Augmented Multivariate Image Analysis Applied to Quantitative

Subscriber access provided by EMORY UNIV

Article

aug-MIA-QSAR modeling of the phytotoxicities of benzoxazinone herbicides and related compounds on problematic weeds Mirlaine R. Freitas, Stella VBG Matias, Renato Luiz Grisi Macedo, Matheus Puggina de Freitas, and Nelson Venturin J. Agric. Food Chem., Just Accepted Manuscript • DOI: 10.1021/jf4024257 • Publication Date (Web): 15 Aug 2013 Downloaded from http://pubs.acs.org on August 20, 2013

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Journal of Agricultural and Food Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 18

Journal of Agricultural and Food Chemistry

1

aug-MIA-QSAR modeling of the phytotoxicities of benzoxazinone

2

herbicides and related compounds on problematic weeds

3

Mirlaine R. Freitas,1* Stella V. B. G. Matias,1 Renato L. G. Macedo,1 Matheus P.

4

Freitas,2 Nelson Venturin1

5

1

Department of Forest Sciences, Federal University of Lavras, 37200-000, Lavras, MG, Brazil

6

2

Department of Chemistry, Federal University of Lavras, 37200-000, Lavras, MG, Brazil

7 8

* e-mail: [email protected]

9 10

Abstract

11

Two of major weeds affecting cereal crops worldwide are Avena fatua L. (wild oat) and

12

Lolium rigidum Gaud. (rigid ryegrass). Thus, development of new herbicides against

13

these weeds is required; in line with this, benzoxazinones, their degradation products

14

and analogues have shown to be important allelochemicals and natural herbicides.

15

Despite earlier structure-activity studies have demonstrated that hydrophobicity (logP)

16

of aminophenoxazines correlates to phytotoxicity, our findings for a series of

17

benzoxazinone derivatives do not show any relationship between phytotoxicity and

18

logP, nor with other two usual molecular descriptors. On the other hand, a quantitative

19

structure-activity relationship (QSAR) analysis based on molecular graphs representing

20

structural shape, atomic sizes and colors to encode other atomic properties performed

21

very accurately for the prediction of phytotoxicities of these compounds against wild

22

oat and rigid ryegrass. Therefore, these QSAR models can be used to estimate the

23

phytotoxicity of new congeners of benzoxazinone herbicides toward A. fatua L. and L.

24

rigidum Gaud.

25

Keywords: QSAR; herbicides; phytotoxicity; Avena fatua L.; Lolium rigidum Gaud.

26

1

ACS Paragon Plus Environment


Page 2 of 18

27

Introduction

28

Despite the use of oats for thousands as food source for humans and livestock,1 the wild

29

oat Avena fatua L. is a major weed in oat farming; likewise, Lolium rigidum Gaud.

30

(rigid ryegrass) affects cereal crops worldwide. The weed control is commonly

31

performed using herbicides, but different cases of resistance have appeared.2-7

32

Accordingly, development of new herbicides is required. Benzoxazinones containing

33

the hydroxamic moiety have gained widespread use in phytochemistry, as well as their

34

degradation products;8 therefore, development of new derivatives of this family of

35

natural allelochemicals present in corn, wheat and rye is of interest, which can be

36

achieved using quantitative structure-activity relationships (QSAR) techniques.

37

The physicochemical and biological properties of herbicides can be related with

38

their hydrophobicity, e.g. the soil sorption of a variety of herbicide families has shown

39

to be linearly correlated with the octanol/water partition coefficient (logP).9 In addition,

40

other molecular descriptors, like molecular weight and volume, determine transport

41

characteristics of molecules.10 However, properties like these do not always correlate

42

significantly with biological properties, e.g. in the QSAR modeling of antifungal

43

activities of some benzothiazole derivatives.11 Thus, more representative descriptors are

44

usually invoked to generate predictive QSAR models.

45

Three-dimensional descriptors are frequently calculated to provide useful

46

quantitative structure-activity relationships,12-14 despite the need for exhaustive data

47

manipulation, such as conformational screening, geometry optimization and three-

48

dimensional alignment of molecules. Otherwise, a method based on 2D molecular

49

representations (chemical structure images), namely MIA-QSAR (multivariate image

50

analysis applied to quantitative structure-activity relationships),15 can provide predictive 2


Page 3 of 18


51

QSAR models. Recently, this method was improved to aug-MIA-QSAR,16 in which

52

new dimensions were introduced to account for atomic size and other properties. In this

53

way, physical, chemical and biological properties of molecules can be appropriately

54

explained by more complex information than hydrophobicity, like 2D molecular shape,

55

atomic sizes and colors to encode other atomic properties.

56

Accordingly, aug-MIA descriptors were used in this study to generate

57

quantitative relationships between the chemical structures of a series of benzoxazinones,

58

their degradation products and analogues with the respective phytotoxicities toward A.

59

fatua L. and L. rigidum Gaud., expressed in terms of percent of root length compared to

60

control. These biological data were obtained in the literature,8 where a reasonable

61

correlation between the phytotoxicities of five aminophenoxazines and logP was found,

62

but it cannot be extended to the whole series of compounds analyzed.

63 64

Materials and Methods

65

A series of 21 benzoxazinones, their degradation products and analogues, together with

66

the corresponding approximate phytotoxicity data toward A. fatua L. and L. rigidum

67

Gaud. (expressed in terms of percent of root length relative to control, exposed to 10-3

68

M of herbicide in DMSO/buffer solution) were obtained from the literature,8 and are

69

given in Table 1. The values of logP, molecular weight and molecular volume were

70

calculated using the Molinspiration program (www.molinspiration.com).

71

The aug-MIA-QSAR model was built according to the procedure described

72

elsewhere;16 thus, only a brief description is given here. A basic framework (a chemical

73

structure) was drawn in the GaussView 5.0 program;17 the molecules were generated by

74

consecutive replacement of substituents, whose atom sizes were proportional to the 3



Page 4 of 18

75

corresponding Van der Waals radii. Images were saved individually as bitmaps in a well

76

defined workspace (of 335×218 pixels size) in the Paint application of the Microsoft

77

Windows. 2D alignment was performed by making the common scaffold of the whole

78

series congruent; the superposed chemical structures used in the aug-MIA-QSAR are

79

shown in Figure 1 to illustrate the data variance. Each image (a combination of pixels)

80

was transformed in numerical values according to the RGB system of colors using the

81

Chemoface program18 and then grouped to give a three-way array of 21×335×218

82

dimension. This array was unfolded to a matrix of 21×73030 dimension, then reduced to

83

21×8269 dimension after removing columns with zero variance. This matrix was

84

regressed against the phytotoxicity data using partial least squares (PLS), giving the

85

calibration model, whose quality was evaluated by analyzing the RMSEC (root mean

86

square error of calibration) and r2, defined as 1 - [(Σ(yi-ŷi)2/Σ(yi-ӯ)2], in which yi

87

corresponds to the experimental phytotoxicity values (in % of root length relative to

88

control), ŷi are the predicted values, and ӯ corresponds to the mean values. The

89

calibration model was validated using leave-one-out cross-validation (LOOCV,

90

statistically evaluated using RMSECV and q2, defined similarly as above) and by a Y-

91

randomization test (mean of 10 repetitions), statistically evaluated using r2

92

additional statistical parameter proposed by Mitra et al.,19 cr2P (Eq. 1), was used to give

93

insight about the statistical difference between r2 and r2

94

acceptable). Values of r2 ≥ 0.8 and q2 ≥ 0.5 are widely recognized as acceptable and

95

indicate that predictive models are then achieved.

96

Y-rand

c 2

r P = r (r2 - r2rand)1/2

Y-rand.

An

(values above 0.5 are

(1)

97 98 4


Page 5 of 18


99

Results and Discussion

100

Hydrophobicity, expressed in terms of the octanol/water partition coefficient (logP), has

101

been found to correlate with biological properties for a long time20 and, for five

102

aminophenoxazines, a good correlation was found with phytotoxicity toward L. rigidum

103

Gaud.8 However, a general extension to the 21 compounds analyzed in this work cannot

104

be done using calculated values of logP, since no correlation with the percent of root

105

length relative to the control was found (Table 2). In addition, there is not any

106

relationship between the weed development with the molecular weight (MW) and

107

volume (MV) of the 21 titled herbicides, neither with all three parameters (logP, MW

108

and MV) together using multiple linear regression. Therefore, the action mechanism of

109

the herbicides is more complex than those based on solubility and transportation

110

through cell membranes. A better relationship with molecular structural changes and,

111

consequently, with specific ligand interactions at the biomolecular level is therefore

112

expected. Thus, a structure-based design accounting for molecular shape and atomic

113

properties was developed using aug-MIA-QSAR to achieve models for the

114

phytotoxicity estimation of congeners of benzoxazinones, their degradation products

115

and analogues.

116

The aug-MIA descriptors for the 21 compounds of Table 1 were submitted to

117

PLS regression against the phytotoxicity data toward A. fatua L. and L. rigidum Gaud.,

118

and model calibrations using 4 PLS components (in which the RMSECV values were

119

minimized) were achieved (Figure 2), giving acceptable statistical results (Table 3),

120

especially for the L. rigidum Gaud. model. The values of r2 above 0.8 do not guarantee

121

that the QSAR models are reliable for the prediction of biological data; thus, model

122

validation was performed using leave-one-out cross-validation (LOOCV). It is worth 5



Page 6 of 18

123

mentioning that external validation was not performed because of the limited amount of

124

samples (only 21) and, therefore, the splitting of the data set into training and test

125

samples would give few compounds, with risk of lack of representativeness in the

126

training set. Despite the high residual for compound 18 in the leave-one-out cross-

127

validation for A. fatua L., outliers were not found using sample leverages and

128

studentized residuals as outlier diagnostic probes; thus, this compound was considered

129

important for the model. The high residual for 18 in the LOOCV is probably due to the

130

R1 substituent, which is the only OH group along with the series in this position. The

131

LOOCV results (q2 > 0.50) attest the good prediction performance of the models, but a

132

Y-randomization test was also performed to guarantee that calibration results were good

133

due to the actual relationship between descriptors and variables rather than by chance

134

correlation or over-fitting. According to this approach, the phytotoxicities block is

135

randomized, while the descriptor matrix is kept intact; bad correlation using PLS is

136

expected if aug-MIA descriptors indeed encode the corresponding biological data, but

137

not the randomized data. The low mean values of r2Y-rand compared to r2 (statistically

138

analyzed using cr2P) indicate that the real calibration is robust.

139

Overall, models for the prediction of herbicide potency of new derivatives of the

140

titled compounds were built and 2D view of chemical structures describes the biological

141

data accurately. The aug-MIA descriptors can also be used to generate pattern

142

recognition models using principal component analysis (PCA), which give insight about

143

chemical properties responsible for the activity profiles. The phytotoxicity levels were

144

divided in three classes: high (-70 to -92%), moderate (-30 to -69%) and low (5 to -

145

29%). PC1 explained the largest data variance (95.4%), but PC2 (1.6%) was capable of

146

clustering compounds with low activities (negative scores in PC2) and with

147

moderate/high activities (positive scores in PC2). Only few samples of a given class 6


Page 7 of 18


148

have fallen within the cluster of a different class (compounds 6 and 7) or have formed a

149

different cluster (compounds 1, 16 and 17), for both weeds (Figure 3); thus, the PCA

150

models were generally good to separate classes of compounds with different

151

phytotoxicity levels toward A. fatua L. and L. rigidum Gaud. The scores analysis in

152

PCA reveals that malonamic acids and most of aminophenoxazines are not promising,

153

potent herbicides, while derivatives based on the benzoxazinone and benzoxazolinone

154

substrutures are potential compounds to be used in the control of weeds A. fatua L. and

155

L. rigidum Gaud.

156 157

Conclusions

158

Hydrophobicity, molecular weight and volume are not enough to describe completely

159

the phytotoxicity data of benzoxazinone herbicides and derivatives toward A. fatua L.

160

and L. rigidum Gaud. aug-MIA descriptors encoding molecular shape and different

161

atomic properties are better to explain the bioactivity behavior of these compounds,

162

indicating that the action mechanism is not only proportional to cell permeation and

163

transportation, but also to specific ligand-receptor interactions. Development of new

164

allelochemical derivatives must be driven by the substructures of benzoxazinones and

165

benzoxazolinones rather than malonamic acids and aminophenoxazines.

166 167

Acknowledgements

168

Authors are thankful to FAPEMIG and CNPq for the financial support, studentship (to

169

M.R.F. and S.V.B.G.M.) and fellowships (to R.L.G.M., M.P.F. and N.V.).

170

7



Page 8 of 18

171

References

172

(1) Watson, L.; Dallwitz, M. J. The grass genera of the world; CAB International:

173

Wallingford, 1994.

174

(2) Andrews, T. S.; Morrison, N.; Penner, A. Monitoring the spread of ACCase

175

inhibitor resistance among wild oat (Avena fatua) patches using AFLP analysis.

176

Weed Sci. 1998, 46, 196-199.

177

(3) Bourgeois, L.; Kenkel, C.; Morrison, N. Characterization of cross-resistance patterns

178

in acetyl-CoA carboxylase inhibitor resistant wild oat (Avena fatua). Weed Sci. 1997,

179

45, 750-755.

180

(4) Colwill, D. F.; Hawkes, R.; Williams, H.; Warner, A. J.; Sutton, B.; Powles, B.;

181

Preston, C. Resistance to glyphosate in Lolium rigidum. Pest. Sci. 1999, 55, 489-491.

182

(5) Gill, G. S. Development of herbicide resistance in annual ryegrass populations

183

(Lolium rigidum Gaud.) in the cropping belt of Western Australia. Aust. J. Exp.

184

Agric. 1995, 35, 67-72.

185 186

(6) De Prado, R.; De Prado, L.; Menendez, J. Resistance to substituted urea herbicides in Lolium rigidum biotypes. Pest. Biochem. Physiol. 1997, 57, 126-136.

187

(7) Seefeldt, S. S. D.; Hoffman, L. D.; Gealy, R.; Fuerst, P. Inheritance of diclofop

188

resistance in wild oat (Avena fatua L.) biotypes from the Willamette Valley of

189

Oregon. Weed Sci. 1998, 46, 170-175.

190

(8) Macías, F. A.; Marín, D.; Oliveros-Bastidas, A.; Castellano, D.; Simonet, A. M.;

191

Molinillo, J. M. G. Structure-activity relationship (SAR) studies of benzoxazinones,

192

their degradation products, and analogues. Phytotoxicity on problematic weeds

8


Page 9 of 18


193

Avena fatua L. and Lolium rigidum Gaud. J. Agric. Food Chem. 2006, 54, 1040-

194

1048.

195

(9) Sabljić, A.; Güsten, H.; Verhaar, H.; Hermens, J. QSAR modelling of soil sorption.

196

Improvements and systematics of logKOC vs. logKOW correlations. Chemosphere

197

1995, 31, 4489-4514.

198

(10) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J. Experimental and

199

computational approaches to estimate solubility and permeability in drug discovery

200

and development settings. Adv. Drug Deliv. Rev. 1997, 23, 4-25.

201

(11) Bitencourt, M.; Freitas, M. P. Bi- and multilinear PLS coupled to MIA-QSAR in

202

the prediction of antifungal activities of some benzothiazole derivatives. Med. Chem.

203

2009, 5, 79-86.

204

(12) Cramer, R. D.; Patterson, D. E.; Bunce, J. D. Comparative molecular field analysis

205

(CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J. Am. Chem.

206

Soc. 1988, 110, 5959-5967.

207

(13) Klebe, G.; Abraham, U.; Mietzner, T. Molecular similarity indices in a comparative

208

analysis (CoMSIA) of drug molecules to correlate and predict their biological

209

acivity. J. Med. Chem. 1994, 37, 4130-4146.

210

(14) Hopfinger, A. J.; Wang, S.; Tokarski, J. S.; Jin, B.; Albuquerque, M.; Madhav, P.

211

J.; Duraiswami, C. Construction of 3D-QSAR models using the 4D-QSAR analysis

212

formalism. J. Am. Chem. Soc. 1997, 119, 10509-10524.

213

(15) Freitas, M. P.; Brown, S. D.; Martins, J. A. MIA-QSAR: A simple 2D image-based

214

approach for quantitative structure-activity relationship analysis. J. Mol. Struct.

215

2005, 738, 149-154. 9



216 217 218 219

Page 10 of 18

(16) Nunes, C. A.; Freitas, M. P. Introducing new dimensions in MIA-QSAR: a case for chemokine receptor inhibitors. Eur. J. Med. Chem. 2013, 62, 297-300. (17) Dennington II, R. D.; Keith, T. A.; Millam, J. M. GaussView 5.0.8, Gaussian, Inc.: Wallingford, 2008.

220

(18) Nunes, C. A.; Freitas, M. P.; Pinheiro, A. C. M.; Bastos, S. C. Chemoface: A novel

221

free user-friendly interface for chemometrics. J. Braz. Chem. Soc. 2012, 23, 2003-

222

2010.

223

(19) Mitra, I.; Saha, A.; Roy, K. Exploring quantitative structure–activity relationship

224

studies of antioxidant phenolic compounds obtained from traditional Chinese

225

medicinal plants. Mol. Simul. 2010, 36,1067-1079.

226 227

(20) Hansch, C.; Fujita, T. ρ-σ-π Analysis. A method for the correlation of biological activity and chemical structure. J. Am. Chem. Soc. 1964, 86, 1616-1626.

228 229 230

10


Page 11 of 18


231

Figures

232 233 234

Figure 1. Superposed structures of the 21 benzoxazinones, their degradation products

235

and analogues, used to generate the aug-MIA descriptors for the QSAR modeling.

236 237

Figure 2. Plots of experimental vs. fitted and predicted phytotoxicity, expressed in

238

terms of root length relative to the control, obtained by the aug-MIA-QSAR models for

239

A. fatua L. and L. rigidum Gaud.

240

Figure 3. Score plots of the principal component analysis of benzoxazinones, their

241

degradation products and analogues, using aug-MIA descriptors and classified

242

according to phytotoxicity levels toward A. fatua L. and L. rigidum Gaud.

243 244

11



245

Page 12 of 18

Tables

246 247

Table 1. Series of compounds analyzed using aug-MIA-QSAR and the corresponding

248

phytotoxicities (% of root length relative to the control) toward A. fatua L. and L.

249

rigidum Gaud. R1

O

R2

R1

R1

O

O

O

N

NHR2

O N

N H

O

R1

8888 1111 ---4444 1111

3333 1111 dddd nnnn aaaa 2222 1111

1111 1111 ---1111 R3

OH

OH O

O NH2

OH

1111 2222

0000 2222 dddd nnnn aaaa 9999 1111

250

Compound 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

R1 H H OCH3 H OCH3 H OCH3 H OCH3 H OCH3 H OCH3 H OCH3 H OCH3 OH H OCH3

R2 O-β-D-glucose OH OH OH OH H H H H H H

H H OAc OAc H

R3 OH OH OH H H H H OH OH OAc OAc

%A. fatua L. -87 -87 -77 -83 -87 -80 -55 -87 -32 -52 -65 -28 -20 -23 -17 -80 -43 -22 5 -22 -80

%L. rigidum Gaud. -87 -92 -72 -85 -80 -65 -20 -90 -28 -60 -60 -48 -32 -15 -13 -92 -50 -18 -13 2 -83

251 12


Page 13 of 18


252

Table 2. Percent of root length of weeds exposed to compounds 1-21 relative to control

253

and calculated molecular descriptors.a Cpd 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

254

a

A. fatua L. % logP -87 -1.451 -87 0.256 -77 0.289 -83 0.328 -87 0.361 -80 0.986 -55 1.019 -87 0.914 -32 0.947 -52 1.278 -65 1.310 -28 1.176 -20 1.209 -23 2.051 -17 2.083 -80 2.471 -43 2.504 -22 1.547 5 0.365 -22 0.398 -80 1.153

MW 343.3 181.1 211.2 165.1 195.2 149.1 179.2 165.1 195.2 207.2 237.2 135.1 165.1 212.2 242.2 270.2 300.3 228.2 195.2 225.2 109.1

MV 278.9 146.8 172.4 138.4 164.0 130.4 155.9 138.8 164.3 175.3 200.8 113.6 139.1 179.7 205.2 225.4 250.9 187.7 167.2 192.8 103.4

L. rigidum Gaud. % logP -87 -1.451 -92 0.256 -72 0.289 -85 0.328 -80 0.361 -65 0.986 -20 1.019 -90 0.914 -28 0.947 -60 1.278 -60 1.310 -48 1.176 -32 1.209 -15 2.051 -13 2.083 -92 2.471 -50 2.504 -18 1.547 -13 0.365 2 0.398 -83 1.153

MW 343.3 181.1 211.2 165.1 195.2 149.1 179.2 165.1 195.2 207.2 237.2 135.1 165.1 212.2 242.2 270.2 300.3 228.2 195.2 225.2 109.1

MV 278.9 146.8 172.4 138.4 164.0 130.4 155.9 138.8 164.3 175.3 200.8 113.6 139.1 179.7 205.2 225.4 250.9 187.7 167.2 192.8 103.4

logP = octanol/water partition coefficient; MW = molecular weight; MV = molecular volume.

255 256

13



257

Page 14 of 18

Table 3. Statistical parameters of the aug-MIA-QSAR models. Parameter PLS components r2 RMSEC q2 RMSECV r2Y-rand c 2 rP

A. fatua L. 4 0.828 12.21 0.504 21.37 0.419 0.582

L. rigidum Gaud. 4 0.858 11.60 0.657 18.15 0.535 0.526

258 259

14


Page 15 of 18


260

261 262

Figure 1

263

15



Page 16 of 18

264

0

20 Calibration LOO cross-validation

Fitted and predicted phytotoxicity (% development relative to control)

Fitted and predicted phytotoxicity (% development relative to control)

20

-20 -40 -60 -80 A. fatua L. -100 -100

-80

-60

-40

-20

0

20

Experimental phytotoxicity (% development relative to control)

0

Calibration LOO cross-validation

-20 -40 -60 -80 L. rigidum Gaud. -100 -100

-80

-60

-40

-20

0

20

Experimental phytotoxicity (% development relative to control)

265 266

Figure 2

267

16


Page 17 of 18


268

269 270

Figure 3

271 272

17



172x131mm (120 x 120 DPI)


Page 18 of 18

Augmented Multivariate Image Analysis Applied to Quantitative

Recommend Documents