Electronic Supplementary Material s46

1 Electronic Supplementary Material

3 Magnetic metal-organic frameworks for the extraction of trace amounts of heavy metal ions

4 prior to their determination by ICP-AES

5 Meysam Safari, Yadollah Yamini1, Mohammad Yaser Masoomi, Ali Morsali, Ahmad Mani-

6 Varnosfaderani

7 Department of Chemistry, Tarbiat Modares University, P.O. Box 14115-175, Tehran, Iran

10 Characterization of the sorbents

11 To gain a better understanding of the morphology Fe3O4@TMU-8, the samples were also

12 characterized using SEM and TEM (Fig. S2A). The SEM images show that micro flower like

13 morphologies have been obtained which are comprised of magnetic nanoparticles, and TEM image

14 revealed that the finally formed Fe3O4@TMU-8 are composed of a Fe3O4 core and a TMU-8 shell.

15 Based on the SEM images, the average diameter of the Fe3O4@TMU-8 studied here is about 40 nm.

16 The comparison between the TGA curves of TMU-8 and Fe3O4@TMU-8 showed the same thermal

17 stability for both of MOFs (Fig. S2B).

18 The magnetic hysteresis loops of the Fe3O4@TMU-8 samples were obtained from VSM

19 measurement at room temperature. Fe3O4, Fe3O4@TMU-8, 5-cycles, 10-cycles, 15-cycles and 20-

20 cycles showed a characteristic superparamagnetic feature with negligible hysteresis at room

21 temperature. Fig. 2SC shows the magnetic properties of Fe3O4 NPs and Fe3O4@TMU-8 at room

22 temperature obtained by VSM within the field range of −10 to 10 KOe. The MOF shell results in the

23 decrease of the magnetic strength of the nanoparticles due to the weight contribution from the

24 nonmagnetic MOF. The large saturation magnetization of Fe3O4, Fe3O4@TMU-8 5-cycles, 10-

-1 25 cycles, 15-cycles and 20-cycles were 65, 53, 41, 25, and 17 emug , respectively. After dispersing 20

26 cycles in water, the Fe3O4@TMU-8 was collected within 15 s using a magnet. This indicates that

27 the magnetic microspheres can be used for magnetic separation.

28 Optimization of extraction conditions

1 1Corresponding author. Phone: +98-21-82883449, Fax: +98-21-82883460, E-mail address:

2 [email protected] (Y. Yamini). 29 The effect of the number of layers on the composite plays a significant role in the adsorption of

30 target metal ions onto Fe3O4@TMU-8. Therefore, the effect of the number of layers on the

31 composite for the extraction of heavy metal ions was investigated. The extraction of the heavy

32 metals increased rapidly when the layers on the composite increased from 3 to 15 and changed

33 slightly when the layers on the composite increased from 15 to 20, which indicated the extraction of

34 heavy metal ions was inadequate when the layers on the composite was smaller than 15. Heavy

35 metal ions were adequately adsorbed on the sorbent when 15 layers was used on the composite and

36 thus the extraction of heavy metal ions changed slightly with a further increase of layers on the

37 composite. To ensure sufficient extraction, 15 layers on the composite were selected for synthesized

38 Fe3O4@TMU-8.

39 The solution pH plays a critical role on the extraction of the analytes. The solution pH not only

40 affects the molecular status of analytes but also influences the charge species and charge density on

41 the surface of the sorbent. The effect of the solution pH on the extraction of heavy metal ions was

42 investigated in a pH range of 4.0 to 11.0. The extraction efficiency increased dramatically as the pH

43 was increased from 4 to 10 and then decreased. This effect can be related to the protonation of the

44 sorbent donor atoms at low pH values and the formation and precipitation of hydroxide species of

45 the target metal ions at high pH values (pH > 10), leading to a decrease in the adsorption of metal

46 ions. According to these results, the pH of the sample solution was adjusted at pH 10 for subsequent

47 experiments.

48 The type of desorption solvent is vital for the desorption efficiency. Thus, the choice of

49 desorption solvent should be carefully taken into account. The effect of the concentration of HNO3

50 solution on the efficiency of desorption of the metal ions was studied in the range of 0.025–1 M.

51 The best desorption of the metal ions is attained by application of 0.5 M HNO3 as the elution

52 solvent.

53 In this approach, the central composite design strategy was used for designing the experiments, and

54 then the BRANN technique was used for nonlinear mapping of the data. Despite other ANN

55 training algorithms, BRANN is more reproducible and predictive. The main advantages of this

56 approach are its reproducibility, the ability to model multiple responses, modeling the nonlinear

57 interactions and unraveling hidden dependencies in data. By using the ED-BRANN strategy, it is

58 possible to investigate the effect of the factors and their interactions on the extraction efficiency of

59 each analyte separately. This advantage was not achievable using a usual polynomial equation

60 commonly used in ED approaches. Moreover, BRANN models are automatically regulated by using

61 the Bayes theorem, which prevents the occurrence of overfitting. Therefore, the generated models

62 are predictive and reliable. In this work a BRANN model with a 4-5-7 architecture has been used

63 for modeling the data obtained after running the CCD experiments 64 As previously mentioned, the effects of the sorbent, extraction time, eluent and sample volume

65 have been investigated using ED-BRANN approach. In this method, the coded levels of the factors

66 for 21 experiments (i.e. data in Table S1) were used as independent variables. Moreover, the peak

67 areas (PAs) values for all 7 analytes have been used as dependent variables. A 4-5-7 network was

68 developed and trained with Bayesian regularization algorithm. The transfer function for the hidden

69 layer and the output layer were sigmoidal and linear, respectively. The statistical parameters for this

70 model are summarized in Table S2. As can be seen in this Table, the model explains more than 90%

71 of the variance of the responses for each analyte. High values of the coefficient of determination

72 together with low values of root mean square error (RMSE) imply that the developed BRANN

73 model has a high potential to describe the experiments. The relative importance of the main effects

74 and their interactions have been investigated using a sensitivity analysis approach [26]. The relative

75 contribution of the sorbent, extraction time, sample volume, eluent and their interactions on the

76 average PAs values of the analytes are illustrated in Fig. 2SB. As can be seen in this figure, the

77 amount of the sorbent, the extraction time, and their interaction are the more contributing factors in

78 this model. All analytes represent the same pattern of responses. The effects of the interaction of the

79 sorbent and extraction time on the PAs values of all seven analytes are illustrated in Fig. 3SA. As

80 can be seen in this figure, there is a positive synergism effect between the extraction time and the

81 amount of sorbent. With increasing the extraction time and the amount of the sorbent, the PAs for

82 all seven analytes increases. By developing the nonlinear ANN model, the pattern search strategy

83 [27] was used to find the best values of the factors in order to get the maximum PAs values for each

84 analyte. A constrained multivariate optimization strategy (Fig. 3SB) has been used to optimize the

85 objective function (inverse of the summation of PAs) under the constraint which the relative

86 standard deviations of the PAs were less than a small threshold. This criterion guarantees that the

87 experiment is optimized in a way which the PA of all analytes are maximized, simultaneously.

88 After the optimization procedure, best values of the parameters were 10 mg for the sorbent, 185 mL

89 of sample volume, 566 µL of eluent and 11 min of extraction time.

92 Table S1 Experimental factors, levels and matrix of the face-centered central composite design

93 (FCCCD) for determination of heavy metal ions by MSPE.

No Time (min) Eluent (µL) Sample Sorbent (mg) volume (mL)

1 3.00 200.00 180.00 4.00 2 6.00 350.00 135.00 7.00

3 6.00 602.27 135.00 7.00

4 6.00 350.00 210.68 7.00

5 3.00 200.00 90.00 4.00

6 9.00 200.00 180.00 10.00

7 6.00 97.73 135.00 7.00

8 9.00 500.00 90.00 4.00

9 6.00 350.00 59.32 7.00

10 11.05 350.00 135.00 7.00

11 6.00 350.00 135.00 12.05

12 0.95 350.00 135.00 7.00

13 6.00 350.00 135.00 7.00

14 3.00 500.00 90.00 10.00

15 6.00 350.00 135.00 7.00

16 3.00 500.00 180.00 10.00

17 6.00 350.00 135.00 7.00

18 6.00 350.00 135.00 1.95

19 9.00 200.00 90.00 10.00

20 6.00 350.00 135.00 7.00

21 9.00 500.00 180.00 4.00

95 96 Table S2 The statistical parameters of the developed 4-3-7 BRANN model, for nonlinear

97 modeling of central composite design experiments.

2 2 1 2 3 Analyte R training R MC-CV RMSE training RMSE MC-CV

Co 0.904 0.890 0.474 0.494

Cu 0.971 0.939 0.192 0.218

Pb 0.943 0.922 0.456 0.482

Cd 0.962 0.930 0.151 0.180

Ni 0.916 0.925 0.238 0.263

Cr 0.918 0.902 0.218 0.227

Mn 0.941 0.913 0.097 0.135

98 MC-CV: Monte Carlo-Cross Validation, 2- RMSE: Root mean square error, 3- The results for MC-

99 CV are the average values for 100 times repetition of the algorithm.

Table S3 Determination of heavy metal ions and recoveries for tap, river and minerals water samples.

Tap Water River Water

Element Added(μg L-1) Found ± SD (μg L-1) Recovery (%) Added (μg L-1) Found ± SD (μg L-1) Recovery (%) Added(μg L

10.0 10.03 ± 0.45 100.3 10.0 14.3 ± 0.43 99.0

0.0 2.9 ± 0.12 - 0.0 4.9 ± 0.31 - Cu 10.0 12.8 ± 0.34 99.0 10.0 14.6 ± 0.42 97.0

Mn 0.0 3.4 ± 0.13 - 0.0

118 Table S3 (continue) Determination of heavy metal ions and recoveries for fruits and tea

Element C u Pb

Ad Found R Added Found R.R Added Found R.R de . d ±SD (μg L-1) ±SD (%) (μg L-1) ±SD (%)

(μg (μg L-1) (μg L-1) (μg L-1) L- 1)

Tomato 0 - _ 0.0 - _ 0.0 - _ .0

1 10.1 ± 0.8 1 10.0 9.0 ± 90.0 10.0 9.6 96.0 0.0 0 0.5 ±0.3

Tea 0 - - 0.0 - _ 0.0 2.9 ± - .0 0.5

1 9.7 ± 0.3 9 10.0 8.8 ± 88.0 10.0 12.8 ± 99.0 0.0 7 0.4 0.8

Watermelon 0 - _ 0.0 - - 0.0 - - .0

1 9.7 ± 0.4 9 10.0 9.1 ± 91.0 10.0 9.6 ± 96.0 0.0 7 0.6 0.4

Cucmber 0 - - 0.0 - - 0.0 - - .0

1 9.5 ± 0.3 9 10.0 9.2 ± 92.0 10.0 10.2 ± 102.0 0.0 5 0.3 0.7

0 2.4 ± 0.2 - 0.0 - - 0.0 1.4 ± - onion .0 0.2

1 12.4 ± 0.7 9 10.0 9.4 ± 94.0 10.0 11.3 ± 99.0 0.0 9 0.7 0.7

0 1.8 ± 0.3 _ 0.0 - _ 0.0 - _ .0 Carrot 1 11.9 ± 0.6 1 10.0 10.0 ± 100.0 10.0 9.5 ± 95.0 0.0 0 0.3 0.4

Apple 0 2.1± 0.4 - 0.0 - - 0.0 - _ .0

1 12.0 ± 0.6 9 10.0 9.0 ± 90.0 10.0 9.6 ± 96.0 0.0 9 0.3 0.5

121 122

124 Fig. S1. Comparison of XRD patterns for TMU-8, Fe3O4@TMU-8 core-shell and

125 Fe3O4/TMU-8 composite.

137 Fig. S2. FT-IR spectra of TMU-8, Fe3O4/TMU-8 and Fe3O4@TMU-8

138 139 A)

140 B)

146 C)

152 Fig. S3. A) SEM images of Fe3O4@TMU-8 NPs and TEM image of Fe3O4@ TMU-8 NPs.

153 B) TGA thermograms of Fe3O4@TMU-8 and TMU-8. C) VSM curves of Fe3O4; 5, 10,15 and

154 20 cycels of Fe3O4@TMU-8 155 A)

156 B)

164 Fig. S4. A) The interaction patterns of the amount of the sorbent and extraction time on the 165 area under peak values of the analytes. a) Cr, b) Co, c) Mn, d) Ni, e) Cu, f) Cd and g) Pb. B) 166 The plot of the values of fitness function against the number of generations for the 167 optimization procedure using constrained genetic algorithm (GA) and BRANN technique. 168 The fitness function equals to (1/) and is minimized using GA.