Bulletin of Environmental Contamination and Toxicology

, Volume 83, Issue 4, pp 520–524

Mechanism-Based Quantitative Structure–Activity Relationships on Toxicity of Selected Herbicides to Chlorella vulgaris and Raphidocelis subcapitata

Authors

  • Guanghui Ding
    • Key Laboratory of Industrial Ecology and Environmental Engineering (MOE), Department of Environmental Science and TechnologyDalian University of Technology
    • College of Environmental Science and EngineeringDalian Maritime University
  • Xue Li
    • Key Laboratory of Industrial Ecology and Environmental Engineering (MOE), Department of Environmental Science and TechnologyDalian University of Technology
  • Fan Zhang
    • Key Laboratory of Industrial Ecology and Environmental Engineering (MOE), Department of Environmental Science and TechnologyDalian University of Technology
    • Key Laboratory of Industrial Ecology and Environmental Engineering (MOE), Department of Environmental Science and TechnologyDalian University of Technology
  • Liping Huang
    • Key Laboratory of Industrial Ecology and Environmental Engineering (MOE), Department of Environmental Science and TechnologyDalian University of Technology
  • Xianliang Qiao
    • Key Laboratory of Industrial Ecology and Environmental Engineering (MOE), Department of Environmental Science and TechnologyDalian University of Technology
Article

DOI: 10.1007/s00128-009-9811-8

Cite this article as:
Ding, G., Li, X., Zhang, F. et al. Bull Environ Contam Toxicol (2009) 83: 520. doi:10.1007/s00128-009-9811-8

Abstract

Four quantitative structure-activity relationships were developed for toxicity of selected photosynthesis (PHS) inhibitors and acetolactate synthase (ALS) inhibitors to Chlorella Vulgaris and Raphidocelis subcapitata using a mechanism-based approach. These models have good fitness and predictive ability. The potential of electron transfer, intermolecular interactions with weak electron-transfer, and intermolecular dispersive interactions between PHS inhibitors and the active site of action are key factors influencing the toxicity of these PHS inhibitors. Intermolecular weak electron-transfer interactions and intermolecular dispersive interactions mainly determine the toxicity of these ALS inhibitors. Sulfonyl is an important functional group governing the toxicity of ALS inhibitors investigated.

Keywords

HerbicidesToxicityQSARsGreen algae

Herbicides provide an effective and economical means of weed control, and have been widely used in agriculture, landscape turf management, gardening, and hard surfaces maintenance (Gerecke et al. 2002; Kempenaar and Spijker 2004). Through a variety of mechanisms, including overspray, drift, surface runoff and leaching, they may enter various aquatic ecosystems (Gerecke et al. 2002; Leu et al. 2004). Adverse effects of these herbicides on non-target organisms of aquatic ecosystems, especially algae, are of special concern, and have to be investigated to assess their environmental risk. Thus, many researches have focused on the toxicity of herbicides to algae (Ma et al. 2002, 2006; Junghans et al. 2003). Among these studies, Ma et al. (2002, 2006) extensively studied 96-h acute toxicity of 40 herbicides to Chlorella vulgaris and Raphidocelis subcapitata.

As experimental determination of toxicity is costly and time-consuming, it is desirable to develop mathematical predictive relationships to theoretically quantify toxicity. Quantitative structure-activity relationship (QSAR) provides a convenient tool for toxicity evaluation and prediction, and can also give some insight into the mechanism of toxic actions (Schultz et al. 2003a; Eriksson et al. 2003). Two approaches, an analog-based approach and a mechanism-based approach, can be used for QSAR modeling. The former develops QSAR models for a series of homologous or congeneric compounds. However, structural similarity does not necessarily imply ecotoxicological similarity, and inclusion of compounds having different toxicity mechanisms may deteriorate the quality of QSAR models (Mekapati and Hansch 2002). Therefore it is appropriate to combine chemicals by their toxicity mechanisms, instead of chemical classes, to develop high-quality QSARs (Escher and Hermens 2002; Schultz et al. 2003a, b). The purpose of this study is to investigate toxicity of selected herbicides to green algae Chlorella Vulgaris and Raphidocelis subcapitata using the mechanism-based QSAR approach.

Materials and Methods

The 96-h toxicity data of 40 herbicides on Chlorella vulgaris and Raphidocelis subcapitata were taken from Ma et al. (2002, 2006). These 40 herbicides have different modes of toxic action. Thus, only two groups of the herbicides could be employed for QSAR establishment for they have comparatively large data sets to warrant regression analyses. These herbicides are 9 photosynthesis (PHS) inhibitors and 9 acetolactate synthase (ALS) inhibitors with toxicity data on Chlorella vulgaris, and 11 PHS inhibitors and 9 ALS inhibitors with toxicity data on Raphidocelis subcapitata. As paraquat (a PHS inhibitor) and bispyribac-sodium (an ALS inhibitor) are ionic compounds, which are different from the other herbicides, they were excluded from QSAR analyses. Thus the following QSAR studies focused on toxicity of selected PHS inhibitors and ALS inhibitors to Chlorella Vulgaris and Raphidocelis subcapitata, the CAS numbers and EC50 values of which are listed in Table 1. The EC50 values were converted to logarithmic form before model development.
Table 1

The CAS numbers and log EC50 values of selected herbicides

No.

Herbicides

CAS number

EC50 (mol/L)a

EC50 (mol/L)b

Photosynthesis (PHS) inhibitors

1

Ametryne

834-12-8

 

5.45 × 10−8

2

Atrazine

1912-24-9

1.92 × 10−6

5.60 × 10−7

3

Bromoxynil

1689-84-5

2.83 × 10−4

2.27 × 10−5

4

Chlorotoluron

15545-48-9

1.19 × 10−7

4.00 × 10−9

5

Cyanazine

21725-46-2

5.35 × 10−7

2.48 × 10−7

6

Diuron

330-54-1

1.84 × 10−8

3.01 × 10−9

7

Isoproturon

34123-59-6

1.09 × 10−7

6.58 × 10−8

8

Methabenzthiazuron

18691-97-9

 

9.46 × 10−8

9

Prometryne

7287-19-6

2.22 × 10−7

4.81 × 10−8

10

Simazine

122-34-9

1.08 × 10−5

3.73 × 10−6

Acetolactate synthase (ALS) inhibitors

11

Bensulfuron-methyl

83055-99-6

4.33 × 10−5

3.31 × 10−5

12

Chlorimuron-ethyl

90982-32-4

4.64 × 10−5

1.33 × 10−5

13

Cyclosulfamuron

136849-15-5

7.03 × 10−7

9.59 × 10−7

14

Ethametsulfuron

111353-84-5

1.64 × 10−4

7.36 × 10−5

15

Flumetsulam

98967-40-9

3.28 × 10−5

7.34 × 10−6

16

Metsulfuron-methyl

74223-64-6

 

6.39 × 10−5

17

Nicosulfuron

111991-09-4

1.06 × 10−5

3.49 × 10−6

18

Pyrazonsulfuron-ethyl

93697-74-6

4.54 × 10−5

2.68 × 10−5

19

Tribenuron

106040-48-6

9.70 × 10−5

 

aEC50 of herbicides to Chlorella Vulgaris

bEC50 of herbicides to Raphidocelis subcapitata

Generally, chemical toxicity is the result of two key steps: partitioning of toxicant into/through the biological membrane, and interactions of the toxicant with the site of action. Thus it is important to select suitable molecular structural descriptors to characterize these processes in QSAR studies.

The partitioning processes of toxicants can be described by the n-octanol/water partition coefficient (KOW). log KOW values of these herbicides were estimated using the software KOWWIN (Version 1.67, EPI Suite, US-EPA). The interactions of the toxicant with the sites of action could be described by 17 molecular structural descriptors, the definitions of which are listed in Table 2. Among the 17 descriptors, 13 quantum-chemical descriptors were computed using PM3 Hamiltonian algorithm of MOPAC (2000), and the other 4 molecular structural descriptors were obtained by software CS Chem3D Ultra (Version 6.0, www.camsoft.com). For ALS inhibitors, two special descriptors were considered to characterize their special structures, which were the net atomic charges of the sulfur atom (qS+) and the average net atomic charges of oxygen atoms on sulfonyl (qO). These two descriptors were also calculated by CS Chem3D Ultra.
Table 2

Definitions of the molecular structural descriptors

No.

Descriptors

Descriptions

1

Mw

Molecular weight (atomic mass units)

2

α

Average molecular polarizability (atomic units)

3

μ

Dipole moment (Debye)

4

ΔHf

Standard heat of formation (kJ mol−1)

5

TE

Total energy (electron Volts, eV)

6

EE

Electronic energy (eV)

7

CCR

Core-core repulsion energy (eV)

8

EHOMO

The energy of the highest occupied molecular orbital (eV)

9

ELUMO

The energy of the lowest unoccupied molecular orbital (eV)

10

qC

The most negative net atomic charges on a carbon atom (atomic charge unit, a.c.u.)

11

qH+

The most positive net atomic charges on a hydrogen atom (a.c.u.)

12

qN

The most negative net atomic charges on a nitrogen atom (a.c.u.)

13

qmax+

The most positive net atomic charges on a atom (a.c.u.)

14

CAA

Connolly accessible area (Å2)

15

CMA

Connolly molecular area (Å2)

16

CSEV

Connolly solvent-excluded volume (Å3)

17

Ov

Ovality is the ratio of the molecular surface area to the minimum surface area

When too many molecular structural descriptors were included in QSAR models, multicollinearity among descriptors may occur, and classical multiple linear regression will not provide predictive and robust models. Thus partial least squares (PLS) regression was adopted, for this method can analyze data with strongly collinear, noisy and numerous predictor variables. In this study, the PLS analyses were carried out by Simca-S (Version 6.0, Umetri AB & Erisoft AB). Simca-S employs “cross validation” to determine the number of PLS components (A). In the end, a statistic \( Q_{\text{cum}}^{2} \) is provided for a PLS model, which denotes the cumulative variance of the dependent variable explained by the extracted PLS components, and is a good measurement of the predictive power and robustness of the model. When \( Q_{\text{cum}}^{2} \) of a model is larger than 0.5, the model is considered to be predictive and robust.

If insignificant descriptors are included in a PLS model, the prediction ability and robustness of the model may decrease, and the interpretation of the model becomes difficult. It is therefore necessary to eliminate redundant descriptors and identify important descriptors. The following variable selection procedure was adopted: At first, correlation analyses were performed between log EC50 and all the predictor variables to find the most significant variable. A simple linear regression between log EC50 and the most significant variable were established. Then one other variable was added and a PLS model was built. This step was repeated until every remaining variable had been added once and only once. \( Q_{\text{cum}}^{2} \) values of this series of models were compared and the model with the highest \( Q_{\text{cum}}^{2} \) was selected to enter the next step. The variable-addition and model-building processes were repeated until the number of PLS components was bigger than n/4 or all predictor variables had been included. The model with the highest \( Q_{\text{cum}}^{2} \) was selected as the optimal model from all the models obtained.

Results and Discussion

Previous studies indicated that narcotic toxicity could be explained by log KOW. Therefore simple regression equations between log EC50 and log KOW were analyzed.

For toxicity of 8 PHS inhibitors to Chlorella vulgaris:
$$ \begin{gathered} \log EC_{50} = - 7.917 + 6.346 \times 10^{ - 1} \log K_{\text{OW}} \hfill \\ {\rm n} = 8,\;r = 0.221,\;SE = 1.401,\;F = 0.307,\;p = 0.599 \hfill \\ \end{gathered} $$
(1)
where r is the correlation coefficient between observed and fitted values, SE is the standard error, F is the F statistic, and p is probability that r equals zero.
For toxicity of 10 PHS inhibitors to Raphidocelis subcapitata:
$$ \begin{gathered} \log EC_{50} = - 7.873 + 3.494 \times 10^{ - 1} \log K_{\text{OW}} \hfill \\ {\rm n} = 10,\;r = 0.128,\;SE = 1.265,\;F = 0.133,\;p = 0.725. \hfill \\ \end{gathered} $$
(2)
For toxicity of 8 ALS inhibitors to Chlorella vulgaris:
$$ \begin{gathered} \log EC_{50} = - 4.634 + 6.562 \times 10^{ - 2} \log K_{\text{OW}} \hfill \\ {\rm n} = 8,\;r = 0.099,\;SE = 0.787,\;F = 0.060,\;p = 0.815. \hfill \\ \end{gathered} $$
(3)
For toxicity of 8 ALS inhibitors to Raphidocelis subcapitata:
$$ \begin{gathered} \log EC_{50} = - 5.011 + 1.255 \times 10^{ - 1} \log K_{\text{OW}} \hfill \\ {\rm n} = 8,\;r = 0.218,\;SE = 0.687,\;F = 0.300,\;p = 0.603. \hfill \\ \end{gathered} $$
(4)

However, F-test showed that log EC50 did not significantly correlate with log KOW for these regression equations. Thus, further statistical analyses were performed. Following the aforementioned variable selection procedure, optimal QSAR models were obtained as follows.

PHS inhibitors on Chlorella vulgaris:
$$ \begin{aligned} \log EC_{50} = & -1.119 \times 10 + 3.904 \times 10^{ - 4} {Mw} + 5.154 \times 10^{ - 3} {TE} - 1.750E_{\text{HOMO}} \\ & -7.164 \times 10^{ - 1} E_{\text{LUMO}} - 3.966q_{\text{N}}^{ - } \\ \end{aligned} $$
(5)
$$ {\rm n} = 8,\;A = 2,\;R_{{X({\text{cum}})}}^{2} = 0.633,\;R_{{Y({\text{cum)}}}}^{2} = 0.979,\;Q_{{({\text{cum)}}}}^{2} = 0.911,\;r = 0.989,\;SE = 0.299 $$
where A is the number of extracted PLS components, \( R_{{X({\text{cum}})}}^{2} \) and \( R_{{Y({\text{cum}})}}^{2} \) stand for the cumulative sum of squares of all the predictor variables and dependent variable explained by all extracted components, respectively.
PHS inhibitors on Raphidocelis subcapitata:
$$ \begin{aligned} \log EC_{50} = & -3.899 - 1.564 \times 10^{ - 3} {Mw} + 4.464 \times 10^{ - 3} {TE} - 1.075E_{\text{HOMO}} \\ & - 8.131 \times 10^{ - 1} E_{\text{LUMO}} -1.183 \times 10q_{ \max }^{ + } \\ \end{aligned} $$
(6)
$$ {\rm n} = 10,\;A = 2,\;R_{{X({\text{cum}})}}^{2} = 0.623,\;R_{{Y ( {\text{cum}})}}^{2} = 0.934,\;Q_{\text{cum}}^{2} = 0.847,\;r = 0.966,\;{SE} = 0.351 $$
ALS inhibitors on Chlorella vulgaris:
$$ \begin{aligned} \log EC_{50} = & -3.357 \times 10 - 1.266 \times 10^{ - 3} {Mw} - 1.591 \times 10^{ - 1} \mu + 1.722 \times 10^{ - 5} {EE} \\ & -1.954 \times 10^{ - 5} {CCR} - 6.524 \times 10^{ - 1} E_{\text{HOMO}} - 3.119q_{C}^{ - } \\ & -3.007 \times 10q_{\text{O}}^{ - } -7.618 \times 10^{ - 4} {CSEV} \\ \end{aligned} $$
(7)
$$ {\rm n} = 8,\;A = 2,\;R_{{X({\text{cum}})}}^{2} = 0.786,\;R_{{Y({\text{cum}})}}^{2} = 0.977,\;Q_{\text{cum}}^{2} = 0.921,\;r = 0.988,\;SE = 0.132. $$
ALS inhibitors on Raphidocelis subcapitata:
$$ \begin{aligned} \log EC_{50} = & -3.733 \times 10 - 7.740 \times 10^{ - 3} \alpha - 9.436 \times 10^{ - 1} E_{\text{HOMO}} - 2.794q_{\text{C}}^{ - } \\ & -2.263q_{\text{S}}^{ + } - 3.548 \times 10q_{\text{O}}^{ - } \\ \end{aligned} $$
(8)
$$ {\rm n} = 8,\;A = 2,\;R_{{X ( {\text{cum}})}}^{2} = 0.624,\;R_{{Y({\text{cum}})}}^{2} = 0.987,\;Q_{\text{cum}}^{2} = 0.887,\;r = 0.993,\;SE = 0.088. $$
From Fig. 1, it can be seen that the predicted log EC50 values agree well with the observed values for these 4 QSAR models. Furthermore, all \( Q_{\text{cum}}^{2} \) values are far higher than 0.5, indicating high predictability and robustness of these models. Considering the paucity of algae toxicity data for many herbicides, the difficulty or high expenditures involved in experimental determinations, these models could serve as a first and fast approximation of toxicity for the relevant herbicides with the same mode of toxic action.
https://static-content.springer.com/image/art%3A10.1007%2Fs00128-009-9811-8/MediaObjects/128_2009_9811_Fig1_HTML.gif
Fig. 1

Plot of predicted versus observed log EC50 values for 4 QSAR models

For each QSAR model, 2 PLS components were extracted. From PLS weights shown in Fig. 2, one can see how predictor variables and response variable combine in PLS components, and how they relate to each other.
https://static-content.springer.com/image/art%3A10.1007%2Fs00128-009-9811-8/MediaObjects/128_2009_9811_Fig2_HTML.gif
Fig. 2

Plots of PLS weights: a for model (5), b for model (6), c for model (7), d for model (8)

For toxicity of the PHS inhibitors, model (5) and (6) are similar, with only one different molecular descriptor. The first PLS components of the two models are mainly related with EHOMO and ELUMO, which correlate with the global readiness of a molecule to donate or accept electron charge (Wang et al. 2004). It is known that the PHS inhibitors are inhibitors of photosynthesis II (PSII), which displace the QB plastoquinone from its binding site on the D1 protein, and then block the electron transfer chain of PSII (Gramatica et al. 2001; Fufezan et al. 2002). So EHOMO and ELUMO characterize the potential of electron transfer of the herbicide molecules in governing the toxicity.

The second PLS component of model (5) is mainly related with qN, while for model (6) is qmax+ and Mw. qN and qmax+ relate with intermolecular interactions with weak electron-transfer, and may characterize hydrogen bonding between the PHS inhibitors and the active site of action. Mw has some important influence on the second PLS component of model (6). Mw is correlated with molecular size, and therefore with intermolecular dispersive interactions or steric hindrance. As Mw is in negative correlation with log EC50, it primarily describes the influence of intermolecular dispersive interactions between PHS inhibitors and the active site of action on the toxicity. For model (5), Mw has certain influence on both PLS components. However, it reveals different effects on the toxicity via these two PLS components. In model (5), these effects were summarized, and Mw takes positive correlation with log EC50. As the coefficient of Mw in model (5) is only 3.904 × 10−4, the influence of Mw on the toxicity is too small to attend. In addition, TE has important influence on both PLS components for model (5) and (6). As TE is significantly negative correlated with molecular size, it could also characterize intermolecular dispersive interactions or steric hindrance. In both models, TE is in positive correlation with log EC50, so it mainly describe intermolecular dispersive interactions between PHS inhibitors and the active site of action, which works together with the above influencing factors.

For toxicity of the ALS inhibitors, model (7) and model (8) have some common descriptors. EHOMO, μ, qC and qO are the most important descriptors for model (7), and EHOMO, qO, qS+ and qC for model (8). EHOMO, qqS+ and qC are all related to intermolecular interactions with weak electron-transfer including hydrogen bond, and μ relates to electrostatic interactions. It is known that the ALS inhibitors inhibit the activity of the acetolactate synthase enzyme and thereby block the biosynthesis of the branched-chain amino acids valine, leucine and isoleucine (Simpson et al. 1995). Therefore these variables characterize the influence of intermolecular weak electron-transfer interactions. qO was included in both models, however, qS+ only in model (8). This means that sulfonyl is an important functional group for toxicity of the ALS inhibitors, especially to Raphidocelis subcapitata.

In addition, Mw, EE, CCR and CSEV have certain contribution to both PLS components for model (7), while α has some contribution to both PLS components for model (8). These descriptors are inter-correlated, and may represent intermolecular dispersive interactions or steric hindrance. From their coefficients, it can be determined that they may mainly describe the influence of intermolecular dispersive interactions on toxicity of these herbicides.

Acknowledgments

The study was supported by the National Basic Research Program of China (973 Project 2006CB403302) and National Natural Science Foundation of P. R. China (No. 20337020).

Copyright information

© Springer Science+Business Media, LLC 2009