Migration kinetics of primary aromatic amines from polyamide kitchenware: Easy and fast screening procedure using fluorescence

Talanta 160 (2016) 46–55

Contents lists available at ScienceDirect

Talanta journal homepage: www.elsevier.com/locate/talanta

Migration kinetics of primary aromatic amines from polyamide kitchenware: Easy and fast screening procedure using fluorescence S. Sanllorente a, L.A. Sarabia b, M.C. Ortiz a,n a b

Department of Chemistry, Faculty of Sciences, University of Burgos, Plaza Misael Bañuelos s/n, 09001 Burgos, Spain Department of Mathematics and Computation, Faculty of Sciences, University of Burgos, Plaza Misael Bañuelos s/n, 09001 Burgos, Spain

art ic l e i nf o

a b s t r a c t

Article history: Received 18 May 2016 Received in revised form 24 June 2016 Accepted 26 June 2016 Available online 27 June 2016

Primary aromatic amines, PAAs, and their derivatives constitute a health risk and control of their migration from food contact materials is the subject of permanent attention by the authorities. 25.1% of notifications made by Rapid Alert System for Food and Feed in the European Union between 2010 and 2015 concerned PAAs, polyamide cooking utensils being a common source. It is thus useful to have fast and efficient analytical methods for their control. In this work a non-separative, easy, fast and inexpensive spectrofluorimetric method based on the second order calibration of excitation-emission fluorescence matrices (EEMs) was proposed for the determination of aniline (ANL), 2,4-diaminotoluene (2,4-TDA) and 4,4′-methylenedianiline (4,4′-MDA) in polyamide cooking utensils. The procedure made it possible to identify unequivocally each analyte. Trilinearity of the data tensor guarantees the uniqueness of the solution obtained through parallel factor analysis (PARAFAC), so the factors of the decomposition match up with the analytes. The three analytes were unequivocally identified by the correlation between the pure spectra and the PARAFAC excitation and emission spectral loadings. The recovery percentages found were, 82.6%, 112.7% and 84.4% for ANL, 2,4-TDA and 4,4′-MDA respectively. The proposed method was applied to carry out a migration test from polyamide cooking utensils, using a 3% (w/v) acetic acid in aqueous solution as food simulant. Detectable levels of 4,4′-MDA were found in food simulant from some of the investigated cooking utensils. Finally, a kinetic model for the migration of 4,4′-MDA has been fitted to experimental data obtained in the migration test. Thanks to the selectivity of PARAFAC calibration, which greatly simplifies sample treatment avoiding the use of toxic solvents, the developed method follows most green analytical chemistry principles. & 2016 Elsevier B.V. All rights reserved.

Keywords: Primary aromatic amines Excitation-emission fluorescence matrices PARAFAC Migration test Unequivocally identification Black nylon utensils

1. Introduction PAAs are a group of compounds that have been used, for many decades, as raw materials and intermediates in a wide variety of

Abbreviations: PAAs, primary aromatic amines; ANL, aniline; 2,4-TDA, 2,4-diaminotoluene; 4,4′-MDA, 4,4′-methylenedianiline; PARAFAC, parallel factor analysis; MDI, methylene diphenyl diisocyanate; FCMs, food contact materials; WHO/IARC, World Health Organization and International Agency for Research on Cancer; LCMS/MS, liquid chromatography-tandem mass spectrometry; UHPLC-MS, ultrahigh-performance liquid chromatography with mass spectrometric; LC–HRMS, liquid chromatography–orbitrap-high resolution mass spectrometry; UPLC-MS/MS, liquid chromatography coupled with triple quadrupole mass spectrometry; PS, polystyrene; PE, polyethylene; HS-SPME-GC/MS, headspace solid-phase microextraction and gas chromatography coupled with mass spectrometry; DART-MS, direct analysis in real time mass spectrometry; RASFF, Rapid Alert System for Food and Feed; EEM, excitation–emission fluorescence matrix; CCα, decision limit; CCβ, capability of detection n Corresponding author. E-mail address: [email protected] (M.C. Ortiz). http://dx.doi.org/10.1016/j.talanta.2016.06.060 0039-9140/& 2016 Elsevier B.V. All rights reserved.

production processes, including pesticides, textiles, and paints (azo-dyes), polymers, pharmaceuticals and cosmetics [1]. As a consequence, many industrial sectors use PAAs. According to the Ref. [2] this use can be distributed as: 25% in the chemical industry, 14% in the industry of polymers and rubber, 8% in agriculture, 9% in adhesives and paintings, 7% in the textile industry, 8% in the pharmaceutical industry and minor percentages in the petroleum, cosmetic and paper industries or in construction. As an example of the growth in PAA production there is the case of ANL. This compound is not only an important organic chemical raw material, but also an important product. There are more than 300 kinds of chemical products and intermediates derived from it. Only in the year 2001, in USA, the production of ANL was of 8.67
108 kg [2]. World consumption of ANL grew at an average annual rate of 3% during 2006–10, and by nearly 7% per year from 2009 to 2014. This increase was driven by the strong recovery in most regions and countries since 2010, and growth in methylene diphenyl diisocyanate (MDI) production, which is used to make polyurethane foam, both rigid and flexible. The global

S. Sanllorente et al. / Talanta 160 (2016) 46–55

production of aniline was about 2.97
109 kg in 2000 and now has exceeded 3.00
109 kg (3000 kt/y). It is predicted that the demand for ANL will have a larger increase with the expansion of demand for MDI. Many of the commercial products, PAA based, can be used in the production of food contact materials (FCMs). Additionally, PAAs may also arise from isocyanates that can be used in the manufacture of FCMs, such as polyurethanes, which are frequently used as adhesives in multilayer films. If the production conditions are not carefully monitored, excess residual PAAs can remain in the final product and potentially migrate into food during use of FCM. PAAs are potentially harmful and suspected to cause cancer and have other adverse effects. Several PAAs have been classified as “possibly carcinogenic to humans” by the World Health Organization and International Agency for Research on Cancer (WHO/IARC) [3]. For example, 2,4-TDA and 4,4′-MDA are listed in group 2B (possibly carcinogenic to humans) while ANL has been classified in group 3 (not classifiable as to its carcinogenicity to humans). Hence, their presence in foodstuffs should be generally avoided. According to present legislation, the European Union has set a legal limit on the permitted level of PAA migration from materials intended to come in contact with food. As laid down by Regulation 10/2011 [4] these articles should not release PAAs above a detection level of 10 μg kg  1 in food or food simulant. One common food contact source of PAAs, polyamide cooking utensils, was identified for several years [5]. These utensils, such as turners, whisks and spoons, are popularly used due to their low cost, high temperature resistance and non-scratch properties. The migration of PAAs from polyamide utensils into foodstuffs can occur [6–9] however, due to remaining residues present from the coloring process (using azo-dyes) and co-monomer addition [9–12]. Many techniques have been described in the literature for the determination of PAAs in extracts of FCMs. These included liquid chromatography-tandem mass spectrometry (LC-MS/MS) in migrates from several samples of FCMs finding that unacceptably high amounts of PAAs are released by black nylon kitchen utensils to the food simulant (the'non-detectable’ limit was exceeded by up to 2100 times) [13]. Aznar et al. [14] described a method for the determination of PAAs based on a solid-phase extraction step using cation-exchange cartridges and the subsequent analysis of the extracts by ultra-high-performance liquid chromatography with mass spectrometry. Brauer et al. [15] proposed a screening method using high pressure liquid chromatography (HPLC) to determine specifically the migration of primary aromatic amines from food contact articles. Lambertini et al. [16] developed a reliable liquid chromatography–mass spectrometry method for investigation of PAA migration from food packaging and during industrial curing of multilayer plastic laminates. Recently [17] by means of a liquid chromatography–orbitrap-high resolution mass spectrometry (LC–HRMS) operating in full scan mode a procedure is developed to determine migrating substances in kitchen utensils including also PAAs. In Ref. [17] a comprehensive analytical method based on ultra-performance liquid chromatography coupled with triple quadrupole mass spectrometry (UPLC-MS/MS) has been applied to determine PAAs in polystyrene and polyethylene master batches for foods. Rubio et al. [18] presents a headspace solid-phase microextraction and gas chromatography coupled with mass spectrometry (HS-SPME-GC/MS) method for trace determination of PAAs in water. Prior to extraction the analytes were derivatized in the aqueous solution by diazotation and subsequent iodination. A PARAFAC or PARAFAC2-based calibration model was carried out for each analyte. In [19] a rapid screening method to determine primary aromatic amines in kitchen utensils using direct analysis in real time mass spectrometry (DART-MS) is developed and it could be used

47

to identify potentially volatile samples, thus limiting the number of samples that would require very time consuming migration testing by means of a chromatographic method. All these methodologies are often laboratory-based and timeconsuming and may lead to large capital costs for multisample analysis. However continued vigilance is needed of the migration of PAAs as a result of the frequent violation of the limits stated. For example, notifications on PAAs given by the Rapid Alert System for Food and Feed, RASFF, relating to the migration from kitchen utensils made of nylon mainly imported from China, report an annual number of notifications that range from 18 to 35 since 2010. Specifically, between 17% and 34% of all the alerts correspond to migrants from FCMs. In most cases, ANL and 4–4′ MDA were identified [20]. In this context, molecular fluorescence spectroscopy is of interest due to its high sensitivity, ease of use and especially because cost reduction of every analysis is possible. These advantages, allow one to increase the sample size in routine control without increasing the expense and without to block the official laboratories with a lot of analysis. Nevertheless, the recorded signals proceeding from the compounds of interest can be overlapped by the presence the signals of other molecules or ions; for this reason, the excitation-emission matrices (EEM) need to be analyzed by means of three ways techniques in order to obtain the separation of the signals of the different fluorophores. This separation is possible even in presence of the quenching effect [21]. When PARAFAC is used, with trilinear data, the mathematical decomposition is unique, so that the factors obtained mathematically correspond with the fluorophores present in the sample. Once the trilinearity of the data tensor has been checked, the unequivocal identification of each analyte is possible by means of two independent ways: both the excitation and emission spectra. In addition, it is possible to quantify in presence of other absent compounds in the calibration standards (the ‘second order’ property). In this work the generalized standard addition method was used, adding to the extract obtained in the migration test, ternary mixtures of three PAAs and using as analytical signal the sample loading related to each amine. As a consequence, other fluorophores which have migrated from the analyzed sample, but different from the amines analyzed, are kept almost constant in the successive additions (also their loading sample), and consequently the data tensor loses the trilinearity. To solve this, a procedure consisting of subtracting the factors related to fluorophores of the matrix from the original tensor was proposed, recovering the trilinearity and so one can identify unequivocally and quantify the PAAs in the samples from the test of migration. This procedure has already been used in our laboratory [22] in the determination of pesticides in a complex matrix. Therefore, in this work, a strategy that couples standard addition and the three-way EEM data with “mathematical separation” based on PARAFAC was developed for the unequivocal identification and quantification of the three PAAs, despite the fact that the signals of the three amines were highly overlapped with each other and with the fluorescent matrix constituents. Finally, the recovery and the capability of detection of the method were also studied. Thanks to the selectivity of PARAFAC calibration, which greatly simplifies sample treatment avoiding the use of toxic solvents, the developed method follows most green analytical chemistry principles.

2. Theory 2.1. PARAFAC According to Leurgans and Ross [23], signals coming from

48

S. Sanllorente et al. / Talanta 160 (2016) 46–55

fluorescence spectroscopy can be described by a tensor of data. For the case of spectroscopy, the fluorescence intensity of a i-th sample, i¼1,…, I, irradiated with a radiation beam whose excitation wavelength is λkex, k ¼1,……, K, when the emission of radiation occurs at a emission wavelength λjem, j¼ 1, …, J, is the one described in Eq. (1):

Once the proper number of factors has been decided, the resultant model is validated. The indices Q and Hotelling's T2 have been used to identify outlier samples: if both indices of a sample exceed the threshold value at a 95% confidence level, that sample will be rejected and the PARAFAC model will be estimated again.

F

x (i, j, k ) =

∑ aif bjf ckf ,

3. Materials and methods

f =1

i = 1, 2, …, I; j = 1, 2, …, J; k = 1, 2, …, K

(1)

where aif is the concentration of the fluorophor f in the i-th sample, bjf is the relative emission at the λjem wavelength and ckf is the relative absorption (extinction coefficient) of fluorophor f at the excitation wavelength λkex. The physical model in Eq. (1) corresponds to the trilinear PARAFAC model in Eq. (2), when the fluorescence intensities are arranged as a three-way tensor X with size (I
J
K), so that: F

x (i, j, k ) =

∑ aif bjf ckf

+ eijk ,

f =1

i = 1, 2, …, I;

j = 1, 2, …, J;

k = 1, 2, …, K

3.1. Chemicals and reagents Aniline (CAS no. 62-53-3; ACS reagent; 99.5% minimum purity), 2,4-diaminotoluene (CAS no. 95-80-7; 98% purity), 4,4′-methylenedianiline (CAS no. 101-77-9; 97% minimum purity), were purchased from Sigma-Aldrich (Steinheim, Germany). The chemical structure and abbreviations are showed in Table 1. Glacial acetic acid (CAS no. 64-19-7; HPLC grade) was obtained from Panreac (Barcelona, Spain). s Methanol (CAS no. 67-56-1; for liquid chromatography LiChrosolv ) was supplied by Merck KGaA (Darmstadt, Germany). Deionised water was obtained by using the Milli-Q gradient A10 water purification system from Millipore (Bedford, MA, USA).

(2)

where eijk is the residual no explained by the trilinear model, xijk is the fluorescence intensity of the i-th sample at the emission wavelength λjem, and excitation wavelength λkex. When a tensor of experimental data is compatible with the structure in Eq. (2) it is said that the data are trilinear and the estimation by least squares of the coefficients in Eq. (2) is unique [24]. Experimental data are compatible with a trilinear model if the factors are the same in all samples differing only in the proportion involved in each of them, that is the emission and excitation spectra of the analyte are the same throughout the I samples. In practice, when the data are trilinear the uniqueness means that the estimates must coincide (except for a scale factor) with the sample profiles, the emission spectra and the excitation spectra of the F fluorophors in the sample. A revision of the theoretical aspects of PARAFAC as well as algorithms and analytical applications can be read in Ref. [25]. The most useful application is that it is possible to determine the analyte of interest in the presence of unknown interferents because, in this case, the interferent(s) will appear as new factor(s) without affecting the rest. This property is known in chemical analysis as the 'second order’ property [26]. The uniqueness property of the PARAFAC model and, consequently, the identification of the underlying phenomenon in the data tensor, caused an increase in their use, especially in Chemometrics. Nevertheless, it is still a challenge to determine the appropriate number of components. It is possible to use resampling techniques such as cross-validation [27], or split-half analysis, but both procedures are often unfeasible when the sample mode includes calibration standards, which are few and different to each other, making the division of the tensor in similar parts difficult. Besides, they can be unattractive because of the intense computations involved and owing to the non-sequential fitting of the PARAFAC model. Therefore, a diagnostic based on single analysis, the Core Consistency Diagnostic index, CORCONDIA, which is an index that measures the degree of trilinearity of the experimental data tensor, was developed by Bro and Kiers [28]. It has already been shown that the core consistency greatly improves when the data which does not follow a trilinear structure is deleted (e. g. the Rayleigh scatter or the null fluorescent intensity in an EEM tensor). That is, if an experimental tensor is composed of a tensor with trilinear structure added to another which is not; then, it is possible to recover the trilinearity removing the tensor with non-trilineal structure.

3.2. Standard solutions and samples Stock solutions of each primary aromatic amine were individually prepared in methanol at a concentration of 500 mg L  1. Intermediate solutions for each analyte at 2.5 mg L  1 were prepared daily by dilution with methanol. Next, solutions of three PAAs for the analysis were prepared daily in methanol from the intermediate solutions. All these solutions were stored at low temperature (4 °C) and protected from light. Cooking utensils made of black nylon and intended for contact with warm food were purchased from several local retail outlets to be tested for PAA migration. 3.3. Instrumental Fluorescence measurements were performed at room temperature on a PerkinElmer LS50B Luminescence Spectrometer (Waltham, MA, USA) equipped with a xenon discharge lamp. In all cases, a 10 mm quartz SUPRASILs cell with cell volume of 3.5 mL by PerkinElmer (Waltham, MA, USA) was used. The corresponding excitation–emission matrices were recorded in the following ranges: emission (295–395 nm, each 1 nm) when the excitation wavelengths vary from 220 to 275 nm (each 5 nm). Excitation and emission monochromator slit widths were both set to 10 nm. The scan speed was 1500 nm min  1. A rotary evaporator (ILMVAC, Ilmenau, Germany) was used for sample preconcentration.

Table 1 Compounds analyzed with name, chemical structure and abbreviations. Name

Chemical structure

Abbrev ANL

Aniline

NH2 4,4'-methylenedianiline

4,4´-MDA

H2N

NH2

2,4-diaminotoluene

2,4-TDA

H2N

NH2

S. Sanllorente et al. / Talanta 160 (2016) 46–55

49

3.4. Software 180

275

160

Excitation wavelenght (nm)

The FL WinLab software (PerkinElmer) was used to recorder the fluorescent signals. The data were imported to Matlab [29] using the INCA software [30] that inserts missing values into the matrix in the wavelengths that correspond to the Rayleigh effect. PARAFAC models were performed with the PLS_Toolbox 6.0.1 [31] for use with MATLAB. The least squares regressions were built and validated with STATGRAPHICS Centurion XVII [32]. Decision limit, CCα, and capability of detection, CCβ, were determined using the DETARCHI program [33].

140

256

120 100 80 238

60 40

4. Experimental procedure

20 220 295

4.1. Sample preparation method

4.2. Standard addition samples The matrix-matched standards were prepared by adding a fixed volume of the final extract, (depending on the required dilution in each stage of this work), the appropriate volume of the solutions (diluted solutions mentioned in Section 3.2) of each analyte into 10 mL volumetric flasks and completed to the mark with methanol, so that the desired concentration of every analyte was achieved for each experiment.

5. Results and discussion

345

370

395

Emission wavelenght (nm)

275

180

Excitation wavelenght (nm)

160 140

256

120 100 80 238

60 40 20

220 295

320

345

370

395

Emission wavelenght (nm)

275

Excitation wavelenght (nm)

PAAs migration was determined for two samples of cooking utensil. Migration experiments were performed without exposure to light. Typically, kitchen utensils were too large, which is why the handle of the test specimens was cut. Dust was removed from the sample by wiping it with a lint-free cloth. The food simulant B [4] (3% (w/v) acetic acid in aqueous solution) was previously heated and then put into contact with the sample. Each sample was placed in a beaker and filled with a given volume of simulant (250 mL), covered with aluminium foil and transferred to a preheated oven. The top of the beaker was covered with a glass plate in order to reduce the loss of the simulant by evaporation. The migration test was conducted for 2 h at 100 °C [34]. After 2 h, the test specimens were removed from the simulant and allowed to cool to room temperature. Fresh simulant was added till the original volume. Then, the simulant was stored at  20 °C until the analysis. Next, a preconcentration step was carried out using a rotary evaporator at 40 °C and 72 mbar. Thus, the volume of the sample was reduced from 100 mL to dryness. The residue was redissolved in methanol. This final extract was collected in an amber bottle and stored under refrigeration at 4 °C. For the recovery study, the extract was prepared following the procedure described above. In this case, instead of a migration extract, a fixed amount was added for each of the three amines. The samples for migration tests were prepared following these steps: i) A new kitchen utensil was placed into a container rinsed with simulant (3% w/v acetic acid) during 15 min at 100 °C ii) After 15 min, the utensil was taken out of the container and placed into a new one which it was previously heated at 100 °C. This procedure was repeated six times.

320

120 100

256 80 60 238

40 20

220 295

320

345

370

395

Emission wavelenght (nm) Fig. 1. Contour plots for each pure analyte in methanol: (a) 20 μg L  1 of ANL, (b) 20 μg L  1 of 4,4′-diaminodiphenylmethane and (c) 200 μg L  1 of 2,4diaminotoluene.

5.1. Reference spectra for analyte identification The unequivocal identification of the analytes using the reference spectra is required by the legislation currently in force. In every stage of this work, the identification of each analyte was carried out through the correlation between its emission and

excitation reference spectra and the emission and excitation loadings estimated from the corresponding PARAFAC mode. The experimental spectra were obtained from the EEMs of the pure analytes. So, standards prepared in methanol, whose composition is described in Section 3.2, were measured. Fig. 1 shows the

50

S. Sanllorente et al. / Talanta 160 (2016) 46–55

0.20

0.50 0.45 0.40

0.15

0.35 0.30

0.10

0.25 0.20 0.15

0.05

0.10 0.05

0.00 295

315

335

355

375

395

0.00 220

230

Emission wavelenght (nm)

240

250

260

270

Excitation wavelenght (nm)

0.20

0.50 0.45 0.40

0.15

0.35 0.30

0.10

0.25 0.20 0.15

0.05

0.10 0.05 0.00 295

315

335

355

375

395

0.00 220

230

Emission wavelenght (nm)

240

250

260

270

Excitation wavelenght (nm)

0.20

0.50 0.45 0.40

0.15

0.35 0.30

0.10

0.25 0.20 0.15

0.05

0.10 0.05

0.00 295

315

335

355

375

Emission wavelenght (nm)

395

0.00 220

230

240

250

260

270

Excitation wavelenght (nm)

Fig. 2. Comparison between the excitation and emission reference spectra and the excitation and emission loadings of the PARAFAC models obtained in solvent calibration (Section 4.2) and in the recovery study (Section 4.3.1) for: a) ANL, b) 4,4′-MDA and c) 2,4-TDA. Emission: left-hand side figures, excitation: right-hand side figures. The reference spectra are represented by a dark blue continuous line; the loadings of the three-factor model obtained in solvent calibration are represented by a red dashed line. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

contour plots of these samples. As can be seen in this figure, the spectra of the three PAAs are highly overlapped. In this work, the maximum fluorescence intensity in excitation spectra appears at 220, 230 and 240 nm for 2,4-TDA, ANL and 4,4′ MDA, respectively. ANL has an emission maximum in the recorded region at 340 nm, 4,4′-MDA at 350 nm and 2,4-TDA at 352 nm. The excitation and emission spectra at the wavelengths of maximum emission and excitation were taken, respectively, for each analyte. These excitation and emission reference spectra are represented by continuous lines in Fig. 2(a) for ANL, in Fig. 2(b) for 4,4′-MDA and in Fig. 2(c) for 2,4-TDA. The graphs corresponding to the emission are on the left and the excitation ones are placed on the right in Fig. 2. It must be noticed that the values have been normalized in this figure to compare them with the spectra obtained in the next stages of this work.

5.2. Solvent calibration First, the analysis was performed in synthetic samples, using pure methanol as solvent. The distribution of concentrations for the three analytes to carry out calibration in solvent was chosen in the following concentration range: 0–20 μg L  1 for aniline, 0–20 μg L  1 for 4,4′-MDA and 0–200 μg L  1 for 2,4-TDA and each analyte was at five levels of concentration. By means of a D-optimal design the 64 ternary mixtures of the complete calibration design were reduced to the 15 that appear in Table 2 representing a significant saving in experimental effort. The EEM matrices of all these samples were arranged, in the order shown in Table 2, and then the EEM matrix corresponding to methanol was subtracted from all the samples, because the methanol provides fluorescent signal in the same range as the three

S. Sanllorente et al. / Talanta 160 (2016) 46–55

Table 2 Distribution of concentrations for the three studied amines used to perform the calibration in solvent. Sample

ANL (μg L  1)

4, 4′-MDA (μg L  1)

2,4-TDA (μg L  1)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

2.5 20 10 5 20 5 10 2.5 10 20 2.5 5 2.5 10 5

5 10 20 2.5 2.5 5 10 20 2.5 5 10 20 2.5 5 10

25 25 25 25 50 50 50 50 100 100 100 100 200 200 200

amines analyzed. In consequence, the tensor obtained was X1 (15
100
12). The first dimension of this tensor refers to the number of samples, the second corresponds to the number of emission wavelengths and the third is the number of excitation wavelengths recorded. The PARAFAC decomposition has been applied to this tensor with the non-negativity constraint imposed for the two last tensor ways, because both the excitation and emission spectra must always be positive. A three-factor PARAFAC model was chosen (CORCONDIA of 98% and explained variance of 99.92%). At the confidence level of 95%, no outlier data were detected by means of the Q and Hotelling's T2 indices. The first three factors represent ANL, 4,4′-MDA and 2,4-TDA, respectively. The loadings of sample, emission and excitation wavelength profiles of these three factors are included in Fig. 3. Fig. 3(a) shows the pattern of calibration samples contained in Table 2. ANL is represented in blue circles, 4,4′- MDA in red squares and 2,4-TDA in green triangles. The correlation coefficients for the emission and excitation profiles, regarding the reference spectra, were 0.996 and 0.996 for ANL, 0.996 and 0.986 for 4,4′-MDA and 0.995 and 0.974 for 2,4TDA, respectively. After the decomposition of every data tensor into the appropriate number of factors and the identification of the factor related to each compound, a LS regression between the sample loadings and the true concentration was built with all the samples. Samples number 12, 13 and 14 for 4,4′-MDA and samples 4 and 8 for 2,4TDA in the calibration lines have a standardized residual greater than 2.5 in absolute value, so they were considered as outliers and removed. A new LS fitting was performed and validated with the remaining data in both cases. In all cases, the regression models were significant and there was not lack of fit at a 95% confidence level. Table 3 shows the parameters of the calibration lines estimated for each analyte, and other figures of merit. According to ISO5725, the term accuracy includes trueness and precision [35]. The accuracy is verified, in a concentration range, with the regression “calculated concentration versus true concentration” that assesses the trueness of the method using the hypothesis tests (for the slope and for the intercept) and evaluates the precision by the standard deviation of regression. As a result, this regression is named “accuracy line”. Table 3 shows the parameters of the accuracy line obtained for each analyte. The mean of the absolute value of the relative errors are also listed in Table 3. The lowest values were obtained for 4,4′-MDA. Two hypothesis tests enable one to verify the accuracy of every calibration model by checking if, at the 95% significance level,

51

there were no statistically significant differences between the values obtained, respectively, for the null hypothesis slope equal to 1, and the intercept equal to 0. The equations of three accuracy lines are collected in Table 3. The p-values for these last two tests are greater than 5% for both tests. So, property of trueness was verified for the three analytes. The lowest values for the decision limit, CCα, (for a probability of false positive, α, fixed at 0.05) and for the capability of detection (the probabilities of false positive and false negative, β, being equal to 0.05) were obtained for 4,4′-MDA; while the highest values were obtained for 2,4-TDA (see Table 3). All decision limits were lower than 10 μg L  1. 5.3. Quantification and identification in cooking utensils In this section, the standard addition method developed has been used to determine the three analytes in extract of simulant. The extract was prepared following the experimental procedure described in Section 4.1. The distribution of concentrations chosen was the same as in Section 5.2. In this case, the matrix-matched standards were prepared with the extract diluted and following the procedure already described in Section 4.1. After subtracting the blank signal, the tensor X2 (21
100
12), which contains the EEM matrices placed in the order of Table 4, was built. The coherence of the PARAFAC model with the experimental knowledge was taken into account to determine the appropriate number of factors. None of the PARAFAC models estimated from this tensor were totally coherent. For example, the factors associated to ANL and 4,4′-MDA were coherent in the three factor model but there was a confusion between the loadings of 2,4-TDA and the loadings of the matrix. On the other hand, the four-factor PARAFAC model was coherent (factor 1-3 corresponding to the three analytes, factor 4 for the matrix) but the CORCONDIA index was 13%, that is to say, this PARAFAC model constructed initially with the tensor X2 is not trilineal. The loadings of the sample mode for the four factors are depicted in Fig. 4. The pattern shown by factor 1 (4,4′-MDA, blue circles), factor 2 (ANL in red squares), factor 3, (2,4-TDA in green triangles). The purple diamond (fourth factor) is constant and corresponds to background. Clearly, it is a compound which is present in the simulant (matrix). A high fluorescent overlapping existed between the three amines and the fluorophores of the matrix (see Fig. 5). Fig. 5(a) shows the contour lines of one sample of extract from the spoon after the migration. Fig. 5b–d show the contour levels from the same sample to which a known amount has been added (the three mixtures marked with asterisk in Table 4). A high degree of overlapping can be observed in the four graphs. The identification of fluorescent matrix signal allows us to substract it from the original tensor. In this way, the new tensor is trilinear. The PARAFAC decomposition of these resulting tensors needs 3 factors and showed a CORCONDIA index equal to 98% (explained variance of 99.93%) the three factors extracted being identical to those obtained in the previous PARAFAC decomposition. For the identification of the amines, the correlation between the PARAFAC spectral loadings and the reference spectra has been used (the correlation coefficients for the emission and excitation profiles, were 0.997 and 0.977 for ANL, 0.997 and 0.989 for 4,4′-MDA and 0.996 and 0.987 for 2,4-TDA, respectively). Once this decomposition has been done, by observing the loadings corresponding to the sample profile (Fig. 4) it is possible to conclude that the only present amine in the extract obtained of the spoon is 4,4′-MDA (samples number 19 and 20, blue circles in Fig. 4). Bearing in mind the dilution factor and the recovery of this procedure, the concentration of 4,4′-MDA that has migrated to the simulant in contact with the spoon was 1351.97455.7 mg L  1.

52

S. Sanllorente et al. / Talanta 160 (2016) 46–55

2500

Loadings

2000

1500

1000

500

0 0

5

10

15

Sample 0.60

0.20

0.18 0.50

0.16

Loadings

Loadings

0.14

0.12 0.10 0.08

0.40 0.30

0.20

0.06 0.04

0.10

0.02 0.00

0.00

295

315

335

355

375

395

220

Emission wavelength (nm)

230

240

250

260

270

Excitation wavelenght (nm)

Fig. 3. Loadings of the three-factor PARAFAC model fitted with the data tensor X1. a) sample mode, b) emission mode and c) excitation mode. ANL is in blue circles, 4,4′MDA in red squares and 2,4-TDA in green triangles. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Table 3 Parameters of the regression line “sample loading versus true concentration” and accuracy line for ANL, 4,4′-MDA y 2,4-TDA in solvent standards. Decision limit (CCα) and capability of detection (CCβ) at x0 ¼0. Probabilities of false positive (α) and false negative (β) fixed at 0.05.

Calibration line

Accuracy line

a

Slope, b1 Intercept, b0 Residual standard deviation, syx Correlation coefficient, ρ Outlier samples Error in calibrationa Slope, b1 Intercept, b0 Residual standard deviation, syx Decision limit, CCα (μg L  1) Capability of detection, CCβ (μg L  1)

Mean of absolute values of relative errors

Aniline

4,4′-MDA

2,4-TDA

101.99  82.21 34.13 0.9976 ― 5.93 (n ¼15) 1 4.8
10  6 0.33 0.65 1.27

102.22  108.40 25.13 0.9990 12, 13 and 14 4.62 (n¼ 12) 1 4.8
10  5 0.25 0.51 0.99

5.36 6.07 27.09 0.9952 4 and 8 8.76 (n¼ 13) 0.999 7.8
10  4 5.06 10 19.59

S. Sanllorente et al. / Talanta 160 (2016) 46–55

Table 4 Distribution of concentrations of the three PAAs, in the matrix-matched standards, to perform the standard addition method. Sample

ANL (μg L  1)

4, 4′-MDA (μg L  1)

2,4-TDA (μg L  1)

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

0 0 2.5 20 10 5 20 5 10 2.5 10 20 2.5 5 2.5 10 5 20 0 0 0

0 0 5 10 20 2.5 2.5 5 10 20 2.5 5 10 20 2.5 5 10 20 0 0 0

0 0 25 25 25 25 50 50 50 50 100 100 100 100 200 200 200 200 0 0 0

that obtained from PARAFAC, of 0.997, 0.998 and 0.976 for sample, emission, and excitation profiles respectively. In this case the calculated concentration obtained for 4,4′-MDA was 1011.97277.7 μg L  1. Also for the ANL two factors were chosen, CORCONDIA index equal to 85%, explained variance of 99.89% and correlation coefficients of 0.9986, 0.9984 and 0.98246 for sample, emission, and excitation profiles respectively. The 2,4-TDA three factors provide a CORCONDIA index lower than zero, but as in Section 5.3, the factor related with the matrix was subtracted, and the new tensor had an index of CORCONDIA equal to 100% with two factors. 0.991, 0.9979 and 0.9373 were the coefficients of correlation for the sample, emission and excitation profiles respectively. Again, the results obtained with the standard addition method carried out over simulant extract, confirmed the absence of the ANL and the 2,4-TDA extract of simulant. 5.4.1. Recovery The procedure for recovery was carried out over a tensor for each amine. The recovery of the samples was evaluated applying to these samples, of known concentration; the same stages made in the migration test stage, that is to say, initially the samples were prepared in 3% (w/v) acetic acid and evaporated to dryness. Finally, the extract is re-constituted in methanol to carry out the instrumental measurements. Later, additions were made to each one of three amines giving rise to three tensors of data (dimensions 12
100
12). In the three cases only one factor was necessary obtaining the following recovery factors, 82.6%, 79.1% and 112.7% for the ANL, 4,4′-MDA and 2,4-TDA respectively.

a

Methanol blank. Diluted extract of spoon. Samples showed in Fig. 5. d Less diluted extract of spoon (two times). b c

3000 2500

5.5. Modeling the Kinetic of the migration of PAAs from polyamide kitchenware

2000

Loadings

53

1500 1000 500

0 0

5

10

15

20

Sample Fig. 4. Loadings of the sample profile of the four-factor PARAFAC model fitted with the data tensor X2. Factor 1 (4,4′-MDA): blue circles, factor 2 (ANL): red squares, factor 3 (2,4-TDA): green triangles, factor 4 (background): purple diamonds. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

5.4. Standard addition method based on PARAFAC and recovery analysis Another way to obtain the concentration of these analytes is to use a standard addition method adding the standards of each amine separately and then to use, as signal, the loadings of each amine. To check this, a fixed amount of each amine was added over the extract obtained. The spoon was put inside acid simulant and the migration test was conducted for 2 h at 100 °C. For this procedure, three tensors of size 12
100
12 were built, one for the ANL, other one for 2,4-TDA and the last for the 4,4′-MDA. This procedure implies making 36 determinations instead of 21 and is only carried out to compare results. These tensors were each analyzed separately obtaining the following results. In the case of 4,4′-MDA, a two-factor PARAFAC model was chosen (CORCONDIA index equal to 100%, explained variance of 99.92% and correlation coefficients, between reference spectra and

For this analysis the extract was obtained from another kitchenware (a spaghetti fork) to which was applied the procedure of migration test described in Section 4.1. Once reconstituted the extract in methanol, the EEM were recorded. In this case, the extracts were taken every 15 min, obtaining the tensor of size (6
100
12). The PARAFAC model obtained by means of the tensor X2 already analyzed in Section 5.3 and shown in Fig. 5 is applied to obtain the concentration of the amines in each sample of the kinetic study. Again the result was the presence in this sample of 4,4′ -MDA and the absence of other two amines analyzed. Later, the procedure for the fitting of the kinetic of migration was carried out. The kinetics of migration has been fitted according to the following Eq. (3) that corresponds to a migration kinetics in which the rate of release of 4,4′-MDA (as a function of time t) from the kitchenware into the food simulant is linearly related to the concentration of 4,4′-MDA in that utensil.

CMDA = a + ( 131. 4−a)e−b( t −15)

(3)

Using the experimental data obtained the fitted model has a coefficient of determination equal to 97.7%, a ¼64.554 and b ¼0.047. In Eq. (3), a represents the concentration of 4,4′-MDA in the whole volume of food simulant after infinite time of migration tests. On the other hand, b is the relative rate of release of 4,4′MDA from the kitchenware. The value of 4,4′-MDA, 131.4 μg L  1, at 15 min is used as reference for the curve. Basically, Eq. (3) shows that the amount of 4,4′-MDA migrated from a certain kitchen utensil decays exponentially at a rate that depends on the parameter b, so its value is characteristic of both the polyamide (bearing in mind its composition and structure) and the experimental conditions for the migration study (the heating

54

S. Sanllorente et al. / Talanta 160 (2016) 46–55

Fig. 5. EEM landscapes for a migration test from: a) spoon sample λexc ¼230, λem ¼ 353 (maximum fluorescence intensity ¼ 60.2; b) spoon sample plus mixture number 5c of Table 4; λexc ¼ 235, λem ¼342 (maximum fluorescence intensity ¼252.7); c) spoon sample plus mixture number 8c; λexc ¼ 240, λem ¼ 348 (maximum fluorescence intensity ¼ 244.1); d) spoon sample plus mixture number 13c of Table 4; λexc ¼ 235, λem ¼ 349 (maximum fluorescence intensity ¼174.3).

Fig. 6. Model fitted for 4,4′-MDA, following Eq. (3), for migration tests performed on the kitchen utensil evaluated.

temperature in this work). The experimental data and fitted curve are shown in Fig. 6. 6. Conclusion A new easy and fast procedure has been proposed. This

procedure allows one to unequivocally identify and quantify three primary aromatic amines (aniline, 4,4′-MDA and 2,4-TDA) by means of PARAFAC decomposition and fluorescent excitationemission signals. Decision limit and capability of detection have been obtained when the probabilities of false positive and false negative are fixed at 0.05. Good recovery in all of them has been found. Furthermore, migration test has been carried out to obtain kinetic equation for 4,4′-MDA, the only one found in the polyamide kitchenware analyzed. Severe problems of overlapping due to interferents (or matrix effects) have been overcome using PARAFAC decompositions in the analysis of the food simulant samples. The procedure proposed greatly simplifies sample treatment avoiding the use of toxic solvents, and saves cost in each analysis.

Acknowledgements The authors thank the financial support provided by project of the Ministerio de Economía y Competitividad CTQ2014–53157-R.

S. Sanllorente et al. / Talanta 160 (2016) 46–55

References [1] H. Kataoka, Derivatization reactions for the determination of amines by gas chromatography and their applications in environmental analysis, J. Chromatogr. A. 733 (1996) 19–34. [2] C. Ostiguy, S. Tranchand, J. Lesage, H. Van Tra, Développement exploratoire d′ une approche pour l′évaluation des amines en milieu de travail et documentation des procédés industriels utilisant ces substances, Montreal, 2003. 〈http://www.irsst.qc.ca/media/documents/PubIRSST/R-338.pdf〉. [3] Monographs on the evaluation of carcinogenic risks to humans: some aromatic amines, organic dyes and related exposures, Lyon (France), 2010. [4] Commission Regulation (EU) 10/2011 on plastic materials and articles intended to come into contact with food, Off. J. Eur. Commun. (2011) 1–89. [5] S.K. Mortensen, X.T. Trier, A. Foverskov, J.H. Petersen, Specific determination of 20 primary aromatic amines in aqueous food simulants by liquid chromatography–electrospray ionization-tandem mass spectrometry, J. Chromatogr. A. 1091 (2005) 40–50. [6] I.S. Arvanitoyannis, L. Bosnea, Migration of substances from food packaging materials to foods, Crit. Rev. Food Sci. Nutr. 44 (2004) 63–76. [7] M. de F.átima Poças, T. Hogg, Exposure assessment of chemicals from packaging materials in foods: a review, Trends Food Sci. Technol. 18 (2007) 219–230. [8] C. Lambert, M. Larroque, J.C. Lebrun, J.F. Gérard, Food-contact epoxy resin: covariation between migration and degree of cross-linking, Food Addit. Contam. 14 (1997) 199–208. [9] G. Lawson, C.T. Barkby, C. Lawson, Contaminant migration from food packaging laminates used for heat and eat meals, Fresenius. J. Anal. Chem. 354 (1996) 483–489. [10] G.L. Baughman, E.J. Weber, Transformation of dyes and related compounds in anoxic sediment: kinetics and products, Environ. Sci. Technol. 28 (1994) 267–276. [11] E. McCall, J. Keegan, B. Foley, Primary aromatic amine migration from polyamide kitchen utensils: Method development and product testing, Food Addit. Contam. – Part A Chem. Anal. Control. Expo. Risk Assess. 29 (2012) 149–160. [12] K. Marchildon, Polyamides - still strong after seventy years, Macromol. React. Eng. 5 (2011) 22–54. [13] X. Trier, B. Okholm, A. Foverskov, M.L. Binderup, J.H. Petersen, Primary aromatic amines (PAAs) in black nylon and other food-contact materials, 2004– 2009, Food Addit. Contam. Part A. Chem. Anal. Control. Expo. Risk Assess. 27 (2010) 1325–1335. [14] M. Aznar, E. Canellas, C. Nerín, Quantitative determination of 22 primary aromatic amines by cation-exchange solid-phase extraction and liquid chromatography–mass spectrometry, J. Chromatogr. A. 1216 (2009) 5176–5181. [15] B. Brauer, T. Funke, Spezifische bestimmung primärer aromatischer amine in wässrigen migraten von bedarfsgegenständen aus kunststoff, Dtsch. Leb. 98 (2002), pp. 405–411. [16] F. Lambertini, V. Di Lallo, D. Catellani, M. Mattarozzi, M. Careri, M. Suman, Reliable liquid chromatography–mass spectrometry method for investigation of primary aromatic amines migration from food packaging and during industrial curing of multilayer plastic laminates, J. Mass. Spectrom. 49 (2014) 870–877. [17] Z. Man, Q. Wang, H. Li, A. Zhang, Simultaneous determination of 33 primary aromatic amines of free state in fine pigments by ultra-performance liquid

[18]

[19]

[20] [21]

[22]

[23] [24] [25] [26] [27]

[28] [29] [30] [31] [32] [33]

[34]

[35]

55

chromatography coupled with triple quadrupole mass spectrometry, Chin. J. Chromatogr. 32 (2014) 1340–1348. L. Rubio, S. Sanllorente, L.A. Sarabia, M.C. Ortiz, Optimization of a headspace solid-phase microextraction and gas chromatography/mass spectrometry procedure for the determination of aromatic amines in water and in polyamide spoons, Chemom. Intell. Lab. Syst. 133 (2014) 121–135. R. Paseiro-Cerrato, G.O. Noonan, T.H. Begley, Development of a rapid screening method to determine primary aromatic amines in kitchen utensils using direct analysis in real time mass spectrometry (DART-MS), Food Addit. Contam. Part A. Chem. Anal. Control. Expo. Risk Assess. 31 (2014) 537–545. Rapid Alert System for Food and Feed, RASFF Portal, 〈https://webgate.ec. europa.eu/rasff-window/portal〉/(accessed 01.01.16). L. Rubio, M.C. Ortiz, L.A. Sarabia, Identification and quantification of carbamate pesticides in dried lime tree flowers by means of excitation-emission molecular fluorescence and parallel factor analysis when quenching effect exists, Anal. Chim. Acta 820 (2014) 9–22. L. Rubio, L.A. Sarabia, M.C. Ortiz, Standard addition method based on four-way PARAFAC decomposition to solve the matrix interferences in the determination of carbamate pesticides in lettuce using excitation-emission fluorescence data, Talanta 138 (2015) 86–99. S. Leurgans, R.T. Ross, Multilinear Models: applications in spectroscopy, Stat. Sci. 3 (1992) 289–310. R.B. Cattell, A cross-validation of primary personality structure in the 16 P.F. by two parcelled factor analyses, Multivar. Exp. Clin. Res. 10 (1994) 181–190. G. Tomasi, R. Bro, Comprehensive Chemometrics, Compr. Chemom. 2 (2009) 411–451. K.S. Booksh, B.R. Kowalski, Theory of analytical chemistry, Anal. Chem. 66 (1994) 782A–791A. S. Gourvénec, G. Tomasi, C. Durville, E. Di Crescenzo, C.A. Saby, D.L. Massart, R. Bro, G. Oppenheim, CuBatch, a MATLABs interface for n-mode data analysis, Chemom. Intell. Lab. Syst. 77 (2005) 122–130. R. Bro, H.A.L. Kiers, A new efficient method for determining the number of components in PARAFAC models, J. Chemom. 17 (2003) 274–286. MATLAB version 7.12.0.635 (R2011a), The Mathworks, Inc., Natick, MA, USA (2011). C.A. Andersson, INCA 1.41, Department of Food Science, University of Copenhagen, Denmark, 〈http://www.models.life.ku.dk/inca〉 (accessed 3.10.14). B.M. Wise, N.B. Gallagher, R. Bro, J.M. Shaver, PLS Toolbox 6.0.1,Eigenvector Research Inc., Wenatchee, WA, USA (2013). STATGRAPHICS Centurion XVI Version 16.1.05 (32bit), Statpoint Technologies, Inc., Herndon, VA, USA, 2010. L. Sarabia, M.C. Ortiz, DETARCHI: A program for detection limits with specified assurance probabilities and characteristic curves of detection, Trac. Trends Anal. Chem. 13 (1994) 1–6. Technical guidelines on testing the migration of primary aromatic amines from polyamide kitchenware and of formaldehyde from melamine kitchenware. EUR24815EN —first ed. (2011). International Organization for Standardization, ISO 5725–2, Accuracy (Trueness and Precision) of Measurement Methods and Results, Part 2: Basic Method for the Determination of Repeatability and Reproducibility of a Standard Measurement Method, Genève, Switzerland (1994).