Unfortunately, the DoE did not give clear results for all the vari-ables because the statistical analysis revealed an unusually highstandard deviation (>10). This means that only factors affectingvery heavily the recoveries of the analytes could be identified.We discovered that both dilution and acidification of the sam-ple after deproteinization and before SPE were necessary, whereasacidification of the sample before extraction did not improve recov-eries significantly. Recoveries also increased as the concentrationof the displacer HCOONH4increased from 5 to 10 mmol L−1and the solvent changes from methanol to methanol:methylene chlo-ride, 1:1 (v/v). Interaction between these two variables is possible,but was not evidenced by the DoE. Solvent-to-sample ratio, as soonas extraction temperature and reconstituting solvent did not showany significative effect. Neither any other effect or interaction couldbe evidenced by applying the DoE to the extraction of the analyteswith methanol and acetonitrile instead of acetone.Owing this difficulty, we investigated the effect of theextraction/deproteinization solvent with univariate study (otherconditions were: extract dilution 1:9 and acidification to pH 2).Results are reported in Table 1, where absolute recoveries of tar-get compounds are shown. Probably, with the exception of PFPeAfor which a breakthrough phenomenon can be hypothesized, themajor source of imprecision arises at this stage of the procedure.Although not clearly better, the results for recoveries obtained withacetone seems at least not worst than that for the other solvents andthe phase separation was easier; in addition, acetone is less toxicthan both methanol and acetonitrile. Optimization of pH and dilu-tion factor of the extracts (the two factors which mainly affectedthe analyte adsorption on GCB) as far as the composition of thedesorbing phase may be possible. The possibility of a fine tuning ofthese parameters, however, would be excluded a priori, owing tothe very large range of polarity and polarizability of the target com-pounds. Although previous experiences with the analysis of linearalkylbenzene sulfonates (LAS) and their carboxylated metabolitesshowed that an increase of pH from 2 to 1 displaced short chain car-boxylated metabolites from the GCB surface [35], as PFAS possesspkahigher than that of carboxylic acids, we tested also this varia-tion, but recoveries of shorter chain compounds deeply decreased.Also the dilution factor sample-to-acidified water could be furtheroptimized, as it was found that it depends on both matrix and com-pound [22,29]. A dilution factor 1:25 did not improve recoveries,whereas decreasing dilution to 1:5 decreased recoveries of PFPeA,PFHeA and PFBS. It is possible that a further increase of methy-lene chloride-to-methanol ratio, and displacer concentration mightfurther increase recoveries of higher isomers (more polarizables),but it might also produce problem of salt solubility and increasecoextracted compounds with a consequent ESI source contamina-tion. Therefore, also by considering that isotopic ISs were used,the selected extraction conditions were those obtained from thescreening DoE and reported in Section 2.5.3.3. Method validationThe performance of the method was evaluated under the opti-mized conditions in terms of matrix effects, range of instrumentallinearity, accuracy, repeatability and linearity at low concentra-tion. The matrix effect was estimated, as reported in Section 2.7, bycomparing the slopes of six point calibration curves of the isotopicalIS in pure solvent and in pooled milk extracts. By using the isotopicIS instead of the analytes, it was not necessary subtracting the peakareas corresponding to the native analytes present in the sampleor blanks. The slope ratio of the regression lines (standard/matrix)ranged between 0.96 and 1.04, therefore matrix effect could beconsidered negligible.Instrumental linearity was evaluated in separate experiments inwhich calibration lines were constructed in a concentration rangewider than that used for quantification of analytes in the sam-ples. For all compounds, linearity in matrix was observed in therange LOQ–10 g L−1. Coefficients of determination R2ranged from0.9982 to 0.9999. Linearity of the method was also tested by usingthe isotopic ISs added to five aliquots of a pooled milk sample in therange of 2.5–100 ng L−1, and submitting the samples to the wholeprocedure. In this case, R2were in the range of 0.81–0.88, but theintercepts did not differ significantly from 0 at p = 0.05 and weresymmetrically distributed.Although our absolute recoveries of analytes were higher thanthose reported in a recent paper for a QuEChERS extractionmethod [5], they showed quite high imprecision, probably dueto one or more uncontrolled factors during extraction from milk.PFAS possess peculiar physical–chemical characteristics, some-times very difficult to be predicted or rationalized. For this reasonthe use of the appropriate isotopical IS is strongly recommended.Standard reference materials for determination of PFAS in cow milkwere not available; therefore, the accuracy and precision of themethod were assessed using milk samples fortified at two con-centration levels, such as 50, and 10 ng L−1(Fig. 2). Six aliquotsfrom three different stock fortified milk samples were individu-ally processed and analyzed.13C4-PFOS was used as IS for PFBS,and13C2-PFDoA for PFTrDA and PFTeDA. The results are shown inTable 2, where instrumental limits of detection (LOD) and quantifi-cation (LOQ), and method detection (MDL) and quantification limits(MQL) are also reported. As can be seen, the use of the isotopic ISsdeeply improve both accuracy and reproducibility.We based the identification criteria on the Commission Decision2002/657/EC. Therefore, a tolerance of 2.5% of tR, and a differenceof less than 20% between the two transitions intensity ratio wereconsidered acceptable for confirmation. However, for PFPeA onlyone transition was available, thus confirmation might be possibleonly with high resolution MS/MS. As consequence of the criteriachosen, based on the concept that a compound can be quantita-tively measured only if it has been unambiguously identified, LODsand MDLs were calculated using the less intense transition. LODsand LOQs were estimated from the signal-to-noise ratio (S/N) inthe standard solution as the quantity that give a S/N = 3 and = 10,respectively. However when a LOQ was apparently lower than thecorresponding LOD, it was considered = LOD. MDLs and MQLs wereestimated to be the blank value +3 and 10 times the standard devi-ation of the blank, respectively. and the same criteria were appliedwhen a MQL was apparently less than the corresponding MDL.Thereby, the estimated MDL values corresponds to CC, (Commis-sion Decision 2002/657/EC), but it should be pointed out that theyare obtained by using the same batch of solvents and materials.Taking into account the fact that blank reproducibility might berather different from laboratory to laboratory, the MDLSand MQLsreported could vary widely, and should be considered as approxi-mate.
Unfortunately, the DoE did not give clear results for all the vari-ables because the statistical analysis revealed an unusually highstandard deviation (>10). This means that only factors affectingvery heavily the recoveries of the analytes could be identified.We discovered that both dilution and acidification of the sam-ple after deproteinization and before SPE were necessary, whereasacidification of the sample before extraction did not improve recov-eries significantly. Recoveries also increased as the concentrationof the displacer HCOONH4increased from 5 to 10 mmol L−1and the solvent changes from methanol to methanol:methylene chlo-ride, 1:1 (v/v). Interaction between these two variables is possible,but was not evidenced by the DoE. Solvent-to-sample ratio, as soonas extraction temperature and reconstituting solvent did not showany significative effect. Neither any other effect or interaction couldbe evidenced by applying the DoE to the extraction of the analyteswith methanol and acetonitrile instead of acetone.Owing this difficulty, we investigated the effect of theextraction/deproteinization solvent with univariate study (otherconditions were: extract dilution 1:9 and acidification to pH 2).Results are reported in Table 1, where absolute recoveries of tar-get compounds are shown. Probably, with the exception of PFPeAfor which a breakthrough phenomenon can be hypothesized, themajor source of imprecision arises at this stage of the procedure.Although not clearly better, the results for recoveries obtained withacetone seems at least not worst than that for the other solvents andthe phase separation was easier; in addition, acetone is less toxicthan both methanol and acetonitrile. Optimization of pH and dilu-tion factor of the extracts (the two factors which mainly affectedthe analyte adsorption on GCB) as far as the composition of thedesorbing phase may be possible. The possibility of a fine tuning ofthese parameters, however, would be excluded a priori, owing tothe very large range of polarity and polarizability of the target com-pounds. Although previous experiences with the analysis of linearalkylbenzene sulfonates (LAS) and their carboxylated metabolitesshowed that an increase of pH from 2 to 1 displaced short chain car-boxylated metabolites from the GCB surface [35], as PFAS possesspkahigher than that of carboxylic acids, we tested also this varia-tion, but recoveries of shorter chain compounds deeply decreased.Also the dilution factor sample-to-acidified water could be furtheroptimized, as it was found that it depends on both matrix and com-pound [22,29]. A dilution factor 1:25 did not improve recoveries,whereas decreasing dilution to 1:5 decreased recoveries of PFPeA,PFHeA and PFBS. It is possible that a further increase of methy-lene chloride-to-methanol ratio, and displacer concentration mightfurther increase recoveries of higher isomers (more polarizables),but it might also produce problem of salt solubility and increasecoextracted compounds with a consequent ESI source contamina-tion. Therefore, also by considering that isotopic ISs were used,the selected extraction conditions were those obtained from thescreening DoE and reported in Section 2.5.3.3. Method validationThe performance of the method was evaluated under the opti-mized conditions in terms of matrix effects, range of instrumentallinearity, accuracy, repeatability and linearity at low concentra-tion. The matrix effect was estimated, as reported in Section 2.7, bycomparing the slopes of six point calibration curves of the isotopicalIS in pure solvent and in pooled milk extracts. By using the isotopicIS instead of the analytes, it was not necessary subtracting the peakareas corresponding to the native analytes present in the sampleor blanks. The slope ratio of the regression lines (standard/matrix)ranged between 0.96 and 1.04, therefore matrix effect could beconsidered negligible.Instrumental linearity was evaluated in separate experiments inwhich calibration lines were constructed in a concentration rangewider than that used for quantification of analytes in the sam-ples. For all compounds, linearity in matrix was observed in therange LOQ–10 g L−1. Coefficients of determination R2ranged from0.9982 to 0.9999. Linearity of the method was also tested by usingthe isotopic ISs added to five aliquots of a pooled milk sample in therange of 2.5–100 ng L−1, and submitting the samples to the wholeprocedure. In this case, R2were in the range of 0.81–0.88, but theintercepts did not differ significantly from 0 at p = 0.05 and weresymmetrically distributed.Although our absolute recoveries of analytes were higher thanthose reported in a recent paper for a QuEChERS extractionmethod [5], they showed quite high imprecision, probably dueto one or more uncontrolled factors during extraction from milk.PFAS possess peculiar physical–chemical characteristics, some-times very difficult to be predicted or rationalized. For this reasonthe use of the appropriate isotopical IS is strongly recommended.Standard reference materials for determination of PFAS in cow milkwere not available; therefore, the accuracy and precision of themethod were assessed using milk samples fortified at two con-centration levels, such as 50, and 10 ng L−1(Fig. 2). Six aliquotsfrom three different stock fortified milk samples were individu-ally processed and analyzed.13C4-PFOS was used as IS for PFBS,and13C2-PFDoA for PFTrDA and PFTeDA. The results are shown inTable 2, where instrumental limits of detection (LOD) and quantifi-cation (LOQ), and method detection (MDL) and quantification limits(MQL) are also reported. As can be seen, the use of the isotopic ISsdeeply improve both accuracy and reproducibility.We based the identification criteria on the Commission Decision2002/657/EC. Therefore, a tolerance of 2.5% of tR, and a differenceof less than 20% between the two transitions intensity ratio wereconsidered acceptable for confirmation. However, for PFPeA onlyone transition was available, thus confirmation might be possibleonly with high resolution MS/MS. As consequence of the criteriachosen, based on the concept that a compound can be quantita-tively measured only if it has been unambiguously identified, LODsand MDLs were calculated using the less intense transition. LODsand LOQs were estimated from the signal-to-noise ratio (S/N) inthe standard solution as the quantity that give a S/N = 3 and = 10,respectively. However when a LOQ was apparently lower than thecorresponding LOD, it was considered = LOD. MDLs and MQLs wereestimated to be the blank value +3 and 10 times the standard devi-ation of the blank, respectively. and the same criteria were appliedwhen a MQL was apparently less than the corresponding MDL.Thereby, the estimated MDL values corresponds to CC, (Commis-sion Decision 2002/657/EC), but it should be pointed out that theyare obtained by using the same batch of solvents and materials.Taking into account the fact that blank reproducibility might berather different from laboratory to laboratory, the MDLSand MQLsreported could vary widely, and should be considered as approxi-mate.
การแปล กรุณารอสักครู่..