cells were seeded in 6-well plates (cell density of 3 × 10⁴ cells per mL for HepaRG and MCF-7 cell lines) in complete culture medium. After 48 h they were treated with 50 μM of compound 16. Untreated cells were used as negative control. At the end of 24 h of incubation, the supernatant of each well was collected, cells were harvested by trypsinization and pooled with the supernatants. The resulting cell suspension was kept on ice, pelleted by centrifugation and resuspended in 400 μL of complete medium. Afterwards, 395 μL of the cell suspension was transferred to a FACS tube and 5 μL of PI was added. A minimum of 10 000 events was acquired using a FACSCalibur flow cytometer in the channels forward scatter (FSC), side scatter (SSC) and fluorescence channel-3 (FL3, for PI). Acquisition and analysis was performed with CellQuest™ Pro Software. In the FSC/FL3 contour plot, two regions were created, one corresponding to viable cells (R1) and another to dead cells (R2) to exclude debris which were not considered in the analysis (data not shown). The percentage of viability is the percentage of cells in R1 as compared to the total number of events in R1 and R2.

Cell cycle. Cell cycle distribution of cells was determined through PI staining of DNA in fixed and permeabilized cells. In brief, 3 mL of cells were seeded in 6-well plates (cell density of 2 × 10⁴ cells per mL for HepaRG and MCF-7 cell lines) in complete culture medium. After 48 h they were treated with 50 μM of compound 16. For comparison, untreated cells were used as negative control and cells treated with 5-FU at 50 μM were used as positive control. After 48 h of incubation, the cells were trypsinized, centrifuged and resuspended in 450 μL of a cold solution of 0.5% bovine serum albumin (BSA; Amresco, USA) in PBS. The resulting cell suspension was kept on ice and then fixed by gently adding ice-cold 70% ethanol (−20 °C) with simultaneous gentle vortex agitation. After at least 2 days at −20 °C, fixed cells were washed twice with PBS and resuspended in a solution of PI (50 μg mL⁻¹) prepared in 0.5% BSA in PBS and sequentially incubated with ribonuclease A from bovine pancreas at a final concentration of 7.1 μg mL⁻¹ (solution in 50% glycerol, 10 mM Tris–HCl, pH 8, Sigma Aldrich, St Louis, MO, USA) for 15 min in the dark. The data were analyzed using ModFit software (Becton Dickinson, San Jose, CA, USA). A region (R1) was created on the FL3-width/FL3-area contour plot to exclude cell aggregates and another region (R2) was created on the FL1-height/FL3-area contour plot to exclude part of the debris.

Statistics. The data are expressed as mean ± standard deviation (SD). Comparison among groups of one factor was analyzed by using the t-student test (two groups) and one-way ANOVA (three groups) followed by Dunnett's post hoc tests to determine significant differences among the means. Difference between groups was considered statistically significant for a p-value lower than 0.05 (p < 0.05). The determination for the IC₅₀ was done by sigmoidal fitting analysis considering a confidence level of 95%.

QSAR study

Data handling. The in vitro cell antiproliferative activities, expressed as the relative cell proliferation in percentage, for the targeted compounds (1–23) at a concentration of 30 μM, against NHDF, HepaRG, Caco-2, MCF-7, T47D and LNCaP cell lines, presented in Table 2, were initially converted to its logarithm base 10. In order to increase the usefulness of the developed model, by reliably predict the log(relative cell proliferation) for the six cell lines tested, the available data for QSAR modelling was increased by combining all activity measurements using a three bit representation system to distinguish between the six cell lines (Table 3). With this, 138 cases were available for QSAR development.²⁶ Using the Kennard–Stone design,^27,28 100 cases were selected for training of the model (training group) and the remaining 38 cases were used to externally validate the QSAR model and assess its predictive performance (test group).

Table 2 Observed and QSAR predicted antiproliferative activities (as the relative cell proliferation in percentage) of the target compounds (1–23) at concentration of 30 μM, against normal human dermal fibroblasts (NHDF) cell line and against hepatic (HepaRG), colon (Caco-2), breast (MCF-7 and T47D) and prostatic (LNCaP) human cancer cell lines^a

Compound	NHDF	HepaRG	Caco-2	MCF-7	T47D	LNCaP
a Results are expressed as means ± SD (standard deviation) after 72 h of treatment, *p < 0.05 versus control; **p < 0.01 versus control; ***p < 0.001 versus control. QSAR predicted values are inside parenthesis, and cases used in the external validation of the QSAR model are in italics. The bold values correspond to the compounds that exhibited strong antiproliferative effects (relative cell proliferation lower than 50%).
1	76.90 ± 4.46* (83.41)	51.95 ± 9.16*** (60.27)	98.25 ± 6.30 (95.74)	94.57 ± 8.07 (86.57)	77.82 ± 13.79 (78.69)	59.82 ± 7.42*** (84.89)
2	92.40 ± 4.10 (86.95)	66.13 ± 7.90* (67.23)	98.83 ± 1.37 (99.66)	98.32 ± 12.02 (89.33)	82.71 ± 7.79 (81.13)	81.67 ± 6.41* (87.80)
3	84.78 ± 3.04*** (90.34)	69.12 ± 3.66** (70.38)	98.29 ± 4.04 (100.88)	84.37 ± 7.30* (93.28)	82.01 ± 4.66 (84.75)	75.09 ± 19.54** (91.62)
4	67.20 ± 3.79* (77.52)	56.56 ± 6.26*** (47.68)	89.41 ± 11.12 (87.33)	97.65 ± 4.02 (81.62)	87.34 ± 3.80 (75.45)	55.16 ± 13.43*** (78.50)
5	83.36 ± 2.40** (82.33)	86.64 ± 8.84* (59.27)	87.07 ± 3.46 (94.86)	87.31 ± 2.68 (86.25)	98.30 ± 2.33 (77.86)	83.13 ± 7.65* (85.15)
6	79.06 ± 7.97 (85.78)	61.12 ± 12.65*** (61.44)	96.41 ± 7.88 (95.74)	91.00 ± 4.37 (90.05)	88.78 ± 8.10 (81.51)	62.15 ± 5.01*** (88.20)
7	60.37 ± 3.44* (65.99)	50.32 ± 4.23*** (54.05)	89.50 ± 8.08 (79.90)	50.86 ± 3.53*** (59.25)	73.94 ± 6.16 (66.02)	50.89 ± 8.55*** (64.14)
8	76.15 ± 1.44*** (68.31)	78.33 ± 9.63** (57.94)	98.48 ± 5.74 (82.10)	50.69 ± 4.74 (61.11)	66.80 ± 6.58 (67.57)	72.27 ± 5.81** (65.47)
9	79.34 ± 6.17** (73.23)	58.51 ± 12.51*** (58.80)	91.63 ± 7.22 (86.44)	58.98 ± 7.61*** (66.71)	70.12 ± 5.37 (70.91)	70.56 ± 12.80*** (68.11)
10	67.04 ± 2.07 (79.71)	55.94 ± 11.67*** (53.83)	88.58 ± 9.33 (88.85)	79.71 ± 5.26** (74.60)	89.83 ± 2.39 (76.79)	57.36 ± 8.13*** (70.48)
11	86.28 ± 5.06 (82.65)	72.01 ± 6.69* (59.07)	89.70 ± 4.28 (92.70)	97.71 ± 34.16 (77.80)	91.31 ± 11.08 (78.60)	79.77 ± 12.58* (73.69)
12	86.94 ± 5.27 (86.52)	72.30 ± 9.95*** (60.46)	85.89 ± 6.87 (95.90)	69.12 ± 14.52 (84.08)	86.32 ± 4.62 (81.92)	85.63 ± 9.60 (78.85)
13	52.91 ± 9.81*** (45.16)	15.19 ± 1.29*** (18.87)	51.74 ± 10.37*** (45.44)	50.94 ± 5.13*** (52.29)	53.45 ± 3.91*** (57.72)	53.74 ± 5.93*** (47.71)
14	60.63 ± 3.04*** (58.26)	36.36 ± 11.00*** (30.15)	54.85 ± 2.73*** (62.72)	53.59 ± 5.73*** (66.97)	50.33 ± 1.55*** (64.73)	64.09 ± 4.41* (65.09)
15	53.26 ± 2.46*** (47.05)	15.50 ± 0.45*** (19.83)	51.43 ± 1.55*** (47.96)	66.94 ± 5.25*** (54.21)	50.22 ± 4.67*** (58.69)	53.09 ± 1.66*** (49.87)
16	54.35 ± 2.38*** (59.90)	36.81 ± 1.81*** (33.37)	63.65 ± 5.73*** (68.39)	65.98 ± 30.33 (67.71)	50.36 ± 4.31* (64.25)	74.95 ± 11.43** (68.96)
17	62.65 ± 1.75 (55.64)	29.62 ± 2.57*** (27.86)	55.57 ± 6.27*** (58.89)	64.17 ± 5.32*** (64.35)	59.11 ± 2.80*** (63.49)	57.68 ± 13.42* (61.82)
18	70.65 ± 3.25*** (66.71)	53.67 ± 8.34*** (57.99)	86.36 ± 6.24*** (87.19)	67.33 ± 17.12** (64.63)	66.86 ± 8.55*** (64.85)	89.78 ± 13.28 (71.98)
19	59.63 ± 3.88*** (69.14)	54.97 ± 12.32*** (60.47)	76.10 ± 2.88*** (86.10)	61.16 ± 19.92 (63.54)	60.17 ± 4.16* (67.21)	60.06 ± 12.29*** (68.49)
20	83.25 ± 4.47*** (72.05)	81.42 ± 7.92 (63.94)	97.56 ± 8.88 (94.08)	82.05 ± 15.06* (73.52)	79.70 ± 10.50** (67.98)	78.17 ± 7.04 (79.96)
21	86.99 ± 16.03* (88.48)	62.87 ± 14.47** (64.97)	83.39 ± 6.73 (97.83)	104.14 ± 4.80 (84.39)	84.25 ± 1.58** (83.38)	85.90 ± 11.42 (79.14)
22	93.72 ± 8.80 (90.50)	52.96 ± 8.18*** (65.59)	96.64 ± 7.31 (98.08)	84.79 ± 14.42 (84.98)	78.73 ± 9.47 (85.74)	74.62 ± 7.19** (78.74)
23	88.17 ± 4.85*** (93.59)	67.80 ± 6.71*** (70.40)	83.17 ± 1.85 (102.05)	92.30 ± 18.21 (92.12)	83.69 ± 7.10 (88.13)	92.20 ± 14.90 (87.12)

---

Table 3 Three bit representation for the NHDF, HepaRG, Caco-2, MCF-7, T47D and LNCaP cell lines used in the QSAR study

Cell lines	Binary variables
	Bit 1	Bit 2	Bit 3
NHDF	1	0	0
HepaRG	0	1	0
Caco-2	0	0	1
MCF-7	1	1	0
T47D	1	0	1
LNCaP	0	1	1

---

In silico calculation of the molecular descriptors. For the in silico calculation of the molecular descriptors, the titled compounds were first manually drawn in ACD/ChemSketch 2015 Pack 2 (ref. 29) and the SMILES notation was obtained for each molecule. Using this notation, tautomer and ionization states were checked and the molecular descriptor log P was calculated (using GALAS algorithm) in ACD/Percepta 2015.³⁰ The remaining molecular descriptors were calculated using E-Dragon online,³¹ and correspond to 10 constitutional descriptors and 5 molecular properties. No molecular descriptor dependent on the 3D conformation of the molecules was calculated, since the antitumoral bioactive conformations of this family of compounds is not known and due to the fact that the simpler selected descriptors are more easily interpreted.^32,33 Additionally, molecular descriptors that were repetitive or insufficiently discriminatory (minimum standard deviation value of 0.05) or highly correlated with other descriptors (maximum correlation coefficient value of 0.9) for QSAR analysis were removed from the training group.

BRANN modelling. QSAR modelling was performed using an in-house developed tool based on Bayesian regularized artificial neural networks (BRANNs) in MATLAB R2014a,³⁴ which allowed process automation, data analysis and the use of a cross-validation procedure. All calculations and modelling were performed on a 3.5 GHz Intel i7 CPU running Windows 7 operating system. Commonly selected parameters for the models were “trainbr” as the training function, pre-processing of input and output variables to the range [−1, 1] “tansig” (sigmoid) transfer function in the neurons of the hidden layer and “purelin” (linear) in the neurons of the output layer. The remaining parameters were kept at their default value. The calculated molecular descriptors and the three bits used for cell line identification were used as independent variables (inputs), and the log(relative cell proliferation) was used as the dependent variable (output).

BRANN models were used for two objectives: the optimization of the model parameters and final predictive model production. For the selection of the most relevant molecular descriptors, a forward selection method was performed using a 10-fold cross-validation procedure with the BRANN models, starting with the three bit inputs. Each iteration was repeated 10 times with random splits of the available data, and the average Q² and RMSE_CV statistics were obtained. At the end of each iteration, the molecular descriptor that produced the best average Q² and RMSE_CV statistics was kept, and the process was repeated until all molecular descriptors have been added to the system. After selection of the most relevant molecular descriptors (the ones that returned the best average cross-validated statistics), the same 10-fold cross-validation procedure with 10 duplicates was used to determine the optimal number of neurons in a single hidden layer between 0 (linear model) to 10 (non-linear models). After optimization of the model parameters regarding the molecular descriptors and model complexity (internal validation), the final QSAR model was finally trained on all available training data.

For the validation of the final QSAR model, the hold-out test group was used by comparison of the QSAR predicted values with the observed experimental values of the cases not used to train the model (external validation). Furthermore, a y-scrambling procedure was also performed to verify the absence of chance correlations between the input and output variables.³⁵ Finally, the applicability domain of the QSAR model was also defined.

Statistical evaluation. For the internal and external validation of the developed QSAR model, the coefficient of determination (R², Q² in cross-validation or R_pred² for the external group) was used as a measurement of the goodness of fit of the model, and the root mean squared error (RMSE) between predicted and observed values as a measurement of accuracy. Also, several statistical requirements for the test group results were used to assure the predictive ability of the QSAR model. These requirements are: R_pred² > 0.6, predicted versus observed activities and observed versus predicted activities for regressions through the origin (R₀² and R′₀², respectively) close to R_pred², slopes k and k′ of regression lines through the origin close to 1 and the difference between R₀² and R′₀² not higher than 0.3.³⁶ Since it has been recently suggested that R² based metrics may be misleading in the external validation of QSAR models, the MAE (mean absolute error) based criteria were also used. According to the MAE based criteria, a QSAR model has good external predictivity when the external group MAE and MAE + 3σ (σ is the standard deviation of the absolute errors for the test group) are within 10 and 20% of the training set response range and a moderate external predictivity when they are within 15 and 25% of the same range, respectively.³⁷

Model interpretation. In order to elucidate the interactions between the selected molecular descriptors and relative cell proliferation in the studied cell lines, the Lek profile method was used.^38,39 This was done for each cell line separately, varying each molecular descriptor across 11 data points (10 equal intervals over the entire range of the molecular descriptor) and holding the remaining molecular descriptors at 5 different range splits value (minimum, percentile 25%, median, percentile 75% and maximum value). The average predicted responses obtained for each data split (fixed variables) across the 11 data intervals (studied variable) are taken as the relationship trend between the molecular descriptor and the response variable. The relative importance of each molecular descriptor was also obtained, being it equal to the maximum difference between the calculated predictions (maximum predicted value − minimum predicted value).⁴⁰

Results and discussion

Chemical synthesis of DHPMs through Biginelli reaction

Twenty-three DHPMs (Table 1) were synthesized by the Biginelli reaction, which consists in a one-pot cyclocondensation reaction among an aldehyde (benzaldehyde, p-tolualdehyde, 4-nitrobenzaldehyde, anisaldehyde, 2,3-dichlorobenzaldehyde, 2,4-dichlorobenzaldehyde, 2,3-difluorobenzaldehyde, furaldehyde), a β-ketoester (ethyl acetoacetate and methyl acetoacetate)/acetylacetone and urea (Fig. 1). The reaction was carried out under solvent-free conditions and the catalyst used to promote the reaction was bismuth(III) nitrate pentahydrate (Fig. 1), similarly to the reported by Khodaei and collaborators.²⁵ The reaction was considered concluded when a complete solidification occurred and it was observed that the reactions occurred fastly (5–30 min) and half of them were considered completed in not more than 7 min. In fact, this reaction usually is very rapid, namely when it occurs in solvent-free conditions, and similar results can be found in the literature.^25,41 The work up was very simple, washing the product with water to remove the catalyst. Then, the dry compounds were recrystallized with ethanol 99.9% to afford the pure products that were subsequently characterized by IR, ¹H- and ¹³C-NMR. Overall the yields of the reactions were moderate to high and the lowest yields were obtained with the compounds containing halogen atoms in its structure (32–58%), which could be related with their high lipophilicity (Table 1). Our results were also in agreement with the work of other authors regarding the good yields obtained for the majority of compounds already described in the literature.^42,43 Interestingly, this synthetic procedure was also extended to other reagents, being the respective products, which are new to the best of our knowledge, successfully synthesized.

Fig. 1 General scheme of one-pot synthesis of 3,4-dihydropyrimidin-2-(1H)-ones catalyzed by Bi(NO₃)₃·5H₂O under solvent-free conditions.

Biological evaluation

Chlorinated Biginelli compounds exhibit relevant antiproliferative effects against the HepaRG cell line. The well-established MTT assay was used to evaluate the antiproliferative effects of the synthesized compounds. In this assay, the amount of the initial soluble yellow MTT converted to the insoluble purple formazan by active mitochondrial lactate dehydrogenase of living cells⁴⁴ is considered equivalent to the number of viable cells after a period of cell growth. The DHPMs as Biginelli products (1–23) were evaluated against five cancer cell lines: HepaRG, Caco-2, MCF-7, T47D and LNCaP. The NHDF cells were included in the study to evaluate the selectivity of these compounds for cancerous versus non-cancerous cells. The commercial anticancer drug 5-FU was also tested for comparison. A single concentration of 30 μM (Table 2) was used to perform the initial screening of the antiproliferative properties and the results correspond to the relative cell proliferation, in percentage, after 72 h of incubation with the compounds of interest. Generally, it was found that the compounds did not exhibit marked cytotoxicity in the normal dermal cells (relative cell proliferation higher than 50% at 30 μM). Additionally, this first screening also showed that they did not have relevant antiproliferative effects regarding the intestinal, prostatic and both breast cancer cell lines. However, the chlorinated compounds appeared to have significant selective cytotoxicity in the hepatic cancer cell line (compounds 13–17 in Table 2). Moreover, to the best of our knowledge, it was the first time that this new hepatic cancer cell line was used in the biological evaluation of Biginelli products. Considering the structure–antiproliferative activity relationship, our results showed that the molecules containing no substituents in the phenyl ring (1, 2 and 3) and a methyl moiety (4, 5 and 6) or a methoxy group (10, 11 and 12) at para-position did not show significant toxicity in all the cell lines assessed. Taking into account that the analogue of compound 9, with the nitro group in C2 exhibited high antiproliferative activity against MCF-7 cells,⁴⁵ and the analogue with a nitro group at C3 was described as displaying inhibition of the cell proliferation higher than the observed with monastrol, a known anticancer 3,4-dihydropyrimidin-2-(1H)-thione, against the rat C6 glioma cell line and the human U138-MG glioma cell line,⁴⁶ it was expected that our compounds with a nitro portion (7, 8 and 9) could display significant antiproliferative activity, but it was not verified. The same was observed when a 5-membered heteroaromatic ring replaced the phenyl ring (21, 22 and 23). On the other hand, the molecules containing chloro atoms in their structure in C2 and C3 positions (13 and 14) and C2 and C4 positions (15, 16 and 17) of the aromatic ring, showed strong antiproliferative effects against the hepatic cells. This suggests that the presence of chloro atoms at these positions play an important role on the effects of these compounds in the in vitro growth of this cell line. Interestingly, the new compounds with a fluoro atom at positions C2 and C3 of the aromatic portion (18, 19 and 20) did not show the cytotoxic properties exhibited by their chloro-analogues. Sachdeva and Dwivedi had already tested DHPMs containing halogens attached to the phenyl ring against MCF-7 cells, showing poor activity.⁴⁷ Actually, although the compounds are different, our results seem to corroborate these findings regarding the halogens-containing DHPMs against breast cancer cell lines.

The in vitro cytotoxic potential of the most toxic compounds 13, 14, 15, 16 and 17 was further investigated by determining the corresponding concentration inducing 50% inhibition of cell growth (IC₅₀) against HepaRG cell line, and the results are summarized in Table 4. From the analysis of the data obtained it can be observed that no marked differences regarding the position of the chloro atoms. However, they suggest that small lateral chains at the position 5 of the pyrimidinone ring (IC₅₀ = 5.47 μM for 14 and IC₅₀ = 5.28 μM for 17) afford higher toxicity than longer lateral chains (IC₅₀ = 9.76 μM for 13, IC₅₀ = 13.33 μM for 15 and IC₅₀ = 15.96 μM for 16). Although the IC₅₀ value for 5-FU (IC₅₀ = 2.02 μM) was lower than the IC₅₀ values for the evaluated compounds, it is noteworthy the selectivity of these Biginelli products against HepaRG cells. On the other hand, the commercial anticancer drug used as control was toxic against all cell lines evaluated (IC₅₀ = 1.71 μM for MCF-7, IC₅₀ = 0.54 μM for T47D, IC₅₀ = 7.80 μM for LNCaP, IC₅₀ = 3.61 μM for NHDF, IC₅₀ = 1.15 μM for Caco-2).

Table 4 Cytotoxicity (IC₅₀ μM) of the most potent compounds against HepaRG cell line^a

Compounds	IC50 (μM)	R^2
a The cells were treated with a variety of concentrations (0.01, 0.1, 1, 10, 50 and 100 μM) during 72 h. The antiproliferative effects were determined by the MTT assay. The data shown are representative of at least two independent experiments.
13	9.76	0.99
14	5.47	0.97
15	13.33	0.99
16	15.96	0.99
17	5.28	0.99
5-FU	2.02	0.93

---

Compound 16 causes a cell cycle arrest of cancer cells at G₀/G₁. To better understand the possible mechanism of cytoxicity of the chlorinated compounds both cell survival and cell cycle distribution were quantified using flow cytometry. As shown in Fig. 2, compound 16 did not cause cell death after 24 h of incubation at the concentration of 50 μM in either HepaRG or MCF-7 cells. In contrast, as shown in Fig. 3, comparing with the control (untreated cells), the treatment of HepaRG cells with compound 16 for 48 h at 50 μM significantly arrested cells in G₀/G₁ phase (either quiescence or growth phase before DNA replication). The same was verified for MCF-7 cells (Fig. 4). This phenomenon was accompanied by a decrease in the number of cells in S phase, (where DNA replication occurs). The effect of 5-FU, an antimetabolite drug known to interfere with DNA replication, was also evaluated for comparison. To the best of our knowledge, it is the first time that it is reported the effect of 5-FU in HepaRG cells. Consistently with its described mechanism of action as antimetabolite, this commercial anticancer drug increased extensively the proportion of cells in S phase, which was accompanied by a significant decrease of the frequency of cells in G₀/G₁ phase, while the G₂/M peak was almost absent (Fig. 3 & panel (B)), although this effect was less marked in MCF-7 cells (Fig. 4 & panel (B)). This can be explained by the arrest of cells in S phase due to the inability of cells to complete DNA replication. In addition, in spite of the fact that compound 16 is chemically related with monastrol, it does not seem to share the same mechanism of action. In fact, monastrol is an inhibitor of kinesin-like protein KIF11 (or EG5) which is essential to mitosis (M phase), where it is responsible to chromosome positioning, centrosome separation and establishing a bipolar spindle during cell mitosis.⁴⁸ Consistently with this mechanism of action, in the hepatic cell line HepG2 monastrol causes the reduction of the G₀/G₁ peak and a clear G₂/M arrest,⁴⁸ a very different effect from the one obtained with compound 16. Additionally, although LaSOM 63, a monastrol derived molecule, reduced glioma cell viability and cell growth, it did not exhibit glioma cell-cycle arrest in G₂/M phase.⁴⁹ The observed cell cycle arrest in G₀/G₁ phase of cell cycle can be related with the interference of one of many proteins participating in the highly regulated cellular mechanisms to standby or initiate DNA replication. Further studies will be necessary to elucidate what is the signalling pathway being affected and to ascertain whether other mechanisms are involved in the cytotoxicity of this compound.

Fig. 2 Percentage of viable cells after 24 h treatment with 50 μM of compound 16 in HepaRG and MCF-7 cell lines through PI flow cytometric assay. The control corresponds to untreated cells. The percentage of survival is the percentage of cells in R1 (live cells) as compared to the total number of events in R1 and R2 (dead cells). Each bar represents the mean ± SD. **p < 0.01 versus control.

Fig. 3 Cell cycle distribution analysis of HepaRG cancer cells after treatment with compound 16 (50 μM) for 48 h. A negative control (untreated cells) and a positive control (5-FU, 50 μM) were included. The analysis of the cell cycle distribution was performed using the PI staining and flow cytometry. (A) Representative cell cycle distribution analysis showing in (a), (d) and (g), gating of singlets by region R1 created on the FL3-width/FL3-area contour plot; in (b), (e) and (h), debris exclusion by region (R2) created on the FL1-height/FL3-area contour plot; and in (c), (f) and (i), cell cycle distribution fit, respectively for negative control, 5-FU and compound 16. (B) Quantification of the proportion of cells in G₀/G₁, S and G₂/M phases of the cell cycle. Each bar represents the mean ± SD of four samples (originating from two independent experiments). **p < 0.01 versus control; ***p < 0.001 versus control.

Fig. 4 Cell cycle distribution analysis of MCF-7 cancer cells after treatment with compound 16 (50 μM) for 48 h. A negative control (untreated cells) and a positive control (5-FU, 50 μM) were included. The analysis of the cell cycle distribution was performed using the PI staining and flow cytometry. (A) Representative cell cycle distribution analysis showing in (a), (d) and (g), gating of singlets by region R1 created on the FL3-width/FL3-area contour plot; in (b), (e) and (h), debris exclusion by region (R2) created on the FL1-height/FL3-area contour plot; and in (c), (f) and (i), cell cycle distribution fit, respectively for negative control, 5-FU and compound 16. (B) Quantification of the proportion of cells in G₀/G₁, S and G₂/M phases of the cell cycle. Each bar represents the mean ± SD of four samples (originating from two independent experiments). **p < 0.01 versus control; ***p < 0.001 versus control.

QSAR study

Model development. Bayesian regularized artificial neural networks (BRANNs) were used in this QSAR study to relate in silico calculated molecular descriptors of the targeted compounds with their experimental in vitro antiproliferative activity in the used cell lines. After development, the QSAR model can be used to predict the antiproliferative activity of hypothetical related compounds, given that the required molecular descriptors are calculated. In this study, BRANN models were coupled with an optimization process for the selection of the most relevant molecular descriptors and ascertainment of the required model complexity.

Regarding the three bit system used in this study (for distinction purposes between cell lines), presented in Table 3, it allowed the use of a larger data set for training, by incorporating all available data in Table 2 in a single output QSAR model, which can generate a better predictive model.³⁶ The resulting QSAR model is of greater utility since it can reliably predict the antiproliferative activity of compounds in a variety of cell lines.

Given the importance of an external test set to ultimately validate a QSAR model,⁵⁰ the available data was split as described in the Experimental section in a training and test group. This was accomplished by the use of the Kennard–Stone design,²⁸ which ensured that the training group was uniformly distributed across the available data space. The resultant test group cases were well distributed both across cell lines and experimental value ranges.

With the initial removal of collinear and insufficiently discriminative molecular descriptors, as described in the Experimental section, 7 out of the initial 16 molecular descriptors were removed, resulting in 9 descriptors available for further optimization. Further removal of molecular descriptors using a forward selection method, by maximizing the 10-fold cross-validated Q², returned three molecular descriptors as the most relevant, since they generated the best average values of Q² and RMSE_CV: log P (GALAS predicted octanol : water partition coefficient), AMR (Ghose–Crippen molar refractivity) and Ss (sum of Kier–Hall electrotopological states). The calculated molecular descriptor values for the targeted compounds are given in Table 5. The Pearson linear correlation matrix for the molecular descriptors and for the tested cell lines is presented in Table 6. It can be observed that the chosen descriptors are poorly correlated between them, which is a requirement for a valid QSAR model.⁵¹ Moreover, to further confirm the orthogonality of the selected molecular descriptors, their variance inflation factor values were calculated.⁵² The obtained values of 2.6, 1.6 and 3.4 for log P, Ss and AMR, respectively, confirm that no relevant multicollinearity exists between the selected molecular descriptors considering a typical limit value of 5.⁵³

Table 5 Values for the in silico calculated molecular descriptors of the titled compounds (1–23) used in the QSAR study

Compound	log P^a	AMR^b	Ss^c
a log P, Galas predicted octanol : water partition coefficient;b AMR, Ghose–Crippen molar refractivity;c Ss, sum of Kier–Hall electrotopological states.
1	1.79	71.43	47.67
2	1.48	66.69	46.17
3	1.46	65.47	42.67
4	2.13	76.47	49.33
5	1.88	71.73	47.83
6	1.84	70.51	44.33
7	1.74	78.76	63.33
8	1.41	74.01	61.83
9	1.34	72.79	58.33
10	1.57	77.90	52.83
11	1.34	73.15	51.33
12	1.38	71.93	47.83
13	3.31	81.04	55.22
14	3.03	75.08	50.22
15	3.25	81.04	55.22
16	2.94	76.29	53.72
17	3.12	75.08	50.22
18	1.94	71.87	63.00
19	1.27	67.12	61.50
20	1.71	65.90	58.00
21	0.87	63.82	47.17
22	0.52	59.08	45.67
23	0.69	57.86	42.17

---

Table 6 Pearson linear correlation matrix for the in silico calculated molecular descriptors and the relative cell proliferation for the cell lines NHDF, HepaRG, Caco-2, MCF-7, T47D and LNCaP

Correlation	log P^a	AMR^b	Ss^c
a log P: Galas predicted octanol : water partition coefficient;b AMR: Ghose–Crippen molar refractivity;c Ss: sum of Kier–Hall electrotopological states.
log P	1.000
AMR	0.764	1.000
Ss	0.212	0.476	1.000
NHDF	−0.797	−0.769	−0.534
HepaRG	−0.740	−0.540	−0.170
Caco-2	−0.789	−0.501	−0.145
MCF-7	−0.471	−0.533	−0.694
T47D	−0.669	−0.414	−0.475
LNCaP	−0.540	−0.657	−0.278

---

Finally, regarding the tested model complexities (0 to 10 neurons in one hidden layer), three neurons returned the best average Q² and RMSE_CV values. Thus, the final model 6–3–1, trained on all training cases, contains six inputs (three bits for cell line identification and the three selected molecular descriptors), three neurons in one hidden layer and one output neuron which returns the logarithm of the relative cell proliferation.

Model validation. The internal (both training and cross-validation) and external validation statistics (coefficient of determination and RMSE) of the developed QSAR model, can be seen in Table 7. The external validation statistics were obtained by comparing the predicted output of the cases left out of the training process, but for which the experimental result is known.

Table 7 Statistical evaluation of the developed QSAR model for the cross-validation, training and test data

Group	Parameter	Value
Train	N	100
	Q^2 (10-fold cross-validation)	0.663
	RMSECV (10-fold cross-validation)	0.071
	R^2 (non-cross-validated)	0.839
	RMSE (non-cross-validated)	0.049
Test	N	38
	Rpred^2	0.740
	RMSE	0.077

---

Analyzing the obtained cross-validated statistics Q² of 0.663 and RMSE_CV of 0.071, and external test set statistics R_pred² of 0.740 and RMSE of 0.077, it can be concluded that the model presents great predictive ability, generating reliable predictions. This can be confirmed in Fig. 5, in which the predicted log(relative cell proliferation) of both training and test group cases is similar to the respective experimental results. In order to verify the normality of the training and test group obtained residuals, the statistical test of Lilliefors was applied.⁵⁴ For both groups, with a significance level of α = 0.05, the null hypothesis of normality was not rejected, which indicated the absence of a systematic contribution to the error.⁵¹

Fig. 5 Plot of the experimental and predicted log(relative cell proliferation) for the developed 6–3–1 BRANN QSAR model. Solid line represents the line of unity, grey marks indicate cases used for training and open circles represent cases used for external testing.

For the y-scrambling procedure, the highest value of Q² achieved in 10 trials was 0.024, which when compared to the Q² value of 0.663 of the final model confirms the absence of chance correlations between the molecular descriptors and the response. Also, all additional statistics proposed in the work of Tropsha for a QSAR model to be predictive are well between the limit values (difference between the R_pred² value and R₀² or R′₀² not higher than 10%, k and k′ values between 0.85 and 1.15 and the absolute difference between R₀² and R′₀² not higher than 0.3), which reinforces the predictive ability of the developed model.³⁶ Finally, regarding the MAE based criteria proposed in the work of Roy, the developed QSAR model is classified as having a moderate predictive capability, when using all absolute error values of the test group (test group MAE and MAE + 3σ within 15 and 25% of the training set response range, respectively).³⁷ This is somewhat concordant with the other statistical evaluation metrics.

In order to enable that the local QSAR model developed can be effectively used to guide future efforts in improving activity of new hypothetical related compounds, its applicability domain has to be defined. The applicability domain of a QSAR model can be defined as the chemical space defined by the molecular descriptors used for training in which the model can produce reliable predictions, and thus it is valid.⁵⁵ The applicability domain of the non-linear model was defined using the Euclidean distances between testing cases and the mean centroid of the model defined by the training data (vector of the average values of the used molecular descriptors). The maximal Euclidean distance found between the training set cases and the mean centroid was defined as the threshold to consider the prediction of an external case reliable.^56,57 All cases belonging to the external validation set had a Euclidean distance to the mean centroid within this threshold, and thus were reliable predictions.

Model interpretation. Despite the “black-box” nature of artificial neural networks, due to the difficult interpretation of the used inputs,⁵⁸ several approaches have been developed to overcome this limitation.⁵⁹ In this work, we employed the Lek profile method to study the trends between the inputs (molecular descriptors) and the output (relative cell proliferation) as explained in the Experimental section. Fig. 6 shows the obtained trends for each cell line between the molecular descriptors log P, Ss and AMR and the log(relative cell proliferation), after application of the QSAR model on the created artificial dataset. Although the trends obtained differ between the cell lines, as expected, there are some similarities between them. As shown in Fig. 6, log P appears to have a parabolic relationship with the output, in which higher values of log P generate lower values of relative cell proliferation. This observation is in accordance with the experimental findings, in which the more lipophilic compounds (compounds 13–17, containing chloro atoms in their structure) are the most bioactive, particularly in the HepaRG cell line. In some cell lines (Caco-2, LNCaP and MCF-7), lower values of log P also produce lower values of relative cell proliferation, and thus, higher cytotoxicity. For the molecular descriptor Ss, in general higher values generate lower values of relative cell proliferation. As an exception, for the cell lines Caco-2 and HepaRG, the relationship is positive (higher values generate higher values of relative cell proliferation). Finally, higher values of AMR tend to produce lower values of relative cell proliferation, with T47D cell line being the exception. Again, this observation is in accordance with the experimental findings, in which the more bulky chloro containing compounds 13–17 are the most bioactive. This is particularly evident in the HepaRG cell line. Thus, in general, this QSAR study suggests that more lipophilic and bulky related molecules, with higher values for Ss, can generate compounds with improved cytotoxicity. This can be partially confirmed in Table 6 by the obtained negative values for the Pearson's linear correlation coefficients between the molecular descriptors and the relative cell proliferation in each cell line (negative relationships). Although log P appears to be negatively correlated with the response variable, as it can be seen in Fig. 6, the trend for all cell lines is not linear, and thus a direct conclusion cannot be drawn from the correlation coefficient directly without further investigation (as done in the Lek profile method).

Fig. 6 Contribution profile of the in silico calculated molecular descriptors log P (Galas predicted octanol : water partition coefficient), Ss (Ghose–Crippen molar refractivity) and AMR (sum of Kier–Hall electrotopological states) to the prediction of the log(relative cell proliferation) by the BRANN QSAR model for the (A) Caco-2, (B) HepaRG, (C) LNCaP, (D) MCF-7, (E) NHDF and (F) T47D cell lines. Each data point is obtained as the average predicted output when each variable is varied across its minimum and maximum value and the remaining variables are fixed at their minimum, first quartile, median, third quartile and maximum value.

In general, log P appears to be the most relevant molecular descriptor, followed by AMR, and being Ss the least important. Nevertheless, high variability exists in the relative relevance of each molecular descriptor in each cell line.

By analyzing the trends obtained in each cell line, some degree of selectivity of the titled compounds 1–23 for hepatic tissue may be argued, since the obtained log(relative cell proliferation) values for the HepaRG cell line are generally lower using the artificial dataset, in comparison with the values obtained for the other cell lines. This is particularly apparent with the experimental results obtained for the chloro containing compounds 13–17.

Conclusions

In conclusion, a series of DHPMs, including several new compounds, was successfully synthesized via the solvent-free Biginelli reaction by condensation of an aldehyde, a β-ketoester/acetylacetone and urea, catalyzed by Bi(NO₃)₃·5H₂O. After performing their in vitro cytotoxicity evaluation, it was found that the molecules containing chloro atoms in their structure demonstrated selective cytotoxicity for hepatic cancer cells. Our data also show that compound 16 led to a G₀/G₁ cell cycle arrest as evidenced by flow cytometric analysis. Moreover, QSAR analysis revealed that, in general, an increase in lipophilicity, bulkiness and the value of the molecular descriptor Ss could lead to increased activity. Nonetheless, although the developed QSAR model did not help in the selection or prioritization of the synthesized compounds, albeit being able to reliably predict the relative cell proliferation in multiple cell lines (NHDF, HepaRG, Caco-2, MCF-7, T47D and LNCaP), requiring only three in silico calculated molecular descriptors (log P, AMR and Ss), it can be used to guide future efforts in improving activity and selectivity by making predictions on hypothetical related compounds inside its applicability domain. Additional studies will be required to better evaluate the mechanisms of action which are involved in the toxicity of these promising molecules as well as structural modifications which could improve the potency without affecting the selectivity.

Acknowledgements

The authors are grateful to Fundação para a Ciência e a Tecnologia (Lisbon, Portugal) for the PhD fellowships of Mariana Matias (SFHR/BD/85279/2012) and Gonçalo Campos (SFHR/BD/95505/2013), involving the POPH–QREN, which is co-funded by FSE and MEC. The authors also acknowledge the support provided by FEDER funds through the POCI – COMPETE 2020 – Operational Programme Competitiveness and Internationalisation in Axis I – Strengthening research, technological development and innovation (Project No. 007491) and National Funds by FCT – Foundation for Science and Technology (Project UID/Multi/00709).

References

1. B. Ganem, Acc. Chem. Res., 2009, 42, 463–472.
2. A. Domling, Chem. Rev., 2006, 106, 17–89.
3. C. Hulme and V. Gore, Curr. Med. Chem., 2003, 10, 51–80.
4. J. E. Biggs-Houck, A. Younai and J. T. Shaw, Curr. Opin. Chem. Biol., 2010, 14, 371–382.
5. C. O. Kappe, Eur. J. Med. Chem., 2000, 35, 1043–1052.
6. H. Kumar and A. Parmar, Ultrason. Sonochem., 2008, 15, 129–132.
7. F. Dong, L. Jun, Z. Xinli, Y. Zhiwen and L. Zuliang, J. Mol. Catal. A: Chem., 2007, 274, 208–211.
8. M. A. Kolosov, V. D. Orlov, D. A. Beloborodov and V. V. Dotsenko, Mol. Diversity, 2009, 13, 5–25.
9. B. R. P. Kumar, G. Sankar, R. B. N. Baig and S. Chandrashekaran, Eur. J. Med. Chem., 2009, 44, 4192–4198.
10. A. N. Chiang, J. C. Valderramos, R. Balachandran, R. J. Chovatiya, B. P. Mead, C. Schneider, S. L. Bell, M. G. Klein, D. M. Huryn, X. S. Chen, B. W. Day, D. A. Fidock, P. Wipf and J. L. Brodsky, Bioorg. Med. Chem., 2009, 17, 1527–1533.
11. D. R. Duguay, M. T. Zamora, J. M. Blacquiere, F. E. Appoh, C. M. Vogels, S. L. Wheaton, F. J. Baerlocher, A. Decken and S. A. Westcott, Cent. Eur. J. Chem., 2008, 6, 562–568.
12. D. R. Godhani, P. B. Dobariya, A. A. Jogel, A. M. Sanghani and J. P. Mehta, Med. Chem. Res., 2014, 23, 2417–2425.
13. T. Gireesh, R. R. Kamble, P. P. Kattimani, A. Dorababu, M. Manikantha and J. H. Hoskeri, Arch. Pharm., 2013, 346, 645–653.
14. N. Gangwar and V. K. Kasana, Med. Chem. Res., 2012, 21, 4506–4511.
15. P. Lacotte, D. A. Buisson and Y. Ambroise, Eur. J. Med. Chem., 2013, 62, 722–727.
16. K. Singh, K. Singh, D. M. Trappanese and R. S. Moreland, Eur. J. Med. Chem., 2012, 54, 397–402.
17. J. Kim, T. Ok, C. Park, W. So, M. Jo, Y. Kim, M. Seo, D. Lee, S. Jo, Y. Ko, I. Choi, Y. Park, J. Yoon, M. K. Ju, J. Ahn, J. Kim, S. J. Han, T. H. Kim, J. Cechetto, J. Namc, M. Liuzzi, P. Sommer and Z. No, Bioorg. Med. Chem. Lett., 2012, 22, 2522–2526.
18. J. M. Goss and S. E. Schaus, J. Org. Chem., 2008, 73, 7651–7656.
19. D. L. da Silva, F. S. Reis, D. R. Muniz, A. L. Ruiz, J. E. de Carvalho, A. A. Sabino, L. V. Modolo and A. de Fatima, Bioorg. Med. Chem., 2012, 20, 2645–2650.
20. J. J. Bariwal, M. Malhotra, J. Molnar, K. S. Jain, A. K. Shah and J. B. Bariwal, Med. Chem. Res., 2012, 21, 4002–4009.
21. J. G. Sosnicki, L. Struk, M. Kurzawski, M. Peruzynska, G. Maciejewska and M. Drozdzik, Org. Biomol. Chem., 2014, 12, 3427–3440.
22. J. C. Dearden, Int. J. Quant. Struct. Prop. Relat., 2016, 1, 1–44.
23. K. Roy, S. Kar and R. N. Das, Understanding the basics of QSAR for applications in pharmaceutical sciences and risk assessment, Academic press, 2015.
24. F. R. Burden and D. A. Winkler, J. Med. Chem., 1999, 42, 3183–3187.
25. M. M. Khodaei, A. R. Khosropour and M. Beygzadeh, Synth. Commun., 2004, 34, 1551–1557.
26. P. Paixao, N. Aniceto, L. F. Gouveia and J. A. Morais, Pharm. Res., 2014, 31, 3313–3322.
27. R. W. Kennard and L. A. Stone, Technometrics, 1969, 11, 137–148.
28. M. Daszykowski, B. Walczak and D. Massart, Anal. Chim. Acta, 2002, 468, 91–103.
29. ACD/ChemSketch, 2015 release, Advanced Chemistry Development, Inc., Toronto, ON, Canada, 2015, http://www.acdlabs.com.
30. ACD/Percepta, 2015 release, Advanced Chemistry Development, Inc., Toronto, ON, Canada, 2015, http://www.acdlabs.com.
31. I. V. Tetko, J. Gasteiger, R. Todeschini, A. Mauri, D. Livingstone, P. Ertl, V. A. Palyulin, E. V. Radchenko, N. S. Zefirov, A. S. Makarenko, V. Y. Tanchuk and V. V. Prokopenko, J. Comput.-Aided Mol. Des., 2005, 19, 453–463.
32. A. Tropsha and A. Golbraikh, Curr. Pharm. Des., 2007, 13, 3494–3504.
33. A. Talevi, L. Gavernet and L. E. Bruno-Blanch, Curr. Comput.-Aided Drug Des., 2009, 5, 23–37.
34. MATLAB and Neural Network Toolbox Release 2014a, The MathWorks, Inc., Natick, Massachusetts, United States.
35. C. Rucker, G. Rucker and M. Meringer, J. Chem. Inf. Model., 2007, 47, 2345–2357.
36. A. Tropsha, Mol. Inf., 2010, 29, 476–488.
37. K. Roy, R. N. Das, P. Ambure and R. B. Aher, Chemom. Intell. Lab. Syst., 2016, 152, 18–33.
38. S. Lek, A. Belaud, P. Baran, I. Dimopoulos and M. Delacoste, Aquat. Living Resour., 1996, 9, 23–29.
39. S. Lek, M. Delacoste, P. Baran, I. Dimopoulos, J. Lauga and S. Aulagnier, Ecol. Modell., 1996, 90, 39–52.
40. J. D. Olden, M. K. Joy and R. G. Death, Ecol. Modell., 2004, 178, 389–397.
41. A. K. Bose, S. Pednekar, S. N. Ganguly, G. Chakraborty and M. S. Manhas, Tetrahedron Lett., 2004, 45, 8351–8353.
42. A. Shaabani, A. Bazgir and S. Arab-Ameri, Phosphorus, Sulfur Silicon Relat. Elem., 2004, 179, 2169–2175.
43. X. Han, F. Xu, Y. Luo and Q. Shen, Eur. J. Org. Chem., 2005, 2005, 1500–1503.
44. T. Mosmann, J. Immunol. Methods, 1983, 65, 55–63.
45. L. M. Ramos, B. C. Guido, C. C. Nobrega, J. R. Correa, R. G. Silva, H. C. B. de Oliveira, A. F. Gomes, F. C. Gozzo and B. A. D. Neto, Chem.–Eur. J., 2013, 19, 4156–4168.
46. R. F. S. Canto, A. Bernardi, A. M. O. Battastini, D. Russowsky and V. L. Eifler-Lima, J. Braz. Chem. Soc., 2011, 22, 1379–1388.
47. H. Sachdeva and D. Dwivedi, Sci. World J., 2012, 2012, 109432.
48. H. Asraf, R. Avunie-Masala, M. Hershfinkel and L. Gheber, PLoS One, 2015, 10, e0129255.
49. F. Figueiro, F. B. Mendes, P. F. Corbelini, F. Janarelli, E. H. F. Jandrey, D. Russowsky, V. L. Eifler-Lima and A. M. O. Battastini, Anticancer Res., 2014, 34, 1837–1842.
50. A. Tropsha, P. Gramatica and V. K. Gombar, QSAR Comb. Sci., 2003, 22, 69–77.
51. J. Dearden, M. Cronin and K. Kaiser, SAR QSAR Environ. Res., 2009, 20, 241–266.
52. B. Lau, Collinearity diagnostics, Version 1.3, MATLAB Central File Exchange, https://www.mathworks.com/matlabcentral/fileexchange/48163.
53. K. Gholivand, A. A. EbrahimiValmoozi, A. Gholami, M. Dusek, V. Eigner and S. Abolghasemi, J. Organomet. Chem., 2016, 806, 33–44.
54. H. W. Lilliefors, J. Am. Stat. Assoc., 1967, 62, 399–402.
55. D. Gadaleta, G. F. Mangiatordi, M. Catto, A. Carotti and O. Nicolotti, Int. J. Quant. Struct. Prop. Relat., 2016, 1, 45–63.
56. N. Minovski, S. Zuperl, V. Drgan and M. Novic, Anal. Chim. Acta, 2013, 759, 28–42.
57. N. Fjodorova, M. Novic, A. Roncaglioni and E. Benfenati, J. Comput.-Aided Mol. Des., 2011, 25, 1147–1158.
58. J. D. Olden and D. A. Jackson, Ecol. Modell., 2002, 154, 135–150.
59. M. Gevrey, I. Dimopoulos and S. Lek, Ecol. Modell., 2003, 160, 249–264.

---