Proficiency Testing has improved the quality of data of total vitamin B2 analysis in liquid dietary supplement. By Fapas Proficiency Testing

Transcription

1 Proficiency Testing has improved the quality of data of total vitamin B2 analysis in liquid dietary supplement By Fapas Proficiency Testing INSPIRING BUSINESSES TO WORK SMARTER AND SUSTAINABILTY THAT DELIVERS REAL VALUE Date: April 2017, Version 1

2 Anal Bioanal Chem (2011) 400: DOI /s ORIGINAL PAPER Proficiency testing has improved the quality of data of total vitamin B 2 analysis in liquid dietary supplement Mark Sykes & Joanne Croucher & Rosemary Ann Smith Received: 4 November 2010 / Revised: 24 January 2011 / Accepted: 25 January 2011 / Published online: 12 February 2011 # British Crown Copyright 2011 Abstract A previously reported proficiency test for the analysis of vitamin B 2 in liquid dietary supplement demonstrated bimodality. The same trend has now been observed in four subsequent tests of this type. The trend would not so easily have been observed without applying a fit-for-purpose standard deviation that is more generous than that predicted by the Horwitz equation. Since originally reporting the bimodal problem and hypothesising its cause by incomplete enzymic digestion of riboflavin-5- phosphate, there has been a general improvement in the reporting of the higher mode. This is thought to correspond to free riboflavin following complete digestion of the sample. Several individual participants appear to have learned from the experience and have changed their reporting of the lower mode to the higher mode. Keywords Proficiency testing. Vitamin analysis. Enzymic digestion Introduction Vitamin analysis is widely acknowledged as being difficult, for a variety of reasons. One reason was highlighted by FAPAS [1] due to the likely difference in chemical form of vitamin B 2 in different matrices. Proficiency test (PT) results for liquid supplement matrix have a bimodal distribution, whereas for breakfast cereal matrix they are unimodal. The hypothesis for the difference is that the liquid supplement generally contains riboflavin-5-phosphate, which M. Sykes (*) : J. Croucher : R. A. Smith The Food and Environment Research Agency, Sand Hutton, York YO41 1LZ, UK mark.sykes@fera.gsi.gov.uk requires an enzymic dephosphorylation to the determinand, riboflavin. Breakfast cereal, by contrast, is fortified with riboflavin in which the extraction is not dependent on the enzymic step. Laboratories that neglect the enzymic step or fail to apply the necessary care with which enzymes must be treated may have incomplete digestion of the sample. The use of acid digestion is a necessary first step in digesting protein-bound riboflavin. However, the acid digestion will not hydrolyse riboflavin-5-phosphate to riboflavin. The activity of enzymes will depend on batch-to-batch variation, age, storage, and ph, temperature and duration of incubation with the sample. The mixture of enzymes may also have an effect on the digestion of some B group vitamins [2]. Vitamin B 2 may be present in different forms, depending on the matrix [3 6]. In foods, this may be as free riboflavin or in combination with its nucleotide forms, the co-enzymes flavin adenine dinucleotide and flavin mononucleotide. The latter is more commonly referred to as riboflavin-5- phosphate. EU legislation [7] allows for riboflavin and riboflavin-5-phosphate to be used in the manufacture of food supplements. Fortified foods or dietary supplements will tend to contain only one form of the vitamin. In solid matrices, such as breakfast cereals or vitamin tablets, this is likely to be as free riboflavin. However, since riboflavin has a limited solubility in water at fortified levels (about 10 mg/100 ml) [6], liquid matrices (liquid dietary supplements and energy drinks) will more likely contain the more soluble riboflavin- 5-phosphate (about 50 g/l). Since the data relating to reference [1] were collated, there have been a further four FAPAS PTs of vitamin analysis in liquid supplement matrix [8]. Although each dataset for vitamin B 2 shows the same bimodality, certain trends are beginning to become apparent. In this follow-up paper to the original report [1], we describe the PT data collated over a 7-year period and discuss the supporting

3 306 M. Sykes et al. evidence that long-term participation in these PTs has been beneficial to some laboratories. Literature Numerous papers have been published on vitamin analysis in a variety of matrices. Some of these are pertinent to the problem highlighted here, in terms of relevance to the observed PT results. The preferred method of analysis for vitamin B 2 is now extraction followed by high performance liquid chromatography (HPLC) determination [9, 10]. This has largely superseded the microbiological method, which is reflected in the method details returned by FAPAS PT participants. Several interlaboratory studies have been conducted to establish performance characteristics for different methods [2, 3, 5, 11, 12]. An early European interlaboratory comparison of methods for water-soluble vitamins asked participants to apply their routine method to certified reference materials [2]. The results highlighted the problem of vitamin B 2 analysis with observed relative standard deviation of reproducibility (RSD R )of28 74% and Horwitz ratio (HorRat) values of 2.6 and 5.9. HorRat values are frequently used in collaborative trials as a measure of the observed precision to the predicted precision using the Horwitz equation. HorRat values are generally considered to demonstrate acceptable precision if they are within the range Most laboratories used a similar extraction method of autoclaving or boiling with acid, followed by dephosphorylation with takadiastase (for laboratories using HPLC determination). Differences between laboratories largely amounted to the detail of time or temperature for extraction. Laboratories using microbiological determination did not use the dephosphorylation step, since the microorganism responds equally to free riboflavin and its phosphate forms. The acid hydrolysis is still required, however, for all determinations to free the vitamin from complexation with proteins. The hydrolysis protocol needs to be rigorous, whether acid or acid and enzymic, and tailored to the foodstuff being analysed. An HPLC-fluorescence method for thiamine and riboflavin [11] evolved the enzymic step by a limited interlaboratory comparison, prior to a full collaborative trial [3]. After some 8 years of routine use in French food control laboratories, the official 1987 method for the determination of vitamins B 1 and B 2 in foods was subjected to collaborative study [3]. This was justified by the observation that recovery rates were sometimes unsatisfactory in practice. The sample types used in the study included food complement (i.e. supplement, not described further). Results were recovery-corrected and RSD R ranged from 4% to 16% (7% for food complement). Chocolate powders, previously shown to be a difficult matrix for recovery, were treated in a separate study. Recovery rates for vitamin B 2 in chocolate powder were lower (75%) than for other foodstuffs (89 94%). It should be noted that the authors of [3] did not include HorRat values but they calculate to a range of , except meal with fruits (0.49) and tube-feeding solution (0.28). A number of studies commented on the efficiency of the enzyme used and variations between different suppliers of enzyme. One laboratory in [2] reported different activity of takadiastase from different suppliers. The authors of [11] indicated that takadiastase is insufficient by itself for complete dephosphorylation. During the development of the method, it had been found that the use of takadiastase on its own gave poor reproducibility, especially for foods containing high levels of naturally phosphorylated vitamins. This may be due to differences in enzymic activity, depending on the enzyme commercial product. The final method [3] used a mixture of takadiastase and β-amylase (10:1) for total dephosphorylation. Possibly, it is the impurities in the amylase that are additionally responsible for the efficiency of the process. A later EU-wide interlaboratory comparison [12] provided an aliquot of takadiastase enzyme to all laboratories from the same batch, which had been characterised for its efficiency. The prescribed optimised method described by van der Berg et al. [12] commenced with hydrochloric acid treatment, followed by incubation with takadiastase for 18 h. The in-house methods for all laboratories in the interlaboratory comparison were similar with principal differences being in the time for acid or enzymic treatment. No raw data were provided but CVs were quoted as being 12 40% for in-house methods and 12 34% for the optimised method. There was generally good agreement in results between using the in-house and the optimised method. The results of a collaborative trial [5] for vitamins B 1 and B 2 in animal feed had RSD R values ranging from 4.283% to 9.006% (HorRat values ) for mean values of μg/g for vitamin B 2, expressed as riboflavin. Variations in the methodology in relation to the need for the enzymic step have been tested [2 4, 10]. In particular, Viñas et al. [4] compared acid hydrolysis alone to acid plus enzymic hydrolysis. Of the eight B group vitamins under investigation, thiamine (B 1 ) and riboflavin (B 2 ) both required the enzymic step as well as the acid step. The recoveries from spiked samples for acid hydrolysis only were 66.5% and 70.3% for B 1 and B 2,respectively, which improved to 99.4% and 99.0% with the additional enzymic step. Blake [10] suggests that dephosphorylation is

4 Proficiency testing has improved the quality of data 307 preferable to attempting to determine riboflavin 5-phosphate separately and to analyse the total as free riboflavin. A variation for the type of enzyme was developed by interlaboratory trial [5]. Although the method variations were aimed mainly at improving the performance of the method for vitamin B 1, the procedure was adopted for B 2 as well. The initial method hydrolysed with hydrochloric acid in an autoclave (121 C for 30 min), followed by takadiastase enzyme incubation at 50 C for 2 h. The first variation in extraction protocol was to substitute takadiastase for clara-diastase enzyme. The second variation was hydrolysed with hydrochloric acid on a steam bath for 60 min and incubated with clara-diastase at 37 C for 16 h. The contrast of analysing dietary supplements was highlighted in the context of multi-vitamin analysis in a single method [6]. Since dietary supplements tend to have the vitamins present in a single form, and without the presence of complex matrix, this made a good test material for a multi-analyte approach. The standard reference material (NIST SRM 1849) used was a milk-based powder with vitamin B 2 present as free riboflavin. Hence, the sample preparation methodology was simplified to a low ph extraction only, with no dephosphorylation step. Free riboflavin, although a water-soluble vitamin, has limited solubility at dietary supplement concentrations. A solid matrix for the supplement is therefore suitable for riboflavin, which does not have to be in the form of (more expensive) riboflavin-5-phosphate. The vitamin standard solutions were tested for long-term stability, and riboflavin was stable for over a year at 4 C. However, the solution was prepared in ph 2.1 phosphate buffer plus 0.1% β- mercaptoethanol preservative. The purity of vitamin standards was thought to be an important factor in the measurement uncertainty (although probably at negligible levels compared to the 20% RSD R used for the FAPAS standard deviation for proficiency). FAPAS PT results and discussion Standard deviation for proficiency PTs are quantitatively assessed by the use of z scores. A result equal to the assigned value corresponds to a z score of zero. Results within two standard deviations of the assigned value correspond to z scores within ±2 or a 95% probability of the result falling within the normal distribution. The percentage of PT z scores within ±2 for breakfast cereal matrix is generally in excess of 80%. By contrast, the liquid supplement matrix PTs only achieve <60% z scores within ±2. Table 1 summarises the PT statistics for the liquid supplement tests being reported here. The previous report [1] was based on the discussion of PT 2139 (June 2006). This PT contained the first use of the mode to set the assigned value, and also the fit-for-purpose standard deviation for proficiency based on an RSD R of 20%, rather than Horwitz. Hence, z scores for PTs 2126 (July 2004) and 2133 (June 2005) were not issued because the statistics based on the Horwitz standard deviation were clearly not suitable. (Other B group vitamins in these tests were evaluated, however.) This is compounded by the u/σ p (observed uncertainty/standard deviation for proficiency) test, which provides a measure of the observed variability on the z scores. Ideally, this should be <0.4 for a negligible effect. PTs 2126 and 2133 evidently demonstrate that the Horwitz standard deviation is not fit-for-purpose. Breakfast cereal PTs, by comparison, continue to use a tighter Horwitz-derived standard deviation, at similar levels of vitamin B 2. A further consequence of the adoption of the fit-forpurpose RSD R is the comparison with collaborative trial data. The collaborative trials referenced above [3, 5] all demonstrate reproducibility precision in keeping with that Table 1 Summary of the results statistics for seven FAPAS vitamin B 2 PTs in liquid supplement matrix PT Date n AV Units RSD R % σ p u u/σ p test z July R μg/g 8.97 H June R μg/g 8.98 H June M μg/g 20 F % 2146 June M μg/g 20 F % 2152 June M μg/g 20 F % 2158 June M mg/100 ml 20 F % 2164 June M mg/100 ml 20 F % n number of valid results used in the calculation of the assigned value (AV), R robust mean, M mode, H Horwitz standard deviation as RSD R, F fit-for-purpose RSD R (expert advice [1]), σ p standard deviation for proficiency, u uncertainty of reported results, z z scores a 2126 and 2133, observation of results only, no z scores issued a a

5 308 M. Sykes et al. predicted by the Horwitz equation. In RSD R terms, this is in the range of about 5 14% (depending on concentration and matrix). Collaborative trials, however, concentrate the laboratories efforts on a single well-defined method. Interlaboratory comparisons [2] and PT, by contrast, may use a range of methodology as used routinely by each individual laboratory. In most cases, the Horwitz standard deviation is entirely fit for the purpose of prescribing the acceptable precision. The experiences of [2] and of the PTs described here demonstrate an interesting deviation from this rule of thumb. Bimodality of vitamin B 2 PT data PT 2139 set the new standard for this test. By adopting a fit-for-purpose standard deviation for proficiency based on 20% RSD R, the multi-modality of the results distribution could now be resolved into a distinct bimodal distribution. The two modes could be regarded as corresponding to free riboflavin and riboflavin-5-phosphate. The subsequent PTs have continued with this approach to issue z scores based on the higher mode of a bimodal distribution. The method of bump-hunting [13] is now well established to determine the modal distribution of PT results. The kernel density plot produced depends on the σ p value, the standard deviation for proficiency. Figure 1a shows the kernel density plot of the PT 2146 dataset using σ p derived using the Horwitz equation (σ p =5.118). Figure 1b, by contrast, shows the kernel density plot of the same dataset but using σ p basedonthefit-for-purposersd R of 20% (σ p =11.82). There is a clear visual improvement in simplifying the modal distribution. The interpretation of the dataset, based on the hypothesis of free riboflavin vs. riboflavin-5-phosphate, appears to be justified. Fig. 1 a PT 2146 kernel density plot using Horwitz standard deviation. Analytical result units are μg/g. b PT 2146 kernel density plot using standard deviation based on 20% RSD R. Analytical result units are μg/g a Vit B2 Horwitz Density Analytical result b Vit B2 FFP Density Analytical result

6 Proficiency testing has improved the quality of data 309 Table 2 Summary of the modes for seven FAPAS vitamin B 2 PTs in liquid supplement matrix PT Units μg/g μg/g μg/g μg/g μg/g mg/100 ml mg/100 ml Lower mode Lower mode density Higher mode Higher mode density Ratio of modes as % 8% 7% 12% 11% 16% 8% 8% Although the datasets from the earlier PTs 2126 and 2133, based on the Horwitz standard deviation statistics, could not issue evaluative z scores, retrospective bumphunting based on 20% RSD R shows the same bimodal trend of PT 2139 and subsequent PTs. Table 2 summarises the PTs and their two modes. It is observed that the lower mode is some 10% of the higher mode. We can surmise two possibilities from this. The first is that the riboflavin-5- phosphate in the test material contains 10% free riboflavin. The second is that the acid hydrolysis step by itself has about 10% dephosphorylation efficiency. In reality, there could well be a combination of these reasons. Evolution of methodology by participants Kernel density plots can be created for all the datasets corresponding to PTs 2126 to Figure 2 shows the ratio of lower mode density to higher mode density. There is a trend towards a decreasing ratio with later PTs, i.e., there is a general move towards reporting the higher mode in more recent PTs, from reporting the lower mode in earlier PTs. It seems that some participants are learning from the PT reports that the method needs to incorporate dephosphorylation in order to report total vitamin B 2 expressed as riboflavin. The identity of participants in FAPAS PTs remains confidential. However, by internally looking at the results of participants who have returned results for more than four of the seven PTs being examined, some observations can be made. Twenty-two laboratories submitted results in four or more PTs. Of these laboratories, ten fairly consistently reported a result corresponding to the higher mode. Four laboratories consistently reported a result corresponding to the lower mode. One laboratory was completely inconsistent in this respect. Seven laboratories, however, showed a general change from reporting the lower mode result in early PTs to the higher mode result in later PTs. The change in results reported by these laboratories coincides with the PT 2139 report, i.e. the higher mode is reported from PT 2146 onwards. This finding is in keeping with the trend in kernel density ratio observed above. It is evident that some laboratories are still not comfortable with the idea that the lower mode is an incomplete result for vitamin B 2 in liquid supplement matrix. Four laboratories have consistently reported the lower mode for at least four PTs, including the change of standard deviation reported in PT There may still be some confusion over the purpose of the two-step hydrolysis procedure. Additional comments submitted by two participants support this idea. Two participants in recent PTs both specified that their result was for riboflavin only, since riboflavin-5-phosphate was not analysed in their laboratory. The FAPAS PTs ask participants for method details, although the submission for this is optional. Unfortunately, the method details returned for these PTs are too insubstantial to compare methodological trends. It may be hypothesised that the same PT based on a solid dietary supplement test material (tablet or powder) would produce a unimodal dataset from the same participants. This is because the solid supplement need not contain the more expensive and more soluble riboflavin-5-phosphate but free riboflavin which would not benefit from the enzymic digestion step. Conclusions Standard deviation based on fit-for-purpose RSD R of 20% permits the modes of vitamin B 2 PT data to be calculated in liquid vitamin supplement. This is not so easily possible using the tighter Horwitz standard deviation. The HorRat Density ratio PT 2126 PT 2133 PT 2139 PT 2146 PT 2152 PT 2158 PT 2164 PT number Fig. 2 Ratio of lower mode density to higher mode density for seven total vitamin B 2 FAPAS PTs in liquid supplement

7 310 M. Sykes et al. values are still reliable indicators of precision in single method collaborative trials. The observation reported in PT 2139 and its hypothesis of incomplete digestion of riboflavin-5-phosphate producing the modal data continues to be supported by later PT data. Some laboratories have learned from the PT exercises and appear to have modified their methods to encompass the necessary enzymic dephosphorylation step. References 1. Earnshaw A, Smith RA, Owen L (2009) How proficiency testing can improve the quality of analytical data using vitamin analysis as an example. Food Chem 113: Hollman PCH, Slangen JH, Wagstaffe PJ, Faure U, Southgate DAT, Finglas PM (1993) Intercomparison of methods for the determination of vitamins in foods Part 2. Water-soluble vitamins. Analyst 118: Arella F, Lahély S, Bourguignon JB, Hasselmann C (1996) Liquid chromatographic determination of vitamins B 1 and B 2 in foods. A collaborative study. Food Chem 56: Viñas P, López-Erroz C, Balsalobre N, Hernández-Córdoba M (2003) Reversed-phase liquid chromatography on an amide stationary phase for the determination of the B group vitamins in baby foods. J Chromatogr A 1007: Analytical Methods Committee (2000) Determination of thiamine and riboflavin in pet foods and animal feedingstuffs. Analyst 125: Goldschmidt RJ, Wolf WR (2010) Simultaneous determination of water-soluble vitamins in SRM 1849 infant/adult nutritional formula powder by liquid chromatography-isotope dilution mass spectrometry. Anal Bioanal Chem 397: Directive 2002/46/EC of the European Parliament and of the Council on the approximation of the laws of the Members States relating to food supplements. 12/07/2002 Official Journal L183: FAPAS Reports 2126, 2133, 2139, 2146, 2152, 2158, FAPAS, Food and environment research agency, York, UK BS EN 14152:2003, Foodstuffs determination of vitamin B 2 by HPLC 10. Blake CJ (2007) Analytical procedures for water-soluble vitamins in foods and dietary supplements: a review. Anal Bioanal Chem 389: Hasselmann C, Franck D, Grimm P, Diop PA, Soules C (1989) High-performance liquid chromatographic analysis of thiamine and riboflavin in dietetic foods. J Micronutr Anal 5: van der Berg H, van Schaik F, Finglas PM, de Froidmont-Görtz I (1996) Third EU MAT intercomparison on methods for the determination of vitamins B-1, B-2 and B-6 in food. Food Chem 57: Lowthian PJ, Thompson M (2002) Bump-hunting for the proficiency tester searching for multimodality. Analyst 127:

8 Want to know more? To find out more about how Fera can be your partner with regards to improving quality compliance and performing proficiency testing in your business, ultimately ensuring accreditation, please contact; Fera Science Ltd (Fera), National Agri-Food Innovation Campus, Sand Hutton, York, YO41 1LZ, United Kingdom T: +44 (0) E: W: and Fera Science Ltd is a company registered in England and Wales with Company Number VAT Registered number GB Products and services availability may change at any time without prior notice given and all content are for illustration purposes only.