Nuclear magnetic resonance for foodomics beyond food analysis

Trends in Analytical Chemistry 59 (2014) 93–102

Contents lists available at ScienceDirect

Trends in Analytical Chemistry j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / t r a c

Nuclear magnetic resonance for foodomics beyond food analysis ☆ Luca Laghi, Gianfranco Picone, Francesco Capozzi * Department of Agri-Food Sciences & Technologies, Alma Mater Studiorum – University of Bologna, Piazza Goidanich 60, 47521 Cesena, FC, Italy

A R T I C L E

I N F O

Keywords: Authenticity Food analysis Foodomics Food science Metabolomics Molecular profile NMR spectroscopy Nuclear magnetic resonance (NMR) Nutrition Traceability

A B S T R A C T

Nuclear magnetic resonance (NMR) spectroscopy has a long tradition as a powerful platform in the hands of modern food scientists, with several applications related to food safety, traceability and authenticity. The continual advances in instrumental sensitivity and electronic stability, together with rapid growth in new, potent algorithms for multivariate data analysis, facilitate the use of NMR spectroscopy as a competitive, complementary analytical platform for observing the food metabolome. By adapting the holistic views of metabolomics research, foodomics emerges as a new discipline bringing food science and nutritional research closer together. This review mostly focuses on recent efforts dedicated to extraction and interpretation of NMR data, rather than providing technical details about their acquisition. With this aim, we present new trends in the exploitation of the information gained by NMR of food matter. We critically describe and illustrate, via representative examples, the limitations and the counterbalancing advantages of the technique. © 2014 Elsevier B.V. All rights reserved.

Contents 1. 2. 3. 4. 5.

6. 7.

From food safety to food for health through foodomics ........................................................................................................................................................................ 93 Foodomics for a holistic approach to food .................................................................................................................................................................................................. 94 NMR for food analysis and foodomics .......................................................................................................................................................................................................... 94 Levels of information of the metabolome ................................................................................................................................................................................................... 95 Exploring new lands in the foodomics space ............................................................................................................................................................................................ 96 5.1. Standardization and merger of the data obtained through different pulse sequences ................................................................................................. 96 5.2. Merging data obtained at different magnetic fields .................................................................................................................................................................. 96 5.3. Sensitivity ................................................................................................................................................................................................................................................. 98 5.4. Sampling inhomogeneous matter .................................................................................................................................................................................................... 98 5.5. Sample handling .................................................................................................................................................................................................................................... 98 5.6. Peaks overlap .......................................................................................................................................................................................................................................... 98 5.7. Peaks alignment ..................................................................................................................................................................................................................................... 99 5.8. Baseline biases ........................................................................................................................................................................................................................................ 99 5.9. Metadata ................................................................................................................................................................................................................................................... 99 Multivariate analysis ........................................................................................................................................................................................................................................ 100 Conclusions and future perspectives .......................................................................................................................................................................................................... 100 Acknowledgements .......................................................................................................................................................................................................................................... 100 References ............................................................................................................................................................................................................................................................ 100

1. From food safety to food for health through foodomics Nuclear magnetic resonance (NMR) spectroscopy is an investigation technique that, with a minimum sample preparation, offers

☆

This article was originally commissioned for the Special Issue ‘Modern Food Analysis and Foodomics’. * Corresponding author. Tel.: +39 0547 338105; Fax: +39 0547 382348. E-mail address: [email protected] (F. Capozzi). http://dx.doi.org/10.1016/j.trac.2014.04.009 0165-9936/© 2014 Elsevier B.V. All rights reserved.

the possibility to obtain quantitative and structural information of any molecule characterized by atoms with an intrinsic magnetic moment and angular momentum. The elements mainly present in foods, such as H, O, C, N, and P, have at least one detectable isotope, thus granting NMR spectroscopy the title “universal detector”. The widespread application of this high-throughput technique, together with mass spectrometry (MS), is leading to a change in the goals of food analysis, as emerging from the systematic examination of the works published in the past 10 years [1]. The first goal of food analysis has traditionally been, and still is, to ensure food safety. Food

94

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

safety has been flanked by food traceability [2] and food authenticity [3], with the current frontier represented by studies on chemically or genetically fortified foods tailored to promote health [1]. The coverage of recent literature on NMR-based analyses is better in and extensively reported by other dedicated papers [4–8]. The success of these studies passes through evolution from a reductionist approach to food towards the observation of food “as a whole” [9]. For a holistic view of a food, ideally one should: • know the structure and the concentration of all its molecules, by analogy to what has been done in the genetics field with the map of the entire genetic code [10]; • gain information about the metabolic networks and fluxes characterizing the food, where fluxes concern the kinetics description of all the transformations occurring within the same metabolic network [11]; and, • understand how environmental factors or technological treatments modulate food composition. In other words, the description of food composition must have the same high definition reserved to the human being (i.e., with details at the level of its genome, proteome and metabolome). Such a degree of definition, indeed, is required when the link between foods and their health effects must be demonstrated with a clear molecular mechanism. Meta-analysis of the information obtained from all omics, applied to both food and human, is the approach best addressing the mechanistic definition of nutrient activity [12]. In this context, foodomics collects the omics information about food products, which is necessary to define their safety, quality and nutritional value comprehensively.

2. Foodomics for a holistic approach to food Foodomics has been defined as “the discipline that studies the food and nutrition domains through the application of advanced omics technologies in order to improve consumer well-being, health and confidence” [13]. Foodomics traditionally takes advantage of genomics, transcriptomics, proteomics and metabolomics data. Among them, metabolomics is the one geared towards providing an essentially unbiased, comprehensive qualitative and quantitative overview of the metabolites present in an organism [14]. The purpose of foodomics is to define food by applying the omics approaches, because the food is the result of, e.g., selection, production, processing, and storage, on the genome-transcriptomeproteome-metabolome of the originating organisms, or parts of them (Fig. 1). Whilst metabolomics pertains to the study of the metabolome of biological systems, foodomics studies the effect of different factors on the genome, the transcriptome, the proteome and the metabolome of the biological systems during their transformation in food. Then, foodomics continues its role by defining the evolution of food during digestion, absorption and interaction with humans, looking at nutrition and health from the food perspective. Foodomics is ideally positioned for use in many areas of food science [4] for two main reasons: • the metabolome can be considered downstream of genome, transcriptome and proteome, and is therefore the best representation of the food phenotype, so it can give a direct view of the substances that interact with our organisms upon eating; and, • it is known from both metabolic control analysis [15] and experiments [16] that concentration changes of individual active enzymes might be expected to have little effect on the

corresponding metabolic fluxes, but significant effects on the concentrations of other numerous individual metabolites due to cascade effects, feedback action or pleiotropy. An example of the latter point is a recent study describing the perturbation caused by the insertion of one or three copies of the same exogenous gene into the metabolome of transgenic grapes [17]. The exogenous DefH9-iaaM construct encodes for tryptophan-2monoxygenase, which is the enzyme regulating the synthesis of the auxin hormone indoleacetic acid. The experiment showed that, whatever the mechanism, the changes occurring in the grape composition were not directly predictable on the basis of type and copy number of the additional genes. For this reason, the metabolome (i.e., the molecular phenotype) could be considered as the monitor integrating the comprehensive perturbations at all omics levels.

3. NMR for food analysis and foodomics The complete characterization and quantification of the molecules constituting the food metabolome can be thought as representing one dimension of the foodomics space, by analogy with and extending the metabolomics space described elsewhere [18]. An analytical technique ideally tailored for its exploration should be characterized by [19]: • ease of quantification and identification; • the high number of metabolites that can be measured through a single-pass, for which automation is important; • short time and low costs needed for analysis, including sample preparation; and, • the possibility to store the data into a database with extensive details and enriched by sufficient descriptors to allow the information to be retrieved by user-specific criteria. The majority of the foodomics studies performed through NMR focus on hydrogen because it gives the highest sensitivity, compared to 31P, 13C, 17O or 15N. Nevertheless, until the potential of ultrasensitive applications, such as dynamic nuclear polarization [20] is expressed, 1H-NMR spectroscopy still has to be considered a poorly sensitive technique compared to other spectrometries [e.g., MS, spectrophotometry, and electron-spin resonance (EPR)]. A second limitation of 1H spectra in the exploration of the foodomics space has been traditionally identified as the relatively reduced resonance window of proton spectroscopy compared to 13C or 31P, so that many signals appear overlapped, especially when complex mixtures are analyzed. A reasonable number of molecules that can be unambiguously identified and simultaneously quantified in a food extract is in the range 50–100 [21]. These limitations are counterbalanced by the fact that the only variables modulating an NMR spectrum are the solvent, the magnetic field, and the pulse sequence employed to transfer magnetization to the observed nuclei, with little instrument drift [22]. The high reproducibility makes NMR spectroscopy the choice for obtaining data that can be directly organized into a database, since pattern recognition and multivariate analyses can be directly applied to raw spectra. To appreciate the importance of the possibility of analyzing unassigned sequences of numbers (such as spectra) or strings of letters (such as codons), it is sufficient to consider the impact that the publication of databases accessible to the entire scientific community had in the genomics field, even at their first appearance when most of the DNA or protein sequences were not yet annotated. The advantage of using raw spectra for the comparison of different metabolomes encouraged the exploitation of NMR foodomics for classification of food products.

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

95

Fig. 1. Foodomics applies omics approaches to define food by looking at the effects of, for example, selection, production, processing, and storage, on the genometranscriptome-proteome-metabolome of the originating organisms, or parts of them. Then, foodomics continues its role by describing the evolution of food during digestion, absorption and interaction with human, looking at nutrition and health from the food perspective.

4. Levels of information of the metabolome Perfect knowledge of the entire foodomics space of a food can be considered a “holistic hope” [14]. To have an impression of the reasons, it is sufficient to consider that the entire pool of molecules that we ingest, typically referred to as the nutrition metabolome [23], is made up of more than 15,000

components – nutrient and non-nutrient, natural and artificial molecules – pertaining to more than 100 major chemical classes, with a concentration ranging over nine orders of magnitude (pM–mM). The most complex foods from the point of view of composition are based on vegetables, whose metabolome probably comprises more than 10,000 different detectable compounds [24].

96

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

The research topics covered by NMR for the definition of the main food attributes can be classified according to increasing levels of exploration of the foodomics space, by means of the definitions set out by Fiehn [25], later summarized by Goodacre [11], as: • Targeted metabolite analysis: analysis restricted to a limited number of metabolites, chosen according to previous knowledge about the system under investigation. For instance, Mannina [26] set up a protocol based on 1H-NMR, followed by the identification of six characteristic resonances, to reveal the presence of refined hazelnut oils in concentrations as low as 10% in admixtures with refined olive oils. • Metabolite profiling: analysis focused on a group of metabolites, for example, a class of compounds [27]. Ohno et al. studied how the growth at different cultivation altitudes affected the phenolic fraction profile of black tea [28]. Both targeted metabolite analysis and metabolite profiling often take advantage of NMR as a quantitative analytical tool [25]. The necessary conditions were described in a milestone review [29]. Many of the works aiming to reveal the presence of contaminants in food pertain to this kind of approach. Schievano et al. set up a rapid method for quantifying histamine in several kinds of cheese with a limit of detection (LOD) of 0.6–1 mg/kg, making it suitable for investigations of the authenticity and the hygienic quality of cheese [30]. • Untargeted metabolite analysis (fingerprinting): nonquantitative observation of the entire metabolome for classification purposes, for example to discriminate samples on the basis of biological or geographic provenance. An example can be found in the work by Mallamace et al., where samples of cherry tomatoes grown in the Sicilian area of Pachino, accredited by the European PGI (Protected Geographical Indication), were differentiated from cherry tomatoes grown in several other areas [31]. • Metabolomics: measure of the fingerprint of biochemical perturbations caused by an external effect. This kind of approach appears in most of the papers dealing with the effect of biotechnological treatments or ageing on food characteristics. Savorani et al. observed how the rearing conditions of gilthead seabream (Sparus Aurata) influenced the metabolic profile assessed on perchloric-acid extracts prepared from their white muscle [32]. Ercolini et al. monitored the evolution of the metabolome of meat samples stored under air, modifiedatmosphere, and vacuum packaging [33]. In each of these works, the application of multivariate data analysis produced a list of molecules exerting the largest variation upon system perturbations, and provided a set of biomarkers able to describe the evolution of the food product under different conditions.

5. Exploring new lands in the foodomics space The previous section showed that many advances in the studies concerning food have been obtained even with a limited view of the foodomics space. Nevertheless, efforts are daily made by the scientific community to understand how each part of the NMR metabolomics pipeline (i.e., the sequence of experimental design, sampling, sample preparation, sample analysis, data pre-processing and analysis) can be tailored to explore new portions of this space (Fig. 2) [34]. The fil rouge of most of these studies can be sensed through a citation that finds space in papers dedicated to omics techniques: “A collection of facts is no more a science than a heap of stones is a house’’ (Henri Poincaré). This sentence has a triple meaning in such a context:

• of the wealth of data that NMR spectroscopy can generate from a sample, only a handful might be needed to describe a problem adequately [11]; • the collection of these data can be transformed into knowledge when useful algorithms are designed; and, • the full potential of this knowledge is exploited when accessible and useful to others. In other words, to cope with the torrent of data originating from foodomics, we need good data, good databases and even better algorithms [2,11]. 5.1. Standardization and merger of the data obtained through different pulse sequences Several parameters modulate even the simplest 1H-NMR pulse sequences and must accompany the NMR spectra [35]. A list of parameters can be obtained from Beckonert et al. [36] for the main pulse sequences employed in metabolomics, together with the corresponding suggested values. A specific complication arises when the water signal has to be saturated to reach an adequate dynamic range allocating all peaks of interest. Potts et al. [37] tested four water signal-suppression schemes (Watergate, WET, presaturation during relaxation, and NOESY presaturation during relaxation and mixing time), and found that each pulse sequence gave peculiar features to the spectra. The main differences that he found were connected to the partial suppression of the signals adjacent to the water peak and to some artifacts in the baseline. To compare the spectra affected by such biases, the spectral regions around the water peak could be removed from the computation, and baseline artifacts could be reduced by ad hoc adjustments, described in sub-section 5.8. In the light of a standard protocol, NOESY pulse sequence, including a presaturation step, appeared to be the best choice because of its highest repetitivity between experiments [37]. 5.2. Merging data obtained at different magnetic fields One of the major limitations, encountered when comparing spectra recorded with instruments operating at different magnetic fields, is represented by the effect of the coupling constants on the signal shape. The obvious remedy is to focus on a single field strength, opting for a good compromise between resolution and affordability, the latter being a sensible choice criterion in the food area. Spraul et al. set up a totally robotized system for the characterization of fruit juices, based on a 400-MHz equipment [38]. When affordability is a major issue, miniaturized NMR instruments, characterized by permanent magnets operating at 45–90 MHz, may be the optimal solution [39]. The given width at half height, as low as 3 Hz, cannot be compared with that of instruments equipped with superconducting magnets. However, in many applications, the spectra registered with superconducting magnets are processed by averaging over portions (i.e., binning or bucketing) prior to multivariate analysis, thus neutralizing the improved resolution given by higher fields. These portions are typically 0.02–0.04 ppm wide, that is 14–28 Hz in a spectrum acquired at a 700-MHz instrument. Even when the highest resolution is not requested, high-field instruments are still recommended for the sensitivity aspect, as the signalto-noise ratio depends on the 3/2 power of the intensity of the magnetic field. The range of magnetic fields presently employed for foodomics applications suggests that there can be space for strategies based on merging the information obtained at different fields. Potts et al. [37] successfully merged data acquired on urines at 500 MHz and 600 MHz by performing principal-component

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102 97

Fig. 2. Some biases that can limit the portion of the metabolomics space observable by nuclear magnetic resonance (NMR).

98

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

analysis (PCA) to find the peaks most affected by the magnetic field and by averaging them over conveniently large bins. The same strategy can analogously be extended to food matrices. J-resolved or pure-shift sequences, described in sub-section 5.6., offer additional solutions to ease the comparison of spectra obtained at different magnetic fields by removing the effects of magnetic coupling. 5.3. Sensitivity In the pool of works aimed at exploring larger portions of the foodomics space, it seems reasonable to include those devoted to deliver NMR spectra with unprecedented sensitivities, through optical pumping [40] or microwave-driven transfer of magnetization from electrons to nearby nuclei by dynamic nuclear polarization [41]. Examples of the application of the latter technique to elucidate metabolic reactions were published recently [42], but it is questionable whether it will be ever developed as a quantitative, high-throughput method [43]. 5.4. Sampling inhomogeneous matter As with data in any other omics techniques, biological and analytical variation within the data should be expected [44], as the data may be confounded with factors of interest [45]. Defernez and Colquhoun [46] recorded 1H spectra for several days on water extracts from potato tubers. Even if they pertained to the same cultivar and were grown, harvested, and analyzed together, the main source of variance was the compositional difference among the tubers themselves. When many samples are analyzed per batch, the unavoidable lapse of time separating sampling and the acquisition of the NMR spectra can be reflected in an altered fingerprint, occurring even during storage at -80°C [14]. Some sources of biological variability can be dealt with by applying suitable mathematical treatments to the spectra. The variability due to different dilution of the samples may be compensated for by normalizing the spectra to the area of an added molecule, often the same added as a chemical-shift reference [47]. When the contamination with extraneous substances is undesirable, a compartmentalized addition can be performed by means of a sealed capillary tube. To keep in consideration the variability introduced by sample preparation, a molecule naturally present at a constant concentration in the studied matrix can be used as an internal standard. This requires that the substance is stable at the experimental conditions, and it has no other interfering signals in all samples of the experimental series. An example of this approach was published by Laghi et al. [48], who considered the concentration of lactate to be constant during the in-vitro digestion of cured beef, thus keeping constant the area of the lactate signal among the sequence of spectra, while calculating the concentration of all the other soluble molecules released during the digestion. Similarly, Ciampa et al. considered the concentration of the entire ATP pool to be constant during storage of fish, because it constituted a closed inter-conversion pathway. Thus, the area summation of all signals arising from ATP and its by-products was considered constant in all fish extracts, while calculating the concentration of all the other metabolites [49]. Referencing a spectrum to the intensity of a single peak transfers its variance to the entire spectrum. The same applies to normalizing a spectrum to a constant area, because a covariance between high- and low-intensity peaks is introduced. A wiser option is represented by protocols, such as Probabilistic Quotient Normalization [50], giving the same importance to each signal of the spectrum in the determination of the optimal multiplicative factor needed to compensate for different dilutions. Another method, proposed by Capozzi et al., was based on a normalization algorithm validated by

TOCSY experiments on tomato samples [51]. A comprehensive paper describing optimal solutions to correct several artifacts introduced by dilution errors was published by Kohl et al. [52]. 5.5. Sample handling Foodomics experiments are generally designed to capture snapshots of food metabolomes. Between sample collection and analysis, care must be taken to avoid food metabolic fluctuations, chemical reactions or microorganisms that alter such a picture. When liquid-state NMR needs to be performed on solid or inhomogeneous samples, an extraction step needs to be performed. This is commonly considered one of the most critical steps of the foodomics pipeline [2]. This is due to the de facto impossibility of totally extracting the molecules of interest from a food matrix, a condition that obligates the recorded fingerprint to be associated to the corresponding solvent. Any chosen solvent, indeed, introduces a specific bias, because of the strong differences in the polarity of the molecules constituting the metabolome. That is why heterogeneous foodstuffs should always be described at least by a couple of molecular profiles, namely hydrophobic and hydrophilic, recorded on aqueous and organic extracts. The sample modifications attributable to the solvent choice can be managed by considering the following three options. (1) The inclusion of the extraction conditions in the metadata accompanying each experiment, with reference to characteristics of the solvent and the mixing conditions for all the heterogeneous phases (and mincing, in the case of solid samples). In the case of fish-muscle tissues, a description of how the disruption method can modulate the information obtained through NMR can be obtained from the work of Lin et al. [53]. The inclusion of the extraction conditions can be sufficient when molecular profiling or untargeted metabolite investigations are performed. Such approaches do not necessarily rely on characterization or quantification of all the observed signals, provided that a sufficiently reproducible extraction protocol is selected. (2) When lengthy sample preparations are allowed (e.g., stable extracts), repeated extractions should be preferred because they lead to more complete extraction and more homogenous samples. Both Wu et al. [54], for liver tissue from adult flatfish, and Capanoglu et al. [55], for methanol extracts from tomato paste, found optimal results in a double-extraction protocol. (3) When a single extraction is the only feasible option, opportune solvents (including a mixture of them) and conditions reducing sample variability should be preferred. For liver tissue from adult flatfish, Wu et al. found that the mixture methanol/ chloroform/water is the preferable choice for metabolomics investigations [54]. Other investigations on the topic have been cited by Lin et al. [53]. The efforts to register quantitative NMR spectra (e.g., the integrals of all signals are proportional to the corresponding actual molar concentrations of the corresponding molecules) may be neutralized by incomplete or irreproducible extraction of the substances of interest. To balance the accuracy of the information obtainable by the exact quantification of all substances of interest, fingerprinting needs more metadata related to the protocol of analysis. 5.6. Peaks overlap In proton-NMR spectra, the peaks of interest typically span less than 10 ppm, corresponding to 6000 Hz, for resonance frequencies at 600 MHz. This range could theoretically allocate 2000 peaks

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

without overlap, assuming 3 Hz width for each one. Keeping in mind that each molecule is the origin of several distinct peaks, with further multiplicity caused by magnetic coupling with neighbor atoms, the possibility to have only well-resolved signals in the spectrum is practically confined to mixtures of a few tens of metabolites with different chemical features. In all other cases (e.g., complex mixtures, such as food extracts), congestion of signals can represent a major problem. Several work-arounds have been devised to cope with poor resolution, increasing the magnetic field representing a costly option, considering that it took 28 years to double the highest field available (from 500 MHz to 1000 MHz for proton) [56]. J-resolved spectra are typically recorded to measure the protoncoupling constants for identification purposes, but the F2 projection can be employed for foodomics investigations. This retains both chemical shift and relative intensity of the signals, but is less crowded than a simple 1D-NMR spectrum due to the collapse of multiplets into singlets [22]. Khatib et al. successfully derived a PCA model from the skyline projection of 2D J-resolved NMR spectra obtained on 2-butanol extracts of beer [57]. Ward et al. studied the carbohydrate region of proton spectra recorded on D2O-CD3OD extracts of Arabidopsis thaliana [22]. Yilmaz et al. used 2D J-resolved NMR spectra and PARAllel FACtor analysis (PARAFAC) to resolve spectra of complex plant extracts [58]. The possibility to obtain homonuclear broad-band decoupled proton spectra has been studied since the 1970s [59] but several drawbacks limited their use. “Pure-shift” versions of 1D, and diffusion-ordered spectroscopy (DOSY) sequences seem to have reached the goal, with a reduction of the superimposition of the peaks as high as 20 times [56,60]. Bidimensional experiments, correlating protons with heteronuclei, are suitable to spread signals much more than a proton 1D spectrum. The generally unacceptably long acquisition times have rarely allowed their application to omics studies [22]. The introduction of ultrafast approaches to nD NMR has the potential to represent a breakthrough in this respect, because it is able to deliver any type of multidimensional spectrum in a single transient [61].

5.7. Peaks alignment Data mining on matrices of NMR spectra is typically performed by means of multivariate analysis techniques, which postulate a relationship between the intensity associated to each point of the spectrum and the concentration of a certain substance. This assumption does not hold when signals belonging to the same substance shift across the spectra of different samples, due to pH, interacting solutes, solvent or temperature variations. This phenomenon adds unwanted sources of variance, which decrease the effectiveness of the multivariate analysis in extracting the relevant information. A remedy for signal misalignment has traditionally been found in averaging the spectra over portions, so-called binning or bucketing. The ideal width of these portions has been the subject of several papers [62], and a certain consensus has grown up around the value of 0.04 ppm, conventionally considered to be sufficient to include the range covered by a signal due to small pH oscillations. When peak shape is informative [63], peaks-alignment protocols, such as icoshift [64], are preferable to binning. In some cases, changes in the position of signals, pertaining to the same species, across the spectra of different samples may carry useful information. Spraul et al. [38] built a model to predict potassium and magnesium from proton-NMR spectra, taking advantage of the relationship between the concentration of those ions and the shift of some peaks belonging to molecules interacting with them.

99

5.8. Baseline biases Even though NMR hardware has improved from any point of view, and the pulse sequences set up for foodomics investigations are tailored to give flat baselines (see sub-section 5.1.), the need to adjust the baseline finely after spectrum acquisition is conserved. Most of the papers concerning food investigations describe baseline adjustments as manually performed, with no further characterization. This hampers the possibility to include such a step in the information accompanying the spectra (reproducible metadata). A second problem is that spectral manual adjustments are often based on the visual inspection, thus introducing an uncontrolled subjective source of variance. This is why algorithms that automatically perform baseline adjustments on well-defined principles seem more convenient than manual modifications of the spectra. In this way, the conversion of the acquired free induction decay (FID) into the corresponding final spectrum is a process that could be reproduced in all laboratories accessing the raw data. The list of algorithms described in the literature is impressively long. It is convenient here to cite the work of Liland et al. [65], who nicely organized in a handy R-package, under the name “baseline”, most of these algorithms. 5.9. Metadata Extracting the relevant information about the properties of a food product, by means of the NMR spectra of its samples, requires a detailed definition of the history of that sample, which begins with experimental design and ends with sample preparation and spectral acquisition. All this information must be reported in a proper way to be inherited by the scientific community. In the past decade there have been a number of initiatives to build consensus on the specific ways that data from omics investigations are organized and reported. The Standard Metabolic Reporting Structures (SMRS) group [44] aims to derive and recommend standards for conducting and reporting metabolomics studies. SMRS suggests that such studies are accompanied by two kinds of metadata: (1) metadata describing the so-called experimental context, such as types of test and control samples, sample identity, storage of the source before sampling, phenotype characteristics, environmental conditions, and sample collection and preparation protocol; and, (2) metadata disclosing the objective and the type of modeling conducted on the samples, together with the steps taken to prepare the raw data. In foodomics studies, the collection and the organization of the metadata accompanying the samples are crucial, due to the dynamic nature of metabolites typically considered [66]. This is why Bino et al. [67] outlined a checklist called MIAMET (Minimum information about metabolomics experiment) to enrich metabolomics data with information suitable to put them in the proper context. However, Jenkins et al. [66] outlined a framework called ArMet, to save metabolomics data rationally. The wise organization of the data is of the utmost importance to ease data retrieval and analysis for purposes that go beyond the purposes for which the data were originally collected. For this purpose, the European COSMOS consortium, developing de facto standard formats where various components are encapsulated, aims to build a suitable repository for NMR- and MSbased metabolomics experiments [68]. Unfortunately, there are no repositories solving the problem of metadata storage in food metabolomics. Although only a few metabolomics studies about vegetable and animal organisms, including humans, have been deposited in dedicated repositories, we expect that the next field of application will be food and nutrition

100

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

studies. At the moment, indeed, there are a few examples of Webbased platforms approaching the collection of studies about plant species, with MS data (http://plantmetabolomics.org/), or both NMR and MS data, although the latter is limited to tomato studies (http:// www.ebi.ac.uk/metabolights). It is worth noting here that there is only one repository collecting NMR metadata about several vegetable species (http:// services.cbib.u-bordeaux2.fr/MERYB/). Despite the limited number of thematic examples, due to the complexity of the problem, the necessity to develop a repository for the results of food analyses is a matter of fact, especially for sampling, storage before analyses, sample preparation and data acquisition. The double opportunity offered by a shared protocol involves harmonizing the data collection and providing scientific communities with raw data that could be analyzed according to the individual experience of each scientist, in a way similar to genome and proteome counterparts, realized previously. 6. Multivariate analysis Foodomics experiments through NMR are likely to generate thousands of data points, of which only a few may bring useful information. The first step of the analysis of data is their exploration, (i.e., the search for underlying structures, an operation referred to as pattern recognition) [46]. This is typically done by means of unsupervised multivariate analysis, particularly helpful to highlight expected but also unexpected characteristics of the data. Algorithms addressed to reveal unexpected variations of molecular composition are important for those applications aimed at disclosing adulterated samples. Indeed, food frauds tend to evolve over the years. For this reason, it is important to build standard models on genuine samples that will be able to capture future, yet unknown, frauds. For an illuminating example, see Spraul et al. [38]. Exploratory multivariate analyses are based on the assumption that most of the data collected through spectroscopic techniques on each sample are characterized by redundancy, so that the information can be reasonably summarized with a limited number of new orthogonal variables [69]. This condensation can be thought of as visualizing the data from a more convenient point of view than the original. Such operation is often referred to as projection pursuit [69]. In PCA, described more than 100 years ago [70] but still widely applied in foodomics, the chosen point of view is the one explaining the highest portion of variance of the original data. The rationale of such choice is that the projections explaining little portions of variance have greater possibilities to show apparent structure that is merely the result of random variation [69]. Such a postulate cannot always be considered true. This is why other projection criteria have been defined [71]. Asimov [72] first suggested finding the optimal point of view by joining multiple views by means of animations (called Grand Tours). When data analysis aims to identify models for possible groups, hierarchical clustering analysis (HCA) or soft independent modeling of class analogy (SIMCA) [73,74] approaches may be the techniques of choice [75,76]. Unsupervised multivariate analysis has been applied to NMR data of meat to predict the effect of environmental conditions on technological properties, such as the water-holding capacity [77]. Supervised methods [78] are usually more powerful, as the classification is based on prior knowledge. The PLS projection method is currently the most popular for multivariate calibration in chemometrics [79]. PLS discriminant analysis (PLS-DA) is a supervised technique that can be used to enhance the separation between groups of observations by rotating PCA components so that a maximum separation among classes is obtained. In PLS methods, the simultaneous decomposition of two matrices

is performed: an analytical signal matrix (X) and a matrix of corresponding chemical indices or descriptors (Y). PLS methods were repeatedly used for screening NMR spectroscopic analysis of foodstuffs and beverages [80]. As a drawback, they can easily overfit the model, a situation in which the training data are represented to nearperfection but similar data not used in the calibration are no longer recognized as similar. Interesting applications that recently appeared in the literature were iPLS [81,82] and, for clustering purposes, iECVA [32,83]. 7. Conclusions and future perspectives NMR-based foodomics is contributing to progressive advances towards a deeper understanding of food, with applications spanning food safety and authenticity to exploration of the interactions between nutrients and the human organism, an omics application still in its infancy. Information about the spatial distribution of metabolites in the food matrix, through time-domain NMR (TD-NMR) [84], could represent a further step, as well as the possibility to follow the release of molecules during digestion. Multivariate data analysis is now shedding new light on NMR spectroscopy, by shifting the applications of choice for this technique, formerly confined to structural elucidation of unknown compounds, to the discovery of metabolic patterns and trajectories, as well as in the definition of metabotypes. These new applications are still in the exploratory phase, but we can predict that similarly promising results will soon also emerge from nutritional studies aimed at finding the relationships between dietary patterns and health status, the latter assessed by metabolomics. In other words, foodomics is a new approach to food and nutrition that studies the food domain as a whole with the nutrition domain to reach the main objective, the optimization of human health and well-being by nutrition. Metabolomics is expected to support genomics and proteomics in providing clear scientific evidence for disease prevention by means of correct nutrition. Acknowledgements This work was supported by the CHANCE Project (EC KBBE-FP7 N. 266331). The authors participate in the COST action FA1005 INFOGEST. References [1] V. García-Cañas, C. Simó, M. Herrero, E. Ibáñez, A. Cifuentes, Present and future challenges in food analysis: foodomics, Anal. Chem. 84 (2012) 10150–10159. [2] L. Mannina, A. Sobolev, S. Viel, Liquid state 1H high field NMR in food analysis, Prog. Nucl, Magn. Reson. Spectrosc. 66 (2012) 1–39. [3] J.C. Moore, J. Spink, M. Lipp, Development and application of a database of food ingredient fraud and economically motivated adulteration from 1980 to 2010, J. Food Sci. 77 (2012) R118–R126. [4] D.S. Wishart, Metabolomics: applications to food science and nutrition research, Trends Food Sci. Technol. 19 (2008) 482–493. [5] J.M. Cevallos-Cevallos, J.I. Reyes-De-Corcuera, E. Etxeberria, M.D. Danyluk, G.E. Rodrick, Metabolomic analysis in food science: a review, Trends Food Sci. Technol. 20 (2009) 557–566. [6] M.F. Marcone, S. Wang, W. Albabish, S. Nie, D. Somnarain, A. Hill, Diverse food-based applications of nuclear magnetic resonance (NMR) technology, Food Res. Intern. 51 (2013) 729–747. [7] A. Sobolev, D. Capitani, D. Giannino, C. Nicolodi, G. Testone, F. Santoro, et al., NMR-metabolic methodology in the study of GM foods, Nutrients 2 (2010) 1–15. [8] C. Piras, F. Cesare, F. Marincola, S.B. Savorani, S. Engelsen, S. Cosentino, et al., NMR metabolomics study of the ripening process of the Fiore Sardo cheese produced with autochthonous adjunct cultures, Food Chem. 141 (2013) 2137– 2147. [9] H. Kitano, Systems biology: a brief overview, Science 295 (2002) 1662–1664. [10] M. Pedro, Emerging bioinformatics for the metabolome, Brief. Bioinform. 3 (2002) 134–145. [11] R. Goodacre, S. Vaidyanathan, W.B. Dunn, G.G. Harrigan, D.B. Kell, Metabolomics by numbers: acquiring and understanding global metabolite data, Trends Biotechnol. 22 (2004) 245–252.

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

[12] F. Capozzi, A. Bordoni, Foodomics: a new comprehensive approach to food and nutrition, Genes Nutr. 8 (2013) 1–4. [13] A. Cifuentes, Special issue: advanced separation methods in food analysis, J. Chromatogr. A 1216 (2009) 7109–7358. [14] R.D. Hall, Plant metabolomics: from holistic hope, to hype, to hot topic, New Phytol. 169 (2006) 453–468. [15] D. Kell, P. Mendes, Snapshots of systems, in: A. Cornish-Bowden, M. Cárdenas (Editors), Technological and Medical Implications of Metabolic Control Analysis, Springer, Netherlands, 2000, pp. 3–25. [16] L.M. Raamsdonk, B. Teusink, D. Broadhurst, N. Zhang, A. Hayes, M.C. Walsh, et al., A functional genomics strategy that uses metabolome data to reveal the phenotype of silent mutations, Nat. Biotechnol. 19 (2001) 45–50. [17] G. Picone, B. Mezzetti, E. Babini, F. Capocasa, G. Placucci, F. Capozzi, Unsupervised principal component analysis of nmr metabolic profiles for the assessment of substantial equivalence of transgenic grapes (vitis vinifera), J. Agric. Food Chem. 59 (2011) 9271–9279. [18] I. Nobeli, H. Ponstingl, E.B. Krissinel, J.M. Thornton, A structure-based anatomy of the E. coli metabolome, J. Mol. Biol. 334 (2003) 697–719. [19] R. Verpoorte, Y.H. Choi, H.K. Kim, NMR-based metabolomics at work in phytochemistry, Phytochem. Rev. 6 (2007) 3–14. [20] T. Prisner, W. Köckenberger, Dynamic nuclear polarization: new experimental and methodology approaches and applications in physics, chemistry, biology and medicine, Appl. Magn. Reson. 34 (2008) 213–218. [21] D.S. Wishart, Quantitative metabolomics using NMR, Trends Analyt. Chem. 27 (2008) 228–237. [22] J.L. Ward, J.M. Baker, M.H. Beale, Recent applications of NMR spectroscopy in plant metabolomics, FEBS J. 274 (2007) 1126–1131. [23] M.J. Gibney, M. Walsh, L. Brennan, H.M. Roche, B. German, B. van Ommen, Metabolomics in human nutrition: opportunities and challenges, Am. J. Clin. Nutr. 82 (2005) 497–503. [24] T.M. Ehrman, D.J. Barlow, P.J. Hylands, Phytochemical Databases of Chinese Herbal Constituents and Bioactive Plant Compounds with Known Target Specificities, J. Chem. Inf. Model. 47 (2007) 254–263. [25] O. Fiehn, Combining genomics, metabolome analysis, and biochemical modelling to understand metabolic networks, Comp. Funct. Genomics 2 (2001) 155–168. [26] L. Mannina, M. D’Imperio, D. Capitani, S. Rezzi, C. Guillou, T. Mavromoustakos, et al., 1H NMR-based protocol for the detection of adulterations of refined olive oil with refined hazelnut oil, J. Agric. Food Chem. 57 (2009) 11550–11556. [27] O. Fiehn, Metabolomics – the link between genotypes and phenotypes, Plant Mol. Biol. 48 (2002) 155–171. [28] A. Ohno, K. Oka, C. Sakuma, H. Okuda, K. Fukuhara, Characterization of tea cultivated at four different altitudes using 1H NMR analysis coupled with multivariate statistics, J. Agric. Food Chem. 59 (2011) 5181–5187. [29] G.F. Pauli, T. Gödecke, B.U. Jaki, D.C. Lankin, Quantitative 1H NMR. development and potential of an analytical method: an update, J. Nat. Prod. 75 (2012) 834–851. [30] E. Schievano, K. Guardini, S. Mammi, Fast determination of histamine in cheese by nuclear magnetic resonance (NMR), J. Agric. Food Chem. 57 (2009) 2647– 2652. [31] D. Mallamace, C. Corsaro, A. Salvo, N. Cicero, A. Macaluso, G. Giangrosso, et al., A multivariate statistical analysis coming from the NMR metabolic profile of cherry tomatoes (The Sicilian Pachino case), Physica A 401 (2014) 112–117. [32] F. Savorani, G. Picone, A. Badiani, P. Fagioli, F. Capozzi, S.B. Engelsen, Metabolic profiling and aquaculture differentiation of gilthead sea bream by 1H NMR metabonomics, Food Chem. 120 (2010) 907–914. [33] D. Ercolini, I. Ferrocino, A. Nasi, M. Ndagijimana, P. Vernocchi, A. La Storia, et al., Monitoring of microbial metabolites and bacterial diversity in beef stored under different packaging conditions, Appl. Environ. Microbiol. 77 (2011) 7372– 7381. [34] M. Brown, W. Dunn, D. Ellis, R. Goodacre, J. Handl, J. Knowles, et al., A metabolome pipeline: from concept to data to knowledge, Metabolomics 1 (2005) 39–51. [35] F. Capozzi, M.A. Cremonini, Nuclear magnetic resonance spectroscopy in food analysis, in: S.E. Otles (Editor), Handbook of Food Analysis Instruments, CRC Press, Boca Roton, FL, USA, 2008, pp. 281–318. [36] O. Beckoner, H.C. Keun, T.M.D. Ebbels, J. Bundy, E. Holmes, J.C. Lindon, et al., Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts, Nat. Protoc. 2 (2007) 2692–2703. [37] B.C.M. Potts, A.J. Deese, G.J. Stevens, M.D. Reily, D.G. Robertson, J. Theiss, NMR of biofluids and pattern recognition: assessing the impact of NMR parameters on the principal component analysis of urine from rat and mouse, J. Pharm. Biomed. Anal. 26 (2001) 463–476. [38] M. Spraul, B. Schütz, E. Humpfer, M. Mörtter, H. Schäfer, S. Koswig, et al., Mixture analysis by NMR as applied to fruit juice quality control, Magn. Reson. Chem. 47 (2009) S130–S137. [39] E. Danieli, J. Perlo, B. Blümich, F. Casanova, Small magnets for portable NMR spectrometers, Angew. Chem. Int. Ed Engl. 49 (2010) 4133–4135. [40] M.S. Albert, G.D. Cates, B. Driehuys, W. Happer, B. Saam, C.S. Springer, et al., Biological Magnetic Resonance Imaging Using Laser-Polarized 129Xe, 1994. [41] T.R. Carver, C.P. Slichter, Polarization of nuclear spins in metals, Phys. Rev. 92 (1953). [42] S. Meier, M. Karlsson, P.R. Jensen, M.H. Lerche, J.Ø. Duus, Metabolic pathway visualization in living yeast by DNP-NMR, Mol. Biosyst. 7 (2011) 2834–2836.

101

[43] F. Savorani, M.A. Rasmussen, M.S. Mikkelsen, S.B. Engelsen, A primer to nutritional metabolomics by NMR spectroscopy, Food Res. Intern. 54 (2013) 1131–1145. [44] J.C. Lindon, J.K. Nicholson, E. Holmes, H.C. Keun, A. Craig, J.T.M. Pearce, et al., Summary recommendations for standardization and reporting of metabolic analyses, Nat. Biotechnol. 23 (2005) 833–838. [45] A. Craig, O. Cloarec, E. Holmes, J.K. Nicholson, J.C. Lindon, Scaling and normalization effects in NMR spectroscopic metabonomic data sets, Anal. Chem. 78 (2006) 2262–2267. [46] M. Defernez, I.J. Colquhoun, Factors affecting the robustness of metabolite fingerprinting using 1H NMR spectra, Phytochemistry 62 (2003) 1009–1017. [47] A. Bordoni, G. Picone, E. Babini, M. Vignali, F. Danesi, V. Valli, et al., NMR comparison of in vitro digestion of Parmigiano Reggiano cheese aged 15 and 30 months, Magn. Reson. Chem. 49 (2011) S61–S70. [48] L. Laghi, E. Babini, A. Bordoni, A. Ciampa, F. Danesi, M. Di Nunzio, et al., Time domain measurements and high resolution spectroscopy are powerful nuclear magnetic resonance approaches suitable to evaluate the in vitro digestion of protein-rich food products, in: J. van Duynhoven (Editor), Time Domain Measurements and High Resolution Spectroscopy are Powerful Nuclear Magnetic Resonance Approaches Suitable to Evaluate the In Vitro Digestion of Protein-rich Food Products, RSC Publishing, Cambridge, 2013. [49] A. Ciampa, G. Picone, L. Laghi, H. Nikzad, F. Capozzi, Changes in the amino acid composition of bogue (Boops boops) fish during storage at different temperatures by 1H-NMR spectroscopy, Nutrients 4 (2012) 542–553. [50] F. Dieterle, A. Ross, G. Schlotterbeck, H. Senn, Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. application in 1H NMR metabonomics, Anal. Chem. 78 (2006) 4281–4290. [51] F. Capozzi, A. Ciampa, G. Picone, G. Placucci, F. Savorani, Normalization is a necessary step in NMR data processing: finding the right scaling factors, in: J.P. Renou, P.S. Belton, E.G.A. Webb (Editors), Magnetic Resonance in Food Science: An Exciting Future, The Royal Society of Chemistry Publ, Cambridge UK, 2011, pp. 147–160. [52] S. Kohl, M. Klein, J. Hochrein, P. Oefner, R. Spang, W. Gronwald, State-of-the art data normalization methods improve NMR-based metabolomic analysis, Metabolomics 8 (2012) 146–160. [53] C. Lin, H. Wu, R. Tjeerdema, M. Viant, Evaluation of metabolite extraction strategies from tissue samples using NMR metabolomics, Metabolomics 3 (2007) 55–67. [54] H. Wu, A.D. Southam, A. Hines, M.R. Viant, High-throughput tissue extraction protocol for NMR- and MS-based metabolomics, Anal. Biochem. 372 (2008) 204–212. [55] E. Capanoglu, J. Beekwilder, D. Boyacioglu, R. Hall, R. de Vos, Changes in antioxidant and metabolite profiles during production of tomato paste, J. Agric. Food Chem. 56 (2008) 964–973. [56] J.A. Aguilar, S. Faulkner, M. Nilsson, G.A. Morris, Pure shift 1H NMR: a resolution of the resolution problem?, Angew. Chem. Int. Ed Engl. 49 (2010) 3901–3903. [57] A. Khatib, E.G. Wilson, H.K. Kim, A.W.M. Lefeber, C. Erkelens, Y.H. Choi, et al., Application of two-dimensional J-resolved nuclear magnetic resonance spectroscopy to differentiation of beer, Anal. Chim. Acta 559 (2006) 264– 270. [58] A. Yilmaz, N.T. Nyberg, J.W. Jaroszewski, Metabolic profiling based on twodimensional J-resolved 1H NMR data and parallel factor analysis, Anal. Chem. 83 (2011) 8278–8285. [59] W.P. Aue, J. Karhan, R.R. Ernst, Homonuclear broad band decoupling and two-dimensional J-resolved NMR spectroscopy, J. Chem. Phys. 64 (1976) 4226–4227. [60] M. Nilsson, G.A. Morris, Pure shift proton DOSY: diffusion-ordered 1H spectra without multiplet structure, Chem. Commun. 9 (2007) 933–935. [61] M. Mishkovsky, L. Frydman, Principles and progress in ultrafast multidimensional nuclear magnetic resonance, Annu. Rev. Phys. Chem. 60 (2009) 429–448. [62] B.-J.M. Webb-Robertson, D.F. Lowry, K.H. Jarman, S.J. Harbo, Q.R. Meng, A.F. Fuciarelli, et al., A study of spectral integration and normalization in NMR-based metabonomic analyses, J. Pharm. Biomed. Anal. 39 (2005) 830–836. [63] F. Savorani, M. Kristensen, F.H. Larsen, A. Astrup, S.B. Engelsen, High throughput prediction of chylomicron triglycerides in human plasma by nuclear magnetic resonance and chemometrics, Nutr, Metabolism 7 (2010) 43. [64] F. Savorani, G. Tomasi, S.B. Engelsen, icoshift: a versatile tool for the rapid alignment of 1D NMR spectra, J. Magn. Reson. 202 (2010) 190–202. [65] K.H. Liland, T. Almøy, B.-H. Mevik, Optimal choice of baseline correction for multivariate calibration of spectra, Appl. Spectrosc. 64 (2010) 1007–1016. [66] H. Jenkins, N. Hardy, M. Beckmann, J. Draper, A.R. Smith, J. Taylor, et al., A proposed framework for the description of plant metabolomics experiments and their results, Nat. Biotechnol. 22 (2004) 1601–1606. [67] R.J. Bino, R.D. Hall, O. Fiehn, J. Kopka, K. Saito, J. Draper, et al., Potential of metabolomics as a functional genomics tool, Trends Plant Sci. 9 (2004) 418–425. [68] K. Haug, R.M. Salek, P. Conesa, J. Hastings, P. de Matos, M. Rijnbeek, et al., MetaboLights – an open-access general-purpose repository for metabolomics studies and associated meta-data, Nucleic Acids Res. 41 (2013) D781–D786. [69] M.C. Jones, R. Sibson, What is projection pursuit?, J. R. Stat. Soc. [Ser. A] 150 (1987) 1–37. [70] K. Pearson, On linear and planes of closest fit to systems of points in space, Philos. Mag. 2 (1901) 559–572. [71] J.H. Friedman, Exploratory projection pursuit, J. Am. Stat. Assoc. 82 (1987) 249–266.

102

L. Laghi et al./Trends in Analytical Chemistry 59 (2014) 93–102

[72] D. Asimov, The grand tour: a tool for viewing multidimensional data, SIAM J. Sci. Stat. Comput. 6 (1985) 128–143. [73] P. Oliveri, G. Downey, Multivariate class modeling for the verification of food-authenticity claims, TrAC-Trend, Anal. Chem. 35 (2012) 74–86. [74] M.I. López, E. Trullols, M.P. Callao, I. Ruisánchez, Multivariate screening in food adulteration: untargeted versus targeted modelling, Food Chem. 147 (2014) 177–181. [75] J. Kim, Y. Jung, B. Song, Y.-S. Bong, D.H. Ryu, K.-S. Lee, et al., Discrimination of cabbage (Brassica rapa ssp. pekinensis) cultivars grown in different geographical areas using 1H NMR-based metabolomics, Food Chem. 137 (2013) 68–75. [76] S. Esslinger, J. Riedl, C. Fauhl-Hassek, Potential and limitations of non-targeted fingerprinting for authentication of food in official control, Food Res. Intern. 60 (2014) 189–204. [77] M. Bianchi, F. Capozzi, M.A. Cremonini, L. Laghi, M. Petracci, G. Placucci, et al., Influence of the season on the relationships between NMR transverse relaxation data and water-holding capacity of turkey breast meat, J. Sci. Food Agric. 84 (2004) 1535–1540. [78] C. Ludwig, U. Gunther, MetaboLab – advanced NMR data processing and analysis for metabolomics, BMC Bioinformatics 12 (2011) 366.

[79] S. Wold, H. Martens, H. Wold, The multivariate calibration in chemistry solved by the PLS methods, in: A. Ruhe, B. Kagstrom (Editors), Lecture Notes in Mathematics, Proceedings of the Conference on Matrix Pencils, Springer-Verlag, Heidelberg, 1983, pp. 286–293. [80] Y.B. Monakhova, T. Kuballa, D.W. Lachenmeier, Chemometric methods in NMR spectroscopic analysis of food products, J. Anal. Chem. 68 (2013) 755– 766. [81] M. Kristensen, F. Savorani, G. Ravn-Haren, M. Poulsen, J. Markowski, F. Larsen, et al., NMR and interval PLS as reliable methods for determination of cholesterol in rodent lipoprotein fractions, Metabolomics 6 (2010) 129–136. [82] E. López-Rituerto, F. Savorani, A. Avenoza, J.H. Busto, J.M. Peregrina, S.B. Engelsen, Investigations of La Rioja terroir for wine production using 1H NMR metabolomics, J. Agric. Food Chem. 60 (2012) 3452–3461. [83] L. Rasmussen, F. Savorani, T. Larsen, L. Dragsted, A. Astrup, S.B. Engelsen, Standardization of factors that influence human urine metabolomics, Metabolomics 7 (2011) 71–83. [84] V. Panarese, L. Laghi, A. Pisi, U. Tylewicz, M.D. Rosa, P. Rocculi, Effect of osmotic dehydration on Actinidia deliciosa kiwifruit: a combined NMR and ultrastructural study, Food Chem. 132 (2012) 1706–1712.