ANALYST, MAY 1993, VOL. 118 475 Intercomparison of Methods for the Determination of Vitamins in Foods Part 1 Fat-soluble Vitamins Peter C. H. Hollman and Jean H. Slangen DLO-State Institute for Quality Control of Agricultural Products (RIKIL T-DLO), Bornsesteeg 45, NL-6708 PD Wageningen, The Netherlands Peter J. Wagstaffe and Uta Faure Commission of the European Communities, Community Bureau of Reference (BCR), Rue de la Loi 200, B- 1049 Brussels, Belgium David A. T. Southgate and Paul M. Finglas AFRC Institute of Food Research, Norwich Laboratory, Norwich Research Park, Colne y Lane, Norwich, UK NR4 7UA An intercomparison of methods involving 18 European laboratories was organized to assess the state-of-the-art of vitamin determination in foods. Each laboratory received identical samples of dry food reference material (homogeneous powders, milk powder, pork muscle and haricot vert beans), which were recently certified for major dietary components and elements.Each laboratory was requested to perform the analyses by its own methods. Results for fat-soluble vitamins are reported. All participants isolated the fat-soluble vitamins by alkaline saponification. For retinol, only high-performance liquid chromatography (HPLC), reversed- or normal-phase, was applied, with both ultraviolet (UV) and fluorescence detection. Results in milk powder showed a relative standard deviation of reproducibility (RSDReprod) of only 10%. Carotene was determined by HPLC (reversed- and normal-phase) and with open-column chromatography at atmospheric pressure. For p-carotene results in milk powder agreed very well; the RSDReprod was 14%. The values reported for haricot vert beans showed poor agreement; the RSDReprod was 52%.A major part of this variability was due to differences in methodological principles. The results for a-tocopherol in milk powder and haricot vert beans agreed very well, with RSDSReprod of 16 and 15%, respectively. Only HPLC (reversed- and normal-phase) with UV and fluorescence detection was applied. Keywords: Intercomparison; food: retinol; /3-carotene; a-tocopherol Vitamins are a large group of compounds, which differ in their chemical composition, physiological action and nutritional importance. From a nutritional point of view a sufficient intake of the fat-soluble (pro)vitamins is of great importance. This is also reflected in the legislation and labelling guidelines for retinol and vitamin D3 (both dangerous in excess), and tocopherols.Therefore, accurate methods of analysis for these vitamins are needed. So far, no data are available in the open literature on the between-laboratory reproducibility for the determination of fat-soluble vitamins in foods using routine methods. Recently, the Community Bureau of Reference of the Commission of the European Communities undertook a programme to improve the quality of vitamin determination in food. The following aspects will be covered: improvements in method- ology, intercomparison of methods and preparation of refer- ence materials. As a first step in this programme, an intercomparison of methods for fat- and water-soluble vit- amins in foods was planned.The purpose of this intercompari- son was to assess the state-of-the-art of vitamin determination and to identify problem areas. The results of this intercomparison of methods for the determination of the fat-soluble (pro)vitamins retinol, car- otene and a-tocopherol in food are reported in this paper. Seventeen laboratories with experience of vitamin determina- tion in food participated in this study. Participants were encouraged t o apply the methods of analysis routinely used in their respective laboratories. A subsequent paper deals with water-soluble vitamins (see Part 2).13* * A report with detailed data can be obtained from Peter C. H. Hollman, DLO-State Institute for Quality Control of Agri- cultural Products (RIKILT-DLO), Bornsesteeg 45, NL-6708 PD, Wageningen, The Netherlands.Experimental Protocol The participating laboratories were invited to use their routine methods for the analysis of each food sample and vitamin. The laboratories had to carry out at least three separate determinations on three separately weighed sub-samples taken from at least two of the sachets provided. Results were expressed on a dry-matter basis as determined by drying under prescribed conditions. The vitamin standards used as cali- brants can be an important source of variation. To be able to assess these errors, a multivitamin reference mixture of known composition was sent with the samples. Determinations on separately weighed sub-samples taken from two units had to be made.Guidelines for the preparation of stock solutions of this multivitamin mixture were given. Materials Three dry food certified reference materials (CRMs) in the form of homogeneous powders [whole milk powder (not enriched, CRM 380), freeze-dried pork muscle (CRM 384) and dried haricot vert beans (CRM 383)] were selected for this intercomparison. Only pork muscle was used for the deter- mination of the water-soluble vitamins. The samples were packed into heat-sealed laminated-foil sachets, flushed with nitrogen. These foods have been developed as CRMs for major nutrients and elements, and proved to be homogeneous with respect to more than 10 elements and major nutrients.2 The homogeneity with respect to retinol, a-tocopherol, vitamin B1 and B2 was also studied in the milk powder and pork muscle (only vitamin B1 and B2) and was found to be adequate.Additional evaluation of the stability study3 showed476 ANALYST, MAY 1993, VOL. 118 that the between-sachet RSD of a-tocopherol and vitamin C content for haricot vert beans was smaller than 4%. Long-term stability at 4 "C was demonstrated for: retinol, a-tocopherol, vitamin B1 and B2 in milk powder; vitamin B1 in pork muscle; and a-tocopherol and vitamin C in haricot vert beans. In addition, short-term stability at 30°C of the most labile vitamins, retinol and a-tocopherol in milk powder and of vitamin C in haricot vert beans was known to be acccptable,3 hence the stability of the vitamins was sufficient to allow normal postal shipment. However, data on the multivitamin mixture and of other vitamins were lacking.Therefore, a linear regression analysis was carried out for all vitamins and samples to study a possible relationship between the concentration (logarithmic concentration) and the date of analysis in the laboratory. Correlation coefficients calculated did not indicate a significant negative correlation (t-distribu- tion according to Fisher, one sided, P >0.05). The multivitamin reference mixture was composed of lactose with vitamins A and E added in the form of beadlets (made of gelatine, starch and saccharose), the other vitamins were added as the pure substances. In order to assess between-sample variation, five bottles were randomly chosen and analysed for a number of fat-soluble vitamins in one laboratory. The variation between the samples for fat-soluble vitamins was rather small (RSD = 2.3-2.7%0) compared with the analytical variation (RSD,,,,,,) of the laboratories (Table 1) and thus the multivitamin mixture can be regarded as homogeneous.Technical and Statistical Evaluation Statistical evaluation followed the principles of the Interna- tional Organization for Standardization (ISO) norm, I S 0 5725,4 to calculate the RSDReprod and the relative standard deviation of repeatability (RSD,,,,,,). Strictly speaking, I S 0 5725 was designed to evaluate collaborative studies with one, well-defined method and, therefore, RSD,,,,,, represents the average repeatability of all methods as applied in this study. This statistical evaluation guided the technical evaluation of the study at a meeting with the participating laboratories.In addition, the Youden rank sum test was applied.5 For vitamin B1 this test identified a laboratory with possible systematic bias. Outlying laboratories identified by the statistical tests were only rejected from the calculations if supported by technical considerations, for instance, inadequate methods or poor laboratory performance. Suppressed results will be discussed subsequently. The aim of this intercomparison was to investigate the influence of different procedures routinely used by different laboratories. Results Laboratories carried out the analyses within a period of 3 months, with instruction to store the samples at 4 "C until use. Table 1 Summary of the variation in the results for fat-soluble vitamins in the multivitamin mixture, and the effect of the correction on the reproducibility Multivitamin mixture Milk powder Haricot vert beans RSD,* RSDR+ RSD,* RSDR RSDR(~~,.~)§ RSDR RSI?,R(~<,~~) Vitamin ( Y o ) (Yo) (Yo) (Yo) ( Y ) (Yo) (/A) - - Retinol 9.1 24 4.7 10 21 Carotene 7.3 35 5.1 14 16 52 46 a-Tocopherol 4.1 19 3.2 16 21 1s 26 * RSD, = RSD,,,,,,.* RSD, = predicted RSDRcprod according to the equation of Horwitz.6 + RSDR = RSDncprod. RSDR(,,,,, = RSDReprod of corrected results. Multivitamin Reference Mixture Table 1 summarizes the precision achieved by the different laboratories in analysing the multivitamin mixture. An indica- tion of the RSDKeprod expected for the determinations in the multivitamin mixture was obtained from the empirical equa- tion of Horwitz:6 RSDReprod = 2 exp(1 - 0.5 logc).This equation related the reproducibility observed in collaborative studies, when the same rigidly defined standardized methods were applied, to analyte concentration, c , expressed as a decimal fraction. When comparing the actual RSDReprod in this study with this estimated RSDReprotl, the results for fat-soluble vitamins are very poor. Inhomogeneity of the multivitamin mixture can be ruled out (see under Materials). Segregation of the beadlets during shipping could have caused inhomogeneity; however, this could not have resulted in the large variation observed, because the sample sizes prescribed (31 g) would have ensured a representative sample. This large RSDRcprod could possibly indicate differences in calibration procedures of the laboratories.To study this effect a correction factor for each vitamin was separately calculated for each laboratory. Division of the theoretical level of the specified vitamin in the multivitamin reference mixture by the mean value determined in the multivitamin mixture by the laboratory concerned, yielded the correction factor for this laboratory. Next, results of the laboratory for each sample and the specified vitamin were multiplied by the correction factor. The corrected values were calculated for all laboratories, food samples and vitamins. Finally, RSDReprod of the corrected results was calculated. By comparing RSDKeprod of the food samples with the RSDReprod of corrected results [RSDR(cor,) in Table 11, no decrease in the variation is noticeable. If differences in the calibration of vitamin standards are an important source of variation between laboratories, RSDR(corr) would be expected to be smaller than the uncorrec- ted RSDReprod; however, correction only increased the variability.Most likely, this was caused by inadequate execution of the extraction procedures in the standard mixture. These procedures were prescribed in the protocol, and participants were not familiar with them. Consequently, this intercomparison was not able to reveal the possible effects on the precision of differences in the calibration procedures of vitamin standards. In order to avoid the type of extraction problems with the powdered standards of the present intercomparison, liquid solutions of vitamin standards should be used, and the laboratories should try out the prescribed procedures before starting the trial.Retinol Results for milk powder show an RSDReprod of 10% (Table 2), and compare well with an interlaboratory study using enriched skimmed milk powder containing >0.3 mg of retinol per 100 8.7 In this International Dairy Federation study, with partici- pants applying a uniform high-performance liquid chromato- graphic method, the following precision data were found: All participants isolated vitamin A by alkaline saponifica- tion, followed by extraction of the retinol. Conditions for extraction varied (Table 3). Laboratories 4 and 17 used solid-phase extraction instead of liquid-liquid extractions used by the other participants. Only high-performance liquid chromatography (HPLC) with reversed- and normal-phase (laboratories 1, 6 , 7, 10 and 17) was used (Tables 4 and 5).Both ultraviolet (UV) and fluorescence detection (labora- tories 1, 4, 15 and 17) were used. Results of one laboratory (not shown) were judged to be unreliable because of insuffi- cient extraction of retinol after saponification and inadequate chromatographic resolution. The material used in this intercomparison was purchased from a commercial supplier, so the cis-isomers (13-cis, 9,13-di-cis and 9 4 s ) can be expected to be present in the RSDReprod = 15%, RSD,,,,,, = 5%.477 ANALYST, MAY 1993, VOL. 118 Table 2 Summary of the results of the intercomparison (expressed as mg per 100 g of dry mass) Number of Mean* RSDrepeat RSDReprod RSDpt Vitamin laboratories (range) (Yo) ("/I (Yo) Ratio$ Milk powder 12 0.267 6.7 10 14 0.7 Retinol- (0.2224293) Carotene- Milk powder 7 0.119 7.3 14 16 0.9 Haricot vcrt beans 9 0.222 7.6 52 14 3.7 (0.098-0.146) (0.063-0.398) a- Tocopherol- Milk powder 11 0.603 7.0 16 12 1.3 Haricot vert beans 10 0.335 8.1 15 12 1.2 (0.503-0.771) (0.25 14.406) * Mean = mean of means of the laboratories.t RSD, = predicted RSDReprod according to the equation of Honvitz.6 Ratio = RSDReProd/predicted RSDReprod. Table 3 Extraction methods used for the determination of fat-soluble (pro) vitamins. All laboratories applied alkaline saponification Extraction Laboratory Rctinol Carotene 1 Diisopropyl ether Diisopropyl ether 2 - Acetone-hexane (2 + 3), acetone, 3 Light petroleum (3x) Diethyl ether, (3x) 4 Extrelut, hexane - 6 Light petroleum-diethyl ether - hexane* (1 + I), (3x1 7 Diethyl ether, (3x) Diethyl ether 8 Diethyl ether Light petroleumt 9 - - 10 Diethyl ether - 11 Dichloroet hane Diisopropyl ethcr 13 - Diethyl ether+ 14 Hexane, (2x) Hexane, (2x) 15 Hexane, (5x) Hexane, ( 5 x ) 16 Light petroleum-diethyl ether Light petroleum-diethyl ether 17 Extrelut, hexane - (1 + I), (2x1 (1 + 11, (2x) * Only haricot beans were analysed, saponification was not applied.t Only haricot beans were analysed. a-Tocopherol Diisopropyl ether Light petroleum (3x) Extrelut, hexane Light petroleum-diethyl ether Diethyl ether Diethyl ether Hexane ( 2 x Dichloroethane Diethyl ether, (4x) Hexane, (2x) Hexane, (5x) Light petroleum-diethyl ether (1 + 1) (1 + I), (3x1 - Extrelut, hexane Table 4 Normal-phase HPLC conditions used for the determination of retinol and Laboratory HPLC 1 Column Eluent Detection Eluent Detection Eluent Detection Eluent Detection Eluent Detection Eluent Detection Eluent Detection Eluent Detection Eluent Detection 3 Column 6 Column 7 Column 10 Column 14 Column 15 Column 16 Column 17 Column Retinol Polygosil Si-60,5 pm, 250 x 4.6 mm Hexane-CH2CI2-propan-2-ol (90 + 9 + 1) Fluorescence, 333/470 nm - - - p-Porasil, 10 pm, 300 x 4.0 mm Isooctane-propan-2-01(985 + 15) UV, 340 nm Kieselgel, 5 pm, 250 x 4.6 mm Propan-2-ol-heptane (gradient) UV, 325 nm LiChrosorb Si-60,5 pm, 250 x 4.6 mm Isooctane-propan-2-01(98.5 + 15) UV, 312 nm - - - - - - - - - APS Hypersil, 3 pm, 100 x 4.6mm Isooctane-butan-2-ol(96 + 4) Fluorescence, 328/477 nm a-tocopherol a-Tocopherol Polygosil Si-60,5 pm, 250 X 4.6 mm Hexane-diisopropyl ether (90 + 1) Fluorescence, 296/320 nm LiChrosorb Si-60,5 pm, 250 x 4.6 mm Hexane-propan-2-01(99 + 1) Fluorescence, 290/325 nm p-Porasil, 10 pm, 300 X 4.0 mm Isooctane-propan-2-01(996 + 4) Fluorescence, 290/325 nm Kieselgel, 5 pm, 250 x 4.6 mm Propan-2-ol-heptane (gradient) Fluorescence, 290/327 nm - Spherisorb SSW, 5 pm, 250 X 4.6 mm Hexane-propan-2-01(985 + 15) UV, 295 nm LiChrospher Si-60,s pm, 125 x 4 mm Hexane-1.4-dioxane (97 + 3) Fluorescence, 293/326 nm LiChrosorb Si-60,5 pm, 250 x 4.6 mm Hexane-propan-2-ol(99.5 + 5 ) Fluorescence, 290/330 nm APS Hypersil, 3 pm, 100 X 4.6 mm Isooctane-butan-2-01(96 + 4) Fluorescence, 29Y327 nm478 ANALYST, MAY 1993, VOL.118 Table 5 Reversed-phase HPLC conditions used for the determination of retinol and a-tocopherol Laboratory HPLC 3 4 8 9 11 13 14 15 16 Column Eluent Detection Column Eluent Detection Column Eluent Detection Column Eluent Detection Column Eluent Detection Column Eluent Detection Column Eluent Detection Column Eluent Detection Column Eluent Detection Retinol p-Bondapak CI8, 10 pm, 250 x 4.6 mm Methanol-water (90 + 10) UV, 325 nm Hypersil ODs, 10 pm, 100 x 4.6 mm Methanol Fluorcscence, 32Y4.50 nm LiChrosorb RPls, 10 pm, 2.50 X 4.6 mm Methanol-water (93 + 7) UV 328 nm - - - Hypersil ODs, 5 pm, 250 X 4.6 mm Methanol-water (97 + 3) UV, 325 nm - - - Spherisorb ODs, 3 pm, 150 X 4.0 mm Methanol-water (95 + 5 ) UV, 325 nm LiChrospher 100 RP18, 5 pm Methanol-water (98 + 2) Fluorescence, 325/470 nm Partisil ODs-2,lO pm, 250 X 4.6 mm Methanol-water (90 + 10) UV, 325 nm a-Tocopherol - - - Hypersil ODs, 3 pm, 100 x 4.6 mm Methanol Fluorescence, 290/330 nm LiChrosorb C18, 10 pm, 250 X 4.6 mm Methanol-water (93 + 7) UV 292 nm LiChrosorb Cls, 5 pm, 250 x 4.6 mm Methanol-water (97 + 3) Fhorescence, 293/326 nm Hypersil ODs, 5 pm, 250 x 4.6 mm Methanol Fluorescence, 290/330 nm C18, 5 pm, 220 x 4.6 mm Acetonitrile-CH2CI2-MeOH (70 + 20 + 10) UV, 294 nm - - - - - - - - - 0.35 - A l l - t r a n s b A l l - t r a n s + 13-cis- T L U 0 o1 0.30 7 c Q 7 0.25 .I ................................................................ b 1 1 i 1 4 i ............. I I . - n I - .- ; I s T - tT I 0.20 1 1 1 1 1 1 1 1 1 1 1 1 6 16 10 7 1 15 17 3 8 14 4 11 Labcode Fig.1 Results of individual laboratories for retinol in milk powder (mg per 100 g dry mass). Data represent the mean k standard deviation of at least three separate determinations for each laboratory samples. Woollard and Indyk8 determined cis-isomers in different commercial samples of milk powder and found 13-cis to be the most predominant isomer in all samples. Levels ranged from 9 to 20% of the all-trans-retinol. Laboratories 1,6,7,10 and 16 reported all-trans values, which were used for the statistical calculations and are shown in Fig. 1. The majority of the laboratories did not separate 134s and all-trans-retinol, so 13-cis is included in their results. However, the contribution of 13-cis to the total value for retinol depends on the type of detection chosen. With UV detection at 325 nm the absorbance of 13-cis is 8% smaller than the absorbance of all-trans, whereas with fluorescence detection at 314/485 nm, the intensity of 13-cis is only one third of the intensity of all-trans.Laboratories 7 and 16 determined 13-cis, and found values of 10 and 13% of the all-trans-retinol content, respectively. This is in agreement with the data of Woollard and Indyk.8 However, it is documented that during the analytical procedure isomerization can occur depending on the type of sample and conditions of saponification and extraction.9~10 Laboratories 3,8,11 and 14 using UV detection reported an average retinol content of 0.28 mg per 100 g, about 10% higher than the average value of 0.26 reported by laboratories 1, 6, 7, 10 and 16 only reporting all-trans.This compares well with the estimated content of 13-cis. It is concluded that the results for retinol in milk powder agreed very well between laboratories. Precision can be improved by taking into account the two isomers of retinol, viz . , 134s and all-trans. p-Carotene Reproducibility of the determination of carotene in haricot vert beans is very poor; results range from 0.063 to 0.398 mg per 100 g of dry mass (Table 2). Collaborative studies, using uniform methods, show an analyte at this level would be expected to give an RSDReprod of about 15% according to the equation of Honvitz,e as opposed to the RSD of 52% obtained. On the other hand, results for milk powder agree very well; the RSDReprod of 14% calculated for milk powder is identical with the predicted value.All laboratories, except No. 2, extracted carotenes after alkaline saponification (Table 3). As is shown in Table 6, most of the participants subsequently used HPLC, both in normal- phase (laboratories 7 and 14), and in reversed-phase mode (laboratories 1, 11, 13, 15 and 16). Laboratories 2, 3 and 8 used methods based on open-column chromatography at atmospheric pressure, and consequently determined the total of all-trans carotenes (a, B, y and 6) and their stereoisomers, calculated as (3-carotene .I1 Of these laboratories, only laboratory 3 carried out analyses in the milk powder. Indyk12 determined carotenoids in milk powder using HPLC and found that B-carotene is the principal carotenoid; no a-carotene was present. Several participants using HPLC confirmed that no &-carotene or other carotenes were present.Consequently, the results of carotene in milk powder obtained with open-column chromatography and HPLC should theoretically agree. In this intercomparison, the results of laboratory 3 agreed with the HPLC results. Laboratory 16 determined the a-carotene content of haricot vert beans and found 0.083 mg per 100 g (21% of the (3-carotene content); probably no y- and &carotene were present. So the results of laboratories 2 , 3 and 8 are expectedANALYST, MAY 1993, VOL. 118 479 0.5 I 1 Table 6 Methods and conditions used for the determination of carotene (for extraction methods see Table 3) Laboratory HPLC Conditions Normal-phase HPLC- 7 Column Kieselgel, 5 pm, 250 X 4.6 mm Eluent Propan-2-01-heptane (gradient) Detection Visible, 4.50 nm 14 Column Partisil, 5 pm, 250 X 4.6 mm Eluent Hexane-ethanol(9999 + 1) Detection Visible, 450 nm Reversed-phase HPLC- 1 Column Hypersil ODs, 5 pm, 250 x 4.6 mm Eluent Acetonitrile-CHCI3-acetone-water Detection Visible, 445 nm Eluent Acetonitrile-CHCI3-methanol-water Detection Visible, 445 nm 13 Column CI8, 5 pm, 220 x 4.6 mm Eluent Acetonitrile-CH2CI2-methanol (70 + 20 + 10) Detection Visible, 450 nm Column Eluent Acetonitrile-methanol-CHZC12 (36 + 40 + 24) Detection Visible, 450 nm Eluent Acetonitrile-CH2C12-methanol (70 + 20 + 10) Detection Visible, 450 nm (750 + 150 + 100 + 20) 11 Column Hypersil ODs, 5 pm, 250 X 4.6 mm (85 + 8 + 5 + 2) 15 LiChrospher 100 RP18, 5 pm, 125 X 4.6 mm 16 Column Zorbax ODs, 5 pm, 250 X 4.6 mm Open-column chromatography- 2 Column Activated magnesia-diatomaceous earth (1 + 1) Eluent Hexane-acetone (9 + 1) Detection Visible, 436 nm Eluent Hexane Detection Eluent Detection Visible, 452 nm 3 Column De-activated aluminium oxide 90 Visible, 4.50 nm, E(1%, 1 cm) = 2590 Diethyl ether-light petroleum (1 + 3) 8 Column Aluminium oxide neutral activity to be higher than the results of HPLC methods (Fig.2). However, in this intercomparison these laboratories using open-column chromatography at atmospheric pressure, found values lower than the trial mean. Laboratory 14 did not succeed in separating a- and @-carotene and probably re- ported a- + p-carotene. Chromatograms for haricot vert beans showing adequate resolution of a- and @-carotene were given by laboratories 11 and 16, reporting the highest p-carotene values, but also by laboratory 13 giving one of the lowest values.So poor resolution of the carotene isomers seems not to be the only problem in the analysis of @-carotene. Summarizing, the results for @-carotene in milk powder agreed very well between laboratories. On the other hand, the reproducibility of the carotene determination in haricot vert beans was very poor. Poor resolution between a- and @-carotene in haricot vert beans as determined by HPLC does not explain the variability found. Methods based on open- column chromatography at atmospheric pressure tended to give the lowest results, even in the haricot vert beans where both a- and @-carotene were present. a-Tocopherol Results for milk powder, with an RSD,,,,,, of 7.0% and an RSDReprod of 16% (Table 2), compare favourably with the results of a collaborative study of tocopherols in vegetable oils and fats organized by the International Union of Pure and Applied Chemistry (IUPAC).13 In this IUPAC study an RSDrepeat of 5% and an RSDReprod of 31% were found for a sample with a tocopherol content of 1.7 mg per 100 g.Values +Open column- HPLC _____( wi 8 0.3 ....... \ S s 0.1 2 u € . . . . . . . . . . . . € m . . . . . . . . . . . . . . . . . . . . . . . . . . 01 I I I I 1 I I I I I 3 2 8 13 14 15 1 11 16 Labcode * Fig. 2 Results of individual laboratories for carotene in haricot vert beans (mg per 100 g dry mass). Data represent the mean k standard deviation of at least three separate determinations for each laboratory v) 1.0 v) T 17 3 7 6 1 5 9 11 8 1 6 1 4 1 4 2 0 .4 l ' I ' ' ' ' I ' ' ' Labcode Fig. 3 Results of individual laboratories for a-tocopherol in milk powder (mg per 100 g dry mass). Data represent the mean k standard deviation of at least three separate determinations for each laboratory reported for haricot vert beans agreed very well. For both products the RSDReprod is comparable to the predicted RSDReprod when uniform methods are used (Table 2).6 All participants used alkaline saponification, mostly iden- tical with the procedure used for retinol and carotene. The tocopherols were extracted with hexane or light petroleum, mixtures of light petroleum and diethyl ether, diisopropyl ether and dichloroethane (Table 3). Laboratories 4 and 17 used solid-phase extraction, and again laboratory 4 (see under Retinol) showed a high intralaboratory variation.All used reversed-phase HPLC (laboratories 4, 8, 9, 11 and 13) or normal-phase HPLC (Tables 4 and 5). Most of the labora- tories used fluorescence detection; only laboratories 8,13 and 14 used UV detection. No effect of this choice of chromato- graphy and detection on the results was apparent (Fig. 3). Results of one laboratory (not shown) applying a continuous- flow method with fluorescence detection were rejected, because for the determination of a-tocopherol in milk powder and haricot vert beans, its method reached the detection limit. High results from laboratory 4 were not included in the statistical evaluation, because of the calibration procedure.This laboratory used a-tocopherol as a standard, without checking the content. Because the tocopherol supplied never has 100% purity, this leads to erroneous results. Laboratory 14 reported difficulties during the extraction of the haricot vert beans caused by formation of emulsions. The results given by this laboratory are very high (2.48 mg per 100 g) and show a very large variation (RSD = 50%) and have not been included. In conclusion, the results for a-tocopherol in milk powder and haricot vert beans obtained with different types of HPLC, agreed well between laboratories.480 ANALYST, MAY 1993, VOL. 118 Discussion A summary of the results of this intercomparison on methods for the determination of fat-soluble vitamins in foods is given in Table 2.Although homogeneous materials were involved, it can be argued that the homogeneity of carotene in these materials was not demonstrated beforehand. As participants carried out separate determinations in at least two different sachets, non-homogeneity will be reflected in the analytical variation (RSD,,,,,,) of the laboratories. Comparing RSD,,,,,, and RSDReprod for carotene (Table 2), it can be concluded that possible non-homogeneity was not an impor- tant factor. Experienced food laboratories participated; in this inter- comparison the choice of method was left to the participants and was only subject to the requirements of achieving the best level of accuracy. The reproducibility ( RSDReprod) shown gives an impression of the state-of-the-art of fat-soluble vitamin determination.In order to evaluate the results the approach of Horwitz et aZ.14 was followed. In this, the ratio (Table 2) was calculated, with as the numerator RSDReprod found in this intercomparison, and as denominator the RSDReprod achievable. From collaborative studies with uni- form methods Honvitz6 had concluded that the KSDReprod achievable is mainly a function of concentration, largely independent of analyte, matrix and method. It represents the analytical variation caused by different laboratories with different operators using different equipment, but using the same well-defined method. In the present intercomparison laboratories used different methods. Therefore, if different procedures used by different laboratories do not have a strong influence on the results, the ratio will be close to 1.This proved to be true for the analysis of retinol and 6-carotene in milk powder and a-tocopherol in milk powder and haricot vert beans. The results agreed well; sometimes it was necessary to exclude one or two laboratories using inadequate methods. This good agreement allowed indicative values for retinol and a-tocopherol to be issued in these food RMs, which have already been certified for major dietary components and elements.2 High-performance liquid chromatography was the method of choice of most of the participants for the determination of fat-soluble vitamins. Future intercomparisons for retinol need to take into account all-trans and 13-czs-retinol, because of the different response depending on the type of detection.As was evident from the results of p-carotene in haricot vert beans, the traditional methods based on open-column chro- matography proved to be biased. In the evaluation of results of a-tocopherol, calibration procedures as a source of variation was emphasized. The present intercomparison failed to identify the role of different calibration procedures, because of the inadequacy of the extraction procedures prescribed for the multivitamin mixture. The skilful participation of the following laboratories is gratefully acknowledged: Tnstituto del Frio, Instituto del Fermentaciones Industriales CSIC, Madrid, Spain; Labora- tory of the Government Chemist, Teddington, UK; The National Food Agency of Denmark, SQborg, Denmark; Schweizerisches Vitaminin-Institut , Basel, Switzerland; State Institute for Quality Control of Agricultural Products (RIKILT) , Wageningen, The Netherlands; Unilever Research, Bedford, UK; VTT Food Research Laboratory, Espoo, Finland; TNO-CIVO Institutes, Zeist, The Nether- lands; Bundesforschungsanstalt fur Ernahrung, Stuttgart, Germany; Servicio de Nutricion, Madrid, Spain; Leatherhead Food R.A., Leatherhead, UK; University College Cork, Cork, Ireland; Federal Dairy Research Institute, Liebefeld- Bern, Switzerland; Swedish National Food Administration, Uppsala, Sweden; Produits Roche, Fontenay sous Bois, France; Food Inspection Service, Maastricht, The Nether- lands; and AFRC Institute of Food Research, Norwich, UK.1 2 3 4 5 6 7 8 9 10 11 12 13 14 References Hollman, P. C. H., Slangen, J. H., Wagstaffe, P. J . , Faure, U . , Southgate, D. A. T., and Finglas, P. M., Analyst, 1993, 118, 469. Hollman, P. C. H., Boenke, A., and Wagstaffe, P. J., Fresenius’ J. Anal. Chem., 1993,345, 174. Hollman, P. C. H., Slangen, J. H., Finglas, P. M., Wagstaffe, P. J., and Faure, U., Fresenius’ J. Anal. Chem., 1993,345,236. International Standard I S 0 5725-1986, International Organiza- tion for Standardization, Geneva, 1986. Youden, W. J., and Steiner, E. H., Statistical Manual of the Association of Official Analytical Chemists, The Association of Official Analytical Chemists, Washington, DC, 1975. Horwitz, W., Anal. Chem., 1982, 54, 67A. IDF Questionnaire 1188/E, International Dairy Federation, Brussels, 1988. Woollard, D. C., and Indyk, H . . J. Micronutr. Anal., 1986, 2, 125. Landers, G. M., and Olson, J. A., J. Assoc. Of5 Anal. Chem., 1986, 69, 50. Steuerle, H., 2. Lebensm. Unters. Forsch., 1985, 181, 400. Brubacher, G. Miiller-Mulot, W., and Southgate, D. A. T., Methods for the Determination of Vitamins in Food, Elsevier, LondonRVew York, 1985. Indyk, H., J . Micronutr. Anal., 1987, 3, 169. Pocklington, W. D., and Dieffenbacher, A., Pure Appl. Chem., 1988, 60, 877. Horwitz, W., Albert, R., and Deutsch, M., J. Assoc. Off. Anal. Chem., 1990, 73, 661. Paper 21061 67H Received November 19, 1992 Accepted January 18, 1993