Analytica Chimica Acta 594 (2007) 257264

Validation according to European Commission Decision 2002/657/EC of a conrmatory method for aatoxin M1 in milk based on immunoafnity columns and high performance liquid chromatography with uorescence detection
Marilena Muscarella a , Sonia Lo Magro a , Carmen Palermo b , Diego Centonze b,
a

Istituto Zooprolattico Sperimentale della Puglia e della Basilicata, Foggia, via Manfredonia 20, 71100 Foggia, Italy b DISACD Dipartimento di Scienze Agro-Ambientali, Chimica e Difesa Vegetale, Universit degli Studi di Foggia, a via Napoli 25, 71100 Foggia, Italy Received 28 March 2007; received in revised form 10 May 2007; accepted 21 May 2007 Available online 25 May 2007

Abstract A high performance liquid chromatographic method with uorimetric detection for the determination of aatoxin M1 (AFM1 ) in milk has been optimized and validated according to Commission Decision 2002/657/EC by using the conventional validation approach. The procedure for determining selectivity, recovery, precision, decision limit (CC ), detection capability (CC ) and ruggedness of the method has been reported. The results of the validation process demonstrate the agreement of the method with the provisions of Commission Regulation 401/2006/EC. The mean recovery calculated at three levels of fortication (0.5, 1.0, and 1.5-fold the MRL) was 91% and the maximum relative standard deviation value for the within-laboratory reproducibility was 15%. Limit of detection (LOD) and limit of quantitation (LOQ) values were 0.006 g kg1 and 0.015 g kg1 while the CC and CC values were 0.058 g kg1 and 0.065 g kg1 , respectively. The relative expanded measurement uncertainty of the method was 7%. The method was not affected by slight variations of some critical factors (ruggedness minor changes) as pre-treatment and clean-up of milk samples, thermal treatment and different storage conditions, as well as by major changes valued in terms of milk produced by different species (buffalo, goat and sheep). The method allowed accurate conrmation analyses of milk samples, resulted positive by the screening method. In fact, the Z-score values attained in a prociency test round were well below the reference value of 1, proving the excellent laboratory performances. 2007 Elsevier B.V. All rights reserved.
Keywords: Aatoxin M1 ; Milk; Validation; High performance liquid chromatography; Fluorescence detection; 2002/657/EC

1. Introduction Aatoxins are secondary metabolites produced by moulds belonging to Aspergillus species (Aspergillus avus and Aspergillus parasiticus). Mutagenic, teratogenic and carcinogenic compounds are highly toxic for animals and humans, producing acute liver damage (e.g., cirrhosis) [1,2]. Aatoxins can be found [3,4] in a wide variety of agricultural products and animal feeds as a result of moulds contamination before or during harvest or improper storage [1]. Among them aatoxin B1 (AFB1 ) is the most carcinogenic and it is classied by the International Agency for Research on Cancer (IARC) as group 1 of human carcinogens [5].

Corresponding author. Tel.: +39 0881 589 104; fax: +39 0881 740211. E-mail address: centonze@unifg.it (D. Centonze).

When animals eat foodstuffs containing AFB1 this toxin is metabolized and accumulated in milk as aatoxin M1 (AFM1 ) [6]. Many studies have demonstrated that also AFM1 displays toxic and carcinogenic effects, therefore it has been included in the IARC group 1 [7]. Unfortunately, AFM1 is relatively stable either to thermal processes like pasteurization and sterilization or during preparation and storage of various dairy products [8]. For these reasons, AFM1 was included in the group B of annex I of Council Directive 96/23/EC [9], which comprises those substances with established maximum residue limits (MRLs). European Community has adopted for AFM1 MRLs of 0.050 g kg1 in milk [10] and 0.025 g kg1 in baby food products [11]. Several procedures for the determination of AFM1 have been developed, comprising immunological methods [1214], such

0003-2670/$ see front matter 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2007.05.029

258

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264

as enzyme linked immunosorbent assays (ELISA) commonly used for screening purposes and new methods based on electrochemical immunosensors [15] and ow injection immunoassay systems [16]. In the last years several methods based on high performance liquid chromatography (HPLC) were reported for the determination of AFB1 , B2 , G1 , and G2 in some foodstuffs [1720] and AFM1 in milk [2127]. Immunoafnity column sample clean-up followed by normal or reversed phase HPLC separation with uorimetric detection is mostly used for quantitative determination of AFM1 because of characteristics of specicity, high sensitivity, and simplicity of operation. In spite of several published conrmatory methods for the control of AFM1 in milk, only few validations are reported [2527], which do not strictly comply with recent European Commission guidelines [2830]. In fact, each ofcial method should be validated not only in a collaborative trial study but also by following the validation guidelines aimed at the attainment of minimum performance criteria, as dened in the recent Commission Regulation 401/2006/EC [28]. Analytical methods for food control should also comply with the provisions of Commission Regulation 882/2004/EC [30], a document about ofcial controls performed to ensure the verication of compliance with feed and food laws, animal health, and animal welfare rules. Therefore, validation procedures of analytical methods are required in order to provide accurate and reproducible results within- and inter-ofcial control laboratories, which are involved in the monitoring and risk-assessment studies, as well as in ofcial controls. In 2002 the European Union issued Decision 2002/657/EC [29] aimed at the ensurement of both the quality and the correct interpretation of the analytical results attained by control ofcial laboratories. In this document performance criteria and procedures for the validation of screening and conrmatory methods are established. Nevertheless, Regulation 401/2006/EC [28] and Regulation 882/2004/EC [30] establish performance criteria for methods of analysis of AFM1 in milk, but the validation procedures are not mentioned. Therefore, we have decided to follow the validation approach of Decision 2002/657/EC [29] verifying minimum criteria described in Regulation 401/2006/EC [28]. The validation protocol mentioned in Decision 2002/657/CE [29] allows evaluating some performance characteristics, such as selectivity, linearity, precision, trueness, decision limit (CC ), detection capability (CC ) and ruggedness, in agreement with Regulation 882/2004/EC [30] and UNI CEI EN ISO/IEC 17025 [31], indicated by the European Commission as accreditation protocol for the competence of testing and calibration laboratories. In this paper we report the validation of an optimized method for the determination of AFM1 in bovine milk, based on immunoafnity chromatography clean-up followed by HPLC separation with uorimetric detection. The optimized analytical method has been validated according to Decision 2002/657/EC [29] by using the conventional validation approach, and the procedure for determining selectivity, recovery, precision, decision

limit, detection capability and ruggedness of the method has been reported. Validation data together with bottom-up approach has been used for the evaluation of measurement uncertainty. Furthermore, laboratory performances were tested by taking part in a prociency test round. 2. Experimental 2.1. Chemicals and standard solutions Acetonitrile, methanol, chloroform and water were of HPLC grade and were purchased from Baker (Deventer, Holland). Acetonitrile and methanol used in ruggedness studies were purchased from Carlo Erba (Milano, Italy). The reference standard of AFM1 (from Aspergillus avus, 10 g) was purchased from SigmaAldrich (Milano, Italy). A 5 g mL1 AFM1 standard solution was prepared by dissolving 10 g of powder in 2 mL of chloroform; this solution was stable for 9 months at 20 C. The concentration was tested by a spectrophotometric method according to the procedure no. 971.22 published in the Ofcial Methods of AOAC [32]. A 200 g L1 diluted solution was prepared in methanol by using a foil-wrapped volumetric ask; this solution was stable for 3 months at 20 C. The following 20 g L1 diluted solution, prepared in mobile phase by using a foil-wrapped volumetric ask, was stable for 1 month at 20 C. Working standard solutions (0.3, 0.6, 0.9, 1.5 and 2.0 g L1 ) were prepared by appropriate dilution in mobile phase and stored at 20 C when not in use; these solutions were stable for at least 3 weeks. 2.2. Equipment and analytical conditions An Agilent 1100 Series (Agilent Technologies, Waldbronn, Germany) consisted of an LC system equipped with a membrane degasser, a quaternary pump, an autosampler, a 20 L loop, a termostated column compartment and a uorescence detector set at 360 nm (emission) and 440 nm (excitation) was used for the analyses. The LC column was a ZORBAX Eclipse XDB-C18, 250 mm 4.6 mm, particle size 5 m, purchased from Agilent. The mobile phase consisted of water (A), acetonitrile (B) and methanol (C) owed at 0.8 mL min1 . The optimized elution gradient was the following: 6 min (06 min) isocratic step at 55% A, 24% B and 21% C; 1 min linear gradient (67 min) to 15% A, 40% B and 45% C; an isocratic step from 7 to 10 min at 15% A, 40% B and 45% C; a 1 min linear gradient (1011 min) to the initial eluent composition (55% A, 24% B and 21% C), at which the system was re-equilibrated for 6 min. 2.3. Sample preparation A 40 g milk sample was centrifuged at 10 C in a polypropylene tube at 3500 rpm for 10 min. 10 g of skimmed milk were puried by an immunoafnity column (RIDA Aatoxin column, R-Biopharm, Darmstadt) at a slow steady volume (ow rate of ca. 2 mL min1 ), which was previously conditioned with

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264

259

2 mL of water. The cartridge was washed with 10 mL water, dried by vacuum for 10 s and eluted with 750 L of methanol. The eluate was evaporated at 45 C with a nitrogen ow (Dubnof Bath BSD/D) to bare dryness. The residue was solubilized in 500 L of mobile phase and ltered by an Anotop 10 LC (0.2 m, 10 mm, Whatman). The procedure resulted in a 20-fold concentration of AFM1 evaluated against a known concentration added to a blank real sample. 3. Results and discussion 3.1. Model independent performance characteristics On the basis of Decision 2002/657/EC [29] some performance characteristics have to be determined before to proceed with the validation approach: selectivity, which is important for discrimination between the analyte and closely related substances (e.g., isomers) or matrix interferents; ruggedness (minor changes), which is the degree of reproducibility of test results obtained by the analysis of the same samples under a variety of normal test conditions. As no certied reference materials (CRMs) were available, instead of trueness the recovery parameter was determined. Moreover, as aatoxin M1 is known to be stable, the stability parameter in real samples and standard solutions was not assessed. 3.2. Selectivity towards interferences Selectivity represents one of the main problems to be solved in order to carry out accurate analyses. For instance the analytical procedure [33], suggested in the ISTISAN report [34], as the reference method for AFM1 analysis in bovine raw milk samples, shows a low selectivity. In fact, in the chromatograms (data not shown) of some blank real samples (negative by ELISA screening) matrix interferences were present at the same analyte retention time. A further problem presented by this approach is a carry over of impurities between runs. Improvement of the method selectivity and column cleaning and re-equilibration after each run has been achieved by the optimization of the elution gradient (see Section 2). In preliminary experiments the most important aatoxins have been tested, including AFM2 that is the most known AFM1 isomer. All investigated compounds showed a retention time signicantly different from that of AFM1 . Selectivity towards naturally present compounds (e.g., metabolites, endogenous substances, etc.) has been established by the analysis of 20 independent blank samples of bovine raw milk, obtained from different producers, and whole milk (U.H.T. and pasteurized) purchased from a local market, found negative by ELISA screening [35,36]. As can be seen in Fig. 1(AD), which shows typical chromatograms obtained for spiked and blank milk samples, the proposed method is able to distinguish between the analyte and other matrix components in each kind of milk analysed; no matrix interfering substances at the retention time of AFM1 (2.5%) were observed.

Fig. 1. Typical chromatograms obtained by the proposed method for a bovine milk sample spiked with AFM1 at 0.050 g kg1 (A), raw (B), pasteurized (C), and UHT (D) blank milk samples.

3.3. Ruggedness (minor changes) Ruggedness can be evaluated in the case of minor changes by a deliberate introduction of minor reasonable variations by the laboratory and the observation of their consequences. The pre-investigative study has been carried out by selecting factors relevant to the sample pre-treatment, clean-up and analysis, which may inuence the measurement results. These factors should include the source and the age of reagents, solvents, standards and sample extracts, the rate of heating, the temperature, as well as other factors that may affect results in the laboratory. They should be modied in an order of magnitude that matches the deviations usually encountered among laboratories. Com-

260

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264

Table 1 Experimental design for the ruggedness study (minor changes) by using the fractional factorial Youden approach Selected factor Experiment number 1 Milk thermal treatment Evaporation temperature Storage Eluate ltration Immunoafnity cartridge lot Lot and commercial company of acetonitrile Lot and commercial company of methanol Observed result Selected factor Milk thermal treatment Evaporation temperature Storage Eluate ltration Immunoafnity cartridge lot Lot and commercial company of acetonitrile Lot and commercial company of methanol
a

2 A B c D e f g T

3 A b C d E f g U Abbreviation A, a B, b C, c D, d E, e F, f G, g

4 A b c d e F G V High level

5 a B C d e F g W

6 a B c d E f G X

7 a b C D e f G Y Low level

8 a b c D E F g Z

A B C D E F G S Units
C

Raw milk 47 C New milk Eluate Anotop ltration Lot A Lot A, J.T. Baker Lot A, J.T. Baker

Pasteurized 42 C Milk stored at +4 C for 1 day No ltration Lot B Lot B, Carlo Erba Lot B, Carlo Erba

Dimensionless.

monly, variations of 10% with respect to the nominal value have been used [29]. In the present work ruggedness was evaluated through the fractional factorial Youden design [29,3638] by selecting seven factors as variables for Youden test: milk thermal treatment (raw or pasteurized), evaporation temperature, storage at +4 C for 1 day, eluate ltration, immunoafnity cartridge lot, and lot and commercial company of methanol and acetonitrile. As displayed in Table 1, according to Scortichini et al. [36], eight experiments were carried out for the evaluation of the seven selected factors by using eight spiked bovine milk samples at the MRL. The effect of each factor was calculated by subtracting the mean result obtained with the variable at high level (reference value) indicated by the capital letter and the mean result achieved with the factor at low level (investigated factor), marked with the corresponding small letter. This difference is indicated as Di , and the standard deviation of the differences SDi has been calculated by the following equation [29]: SDi = 2
2 Di 7

where n is the number of experiments carried out at each level for each parameter (n = 4 in our experimental plan) and the reproducibility standard deviation (4.6 ng kg1 ) at MRL is indicated as S.D. For all investigated factors, the obtained t-value was compared with the 2-tailed t-critical value (tcrit ) at n 1 degrees of freedom and = 0.05 (95% condence level). If the t-value is higher than tcrit the investigated factor shows a signicant inuence and the method is not sufciently robust against the chosen modication. The results of ruggedness studies, performed by the experimental plan shown in Table 1, are reported in Table 2. The calculated standard deviation of difference SDi (3.9 ng kg1 ) was lower than the estimated method precision at MRL (4.6 ng kg1 ). This result demonstrates that all together selected factors have no effect on the result and the method is robust. Moreover, each t-value is lower than tcrit (2.11, = 17), conrming that the method is not affected by slight variations of each critical factor as pre-treatment, clean-up, and different storage conditions of milk samples. Furthermore, in independent experiments no appreciable differences were observed for
Table 2 Results of the ruggedness study (minor changes) Selected factor Difference (D) of %absolute recovery values 5.0 1.2 2.5 0.9 3.5 1.3 1.6 t-Value

If SDi is signicantly greater than the standard deviation of the method, it means that all together the selected factors have an effect on the result, even if every single factor does not show a signicant inuence. Inuence of each investigated factors has been determined by using a t-test. The experimental t-value is given by the equation: n |Di | t= 2 S.D.

Milk thermal treatment Evaporation temperature Storage Eluate ltration Immunoafnity cartridge lot Lot and commercial company of acetonitrile Lot and commercial company of methanol

1.6 0.4 0.8 0.3 1.2 0.4 0.5

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264

261

both U.H.T. and raw milk stored at 20 C for 2 and 20 days, respectively. 3.4. Calibration curves and limits of detection and quantitation Method linearity was determined by three series of analyses, performed on three different days by the injection of ve standard solutions of AFM1 at concentrations of 0.3, 0.6, 0.9, 1.5, and 2.0 g L1 , corresponding to 15, 30, 45, 75, and 100 ng kg1 in matrix. Prior to the calculation of regression line and relevant parameters the absence of outliers was controlled by the Dixon test and any measurement was eliminated. The calibration curve, obtained by plotting peak area of the three series of analyses, expressed in luminescence unit versus AFM1 concentration in the range 0.32.0 g L1 gave the linear regression equation y = 0.702x + 0.013 with a determination coefcient (r2 ) of 0.999. The goodness-of-t of the data to the curve is obtained in terms of response factor distribution yi /xi that falls in the range y/xmean 10% [39]. Furthermore, any systematic instrumental bias can be ruled out since the intercept includes the zero value at 95% condence level. Slope and intercept of the calibration curve were statistically compared by a t-Student test at 95% condence level with those obtained by the analysis of spiked blank milk samples, while residual standard deviations were tested by the F-test at 95% condence level. Comparisons showed no difference between calibration curves obtained from standard solutions and spiked blank samples. Therefore, for routine analysis calibration curves from standard solutions were used. AFM1 detection limit (LOD) and quantitation limit (LOQ) were calculated by using the slope (b) of the matrix calibration curve and the residual standard error (sy/x ) by means the following equations [37]: LOD = 3.3 sy/x /b; LOQ = 10 sy/x /b. LOD and LOQ values were 0.12 g L1 and 0.30 g L1 , respectively, corresponding to 0.006 g kg1 and 0.015 g kg1

in matrix. LOQ is satisfactory with respect to MRLs of 0.050 g kg1 and 0.025 g kg1 that European Community has adopted for AFM1 in milk and baby food products, respectively. 3.5. Conventional validation procedure 3.5.1. Precision and recovery Precision and method recovery were determined, according to Decision 2002/657/EC [29], by performing tests on three sets of blank raw milk samples (six replicates each) fortied with AFM1 at concentration of 0.5, 1.0, and 1.5 times the MRL (i.e., 0.025, 0.050 and 0.075 g kg1 ). Samples were analyzed on three different days with the same instrument but by three different operators, corresponding to a total number of 54 samples. Recovery values were improved signicantly by the optimization of the elution volume in the immunoafnity column clean-up (see Section 2). Precision has been calculated either in terms of repeatability (R.S.D.r ), the variability of independent test results obtained with the same method on identical test items in the same laboratory by the same operator using the same equipment, or in terms of within-laboratory reproducibility (R.S.D.R ), the variability of independent test results obtained with the same method on identical test items in the same laboratory by different operators in different time, using the same equipment. Precision and recovery results, reported in Tables 3 and 4, respectively, have been previously processed by the ShapiroWilk test to verify the normal distribution and by the Dixon test to reject possible outliers. Cochran test was then applied to assess the variance homogeneity in the range of concentration considered. Finally, ANOVA test was performed in order to verify the absence of systematic errors. The acceptability of these tests allowed to calculate relative standard deviations (R.S.D.s ) over the entire range of replicate measurements. Regulation 401/2006/EC [28] issued that the permitted value of experimental R.D.S. for each concentration value must be

Table 3 Repeatability and within-laboratory reproducibility for the determination of AFM1 in spiked milk samples Fortication level ( g kg1 ) 0.025 (0.5 MRLc ) Parameter Meand ( g kg1 ) S.D.e ( g kg1 ) R.S.D. (%) Mean ( g kg1 ) S.D. ( g kg1 ) R.S.D. (%) Mean ( g kg1 ) S.D. ( g kg1 ) R.S.D. (%) Opa 1 0.025 0.003 12.0 0.045 0.005 10.6 0.070 0.006 8.8 Op 2 0.022 0.003 14.4 0.046 0.005 11.1 0.072 0.006 7.9 Op 3 0.024 0.004 16.8 0.047 0.005 9.4 0.067 0.005 8.0 Overallb 0.024 0.004 15.0 0.046 0.005 10.0 0.070 0.006 8.4

0.050 (1.0 MRL)

0.075 (1.5 MRL)

a b c d e

Op, operator: relative standard deviation under repeatability conditions (R.S.D.r ) (six replicates at each level). Relative standard deviation under within-laboratory reproducibility conditions (R.S.D.R ) (eighteen replicates at each level). MRL, maximun residue limit. Mean concentration calculated from the calibration curve. S.D., standard deviation.

262

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264

Table 4 Recoveries of AFM1 from spiked blank milk samples Fortication level ( g kg1 ) 0.025 (0.5 MRLc ) 0.050 (1.0 MRL) 0.075 (1.5 MRL)
a b c d

Parameter Mean recovery (%) R.S.D.d (%) Mean recovery (%) R.S.D. (%) Mean recovery (%) R.S.D. (%)

Opa 1 95.9 12.0 87.9 10.6 91.3 8.8

Op 2 83.5 14.4 88.4 11.1 93.6 7.9

Op 3 97.2 16.8 94.8 9.4 89.2 8.0

Overallb 92.2 14.4 90.4 10.4 91.4 8.2

Op, operator. Six replicates at each level. Eighteen replicates at each level. MRL, maximun residue limit. R.S.D., relative standard deviation.

below twice-fold the value derived by Horwitz equation, which provides the expected R.S.D.% only on the basis of the concentration, independently of the matrix and analytical method used. Nevertheless, for mass fractions lower than 120 g kg1 , the R.S.D. can be predicted in a better manner by the Thompson equation [40], which provides a value of 44% as maximum permitted inter-laboratory reproducibility R.S.D. and of 29% as maximum permitted repeatability. As can be seen in Table 3, both R.S.D.R and R.S.D.r values are lower than calculated repeatability by Thompson equation; this result indicates that the method satises minimum performance criteria established by commission Regulation 401/2006/EC [28]. The recoveries determined on three different days at the concentration levels of 0.025, 0.050 and 0.075 g kg1 were in the range of 83.597.2% with R.S.D.s in the range of 7.116.3%. The inter-day and inter-level mean recovery value, which has been used to correct routine analysis results, was of 91%. The recovery values achieved were all in agreement with performance criteria dened in Regulation 401/2006/EC [28] that recommends values in the range of 60120% for concentrations in the range of 0.0100.050 g kg1 , and 70110% for concentrations above 0.050 g kg1 . The short term stability of test results was assessed by the determination of recoveries in three different days. As can be seen from overall recoveries and relevant standard deviations, the method resulted stable over the tested period of time. 3.5.2. Limit of decision (CC ) and detection capability (CC ) Decision 2002/657/EC [29] introduces two new parameters, CC (decision limit) and CC (detection capability), to replace the old concepts of limit of detection and limit of quantitation. CC was dened as the limit above which samples are concluded to be non-compliant, with an error probability of 5%, while CC was dened as the smallest content of the substance that may be detected, identied and/or quantied in a sample with an error probability of 5%. In the case of substances with MRLs, decision limit and detection capability must be greater than MRL and and errors must be equal or less than 5%. The limit of decision represents an index of results dispersion and, as underlined by regulation 401/2006/EC [28], is possible to estimate the measured uncertainty by using CC .

The critical concentration for MRL compliance (CC , = 0.05) were calculated, according to Decision 2002/657/EC [29], from the MRL value of 0.050 g kg1 plus 1.64 times the standard deviation of the fortied samples at the MRL. CC is obtained adding to CC 1.64 times the same standard deviation. CC and CC values, calculated as above reported, were 0.058 g kg1 and 0.065 g kg1 , respectively. 3.5.3. Ruggedness (major changes) The analytical method was also tested by using buffalo, goat and sheep milk samples and method ruggedness under conditions of major changes was evaluated by using the Youden experimental design. To evaluate the method ruggedness in analyses of buffalo, goat and sheep milk samples, two independent Youden tests were performed. As reported in the case of ruggedness in conditions of minor changes, eight experiments were carried out with spiking milk samples at the MRL, and the seven factors chosen as variables for the Youden test were the species and six ctitious factors, respectively. The use of a ctitious variable means no variation of analysis conditions. In this case the Youden experimental design requires eight independent experiments: four with bovine milk and four with buffalo, goat or sheep milk. The effect of each variable was calculated, as in the case of minor changes, by subtracting the mean result obtained with the variable at high level indicated by the capital letter, and the mean result achieved with the variable at low level, indicated with the corresponding small letter. In this case only the effect of different species was measured. Each species gave a calculated standard deviation of difference SDi (buffalo milk: 3.2 ng kg1 ; goat and sheep milk: 4.4 ng kg1 ) lower than the estimated method precision at the MRL (4.6 ng kg1 ). This result demonstrates that the variation of species has no effect on the result and the method is also applicable to buffalo and goat and sheep milk samples. 3.6. Expanded measurement uncertainty The evaluation of uncertainty of analytical results is compulsory for laboratories accredited according to ISO 17025 [31], and several methods for the determination of these parameters were proposed [4143]. A possible approach referred as bottom-

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264

263

up or deconstructive, developed by metrologists and physicists, allows to obtain the measurement uncertainty of results by combining the contributions of all uncertainty sources. Nevertheless, for more complex chemical methodologies, the quantication of all relevant uncertainty contributions is not an easy task [44,45]. For instance, some unidentiable uncertainty sources can affect the accuracy determined by the validation. EURACHEM/CITAC Guide [42] suggested the use of validation data and quality assurance results to evaluate the measurement uncertainty. This possibility represents a very practical solution for analytical chemists who have to validate methods [45]. In this work we have used the bottom-up method together with validation data attained from each step of the analytical procedure [45]. Uncertainties can be listed as class A, due to casual errors expressed in terms of standard deviation attained for replicate measurements, and class B derived from external sources, as for instance previous data, instrumental calibration, reference materials and standard purity, expressed as standard deviation indicated by the supplier. The components of uncertainty have been then identied and classied as groups A and B (see Table 5). On the basis of uncertainties propagation law, the AFM1 concentration relative uncertainty has been calculated by the equation: u(CAFM1 ) = (u(C))2 + (u(Vf ))2 + (u(w))2

calculated as the difference between CC and MRL value (0.050 g/kg). The relative uncertainty valued as percent ratio of this difference with respect to MRL was 16%, resulting very close to the expanded relative uncertainty calculated with classical method for 1 replicate (17%). 3.7. Prociency test round The prociency test round was organised by Milk Standard Laboratory of A.I.A. (Italian Association of Breeders) in May 2006. Four unknown milk samples were analyzed in duplicate and the results were evaluated in terms of the Z-score, as calculated by the prociency test organizer. The Z-scores obtained in the A.I.A. ring test performed on unknown milk samples analyzed in duplicate (Z1 = 0.085, Z2 = 0.460, Z3 = 0.332, Z4 = 0.000) and the value of the laboratory Z-score (ZLAB = 0.186) were all below the reference value of 1, proving the technical competence of the laboratory. 4. Conclusion An optimized analytical method based on immunoafnity column clean-up followed by a liquid chromatographic separation with uorimetric detection, for the determination of AFM1 in milk was validated according to Decision 2002/657/EC [29], in terms of selectivity, recovery, precision, decision limit, detection capability and ruggedness (minor and major changes). The result of validation process demonstrated the agreement of method performances with the provisions of Regulation 401/2006/EC [28] (mean recovery of 91% and maximum R.S.D.R under within-laboratory conditions of 15%), then allowing accurate conrmation analyses of milk samples resulted positive by the screening method. The expanded measurement uncertainty value, obtained by validation data with the bottom-up approach, and excellent results achieved in a prociency test round conrmed the laboratory technical competence in the determination of AFM1 in milk samples, with the accuracy level corresponding to the requirements of the recent European regulation. Acknowledgements Istituto Zooprolattico Sperimentale di Puglia e BasilicataFoggia is gratefully acknowledged for providing the nancial support. Mr. Pasquale DAntini, from Istituto Zooprolattico Sperimentale di Puglia e Basilicata-Foggia, is gratefully acknowledged for technical assistance. The authors are grateful to Dott. Domenico Palermo, Head of Department of Chemistry, Istituto Zooprolattico di Puglia e Basilicata-Foggia, for his valuable advice and encouragement. References
[1] M.J. Sweeney, A.D.W. Dobson, Int. J. Food Microbiol. 43 (1998) 141158. [2] H.S. Hussein, J.M. Brasel, Toxicology 167 (2001) 101134. [3] M. Peraica, B. Radic, A. Lucic, M. Pavlovic, World Health Organ. 77 (1999) 754756.

where u indicates the relative uncertainty, CAFM1 is the ana lyte concentration in the sample, Vf is the volume of the nal extract, and w is the sample weight. The determination of u(C) has been performed by considering four sources of uncertainty: (a) preparation of the standard; (b) method reproducibility; (c) method recovery; (d) calibration curve. A relative expanded measurement uncertainty of 7% was calculated by using a coverage factor k of 2, corresponding approximately to a 95% condence level [43]. In the routine analysis the relative measurement uncertainties of class A can be corrected considering the number of replicates for each samples; for instance measurement uncertainties of 17% and 13% were achieved in the case of 1 and 2 measurement replicates, respectively. As reported in Regulation 401/2006/EC [28] for food of animal origin and in accordance with commission Decision 2002/657/EC [29], the measurement uncertainty can also be
Table 5 Classication of measurement uncertainty components and relevant values expressed as relative standard deviation Class A Component Intra-laboratory reproducibility uncertainty Calibration curve uncertainty Method recovery uncertainty Volume uncertainty Weight uncertainty Standard uncertainty Uncertainty value 0.026 0.005 0.015 0.006 Negligible 0.01

264

M. Muscarella et al. / Analytica Chimica Acta 594 (2007) 257264 [26] P. Gallo, A. Salzillo, C. Rossini, V. Urbani, L. Serpe, Ital. J. Food Sci. 3 (18) (2006) 251259. [27] A.C. Manetta, L. Di Giuseppe, M. Giammarco, I. Fusaro, A. Simonella, A. Gramenzi, A. Formigoni, J. Chromatogr. A 1083 (2005) 219222. [28] European Commission, Regulation (EC) No. 401/2006 of 23 February 2006, Off. J. European Union L70 (2006) 1234. [29] European Commission, Decision 2002/657/EC of 12 August 2002, Off. J. European Union L221 (2002) 836. [30] European Commission, Regulation (EC) No. 882/2004 of 29 April 2004, Off. J. European Union L165 (2004) 1141. [31] UNI CEI EN ISO/IEC 17025 (2000). [32] Association of Ofcial Analytical Chemists (AOAC), EDN sixteenth ed. Ofcial Methods of Analysis of AOAC, II n.971.22, AOAC, Gaithersburg, 1997. [33] D.N. Mortimer, J. Gilbert, M.J. Shepherd, J. Chromatogr. 407 (1987) 393398. [34] ISTISAN Report 96/34, ISSN 11233177 (1996) 218219. [35] I. Pecorelli, R. Bibi, L. Fioroni, R. Galarini, J. Chromatogr. A 1032 (2004) 2329. [36] G. Scortichini, L. Annunziata, M.N. Haouet, F. Benedetti, I. Krusteva, R. Galarini, Anal. Chim. Acta 535 (2005) 4348. [37] E.J.C. Miller, J.N. Miller, Statistics for Analytical Chemistry, third ed., Ellis Horwood TR Prentice-Hall, West Sussex, UK, 1993, pp. 115 117. [38] W.J. Youden, E.H. Steiner, Statistical Manual of AOAC, Association of Ofcial Analytical Chemists, Gaithersburg, 1975, pp. 3335. [39] P. Van Wiele, F. Van Hoo, A. Bruchet, I. Schmitz, J.L. Guinumant, F. Acobas, A. Ventura, F. Sacher, I. Bobeldijk, M.H. Marecos do Monte, in: A. Fajgelj, A. Ambrus (Eds.), Optimization and Evaluation of Multi-Residue Methods for Priority Pesticides in Drinking and Related Waters (Principles and Practices of Method Validation), The Royal Society of Chemistry, Cambridge, 2000, pp. 918. [40] M. Thompson, Analyst 125 (2000) 385386. [41] EURACHEM Guide, in: S.L.R. Ellison, M. Rosslein, A. Williams (Eds.), Quantifying Uncertainty in Analytical Measurement, rst ed., 1995. [42] EURACHEM/CITAC GUIDE CG 4, in: S.L.R. Ellison, M. Rosslein, A. Williams (Eds.), Quantifying Uncertainty in Analytical Measurement, second ed., 2000. [43] Nordic Committee on Food Analysis (NMKL), Estimation and expression of measurement uncertainty in chemical analysis, NMKL Secretariat, Finland procedure No. 5, second ed., 2003. [44] E. Hund, D.L. Massart, J. Smeyers-Verbeke, Trends Anal. Chem. 20 (2001) 394406. [45] I. Pecorelli, R. Bibi, L. Fioroni, A. Piersanti, R. Galarini, Anal. Chim. Acta A 529 (2005) 1520.

[4] D.L. Eaton, J.D. Groopman (Eds.), The Toxicology of Aatoxins, Academic Press, New York, 1994, pp. 383426. [5] International Agency for Research on Cancer (IARC), IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, 56, Lyon, 1993, pp. 245362. [6] H.P. van Egmond, Mycotoxins in Dairy Products, Elsevier Applied Science, London, New York, 1989, pp. 154. [7] International Agency for Research on Cancer (IARC), IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, 56, Lyon, 2002, pp.171176. [8] L. Stoloff, J. Food Prot. 43 (1980) 226230. [9] European Parliament and Council, Directive 96/23/EC of 23 May 1996, Off. J. European Union L125 (1996) 1016. [10] European Commission, Regulation (EC) No. 466/2001 of 8 March 2001, Off. J. European Union L77 (2001) 113. [11] European Commission, Regulation (EC) No. 683/2004 of 13 April 2004 amending Regulation (EC) No. 466/2001, Off. J. European Union L106 (2004) 35. [12] K. Thirumala-Devi, M.A. Mayo, A.J. Hall, P.Q. Craufurd, T.R. Wheeler, F. Waliyar, A. Subrahmanyam, D.V.R. Reddy, J. Agric. Food Chem. 50 (2002) 933937. [13] M. Magliulo, M. Mirasoli, P. Simoni, R. Lelli, O. Portanti, A. Roda, J. Agric. Food Chem. 53 (2005) 33003305. [14] J. Stroka, E. Anklam, Trends Anal. Chem. 21 (2002) 9095. [15] L. Micheli, R. Greco, M. Badea, D. Moscone, G. Palleschi, Biosens. Bioelectron. 21 (4) (2005) 588596. [16] M. Badea, L. Micheli, M.C. Messia, T. Cardigliota, E. Marconi, T. Mottram, M. Valesco-Garcia, D. Moscone, G. Palleschi, Anal. Chim. Acta 520 (2004) 141148. [17] J. Blesa, J.M. Soriano, J.C. Molto, R. Marin, J. Manes, J. Chromatogr. A 1011 (2003) 4954. [18] G. Tav ar-Kalcher, K. Vrta , U. Pestevek, A. Vengut, Food Control 18 c c s s (2007) 333337. [19] D. Chan, S.J. MacDonald, V. Boughtower, P. Brereton, J. Chromatogr. A 1059 (2004) 1316. [20] H.K. Abbas, W.P. Williams, G.L. Windham, H.C. Pringle, W. Xie, W.T. Shier, J. Agric. Food Chem. 50 (2002) 52465254. [21] C. Cavaliere, P. Foglia, E. Pastorini, R. Samperi, A. Lagan` , J. Chromatogr. a A 1101 (1/2) (2006) 6978. [22] J. Gilbert, E. Anklam, Trends Anal. Chem. 21 (2002) 468486. [23] P. Simon, P. Delsaut, M. Lafntaine, Y. Morele, T. Nicot, J. Chromatogr. B 712 (1998) 95104. [24] E. Chiavaro, C. DallAsta, G. Galaverna, A. Biancardi, E. Gambarelli, A. Dossena, R. Marchelli, J. Chromatogr. A 937 (2001) 3140. [25] J. Gilbert, E. Anklam, Trends Anal. Chem. 21 (6/7) (2002) 468486.