Food Chemistry 162 (2014) 176–185
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Thermal degradation of cloudy apple juice phenolic constituents D. De Paepe a,b,c,⇑, D. Valkenborg c,d, K. Coudijzer a, B. Noten c, K. Servaes c, M. De Loose b, S. Voorspoels c, L. Diels c, B. Van Droogenbroeck b a
Institute for Agricultural and Fisheries Research (ILVO), Technology and Food Science Unit (T&V), Food Pilot, Brusselsesteenweg 370, 9090 Melle, Belgium Institute for Agricultural and Fisheries Research (ILVO), Technology and Food Sciences Unit (T&V), Product Quality and Innovation (PI), Burgemeester Van Gansberghelaan 115/1, 9820 Merelbeke, Belgium c Flemish Institute for Technological Research (VITO), Business Unit Separation and Conversion Technology (SCT), Boeretang 200, 2400 Mol, Belgium d Interuniversity Institute for Biostatistics and Statistical Bioinformatics, Hasselt University, Agoralaan 1, 3590 Diepenbeek, Belgium b
a r t i c l e
i n f o
Article history: Received 6 November 2013 Received in revised form 26 March 2014 Accepted 1 April 2014 Available online 13 April 2014 Keywords: Phenolics Thermostability Cloudy apple juice Pasteurization
a b s t r a c t Although conventional thermal processing is still the most commonly used preservation technique in cloudy apple juice production, detailed knowledge on phenolic compound degradation during thermal treatment is still limited. To evaluate the extent of thermal degradation as a function of time and temperature, apple juice samples were isothermally treated during 7200 s over a temperature range of 80–145 °C. An untargeted metabolomics approach based on liquid chromatography–high resolution mass spectrometry was developed and applied with the aim to find out the most heat labile phenolic constituents in cloudy apple juice. By the use of a high resolution mass spectrometer, the high degree of in-source fragmentation, the quality of deconvolution and the employed custom-made database, it was possible to achieve a high degree of structural elucidation for the thermolabile phenolic constituents. Procyanidin subclass representatives were discovered as the most heat labile phenolic compounds of cloudy apple juice. Ó 2014 Published by Elsevier Ltd.
1. Introduction Despite the emerging of novel non-thermal preservation techniques, conventional thermal processing is still the most commonly used preservation technique in cloudy apple juice production (Rupasinghe & Yu, 2012). This practice is a result of its efficient inactivation of microorganisms and enzymes responsible for deterioration (Awuah, Ramaswamy, & Economides, 2007). Batch heating at 63–65 °C for 30 min. is the most traditional method (D’Amico et al., 2006). However, through the years, process optimisation was conducted by reducing processing times at high(er) temperatures in order to avoid undesirable quality changes during this process (Awuah et al., 2007). Currently, high temperature – short time (HTST) pasteurisation at 77–88 °C for 25–30 s is the most commonly used method for heat treatment of cloudy apple juice (Aguilar-Rosas, Ballinas-Casarrubias, Nevarez-Moorillon, Martin-Belloso, & Ortega-Rivas, 2007).
⇑ Corresponding author at: Institute for Agricultural and Fisheries Research (ILVO), Technology and Food Sciences Unit (T&V), Product Quality and Innovation (PI), Burgemeester Van Gansberghelaan 115/1, 9820 Merelbeke, Belgium. Tel.: +32 09 272 28 38; fax: +32 09 272 28 01. E-mail address:
[email protected] (D. De Paepe). http://dx.doi.org/10.1016/j.foodchem.2014.04.005 0308-8146/Ó 2014 Published by Elsevier Ltd.
However, thermal processing can promote reactions that could affect colour, odour, flavour, texture and health-effect, which all could be linked to the change in phenolic profile (Niu et al., 2010). To predict the change in phenolic profile of cloudy apple juice during heat treatment, the knowledge of the phenolic composition as well as the kinetics of phenolic compound degradation, including the reaction rate as a function of temperature, are required (Vikram, Ramesh, & Prapulla, 2005). In the past, kinetic studies regarding phenolic constituents of apple juice were mostly carried out in aqueous model solutions. For such solutions, it was already demonstrated that the heat sensitivity of phenolic compounds depends on their structure, in other words, to the phenolic subclass to which they belong (Buchner, Krumbein, Rohn, & Kroh, 2006). This tendency was confirmed in enriched apple juice during accelerated storage experiments (van der Sluis, Dekker, & van Boekel, 2005). However, it is difficult to use such intrinsic kinetic data for process optimisation given the fact that the matrix can protect against heat or promote the degradation (Ioannou, Hafsa, Hamdi, Charbonnel, & Ghoul, 2012). Furthermore, in all available studies, only a limited set of wellknown and often most abundant phenolic compounds are studied as a consequence of the employed ‘targeted approach’. Following such an approach, it is very time consuming to find out the identity
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
of ‘the most heat labile phenolic constituents’, while these molecules could be just the ones that are best suited to be used as ‘quality targets’ during process optimisation. A non-targeted approach imposes itself. Due to recent technological advances in the field of High-Resolution Mass Spectrometry (HRMS), untargeted metabolomics approaches have become a feasible approach to analyse simultaneous behaviour a large amount of molecular entities in complex matrices. Therefore, the objective of the present study was to find out the most heat labile phenolic constituents in cloudy apple juice by the application of a targeted metabolomics approach. Attention will be given to the pitfalls inherent to the large-scale liquid chromatography and mass spectrometry based untargeted screening method for gathering kinetic information.
177
2.3. Sample preparation The sealed capillary tubes were degreased with 100% ethanol whereupon the tips of the tubes were broken and the samples were released in sterile tubes (Eppendorf, Nijmegen). Subsequently, the samples were centrifuged for 10 min at 12,000 rpm (13,000g) using a Galaxy 16DH ultracentrifuge (VWR, Leuven, Belgium). Of each obtained supernatant, 100 lL was diluted serially in a microvial by adding 150 lL U-HPLC grade methanol containing a 40 mM ammonium formate buffer. A quality control (QC) sample, assembled by pooling 20 lL of each obtained supernatant (in total 288 juice samples, coming from 3 repetitions of the eight temperatures consisting of 12 different time points), was divided into 24 aliquots and diluted in the same manner. All samples were stored at 4 °C prior to injection into the U-HPLC-DAD/ESI-am-MS system.
2. Materials and methods 2.4. Analytical platform 2.1. Pilot scale cloudy apple juice production Storage apples cv. ‘Jonagored’ (ULO storage for 10 months, caliber 65–70 mm), were purchased from a Belgian fruit auction and stored at normal atmosphere in a cold room (0 °C) until use. A combined washer/elevator/rasp mill combination (KWEM 1000, Kreuzmayr, Wallem, Germany) was used for washing and shredding the apples into mash. Subsequently, the apple mash was pressed by means of a spiral-filter pressed (VacuLIQ 1000, VacuLIQ, Hamminkeln, Germany) whereof optimised conditions were used (De Paepe, Noten, De Loose, Van Droogenbroeck, & Voorspoels, 2013). The obtained cloudy apple juice was collected in transparent plastic bags and vacuum-packed (8 mbar) just after sampling (K5N, VC999 verpackungssysteme AG, Herisau, Switzerland). Soluble solids content and pH were measured by means of digital refractometry (RM 40, Mettler-Toledo, Greifensee, Switzerland) and potentiometry (SevenCompact, Mettler Toledo, Greifensee, Switzerland), respectively. 2.2. Isothermal heat treatments To evaluate the extent of heat treatment as a function of time and temperature, apple juice samples were isothermally treated at different time–temperature combinations. The studied temperature ranges included HTST temperatures (80, 85 and 90 °C) and ultra high temperatures (UHT, 135, 140 and 145 °C) and some temperatures in between both ranges (100 and 120 °C). To reach isothermal conditions fast, sterile capillary glass tubes (150 1.55 mm, VWR, Haasrode, Belgium) were used. They were filled with 150 lL of pasteurised apple juice prepared as described above and were taken through a small hole in packaging. Subsequently, each tube was brought horizontal whereby a headspace of 20 mm arose at both ends, protecting the juice for warm up during sealing. The tubes were sealed using a Bunsen burner (Sigma– Aldrich, Bornem, Belgium). Only the extreme top of the tube was brought in the flame for 1 s. For the heat treatment, the capillary tubes were completely submerged in an oil bath (Immersion circulator 1127P, VWR, Haasrode, Belgium) of which the temperature was electronically controlled. After the desired heating time (0 or untreated, 10, 20, 40, 60, 120, 300, 600, 900, 1800, 3600, 5400, and 7200 s), the tubes were removed from the oil bath and immediately cooled in ice water to stop any further reaction. The experiment as a whole (8 temperatures, 12 time points per temperature) was repeated three times on three consecutive days in a randomised block design, wherein each block represents a time series at a certain temperature.
The LC system consisted of an Accela™ quaternary solvent manager, a ‘Hot Pocket’ column oven (Thermo Fisher Scientific, Bremen, Germany) and a CTC PAL™ autosampler (CTC Analytics, Zwingen, Switzerland). A reversed phase separation was performed on a Waters Acquity UPLC™ BEH SHIELD RP18 column, with dimensions 3.0 150 mm, 1.7 lm (Waters, Milford, MA). To protect the U-HPLC column, an Acquity BEH C18 VanGuard pre-column, with dimensions 1.7 lm, 2.1 5 mm (Waters, Milford, MA) was coupled with the analytical column. The mobile phase consisted of water + 0.1% formic acid (solvent A) and acetonitrile + 0.1% formic acid (solvent B). The gradient was varied linearly from 0% to 26% B (v/v) in 9.91 min, to 65% B at 18.51 min, and finally to 100% B at 18.76 min and held at 100% B to 20.76 min. Afterwards, the initial conditions of 100% A were re-equilibrated from 20.88 min to 23.00 min prior to the next injection. The flow rate was 500 lL min1 and the column temperature was set at 40 °C. Aliquots of 5 lL of the sample extract were injected into the chromatographic system. The UV spectra of all selected phenolic compounds were recorded in the range of 200–400 nm for tentative identification using an Accela™ photo diode array (PDA) detector. An Orbitrap mass spectrometer (Exactive™, Thermo Fisher Scientific, Bremen, Germany) operating with an Ion Max™ ESI source (Thermo Fisher Scientific, Bremen, Germany) in negative ionisation mode (ESI-) was used with the following operation parameters: spray voltage 2.5 kV; sheath gas (N2, >99.99%) 47 (adimensional); auxiliary gas (N2, >99.99%) 15 (adimensional); skimmer voltage 25 V; tube lens voltage 110 V; and capillary temperature 350 °C. The mass spectra were acquired using an acquisition function as follows: resolution, high (equivalent to a mass resolving power of 50,000 FWHM at m/z 200); automatic gain control (AGC), balanced (target value of 1 106), and scan speed, 2 Hz. Mass range in the full scan experiments was set at m/z 90– 1800. Data acquisition and instrument control were performed using Xcalibur 2.2.1 software (Thermo Fisher Scientific, Bremen, Germany). 2.5. Measurement design A total of 288 apple juice samples + 24 QC were analysed during two consecutive measurement series. The first measurement series consisted of the time series at the 4 lowest studied temperatures (80, 85, 90 and 100 °C), while the second measurement series consisted of the time series at the 4 highest studied temperatures (120, 135, 140 and 145 °C). Within each measurement series, the time series were measured in a randomised block design. Each time
178
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
series includes an untreated apple juice sample and a pooled QC sample, surrounded by two solvent blanks. 2.6. Data processing The global strategy followed to determine the most heat unstable constituents of cloudy apple juice is presented as a funnel diagram in Fig. 1. The employed workflow was based on the modified procedure for non-targeted metabolic profile analysis using LC–MS (Matsuda et al., 2009) with some additional steps. In what follows, each sub-step will be discussed in detail. 2.6.1. Data pretreatment The ‘.raw’ extension data files acquired from analysed samples were converted to ‘.mzXML’ file format by the msconvert tool of ProteoWizard 3.0.4416 (Kessner, Chambers, Burke, Agus, & Mallick, 2008). These raw chromatogram data were organised into a matrix structure (hereafter referred to as a ‘peak intensity table’), using MetAlign™ 2.0 software package (De Vos et al., 2007; Lommen et al., 2007). In this matrix, peak intensity data derived from a chromatographic peaks commonly observed among the samples (eluted at similar retention times with identical mass numbers) were recorded in a single row. The baselines of the converted raw chromatogram data were removed by automated baseline correction by means of the ‘Baseline correction module of MetAlign™.’ The MetAlign™ settings for the baseline correction were tuned according the MetAlign protocol for untargeted large-scale plant metabolomics using liquid chromatography coupled to mass spectrometry (De Vos et al., 2007). The values for the instrument-dependent parameters of MetAlign™ are the retention time region included in the analysis, i.e., scan number 50–1750. Further, the maximum amplitude of the mass spectrometer was set at 108 arbitrary units and the chromatographic average peak width equal to a FWHM of 4 s (8 scans). These were all determined from the Qual Browser module of the Xcalibur 2.2.1 acquisition software (Thermo Fisher Scientific, Bremen, Germany) and were found to similar for all chromatograms of the data set (De Paepe et al., 2013). Other baseline correction parameters were: peak slope factor = 1, peak threshold factor = 2, and peak threshold = 104.
Raw data
Baseline correction Peak alignment Prefiltering Intensity filtering Deconvolution Phenolic filtering
Phenolic database
Kinetic filtering Identification Parameter estimation
Top 5 most heat labile phenolic constituents Fig. 1. Funnel diagram presenting the applied strategy to determine the most heat unstable constituents of cloudy apple juice.
To correct for retention time drift, the ‘Alignment’ module of MetAlign™ was applied on the baseline corrected chromatograms (De Vos et al., 2007; Lommen et al., 2007). The chromatographic shift parameters for alignment (‘iterative procedure’) were the following: maximum shift (30 per 100 scans), minimum factor (3 first and 2 s iteration), and minimum number of masses (5 first and 3 s iteration) were optimised by trial-and-error according to the parameter selection procedure for peak alignment in chromatographic sample profiling (Lee et al., 2012; Peters, van Velzen, & Janssen, 2009). In this procedure, the nonsymmetrical distributed intensities of the samples were converted to a normal distribution by means of Briggsian log transformation. Sequentially, the dataset was quantile normalised to remove systematic errors which allows the comparison of the intensity profiles of samples (Lee et al., 2012). Subsequently, the quality of the alignment was investigated by the Pearson’s correlation coefficient (R2) between intensity profiles of sample replicates. For each sample, the average correlation coefficient regarding the replicates was calculated and Z-transformed Zcorr = 0.5 Ln((1+R2)/(1R2))(n3)1/2, where n is the number of peaks in the peak intensity table. The normality of the Zcorr values enables both comparison of Zcorr values across different alignment outputs with different numbers of detected peaks and selection of the best output according to the p-value of Zcorr deviation from the N(0,1) null hypothesis (Lee et al., 2012). Quantile normalisation and quality check of the alignment was performed with R 2.15.2 and the ‘DNAMR’ package. To correct for drift in signal intensities between and within the measurement series, a standard linear normalisation method was conducted, based on the replicate measurements of the pooled QC samples throughout the measurement series (baseline matrix). Within each time series and for each mass peak, the ion intensity of each sample was divided by the ion intensity of the pooled QC sample (acquired at the end of each time series) and subsequently multiplied by the mean ion intensity of all 24 pooled QC samples and Briggsian log transformed. The coefficient of variation of intensities (ratio of the standard deviation and mean) of a specific mass peak across the untreated (t = 0) apple juice replicates was used to provide an indication for the quality of normalisation (Lämmerhofer & Weckwerth, 2013). In addition to this numerical variability measure, box-whisker plots for the coefficient of variation and Briggsian log intensity distributions of the same replicates before and after normalisation were constructed with R 2.15.2 and the ‘DNAMR’ package. After normalisation, the MetAlign™ peak intensity table was further processed by means of Perl/Tk written software package NToolbox™ (Matsuda et al., 2009). To remove the low quantitative peaks eluted near the void volume, as well as the broad peaks eluted at the end of the chromatogram, the peaks eluted before 5 min (scan number 100) and after 20 min (scan number 1800) were discarded using the ‘Nprefilter.pl’ function. Low-intensity mass peaks were filtered by ‘Nfilter.pl’. Therefore, the thermal treated samples in the peak intensity table (8 temperatures 12 time points 3 replicates = 288 samples) were divided into 96 groups (8 temperatures 12 time points) of 3 corresponding replicates. Rows including at least one group in which the intensity values of all the 3 replicates were above the cutoff value (5000) were kept in the matrix, while the others were removed. The cutoff value was calculated from the MetAlign™ noise output file and corresponded with an S/N ratio of 5. Further reduction of the data complexity and redundancy and peak deconvolution by means of multivariate mass spectra reconstruction (MMSR) on the peak intensity table was also performed (Tikunov et al., 2005). Each molecular entity with a unique m/z and retention time (feature) was combined with his putative corresponding in-source fragments, adducts and isotopes based on the correlation between them throughout the samples, by means of
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
the ‘Nisotoperemover.pl’. The parameters to deconvolute the isotope peaks were: threshold for m/z shift = 2, retention time window = 3 scans, threshold for correlation between the expression patterns = 0.9. In-source fragments and adducts were deconvoluted with the following parameters: threshold for correlation between the expression patterns = 0.9, retention time window = 3 scans, and threshold for intensity = 105. Due to the fact that deconvolution by means of MMSR was performed on centroid data (peak intensity table), peak shape similarities between putative corresponding ions were manually checked by means of the Qual Browser module of the Xcalibur 2.2.1 acquisition software (Thermo Fisher Scientific, Bremen, Germany) during the identification process. Doing so, ions with both the same apex location and similar peak shape are clustered into one molecular entity. MMSR clustered ions with a deviating apex location and/or peak shape were eliminated. 2.6.2. Database development A database containing 7050 phenolic compounds (6534 flavonoids, 129 phenolic acids and 387 condensed tannins) was build based on two comprehensive online databases (www.lipidmaps.org, www.metabolomics.jp). Phenolic compounds were first organised into appropriate subclasses based on their backbone structure. For all flavonoids, potential in-source fragments originating from their substitution pattern were generated by means of R-group decomposition implemented in JChem for Excel (Chemaxon, Budapest, Hungary). Neutral loss and backbone fragmentation for the flavonoid subclasses flavones, flavonols, and flavanones (4124 compounds) were simulated based on their subclass specific fragmentation patterns described in literature (Abranko, Garcia-Reyes, & Molina-Díaz, 2011). The developed database allowed a reduction of the number of hits after accurate mass filtering by subclass elucidation prompted by the observed UV/Vis spectrum. Furthermore, theoretical calculated fragments give direct insight in the expected in-source fragments which is not the case for the databases publicly available online. Finally, a high degree of completeness permits the calculation of molecular formula properties (allowed elements, element limits and element ratio’s) typical for phenolic compounds. 2.6.3. Phenolic compound filtering With the aim to identify the most heat labile phenolic compounds in cloudy apple juice, and to minimize the identification workload, potential phenolic compounds were selected from other compounds before identification was started. For this purpose, several criteria based on the molecular formula properties of phenolic compounds were applied during the prediction of the elemental composition of the molecular entities included in the deconvoluted peak intensity table. The prediction of the elemental composition was based on the seven golden rules for heuristic filtering of molecular formulas from accurate mass data (Kind & Fiehn, 2007). With the intention to focus on phenolic compounds, the list of allowed chemical elements was limited to carbon (12C, 13C), hydrogen (1H) and oxygen (16O) based on the developed database. Furthermore, by using the full set of molecular formulas inside the developed database, maximum element count for 12C (79), 1H (91) and 16O (45) were determined and used as absolute element limits. The maximum element count for 13C was set to 1. To allow calculation of the molecular formulas of the fragments, the minimum element count for 12C, 13C, 1H and 16O were set to 0. The allowed mass deviation was set to 5 ppm. To check if a molecular graph (a chemically existent species) could be built from a specific formula, the detected ionic species were first ‘neutralized’ by determining the adduct formation ([MH], [M2H]2/2) based on the isotope distribution and correcting for it. The isotope distribution was manually checked by
179
means of Xcalibur 2.2.1 software (Thermo Fisher Scientific, Bremen, Germany). Subsequently, the obtained molecular formulas were subjected to the ‘Lewis rule’ and ‘Senior rule’ by using MOLGEN version 5.0 to verify the electronic structure (Braun et al., 2004). Finally, based on the developed database, the allowed hydrogen/carbon ratio range (0.738–2.250) and oxygen/carbon ratio range (0.067–1.267) were used to check whether it was possibly a phenolic compound. A shortlist of molecular entities with molecular formula properties of phenolic compounds was obtained. 2.6.4. Kinetic plot filtering To find out the most heat labile molecular entity in the filtered and deconvoluted peak intensity table, the kinetic curves for each molecular entity obtained after deconvolution and filtering were plotted (intensity vs. time). The molecular entities exhibiting a single-step degradation trending at all studied temperatures were separated by visual evaluation of the printed kinetic curves from those which demonstrates a consecutive degradation trend (formation followed by degradation) throughout the studied temperatures. 2.6.5. Kinetic parameter determination For the resulting set of molecular entities which exhibit a single-step degradation behaviour, a first-order differential equation was used to determine the kinetic parameters. The use of a first order model to describe the thermal degradation of phenolic compounds was previously reported (Ioannou et al., 2012). It was technical impossible to monitor the cloudy apple juice temperature during heat treatment, but based on the use of thin-walled capillary glass tubes, the ramp-up and ramp-down times were assumed to be negligible compared to the length of the holding phase. Therefore, an isothermal method was used to estimate the degradation kinetics. The first-order kinetic rate constants (k) at each temperature were calculated by means of linear regression through origin using Eq. (1), with I the intensity of the molecular entity at time t, I0 the initial intensity of the molecular entity, t the treatment time (s) and k the first-order degradation rate constant (s1).
ln
I ¼ kt I0
ð1Þ
The effect of temperature on the degradation rate constants was expressed by the Arrhenius equation, in which the temperature dependency of k is quantified by the activation energy Ea (kJ mol1). Estimation of Ea was conducted using linear regression using Eq. (2), with k the degradation rate constant at temperature T, krefT the degradation rate constant at temperature Tref (120 °C) and R the universal gas constant (8.314 J K1 mol1).
ln
k krefT
¼
Ea 1 1 R T T ref
ð2Þ
Parameter estimation was performed using R statistical computing software (V2.15.1, R Foundation, Auckland, New Zealand) and in-house R scripts. 2.6.6. Identification process To identify the most heat labile molecular entities, both UV/Vis spectra and MS data (accurate mass, isotopic distribution, and fragmentation pattern in negative-ion mode) were employed. UV/Vis spectra were used to ascertain the phenolic nature and subclass. The fragmentation patterns were employed to provide further evidence for the characterization of the heat labile molecular entities. For in-source fragments that contain the intact A- and B-ring of flavonoid aglycone skeleton, the common nomenclature system
100 y 1:109 x
ð3Þ
with n number of carbon atoms, x intensity of the 12C isotope and y the intensity of the corresponding 13C isotope. The result was rounded to the nearest integer. The mass accuracy d (ppm) was calculated with the formula (Eq. (4)):
d¼
me ma 106 me
70
n¼
Maximum shift [scans]
was used in which ions were labelled as i,jA and i,jB with superscripts i and j indicating the position of bonds broken in the C-ring. 13 C isotopes were used to determine the charge state (relative distance between 12C and 13C isotope) and to verify the number of carbon atoms by using the following formula (Eq. (3)):
614.64
610.97
615.23
617.29
50
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
614.93
614.75
994.86
615.79
30
180
463.37
560.44
591.38
591.58
15
30
60
ð4Þ
with me the exact m/z (theoretical m/z) and ma the accurate m/z (observed peak average m/z value). Some high abundant phenolic compounds were unambiguously identified by comparison of retention time, UV/Vis spectrum and fragmentation pattern with a commercial standard. 3. Results and discussion
10
Maximum shift per 100 scans [scans]
Fig. 2. Z-transformed average correlation coefficients (Zcorr) indicating the quality of alignment.
3.1. Data pretreatment For the extraction of relevant information from the obtained chromatograms by means of an untargeted metabolomics approach, in our case, the variation in concentration of cloudy apple juice (13.8 °Brix, pH = 3.54) phenolic constituents of as a function of time and temperature (‘thermally induced variation’), several pitfalls inherent to large-scale LC–MS metabolomics should be circumvented. First, the presence and variation (drift) of the chromatograms baseline offset hampers a meaningful signal analysis since background is added to the real signal. Hereby, peak height calculations yield erroneous results with the magnitude of the error depending on the relative severity of the local baseline offset (Lämmerhofer & Weckwerth, 2013). Although baseline offset and drift were found to be minimal due to the proper tuning of the employed mass spectrometer, the baselines of all obtained chromatograms were zero-centred throughout each run before subsequent analysis (De Paepe et al., 2013). Second, another chromatogram related issue that can obscure the extraction of ‘thermally induced variation’, is the presence of retention time drift. Peaks from the same molecular entity are expected to appear at the same retention time. However, variation in retention times caused by small variations in run conditions (further called ‘misalignment’) may occur, even when samples run close together temporally (Lämmerhofer & Weckwerth, 2013). Misalignment may give rise to erroneous assignment of the observed peaks. Correcting for misalignment (alignment) was conducted by a peak-table-based approach wherefore the parameters were optimised according to the procedure for peak alignment in chromatographic sample profiling (Lee et al., 2012; Peters et al., 2009). In accordance with Peters et al., 2009, the alignment outcome did not change by varying the peak slope factors or peak threshold factors around reasonable values (data not shown). However, it was found that the alignment result was largely influenced by the ‘maximum shift’ restricting the retention time shift over all chromatograms (used for calculating the retention shift profile) and ‘maximum shift per 100 scans’ capping the change in retention time shifts in the retention shift profile. The studied parameter combinations and corresponding Z-transformed average correlation coefficients (Zcorr) that indicate the quality of alignment are given in Fig. 2. The best found alignment output
contained 15,168 mass peaks with a Zcorr of 995. The corresponding peak intensity table was used for further processing. Third, quantitative comparison across samples of different measurement series is hampered by the lack of reproducible ion intensities. Just as described above for misalignment, drift in signal intensities between and within the measurement series result in an incorrect evaluation of the tendency of molecular entities throughout the samples and incorrect estimation of kinetic parameters. Fig. 3a and b presents the box-whisker plots for the Briggsian log intensity distribution of the untreated (t = 0) apple juice replicates before and after normalisation, acquired during the two measurement blocks. It is clearly shown that normalisation removes the systematic block effects seen in the original data. From the box-whisker plots of the coefficient of variation distribution, a significant difference was observed between the normalised and nonnormalised data indicating a reduction of the spread in the pooled QC dataset after normalisation (Fig. 3c). It is important to stress that normalisation only corrects for systematic variation between the measurement blocks and not for random errors caused by sample preparation and measurement instrument instabilities. Strictly speaking, one can start with the identification and analysis of the thermally induced variation after the normalisation process. However, the normalised peak intensity table containing 151,168 mass peaks was characterised by different kinds of data redundancy which hampers both processes. A data reduction workflow was therefore meaningful (Fig. 1). A first category of data redundancy was the presence of poorly shaped mass peaks eluting at the beginning/end of the chromatograms whereof the removal lead to a mass peak count of 127,572. Furthermore, due to the fact that the peak-picking parameters of MetAlign™ were selected for sensitive detection of low-intensity peaks with the view to identify fast degrading phenolic constituents, the peak intensity table contains a large amount of entities whereof the intensity through all the studied samples does not exceed the corresponding noise level (MetAlign™ noise table). Eliminations lead to a peak intensity table consisting of 15,864 entities. Moreover, it was observed that several types of ions with different mass numbers, such as in-source fragments, adducts, and isotope ions, were recognised as distinctive peaks and recorded in different rows in the matrix. Deconvolution by means of the
181
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
b 8
7
7
6
5
4
c 0.20
0.15 Coefficient of variation [-]
Briggsian log peak intensity [-]
Briggsian log peak intensity [-]
a 8
6
5
4
0.10
0.05
2
2
0.00 _80dc.0s.1 _80dc.0s.2 _80dc.0s.3 _85dc.0s.1 _85dc.0s.2 _85dc.0s.3 _90dc.0s.1 _90dc.0s.2 _90dc.0s.3 100dc.0s.1 100dc.0s.2 100dc.0s.3 120dc.0s.1 120dc.0s.2 120dc.0s.3 135dc.0s.1 135dc.0s.2 135dc.0s.3 140dc.0s.1 140dc.0s.2 140dc.0s.3 145dc.0s.1 145dc.0s.2 145dc.0s.3
3
_80dc.0s.1 _80dc.0s.2 _80dc.0s.3 _85dc.0s.1 _85dc.0s.2 _85dc.0s.3 _90dc.0s.1 _90dc.0s.2 _90dc.0s.3 100dc.0s.1 100dc.0s.2 100dc.0s.3 120dc.0s.1 120dc.0s.2 120dc.0s.3 135dc.0s.1 135dc.0s.2 135dc.0s.3 140dc.0s.1 140dc.0s.2 140dc.0s.3 145dc.0s.1 145dc.0s.2 145dc.0s.3
3
Before
After
Fig. 3. Box-whisker plots for the Briggsian log intensity distribution of the untreated (t = 0) apple juice replicates before (a) and after (b) normalisation and coefficient of variation distribution of the normalised and non-normalised dataset (c).
Metalign™ output compatible multivariate mass spectra reconstruction (MMSR) combines deprotonated molecular ions with all putative corresponding in-source fragments, adducts and isotopes based on the strong positive correlation between them (throughout the samples) within a certain retention time window (Tikunov et al., 2005). By using this method, it has been observed that several explainable fragments, adducts and isotopes of some expected compounds (isoquercetin, phloridzin, procyanidin B2 etc.) could be combined. After deconvolution, a peak intensity table containing 973 molecular entities was obtained.
3.2. Identification Due to the fact that the objective of this study was to look for the most thermally labile phenolic compounds in cloudy apple juice, the intensity table was further reduced by selecting those compounds whereof the accurate mass and corresponding elemental composition matched with entities in the custom-made phenolic database. Doing so, a final shortlist of 775 potential phenolic constituents was acquired whereof the kinetic curves were plotted (Supplementary Data 1). The molecular entities exhibiting a singlestep degradation (e.g. procyanidins and quercetin glycosides) trend at all studied temperatures were separated by visual inspection from those which demonstrate a consecutive degradation (e.g. quercetin aglycone) trend throughout the studied temperatures leading to a final list of 39 entities which are highly susceptible to thermal degradation. Verification of the phenolic nature of 42 selected thermolabile putative phenolic compounds was done by a detailed identification study summarized in Supplementary Data 2. The peak identification information includes retention time, absorbance maxima in the UV/Vis range (undetectable absorbance is represented by ‘‘’’), accurate m/z of the molecular ion, in-source fragments/ adducts and 13C isotope (of molecular adducts [MH], [2MH] and [M2H]2) mass deviation (ppm), the predicted formula for the detected deprotonated molecular ions, and the tentative assignment for this formula. Unambiguous structure assignment was hampered by several constraints. First, as a result of the complexity of the spectra, peak resolution was not complete in many cases and thus the UV/Vis spectra of some compounds might hide the spectra of other analytes. Furthermore, the UV/Vis spectra of trace compounds could not be clearly assigned, which is also the case for in-source fragments and adducts. Moreover, due to the fact that phenolic
compounds can exist as different constitutional isomers, it was impossible to assign unambiguously the structure of these isomers either. However, by the use of a high-resolution mass spectrometer, the high degree of in-source fragmentation, the quality of deconvolution and the employed database it was possible to achieve a high degree of structural elucidation bordering the limits of the mass spectrometry. 3.2.1. Phenolic acids Some conjugated forms of hydroxybenzoic acids and hydroxycinnamic acids (caffeic acid, p-coumaric acid), were identified as heat labile phenolic compounds (Supplementary Data 2). Based on their backbone structure, they could be easily distinguished from flavonoids during untargeted phenolic compound analysis. The benzoic acids containing a seven carbon atom (C6–C1) backbone, while cinnamic acids, comprising nine carbon atoms (C6–C3). Furthermore, a distinction could be made between hydroxybenzoic acids and hydroxycinnamic acids based on their corresponding UV absorption spectra: hydroxybenzoic acids are represented by a kmax around 280 nm, while hydroxycinnamic acids have a kmax between 305 and 330 nm and a shoulder between 290 and 300 nm (Abad-García, Berrueta, Garmon-Lobato, Gallo, & Vicente, 2009). The in-source fragmentation spectra of the putative identified hydroxybenzoic acids (Table 1, RO 31 and RO 33) are characterised by the neutral loss of CO2, as well as losses of CH3, typical when a methoxyl groups is present. By the absence of product ions generated through the cleavage of the sugar ring C–C bound, the type of the sugar unit could only be deduced from the difference in elemental composition of the molecular ion [MH] and the corresponding aglycone moiety [Y0]-. In the case of esterified hydroxycinnamic acids, p-coumaroylquinic acid (pCQA) and 3O-caffeoylquinic acid (CQA) (Table 1, RO 37 and RO 37) neutral losses of H2O, CO2 and C3H2O2 were observed. Both have m/z 191.056 in common: quinic acid. The quinic acid unit cleaved extensively into fragments, all a result of neutral loss of H2O and CO and combinations of both. CQA and pCQA were unambiguously identified by direct comparison with analytical standards. 3.2.2. Flavonoids Several flavonoids belonging to the subclasses dihydrochalcones and flavonols were identified as well. Due to the fact that each flavonoid subclass exhibited different characteristic UV absorption bands, subclass elucidation prompted by the observed
RO
Identification
1
145,669
2
145,668
3
145,666
4
141,869
5
141,442
6
141,861
7
98,367
8
105,625
9
98,368
10
149,272
11
60,050
12 13
125,489 145,744
14
125,116
15 16
87,755 29,816
17
125,118
18 19 20 21 22
94,484 115,484 98,366 97,938 101,735
23 24 25 26 27 28
122,341 115,487 142,629 108,418 97,650 105,621
Procyanidin hexamer I (B-type) Procyanidin hexamer II (Btype) Procyanidin hexamer III (Btype) Procyanidin pentamer derivative Procyanidin pentamer I (Btype) Procyanidin derivative Quercetin-3-Oarabinoside* Quercetin-3-Oglucoside* Quercetin-3-Oxyloside* Procyanidin tetramer I (B-type) Procyanidin tetramer II (Btype) Unknown Procyanidin trimer (B-type) Methylated procyanidin dimer I (A-type) Unknown Hydroxycinnamic acid hexoside Methylated procyanidin dimer II (A-type) Unknown Unknown Unknown Unknown Quercetin-3-Orhamnoside* Unknown Unknown Unknown Unknown Unknown Quercetin-3-O-
k (102 min1)
Ea (kJ mol1)
85 °C
90 °C
100 °C
120 °C
135 °C
140 °C
145 °C
0.33 ± 0.02 (0.960)
0.56 ± 0.034 (0.964)
1.58 ± 0.014 (0.999)
10.6 ± 0.02 (0.976)
39.4 ± 0.0079 (0.993)
59.6 ± 0.0087 (0.988)
89.5 ± 0.009 (0.969)
116 ± 0.017 (0.999)
0.28 ± 0.017 (0.964)
0.47 ± 0.023 (0.975)
1.22 ± 0.014 (0.998)
7.26 ± 0.0065 (0.979)
24.8 ± 0.0037 (0.996)
36.6 ± 0.007 (0.978)
53.6 ± 0.0098 (0.908)
109 ± 0.04 (0.999)
0.22 ± 0.017 (0.954)
0.39 ± 0.015 (0.988)
1.09 ± 0.017 (0.998)
7.36 ± 0.012 (0.995)
25.6 ± 0.018 (0.985)
38.0 ± 0.0093 (0.991)
55.9 ± 0.021 (0.969)
117 ± 0.98 (0.997)
–
0.90 ± 0.023 (0.988)
2.05 ± 0.014 (0.999)
9.47 ± 0.009 (0.976)
29.9 ± 0.0042 (0.997)
43.0 ± 0.0042 (0.994)
61.4 ± 0.0047 (0.999)
93.3 ± 1.44 (0.991)
–
0.59 ± 0.025 (0.964)
1.47 ± 0.016 (0.998)
8.02 ± 0.0045 (0.992)
28.1 ± 0.018 (0.998)
41.8 ± 0.0091 (0.972)
61.6 ± 0.012 (0.992)
104 ± 1.21 (0.995)
–
0.45 ± 0.016 (0.983)
1.15 ± 0.025 (0.994)
6.61 ± 0.0061 (0.978)
23.2 ± 0.0066 (0.968)
34.6 ± 0.0093 (0.972)
51.1 ± 0.011 (0.987)
106 ± 0.93 (0.997)
–
0.37 ± 0.015 (0.978)
0.72 ± 0.016 (0.995)
2.53 ± 0.0042 (0.995)
6.48 ± 0.011 (0.996)
8.74 ± 0.03 (0.980)
11.7 ± 0.054 (0.996)
75.9 ± 1.38 (0.988)
–
0.33 ± 0.019 (0.957)
0.79 ± 0.025 (0.989)
3.88 ± 0.0071 (0.978)
11.7 ± 0.0089 (0.976)
16.5 ± 0.0058 (0.980)
23.2 ± 0.0046 (0.992)
97.3 ± 0.12 (0.999)
–
0.31 ± 0.019 (0.966)
0.8 ± 0.025 (0.990)
4.65 ± 0.006 (0.956)
15.5 ± 0.0037 (0.990)
22.7 ± 0.012 (0.991)
32.9 ± 0.0093 (0.998)
107 ± 0.094 (0.999)
–
–
0.93 ± 0.017 (0.994)
5.68 ± 0.015 (0.987)
22.5 ± 0.0082 (0.996)
34.9 ± 0.01 (0.987)
53.4 ± 0.0099 (0.992)
111 ± 1.88 (0.989)
–
–
0.87 ± 0.033 (0.968)
5.33 ± 0.0076 (0.986)
23.3 ± 0.0032 (0.956)
37.1 ± 0.012 (0.979)
58.5 ± 0.0047 (0.971)
110 ± 3.32 (0.967)
– –
– –
0.71 ± 0.012 (0.995) 0.50 ± 0.01 (0.995)
2.79 ± 0.017 (0.985) 5.75 ± 0.03 (0.989)
8.51 ± 0.036 (0.924) 30.9 ± 0.019 (0.993)
12.1 ± 0.0085 (0.991) 52.6 ± 0.0076 (0.998)
17.1 ± 0.024 (0.984) 88.5 ± 0.029 (0.971)
82.8 ± 2.71 (0.961) 149 ± 0.069 (0.999)
–
–
0.49 ± 0.017 (0.983)
2.73 ± 0.024 (0.97)
9.67 ± 0.027 (0.978)
14.4 ± 0.049 (0.932)
21.4 ± 0.033 (0.940)
105 ± 1.08 (0.996)
– –
– –
0.44 ± 0.027 (0.970) 0.38 ± 0.025 (0.952)
1.72 ± 0.017 (0.963) 5.36 ± 0.014 (0.955)
4.20 ± 0.038 (0.926) 31.0 ± 0.0076 (0.992)
5.57 ± 0.05 (0.967) 54.2 ± 0.0058 (0.986)
7.35 ± 0.047 (0.932) 93.3 ± 0.0055 (0.976)
83.1 ± 0.56 (0.998) 162 ± 1.02 (0.998)
–
–
–
3.63 ± 0.027 (0.959)
12.9 ± 0.01 (0.972)
19.2 ± 0.03 (0.941)
28.5 ± 0.018 (0.99)
107 ± 0.96 (0.997)
– – – – –
– – – – –
– – – – –
1.76 ± 0.0049 (0.997) 1.44 ± 0.011 (0.978) 1.1 ± 0.0047 (0.989) 0.86 ± 0.0063 (0.979) 0.77 ± 0.0081 (0.957)
5.05 ± 0.019 (0.987) 4.40 ± 0.016 (0.987) 5.98 ± 0.0036 (0.999) 3.12 ± 0.0054 (0.994) 3.2 ± 0.0096 (0.997)
7.06 ± 0.025 (0.975) 6.27 ± 0.014 (0.989) 10.2 ± 0.065 (0.935) 4.70 ± 0.0071 (0.986) 5.02 ± 0.029 (0.988)
9.8 ± 0.016 (0.984) 8.87 ± 0.0071 (0.996) 17.2 ± 0.0074 (0.999) 7.01 ± 0.0056 (0.996) 7.81 ± 0.0092 (0.997)
87.1 ± 1.13 (0.994) 85 ± 2.32 (0.973) 124 ± 4.54 (0.952) 106 ± 1.57 (0.992) 169 ± 6.8 (0.942)
– – – – – –
– – – – – –
– – – – – –
0.75 ± 0.0042 (0.988) 0.72 ± 0.006 (0.973) 0.70 ± 0.01 (0.958) 0.61 ± 0.0065 (0.957) 0.49 ± 0.0053 (0.956) 0.30 ± 0.0055 (0.950)
2.76 ± 0.0044 2.13 ± 0.0077 2.81 ± 0.0084 2.31 ± 0.0067 1.90 ± 0.0076 1.69 ± 0.0037
4.18 ± 0.011 (0.961) 3.00 ± 0.016 (0.988) 4.38 ± 0.012 (0.987) 3.53 ± 0.012 (0.990) 2.92 ± 0.02 (0.977) 2.91 ± 0.011 (0.971)
6.26 ± 0.0049 (0.996) 4.20 ± 0.016 (0.989) 6.74 ± 0.0059 (0.995) 5.33 ± 0.0043 (0.996) 4.44 ± 0.0088 (0.993) 4.95 ± 0.017 (0.979)
86.2 ± 4.91 (0.989) 90.7 ± 0.77 (0.997) 110 ± 2.1 (0.986) 108 ± 1.84 (0.989) 95.3 ± 3.48 (0.952) 162 ± 1.65 (0.996)
(0.980) (0.995) (0.986) (0.997) (0.994) (0.967)
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
PN
182
Table 1 Kinetic parameters (±standard error of regression) for the isothermal degradation of the discovered heat labile phenolic constituents in cloudy apple juice for the studied temperatures 85–145 °C, estimated by means of a first-order kinetic model via 2 step regression. k = degradation rate coefficient, Ea = activation energy. Between the brackets, the Pearson’s correlation coefficient is indicated. The entities are sorted based on the cut-off temperature whereupon a significant (p < 0.001) degradation was observed in the studied time range of 7200 s, and the value of the reaction rate coefficient (in decreasing order). RO = rank order, PN = peak number from peak intensity table.
65.8 ± 8.79 (0.996) 107 ± 3.73 (0.993) 2.16 ± 0.012 (0.972) 1.18 ± 0.0051 (0.970) 1.40 ± 0.027 (0.998) 0.80 ± 0.012 (0.969) – – – – – –
26,161
126,352 26,148
124,456
122,340 99,463
73,356
115,458 77,705
31
32 33
34
35 36
37
38 39
Identification confirmed by comparison with authentic reference material: I, II, and II different isomers. *
– – – –
– –
– –
0.90 ± 0.012 (0.995) 0.53 ± 0.004 (0.939)
67.8 ± 5.31 (0.953) 1.33 ± 0.021 (0.978) 0.98 ± 0.006 (0.995)
1.99 ± 0.032 (0.997)
67.2 ± 9.57 (0.965) 140 ± 2.08 (0.999) 2.96 ± 0.013 (0.992) 2.82 ± 0.0038 (0.997) 1.86 ± 0.0082 (0.994) 1.76 ± 0.011 (0.983) 1.15 ± 0.014 (0.966) 1.08 ± 0.005 (0.993) – –
124 ± 0.97 (0.999) 3.13 ± 0.0073 (0.991) 2.03 ± 0.017 (0.971) 1.31 ± 0.0036 (0.988) – –
–
–
75.9 ± 9.02 (0.952) 180 ± 14.8 (0.997) 2.88 ± 0.0051 (0.975) 28.6 ± 0.0046 (0.965) 1.77 ± 0.0071 (0.978) 11.2 ± 0.0068 (0.957) 1.08 ± 0.0058 (0.983) 4.30 ± 0.012 (0.991) – – – –
– –
0.23 ± 0.0036 (0.959) –
84.3 ± 8.49 (0.922) 3.00 ± 0.013 (0.963) 1.86 ± 0.023 (0.907) 1.14 ± 0.0052 (0.966) 0.24 ± 0.0059 (0.909) – – –
– – 95,228 102,705 29 30
galactoside* Unknown 30 Hydroxyphloretin20 -O-glucoside Methoxybenzoic acid hexoside Unknown Methoxybenzoic acid hexoside 30 Hydroxyphloretin 20 -O-xylosylglucoside Unknown Phloretin-20 -Oglucoside* p-Coumaroyl quinic acid Unknown 3-OCaffeoylquinic acid*
– –
– –
0.29 ± 0.012 (0.996) 0.26 ± 0.014 (0.963)
1.06 ± 0.0058 (0.983) 1.01 ± 0.0053 (0.929)
1.60 ± 0.0081 (0.955) 1.54 ± 0.017 (0.944)
2.38 ± 0.0045 (0.993) 2.33 ± 0.0056 (0.991)
75.7 ± 6.01 (0.907) 120 ± 2.44 (0.998)
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
183
UV/Vis spectrum could be employed: flavonols presented an intense band A at 370 nm (associated with cinnamoyl system) which is absent for flavan-3-ols and dihydrochalcones. The latter show an intense band at 280 nm, associated with benzoyl systems (Gomez-Romero et al., 2010). Some quercetin glycosides susceptible to degradation were elucidated (Supplementary Data 2). All have m/z 301.0.36 (aglycon after heterolytic saccharide cleavage, [Y0]) in common. The type of sugar was established by the mass difference between the glycoside and the corresponding aglycone: a difference of 132 amu for pentose (xylose, arabinose), 146 amu for deoxyhexose (rhamnose) and 162 amu for hexose (glucose, galactose). Furthermore, several quercetin characteristic fragments were found: m/z 300.027 (radical anion aglycone moiety after homolytic saccharide cleavage, [Y0H]), m/z 273.04140 (loss of CO from [Y0]), m/z 271.025 (quinone formation subsequent to the loss of CO from [Y0H]), m/z 255.030 (loss of H2O and CO from [Y0]) (Abad-García et al., 2009). Furthermore, retro Diels–Alder cleavage (RDA) of C-ring bonds resulting in 2 diagnostic fragments: m/z 151.003 indicates the presence of a 4-keto group in the C-ring, m/z 178.998 demonstrates the presence of a hydroxyl substituent on the 3 position of the C-ring. Quercetin-3-O-arabinoside, quercetin-3-O-xyloside, quercetin-3-O-rhamnoside, quercetin-3-O-galactoside, quercetin3-O-glucoside were unambiguously identified by direct comparison with analytical standards. Besides flavonols, several dihydrochalcones (DHC) and hydroxydihydrochalcones (HDHC) were unmasked as heat-sensitive phenolic compounds. Both phloretin and hydroxyphloretin conjugates were found. The type of sugar was established by the mass difference between the glycoside and the corresponding aglycone: 162 amu for hexose, 294 amu for hexose + pentose. Elucidation of the aglycone backbone and its substitution pattern was based on a decision tree for the identification of DHC/HDHC by UV and negative Atmospheric-Pressure Chemical Ionization (APCI) mass spectrometry, because previous published work dealing with the fragmentation of DHC/HDHC by ESI and APCI show very similar structurally informative fragment ions. Maximum absorption bands around 230 and 290 were an indication of the DHC/HDHC or flavanone nature of the molecular entities. A distinction between flavanone equivalents for the same aglycone accurate mass was made based on the occurrence of H2O neutral loss of the aglycone structure: m/z 255.065 (DHC) and m/z 271.065 (HDHC) (Portet et al., 2008). Determination the hydroxylation of the A-ring was based on the RDA fragments m/z 167.035 ([2A+H]), m/z 151.003 ([3AH], [2A–CH3–H]) and m/z 125.024 ([4A + H]) which have DHC and HDHC in common. Other observed RDA fragments [2B] (DHC: m/z 107.050; HDHC: m/z 123.045) and [3B + H] (DHC: m/z 93.034; HDHC: m/z 109.029) were used to distinguish between DHC and HDHC. Phloretin-20 -O-glucoside was unambiguously identified by direct comparison with an analytical standard. 3.2.3. Condensed tannins Several A-type and B-type procyanidins were tentatively identified (Supplementary Data 2). The subsequent oligomeric procyanidin B-type oligomeric procyanidin dimers, trimers, tetramers, pentamers and hexamers built up by epi(catechin) subunits give deprotonated ions at m/z 577.135, m/z 865.198, m/z 1353.261, m/z 1441.325, and m/z 1729.388, respectively. The tendency of formation of multiple charged ions was observed for tetramers and higher: m/z 576.125 (tetramers), m/z 720.159 (pentamers), and m/z 864.192 (hexamers). Analysis of the fragments obtained by in-source RDA fission, heterocyclic ring fission (HRF), and quinone methide (QM) fission enabled tentative identification (Cheng, Wan, Li, & Qi, 2011). The main fragmentation pathways of dimers in the negative mode (m/z 577.135) include a RDA cleavage of the heterocyclic
184
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185
C-ring of the upper flavanols-monomer resulting in a neutral loss of C8H8O3. This 152 Da neutral loss indicates that the B ring of the upper unit has a catechol group. When ionised, this 425.088m/z fragment will loose a molecule of water, producing the product ion at m/z 407.077. Furthermore heterocyclic ring fission (HRF) leads to a C6H6O3 neutral loss (m/z 451.104). In addition, cleavage of the interflavanoid linkages was observed leading to an upper unit corresponding to (epi)catechin monomer (m/z 289.072) and a lower unit corresponding to a monomer with additional desaturation, respectively (m/z 287.056). Similar fragmentation behaviour was found for trimeric (DP3) and tetrameric (DP4) analogues: at m/z 739.168 (DP3) and 1027.229 (DP4) from the loss of an HRF fragment (126 Da) at m/z 713.152 (DP3) and m/z 1001.216 (DP4) from the loss of an RDA fragment, and at m/z 577.136 (DP3) and m/z 865.198 (DP4). The latter fragment shows the same fragmentation behaviour as dimeric and trimeric B-type procyanidins respectively. Only QM fission was observed for pentameric and hexameric analogues. 3.3. Thermal stability Kinetic parameters (reaction rate coefficients and activation energies) for the 42 discovered thermolabile constituents are given in Table 1. The entities are sorted based on the cut-off temperature whereupon a significant (p < 0.001) degradation was observed in the studied time range of 7200 s, and the value of the reaction rate coefficient (in decreasing order). Below 85 °C, no significant difference (p < 0.001) was observed for all discovered thermal depleting compounds. R2 values were higher than 0.95, indicating a good fit to the first-order kinetic model. It could be clearly observed that procyanidins oligomers represent the most heat labile phenolic constituents in cloudy apple juice: the top 10 of the most heat sensitive phenolic compounds contains 7 procyanidin oligomeric compounds, mainly B-type procyanidins (Table 1). Furthermore, it could be clearly observed that the thermal resistance increased with a decreasing degree of polymerization (DP). The same tendency was already observed during thermal processing of peaches (Hong, Barrett, & Mitchell, 2004) and grape and blueberry pomace (Khanal, Howard, & Prior, 2010). The heat sensitivity of procyanidins does not have to be experienced as disadvantageous. High DP procyanidins hydrolyze under the influence of temperature into monomers ((+)-catechin and ()-epicatechin) and dimeric compounds, such as procyanidin B2. A remarkable increase could be demonstrated for (+)-catechin and ()-epicatechin (from 90 °C) and for procyanidin B2 (from 100 °C) (Supplementary Data 3). An in vivo study performed with ileostomy patients who ingested apple juice, has demonstrated that procyanidins will pass the stomach and reach the colon intact, independent of DP (Kahle et al., 2007). However, a higher gastrointestinal absorption of procyanidins dimers and monomers compared to higher DP oligomers was observed (Appeldoorn, 2009). Also some quercetin glycosides were identified, in a greater or lesser extent, as thermolabile (Table 1). A clear difference in degradation rate between the various glycosides was observed. The stability decreased in the order: quercetin-3-O-galactoside > quercetin-3-O-rhamnoside > quercetin-3-O-xyloside > quercetin-3O-glucoside > quercetin-3-O-arabinoside which is in accordance with earlier findings (van der Sluis et al., 2005). This order could be explained by the different susceptibility of the glycosidic bond to hydrolysis in an acidic environment. The most heat-resistant compounds observed in this study belong to hydroxycinnamic acids and dihydrochalcone classes. Especially the high abundant phloretin-2’-O-glucoside, and 3-O-caffeoylquinic acid were found to be very resistant to thermal degradation, which is in accordance with earlier studies (van der Sluis et al., 2005).
4. Conclusion The research described in this paper demonstrates the high applicability of untargeted metabolomics approach in process optimisation with the aim to retain bioactive phenolic constituents in cloudy apple juice. The developed workflow consisting of baseline correction, alignment, filtering and deconvolution, made it possible to gather kinetic information from all detected putative phenolic constituents. Furthermore, by the use of a high resolution mass spectrometer, the high degree of in-source fragmentation, the quality of deconvolution and the employed custom-made database, it was possible to achieve a high degree of structural elucidation for the thermolabile phenolic constituents. By means of this holistic approach, a clear difference in degradation susceptibility between the phenolic subclasses presented in cloudy apple juice could be observed. Several procyanidin representatives were found to be the most heat labile phenolic compounds of cloudy apple juice. Although 42 compounds were identified as susceptible to thermal degradation, kinetic parameters indicate a low impact of the HTST pasteurisation (typical 85 °C, 30 s; 90 °C, 15 s) applied in industry on the phenolic content of cloudy apple juice.
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodchem.2014. 04.005.
References Abad-García, B., Berrueta, L. A., Garmon-Lobato, S., Gallo, B., & Vicente, F. (2009). A general analytical strategy for the characterization of phenolic compounds in fruit juices by high-performance liquid chromatography with diode array detection coupled to electrospray ionization and triple quadrupole mass spectrometry. Journal of Chromatography A, 1216, 5398–5415. Abranko, L., Garcia-Reyes, J. F., & Molina-Díaz, A. (2011). In-source fragmentation and accurate mass analysis of multiclass flavonoid conjugates by electrospray ionization time-of-flight mass spectrometry. Journal of Mass Spectrometry, 46, 478–488. Aguilar-Rosas, S. F., Ballinas-Casarrubias, M. L., Nevarez-Moorillon, G. V., MartinBelloso, O., & Ortega-Rivas, E. (2007). Thermal and pulsed electric fields pasteurization of apple juice. Effects on physicochemical properties and flavour compounds. Journal of Food Engineering, 83, 41–46. Appeldoorn M. (2009). Dietary A- and B-type procyanidins: Characterization and biofunctional potential of an abundant and diverse group of phenolics. PhD thesis Wageningen University, The Netherlands. Awuah, G. B., Ramaswamy, H. S., & Economides, A. (2007). Thermal processing and quality: Principles and overview. Chemical Engineering and Processing: Process Intensification, 46, 584–602. Braun, J., Gugisch, R., Kerber, A., Laue, R., Meringer, M., & Rücker, C. (2004). MOLGEN-CID – A canonizer for molecules and graphs accessible through the internet. Journal of Chemical Information and Computer Sciences, 44, 542–548. Buchner, N., Krumbein, A., Rohn, S., & Kroh, L. W. (2006). Effect of thermal processing on the flavonols rutin and quercetin. Rapid Communications in Mass Spectrometry, 21, 3229–3235. Cheng, X. L., Wan, J. Y., Li, P., & Qi, L. W. (2011). Ultrasonic/microwave assisted extraction and diagnostic ion filtering strategy by liquid chromatography– quadrupole time-of-flight mass spectrometry for rapid characterization of flavonoids in Spatholobus suberectus. Journal of Chromatography A, 1218, 5774–5786. D’Amico, D. J., Silk, T. M., Wu, J., & Guo, M. (2006). Inactivation of microorganisms in milk and apple cider treated with ultrasound. Journal of Food Protection, 69, 556–563. De Paepe, D., Noten, B., De Loose, M., Van Droogenbroeck, B., & Voorspoels, S. (2013). An improved mass spectrometric method for identification and quantification of phenolic compounds in apple fruits. Food Chemistry, 136, 368–375. De Vos, C. H. R., Moco, S., Lommen, A., Keurentjes, J. J. B., Bino, R. J., & Hall, R. D. (2007). Untargeted large-scale plant metabolomics using liquid chromatography coupled to mass spectrometry. Nature Protocols, 2, 778–791. Gomez-Romero, M., Segura-Carretero, A., & Fernandez-Gutierrez, A. (2010). Metabolite profiling and quantification of phenolic compounds in methanol extracts of tomato fruit. Phytochemistry, 71, 1848–1864. Hong, Y. J., Barrett, D. M., & Mitchell, A. E. (2004). Liquid chromatography/mass spectrometry investigation of the impact of thermal processing and storage on peach procyanidins. Journal of Agricultural and Food Chemistry, 52, 2366–2371.
D. De Paepe et al. / Food Chemistry 162 (2014) 176–185 Ioannou, I., Hafsa, I., Hamdi, S., Charbonnel, C., & Ghoul, M. (2012). Review of the effects of food processing and formulation on flavonol and anthocyanin behavior. Journal of Food Engineering, 111, 208–217. Kahle, K., Huemmer, W., Kempf, M., Scheppach, W., Erk, T., & Richling, E. (2007). Polyphenols are intensively metabolized in the human gastrointestinal tract after apple juice consumption. Agricultural and Food Chemistry, 55, 10605–10614. Kessner, D., Chambers, M., Burke, R., Agus, D., & Mallick, P. (2008). ProteoWizard: Open source software for rapid proteomics tools development. Bioinformatics, 24, 2534–2536. Khanal, R. C., Howard, L. R., & Prior, R. L. (2010). Effect of heating on the stability of grape and blueberry pomace procyanidins and total anthocyanins. Food Research International, 43, 1464–1469. Kind, T., & Fiehn, O. (2007). Seven golden rules for heuristic filtering of molecular formulas obtained by accurate mass spectrometry. BMC Bioinformatics, 105. Lämmerhofer, M., & Weckwerth, W. (2013). Metabolomics in practice. Successful strategies to generate and analyze metabolic data (pp 312–314). Weinheim: Wiley-VCH. Lee, J., Park, J., Lim, M. S., Seong, S. J., Seo, J. J., Park, S. M., et al. (2012). Quantile normalization approach for liquid chromatography–mass spectrometry-based metabolomic data from healthy human volunteers. Analytical Sciences, 28, 801–805. Lommen, A., van der Weg, G., van Engelen, M. C., Bor, G., Hoogenboom, L. A., & Nielen, M. W. (2007). An untargeted metabolomics approach to contaminant analysis: Pinpointing potential unknown compounds. Analytica Chimica Acta, 584, 43–49. Matsuda, F., Yonekura-Sakakibara, K., Niida, R., Kuromori, T., Shinozaki, K., & Saito, K. (2009). MS/MS spectral tag-based annotation of non-targeted profile of plant
185
secondary metabolites. The Plant Journal: For Cell and Molecular Biology, 57, 555–557. Niu, S., Xu, P., Fang, Y., Zhang, L., Yang, Y., Liao, X., & Hu, X. (2010). Comparative study on cloudy apple juice qualities from apple slices treated by high pressure carbon dioxide and mild heat. Innovative Food Sciemce and Emerging Technologies, 11, 91–97. Peters, S., van Velzen, E., & Janssen, H. G. (2009). Parameter selection for peak alignment in chromatographic sample profiling: objective quality indicators and use of control samples. Analytical and Bioanalytical Chemistry, 394, 1273–1281. Portet, B., Fabre, N., Rozenberg, R., Habib-Jiwanb, J. L., Moulisa, C., & QuetinLeclercq, J. (2008). Analysis of minor flavonoids in Piper hostmannianum var. berbicense using liquid chromatography coupled with atmospheric pressure chemical ionization mass spectrometry. Journal of Chromatography A, 1210, 45–54. Rupasinghe, H. P. V., & Yu, L. J. (2012). Emerging preservation methods for fruit juices and beverages. In Y. El-Samragy (Ed.), Food additive (pp. 55–77). Rijeka (Croatia): InTech Europe. Tikunov, Y., Lommen, A., de Vos, C. H. R., Verhoeven, H. A., Bino, R. J., Hall, R. D., et al. (2005). Novel approach for nontargeted data analysis for metabolomics: Large-scale profiling of tomato fruit volatiles. Plant Physiology, 139, 1125–1137. van der Sluis, A. A., Dekker, M., & van Boekel, M. A. (2005). Activity and concentration of polyphenolic antioxidants in apple juice. Stability during storage. Journal of Agricultural and Food Chemistry, 53, 1073–1080. Vikram, V. B., Ramesh, M. N., & Prapulla, S. G. (2005). Thermal degradation kinetics of nutrients in orange juice heated by electromagnetic and conventional methods. Journal of Food Engineering, 69, 31–40.
