Accepted Manuscript

Hepatoprotection of sesquiterpenoids: a quantitative structure-activity relation‐

ship (QSAR) approach

Juliana Vinholes, Alisa Rudnitskaya, Pedro Gonçalves, Fátima Martel, Manuel

A. Coimbra, Sílvia M. Rocha

PII: S0308-8146(13)01291-0

DOI: http://dx.doi.org/10.1016/j.foodchem.2013.09.039

Reference: FOCH 14667

To appear in: Food Chemistry

Received Date: 17 May 2013

Revised Date: 27 June 2013

Accepted Date: 5 September 2013

Please cite this article as: Vinholes, J., Rudnitskaya, A., Gonçalves, P., Martel, F., Coimbra, M.A., Rocha, S.M.,

Hepatoprotection of sesquiterpenoids: a quantitative structure-activity relationship (QSAR) approach, Food

Chemistry (2013), doi: http://dx.doi.org/10.1016/j.foodchem.2013.09.039

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers

we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and

review of the resulting proof before it is published in its final form. Please note that during the production process

errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Hepatoprotection of sesquiterpenoids: a quantitative structure-activity 1

relationship (QSAR) approach 2

3

4

5

Juliana Vinholes1, Alisa Rudnitskaya2, Pedro Gonçalves3, Fátima Martel3, Manuel A. 6

Coimbra1 and Sílvia M. Rocha1* 7

8

1 QOPNA, Chemistry Department, University of Aveiro, 3810-193 Aveiro, Portugal 9

2 CESAM, Chemistry Department, University of Aveiro, 3810-193 Aveiro, Portugal 10

3 Department of Biochemistry (U38-FCT), Faculty of Medicine, University of Porto, 11

4200-319 Porto, Portugal 12

13

14

15

16

17

18

19

* Corresponding author. Tel. + 351 234401524; Fax. + 351 234370084 20

E-mail address: smrocha@ua.pt (Sílvia M. Rocha) 21

2

ABSTRACT 22

The relative hepatoprotection effect of fifteen sesquiterpenoids, commonly 23

found in plants and plant-derived foods and beverages was assessed. Endogenous lipid 24

peroxidation (assay A) and induced lipid peroxidation (assay B) were evaluated in liver 25

homogenates from Wistar rats by the thiobarbituric acid reactive species test. 26

Sesquiterpenoids with different chemical structures were tested: trans,trans-farnesol, 27

cis-nerolidol, (-)-α-bisabolol, trans-β-farnesene, germacrene D, α-humulene, β-28

caryophyllene, isocaryophyllene, (+)-valencene, guaiazulene, (-)-α-cedrene, (+)-29

aromadendrene, (-)-α-neoclovene, (-)-α-copaene, and (+)-cyclosativene. Ascorbic acid 30

was used as a positive antioxidant control. With the exception of α-humulene, all the 31

sesquiterpenoids under study (1 mM) were effective in reducing the malonaldehyde 32

levels in both endogenous and induced lipid peroxidation up to 35 % and 70 %, 33

respectively. The 3D-QSAR models developed, relating the hepatoprotection activity 34

with molecular properties, showed good fit (Radj2 0.819 and 0.972 for the assays A and 35

B, respectively) with good prediction power (Q2 > 0.950 and SDEP < 2%, for both 36

models A and B). A network of effects associated with structural and chemical features 37

of sesquiterpenoids such as shape, branching, symmetry, and presence of 38

electronegative fragments, can modulate the hepatoprotective activity observed for these 39

compounds. 40

41

Keywords: 42

Hepatoprotection 43

Lipid peroxidation 44

Quantitative structure-activity relationship 45

Rat hepatocytes 46

Sesquiterpenoids 47

TBARS 48

3

1. Introduction 49

50

Sesquiterpenoids are secondary plant metabolites and have been the subject of 51

considerable research in recent years. Such interest has been spurred mainly by their 52

presence in several medicinal plants with different health benefits properties 53

(Petronilho, Maraschin, Coimbra & Rocha 2012). Besides, these compounds are also 54

present in several fruits, namely in grapes from Vitis vinifera L., where they have been 55

reported as potential contributors to the aroma characteristics and health benefits of 56

wine (Rocha, Coelho, Vinholes & Coimbra, 2006). In fact, sesquiterpenic compounds 57

are described as possessing anti-inflammatory properties (Lim et al., 2005), antibacterial 58

properties including enhancement of bacterial susceptibility to antibiotics (Gonçalves, 59

Pereira, Gonçalves, Mendo, Coimbra & Rocha, 2011), anti-carcinogenic (Tatman & 60

Mo, 2002) and antioxidant effects (Lim et al., 2005). Ruberto & Baratta (2000) 61

reported the antioxidant activity of different sesquiterpenoids present in different 62

essential oils. Among the twenty one sesquiterpenoids tested, the oxygenated 63

sesquiterpenes showed higher antioxidant activity, farnesol (mixture of isomers), 64

trans,trans-farnesol, (+)-8-(15)-cedren-9-ol and guaiol being the most active ones. 65

Additionally, in vivo studies with trans,trans-farnesol have demonstrated its protective 66

effect against oxidative damage caused by 1,2-dimethylhydrazine in the colon of Wistar 67

rats (Khan & Sultana, 2011) and cigarette smoke toxicants (Qamar & Sultana, 2008). 68

Guaiazulene, a sesquiterpenoid present in Matricaria chamomilla L., has been reported 69

to efficiently inhibit membrane lipid peroxidation in vitro (Kourounakis, Rekka & 70

Kourounakis, 1997a). This compound also showed potential protection against 71

paracetamol hepatointoxication in vivo. When using this sesquiterpene, a decrease in the 72

metabolic activation of paracetamol was observed, which was attributed to its ability to 73

4

act as a chain-breaking antioxidant, preventing cytochrome P450 activity which is 74

involved in the metabolic activation of paracetamol to the toxic metabolite N-acetyl-p-75

benzoquinone imine (NAPQI) (Kourounakis, Rekka & Kourounakis, 1997b). These 76

protective effects are dependent on different factors, such as the compounds structure, 77

concentration (varying from μM to mM) and biological models used. 78

β-Caryophyllene, a sesquiterpenoid commonly found in plants and plant-derived 79

foods and beverages, has been reported with inhibitory activity against the enzyme 5-80

lipoxygenase, reducing the in vitro formation of lipid peroxides, and with ability to 81

protect hepatic stellate cells against the lipid peroxidation caused by exposure to carbon 82

tetrachloride (Calleja et al., 2013). β-Caryophyllene also showed the highest 83

neuroprotective activity against oxidative damage among a series of eleven terpenoids 84

(Chang, Kim & Chun, 2007). The structure activity relationship (QSAR), established by 85

Chang et al. (2007), was quantified by developing a mathematic model that related the 86

neuroprotective activity with particular molecular characteristics. The use of QSAR 87

models is very popular for the estimation of the compounds druglikeness, aiming to 88

identify chemical structures associated with bioactivity and bioavailability. 89

It is already known that antioxidant compounds play an important role in 90

protecting cells against damage from free radicals and reactive oxygen species (ROS). 91

When ROS exceeds cellular antioxidant capacity, this may result in the damage of 92

proteins, DNA and lipids. The alterations caused by ROS on cell membrane lipids lead 93

to changes in its permeability. Thus, cell membrane functions such as controlled flux of 94

ionic and non ionic substances, including toxic substances is lost, and DNA alterations 95

and low-density lipoprotein oxidation can occur. This process can result in cell death 96

and neoplasia, which probably contribute to many human diseases, such as 97

cardiovascular diseases and cancer, and even in the acceleration of aging. The process 98

5

of lipid peroxidation involves different pathways: (a) non-enzymatic free radical-99

mediated chain reaction, involving the participation of ROS, transition metals and other 100

free radicals; (b) enzymatic reaction, involving cyclooxygenases and lipoxygenases on 101

the oxidation of membrane lipids and (c) non-enzymatic and non-radical mechanisms, 102

involving singlet oxygen and ozone (Niki, Yoshida, Saito & Noguchi, 2005). After 103

initiation of the lipid peroxidation cycle, the propagation process can only be broken by 104

the intervention of free radical scavengers and antioxidants. The mechanism of 105

antioxidant action involves membrane stabilisation and neutralisation of free radicals 106

(Wiseman, Quinn & Halliwell, 1993). 107

The present study aimed to investigate the potential hepatoprotection activity of 108

sesquiterpenoids, commonly found in plants and plant-derived foods and beverages 109

against lipid peroxidation using liver homogenate from Wistar rats in an in vitro study. 110

In order to establish the structure-activity relationships for future hepatoprotection 111

prediction, fifteen sesquiterpenoids representing different chemical classes were used: 112

linear alcohols, a cyclic alcohol, a linear hydrocarbon, cyclic hydrocarbons, bicyclic 113

hydrocarbons, tricyclic hydrocarbons and tetracyclic hydrocarbons (Fig. 1). As far as we 114

known this is the first attempt to achieve such relationship using these set of compounds 115

commonly found in nature. 116

117

2. Materials and methods 118

119

2.1. Materials 120

Ethanol (purity ≥ 99.8 %) was from Riedel-de Haën (Seelze, Germany), 3-121

amino-1,2,4-triazole (95 %), mercaptosuccinic acid (97 %), and tert-butyl 122

hydroperoxide (tert-BuOOH) were purchased from Aldrich Chemical Co (Milwaukee, 123

6

WI, USA). Pentobarbital sodium salt, potassium phosphate monobasic (> 99 %), and 124

sodium phosphate dibasic (> 99 %) were purchased from Sigma (St. Louis, MO, USA). 125

Triton X-100 was purchased from Merck (Darmstadt, Germany). The ascorbic acid was 126

purchased from AnalaR BDH Chemical Ltd. (London, United Kingdom). 2-127

Thiobarbituric acid (TBA) (≥ 98 %) was purchased from Fluka (Buchs, Switzerland). 128

Fifteen sesquiterpenic compounds were used (Fig. 1): β-caryophyllene (≥ 98.5 % GC) , 129

(-)-α-cedrene (≥ 99 %), (+)-aromadendrene (> 97 % GC), α-humulene (>98% GC), (-)-130

α-copaene (≥ 90 %), (+)-cyclosativene (≥ 99 %), isocaryophyllene (> 98 %), (-)-α-131

neoclovene (≥ 95 %), (+)-valencene (≥ 70 %), trans-β-farnesene (90 %), trans,trans-132

farnesol (≥ 95 %), and (-)-α-bisabolol (≥ 95 %) were purchased from Fluka (Buchs, 133

Switzerland). cis-Nerolidol (≥ 96 %) was purchased from Sigma-Aldrich (St. Louis, 134

Mo.), guaiazulene (> 98 %) was purchased from TCI Europe N.V. (Zwijndrecht, 135

Belgium), and germacrene D (40%, natural extract enriched in germacrene D, this 136

extract containing also other sesquiterpenoids, where β-farnesene ca. 10 % is the second 137

most abundant) was kindly offered by Tecnufar Ibérica, (Madrid, Spain). Stock 138

solutions of each sesquiterpenic compound (10 mM) were prepared in ethanol. 139

140

2.2. Animals and hepatocytes isolation 141

Female Wistar rats weighting ca. 400-600 g were used in the experiments. 142

Animals were kept one per cage under controlled environmental conditions: 12.00 h 143

light - dark cycle, room temperature (24 ºC), and food and tap water were allowed ad 144

libitum. Animals were anesthetized with pentobarbital (50 mg/kg). The livers were 145

removed and placed (0.5 g/ml) in a glass tube with homogenization buffer (KH2PO4 146

62.5 mM, Na2HPO4 50.0 mM and Triton X-100 0.1 %). For the experiments, the liver 147

tissue was homogenized in a glass-teflon homogenizer and kept continuously on ice. 148

7

Rat liver homogenate (RLH) inhibition of catalase and glutathione peroxidase activities 149

was achieved by the addition of 3-amino-1,2,4-triazol and mercaptosuccinic acid (both 150

at 10 mM), respectively. 151

152

2.3. Evaluation of sesquiterpenoids hepatoprotection potential 153

The individual relative hepatoprotection of the fifteen sesquiterpenoids was 154

evaluated for endogenous (assay A) and induced (assay B) lipid peroxidation effects, as 155

shown in Fig. 2, according to Fernandes, Carvalho, Remião, Bastos, Pinto & Gottlieb 156

(1995). Relative hepatoprotection of ascorbic acid was evaluated and used as positive 157

antioxidant control. In order to assess the molecular structure - activity relationships, 158

sesquiterpenoids were tested at the same molar concentration. The procedure consisted 159

of individual additions of 40 μl from each sesquiterpenoid solution (1 mM, final 160

concentration) or ascorbic acid (1 mM, final concentration) to aliquots of 400 μl of RLH 161

and incubation at 37 ºC for 1 hour. Compounds were tested at 1 mM, based on the 162

protective effects reported in the literature for some of them (Kourounakis et al., 1997b; 163

Rocha et al., 2011). In the initial step a concentration 10 times lower (0.1 mM) was 164

tested, but none of the compounds under study were able to reduce the MDA levels 165

(data not shown). Thus, the hepatoprotection evaluation effect was performed using 166

standards at 1mM. Subsequently, tert-BuOOH (1 mM) was added to the assay B to 167

induce lipid peroxidation, and the assay was kept at the same temperature for one more 168

hour. Negative controls were run for all treatments (ethanol for all sesquiterpenoids and 169

water for ascorbic acid assay). 170

After incubation, the level of malonaldehyde (MDA) in the assays A and B was 171

measured using the thiobarbituric acid reactive substances (TBAR’s) assay adapted 172

from Fernandes et al. (1995). Briefly, the homogenates were centrifuged at 13000 rpm 173

8

for 10 min at 4 ºC, and the supernatant (100 μl) was collected and precipitated with 200 174

μl of TCA 10 %. Centrifugation was then carried out at 13000 rpm for 2 min at 4 ºC and 175

100 μl of supernatant was added to 100 μl of TBA (1 %). Samples were kept in a water 176

bath at 96 ºC for 10 min and, after cooling to room temperature, the absorbance at 535 177

nm was recorded. The concentration of MDA was determined using the extinction 178

coefficient of 1.56 × 105 1/M cm, and results were expressed as percentage of reduction 179

of nmol MDA/mg protein in relation to negative controls. Four independent 180

experiments were repeated at least three times on 4 different days. 181

182

2.4. Protein determination 183

The concentration of MDA formed during lipid peroxidation was normalized by 184

the protein content in homogenate, which was quantified as described by Bradford, 185

1976, using human serum albumin as a standard. 186

187

2.5. Molecular descriptors calculation 188

The 3D-structures of all the compounds were drawn and their minimum energy 189

conformations were obtained by the mechanic method of Allinger (MM2) using 190

HyperChem software evaluation version 7.0 (Hypercube, Inc. Gainesville, Florida, USA 191

2002). Then, they were transferred into the Dragon program evaluation version 5.4 192

(Talette SRL, Milan, Italy) and the molecular descriptors were calculated. The 193

molecular descriptors are organized in twelve groups: (a) constitutional; (b) molecular 194

properties; (c) atom-centered fragments; (d) topological; (e) connectivity indices; (f) 195

information indices; (g) functional group counts; (h) geometrical; (i) Radial Distribution 196

Functions (RDF); (j) Molecule Representation of Structures based on Electron 197

diffraction (3D-MoRSE); (k) Weighted Holistic Invariant Molecular descriptors 198

9

(WHIM) and (l) Geometry,Topology, and Atom-Weights AssemblY (GETAWAY). 199

The meaning of these molecular descriptors and the calculation procedures are 200

summarized by Todeschini & Consonni (2000). 201

202

2.6. QSAR model development 203

A total of 803 molecular descriptors were calculated by the Dragon software. 204

Correlation coefficients between the activity (dependent variable) and the descriptor or 205

descriptors were determined by correlation analysis. Reduced descriptors were obtained 206

by discarding highly inter-correlated (r > 0.9) descriptors and molecular descriptors 207

with constant or near constant values, resulting in 80 molecular descriptors. Further, 208

selection of the best subset of descriptors for the prediction of sesquiterpenoids 209

hepatoprotection activity was done using genetic algorithm (GA) and Principle 210

Component Regression (PCR). GA is an optimization technique based on the principles 211

of evolutionary selection. GA is effective for finding global minimum (or maximum) 212

for complex tasks, in particular ones that involve a large number of independent 213

variables (Liu & Long, 2009). GA has been widely applied to the selection of the best 214

subset of the molecular descriptors in QSAR and quantitative structure-property 215

relationships analysis in different fields (Liu & Long, 2009). 216

The reason for optimization using GA was to minimise the prediction error for 217

the QSAR model. Thus, Root Mean Square Error in Cross-Validation (RMSECV) 218

calculated using PCR was used as objective function. PCR was employed for 219

calculating calibration models during the optimization step due to the large number of 220

descriptors compared to the number of samples. Optimization was done using activity 221

data from both assays A and B. 222

Calibration models with respect to the hepatoprotection activities were build 223

10

using the best subsets of molecular descriptors and Multilinear Regression (MLR). 224

MLR calibration models were validated using leave-one-out cross-validation and 225

evaluated using the following statistics: adjusted R2 and Q2, Standard Deviations Errors 226

in Calculation (SDEC) and Prediction (SDEP). As the number of samples is relatively 227

small compared to the number of the descriptors necessary to achieve the best model, 228

statistics adjusted for the number of the variables were used as recommended by 229

Todeschini (2010). All calculations were done in MATLAB 7.3.0. GA was 230

implemented using GA toolbox v. 1.2 available from the Department of Automatic 231

Control and Systems Engineering of The University of Sheffield, UK (Chipperfield & 232

Fleming, 1995). 233

234

3. Results and discussion 235

236

3.1. Hepatoprotection of sesquiterpenoids based on MDA values 237

Relative hepatoprotection of fifteen sesquiterpenoids with different molecular 238

structures (Fig. 1) was evaluated by the estimation of MDA levels in rat liver 239

homogenates for endogenous (assay A) and induced (assay B) lipid peroxidation assays. 240

The amounts of MDA formed in assay A, for water and ethanol controls, were 241

4.47 ±0.27 and 4.23 ±0.38 nmol MDA/mg protein, respectively. These values were 242

three and seven times higher on controls exposed to tert-BuOOH (assay B), being 243

15.51 ±2.02 nmol MDA/mg protein for water control, and 29.50 ±0.59 nmol MDA/mg 244

protein for ethanol control. The percentage of MDA reduction, displayed on Fig.3, for 245

ascorbic acid and sesquiterpenoids compounds was calculated against their respective 246

controls. 247

For endogenous hepatoprotection (assay A), (-)-α-neoclovene showed the 248

11

strongest reduction (35.46 ±1.25 %), followed by cis-nerolidol (25.60 ±0.98 %), 249

trans,trans-farnesol (23.82 ±1.03 %), (-)-α-copaene (23.66 ±0.71 %) and trans-β-250

farnesene (21.82 ±1.94 %) (Fig. 3). Moreover, lower activities were observed for 251

guaiazulene (17.20 ±0.61 %), (+)-valencene (14.77 ±0.49 %), β-caryophyllene 252

(14.61 ±1.43 %), germacrene D (13.55 ±1.50 %), (-)-α-cedrene (8.80 ±0.76 %), (+)-253

cyclosativene (7.60 ±0.37 %) and (+)-aromadendrene (4.46 ±0.45 %). Isocaryophyllene 254

(1.50 ±0.19 %) and (-)-α-bisabolol (0.32 ±0.06 %) had very low activity and for α-255

humulene no protection activity was observed. 256

The sesquiterpenoids hepatoprotection for assay B ranged from 23.70 % to 69.72 257

% (Fig. 3). trans,trans-Farnesol, guaiazulene and trans-β-farnesene showed the highest 258

hepatoprotector effects with 69.72 ±1.82 %, 69.53 ±1.42 % and 64.55 ±2.20 % of MDA 259

reduction, respectively. Lower reduction effects were observed for (+)-valencene (56.97 260

±1.31 %), (-)-α-copaene (54.19 ±1.66 %), (-)-α-bisabolol (47.86 ±1.21 %), 261

isocaryophyllene (47.31 ±4.77 %), (-)-α-neoclovene (43.50 ±2.65 %), (+)-cyclosativene 262

(41.68 ±1.01 %), (+)-aromadendrene (37.13 ±1.56%), (-)-α-cedrene (36.60 ±1.33 %), 263

cis-nerolidol (36.50 ±4.47 %), germacrene D (25.77 ±1.55 %), β-caryophyllene (23.70 264

±2.02 %) and α-humulene (-6.48 ±2.49%). 265

The comparison between results obtained for the assays A and B showed that 266

under oxidative stress conditions the sesquiterpenoids hepatoprotection activity was 267

higher. The decrease of MDA formation in assay B was 150 times higher compared to 268

assay A for α-bisabolol (47.86 % and 0.32 %, respectively), while for isocaryophyllene 269

the decrease was 30 times higher (47.31 % and 1.49 %, respectively). Other 270

sesquiterpenoids showed smaller differences (1 to 8 times higher) though the same 271

tendency was maintained. The results also show that the sesquiterpenoids are able to 272

protect the hepatocytes from hydrophobic oxidants more efficiently than ascorbic acid 273

12

(which produced a 44.19 ±0.92 % reduction), possibly due to their lipophilicity. 274

However, for endogenous hepatoprotection none of the sesquiterpenoids were as 275

efficient as ascorbic acid, as this compound acts mainly in the aqueous phase by 276

increasing the serum oxygen-radical absorbance and inhibiting peroxyl radicals (Frei, 277

England, & Ames, 1989). 278

Most of the hepatoprotective drugs belong to the group of free radical 279

scavengers or antioxidants, and their action involves membrane stabilisation and 280

neutralisation of free radicals (Wiseman et al., 1993). To achieve membrane 281

stabilisation the compound should be able to penetrate it, which is a physico-chemical 282

process highly dependent on its solubility and diffusion across the lipid bilayer. 283

Compound size, lipophilicity and shape are parameters that highly influence that 284

process (Marrink & Berendsen, 1996). Since the largest portion of the cell membrane is 285

lipophilic, simple diffusion for lipophilic molecules can be attained with few 286

restrictions. The lipophilicity of the molecule depends on various physical and chemical 287

characteristics, such as molecular surface area, molecular volume and polarity (Leo, 288

Hansch & Jow, 1976). The sesquiterpenoids under study present a lipophilic character, 289

with a ALogP (Ghose-Crippen octanol-water partition coeff. (logP)) ranging from 3.6 to 290

5.7. This fact was determinant when the oxidative stress was induced by tert-BuOOH 291

on assay B, since both sesquiterpenoids and the membrane-permeant oxidant tert-292

BuOOH have high affinity to the cell membrane lipid fraction. 293

The obtained results are also in agreement with the data of lipid peroxidation 294

reduction and antioxidant effects determined for some of the sesquiterpenoids under 295

study using different methodological approaches. Ruberto & Baratta (2000) found that 296

the inhibition of lipid peroxidation follows this order: farnesol (mixture of isomers) > 297

trans,trans-farnesol > (±)-α-bisabolol > (+)-valencene > α-cedrene > (+)-298

13

aromadendrene > α-humulene. 299

Guaiazulene exhibits the highest activity in assay B. This compound was 300

reported as protecting against the hepatotoxicity of paracetamol in vivo by decreasing its 301

metabolic activation by preventing both cytochrome P450 activity and NAPQI-induced 302

glutathione depletion (Kourounakis et al., 1997b). This property was attributed to the 303

ability of guaiazulene to act as a chain-breaking antioxidant. Additionally, very efficient 304

inhibition by guaiazulene of membrane lipid peroxidation in vitro has been reported, 305

which is in agreement with the high activity observed for this compound in assay B 306

(Kourounakis et al., 1997a). Also, (−)-α-bisabolol was shown to protect the gastric 307

mucous membrane of male Swiss mice against injuries caused by ethanol by reducing 308

lipid peroxidation and increasing the superoxide dismutase activity, which is in 309

agreement with the activity of this compound in the assay B (Rocha et al., 2011). 310

trans,trans-Farnesol, possessing also high hepatoprotection activity in assays A 311

and B, has been shown to reduce the lipid peroxidation on egg yolk homogenates 312

(Ruberto & Baratta, 2000). Moreover, in vivo protection against oxidative damage 313

caused by 1,2-dimethylhydrazine in the colon of Wistar rats (Khan & Sultana, 2011), 314

and against cigarette smoke toxicants in the trachea of Wistar rats (Qamar & Sultana, 315

2008) has also been reported for trans,trans-farnesol. 316

β-Caryophyllene, with mild MDA reduction in both A and B assays, has been 317

reported to have antioxidant activity, which explains its role as a neuroprotector (Chang 318

et al., 2007). It also has a scavenger ability in relation to hydroxyl and superoxide anion 319

radicals, and also inhibits the enzymes xanthine oxidase and 5-lipoxygenase, which are 320

involved in the initiation of the lipid peroxidation (Calleja et al., 2013). Neuroprotective 321

activity was related to the lipophilicity, shape and electrostatic parameters (Chang et al., 322

2007). 323

14

Another mechanism of hepatoprotection may involve the interaction of these 324

compounds with the cell defense system. The protection against lipid peroxidation in 325

biological models involves different biochemical molecules such as the endogenous 326

antioxidants (glutathione, ascorbic acid, tocopheryl) and the detoxification enzyme 327

system (i.e. catalase, glutathione peroxidase, superoxide dismutase, glutathione-S-328

transferase redox system, quinone reductase) (Valko, Leibfritz, Moncol, Cronin, Mazur 329

& Telser, 2007). The treatment of cells with an oxidant agent such as tert-BuOOH, that 330

is metabolized originating tert-butoxyl, peroxyl and methyl radicals, results in a 331

significant decrease of molecules involved in the cell defense mechanism in order to 332

overcome the injury, and consequently, an increase in lipid peroxidation is observed 333

(Yang, Hong, Lee, Kim & Lee, 2013). Different studies reported the increase of 334

glutathione levels and induction of the enzymatic detoxification system by trans,trans-335

farnesol and β-caryophyllene when biological models are pre-treated with this 336

compound before injury by different agents (Calleja et al., 2013; Khan & Sultana, 2011; 337

Qamar & Sultana, 2008). 338

339

3.2. QSAR model for sesquiterpenoids hepatoprotection 340

A QSAR model based on three-dimensional molecular descriptors for the fifteen 341

sesquiterpenoids (Tables 1 and S1) for the prediction of the relative MDA reduction 342

(Fig. 3) was developed. First, the number of descriptors was reduced using GA, 343

resulting in two optimal subsets of molecular descriptors for the assays A and B. 344

Optimal descriptors subsets are shown in the Table 1. The MLR models were validated 345

by leave-one-out validation, resulting in the following equations and respective 346

statistical data: 347

348

15

Model A= 66 + 1.1e+03R6u+ - 631HATS2v - 431R4p+ + 288R4e+ - 287Gm - 8HATS6v + 166R4u+ + 219R3m+ 349

SDEC= 0.82, SDEP= 2.09, R2LOO= 0.819, Q2

LOO= 0.961 350

351

Model B= -83 + 1.8e+03R5e+ - 760HATS2v - 531R1m+ + 460R3u+ + 415R1u+ + 332R8u+ - 322R4e+ + 312G2e + 352

109R7u+ - 66R6m+ + 3Mor24p + 2Mor02m 353

SDEC= 0.09, SDEP= 1.45, R2LOO=0.972, Q2

LOO=0.994 354

355

The resulting models showed high correlations coefficients, both to calibration 356

(R2LOO= 0.819 and R2

LOO=0.972, for assays A and B, respectively) and prediction 357

(Q2LOO= 0.961 and Q2

LOO=0.994), as well as low errors (SDEC and SDEP < 2.09 %). 358

The scatter plots of the experimental data (Relative hepatoprotection obs.) versus the 359

predicted one (Relative hepatoprotection calc.) using the obtained equations are shown 360

in Fig. 4 A and B for the models A and B, respectively. The agreement observed 361

between the predicted and experimental values confirmed the efficiency of these QSAR 362

methods. 363

364

3.3. QSAR model interpretation 365

According to the obtained models, the most relevant molecular descriptors 366

related to the hepatoprotection properties belong to three groups: the 3D GEometry, 367

Topology, and Atom-Weights AssemblY (GETAWAY), Weighted Holistic Invariant 368

Molecular (WHIM) and Molecule Representation of Structure based on Electron 369

diffraction (MoRSE) descriptors (Tables 1 and S1). Due to the complexity of the 3D 370

descriptors, interpretation is usually done on a series of structurally similar compounds 371

(Freitas, Paz & Castilho, 2009). 372

GETAWAY descriptors encode both geometrical and topological information. 373

While geometrical information is given by the influence molecular matrix (H-374

GETAWAY descriptors), the topological one is given by the influence/distance matrix 375

16

R. This matrix R combines the molecular matrix influence with the molecular inter-376

atomic geometrical distance weighted by physicochemical properties such as atomic 377

mass, polarizability, van der Waals volume and electronegativity (R-GETAWAY 378

descriptors) (Cossoni, Todeschini & Pavan, 2002). 379

The models obtained for assays A and B included several R-GETAWAY 380

descriptors, among which descriptors R6u+ and R5e+ had the highest positive 381

regression coefficients, for each model, respectively. Larger values of the 382

autocorrelation descriptors at topological distance (lag) correspond to the most external 383

atoms that are simultaneously next to each other in the molecular space, i.e. terminal 384

atoms located next to other terminal atoms. Moreover, they can be related to the 385

molecular shape, where higher values are associated with compact molecules and lower 386

values are associated with more linear conformations (Cossoni et al., 2002). A tendency 387

in the value of the R6u+ descriptor can be observed among the compounds with higher 388

activity in assay A, indicating that an increase in molecular compactness may be related 389

with the increase of the activity in assay A. Thus, molecules possessing higher R6u+ 390

values, for example α-neoclovene (R6u+=0.051), (-)-α-copaene (R6u+=0.039) and 391

trans-β-farnesene (R6u+=0.036), were more efficient than α-humulene (R6u+=0.027) 392

and (-)-α-bisabolol (R6u+=0.024) that have low values of this descriptor. 393

The most important descriptor in model B is R5e+, which is a R index weighted 394

by atomic Sanderson electronegativity. The weighting of the R indexes encode 395

information about substituents differently from unweighted indexes. In this case, largest 396

values of this descriptor can be expected when high electronegative atoms are situated 397

far from the centre of the molecule at a topological distance of 5 bonds. Since this 398

descriptor had a positive contribution to the model, it is expected that the 399

hepatoprotective activity increases with the increase of its values. Therefore, 400

17

compounds with MDA reduction above 50% such as guaiazulene, trans,trans-farnesol, 401

trans-β-farnesene, (+)-valencene and (−)-α-copaene, had higher values of the R5e+ 402

descriptor (with values between 0.042 and 0.049) compared to molecules with lower 403

hepatoprotective activity such as α-humulene and germacrene D (R5e+= 0.033 and 404

0.036). trans,trans-Farnesol has a terminal hydroxyl group while trans-β-farnesene and 405

(+)-valencene have a terminal alkene, which can explain the higher values of 406

electronegativity observed for the R5e+ descriptor. Recently, the antioxidant activity of 407

di(hetero)arylamines derivatives of benzo[b]thiophenes has been related to the 408

descriptors encoding atomic electronegativities, and the effect was evidenced to be due 409

to the presence of electron-donating substituents (Abreu, Ferreira & Queiroz, 2009). 410

The second most important descriptor in both models is HATS2v, which belongs 411

to the H-GETAWAY group. H-GETAWAY descriptors are calculated using diagonal 412

elements of the molecular influence matrix accounting for the relative position of each 413

atom in the 3D molecular space weighted by different atomic properties, in this case 414

atomic van der Waals volume (Cossoni et al., 2002). High values of this descriptor 415

indicate higher ramification of the molecules, while low values of this descriptor are 416

typical for more linear molecules. The negative influence of HATS2v on both 417

hepatoprotection models indicates that molecules possessing lower values of this 418

descriptor are more active. The most active compounds in models A and B were 419

trans,trans-farnesol, trans-β-farnesene, (-)-α-neoclovene, cis-nerolidol, (-)-α-copaene 420

and guaiazulene (HATS2v ranging from 0.058 to 0.089), while α-humulene, the less 421

active compound for both assays, has an HATS2v value of 0.097. 422

The two WHIM descriptors involved in models A and B, Gm and G2e, 423

respectively, represent different sources of chemical information for the whole 3D 424

molecular structure in terms of size, shape, symmetry, and atom distribution. They are 425

18

calculated by performing a PCA on a weighted covariance matrix of the centered 426

Cartesian coordinates of a molecule, obtained from different weighting schemes for the 427

atoms (Todeschini & Gramatica, 1997). G2e descriptor encodes the symmetry of the 428

molecules along the second component weighted by electronegativity, while Gm is the 429

total symmetry that tends to 1 as the molecule shows a central symmetry along each 430

axis and to 0 when there is a decrease in the symmetry along at least one axis 431

(Todeschini & Consonni, 2000). For model A, Gm showed a negative effect, indicating 432

that the reduction of MDA increases with the decrease of symmetry. For instance, the 433

less active compound, α-humulene, has the highest value (Gm=0.191), while 434

compounds with activity higher than 20% showed lower values of this descriptor, 435

namely, trans-β-farnesene, trans,trans-farnesol , (−)-α-copaene, cis-nerolidol and (-)-α-436

neoclovene (with Gm ranging from 0.163 to 0.186). 437

In model B, the hepatoprotection activity increases for more symmetric 438

molecules, considering the second component and an electronegative fragment. Thus, 439

the compounds with more than 50% MDA reduction such as guaiazulene, trans,trans-440

farnesol, trans-β-farnesene and (−)-α-copaene, all with high values of the G2e 441

descriptor (values ranging from 0.173 to 0.205), were more active than α-humulene 442

(G2e=0.159). 443

From the structural features described by the QSAR models, that explain 444

scavenger activity, three key observations can be made: 1. (−)-α-neoclovene and (−)-α-445

copaene have one allylic hydrogen atom that can be abstracted giving rise to a more 446

stable radical; 2. in the case of (−)-α-copaene an extremely stable radical can be formed 447

due to the presence of an allylic hydrogen on a tertiary carbon; 3. cis-nerolidol, 448

trans,trans-farnesol and trans-β-farnesene also have additional allylic hydrogens and 449

the first two also have a hydroxyl group from which the hydrogen can be abstracted. 450

19

Besides, they are polyunsaturated compounds with a lipophilic backbone similar to the 451

lipid core of the cell membrane, being particularly susceptible to oxidation damage. 452

The high activity observed for guaiazulene in assay B can be associated with its 453

rigid and therefore less ramified structure, which is reflected in a relatively low value of 454

the HATS2v descriptor, and its symmetry in the second component, reflected by the 455

high value of the G2e descriptor. Another important characteristic for the antioxidant 456

activity of guaiazulene is the number of double bounds and their conjugated position, 457

which allows easier free radical stabilization. 458

459

4. Conclusions 460

461

This is the first attempt to develop 3D-QSAR models of the hepatoprotective 462

activity of sesquiterpenoids with different backbone structures using an in vitro model 463

system. The developed models allowed the extraction of relevant information 464

suggesting that sesquiterpenoids possessing more compact molecular structures ((-)-α-465

neoclovene and (-)-α-copaene), low ramification (trans,β-farnesene, trans,trans-466

farnesol and cis-nerolidol) and less symmetric (according to Gm- total symmetry of the 467

molecule) (trans-β-farnesene, trans,trans-farnesol, (−)-α-copaene, cis-nerolidol and (-)-468

α-neoclovene) will be more effective for endogenous hepatoprotection. Otherwise, 469

compounds with electronegative substituents (guaiazulene, trans,trans-farnesol, trans-470

β-farnesene, (+)-valencene and (−)-α-copaene), less ramified structures (trans,trans-471

farnesol, trans-β-farnesene, (-)-α-copaene and guaiazulene) and with more symmetry 472

(according to G2e- symmetry considering the second component) with an 473

electronegative terminal fragment (guaiazulene, trans-β-farnesene and trans,trans-474

farnesol) seem to be more effective for the induced hepatoprotection. This knowledge 475

20

can be useful for future valorisation of plants or related materials containing such 476

structures. Finally, it is important to point out that the concentration tested (1 mM) to 477

evaluate the hepatoprotective effects is higher than that usually observed for these 478

compounds in plant materials. Thus, this study supports the current tendencies of 479

valorisation of natural products as a source of bioactive compounds for the formulation 480

of foods and/or nutraceuticals enriched extracts. 481

482

Acknowledgement 483

J. Vinholes thanks the PhD grant from Fundação para a Ciência e a Tecnologia 484

(SFRH/BD/25338/2005). The authors would like to give their thanks for the financial 485

support of Research Unit 62/94, QOPNA (project PEst-C/QUI/UI0062/2011) 486

487

488

21

References 489

Abreu, R. M. V., Ferreira, I. C. F. R., & Queiroz, M. J. R. P. (2009). QSAR model for 490 predicting radical scavenging activity of di(hetero)arylamines derivatives of 491 benzo[b]thiophenes. European Journal of Medicinal Chemistry, 44, 1952-1958. 492

Bradford, M. M. (1976). A rapid method for the quantitation of microgram quantities of 493 protein utilizing the principle of protein-dye binding. Analytical Biochemistry, 494 72, 248-254. 495

Calleja, M. A., Vieites, J. M., Montero-Meterdez, T., Torres, M. I., Faus, M. J., Gil, A., 496 & Suárez, A. (2013). The antioxidant effect of β-caryophyllene protects rat liver 497 from carbon tetrachloride-induced fibrosis by inhibiting hepatic stellate cell 498 activation. British Journal of Nutrition, 109, 394-401. 499

Chang, H. J., Kim, H. J., & Chun, H. S. (2007). Quantitative structure-activity 500 relationship (QSAR) for neuroprotective activity of terpenoids. Life Sciences, 501 80, 835-841. 502

Chipperfield, A. J., & Fleming, P. J. (1995). The MATLAB genetic algorithm toolbox. 503 In: Proceedings of IEE Colloquium on Applied Control Techniques Using 504 MATLAB, pp 10/1-10/4. 505

Consonni, V., Todeschini, R., & Pavan, M. (2002). Structure/response correlations and 506 similarity/diversity analysis by GETAWAY descriptors. 1. Theory of the novel 507 3D molecular descriptors. Journal of Chemical Information and Computer 508 Sciences, 42, 682-692. 509

Fernandes, E. R., Carvalho, F. D., Remião, F. G, Bastos, M. L., Pinto, M. M., & 510 Gottlieb, O. R. (1995). Hepatoprotective activity of xanthones and 511 xanthonolignoids against tert-butylhydroperoxide-induced toxicity in isolated rat 512 hepatocytes-comparison with silybin. Pharmaceutical Research, 12, 1756-1760. 513

Frei, B., England L., & Ames, B. N. (1989). Ascorbate is an outstanding antioxidant in 514 human blood plasma. Proceedings of the National Academy of Sciences of the 515 United States of America, 86, 6377-6381. 516

Freitas, H. F., Paz, O. S., & Castilho, M. S. (2009). Estudos de QSAR 3D para um 517 conjunto de inibidores de butirilcolinesterase humana. Química Nova, 32, 2114-518 2121. 519

Gonçalves, O., Pereira, R., Gonçalves, F., Mendo, S., Coimbra, M. A., & Rocha, S. M. 520 (2011). Evaluation of the mutagenicity of sesquiterpenic compounds and their 521 influence on the susceptibility towards antibiotics of two clinically relevant 522 bacterial strains. Mutation Research/Genetic Toxicology and Enviromental 523 Mutagenesis, 723, 18-25. 524

Khan, R., & Sultana, S. (2011). Farnesol attenuates 1,2-dimethylhydrazine induced 525 oxidative stress, inflammation and apoptotic responses in the colon of Wistar 526 rats. Chemico-Biological Interactions, 192, 193-200. 527

Kourounakis, A. P., Rekka, E. A., & Kourounakis, P. N. (1997a). Antioxidant activity 528 of guaiazulene and protection against paracetamol hepatotoxicity in rats. Journal 529 of Pharmacy and Pharmacology, 49, 938-942. 530

Kourounakis, A. P., Rekka, E. A., & Kourounakis, P. N. (1997b). Effect of guaiazulene 531

22

on some cytochrome P450 activities. Implication in the metabolic activation and 532 hepatotoxicity of paracetamol. Archiv der Pharmazie, 330, 7-11. 533

Leo, A., Hansch, C., & Jow, P. Y. C. (1976). Dependence of hydrophobicity of apolar 534 molecules on their molecular volume. Journal of Medicinal Chemistry, 19, 611-535 615. 536

Lim, C. S., Jin, D.-Q., Mok, H., Oh, S. J., Lee, J. U., Hwang, J. K., Ha, I., & Han, J.-S. 537 (2005). Antioxidant and antiinflammatory activities of xanthorrhizol in 538 hippocampal neurons and primary cultured microglia. Journal of Neuroscience 539 Research, 82, 831-838. 540

Liu, P., & Long, W. (2009). Current mathematical methods used in QSAR/QSPR 541 studies. International Journal of Molecular Sciences, 10, 1978-1998 542

Marrink, S. J. A., & Berendsen, H. J. C. (1996). Permeation process of small molecules 543 across lipid membranes studied by molecular dynamics simulations. The Journal 544 of Physical Chemistry, 100, 16729-16738. 545

Niki, E., Yoshida, Y., Saito, Y., & Noguchi, N. (2005). Lipid peroxidation: 546 Mechanisms, inhibition, and biological effects. Biochemical and Biophysical 547 Research Communications, 338, 668-76. 548

Petronilho, S., Maraschin, M. Coimbra, M. A., & Rocha, S. M. (2012). In vitro and in 549 vivo studies of natural products: a challenge for their valuation. The case study 550 of Chamomile (Matricaria recutita L.). Industrial Crops and Products, 40, 1-12. 551

Qamar, W., & Sultana, S. (2008). Farnesol ameliorates massive inflammation, oxidative 552 stress and lung injury induced by intratracheal instillation of cigarette smoke 553 extract in rats: An initial step in lung chemoprevention. Chemico-Biological 554 Interactions, 176, 79-87. 555

Rocha, S. M., Coelho, E., Vinholes, J., & Coimbra, M. A. (2006). Grapes and wine 556 from Vitis vinifera L. as a potential source of sesquiterpenoids. In V. K., Singh, 557 J. M., Govil, K., Ahmade, & R. K. Sharma (Eds.), Recent Progress in Medicinal 558 Plants (pp. 253-272). Houstoun: Studium Press LLC. 559

Rocha, N. F. M., Oliveira, G. V. d., Araújo, F. Y. R. d., Rios, E. R. V., Carvalho, A. M. 560 R., Vasconcelos, L. F., Macêdo, D. S., Soares, P. M. G., Sousa, D. P. D., & 561 Sousa, F. C. F. d. (2011). (-)-α-Bisabolol-induced gastroprotection is associated 562 with reduction in lipid peroxidation, superoxide dismutase activity and 563 neutrophil migration. European Journal of Pharmaceutical Sciences, 44, 455-564 461. 565

Ruberto, G., & Baratta, M. T. (2000). Antioxidant activity of selected essential oil 566 components in two lipid model systems. Food Chemistry, 69, 167-174. 567

Tatman, D., & Mo, H. (2002). Volatile isoprenoid constituents of fruits, vegetables and 568 herbs cumulatively suppress the proliferation of murine B16 melanoma and 569 human HL-60 leukemia cells. Cancer Letters, 175, 129-139. 570

Todeschini, R. (2010). Useful and unuseful summaries of regression models. URL 571 http://www.moleculardescriptors.eu/tutorials/T5_moleculardescriptors_models.p572 df. Accessed 08.07.12. 573

Todeschini, R., & Consonni, V. (2000). Handbook of Molecular Descriptors, I, 574 Mannhein: Wiley-VCH. 575

23

Todeschini, R., & Gramatica, P. (1997). SD-modelling and prediction by WHIM 576 descriptors. Part 5. Theory development and chemical meaning of WHIM 577 descriptors. Quantitative Structure-Activity Relationships, 16, 113-119. 578

Valko, M., Leibfritz, D., Moncol, J., Cronin, M. T. D., Mazur, M., Telser, J. (2007). 579 Free radicals and antioxidants in normal physiological functions and human 580 disease. The International Journal of Biochemistry & Cell Biology, 39, 44-84 581

Wiseman, H., Quinn, P., & Halliwell, B. (1993). Tamoxifen and related compounds 582 decrease membrane fluidity in liposomes: Mechanism for the antioxidant action 583 of tamoxifen and relevance to its anticancer and cardioprotective actions? FEBS 584 Letters, 330, 53-56. 585

Yang, S.-Y., Hong, C.-O., Lee, G. P., Kim, C.-T., & Lee, K.-W. (2013). The 586 hepatoprotection of caffeic acid and rosmarinic acid, major compounds of 587 Perilla frutescens, against t-BHP-induced oxidative liver damage. Food and 588 Chemical Toxicology, 55, 92-99. 589

24

Figure legends 590

Figure 1. Chemical structures of fifteen sesquiterpenoids tested: trans,trans-farnesol, 591

cis-nerolidol, (-)-α-bisabolol, trans-β-farnesene, germacrene D, α-humulene, β-592

caryophyllene, isocaryophyllene, (+)-valencene, guaiazulene, (-)-α-cedrene, (+)-593

aromadendrene, (-)-α-neoclovene, (-)-α-copaene and (+)-cyclosativene. 594

595

Figure 2. Experimental procedure for hepatoprotection evaluation in rat liver 596

homogenate. Standards tested: ascorbic acid, trans,trans-farnesol, cis-nerolidol), (-)-α-597

bisabolol, trans-β-farnesene, germacrene D, α-humulene, β-caryophyllene, 598

isocaryophyllene, (+)-valencene, guaiazulene, (-)-α-cedrene, (+)-aromadendrene, (-)-α-599

neoclovene, (-)-α-copaene and (+)-cyclosativene. 600

601

Figure 3. Relative hepatoprotection activity of ascorbic acid and sesquiterpenoids 602

(1mM) in endogenous and induced tert-butyl hydroperoxide (1mM) assays. Mean 603

activity values with standard errors of mean (SEM) from four independent experiments 604

performed in triplicate in different days are plotted. 605

606

Figure 4. Scatter plots of observed vs. calculated hepatoprotection for assay A (A) and 607

assay B (B). Results are shown for the cross-validation data. MLR models A and B 608

were calculated using optimized descriptor subsets (Table 1S). 609

610

611

612

613

614

25

Linear alcohols Cyclic alcohol

trans,trans-farnesol cis-nerolidol (-)-α-bisabolol

Linear hydrocarbon Cyclic hydrocarbons

trans-β-farnesene germacrene D α-humulene

Bicyclic hydrocarbons

β-caryophyllene (-)-isocaryophyllene (+)-valencene Guaiazulene

Tricyclic hydrocarbons

(-)-α-cedrene (+)-aromadendrene (-)-α-neoclovene (-)-α-copaene

Tetracyclic hydrocarbons

(+)-cyclosativene 615

26

Liver homogenate

400μL

Assay A 40μL of each compound

(1mM), 2h incubation

Assay B40μL of each compound

(1mM), 1h incubatiion

tert-BuOOH

(1mM), 1h Incubation 616

27

-10

0

10

20

30

40

50

60

70

80

MD

A r

edu

ctio

n (

%)

endogenous

tBuOOH induction

endogenous (Assay A)

induced tert-BuOOH (Assay B)

617 618

619

620

28

(+)-valencene

α-neoclovene

β-caryophyllene

α-humulene

guaiazulene

trans,trans-farnesol

trans-β-farnesene

cis-nerolidol

(-)-α-bisabolol

(-)-α-copaene

(+)-cyclosativeneisocaryophyllene

germacrene D

(-)-cedrene

(+)-aromadendrene

trans,trans-farnesol

guaiazulene

trans-β-farnesene

(-)-α-bisabolol(+)-valencene

α-neoclovene

(-)-α-copaeneisocaryophyllene

cis-nerolidol

(+)-aromadendrene(+)-cyclosativene

(-)-cedrene

β-caryophyllene

α-humulene

germacrene D

y= 0.9999x + 0.0059R2=0.9999

y= 1.0086x - 0.2385R2=0.9929

621 622

623

29

Table 1 624

625

Subsets of 3D molecular descriptors, selected by genetic algorithm, of the QSAR 626

regression models reported in this study 627

Descriptor* Meaning Model

GETAWAY HATS6v leverage-weighted autocorrelation of lag 6 / weighted by atomic van der Waals volumes A HATS2v leverage-weighted autocorrelation of lag 2 / weighted by atomic van der Waals volumes A and B R1u+ R maximal autocorrelation of lag 1 / unweighted B R3u+ R maximal autocorrelation of lag 3 / unweighted B R4u+ R maximal autocorrelation of lag 4 / unweighted A R6u+ R maximal autocorrelation of lag 6 / unweighted A R7u+ R maximal autocorrelation of lag 7 / unweighted B R8u+ R maximal autocorrelation of lag 8 / unweighted B R1m+ R maximal autocorrelation of lag 1 / weighted by atomic masses B R3m+ R maximal autocorrelation of lag 3 / weighted by atomic masses A R6m+ R maximal autocorrelation of lag 6 / weighted by atomic masses B R4e+ R maximal autocorrelation of lag 4 / weighted by atomic Sanderson electronegativities A and B R5e+ R maximal autocorrelation of lag 5 / weighted by atomic Sanderson electronegativities B R4p+ R maximal autocorrelation of lag 4 / weighted by atomic polarizabilities A WHIN

G2e 2st component symmetry directional WHIM index / weighted by atomic Sanderson electronegativities

B

Gm total symmetry index / weighted by atomic masses A 3D-MORSE Mor02m 3D-MoRSE - signal 02 / weighted by atomic masses B Mor24p 3D-MoRSE - signal 24 / weighted by atomic polarizabilities B

*Calculated using Dragon program evaluation version 5.4. 628

629

630

30

Highlights 631 632 • The hepatoprotection of fifteen sesquiterpenoids was evaluated using Rat 633

hepatocytes. 634

• Malonaldehyde level was reduced up to 35% (endogenous) and 70% (induced 635 assays). 636

• 3D-QSAR models were developed relating compounds activity with molecular 637 properties. 638

• Models showed good fit (Radj2 >0.819) and prediction power (Q2 >0.950). 639

• A network of sesquiterpenoids structural features modulates the hepatoprotective 640 activity. 641

642