Journal of Chromatography A, 1279 (2013) 86–91

Contents lists available at SciVerse ScienceDirect

Journal of Chromatography A
journal homepage: www.elsevier.com/locate/chroma

Determination of disease biomarkers in Eucalyptus by comprehensive

two-dimensional gas chromatography and multivariate data analysis
Leandro Wang Hantao a,b , Helga Gabriela Aleme a,b , Martha Maria Passador c , Edson Luiz Furtado c ,
Fabiana Alves de Lima Ribeiro a,b , Ronei Jesus Poppi a,b , Fabio Augusto a,b,∗
a
Institute of Chemistry – University of Campinas (Unicamp), CP 6154, 13084-970 Campinas, São Paulo, Brazil
b
Instituto Nacional de Ciência e Tecnologia em Bioanalítica (INCTBio), CP 6154, 13084-970 Campinas, São Paulo, Brazil
c
Faculty of Agronomic Sciences – State University Julio de Mesquita Filho (UNESP), Botucatu, São Paulo, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: In this paper is reported the use of the chromatographic profiles of volatiles to determine disease markers
Received 22 November 2012 in plants – in this case, leaves of Eucalyptus globulus contaminated by the necrotroph fungus Teratosphaeria
Received in revised form 3 January 2013 nubilosa. The volatile fraction was isolated by headspace solid phase microextraction (HS-SPME) and
Accepted 4 January 2013
analyzed by comprehensive two-dimensional gas chromatography–fast quadrupole mass spectrome-
Available online 12 January 2013
try (GC × GC–qMS). For the correlation between the metabolic profile described by the chromatograms
and the presence of the infection, unfolded-partial least squares discriminant analysis (U-PLS-DA) with
Keywords:
orthogonal signal correction (OSC) were employed. The proposed method was checked to be independent
Comprehensive two-dimensional gas
chromatography
of factors such as the age of the harvested plants. The manipulation of the mathematical model obtained
Mass spectrometry also resulted in graphic representations similar to real chromatograms, which allowed the tentative iden-
Metabolic profiling tification of more than 40 compounds potentially useful as disease biomarkers for this plant/pathogen
Orthogonal signal correction, Unfolded pair. The proposed methodology can be considered as highly reliable, since the diagnosis is based on the
partial least squares discriminant analysis whole chromatographic profile rather than in the detection of a single analyte.
Solid phase microextraction © 2013 Elsevier B.V. All rights reserved.

1. Introduction diseases in plants and vegetables. Traditionally, direct methods are

Eucalyptus is an important commercial plant, extensively used
for production of cellulose, paper and coal and as source of essential
oils used for the production of aroma and fragrance products. Addi-
tionally, plants naturally have defense mechanisms against most
pathogens [1], however, when infected the extent of their interac-
tion with the parasite can often lead to the death of the plant, as it is
for the case of the interaction between Eucalyptus globulus and the
necrotroph fungus Teratosphaeria nubilosa. During the early stages
of this infection, their leaves are readily affected by the pathogen’s
presence, and, thus, defoliation and reduction of the growth rate
occurs leading to an overall loss in productivity of the industrial
sector. These symptoms can be potentially enhanced when dealing
with inappropriate diagnosis methods.
As a consequence, attention and resources from research and
development areas, in these specific companies, are being shifted
toward finding more efficient assays for the early diagnosis of
∗ Corresponding author at: Instituto de Química – Universidade Estadual de
Campinas (Unicamp), CP 6154, 13084-971 Campinas, São Paulo, Brazil.
Tel.: +55 19 3521 3057; fax: +55 19 3521 3023.
E-mail address: augusto@iqm.unicamp.br (F. Augusto).
used for the detection and identification of the pathogen [2]. The
very first described, and still widely accepted, method for this pur-
pose is the microscopic inspection of the structural morphology of
the fungus, or bacteria. Then, a comparison between the obtained
images is manually performed against an extensive database [2].
However, direct molecular methods have also played an impor-
tant role in confirming the identity of the pathogen either by the
analysis of its genetic material content, often accomplished by the
use of the polymerase chain reaction (PCR), or by specific inter-
actions between antigens and antibodies, such as enzyme-linked
immunosorbent assay (ELISA). However, the major limitation of
these direct methods is the minimum amount of pathogen in the
sampled material required to achieve the method’s detection lim-
its [2]. As reported by Bazzini et al. [3], in their case study, while
the visual symptoms of the infection were evolving very quickly
after just a few days post infection, the systemic accumulation of
pathogen in the plant’s leaves was still incipient. Consequently,
for the purpose of early diagnosis the direct methods are most
likely limited by the time of cultivation of the pathogen mate-
rial prior analysis [2], and, very often, this step can surpass a time
frame of days. Consequently, after just a period of 2 weeks the
amount of infected plants, originally healthy, can reach almost
100%.

0021-9673/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.chroma.2013.01.013
 L.W. Hantao et al. / J. Chromatogr. A 1279 (2013) 86–91
In this context, with the introduction of more modern analytical
platforms the study of metabolites – coined metabolome – is being
extensively used for the diagnosis of diseases in plants. However,
because of their lack of motility the plants have more pronounced
metabolic responses to stress than mammals [4]. This makes the
inspection of the metabolome for the diagnosis of diseases in plants
a particularly interesting and challenging task.
In phytopathology, the volatile and semi-volatile fraction is
frequently assessed and very often terpenes and other classes of
compounds are implicated as potential biomarkers [5]. Hence, the
use of solid phase microextraction (SPME) is an interesting alterna-
tive since it can be performed in most laboratories without the need
for specific instrumentation [6]. The feasibility of using SPME for
these studies arises from the combination of sampling and sample
preparation into a simple and unique step [6]; and under relatively
mild conditions the isolation of the biogenic volatile organic com-
pounds (BVOC) can be performed from the matrix with some degree
of analyte pre-concentration [7,8].
Despite the inherent advantages of SPME for BVOC sampling,
there are essentially two issues that must be met for proper
metabolic profiling. Firstly, due to the amount of sorbent mate-
rial available in the SPME fiber the maximum extracted amount
of analytes will be significantly less, when compared to stir-
bar sorptive extraction (SBSE) or needle trap devices (NTD), for
instance. Secondly, the structural and functional diversity of the
volatiles present in the BVOC ranges from aliphatic alcohols,
aldehydes, methylketones, acids, lactones, esters, terpenoids, fura-
nones and pyrones [5]. Consequently, some degree of coelution
has to be expected. Hence, comprehensive two-dimensional gas
chromatography–mass spectrometry (GC × GC–MS) is probably the
ideal companion for SPME because it presents higher detectability
– achieved by band compression in the modulator – and sensitivity
offered by the MS. Also, GC × GC–MS provides at least two tunable
dimensions for separation [9]. Thus, GC × GC has being recognized
as a potential tool for metabolomics.
The main advantage of using the metabolic profile obtained
by SPME and GC × GC–MS is the possibility to diagnose the infec-
tion during the initial stage of the disease. However, many other
sources that can cause stress to the plant will also have an impact
on the measured chromatographic profile. This fact makes visual
distinction/identification of biomarkers an extremely difficult and
time-consuming task. In this sense, many partial least squares (PLS)
based approaches have already been reported for GC × GC data
[10–13]. Briefly, the use of partial least squares for discriminant
analysis (PLS-DA) is an interesting alternative for the differen-
tiation of classes based on the metabolic profile, because it is
able to discriminate the within group variation (12 ) from the
between group variation (22 ) [14]. Hence, if classical principal
component analysis (PCA) was used for the pattern recognition
of metabolic profiles, where 12 > 22 it would potentially fail to
detect the difference between these classes, while the PLS-DA
approach could be employed successfully [14]. An example of
PLS-DA analysis of GC × GC–TOFMS (TOFMS – mass spectrome-
try with a time-of-flight mass analyzer) data for metabolomic
purposes was successfully carried out by Gröger and Zimmer-
mann [11]. However, pre-processing techniques such as orthogonal
signal correction (OSC) can also be used for the same purpose,
thus it often provides an optimal platform for PLS modeling
allowing for better predictive ability and interpretation [15].
Remarkably, Li et al. [12] reported the use of OSC-PLS-DA for the
analysis of GC × GC–TOFMS data to distinguish between healthy
and diabetic patients; and the implication of potentially new
biomarkers.
In this context, the combination of SPME with GC × GC–qMS
(qMS – mass spectrometry with rapid quadrupolar mass analyzer)
for the metabolic profiling of plants is proposed. So, in the following
87
sections OSC processing prior to unfolded partial least squares-
discriminant analysis (U-PLS-DA) will be described as well as some
non-conventional strategies for data interpretation. In addition, a
careful inspection and interpretation of biomarkers, as well as their
tentative identification, will be presented.
2. Theory
The correlation between the chromatograms and presence of
fungal infection on the plants was determined through U-PLS-DA.
PLS is a very useful tool to model dependent variables (denoted
by Y) as a function of independent responses (denominated X). In
this case, a two-way data set X (I × J), where I corresponds to the
number of chromatograms and J is the number of scans in the chro-
matogram, is decomposed into a score vector t related to the first
order of the data and in a weight vector w assigned to the sec-
ond order of the data, as seen in Eqs. (1) and (2). These vectors
are defined to have the maximum covariance with the dependent
variable Y, Eq. (3):

z
X= tz wzT + E (1)
Z=1

z
Y= uz qTz + E (2)
Z=1
⎡  ⎛ ⎞⎤

 I  J
max ⎣cov(tz , y) min ⎝ (xij − tz,i wz,j )⎠ ⎦ (3)
wz
 i=1 j=1
where tz and uz are scores vector, wz and qz are loading vectors,
z is the number of latent variables (LV) and E is the residual matrix
[16]. In addition, DA-based methods correlate the input data to
binary or other discrete response sets – i.e., the samples are discrim-
inated between two groups separated by a discriminant hyperplane
in the X-space.
However, very often much information present in the chro-
matograms is not correlated to the property of interest (Y), because
several sources contribute to variation present in the experimen-
tal measurements. Hence, OSC can be used for preprocessing of
the chromatographic data in order to eliminate this non-correlated
variation from the descriptor variables (X). Briefly, the OSC algo-
rithm uses a PCA/PLS-related solution that removes only so much
of X as is unrelated (orthogonal) to Y, a more comprehensive dis-
cussion was reported by Wold et al. [17].
3. Materials and methods
Samples. All samples were collected at a commercial forest (Pin-
heiro Machado, RS, Brazil). Leaves from E. globulus, both healthy
and infected with the fungal pathogen T. nubilosa and from plants
with varied ages, were harvested. The detection of the infection, in
the E. globulus plants, was performed by the microscopic inspec-
tion of Mycosphaerella leaf disease’s symptoms [18,19]. Also, the
identification, and confirmation, of the pathogen was performed
by analyzing its genetic material. The samples were immediately
frozen with liquid nitrogen (LN2 ) after removal from the plant, to
cease all physiological processes, and kept under refrigeration (dry
ice, −78 ◦ C) until analysis.
Reagents and materials. For the determination of the 1 D lin-
ear temperature programmed retention indices (LTPRI), a mix of
C7–C20 n-alkanes was used (Sigma–Aldrich – St. Louis, MO, USA).
For the HS-SPME procedure, glass v-vials (Wheaton Science Prod-
ucts – Millville, NJ, USA), magnetic stirrers and appropriate screw
caps with PTFE/silicon septa (Sigma–Aldrich) were employed, as
88 L.W. Hantao et al. / J. Chromatogr. A 1279 (2013) 86–91
well as SPME fibers coated with 100 ␮m poly(dimethylsiloxane)
(PDMS) (Sigma–Aldrich).
HS-SPME procedure. Prior to the isolation of the volatiles, ca. 4–5
frozen leaves were ground in a sterile mortar in the presence of
LN2 and kept on sealed flasks until it reached ambient temperature.
Aliquots of (300 ± 5) mg of ground leaves were weighed directly in
the v-vials and 2 mL of 18.5% (m/v) aqueous sodium chloride were
added. A 100 ␮m PDMS SPME fiber was exposed to the headspace
of the suspension for 30 min at 41.5 ◦ C, under magnetic stirring
(800 rpm). The extracted analytes were immediately desorbed, sep-
arated and detected using GC × GC–qMS.
GC × GC. All analyses were performed on a lab-made
GC × GC–qMS prototype based on a QP2010+ GC (Shimadzu
Corp., Tokyo, Japan) fitted with a split/splitless injector and a
miniaturized sealed two-stage cryogenic modulator, previously
described [20]. The modulator was controlled by a low cost 8-bit
Duemilanove microcontroller board (Arduino, Ivrea, Italy) [21].
The column set consisted of a 30 m × 0.25 mm × 0.25 ␮m HP-5MS
(Agilent Technologies – Palo Alto, CA, USA) column fitted to a
100 cm × 0.10 mm × 0.10 ␮m SulpecoWax 10 (Sigma–Aldrich)
with a SilTite mini-union (SGE Inc., Austin, TX, USA). The injection
port and transfer line were kept at 250 ◦ C and H2 at 0.6 mL min−1
was used as carrier gas. For all runs, the oven temperature pro-
gramming was set from 60 ◦ C to 175 ◦ C at 3 ◦ C min−1 followed by
a secondary ramp of 20 ◦ C min−1 to 210 ◦ C. The modulation period
for all analyses was 6.0 s. The MS ionization source was kept at
200 ◦ C, and the mass scan range was set from 40 to 383 m/z units to
achieve a data collection rate of 25 spectra s−1 . Peak identification
was performed by matching against NIST 2010 (NIST, Gaithersburg
– MD, USA) and FFNSC (Chromaleont, Messina, Italy) spectra
libraries, combined with LTPRI inspection. Whenever available,
analytical standards were employed for the confirmation of the
analytes’ identities. Complementary data analysis and generation
of chromatograms was performed using GCImage software (Zoex
Corp., Houston, TX, USA).
Multivariate data analysis. The unfolded total ion chro-
matograms (TIC) were used in order to find mathematical models
correlating the profiles with the infection by T. nubilosa. All data
analysis was performed using routines programmed on the plat-
form Matlab (MathWorks, Natick, MA, USA) using the 6.2 PLS
Toolbox (Eigenvector Research, Wenatchee, WA, USA). Prior to
data processing, the chromatograms were aligned using correlation
optimized warping algorithm (COW) [22], as previously reported
[23], and filtered using OSC [17].
The chromatograms were split into two subsets: a calibration
set consisting of the chromatograms from 24 samples (Xcal ) and an
external validation set with 16 samples (Xval ). For both subsets, half
of the chromatograms corresponded to leaves from young plants.
In addition, a value of zero (0) was attributed, as response, for the
chromatograms that originated from healthy plants; while one (1)
was assigned to samples from infected plants. The U-PLS model
with leave-one-out cross-validation was performed [24] and, sub-
sequently, the classification was then based on a Bayesian approach
using the scores obtained from PLS.
Biological assay. The spores of T. nubilosa were grown in steril-
ized Petri dishes with agar as nutrient. Three sets of experiments
were performed: a control group, and two other groups with
growth in the presence of analytical standards #1 and #2. The
first evaluated solution (#1) was pure ␣-pinene, while the sec-
ond solution (#2) evaluated consisted of 12.5% (w/w) carvone,
33.7% (w/w) ␣-terpineol, and 53.8% (w/w) geraniol. During the
preparation of the solutions no organic solvents were used. Also,
the liquid standards were kept isolated from the growth medium,
but the volatile fraction was allowed to equilibrate through the
entire system. To evaluate the effect of the volatiles, the germina-
tion of the spores was assessed. The experiment was performed
Fig. 1. Illustration of a GC × GC–qMS chromatogram of leaf samples collected from
healthy E. globulus plants. The sampling and BVOC isolation was performed through
the static headspace by a 100 ␮m PDMS coated SPME fiber.
in duplicates and the temperature and photoperiod was carefully
controlled.
4. Results and discussion
During optimization of the chromatographic conditions, the
method was optimized to increase the detectability of the trace
analytes, while jeopardizing the peak symmetry for a few major-
ity analytes. Hence, some tailing in the 1 D and 2 D dimensions was
observed for these majoritary analytes. However, these regions of
the chromatograms were not removed prior to multivariate data
analysis, because preliminary results showed that the tailing did
not influence negatively the analytical results. Furthermore, the
modulation period was set to 6 s for it allowed sufficient time for
the separation in the 2 D and exhibited an average modulation ratio
of 4. In Fig. 1 is illustrated the chromatogram of the metabolic
profile obtained by the combination of SPME and GC × GC–qMS
of leaf samples collected from healthy E. globulus plants. At first
sight, it can be readily expected that an attempt for the identifi-
cation of potential biomarkers through visual comparison would
be a very time-consuming task. Therefore, the development of an
OSC-U-PLS-DA approach was carried out.
Prior to multivariate data analysis, the chromatograms were
inspected for retention time shifts. It was found no retention time
shifts in the first dimension, but occasionally shifts in the second
dimension were detected. Whenever these shifts, in the second
dimension, were greater than 100 ms, the corresponding sample
was re-injected and analyzed. For the remaining samples, where
retention time shifts were detected and were smaller than 100 ms,
the two-dimensional arrays were unfolded into one-dimensional
arrays. Hence, COW was used to correct these shifts in the data
vector. The segment length and slack parameter were empirically
optimized. The former was limited to a value greater than the reten-
tion time shift observed, but smaller than value of the modulation
period. The value of the segment slack was defined as the average
retention time shift observed in the some chromatograms.
In order to determine the number of LV to be used for the
construction of the calibration model, a “leave-one-out” cross-
validation was performed. The best model was selected based on
minimum values for LV and root mean squared error of cross-
validation (RMSECV). The best model was obtained with 2 LV, which
explained 90% of the Y-axis variance and 72% of the X-axis variance.
Leverage and Q residuals analysis did not reveal the presence of
outlier samples [25]. This last observation shows that even for very
complex samples such as BVOC from plants – where the effect of
abiotic stress, and other variables, has a significant effect upon its
metabolic profile – the proposed tool for the diagnosis of T. nubilosa
 L.W. Hantao et al. / J. Chromatogr. A 1279 (2013) 86–91 89

Fig. 2. Illustration of the (A) scores graph for the first and second latent variables; (B) is the predicted response for the samples in the calibration (䊉) and external validation
() data set. The calculated threshold value was of 0.46. A prediction efficiency of 100% was achieved.

in E. globulus is robust to most non-relevant variations, including
age.
Fig. 2(A) presents the score plot from U-PLS-DA modeling. It
can be readily noticed that the distinction of samples from healthy
E. globulus plants from diseased ones can be performed with the
information contained in the first LV. In a quantitative sense, a
threshold value was determined for proper distinction between
samples taken from healthy and infected plants, at a confidence
level of 95%. In this sense, in Fig. 2(B) is illustrated a classification
plot with the predicted values for the E. globulus samples. Through
the inspection of the calibration set (䊉) and prior knowledge of the
samples, it can be observed that a 100% prediction was achieved, as
no sample was misclassified. However, to effectively test the model
an external validation set of samples was used. The evaluation of
the validation samples () showed that, also, no misclassification
was performed, thus, a prediction efficiency of 100% was obtained.
Although ideally the fitted prediction values should be zero for
healthy samples and one for infected samples, due to presence of
non-relevant information a small deviation is to be expected. The
responses provided by the U-PLS-DA model should be compared
to the 95% confidence critical cutoff value provided by the algo-
rithm. The threshold was determined, by the U-PLS-DA algorithm,
as being equal to 0.46. Hence, the key for the classification by U-PLS-
DA is that the predicted value has to be below this 0.46 threshold for
healthy samples, and above this same value for samples taken from
diseased plants. Consequently, it can be seen that for all samples in
the calibration and validation set the diagnosis results were consis-
tent. Thus, the proposed combination of SPME and GC × GC–qMS
with multivariate analysis can be used as a potential tool for the
accurate diagnosis of diseases in plants.
Additionally, because the information responsible for the differ-
entiation between healthy and diseased samples from E. globulus
was contained in the first LV; a non-conventional method was con-
ceived for the analysis of the loadings from this LV. It essentially
consisted of a simple routine that sorts the variables from the load-
ing by its value. So, variables with positive values were displaced
into a vector l1 while the modules of the negative values were
placed into a second vector l2 . However, during the construction
of these vectors (l1 and l2 ), whenever the conditional statement
was violated, a value of zero would be placed in the correspond-
ing variable in this new vector. Then, each vector was exported in
ASCII format and the contour plot was constructed using GCImage
software, as illustrated in Fig. 3.
Remarkably, the proposed visualization of the loadings graph
allows an easier interpretation of their information because it sorts
the contributions of both classes (healthy and infected samples)
and it resembles a chromatogram – which in turn makes peak
assignment a readily accomplishable task. A careful inspection of

Fig. 3. Loadings graph of the first LV. In (A) are illustrated the analytes present in higher intensities in samples from healthy E. globulus; in (B) are the analytes in higher
intensities in the samples from E. globulus infected with T. nubilosa. The tentative identification is illustrated in Table 1.
90 L.W. Hantao et al. / J. Chromatogr. A 1279 (2013) 86–91

Table 1
Tentative identification of the analytes found in the loadings graph. The peak assignment is illustrated in Fig. 3.

# Analyte identity M.F. LTPRI Similarity

Exp. Lit.

1 2,2-Dimethyl-valeraldehyde C7 H14 O 821 821 91

2 (2E)-2-hexenal C6 H10 O 845 850 85
3 2,4-Hexadien-1-ol C6 H10 O 874 876 88
4 ␣-Pinenea C10 H16 953 948 97
5 ␤-Terpinenea C10 H16 986 993 89
6 ␤-Myrcene C10 H16 996 991 96
7 Careen C10 H16 1012 1009 91
8 Eucalyptola C10 H18 O 1036 1032 95
9 5-Ethylidene-1-methyl-cycloheptene C10 H16 1098 1095 87
10 Solusterol C10 H20 O2 1113 1109 92
11 Perillena C10 H14 O 1122 1125 82
12 Trans-3-caren-2-ol C10 H16 O 1137 1136 85
13 Camphene hydrate C10 H18 O 1156 1156 84
14 ␦-Terpineola C10 H18 O 1175 1170 92
15 ␣-Terpineola C10 H18 O 1199 1195 94
16 (2E)-2-butenoic acid 2-methyl-3-methylbutyl ester C10 H18 O2 1198 1196 81
17 Myrtenal C10 H14 O 1203 1197 85
18 Carveol C10 H16 O 1228 1223 87
19 3(Z)-hexenyl-isovalerate C11 H20 O2 1238 1235 83
20 Carvonea C10 H14 O 1250 1246 91
21 Geraniola C10 H18 O 1262 1255 81
22 Geranial C10 H16 O 1276 1268 81
23 1,7,7-Trimethylbicyclo[2.2.1]hept-2-yl acetate C12 H20 O2 1289 1277 89
24 Geranyl formate C11 H18 O2 1306 1300 86
25 ␦-Terpinyl acetate C12 H20 O2 1320 1314 88
26 Epoxy-␣-terpenyl acetate C12 H20 O3 1346 1346 83
27 ␣-Terpinyl acetate C12 H20 O2 1354 1349 93
28 ␣-Copaene C15 H2 4 1379 1375 93
29 Geranyl acetate C12 H20 O2 1387 1380 92
30 ␤-Bourbonene C15 H24 1389 1382 92
31 Benzyl-isovalerate C12 H16 O2 1399 1399 86
32 Aromadendrene C15 H24 1401 1396 90
33 ␣-Gurjunene C15 H24 1413 1406 94
34 Aromadendrene C15 H24 1438 1435 91
35 ␥-Muurolene C15 H24 1448 1449 88
36 ␣-Himachalene C15 H24 1460 1454 88
37 ␣-Humulene C15 H24 1468 1461 89
38 Valencene C15 H24 1478 1474 90
39 ␣-Bulnesene C15 H24 1493 1490 91
40 Phenethyl isovalerate C13 H18 O2 1496 1493 88
41 Germacrene C15 H24 1518 1515 91
42 1,2,3,4,4a,7-Hexahydro-1,6-dimethyl-4-(1-methylethyl)-, (1r,4s,4ar)-naphthalene C15 H24 1537 1536 84
43 ␣-Cadinene C15 H24 1542 1538 81
44 ␣-Calacorene C15 H20 1547 1547 85
45 Nerolidol C15 H26 O 1566 1564 83
46 Ledol C15 H26 O 1573 1574 86
47 Viridiflorol C15 H26 O 1600 1594 82
48 ␤-Eudesmol C15 H26 O 1658 1656 85

M.F., molecular formula; LTPRI, 1 D linear temperature programmed retention indices determined experimentally (exp.) and found in the MS databases (lit.).
a
Analytes whose identity were confirmed with analytical standards.

the peak regions which were implicated, by the model, as potential
biomarkers showed that the negative values in the original loading
vector corresponded to the peaks that were in higher intensities
in the healthy samples [Fig. 3(A)], while the positive values cor-
responded to peaks that were in higher intensities in the infected
samples [Fig. 3(B)]. At first sight, no immediate hypothesis could
be inferred without further knowledge of the system. However, it
is known that E. globulus is susceptible to infection by T. nubilosa.
Hence, the infection of E. globulus plants by T. nubilosa leads
to the higher production of the semi-volatile compounds, such
as sesquiterpenes, when compared to more volatile species, as
terpenes. The last observation could probably be supported by the
findings reported by Bollina et al. [26] where flavonoids were also
implicated as biomarkers; and it is known that heavier terpenoids
play an important role in the biological synthesis of flavonoids
[27] and in the defense mechanism [28] of plants. Lastly, most
analytes were tentatively identified by the combination of the
mass spectral similarity searches and LTPRI comparison. As a
result, over 40 biomarkers were accurately determined by the
proposed model and identified. The tentative identification of the
analytes, indicated in Fig. 3(A) and (B), is listed in Table 1.
In addition, to evaluate the role of theses analytes in the defense
mechanism of E. globulus, two standard solutions were used in the
in vitro assays with spores of the pathogen, T. nubilosa. The first
solution (#1) consisted of pure ␣-pinene, related to both healthy
and infected plants (Fig. 3); while the second solution consisted of
namely carvone, ␣-terpineol and geraniol, related to the infected
plants. To evaluate the biological activity of these volatiles, the ger-
mination of the spores was analyzed. In Fig. 4(A) is illustrated the
germination of the spores in the absence of the analytes in the
system’s headspace. In Fig. 4(B) is illustrated the inhibition of the
germination of the pathogen’s spores when the headspace was sat-
urated with analytes from standard solution #2. In addition, in vitro
assay with standard solution #1 also inhibited the germination
 L.W. Hantao et al. / J. Chromatogr. A 1279 (2013) 86–91
Fig. 4. Germination images obtained from the in vitro assays with the pathogen’s spores, T. nubilosa. In (A) is illustrated the growth of the spores with no analytes in the
system’s headspace. In (B) is illustrated the inhibition of the germination of the spores when the headspace was saturated with analytes, from standard solution #2.
of the spores. Consequently, the in vitro assays showed that both
standard solutions (#1 and 2#) inhibited in 100% the germination
of the spores from T. nubilosa. Hence, it was confirmed that most
likely some terpenes and sesquiterpenes, indicated by the loadings
graph, are indeed related to the defense mechanism of E. globulus.
5. Conclusions
The combination of SPME and GC × GC–qMS was successfully
used for the assessment of BVOC produced by plants. Additionally,
this combination was successfully employed as a tool for the diag-
nosis of T. nubilosa infection in E. globulus. However, the proposed
methodology can be extended to most pathosystems available.
Also, once the OSC-U-PLS-DA model is established, the diagnosis
of the disease can be performed within hours after sample collec-
tion. In addition, this tool for diagnosis was proven to be robust
against difference in age from the harvested plant material. Even
more, the inspection of the loadings graph allowed distinguishing
between BVOC produced in higher amounts by healthy plant from
infected plants. Thus, the proposed model allowed for the tentative
identification of over 40 potential biomarkers and a further under-
standing of defense mechanism of E. globulus during infection by T.
nubilosa.
Acknowledgments
This work was funded by FAPESP (Foundation for Research Sup-
port of the State of São Paulo) and CNPq (Brazilian National Council
for Research and Technological Development) through the National
Institutes of Science and Technology Program (INCT). Hantao grate-
fully thanks FAPESP for a doctoral grant. Aleme and Ribeiro thank
CAPES (Brazilian Ministry of Education Agency for Improvement of
Graduate Personnel) and Passador thanks CNPq for the postdoctoral
grants.
91
References
[1] G.N. Agrios, Plant Pathology, 5th ed., Academic Press, San Diego, 2005.
[2] R. Burns, Plant Pathology: Techniques and Protocols, 1st ed., Humana Press,
New York, 2008.
[3] A.A. Bazzini, C.A. Manacorda, T. Tohge, G. Conti, M.C. Rodriguez, A. Nunes-Nesi,
S. Villanueva, A.R. Fernie, F. Carrari, S. Asurmedi, PLoS One 6 (2011) 1.
[4] A.R. Fernie, Phytochemistry 68 (2007) 2861.
[5] W. Schwab, R. Davidovich-Rikanati, E. Lewinsohn, Plant J. 54 (2008) 712.
[6] F. Augusto, A.L.P. Valente, Trends Anal. Chem. 21 (2002) 428.
[7] F. Augusto, A. Leite e Lopes, C.A. Zini, Trends Anal. Chem. 22 (2003) 160.
[8] C.A. Zini, F. Augusto, E. Christensen, E.B. Camarão, J. Pawliszyn, J. Agric. Food
Chem. 50 (2002) 7199.
[9] J.B. Phillips, J. Beens, J. Chromatogr. A 856 (1999) 331.
[10] L.A.F. de Godoy, M.P. Pedroso, L.W. Hantao, R.J. Poppi, F. Augusto, Talanta 83
(2011) 1302.
[11] T. Gröger, R. Zimmermann, Talanta 83 (2011) 1289.
[12] X. Li, Z. Xu, X. Lu, X. Yang, P. Yin, H. Kong, Y. Yu, G. Xu, Anal. Chim. Acta 633
(2009) 257.
[13] K.J. Johnson, B.J. Prazen, D.C. Young, R.E. Synovec, J. Sep. Sci. 27 (2004) 410.
[14] A.J. Hobro, J. Kuligowski, M. Döll, B. Lendl, Anal. Bioanal. Chem. 398 (2010) 2713.
[15] S. Wold, J. Trygg, A. Berglund, H. Antti, Chemom. Intell. Lab. Syst. 58 (2011) 131.
[16] R.G. Brereton, Analyst 125 (2000) 2125.
[17] S. Wold, H. Antti, F. Lindgren, J. Öhman, Chemom. Intell. Lab. Syst. 16 (2002)
283.
[18] G. Pérez, G.C. Hunter, B. Slippers, C. Pérez, B.D. Wingfield, M.J. Wingfield, Eur.
Plant Pathol. 125 (2009) 109.
[19] P.W. Crous, U. Braun, J.Z. Groenwald, Stud. Mycol. 58 (2007) 1.
[20] M.P. Pedroso, E.C. Ferreira, L.W. Hantao, S. Bogusz Jr., F. Augusto, J. Sep. Sci. 34
(2011) 1547.
[21] S. Bogusz Jr., L.W. Hantao, S.C.G.N. Braga, V.C.R.M. França, M.F. Costa, R.D.
Hamer, F. Augusto, J. Sep. Sci. 35 (2012) 2438.
[22] G. Tomasi, F. van den Berg, C. Andersson, J. Chemom. 18 (2004) 231.
[23] L.A.F. de Godoy, L.W. Hantao, M.P. Pedroso, R.J. Poppi, F. Augusto, Anal. Chim.
Acta 699 (2011) 120.
[24] M. Barker, W. Rayens, J. Chemom. 17 (2003) 166.
[25] D.L. Massart, B.G.M. Vandeginste, S.N. Deming, Y. Michotte, L. Kaufman, Chemo-
metrics: A Textbook, 1st ed., Elsevier, Amsterdam, 1988.
[26] V. Bollina, G.K. Kumaraswamy, C.K. Ajjamada, T.M. Choo, Y. Dion, S. Rioux, D.
Faulbert, H. Hamzehzarghani, Mol. Plant Pathol. 11 (2010) 769.
[27] E. Breitmaier, Terpenes: Flavors, Fragrances, Pharmaca, Pheromones, 1st ed.,
Wiley-VHC, Weinheim, 2006.
[28] K.A. Leiss, Y.H. Choi, R. Verpoorte, P.G.L. Klinkhamer, Phytochem. Rev. 10 (2011)
205.