Comparative Untargeted Metabolic Profiling of Different Parts of Citrus sinensis Fruits via Liquid Chromatography–Mass Spectrometry Coupled with Multivariate Data Analyses to Unravel Authenticity

Differences between seven authentic samples of Citrus sinensis var. Valencia peel (albedo and flavedo) and juices from Spain and Uruguay, in addition to a concentrate obtained from Brazil, were investigated by untargeted metabolic profiling. Sixty-six metabolites were detected by nano-liquid chromatography coupled to a high-resolution electrospray-ionization quadrupole time-of-flight mass spectrometer (nLC-ESI-qTOF-MS) belonging to phenolic acids, coumarins, flavonoid glycosides, limonoids, terpenes, and fatty acids. Eleven metabolites were detected for the first time in Citrus sinensis and identified as citroside A, sinapic acid pentoside, apigenin-C-hexosyl-O-pentoside, chrysoeriol-C-hexoside, di-hexosyl-diosmetin, perilloside A, gingerol, ionone epoxide hydroxy-sphingenine, xanthomicrol, and coumaryl alcohol-O-hexoside. Some flavonoids were completely absent from the juice, while present most prominently in the Citrus peel, conveying more industrial and economic prospects to the latter. Multivariate data analyses clarified that the differences among orange parts overweighed the geographical source. PCA analysis of ESI-(−)-mode data revealed for hydroxylinoleic acid abundance in flavedo peel from Uruguay the most distant cluster from all others. The PCA analysis of ESI-(+)-mode data provided a clear segregation of the different Citrus sinensis parts primarily due to the large diversity of flavonoids and coumarins among the studied samples.


Introduction
Citrus trees bear some of the most popular fruits and are grown globally for food, medicinal and other industrial applications, with a total annual production of nearly 85 million tons [1]. Various species of Citrus genus are valuable, such as C. medica (citron), C. limon (lemon), C. aurantium (sour orange), C. reticulata (mandarin, tangerine), C. paradisi (grapefruit), C. clementina (clementine) and C. sinensis (sweet orange) [2]. They are either consumed as fresh fruits or after processing to juices, beverage products, and jams, with the peel being the main by-product of processing. Anatomically, the fruit consists of two parts-the outer peel and the pulp with juice sac glands [2]. The peel main parts include the outer pigmented flavedo with parenchymatous cells and cuticle, and the white albedo part lying beneath the flavedo [3]. Different plant parts find wide application in many countries in recipes to treat stomach disorders, skin inflammation, cough, muscular pain, and nausea, as well as being used as a slimming agent [4].

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. Af orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Ald heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked with hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-fl MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operated ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion masse retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas of pound after normalization to internal standard. Aiming at comparing the rel dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to e biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal proj tent structures-discriminant analysis (OPLS-DA) was applied to a metabolit MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vario characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were us the validity of models, while all variables were mean centered and Pareto sca Albedo peel Spain AU

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. A orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Ald heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked wit hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-fl MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operated ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion masse retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas o pound after normalization to internal standard. Aiming at comparing the re dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to e biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal proj tent structures-discriminant analysis (OPLS-DA) was applied to a metabolit MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vario characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were us the validity of models, while all variables were mean centered and Pareto sca Albedo peel Uruguay CB

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. A orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Al heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked wit hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-f MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operate ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion masse retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas o pound after normalization to internal standard. Aiming at comparing the re dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal pro tent structures-discriminant analysis (OPLS-DA) was applied to a metabolit MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vari characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were u the validity of models, while all variables were mean centered and Pareto sca Juice concentrate Brazil FS

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. A orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Ald heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked wit hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-fl MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operate ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion masse retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas o pound after normalization to internal standard. Aiming at comparing the re dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to e biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal proj tent structures-discriminant analysis (OPLS-DA) was applied to a metabolit MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vari characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were u the validity of models, while all variables were mean centered and Pareto sca

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. A orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Al heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked wit hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-f MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operate ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion mass retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas o pound after normalization to internal standard. Aiming at comparing the re dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal pro tent structures-discriminant analysis (OPLS-DA) was applied to a metaboli MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vari characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were u the validity of models, while all variables were mean centered and Pareto sca

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. A orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Al heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked wit hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-f MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operate ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion mass retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas o pound after normalization to internal standard. Aiming at comparing the re dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal pro tent structures-discriminant analysis (OPLS-DA) was applied to a metaboli MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vari characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were u the validity of models, while all variables were mean centered and Pareto sca Juice Spain JU

Orange Samples and Chemicals
Valencia oranges (C. sinensis) from Spain and Uruguay, provided (Holzminden, Germany) as depicted in Table 1, were cleaned and squeezed t juice, then flavedo and albedo were manually stripped by a fruit peeler. A orange samples, including a Brazilian orange concentrate acquired from Sy lyophilized and separately ground to fine powder by a mill rotor cyclone racicaba, Brazil, TE-651/2). Comminuted orange samples were stored at −80 ° sequent analysis. All chemicals and solvents were purchased from Sigma Al heim, Germany).

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked wit hesperidin followed by sonication for 30 min, then centrifugation for 20 min to eliminate any leftover debris. Solid-phase extraction was applied to each e a C18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. T extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen equipped with a reversed phase column (150 × 0.1 mm, particle size 3 μm; M resources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-f MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operate ative and positive ion modes under conditions as previously reported [18]. Id of metabolite mass fingerprints was carried out using exact parent ion mass retention data, reference literature, fragmentation patterns, and the Phytoch tionary of Natural Products Database (https://dnp.chemnetbase.com/ a (16.3.2022)). Semi-quantification was based on the integrated peak areas o pound after normalization to internal standard. Aiming at comparing the re dance of a given compound in the seven different samples, the determination concentrations by an external calibration of every compound was not require dependent replicates of each orange specimen were analyzed in parallel to biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal pro tent structures-discriminant analysis (OPLS-DA) was applied to a metaboli MS abundances produced by nLC-MS either in the negative or positive ion m SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint vari characterizing each group declared with correlation (pcor) and covariance (p tive permutation testing and diagnostic indices, viz. R2 and Q2 values, were u the validity of models, while all variables were mean centered and Pareto sca Juice Uruguay

Metabolite Mass Fingerprinting
Orange specimens (2 mg), extracted by 1 mL methanol, were spiked with 4 µg mL −1 hesperidin followed by sonication for 30 min, then centrifugation for 20 min at 15,000× g to eliminate any leftover debris. Solid-phase extraction was applied to each extract using a C 18 cartridge (JT Baker, Phillipsburg, NJ, USA) as previously reported [17]. The resulting extracts were injected into a nano-LC system EASY-nLC II (Bruker, Bremen, Germany) equipped with a reversed phase column (150 × 0.1 mm, particle size 3 µm; Michrom Bioresources, Auburn, CA, USA) coupled to maXis impact quadrupole-time of-flight (qTOF) MS (Bruker, Bremen, Germany). A captive nano-spray ionization was operated in the negative and positive ion modes under conditions as previously reported [18]. Identification of metabolite mass fingerprints was carried out using exact parent ion masses as well as retention data, reference literature, fragmentation patterns, and the Phytochemical Dictionary of Natural Products Database (https://dnp.chemnetbase.com/ accessed on (16 March 2022)). Semi-quantification was based on the integrated peak areas of each compound after normalization to internal standard. Aiming at comparing the relative abundance of a given compound in the seven different samples, the determination of absolute concentrations by an external calibration of every compound was not required. Three independent replicates of each orange specimen were analyzed in parallel to evaluate the biological variance.

Multivariate Data Analyses (MVA)
Modeling viz. principal component analysis (PCA) and orthogonal projection to latent structures-discriminant analysis (OPLS-DA) was applied to a metabolite dataset of MS abundances produced by nLC-MS either in the negative or positive ion mode via the SIMCA-P+ 13.0 software package (Umetrics, Umeå, Sweden) to pinpoint various markers characterizing each group declared with correlation (pcor) and covariance (p). The iterative permutation testing and diagnostic indices, viz. R2 and Q2 values, were used to assess the validity of models, while all variables were mean centered and Pareto scaled. (Figure 1) of Citrus samples resulted in the identification of 66 metabolites, categorized into seven classes: phenolic acids, terpenes, limonoids, coumarins, flavonoids, fatty acids and nitrogenous compounds. The elution order of metabolites followed a sequence of decreasing polarity, whereby phenolic acids eluted first, followed by coumarins, flavonoid glycosides, limonoids, free aglyca and fatty acids. Samples were analyzed in both the negative and positive ionization modes to provide a greater coverage of the metabolome. Fatty acids and flavonoids were preferentially ionized under negative ionization conditions, while coumarins, limonoids and nitrogenous compounds showed better ionization in the positive mode. The list of identified compounds along with their retention time, characteristic molecular and fragment ions and occurrence is presented in Table 2.    Figure S2) [19].  2017) showing that the highest coumarin content was found in the flavedo part among different studied Citrus cultivars [21], conveying a particular interest regarding the presumed anticancer and antidiabetic effects of this part.     Figure S4). In the studied samples, flavedo peel from Uruguay was found more enriched in apigenin derivatives, while naringenin derivatives were found most abundant in the albedo part of both suppliers.  Figure S5). It was annotated as hesperidin, previously reported as the main flavanone in the peel of Citrus sinensis L. varieties, which suffers dramatic losses in filtered peel juice due to its relatively low water solubility [48]. Hesperidin is well known for its supposed effects on health including antimicrobial, anticancer, antihypertensive and antiulcer effects, thus attracting medicinal interest to orange peel [49].

Identification of Limonoids and Terpenes
Limonoids are tetranortriterpenoids, found extensively in Rutaceae and Meliaceae [54]. They are widely distributed in different Citrus fruits, such as grapefruit (Citrus paradisi), sweet orange (Citrus sinensis), sour orange (Citrus aurantium), lemon (Citrus limon) and lime (Citrus aurantiifolia) [55]. Water-insoluble limonoid aglycones are mainly distributed within seeds and peels. In contrast, the water-soluble limonoid glycosides are more abundant within juices and pulps [56]. In the studied Citrus samples, limonoids occur in significant amounts as aglycone and glycoside forms.  [57] and was identified as deacetylnomilin (Suppl. Figure S6). It was found more abundant in the peel than in juice or concentrate; this may be related to its relatively low water solubility.  [58]. Limonin was detected in more abundant content in albedo from Spain (AS), suggestive to have a slightly bitter taste compared to albedo from Uruguay (AU) upon providing a likely sustainable source of food additive [59].
Terpenes are important for plant aroma and flavor playing key roles in fruit quality, plant defense and pollinator attraction. A total of three terpenes was detected ionized most preferentially in the negative ionization mode, from which a gluco-conjugated megastigmadienone was identified as peak ] was an apo-carotenoid monoterpene identified as ionone epoxide, a flavoring substance with a fruity and woody flavor previously identified in many foods, such as apricot, raspberry, tea and lemon balm [43].

Identification of Fatty Acids and Fatty Acid Amides
Saturated and unsaturated fatty acids in addition to low molecular mass amide compounds were detected in several orange peel compartments. For details, refer to Supplementary Materials [28,35,44].

Identification of Nitrogenous Compounds
Other metabolites containing nitrogen were detected at trace levels. For details, refer to Supplementary Materials [34].

Multivariate Data Analyses of Orange Samples
The datasets encompassed a total of 42 nLC-ESI-MS/MS runs (seven different orange samples with three biological replicates each in both the negative and positive ionization modes). Thus, unsupervised and supervised multivariate data analyses were adopted to simplify interpretation of such complex datasets allowing better biomarkers characterization and sample classification [62]. Principal component analysis (PCA), as an unsupervised approach, was reported to evaluate the variance within various samples involving no prior knowledge of the datasets [63]. Table 1 shows the color-coded source of the Citrus sinensis specimens compared.
PCA on negative ionization runs as depicted in Figure 2A-C for all orange samples was illustrated by PC1 and PC2 accounting for 87% of the total variance (Figure 2A). Along PC1, there was a clear separation between flavedo specimens from Spain (FS) and Uruguay (FU), while the latter was distinguished from the other samples and located at upper right quadrant with positive score values, suggesting the impact of geographical source based on their metabolites. The albedo samples were positioned at the lower right quadrant, whereas juice samples were located at upper left quadrant with negative PC1 values. Conversely, orange concentrate from Brazil (CB) appeared near the origin (Figure 2A). The respective loading plot demonstrated that mono-methoxy flavonoids, i.e., hesperetin and sakuranetin along with hydroxy-oxohexadecanoic acid were found more abundant in albedo specimens ( Figure 2B). On the other hand, hydroxylinoleic acid was major contributor to flavedo from Uruguay (FU) being the most distant data point. Naringenin was responsible for the clustering of orange juice of both suppliers and flavedo from Spain (FS) in the upper left quadrant, while N-phenylacetylglycine was more abundant in juice from Spain (JS) and orange concentrate (CB). The dendrogram of HCA (Hierarchical clustering analysis), as depicted in Figure 2C, revealed that flavedo from Uruguay (FU) represented one cluster, whereas the other cluster was divided into albedo specimens at one sub-group and the other sub-group encompassed the remaining samples. Likewise, PCA on positive ionization as illustrated in Figure 2D-F revealed 86% of the total variance, albeit with better segregation ( Figure 2D) compared to its negative ionization counterpart. Particularly, all replicates from each orange sample were tightly clustered indicating the superb reproducibility of the experimental analysis in both the negative and positive ionization modes. In agreement with the previous PCA model, flavedo from Uruguay (FU) was also the most distant group. Flavedo specimens showed positive Likewise, PCA on positive ionization as illustrated in Figure 2D-F revealed 86% of the total variance, albeit with better segregation ( Figure 2D) compared to its negative ionization counterpart. Particularly, all replicates from each orange sample were tightly clustered indicating the superb reproducibility of the experimental analysis in both the negative and positive ionization modes. In agreement with the previous PCA model, flavedo from Uruguay (FU) was also the most distant group. Flavedo specimens showed positive score values separable from other orange parts, while orange juice and concentrate had negative score values. Albedo from both countries (AU and AS) was separated in the upper left quadrant. The respective loading plot ( Figure 2E) demonstrated high levels of nobiletin, tangeretin, dihydroxytrimethoxyflavone and heptamethoxyflavone in flavedo samples. Albedo specimens were enriched in sakuranetin-O-hexosyl-O-deoxyhexoside and hesperetin-O-deoxyhexoside, whereas high levels of hesperidin were observed in orange juice and concentrate. The HCA dendrogram, as depicted in Figure 2F, revealed that the flavedo from both suppliers represented one cluster, while the other cluster included the remaining orange parts.
OPLS-DA (orthogonal projection to latent structures-discriminant analysis) as a supervised approach was reported to possess a great potentiality maximizing the segregation of overlapping sample groups by identifying chemical determinants [64]. Being the most distant cluster, flavedo from Uruguay (FU) was further subjected to OPLS-DA against all other specimens ( Figure 3) to assess sample discrimination with p value less than 0.001. The first OPLS model on negative ionization ( Figure 3A) exhibited 0.96 model predictability (Q2) and 97% total variance (R2). The relevant loading S-plot showed that FU included abundant levels of hydroxylated fatty acids, i.e., hydroxylinoleic, trihydroxylinoleic, and dihydroxyoctadecadienoic acids ( Figure 3B). Conversely, the second OPLS model on positive ionization ( Figure 3C) exhibited 0.62 model predictability (Q2) and 72% total variance (R2). The relevant loading S-plot revealed that nobiletin, dihydroxytrimethoxyflavone, demethylnobiletin and tangeretin were detected in higher levels in FU (flavedo from Uruguay) ( Figure 3D). Then, several OPLS models were constructed (Suppl. Figures S7-S10) with p value less than 0.001 to assess chemical determinants in orange parts derived from the various countries. Hence, a model of flavedo from Uruguay (FU) against its counterpart from Spain (FS) was performed (Suppl. Figure S7). On negative mode, OPLS model (Suppl. Figure S7A) revealed the enrichment of flavedo from Uruguay (FU) in hydroxyl-linoleic acid, dimethylkaempferol, apigenin-di-O-hexoside and citropten (Suppl. Figure S7B). Notably, hydroxyl fatty acids other than hydroxylinoleic acid did not appear in this model, which was attributed to their similar levels present in both samples. Conversely, OPLS model in positive mode (Suppl. Figure S7C) exhibited the enrichment of flavedo from Uruguay (FU) in nobiletin, demethylnobiletin, and tangeretin, whereas heptamethoxyflavone was abundant in FS (Suppl. Figure S7D). Another OPLS model was performed with albedo from Uruguay (AU) versus its counterpart from Spain (AS) (Suppl. Figure S8). OPLS model in negative ionization (Suppl. Figure S8A) demonstrated the particular abundance of naringenin-O-hexoside and linoleic acid in albedo from Uruguay (AU), while hesperetin, sakuranetin and hydroxy-oxohexadecanoic acid were found more abundant in albedo from Spain (Suppl. Figure S8B). In positive ionization, the OPLS model (Suppl. Figure S8C) showed significantly higher levels of glycosyl flavonoids, i.e., sakuranetin-O-hexosyl-O-deoxyhexoside, limonin and hesperetin-O-deoxyhexoside in AS (albedo from Spain) (Suppl. Figure S8D). Further and for better discrimination assessment of orange juices derived from various sources, the OPLS model in negative mode (Suppl. Figure S9A) revealed abundant levels of N-phenylacetylglycine, nomilinhexoside and gingerol in orange juice from Spain (JS) (Suppl. Figure S9B). Lastly, the OPLS model including orange juice from both suppliers and orange concentrate from Brazil (CB) was implemented in positive mode (Suppl. Figure S10A). The relevant loading plot (Suppl. Figure S10B) showed that hesperidin, hydroxy-sphingenine and pyridoxamine phosphate amounted for the major metabolites in orange concentrate (CB). The obvious segregation between orange juice from Uruguay (JU) and Spain (JS) was ascribed to the abundance of heptyl caffeate in the former, while the later was enriched in nomilinhexoside and sakuranetin-O-hexosyl-deoxyhexoside.
(Suppl. Figure S9A) revealed abundant levels of N-phenylacetylglycine, nomilin-hexoside and gingerol in orange juice from Spain (JS) (Suppl. Figure S9B). Lastly, the OPLS model including orange juice from both suppliers and orange concentrate from Brazil (CB) was implemented in positive mode (Suppl. Figure S10A). The relevant loading plot (Suppl. Figure S10B) showed that hesperidin, hydroxy-sphingenine and pyridoxamine phosphate amounted for the major metabolites in orange concentrate (CB). The obvious segregation between orange juice from Uruguay (JU) and Spain (JS) was ascribed to the abundance of heptyl caffeate in the former, while the later was enriched in nomilinhexoside and sakuranetin-O-hexosyl-deoxyhexoside.

Conclusions
In this study, the metabolites in the orange peels (albedo and flavedo parts) together with the juice and concentrate from different suppliers were systematically analyzed and identified using state-of-the-art nLC-ESI-MS/MS-based, widely non-targeted metabolome analysis. A total of 66 metabolites were annotated, 37 of which were compounds shared by all samples. Furthermore, 29 differential metabolites were detected, 15 of which were mainly flavonoids and completely absent from the juice. A total of eleven metabolites were detected for the first time in Citrus sinensis: citroside A, sinapic acid pentoside, di-hexosyl-diosmetin, apigenin-C-hexosyl-O-pentoside, chrysoeriol-C-hexoside, perilloside A, hydroxy-sphingenine, xanthomicrol, coumaryl alcohol-O-hexoside, gingerol and ionone epoxide. The annotation of the novel metabolites warrants in-depth functional genomic mining to identify their various biosynthetic pathways. Comparing different fruit parts, a number of flavonoids with proposed preventive therapies against some diseases, i.e., naringenin-O-hexoside, hesperetin-O-deoxyhexoside, sakuranetin-O-hexosyl-O-deoxyhexoside, and demethylnobiletin, were completely absent from the juices, but present most prominent in the peel. Citrus peel is thus considered a renewable bio-resource of functional foods. In addition, hesperetin-O-deoxyhexoside, sakuranetin-O-hexosyl-Odeoxyhexoside and hydroxylinolenic acid were detected only in albedo and flavedo of both suppliers, albeit absent in orange juices and concentrate, suggestive to be a distinctive marker for orange peel and to increase the potential transform of Citrus by-products into valuable food ingredients, nutraceuticals, and perhaps even pharmaceuticals.
The comprehensive nLC-ESI-MS/MS metabolic profiling followed by multivariate data analyses suggested that the differences between orange parts were much more obvious than the geographical source. Generally, albedo was richer in mono-methoxylated flavonoids, while flavedo was richer in poly-methoxylated flavonoids and hydroxylated fatty acids. Moreover, further research is required to more accurately evaluate the effectiveness, the toxicity, and the mechanism of action of many PMFs as well as to increase their bioavailability by using readily accessible and appropriate drug delivery vehicles.
Further analysis of more orange samples from other sources has yet to be evaluated to obtain a bigger picture of the entire respective population of orange samples. In addition, other factors, i.e., seasonal variation, storage conditions and agricultural practices may be assessed using the same platform.