Egg product freshness evaluation: A metabolomic approach

Abstract Egg products' freshness is a crucial issue for the production of safe and high‐quality commodities. Up to now, this parameter is assessed with the quantification of few compounds, but the possibility to evaluate more molecules simultaneously could help to provide robust results. In this study, 31 compounds responsible of freshness and not freshness of egg products were selected with a metabolomic approach. After an ultrahigh‐pressure liquid chromatography–high‐resolution mass spectrometry (UHPLC‐HRMS) analysis, different chemometric models were created to select gradually the most significant features that were finally extracted and identified through HRMS data. Sample lots were collected directly from their arrival at the production plant sites, extracted immediately after, then left at room temperature, and extracted again after 24 and 48 hours (first day and second day, respectively). A total amount of 79 samples was used for the model creation. Furthermore, the same compounds were detected in seven new egg products sample lots not used for the model creation and treated with the same experimental design (total amount of samples, 21). The results obtained clearly demonstrate that these 31 molecules can be considered real freshness or not freshness chemical markers. Furthermore, this UHPLC‐HRMS metabolomic approach allows for the detection of a larger set of metabolites clearly related to possible microbial growth over time, which is a relevant point for also ensuring food safety.

(≤1000 mg/kg of dry egg) and succinic acid (≤25 mg/kg of dry egg), 1 by the current laws consider only lactic acid as reliable marker. 2 These two compounds could act sometimes as "tardive" markers of eggs ageing, and their increase is often not linear during the time; moreover, the use of only one/two molecule(s) for the evaluation of such critical parameter can be not sufficient in some circumstances; at the same time, more "not freshness" compounds should be monitored.
Up to now, major attention was paid to the development of rapid techniques able to assess egg products freshness, eg, using electronic noses [3][4][5] or spectroscopic analyses. [6][7][8] These techniques are particularly suitable for industrial screening purposes, due to the reduced time and costs.
However, the identification of robust "markers of freshness" (ie, compounds that decrease their intensity during the egg products storage) could strongly support the evaluation of egg products ageing, with a reduction of the risk for the final consumer.
In this context, the use of "nontargeted" methods based on a metabolomic approach-which are considered emerging methodologies for detecting food frauds 9 -will offer the opportunity to identify and validate proper markers.
In consideration of the ageing process, chances in the volatile profile of egg products can be expected. Therefore, among possible analytical techniques, gas chromatography-mass spectrometry could be used for the identification of these changes, in agreement with previous applications in the field of food frauds, ie, tomato cultivars discrimination, 10 honey authenticity, 11 or geographical origin of saffron. 12 However, the nonvolatile fraction of food may be of great interest for the discrimination of ageing.
Scientific papers suggest that other rapid mass spectrometry approaches, as, for example, direct analysis real-time-mass spectrometry 13 or rapid evaporative ionization-mass spectrometry, 14 could lead to the characterization of raw materials, avoiding the chromatographic separation.
In any case, liquid chromatography-high-resolution mass spectrometry (LC-HRMS) is the technique of choice for the execution of a metabolomic study: Thanks to its versatility, a huge amount of information related to the nonvolatile profile can be extracted from the chromatographic fingerprints.
Literature presents applications of this approach for the detection of frauds related to a wide range of raw materials, from fruit juices 15 to wheat 16 and from extra virgin olive oil 17 to spirit drinks. 18 Independently from the analytical technique used, the creation of robust chemometric models is a crucial step for the selection of new marker compounds responsible of the target fraud. 19 Based on what described above, this work presents a metabolomic study on egg products samples able to select and identify a group of new compounds that can be used as "freshness" or "not freshness" markers.
Liquid chromatography-high-resolution mass spectrometry was exploited for fingerprints recording; subsequently, robust data elaboration was executed, creating different chemometric models able to select some marker compounds that were then studied and identified.
The final step of this workflow was the validation of these molecules, with the aim to assess their value as chemical markers, independently from the chemometric model used to select them.
Water was purified using a Milli-Q system (Millipore, Bedford, Massachusetts).
Egg products samples were directly collected from batches that daily arrive at production plant sites; to increase the variability of the model, the sampling period lasted 6 weeks, and samples coming from different batches, different suppliers, and with different amounts of carotenoids (in the range of tenths of milligrams per kilogram, authorized for improving appearance and correspondent consumer preference in finished products) were selected.

| Experimental design
The first step of the adopted workflow, as suggested by US Pharmacopoeia, 20 required the creation of chemometric models able to discriminate between fresh and not fresh samples; subsequently, the features responsible of this clusterization were selected, and a tentative of compounds identification was performed.
Finally, marker compounds identified were validated, that means that the target molecules were searched in new egg products batches subject to the same ageing applied during the model creation. 15 For the creation of the model, 29 egg products batches were collected and extracted immediately after the reception at the production plant site; subsequently, these raw materials were left at room temperature and extracted again after 1 day and after 2 days, according to the sampling plan described in Table 1. Summarizing, a global amount of 79 samples was used for the creation of the chemometric model.
For the validation of the marker compounds, seven new egg products batches were collected, extracted, and analyzed following the same experimental design described for the model creation: The total amount of samples was 21. For the evaluation of method reproducibility, 20% of the samples were double prepared, as detailed in Table 1.

| Sample preparation
During each extraction session, the same procedure was performed also into empty tubes, in which all the steps were executed without the egg product addition. These samples were labelled as "extraction blanks." A 500-ng/mL chloramphenicol solution in acetonitrile to water ratio of 80:20 was prepared and named "standard solution." Additionally, two quality control (QC) samples were prepared mixing 10 μL of each extract sample.

| LC-HRMS analysis
High-performance liquid chromatography analysis was performed with a Dionex UltiMate 3000 ultrahigh-performance liquid chromatography (Thermo Fisher Scientific, Inc, Waltham, Massachusetts) equipped with a Luna Omega 150 mm × 2.1 mm, 1.6 μm particle size analytical column (Phenomenex, Torrance, California) with a precolumn security guard ultra C18, both maintained at 30°C. Gradient elution was performed using FA and AF as mobile phase modifiers with a constant flow rate of 0.3 mL/min. Gradient conditions are the following: After 1 minute with 95% of mobile phase A (0.1% FA and 5mM AF in water) and 5% of mobile phase B (0.1% FA and 5mM AF in methanol), the percentage of solvent B increased to 95% in 24 minutes and then was maintained at this percentage for 5 minutes before column re-equilibration (10 min).
Autosampler was maintained at 5°C, and the injection volume was 5 μL. Samples, together with the "extraction blanks," were randomly injected to avoid systematic bias. 16 The "standard solutions" and the QC samples were injected at the beginning of the sequences and every 10 sample injections.

| Data treatments and statistics
Ultrahigh-performance liquid chromatography-high-resolution mass spectrometry raw data were acquired using Xcalibur software (version 3.0, Thermo Fisher Scientific, Waltham, Massachusetts); peaks alignment, "extraction blanks" subtraction, and features extraction were performed using Compound Discoverer software (version 2.0 Thermo Fisher Scientific, Waltham, Massachusetts); the mass range inspected was between 75 m/z and 1000 m/z from 1 to 30 minutes of the chromatographic runs.
The values of the critical parameters for features extractions are the following: precursor ion deviation of 5 ppm (for the positive runs) and 10 ppm (for the negative runs); maximum retention time shift of 0.3 minutes; minimum peak intensity for a peak to be selected 1 000 000 AU; relative intensity tolerance used for isotope search 30%.
Structures prediction was also performed using "ChemSpider" databases, setting a maximum mass shift of 5 ppm for positive acquisitions and 10 ppm for negative acquisitions.
For the "m/z CLOUD" MS/MS library search, the precursor mass tolerance used was 0.05 Da while the fragments mass tolerance was 10 ppm.  Afterwards, supervised orthogonal partial least square discriminant analysis (OPLS-DA) models were built comparing the "fresh" samples against the "1 day" samples, against the "2 days" samples and against the union of "1 day" and "2 days" samples.
Thanks to the S-plots, statistically significant markers responsible of the clusterization were selected: The ions furthest away from the origin contribute more significantly to the separation between the groups and may therefore be regarded as the differentiating ones. 21 This approach was used for the six OPLS-DA models to be sure that all the most discriminative features could be selected.
In addition, their VIP (variable influence on projection) values were evaluated to assess their relevance for the chemometric model (VIP values had to be >1.4). 22 Features that survived this process were studied, and a tentative of compounds identification was performed.
In all the chemometric models created, an internal leave 1/7 out cross-validation was executed.

| Compounds identification workflow
Compounds identification was performed according to the following steps: 1) Chemical formula hypothesis, taking into account also adducts, if present.
3) Evaluation of the mass shift between the experimental and the theoretical values.

| Markers validation procedure
The compounds identified with the workflow detailed above were searched in seven new egg products batches not used for the model creation; they were collected, extracted, and analyzed with the same experimental design previously described.
The maximum reliability of these compounds as "freshness" or "not freshness" markers implies they have to be confirmed not only with their presence or absence in the new samples but also highlighting the same increasing or decreasing trend through the time points.

| Extraction procedures and sequences evaluation
Before starting the real data elaboration, different evaluations on the internal standard results were performed.
Chloramphenicol is an exogenous compound that was added to every sample with the aim of monitoring the extraction procedures and the goodness of the analytical sequences. 25 The injection of the "standard solution" periodically during the sequence provided information about the sequence trends and helped in the detection of potential decrease in signal's intensity. For the sequence analyzed in positive ionization mode, the areas CV was 7.2%, and the retention time CV was 1.6% and, for the negative ionization mode, 10.5% (areas) and 3.1% (retention time).
According to the results obtained, we can deduce that no issues occurred during sample extraction procedures; moreover, the slight shift in the retention times is a further proof of the goodness of the analytical sequences.

| Chemometric data elaboration
The preliminary PCA obtained with the areas of the entire data set highlighted a separation between the fresh samples and the not fresh ones with a clustering of the QC samples for both positive and negative ionization modes (data not shown); the first two components described more than 45% of the variance for both the models.
After data filtering (obtained by selecting in the QC samples the features with CV lower than 40% in the area values), an improvement in the clusterization of the samples was obtained, with a tight clustering of the QC samples (Figures 1 and 2). In addition, the overlay in the scores plot of the replicate extracted samples was essentially complete.
The tight clustering of the QC samples in the center of the score plot is a further confirmation of the robustness of the analytical proce-dure; moreover, this ensure that the separation through the groups is not random but due to a real variability. 26 The overlay in the scores plot of the replicate samples certify that the extraction procedure is repeatable not only for the chloramphenicol compound but for the whole fingerprint of the egg product.
The final PCA, created with the average value of the replicates and without the QC samples, clearly highlights the separation between the fresh samples and the not fresh ones (Figures 3 and 4).
All the OPLS-DA-supervised models, as expected, extraordinarily increase the separation between the two groups. As example, Fig  The results obtained strongly indicate that the models created for both positive and negative ionization modes are reliable and robust.

| Compounds identification
Features responsible of the clusterization were selected, and a group of them was identified according to what described in the "compounds identification workflow" paragraph.
Both markers of "not freshness" (that drastically increase their intensity or even appear during the eggs ageing) and of "freshness" (that drastically decrease their intensity or even disappear during the eggs ageing) were identified.
Tables 4 and 5 summarize the compounds responsible of the "freshness" and "not freshness" classification, ranked according to the level of identification suggested by the Standard Initiative in Metabolomics. 23,24 Twelve compounds were identified with their respective reference standards, and a total amount of 31 markers (15 of freshness and 16 of not freshness) was selected. Table 6 resumes the number of features filtered out through each statistical step.
As highlighted in the markers lists, lactic acid and succinic acid were undoubtedly detected, but together with many other compounds. This could lead to a more complete evaluation of the freshness issue: potentially 31 m/z values can be simultaneously considered for this topic.

| Markers validation
As also suggested by the US Pharmacopoeia, 20 predictive chemometric models should be always validated with an external set of samples (possibly collected in different periods) that have to be treated as unknown; their classification should be predicted, to certify that the    Tables 7 and 8.
All the target molecules were detected with almost the same trend through the time points (with exception of a couple of markers but only in terms of relative ratio): These results prove that the compounds identified are real markers of "freshness" or "not freshness" in egg products and are not related only to a misleading overfitting of the chemometric model.

| Markers interpretation
Most of the markers reported in this work have not been previously identified in eggs shelf life studies, because of the analytical approach.
While egg freshness is commonly followed by volatile profiling or by e-nose fingerprinting, the use of high-resolution liquid chromatography-mass spectrometry allowed to enlarge the identification to medium-polar nonvolatile compounds.
Volatile markers are actually based on the formation of volatile aldehydes from fatty acids oxidative degradation. 5 In our study,

| CONCLUSIONS
In the current study, new compounds related to freshness in egg products samples were identified using a nontargeted metabolomic

Methionine sulfoxide
Methylhistidine N-Acetyl-α-D-galactosamine or N-acetyl-α-D-glucosamine Threonine a Time points: first bar, "t zero"; second bar, "1 day"; third bar, "2 days." approach. After model creation, the robustness of the identified markers of freshness was assessed through the analysis of a validation set of not previously used egg products.
Along with lactic acid and succinic acid, the reference compounds already considered in the EU legislation, other 29 compounds were detected as discriminant features, allowing a more robust evaluation of freshness.
Further improvements of the results presented in this study should lead to two directions. Firstly, target methods on the identified compounds could be developed for a quantitative evaluation: Quality control laboratories should only be able to detect these 31 molecules, avoiding the need of high-resolution mass spectrometry and chemometric software, that are much more expensive than a simple single-stage liquid chromatography-mass spectrometry instrument.
Secondly, this group of compounds, and generally the recorded fingerprints, could be helpful also for both the safety evaluations on microbial growth perspectives and the detection of other frauds related to egg products food chain, as, for example, the illegal use of incubated eggs.