Food Chemistry 148 (2014) 7–17

Contents lists available at ScienceDirect

Food Chemistry
journal homepage: www.elsevier.com/locate/foodchem

Analytical Methods

Assessment of fruit juice authenticity using UPLC–QToF MS:

A metabolomics approach
Z. Jandrić a,⇑, D. Roberts b, M.N. Rathor a, A. Abrahim a, M. Islam a, A. Cannavan a
a
Food and Environmental Protection Laboratory, FAO/IAEA Agriculture and Biotechnology Laboratories, Joint FAO/IAEA Division of Nuclear Techniques in Food and
Agriculture, International Atomic Energy Agency, Vienna, Austria
b
Waters Corporation, Atlas Park, Simonsway, Manchester M22 5PP, United Kingdom

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

Article history: In recent years, with the growing complexity of global food supply chains and trade, food fraud, including
Received 1 April 2013 adulteration of high value foods with cheaper substitutes, has become an increasingly important issue. A
Received in revised form 13 September metabolomics approach can be applied to discover biomarkers that can be used to trace food adultera-
2013
tion. A study was undertaken to discover novel, potential biomarkers for the rapid detection of the adul-
Accepted 1 October 2013
Available online 11 October 2013
teration of fruit juices with cheaper alternatives. Pineapple, orange, grapefruit, apple, clementine, and
pomelo were investigated. Untargeted metabolite fingerprinting was performed by UPLC–QToF MS with
multivariate data analysis. Twenty-one differential metabolites were selected, contributing to the sepa-
Keywords:
Fruit juices
ration between pineapple, orange and grapefruit juices, and their admixtures down to 1% adulteration
Adulteration level. A targeted metabolomics method was then optimised and adulteration could be detected at 1%.
Metabolomics The results demonstrate that metabolomics has potential as a screening tool for the rapid detection of
Fingerprinting food adulteration.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction $37 million in the USA (http://www.gpo.gov/fdsys/pkg/GAORE-

PORTS-RCED-96-18/pdf/GAOREPORTS-RCED-96-18.pdf). Food
Adulteration of food and beverages is a growing problem in fraud has often been considered to be mainly an economic issue
today’s global market. One highly prized food commodity popular and less a concern of the traditional food safety or food protection
as a target for adulteration and fraud is fruit juice. Fruit juices intervention and response infrastructure (Moore et al., 2012).
(orange and apple juice) where in the top 7 foods reported from However, the high nutritional value of fruit juices is usually low-
1980 to 2010 as the most common targets for adulteration (Moore, ered with adulteration and some substituted ingredients can cause
Spink, & Lipp, 2012). Fruit juices, fruit juice drinks, and fruit nectars allergic reactions in consumers, or may be toxic.
are consumed worldwide, and have become very popular due to According to European legislation, orange juice consists of the
their many reported health benefits (essential vitamin and mineral pure juice of sweet oranges (Citrus sinensis) while pineapple juice
content, aid in the prevention of cancer, aid digestion, anti- is considered as pure juice obtained from Ananas comosus (Euro-
inflammatory properties, increase bone strength, etc.). pean Commission, 2009). The Codex Alimentarius gives a defini-
The most common fruit juice adulteration practices are dilution tion of orange juice and concentrated orange juice which is
with water and addition of sugars, pulp wash, or other fruit juices obtained from C. sinensis and may contain up to 10% of mandarin
(Ashurst, 2005; Muntean, 2010; Simpkins & Harrison, 1995). These juice (Citrus reticulata) (Codex Alimentarius, 1992). The Food and
procedures are applied alone, or in combination in order to make Drug Administration (FDA) in the US also permits the addition of
the detection of adulteration more difficult. Estimated rates of mandarin and/or tangerine juice up to 10%, and up to 5% juice from
adulteration for orange juice can range from as low as 1% for juice the sour orange (Citrus aurantium) to frozen concentrated orange
sold in the retail market to as high as 20% for juice sold to institu- juice (Ooghe, 1999). The market price of orange juice fluctuates,
tions, such as schools (http://www.gpo.gov/fdsys/pkg/GAORE and when it is low in relation to other juices, orange juice can be
PORTS-RCED-96-18/pdf/GAOREPORTS-RCED-96-18.pdf). In recent also used as an adulterant. Therefore it is necessary to be able to
years, estimates of the magnitude of the fraud associated with detect the addition of orange to, for example, grapefruit juice
the individual cases prosecuted range from about $2 million to (Rouseff, 1988).
A number of analytical methods have been used to identify dif-
⇑ Corresponding author. Tel.: +43 (1) 2600 28373; fax: +43 (1) 2600 28222.
ferent types of adulteration, from simple techniques (soluble sol-
E-mail address: [email protected] (Z. Jandrić).
ids, colour score, suspended pulp, etc.) (Redd, Hendrix, &

0308-8146/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.foodchem.2013.10.014
8 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17

Hendrix, 1992) to more complex techniques based on profile anal- a hybrid triple quadrupole/linear ion trap (QqQ/LIT) mass analyser
ysis of sugars, organic acids or flavonoids, as well as analysis of and marker characterisation using quadrupole–quadrupole-time-
minerals, trace metals, and isotopes using techniques such as high of-flight mass spectrometry (QqTOF MS) (Vaclavik, Schreiber, Laci-
performance liquid chromatography (HPLC) or gas chromatogra- na, Cajka, & Hajslova, 2011). The paper demonstrated the potential
phy (GC) coupled to various types of detectors (Ehling & Cole, of this type of approach for the detection of adulteration. However,
2011; Gómez-Ariza, Villegas-Portero, & Bernal-Daza, 2005; Munte- the method failed to detect some fruit juice adulteration at levels
an, 2010), capillary electrophoresis (Saavedra, Rupérez, & Barbas, lower than 15% (addition of apple and grapefruit to orange juice)
2001), spectroscopy (Cuny et al., 2008), inductively coupled plasma and 10% (the presence of orange juice in grapefruit and apple
mass spectrometry (Schwartz & Hecking, 1991), neutron activation juice), or to confirm the identity of the markers. Only 55% of the
analysis (Anderson, Cunningham, & Alvarez, 1992), and isotope ra- markers used were tentatively identified and none of the markers
tio mass spectrometry (Rossmann, 2001). The total cost of analysis was confirmed using a reference compound.
for fruit juice adulteration can be very expensive, from $15 for a The feasibility of applying ultra-performance liquid chromatog-
basic test to identify dilution with water, to $800 for a test to iden- raphy–quadrupole time of flight mass spectrometry (UPLC–QToF
tify the presence of pure beet sugar (http://www.gpo.gov/fdsys/ MS) coupled with multivariate statistical analysis for the assess-
pkg/GAOREPORTS-RCED-96-18/pdf/GAOREPORTS-RCED-96- ment of the authenticity of fruit juices was explored in the present
18.pdf). The most frequently used methods to detect one of the study. The aim of the study was to identify potential robust bio-
most difficult types of adulteration to control, addition of inexpen- markers to detect fruit juice adulteration using an untargeted met-
sive juices to authentic, higher value fruit juices, are based on pro- abolomics approach and to develop a generally reliable, rapid,
filing and/or targeted analysis of a number of compounds that may sensitive, and confirmatory analytical method for adulteration
be from one chemical class or from different classes. Disadvantages detection at the low levels typically used in fraudulent practice.
of these methods are that usually only known (targeted) com- Pineapple, orange, and grapefruit juices, as high value juices and
pounds can be detected as indicators of certain type of adultera- the most popular fruit juices consumed globally, frequently sold
tion, and the analysis cost is high. as 100% pure, were chosen as ‘model juices’ for the study. The
With today’s testing technology, it is not possible to detect all capability to detect the addition of orange and grapefruit to pine-
adulterants at once in juice. It has been reported by the United apple, grapefruit to orange and orange to grapefruit was also
States General Accounting Office (US GAO, 1995) that current tests investigated.
cannot effectively detect adulteration (dilution with water, sugars
or other fruit juices) at levels below 10% (http://www.gpo.gov/
fdsys/pkg/GAOREPORTS-RCED-96-18/pdf/GAOREPORTS-RCED-96- 2. Materials and methods
18.pdf). Detection of adulteration can be a difficult and complex
task due to the natural variation in the cultivars, as well as differ- 2.1. Chemicals and reagents
ences that occur with different growing regions, storage condi-
tions, sample treatment and processing techniques. Therefore, Ultrapure water was obtained from a Milli-Q purification sys-
there is a need for suitable, comprehensive analytical methods to tem (Millipore, Molsheim, SA, France). HPLC grade acetonitrile,
control authenticity parameters (European Commission, 2009). ammonium acetate (P99%), hesperidin (P95%), neohesperidin
A relatively new discipline used to explore and characterise the (P95%), and naringin (P95%) were supplied by Sigma–Aldrich
complexity of biological pathways, metabolomics, has been shown (St. Louis, MO, USA). Limonin glucoside (P95.7%) was obtained
to be of great promise in different scientific areas, including food from LKT, Laboratories, Inc. (St. Paul, MN) and narirutin (P99%)
fingerprinting. Metabolomics is the study of low molecular weight was purchased from Extrasynthese (Genay Cedex, France).
(<1 kDa) molecules, mostly carried out in biological samples using
bioanalytical and bioinformatic tools (Viant, 2007). Generation of
2.2. Instrumentation
the metabolic chemical profile of a food will include many un-
known components as well as the easily identified components
Chromatographic experiments were performed on a Waters
that would usually be used for targeted analysis, potentially pro-
ACQUITY™ UPLC™ system connected to Xevo G2 Q-ToF MS
viding a more efficient and powerful tool for food fingerprinting.
equipped with an electrospray ionisation source (Waters Corp.,
Thousands of unknown metabolites can be found in plants. As
Milford, MA, USA). A high speed centrifuge (Sigma–Aldrich, St.
the metabolites show high diversity in chemical and physical prop-
Louis, MO, USA) was used for sample preparation.
erties, it is necessary to use high-throughput techniques to profile
the large numbers of these unknown compounds using an untar-
geted metabolomics approach. Additionally, statistical treatment 2.3. Data acquisition and processing
of often very complex datasets is necessary to identify and charac-
terise the metabolites. The data processing steps are very impor- Various software packages associated with the Xevo G2 QToF
tant in any metabolomics study. The most frequently employed MS were used for data acquisition and processing. These included
methods are principal component analysis (PCA), orthogonal pro- MassLynx 4.1., MarkerLynx XS, EZinfo, EleComp, MassFragment
jection to latent structures discriminant analysis (OPLS-DA), sup- and ChromaLynx XS (all from Waters Corp., Milford, MA, USA).
port vector machine (SVM), and artificial neural networks (ANNs) Data were acquired using MassLynx 4.1. Processing and analysis
(Berruta, Alfonso-Salces, & Herberger, 2007). Despite the high (peak detection, data mining, alignment and normalisation) of
interest in the development of metabolomics in recent years, a the data was performed using MarkerLynx XS software. The fol-
number of substantial issues still remain to be elaborated in this lowing parameters were applied: initial retention time 1 min, final
field from a purely methodological view (Antignac et al., 2011). retention time 10 min, mass tolerance 0.02 Da, mass window
Metabolomics has the potential to be used as more than a research 0.02 Da, retention time window 0.1 min, noise elimination level
tool, and may also provide a new platform that will help in routine 6. Multivariate statistical analysis (PCA and OPLS-DA) was per-
monitoring. formed on mass spectral data sets using EZinfo software. Prior to
The authentication of some fruit juices (orange, grapefruit, ap- the PCA, the data were pre-processed using the Pareto scaling.
ple, and their admixtures) has been reported using a metabolo- The quality of the models was described by R2 (goodness of fit)
mics-based HPLC–mass spectrometry (MS) technique employing and Q2 (predictability) values.
 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17 9

Markers identified by EZinfo software were transferred into Ele- separation, mobile phase A was 10 mM aqueous ammonium ace-
Comp software, Chemspider, MassBank, and Metlin online dat- tate and mobile phase B was acetonitrile, with a linear gradient
abases, in order to obtain the elemental composition and elution: 0–0.75 min, A: 99%; 0.75–2.0 min, A: 99–95%; 2.0–
potential molecular structures for the markers. The final confirma- 3.0 min, A: 95%; 3.0–6.5 min, A:95–45%; 6.5–8.5 min, A: 45–
tion of chosen formulae was carried out using MassFragment soft- 10%; 8.5–9.0 min, A: 10%; 9.0–9.1 min, A: 10–99%; 9.1–10 min,
ware. ChromaLynx XS software was used for confirmation and A: 99%. The flow rate of the mobile phase was 0.4 mL/min, and
validation of characterised markers in the targeted metabolomics the column temperature was maintained at 45 °C. The injection
approach. The parameters of the optimised ChromaLynx XS meth- volume was 3 lL. For the HILIC column the same gradient pro-
od were as follows: tolerance range of retention time 0.1 min, mass file was used but mobile phase A was acetonitrile and B was
tolerance 0.002 Da, and threshold of peak absolute height 500. 10 mM aqueous ammonium acetate.
The Student t-test was applied for comparative statistical data
evaluation using Analysis ToolPak (Microsoft Excel 2010).
2.6. QToF MS conditions
2.4. Sample preparation
The MS system was operated using both the positive-ion (ESI+)
Fruit juices were prepared in the laboratory from pineapple (A. and negative-ion (ESI) modes with the mass range set at m/z 50–
comosus), orange (C. sinensis), apple (Malus domestica), clementine 1200 in full scan resolution mode. The ionisation source conditions
(C. reticulata), grapefruit (Citrus paradisi), pomelo (Citrus maxima). were as follows: capillary voltage of 2 kV, source temperature of
The fruits were purchased from Austrian retail markets. Fresh- 130 °C and desolvation temperature of 450 °C. The sampling cone
pressed juices (n = 3) and juices prepared from concentrate voltage was set at 42 V and the extraction cone at 4 V. Nitrogen
(n = 5), produced by various companies, were also purchased from and argon were used as cone and collision gases, respectively.
Austrian retail markets. The cone and desolvation gas flows were 10 and 900 L/h,
Mixtures of pineapple juice adulterated with orange, apple, respectively.
grapefruit and clementine were prepared at 1%, 5%, 10% and 15% The MSE experiment, which records exact mass precursor
adulteration. Orange juice was adulterated with apple, grapefruit and fragment ion information in the same run, was performed
and clementine at the same adulteration levels. with the low collision energy set to 4 eV and the high collision
Fruit juice samples were prepared in the laboratory by pressing energy ramped from 15 to 55 eV. A scan time of 0.1 s was
fresh fruits. Aliquots (2 ml) of fruit juice were centrifuged (10 min, used throughout. The mass spectrometric data were collected
36,670g, 20 °C) and filtered through 0.2 lm PTFE membrane filter in the full scan mode from m/z 50 to 1200, in centroid
before injection. format.
All the data were acquired using an independent reference
2.5. Chromatographic conditions lock mass via the LockSpray interface to ensure accuracy and
reproducibility during the MS analysis. Leucine enkephalin was
UPLC separation was achieved using ACQUITY UPLC™ BEH C18 used as the reference compound ([MH] = 554.2615 and
and BEH HILIC columns (100  2.1 mm, 1.7 lm). For the C18 [M+H]+ = 556.2771) and was continually introduced along with

Fig. 1. The PCA 3-D scores plot obtained in negative (A) and positive (B) mode from pineapple admixtures at 1% adulteration level (n = 6). The ellipse represents the 95%
confidence region for Hotelling’s T2.
10 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17

Fig. 2. The PCA 3-D scores plot obtained in negative (A) and positive (B) mode from orange admixtures at 1% adulteration level (n = 6). The ellipse represents the 95%
confidence region for Hotelling’s T2.

the LC stream for accurate mass calibration. The Lockspray fre- 3. Results and discussion
quency was set at 15 s and the scan time at 0.3 s. The time-of-
flight MS was calibrated daily according to the manufacture’s 3.1. UPLC–QToF MS metabolomic analysis
recommendations.
For sample analysis, replicate injections (n = 6) were used in a For method optimisation in untargeted metabolomics, a generic
random sequence. UPLC–QToF MS method was developed to cover a wide range of
metabolites with diverse chemical and physical properties.

Fig. 3. The PCA 3-D scores plot obtained in negative (A) and positive (B) mode from grapefruit admixtures at 5% adulteration level (n = 6). The ellipse represents the 95%
confidence region for Hotelling’s T2.
 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17 11

Fig. 4. S- and loadings plot of pure pineapple versus pineapple adulterated with orange. Each data point is an exact mass/retention time pair (EMRT) and the points on the top
right and bottom left are potential marker compounds for pineapple (left) and pineapple–orange (right).

Fig. 5. Trend plots of four potential markers extracted from the S-plot in Fig. 3. The four coloured plots show the variation in intensity of each of the EMRT markers in the
samples (p_100% – pineapple pure; p_or_15% – pineapple with 15% orange; p_gr_15% – pineapple with 15% grapefruit; p_md_15% – pineapple with 15% clementine;
p_pm_15% – pineapple with 15% pomelo; (n = 6).
12 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17

Score plots from the unsupervised PCA show clear segregation

between the sample groups (pineapple and pineapple mixed with
orange, grapefruit, apple, and clementine) at the 1% adulteration
level, as shown on the three-dimensional scores plot (3-D), in both
positive (Rx2 = 0.48; Q2 = 0.30) and negative mode (R2x ¼ 0:60;
Q2 = 0.42) (Fig. 1).
Separation of orange and its admixtures (at 1% adulteration le-
vel) was observed on the PCA 3-D scores plot with strong predic-
tion capabilities in both positive (R2x ¼ 0:62; Q2 = 0.47) and
negative (R2x ¼ 0:55; Q2 = 0.47) modes (Fig. 2). This demonstration
of the potential to identify the markers responsible for clear cluster
group segregation at low adulteration level was significant, since
successful detection of C. sinensis adulteration with C. reticulata is
a very complex task and was based until now mostly on determi-
nation of ratios between some flavonoids (Ooghe, 1999) or use of
Fig. 6. Variable trend plots for two unique compounds found in the authentic DNA techniques (Moreau & Canivenc, 2008).
pineapple and pineapple adulterated with orange (at 5%, 10%, and 15%). The lower The PCA score plot of grapefruit admixtures showed clear sepa-
levels of the two markers presented (4.73_498.1542; 4.82_385.1135) in the ration in negative mode with stronger model prediction values
adulterated samples indicate dilution of the authentic pineapple juice. Data points (R2x ¼ 0:45; Q2 = 0.28) than in positive mode (R2x ¼ 0:50; Q2 = 0.13)
represent mean intensity values (n = 6) ± standard deviation.
at 5% adulteration level (Fig. 3). For these admixtures, separation
at 1% adulteration level was not achieved (data not shown).
Chromatographic separation was investigated using reversed It is apparent that more pronounced clustering and significantly
phase (C18) and polar (HILIC) stationary phase to allow separation better differentiation amongst sample groups was obtained for
of a range of compounds of different polarities, from low to high negative ionisation data compared to those acquired in the positive
polarity. The UPLC BEH C18 column incorporates trifunctional li- mode. Therefore, only data acquired in negative ionisation mode
gand bonding chemistries on 1.7 lm BEH particles that produce were further used in this study.
stability over a wide pH range. On the other hand, the UPLC HILIC To understand the interclass separation and identify potential
column contains 1.7 lm unbonded BEH particles and is more sui- characteristic markers for each juice, an OPLS-DA model was used.
ted to retaining and separating very polar basic compounds. The Potential markers of interest were extracted from the S- and/or
information obtained using the BEH column (chromatographic Loadings-plots, which were constructed from OPLS analysis for
separation and peak intensity) was more useful than data obtained all pineapple and orange admixtures. Markers were chosen based
using the HILIC column (data not shown) and therefore the results on their contribution to the variation and correlation within the
from the BEH column were used for the study. data set. From the corresponding plots, the ions furthest away from
The MSE capability of the instrument was used to provide the the origin contribute more significantly to the separation between
exact mass precursor and fragment ion information in one run, en- the groups and may therefore be regarded as the differentiating
abling identification the elucidation of potential structure un- markers. The UPLC–MS analysis platform provides exact mass
known metabolites. This method alternates between two retention time (EMRT) pairs, consisting of the retention time
functions: function 1, when acquiring low-energy exact mass pre- (RT), exact molecular mass and MS/MS data for the structural iden-
cursor ion spectra and function 2 when acquiring high energy exact tification of biomarkers.
mass fragment ion information with an energy (voltage) ramp. MS An example of an S-plot, obtained from OPLS-DA for pure pine-
analysis was done in both positive and negative ionisation modes apple juice and a pineapple–orange admixture (at 5% addition of
to ensure that metabolites amenable to either positive or negative orange juice), is shown in Fig. 4. The S-plot identified the EMRT
ionisation were covered, in order to monitor as many ions as pairs that contribute to the most significant differences between
possible. pure and adulterated pineapple juices (data points outlined on
Sample preparation can be a critical factor in metabolite deter- the plot).
mination and was therefore kept to an absolute minimum (centri- Candidate markers were selected and a trend plot was produced
fugation and filtering) to prevent any substantial loss of possible showing the intensity of characteristic markers in each sample
markers. (Fig. 5). The trend plot is a useful screening tool to indicate possible
Replicate injections (n = 6) of the samples were used in random adulteration. As can be seen on the trend plot, markers with EMRT
sequence to ensure that any experimental trends observed were (4.84_649.2499; 5.39_579.1710; 5.95_593.1867; 5.47_609.1807)
directly associated with the sample and not due to any change in have detected in orange, grapefruit, clementine, and pomelo but
the instrument’s performance over time. A pooled sample (a sam- they are not detected in pineapple pure juice. These markers can
ple comprising an aliquot of every sample in the study) followed by be considered as unique markers and can be used to detect adulte-
a blank (deionized water) were injected in order to ensure the sta- ration of pineapple juice with citrus juices (orange, grapefruit,
bility and repeatability of the LC–MS system. clementine, and pomelo). Four unique markers were also charac-
terised for pineapple juice.
The analysis of markers extracted from the S-plot for orange
adulterated with clementine juice (5%) did not result in the discov-
3.2. Analysis of metabolic pattern ery of any unique markers. Some of the selected markers had high
intensity in clementine but were still present in orange juice at low
The MS data were processed using MarkerLynx XS and the mul- intensity. The separation of the authentic and adulterated orange
tivariate data matrix was analysed using EZinfo software. To eval- juice sample by PCA (Fig. 2) was probably due to the total meta-
uate the capability of the UPLC–MS based metabolomics approach bolomics fingerprints of the juices and characteristic high intensity
for differentiating between fruit juice admixtures, PCA and OPLS- markers, as unique markers were not detected.
DA analysis were further carried out in the study using the EZinfo Unique markers presented using variable trend plots could
software. potentially be used also as a tool to detect some other types of
 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17 13

Fig. 7. The PCA 3-D scores plot obtained in negative (A) and positive (B) mode from orange juice prepared from concentrate and fresh squeezed orange juice obtained from
the market (from different producers, labelled as 100% and 20% pure juice, and juice produced in the laboratory (from oranges grown in Italy, Spain, Greece) (n = 6). The ellipse
represents the 95% confidence region for Hotelling’s T2.

adulteration that cause dilution of authentic fruit juice (with concentrate and fresh squeezed orange juice obtained from the
other fruit juices, water, sugars, etc.). An example of two unique market (from different producers), labelled as 100% pure juice,
markers (4.73_498.1542 and 4.82_385.1135) found in pineapple and juice produced in laboratory (from oranges grown in Italy,
juice exhibiting a dilution effect in three juice samples (pineap- Spain, Greece) were analysed (Fig. 7A and B). The PCA scores plot
ple adulterated with orange at 5%, 10%, and 15%) is shown in in negative mode showed separation between juice prepared in
Fig. 6. Based on the changes in the intensity of these markers the laboratory (inside black circle) and juices from the market
one can conclude that the pure juice has been adulterated (fresh squeezed or prepared from concentrate, red circle) with
(diluted). The differences in the intensities of both markers be- good prediction abilities (R2x ¼ 0:68 and Q2 = 0.61). A mixed prod-
tween the pure pineapple juice and the pineapple adulterated uct labelled as containing 20% orange juice from concentrate (red
with 5% orange juice were significant at 99% confidence level triangles) and fresh squeezed blood orange juice were separated
using a paired t test (n = 6). However, no conclusion on the type from the other groups (orange triangles)1 (Fig. 7A). In positive mode
or level of adulteration can be drawn. This could be a useful tool (Fig. 7B), separation was obtained between commercial fresh
to screen for adulteration with unknowns using authentic squeezed orange juices (blue circle), juices from concentrate (black
samples as a reference. circle), juices made in laboratory (red circle), and between commer-
Finally, all potential unique and high intensity biomarkers giving cial fresh squeezed blood orange juice (orange triangle) and a mixed
a significant contribution were characterised and are listed in Ta- product labelled as containing 20% orange juice from concentrate
ble 1. The confirmation of the unique biomarkers characteristic for (red triangle), with R2x ¼ 0:84 and Q2 = 0.79. One brand of juice pre-
different fruit juices was achieved by extracting the accurate mass pared from concentrate (blue triangle) fell outside the grouping of
of corresponding peaks from the raw data set and constructing juices from concentrate (black circle).
trend-line plots, as shown in Fig. 5. Markers for the other fruit juices Separation by origin (Spain, Italy, and Greece) of Navel orange
investigated were identified using the same principle. juice produced in the laboratory was possible, with a prediction
capability of the PCA model of R2x ¼ 0:59; Q2 = 0.47 and
R2x ¼ 0:58; Q2 = 0.47 for negative and positive mode, respectively
3.3. Analysis of orange juices from different origin (Fig. 8). However, a larger data set would be required to validate
this model and assign an appropriate confidence level before it
Differences in the species of the fruit, its origin, and other fac- could be concluded that juice produced from Navel orange was
tors such as different production practices could cause differences
in the metabolic patterns of commercially available fruit juices. To 1
For interpretation of color in Fig. 7 , the reader is referred to the web version of
evaluate the effect of these factors, orange juice prepared from this article.
14 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17

Table 1
The overview of characteristic marker compounds detected in evaluated fruit juices.

Marker RTa m/z (Precursor Ion m/z (Fragment Probable ion elemental Mass error Identification (confirmedb and Juice
number (min) ion) ion) formula (mDa) tentativec) type
1 5.46 609.1827 [MH] 301.0718; C28H34O15 0.8 Hesperidineb O, C
475.2139
 b
2 5.53 609.1825 [MH] 301.0718; C28H34O15 0.6 Neohesperidine G
475.2139
3 4.84 649.2497 [MH] 443.2072; C32H42O14 0.1 Limonin-17-b-D-glucopyranosideb O, G, C
605.2596
4 5.32 693.2751 [MH] 271.0603; C34H46O15 0.7 Nomilinin-17-b-D-glucopyranosidec O, G, C
556.2641

5 5.15 651.2651 [MH] 607.2739 C32H44O14 0.2 Obacunoic acid-17-b-D- O, G
glucopyranosidec

6 4.93 711.2855 [MH] 607.2737 C34H48O16 0.9 Nomilinic acid-17-b-D- O, G, C
glucopyranosidec
7 5.34 579.1718 [MH] 271.0807; C27H32O14 0.4 Narirutinb O, G, C
1510037
8 5.40 579.1712 [MH] 271.0605; C27H32O14 0.2 Naringinb G, PO
151.0030
 c
9 5.95 593.1870 [MH] 285.0759; C28H34O14 0.6 Isosakuranentin-7-O-rutinoside O, C
164.0108
10 4.58 669.2747 [MH] 609.2548; C32H46O15 1.1 Daecetyl nomilinic acid-17-b-D- O
205.0349 glucopyranosidec
11 5.18 593.1498 [MH] 391.1599; C27H30O15 0.8 Unknown O, G, C
353.1445

12 5.52 463.1232 [MH] 301.0710; C22H24O11 0.8 Unknown O
164.0112
 c
13 4.73 498.1541 [MH] 306.0755; C21H29N3O9S 0.5 S-synapiylgluthathione P
143.0454
14 4.63 441.1301 [MH] 128.0345; C19H28N2O8S 1.7 N-L-glutamyl-S-synapyl-L-cysteinec P
249.0534
16 5.48 407.1917 [MH] 125.0249; C15H26O10 0.2 Unknown A
263.1483

17 2.79 164.0712 [MH] 147.0441 C9H11NO2 0.6 L-()-Phenylalanine P
18 4.81 385.1129 [MH] 205.0495 C17H22O10 0.1 1-O-b-D-glucopiranoside sinapate P
19 4.90 441.1967 [MH] 249.133 C18H33O12 0.5 Unknown A
20 4.74 323.1329 [MH] 101.06 C13H23O9 0.9 Unknown A
21 4.35 351.1290 [MH] 101.0602 C14H23O10 0.1 Unknown A

O, orange; G, grapefruit; C, clementine; P, pineapple; A, apple; PO, pomelo.

The mass error shows the difference between the calculated and the specified mass.
a
Retention time.
b
Confirmed by standard.
c
Tentative.

influenced by differences in origin (environmental conditions such bo_search_alt2.php) databases were used. The search for the
as diverse soil types, cultivation environments, altitude, etc.). molecular formula can be complicated as multiple hits can be
matched with the same exact mass. In this case the use of the
3.4. Biomarker identification/characterisation fragmentation pattern of each molecule obtained with the MSE
functionality increased the confidence that the correct formulas
The UPLC QTof MS analysis platform provides the retention were being selected. The experimental MS/MS spectra were
time, precise molecular mass and MS/MS fragmentation data for compared to reference spectra obtained from MassBank. For the
the structural elucidation of biomarkers. compounds that were not found in MassBank, the most likely
The accurate mass information of precursor and product ions compounds were chosen from other databases, taking into account
was used to calculate the proposed molecular formula of each mar- consistency with the samples under study and the classes of
ker using EleComp software. Atoms considered for the molecular compounds expected in those samples, and their structures were
formula calculation, were as follows: C (n 6 50), H (n 6 100), N analysed further with fragments obtained previously.
(n 6 10), O (n 6 20), and S (n 6 5). This software proposes the for- In the next step, the structures of tentatively identified com-
mula based on matching of the experimental and theoretical isoto- pounds were confirmed using MassFragment software. The Mass-
pic pattern of the marker. Suggested formulas are sorted according Fragment application manager facilitates the MS/MS fragment
to i-FIT that indicates the confidence in the accuracy of the sug- ion analysis process by way of chemically intelligent peak-match-
gested composition. The i-FIT score is a comparative measure of ing algorithms, allowing structural confirmation of identified com-
how well a cluster of peaks matches the pattern of each set of pre- pound. For the defined operational conditions, the identified
dicted isotope peaks; the lower the i-FIT value is, the better the fit. markers, from the 21 markers characterised, are shown in Table 1.
All suggested formulas have very good mass accuracy, less than 2 Finally, these biomarkers were selected as candidate markers
mDa, thus increasing the confidence in predicted formulas for further confirmation and validation by Chromalynx.
(Table 1).
The molecular formulas obtained were entered in online chem- 3.5. Biomarker confirmation
ical databases to search for reference candidates. The Chemspider
(http://pubchem.ncbi.nlm.nih.gov/), MassBank (http://www.mass- To confirm some of the identified specific markers, commer-
bank.jp/?lang=en/), and Metlin (http://metlin.scripps.edu/meta- cially available reference standards were analysed and their mass
 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17 15

Fig. 8. The PCA 3-D scores plot obtained in negative (A) and positive (B) mode from juice produced in the laboratory (from Navel oranges grown in Italy, Spain, Greece) (n = 6).

Fig. 9. (A) Extracted ion chromatograms of (top to bottom) neohesperidin standard, hesperidin standard, and ions at m/z 609.1816 and 609.1814 in grapefruit juice
adulterated with orange juice at 10% and (B) the MSE fragmentation data correlated to the standards and the sample.
16 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17

Fig. 10. (A) Extracted ion chromatograms of (top to bottom) naringin standard, narirutin standard, and the ions at m/z 579.1708 and 579.1703 in grapefruit juice adulterated
with orange juice at 10%, and (B) the MSE fragmentation data correlated to the standards and the sample.

spectra and chromatographic retention times were compared to out using ChromaLynx XS software. Using this software it was also
those obtained from the samples. possible to test and confirm the robustness of the discovered
For two compounds that eluted at retention times 5.37 and biomarkers and their presence in fruit juices obtained from the
5.43 min with m/z 609.1827 and 609.1825, respectively, the Elem- market without using reference compounds. ChromaLynx XS soft-
Comp software suggested the same formula C27H31O14 (Fig. 9). In ware identifies the presence of multiple components, applies
this instance ChemSpider proposed two potential hits: hesperidin deconvolution to accurately profile complex chromatograms to
and neohesperidin. These two compounds are isobaric and have produce pure spectra for each component, estimates component
the same molecular formula. The high resolving power of the concentrations by performing semi-quantative determinations,
instrument allowed separation of these compounds and final iden- and calculates and reports peak areas and relative peak areas.
tification by comparing MS high resolution data with the data ob- The biomarker confirmation was based on exact-mass scoring
tained from injected standard compounds. (match of exact and fragment masses, and retention time) using
The situation was similar for compounds eluting at 5.34 and a library search. The library was built up from characteristic mark-
5.40 min, where EleComp software suggested narirutin and narin- ers obtained using the untargeted metabolomics approach.
gin (Fig. 10). Thus, two unique markers (neohesperidine and narin- Most of the markers were detected in concentrate and fresh
gin) characteristic for grapefruit juice were detected. fruit juices obtained from the market at 1% adulteration level. In
These examples show the advantage of the use of a high resolu- this case, the selected markers were shown to be robust even if dif-
tion mass spectrometry untargeted metabolomic method com- ferent varieties of fruit were used in production, or different pro-
pared to the more commonly used targeted chromatographic duction practices and storage conditions were applied.
methods, which would not have allowed the resolution and identi- Metabolic profiling using UPLC–QToF MS is a novel, highly sen-
fication of these unique markers. Metabolites frequently have very sitive and specific biomarker tool for tracing fruit juice authentic-
complex chemical structures and may be difficult to synthetize, so ity. In the present study, we focused on the adulteration of
commercial standards are not available. To overcome these disad- pineapple, orange, and grapefruit juices (as ‘model juices’) at low
vantages of the untargeted metabolomics approach, a targeted adulteration levels with other fruit juices (grapefruit, orange, ap-
metabolomics method was optimised and analysis was carried ple, clementine, and pomelo) using an untargeted metabolomics
 Z. Jandrić et al. / Food Chemistry 148 (2014) 7–17 17

approach, and a targeted method for biomarker confirmation. The Antignac, J. P., Courant, F., Pinel, G., Bichon, E., Monteau, F., Elliot, C., et al. (2011).
Mass spectrometry-based metabolomics applied to the chemical safety of food.
development of a metabolomics method could facilitate the rapid
Trends in Analytical Chemistry, 30, 292–301.
detection, risk assessment and routine monitoring of fruit juice Ashurst, P. R. (2005). Chemistry and technology of soft drinks and fruit juices (2nd ed.).
adulteration. The study illustrated the ability of the UPLC–QToF John Wiley & Sons.
MS untargeted metabolomics approach to identify potential mark- Berruta, L. A., Alfonso-Salces, R. M., & Herberger, K. (2007). Supervised pattern
recognition in food analysis. Journal of Chromatography A, 1158, 196–214.
ers that could be useful for a targeted metabolomics method. The Codex Alimentarius Commission. (1992). Fruit juices and related products. Rome.
present study also indicates that metabolomics may be a feasible Cuny, M., Vigneau, E., Le Gall, G., Colquhoun, I., Lees, M., & Rutledge, D. N. (2008).
approach to investigate the dynamic metabolic alterations that Fruit juice authentication by 1H NMR spectroscopy in combination with
different chemometric tools. Analytical and Bioanalytical Chemistry, 390,
can occur during fruit juice production (extraction, pasteurisation, 419–427.
and evaporation) and storage. Ehling, S., & Cole, S. (2011). Analysis of organic acids in fruit juices by liquid
The results not only indicated that the metabolomics method chromatography–mass spectrometry: An enhanced tool for authenticity testing.
Journal of Agricultural and Food Chemistry, 59, 2229–2234.
had sufficient sensitivity and specificity to distinguish between European Commission Directive 2009/106/EC amending Commission Directive
authentic fruit juices and their admixtures, including dilution, 2001/112/EC relating to fruit juices and certain similar products intended
but also that metabolic fingerprinting of fruit juices has the poten- to human consumption. Official Journal of the European Communities,
L 212/42.
tial to be developed into useful tool in fruit juice forensics for dis- Gómez-Ariza, J. L., Villegas-Portero, M. J., & Bernal-Daza, V. (2005). Characterization
covering new adulteration practices, as more advanced and and analysis of amino acids in orange juice by HPLC–MS/MS for authenticity
sophisticated adulteration methods are continuously developing. assessment. Analytica Chimica Acta, 540, 221–230.
Moore, J. C., Spink, J., & Lipp, M. (2012). Development and application of a database
of food ingredient fraud and economically motivated adulteration from 1980 to
4. Conclusions 2010. Journal of Food Science, 77, R118–R126.
Moreau, F., & Canivenc, G. (2008). DNA analysis as a tool for Citrus reticulate
adulteration detection and variety identification in commercial orange juices.
Metabolomics provides a powerful approach to discover robust, Fruit Processing, 156–159.
unique and high intensity biomarkers by obtaining a metabolic fin- Muntean, E. (2010). Simultaneous carbohydrate chromatography and unsuppressed
gerprint of fresh prepared fruit juices. Here, for the first time to our ion chromatography in detecting fruit juices adulteration. Chromatographia, 71,
S69–S74.
knowledge, we report the identification of markers within a meta- Ooghe, W. (1999). Flavonoids as authenticity markers for Citrus sinensis juice. Fruit
bolomic pattern of a range of fruit juices using UPLC–QToF MS. Processing, 308–313.
Metabolic fingerprint analysis was shown to be an effective ap- Redd, J. B., Hendrix, D. L., & Hendrix, C. M. (1992). Quality control manual for citrus
processing plants (Vol. 2). Auburandale, FL: Agscience.
proach in differentiating between authentic fruit juices and their
Rossmann, A. (2001). Determination of stable isotope ratios in food analysis. Food
adulterated mixtures down to 1% adulteration level. We character- Reviews International, 17(3), 347–381.
ised 21 metabolites, 15 of which were identified, that were used in Rouseff, Russel L. (1988). Adulteration of fruit juices beverages. In S. Nagy, M. E.
Rhodes, & J. A. Attaway (Eds.), Differentiating citrus juices using flavanone
the validation process. It is suggested that the UPLC–MS metabolo-
glycoside concentration profiles (pp. 49–64). New York and Basel: Marcel Dekker
mics approach is highly effective for biomarker identification and Inc..
could thus represent a new strategy in food forensics. To further Saavedra, L., Rupérez, F. J., & Barbas, C. (2001). Capillary electrophoresis for
develop and investigate such a strategy, more different fruit juice evaluating orange juice authenticity: A study on Spanish oranges. Journal of
Agricultural and Food Chemistry, 49, 9–13.
varieties must be analysed to build up the database of metabolo- Schwartz, R. S., & Hecking, L. T. (1991). Determination of geographic origin of
mic patterns. agricultural products by multivariate analysis of trace element composition.
In conclusion, the results demonstrate that metabolomics, in Journal of Analitical Atomic Spectrometry, 6, 637–642.
Simpkins, W., & Harrison, M. (1995). The state of the art in authenticity testing.
conjunction with a comprehensive database, has potential as a Trends in Food Science and Technology, 10, 6.
screening tool for the detection of food fraud by adulteration, GAO (United States General Office). (1995). Fruit juice adulteration. <http://
and may be used in the future to enable a rapid reaction in the glo- www.gpo.gov/fdsys/pkg/GAOREPORTS-RCED-96-18/pdf/GAOREPORTS-RCED-
96-18.pdf> Accessed 15.06.12.
bal fruit juice market and to help regulators to stay one step ahead Vaclavik, L., Schreiber, A., Lacina, O., Cajka, T., & Hajslova, J. (2011). Liquid
of fraudsters. chromatography–mass spectrometry-based metabolomics for authenticity
assessment of fruit juices. Metabolomics. http://dx.doi.org/10.1007/s11306-
011-0371-7. ISSN: 1573-3882, IF 3,608.
References
Viant, M. R. (2007). Metabolomics of aquatic organisms: The new ‘omics’ on the
block. Marine Ecology-Progress Series, 332, 301–306.
Anderson, D. L., Cunningham, W. C., & Alvarez, G. H. (1992). Multielement analysis
of foods by neutron capture prompt c-ray activation analysis. Journal of
Radioanalytical and Nuclear Chemistry, 167, 139–144.