Varietal and Geographic Classification French Red Wines Terms Elements, Amino Acids and Aromatic Alcohols
Varietal and Geographic Classification French Red Wines Terms Elements, Amino Acids and Aromatic Alcohols
Varietal and Geographic Classification French Red Wines Terms Elements, Amino Acids and Aromatic Alcohols
Jean-Claude Bouvier
INRA, Institut des Produits de la Vigne, Station Experimentale de Pech-Rouge,
11430 Gruissan, France
Patrick Symonds
Laboratoires associes de Recherches Agricoles (LARA), 271 avenue de Grande-Bretagne,
3 1300 Toulouse, France
and
Alain Bertrand
Universite de Bordeaux 11, Institut d’oenologie, 351 cows de la Liberation,
33405 Talence Cedex, France
ABSTRACT
Thirty-four French red wines @om three regions already studied for their
anthocyanin and jlavonoid constituents have been jiwther analysed for
elements, amino acids and aromatic alcohols. An interpretation of the
differences between wines related to their different geographic and varietal
origins has been made @om the results of statistical analyses: F statistic,
principal component analysis (PCA) and stepwise discriminant analysis
(SDA).
Wine samples produced near Bordeaux were found to be characterised by
higher rubidium and lower lithium and calcium concentrations. Differences
25
J Sci Food Agric 0022-5l42/88/.$03.500 1988 Society of Chemical Industry. Printed in Great Britain
26 P E t i b a n t , P Schlich, J-C Bouoier, P Symonds, A Bertrand
between wine samples made from the same grape variety or produced in the
same region are mainly related to differences in potassium and ash contents.
The PCA demonstrates that the concentration of most amino acids in wines is
mainly a function of the technology used in wine production. Moreover, it
shows that the concentration of proline, hydroxyproline and ethanolamine
are generally lower in wines prepared from Carignan, Cinsault and Grenache
grapes than in wines from Cabernet Franc, Cabernet Sauvignon or Merlot
grapes. Finally, it shows clear differences between wines according to the
latitude of the production centre. These differences correspond to increasing
total nitrogen content from south to north, ie Narbonne and Bordeaux versus
Angers.
In discriminant analysis Bordeaux wines were characterised by high
rubidium and low arginine concentrations, Angers wines by a lower arginine
content and Narbonne wines by their abundance of magnesium and dejiciency
of ethanolamine.
1 INTRODUCTION
growth phase, some are excreted by living yeasts at the end of the fermentation or
released by proteolysis during the autolysis of dead yeasts, and others are produced
by enzymic degradation of the grape proteins. Moreover, it is well known that the
amino acid content of grapes is dependent upon the fertilisation and climatic
conditions, and on duration of skin maceration in the must. In spite of these
different sources of variation, some authors successfully used amino acids to
differentiate wines made according to grape variety used (Ough and Stashak 1974;
Ooghe et a1 1981;Cantagrel et a/ 1982; Symonds and Cantagrel 1982; Schaefer et a1
1983; Rizzon 1985) or to production area (Ooghe et a1 1981; Forina et a1 1986;
Tapias et a1 1987).
2 EXPERIMENTAL
2.1 Materials
The wines were described in a previous publication (Etievant et a1 1988). Their
origins (varieties and production areas) and code names are itemised in Table 1
together with the weight of grapes used for fermentation.
2.2 Methods
Wines were analysed using methods described previously (Moore et a1 1958;
Ribereau-Gayon et a1 1976; Medina 1978; Cantagrel et a1 1982; Lacasta 1982).
Chloride and ash were determined as described by Ribereau-Gayon et a1 (1976).
Potassium, sodium, calcium, lithium and rubidium were quantified by flame
emission spectrophotometry (Lacasta 1982)using a Varian AA175, and magnesium
and manganese by flameless absorption spectroscopy (Medina 1978)using the same
instrument.
Acid, neutral and basic amino acids were separated on two different columns
packed with a cation exchange phase as described previously by Moore er a1 (1958),
using an amino acid autoanalyser (T'echnicon). T'yrosol and tryptophoi were
separated by high performance liquid chromatography on a reverse phase column
((218) as previously described (Salagoity-Auguste and Ekrtrand 1984). Total
nitrogen was estimated by the Kjeldahl method (Anon 1970).
Data were processed using the same computer and the same statistical packages
as previously described (Etievant er al 1988). Analyses of the 34 wines were
performed on duplicate samples. Sixty-eight results were therefore used for
statistical analyses (F statistic, principal component analysis (PCA) and stepwise
discriminant analysis (SDA)),but sample plots of PCA show only the position ofthe
average of the duplicate samples. Nine of the wines were analysed a third time as
'test wines' by SDA in order to check the efficiency of the discriminating model.
The names of the variables quantified and their associated code names are itemised
in Table 2 for trace elements and ash, and in Table 3 for amino acids, ethanolamine,
28 P Etievant, P Schlich, J-C Bouvier, P Symonds, A Bertrand
TABLE 1
Identities of the Wines Studied
Area of production
Grenache 273 G. 1
100 G.2
100 G.3
382 G.4
1734 G.5
Carignan 100 A.l
102 A.2
417 A.3
5974 A.4
102 AS
90 A.6
Cinsault 100 1.1
501 1.2
4300 1.3
a
Merlot I00 M. 1 M* 1
a
100 M.2 M*2
100 M.3 (I
M*3
Cabernet -Sauvignon 256 s.l (1
a
s*1
100 s.2 5*2
a
114 s.3 5*3
Cabernet -Franc II
F* 1 50 F+1
a
F*2 50 F+2
a
F*3 50 F+3
50 F+4
50 F+5
aromatic alcohols and total nitrogen. Coefficients of variation observed for replicate
analyses are lower than or equal to 7%, thus demonstrating the fairly good
repeatability of the measurements (Tables 2 and 3).
3.1 Elements
A selection of some of the inorganic constituents of wine was first made. Some
elements, the concentration of which can be modified by contamination in the field
or during processing (eg copper, which can arise in wine from copper pipes and taps,
and iron, which can increase drastically when rusted wires are in use in the
vineyard), were disregarded. Eight ions and the weight of ash were finally selected
for the study (Table 2).
The variations in concentration of these elements in the different wines studied
Classijccation of French red wines 29
TABLE 2
Inorganic Substances, Names and Codes of the Variables Quantified
~~ ~
are less than 100% of the lowest value, with maximum variations for lithium (97 %),
rubidium (95%) and sodium (78%). The Fisher's test values show that all the
elements studied, except manganese can globally discriminate wines made from
different grape varieties (Table 4), and that Rb and Li are the only variables that can
provide a statistically significant difference (P< 001) between wines produced in
different areas (Table 5).
Ash and potassium are the only variables that show a high correlation (0.92).This
correlation is easily explained by the high level of potassium among cations in wines
( 275 %). Correlations found by Rizzon (1985)between K and Rb or between Li and
Rb when analysing wines from three different grape varieties (Merlot, Cabernet
Sauvignon and Malbec) and produced in the Bordeaux areas were not confirmed by
the present investigation in which four other varieties and two extra production
areas were studied.
Sample differences are illustrated in the first sample plot of the PCA (Fig 1) in
which 57 % of the variance is projected. All variables except rubidium are positively
loaded on the first principal component (PC1). PC2 shows contrasts between
correlated Li and Ca, and the negatively correlated Rb and C1 (Fig 2). A clear
difference between wines produced near Bordeaux and Narbonne can be noticed
along the second axis. The corresponding factor loadings (Fig 2) and the raw data
(Table 5) demonstrate that the particular position of Bordeaux wines is due to
higher Rb and C1 and to lower Li and Ca concentrations. The concentrations of Rb
in wines produced in Bordeaux (2.9 mg litre-' for Cabernet Sauvignon and 1.8 mg
litre-' for Merlot) are in good agreement with the values published by Rizzon
(1985). As the concentration of this trace element statistically proved to be
unmodified when the time of skin maceration in the must varied from 0 to 240 h
30 P Etievant, P Schlich, J-C Bouvier, P Symonds, A Bertrand
TABLE 3
Amino Acids and Aromatic Alcohols, Names and Codes of the Variables Quantified
Proline Pro L 5
Hydroxyproline HYP L <1
Arginine Arg L 2
Glutamic acid Glu L 1
Gamma-amino butyric acid Aha L 2
Ornithine Orn L 2
Alanine Ala L 3
Serine Ser L 2
Glycine GlY L 3
Valine Val L 2
Leucine Leu L 4
Aspartic acid ASP L 3
Asparagine AsPg L 3
Threonine Thr L 2
Isoleucine Ile L 4
Methionine Met L 3
Lysine LYS L 3
Tyrosine TYr L 2
Phenylalanine Phe L 3
Histidine His L 2
Ethanolamine Eami L 2
Tyrosol Tyro B 6
Tryptophol TVP B 3
Total nitrogen TN DU 6
TABLE 4
Inorganic Substances Measurements. Mean Values Calculated for Each Variety and
Significance of the Difference (Ash, g litre-’; Elements, mg litre-‘)
~~
TABLE 5
Inorganic Substances Measurements. Mean Values Calculated for Each Area of Production
and Significance of the Difference (Ash, g litre-'; Elements, mg litre-')
20% 2
A.5
G.2
I.3 h2
.
s'2A.1
G.1
M.l I" G.f 5.1
I .2 F.1 6.3
I% p:$ M.2 1
M.3
F. 3
5*3 A.3
-
37%
Ma1 A.4
Fig 1. Element measurements. PCA sample plot along the first two axes. Codes refer to Table 1
32 P Erievant, P Schlich, J-C Bouvier, P Symonds, A Bertrand
t'
Fig 2. Element measurements. First and second vector loadings of the PCA.
LOW
_ _92.0-
___ 800-
-- 60.0- SO.0- 91.0- -
SO.0 LOO -
- -
EZQ
61.0 610 WQ Eli) ZZQ 9ZQ LZ.0 XOQ PZ.0
PIQ PZO EI.0 ZZQ 51.0 IZQ LIO- OJd
ISQ 8PQ EEQ 8EQ OS.0 PS.0 ES.0 SE0 YSQ LPU 950 IPQ SSQ LZ.0 IZQ 50.0- UL
__ __ -
6P.0 ZPQ 6EQ ZPQ 55.0 6P.0 LSQ SEQ ISQ LPQ 190 EPQ PSU -
SPQ 09Q 91.0- eqV
- -
L6.0 E60 68.0 98.0 06.0 060 16.0 98.0 68.0 E6Q 88Q 58.0 LLQ 6PQ 20.0 SIQ SIQ- EZQ JJS
03.1 S6.0 260 06.0 16.0 16.0 E6Q 16.0 16.0 P6.0 P6Q 68 Q 6L.0 090 PI0 8ZQ LIQ- EZ.0 '91
00.1 16.0 16.0 260 E6.0 Z6Q 060 L8.0 88.0 06.0 980 89Q 8S.0 51.0 Sl0 81.0- 91.0 dSV
00.1 560 S6Q 161) 260 06.0 YL.0 S8.0 580 58.0 19.0 nso 60.0 ZZQ E1.0- 81.0 JlI
00.1 56 16.0 56-0 060 SL0 180 98.0 98.0 65.0 090 SIQ IE.0 Ll.0- 91.0
00 961) P6Q S80 9LQ 18Q S8.0 58.0 p9.0 55.0 E 1.0 ZEQ PZQ- 11.0 rial
00.1 06.0 P8Q 08.0 ESQ P8 .0 18.0 Y9.0 IS.0 80.0 PZ.0 02.0- 11.0 s h
00. I 98.0 8LQ 98.0 16.0 16.0 ILQ 99.0 IZQ 8E.0 SZQ- 21.0 alld
00.1 18.0 Y8Q L8Q E8Q W.0 LSQ I 1.0 PZQ 01.0- 82.0 IRA
00.1 C6.0 16.0 bL.0 08.0 SQ.0 91.0 810 PZ.0- 82.0 "If)
00.1 160 180 PRQ zso 91.0 LZQ 619
00.I 18.0 6LQ 89.0 ZEQ ZZQ el v
00.1 09.0 8L.0 EZQ so0 JKl
00.I OfQ
__ LZ.0 dSV
00.I 010- BS!H
00.1 LSQ PEQ- %JV
ZE.0- LOO- "
'0
34 P Etievant, P Schlich, 3-C Bouvier, P Symonds, A Bertrand
TABLE 7
Amino Acids and Aromatic Alcohols Measurements. Mean Values Calculated for Each
Variety and Significance of the Difference (mg litre- ’)
TABLE 8
Amino Acids and Aromatic Alcohols Measurements. Mean Values Calculated for Each Area
of Production and Significance of the Difference (rng litre-’)
varieties and production areas (Tables 7 and 8). Proline (Pro), 4-hydroxyproline
(Hyp) and ethanolamine (Eami) are interesting variables with high F values for both
variation factors. Moreover, proline and its hydroxylated derivative have earlier
been cited by several authors as discriminating wines made from different grape
varieties and produced in different areas (Ough and Stashak 1974; Ooghe er all981 ;
Symonds and Cantagrel 1982). These authors attribute this interesting property to
proline’s not usually being metabolised by yeasts during the fermentation. It should
therefore be possible to find in the wine some differences originally present in the
grape juice but not subsequently masked by the amount of proline excreted by the
yeasts at the end of the fermentation, or liberated by yeast autolysis. It can be
noticed that gamma-aminobutyric acid (Aba), which has previously been shown by
several authors to discriminate varieties in red wines (Ooghe el a1 1981; Polo et a1
1984; Rizzon 1985), also exhibits significant F values for both variation factors
studied. Its use in wine authentication must, however, be undertaken with care since
this amino acid is metabolised, to some degree, by yeasts during the fermentation.
The two first principal components of the PCA of the amino acid data explain
36 P Etievant, P Schlich, J-C Bouvier, P Symonds, A Bertrand
14% A 2
F.4F.3
F+1
F*2 F*5 5*3
Ma3
5.2
s.1 M.3m3 6.5
ML1 1
Ma2
b
5.3 M1 s.l 5.2
%
I? A.6 56%
1 . 2 ~ 2 Go4 G-3
A.lG.1 I.1 1.3
A.2 A.5
A.3 A.4
Fig 3. Amino acid measurements. PCA sample plot along the first two axes. Codes refer to Table 1.
70% of the total variance between samples (Fig 3). PC1 is a function of a large
number of positively loaded variables including the 13 heavily correlated amino
acids already mentioned and asparagine (Aspg) plus histidine (His). PC2 is
positively correlated with proline (Pro), hydroxyproline (Hyp), arginine (Arg),
gamma-aminobutyric acid (Aba), ethanolamine (Eami) and total nitrogen (TN).
The corresponding sample plot shows a large dispersion of the wine samples
along the first axis, mainly related positively to the different amino acids cited above
as highly correlated (Figs 3 and 4, Table 6). Since most of these components are
known to be utilised by yeasts, their concentrations in wine must vary mainly with
the conditions of the fermentation (yeast strain, temperature, time of storage over
yeast, NH: added, etc). However, they vary also with the maturity of the grapes.
This axis can consequently be considered as a ‘technological axis’. For instance, A.3
and A.4, G.4 and G.5, and 1.2 and 1.3 are made from grapes harvested at the same
time in similar vineyards, and they differ only by the amount of grapes used for
fermentation (around 400 kg, for A.3, G.4 and 1.2, and more than 2000 kg for A.4,
G.5 and 1.3). It is clear (see locations of these wines on Fig 3) that PCl is positively
correlated with the amount of grapes and therefore with the conditions of
fermentation, thus confirming the technological interpretation.
A rough classification of wines into two groups, related to the grape varieties, can
be noticed along PC2 (Figs 3 and 4). The first group of wines made from Carignan,
Cinsault and Grenache grapes contains lower concentrations of proline,
Classijcation of French red wines 37
Fig 4. Amino acid measurements. First and second vector loadings of the PCA.
TABLE 9
Individual Stepwise Discriminant Analyses
Type of Variation Number oj‘ <: oj’ correctly ”/, of correctly
cariable factor selected cla.s.s1$ed classified
oariahles base wines test wines
from Angers containing less ornithine and arginine, which are positively loaded on
PC3. N o explanation can be given for this fact. However, it is not surprising that
these two variables vary together as they have similar biochemical pathways
(Louisot 1983).
3.3 Selection of discriminating variables
Four different SDA were first performed using, separately, elements and amino
acids to discriminate grape varieties or production areas.
The few conclusions obtained from PCA concerning varietal characteristics and
the poor results of the individual SDA performed on the same criterion of variation
(Table 9) led to the focusing of interest on production-area discrimination.
As almost no redundancy occurred between variables of the two sets of data
(serine, aspartic acid and threonine are correlated with potassium and consequently
ash 0 6 1 < r < 0 . 6 6 ) , an attempt was then made to improve the discrimination by
combining all variables in an SDA. The efficiency of this last procedure is
demonstrated as 94% of base wines, and 100% of test wines were well classified
using as few as four variables (Table 9).
This fifth SDA is described in Table 10 and Fig 5. I t will be noted that two
elements (rubidium and magnesium) and two amino acids (ethanolamine and
arginine) were selected. As previously concluded from PCA, Bordeaux wines are
mainly characterised by a high concentration of rubidium, positively correlated
with the first discriminant factor (DFI), and also by a deficiency of arginine
negatively loaded on DF2. Moreover, a clear discrimination can be observed along
the same axis for Angers wines which are characterised by higher arginine
concentrations.
The other two selected variables allow the characterisation of the Narbonne
wines by the relative deficiency in ethanolamine and the relative abundance of
magnesium. It is worth noticing that the values of Fisher’s statistic for the
discriminant factors are both higher than the individual values calculated for the
Classification of French red wines 39
TABLE 10
Stepwise Discriminant Analysis. Correlations between Selected Variables and Discriminant
Factors
t2
N
N
A
A A kA
A
A
A
Fig 5. Discriminant analysis of winrs made for different arcas of production: sample plot along the two
discriminant factors (DF). N = Narbonne, B = Bordeaux, A = Angers.
two amino acids and the two elements selected. This demonstrates first the
advantages of the multidimensional approach provided by SDA, and secondly the
complementariness of amino acids and elements for characterising the vine
production areas.
ACKNOWLEDGEMENTS
Thanks are due to: Pr Feuillat (Universite de Dijon) for total nitrogen analyses, to
Messrs C Flanzy, P Benard, J-M Barillere, C Asselin and R Morlat (INRA), and
40 P Erii'cuni, P Schlich. J-C Boucier, P Spmonds, A Bertrund
REFERENCES
Amerine M 1958 Composition of wines. 11. Inorganic constituents. Adv Food Res 8 133-225.
Anon 1970 Methodes oficielles d'analyse physiques et chimiques du lait. J offic Rep
Francaise 25 Jan 1970.
Bermejo F, Baluja Santos C 1974 Deterrninacion de litio en 10s acidos aminopolicarboxilicos
como reguladores espectrofotometricos. Actu Cientifica Compostelana 11 3-1 2.
Cantagrel R, Symonds P, Carles J 1982 Composition en acides amines du moSt en fonction
du cepage et de la technologie et son influence sur la qualite du vin. Sci Aliments 2 109--142.
Etievant P, Schlich P, Bertrand A, Symonds P, Bouvier J C 1988 Varietal and geographic
classification of French red wines in terms of pigments and flavonoid compounds. J Sci
Food Agric 42 39-54.
Forina M, Armanino C, Castino M, Ubigli M 1986 Multivariate data analysis as a
discriminating method of the origin of wines. Vitis 25 189-201.
Lacasta F 1982 Dosage de quelques metaux dans les vins par spectrometrie d'adsorption et
d'emission de flamme, Rapport de stage BTAO, Station Agronomique et Oenologique,
University of Bordeaux 11.
Louisot P 1983 Catabolisme des proteines et metabolisme des amino-acides. In: Biochimie
generale et midicale/structurale, metabolique, semeiologique. 4: Catabolismes, energiiique
cellulaire, biosyntheses, hormones. Simep, Bruxelles, p 735.
Maarse H, Schlump P, Tas A C , Schaefer J 1987 Classification of wines according to type and
region based on their composition. Z Lehensm L'ntrrs Forsch 184 198-203.
Medina B 1978 Application de la spectrophometrie d'adsorption atomique sans flamme au
dosage de quelques metaux dans le vin. Thesis, University of Bordeaux 11.
Medina B, Sudraud P 1979 Teneur des vins en lithium. Ann Fals Exp Chim 72 65-71.
Medina B, Van Zeller A 1984 Differenciation des vins de trois regions d e France. Conn Vigne
Vin 4 225-235.
Moore S, Spackman P H, Stein W H 1958 Chromatography of amino acids on sulfonated
polystyrene resins, an improved system. Anal Chem 30 1185-1 190.
Moret I, Scarponi G, Zanin S, Camaiani G , Toniolo A 1980 Analytical parameters in the
characterization of three Venetian wines. Application of the statistical linear discriminant
analysis. Am J Enol Vitic 31 245-249.
Moret I, Di Leo F, Giromini V, Scarponi G 1984 Multiple discriminant analysis in the
analytical differentiation of Venetian white wines. 4. Application t o several vintage years
and comparison with the k nearest-neighbor classification. J Ayric Food Chem 32 329-333.
Ough C , Stashak R 1974 Further studies on proline concentration in grapes and wines. Am J
E n d Vitic 25 7-12.
Ooghe W, Kastelijn H , De Waele A 1981 Determination de l'origine d'un vin rouge a h i d e
du spectre des acides amines. Ann Fals E x p Chim 789 381-408.
Polo M, Martin-Corder0 P, Cabezudo M 1984 Influence des caracteristiques varietales de
mohts de cepages differents sur la fermentation alcoolique par une souche de levure
selectionnee. Bull 01 V 638 313-321.
Ribereau-Gayon J, Peynaud E, Ribereau-Gayon P, Sudraud P 1976 Matieres minerales. In:
Traite d'0enoloyie. Sciences et techniques du cin. Tome I , Anulyse et contrsle des vins.
Dunod, Paris, pp 237-286.
Rizzon L 1985 Incidence de la maceration sur la composition chimique des vins. Thesis,
University of Bordeaux 11.
Salagoity-Auguste M H , Bertrand A 1984 Wine phenolics. Analysis of low molecular weight
components by high performance liquid chromatography. J Sci Food Ayric 35 1241-1247.
Classification of French red wines 41