Plant Cell Rep (2004) 23:246–250

DOI 10.1007/s00299-004-0811-1

PHYSIOLOGY AND BIOCHEMISTRY

S. W. Kim · S. H. Ban · H. Chung · S. Cho ·

H. J. Chung · P. S. Choi · O. J. Yoo · J. R. Liu

Taxonomic discrimination of flowering plants by multivariate analysis

of Fourier transform infrared spectroscopy data

Received: 17 October 2003 / Revised: 19 April 2004 / Accepted: 20 April 2004 / Published online: 10 July 2004

Springer-Verlag 2004

Abstract Fourier transform infrared spectroscopy (FTIR)
provides biochemical profiles containing overlapping sig-
nals from a majority of the compounds that are present
when whole cells are analyzed. Leaf samples of seven
higher plant species and varieties were subjected to FTIR
to determine whether plants can be discriminated phylo-
genetically on the basis of biochemical profiles. A hier-
archical dendrogram based on principal component anal-
ysis (PCA) of FTIR data showed relationships between
plants that were in agreement with known plant taxono-
my. Genetic programming (GP) analysis determined the
top three to five biomarkers from FTIR data that dis-
criminated plants at each hierarchical level of the den-
drogram. Most biomarkers determined by GP analysis at
each hierarchical level were specific to the carbohydrate
fingerprint region (1,200–800 cm1) of the FTIR spec-
trum. Our results indicate that differences in cell-wall
composition and structure can provide the basis for che-
motaxonomy of flowering plants.
Communicated by J.M. Widholm
S. W. Kim · S. H. Ban · H. J. Chung · J. R. Liu ())
Laboratories of Plant Genomic Services
and Plant Cell Biotechnology,
Korea Research Institute of Bioscience
and Biotechnology (KRIBB),
52 Eoun-dong, Yuseong-gu, Daejeon, 305-333, South Korea
e-mail: jrliu@kribb.re.kr
Fax: +82-42-8604608
H. Chung · S. Cho
Department of Chemistry, Hanyang University,
17 Haengdang-dong, Seongdong-gu, Seoul, 133-791, South Korea
P. S. Choi
Laboratory of Functional Genomics
for Plant Secondary Metabolism (National Research Laboratory),
Eugentech, Inc., 52 Eoun-dong, Yuseong-gu, Daejeon,
305-333, South Korea
O. J. Yoo
Department of Biological Science,
Korea Advanced Institute of Science and Technology,
373-1 Kusong-dong, Yuseong-gu, Daejeon, 305-701, South Korea
Keywords Dendrogram · Genetic programming
analysis · Phylogenetic relationship · Principal component
analysis
Abbreviations FTIR: Fourier transform infrared
spectroscopy · GP: Genetic programming ·
PCA: Principal component analysis · PyMS: Pyrolysis
mass spectrometry
Introduction
Fourier transform infrared spectroscopy (FTIR) is a rapid,
simple, high-resolution analytical method that is based on
the vibrations of functional groups and highly polar bonds
of the components analyzed. Thus, FTIR provides bio-
chemical profiles containing overlapping signals from a
majority of the compounds that are present in the cell
when whole cells are analyzed. The biochemical profiles
of FTIR from whole cell samples are extremely high-
density data sets and, consequently, FTIR data must be
analyzed by means of multivariate analysis when multiple
whole cell samples are compared. Using multivariate
analysis, FTIR has been used for discriminating closely
related microbial strains (Freeman et al. 1994; Goodacre
et al. 1998; Wenning et al. 2002).
FTIR has only been used in plant biology in a limited
number of studies, including those on discriminating be-
tween cell-wall mutant plants (Stewart et al. 1997; Chen
et al. 1998). These studies have demonstrated that FTIR is
robust in identifying structural and architectural alter-
ations in cell walls. However, it remains to be determined
whether the results of this approach are valid represen-
tations of phylogenetic relationships between higher plant
species. The purpose of the investigation reported here
was to determine whether multivariate analysis of FTIR
data from flowering plants can be used to discriminate
plants phylogenetically.

Materials and methods
Plant materials
Leaf tissues of seven species and varieties (Table 1) were subjected
to FTIR analysis. All plants were grown in a greenhouse at ap-
proximately 30/20C(day/night) under a 14/10-h (day/night) pho-
toperiod with light provided at an intensity of 500 mmol m2 s1.
Fully expanded leaves were excised from plants at the flowering
stage and immediately plunged into liquid nitrogen before grinding
with a mortar and pestle. Ground samples were then freeze-dried
and stored at 70C until use.
FTIR spectroscopy
Five milligrams of freeze-dried, powdered leaves was mixed with
80 mg of KBr and the mixture ground to reduce the particle size to
less than 100 mm in diameter. The finely ground mixture was then
compressed in a pellet press in order to form a pellet through which
the beam of the spectrometer was able to pass. Infrared analysis
was performed using a BOMEM FTIR spectrophotometer, Model
DA 3.2, equipped with a liquid nitrogen cooled mercury-cadmium-
telluride detector and a KBr beam-splitter. To improve the signal-
to-noise ratio, we co-added 32 interferograms and averaged these
with the analytical results. Infrared spectra were obtained by sub-
traction of the plate spectra (background) used for deposition of the
samples. Spectral resolution was 4 cm1, and spectra collected over
the wave number ranged from 8,000 cm1 to 400 cm1. Spectra
were processed using the GRAMS/386 program (Galactic Indus-
tries, Salen, N.H.). Samples were run in triplicate using ground leaf
discs from three different individual plants of each kind.
Data normalization and statistical analysis
Procedures were implemented to minimize problems arising from
baseline shifts. Spectra were first subjected to path-length correc-
tion, and then the spectra were baseline corrected so that the
smallest absorbance (2,000 cm1) was equal to 0. The smoothed
second derivatives of these normalized spectra were calculated
using the Savitzky-Golay algorithm (Savitzky and Golay 1964)
with five-point smoothing. These data were then subjected to
multivariate analysis.
A hierarchical dendrogram was constructed from principal
component analysis (PCA) of FTIR data by the unweighted pair
group method with arithmetic mean analysis using the multivariate
statistical package (mvsp 3.13) of Kovach Computing Services.
Genetic programming (GP) analysis was used to determine bio-
marker variables that discriminated plants at each hierarchical level
of the dendrogram. gmax-bio software (Aber Genomic Computing,
Aberystwyth, Wales, UK) was used with the default parameters of a
population size of 1,000; a maximum program length of 44 nodes;
fitness based on tournament selection/Gmax(v); a crossover operator
used 80% of the time, and terminals of the mutations were selected
20% of the time. The operators used were the default numeric (0.1,
1, 3, 5, and rand) and arithmetic (1, 2, /, and *) operators.
Table 1 Plant materials used for FTIR spectroscopy
Order Family Genus Species
Rosales Rosaceae Rosa multiflora
Rosa multiflora
Rosa rugosa
Crassulaceae Sedum kamtschaticum
Gentianales Apocynaceae Catharanthus roseus
Catharanthus roseus
Liliales Liliaceae Lilium longiflorum
a
Identifiers of plant materials for FTIR spectra and PCA plots in Fig. 1.
247
Results
PCA analysis
Quantitative FTIR data for each sample are given in
Fig. 1a. PCA of FTIR data is displayed in a two-dimen-
sional plot using the first two principal components
(Fig. 1b). Three replicate samples of each species and
variety were grouped in discrete clusters, indicating that
PCA is able to discriminate different species and vari-
eties. A hierarchical dendrogram was generated to display
the relationships between plant species based on PCA of
FTIR data (Fig. 1c). The dendrogram divided seven spe-
cies and varieties into two groups. Lilium longiflorum and
the dicotyledonous plants were placed at the top level,
indicating that the monocot species were separate from
the dicotyledonous plants. The dicotyledonous plants
were subsequently divided into two groups. Two Catha-
ranthus roseus varieties and the Rosales plants were
placed at the second level. Sedum kamtschaticum was
subsequently separated from three Rosa plants, which
were further discriminated into Rosa rugosa and Rosa
multiflora varieties at the lowest level. The separation of
species in the dendrogram was in agreement with the
known taxonomy of the plants, indicating that discrimi-
nation of higher plant species and varieties by FTIR re-
flects phylogenetic relationships between plants.
GP analysis
GP analysis revealed the top three to five biomarkers
from FTIR data that contributed most to the discrimina-
tion of plants at each hierarchical level of the dendrogram
(Fig. 2). We determined a total of 15 biomarkers that
included three biomarkers [V116 (1,100 cm1), V76
(1,022 cm1), and V66 (1,003 cm1)] discriminating Lil-
ium from the dicotyledonous plants, three biomarkers
[V125 (1,117 cm1), V124 (1,115 cm1), V16 (906 cm1)]
discriminating Catharanthus roseus cultivars from the
Rosales plants, two biomarkers [V37 (948 cm1), V25
(924 cm1)] discriminating Sedum kamtschaticum from
the Rosa plants, and three biomarkers [V139 (1,143 cm1),
V85 (1,040 cm1), V82 (1,034 cm1) discriminating Rosa
rugosa from R. multiflora varieties (Fig. 2). These 11
biomarkers out of the 15 were deconvoluted to cell-wall
components including cellulose (Chen et al. 1997), pectin
Variety Cultivar Identifera
a
platyphylla b
c
d
Cooler grape e
Cooler peppermint f
Casablanca g
248
Fig. 1a–c FTIR spectra, PCA of FTIR data, and a dendrogram
based on PCA of FTIR data from seven plants. a Representative
FTIR spectra of seven plants, b PCA of FTIR data from seven
(Wilson et al. 2000), polygalacturonic acid, and rhamno-
galacturonan (Kaurkov et al. 2000). One biomarker of
V390 (1,627 cm1), discriminating C. roseus cultivars
from the Rosales plants, and the other biomarker of V380
(1,608 cm1), discriminating R. rugosa from R. multiflora
varieties, are deconvoluted to amides associated to pro-
teins (Nelson 1991; Williams and Fleming 1996).
Discussion
Seven different plants including one variety and two
cultivars were chosen in this study. L. longiflorum was
chosen to determine whether this monocot plant could be
discriminated from the dicot plants by FTIR. One variety
and two cultivars within two different species were cho-
sen to determine whether genotypes within the same
species could be discriminated by FTIR. Plants that be-
plants, c a dendrogram based on PCA of FTIR data from seven
plants. Refer to Table 1 for the identifiers a, b, c, d, e, f, and g. The
numbers next to the identifiers 1, 2, and 3 indicate three replicates
long to two orders and three families were chosen to
determine whether they could be systematically discrim-
inated by FTIR. Therefore, these seven different plants
represent a sample of the flowering plants that can be
discriminated by FTIR.
FTIR is an excellent method for determining phylo-
genetic relationships between flowering plants based on
its ease of use and quick results. This method has been
widely used to discriminate closely related microbial
strains (Freeman et al. 1994; Goodacre et al. 1998;
Wenning et al. 2002). FTIR has been used in plant bi-
ology in a number of studies that include discrimination
of cell-wall mutant plants (Stewart et al. 1997; Chen et
al. 1998), cell-wall composition and architecture (Sn et
al. 1994; McCann et al. 1997), mechanical properties and
molecular dynamics of plant cell-wall polysaccharides
(Wilson et al. 2000), and determination of the fruit
content in processed foods (Wilson et al. 1993). Sn et

Fig. 2 Biomarkers (arrows) from FTIR data identified by GP
analysis. a Three biomarkers [V116 (1,100 cm1), V76 (1,022 cm1),
and V66 (1,003 cm1)] discriminating Lilium from the dicotyle-
donous plants and spectra averaged from the whole data set of L.
longiflorum (broken line) and the dicotyledonous plants (solid line).
b Four biomarkers [V390 (1,627 cm1), V125 (1,117 cm1), V124
(1,115 cm1), and V16 (906 cm1)] discriminating Catharanthus
roseus cultivars from the Rosales plants and spectra averaged from
the whole data set of C. roseus cultivars (broken line) and the
al. (1994) showed differences in the plant cell walls of
five angiosperms by infrared and Raman spectroscopies,
thereby indicating that taxonomic classification was
possible. However, they did not attempt to discriminate
between the different flowering plants using multivariate
analysis of FTIR data for taxonomic classification. Wil-
son et al. (1993) reported that multivariate analysis of
FTIR data for fruit jam enables the discrimination be-
tween differing fruit content.
FTIR detects all compounds, including polymers and
low-molecular weight compounds in whole cell samples,
subsequently providing biochemical profiles of extremely
high-density data sets. PCA, an unsupervised learning
method requiring no prior knowledge of class structure
within the data set, was used to display the natural rela-
249
Rosales plants (solid line). c three biomarkers [V437 (1,720 cm1),
V37 (948 cm1), and V25 (924 cm1)] discriminating Sedum
kamtschaticum from the Rosa plants and spectra averaged from the
whole data set of S. kamtschaticum (broken line) and the Rosa
plants (solid line). d Five biomarkers (V380 (1,608 cm1),
V175(1,213 cm1), V139 [1,143 cm1), V85 (1,040 cm1), and V82
(1,034 cm1)] discriminating Rosa rugosa from R. multiflora va-
rieties and spectra averaged from the whole data set of R. rugosa
(broken line) and R. multiflora varieties (solid line)
tionship between plants in a two-dimensional plot, lead-
ing to construction of a dendrogram that provided hier-
archical levels of plant groupings. GP analysis, a super-
vised learning method for the production of mathematical
rules that enable the easy identification of data selected to
perform the classification, led to the identification of
FTIR biomarkers that discriminated between plants at
each hierarchical level of the dendrogram. Biomarkers
were commonly found in the carbohydrate fingerprint
region where differences in cell-wall composition and
structure are reflected. Thus, differences in cell-wall
composition and structure can provide the basis for
chemotaxonomy of flowering plants. We have previously
reported that pyrolysis mass spectrometry (PyMS) dis-
criminates plants phylogenetically in a similar manner
250
(Kim et al. 2004). However, biomarkers determined by
GP of PyMS data cannot be deconvoluted to identify the
chemical compounds.
Acknowledgements This work was supported by grants to JRL
from the National Research Laboratory Program (M10104000234-
01J000-10710), from the Strategic National R&D Program through
the Genetic Resources and Information Network Center (no.
BDM0100211), from the Korea Science and Engineering Founda-
tion through the Plant Metabolism Research Center of the Kyung
Hee University funded by the Korean Ministry of Science and
Technology, and from the KRIBB Research Initiative Program.
References
Chen L, Wilson RH, McCann MC (1997) Infrared microspec-
troscopy of hydrated biological systems: design and construc-
tion of a new cell with atmospheric control for the study of
plant cell walls. J Microsc 188:62–71
Chen L, Carpita NC, Reiter WD, Wilson RH, Jeffries C, McCann
MC (1998) A rapid method to screen for cell-wall mutants
using discriminant analysis of Fourier transformation infrared
spectra. Plant J 16:385–392
Freeman R, Goodacre R, Sisson PR, Magee JG, Ward AC, Light-
foot NF (1994) Rapid identification of species within the My-
cobacterium tuberculosis complex by artificial neural network
analysis of NMR data. J Med Microbiol 40:170–173
Goodacre R, Timmins M, Burton R, Kaderbhai N, Woodward AM,
Kell DB, Rooney PJ (1998) Rapid identification of urinary tract
infection bacteria using hyperspectral whole-organism finger-
printing and artificial neural networks. Microbiology 144:
1157–1170
Kaurkov M, Capek P, Sasinkov V, Wellner N, Ebringerov A
(2000) FT-IR study of plant cell wall model compounds: pectic
polysaccharides and hemicelluloses. Carbohydr Polym 43:195–
203
Kim SW, Ban SH, Chung HJ, Choi DW, Choi PS, Yoo OJ, Liu JR
(2004) Taxonomic discrimination of higher plants by pyrolysis
mass spectrometry. Plant Cell Rep 22:519–522
McCann MC, Chen L, Roberts K, Kemsley EK, Sn C, Carpita
NC, Stacey NJ, Wilson RH (1997) Infrared microspectroscopy:
sampling heterogeneity in plant cell-wall composition and ar-
chitecture. Physiol Plant 100:729–738
Nelson WH (1991) Modern techniques for rapid microbiological
analysis. VCH, New York
Savitzky A, Golay MJE (1964) Smoothing and differentiation of
data by simplified least squares procedures. Anal Chem 36:
1627–1633
Sn CFB, McCann MC, Wilson RH, Grinter R (1994) Fourier-
transform Raman and Fourier-transform infrared spectroscopy:
an investigation of five higher plant cell walls and their com-
ponents. Plant Physiol 106:1623–1631
Stewart D, Yahiaoui N, McDougall GJ, Myton K, Marque C,
Boudet AM, Haigh J (1997) Fourier-transform infrared and
Raman spectroscopic evidence for the incorporation of cin-
namaldehydes into the lignin of transgenic tobacco (Nicotiana
tabacum L.) plants with reduced expression of cinnamyl alco-
hol dehydrogenase. Planta 201:311–318
Wenning M, Seiler H, Scherer S (2002) Fourier-transform infrared
microspectroscopy, a novel and rapid tool for identification of
yeast. Appl Environ Microbiol 68:4717–4721
Williams DH, Fleming I (1996) Spectroscopic methods in organic
chemistry, 5th edn. McGraw-Hill, London
Wilson RH, Slack PT, Appleton GP, Sun L, Belton PS (1993)
Determination of the fruit content of jam using Fourier trans-
form infrared spectroscopy. Food Chem 47:303–308
Wilson RH, Smith AC, Kacurakova M, Saunders PK, Wellner N,
Waldron KW (2000) The mechanical properties and molecular
dynamics of the plant cell wall polysaccharides studied by
Fourier-transform infrared spectroscopy. Plant Physiol 124:
397–405