NEAR INFRARED SPECTROSCOPIC ANALYSIS OF VITAMINS The present invention relates to a novel process for the qualitative and quantitative analytical determination of a vitamin or a mixture of vitamins and in particular to a process for the qualitative and quantitative determination of vitamins using near infrared spectroscopy.

It has been sought for a long time to optimize the analysis of vitamins in a mixture without the need to extract each of them individually from the constituents of the mixture, in order to be able to determine the presence of the main vitamins both in a vitamin mixture and in a premix primarily intended for human food or animal feed.

Specifically, premixes intended for animal feed can be in the form of liquid mixtures or in a solid form. When the vitamin premix is in the liquid form, it often contains diluents, surfactant and stabilisers. When the premix is in solid form, the composition often contains a support for example of mineral or plant origin, as well as food additives such as amino acids and/or vitamins. After preparation of the premix, it is preferable to check the composition and/or homogeneity of these mixtures by means of a rapid and non-destructive analytical technique. It is also often advantageous to determine, after a certain period of storage under occasionally non-ideal

conditions, for vitamins which are temperature-unstable and moisture-unstable compounds, the compliance of the activity of these vitamins without undertaking sophisticated analytical methods such as the extraction techniques described below, which are always difficult to carry out at the storage premises.

Among the techniques currently used to analyse the various vitamins present either in vitamin complexes or in premixes or in whole foods the following is a representative sample: for vitamins A and D3, a saponification with potassium hydroxide is carried out, followed by a dilution in 2-propanol and ethanol, respectively, followed respectively by UV spectroscopic analysis and liquid chromatography with a UV detector; for vitamins Bl, B6, B9 and B12, extraction is carried out with a methanol, acetonitrile, acetic acid, water and surfactant mixture or a methanol, acetic acid, water and surfactant mixture, respectively, or an extraction with a dilute aqueous solution of ammonia and sodium perchlorate or an extraction with acetonitrile followed by a liquid- phase chromatography with detection in the UV range; for vitamins H and K3, the analysis is carried out by colorimetry with dimethylaminocinnamaldehyde and ethyl cyanoacetate, respectively; for vitamin E, the extraction is carried out with ether followed by a dilution with hexane and a gas-chromatographic analysis; for vitamin B2, the dissolution is carried

out in sodium hydroxide, followed by a neutralization with acetic acid and spectroscopy in the visible range; for vitamins B5 and B3, the analysis consists in carrying out an acid-base assay.

It is almost impossible to implement all of these assays systematically, batch by batch, during the preparation of a vitamin premix, or alternatively, this is achievable at such a cost that it is only implemented at random at a low sampling rate.

Spectroscopy can be used as an analytical tool and there exists publications describing"in vitro"analyses combined with near-infrared spectroscopy. One such publication is the one by Dardenne, Andrieu, Barrière, Biston, Demarquilly, Femenias, Lila, Maupetit, Rivière and Ronsin, Ann.

Zootech (1993) 42,251-270 who established calibration curves for the following analyses: solids content, ash content, total proteins, water-soluble carbohydrates, celluloses, cell wall constituents, total fibres, acidic and neutral detergent fibres, lignin, enzymatic solubilities determined by the Aufrère, Limagrain and Lila methods. This combination of"in vitro"analyses has made it possible to carry out near-infrared analyses and, with the aid of a statistical analysis, to establish correlation curves which henceforth allow a direct near-infrared spectroscopy determination of the digestibility value"in vivo".

None of these methods has ever made it possible to date to determine the presence of vitamins in a mixture with each other or in a foodstuff by means of near-infrared spectroscopic analysis. This determination was not obvious on reading any of this essentially metabolic and mathematical prior art, since vitamins are hydrocarbon-based chemical species with a complex structure which is entirely different from the structure of materials such as fibres, cell walls, ash, etc.

We have now developed a method of determining the qualitative and quantitative presence of a vitamin in a vitamin-containing composition through the use of infra red spectroscopy.

Accordingly, the present application provides a method for the determination of the presence of one or more vitamins in a composition containing the vitamin characterised in that the presence is determined by near infrared spectroscopy.

In particular, there is provided a method for the determination of the presence of one or more vitamins in a composition containing the vitamin wherein the method comprises (a) obtaining the chemical analytical data and the near infrared spectroscopy data from known vitamins and vitamin mixtures,

(b) establishing a data base comprising the chemical analytical data and the near infrared spectra, (c) establishing a collaboration plot from the data of the data base, (d) recording the near infrared spectrum of the composition, (e) comparing the spectrum obtained in step (d) with the spectra of the data base to identify the vitamin and determining the concentration of vitamin from the collaboration plot.

By the presence is meant the identity of the vitamin and the amount of the vitamin.

By near infrared is meant the region of the spectrum ranging from 700 to 2500 nm.

The present invention relates to a novel method for the direct determination of the presence of one or more vitamins in a composition containing at least one vitamin. The method relates in general to the direct determination of the emission or transmission spectrum of the composition containing the vitamin in the near-infrared range and then in applying chemometric algorithms to the spectrum in order to determine the vitamin concentration. The present method is generally not affected by the presence of the matrix in the composition even if the composition is a complex mixture, such as the various liquid supports containing

the mixture of vitamins or the various solid supports constituting the basic vitamin premix to which the vitamins are added. The present invention is a non- invasive technique which avoids the need to extract the vitamins, which always carries uncertainty regarding the quality of the extraction, and avoids the need to undertake subsequent reactions which consume time and thus money.

The subject of the present invention is the qualitative and/or quantitative determination of the presence of vitamins which may be in a liquid or solid vitamin complex; a mixture of vitamins and/or in a vitamin premix which, in this case also, may be liquid or solid, this complex or mixture being intended to be incorporated into whole foods for man or feeds for rearing animals, by near-infrared spectroscopy.

The method of the present invention is used to determine the presence of at least one vitamin in a vitamin-containing composition. The method is particularly suitable for the determination of vitamin E or tocopherol in esterified form (acetate), vitamin B5 or calcium D-pantothenate, vitamin D3 or cholecalciferol, vitamin PP or niacin in the acid or amide form, vitamin B1 or thiamine mononitrate, vitamin B2 or riboflavin, vitamin B6 or pyridoxine hydrochloride, vitamin H or biotin, vitamin B12 or cyanocobalamin, vitamin C or ascorbic acid, vitamin A or retinol in esterified form (acetate, palmitate or

propionate), vitamin K3 or menadione and vitamin B9 or folic acid. Among these vitamins, the most suitable for characterisation using the present method, are vitamins A, E, B2, B5 and PP.

The method of determination may be applied to a composition containing an individual vitamin or to a composition comprising a mixture of vitamins. The composition may comprise a liquid and/or solid support which may be liquid or solid. The support used in the solid premixes can be chosen from supports of mineral or plant origin. Among the mineral supports which may be mentioned are calcium carbonate, dicalcium phosphate and sodium chloride. Among the plant supports which may be mentioned, in a non-limiting manner, are wheat or rice bran, middlings, soya cake and corn husks. These premixes can also contain microelements such as iron, cobalt, copper, manganese, molybdenum, selenium and zinc salts with various hydration contents.

The method of the present invention comprises the creation of a data base. This procedure involves the accumulation of chemical analytical data for a number of known proteins and/or protein mixtures. The method for chemical determination of the vitamin consists in carrying out a conventional chemical analysis of a population of representative samples of the variation range in terms of composition which may be expected in practice. For reasons of statistical reliability, a minimum of 80, preferably at least 120

samples may be used to start the construction of such a type of mathematical model.

After obtaining the chemical results, the corresponding near infrared spectrum for each sample is recorded by means of a spectrophotometer working in the near-infrared range (i. e. between 700 and 2500 nm). The measurement taken is in the form of a spectrum which in fact bears information regarding the physicochemical composition of the sample.

The two sets of results are recorded and stored, thus establishing a database of the results obtained by chemical analysis and the absorbance values measured by near infrared spectroscopy.

A calibration plot is then prepared, by mathematical processing of the infra red spectroscopy data, using appropriate computer software. The mathematical processing provides a linear regression which makes it possible to establish a prediction curve. The management of these operations can be carried out by means of any suitable commercially available software system.

Once the calibration plot has been prepared, the accuracy of the relationship between the near infrared analysis and the chemical analysis may be verified by recording the spectrum of samples of vitamin of known concentration using samples which have not been used to establish the calibration curve. The

results obtained after the mathematical treatment may be compared with the calibration curve.

The next step of the method of the present invention comprises recording the infra red spectrum of the vitamin-containing composition. For the qualitative determination, the spectrum, thus obtained, is compared with the spectra stored in the date base.

For the quantitative determination, the spectrum, thus obtained, is subjected to the same mathematical treatment as the calibration samples to provide a number. As proven from the calibration plot, this number corresponds to the concentration of vitamin.

These operations allow, by means of a simple and inexpensive method which does not destroy the starting material, a qualitative and quantitative determination of the presence of one or more vitamins in a vitamin mixture or in a solid and/or liquid premix in a very short space of time. This method can be used with ease by vitamin manufacturers, food manufacturers and breeders wishing to optimize the feed ration which they will give to their animals. This method can allow greater quality control at the manufacturing level and thus optimize the traceability of the various vitamins.

This method can also be used to monitor the quality of the vitamin-containing composition by regularly measuring the vitamin concentration. Thus according to another aspect of the present invention there is provided a method for implementing a quality

control check of a vitamin-containing composition which comprises: (a) determining a specification for the average vitamin concentration and maximum acceptable vitamin concentration in a sample, (b) establishing a theoretical coefficient of variation and standard deviation for the composition, (c) recording the near infrared spectrum of a plurality of samples of the composition, (d) determining the concentration of vitamin by the procedure as herein before described, (e) determining the coefficient of variation and standard deviation for the samples, (f) repeating steps (c) to (e) at regular intervals and comparing the results, The quality control method may be carried out by the vitamin manufacturer or the breeder.

It is also possible to establish a spectral fingerprint of a series of standard vitamin mixtures which will serve as a reference. This may be done without proceeding via the calibration and optional validation steps. For each new batch of a mixture prepared, it will be possible to determine whether it complies with one of the standard mixtures or whether it does not correspond to any of these and thus does not comply. In the case of non-compliance, the mixture is then analysed quantitatively as regards each of the

vitamins which is to be included in the said mixture.

The principle of this approach is the discriminating analysis, which can also be referred to as a "fingerprint".

The present invention will now be illustrated with reference to the following example: Step 1: Construction of a calibration population A calibration population of 120 samples of vitamin-containing compositions was prepared which involved the preparation of data base of chemical analysis results and near infrared spectra.

The chemical analysis of the vitamin- containing composition was carried out for each sample as per the standard routine analysis for the particular vitamin to qualitatively and quantitatively identify either the vitamin and/or the vitamins in the composition The samples were also analysed using near infra red spectroscopy wherein the spectrum of each sample was obtained in the region of 700 to 2500nm.

The spectrum of each sample was recorded using a grating spectrophotometer of NIRS 6500 type (sold by FOSS Int.) operating in"reflectance"mode. A particular specificity of the instrument was that it worked over the entire near-infrared range rather than

over a number of specific wavelengths, as is the case for certain more conventional machines.

The sample was packaged in a rectangular transportation cell composed of three opaque faces and a face fitted with a quartz crystal. After positioning the cell in the spectrophotometer, the light reflected by the sample, which bears information regarding the chemical composition of this sample, was recorded in the form of a spectrum. Each type of product, and consequently each sample, has a different spectrum.

The results from the chemical and spectroscopic analyses were stored in a data base and identified as the calibration file. The calibration file thus contained a combination of numerical and spectral data.

The entire calibration file constructed was subjected to a statistical calculation using suitable software (NIRS 2, version 3.00 distributed by Infrasolt International I. S. I.) which made it possible to detect the presence of non-standard spectra which have excessively large Mahalanobis distances (statistical distance H) compared with the average spectrum of the population (centre of gravity). Once detected, these samples were eliminated in order to obtain homogeneous population.

The calibration file containing all the samples was then subjected to a statistical analysis, of which three different types exist:"Modified Partial

Least Square" (MPLS), Partial Least Square (PLS), which is a general form of regression by main components, or "step up", which corresponds to a simpler linear regression. After choosing the appropriate regression method, it was preferable to carry out mathematical processing by differentiation for all the analyses.

This mathematical processing was determined by an evaluation sequence a, b, c, d: a = order of the differential, b = interval in which the differential calculation covers, c = smoothing constant 1, d = smoothing constant 2. After initialization by means of the preceding choices, a calibration phase was commenced.

Six calibration curves are provided. Figures 1/6 to 4/6 show, respectively, the calibration curves for vitamins A, E, B5 and PP in vitamin mixtures, and Figures 5/6 to 6/6 represent calibration curves for vitamins E and A established on food premixes on an organic support containing various mineral salts.

Step 2 Determination of Unknown Vitamin Sample The near infrared spectrum of the unknown sample was recorded and entered into the data base to find a corresponding spectrum. The vitamin was therefore identified qualitatively.

To determine the quantity of vitamin, the spectrum was subjected to the same statistical calculation as the database samples to obtain a figure

which corresponds to the concentration of the vitamin as verified by the calibration.

In particular, the determination by infrared reading on an unknown sample was carried out using the prediction equations which give the highest quality of performance. This selection was carried out as a function of various statistical parameters such as the correlation coefficient (R2) or the standard error of prediction (SEP).

The following calibration equations were determined for the following vitamins: A, E, B5, PP.

'Vit A SEP = 26,809 R2= 0.81 STDVref = 62,109 (x IU/kg) 'Vit E SEP = 15,132 R= = 0.95 STDVref = 71,150 (x IU/kg) 'Vit B5 SEP = 19,503 R2 = 0.86 STDVref = 51,817 (x IU/kg) * Vit PP SEP = 12,731 R2 = 0.98 STDVref = 89,474 (x IU/kg) SEP = standard error of prediction STDref = standard deviation of the reference population R2 = correlation coefficient for the validation.

The following equations were determined for premixes containing various vitamins: 'Vit E SEP = 3.82 R2 = 0.84 STDVref = 5.70 (x104IU) * Vit A SEP = 3.01 R = 0.30 STDVref = 3.30 (x106IU) in which SEP, STDref and R2 have the same meaning as above.

Quality Control Method The following procedure was used to monitor the vitamin level in samples prepared: (1) The procedure as detailed above was carried out to determine the identity of the sample; (2) A specification for the average vitamin level and maximum acceptable vitamin level in the composition were established; (3) Using the appropriate computer software, the coefficient of variation and standard deviation for the vitamin composition are determined. A maximum coefficient of deviation was also identified; (3) Seven random composition samples were taken, numbered and the near infrared spectrum recorded; (4) Using the appropriate computer software, the coefficient of variation and standard deviation for the seven samples were determined. The results were plotted.

(5) The above procedure was repeated 30 days later and 60 days later to provide three sets of results and the average of the three results was determined.

(6) The procedure was repeated again 60 days later and a t-test was conducted to determine if the new average result differed significantly from the initial result determined in step (5).

(7) A chi-square statistic was generated and an F- test is conducted to determine if significant

differences exist between the the initial and current sample standard deviations.

(8) The above procedure is repeated every 90 days.

(9) A line is fitted through the last three measurements of coefficient of deviation by the method of least squares. (Figure 7) If the slope of the line is positive, the time when the line would pass through the maximum acceptable coefficient of variation is estimated. If that time (t*) is less than 90 days, the procedure is repeated in half of t*.

(10) If the procedure is repeated in t* and the slope remains positive, vitamin formulation maintenance is recommended.