Improvement in the identification and quantification of UV filters and additives in sunscreen cosmetic creams by gas chromatography/mass spectrometry through three-way calibration techniques

Accepted Manuscript Improvement in the identification and quantification OF UV filters and additives in sunscreen cosmetic creams by gas chromatography/mass spectrometry through three-way calibration techniques L. Rubio, L. Valverde-Som, L.A. Sarabia, M.C. Ortiz PII: S0039-9140(19)30782-9 DOI: https://doi.org/10.1016/j.talanta.2019.120156 Article Number: 120156 Reference: TAL 120156 To appear in: Talanta Received Date: 3 June 2019 Revised Date: 10 July 2019 Accepted Date: 14 July 2019 Please cite this article as: L. Rubio, L. Valverde-Som, L.A. Sarabia, M.C. Ortiz, Improvement in the identification and quantification OF UV filters and additives in sunscreen cosmetic creams by gas chromatography/mass spectrometry through three-way calibration techniques, Talanta (2019), doi: https://doi.org/10.1016/j.talanta.2019.120156. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 1/31 IMPROVEMENT IN THE IDENTIFICATION AND QUANTIFICATION OF UV FILTERS AND ADDITIVES IN SUNSCREEN COSMETIC CREAMS BY GAS CHROMATOGRAPHY/MASS SPECTROMETRY THROUGH THREE-WAY CALIBRATION TECHNIQUES L. Rubio a , L. Valverde-Som a , L.A. Sarabia b , M.C. Ortiz a,1 a Department of Chemistry, b Department of Mathematics and Computation Faculty of Sciences, Universidad de Burgos Plaza Misael Bañuelos s/n, 09001 Burgos (Spain) Abbreviations 2 Abstract The simultaneous determination of 2,6-di-tert-butyl-4-methyl-phenol (BHT), benzophenone (BP), benzophenone-3 (BP3) and diisobutyl phthalate (DiBP) in seven sunscreen creams was carried out by gas chromatography/mass spectrometry (GC/MS) using DiBP-d 4 as internal standard. The content of BP3, which is a UV filter, must not exceed 6% (w/w) in 1 Corresponding author. Telephone number: +34-947-259571. E-mail address: [email protected] (M.C. Ortiz). 2 2,6-di-tert-butyl-4-methyl-phenol (BHT), benzophenone (BP), benzophenone-3 (BP3), capability of detection (CCβ), core consistency diagnostic (CORCONDIA), decision limit (CCα), diisobutyl phthalate (DiBP), electron impact (EI), elliptical joint confidence region (EJCR), gas chromatography/mass spectrometry (GC/MS), internal standard (IS), multivariate curve resolution coupled to alternating least squares (MCR-ALS), parallel factor analysis (PARAFAC), principal component analysis (PCA), probability of false positive (α), probability of false negative (β), programmed temperature vaporizer (PTV), single ion monitoring (SIM), sun protection factor (SPF), total ion chromatogram (TIC), ultraviolet (UV). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 2/31 cosmetic products according to Regulation (EU) 2017/238 and the use of DiBP in cosmetic products shall be prohibited according to Regulation (EC) No 1223/2009. The conclusions obtained with the univariate standard methodology in the identification of the analytes contained in the creams were wrong. However, a calibration based on PARAFAC or PARAFAC2 decompositions, where the samples of the prediction set were projected on the model obtained previously with the calibration set, enabled the unequivocal identification and quantification of the analytes even in the presence of interferents not considered in the calibration model. The PARAFAC2 decomposition was used to overcome the shifts in the retention time of BP and BP3. These three-way calibration techniques are needed to avoid false negative results. The method had not proportional or constant bias. The presence of BHT was detected in the seven sunscreen creams analysed at an amount of 6.48 . 10 -2 %, 8.53 . 10 -2 %, 1.70 . 10 -4 %, 1.11 . 10 -4 %, 2.51 . 10 -3 %, 3.20 . 10 -5 % and 6.35 . 10 -3 %. The concentrations of DiBP found in four creams were 3.49 . 10 -2 %, 3.19 . 10 -2 %, 3.26 . 10 -2 % and 2.51 . 10 -2 %. On the other hand, BP was only detected in two of the cosmetic creams analysed at an amount of 7.84 . 10 -3 % and 1.04 . 10 -2 %. In addition, BP3 was detected in six of the creams at an amount of 4.73%, 3.49%, 4.94 . 10 -3 %, 1.98 . 10 -3 %, 6.62 . 10 -1 % and 1.73%. Therefore, none of the cosmetic creams contained BP3 in an amount higher than 6%. Keywords: Benzophenone-3; PTV-GC/MS; PARAFAC; PARAFAC2; UV filter; sunscreen cream. 1. Introduction The use of sunscreen cosmetic creams protects the skin from the negative effects of ultraviolet (UV) rays such as sunburn or skin cancer. Unfortunately, some of the additives contained in these creams could interfere in the hormone levels of the human body [1,2,3]. Regulation (EC) No 1223/2009 [4] on cosmetic products establishes rules to be complied with by any cosmetic product made available on the market, in order to ensure the MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 3/31 functioning of the internal market and a high level of protection of human health. The content of oxybenzone or benzophenone-3 (BP3), with the chemical name 2-hydroxy-4methoxybenzophenone, was modified in Commission Regulation (EU) 2017/238 [5] and must not exceed 6% (w/w). In addition, the label of the cosmetic product must include the wording "contains benzophenone-3" when the concentration is upper than 0.5% (w/w) and it is not used for product protection purposes. This range does not pose a risk to human health, apart from its contact which could produce allergy [6]. BP3 is a sunscreen agent used to absorb UV radiation in plastics and in personal care products. In addition, it is used as a photo-stabilizer to minimize the colour and odour changes of the cosmetic product [7]. The concentration of this compound is regulated because it is harmful to human health [7] and causes allergy [8,9]. Benzophenone-type UV filters also induce endocrine disrupting effects [10,11,12]. BP3 penetrates the skin and 1-2% of the sunscreen is absorbed in humans [13,14,15]. BP3 has been detected in blood plasma [16,17], in human breast milk [18,19] and in urine since the compound is excreted [1,13,14,20]. On the other hand, most of the compounds present in creams are soluble in water so negative environmental effects appear [7] such as fish contaminations which cause the appearance of these compounds in food [21,22]. In addition, the presence of BP3 in swimming pool water could produce hazardous products by reaction with chlorine [7,23] being a problem to human health. Other additives are added to sunscreen cosmetic creams such as benzophenone (BP) and 2,6-di-tert-butyl-4-methyl-phenol (BHT). BP is another UV filter [2,24], whereas BHT is used as antioxidant to prevent rancidity or to inhibit oxidation in cosmetic formulations [2,25]. On the other hand, the presence of phthalates such as diisobutyl phthalate (DiBP) in these creams can be due to manufacturing process or to their migration from packaging when plastic is used [26]. The control and analysis of these compounds is important because they have harmful effects on health too [27,28]. DiBP has been classified as carcinogenic, mutagenic or toxic substance to reproduction (category 1B) in [29]. Therefore, the use of this MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 4/31 compound in cosmetic products shall be prohibited as stated in [4]. However, the nonintended presence of a small amount of this substance shall be permitted if its presence is technically unavoidable, that is, if it comes from impurities of natural or synthetic ingredients, the manufacturing process, storage or migration from packaging. Many analysis methods have been developed to detect these compounds [2,30]. The extraction of the analytes from the cosmetic product can be performed with organic solvents such as methanol [24,31], ethanol [32,33], among others [26,34]. The use of an ultrasonic bath or a vortex can accelerate the solubilisation of those compounds. Then, the extract could be filtered or centrifuged to extract the fraction of interest and remove the insoluble fraction of the cosmetic matrix. Other extractive approaches are based on solid phase microextraction [35], or liquid-liquid microextraction [24]. The analytical techniques employed to determine UV filters in cosmetics [30] and other additives [2] are: (i) chromatographic techniques with different detectors; (ii) spectroscopic techniques; and (iii) electrochemical techniques. The two latter have been less used than chromatographic techniques. The use of liquid chromatography with different detectors in this determination has lately increased [32,33,36]. However, gas chromatography has been less used since a derivatization step with silylating reagents is sometimes required to increase the volatility and sensitivity of the compounds [35,37,38]. The determination of phthalates and sunscreen agents in cosmetic products is carried out using a gas chromatograph coupled to a mass spectrometer detector without derivatization in [26,39]. Several three-way algorithms can be used with chromatographic signals. Previous works [40,41,42] have demonstrated the usefulness of three-way calibrations based on the PARAFAC decomposition using chromatographic data obtained with different detectors that provide multivariate signals (mass spectrometers or diode array detectors) [43]. These works highlight the advantage of using the abundances recorded at all the ions selected (or the absorbance spectrum) when quantifying or identifying the analytes according to the requirements stated in European regulations. In addition, these calibrations are useful for the MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 5/31 optimization and evaluation of the robustness of analytical methods [42] since the secondorder property enables the identification of the analyte of interest as a single factor independently of the change in the instrumental factors, even in the presence of unknown interferents, which coelute with the analytes. The PARAFAC2 decomposition overcomes some deviations in the chromatographic signals [43,44]. PARAFAC2 has the second-order property if the correlation between the time profiles is the same in all the samples, which is a weaker condition than the equality of chromatographic profiles imposed by the PARAFAC model. If the loss of trilinearity is important, then multivariate curve resolution techniques are a useful alternative, because their signal-related requirements are weaker than those demanded by PARAFAC or PARAFAC2. Multivariate curve resolution coupled to alternating least squares (MCR-ALS) has been widely applied in analytical chemistry and its related fields [43,45,46], and it has been used to resolve coeluted compounds. Its major limitation in identifying and quantifying an analyte is the presence of rotational ambiguities and non-unique solutions. However, the non-uniqueness problem can be alleviated or totally avoided in some cases through the intelligent use of the data structure and appropriate constraints. This problem is discussed in depth in [46]. In this work, the simultaneous determination of BHT, BP, BP3 and DiBP in seven sunscreen cosmetic creams, using DiBP-d 4 as internal standard (IS), was carried out by means of gas chromatography/mass spectrometry (GC/MS) with a single quadrupole mass analyser in selected ion monitoring (SIM) mode. Parallel factor analysis (PARAFAC) or PARAFAC2 decomposition methods were used to discover if a coeluent that shares ions with the analyte of interest is present [47,48] and to identify unequivocally the compounds present by their chromatographic and spectral profiles following the criteria established in Decision 2002/657/EC for residues of veterinary medicinal products [49] which are stricter than other official regulations and guidelines [49,50,51]. In this case, at least a minimum of 3 identification points is needed for the confirmation of each compound; in this work 5 ions MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 6/31 were selected. The PARAFAC2 decomposition overcome the shifts in the retention time of BP and BP3 in the samples and the analytes were unequivocally identified although the shift of the retention time is limited in regulated analyses. The results of the identification were compared with the ones obtained with the univariate standard methodology. The determination of the compounds in the sunscreen creams analysed was performed through the projection of the samples of the prediction set on the corresponding PARAFAC or PARAFAC2 model. Other advantages of this work with respect to analytical methods previously reported are the determination of different compounds, that affect human health, in sunscreen cosmetic creams and the determination of BP3 by GC/MS without a derivatization reaction, which is not very usual. 2. Material and methods 2.1. Chemicals Benzophenone (CAS no. 119-61-9, purified by sublimation, ≥ 99% purity), 2-hydroxy-4methoxybenzophenone (benzophenone-3, CAS no. 131-57-7, 98% purity), 2,6-di-tert-butyl-4methyl-phenol (CAS no. 128-37-0, ≥ 99% purity), diisobutyl phthalate (CAS no. 84-69-5, 99% purity) and diisobutyl phthalate-3,4,5,6-d 4 (CAS no. 358730-88-8, analytical standard, 99.7% purity) were purchased from Sigma-Aldrich (Steinheim, Germany). Ethanol (96% vol., CAS no. 64-17-5, HiPerSolv CHROMANORM®, gradient grade for HPLC) was supplied by VWR International (Radnor, Pennsylvania, USA). N-hexane (CAS no. 11054-3) and acetone (CAS no. 67-64-1) for liquid chromatography Lichrosolv® were from Merck KGaA (Darmstadt, Germany). 2.2. Standard solutions MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 7/31 Stock solutions of BHT at 1 g L -1 , of BP at 5 g L -1 , of DiBP at 2 g L -1 , and of DiBP-d 4 at 0.45 g L -1 were prepared individually in hexane. Intermediate solutions were prepared from the former ones by dilution in the same solvent. A stock solution of BP3 at 30.5 g L -1 were prepared in ethanol and intermediate solutions at concentrations of 1, 5, 7.5, 10, 15, 20 and 25 g L -1 were prepared from that stock solution in ethanol. All those solutions of BP3 were diluted 25000 times to prepare the corresponding calibration standards in hexane. The stock and intermediate solutions of BP3 were stable for 15 days. The rest of the solutions of this analyte were prepared daily since this compound was not stable. All the stock and intermediate solutions, which weight was controlled to verify that the solvent had not evaporated, were stored in crimp vials at 4ºC and protected from light. The laboratory glassware used was thorough cleaned and plastic consumables were avoided as far as possible. The number and type of the samples analysed together with the concentration ranges of the solvent standards (analyte standards prepared in solvent) for each analyte in each stage are collected in Table S1 in the Supplementary Material. 2.3. Sunscreen cream samples Seven different sunscreen creams were purchased at local stores and pharmacies (Burgos, Spain). These cosmetic products were: i) cream 1 (sun protection factor (SPF) 50+), ii) cream 2 (SPF 30), iii) cream 3 (SPF 50+), iv) cream 4 (SPF 30), v) cream 5 (SPF 30), vi) cream 6 (SPF 50+), and vii) cream 7 (SPF 15). Cream 1 and 2 belong to the same cosmetic brand, whereas cream 4 and 7 belong to another one. The sunscreen product’s label of cream 1, 2 and 7 specified that BP3 was contained in their formulation and the label of cream 1, 2, 5 and 7 specified that BHT was one of their ingredients. 2.4. Sample preparation method MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 14/31 number of samples, while the second and third ones were the number of scans and ions, respectively. 3.1.3.1. Tolerance intervals (step ii) To estimate the tolerance intervals, the reference standards measured in Section 3.1.1 were used. In addition, three reference standards that contained the IS at three concentration levels and the analytes at a fixed intermediate concentration (see Table S1 in the Supplementary Material, third row) were also considered. The chromatograms of all these samples were equally fragmented around the retention time of each analyte after a baseline correction. The resulting data matrices were arranged in a three-way array, X 0 , for each analyte except for DiBP and DiBP-d 4 peaks for which a joint array was considered for both compounds. The dimension of the four three-way arrays built are given in the second row of Table 1, whereas the samples included in those arrays are detailed in rows 1 to 3 of Table S1 in the Supplementary Material. The first dimension corresponds to the number of scans considered, the second one refers to the number of ions recorded and the third one is the number of samples. Then, a PARAFAC decomposition was performed for each of the arrays. Table 1 (rows 1-7) also contains the features of the models obtained in each case. The core consistency diagnostic (CORCONDIA) [60] measures the trilinearity degree of the experimental three-way array when there are more than 2 factors in the model. If the array is trilinear, then the maximum CORCONDIA value of 100 is found. For BP and BP3, a PARAFAC2 decomposition was carried out since there were shifts in the retention time of those analytes in the samples. In addition, the variance explained by these PARAFAC2 models was higher than the one obtained with a PARAFAC model. The PARAFAC model for DiBP and DiBP-d 4 required three factors where the third one was related to an interferent (characteristic m/z ratio: 223) that eluted before DiBP-d 4 . MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 15/31 The tolerance intervals for the relative retention time (see Table 2A, fourth column) and for the relative ion abundances (see Table 2B, fifth column) were estimated from the loadings of the chromatographic and spectral profiles, respectively, of the PARAFAC or PARAFAC2 models. These two intervals were built following the requirements established in [49] and were used as reference for the unequivocal identification of the analytes. PARAFAC decompositions provide a unique chromatographic profile for each compound that is common to all the samples, whereas PARAFAC2 decompositions provide a chromatographic profile of each compound for every sample. When a PARAFAC2 decomposition was considered, the median of the retention times of the analyte obtained from the chromatographic profile was used to calculate the relative retention time (the ratio of the chromatographic retention time of the analyte to that of the internal standard). Table 2A contains the retention times (second column of this table) and relative retention times (third column of this table) for each analyte obtained from the models estimated with the array that contained the reference standards. The tolerance intervals for the relative retention time were built with a tolerance margin of ± 0.5% as [49] established. It has also been checked that the relative retention times of BP and BP3 in all the samples obtained through the PARAFAC2 decomposition were within the tolerance intervals estimated with the retention times of these analytes in each sample. Therefore, the use of the median to estimate a representative retention time for BP and BP3 was adequate. On the other hand, PARAFAC and PARAFAC2 decompositions provide a unique spectral profile for each compound common to all the samples. The relative ion abundances of each m/z ratio used to determine the tolerance intervals according to ref. [49] were calculated with the corresponding spectral loading (see Table 2B, third column) with respect to the one of the corresponding base peak of the analyte. 3.1.3.2. Calibration models and projection (steps iii to viii) MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 16/31 Then, a three-way data array was built for the calibration set (X 1 ) and another one for the prediction set (X 2 ) for each analyte, except for DiBP and DiBP-d 4 peaks for which a joint array was considered again in both cases. The samples that made up the calibration set were: four system blanks (empty vials without solvent), the first solvent blank measured, the first calibration batch together with the two replicates injected after it and two calibration standards of the second and of the third calibration batches. The prediction set was constituted with the rest of the samples (the rest of the calibration standards, solvent blanks, blank extracts and the diluted extracts of the creams). Some calibration standards were included in the prediction set to guarantee the feasibility of the projection of the samples. The dimension of X 1 and X 2 is included in Table 1 (rows 9 and 15, respectively). Then, PARAFAC decompositions (or PARAFAC2 decompositions in the case of BP and BP3) were performed with the three-way array that contained the calibration set. When a PARAFAC decomposition was carried out, a change in the order of the dimension of the three-way arrays for the calibration and prediction sets was carried out (see Table 1). In the case of the array of DiBP and DiBP-d 4 , some additional solvent standards were added as can be seen in Table 1 since both compounds were completely overlapped and the abundance of DiBP-d 4 was much lower than the abundance of DiBP, so PARAFAC needs a greater variation of DiBP-d 4 . Therefore, a standard containing all the analytes and a higher amount of DiBP-d 4 (100 µg L −1 ) together with two standards that only contained 100 µg L −1 of DiBP-d 4 were added to that three-way array. The characteristics of the PARAFAC and PARAFAC2 models obtained in each case are listed in Table 1 (rows 8-15, columns 2-5). An unconstrained one-factor model was needed for BHT, BP and BP3, whereas the decomposition of the common array for DiBP and DiBPd 4 needed three factors (CORCONDIA index of 100%). Once these models were obtained, the samples of the prediction set were projected on the corresponding model. The extract obtained from cream 6 diluted 10 times exceeded the threshold value of the Q and Hotelling’s T 2 statistics at the 95% confidence level when it was MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 17/31 projected on the PARAFAC2 model for BP3. Therefore, it was considered an outlier and removed from the three-way array that contained the samples of the prediction set (see the dimension of this array in Table 1). Fig. 3 shows the chromatographic, spectral and sample profiles of the one-factor PARAFAC2 model obtained for BP3. As can be seen in Fig. 3 (c), the sample loadings of BP3 were zero in the solvent blanks measured after the injection of each extract of cream (samples number 23, 25, 27, 30, 33, 36, 39, 49, 51, 53, 56, 59, 62 and 65) so the cleanliness of the GC/MS system was guaranteed. By way of example, the loadings of the three-factor PARAFAC model for DiBP and DiBP-d 4 are shown in Fig. 4. The sample loadings for the factor corresponding to DiBP Fig. 4 (a)) increased with the concentration of the calibration standards as expected and the replicates were similar. The sample loadings of this analyte in the extracts of cream 3 and 4 diluted 10 times were outside the calibration range (samples number 80 and 82 not shown in that figure) so the amount of DiBP in those dilutions could not be quantified. The loadings of the sample profile for the factor associated to the internal standard (Fig. 4 (b)) were zero in the solvent and system blanks, whereas they remained nearly constant in the rest of the samples except for the additional samples added to the three-way array where the loadings were higher as expected. On the other hand, the sample loadings for the third factor (Fig. 4 (c)) were very high in the extracts of cream 3 and 4 diluted 10 times. Some of the m/z recorded for DiBP were shared with this factor being 223 its most characteristic m/z (see Fig. 4 (d)). This factor was attributed to an unidentified interferent eluting near the beginning of the DiBP-d 4 peak as can be seen in red in Fig. 4 (e). The analytes were unequivocally identified since the relative retention times (see Table 2A, fifth column) and the relative ion abundances (see Table 2B, sixth column) estimated from the loadings of the chromatographic and spectral profiles obtained in this analysis, respectively, were within their corresponding tolerance intervals (see Table 2A (fourth column) and Table 2B (fifth column), respectively). As can be seen in Fig. 1, the relative abundances estimated from the spectral PARAFAC loadings of DiBP obtained were within MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 18/31 the tolerance intervals for all the m/z ratios. Therefore, the conclusion that DiBP was not present in the creams stated in Section 3.1.1 was wrong since a coeluting interferent (see Fig. 2 (b)) present in those samples shared some m/z ratios with DiBP. This interferent did not appear in the PARAFAC model considered (see Table 1) since it was not present in the samples of the calibration set and the extracts of the creams were projected on the model obtained with the calibration set. Therefore, there was no problem in the identification of DiBP using PARAFAC. It is important to bear in mind that one of the advantages of the PARAFAC decomposition over the univariate standard methodology is that this three-way technique provides a unique spectral profile for each analyte that is common to all the samples. The results of the steps ix, x and xi of the procedure will be shown in the following section although many approaches can be used to perform a calibration based on PARAFAC or PARAFAC2 decompositions [61]. 3.2. Quantification using three-way techniques The sample loadings of each analyte were numerically high since they came from the first mode of a PARAFAC decomposition or from the third mode of a PARAFAC2 decomposition and these modes are not normalized in the decomposition. Therefore, they were manually normalized prior to standardization. Once the sample loadings for each analyte were standardized by dividing each of them by that of the internal standard, calibration models “standardized sample loading versus true concentration” using the standards contained in the calibration set were fitted and validated. The parameters of the regression models estimated for each analyte are included in Table S2 in the Supplementary Material. A quadratic regression model was considered for all the analytes except for DiBP. One outlier was detected in the calibration models for BP and BP3 since these data had a studentized residual greater than 3 in absolute value. Therefore, those outliers were removed, and a new fitting was carried out with the remaining data in both cases (see the calibration models MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 19/31 obtained in Table S2 in the Supplementary Material). The lowest mean of the absolute value of the relative errors in calibration was 1.97% (n = 11) for BP, whereas the highest value was 6.86% (n = 8) for BP3 when the samples with predicted concentration lower than the corresponding CCβ had been excluded. In addition, the mean of the absolute value of the relative errors in prediction for the second and third calibration batches were calculated. Taking these values into account, it was concluded that BP3 was not stable over time since the errors increased from the second calibration batch to the third one. Therefore, the solutions were prepared daily. Table S2 in the Supplementary Material also contains the parameters of the corresponding accuracy lines built with the calibration standards, that is the regressions “predicted concentration versus true concentration”. The elliptical joint confidence region (EJCR) test was computed and Fig. S2 in the Supplementary Material shows the confidence ellipses, at a 95% confidence level, for the slope and the intercept of the accuracy line estimated for each analyte. All the confidence ellipses contained the point (0,1). In addition, Table S2 in the Supplementary Material contains the p-values of this test. These p-values were higher than 0.05 (0.973 for BHT, 0.960 for BP, 1.000 for DiBP and 0.123 for BP3) so the intercept and the slope were significantly not different from 0 and 1, respectively. Therefore, the method had not constant or proportional bias at a 95% confidence level. The values of decision limit (CCα) and capability of detection (CCβ) for each analyte with the probabilities of false positive (α) and false negative (β) fixed at 0.05 are listed in the first two rows of Table 3. The blank extract contained DiBP in all its dilutions except for the one diluted 8000 and 10000 times, whereas the amount of the rest of the analytes in the blank extract was below the corresponding CCα values. The amount of each analyte found in the sunscreen creams together with the corresponding 95% confidence interval are detailed in Table 3. The label of cream 1, 2, 5 and 7 specified that BHT was contained in their formulation but BHT was detected in the seven creams analysed. On the other hand, BP3 was found in all the sunscreen creams except for cream 4 since the confidence interval for MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 20/31 that analyte in that cream contained zero and the values were below CCα as can be seen in Table 3. The creams 1, 2, 6 and 7 were the ones which contained a concentration of BP3 between 0.5% and 6%. However, only the product’s label of creams 1, 2 and 7 specified that BP3 was one of their ingredients as required in [5]. In addition, DiBP and BP were only detected in four and two cosmetic creams, respectively. The amount found of DiBP (below 0.035%) may come from impurities of natural or synthetic ingredients, the manufacturing process, storage or migration from packaging of the cosmetic product. 4. Conclusions The projection of the samples of the prediction set on the corresponding PARAFAC or PARAFAC2 model enabled the determination of BHT, BP, DiBP and BP3 in the sunscreen cosmetic creams analysed even in the presence of interferents not considered in the calibration model. PARAFAC2 decomposition overcame the problems due to the shifts in the retention time of BP and BP3. In addition, the unequivocal identification and quantification of each analyte according to the requirements established by EU regulations were possible using a PARAFAC or PARAFAC2 decomposition despite some of the m/z ratios of a coeluting interferent were shared with DiBP. In fact, the unequivocal identification of this analyte could never have been achieved from the mass spectrum recorded at its retention time due to the presence of a coeluting interferent present in the sunscreen creams analysed. The presence of BHT was detected in the seven sunscreen creams analysed, whereas BP3, DiBP and BP were detected in some of these creams. None of the cosmetic creams contained BP3 in an amount higher than 6% as established in Regulation 2017/238 [5]. 5. Acknowledgments The authors thank the financial support provided by Spanish MINECO (AEI/FEDER, UE) through project CTQ2017‐88894‐R and by Junta de Castilla y León through project MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 21/31 BU012P17 (both co‐financed with European FEDER funds). L. Valverde-Som thanks JCyL for her postdoctoral contract through BU012P17 project. CONFLICT OF INTEREST The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. 6. References [1] K.N. Jallad, Chemical characterization of sunscreens composition and its related potential adverse health effects, J. Cosmet. Dermatol-US. 16 (2017) 353-357. https://doi.org/10.1111/jocd.12282. [2] M. Lores, M. Llompart, G. Alvarez-Rivera, E. Guerra, M. Vila, M. Celeiro, J.P. Lamas, C. Garcia-Jares, Positive list of cosmetic ingredients: Analytical methodology for regulatory and safety controls - A review, Anal. Chim. Acta. 915 (2016) 1-26. https://doi.org/10.1016/j.aca.2016.02.033. [3] T. Wong, D. Orton, Sunscreen allergy and its investigation, Clin. Dermatol. 29 (2011) 306-310. https://doi.org/10.1016/j.clindermatol.2010.11.002. [4] Regulation (EC) No 1223/2009 of the European Parliament and of the Council of 30 November 2009 on cosmetic products, Off. J. Eur. Union, L342 (2009) 59-209. [5] Commission Regulation (EU) 2017/238 of 10 February amending Annex VI to Regulation (EC) No 1223/2009 of the European Parliament and of the Council on cosmetic products, Off. J. Eur. Union, L36 (2017) 37-38. [6] Scientific Committee on Consumer Products, Health & Consumer Protection Directorate-General, Opinion on benzophenone-3. COLIPA Nº S38, European Commission, 2008. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 22/31 [7] J.C. DiNardo, C.A. Downs, Dermatological and environmental toxicological impact of sunscreen ingredient oxybenzone/benzophenone-3, J. Cosmet. Dermatol. 17 (2018) 15-19. https://doi.org/10.1111/jocd.12449. [8] A.R. Heurung, S.I. Raju, E.M. Warshaw, Benzophenones, Dermatitis 25 (2014) 310. https://doi.org/10.1097/DER.0000000000000025. [9] M.D. Palm, M.N. O'Donoghue, Update on photoprotection, Dermatol. Ther. 20 (2007) 360-376. https://doi.org/10.1111/j.1529-8019.2007.00150.x. [10] J. Wang, L. Pan, S. Wu, L. Lu, Y. Xu, Y. Zhu, M. Guo, S. Zhuang, Recent advances on endocrine disrupting effects of UV filters, Int. J. Env. Res. Pub. He. 13 (2016) 111. https://doi.org/10.3390/ijerph13080782. [11] M. Krause, A. Klit, M.B. Jensen, T. Søeborg, H. Frederiksen, M. Schlumpf, W. Lichtensteiger, N.E. Skakkebaek, K.T. Drzewiecki, Sunscreens: are they beneficial for health? An overview of endocrine disrupting properties, Int. J. Androl. 35 (2012) 424-436. https://doi.org/10.1111/j.1365-2605.2012.01280.x. [12] M. Schlumpf, P. Schmid, S. Durrer, M. Conscience, K. Maerkel, M. Henseler, M. Gruetter, I. Herzog, S. Reolon, R. Ceccatelli, O. Faass, E. Stutz, H. Jarry, W. Wuttke, W. Lichtensteiger, Endocrine activity and developmental toxicity of cosmetic UV filters - an update, Toxicology 205 (2004) 113-122. https://doi.org/10.1016/j.tox.2004.06.043. [13] H. Gonzalez, A. Farbrot, O. Larkö, A.M. Wennberg, Percutaneous absorption of the sunscreen benzophenone-3 after repeated whole-body applications, with and without ultraviolet irradiation, Brit. J. Dermatol. 154 (2006) 337-340. https://doi.org/10.1111/j.1365-2133.2005.07007.x. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 23/31 [14] H.G. Gonzalez, A. Farbrot, O. Larkö, Percutaneous absorption of benzophenone-3, a common component of topical sunscreens, Clin. Exp. Dermatol. 27 (2002) 691694. https://doi.org/10.1046/j.1365-2230.2002.01095.x. [15] C.G Hayden, M.S. Roberts, H.A. Benson, Systemic absorption of sunscreen after topical application, Lancet. 350 (1997) 863-864. https://doi.org/10.1016/S01406736(05)62032-6. [16] N.R. Janjua, B. Kongshoj, A.M. Andersoon, H.C. Wulf, Sunscreens in human plasma and urine after repeated whole-body topical application, J. Eur. Acad. Dermatol. 22 (2008) 456-461. https://doi.org/10.1111/j.1468-3083.2007.02492.x. [17] V. Sarveiya, S. Risk, H.A.E. Beson, Liquid chromatographic assay for common sunscreen agents: application to in vivo assessment of skin penetration and systemic absorption in human volunteers, J. Chromatogr. B 803 (2004) 225-231. https://doi.org/10.1016/j.jchromb.2003.12.022. [18] D. Molins-Delgado, M.M. Olmo-Campos, G. Valeta-Juan, V. PleguezuelosHernández, D. Barceló, M.S. Díaz-Cruz, Determination of UV filters in human breast milk using turbulent flow chromatography and babies' daily intake estimation, Environ. Res. 161 (2018) 532-539. https://doi.org/10.1016/j.envres.2017.11.033. [19] R. Rodríguez-Gómez, A. Zafra-Gómez, N. Dorival-García, O. Ballesteros, A. Navalón, Determination of benzophenone-UV filters in human milk samples using ultrasound-assisted extraction and clean-up with dispersive sorbents followed by UHPLC-MS/MS, Talanta 134 (2015) 657-664. https://doi.org/10.1016/j.talanta.2014.12.004. [20] M. Schlumpf, B. Cotton, M. Conscience, V. Haller, B. Steinmann, W. Lichtensteiger, In vitro and in vivo estrogenicity of UV screens, Environ. Health Persp. 109 (2001) 239-244. https://doi.org/10.1289/ehp.01109239. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 30/31 FIGURE CAPTIONS Fig. 1 Tolerance intervals for the relative ion abundances of DiBP estimated for each m/z ratio with three reference samples that contained a different concentration level of DiBP: 25 µg L -1 (in blue), 75 µg L -1 (in red) and 125 µg L -1 (in green). Relative abundance for each m/z ratio obtained with: the PARAFAC models obtained with the reference samples (light blue circles) and with all the samples of the analysis of the creams (black cross), the extract obtained from cream 1 diluted 10000 times (pink triangle), the extract obtained from cream 2 diluted 10000 times (purple diamond), the extract obtained from cream 3 diluted 10000 times (red square) and the extract obtained from cream 6 diluted 8000 times (light green star). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article). Fig. 2 Chromatographic profiles of the PARAFAC models obtained with the three-way arrays that contained the calibration and predictions sets together in the same array for: (a) BHT, (b) DiBP and DiBP-d 4 , (c) BP3 (a PARAFAC2 decomposition of the array was carried out in this last case). Fig. 3 One-factor PARAFAC2 model obtained with the three-way array that contained the calibration set for BP3. Loadings of the: (a) chromatographic profile, (b) spectral profile and (c) sample profile. The sample loadings of the 20 samples of the calibration set are represented by grey circles, whereas the sample loadings of the 62 samples of the prediction set, which have been projected on the PARAFAC2 model, are represented by red diamonds. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 31/31 Fig. 4 Loadings of the sample profile for: a) factor 1 (DiBP), b) factor 2 (DiBP-d 4 ) and c) factor 3 (unknown interferent) of the three-factor PARAFAC model fitted with the common three-way array for DiBP and DiBP-d 4 that only contained the calibration set. The sample loadings of the 24 samples of the calibration set are represented by grey circles, whereas the ones of the 63 samples of the prediction set, which have been projected on the model, are represented by red diamonds. d) Loadings of the spectral profile and e) loadings of the chromatographic profile. In figures (d-e), factor 1 (DiBP) is in light blue, factor 2 (DiBP-d 4 ) is in light green, while factor 3 (interferent) is in red. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Table 1 Dimensions (scans × ions × samples) of the three-way arrays built for every analyte in each experimental stage and characteristics of the PARAFAC or PARAFAC2 models (number of factors, constraints imposed, explained variance and CORCONDIA index) obtained from the decomposition of each array. Analytical stage BHT BP DiBP and DiBP-d 4 BP3 Tolerance intervals Dimension of X 0 21 × 5 × 9 49 × 5 × 9 41 × 10 × 9 56 × 5 × 9 Model PARAFAC PARAFAC2 PARAFAC PARAFAC2 # Factors 1 1 3 1 Non-negativity constraints None None In modes 1 and 2 None Expl. Var (%) 99.71 99.88 99.45 99.79 CORCONDIA (%) a --- --- 100 --- Analysis of the sunscreen creams Dimension of X 1 (calibration set) 20 × 21 × 5 b 49 × 5 × 20 24 × 41 × 10 b 56 × 5 × 20 Model PARAFAC PARAFAC2 PARAFAC PARAFAC2 # Factors 1 1 3 1 Non-negativity constraints None None In the three modes None Expl. Var (%) 99.74 99.87 99.08 99.82 CORCONDIA (%) a --- --- 100 --- Dimension of X 2 (prediction set) 63 × 21 × 5 b 49 × 5 × 63 63 × 41 × 10 b 56 × 5 × 62 Expl. Var.: Explained variance by the model. a The CORCONDIA index cannot be calculated in the PARAFAC or PARAFAC2 decomposition with only one factor. b In this case, the dimension of the array corresponds to the number of samples × scans × ions. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Table 2 Tolerance intervals for: A) the relative retention time and for B) the relative ion abundances estimated from the loadings of the chromatographic and spectral profiles, respectively. Identification of every analyte in the analysis of sunscreen creams. In the case of BP and BP3, the median of the retention times obtained in the corresponding PARAFAC2 decomposition was considered. A) Retention time Analyte t R (min) Relative t R Tolerance interval Identification in the analysis of the creams Relative t R BHT 8.229 0.806 (0.802-0.810) 0.806 BP 9.016 0.883 (0.879-0.887) 0.883 DiBP-d 4 10.211 1.000 (0.995-1.005) 1.000 DiBP 10.220 1.001 (0.996-1.006) 1.001 BP3 11.120 1.089 (1.083-1.094) 1.089 B) Diagnostic ions Analyte m/z ratio Spectral loading Relative abundance (%) Tolerance interval (%) b Identification in the analysis of the creams Relative abundance (%) BHT 91 6.85 . 10 - 2 7.13 (3.57-10.70) 7.08 145 1.15 . 10 - 1 11.94 (9.55-14.33) 11.94 177 7.67 . 10 - 2 7.98 (3.99-11.97) 8.03 205 a 9.61 . 10 - 1 100 - 100 220 2.29 . 10 - 1 23.81 (20.24-27.38) 23.94 BP 51 1.37 . 10 - 1 17.75 (14.20-21.30) 17.42 77 4.38 . 10 - 1 56.77 (51.09-62.45) 56.63 105 a 7.71 . 10 - 1 100 - 100 152 3.27 . 10 - 2 4.24 (2.12-6.36) 4.34 182 4.40 . 10 - 1 57.05 (51.35-62.76) 57.61 DiBP-d 4 80 5.65 . 10 - 2 5.68 (2.84-8.52) 5.70 153 a 9.94 . 10 - 1 100 - 100 171 2.48 . 10 - 2 2.50 (1.25-3.75) 2.57 209 1.17 . 10 - 2 1.18 (0.59-1.77) 1.23 227 4.95 . 10 - 2 4.98 (2.49-7.47) 5.14 DiBP 104 7.82 . 10 - 2 7.86 (3.93-11.79) 7.66 149 a 9.95 . 10 - 1 100 - 100 167 2.75 . 10 - 2 2.76 (1.38-4.14) 2.75 205 1.37 . 10 - 2 1.38 (0.69-2.07) 1.39 223 5.24 . 10 - 2 5.27 (2.64-7.91) 5.27 BP3 77 1.56 . 10 - 1 22.86 (19.43-26.29) 22.33 105 8.16 . 10 - 2 11.99 (9.59-14.39) 11.74 151 5.73 . 10 - 1 84.19 (75.77-92.61) 83.43 227 a 6.81 . 10 - 1 100 - 100 228 4.21 . 10 - 1 61.86 (55.67-68.05) 61.80 a Base peak. b According to ref. [42], the estimation of the tolerance interval is different depending on the value of the relative abundance of the corresponding m/z ratio. The tolerance margin was ±50% for relative intensities lower or equal to 10%, a ± 20% for relative intensities from 10% to 20%, a ± 15% for relative intensities from 20% to 50% and ± 10% for relative intensities higher than 50%. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Table 3 Decision limit (CCα), capability of detection (CCβ) at x 0 = 0 and concentration of each analyte found in the sunscreen creams together with the corresponding confidence intervals for the predicted concentration at a 95% confidence level. BHT BP DiBP BP3 CCα (x 0 = 0) (µg L − 1 ) 4.03 6.32 5.93 142.67 CCβ (x 0 = 0) (µg L − 1 ) a 7.93 12.40 11.65 279.80 Sunscreen cream 1 Concentration (% w/w) 6.48 . 10 - 2 < CCα b 3.49 . 10 - 2 4.73 Interval at a 95% confidence level (% w/w) (4.62 . 10 - 2 , 8.21 . 10 - 2 ) b (6.75 . 10 - 3 , 6.20 . 10 - 2 ) b (4.17, 5.52) b Sunscreen cream 2 Concentration (% w/w) 8.53 . 10 - 2 < CCα b 3.19 . 10 - 2 3.49 Interval at a 95% confidence level (% w/w) (6.70 . 10 - 2 , 1.02 . 10 - 1 ) b (3.75 . 10 - 3 , 5.91 . 10 - 2 ) b (2.90, 4.17) b Sunscreen cream 3 Concentration (% w/w) 1.70 . 10 - 4 < CCα b 3.26 . 10 - 2 4.94 . 10 - 3 Interval at a 95% confidence level (% w/w) (1.46 . 10 - 4 , 1.92 . 10 - 4 ) c (4.50 . 10 - 3 , 5.98 . 10 - 2 ) b (4.17 . 10 - 3 , 5.95 . 10 - 3 ) c Sunscreen cream 4 Concentration (% w/w) 1.11 . 10 - 4 7.84 . 10 - 3 < CCα d 8.41 . 10 - 4 Interval at a 95% confidence level (% w/w) (8.68 . 10 - 5 , 1.34 . 10 - 4 ) c (3.92 . 10 - 3 , 1.14 . 10 - 2 ) d (-1.56 . 10 - 4 , 1.56 . 10 - 3 ) c Sunscreen cream 5 Concentration (% w/w) 2.51 . 10 - 3 1.04 . 10 - 2 < CCα e 1.98 . 10 - 3 Interval at a 95% confidence level (% w/w) (5.75 . 10 - 4 , 4.27 . 10 - 3 ) e (7.49 . 10 - 3 , 1.31 . 10 - 2 ) e (1.08 . 10 - 3 , 2.75 . 10 - 3 ) c Sunscreen cream 6 Concentration (% w/w) 3.20 . 10 - 5 < CCα f 2.51 . 10 - 2 6.62 . 10 - 1 Interval at a 95% confidence level (% w/w) (6.87 . 10 - 6 , 5.56 . 10 - 5 ) c (2.40 . 10 - 3 , 4.67 . 10 - 2 ) f (3.84 . 10 - 2 , 1.06) f Sunscreen cream 7 Concentration (% w/w) 6.35 . 10 - 3 < CCα g < CCα g 1.73 Interval at a 95% confidence level (% w/w) (5.40 . 10 - 4 , 1.17 . 10 - 2 ) g (1.56, 2.00) g a α = β = 0.05. b The concentration value was calculated using the extract diluted 10000 times (n = 2). c The concentration value was calculated using the extract diluted 10 times (n = 1). d The concentration value was calculated using the extract diluted 1300 times (n = 2). e The concentration value was calculated using the extract diluted 1000 times (n = 2). f The concentration value was calculated using the extract diluted 8000 times (n = 2). g The concentration value was calculated using the extract diluted 3000 times (n = 2). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Fig. 1 Relative abundance (%) MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Fig. 2 8.20 8.21 8.22 8.23 8.24 8.25 8.26 Time (min) 0 2 4 6 8 10 12 104 (a) BHT Interferent 1 Interferent 2 10.18 10.19 10.20 10.21 10.22 10.23 10.24 10.25 Time (min) 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 106 (b) DiBP IS Interferent Chromatographic loading MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Fig. 3 Chromatographic loading 77 105 151 228 m/z ratio 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 (b) 227 10 20 30 40 50 60 70 80 Sample number 0 2 4 6 8 10 12 14 104 (c) PREDICTION SET CALIBRATION SET MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT Fig. 4