Bull Environ Contam Toxicol (2014) 92:143–147 DOI 10.1007/s00128-013-1184-3
Three-Parameter Modeling of the Soil Sorption of Acetanilide and Triazine Herbicide Derivatives Mirlaine R. Freitas • Stella V. B. G. Matias • Renato L. G. Macedo • Matheus P. Freitas • Nelson Venturin
Received: 18 July 2013 / Accepted: 18 December 2013 / Published online: 28 December 2013 Ó Springer Science+Business Media New York 2013
Abstract Herbicides have widely variable toxicity and many of them are persistent soil contaminants. Acetanilide and triazine family of herbicides have widespread use, but increasing interest for the development of new herbicides has been rising to increase their effectiveness and to diminish environmental hazard. The environmental risk of new herbicides can be accessed by estimating their soil sorption (logKoc), which is usually correlated to the octanol/water partition coefficient (logKow). However, earlier findings have shown that this correlation is not valid for some acetanilide and triazine herbicides. Thus, easily accessible quantitative structure–property relationship models are required to predict logKoc of analogues of the these compounds. Octanol/water partition coefficient, molecular weight and volume were calculated and then regressed against logKoc for two series of acetanilide and triazine herbicides using multiple linear regression, resulting in predictive and validated models. Keywords Environmental risk Herbicides QSPR modeling Soil sorption
Persistent organic pollutants (POPs) are chemicals that persist for very long periods of time in the environment and consequently may accumulate to a high level in the food M. R. Freitas (&) S. V. B. G. Matias R. L. G. Macedo N. Venturin Department of Forest Sciences, Federal University of Lavras, Lavras, MG 37200-000, Brazil e-mail: [email protected] M. P. Freitas Department of Chemistry, Federal University of Lavras, Lavras, MG 37200-000, Brazil
chain, causing toxic effects like problems in reproduction, development and immunological functions (Corsonlini et al. 2005; Domingo 2004; Giesy et al. 1994; Kavlock et al. 1996; Kelce et al. 1995; Ratcliffe 1967, 1970). POPs are often delivered in water and soil, while herbicides (a specific type of POP) are an important source of contaminants in soil. For example, the triazine herbicide Atrazine was banned in the European Union in 2004 because of its persistent groundwater contamination (Ackerman 2007). The soil/water partition coefficient normalized to organic carbon (Koc) is an important property for accessing fate and persistence of POPs. Therefore, it is usually correlated to Kow, the octanol/water partition coefficient, which can be experimentally determined or estimated using calculation models based on libraries of compounds. Indeed, a variety of compounds has been analyzed by comparing logKoc with logKow, but also with other parameters (Schu¨u¨rmann et al. 2006). However, some models based on Kow fail in describing the soil sorption of some classes of herbicides, like acetanilides and triazines (Fig. 1). According to Sabljic´ et al. (1995), the linear regression between logKow and logKoc gave very poor correlations, with r2 B 0.5, while the corresponding values for other classes of herbicides were stronger (e.g. phosphates, alcohols, anilines and phenols). In quantitative structure–property relationship (QSPR), molecular properties are used to correlate chemical structures with the corresponding physical, chemical and/or biological properties, including soil sorption. In this case, the property is assumed to be dependent on the molecular structure and atomic composition. Some three-dimensional QSPR methods have been developed to generate descriptors that correlate with molecular properties, like CoMFA (Cramer et al. 1988) and CoMSIA (Klebe et al. 1994). However, Estrada et al. (2001), and Brown and Martin
123
144
Bull Environ Contam Toxicol (2014) 92:143–147
O R1
N
Table 1 Series of acetanilide (1–21) and triazine (22–36) herbicides, experimental logKoc obtained from the literature (Sabljic´ et al. 1995) and calculated molecular descriptors
R1 CH3
N R2
R2
N R3
acetanilide-type herbicides
N N
N
R4
R5
1,3,5-triazine-type herbicides
Fig. 1 Acetanilide and triazine basic structures used in the QSPR modeling
(1997) pointed out that 2D molecular structural descriptors, possessing high reappearance, are not inferior to 3D ones, at the least in many practical cases. In addition, images representing molecular structures have also been shown to correlate with biological properties of herbicides (Freitas et al. 2013). Thus, QSPR strategies can be useful for the development of predictive models to estimate herbicide properties. Acetanilide herbicides comprise compounds including the widely used Alachlor; sorption in soils and sediments is an important factor controlling the migration and bioavailability of these herbicides, while microbial degradation is the most important factor in determining their overall fate in the environment (Stamper and Tuovinen 1998). Triazine herbicides are generally of low acute toxicity for birds and mammals, although certain species show unexpected vulnerability for some compounds (Prosen 2012). Thus, this work aims at using few molecular descriptors to generate a QSPR model with improved prediction performance if compared to existing uniparametric models, in order to estimate soil sorption of a series of acetanilide and triazine herbicides. In this way, development of novel herbicides of these families can be focused not only on the efficacy for the control of unwanted plant, but also considering the environmental hazard of a designed compound.
Materials and Methods The sets of 21 acetanilide (1–21) and 15 triazine (22–36) herbicides, as well as the corresponding soil sorption (logKoc) values (Table 1) were obtained from the literature (Sabljic´ et al. 1995; Gerstl 1990), where the soil conditions and the method for data collection and treatment are described. The values of octanol/water partition coefficient (logKow), molecular weight (MW) and molecular volume (MV) were calculated using the property prediction module of the Molinspiration freeware program (www.
123
#
Compound
logKoc
logKow
MW
MV
1
Acetanilide
1.43
1.156
135.2
132.0
2
2-Cl-acetanilide
1.58
1.786
169.6
145.5
3 4
3-CH3-acetanilide 3-F-acetanilide
1.45 1.57
1.581 1.296
149.2 153.2
148.6 136.9
5
3-Cl-acetanilide
1.86
1.810
169.6
145.5
6
3-Br-acetanilide
2.01
1.941
214.1
149.9
7
3-CF3-acetanilide
1.75
2.028
203.2
163.3
8
3-NO2-acetanilide
1.94
1.091
180.2
155.3
9
4-F-acetanilide
1.48
1.320
153.2
136.9
10
4-Br-acetanilide
1.95
1.965
214.1
149.9
11
4-OCH3-acetanilide
1.40
1.213
165.2
157.5
12
Butyranilide
1.71
2.550
163.2
165.6
13
Propachlor
2.42
2.639
211.7
196.1
14
3,4-diCl-acetanilide
2.34
2.440
204.0
159.1
15
3-Cl-4-OCH3-acetanilide
1.95
1.819
199.6
171.1
16
Alachlor
2.28
3.671
269.8
255.2
17
Butachlor
2.86
5.109
311.8
305.6
18 19
Norfluorazon Acetchlor
3.28 2.32
2.886 3.580
303.7 269.8
229.4 255.2
20
Metholachlor
2.46
3.548
283.8
271.8
21
Matalaxyl
1.57
2.339
279.3
269.2
22
Simazine
2.10
2.251
201.7
176.6
23
Propazine
2.40
2.845
229.7
209.8
24
Ametryn
2.59
2.728
213.3
197.8
25
Terbutryn
2.85
3.537
241.4
230.6
26
Prometron
2.60
3.033
225.3
221.8
27
Atrazine
2.24
2.548
215.7
193.2
28
Ipazine
2.91
3.916
257.8
243.4
29
Trietazin
2.76
3.619
243.7
226.8
30
Dipropetryn
3.07
3.698
255.4
247.8
31
Terbuthylazine
2.32
3.061
229.7
209.5
32
Prometryn
2.85
3.322
241.4
231.0
33
Metribuzin
1.71
1.460
214.3
192.2
34 35
Cyanazine Sec-Bumeton
2.28 2.78
2.470 3.272
240.7 225.3
209.8 222.0
36
Metamitron
2.17
0.357
202.2
179.3
logarithm of the octanol/water partition coefficient—logKow; molecular weight—MW; molecular volume—MV
molinspiration.com) and then regressed against the experimental values of logKoc using multiple linear regression (MLR). The statistical results of calibration were analyzed on the basis of RMSEC (root mean square error of calibration) h i and r2, defined as 1 ðRðyi y^i Þ2 =Rðyi yÞ2 , in which yi corresponds to the experimental logKoc values, yˆi are the
Bull Environ Contam Toxicol (2014) 92:143–147
145
Results and Discussion The soil sorption of organic compounds is expected to be correlated with logKow, because sediments and mainly organic matter in soil have been shown to interact strongly with progressively more hydrophobic compounds. However, logKow values for acetanilide and triazine herbicides were found elsewhere to be poorly correlated to logKoc (r2 of 0.491 and 0.273, respectively) (Sabljic´ et al. 1995). logKow values can be calculated using a variety of programs; for example, the calculated logKow obtained for the set of acetanilides and triazines using the Molinspiration program correlates significantly well with the corresponding experimental data (r2 of 0.766 and 0.873, respectively). However, the calculated logKow values do not fit appropriately with the experimental logKoc data, despite the higher coefficient of determination (r2 of 0.565 and 0.656, respectively) than those obtained elsewhere (Sabljic´ et al. 1995) using the following formulae (Eqs. 2 and 3): For acetanilides : log Koc ¼ 1:12 þ 0:40 log Kow ; r 2 ¼ 0:491; q2 ¼ 0:436
ð2Þ
For triazines : log Koc ¼ 1:50 þ 0:30 log Kow ; r 2 ¼ 0:273; q2 ¼ 0:045
ð3Þ
This corroborates that hydrophobicity does not completely explain the soil sorption profile of acetanilide and triazine herbicides. Therefore, molecular weight (MW) and volume (MV) were calculated to be added in the models. The linear three-parametric model for the acetanilide series based on logKow, MW and MV gave an improved correlation with
3.5
Fitted and predicted logK oc
predicted logKoc values, and y^i corresponds to the mean logKoc values. The model was validated through leave-oneout cross-validation (statistically evaluated using RMSECV and q2, defined similarly as above). QSPR studies in the literature indicate that r2 C 0.8 and q2 C 0.5 are satisfactory. In order to guarantee that good models were not obtained by chance correlations, the Y-block (the logKoc column vector) was randomized and a new regression was carried out with the intact X-matrix (mean of 10 repeti2 tions); low values of rYrand indicate that the real calibration is not a chance correlation. Accordingly, an additional statistical parameter proposed by Mitra et al. (2010), namely rP2 (Eq. 1), was calculated to give insight about the 2 statistical difference between r2 and rYrand (values above 0.5 are acceptable). These calculations were performed using the Chemoface program (www.ufla.br/chemoface) (Nunes et al. 2012). 1=2 2 rP2 ¼ r 2 r 2 rrand ð1Þ
Calibration Leave-one-out cross-validation
3.0
2.5
2.0
1.5
1.0 1.0
1.5
2.0
2.5
3.0
3.5
Experimental logKoc Fig. 2 Plot of experimental versus fitted and predicted logKoc for the series of acetanilide herbicides. Segmented line for calibration and solid line for LOO cross-validation
logKoc (Eq. 4), as illustrated in Fig. 2. The model is slightly improved after removing compound 18 (not an acetanilide in fact), with r2 = 0.795 and q2 = 0.633, but Eq. 4 comprising all 21 derivatives is sufficiently predictive to estimate logKoc of acetanilide-type herbicides with reasonable accuracy. The recognition of compound 18 as an outlier indicates that the QSPR model is robust and that it really encodes chemical information about the series of acetanilide herbicides. logKoc ¼ 0:75 þ 0:33 logKow þ 0:011 MW 0:010 MV
ð4Þ
N = 21, r2 = 0.788, RMSEC = 0.225, q2 = 0.554, 2 RMSECV = 0.334, rYrand = 0.141, rP2 = 0.616 Despite the apparent collinearity between MW and MV (r2 = 0.848), the absence of one of these molecular descriptors to give a two-parametric model does not provide highly predictive QSPR models, i.e. r2 of 0.661 (RMSEC = 0.285) and q2 of 0.424 (RMSECV = 0.382) when using logKow and MW, and r2 of 0.567 (RMSEC = 0.322) and q2 of 0.390 (RMSECV = 0.395) when using logKow and MV. Molecular weight and volume do not necessarily reflect hydrophobicity; for instance, the polar nitro substituent in 8 leads to a larger molecular weight and volume than the methyl derivative (3), but it is less hydrophobic (logKow = 1.091 for 8 and 1.581 for 3). Thus, the chemical information explained by the three descriptors is complementary, suggesting that logKow alone is not enough to explain accurately the sorption mechanism and mobility of acetanilides in the soil. Similar analysis was carried out for the set of 15 triazines (22-36). The three-parametric QSPR model is significantly better when compared to the monoparametric
123
Bull Environ Contam Toxicol (2014) 92:143–147 3.2
Calibration Leave-one-out cross-validation
3.0 2.7 2.4 2.1 1.8 1.5
a
1.2 1.2
1.5
1.8
2.1
2.4
2.7
3.0
Fitted and predicted logKoc
Fitted and predicted logKoc
146
Calibration Leave-one-out cross-validation
3.0 2.8 2.6 2.4 2.2
b 2.0 2.0
Experimental logK oc
2.2
2.4
2.6
2.8
3.0
3.2
Experimental logK oc
Fig. 3 Plot of experimental versus fitted and predicted logKoc for the series of triazine herbicides. a all triazine compounds included; b outliers 33 and 36 removed. Segmented line for calibration and solid line for LOO cross-validation
model based only on logKow and can be expressed by Eq. 5 and Fig. 3. Since the structural scaffold of compounds 33 and 36 corresponds to 1,2,4-triazinones rather than 1,3,5triazines of the remaining 13 compounds, they behave as outliers (as identified by Student’s residuals and leverage diagnostics). Therefore, the QSPR model for the 13 1,3,5triazines (33 and 36 removed) gave an r2 of 0.91, which is highly predictive and reliable for estimation of logKoc of different triazine analogues, as supported by the validation data (Eq. 6 ).
while molecular weight also plays a significant role in these models. Accordingly, norfluorazon, which contains the bulky substituents chlorine, ethyl and ethoxymethyl groups at the acetanilide framework, is the most persistent organic herbicide along with the series, whilst one of the least hydrophobic derivatives (11, containing only an additional methoxy group at acetanilide) interacts poorly with soil compared to other herbicides.
logKoc ¼ 0:68 þ 0:100 logKow 0:012 MW þ 0:021 MV
Conclusions ð5Þ
N = 15, r2 = 0.817, RMSEC = 0.154, q2 = 0.551, 2 RMSECV = 0.287, rYrand = 0.231, rP2 = 0.625 logKoc ¼ 0:79 þ 0:201 logKow 0:010 MW þ 0:017 MV
ð6Þ
N = 13, r 2 = 0.908, RMSEC = 0.088, q2 = 0.829, 2 = 0.353, rP2 = 0.676 RMSECV = 0.123, rYrand Similarly to described before for the acetanilide-type herbicides, two-parametric modeling using only logKow and MW, or logKow and MV, achieved very modest correlations with logKoc and poor statistical results of validation, i.e. r2 of 0.685 (RMSEC = 0.202) and q2 of 0.369 (RMSECV = 0.349) when using logKow and MW, and r2 of 0.776 (RMSEC = 0.171) and q2 of 0.501 (RMSECV = 0.305) when using logKow and MV. Overall, while logKow does not completely explain the soil sorption of acetanilide and triazine herbicides, introduction of two simple and easily accessible molecular descriptors in QSPR models provides enhanced predictive abilities for environmental risk assessment. Compounds with low sorption in soil are characterized by low logKow values compared to the average, and compounds with high soil sorption are characterized by high molecular volume,
123
The soil sorption of acetanilide and triazine herbicides can be accurately predicted using molecular volume, molecular weight and logKow values rather than only one of these molecular descriptors. A single model to predict logKoc and other properties of different organic classes of compounds remains to be investigated in the future. Acknowledgments Authors are thankful to FAPEMIG and CNPq for the financial support, studentship (to M.R.F. and S.V.B.G.M.) and fellowships (to R.L.G.M., M.P.F. and N.V.).
References Ackerman F (2007) The economics of atrazine. Int J Occup Environ Health 13:437–445 Brown RD, Martin YC (1997) The information content of 2D and 3D structural descriptors relevant to ligand-receptor binding. J Chem Inf Comp Sci 37:1–9 Corsonlini S, Ademollo N, Romeo T, Greco S, Focardi S (2005) Persistent organic pollutants in edible fish: a human and environmental health problem. Microchem J 79:115–123 Cramer RD, Patterson DE, Bunce JD (1988) Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc 110:5959–5967 Domingo JL (2004) Human exposure to polybrominated diphenyl ethers through the diet. J Chromatogr A 1054:321–326
Bull Environ Contam Toxicol (2014) 92:143–147 Estrada E, Molina E, Perdomo-Lo´pez I (2001) Can 3D structural parameters be predicted from 2D (topological) molecular descriptors? J Chem Inf Comp Sci 41:1015–1021 Freitas MR, Matias SVBG, Macedo RLG, Freitas MP, Venturin N (2013) Augmented multivariate image analysis applied to quantitative structure-activity relationship modeling of the phytotoxicities of benzoxazinone herbicides and related compounds on problematic weeds. J Agric Food Chem 61:8499–8503 Gerstl Z (1990) Estimation of organic chemical sorption by soils. J Cont Hydrol 6:357–375 Giesy JP, Ludwig JP, Tillitt DE (1994) Deformities in birds of the Great-Lakes region assigning causality. Environ Sci Technol 28:A128–A135 Kavlock RJ, Daston GP, Derosa C, Fenner-Crisp P, Gray LE, Kaattari S, Lucier G, Luster M, Mac MJ, Maczka C, Miller R, Moore J, Rolland R, Scott G, Sheehan DM, Sinks T, Tilson HA (1996) Research needs for the risk assessment of health and environmental effects of endocrine disruptors: a report of the US EPAsponsored workshop. Environ Health Persp 104:715–740 Kelce WR, Stone CR, Laws SC, Gray LE, Kemppainen JA, Wilson EM (1995) Persistent DDT metabolite P, P’-DDE is a potent androgen receptor antagonist. Nature 375:581–585 Klebe G, Abraham U, Mietzner T (1994) Molecular similarity indices in a comparative analysis (CoMSIA) of drug molecules to correlate and predict their biological activity. J Med Chem 37: 4130–4146
147 Mitra I, Saha A, Roy K (2010) Exploring quantitative structureactivity relationship studies of antioxidant phenolic compounds obtained from traditional Chinese medicinal plants. Mol Simul 36:1067–1079 Nunes CA, Freitas MP, Pinheiro ACM, Bastos SC (2012) Chemoface: a novel free user-friendly interface for chemometrics. J Braz Chem Soc 23:2003–2010 Prosen H (2012) Fate and determination of triazine herbicides in soil, herbicides: properties, synthesis and control of weeds, ed. by Hasaneen MN, InTech. Available from: http://www.intechopen. com/books/herbicides-properties-synthesis-and-control-of-weeds/ fateand-determination-of-triazine-herbicides-in-soil Ratcliffe DA (1967) Decrease in eggshell weight in certain birds of prey. Nature 215:208–210 Ratcliffe DA (1970) Changes attributable to pesticides in egg breakage frequency and eggshell thickness in some British birds. J Appl Ecol 7:67–115 Sabljic´ A, Gu¨sten H, Verhaar H, Hermens J (1995) QSAR modelling of soil sorption. Improvements and systematics of logKOC versus logKOW correlations. Chemosphere 31:4489–4514 Schu¨u¨rmann G, Ebert R-U, Ku¨hne R (2006) Prediction of the sorption of organic compounds into soil organic matter from molecular structure. Environ Sci Technol 40:7005–7011 Stamper DM, Tuovinen OH (1998) Biodegradation of the acetanilide herbicides alachlor, metolachlor, and propachlor. Crit Rev Microbiol 24:1–22
123
Three-Parameter Modeling of the Soil Sorption of Acetanilide and Triazine Herbicide Derivatives Mirlaine R. Freitas • Stella V. B. G. Matias • Renato L. G. Macedo • Matheus P. Freitas • Nelson Venturin
Received: 18 July 2013 / Accepted: 18 December 2013 / Published online: 28 December 2013 Ó Springer Science+Business Media New York 2013
Abstract Herbicides have widely variable toxicity and many of them are persistent soil contaminants. Acetanilide and triazine family of herbicides have widespread use, but increasing interest for the development of new herbicides has been rising to increase their effectiveness and to diminish environmental hazard. The environmental risk of new herbicides can be accessed by estimating their soil sorption (logKoc), which is usually correlated to the octanol/water partition coefficient (logKow). However, earlier findings have shown that this correlation is not valid for some acetanilide and triazine herbicides. Thus, easily accessible quantitative structure–property relationship models are required to predict logKoc of analogues of the these compounds. Octanol/water partition coefficient, molecular weight and volume were calculated and then regressed against logKoc for two series of acetanilide and triazine herbicides using multiple linear regression, resulting in predictive and validated models. Keywords Environmental risk Herbicides QSPR modeling Soil sorption
Persistent organic pollutants (POPs) are chemicals that persist for very long periods of time in the environment and consequently may accumulate to a high level in the food M. R. Freitas (&) S. V. B. G. Matias R. L. G. Macedo N. Venturin Department of Forest Sciences, Federal University of Lavras, Lavras, MG 37200-000, Brazil e-mail: [email protected] M. P. Freitas Department of Chemistry, Federal University of Lavras, Lavras, MG 37200-000, Brazil
chain, causing toxic effects like problems in reproduction, development and immunological functions (Corsonlini et al. 2005; Domingo 2004; Giesy et al. 1994; Kavlock et al. 1996; Kelce et al. 1995; Ratcliffe 1967, 1970). POPs are often delivered in water and soil, while herbicides (a specific type of POP) are an important source of contaminants in soil. For example, the triazine herbicide Atrazine was banned in the European Union in 2004 because of its persistent groundwater contamination (Ackerman 2007). The soil/water partition coefficient normalized to organic carbon (Koc) is an important property for accessing fate and persistence of POPs. Therefore, it is usually correlated to Kow, the octanol/water partition coefficient, which can be experimentally determined or estimated using calculation models based on libraries of compounds. Indeed, a variety of compounds has been analyzed by comparing logKoc with logKow, but also with other parameters (Schu¨u¨rmann et al. 2006). However, some models based on Kow fail in describing the soil sorption of some classes of herbicides, like acetanilides and triazines (Fig. 1). According to Sabljic´ et al. (1995), the linear regression between logKow and logKoc gave very poor correlations, with r2 B 0.5, while the corresponding values for other classes of herbicides were stronger (e.g. phosphates, alcohols, anilines and phenols). In quantitative structure–property relationship (QSPR), molecular properties are used to correlate chemical structures with the corresponding physical, chemical and/or biological properties, including soil sorption. In this case, the property is assumed to be dependent on the molecular structure and atomic composition. Some three-dimensional QSPR methods have been developed to generate descriptors that correlate with molecular properties, like CoMFA (Cramer et al. 1988) and CoMSIA (Klebe et al. 1994). However, Estrada et al. (2001), and Brown and Martin
123
144
Bull Environ Contam Toxicol (2014) 92:143–147
O R1
N
Table 1 Series of acetanilide (1–21) and triazine (22–36) herbicides, experimental logKoc obtained from the literature (Sabljic´ et al. 1995) and calculated molecular descriptors
R1 CH3
N R2
R2
N R3
acetanilide-type herbicides
N N
N
R4
R5
1,3,5-triazine-type herbicides
Fig. 1 Acetanilide and triazine basic structures used in the QSPR modeling
(1997) pointed out that 2D molecular structural descriptors, possessing high reappearance, are not inferior to 3D ones, at the least in many practical cases. In addition, images representing molecular structures have also been shown to correlate with biological properties of herbicides (Freitas et al. 2013). Thus, QSPR strategies can be useful for the development of predictive models to estimate herbicide properties. Acetanilide herbicides comprise compounds including the widely used Alachlor; sorption in soils and sediments is an important factor controlling the migration and bioavailability of these herbicides, while microbial degradation is the most important factor in determining their overall fate in the environment (Stamper and Tuovinen 1998). Triazine herbicides are generally of low acute toxicity for birds and mammals, although certain species show unexpected vulnerability for some compounds (Prosen 2012). Thus, this work aims at using few molecular descriptors to generate a QSPR model with improved prediction performance if compared to existing uniparametric models, in order to estimate soil sorption of a series of acetanilide and triazine herbicides. In this way, development of novel herbicides of these families can be focused not only on the efficacy for the control of unwanted plant, but also considering the environmental hazard of a designed compound.
Materials and Methods The sets of 21 acetanilide (1–21) and 15 triazine (22–36) herbicides, as well as the corresponding soil sorption (logKoc) values (Table 1) were obtained from the literature (Sabljic´ et al. 1995; Gerstl 1990), where the soil conditions and the method for data collection and treatment are described. The values of octanol/water partition coefficient (logKow), molecular weight (MW) and molecular volume (MV) were calculated using the property prediction module of the Molinspiration freeware program (www.
123
#
Compound
logKoc
logKow
MW
MV
1
Acetanilide
1.43
1.156
135.2
132.0
2
2-Cl-acetanilide
1.58
1.786
169.6
145.5
3 4
3-CH3-acetanilide 3-F-acetanilide
1.45 1.57
1.581 1.296
149.2 153.2
148.6 136.9
5
3-Cl-acetanilide
1.86
1.810
169.6
145.5
6
3-Br-acetanilide
2.01
1.941
214.1
149.9
7
3-CF3-acetanilide
1.75
2.028
203.2
163.3
8
3-NO2-acetanilide
1.94
1.091
180.2
155.3
9
4-F-acetanilide
1.48
1.320
153.2
136.9
10
4-Br-acetanilide
1.95
1.965
214.1
149.9
11
4-OCH3-acetanilide
1.40
1.213
165.2
157.5
12
Butyranilide
1.71
2.550
163.2
165.6
13
Propachlor
2.42
2.639
211.7
196.1
14
3,4-diCl-acetanilide
2.34
2.440
204.0
159.1
15
3-Cl-4-OCH3-acetanilide
1.95
1.819
199.6
171.1
16
Alachlor
2.28
3.671
269.8
255.2
17
Butachlor
2.86
5.109
311.8
305.6
18 19
Norfluorazon Acetchlor
3.28 2.32
2.886 3.580
303.7 269.8
229.4 255.2
20
Metholachlor
2.46
3.548
283.8
271.8
21
Matalaxyl
1.57
2.339
279.3
269.2
22
Simazine
2.10
2.251
201.7
176.6
23
Propazine
2.40
2.845
229.7
209.8
24
Ametryn
2.59
2.728
213.3
197.8
25
Terbutryn
2.85
3.537
241.4
230.6
26
Prometron
2.60
3.033
225.3
221.8
27
Atrazine
2.24
2.548
215.7
193.2
28
Ipazine
2.91
3.916
257.8
243.4
29
Trietazin
2.76
3.619
243.7
226.8
30
Dipropetryn
3.07
3.698
255.4
247.8
31
Terbuthylazine
2.32
3.061
229.7
209.5
32
Prometryn
2.85
3.322
241.4
231.0
33
Metribuzin
1.71
1.460
214.3
192.2
34 35
Cyanazine Sec-Bumeton
2.28 2.78
2.470 3.272
240.7 225.3
209.8 222.0
36
Metamitron
2.17
0.357
202.2
179.3
logarithm of the octanol/water partition coefficient—logKow; molecular weight—MW; molecular volume—MV
molinspiration.com) and then regressed against the experimental values of logKoc using multiple linear regression (MLR). The statistical results of calibration were analyzed on the basis of RMSEC (root mean square error of calibration) h i and r2, defined as 1 ðRðyi y^i Þ2 =Rðyi yÞ2 , in which yi corresponds to the experimental logKoc values, yˆi are the
Bull Environ Contam Toxicol (2014) 92:143–147
145
Results and Discussion The soil sorption of organic compounds is expected to be correlated with logKow, because sediments and mainly organic matter in soil have been shown to interact strongly with progressively more hydrophobic compounds. However, logKow values for acetanilide and triazine herbicides were found elsewhere to be poorly correlated to logKoc (r2 of 0.491 and 0.273, respectively) (Sabljic´ et al. 1995). logKow values can be calculated using a variety of programs; for example, the calculated logKow obtained for the set of acetanilides and triazines using the Molinspiration program correlates significantly well with the corresponding experimental data (r2 of 0.766 and 0.873, respectively). However, the calculated logKow values do not fit appropriately with the experimental logKoc data, despite the higher coefficient of determination (r2 of 0.565 and 0.656, respectively) than those obtained elsewhere (Sabljic´ et al. 1995) using the following formulae (Eqs. 2 and 3): For acetanilides : log Koc ¼ 1:12 þ 0:40 log Kow ; r 2 ¼ 0:491; q2 ¼ 0:436
ð2Þ
For triazines : log Koc ¼ 1:50 þ 0:30 log Kow ; r 2 ¼ 0:273; q2 ¼ 0:045
ð3Þ
This corroborates that hydrophobicity does not completely explain the soil sorption profile of acetanilide and triazine herbicides. Therefore, molecular weight (MW) and volume (MV) were calculated to be added in the models. The linear three-parametric model for the acetanilide series based on logKow, MW and MV gave an improved correlation with
3.5
Fitted and predicted logK oc
predicted logKoc values, and y^i corresponds to the mean logKoc values. The model was validated through leave-oneout cross-validation (statistically evaluated using RMSECV and q2, defined similarly as above). QSPR studies in the literature indicate that r2 C 0.8 and q2 C 0.5 are satisfactory. In order to guarantee that good models were not obtained by chance correlations, the Y-block (the logKoc column vector) was randomized and a new regression was carried out with the intact X-matrix (mean of 10 repeti2 tions); low values of rYrand indicate that the real calibration is not a chance correlation. Accordingly, an additional statistical parameter proposed by Mitra et al. (2010), namely rP2 (Eq. 1), was calculated to give insight about the 2 statistical difference between r2 and rYrand (values above 0.5 are acceptable). These calculations were performed using the Chemoface program (www.ufla.br/chemoface) (Nunes et al. 2012). 1=2 2 rP2 ¼ r 2 r 2 rrand ð1Þ
Calibration Leave-one-out cross-validation
3.0
2.5
2.0
1.5
1.0 1.0
1.5
2.0
2.5
3.0
3.5
Experimental logKoc Fig. 2 Plot of experimental versus fitted and predicted logKoc for the series of acetanilide herbicides. Segmented line for calibration and solid line for LOO cross-validation
logKoc (Eq. 4), as illustrated in Fig. 2. The model is slightly improved after removing compound 18 (not an acetanilide in fact), with r2 = 0.795 and q2 = 0.633, but Eq. 4 comprising all 21 derivatives is sufficiently predictive to estimate logKoc of acetanilide-type herbicides with reasonable accuracy. The recognition of compound 18 as an outlier indicates that the QSPR model is robust and that it really encodes chemical information about the series of acetanilide herbicides. logKoc ¼ 0:75 þ 0:33 logKow þ 0:011 MW 0:010 MV
ð4Þ
N = 21, r2 = 0.788, RMSEC = 0.225, q2 = 0.554, 2 RMSECV = 0.334, rYrand = 0.141, rP2 = 0.616 Despite the apparent collinearity between MW and MV (r2 = 0.848), the absence of one of these molecular descriptors to give a two-parametric model does not provide highly predictive QSPR models, i.e. r2 of 0.661 (RMSEC = 0.285) and q2 of 0.424 (RMSECV = 0.382) when using logKow and MW, and r2 of 0.567 (RMSEC = 0.322) and q2 of 0.390 (RMSECV = 0.395) when using logKow and MV. Molecular weight and volume do not necessarily reflect hydrophobicity; for instance, the polar nitro substituent in 8 leads to a larger molecular weight and volume than the methyl derivative (3), but it is less hydrophobic (logKow = 1.091 for 8 and 1.581 for 3). Thus, the chemical information explained by the three descriptors is complementary, suggesting that logKow alone is not enough to explain accurately the sorption mechanism and mobility of acetanilides in the soil. Similar analysis was carried out for the set of 15 triazines (22-36). The three-parametric QSPR model is significantly better when compared to the monoparametric
123
Bull Environ Contam Toxicol (2014) 92:143–147 3.2
Calibration Leave-one-out cross-validation
3.0 2.7 2.4 2.1 1.8 1.5
a
1.2 1.2
1.5
1.8
2.1
2.4
2.7
3.0
Fitted and predicted logKoc
Fitted and predicted logKoc
146
Calibration Leave-one-out cross-validation
3.0 2.8 2.6 2.4 2.2
b 2.0 2.0
Experimental logK oc
2.2
2.4
2.6
2.8
3.0
3.2
Experimental logK oc
Fig. 3 Plot of experimental versus fitted and predicted logKoc for the series of triazine herbicides. a all triazine compounds included; b outliers 33 and 36 removed. Segmented line for calibration and solid line for LOO cross-validation
model based only on logKow and can be expressed by Eq. 5 and Fig. 3. Since the structural scaffold of compounds 33 and 36 corresponds to 1,2,4-triazinones rather than 1,3,5triazines of the remaining 13 compounds, they behave as outliers (as identified by Student’s residuals and leverage diagnostics). Therefore, the QSPR model for the 13 1,3,5triazines (33 and 36 removed) gave an r2 of 0.91, which is highly predictive and reliable for estimation of logKoc of different triazine analogues, as supported by the validation data (Eq. 6 ).
while molecular weight also plays a significant role in these models. Accordingly, norfluorazon, which contains the bulky substituents chlorine, ethyl and ethoxymethyl groups at the acetanilide framework, is the most persistent organic herbicide along with the series, whilst one of the least hydrophobic derivatives (11, containing only an additional methoxy group at acetanilide) interacts poorly with soil compared to other herbicides.
logKoc ¼ 0:68 þ 0:100 logKow 0:012 MW þ 0:021 MV
Conclusions ð5Þ
N = 15, r2 = 0.817, RMSEC = 0.154, q2 = 0.551, 2 RMSECV = 0.287, rYrand = 0.231, rP2 = 0.625 logKoc ¼ 0:79 þ 0:201 logKow 0:010 MW þ 0:017 MV
ð6Þ
N = 13, r 2 = 0.908, RMSEC = 0.088, q2 = 0.829, 2 = 0.353, rP2 = 0.676 RMSECV = 0.123, rYrand Similarly to described before for the acetanilide-type herbicides, two-parametric modeling using only logKow and MW, or logKow and MV, achieved very modest correlations with logKoc and poor statistical results of validation, i.e. r2 of 0.685 (RMSEC = 0.202) and q2 of 0.369 (RMSECV = 0.349) when using logKow and MW, and r2 of 0.776 (RMSEC = 0.171) and q2 of 0.501 (RMSECV = 0.305) when using logKow and MV. Overall, while logKow does not completely explain the soil sorption of acetanilide and triazine herbicides, introduction of two simple and easily accessible molecular descriptors in QSPR models provides enhanced predictive abilities for environmental risk assessment. Compounds with low sorption in soil are characterized by low logKow values compared to the average, and compounds with high soil sorption are characterized by high molecular volume,
123
The soil sorption of acetanilide and triazine herbicides can be accurately predicted using molecular volume, molecular weight and logKow values rather than only one of these molecular descriptors. A single model to predict logKoc and other properties of different organic classes of compounds remains to be investigated in the future. Acknowledgments Authors are thankful to FAPEMIG and CNPq for the financial support, studentship (to M.R.F. and S.V.B.G.M.) and fellowships (to R.L.G.M., M.P.F. and N.V.).
References Ackerman F (2007) The economics of atrazine. Int J Occup Environ Health 13:437–445 Brown RD, Martin YC (1997) The information content of 2D and 3D structural descriptors relevant to ligand-receptor binding. J Chem Inf Comp Sci 37:1–9 Corsonlini S, Ademollo N, Romeo T, Greco S, Focardi S (2005) Persistent organic pollutants in edible fish: a human and environmental health problem. Microchem J 79:115–123 Cramer RD, Patterson DE, Bunce JD (1988) Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc 110:5959–5967 Domingo JL (2004) Human exposure to polybrominated diphenyl ethers through the diet. J Chromatogr A 1054:321–326
Bull Environ Contam Toxicol (2014) 92:143–147 Estrada E, Molina E, Perdomo-Lo´pez I (2001) Can 3D structural parameters be predicted from 2D (topological) molecular descriptors? J Chem Inf Comp Sci 41:1015–1021 Freitas MR, Matias SVBG, Macedo RLG, Freitas MP, Venturin N (2013) Augmented multivariate image analysis applied to quantitative structure-activity relationship modeling of the phytotoxicities of benzoxazinone herbicides and related compounds on problematic weeds. J Agric Food Chem 61:8499–8503 Gerstl Z (1990) Estimation of organic chemical sorption by soils. J Cont Hydrol 6:357–375 Giesy JP, Ludwig JP, Tillitt DE (1994) Deformities in birds of the Great-Lakes region assigning causality. Environ Sci Technol 28:A128–A135 Kavlock RJ, Daston GP, Derosa C, Fenner-Crisp P, Gray LE, Kaattari S, Lucier G, Luster M, Mac MJ, Maczka C, Miller R, Moore J, Rolland R, Scott G, Sheehan DM, Sinks T, Tilson HA (1996) Research needs for the risk assessment of health and environmental effects of endocrine disruptors: a report of the US EPAsponsored workshop. Environ Health Persp 104:715–740 Kelce WR, Stone CR, Laws SC, Gray LE, Kemppainen JA, Wilson EM (1995) Persistent DDT metabolite P, P’-DDE is a potent androgen receptor antagonist. Nature 375:581–585 Klebe G, Abraham U, Mietzner T (1994) Molecular similarity indices in a comparative analysis (CoMSIA) of drug molecules to correlate and predict their biological activity. J Med Chem 37: 4130–4146
147 Mitra I, Saha A, Roy K (2010) Exploring quantitative structureactivity relationship studies of antioxidant phenolic compounds obtained from traditional Chinese medicinal plants. Mol Simul 36:1067–1079 Nunes CA, Freitas MP, Pinheiro ACM, Bastos SC (2012) Chemoface: a novel free user-friendly interface for chemometrics. J Braz Chem Soc 23:2003–2010 Prosen H (2012) Fate and determination of triazine herbicides in soil, herbicides: properties, synthesis and control of weeds, ed. by Hasaneen MN, InTech. Available from: http://www.intechopen. com/books/herbicides-properties-synthesis-and-control-of-weeds/ fateand-determination-of-triazine-herbicides-in-soil Ratcliffe DA (1967) Decrease in eggshell weight in certain birds of prey. Nature 215:208–210 Ratcliffe DA (1970) Changes attributable to pesticides in egg breakage frequency and eggshell thickness in some British birds. J Appl Ecol 7:67–115 Sabljic´ A, Gu¨sten H, Verhaar H, Hermens J (1995) QSAR modelling of soil sorption. Improvements and systematics of logKOC versus logKOW correlations. Chemosphere 31:4489–4514 Schu¨u¨rmann G, Ebert R-U, Ku¨hne R (2006) Prediction of the sorption of organic compounds into soil organic matter from molecular structure. Environ Sci Technol 40:7005–7011 Stamper DM, Tuovinen OH (1998) Biodegradation of the acetanilide herbicides alachlor, metolachlor, and propachlor. Crit Rev Microbiol 24:1–22
123
