FLAVOUR AND FRAGRANCE JOURNAL

Flavour Fragr. J. 2000; 15: 1±6

Multivariate statistical analysis of the essential oil

composition of Thymus praecox Opiz ssp. polytrichus
(Kern. ex Borb.) Ronn. collected in the Tyrolean Alps
C. Bischof-Deichnik, J. Holtuijzen and E. Stahl-Biskup*
Lehrstuhl fuÈr Pharmazeutische Biologie der UniversitaÈt, Bundesstrasse 43, D-20146 Hamburg, Germany

Received 30 July 1998

Accepted 30 November 1998

ABSTRACT: The essential oils of 141 individual plants of Thymus praecox Opiz ssp. polytrichus (Kern. ex Borb.)
Ronn., emend. Jalas, collected at 16 localities in the Austrian and Italian Alps (Tyrol and South Tyrol), were
investigated concerning their composition and the variations therein. In total, 60 components, almost exclusively
monoterpenes and sesquiterpenes, were identi®ed by GC and GC±MS. The patterns of the oils were processed by
means of multivariate statistical analysis (MSA) including factor analysis (FA), cluster analysis (CA) and
discriminant analysis (DA). The most plausible result, revealing 12 clusters, was obtained by a modi®ed CA
method based on the average linkage method with the Pearson correlation. Considering all other statistical
results obtained, 12 oil types became evident with the following type-characterising compounds: thymol,
geraniol/geranyl acetate, trans-sabinene hydrate/terpinen-4-ol, a-terpineol, linalol, linalol/linalyl acetate,
borneol, trans-nerolidol, hedycaryol, T-cadinol, germacra-1(10),5-dien-4-ol, and germacra-1(10),4-dien-6-ol.
All these oil types have been found before in other Thymus species. The thymol type, found for 46 samples,
represented the most dominant oil type. The chemical similarity with Mediterranean and North Atlantic Thymus
species indicates a mixing of southern and northern elements in the Alps, which is also re¯ected in the
overlapping morphological characters of T. praecox ssp. polytrichus. Copyright # 2000 John Wiley & Sons, Ltd.

KEY WORDS: Thymus praecox Opiz ssp.; polytrichus (Kern. ex. Borb.) Ronn. emend. Jalas; Lamiaceae; Alps;
essential oil; oil types; thymol type; statistical analysis

Introduction vegetation on ¯at, rocky surfaces. In the northern Alps,

hybrids with T. pulegioides L. can be found; at the
Thymus species occurring in the Alps have turned out southern edge the ingression of characteristics of the
to be extremely polymorphic in leaf-shape and indu- Mediterranean Thymus species makes a clear de®nition
mentum. Therefore, taxonomists have classi®ed them in more dicult.
a wealth of subspecies, varieties or formae. According The only chemotaxonomic study of the alpine
to a more sensible concept, Jalas pursued the idea of a Thymus species was published in 1948, presenting poor
broader description of the Thymus species, and chemical data.3 Nevertheless three di€erent oil types
assigned the Thymus plants of the mountains of were observed at that time, namely essential oils rich in
southern and southern central Europe to Thymus phenols (thymol or carvacrol), oils poor in phenols,
praecox Opiz ssp. polytrichus (Kern. ex. Borb.) Ronn. and citral-containing oils with a characteristic lemon
emend. Jalas.1,2 It is one of the ®ve subspecies of the scent. This indicates that the phenomenon of chemical
T. praecox group including plants with long, somewhat polymorphism also exists among the alpine popula-
woody, creeping branches with a terminal in¯orescence. tions, as was found for many other Thymus species
The morphological features characterizing the sub- growing elsewhere.4
species polytrichus include the presence of stems which In order to investigate the essential oil pro®le of the
are hairy mostly on two sides only, and the fact that the alpine Thymus populations as well as the range of the
upper cauline leaves are larger. chemical polymorphism, 141 individual plants were
T. praecox ssp. polytrichus is widespread in all collected at 16 di€erent localities in the alpine regions
altitudes of the Austrian Alps, being often the main of Tyrol and South Tyrol. The essential oils were
isolated by hydrodistillation and analysed by GC and
* Correspondence to: E. Stahl-Biskup, Lehrstuhl fuÈr Pharmazeutische Biologie
GC±MS. For the chemical classi®cation of the oils, the
der UniversitaÈt, Bundesstrasse 43, D-20146 Hamburg, Germany. complete GC data set was processed by means of

Copyright # 2000 John Wiley & Sons, Ltd. Flavour Fragr. J. 2000; 15: 1±6
2 C. BISCHOF-DEICHNIK, J. HOLTHUIJZEN AND E. STAHL-BISKUP

multivariate statistical analysis (MSA). Similar Peak Identification

techniques had been applied to the essential oils
of T. willkommii,5 T. pulegioides ssp. chamaedrys,6 Peak identi®cation was accomplished by comparison of
T. moroderi,7 T. antoninae,7 T. funkii,8 T. hyemalis,9 the mass spectra with those of authentic substances as
T. carnosus,10 T. zygis,11 T. villosus ssp. villosus,12 and well as by means of comparison of retention times of
T. camphoratus.13 authentic substances on two di€erently polar GC
phases.

Experimental
Multivariate Statistical Analysis (MSA)
Plant Material
The MSA was accomplished by SPSS (Superior
The aerial parts of 141 individual plants of T. praecox Performing Software Systems) 6.1.3 for Windows.
ssp. polytrichus were collected at 16 localities in the The aptitude of the complete correlation matrix for
Alps during the ¯owering period in June 1992. OÈtztaler MSA was checked by using the Kaiser±Meyer±Olkin
Alpen (Tyrol, Austria), eight localities (1890 m, criterion (KMO ˆ 0.772).
2400 m, 2260 m, 2400 m, 2020 m, 1920 m, 2050 m, The factor analysis (FA) was based on a principle
1890 m above sea level); Inntal (Tyrol, Austria), one component analysis including a Varimax rotation. The
locality (650 m); Karwendel (Tyrol, Austria), four extraction of the factors was completed on the basis of
localities (1770 m, 1900 m, 2  1700 m); Region Eigenvalue 4 1. Twenty-four variables were eliminated
Trentino-Alto Adige (Province Bolzano, Italy), three from the FA because of their low measure of sampling
localities (1210 m, 1300 m, 600 m). The fresh plants adequacy (MSA 5 0.5). For the ®nal cluster analysis
were carefully dried at room temperature. Herbarium (CA), the results of the FA ( factor values) were
samples have been deposited in the Department of combined with the original data (variable values, both
Pharmaceutical Biology, University of Hamburg. z-transformed).
For the CA, three algorithms were applied: the single
Isolation of Essential Oils linkage method for ®nding the runaways; the average
linkage; and the Ward method in order to ®nd the oil
The dried aerial parts of the plants were submitted to groups. As similarity measures, the Cosinus and
hydrodistillation in a `Karlsruher apparatus' for 2 h Pearson correlations (average linkage method) were
using n-hexane (2 ml) as a solvent. used, and as distance measures, the Euclidian distance
and the squared Euclidian distance (Ward method).
The results of the CA were presented in dendrograms.
Gas Chromatography The de®nition of the oil types was based on a special
CA, using the average linkage method with the Pearson
A Varian 4270 gas chromatograph equipped with an correlation and processing both the factor values (eight
FID, a split-injection system and a FFAP fused-silica factors) plus the variable values of the 24 variables that
column (25 m  0.25 mm i.d.; ®lm thickness 0.25 mm) were eliminated from the FA because of their low MSA.
was used. Carrier gas: nitrogen, 1 ml/min; oven The discriminative variables of the groups were
temperature programmed at 608C for 5 min, then at found by discriminant analysis (DA). For every
38C/min up to 2208C; injector temperature, 2208C; dendrogram a stepwise variable selection was carried
detector temperature; 2508C; injection volume 1±2 ml; out, as the selection rule the Wilks' Lambda was
split ratio, 1:30. For peak identi®cation, a second minimized.
column was used ( fused-silica 30 m  0.25 mm i.d.,
coated with DB-1, ®lm thickness 0.25 mm). Quantita-
tive peak estimation was achieved by means of a Varian
4270 integrator, method 0). Oil composition was given Results and discussion
in peak area percentages.
The GC and GC±MS analyses of the 141 individual oils
of the alpine thyme plants revealed a chemical pattern
Gas Chromatography±Mass Spectrometry of 60 compounds (Table 1). Obviously, the Thymus
populations of the Alps combined elements of
GC±MS analyses were performed using a Finnigan Mediterranean as well as North Atlantic Thymus
MAT ITD 800 mass spectrometer coupled to a species. Thymol, carvacrol, geraniol, terpinen-4-ol,
United Technologies Packard 438 gas chromatograph trans-sabinene hydrate, a-terpineol and borneol have
equipped with a 25 m  0.25 mm i.d. fused-silica often been found in the Mediterranean species, whereas
column coated with DB-Wax ( ®lm thickness 0.25 mm). linalyl acetate, linalol and the sesquiterpenes are more

Copyright # 2000 John Wiley & Sons, Ltd. Flavour Fragr. J. 2000; 15: 1±6
 MULTIVARIATE ANALYSIS OF THYMUS PRAECOX ESSENTIAL OIL 3

Table 1. Constituents of the essential oils isolated from Thymus praecox Opiz ssp. polytrichus (Kern. ex. Borb.)
Ronn. collected in the Tyrolean Alps, and their occurrence in factors 1±8 by factor analysis (FA)
Composition of factors 1±8*
Factor 1 Factor 3 Factor 5
g-Terpinene Geranial allo-Aromadendrene
p-Cymene Neral a-Cadinol
Thymol Geranyl acetate g-Cadinene
a-Pinene Citronellol T-Muurolol
Thymyl acetate Nerol Unknown STHC
Carvacrol Neryl acetate
Factor 6
Factor 2 Factor 4 b-Pinene
Terpinen-4-ol Unknown STHC 1,8-Cineole
Terpinolene Germacrene-D Camphene
trans-Sabinene hydrate b-Bourbonene
b-Cubebene b-Sesquiphellandrene Factor 7
a-Terpinene Germacrene-A cis-b-Ocimene
Sabinene a-Farnesene Oct-1-en-3-ol
trans-b-Ocimene
Factor 8
Thymyl methyl ether
Variables excluded from the FA**
Aromadendrene cis-b-Farnesene Linalyl acetate
b-Bisabolene Geraniol 6-Methyl-heptan-3-one
Borneol Germacra-1(10),4-dien-6-ol a-Muurolene
Bornyl acetate Germacra-1(10),5-dien-4-ol Myrcene
T-Cadinol Hedycaryol trans-Nerolidol
Camphor a-Humulene Octan-3-ol
b-Caryophyllene Limonene Spathulenol
Caryophyllene epoxide Linalol a-Terpineol

STHC ˆ sesquiterpene hydrocarbon. * Components in order of their loading on their factor. ** Components in alphabetical order.

typical of the Thymus species of the Scandinavian 0.772 (middling). The eight factors found explain
countries.4 70.9% of the total variance of possible combinations
As could already be discerned by sensoric testing, of variables, which is fairly good when plant material is
which showed strongly aromatic, lemon-scented, sweet investigated.
and scentless types, there was a wide range of chemical For a better understanding of Table 1, it is important
polymorphism among the alpine thymes. In order to to know that variables within one factor are correlated.
achieve an objective grouping within the individual oils, The loading of the individual variables on the factors
the quantitative data were processed by a combination decreases in the order from the uppermost to the lowest.
of MSA methods. As a ®rst step, factor analysis was The order of the factors indicates the strength of the
carried out; followed by cluster analysis and discrimi- correlation of the variables within a factor, factor 1
nant analysis. being the strongest. Correlations can be coincidental
but they can also be explained by biosynthetic relations
between variables, which is obvious for some variables
Factor Analysis (FA) of factors 1, 2, 3 and 4. For every factor a `factor value'
was calculated which replaced the variable values (ˆ%
Within the total data set, the FA searches for correlated concentration) in the cluster analysis.
variables (variables ˆ individual oil components) and
unites them as bundles, so-called `factors', which are
independent of each other. Quantitatively correlated Cluster Analysis (CA)
variables distort the following cluster analysis. In order
to obtain a more signi®cant result, variables with weak The CA completes the grouping of the cases
correlations (MSA 5 0.5) have to be eliminated from (ˆindividual oils). Concerning T. praecox ssp. poly-
the FA beforehand. In the case of T. praecox ssp. trichus, two di€erent algorithms were used with two
polytrichus, 24 variables were excluded (Table 1, lower di€erent measures each: the average linkage method
part). The remaining 36 variables were assigned to eight (considering the pattern of the oils) with two similarity
factors (Eigenvalue 4 1; Table 1, upper part). By using measures each (Cosinus and Pearson correlations), and
this procedure the proportion of the number of the Ward method (best grouping when distance
variables to cases (cases ˆ individual oils) became measures are used) with the Euclidian distance measure
more advantageous for the subsequent cluster analysis. and the squared Euclidian distance. Both methods were
Indeed, the KMO increased from 0.591 (miserable) to applied using the original variable values ( percentage

Copyright # 2000 John Wiley & Sons, Ltd. Flavour Fragr. J. 2000; 15: 1±6
4 C. BISCHOF-DEICHNIK, J. HOLTHUIJZEN AND E. STAHL-BISKUP

of 60 variables) as well as the factor values (eight The geraniol/geranyl acetate type (cluster e) is a
factors). Beforehand, the single linkage method was lemon-scented type containing in total 30±85% of the
used and three oils were found to be runaways, compounds geraniol and geranyl acetate. The charac-
consequently they were excluded from the CA. teristic smell is also caused by minor concentrations of
Eight di€erent dendrograms were obtained with neral, geranial and citronellol, also belonging to factor
similar results, revealing four±®ve clusters on a high 3. The 18 plants showing this oil type were distributed
level and 11±13 clusters on a lower level (elbow over six di€erent localities.
criterion). Every dendrogram was analysed by discri- A trans-sabinene hydrate/terpinen-4-ol type (cluster c
minant analysis (DA), and the factors and variables plus 3 individual oils) was found for 15 oils from six
with the highest discriminative power were determined. di€erent localities. Both oil components appeared in
Afterwards, on the basis of these statistical results, a factor 2, which makes a biosynthetic correlation
modi®ed CA method was applied, processing the factor evident. Three further oil types with oxygenated mono-
values plus the 24 variables which had been excluded terpenes as characteristic compounds were found,
from the FA (both z-transformed). The average linkage namely an a-terpineol type (cluster b, four oils, two
method in combination with the Pearson correlation localities), a linalol type (cluster h, six oils, three
gave the most reliable result. The corresponding den- localities) and a linalol/linalyl acetate type (cluster f,
drogram made 12 clusters plausible (clusters a to l) and two oils, two localities). These three oil types had a
formed the basis for the de®nition of the oil types. sweet character, contrasting with the strongly aromatic
However, in order to interpret the results completely trans-sabinene hydrate/terpinen-4-ol and thymol types.
it is also necessary to consider the results of the other A group of oils with oxygenated sesquiterpenes were
CA methods, as well as those of the DA. In this way it widespread and gave good reason to de®ne ®ve distinct
became obvious that cluster a should be separated into oil types. The type-characterizing compounds were
two di€erent oil types (a1 and a2), and that the high trans-nerolidol (cluster g plus three individual oils,
correspondence of the clusters j, k and l made it logical 11 oils, four localities), hedycaryol (cluster a1 , 11 oils,
to unite them to one single oil type. In addition, after one locality), T-cadinol (cluster a2 , ®ve oils, two local-
analysing the clusters it was found that the assignments ities), germacra-1(10),5-dien-4-ol (cluster d, nine oils,
of 10 cases (individual oils) were ambiguous. Such oils seven localities) and germacra-1(10),4-dien-60-ol
were transferred `by hand' to their correct clusters made (cluster i, 11 oils, four localities). The corresponding
plausible by DA. As a ®nal result, 12 di€erent oil types diagrams show that the oxygenated sesquiterpenes
of T. praecox ssp. polytrichus collected in the Tyrolean occurred in the oils in relatively high percentages (mean
Alps became evident. The compositions of these oil values 25±60%).
types are given in Figure 1, showing in diagrams the Finally, the three oils that were mathematically
mean values ( peak area percentages) of components considered as runaways must be discussed. The ®rst
with more than 3%. In addition, Figure 1 gives contained borneol in a high concentration (64%). It
information on the number of oils belonging to an was found to be a runaway because the typical oil
oil type. pattern occurred only once. This does not necessarily
mean that such an oil type does not exist. Together with
another oil with a high borneol content, mathe-
Definition of the Oil Types matically assigned to the thymol type because of other
components of factor 1, they formed a borneol type.
With 46 individual samples the thymol oil type (united The second runaway could be included in the thymol
clusters j, k and l) was dominant within the alpine T. type because of a high thymol content. Mathematically
praecox ssp. polytrichus (one-third of the plants). It was it was treated as a runaway because other components
found at nine of the 16 localities. The oil composition of the oil proved to be highly discriminative (e.g. oct-1-
of this type resembles one of those of T. vulgaris en-3-ol and trans-b-ocimene). The third oil recognized
containing thymol and its biosynthetically related as a runaway could not be integrated in any of the
compounds, carvacrol, p-cymene, g-terpinene, thymyl types. It contained a very unusual combination of
acetate and thymyl methyl ether. Except for the latter components of the factors 1 and 3, which may indicate
compound, they can all be found together with thymol that still a further oil type exists.
within Factor 1. Further important constituents were
borneol and germacrene-D, which occurred in higher
concentrations in some oils and were therefore respon- Conclusion
sible for the splitting of clusters j, k and l. However, it
was not plausible to consider these oils as separate oil The essential oil of Thymus praecox ssp. polytrichus is
types, as became obvious after a more detailed study of not characterized by special main components and
the oil compositions. therefore it does not allow an identi®cation of the

Copyright # 2000 John Wiley & Sons, Ltd. Flavour Fragr. J. 2000; 15: 1±6
 MULTIVARIATE ANALYSIS OF THYMUS PRAECOX ESSENTIAL OIL 5

Figure 1. Composition of the essential oil of Thymus praecox Opiz ssp. polytrichus (Kern. ex Borb.) Ronn. collected in
the Tyrolean Alps showing 12 oil types. The heights of the bars correspond with the mean values of the peak area
percentages (y-axes). The numbers of the x-axes correspond with oil components as follows: 1 ˆ a-pinene,
2 ˆ myrcene, 3 ˆ 1,8-cineole, 4 ˆ g-terpinene, 5 ˆ p-cymene, 6 ˆ trans-sabinene hydrate, 7 ˆ linalol, 8 ˆ linalyl
acetate, 9 ˆ b-caryophyllene, 10 ˆ thymyl methyl ether, 11 ˆ terpinene-4-ol, 12 ˆ cis-b-farnesene, 13 ˆ borneol,
14 ˆ germacrene-D, 15 ˆ a-terpineol, 16 ˆ b-bisabolene, 17 ˆ a-muurolene, 18 ˆ a-farnesene, 19 ˆ g-cadinene,
20 ˆ geranyl acetate, 21 ˆ geraniol, 22 ˆ trans-nerolidol, 23 ˆ germacra-1(10),5-dien-4-ol, 24 ˆ hedycaryol,
25 ˆ T-cadinol, 26 ˆ thymol, 27 ˆ a-cadinol, 28 ˆ germacra-1(10),4-dien-6-ol

Copyright # 2000 John Wiley & Sons, Ltd. Flavour Fragr. J. 2000; 15: 1±6
6 C. BISCHOF-DEICHNIK, J. HOLTHUIJZEN AND E. STAHL-BISKUP

plants on a chemical basis. Most of the oil components References

have been found earlier in the oils of other Thymus
species. On analysing a large number of individual 1. J. Jalas, Bot. J. Linn. Soc., 64, 247 (1971).
2. T. G. Tutin, V. H. Heywood, N. A. Burges, D. M. Moore and
plants, we found the alpine populations to be chemi- D. H. Valentine, Flora Europaea, Vol. III, Cambridge University
cally highly polymorphous, manifesting itself in 12 Press, Cambridge (1972).
di€erent oil types, apparently distributed irregularly all 3. R. Hegnauer, Ber. Schweiz. Bot. Ges., 58, 391 (1948).
4. E. Stahl-Biskup, J. Essent. Oil Res., 3, 61 (1991).
over the 16 localities investigated. Surprisingly, the oil 5. T. Adzet, S. CanÄigueral, N. Gabalda, C. IbanÄez, X. TomaÁs and R.
patterns of all of these types have been detected before Vila, Phytochemistry, 30, 2289 (1991).
in other Thymus species, e.g. in T. vulgaris (thymol, 6. P. MaÂrton®, J. Essent. Oil Res., 4, 173 (1992).
7. S. CanÄigueral, R. Vila, G. Vicario, X. TomaÁs and T. Adzet,
geraniol, linalol, trans-sabinene hydrate, a-terpineol),14 Biochem. Syst. Ecol., 22, 307 (1994).
T. carnosus (borneol),10 and in the North Atlantic 8. R. Vila, B. Freixa, S. CanÄigueral, T. Adzet, X. TomaÁs and J. J.
Thymus species, namely T. praecox ssp. arcticus Molins, Flavour Fragr. J., 10, 379 (1995).
9. F. SaÂez, Biochem. Syst. Ecol., 23, 431 (1995).
(linalol/linalyl acetate),15,16,17 T. serpyllum ssp. tanaen- 10. L. Salgueiro, R. Vila, X. TomaÁs, F. Tomi, S. CanÄigueral, J.
sis, and T. serpyllum ssp. serpyllum (germacra-1(10), Casanova, A. ProencËa da Cunha and T. Adzet, Phytochemistry,
4-dien-6-ol, germacra-1(10),5-dien-4-ol).18 Regarding 38, 391 (1995).
11. F. SaÂez, Phytochemistry, 40, 819 (1995).
the northern thyme populations, the existence of a 12. L. Salgueiro, A. ProencËa da Cunha, X. TomaÁs, S. CanÄigueral, T.
linalol/linalyl acetate type is most remarkable because Adzet and R. Vila, Flavour Fragr. J., 12, 117 (1997).
this type was found to be the most abundant oil type 13. L. Salgueiro, R. Vila, X. TomaÁs, S. CanÄigueral, J. Casanova, A.
ProencËa da Cunha and T. Adzet, Phytochemistry, 45, 1177 (1997).
there. All in all, the chemical similarity with Mediter- 14. R. Granger and J. Passet, Phytochemistry, 12, 1683 (1973).
ranean and North Atlantic Thymus species indicates a 15. E. Stahl, Planta Med., 50, 157 (1984).
mixing of southern and northern elements in the Alps, 16. E. Stahl, Nord. J. Bot., 4, 597 (1984).
17. E. Stahl-Biskup, Planta Med., 52, 36 (1986).
which is also re¯ected in the overlapping morphological 18. E. Stahl-Biskup and I. Laakso, Planta Med., 56, 64 (1990).
characters of T. praecox ssp. polytrichus.

Copyright # 2000 John Wiley & Sons, Ltd. Flavour Fragr. J. 2000; 15: 1±6