Real-time Assessment of Fruit Juices Adulteration by Rapid Evaporative Ionization Mass Spectrometry with IKnife Based on Metabolomics

WANG Hui-jun,WEN Wen,CAO Xue-li,HAN Tian

(Beijing Advanced Innovation Centre for Food Nutrition and Human Health,Beijing Technology & Business University (BTBU),Beijing 100048，China)

Abstract:In recent years,fruit juice adulteration has become an increasingly important issue.Rapid evaporative ionization tandem mass spectrometry(REIMS),as a novel method for real-time discrimination of samples,could solve this problem without any sample preparation.In this paper,a metabolomics fingerprinting approach for real-time authentication of orange,apple and grapefruit juices was developed.A discriminate model established by principal component analysis(PCA) and linear discriminate analysis(LDA) was able to distinguish different types of fruit juice with a correct classification of 97.28% in leave-20%-out cross-validation and a 100% burns correctly identified in real-time recognition.Characteristic m/z markers were detected and some of them were identified.Adulterated orange juices with apple and grape juices at different percentages could be well distinguished from the level of 50% down to 10%.Furthermore,the adulterated samples were analyzed by optimized single MS and tandem MS/MS REIMS(REIMS and REIMS/MS).Partial least squares analysis was successfully used to predict the level of the adulterants(both R2 and Q2>0.82 for all models).REIMS/MS could offer higher prediction accuracy compared with the conventional REIMS.

Key words:iKnife;rapid evaporative ionization mass spectrometry(REIMS);metabolomics;fruit juice;adulteration;rapid evaporative ionization tandem mass spectrometry(REIMS/MS)

Nowadays consumers favor healthy diet consumption including a lot of fruits and vegetables.Under this situation,juice markets are steadily growing across developing and developed countries[1].Taking into consideration the large fruit juice quantities consumed worldwide,fruit juice is one of the most common food commodities subject to adulteration and fraud[2].The most frequent procedures within the fruit juice industry,includs dilution with water,addition of sugars,pulp wash or other cheaper juices such as apple and grape juices[3-5].Various adulteration techniques make the detection and prevention of juice adulteration very complicated.

A number of analytical methods have been used to identify fruit juice adulteration.The most frequently used method,targeted approach,focuses on the detection and identification of a particular compound or class of compounds,including amino acids,organic acids,sugars,etc[6-9].However,chemicals which are not selected beforehand cannot be detected.Metabolomics,as one of the most recent advanced analytical platforms,could be used to obtain the metabolic chemical profile of food including many unknown components and provide a more efficient and powerful tool for food fingerprinting.To process and interpret complex data,a method combining chemometrics with metabolomics gives possibility to explore data more thoroughly and can provide an effective investigative tool and classification technique for food authenticity verification[10].Particularly,metabolomics based on high-throughput mass spectrometry analytical platforms has proven their efficiency in adulteration detection and highlighting candidate biomarkers of fruit juice[4,11-12].However,complicated sample preparation and chromatographic separation are time-consuming and laborious.Therefore,the importance to develop a simple,rapid and reliable technique has increased for both producers and regulators to realize early detection successfully.

Rapid evaporative ionization mass spectrometry(REIMS) is a novel and rapid analytical technique developed by Takats[13].An electrometric knife,iKnife,enables direct ionization and desorption of lipid by generating aerosol during electrosurgical dissection[14].Without sample preparation,REIMS analysis takes 2-3 s to realize real-time and accurate identification of samples.Up to now,REIMS has received increasing interests and been widely used as an alternative approach for screening food adulteration[15-16].The majority of applications of REIMS technique,such as meat[16-18]and fish[15,19-20],deal with the untargeted profile of lipid fraction.Even the latest study for authenticity assessment of pistachio,a non-animal food,have focused on the comparison of lipid profile[21].Recently,we have demonstrated that this technology can be deployed on honey,a non-lipid food[22].However,its application in non-liquid food with higher water content is not yet known.Moreover,to the best of our knowledge,all research published to date in the application of food using REIMS to perform classification or identification has utilised mass spectrometers in MS-only mode.Studies of Paul Abu-Rabie showed the importance of utilising optimized tandem MS/MS REIMS,REIMS/MS,methodology to increase the discrimination power in tissue and cell samples[23].While,the use of REIMS/MS in the food adulteration has not been investigated.

In this study,a metabolomics approach for authentication of fruit juices(orange,apple and grape juices) by iKnife coupled with REIMS was explored.The aim was to assess the feasibility of this technology to detect juice-to-juice adulteration.Orange juice,as high value juice and most popular fruit juices consumed globally,was chosen as model juices for study.The secondary aim was to further demonstrate comparison of REIMS and REIMS/MS methodologies to investigate which mode produces maximum discrimination power for detecting the adulteration level.

1 Experimental

1.1 Juice samples

Three types of commercial juices(orange,apple and grape) purchased from retail markets were used in this study(four for each type of fruit). No sugars were added in the pulp free juices.Collected samples represented both fresh-pressed and concentrated juice,produced in various countries(China,Cyprus and Spain).Besides,fresh fruits(two for each type of fruit) were purchased and immediately processed to juice samples in a home fruit processor.Three fresh fruit juice samples(one for each type of fruit) were employed as validation samples.Simulated adulterate samples were made by adding lab made apple juice and grape juice to orange juice at 10%,20%,30%,40%,and 50%(by volume),respectively.

1.2 Chemicals

DL-malic acid(99.5%),citric acid(99.5%) andD-(-)-tartaric acid(99%) were purchased from Macklin(Shanghai,China).Stock solutions were prepared by dissolving 0.2 g each chemicals in 10 mL ultrapure water,respectively.

Fig.1 REIMS experimental setup for sampling using an iKnife

1.3 REIMS analysis

The experimental setup was described in Fig.1.A commercially available iKnife device(WSD71,Weller,Germany) was used to generate aerosol at 225 ℃.The sampling was operated by holding an iKnife in one hand and a smoke evacuator in the other.When the tip of iKnife touched liquid surface,the liquid boiled and produced aerosol vapor.Each sample was burnt ten times for 2 s to register a meaningful average mass spectrum.Contacting time was controlled by following the rhythm instructed from a metronome.The generated evaporative aerosol was aspirated into the MS via smoke evacuator,a polytetrafluoroethylene(PTFE) tube,by a Venturi pump with 2 bar nitrogen.MS measurements were carried out using a Quadrupole-Time of Flight(Q-TOF) mass spectrometer(Xevo G2-XS,Waters Co.,Ltd.,Wilmslow,UK) coupled with a REIMS interface,and was operated in negative ion mode with mass range set atm/z50-1 200 in sensitivity mode with a scan time of 1 s/scan.The auxiliary solvent of isopropyl alcohol was introduced into a heated helix collision surface by a stainless steel capillary with leucine encephalin(m/z554.261 5,[M-H]-) added for lock mass correction.The mass resolution was approximately 20 000 FWHM over the mass range of interest.For each day of operation,mass spectrometer was calibrated using sodium formate.For instrument cleaning,the disassembled Venturi pump unit was cleaned at the end of each running day through sonication for 20 min in methanol.Other parameters were operated as follows:cone voltage:10 V;heater bias voltage:30 V,flow rate of auxiliary solvent:200 μL/min.

1.4 Data processing and chemometric analysis

Data were acquired using MassLynx 4.1.Processing and analysis(peak detection,data mining,alignment and normalisation) of the data was performed using Progenesis QI software.Multivariate statistical software package LiveIDTMwas used for principal component analysis(PCA) and linear discriminant analysis(LDA) as previously described[22].PCA,an unsupervised method,was used to get a better visualization by projecting the objects of data set into the space of the first few components.Unsupervised PCA followed by supervised LDA was found to reduce the chance of over-fitting that may occur with a pure LDA model.Insilicofive-fold stratified validation,leave-20%-out,was performed to determine the predictive accuracy of this model.Then,the model was employed with training set samples for real-time identification of fruit juice.

Additional and complementary PCA and partial least squares-discriminant analysis(PLS-DA) methods were performed using SIMCA 14.1(Umetrics Sartorius Stedim Biotech,Sweden) to determine candidate biomarkers and percentage of adulteration.R2andQ2are usually used to evaluate the quality and reliability of these models.Generally,their values close to 1.0 indicate an excellent fitness and predictive capability for the model.Permutation test was applied for PLS-DA model validation and testing of the model over fitting.The biomarkers were filtered by the results of variable importance in projection(VIP) analysis.Regression model was constructed by PLS to examine relation between metabolite profiling and level of adulteration.

Selectedm/zmarkers were transferred into Elemental Composition software and FooDB online database in order to obtain the elemental composition and potential molecular structures for markers.The experimental MS/MS data were compared to reference data obtained from MassBank online databases.In addition,fragmentation information of the selected compound was verified using MassFragment software by analyzing MS/MS spectra.The molecular structures of compounds were imported into MassFragment for theoretical fragmentation prediction and matching.Furthermore,authentic standards were also analyzed to verify the identification.

Fig.2 Effects of iKnife temperature on the signal intensity of three selected ion and overall peaks

2 Results and discussion

2.1 Optimization of IKnife conditions

Previous research has established that the most important physical factor for REIMS was the heating rate[24].Since the heating rate is largely determined by the temperature of iKnife,effect of this setting was studied in detail.In the view of signal intensity of overall and three ions,includingm/z133.014 6(malic acid,[M-H]-),149.009 3(tartaric acid,[M-H]-) and 191.019 6(citric acid,[M-H]-) in negative ion mode,the temperature of iKnife was optimized in the range of 200-300 ℃.For the optimization experiments,five replicate burns were carried out per parameter.The results are depicted in Fig.2.The relative standard deviation(RSD) of ion intensity is less than 30%,thus indicating a satisfactory repeatability of manual burns[21].Overall signal intensity enlarged from 200 ℃ to 225 ℃ and reached the highest value at 225 ℃.More by-product was easier generated by caramelization and Maillard reaction at higher temperature.Finally,a temperature of 225 ℃ was chosen.

These three organic acids [M-H]-ion formation is based on the rapid thermal evaporation of the juice material.The ion intensities showed strong dependence on evaporation temperature.Closer inspection of Fig.2 showed that signal intensities of the three selected ions had a clear variation trend with increasing iKnife temperature.Maximal signal intensities ofm/z133.014 6,149.009 3 and 191.019 6 were acquired at 200 ℃,250 ℃ and 300 ℃,respectively.Higher evaporation rate at higher temperature may improve the ionization and disintegration efficiencies of fruit juice.While extreme high temperature may cause decline of the ionization efficiency attributed to thermal degradation.The different ionization efficiencies of the three ions can be tentatively associated with their different ionic characters and desolvation(dehydration) enthalpies[13].

2.2 Chemometric analysis

After background subtraction and removal of lock mass compound peak(m/z554.261 5),typical total ion count chromatograms and representative mass spectra of orange,apple and grape juices are shown in Fig.3.The spectra adjacent to peak was selected as background.In order to evaluate the smoke profiles differences between orange,apple and grape juices,PCA approach was employed.PCA represents one of the most frequently used chemometric tools for initial data overview.Fig.4A and Fig.4B showed PCA score plot and loading plot generated by LiveIDTM.The first two principal components(PCs) explained the majority of variation:PC1 accounts for 40.7% and PC2 for 37.0%,respectively.Leave-20%-out cross-validation of PCA-LDA model resulted in 97.28% correct classification.Real-time identification of three fresh juices,which had not been previously used to generate the chemometric models,were obtained near-instantaneously using “live-recognition” function with 100% burns correctly identified(Fig.4C).Each sample was burnt for five times.

LiveIDTMis a novel software developed for REIMS applications.Another software SIMCA 14.1,one of the most employed software for multivariate statistical processing,was also used for comparison purpose.Pareto scaling was conducted to preprocess the metabolomics data before PCA.Score plots generated by SIMCA(Fig.5A) and LiveIDTM(Fig.4A) softwares showed similar pronounced clustering and significantly differentiation.PC1 and PC2 explained 33.0% and 25.8% of total variance.Next,loading plot of supervised PLS-DA was used to investigate the most differential metabolites among the three juices(R2=0.918 andQ2=0.907)(Fig.5B).Cross-validation with 100 permutation tests indicated that the PLS-DA model was reliable(intercepts ofR2andQ2were 0.04 and -0.28,respectively)(Fig.5C).In total,eleven ions(VIP≥2) were selected as potential characteristic biomarkers in different juices(Fig.5D),and an overview of these markers information is provided in Table 1.

Fig.4 Multivariate statistical analysis of fruit juices(orange,apple and grape) generated by LiveIDTMA.PCA score plot;B.PCA loading plot;C.real-time recognition of juice type

Table 1 Compounds with VIP≥2 between orange,apple and grape juices based of the PLS-DA

2.3 Identification of marker compounds

Identification of marker compounds generated by REIMS represents extraordinary laborious and time-consuming.Without chromatographic separation,all compounds are detected simultaneously.In addition,matrix produces a series of complex substances due to thermal degradation after contact heating at high temperature.

Thanks to the high sensitivity and resolution of Q-TOF/MS instrument,6 of 11 marker ions were identified according to authentic standards,accurate mass,MS/MS spectra,online databases and literatures.For the ion ofm/z133.014 6,Elemental Composition software suggested formula C4H5O5,with a mass error of 2.49 ppm.This molecular formula corresponded to two potential candidates in the FooDB.Experimental MS/MS spectra were compared to the MS/MS spectra of malic acid reported in MassBank,and five common fragments were revealed.In addition,aqueous solution of malic acid reference standard was burnt and both MS and MS/MS spectra was obtained.Thus,its presence in the samples was confirmed.However,in most cases,m/zvalue could not be assigned to one single compound.In the absence of separation methods,it is hard to distinguish the isomers.

Previous studies reported that organic acids showed different profiles in fruit juices providing useful “fingerprints” in the authenticity testing[7,25].For instance,tartaric acid is usually considered as an indicator of grape juice addition to a more expensive juice.Intensity of ionm/z149.009 3(tartaric acid) was relatively high in grape juice samples,indicating that tartaric acid could be used as a potential characteristic marker of grape juice.Similarly,intensities of ionm/z325.183 8 andm/z133.014 6(malic acid) were relatively high in orange and apple juices respectively,but undetectable in other juice samples,indicating the possibility to be used as characteristic markers.

2.4 Identification of juice-to-juice adulteration

Orange juice is a widely consumed and high-volume product that makes it a potential target for economically motivated fraud.One of the most frequent procedures is adding other cheaper fruit juices,typically apple and grape juices.The capability to detect the orange juice adulteration of apple and grape juice with different level(10%,20%,30%,40%,50%) was investigated.PCA-LDA analyses of these adulterated samples are shown in Fig.6A and Fig.6B.Pure orange juice was placed in negative region of LD1,while pure apple and grape juice were placed in positive region.The highest level(50%) was farthest away from pure orange juice,whereas the lowest level(10%) was closest to it.Moreover,the proportion of adulteration increases gradually from negative to positive in the LD1 axis in a near linear tendency.It can be seen that even the orange juices adulterated with 10% apple and grape juices can be distinguished from the pure orange juice samples.The results indicated the potential of REIMS untargeted approach coupled with chemometric analysis in the investigation of juice adulteration.

2.5 Comparison of single MS and tandem MS/MS REIMS methodologies

In MS-only mode,quadrupole and collision cell of Q-TOF/MS was effectively unused(default setting,6 V).Further work was undertaken to investigate the feasibility of using MS/MS approach for the prediction of orange juice adulterations with apple and grape juices at different percentages(10%,20%,30%,40%,50%).

Preliminary REIMS/MS work involved optimization of collision energy(CE) settings.Orange juice was analyzed using REIMS/MS with CE of 2,5,10,15,20,25,30,40,50,60,70,80,90,100,120,140,160,180 and 200 V.Three replicate burns were performed per CE setting.PLS-DA score plot demonstrated clear discrimination of CE settings that lower CE values grouped on the left side of the plot and higher CE value grouped on the right side(Fig.6C).Low CE(CE=15 V) resulted in spectra where the magnitude of fragment ions and parent ions were as equal as possible.High CE(CE=60 V) appeared to be the point at which fragmentation was effectively maximized.It was decided that subsequent experiments should be run using three discrete MS modes:(ⅰ) MS only,as a control;(ⅱ) MS/MS with a low CE of 15 V;(ⅲ) MS/MS with a high CE of 60 V.

PLS regression model was applied to examine relation between metabolite profiling and adulteration level,and one of them was plotted in Fig.6D,as predictedvs.measured values of the adulteration levels were plotted together in order to evaluate the performance of the created prediction linear model.Table 2 showed theR2,Q2,root mean squared error of estimation(RMSEE) and prediction(RMSEP) for the six models.BothR2andQ2values were greater than 0.82 for all models.The best adulteration predictive model with the highest prediction accuracy was obtained from orange-apple adulteration model using low CE REIMS/MS mode in which RMSEP was lowest(4.488).

Furthermore,it was also found that low CE REIMS/MS mode produced better prediction in the same adulteration types compared with the MS-only and high CE MS/MS modes.REIMS/MS mode enables either greater emphasis of the key ions that contribute to class discrimination in multivariate analysis,or the production of different ions not observed using REIMS in MS-only mode[22].As different juices produced characteristically different spectral fingerprints largely based on the organic acid components which were located in the mass ranges ofm/z100-200,high CE MS/MS mode may result in decreased sensitivity for low-mass ions.

Table 2 PLS prediction results obtained from validation data for the adulteration of orange juice with apple juice and grape juice at different percentages(10%,20%,30%,40% and 50%) using three separate MS modes:MS only(CE=6 V),low collision energy MS/MS(CE=15 V) and high collision energy MS/MS(CE=60 V)

3 Conclusions

Within this context,untargeted iKnife coupled to REIMS based metabolomics was demonstrated to be a powerful tool for discrimination and authentication of fruit juices.No sample preparation,accurate and near-instantaneous results were obtained.Moreover,some characteristic chemical constituents responsible for the separation between different juices were identified.The discrimination of juice-to-juice adulteration was achieved at 10% adulteration level.Furthermore,the results of comparison between single MS and tandem MS/MS REIMS methodologies showed that PLS regression models were capable of predicting the adulteration level in different juice types.REIMS/MS offered higher prediction accuracy compared with conventionally used REIMS.For these reasons,this new analytical platform is very promising for other applications in lipid and non-lipid food safety or quality.

Acknowledgments:We are much grateful to Dr.LIN Ji-hong and FU Ti-peng from Waters Corporation for their technical support.