A study on the decolorization of methylene blue by Spirodela polyrrhiza: experimentation and modeling

Abstract

Phytoremediation is a cutting-edge technology applied for the purpose of biological treatment of contaminated soil, water and air. We studied the Spirodela polyrrhiza-assisted removal of methylene blue (MB), and then assessed the effect of dye concentration, pH, plant weight and exposure duration on pollutant elimination in a 7 day time span. Response surface methodology (RSM) was used to design the experiments on a daily sampling basis. The statistical analysis of data shows a promising capability of Spirodela polyrrhiza for bioremediation. The percentage removal of methylene blue in some samples was approximately 90–95%. Results show that the duckweed-aided phytoremediation is highly effective when experiments are run in low concentration of methylene blue, pH of 6–7, and high plant weight. We then came up with an artificial neural network (ANN) model to simulate the dye removal process and analyzed the sensitivity of the system towards various factors. It turned out that the exposure time is the most important factor in the phytoremediation of methylene blue by Spirodela polyrrhiza.

---

1. Introduction

Population growth and increasing environmental pollution have introduced detrimental influences on the ecology of the planet to the extent where some regions have been depleted of potable water.¹ Thus far, a lot of effort has gone into the improvement of water recovery strategies adopted by textile industries. The wastewater surge into the environment directly from the dyeing units of textile industry contains residuals of dye and detergent. These chemicals are amongst the main pollutants that endanger the ecology of rivers and lagoons.² Several methods, such as precipitation,^3,4 reverse osmosis,^5,6 adsorption^7,8 and chemical oxidation,^9,10 have been devised to address the issue. Although, in some cases, these physical and chemical methods have been proven useful, there are some drawbacks associated with them amongst which cost-ineffectiveness and the difficulty of the process predominate.¹¹

One of the important mechanisms that has recently attracted the attention of scientists is the accumulation of dyes in plants. This technique, called phytoremediation, is one of the modern and novel methods of biological refining.^12–15 Hydrocarbons, heavy metal ions and dyeing wastewater generated from cane molasses fall in the category of chemicals capable of being removed by this method.^16–18 Methylene blue is one of the organic dyes frequently used in industries such as textile, wood, paper and dyehouses.¹⁹ The effluent of such industries contains large amounts of this compound as incompetent treatment results in water and soil contamination.²⁰ Methylene blue is considered to have less toxicity compared to other toxins; however, high concentrations of the pollutant could cause headache, nausea, and increased heart rate.¹⁹ The chemical structure of methylene blue is shown in Fig 1. Aquatic plants are able to absorb and accumulate pollutants in their roots, pedicles and leaves.^21,22 Spirodela polyrrhiza is a species in the duckweed family that can easily absorb dyes and heavy metal ions owing to its vascular structure.^23,24 Waranusantigul et al.²³ investigated the capacity of Spirodela polyrrhiza for the removal of methylene blue and also studied the impact of main factors (e.g. adsorption time, pH and amount of plant). Their results show that Spirodela polyrrhiza can take up the pollutant with a high absorption rate of 80% in the first two days of experiments.

Fig. 1 Chemical structure of methylene blue.

The main objectives of the present study are to determine the potential of methylene blue removal in aqueous solution by Spirodela polyrrhiza, to launch an investigative study on the factors influencing the removal rate such as contact time, initial dye concentration, pH and plant weight, and also to mathematically model the process. Bioremediation is by nature a sophisticated process; thus, simple mathematical models fall short in the determination of the effective parameters and their impact on the process, and they cannot represent the process as a consequence. Moreover, the multiplicity of parameters causes long running set of experiments when observing all possible combinations of factor levels. Therefore, central composite design²⁵ (CCD), one of the most popular classes of RSM,²⁶ was used to design the experiments and analyze the importance of the factors. A multilayer perceptron neural network associated with a back-propagation (BP) learning algorithm was used to simulate the process, and finally, the experimental results were compared with those obtained by the proposed model.

2. Materials and methods

2.1 Materials

Aquatic Spirodela polyrrhiza was obtained from a lagoon around the city of Khomam, Gilan Province in northern Iran. Deionized double-distilled water was used for the preparation of samples and cultures. Chemicals used were purchased from Merck and Millipore.

2.2 Preparation of culture

After the initial washing and removal of the impurities, plants were washed by both 0.5% ethanol and 0.5% sodium hypochlorite and pure water consecutively.^27,28 In order to obtain the plants adjusted to their new environment, they were cultivated in a nutritive solution of Hoagland²⁹ (303.3 g KNO₃; 330 g Ca(NO₃)₂; 57.5 g NH₄H₂PO₄; 120.4 g MgSO₄; 0.004 g H₂MoO₄; 0.033 μmol ZnSO₄; 0.016 μmol CuSO₄; 0.04 μmol MnSO₄; 0.1 μmol H₃BO₃; 3.5 μmol EDTA) at 25 °C for 4 weeks. Afterwards, pH was set to 6.5–7 by a buffer solution (0.1 N HNO₃ and 0.1 N NaOH).²⁸

2.3 Design of experiments

The design of experiments for a 7 day run was performed by a CCD method. Each factor (initial dye concentration, pH and plant weight) was considered in five different levels (Table 1), and the central point run was replicated 3 times as the overall number of tests count 18. Design Expert 7.0 (DX) software³⁰ was used to design the experiments. Having designed the experiments, we planned and then installed a pilot setup: 20 transparent glass vessels with the volume of 1 liter equipped with an aeration system. A schematic of the installation is shown in Fig. 2. A 40 ppm solution of methylene blue was made by the addition of the pollutant to the nutritive solution. After filling the vessels with 750 ml of solutions with different concentrations, the pH was adjusted by a buffer solution, and the vessels were filled with Spirodela polyrrhiza. The experiments were performed to closely provide the original greenhouse condition via a uniform irradiation of sunlight achieved by the daily replacement of containers, uniform aeration, daily temperature control and appropriate air conditioning. Pure water was added to the system in order to make up for water loss due to evaporation. Two vessels with no plant inside were prepared randomly to represent the blank samples. The levels of the three factors for each run along with the final results of the removal are shown in Table 2.

Table 1 Levels of the independent test variables

Variable	Low axial (−2)	Low factorial(−1)	Center (0)	High factorial (+1)	High axial (+2)
A: Initial concentration (ppm)	1.97	3.5	5.75	8	9.53
B: pH	2.98	4	5.5	7	8.02
C: Plant weight (g)	3.49	4	4.75	5.5	6.01

---

Fig. 2 A schematic diagram of the pilot.

Table 2 Central composite design (CCD)

Run	A: Initial concentration (ppm)	B: pH	C: Plant weight (g)	Real removal (%)	Predicted removal (%)
1	5.75	5.5	4.75	90.15	90.47
2	5.75	5.5	4.75	92.42	90.47
3	5.75	5.5	4.75	88.90	90.47
4	5.75	5.5	4.75	91.32	90.47
5	5.75	5.5	3.49	81.70	84.41
6	5.75	8.02	4.75	76.31	71.81
7	5.75	5.5	6.01	97.26	96.53
8	5.75	2.98	4.75	57.38	58.07
9	8	7	5.5	81.62	80.46
10	8	4	5.5	65.38	64.74
11	8	7	4	80.84	78.92
12	8	4	4	66.00	63.20
13	3.5	7	5.5	94.73	97.32
14	3.5	7	4	85.48	84.44
15	3.5	4	4	77.13	78.11
16	3.5	4	5.5	90.75	96.46
17	1.97	5.5	4.75	98.63	96.54
18	9.5	5.5	4.75	66.42	69.86

---

2.4 Sampling and analytical methods

On a daily basis, samples were extracted by a syringe, centrifuged for 20 minutes at 4500 rpm, and spectrophotometrically examined at the maximum wavelength of 670 nm. The daily percentage removal of methylene blue was determined by the following equation:¹⁴where C and C₀ are the initial and final concentration of methylene blue, respectively. The daily values are listed in Table 2.

3. Results and discussion

Upon the completion of experiment, a sample of Spirodela polyrrhiza plant abundant in dye together with a fresh Spirodela polyrrhiza plant was selected for X-ray microscopy. As seen in Fig. 3, absorption of dye extended from the roots to the xylem to a degree that saturation was almost achieved. It can also be seen that the plant adapted to the new environment and started to grow small colorless sprouts adjacent to the old colored leaves. As can be seen in Fig. 4, the rate of removal in the first two days was high, but it dwindled thereafter. This was probably due to the germination and reproduction of the plant. The highest removal rate was achieved when low concentrations of pollutant, a roughly neutral pH, and the highest weight of the plant were adopted.

Fig. 3 Microscopic images for Spirodela polyrrhiza (a) before test and (b) after test.

Fig. 4 Dye removal rate in seven days.

3.1 Analysis of variance

The analysis of variance (ANOVA) is concerned with the determination of the impact factor of variable parameters. Table 3 is known as ANOVA table for responses. The F value, which is a statistical measure used to compare different models, is the ratio of factor variance to error variance. A factor is likely to be insignificant provided its corresponding F value approaches unity. In our case of study, the F value of 41.76 for the proposed model indicates an effective model. The statistical measures of ANOVA for each factor along with the F and p values are compared in Table 3, according to which factors A, B, C and interactions AB, AC, A² and C² are of considerable significance. The R² values for both original and reduced models, obtained after the removal of insignificant terms, are 0.9792 and 0.9786, respectively. A lack of fit with the value of 2.91 ensures the high accuracy of the developed model.

Table 3 Statistical measures of the ANOVA for the removal of methylene blue by Spirodela polyrrhiza

Source	Sum of square	Mean square	F value	p-value prob > F
Model	0.25	0.027	41.76	<0.0001
A	0.086	0.086	131.88	<0.0001
B	0.041	0.041	63.51	<0.0001
C	0.018	0.018	27.18	0.0008
AB	4.409 × 10^−3	4.409 × 10^−3	6.75	0.0317
AC	6.420 × 10^−3	6.420 × 10^−3	9.83	0.0139
BC	1.122 × 10^−4	1.122 × 10^−4	0.17	0.6894
A^2	8.806 × 10^−3	8.806 × 10^−3	13.18	0.0067
B^2	0.084	0.084	128.70	<0.0001
C^2	2.802 × 10^−5	2.802 × 10^−5	0.043	0.8411
Residual	5.225 × 10^−3	6.532 × 10^−4
Lack of fit	4.537 × 10^−3	9.074 × 10^−4	3.95	0.1436
Pure error	6.885 × 10^−3	2.295 × 10^−4
Core total	0.25

---

A normal probability curve is plotted against the standardized residuals in Fig. 5. The points approaching the straight line in the diagram represent the normal distribution of errors with an average value of either zero or infinitesimal. In addition, the residuals shown in Fig. 6 illustrate a randomly scattered structure, which is uniformly distributed above and below the horizontal axis and denote the sufficiency of the model. The reduced equation of removal rate is as follows:

Removal (%) = 90 − 7.9A + 5.5B + 3.6C + 2.3AB − 2.8AC − 2.6A² − 8.1B² (2)

Fig. 5 Normal probability against standardized residuals (ANOVA).

Fig. 6 Residual for statistical model.

3.2 Effect of MB initial concentration pH and their interaction

The effect of the initial concentration of methylene blue on percentage removal is shown in Fig. 7(a). An increase in concentration from 3.5 to 8 ppm results in the reduction of percentage removal from 95% to 78% probably due to slow growth of plant as a consequence of increased concentration. As discussed previously, in cases where growth and reproduction differs, removal rate changes dramatically during the experiments. Fig. 7(b) shows the effect of pH on the removal percentage. With an increase of pH from 4 to 7, the percentage removal increases from about 77% to 90%. For pH values higher than 5.5, a further increase in pH decreases the percentage removal. It can be inferred that increasing the pH up to a certain point brings about a positive change in removal rate, which is then succeeded by an unsought decrease. This can justified by the inadequate plant growth in high and low pH, and the fact that plant possesses a higher removal capacity when the pH is neutral. Fig. 8 shows the 3D surface and contour plot of the interaction between initial MB concentration and pH at a constant plant weight of 4.75 g. The maximum removal percentage was obtained in the minimum concentration of the pollutant and a pH value of 6. In order to set the optimal pH, the significance of the other two factors must be carefully investigated.

Fig. 7 The main effects of independent variables on the removal (%): (a) initial concentration of MB and (b) pH.

Fig. 8 The effect of the Initial concentration of MB and pH on the removal rate (%): (a) 3D plot and (b) contour plot.

3.3 Effect of plant load and its interaction with initial concentration of MB

Fig. 9 shows the effect of plant weight on the percentage removal of methylene blue. When the plant weight increased from 4 to 5.5 g, the removal rate increased from 86% to 93%. According to the results reported in Fig. 9, increased plant load clearly has a positive effect on the percentage removal of MB. This might be due to an improved capacity of the plant's xylem for the absorption of the pollutant and also better plant growth. Results also indicate that by increasing the initial concentration of MB, the rapid drop in percentage removal can be balanced by an increase in plant load.

Fig. 9 The effect of plant weight on the removal rate (%).

3.4 Modeling

Recently, simulation of novel systems with unknown mechanisms by means of input and output data of the process has become widely practiced. Artificial neural network, which is conceptually founded on the biological behavior of the brain, is one of the current methods applied to study bioprocesses.¹⁴ It is usually presented as a system of interconnected neurons that compute output values by feeding the network through the input information. In other words, ANNs ease the simulation of processes wherein no physical conception or mechanism is at hand. Several ANN training methods have been developed, amongst which back-propagation (BP) method has been extensively used. Taking advantage of this method, Khataee et al.¹⁴ modeled a duckweed-assisted degradation of AB92 in aqueous solution with acceptable results.

3.4.1 Selection of back-propagation (BP) algorithm and optimum neurons number. In the first place, ten different back-propagation algorithms were compared to select the best algorithm. An artificial neural network model with a hidden layer consisting of ten neurons, each coupled with a distinct tangent sigmoid transfer function (transig) and a linear transfer function-associated output layer was proposed. Afterwards, the model was trained by all back-propagation algorithms, and the Levenberg–Marquardt minimization search method with minimum mean square error (MSE) was found to be the best algorithm (Table 4). Subsequently, the number of neurons in the hidden layer varied from 2–20, and the corresponding errors were compared to find a suitable configuration. Fig. 10 plots the error against the number of neurons. As can be seen, the MSE value follows a descending trend and levels off around 15 to a point that no significant change is observable. Although a configuration with ten neurons was reasonable, 15 neurons were preferred in order to obtain more accurate results, even at the expense of higher complexity. Fig. 11 shows the structure of an optimized ANN model with 15 neurons in the hidden layer.

Table 4 Comparison of different back-propagation algorithms

Back-propagation algorithm	Function	Mean square error	Epoch	Correlation coefficient (R^2)	Best linear equation
Levenberg–Marquardt back propagation	Trainlm	7.92 × 10^−6	12	0.9913	Y = 0.98T + 0.0037
Scaled conjugate gradient back propagation	Trainscg	3.89 × 10^−5	50	0.9627	Y = 0.92T + 0.021
BFGS quasi-Newton back propagation	Trainbfg	3.29 × 10^−5	85	0.9553	Y = 0.89T + 0.028
One-step secant back propagation	Trainoss	1.78 × 10^−4	73	0.8534	Y = 0.8T + 0.05
Batch gradient descent back propagation	Traingd	2.38 × 10^−3	1000	0.4623	Y = 0.61T + 0.038
Variable learning rate back propagation	Traingdx	5.4 × 10^−5	264	0.9440	Y = 0.89T + 0.026
Batch gradient descent with momentum back propagation	Traingdm	4.53 × 10^−5	256	0.9597	Y = 0.93T + 0.018
Fletcher–Reeves conjugate gradient back propagation	Traincgf	3.18 × 10^−5	58	0.9641	Y = 0.97T + 0.009
Polak–Ribiere conjugate gradient back propagation	Traincgp	2.12 × 10^−5	42	0.971	Y = 0.96T + 0.011
Powell–Beale conjugate gradient back propagation	Traincgb	3.52 × 10^−5	45	0.9612	Y = 0.91T + 0.022

---

Fig. 10 Dependence of MSE on the number of neurons (ANN model).

Fig. 11 A schematic of the proposed ANN-based model.

3.4.2 Training and validation of the model. To avoid data scatter, input and output data were normalized:where x stands for either input or output data, and θ is a dimensionless variable. Fig. 12 shows the regression analysis of the proposed network and compares the actual values to those predicted by the model (Y = 0.99T + 0.0027, R = 0.9931 and MSE = 2.8 × 10⁻⁶). A correlation coefficient near one assures a good agreement between predicted and experimental values.

Fig. 12 Comparison between experimental data and the predicted results by back-propagation neural network.

3.4.3 Sensitivity analysis. One of the most remarkable advantages of an intelligent neural network is its ability to determine the most important input variables. In this study, two different methods for the evaluation of relative sensitivity of input variables were investigated and the results were compared. The first method, Garson equation,³¹ is based on the neural weights matrix. According to this method, the relative sensitivity of the input variable (I_j) is evaluated by eqn (4), which is a function of the weight of connections between neurons in each layer:where N_i, N_h and W are the number of inputs neurons, hidden neurons and connectivity weight function, respectively. Subscripts i, h and o represent the input, hidden and output layers, while k, m and n stand for input, hidden and output neurons, respectively. By introducing the connection weights, Garson equation (Table 5) was solved for relative importance. Table 6 shows the relative importance of each variable according to which time and initial concentration of pollutant give the most important effects. The effects of pH and plant weight are less significant, although they cannot be neglected; thus, all variables and their interactions significantly affect the phytoremediation process. The second investigation of sensitivity analysis was done by performance evaluation³² of the possible combinations of variables. Results of the performance evaluation for combinatorial groups of one to four variables, optimized by an ANN model are summarized in Table 7. Fig. 13 shows the trend of MSE for all combinations. As can be seen in Fig. 13, the highest and lowest errors were generated by the 4-parameter group and single parameter group respectively.

Table 5 Connection weights between input and hidden layer (W₁) and between hidden layer and output layer (W₂)

Neuron	W1				W2
	Initial variables				Output
	Time (day)	Initial concentration (ppm)	pH	Plant weight (g)	Removal (%)
1	1.3484	1.7998	0.7481	2.2180	0.0814
2	0.1313	0.7377	−0.0462	−2.22119	−0.3763
3	−1.6581	0.4460	−1.4419	−1.1351	0.0500
4	−1.4362	0.5092	1.3919	2.0266	−0.2127
5	−1.2144	−1.1366	0.9267	2.0949	0.1027
6	0.0677	3.1304	−2.7372	−0.5845	−0.6487
7	2.6983	−0.8047	−0.2210	0.3807	−0.0837
8	−1.3796	−0.4249	1.6627	0.8003	−0.1409
9	0.00223	0.5596	1.3447	0.2352	0.5852
10	0.1665	1.7340	−1.1186	0.6784	0.5052
11	−0.0253	−3.2792	2.2829	2.3252	0.5926
12	0.0460	−0.2390	−1.9385	−1.377	0.4093
13	2.6970	1.0257	−0.2170	0.1804	0.5389
14	−0.5016	1.1988	1.2975	−1.4991	0.0954
15	−2.0834	1.1502	0.4152	−0.7845	−0.3626

---

Table 6 Relative importance for input variables

Initial variables	Importance (%)
Time (day)	47
Initial concentration (ppm)	25.93
pH	9.37
Plant weight (g)	17.7
Total	100

---

Fig. 13 Best performance of every group of parameters. The * sign represents the best group of parameters.

Table 7 Performance evaluation for the combination of input variables^a

Combination	Mean square error (MSE)	Epoch	Correlation coefficient (R^2)	Best linear equation
a (a) Time. (b) Initial concentration of MB. (c) pH. (d) Plant weight. The * sign represents the best group of parameters.
*a	3.29 × 10^−4	2	0.5683	Y = 0.29T + 0.12
b	4.25 × 10^−4	2	0.4669	Y = 0.25T + 0.2
c	4.65 × 10^−4	3	0.3442	Y = 0.17T + 0.22
d	4.38 × 10^−4	2	0.4270	Y = 0.21T + 0.21
*a + b	2.18 × 10^−4	8	0.7457	Y = 0.5T + 0.13
a + c	2.68 × 10^−4	8	0.604	Y = 0.44T + 0.14
a + d	2.38 × 10^−4	8	0.7140	Y = 0.47T + 0.16
b + c	2.66 × 10^−4	5	0.6390	Y = 0.41T + 0.1
b + d	3.37 × 10^−4	4	0.6360	Y = 0.42T + 0.15
c + d	3.13 × 10^−4	4	0.5946	Y = 0.40T + 0.17
a + b + c	1.61 × 10^−4	14	0.8833	Y = 0.68T + 0.065
a + b + d*	1.45 × 10^−4	9	0.9213	Y = 0.78T + 0.05
a + c + d	1.73 × 10^−4	8	0.8593	Y = 0.64T + 0.088
b + c + d	1.92 × 10^−4	8	0.8283	Y = 0.61T + 0.089
a + b + c + d*	2.80 × 10^−4	14	0.9931	Y = 0.99T + 0.0027

---

4. Conclusion

In the present study, the phytoremediation of methylene blue by Spirodela polyrrhiza was investigated. The elimination of methylene blue by duckweed and the effects of initial concentration of pollutant, pH and plant weight on the process of elimination were studied over a 7 day time span. The process was also simulated by a statistical and ANN-based model.

Results indicate that Spirodela polyrrhiza is highly capable of eliminating the pollutant from aqueous solution by adsorption and accumulation mechanisms. At the beginning of the experiments, the rate of elimination was very high for all samples. After 2 days of adaptation to the new environment, the plant kept growing and the percentage removal of methylene blue increased up to 97% in some samples despite the plant being saturated. Although pH values lower than 4 inflict significant damage on the plant, the absorption continued. Elimination efficiency of the plant increased at higher pHs, and the best results were obtained in the neutral pH range of 6–7. As expected, in addition to the percentage removal of the pollutant, the rate of growth and reproduction were higher when high amounts of plant were used. The results obtained indicate that the Spirodela polyrrhiza tissues, either alive or dead, are suitable to be used as an absorbent for decolorization. It was also observed that the percentage removal of methylene blue is more effective in samples with low concentration, pH ranging from 6–7 and high plant weight.

The experiments were statistically designed and modeled by response surface methodology. The statistical analysis of the experimental data, which is tabulated in the ANOVA table, indicates that the most important factor affecting the elimination process is the initial concentration of the pollutant (F value = 147). Results by the statistical design, analysis of the process factors and their interaction, and also comparison of the actual values with the results predicted confirm the suitable accuracy of the proposed model. An intelligent neural network model configured by a 15-neuron hidden layer was proposed to model the experimental data. The proposed model can accurately model the experimental data. The evaluation of the sensitivity of results by Garson suggests that the most important factor of the process is time.

References

1. G. De Marsily, Ecohydrol. Hydrobiol., 2007, 7, 147–155.
2. L. R. Gravelet-Blondin, S. J. Barclay, C. M. Carliell and C. A. Buckley, Water Sci. Technol., 1997, 36, 303–310.
3. F. El-Gohary and A. Tawfik, Desalination, 2009, 249, 1159–1164.
4. V. Golob, A. Vinder and M. Simonič, Dyes Pigm., 2005, 67, 93–97.
5. N. Al-Bastaki, Chem. Eng. Process., 2004, 43, 1561–1567.
6. C. Suksaroj, M. Héran, C. Allègre and F. Persin, Desalination, 2005, 178, 333–341.
7. S. Karcher, A. Kornmüller and M. Jekel, Water Res., 2002, 36, 4717–4724.
8. A. Pala and E. Tokat, Water Res., 2002, 36, 2920–2925.
9. S.-F. Kang and H.-M. Chang, Water Sci. Technol., 1997, 36, 215–222.
10. M. Pérez, F. Torrades, X. Domènech and J. Peral, Water Res., 2002, 36, 2703–2710.
11. K. Lu, X.-L. Zhang, Y.-L. Zhao and Z.-L. Wu, J. Hazard. Mater., 2010, 182, 928–932.
12. K. Amjad Ali and H. Qayyum, J. Environ. Sci., 2007, 19, 396–402.
13. E. Forgacs, T. Cserháti and G. Oros, Environ. Int., 2004, 30, 953–971.
14. A. R. Khataee, A. Movafeghi, S. Torbati, S. Y. Salehi Lisar and M. Zarei, Ecotoxicol. Environ. Saf., 2012, 80, 291–298.
15. A. V. Patil and J. P. Jadhav, Chemosphere, 2013, 92, 225–232.
16. A. I. Zouboulis and P. A. Moussas, in Encyclopedia of Environmental Health, ed. O. N. Jerome, Elsevier, Burlington, 2011, pp. 1037–1044.
17. M. R. T. Palmroth, J. Pichtel and J. A. Puhakka, Bioresour. Technol., 2002, 84, 221–228.
18. D. Pant and A. Adholeya, World J. Microbiol. Biotechnol., 2009, 25, 1793–1800.
19. H. Deng, J. Lu, G. Li, G. Zhang and X. Wang, Chem. Eng. J., 2011, 172, 326–334.
20. R. Abdallah and S. Taha, Chem. Eng. J., 2012, 195–196, 69–76.
21. T. Horvat, Ž. Vidaković-Cifrek, V. Oreščanin, M. Tkalec and B. Pevalek-Kozlina, Sci. Total Environ., 2007, 384, 229–238.
22. Q. Wang, Y. Cui and Y. Dong, Acta Biotechnol., 2002, 22, 199–208.
23. P. Waranusantigul, P. Pokethitiyook, M. Kruatrachue and E. S. Upatham, Environ. Pollut., 2003, 125, 385–392.
24. S. Pandey, R. K. Asthana, A. M. Kayastha, N. Singh and S. P. Srngh, J. Plant Physiol., 1999, 154, 634–640.
25. B. Ebrahimi, S. Shojaosadati and S. Mousavi, Biotechnol. Bioprocess Eng., 2010, 15, 383–391.
26. F. Amiri, S. M. Mousavi, S. Yaghmaei and M. Barati, Biochem. Eng. J., 2012, 67, 208–217.
27. W. Drost, M. Matzke and T. Backhaus, Chemosphere, 2007, 67, 36–43.
28. S. Megateli, S. Semsari and M. Couderchet, Ecotoxicol. Environ. Saf., 2009, 72, 1774–1780.
29. The water culture method for growing plants without soil, ed. D. R. Hoagland and D. I. Arnon, College of Agriculture, University of California, California, Berkeley, 1950.
30. R. Mafi Gholami, S. M. Mousavi and S. M. Borghei, J. Ind. Eng. Chem. (Amsterdam, Neth.), 2012, 18, 218–224.
31. A. Aleboyeh, M. B. Kasiri, M. E. Olya and H. Aleboyeh, Dyes Pigm., 2008, 77, 288–294.
32. K. Yetilmezsoy and S. Demirel, J. Hazard. Mater., 2008, 153, 1288–1300.

---