A preliminary study in determining the geographical origin of wheat using isotope ratio inductively coupled plasma mass spectrometry with ¹³C, ¹⁵N mass spectrometry

Abstract

Commercially, wheat functionality and usage are dependent on the species and cultivars with each country tending to favour particular ones. Certain geographical origins attract a premium price so there may be a financial incentive to misdescribe grain or extend it via adulteration with cheaper varieties. A preliminary investigation was carried out to examine the utility of ICP-MS elemental isotope analysis of Cd, Pb, Se and Sr plus stable isotope gas analysis of ¹³C and ¹⁵N with multivariate statistics in identifying the country of origin of Triticum aestivum. The emphasis of this work was on differentiating USA, Canadian and European samples on the basis of their fingerprint due to underlying geogenic and anthropogenic differences rather than legislative borders. It was found that the samples in this pilot study could be geographically classified using a single analyte, δ¹³C and that the methodology shows some potential in identifying different cultivars and blends within a region using discriminant analysis. Further work, however, is required to confirm this and to determine if the methodology can be extended to identify the degree of sample adulteration.

---

Introduction

Man has cultivated wheat for more than 10,000 years, originally in the fertile crescent of the Middle East, but in modern times in most of the countries of the world. World production of wheat is in excess of 520 million tonnes.¹

Commercially, wheat functionality and usage is dependant on the species, with, for instance, Triticum aestivum being used in breadmaking and Triticum durum being used in pasta manufacture. Within each species there are numerous cultivars, with each country favouring particular ones. The variation in country of origin and cultivars gives a range of performance for the flour milled from the wheat, with certain origins attracting premium prices. As such, there is a financial incentive to misdescribe grain or extend it via adulteration with cheaper varieties.²

In this paper we report the results of a pilot study that examined the use of combining elemental isotope analysis with multivariate statistics to identify the country of origin of T. aestivum, with emphasis on differentiating North American (USA and Canada) from European samples. The Canadian crop, in contrast to what happens in the US, is pooled and blended. As a result Canadian Western Red Spring (CWRS) and Teal Columbus Pasqua (TCP) samples can come from a wide geographical area. Differentiating USA and Canadian wheat samples is thought to be the most challenging task since there is the potential for the wheats to be grown in close proximity to each other along the border that divides the two countries politically not geologically.

The application of trace metal analysis to food authenticity is not new. Winefordner and co-workers³ used trace element data to compare Florida and Brazilian juices as early as the 1970’s. Subsequently ICP-MS data has been used to differentiate orange juices from six or more different countries.^4,5 This application has been extended to studies of wine vintage, when trace metal data has been combined with ¹⁸O, ²H and ¹³C data to discriminate between wines grown in the same year but different regions in the same country.^6-8 Multivariate analysis of metal data has also been used to classify honey type,⁹ geographical origin of potatoes¹⁰ and origin of macadamia and pistachio nuts as part of a programme to prevent imports from countries on which sanctions have been imposed.¹¹ Isotope ratios have been used to determine the geographical origin of olive oil¹² and Pecorino cheese.¹³

Trace element rationale

The trace element components of wheat are primarily governed by the water and soil layer in which they grow and can to some extent be influenced by anthropogenic practices such as the addition of fertilizers. The elemental/isotopic composition of the wheat grain will therefore be a reflection of the chemical composition of this uppermost soil layer which in turn is influenced by the underlying rock types from which the soil is derived and through which the water passes. Differences in the rock types across the earth have lead to a wide variety in the trace element signatures of soil and water. Processes of erosion mean that whilst the composition of the soil may not be directly analogous to the underlying lithology it will be a reflection of the regional chemical signature. This variation will continue into plant tissues grown in different geographical locations and can be sufficiently large to distinguish geographical areas.

Following a prescreen to identify elements with suitable concentrations in flour, four elemental and isotopic systems were identified as potential means by which geographical variations could be observed. These were the trace elements cadmium and selenium and the natural isotopic systems of strontium and lead.

Cadmium, Cd, is a toxic trace element with no known metabolic function. It naturally occurs at low levels in rock, soil and water.¹⁴ Cd is also a major economic element used in a wide variety of industrial processes so that the total budget of Cd in the surface environment is also a reflection of pollution. It is estimated that up to 99% of Cd released into the atmosphere each year is now of anthropogenic origin.¹⁵ The major sources of this anthropogenic Cd are from mining, fossil fuels, waste incineration and cement manufacture. It was proposed that the combination of natural and anthropogenic Cd levels can be used to distinguish different geographic regions.

Selenium is an essential trace element in human health and its level is known to vary significantly between Europe and North America.^14,16,17 It could therefore be a useful marker but is unlikely to be sufficient on its own to separately classify the American and Canadian samples.

Radiogenic isotopic variations of Sr and Pb were also used in this study. Pb and Sr both comprise a series of isotopes the relative abundances of which vary significantly in nature.

Lead comprises four stable naturally occurring isotopes. With the exception of ²⁰⁴Pb, part of the total abundance of each isotope results from the decay of radioactive isotopes of uranium (for ²⁰⁷Pb and ²⁰⁸Pb) and thorium (²⁰⁶Pb) and the intermediate radionuclides. Differences in the decay rate of ²³⁵U, ²³⁸U and ²³²Th and differences in the chemical reactivity and mobility of U and Th during rock forming processes mean that rocks of different ages and origins have distinct lead isotopic ratios. As is the case for Cd, Pb is also an important industrial metal and it was proposed that regional variations might also reflect differences in industrialisation between growing regions.

Strontium comprises four naturally occurring isotopes ⁸⁴Sr, ⁸⁶Sr, ⁸⁷Sr and ⁸⁸Sr. Strontium is a ubiquitous trace metal in soils, water and rock. ⁸⁷Sr is, in part, the daughter decay product of ⁸⁷Rb. As for lead, variations in the ratio ⁸⁷Sr/⁸⁶Sr reflect different time-periods since rock formation and differing mobility of Rb and Sr. Unlike lead, strontium is not significantly affected by anthropogenic pollution but by geogenic pollution (erosion).

The variation in Pb and Sr isotopic abundances are the foundations of a wide range of earth processing modelling and geochronology.¹⁸ Here, the aim was not to uniquely determine these processes but rather to take a broader regional scale view looking for gross scale differences that could offer a robust test of geographic origin.

Previously, strontium isotopes have been successfully applied in determining the origin and migratory patterns of animals such as salmon¹⁹ and for archaeological samples such as prehistoric human bone.²⁰⁸⁷Sr/⁸⁶Sr ratios have also been used to distinguish wines of differing geographic regions.²¹

Lead isotopic variations have also been successfully applied in determining geographic origins of archaeological materials.²² Lead isotopes have been successfully used to identify anthropogenic inputs from industrial processes.^23,24

In addition the δ¹³C and δ¹⁵N values were determined. It is thought that these analytes reflect the isotopic composition of the organic matter which itself maybe influenced by environmental factors and cultivar.^25,26

In this paper we report the results of what we believe to be the first use of trace element and isotope ratio data to determine the origin of wheats from Canada, the USA and Europe.

Experimental

Sample preparation

Twenty, 1 kg, wheat samples were obtained from Rank Hovis Ltd., consisting of nine North American wheats (7 Dark Northern Spring wheats (DNS) plus two others), seven Canadian (four CWRS, three TCP) and four European (three French, one German).

Each wheat sample (300 g) was placed on a sieve and washed with deionised water to remove surface contamination. The sample was then dried before being milled using a small Quadramat mill. The mill was carefully cleaned between each milling to avoid cross contamination. Approximately 50 g of the white flour obtained from each sample was then placed into sealable polythene bags before being sent for trace metal and isotope ratio analysis.

Sample analysis

Trace element sample preparation. High purity acids were obtained for Romil (UK) and used throughout the procedure. Dilution water was obtained directly from an Elgar water purification system (18.2 mΩ purity). Calibration standards and quality control solutions were prepared from high purity concentrated stock solutions (Alfa Aesar, Karlsruhe, Germany). The isotopic standards were obtained from NIST (Gaithersburg, MD, USA).

Samples were prepared by closed vessel microwave digestion (Multiwave, Perkin Elmer, Beaconsfield UK). 0.5 g aliquots of the pre-milled flour samples were accurately weighed (to 4 significant figures) into pre-cleaned quartz vials. Digestion was achieved by the addition of 5 g HNO₃ and 0.5 g HCl and heating at 1000 W to ∼200 °C and <75 bar. For each analysis total procedural blanks were prepared and the signal intensities subtracted at the time of analysis.

For Cd and Se analyses the digested samples were accurately diluted with ultra-pure water. For Pb analysis the samples were iteratively diluted with ultrapure water to yield equal signal intensities of ²⁰⁸Pb to minimise the effect of detector linearity.

Further processing was required for Sr analysis to remove the isobaric interference of ⁸⁷Rb. The ion separation is a simplification of that routinely used for thermal ionisation mass spectrometry (TIMS) Sr analysis.²⁷ In brief, the digested samples were transferred into pre-cleaned Teflon vessels and converted to the chloride form by repeat drying and dilution with 6 M HCl. For ion separation the samples were converted to 2 ml 2.5 M HCl and loaded onto 4 ml preconditioned columns of Dowex AG50W X8 200-400 mesh resin. The Sr fraction was heated to dryness prior to final preparation in 2% HNO₃ and iteratively diluted to produce equivalent signal intensities.

Trace element instrumentation and operating conditions

All trace element abundances and ratios were analysed using a Finnigan MAT Element single collector ICP-MS. The Element has an argon plasma source coupled to a magnetic sector mass spectrometer with reverse Niers–Johnson geometry.²⁸

Such instruments are typically referred to as high resolution ICP-MS (HR-ICP-MS). Standard operating conditions are shown in Table 1 along with a complete listing of the scanned isotopes.

Table 1 Standard operating conditions of the Finnigan Element HR-ICPMS for analysis of flour

Finnigan MAT Element magnetic sector ICP-MS
Element	Cd	Se	Sr	Pb
Isotopes monitored	^114Cd	^82Se	^85Rb, ^86Sr, ^87Sr	^202Hg, ^204Pb, ^206Pb, ^207Pb, ^208Pb
Internal standard	^103Rh	^103Rh	—	—
Isotopic standard	—	—	SRM987	SRM981
Samples per peak	20	20	40	40
Number of sweeps	5 × 20	5 × 20	5 × 1000	5 × 1000
Mass window	100%	100%	20%	20%
Integration window	20%	20%	100%	100%
Forward power/W	1150
Ar coolant gas flow rate/l min^−1	14
Ar auxiliary gas flow rate/l min^−1	1.0
Ar nebulizer gas flow rate/l min^−1	1.0
Nebulizer	Glass expansion micro-flow concentric nebulizer
Nebulizer uptake/cm^3 min^−1	0.5
Sampling time/ms	5
Scan type	Escan

---

Under low-resolution conditions (R = 300, 10% valley definition) the magnetic ion separation and focusing generates a flat-topped peak geometry and improved sensitivity over conventional quadrupole instruments. Rapid scanning of the plateau (Escan) can therefore generate precise isotope ratio measurements to within ± 0.1%. This precision is not comparable to that which can be generated by multi-collector instruments such as TIMS or multi collector ICP-MS (MC-ICP-MS). However, in this study the aim was to identify gross regional variations as opposed to the more subtle methods required by geochronology. In addition, single collector instruments are also more widely available and robust and as such represent a realistic measure of any technique designed for routine analytical testing.

Cd was analysed by calibration analysis monitoring the ¹¹⁴Cd isotope. The contribution from ¹¹⁴Sn was observed to be minimal. Instrumental drift was determined though online addition of a ¹⁰³Rh standard. Data quality assurance was maintained by repeat measurement of a gravimetric preparation of Cd diluted to equivalent signal intensity and acid concentration.

Se was determined by calibration measuring the minor Se isotope ⁸²Se (9.2% relative abundance). Interferences acting upon Se can only be resolved at high resolution (R > 9000) but under such conditions the signal intensity is significantly attenuated. It was determined that careful tuning of the plasma conditions could significantly reduce the effects of interferences such as ⁴⁰Ar⁴²Ca so that their effect upon the results was negligible. The ArCa interference acting upon ⁸²Se is relatively minor, ⁴²Ca comprising only 0.6% of total Ca. Tuning of the m/z 82 signal was coupled with monitoring of the m/z 84 (⁴⁰Ar⁴⁴Ca and ⁸⁴Sr) and m/z 86 and 88 (Sr) until such time that the Sr ratios were close to natural (allowing for instrumental bias). This was largely achieved by tuning the Z-focus of the plasma.

Data quality assurance was maintained by repeat measurement of a gravimetric preparation of Se diluted to equivalent signal intensity and acid concentration.

All Pb isotopes were monitored. In addition, the interference from ²⁰⁴Hg acting upon ²⁰⁴Pb was accounted for by monitoring ²⁰²Hg. The effects of instrumental mass bias were corrected by bracketing sample of SRM 981 (NIST, Gaithersburg, USA). Data quality assurance was maintained through repeat analysis of an in-house Pb isotopic standard.

For Sr analysis, the residual interference from Rb was corrected by monitoring ⁸⁵Rb.

Instrumental mass bias and long-term precision was achieved through repeat measurement of Sr isotopic standard SRM 987 (NIST, Gaithersburg, USA).

Gas isotope instrumentation and operating conditions

The elemental analyser Finnigan CN (Finnigan, Bremen, Germany) with the combustion interface ConFlo II was used. Details of the instrumentation were described elsewhere.²⁹ The Finnigan MAT 252 (Finnigan, Bremen, Germany) mass spectrometer was used to record the different isotope amount ratios. Operating conditions are given in Table 2. The voltages generated by the ion currents were digitised and processed by ISODAT software (Finnigan MAT, Bremen, Germany) in order to get the results as δ¹³C and δ¹⁵N. Pre-milled flour samples were loaded into tin capsules prior to analysis. Flour contains about 20 times less N than C, due to its starchy nature, and the ionisation efficiency for N₂ is smaller than for CO₂. As a consequence, it was not possible to analyse C and N within the same run without overloading the detectors with C and disturbing the baseline. The amount of flour loaded into each tin capsule was therefore iteratively adjusted to optimise signal intensity to within the linear range of the instrument.

Table 2 Standard operating conditions of the elemental analyser and Finnigan MAT 252 mass spectrometer analysis of flour

Elemental analyser conditions
He flow rate	80 mL min^−1
Oxidation tube	Chromium oxide/silvered cobaltous oxide at 1020 °C
Reduction tube	Copper/copper oxide at 650 °C
GC	Poraplot Q (1 m)
Mass spectrometer conditions
High voltage	10 kV
Emission current	1.5 mA
Analyser pressure	10^−6 mbar
Source pressure	10^−5 mbar

---

For carbon analysis, the CO₂ working reference gas used was calibrated against the oil reference material NBS 22 (δ¹³C = −29.73 ± 0.09‰) (NIST, Gaithersburg, MD, USA), and was found to have a value of δ¹³C = 32.54 ± 0.19‰ (based on a standard deviation of ten measurements). This gas was used as reference gas for all of the flour measurements. The linearity region for the isotope amount ratio n(⁴⁵CO₂)/n(⁴⁴CO₂) as a function of the intensity of m/z = 44 was 1 to 6 V. Only analyses within this range were used in the final values.

Each flour sample was analysed in five replicates, and the reference, NBS 22, was analysed once between each sample. During each run, the reference gas CO₂ was introduced before and after the sample peak. The first reference peak was used as the calibration for the delta values, the second was used as a drift monitor.

The reference working gas N₂ used at LGC was nitrogen from air, which is considered to have a value δ¹⁵N = 0.00‰. The linearity region for the isotope amount ratio n(²⁹N₂)/n(²⁸N₂) as a function of the intensity of m/z = 28 was 0.5 to 4.5V. Signal intensities outside of this range were not normally included in the final calculation.

Five replicates were analysed for each flour sample, and caffeine was analysed once between each sample as a control sample for drift. During each run, the reference gas N₂ was introduced twice before and twice after the sample peak. The first reference peak was used as the calibration for the δ values, the second one and the fourth one were used as a checks for any possible drift or problems. The third peak was not considered because of an interference from a peak at m/z = 28 due to CO gas eluted after the nitrogen from the sample.

Data analysis. Data analysis was carried out using STATISTICA v5.5 (StatSoft, Inc., 1999, Tulsa, USA). The method chosen was forward stepwise discriminant analysis (DA)^30,31 which can be used to produce statistical models that use the input variables (in this case the elemental profile of the samples) to discriminate between two or more naturally occurring groups. The stepwise selection process was used to eliminate from the model those variables which do not contribute significantly to the classification.

With multivariate methods such as DA there is always the potential to over fit the data and produce models which discriminate well for the data set used to create the model(s) but have no predictive power to classify future unknown samples. To reduce this potential over-fitting two approaches can be used; (1) leave-one-out cross validation or; (2) the calibration model can be checked using a validation data set. We have used the latter approach in this pilot study. The calibration set consisted of 15 samples and the validation set of five samples from one year's harvest.

Results

The geographical origin of the samples, an indicator variable showing if the sample was used in the calibration or validation set (Cal/Val) and the mean results from the replicate trace element and isotope analysis are shown in Table 3. As a first step the individual trace metal elements results were assessed separately to determine if they are sufficient to correctly characterise the geographic origins of the white wheat flours. For example, it is well known that selenium concentrations can be used to discriminate between North American wheats and their European counterparts.¹⁷ However, the differences between USA and Canadian wheats are not sufficient to be able to use this one element to classify all the samples (Fig. 1).

Fig. 1 Selenium concentration by geographical origin.

Table 3 Sample details and results of the isotope analysis of 20 white wheat flours

Sample	Cal/Val	Origin	Sub group	Se/ng g^−1	Cd/ng g^−1	^208Pb/^206Pb	^207Pb/^206Pb	^87Sr/^86Sr	δ ^13C (‰)	δ ^15N (‰)
a 3 samples were rejected for Sr analysis because the Rb separation failed.
AU286	0	USA	DNS	239.9	19.8	2.093	0.895	0.8649	−25.515	17.038
AU294	1	USA	DNS	251.5	19.2	2.096	0.903	0.8212	−25.418	17.792
AU319	0	USA	DNS	309.8	21.6	2.077	0.890	—^a	−25.533	16.698
AU320	0	USA	DNS	346.5	22.8	2.053	0.875	0.7240	−25.423	16.590
AU322	0	USA	DNS	354.1	22.0	2.078	0.886	0.7185	−25.388	16.556
AU330	0	USA	DNS	358.7	19.1	2.074	0.892	0.7113	−25.284	16.924
AU332	0	USA	DNS	298.5	22.9	2.075	0.876	0.8312	−25.424	17.424
AU334	1	USA	Unknown	510.2	23.2	2.086	0.891	0.7744	−25.330	15.453
AU335	0	USA	Unknown	193.7	19.6	2.077	0.884	0.7706	−25.890	14.574
AU284	0	Canadian	CWRS	515.5	22.1	2.083	0.899	0.7131	−24.528	16.995
AU317	1	Canadian	CWRS	302.0	229	2.079	0.884	0.7374	−24.502	15.797
AU327	0	Canadian	CWRS	636.1	25.5	2.079	0.884	0.7300	−24.750	17.198
AU331	0	Canadian	CWRS	772.4	23.9	2.083	0.899	—^a	−24.432	17.322
AU308	1	Canadian	TCP	291.0	17.4	2.041	0.881	0.7429	−25.022	16.164
AU326	0	Canadian	TCP	383.0	47.9	2.071	0.891	0.7683	−24.896	15.930
AU329	0	Canadian	TCP	358.5	20.8	2.051	0.876	—^a	−24.844	16.150
AU318	1	Europe	France	25.2	29.6	2.037	0.861	0.7286	−27.205	15.096
AU321	0	Europe	France	39.0	19.8	2.087	0.886	0.7178	−26.836	14.769
AU324	0	Europe	France	44.7	22.4	2.113	0.912	0.7357	−26.314	14.880
AU328	0	Europe	Germany	29.9	29.0	2.100	0.901	0.7357	−26.976	13.560

---

Significant variations were observed in both the Pb and Sr isotopic results for the suite of samples. The most elevated Sr ratios were observed in some of the USA samples which may be linked to high ⁸⁷Sr/⁸⁶Sr values for old continental rock types and from Rb rich metamorphic mineral phases such as biotite which will be less prevalent in the European samples. However, significant overlap occurs between the lower end of the North American array and the European samples rendering the occurrence of some high ratios an inconclusive test alone. Likewise the variation in Pb ratios does not follow with geographic locations. It is possible that the slight elevation in ²⁰⁸Pb/²⁰⁶Pb in some European samples could be indicative of Pb pollution from refined lead (until recently used in petrol for example) but this is far from conclusive.

It is also of note that whilst the majority of samples exhibit near identical Cd results one sample shows an elevated level which may possibly be attributed to localised anthropogenic pollution.

The variation in delta values for carbon and nitrogen by geographic origin are shown in Fig. 2. It is clear from these plots that δ¹³C is a good indicator of origin but δ¹⁵N is not.

Fig. 2 δ ¹³C and δ¹⁵N values by geographical origin.

Several DA models using combinations of the trace metals, isotope ratios and δ¹³C, δ¹⁵N values, predicted with 100% accuracy the origin of all the white flour samples (in both calibration and validation sets). Result not reported here. The most parsimonious DA model, however, involved only one variable, the mean δ¹³C-isotope value. This model consists of three classification functions:

Class 1 (USA) = −429.25δ¹³C − 5471.35 (1)

Class 2 (Canadian) = −416.41δ¹³C − 5148.87 (2)

Class 3 (European) = −449.82δ¹³C − 6008.17 (3)

The predicted geographical origin is found by calculating the three class scores for the determined mean δ¹³C value. The highest score of the three indicates the country of origin.

For example, for a mean δ¹³C value equal to −25.423 ‰ the class scores are:

Class 1 score (USA) = −429.25(−25.423) − 5471.35 = 5441.47

Class 2 score (Canadian) = −416.41(−25.423) − 5148.87 = 5437.52

Class 3 score (European) = −449.82(−25.423) − 6008.17 = 5427.60

The highest score is 5441.47 so the predicted country of origin is the USA.

Of course a DA model is not required in this case as the δ¹³C values can be used directly to classify the samples

Within each country of origin there are, however, sub-groups, which may also be classifiable using DA but, because of the small size of the data set in this pilot study it is not possible to validate the model with a test set. In particular there is only one German and two USA samples of unknown origin. The best DA model for the subgroups is based on two variables (δ¹³C and δ¹⁵N). This model predicts all the sub-groups correctly (see the canonical analysis plot, Fig. 3) but cannot be validated so it is possible over-fitting has occurred.

Fig. 3 Canonical plot of within geographical region wheat sub-groups.

The best DA model using trace metals data only required selenium and cadmium. This model, however, incorrectly classified 50% of the USA and Canadian wheats in the test set.

Discussion and conclusions

A limited number of wheat samples of known origin have been used in this pilot study in order to evaluate the methodologies potential to distinguish between wheat grown in different geographical regions. It was found that only a single analyte, δ¹³C, is required to discriminate correctly between wheat grown in Europe, Canada and the USA.

The method also shows potential for extension to different wheat cultivars and blends within each geographical area but, its potential to detect adulteration has not been evaluated as this would require using a larger set of samples which include deliberately adulterating North American samples with cheaper European samples. A follow-up study using a larger set of samples will also allow wheat from other European countries, including the UK, to be fitted into the model.

With more geographically diverse samples and adulterated samples available it may be necessary to include some of the trace metal isotope variables which were eliminated in this pilot study. At this stage, therefore, we cannot categorically state that any of the analytes measured in this study will not be useful. In particular adulteration of the more expensive North American wheat by cheaper European wheat may be detectable by a fall in selenium levels in combination with other elements.

Acknowledgements

This work carried out was supported by Rank Hovis (Central Laboratories) and under contract with the Department of Trade and Industry as part of the National Measurement System Valid Analytical Measurement Programme. The authors also thank Stan Timms (RHM Technology) for preparing the samples and the referees for their useful comments.

References

1. L. Lasztityin The Chemistry of Cereal Proteins, CRC Press, Boca Raton, FL, 2nd edn. 1996.
2. Food Authenticity: Issues and Methodology, Final Publication of Concerted Action n° AIR3-CT94–2452, Eurofins Scientific, Nantes, 1998, p. 26-32.
3. J. A. Michard, S. J. Foulk and J. D. Winefordner, J. Agric. Food Chem, 1979, 27(6), 1326.
4. G. J. Martin, J. B. Founier, P. Allain and Y. Mauras, Analusis, 1997, 25(1), 7.
5. h. P. Dettmar, G. S. Barbour, K. T. Blackwell, T. P. Vogl, D. L. Alkon, F. S. Fry, J. E. Totah and T. L. Chambers, Comput. Chem., 1996, 20(2), 261.
6. M. Day, B. L. Zhang, G. J. Martin, C. Asselin and R. Morlat, J. Int. Sci. Vigne Vin, 1995, 29(2), 75.
7. J. D. Greenough, H. P. Longerich and S. E. Jackson, Can. J. Appl. Spectrosc., 1996, 41(23), 76.
8. M. Plessi and A. Monzani, Atti Soc. Nat. Mat. Modena, 1992, 123, 85.
9. M. J. Latorre, R. Pena, C. Pita, A. Botana, S. Garcia and C. Herrero, Food Chem., 1999, 66(2), 263.
10. K. A. Anderson, B. A. Magnuson, M. L. Rschirgi and b. Smith, J. Agric. Food Chem., 1999, 47(4), 1568.
11. R. S. Schwartz and L. T. Hecking, J. Anal. At. Spectrom., 1991, 6(8), 637.
12. F. Angerosa, O. Breas, S. Contento, C. Guillo, F. Reneiro and E. Sada, J. Agric. Food. Chem., 1999, 47(3), 1013.
13. G. Manca, F. Camin, G. C. Coloru, A. Del Caro, D. Depentori, M. A. Franco and G. Versini, J. Agric. Food Chem., 2001, 49(3), 1404.
14. R. Herber, in Handbook on Metals in Clinical and Analytical Chemistry. H. Seiler, A. Sigel, H. Sigel., Marcel Dekker, New York, 1994.
15. L. Friberg, C. Elinder and T. Kjellstrom, Environmental Health Criteria 134. Cadmium, World Health Organisation, Geneva, 1992.
16. M. Lamand, J. Tressol and J. Bellanger, in Trace Elements and Free Radicals in Oxidative Diseases, ed. A. Favier, J. Neve and P. Faure, American Oil Chemists Society, Champaign, IL, 1993, p. 276–283.
17. M. P. Rayman, Lancet, 2000, 356, 233–241.
18. G. Faure, Principles of Isotope Geology, John Wiley and Sons, New York, 2nd edn., 1986.
19. B. L. Ingram and P. K. Weber, Geology, 1999, 27(9), 851–854.
20. C. Latkoczy, T. Prohaska, G. Stingeder and M. Teschler-Nicola, J. Anal. At. Spectrom., 1998, 13(6), 561–566.
21. C. Marisa Almeida, M. Teresa and S. Vasconcelos, J. Anal. At. Spectrom., 2001, 16, 607–611.
22. G. Wannemacker, F. Vanmhaecke, L. Moens, A. Mele and H. Thoen, J. Anal. At. Spectrom., 2000, 15, 323–327.
23. K. J. R. Rosman and W. Chisholm, Analysis, 1994, 22(7), M51–M5.
24. M. Hamester, H. Stechmann, P. Krause and W. Dannecker, Anal. Proc., 1992, 29(7), 293–296.
25. H. W. Yeh and W. M. Wang, Proc. Natl. Sci. Counc. Repub. China., Part B, 2001, 25(3), 137–147.
26. T. Yoneyama, H. Fujiwara and W. M. Engelaar, J Exp. Bot., 2000, 51(344), 559–566.
27. F. McDermott and C. Hawkesworth, Earth Planet. Sci. Lett., 1991, 104, 1–15.
28. U. Giessmann and U. Greb, Fresenius’ J. Anal. Chem., 1994, 350, 186–193.
29. D. A. Merritt, K. H. Freeman, M. P. Ricci, S. A. Studley and J. M. Hayes, Anal. Chem., 1995, 67, 2461.
30. P. A. Lachenbruch, Discriminant Analysis, Hafner Press, New York, 1975.
31. E. K. Kemsley, Discriminant analysis and class modelling of spectroscopic data, John Wiley and Sons, New York, 1998.

---

Footnote

† Presented at the Eleventh British National Atomic Spectroscopy Symposium (BNASS), Loughborough University, UK, July 8–10, 2002.

---