**Cross-Species Comparison of Fruit-Metabolomics to Elucidate Metabolic Regulation of Fruit Polyphenolics Among Solanaceous Crops**

#### **Carla Lenore F. Calumpang, Tomoki Saigo, Mutsumi Watanabe and Takayuki Tohge \***

Graduate School of Science and Technology, Nara Institute of Science and Technology (NAIST), Ikoma, Nara 630-0192, Japan; calumpang.carla\_lenore.bw2@bs.naist.jp (C.L.F.C.); saigo.tomoki.sn6@bs.naist.jp (T.S.); mutsumi@bs.naist.jp (M.W.)

**\*** Correspondence: tohge@bs.naist.jp; Tel.: +81-743-72-5480

Received: 20 March 2020; Accepted: 14 May 2020; Published: 19 May 2020

**Abstract:** Many solanaceous crops are an important part of the human daily diet. Fruit polyphenolics are plant specialized metabolites that are recognized for their human health benefits and their defensive role against plant abiotic and biotic stressors. Flavonoids and chlorogenates are the major polyphenolic compounds found in solanaceous fruits that vary in quantity, physiological function, and structural diversity among and within plant species. Despite their biological significance, the elucidation of metabolic shifts of polyphenols during fruit ripening in different fruit tissues, has not yet been well-characterized in solanaceous crops, especially at a cross-species and cross-cultivar level. Here, we performed a cross-species comparison of fruit-metabolomics to elucidate the metabolic regulation of fruit polyphenolics from three representative crops of Solanaceae (tomato, eggplant, and pepper), and a cross-cultivar comparison among different pepper cultivars (*Capsicum annuum* cv.) using liquid chromatography-mass spectrometry (LC-MS). We observed a metabolic trade-off between hydroxycinnamates and flavonoids in pungent pepper and anthocyanin-type pepper cultivars and identified metabolic signatures of fruit polyphenolics in each species from each different tissue-type and fruit ripening stage. Our results provide additional information for metabolomics-assisted crop improvement of solanaceous fruits towards their improved nutritive properties and enhanced stress tolerance.

**Keywords:** polyphenolics; solanaceous crops; *capsicum annuum*; pepper; tomato; eggplant; fruit ripening; metabolomics; tissue-specificity; flavonoid

#### **1. Introduction**

Solanaceae (nightshade family) is an agronomically- and botanically-diverse plant taxonomic group, with members ranging from vegetable crops through medicinal plants to ornamentals. Few representative crops of economic importance include tomato (*Solanum lycopersicum*), eggplant (*Solanum melongena*), and pepper (*Capsicum annuum*). With available genome sequences [1,2] and genetic resources from different tomato varieties and natural mutants [3], tomato has become the first model crop for fleshy fruit ripening, fruit pigmentation, specialized (secondary) metabolism, and plant defense [4–9]. Subsequently, metabolomic data of specialized metabolites from tomato fruits [4,10–12] and a tomato metabolite database [5,13–15] have been published and developed via a mass-spectrometry (MS)-based metabolomic analysis for fruit-omics approach. The metabolomics approach, focusing on specialized metabolism, is also being used on other solanaceous plants, including tobacco (*Nicotiana* spp.) [16,17], potato (*Solanum tuberosum*) [18–20], petunia (*Petunia* spp.) [21], and *Atropa* spp. [22]. However, developing similar resources for related crops is still a goal of the scientific community [23].

In the point of fruit-omics, pepper (*Capsicum* spp.) is increasingly gaining recognition as an excellent model plant for solanaceous fruit-omics [8,12,24,25]. Pepper (*Capsicum* spp.) has a world yield of 184,742 hg/ha and a world production of almost 40 million tons in 2018 [26]. Consumed either raw, cooked, as a spice, or food colorant [27], pepper have known human health benefits, such as weight reduction [28], pain relief [29], and cancer prevention [30]. Peppers can also be used as sprays for crowd control and personal defense since some cultivars cause skin irritation [31,32]. Capsaicinoids (e.g., capsaicin, 8-methyl-*N*-vanillyl-6-nonenamide) have been the main focus of metabolic diversity studies in pepper due to their dramatic bioactivities; however, pepper polyphenols are less highlighted, since fruit metabolomics focusing on polyphenols have already been well-investigated in tomato species. Furthermore, tissue-specific patterns of polyphenol distribution in pepper fruit have not been well-characterized during fruit ripening, specifically in the peel and pericarp. There is still a need to investigate metabolic shifts of polyphenols among different fruit tissues during different stages of ripening in pepper through fruit-omics analysis, because of the fact that significant metabolic shift of polyphenolic compounds during fruit ripening is observed in different fruit tissues, such as in the case of tomato species [10,33–35]. Furthermore, capsaicinoids share the same biosynthetic precursors, e.g., phenylalanine and *p*-coumarate, which can cause differences in metabolic shifts of polyphenols in pepper versus tomato. The pungent properties in pepper are due to the bioactivity of capsaicinoids, an exclusive trait amongst capsicum species and is not exhibited by other solanaceous crops. Pepper also has a variety of fruit colors among cultivars during fruit ripening, which mainly indicate variation in fruit carotenoids and chlorophylls, but also in fruit polyphenolics, in the case of purple pepper cultivars. While total amount of flavonoid aglycones have been compared between pungent and non-pungent pepper cultivars [36], further metabolomic analysis of fruit polyphenolics in terms of their tissue specificity and metabolic shift during fruit ripening should be highlighted.

On the other hand, fruit-metabolomics on eggplant polyphenolics is still in progress with only a few studies having been performed. The metabolomic analysis of different *Solanum* species, including five eggplant (*S. melongena*) accessions, three accessions from an eggplant wild ancestor (*Solanum insanum*), and two from scarlet eggplant (*Solanum aethiopicum*), through a LC-MS based strategy revealed metabolic diversity of anthocyanins, chlorogenic acid derivatives, flavonoids, triterpenoid alkaloids and triterpenoids, and novel biosynthetic frameworks [37]. In the untargeted metabolomics of the fruit of twenty-one eggplant (*S. melongena*) accessions using GC-MS and LC-MS, some accession-specific specialized metabolites were putatively identified [38]. Moreover, the metabolic quantitative trait locus (mQTL) from eggplant fruit was investigated for identification of a genomic region of productivity for chlorogenic acid and two anthocyanin pigments [39].

Plant polyphenols are a large group of plant specialized metabolites, which can be subdivided into several large sub-groups of major phyto-antioxidants, for example flavonoids and hydroxycinnamates [40,41]. In solanaceous plants, flavonoids involved in the stress resistance against abiotic and biotic stressors, such as pathogen infection [42,43], ultraviolet light (UV) [44,45], nitrogen deficiency [46], and cold temperature [47]. Flavonoids are also involved in plant reproduction, such as attracting insect pollinators and seed dispersers [48]. Pigment flavonoids, such as anthocyanins, upon absorbing visible (VIS) light contribute to the red and purple pigmentation of solanaceous fruits [8]. Intake of flavonoids and hydroxycinnamates also reduce the risk of human disease due to their anti-cancer and antioxidant activities based on in vitro assays [49–51]. Chlorogenates have recognized human health benefits, such as antioxidant, antiviral, hepatoprotective, and anti-hypoglycemic properties [52]. With their beneficial roles in human health, tissue-specific accumulation patterns of polyphenols in pepper fruit during ripening can be focused. Since polyphenols contribute a role in plant stress response, there is a propensity for such compounds to accumulate more in the fruit peel than in the pericarp, which has already been observed in tomato fruit [10,53–56].

To develop a baseline on the polyphenolic compounds already detected in our solanaceous crops of study, a phytochemical survey of polyphenolics reported in the fruits of tomato, eggplant, and pepper was investigated via a phytochemical database KNApSAcK (http://www.knapsackfamily. com/KNApSAcK\_Family) and literature search of fruit metabolomic analyses conducted on these species [4,10,25,57–59] (Figure 1A, Table 1). In tomato species, major polyphenolic compounds include derivatives of hydroxycinnamates, flavonols, and anthocyanins [5,10,13–15,57]. For eggplant, its fruit peel is rich in anthocyanins, mainly delphinidin derivatives [60], while the pericarp consists mainly of chlorogenates (CGAs) [61]. Flavonol glycosides have also been detected in the pericarp of eggplant fruit [62]. Polyphenolic compounds detected from pepper are much more diverse, with flavonoids being the main compound group and flavones and flavonols the major flavonoid subfamilies detected. Chlorogenates have also been identified from pepper fruit [59]. Based on the data for all three solanaceous crops of study, common polyphenolic compounds present in their fruit include hydroxycinnamates (specifically chlorogenates and their positional-isomers), chalconoids/stilbenoids, flavanones, and flavonol derivatives, such as rutin (quercetin-3-*O*-rutinoside) and kaempferol-3-*O*-glucosides (Figure 1A, Table 1). A survey of representative polyphenolic compounds have been reported in the metabolic analysis of the solanaceous fruits, a metabolic framework of major polyphenolics among three solanaceous species is illustrated (Figure 1B).

**Figure 1.** Major polyphenolic compounds reported in mature fruit of solanaceous crops. (**A**) chemical structure of major polyphenolic specialized metabolites found in solanaceous crops. Chlorogenates (CGAs), naringenins, and flavonols are shown. (**B**) the biosynthetic framework of polyphenolic compounds in solanaceous crops. Biosynthetic pathways were constructed by an online database search and literature review of major polyphenolic compounds in solanaceous crops. Color: Blue, hydroxycinnamates; black, stilbenoids/flavanones and chalcones; purple, anthocyanins; and green, flavonols/flavones; orange, amino acid. Yellow color inside of circle indicates major accumulation form in the plant. Abbreviations: Phe, phenylalanine; cou, coumaroyl; caf, caffeoyl; fer, feruloyl; sin, sinapoyl; glyc, glycoside; deriv, derivative. K, kaempferol; Q, quercetin; Narichal, naringenin chalcone; CGA, chlorogenate; DiCGA, dicaffeoyl-chlorogenate; Glc, glucose; Rha, rhamnose; Api, apiose; and hex, hexose.

Peppers consist of approximately 35 species [63] with five of them domesticated independently (*C. annuum*, *C. chinense*, *C. frutescens*, *C. baccatum,* and *C. pubescens*). *C. annuum* is the largest and most widely-cultivated species having both spicy (chili) and sweet varieties with different kinds of pigmentation in their fruits [8,64]. However, due to hybridization, pepper currently has around 50,000 varieties, providing a wide range of chemical variability within the same species exhibiting similar physical characteristics [12]. Despite such physical similarity within species, such genetic diversity still provides a large pool of chemical variability within species in terms of tissue-specificity, developmental ripening stage, and cultivar. Furthermore, the process of fruit ripening involves a tight metabolic regulation in conjunction with developmental stage [65], involving biochemical reactions resulting in changes in fruit flavor, texture, aroma, hardness, nutrient composition, and color [25,66]. Since some polyphenolic compounds are involved with flavor and fruit color, changes in polyphenolic content can occur during fruit ripening [15,67]. Accumulation patterns of major flavonoid glycosides in terms of subspecies and cultivar–specificity were evaluated from ripe fruits across thirty-two pepper accessions, including *C. annuum*, *C. chinense*, *C. frutescens*, and *C. baccatum* [25]. As one of the results given by this research, main pepper flavonoid decorations were observed as a metabolic polymorphism within these pepper species. Previously, primary metabolites (sugars, amino acids, and organic acids) were compared between tomato (*S. lycopersicum*) and pepper (*Capsicum chilense*) during fruit ripening [66,68] but not specialized metabolites, specifically polyphenols. Since some metabolomic studies of primary metabolites have been conducted for pepper fruit, such fruit-metabolomics studies should be expanded to include tissue-specificity of polyphenol accumulation in pepper peel versus pericarp during different stages of fruit ripening among various pepper cultivars. Furthermore, after metabolic polymorphisms of polyphenolic profiling were characterized through metabolic profiling with acid hydrolysis [49], a complete representation of the polyphenolic profile for this pepper variety can additionally be focused with metabolic profiling of the individual derivative forms. Meanwhile in hot pepper (*Capsicum annuum* "CM334") fruit, only the pericarp was analyzed for polyphenolic accumulation patterns during ripening [69]. Additionally, the identification of polyphenolic compounds specific for pepper against other solanaceous crops has not yet been performed, nor have the patterns of accumulation in relation with fruit tissue type been reported in some pepper species. Therefore, polyphenolic fruit-omics data developed in tomato research can possibly be used for the extension to other solanaceous crops, such as pepper species and cultivars through other fruit-omics approaches.


**Table 1.** Major polyphenolic compound groups reported in the solanaceous fruits.

Here, we performed a cross-species comparison of representative phenolic compounds among three solanaceous crops (tomato, eggplant, and pepper) during fruit ripening, additionally including different pepper cultivars to address metabolic regulation of fruit polyphenolic metabolism among pepper varieties exhibiting differences in pigmentation and pungency. We focused on differences in the metabolic accumulation pattern between tissue types (peel and pericarp) and three fruit ripening stages (immature, green/purple mature, and final mature) to develop a better understanding of metabolic trade-off in fruit polyphenolics. Such information can allow breeders to optimize the biosynthesis of health-related polyphenolic compounds in pepper during their developmental stage of harvest. Metabolic signatures provided here, provide significant information for future functional genomics approach as well as for the metabolomics-assisted crop breeding of solanaceous fruits towards their nutritive improvement and stress tolerance enhancements.

#### **2. Results**

#### *2.1. Metabolite Profiling of Major Polyphenolic Compounds in Fruit Tissues Among Solanaceous Crops*

Metabolomic profiling of representative polyphenolic compounds from the fruit extracts of two tissue types (peel and pericarp) was conducted on three major solanaceous crops through liquid-chromatography-mass spectrometry (LC-MS) (Supplemental Table S1). Tomato, eggplant, and pepper were cultivated in the field and harvested for fruit-metabolomics analysis. Jalapeño pepper, a variety of chili pepper producing capsaicinoid compounds [70], was chosen for our cross-species analysis in order to understand tissue-specific metabolic trade-offs during metabolic shifts in fruit polyphenolics. Due to available studies on tissue-specific accumulation patterns of polyphenolic compounds in tomato fruit, tomato was used for the reference extracts for the peak annotation and analysis of metabolic change during fruit ripening. In our analysis, thirty-nine polyphenolic compounds were detected and annotated (Figure 2A; log2FC (mature/immature), up, > 2.0; down, < 0.5), including eggplant and pepper-specific flavonoid-derivatives. An upregulation of 19 polyphenolic compounds in the peel (4 chalcones, 9 flavonols, and 6 hydroxycinnamates) and 14 polyphenols in the pericarp (4 chalcones, 5 flavonols, and 5 hydroxycinnamates), were observed (Figure 2B) in tomato. Additionally, 8 polyphenolic compounds had higher upregulation in the peel than in the pericarp for tomato, which include one hydroxycinnamate (cou\_hex\_II, coumaroyl-hexoside II) and 7 flavonols (Q3G2A6R7G, quercetin-3-*O*-(2-*O*-apiosyl-6-*O*-rhamnosyl)glucoside-7-*O*-glucoside; Q3G2A6R, quercetin-3-*O*- (2-*O*-apiosyl-6-*O*-rhamnosyl)glucoside; K3G2A6R, kaempferol-3-*O*-(2-*O*-apiosyl-6-*O*-rhamnosyl) glucoside; rutin; K3G6R, kaempferol-3-*O*-(6-*O*-rhamnosyl)glucoside; Q3G, quercetin-3-*O*-glucoside; Q3R, quercetin-3-*O*-rhamnoside; and K3R, kaempferol-3-*O*-rhamnoside), as reported by Mintz-Oron et al. 2008 [10] in metabolomic analysis of tomato cv. Ailsa Craig. As reported in the other reports [53–56], the most abundant flavonol-glycoside, rutin, had more elevated levels in the peel than the pericarp (Figure 2A).

A cross-species comparison of the accumulation patterns of representative polyphenolic compounds between two tissue types from the fruits of three solanaceous crops was conducted. Two hydroxycinnamates (cou\_hex\_II; caf\_hex\_III, caffeoyl-hexoside III) were upregulated in the fruit peel among all three crops, while none were commonly upregulated in the pericarp. Between tomato cv. M82 and Jalapeño pepper, six polyphenols were commonly upregulated among their fruit peel while none were upregulated in the pericarp. Out of the six polyphenols upregulated in the fruit peel of both tomato cv. M82 and jalapeño pepper, two were hydroxycinnamates (caff\_hex\_I and caf\_hex\_II), two were flavonols (Q3G6R7G; K3G6R), and two were chalcones (narichal, naringenin chalcone; narichal\_hex\_I, naringenin chalcone-hexoside I). Meanwhile, a comparison between jalapeño pepper and eggplant indicated an upregulation of three hydroxycinnamates (cou\_hex\_III; feru\_hex\_I, feruloyl-hexoside I; feru\_hex\_II) in both crops for their fruit peel, while only feru\_hex\_I was upregulated in the pericarp for both. Species-specific polyphenols were also upregulated in jalapeño pepper, wherein coumaroyl-hexoside I (cou\_hex\_I) was specific for pepper in both the fruit peel and pericarp. Two other hydroxycinnamates were specifically upregulated only in pepper, with 5-chlorogenate (5CGA) upregulated in the peel and feru\_hex\_II in the pericarp. Jalapeño pepper showed a greater number of downregulated polyphenols in the peel (14) and pericarp (26), with an especially significant reduction of flavonol-derivatives in both peel and pericarp at fruit mature stage. Our cross-species comparison of fruit-omics with tissue specificity, indicated that pepper polyphenolic metabolism clearly has an opposite metabolic shift between hydroxycinnamate and flavonoid biosynthesis during fruit ripening in both peel and pericarp (Figure 2).

**Figure 2.** Metabolite profiling of major polyphenolic compounds between fruit tissues among three major solanaceous crops. (**A**) metabolic shift during fruit ripening (log2FC(mature/immature)) in peel and pericarp of tomato, eggplant and pepper fruits is presented. (**B**) venn's diagrams show conserved metabolic changes within and between fruit tissues among three solanaceous crops. Metabolites which are commonly up- (>2.0) or down-regulated (<0.5) are shown. MeV software (http://www.mev.tm4.org/) was used for data visualization. Abbreviations: cou, coumaroyl; caf, caffeoyl; fer, feruloyl; CGA, chlorogenate; DiCGA, dicaffeoyl-chlorogenate; K, kaempferol; Q, quercetin; Glc/G, glucose; Rha/R, rhamnose; Api/A, apiose; hex, hexose; Phi, phloretin; narichal, naringenin-chalcone; flv, putative flavonol-derivatives.

#### *2.2. Metabolic Shifts of Polyphenolics Di*ff*erent Pepper Cultivars During Fruit Ripening*

The metabolic analysis of polyphenolic compounds in pepper fruits was extended from the pungent jalapeño pepper to include other sweet pepper cultivars. Due to the high variability among pepper cultivars in terms of color, shape, and pungency, six pepper cultivars of different visible phenotypes caused by differences in pigment composition (elphinidinnone, carotenoid-type, and anthocyanin-type) were selected. Six different cultivars of pepper (*C. annuum*) including four sweet cultivars (green paprika, green pepper, yellow paprika, and red paprika), a pungent cultivar (jalapeño pepper), and an anthocyanin-type pepper (purple pepper) were selected for comparison during three fruit ripening stages (immature, green mature, and red or final mature) (Figure 3). In terms of color, five cultivars had red-colored final mature fruits while yellow paprika was yellow at final maturity. During the immature and green mature stages of ripening, purple pepper was purple-colored while the other five cultivars were colored green. In terms of pungency, jalapeño pepper was the only pungent cultivar and the other cultivars were non-pungent or sweet. In terms of fruit shape, purple and jalapeño peppers were pointed while the other cultivars were bell-shaped. Purple pepper differs from other sweet pepper cultivars since most sweet cultivars are bell-shaped [71].

Upon comparing the polyphenol distribution patterns between fruit peel and pericarp, no clear common pattern was observed among the six pepper cultivars, however they are slightly separated in each tissues and cultivars (Figure 4A). All anthocyanins which are involved in pigmentation, specifically purple to black pigmentation in pepper tissues [72], were only detected in purple pepper and not in other pepper cultivars, since purple pepper is the only cultivar exhibiting a developmental stage with purple color fruit (Figure 4B and Supplemental Table S1).

**Figure 3.** Pepper cultivars used in this study. Fruits from jalapeño pepper cv. "Jalapeño", red paprika pepper cv. "Flupy Red EX", purple pepper cv. "Nara Murasaki", green pepper cv. "Miogi", yellow paprika cv. "Sonia Gold", and green paprika pepper cv. "New Ace", were cultivated for our analysis.

In the pungent cultivar (jalapeño pepper), most polyphenols were upregulated in the peel during both green mature (14 polyphenols) and final mature (14 polyphenols) stages. Meanwhile, most polyphenols were upregulated in the pericarp during its immature stage (24 polyphenols). More hydroxycinnamates were upregulated in the peel (9) than in the pericarp (4) during its red mature stage. An elevation of eight hydroxycinnamates, namely coumaroyl-hexosides (cou\_hex I, II, III), caffeoyl-hexosides (caff\_hex I, II, III), and feruloyl-hexosides (feru\_hex I and II), as well as a reduction of almost all flavonoids except rutin and putative flavonoid one (flv\_1), were suggested as candidate fruit ripening markers in the peel in pungent jalapeño pepper. However, these metabolic shifts were not observed in the other pepper cultivars except purple pepper. For the non-pungent cultivar (green pepper), more hydroxycinnamates were upregulated in the pericarp (11) than in the peel (1) at its red mature stage. Reduction of chlorogenates (3CGA, 4CGA, and 5CGA), were observed in the peel, but CGAs were elevated in the pericarp. Di-caffeoyl-chlorogenate\_I (DiCGA\_I) is the product of both 3CGA and 4CGA while DiCGA\_II is the product of both 3CGA and 5CGA. The upregulation of both DiCGA\_I and DiCGA\_II suggested to explain the absence of 3CGA from the pericarp. In the sweet green paprika pepper, flavonoids were upregulated in the pericarp during its red mature stage. Furthermore, more hydroxycinnamates were upregulated in the pericarp (11) than in the peel (1) at its red mature stage. Eleven candidate ripening markers were identified for green paprika pepper, which include: Three chlorogenates (3CGA, 4CGA, and 5CGA), three di-chlorogenates (DiCGA I, II, III), two coumaroyl-hexosides (cou\_hex II, III), one caffeoyl-hexoside (caff\_hex\_III), and two feruloyl-hexosides (feru\_hex I, II). In red paprika pepper, most polyphenols were upregulated in both peel (21 polyphenols) and pericarp (22 polyphenols) during its immature stage. Only two hydroxycinnamates were upregulated its red mature stage in the peel (cou\_hex\_I and cou\_hex\_III) while one hydroxycinnamate accumulated more at its red mature stage in the pericarp (cou\_hex\_I). In yellow paprika pepper, only two hydroxycinnamates (3CGA and DiCGA\_III) were upregulated in the peel, while three hydroxycinnamates (DiCGA\_III, fer\_hex\_I and fer\_hex\_II) were upregulated in the pericarp at its red mature stage. Flavonoid mono- and di-glycosides were upregulated during the course of ripening for both tissues, specifically K3G6R in the peel and K3R in the pericarp. In purple pepper, more hydroxycinnamates were upregulated in the peel (7) than in the pericarp (3) in its red mature stage. Finally, one chlorogenate (3CGA), two di-chlorogenates (DiCGA I, III), one

coumaroyl-hexoside (cou\_hex\_I), two caffeoyl-hexosides (caff\_hex I, II), and one feruloyl-hexoside (fer\_hex\_II), were selected as ripening marker for polyphenolics of purple pepper (Figure 4B). K3G6R, Q3R, rutin, and Q3G2A6R were upregulated in both tissues during immature stage in purple pepper. Finally, the metabolic shifts observed in the cross-species comparison was the opposing direction of metabolic shift between hydroxycinnamate and flavonoid biosynthesis during fruit ripening in both the pungent cultivar (jalapeño pepper) and anthocyanin-producing purple pepper, but not in other carotenoid-type pepper cultivars.

**Figure 4.** Metabolic shifts of polyphenolics in different fruit tissue among six pepper cultivars during fruit ripening**. (A**) principal component analysis (PCA) of polyphenolics in different pepper cultivars during fruit ripening. The plots were applied for the 39 metabolites with the average values from 3 biological replications. PCA was conducted by the MultiExperiment Viewer [73]. Principal component (PC) triangles and circles indicate peel and pericarp, respectively. Coefficient correlation was estimated by person correlation method using MeV software. (**B**) heatmap visualization of metabolite data is normalized and scaled by log2FC (mean/average\_mean) for each metabolite. Abbreviations: cou, coumaroyl; caf, caffeoyl; fer, feruloyl; CGA, chlorogenate; DiCGA, dicaffeoyl-chlorogenate; Glc/G, glucose; Rha/R, rhamnose; Api/A, apiose; hex, hexose; Phi, phloretin; flv, putative flavonol-derivatives; Narichal, naringenin chalcone; Im, immature stage; Gm(Pm), green mature (or purple mature stage); and Rm(Ym), red mature (or yellow mature stage).

#### **3. Discussion**

Polyphenols are primarily involved in physiological response against abiotic and biotic stressors [42, 44,46,47] and occasionally in plant reproduction [48]. Due to the phytochemical functions of these compounds, polyphenolic which are accumulating on the surface of the fruit, have been focused. Tomato polyphenols including flavonoids and hydroxycinnamates indeed were identified at higher levels in the fruit peel than in the pericarp [10]. Because of the recognized human health benefits of these polyphenolic compounds, their accumulation patterns during different stages of ripening in economically-important crops would be important to both farmers and consumers. Comparison of representative polyphenolic compounds that accumulate in the peel and pericarp during three different stages of fruit ripening, particularly polyphenols that are specifically up-regulated only in jalapeño pepper. These polyphenolic compounds are also upregulated in the other sweet pepper cultivars, but at different stages of fruit ripening (Figure 4).

Comparison of individual polyphenolic compounds suggest possible metabolic markers that are specifically upregulated during fruit ripening. In the pungent pepper cultivar (jalapeño pepper), we observed an elevation of three coumaroyl-hexosides, three caffeoyl-hexosides, and two feruloyl-hexosides, as well as reduction of almost all flavonoids in the peel at the final mature stage (Figure 4B). However, these metabolic shifts were not observed in the other pepper cultivars except purple pepper, which happens to be non-pungent but is the only cultivar in our study that has purple pigmentation before reaching its final maturity stage. In the previous study of hot and semi-hot pepper cultivars (*C. annuum* cvs. Cyklon, Bronowicka Ostra, Tajfun, and Tornado), individual hydroxycinnamate content increased in the pericarp from green mature to red mature stage of development in all cultivars [74]. Additionally, in the analysis of a sweet bell pepper cultivar (*C. annuum* L. cv. Vergasa) exhibited a decrease in total hydroxycinnamate content in the pericarp from their immature green to green mature stage and then slightly increasing to immature red and red mature stages [75]. Taking into both results and our result in this study, results from the previous studies are consistent with our results such that hydroxycinnamate content increased in pungent cultivars from green to red stages of fruit ripening while hydroxycinnamate content decreased in sweet cultivars from immature to mature stages. Interestingly, purple pepper which is an anthocyanin-producing type of pepper cultivar showed similar metabolic changes with the pungent pepper cultivar, wherein the accumulation of hydroxycinnamates is inversely related with that of flavonoids. In both cultivars, hydroxycinnamates were generally upregulated while flavonoids were downregulated in the peel during their red mature stage. The pungent nature of jalapeño pepper due to its production of capsaicinoids could explain its downregulation of flavonoids during its red mature stage, given that both compounds groups share the same biosynthetic precursor.

Previously, in the metabolic analysis of fruit pericarp of the pungent hot pepper (*Capsicum annuum* "CM334") during six stages of fruit ripening, the flavonoid, Q3R, had high levels during earlier stages (16 and 25 DPA) and then gradually decreased until their last stage (48 DPA) [69]. In our study, Q3R in the pericarp of the hot cultivar (jalapeño pepper) was upregulated during its immature stage and then downregulated at its green and red mature stages, which are consistent with the decrease in Q3R observed from the previous study. Total flavonoid content, total *O*-glycosylflavone content and total *C*-glycosylflavone content in the pericarp of the sweet cultivar (*C. annuum* L. cv. Vergasa) decreased during four stages of ripening. Q3R decreased in the pericarp during the four stages of ripening [75], while in our study, Q3R was upregulated in the pericarp during immature stage and then downregulated in the middle stage (green or purple mature) and final mature (red or yellow mature) stages in the sweet cultivars (red paprika, yellow paprika), a non-pungent cultivar (purple pepper), and a pungent cultivar (jalapeño pepper). However, in another sweet cultivar (green paprika), Q3R was downregulated in immature and green mature stages and upregulated during red mature stage. In another non-pungent cultivar (green pepper), Q3R was upregulated during immature and red mature stages and downregulated in green mature stage of ripening. Variability in Q3R regulation among the six pepper cultivars is consistent with the subspecies or genotype-specific accumulation

pattern of flavonoid glycosides among 32 pepper accessions from a previous study, wherein specific accessions contained higher levels of flavones, flavanones, and flavonol glycosides [25].

In our results, flavonol mono-glycosides (Q3G, Q3R, and K3R) were generally decreased during ripening in the peel for almost all pepper cultivars. This metabolic shift pattern was observed inversely in the level of naringenin chalcones which increased in the peels during fruit ripening. In spite of the metabolic changes of flavonol mono-glycosides in peel, these flavonols were increased in green paprika pericarp during fruit ripening which also the same for the naringenin chalcones, including Phi35diGlc. Rutin and flavonol-tri-glycosides showed a slight shift in decrease at late stages among all pepper cultivars for both peel and pericarp during ripening. In green pepper, flavonol di- and tri-glycosides decreased in the peel during ripening, were increased in the pericarp during fruit development. The flavonol-tetra-glycosides, Q3G2'A6R7G which was detected in tomato fruits, was not detected in all pepper species in both tissue types. In red paprika pepper, flavonol di- and tri-glycosides decreased in the peel during fruit ripening, while in the pericarp, flavonol di-glycosides also decreased. There were nine putative flavonoids identified during analysis with most of them decreasing in the peel and pericarp of three cultivars (red paprika, jalapeño, and purple peppers). For green pepper, most of the putative flavonoids decreased in the fruit peel during ripening but increased in the pericarp. No obvious pattern among the nine flavonoids were observed for green and yellow paprika peppers. Flavonol mono-glycosides (Q3G, Q3R, and K3R) decreased in the peel during fruit development for most pepper cultivars, except in green paprika pepper which exhibited highest accumulation during its green mature stage. Meanwhile, in pepper pericarp, flavonol mono-glycosides decreased during maturity only for both red paprika and jalapeño peppers, while no concrete pattern was present for the other pepper cultivars.

Anthocyanin derivatives (anthocyanin\_I, delphinidin-3-*O*-(-feruloyl) rutinoside; anthocyanin\_II, delphinidin-3-*O*-(-*p*-coumaroyl) rutinoside-5-*O*-glucoside) were upregulated in purple pepper during either the purple immature or purple mature stages in both fruit tissues (Figure 4). Anthocyanin II (delphinidin-3-*O*-(-*p*-coumaroyl)rutinoside-5-*O*-glucoside) was also previously detected in immature purple pepper whole fruit [76] and peel [77]. Anthocyanin I being one of the major anthocyanins identified from eggplant fruit peel [78]. Anthocyanins I, II, and III were also upregulated in eggplant peel in our study (Figure 2). Anthocyanins are involved in pigmentation, specifically purple to black pigmentation in pepper fruit tissues [72] and in the purple pigmentation in the peel of eggplant fruit [78]. Metabolic changes of these anthocyanins are slightly different in terms of ripening stage in peel and pericarp. Anthocyanins I and II were upregulated in the peel during immature stage in purple pepper while they were upregulated in the pericarp during purple mature stage. Purple pepper is the only cultivar that was unable to detect more than one of the putative flavonoid compounds and the only cultivar that detected anthocyanins, indicating a possible inverse relationship between the two compound groups.

Most naringenin chalcones and Phi35diGlc (phloretin-3 ,5 -di-*C*-glucoside) were upregulated in the fruit peel in sweet and non-pungent cultivars during their final mature stage (red or yellow mature). Most naringenin chalcones and Phi35diGlc were upregulated in the peel of pungent jalapeño pepper during green mature stage. Due to the pungent nature of jalapeño pepper, differences in metabolic regulation with sweet cultivars could account for the down-regulation of most naringenin chalcones during red mature stage in jalapeño pepper. The pattern of accumulation is different in the fruit pericarp, with all three naringenin chalcones being upregulated at final maturity only for green pepper and green paprika peppers. Only naringenin-chalcone-hexose and phloretin dihexoside have been previously detected in non-pungent pepper (*C. annuum)* cultivars [79]. Accumulation of naringenin chalcones in the fruit peel is related with their functions of attracting seed dispersers, moderating damage against UV-light [55], and providing a structural role in the cuticle by controlling water movement [80]. Concentration of naringenin chalcone and its derivatives might have important roles after fruit maturation, particularly in the non-pungent and sweet cultivars where naringenin chalcones are upregulated in the peel during final mature stage of ripening.

Environmental stress, such as high temperature and UV-light, can increase reactive oxygen species (ROS) concentration resulting in fruit oxidative damage. Specialized metabolites can act as antioxidants protecting the fruit against photooxidative damage [81,82]. Polyphenols, such as flavonoids and hydroxycinnamates, act as antioxidants [40,41]. In solanaceous crop tomato (*S. lycopersicum*), total antioxidant activity in ripe fruit was in agreement with their total polyphenol content which corroborates that polyphenols can act as antioxidants [83]. In the perennial shrub *Ribes stenocarpsum* Maxim, polyphenol content was significantly higher in immature than in mature fruits, with antioxidant activities consequently being higher in immature fruits [84]. In Alphonso mango (*Mangifera indica*) fruit, a positive correlation between phenolic and antioxidant content in healthy tissues was present during the course of fruit ripening [85]. Ripening in sweet pepper (*C. annuum*) is associated with oxidative stress due to an increase in lipid peroxidation [86,87] and decrease in antioxidant enzymes during its red ripe stage [86]. In our study, polyphenols were generally upregulated during red mature stage in the pericarp of sweet green paprika (*C. annuum*) and non-pungent green pepper *(C. annuum),* which is in agreement that fruit ripening in sweet pepper is related with oxidative stress since polyphenols can act as antioxidants. In tomato, total phenolics and free radical scavenging activity increased during ripening for all cultivars under study [88]. Decrease in antioxidant enzyme activity during fruit ripening involved the enzymes catalase [89] and ascorbate peroxidase [86] which are not relevant with polyphenol biosynthesis.

Comparing the number of polyphenolic compounds that accumulated more in the peel against the pericarp showed difference between fruit tissue type for the six pepper cultivars. However, not in all cases were the polyphenolic compounds more abundant during the same stage of development between tissue types. In peel, the compounds upregulated were 5CGA and fer\_hex\_II. 5CGA only increased during fruit ripening for the pungent cultivar and decreased for the sweet cultivars. Fer\_hex\_I and fer\_hex\_II increased remarkably in the peel of the pungent cultivar (jalapeño pepper) although some increase was observed in purple pepper as well. In the fruit pericarp, fer\_hex\_I increased noticeably in the pungent jalapeño pepper; however, there was also some increase in yellow and green paprika peppers. To determine whether such accumulation patterns are specific to pungent cultivars or whether this accumulation is only cultivar-specific, accumulation patterns for these polyphenolic compounds need to be conducted among pungent pepper cultivars during fruit ripening. We also observed that similar metabolic shift of pepper specific flavonoid derivatives between green pepper and green paprika, but these metabolic changes were negatively correlated to the metabolic shift in the jalapeño pepper which is the capsaicinoids producing-type of cultivar. This result provides a metabolic trade-off of fruit polyphenolics metabolism in capsaicinoids-producing cultivars, since polyphenolics and capsaicinoids share the biosynthetic precursors. In contrast to this point, jalapeño pepper showed clear elevation of hydroxycinnamates, such as caffeoyl-hexosides, in both peel and pericarp. However, these metabolic changes were also observed in the peel of green pepper and green paprika, but not in the other red, yellow, and purple peppers. Taking into account both metabolic changes of polyphenolic subgroups, changes of metabolic flux in polyphenolics are specific metabolic regulation in each types of peppers with different tissue specific manners. Finally, our results may have been convoluted by the absence of any pattern among polyphenols in terms of maturity stage and cultivar type, however integration of metabolomics data with previous studies [24] will provide other novel insights to understand this convoluted metabolic regulations. Importantly, naringenin chalcone which is one of the major ripening marker metabolites in tomato fruits, was conserved among almost all pepper species. Metabolomics analysis presented here suggested metabolic shift including convoluted metabolic trade-off in the solanaceous crops and provided hints for metabolomics-assisted crop improvement for the polyphenolic metabolism in the solanaceous fruits for the improvement of nutritive properties and enhanced stress tolerance.

#### **4. Materials and Methods**

#### *4.1. Plant Material and Sampling*

Fruits from tomato (*S. lycopersicum* cv. "M82") (TGRC, Tomato Genetics Resource Center, Davis, CA, USA), eggplant cv. "Ryoma" (*S. melongena*) (Takii, Japan), and different pepper (*C. annuum*) cultivars (green paprika cv. "New Ace" (Takii, Japan), red paprika cv. "Flupy Red EX" (e-taneya, Japan), purple pepper cv. "Nara Murasaki" (Tsurushin, Japan), green pepper cv. "Miogi" (e-taneya, Japan), yellow paprika cv. "Sonia Gold" (Sakata-no-Tane, Japan), and jalapeño pepper cv. "Jalapeño" (Marche, Japan)) were grown from May–December 2019 in standard soil under open field (longitude, 34.734433; latitude 135.736754) conditions. Three developmental stages were defined for analysis from fruit peel and flesh. Immature fruit, mid-stage, and final mature stages were selected and defined based on the height of each fruit and the color of peel and seeds. Immature fruit are half the height of mature fruit and green in color. In the case of purple pepper and eggplant, the immature fruit color was purple. Green mature fruit had approximately the same height as the final mature stage but still green in color. Purple pepper was also purple during its green mature stage. The final mature stage is when peel color has completely changed from green to its ripe fruit color (red, yellow). Mature eggplant fruit were distinguished through their dark brown seed color. Three independent biological replicates per plant tissue during each ripening stage were harvested and used for metabolomic analysis. After tissue separation, fruit peel and pericarp were grounded into powder using liquid nitrogen and stored at −80 ◦C for further analysis.

#### *4.2. Sample Extraction and LC-MS Analysis*

Metabolite extraction was performed as described previously [90,91]. Fifty milligrams of powdered frozen tissue sample were aliquoted and weighed in a 2.0 mL centrifuge tube. Two hundred fifty microliters of extraction solvent (80% methanol in LC-MS grade water with 5 μg/mL isovitexin standard) were added per tissue sample in liquid nitrogen and a Zirconia bead. All frozen tissue samples were ground into powder using Mixer Mill TissueLyser II (Qiagen, Hilden, Germany) for 3 min at 25l/s at room temperature, and centrifuged at 15,366× *g* at 4 ◦C for 10 min. After centrifugation, to take an additional cleaning step of extracts to exclude plant tissues, two hundred microliters of supernatant per sample were transferred into a 1.5 mL centrifuge tube then all 1.5 mL tubes were centrifuged at 15,366× *g* at 4 ◦C for 10 min. One hundred microliters supernatant per sample were transferred into LC-MS vials and stored at 4 ◦C in the dark until analysis. For detection of polyphenolic metabolites, LC-electrospray ionization (ESI)-MS was used. Nanoflow-HPLC "Paradigm MS4 system" (Michrom BioResources, Inc., Auburn, CA, USA), equipped with a Luna C18 column (150 by 2.00 mm i.d. 3 micron particle size, Phenomenex, Torrance, CA, USA) operated at a temperature of 25 ◦C was used. The mobile phases consisted of 0.1% formic acid in water (Solvent A) and 0.1% formic acid in acetonitrile (Solvent B). The flow rate of the mobile phase was 200 μL/min, and 10 μL sample were loaded per injection. The following gradient profile was applied: The concentration of mobile phase A was 100% at 0 min, 93% at 1 min, decreased to 80% at 8 min, 60% at 17 min, 15% at 21 min, and 0% at 25 min and 28 min for column wash, then increased to 100% at 28.01 min and 31 min for the equilibration of the column in the gradient description. The LC was connected to an MS TSQ Vantage (Thermo Fisher Scientific, San Jose, CA, USA). The spectra were recorded using full scan mode, covering a mass range from *m*/*z* 200–1500 by both positive and negative ion detection. The transfer capillary temperature was set to 350 ◦C. The spray voltage was fixed at 3.00 kV. Peak identification of major polyphenolic compounds (rutin, quercetin-3-*O*-rutinoside; Q3G, quercetin-3-*O*-glucoside; and 3CGA, chlorogenate) was performed using standard compounds. Peak annotation of major polyphenolic compounds in Solanaceae species was performed via combination approach of co-elution profile of common tomato fruit extracts and phytochemical database (KomicMarket [14] and MotoDB [13,92]) as well as metabolites table in the literatures [5,15].

#### *4.3. Data Analysis*

Molecular masses, retention time, and associated peak intensities were extracted from the raw files using the Xcalibur software version 4.1.31.9 (Thermo Fisher Scientific, San Jose, CA, USA). Compounds were identified and putatively identified by comparing with corresponding retention time (minutes) and molecular weight with those provided by tomato cv. M82 and eggplant. Previous information on polyphenolic compounds identified from different pepper species and varieties from published journal articles were also used for cross-referencing. Peak picking in the Xcalibur software was performed with the parameter of RT tolerance window (20 s), base window 80, area noise factor 5.0, peak noise factor 10, and "nearest RT". MeV software version 4.9.0 (http://www.mev.tm4.org/, Dana Farber Cancer Institute, Boston, MA, USA) was used for data visualization and PCA analysis. The plots were applied for the 39 metabolites with the average values from 3 biological replications. Heatmap visualization of metabolite data is normalized and scaled by log2FC (mean/average\_mean) for each metabolite. Coefficient correlation was estimated by person correlation method using MeV software.

#### **5. Conclusions**

With several fruit-metabolomics studies on the specialized metabolism of tomato recently available, such an approach could, therefore, be extended to generate information on other members from Solanaceae. Comparing the number of polyphenolic compounds that accumulated more in the peel against the pericarp showed differences between fruit tissue type in the six pepper cultivars. However, not in all cases were the polyphenolic compounds more abundant during the same stage of development between tissue types. We also observed that a similar metabolic shift of pepper-specific flavonoid derivatives between green pepper and green paprika cultivars, but these metabolic changes were negatively correlated to the metabolic shift in jalapeño pepper, which biosynthesizes capsaicinoids. This result exhibits a metabolic trade-off in fruit polyphenolics metabolism in the capsaicinoid-producing cultivar, since polyphenolics and capsaicinoids share the biosynthetic precursors. In support of this point, hydroxycinnamates in the pungent jalapeño and anthocyanin-producing purple pepper cultivars were clearly elevated in both peel and pericarp during red mature stage. However, flavonoids from both cultivars were downregulated during red mature stage suggesting a metabolic trade-off between both compound groups during fruit development. Taking into account both metabolic changes of polyphenolic subgroups, changes of metabolic flux in polyphenolics are specifically regulated in each pepper cultivar with different tissue specific manners. Finally, our results may have been convoluted by the absence of any accumulation patterns of polyphenol in terms of ripening stage and cultivar type, however integration of metabolomics data with previous studies [24] will provide other novel insights to understand these convoluted metabolic regulations. Importantly, naringenin chalcone which is one of the major ripening marker metabolites in tomato fruits, was conserved among almost all pepper cultivars. Metabolomic analysis presented here suggested metabolic shift including convoluted metabolic trade-off in three solanaceous crops and provided hints for metabolomics-assisted crop improvement of polyphenolic metabolism in three solanaceous crops towards their improved nutritive properties and enhanced stress tolerance.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2218-1989/10/5/209/s1, Figure S1: LC-MS TIC (total ion chromatograms) of fruit extracts of tomato, eggplant and pepper; Supplemental Table S1: Metabolite table of peaks detected in this study using LC-MS.

**Author Contributions:** Conceptualization, M.W. and T.T.; software, C.L.F.C. and T.T; validation, C.L.F.C., T.S. and T.T.; formal analysis, C.L.F.C.; project administration, M.W. and T.T.; supervision, M.W. and T.T.; investigation, C.L.F.C.; writing—original draft preparation, C.L.F.C. and T.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Japan Society for the Promotion of Science (JSPS) Scientific Research B (19H03249) and C (19K06723).

**Acknowledgments:** C.L.F.C.: T.S., M.W. and T.T gratefully acknowledge partial support by NAIST. We are grateful Jay Camisora Delfin who helped for growing tomato plants. Research activity of M.W and T.T. were additionally supported by JSPS KAKENHI Grant-in-Aid for Scientific Research B (19H03249) and C (19K06723).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Metabolite Changes during Postharvest Storage: E**ff**ects on Fruit Quality Traits**

#### **Delphine M. Pott, José G. Vallarino \*,**† **and Sonia Osorio \***

Departamento de Biología Molecular y Bioquímica, Instituto de Hortofruticultura Subtropical y Mediterránea "La Mayora", Universidad de Málaga-Consejo Superior de Investigaciones Científicas, Campus de Teatinos, 29071 Málaga, Spain; dpott@uma.es

**\*** Correspondence: vallarino@uma.es (J.G.V.); sosorio@uma.es (S.O.); Tel.: +34-952134271 (J.G.V. & S.O.)

† Current address: Max-Planck-Institut für Molekulare Pflanzenphysiologie, 14476 Potsdam-Golm, Germany.

Received: 14 April 2020; Accepted: 6 May 2020; Published: 8 May 2020

**Abstract:** Metabolic changes occurring in ripe or senescent fruits during postharvest storage lead to a general deterioration in quality attributes, including decreased flavor and 'off-aroma' compound generation. As a consequence, measures to reduce economic losses have to be taken by the fruit industry and have mostly consisted of storage at cold temperatures and the use of controlled atmospheres or ripening inhibitors. However, the biochemical pathways and molecular mechanisms underlying fruit senescence in commercial storage conditions are still poorly understood. In this sense, metabolomic platforms, enabling the profiling of key metabolites responsible for organoleptic and health-promoting traits, such as volatiles, sugars, acids, polyphenols and carotenoids, can be a powerful tool for further understanding the biochemical basis of postharvest physiology and have the potential to play a critical role in the identification of the pathways affected by fruit senescence. Here, we provide an overview of the metabolic changes during postharvest storage, with special attention to key metabolites related to fruit quality. The potential use of metabolomic approaches to yield metabolic markers useful for chemical phenotyping or even storage and marketing decisions is highlighted.

**Keywords:** fruit; postharvest; metabolomics; quality traits; stress; biomarkers

#### **1. Introduction**

Fruit growth, ripening and senescence are complex processes, controlled by multiple developmental and environmental signals, and their molecular mechanisms remain unclear [1]. Fruits undergo important metabolic changes during ripening, including chlorophyll breakdown, anthocyanin or carotenoid pigment accumulation, cell wall degradation and the synthesis of low-weight metabolites (such as sugars, acids and volatiles), which function to increase their attractiveness to seed dispersers [2]. Once fruits are removed from the plant and until they reach consumers on the market, a period known as postharvest ripening or senescence occurs—the duration of which is variable (from days to weeks) and the effects of which mainly depend on fruit metabolism and ripening status at harvest. Indeed, climacteric fruits, such as tomatoes, kiwi or avocados, which exhibit a concomitant peak of ethylene production and a sudden rise in respiration at the onset of ripening [3], can ripen after harvest. In this sense, the control of ethylene production is fundamental to optimize the postharvest storage of these types of fruits [4]. On the other hand, non-climacteric fruits, e.g., strawberries and grapes, do not exhibit respiration and ethylene production peaks, and have to be harvested (almost) fully ripe. Postharvest storage initiates fruit senescence—the effects of which on biological processes are unavoidable and largely negative. Senescence leads to protein, lipid and nucleic acid degradation and cell dysfunction, disintegration and death [5]. Several factors influence and accelerate fruit senescence, with the most relevant being respiration, providing energy for maintaining metabolism, dehydration

and fungal activity [6]. Consequently, the degradative processes associated with postharvest senescence impact fruit quality traits, i.e., aspect, texture, taste, aroma and nutritional characteristics, leading to consumer rejection and important economic losses for the fruit industry.

Currently, depending on fruit crops, different postharvest strategies are commercially practiced in order to adapt ripening to industry needs, delay senescence, maintain fruit quality attributes and, thus, prolong shelf-life. In general, fruits are highly perishable at ambient temperature. Thus, refrigerated storage is the most common method used to delay ripening, fruit respiration, enzymatic activities, and the development of pathogen infections, and, therefore, extend fruit shelf-life [7]. However, cold storage can provoke the development of a physiological disorder called chilling injury (CI). Although CI symptoms are species dependent, CI includes internal and external browning, mealiness, flesh bleeding, pitting or the inability to soften. These physiological disorders tend to appear once the fruits are acquired by consumers, having a negative impact on palatability and acceptance [8,9]. To reduce CI symptoms and depending on the type of fruit, the industry combines low-temperature storage with some complementary strategies. For example, prior heat treatment to cold storage is widely used in several crops, including *Citrus* and loquat [10], while controlled atmosphere (CA, increased CO2 and decreased O2 levels) is commonly applied to apple, strawberry, peach and pear, among others [11]. Additionally, delayed cooling has been successfully applied in apple to reduce soft scald, a chilling-dependent physiological disorder [12]. While CA reduces fruit respiration, heat treatment has a protective effect by acting on membrane integrity and heat shock protein accumulation and by promoting antioxidant and sugar metabolism [13]. Further, it is known that heat treatment induces defense mechanisms and induces physiological changes that allow *Citrus* fruit to withstand stressful conditions during storage. For example, GC–MS analysis in heat-treated oranges during storage showed a higher concentration of sugars while no changes were observed in organic acid levels [14].

In the case of climacteric fruits, such as tomato or banana, the application of ethylene antagonist 1-methylcyclopropene (1-MCP) is commonly used to increase shelf-life [15]. However, the aforementioned strategies have different degrees of effectiveness at reducing CI and prolonging shelf-life, depending on fruit species and varieties. In addition, it must be highlighted that these postharvest techniques constitute abiotic stresses for the fruits, which have to adapt their metabolism to maintain homeostasis [16]. In particular, stress situations induce the synthesis of compounds involved in plant protection, and trigger the accumulation of compatible metabolites, reactive oxygen species (ROS)-scavenging enzymes and changes in carbon metabolism [17,18]. In this sense, metabolomic platforms, allowing the simultaneous detection and quantification of hundreds of metabolites, offer the possibility to improve our knowledge about the molecular mechanisms underlying fruit senescence under commercial storage conditions.

#### **2. Metabolomic Platforms in Postharvest Studies**

The plant metabolome comprises a wide range of small molecules, with a large variety in physico-chemical properties and extremely variable concentrations. Metabolomics is defined as the field of the research that generates a profile of small molecules in a biological system. Thus, it can directly reflect the outcome of complex networks of biochemical reactions and, therefore, provides essential information about the underlying biological status on the system in question.

For these reasons, successful analysis of the complex network of fruit metabolites requires highly sensitive and selective analytical techniques, with each displaying both advantages and limitations and showing differential coverage depending on the nature of the metabolite. In particular, mass spectrometry (MS) coupled with gas chromatography (GC), liquid chromatography (LC) and, to a lesser extent, capillary chromatography (CE) and nuclear magnetic resonance spectroscopy (NMR) have been the most extensively applied methodologies to study the plant metabolome, including its reconfiguration during fruit postharvest senescence [17,19,20].

GC–MS is the technique of choice for measuring small polar metabolites, which are thermally stable and can be made volatile through a derivatization approach [21]. The main advantages of GC–MS are its robustness and reproducibility, which have allowed the establishment of libraries and databases facilitating the identification of metabolites. As a result of its characteristics, GC–MS is mainly used in plant metabolomic studies to investigate central primary metabolism, which includes sugars, sugar alcohols, amino acids, organic acids and polyamines [17,19]. In addition, GC–MS can be coupled with headspace solid-phase microextraction (HS-SPME), which allows the detection of specific volatiles present in a sample [22]. Both volatile and primary metabolite changes occurring during fruit postharvest storage have been extensively studied, as they are key compounds of fruit taste and aroma.

To overcome the limitations of GC–MS, which is restricted to volatile and thermally stable molecules, LC–MS is broadly used to detect a wider range of metabolites. In particular, the enormous diversity of plant secondary metabolites, which includes tens of thousands of different compounds [23], is mainly studied using LC–MS due to its versatility. However, and as a consequence of LC–MS flexibility, metabolite identification remains difficult, as no universal mass spectral library has been created [24].

Another technique used to study plant metabolomics, although rather uncommon, is capillary electrophoresis (CE)–MS. This technique allows the detection of a wide range of highly polar or charged metabolites by separating them based on their mass-to-charge ratio [25]. In this sense, this method has been proposed as a valuable complementary approach for samples that cannot be readily resolved by the more established GC– and LC–MS platforms [26].

Even if NMR presents a low sensitivity compared to that of MS approaches, it provides a series of advantages over the previously mentioned approaches by providing structural information, involving non-destructive sample preparation and providing rapid metabolite screening [19,27]. Integrated NMR platforms, allowing the monitoring of changes in both primary and secondary metabolites, have been developed and can be useful to study metabolic shifts in senescent fruits during postharvest [28].

#### **3. Primary Metabolic Pathways A**ff**ected by Postharvest Storage: E**ff**ects on Fruit Texture and Taste**

Fruit organoleptic quality is a complex trait that is influenced by taste, aroma, color and texture. In particular, fruit acceptance by consumers is directly influenced by sugar and acid content and the ratios of both groups of primary metabolites [29,30]. Fruit respiration during postharvest storage directly affects primary metabolic pathways, such as glycolysis, starch metabolism, and the tricarboxylic acid cycle (TCA), which account for changes in sugar, amino and organic acid levels. Indeed, carbohydrates, organic acids, proteins and fats are the main respiratory substrates during fruit storage. Furthermore, they are involved in gluconeogenesis, a process which has been described to be upregulated during the postharvest in orange and apple fruits [31–33]. Thus, it contributes to fruit depletion and also to important changes in primary metabolite composition. In the next paragraphs, we described alterations in sugars, organic and amino acids as a consequence of postharvest storage.

#### *3.1. Sugars and Sugar Derivatives*

Sugar content, which is commonly estimated by the soluble solid content (SSC), shows differential behavior during postharvest, depending mainly on the species and storage conditions. The SSC trend normally coincides with changes in the main sugar profiles present in ripe fruits, i.e., glucose, fructose and sucrose.

While main carbohydrates tend to decrease in some species, as profiled by GC–MS and NMR analysis in tomatoes kept at room temperature [34] or blackberries stored at 4 ◦C [35], in other fruits, such as bananas and kiwis, their level increases as a consequence of starch hydrolysis, which takes place during postharvest storage [36,37]. In turn, sucrose can be hydrolyzed, leading to a concomitant increase in hexoses, as monitored by GC–MS in Powell oranges stored at room temperature [33]. Interestingly, growing evidence seems to point to sugars playing a regulatory role in senescence

processes [38,39]. During fruit ripening and senescence, cross-talk between sugars and hormones involved in ripening and senescence processes, such as abscisic acid, ethylene and auxin, has been described [40–43], and sucrose degradation during postharvest storage can be crucial for inducing senescence [33,44]. Furthermore, sugar uptake during fruit ripening may affect postharvest water loss by interfering with cuticle development. Indeed, stable silencing of the cell wall invertase *LIN5*, a key determinant of SSC content, led to a diminished water loss rate and wrinkling in transgenic tomato fruits kept at room temperature for 12 days. Even though the complete molecular mechanism has not been described, it was clearly established that sugar entry during fruit development impacts the cell wall and cuticle structure, resulting in a radical effect on tomato senescence [45].

Apart from the most abundant sugars, i.e., sucrose, glucose and fructose, fruits also contain minor sugar and alcohol derivatives, such as sorbitol, galactinol, raffinose, *myo*-inositol and trehalose [46]. Even if those compounds may be at low concentrations, they seem to be crucial for fruit behavior during storage, as they can alleviate the negative effects of the abiotic stresses underlying postharvest conditions. Indeed, soluble sugars are important metabolites in ROS metabolism, being the primary carbon and energy source and contributing to the generation of reducing power generation via the oxidative pentose phosphate pathway [41,47,48]. Furthermore, they play key roles in osmoprotection and cell membrane stabilization [49–51]. As an example, important increases in raffinose and galactinol levels were measured by GC–MS in peaches after heat treatment (three days at 39 ◦C) followed by storage at 0 ◦C for two days and may confer improved tolerance to CI [52]. Moreover, comparing the levels of galactinol (detected by LC–MS/MS), raffinose, trehalose and *myo*-inositol (identified by NMR) in climacteric and non-climacteric plum varieties during postharvest storage at 20 ◦C and in presence of 1-MCP, propylene (ethylene analogue) or control air, Farcuh et al. [46] noticed that the levels were more enhanced in the latter variety. These data could explain the capacity of the non-climacteric variety to cope better with postharvest stress conditions, and the identified sugars could be used as biomarkers to evaluate fruit physiological status during storage (Table 1).

Softening during postharvest storage is a key physiological process leading to ripe fruit firmness; however, excessive loss of firmness as a consequence of overripening can prompt physical damage and pathogen attack, and consequently lead to an important decrease in fruit quality. Softening is the result of several factors, including cell wall disassembling metabolism. Metabolites originated from cell wall disassembly, mainly monosaccharides, can be monitored by primary metabolite profiling. Indeed, in pitaya fruit, the content of several monosaccharides, including xylose, galactose, arabinose, and mannonic acid and glucuronic acid, which originate from cell wall disassembly, was measured by GC–MS [53]. Interestingly, these metabolites were decreased after blue light treatment (2 h at 25 ◦C under blue light emitting diode) compared to control fruits kept in the dark, suggesting that this treatment has a significant effect in delaying cell wall degradation and postharvest decay of pitaya fruit [53]. Another study, using LC coupled with tandem MS (LC–MS/MS), detected an increase in glucuronic acid, a component of pectin, among the major elements of plant cell wall, in pears stored 18 days at room temperature [54]. Pectin de-polymerization and de-esterification were also evidenced by the detection of galacturonic acid by two-dimensional GC–MS (GC x GC–MS) in overripe kiwi fruits stored at 20 ◦C and treated for 24 h with 200 ppm ethylene. On the contrary, no increase in galactose was observed using GC–MS measurements, suggesting that this sugar is directly metabolized after its release from cell wall, or that it is liberated as different form [55]. Among the main symptoms of CI in peach is mealiness, which is the result of a cell wall metabolism disorder. Xylose, the central constituent of hemicellulose, among the key components of plant cell wall, was increased during cold storage in peach chilling-susceptible genotypes, but not in the varieties resistant to CI, confirming a link between cell wall disassembly and mealiness in sensitive cultivars [56].



*Metabolites* **2020**, *10*, 187

#### *3.2. Organic Acids*

Several organic acids are related to fruit postharvest metabolism. Surprisingly, particularly in tomato, the levels of malate, among the most abundant organic acids, impact fruit shelf-life (Table 1). The malate content, measured by GC–MS, decreases during the ripening and postharvest storage at room temperature of several tomato genotypes, including *delayed fruit deterioration*, *non-ripening* and *ripening inhibitor* mutants as well as genotypes that are commercially used because of their delayed maturation and senescence. Interestingly, it was also shown that malate levels were lower in mature tomato fruits that ripened on the vine than off the vine [34]. However, when the malate concentration in tomato fruits was manipulated by reducing the expression of two TCA cycle enzymes (*fumarase* and *malate dehydrogenase* (*MDH*)), a differential postharvest behavior was observed compared to that in wild-type fruits at room temperature. Interestingly, fruits of the MDH-deficient genotype showed higher malate content and poorer postharvest behavior than non-transformed fruits by losing more water and being more susceptible to opportunistic fungal infections and *Botrytis cinerea* spores. In contrast, the fumarase-deficient genotype, with a relatively low malate content, presented a decrease in water loss and in susceptibility to opportunistic fungi [57]. The mechanism underlying malate's role in postharvest responses could not be clearly explained; however, the authors suggested a role for SSC, which changes in the opposite manner in *MDH*- and *fumarase*-silenced lines, in osmotic potential and subsequent water loss during storage. Another study using recombinant inbred lines originating from the cultivated tomato *Solanum lycopersicum* and the wild-type species *Solanum pimpinellifolium* also pointed out the association between malate content, fruit firmness and shelf-life [58]. A comprehensive polar metabolite profiling was performed by GC–MS and NMR and a combination of neuronal clustering and network construction displayed a strong correlation between glycerate and malate content and postharvest, which also showed a negative correlation with fructose levels [58]. This association between metabolites and agronomic traits such as firmness and storage behavior suggested that malate could be a good biomarker to select genotypes with enhanced quality traits, such as improved postharvest life [58].

By performing a GC–MS metabolic characterization of *S. lycopersicum* cv. 'Plaisance' fruits during ripening and postharvest stages, Oms-Oliu et al. [59] showed that one sugar (mannose) and three organic acids (citramalate, gluconate and keto-gulonate) were strongly increased once the fruit was removed from the vine and that these compounds could be indicators of metabolic shifts during postharvest storage [59] (Table 1). As an example, the enhanced gluconate levels could be a consequence of tartarate biosynthesis from ascorbate degradation or energy balance changes during tomato storage [62,63]. Free mannose levels are generally low, as this monomer usually composes carbohydrate polymers. However, it can be found in a free form as a result of cell wall disassembly and hemicellulose breakdown during fruit senescence, as described in tomato, apple and pear [59,62,64].

Organic acids, particularly citric acid, accumulate at high levels in *Citrus* fruits, such as lemons, oranges, grapefruits or pummelos. A study on 'Hirado Buntan' pummelo focused on the relationship between organic acids, measured by high-performance capillary electrophoresis, and fruit senescence during postharvest storage at both ambient and cold temperatures. The authors observed a general decrease in malate, citrate, aconitate and fumarate during storage, accompanied by important fluctuations in their levels; this decrease was associated with a loss of fruit quality [31]. The combination of transcriptomic analysis paralleled the metabolomic data, suggesting that the *peroxisomal MDH*—the expression of which correlated with malate levels—is responsible for organic acid metabolism regulation during postharvest. This result indicated that the glyoxylate cycle, which occurs in peroxisomes and glyoxysomes, is central to organic acid regulation by supplying succinate for the TCA cycle [31]. Tang et al. [33] also observed a decrease in several organic acids analyzed by GC–MS, such as malate, citrate and α-ketoglutarate, during postharvest storage of 'Powell' oranges at room temperature. In this case, they suggested that malate could be used as a substrate for gluconeogenesis, being converted into phosphoenolpyruvate (PEP) by the action of two enzymes upregulated in oranges kept at room temperature: PEP carboxykinase and pyruvate orthophosphate dikinase (PPDK). Similarly, an increased abundance of PPDK proteins associated with decreased malate content was observed in peaches subjected to heat treatment followed by storage at 20 ◦C [65]. Another study using GC–MS analysis in different varieties from the *Citrus* genus suggested that a conversion of organic acids to sugars during fruit postharvest senescence at ambient temperature occurs, as negative correlations were frequently observed between metabolites belonging to the two groups and that the SSC/titratable acidity ratio increased during storage [66]. The succinate content increased during pummelo postharvest storage, showing a positive correlation with GABA and glutamine [31]. In addition, GABA increased during the postharvest senescence of Powell oranges, matched by an upregulation of the genes involved in the GABA shunt [33]. In this sense, the GABA shunt was outlined as an important pathway for organic acid catabolism and for balancing organic acid and amino acid levels. Indeed, superfluous citrate can be converted into amino acids via the GABA shunt [33,67]. Moreover, Sun et al. [31] observed an increase in ROS during pummelo storage, which correlated with enhanced mitochondrial damage. Cross-talk between ROS and organic acids could occur during postharvest senescence, as TCA enzymes have been described to be very sensitive to inhibition by ROS [68,69], while organic acids could be involved in the direct ROS scavenging [70,71].

#### *3.3. Amino Acids*

Amino acid content is also affected to a large extent by postharvest storage, as these compounds are involved in several pathways induced during fruit ripening [47]. In particular, during senescence, amino acid catabolism can counteract the reduction in electron supply from the TCA cycle [72]. Ubiquitination of proteins controls their degradation to free amino acids, and upregulation of the ubiquitin pathway has been reported in stored peaches that were previously were heat treated [73]. Dopamine, a derivative of the aromatic amino acid tyrosine, has been proposed as a postharvest marker in banana fruit stored at 25 ◦C [36] (Table 1). Indeed, NMR-based metabolite profiling of the senescence of bananas stored at room temperature showed that dopamine levels were undetectable at the last postharvest stage. Concomitant with dopamine disappearance was the sudden appearance of salsolinol, which has been described to originate from dopamine and acetaldehyde, the latter formed from ethanol, which is also generated in the late postharvest stage [36,74]. The authors concluded that the conversion of dopamine to salsolinol led to a decrease in fruit quality, making bananas less fit for consumption [36].

Additionally, several amino acids play a key role in tolerance to abiotic stresses in fruits during postharvest senescence. Indeed, a GC–MS comparative study between pineapple varieties tolerant and susceptible to CI stored at 10 ◦C outlined that amino acid increases during chilling stress may be associated with a delay in symptom appearance, such as internal browning, by presumably contributing to the synthesis of enzymes involved in tissue repair and, in the case of cysteine, aspartate and valine, by acting as osmoprotectants [75]. Proline is a well-documented stress-related amino acid and among the main osmolytes that are accumulated during plant stresses, playing important membrane protection and ROS-scavenging functions [76,77] (Table 1). Grape storage in a CO2-enriched atmosphere resulted in a threefold endogenous proline increase when compared to that in air-stored grapes [60], and proline accumulation is a common trend in postharvest fruits subjected to treatments to attenuate CI, such as zucchinis [61], mangoes [78], bananas [79], pears [80] and loquats [81]. However, the possible role of amino acids in counteracting CI seems to be species dependent, as GABA, aspartate, phenylalanine and proline increase in peach stored at 0 ◦C for 21 days was not associated with CI protection, since their levels, quantified by GC–MS, were enhanced in both resistant and susceptible genotypes [56].

A recent study in strawberry also outlined the possible role of amino acids in plant defense, as pathogen resistance mechanisms implicated this group of metabolites [82]. The increase in asparagine, aspartic acid, threonine, glutamic acid, glutamine, alanine and glycine in CO2-treated strawberries compared to control fruits could, at least partially, explain the lower fungal decay observed in the first group [83].

#### **4. Postharvest Impact on Secondary Metabolites**

The two main families of secondary metabolites present in fruits are polyphenol and terpenoid compounds, responsible for their appealing color and also important for their organoleptic and nutritional characteristics [84]. Apart from their importance in the human diet, these molecules are involved in plant defense and responses against biotic and abiotic stresses. In particular, metabolomic approaches have helped in deciphering their role during fruit storage and how different postharvest strategies impact on them. Here, impact on polyphenols, including anthocyanins, and carotenoids during fruit shelf-life is discussed in the next paragraphs.

#### *4.1. Polyphenol Compounds*

Dynamic metabolite changes, profiled by high-performance LC–MS (HPLC–MS), were observed during grape postharvest ripening and dehydration, the metabolic responses being genotype dependent [85]. A particular feature was the cultivar-specific accumulation of stilbenes, a class of phenylpropanoid compounds, with antifungal activity. On the other hand, anthocyanins and other flavonoids, belonging to another phenylpropanoid class, were depleted along postharvest dehydration [85]. An untargeted HPLC–MS profiling during grape ripening and withering (postharvest drying), combined with transcriptomic and proteomic data integration, also correlated the presence of stress-related secondary compounds (stilbenes and acylated anthocyanins) with the postharvest phase. The synthesis of defense molecules could be a response to abiotic stress (dehydration) or biotic stress (eventual pathogen attack). In addition, three metabolites (two taxifolins and tetrahydroxyflavanone-O-deoxyhexoside), belonging to the flavonoid class, have been proposed as putative markers in order to assess berry fruit quality traits (Table 2) [86]. In grapes, the accumulation of different stilbenes during cold postharvest storage was monitored by UHPLC–MS/MS [87]. This increase was also observed when grape fruits were kept at high CO2 [87]. In contrast, CA storage has been described to have negative effects on anthocyanin accumulation in strawberry fruits, compounds responsible for the color of the ripe fruit [88]. In this sense, postharvest cold storage is a mandatory strategy to enhance anthocyanin content in some fruits such as blood oranges, some varieties of plums and anthocyanin-rich tomatoes [89,90]. Interestingly, it has been described that tomato anthocyanin-rich lines are able to maintain fruit quality for longer during storage, mainly by reducing their susceptibility to *Botrytis cinerea* [91,92].

In mandarins, heat treatment previous to storage at 12–16 ◦C positively impacts polyphenol metabolism by increasing flavonoids and lignin content (flavonoids measured by HPLC–MS). The effect of this postharvest strategy can be seen as a modulation of fruit defense against biotic and abiotic stress during postharvest storage, by supplying chemical (flavonoids) and physical barriers against pathogen attack [93]. The relationship between polyphenol content and resistance to postharvest decay caused by *Penicillium expansum* has also been described in apple; indeed, resistant and susceptible apple genotypes could be discriminated based on polyphenol content, measured by UPLC–MS (Table 2) [94]. However, a general polyphenol increase during fruit shelf-life does not always occur, as described by untargeted UHPLC–MS in several mango varieties stored at room temperature during six days, in which it was found that only gallic acid and epicatechin content was enhanced after storage [95].


β-cryptoxanthin

pool in mature oranges

12 ◦C up to 7 w

 Increase

 sweet orange

 HPLC

 [96]

**Table 2.**

Secondary metabolites

(polyphenols

 and carotenoids)

 identified as putative biomarkers

 by

metabolomic

 profiling studies to assess fruit quality changes

#### *4.2. Carotenoids*

Carotenoids are an important class of terpenoids, responsible for the attractive color of many fruits and vegetables. While their low stability during postharvest storage, mainly due to a rapid turnover of β-carotene, has been described in many staple crops, postharvest accumulation in *Citrus* and tomato seems to be temperature dependent [96,97]. Carotenoid levels in grapefruit, determined by HPLC, stored at 2 and 12 ◦C established a link between carotenoid content and CI symptom suppression, suggesting that they play a role in preventing cold damage by protecting plastid structures [98,99]. Furthermore, the ratio between 9-Z-violaxanthin (yellow hues) and β-citraurin (orange-red pigments), responsible for the external orange fruit color, was lower in sweet oranges stored at 12 ◦C than at 2 ◦C, outlining that this important quality indicator is better maintained at moderate temperatures. Additionally, increased levels of β-cryptoxanthin in orange pulp stored at 12 ◦C should be pointed out, due to health-beneficial provitamin A activity (Table 2) [96]. Carotenoid content, measured by HPLC, was also drastically increased during postharvest storage of winter squash at 21 ◦C, even if no induction of the biosynthetic genes could be observed. Starch degradation during winter squash storage, with the concomitant release of soluble sugars which may act as substrates for terpenoid synthesis, and downregulation of genes involved in carotenoid turnover, could be the explanation of their enhanced content [100]. In other fruits, such as green pepper, carotenoid accumulation during postharvest storage has a negative impact on consumer acceptance. Pepper reddening depends on the metabolic dynamic of chlorophyll degradation and active synthesis of carotenoids, such as β-carotene and capsanthin, as depicted by HPLC-based profiling of these pigments [101]. Quantification of chlorophyll by spectrophotometry has also pointed out its breakdown as a deterioration factor occurring during pear or lime shelf-life [102,103]. In this sense, postharvest strategies, such as chlorine dioxide fumigation or hot water treatment, may be effective in downregulating genes involved in chlorophyll-degrading enzymes [101,102].

#### **5. Volatile Profiles during Postharvest and Their Impact on Fruit Aroma**

In fruits, there are three major classes of metabolites responsible for flavor: sugars, acids, and volatile. While fruit taste is mostly dependent on the ratio of sugars and acids, it is the volatiles that determinate the unique flavor of fruits. Most volatiles present in mature fruits originate from terpenoid and phenylpropanoid pathways or are fatty and amino acid derivatives [104]. Volatile profiling is typically achieved by extracting them from the headspace (HS), i.e., the airspace around the fruit, and detecting them by GC–MS. Sampling from headspace is most often performed by the adsorption of the volatiles on a stationary phase coated on a fused silica fiber and is known as solid-phase microextraction (SPME) [104]. Another GC–MS-based strategy for volatile profiling is their collection from chopped fruits on a Super Q column, followed by elution with methylene chloride [105]. To overcome metabolite co-elution by one-dimensional GC, GC x GC–MS has been implemented to increase separation efficiency and volatile detection [106,107]. As not all volatiles impact fruit aroma, a complementary approach, known as GC—olfactometry, can be used to determine odor-active compounds [108]. During postharvest, it could be established that important shifts in fruit volatile profiles are normally observed and are often responsible for the decreased sensory acceptability after prolonged storage. For instance, general trends profiled by GC–olfactometry, describe a loss of 'green' or 'fresh' notes and a concomitant increase in 'fruity', 'overripe' or 'musty' aromas [109]. Changes in aroma are a consequence of metabolic pathways that are active during postharvest and, in turn, appear to be largely depend on the storage strategies used by industry. For example, among the symptoms related to CI is the negative impact perceived on aroma production, a phenomenon described in many species, such as strawberries [110], kiwifruit [111], tomatoes [112] and peaches [113]. Tomatoes stored at 5 ◦C for 7 days were significantly less palatable than fruits recently harvested, and this decrease in consumer acceptance, established by taste panels, was a consequence of changes in volatile emissions [105]. Furthermore, a higher increase in 'musty' and 'damp' aroma notes was observed in tomatoes stored

at 10 ◦C than in those stored at 12.5 ◦C, suggesting that the latter temperature storage was able to maintain better sensory attributes [114].

Fermentation metabolism and amino acid and fatty acid catabolism are of great importance regarding the production and accumulation of volatiles in harvested fruits. Indeed, the activation of amino acid and fatty acid degradation to generate TCA cycle acetyl-CoA precursors and thus maintain energy production leads to the accumulation of specific substrates for volatile formation. In mandarin, a combination of metabolomic and transcriptomic data outlined the upregulation of genes involved in branched-chain amino acid catabolism, fatty acid cleavage and ethanol fermentation, which suggested that central metabolism modifications are accountable for the increase in branched-chain esters ('fruity', 'overripe' aroma), fatty acid-derived volatiles ('musty' notes) and ethanol [115]. The activation of anaerobic fermentative metabolism due to postharvest abiotic stress is especially important in 'off-aroma' compound generation and has been described in fruits of several species, including strawberries [116–119], apples [120], mandarins [109] and peaches [121], among others. Indeed, the glycolysis end-product pyruvate can alternatively serve as a substrate for anaerobic respiration and ATP production under O2-limiting conditions, which produces a shift from aerobic respiration to the fermentation pathway [16,120,122,123]. As a consequence, off-aroma volatiles, namely ethanol, acetaldehyde and ethyl acetate, accumulate, playing a key role in fruit quality decline [117,124] (Table 3).


*Metabolites* **2020**, *10*, 187

Fermentative metabolism activation and off-flavor compound formation are mainly associated with low oxygen concentration under CA storage [118,123,124,131]. However, the production of ethanol via fermentation may also be a consequence of a decline in cellular energy status [132]. As long as energy demand is maintained, fermentation can be endured; nevertheless, failure of cellular homeostasis, such as an imbalance in the pH or ROS production, will lead to storage-induced disorders, strongly affecting fruit quality [133]. Understanding how or when fermentation occurs can help to limit ethanol production. Metabolomic approaches using 1H NMR and GC–MS profiling were used to assess metabolite gradients within the fruit, which may be related to in situ hypoxia in the central part of the ripening fruit [134,135]. CA is of special importance for the long-term storage of fruits such as apples, and could maintain a better aroma quality [136,137]. It appears that the low-oxygen pressure employed during CA affects volatile emissions in a genotype-dependent manner [11]. Indeed, a multiplatform metabolomic approach (proton-NMR, GC–MS and HS-SPME–GC–MS) comparing 'Red Delicious' and 'Granny Smith' apple varieties showed strong activation of fermentative metabolism in the former, with ethanol and acetaldehyde accumulation, while the latter dealt with hypoxia by a reconfiguration of nitrogen metabolism through the intensification of alanine levels to prevent excessive accumulation of pyruvate [11]. Low oxygen may induce changes in metabolite concentrations that reflect a decrease in biosynthetic process, inhibition of the TCA cycle, and activation of anaerobic metabolism, which means accumulation of sucrose and organic acids and diversion of pyruvate to ethanol and alanine [134]. Table grapes stored under elevated CO2 concentrations (5 kPa O2 and 15 kPa CO2) showed an upregulation of genes involved in pyruvate synthesis (*pyruvate kinase*, *PEP carboxykinase* and *NADP-dependent malic acid enzyme*) and a concomitant increase in volatiles, detected by HS-SPME–GC–MS, derived from pyruvate degradation—some of which were suspected to generate 'off flavor'. Additionally, the increased expression of a specific *alcohol dehydrogenase* gene (*ADH*) under anaerobic atmospheric conditions enhanced the accumulation of off-aroma volatiles, including ethanol, acetaldehyde and ethyl acetate [87].

Metabolic reconfiguration during postharvest affects volatile patterns beyond the generation of off-aroma compounds, and changes occurring in most important volatile classes are described in the next sections.

#### *5.1. Fatty and Amino Acid-Derived Volatiles*

Fatty acid-derived volatiles, responsible for aldehyde, alcohol and ester accumulation, the last being the predominant class of aromatic compounds in fruits of several species, seem to be strongly impacted by low-temperature storage [110,138,139]. Free fatty acids such as linoleic acid and linolenic acid are reduced to aldehydes by the lipoxygenase pathway (LOX). Next, aldehydes are reduced to alcohols followed by alcohols to esters by ADH and alcohol acyltransferase (AAT), respectively (for a review, see [84]). Interestingly, correlations among LOX, ADH and AAT activities, gene expression and decreased volatile production under refrigerated postharvest conditions have been established in several fruit-bearing species [140,141]. In particular, a relationship between a reduction in ADH activity and decreased ester content, monitored by SPME–GC–MS technology, during pear cold storage has been established [138]. In tomatoes, ADH activity was diminished as a consequence of refrigerated conditions at both 10 and 12.5 ◦C, and storage was associated with an increase in the aldehyde/alcohol ratio at 10 ◦C [112,114]. Furthermore, a decrease in *ADH2*, *LoxC* and *AAT1* transcripts after 8 days of cold storage (5 ◦C) was associated with lower levels of C6 and C5 (fatty acid-derived) volatiles in chilled tomatoes [105]. Additionally, low temperature also seems to affect upstream lipid catabolism by downregulating the expression of several genes involved in the formation of the unsaturated free fatty acids linoleic acid and linolenic acid, limiting substrate availability for ester biosynthesis [142]. Furthermore, membrane damage during cold storage has also been suggested to impair ester synthesis, as a relatively high leakage rate, a commonly used marker for membrane permeability, was measured in pears stored for a long time under refrigerated conditions [138]. The impact of cold storage on the aromatic compound profile reaches further than that on the pattern of lipid-derived volatiles. Branched-chain volatiles derived from the direct precursors of branched-chain amino acids, measured by GC, after methylene chloride extraction, were also shown to decrease during tomato cold storage and are correlated with a lower expression of two branched-chain aminotransferases (*BCAT1* and *BCAT7*) involved in the first step of the catabolism of these amino acids [105]. Additional treatments, such as hot air or UV-C, combined with cold storage could counteract the negative effect on ester biosynthesis by promoting the *LOX* pathway, as has been demonstrated in peaches [143]. Similarly, a pre-chilling heat treatment (52 ◦C, 5 min) has been shown to alleviate the depletion of important volatiles for tomato aroma quality during its postharvest storage; in this case, the volatiles include amino acidand carotenoid-derived compounds profiled by HS-SPME–GC–MS [144]. Fatty acid-derived alcohols were also higher under elevated CO2 concentrations compared to those of recently harvested grapes and cold-stored berries under atmospheric conditions due to the upregulation of the *LOX* pathway, together with *ADH* [87].

Low-oxygen storage has a broader impact on volatile content than ethanol and off-aroma compound generation, as demonstrated by the different content of ethyl esters between 'Granny Smith' and Red Delicious' apples [11]. Indeed, ethanol can serve as a substrate for ethyl esters, enhancing their synthesis [136,145,146] and competitively inhibiting the formation of esters originated from other alcohols [147]. As a consequence, an imbalance between the ratio of ethyl and the remaining esters occurs during postharvest storage of fruits of ethanol-accumulating apple varieties and those of many other fruit-bearing species, most likely affecting aroma perception. The fruits of 'Granny Smith' and 'Royal Gala' apple varieties did not seem to accumulate ethanol under low-oxygen-pressure storage; however, a negative effect on ester synthesis, in particular straight-chain esters, was observed, with the impact proportional to the decrease in O2 pressure [148–154]. This decrease can be explained by the fact that the LOX pathway requires the presence of oxygen. This effect has also been described in other apple varieties. [137,153]. In contrast, the concentration of branched-chain esters, monitored by HS-SPME–GC–MS, did not seem to be negatively affected by low oxygen, possibly because branched-chain amino acid levels were unaltered [154]. Furthermore, low oxygen suppresses the production of the hormone ethylene, which is involved in ester synthesis, as demonstrated during apple or banana storage in the presence of its antagonist 1-MCP [155–157].

#### *5.2. Terpenoid Volatiles*

Several studies in *Citrus* have highlighted important changes in terpenoid volatiles monitored by HS-SPME–GC–MS. These changes were related to CI and tolerance to cold storage. In mandarins, the accumulation of terpenoid volatiles is associated with chilling-sensitive fruits [158]. This increase is temperature dependent, and the authors suggest that it was responsible for decreased fruit palatability, as terpenes can contribute to an unpleasant aroma, providing 'musty', 'resinous' and 'oily' notes.

Another study on volatile emission by intact grapefruits stored at 12 and 2 ◦C for 7 weeks outlined important differences in the profiles of the terpenoid volatiles as a consequence of temperature [98]. Interestingly, grapefruits stored at 2 ◦C experienced a strong increase in monoterpene content, particularly in limonene and β-myrcene levels; this group of volatiles was strongly decreased in fruits stored at 12 ◦C at the beginning of the postharvest period, after which their content remained unchanged. In contrast, sesquiterpene emissions were predominant in the fruits stored at 12 ◦C [98]. While the accumulation of the monoterpene β-myrcene under 2 ◦C can negatively impact negatively consumer acceptance, by providing 'musty' and 'wet soil' aroma notes to the fruits [108], sesquiterpene ketone nootkatone levels, an important volatile in grapefruit aroma, seems to be promoted (in term of it content) under moderate–intermediate-temperature storage, but not refrigerated conditions [98,159] (Table 3). Taken together, these data suggest that the aroma quality of grapefruits could be better maintained during intermediate-temperature storage, as has been previously demonstrated in mandarins [126,149].

The trend in the accumulation of the monoterpene limonene and the sesquiterpene α-farnesene, which showed a transient increase after one week of storage under the two different temperatures, could be related to cold-induced responses. Limonene release, measured both by HS-SPME–GC–MS and

capillary gas–liquid chromatography–MS, has also been described in other *Citrus* species [125–127,160] and could be a consequence of cell wall and plasmatic membrane disruptions in the oil glands [99,125]. The degree of accumulation of α-farnesene is correlated with the susceptibility of different cold-sensitive *Citrus* species to CI development at 0 ◦C storage [128]. Interestingly, α-farnesene stopped being emitted by grapefruits after 3 weeks of storage at 12 ◦C, which coincided with the decrease in the observed CI symptoms, i.e., peel injury. At 2 ◦C, α-farnesene emissions were maintained during the whole postharvest period, concomitant with the CI symptom progression, confirming the relation between the detection of this volatile and CI manifestation. In this sense, the detection of α-farnesene by metabolome-driven approaches could be of high value as a potential biomarker to assess *Citrus* quality during postharvest (Table 3). Similarly, the monoterpene linalool, a key component of the aroma of papaya, was negatively affected by cold postharvest storage at 10 ◦C, and a concomitant downregulation of linalool synthase expression was also observed, suggesting that this volatile could be used as a marker to define papaya quality during postharvest storage [129] (Table 3). Linalool is also the predominant compound responsible for flavor in Muscat table grapes, a highly appreciated quality trait [161]. The postharvest storage of 'Shine Muscat' grapes at different temperatures between 0 and 10 ◦C showed that the decrease in linalool, profiled by HS-SPME–GC–MS, was enhanced at relatively low (0, 2 and 5 ◦C) temperatures in both fruit skin and flesh. Concomitant with linalool levels, grapes stored for four weeks at 10 ◦C presented a higher Muscat flavor than grapes stored at 0 ◦C for the same duration [130]. A possible effect of temperature on linalool synthesis or on the interconversion of free (aroma-producing) linalool and its glycosidically (odorless) bound form could be responsible for the observed differences in its concentration. In this sense, optimal storage at 10 ◦C for a short period or at relatively low temperatures followed by poststorage conditioning at 10 ◦C is fundamental for maintaining aroma quality for consumers [130] (Table 3).

The combination of 1-MCP with high O2 or high CO2 seemed to favor terpene content, as did the CO2-enriched atmosphere in lemon [162]. CA storage under an elevated CO2 atmosphere (8% CO2 and 2%–3% O2) could also promote terpene accumulation in mango. Indeed, fruit injury as a result of CA can enhance the activity of glycosidases, releasing monoterpenes, such as linalool or terpineol, from their glycoside-bound forms [163]. However, all tested CA treatments resulted in a reduction in total sesquiterpenes and also enhanced levels of ethanol, acetaldehyde and esters compared to those under atmospheric conditions. In addition, it was established that mangoes should not be stored under 3%–5% O2 to avoid excess fermentative compound accumulation and maintain fruit aroma quality [163,164].

#### **6. Conclusions**

Although fruit responses to postharvest storage conditions are species and even cultivar dependent, making them especially complicated to study, metabolomic approaches alone or combined with transcriptomic/proteomic analyses are highly useful for understanding how metabolic changes affect quality traits. In particular, reconfiguration of fruit metabolism as a consequence of the abiotic/biotic stress encountered during postharvest storage conditions (cold, hypoxia, pathogens, etc.) has a direct impact on the accumulation of taste- and aroma-producing metabolites, which are decisive attributes for consumers and thus for the fruit industry. Even though many molecular mechanisms active during fruit postharvest storage and senescence remain elusive, future omic studies will shed light on them to optimize fruit storage conditions.

Furthermore, the recent advances in metabolomic-driven technology allows the identification of valuable biomarkers that can be employed by the fruit industry to tightly monitor changes in quality attributes during postharvest storage. In this sense, the use of multiplatform approaches offers the possibility to select a set of metabolite markers, which could better depict the impact of postharvest storage on aroma, taste, appearance and nutritional value [165].

**Author Contributions:** All the authors did the literature searching, writing and original draft preparation. All authors have read and agreed to the submitted version of the manuscript.

**Funding:** The authors want to acknowledge funding through grants RTI2018-099797-B-100 (Ministerio de Ciencia, Innovación y Universidades, Spain) and UMA18-FEDERJA-179 (FEDER-Junta Andalucía). In addition, we acknowledge partial funding by the University of Malaga (Plan Propio) and the European Union's H2020 Programme (GoodBerry; grant number 679303). D.P. acknowledges the support by the Spanish Ministry of Science and Innovation (BES-2013–062856).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Network Analysis Provides Insight into Tomato Lipid Metabolism**

**Anastasiya Kuhalskaya 1,2, Micha Wijesingha Ahchige 1, Leonardo Perez de Souza 1, José Vallarino 1, Yariv Brotman 1,2 and Saleh Alseekh 1,3,\***


Received: 14 February 2020; Accepted: 11 April 2020; Published: 14 April 2020

**Abstract:** Metabolic correlation networks have been used in several instances to obtain a deeper insight into the complexity of plant metabolism as a whole. In tomato (*Solanum lycopersicum*), metabolites have a major influence on taste and overall fruit quality traits. Previously a broad spectrum of metabolic and phenotypic traits has been described using a *Solanum pennellii* introgression-lines (ILs) population. To obtain insights into tomato fruit metabolism, we performed metabolic network analysis from existing data, covering a wide range of metabolic traits, including lipophilic and volatile compounds, for the first time. We provide a comprehensive fruit correlation network and show how primary, secondary, lipophilic, and volatile compounds connect to each other and how the individual metabolic classes are linked to yield-related phenotypic traits. Results revealed a high connectivity within and between different classes of lipophilic compounds, as well as between lipophilic and secondary metabolites. We focused on lipid metabolism and generated a gene-expression network with lipophilic metabolites to identify new putative lipid-related genes. Metabolite–transcript correlation analysis revealed key putative genes involved in lipid biosynthesis pathways. The overall results will help to deepen our understanding of tomato metabolism and provide candidate genes for transgenic approaches toward improving nutritional qualities in tomato.

**Keywords:** lipophilic compounds; lipid-related genes; lipid metabolism

#### **1. Introduction**

Plants produce a wide variety of biochemical compounds, starting at the central or primary metabolism which generates compounds absolutely vital for plant survival and continuing through the pathways of specialized or secondary metabolism [1–5]. Specialized metabolites display a tremendous diversity, are often specific to certain plant lineages, and play many different roles in adaptation to the environment [6]. Volatile organic compounds (VOCs) are often discussed as a subgroup of secondary metabolites, with their low molecular weight enabling movement across cell membranes and release into the surrounding environment [7]. Secondary metabolites can act as direct or indirect defense agents by deterring herbivores, fending off pathogens, and/or attracting predators or pollinators [8]. When it comes to human consumption of plants, secondary metabolites also fulfill an important role, since they can have beneficial health effects and some of them contribute to flavor [9,10].

Since lipids may either be primary (e.g., glycerolipids, phospholipids) or secondary (derived from the isoprenoid pathway) metabolites, they are often discussed in the literature separately from other metabolites [11–13]. Lipids fulfill many different functions, ranging from carbon storage via cell-membrane components to signaling molecules [14].

The cultivated tomato (*Solanum lycopersicum*) has been widely used for metabolomic studies [15]. With production of over 150 million metric tonnes in 2017, tomato is the second most consumed vegetable in Europe after potato and the first by market value (www.fao.org/faostat/en/#home). The drought-tolerant green-fruited relative, *Solanum pennellii*, has been successfully hybridized with the cultivated tomato, and the offspring from that cross has been used to identify a wide range of phenotypic and metabolic quantitative trait loci (QTL) [16–18]. The importance of tomatoes for human diet, combined with the availability of genetic diversity from wild relative species, makes tomato an optimal model crop for studying different aspects of plant physiology [19]. In recent years, tomato metabolism has been intensively studied. Tomato's specialized metabolites in particular have received much attention, since many compounds are known to have positive health effects [20], serving among others as antioxidants. Many efforts have been undertaken to understand the genetic basis of their biosynthesis and to increase the production of these compounds in fruits [21–26].

However, due to its immense complexity and interconnection, it is a great challenge to understand plant metabolism as a whole. QTL mapping is a commonly used approach to dissect plant metabolism and identify genes in the corresponding pathways [27]. Although QTL mapping is useful for the elucidation of individual pathways, this approach is still not sufficient for dissecting metabolism in its entirety due to the latter's complexity [28]. Thus, metabolite correlation network analysis has been suggested as an additional method for elucidating novel connections in plant metabolism [29]. Several studies have used a network approach to display the correlation between metabolic compounds. Schauer et al. (2006) combined mQTL and network analysis to elucidate the relationship between metabolic and yield-related traits over two seasons of a tomato introgression-line (IL) population [30]. The study revealed a modular network with intra-modular connections of amino acids, sugars, and phenotypic traits, and highlighted the connections between metabolic and phenotypic traits. A similar approach was used to obtain several novel insights concerning the interrelation of seed primary [31] and fruit secondary metabolites with yield-associated traits [22]. Production of seed oils with desirable fatty-acid content raised the interest of oilseed breeders and biotechnologists as a result of the growing demand for vegetable oil [13,32–35]. Studies also focused on tomato seeds because their oil is an excellent source of important fatty acids involved in plant growth [36,37]. Following research has also been extended to other tissues, e.g., tomato leaves [31,38].

Despite the large-scale tomato cultivation and the number of studies related to the fruit's chemical composition and its quality, the majority of tomato metabolome research has focused on polar (non-lipid) primary and secondary compounds: sugars and their derivatives, amino acids, and organic acids. Traits such as aroma, color, and nutritional values, all deriving from secondary metabolites, have been extensively investigated, too [1,39]. However, lipid metabolism is still poorly studied in tomato, with several investigations focusing on cuticular waxes [40–43] and the characterization of genes involved in fatty-acid production [44].

In this study we performed a global correlation-based network analysis of metabolic traits, including lipids, volatiles, primary, and secondary metabolites from tomato fruit pericarp, and in addition plant phenotypic traits across a *S. pennellii* population in three field studies. This allowed us to expose the connectivity network between lipids and other branches of metabolism and key agronomical traits. We further generated a correlation network between the levels of lipophilic compounds and the expression of 1431 lipid-related genes in fruit and leaf. We were able to identify 123 potential lipid-related genes and obtained several novel insights concerning lipid metabolism.

#### **2. Results**

#### *2.1. The Overall Metabolic Network in S. pennellii ILs*

*S. pennellii* ILs have contributed over the years to the investigation of more than 2500 QTLs associated with plant morphology and to our understanding of the genetic regulation of primary, secondary metabolism, and lipids [31]. Here, in order to decipher the underlying regulatory organization of lipid metabolism together with other metabolic traits, we created a correlation-based network using already existing data of primary metabolites [45], secondary metabolites [22], lipophilic compounds [44], volatile compounds [10], and yield-related traits [30] across *S. pennellii* ILs in three independent seasons: 2001, 2003, and 2004 (see Section 4, Supplementary material Table S1). We correlated all metabolic and phenotypic traits against each other using Spearman's rank correlation coefficient. Figure 1 shows an overall view of a correlation-based network of all collected data in *S. pennellii* ILs. As expected, high correlations occur more often within the same group of metabolites or phenotypic traits than between different groups. The vast majority of overall correlations in the network were positive (84.4%) (Supplementary material Figure S1 and Table S2).

**Figure 1.** Overall metabolic network. Each node represents a metabolite or a whole plant phenotypic trait; edges connecting two nodes show an association between two traits. In total, the network is composed of 455 nodes and about 15,000 edges assembled into five large groups: lipophilic metabolites comprise of 171 nodes, primary and secondary metabolites have 89 and 147 nodes, respectively, phenotypic traits have 38 nodes, and the smallest group consists of 10 nodes and represents volatile compounds (Supplementary material Table S1).

#### *2.2. Lipophilic Compounds Expose Weak Correlation with Phenotypic Traits and Primary Metabolites*

We initially focused on correlations between lipids and phenotypic traits, volatiles, primary, and secondary metabolites (see Section 4).

Few correlations between different lipid classes and yield-related traits such as brix and plant weight were identified (Figure 2a). Specifically, galactolipids were correlated negatively to plant weight. Our results showed correlations between phenotypic traits and other metabolic classes, with 46% of traits being connected to primary metabolites. Brix revealed more connections to primary metabolites (primarily sugars) than to phenotypic traits. Brix reflects the total amount of soluble solids and therefore cannot be truly regarded as a phenotypic trait.

**Figure 2.** (**a**) Correlation between phenotypic traits and lipophilic compounds in *Solanum pennellii* introgression-lines (ILs). Yield-related traits showed three connection with several lipid classes such as phospho- and galactolipids. DAG—diacylglycerol; TAG—triacylglycerol; DGDG—digalactosyldiacylglycerol; MGDG—monogalactosyldiacylglycerol. (**b**) Correlation between phenotypic traits and volatiles in *S. pennellii*ILs. Various phenotypic traits are linked to each volatile class.

The network showed negative and positive correlations among phenotypic traits and various volatile compounds (Figure 2b). Lipid-, amino-acid-, and carotenoid-derived volatiles showed 69.2%, 55%, and 90% positive correlations with several phenotypic traits, respectively. Lipid-derived volatile compounds (hexanal, trans-2-hexenal, cis-3-hexen-1-ol) showed some positive correlations with flower-, seed-, and yield-related traits. In addition, yield-related traits showed a high number of negative correlations with amino-acid-derived volatile isovaleronitrile. Two secondary metabolites (calystegine A3 and calystegine B2) showed several correlations with phenotypic traits (Supplementary material Table S2).

Our data show strong, mainly positive (92.1%) correlations within the class of primary metabolites. In agreement with previous studies, we observed a highly interconnected amino-acid module (Gly, Ile, Val, Thr, and Ser) [31]. However, primary metabolites such as galactinol and guanidine highly connected with specialized metabolites. In addition, two alkaloids (calystegine A3 and calystegine B2), belonging to the class of secondary metabolites, presented a high number of correlations with primary metabolites.

We observed seven links between primary metabolites and lipophilic compounds such as phospholipids, glycerolipids, and galactolipids.

Analysis of primary metabolites against volatile compounds indicated no significant correlations between amino-acid-derived volatiles like isovaleronitrile, benzaldehyde, 3-methylbutanal, and 2-phenylethanol and their upstream biosynthesis substrates such as leucine and phenylalanine. However, guanidine showed correlations with all amino-acid-derived volatiles. In total, amino-acidderived volatiles revealed 80.8% positive correlations with primary metabolites. The majority of correlations between lipid-derived volatiles (trans-2-hexenal, hexanal, cis-3-hexen-1-ol) and carotenoidderived volatiles (geranylacetone and b-ionone) with primary metabolites were positive.

#### *2.3. Lipids Are Highly Correlated with Specialized Metabolites and Show Some Negative Correlations with Volatiles*

Our results showed a high number of correlations between various lipophilic compounds and secondary metabolites, with 4588 correlations representing 30.8% of all edges shown in the network (Supplementary Material Table S2). Our data introduced for the first time, a high correlation between specialized metabolites and lipophilic compounds. For example, phospholipids were correlated to almost all secondary-metabolite subclasses. Glycerolipids showed 1725 positive and 335 negative correlations with secondary metabolites. A high number of positive correlations (96%) was also found between galactolipids and secondary metabolites (Table 1, Supplementary material Figure S2).


**Table 1.** Percentage of positive correlations between lipids and secondary metabolites.

We identified 3249 correlations among various groups of secondary metabolites. Only 2% of all identified correlations were negative. Results indicated a high connection between and within different specialized-compound subclasses.

Correlation analysis revealed 10 positive and five negative interactions between lipid-derived volatiles and secondary metabolites. Correlations between amino-acid-derived volatiles and secondary metabolites showed an almost equal amount of positive and negative connections. Our observations revealed 58.7% (27 out of 46) of negative connections between carotenoid-derived volatiles and specialized metabolites (Figure 3a, Supplementary material Table S2). Additionally, a high number of negative correlations between triacylglycerols and volatiles was observed. Other lipid classes were connected to volatiles mostly positively (Figure 3b, Supplementary material Table S2).

**Figure 3.** (**a**) Connection between secondary metabolites and volatiles in *Solanum pennellii* ILs. Volatile compounds are linked to almost all classes of specialized metabolites. (**b**) Correlation between lipophilic compounds and volatiles in *Solanum pennellii* ILs. Lipid class of TAGs showed numerous negative connections with all types of volatile compounds. DAG—diacylglycerol; TAG—triacylglycerol; DGDG—digalactosyldiacylglycerol; MGDG—monogalactosyldiacylglycerol.

Within all lipid classes we identified 5841 correlations, representing 39.2% of all edges shown in the network, with 1012 (17.3%) being negative. Lipids belonging to the same subclass showed strong, mainly positive connections with other members of the same class (Table 2). For example, among different triacylglycerols (TAGs), out of 1229 identified correlations 1102 (89.7%) were positive. There was also a minor number of negative correlations within the monogalactosyldiacylglycerol (MGDG) and digalactosyldiacylglycerol (DGDG) subclasses. Out of all identified correlations within the phospholipids subclass, 78.1% were positive. Moreover, there are no negative correlations between diacylglycerols (DAGs) and DGDGs. The highest percentage of negative correlations was observed between phospholipids and TAGs (Table 2, Supplementary material Figure S3).


**Table 2.** Percentage of positive correlations between different lipid subclasses.

#### *2.4. Fruit-Specific Lipid-Related Genes Show a Mainly Positive Pattern of Change with Lipophilic Compounds*

For further exploration of lipid biosynthesis in tomato fruit pericarp, we correlated lipid profiles across 76 introgression lines [44] with multiple next-generation-sequencing gene-expression datasets from the same introgression lines. We used data consisting of 188 and 117 different lipophilic compounds from leaves and fruits of 74 ILs, respectively, and gene-expression data of lipid-related genes (1431) to perform the correlation analysis.

Using Spearman rank correlation coefficient, we identified 59 positive connections out of 173 in fruits and 33 positive correlations out of 73 in leaves. Figure 4 shows the significant correlation values between gene-expression levels and different lipid concentrations in fruits (Figure 4a) and leaves (Figure 4b).

**Figure 4.** Heatmap of correlation values between lipid-related genes expression and lipid levels in tomato (**a**) fruits and (**b**) leaves. The number of lipid-related genes is higher in the fruit dataset compared to the leaf dataset (81 and 42, respectively). In tomato fruit, lipid-related genes linked mostly to TAGs, while in tomato leaf predominantly to phospho- and galactolipids. DAG—diacylglycerol; TAG triacylglycerol; DGDG—digalactosyldiacylglycerol; MGDG—monogalactosyldiacylglycerol; SQDG sulfoquinovosyldiacylglycerol.

Our analysis identified 123 potential lipid-related genes in both fruit and leaf datasets (Supplementary material Table S3). In agreement with previous results [44], we observed that lipase (*Solyc12g055730)* showed a significant high correlation with different levels of TAGs (Figure 4a).

Our fruit data showed that the expression of the gene for 1-acyl-sn-glycerol-3-phosphate acyltransferase (GPAT) (*Solyc11g065890*) is positively correlated with 18 unsaturated TAGs, four of which showing significant mQTLs in IL 11-2 and IL 11-3 [44]. Further, expression analysis revealed that in IL 11-2 this gene is expressed 1.5 times lower compared to M82, and that in IL 11-3 expression of the gene is 0.69 times higher than in M82. Moreover, various TAGs display correlations with lipid-related genes putatively annotated as phospholipase D (*Solyc01g103910*)*,* non-specific lipid transfer protein (*Solyc10g075150*), acyl-ACP thioesterase (*Solyc12g006930*), and lipoxygenase (*Solyc03g122340*), whereas another lipoxygenase (*Solyc01g099210*) showed five positive correlations with various phospholipids. We also observed connections of galactolipids to several lipid-related genes (Supplementary material Table S3).

Finally, we identified strong correlations between lipophilic compounds and the expression of a class III lipase (*Solyc09g091050*) located on chromosome 9. The gene was negatively correlated with DGDG 36:4 (–0.46). Our previous QTL data showed a significant mQTL for galactolipids and phospholipids in this region. The QTL was mapped to a narrow overlapping region of IL 9-3, IL 9-3-1, and IL 9-3-2. The levels of DAGs, DGDGs and MGDGs, and TAGs were significantly decreased in IL 9-3, IL 9-3-1, and IL 9-3-2 compared to M82. Furthermore, gene expression was 4.6-fold higher in cultivated tomato compared to introgression lines harboring the gene (Figure 5).

**Figure 5.** Transcript level of class III lipase (*Solyc09g091050*) in red ripe fruits of M82, *Solanum pennellii*, IL 9-3, IL 9-3-1, IL 9-3-2. Expression of the gene in wild tomato species *Solanum pennellii* as well as in ILs carrying the same allele version is lower in comparison to cultivated tomato variety M82. Asterisks indicate significant differences (\* *p* < 0.05).

To provide additional support for the observed results, we performed genomic sequence analysis of the promoter region of the lipase (*Solyc09g091050*) and compared the promoter sequences of *S. lycopersicum cv. M82* and *S. pennellii* by genome alignment [17]. Results showed several small deletions and nucleotide substitutions in the promotor region, while the coding region showed 99% similarity between *M82* and *S. pennellii*.

Additionally, we investigated several significant correlations between different lipid classes and other lipid-related genes in tomato fruits (Supplementary material Table S3).

#### *2.5. Leaf-Specific Lipid-Related Genes Show Many Negative Correlations with Lipophilic Compounds*

Similar to the above, we extended our analysis and combined leaf lipid profiling of ILs [44] with transcriptomic data from the same lines [46,47]. Using Spearman rank correlation coefficient, we identified 45.2% (33 out of 73) and 54.8% (40 out of 73) positive and negative correlations, respectively (Figure 4a). Our results show correlations between lipid transfer protein (*Solyc03g079880*), 3-ketoacyl CoA thiolase 1 (*Solyc09g061840*), and glycerophosphoryl diester phosphodiesterase (*Solyc11g045040*) with different DAGs and phospholipids. Phospholipids additionally were linked to phospholipid-translocating flippase (*Solyc01g011100*), diacylglycerol kinase (*Solyc01g096500*), and lipid transfer proteins (*Solyc03g119210* and *Solyc10g075070*). Moreover, one transfer protein, *Solyc03g119210*, was correlated with galactolipids (DGDG 32:3 and SQDG 32:1). Many other significant correlations between different lipid classes and lipid-related gene candidates were identified (Supplementary material Table S3).

The cholesterol acyltransferase gene (*Solyc05g050710*), located on chromosome 5 (IL 5-3), predicted to take part in lipid catabolism, exhibited significant positive correlation with PC 32:0 (0.4). In the same region, several significant mQTLs for different lipid classes were identified [44]. In addition, the *Solyc05g050710* expression is almost 20-fold higher in *S. pennellii* compared to the cultivated variety (M82) (Figure 6a). Comparison of the coding sequence between *S. pennellii* and M82 revealed 99% identity. However, there are many differences in the promotor region, including a 17-bp deletion in *S. pennellii* compared to the cultivated variety, and additionally a 47-bp deletion in M82 compared to the allele derived from the wild species. This may account for the difference in the expression levels of *Solyc05g050710* between M82 and its wild relative *S. pennellii* (Figure 6b).

**Figure 6.** (**a**) Transcript levels of the cholesterol acyltransferase gene (*Solyc05g050710)* in red ripe fruits of M82, *Solanum pennellii*, IL5-3. Expression of the gene in wild tomato species *Solanum pennellii*, similar to IL 5-3 expressing the same allele version, is higher compared to cultivated tomato variety M82. (**b**) Comparison between the promoter regions of allelic versions of *Solyc05g050710* derived from cultivated tomato variety M82 and wild tomato *Solanum pennellii*. Deletions of 47 and 17 bp can be found in the promotor sequence of *Solanum lycorepsicum* and *Solanum pennellii* compared to the respective other. Asterisks indicate significant differences (\* *p* < 0.001).

#### **3. Discussion**

#### *3.1. Network Analysis: Correlation between Metabolic and Phenotypic Traits*

Numerous efforts have been made to identify and characterize metabolic quantitative trait loci (mQTLs) in tomato, with a focus on primary and secondary metabolites [10,22,30,44,48,49]. In addition, QTLs for volatile organic compounds from tomato fruit [24,50–52] and acyl sugars in tomato leaf trichomes have been defined [53,54], with further studies focusing on natural variation [54–58] and cuticle composition [59]. A series of genome-wide association studies (GWAS) contributed to assessing the effects of domestication and crop improvement on the fruit metabolome [15,46,47,59–62]. Despite all these studies, relatively few investigations have hitherto aimed at understanding the genetic basis of lipid composition in tomato fruit, with few studies mainly focused on cuticular lipids [17,59,63,64].

Here, we investigated correlations between different classes of metabolites and phenotypic traits in fruits, combined with expression analysis of lipid-related genes in tomato fruits and leaves. The generated network partially validates previously discovered correlations and presents new ones. For example, a QTL for brix (*Brix9*-*2*-*5*) deriving from green-fruited tomato species increases glucose

and fructose contents in cultivated tomato fruits [65]. We identified in our network significant positive correlations of brix with different sugars (Supplementary material Table S2). Additionally, we observed strong correlations between amino acids (Gly, Ile, Val, Thr, and Ser), with an average *r*-value of 0.78 and small percentage of negative interactions of correlations within the class of secondary metabolites, in agreement with previous studies [22,31,65].

Large classes of lipophilic compounds, similarly as specialized metabolites, display a considerable diversity across different plant species, e.g., isoprenoid-derived compounds are considered "secondary" metabolites produced in a cell-specific manner and are not directly involved in cell growth and development [13]. Our results showed a high number of significant correlations (30.8% of all) between lipophilic and specialized compounds. The number of significant correlations between lipids and other metabolic traits is smaller.

Several yield-related traits showed correlations to galactolipids and to one phospholipid. Tomato fruit as an organ, unlike maize cobs, for example, does not store much lipids. Unlike seeds, tomato fruit and leaf cells do not accumulate high amounts of storage lipids. Lipophilic compounds in tomato fruits and leaves participating mainly in signaling, membrane structure, and development [66–69].

Furthermore, lipids, similarly to secondary metabolites, were less correlated with phenotypic traits (Figure 1). It has been shown experimentally that the variability in secondary metabolites does not impact morphological and yield-related traits [21,70,71]. Therefore, the ability to change lipid composition or levels in tomato fruits would be a valuable tool for improving fruit quality and flavor. For example, the composition of fatty acids can be significantly changed without altering the overall plant morphology [72]. Nevertheless, versatility of lipophilic compounds might indirectly affect traits like shelf-life, if cuticle lipids are changed [73]. Until now, the influence of lipophilic compounds on overall plant phenotype remains unclear.

The connection between secondary metabolites and lipids seems to be more direct since changes in one compound class can have an effect on the other. Experimental evidence has already highlighted the close connection between secondary metabolites and cuticle lipids [42], however, no experimental validation exists of the relationship between those two metabolic classes in fruit pericarp.

In our data, we identified several negative correlations between lipid-derived volatiles and various TAGs. The observations here confirmed previous finding in tomato fruit describing linkage between decreasing levels of TAGs and simultaneously increasing levels of volatiles originating from lipids [44]. In tomato fruit some volatiles are linked to overall liking and flavor intensity [25].

In our network, we observed strong correlations within the lipid class. Lipophilic compounds correlate mostly positively between each other. For example, most of the correlations between galactolipids and glycerolipids are positive. However, phospholipids showed numerous negative correlations with all other classes. These results indicate that phospholipids are functionally the most distinct lipid class.

#### *3.2. Network Analysis: Combining Metabolite Profiling and Expression for Gene Discovery*

Plant lipid metabolism is constantly under investigation. In the well-studied model plant *A. thaliana* only around 40% of the 700 genes putatively annotated as lipid related are functionally characterized. In other plant species like tomato, that number is much smaller [74].

Recently we applied a quantitative genetic strategy to a *S. pennellii* IL population and mapped more than 160 various lipid species belonging to 10 different classes, with a total of 1528 and 428 mQTLs in fruit and leaf, respectively [44]. Here, we combined lipid profiling in leaf and fruit tissues across 76 ILs with gene expression analysis in order to identify genes involved in lipid biosynthesis.

To validate our approach, we checked whether our data confirmed previously proven correlations. We identified connections between a class III triacylglycerol lipase (*Solyc12g055730*) and various TAGs (TAG 48:2, TAG 58:0, TAG 48:3). It has been suggested that the enzyme catalyzes TAGs for further volatiles biosynthesis [44].

Our results highlighted several other lipid-related candidate genes in fruits and leaves (Figure 4). In tomato fruit, for example, the eQTL of lipase (*Solyc09g091050*) and mQTL of DGDGs and phospholipids confirmed high correlation between the gene and DGDG 36:4 (–0.46). Further validation of the gene's function is required.

Furthermore, we identified a high number of lipid-related genes that correlated positively with TAGs. For example, the gene putatively annotated as acyl-ACP thioesterase (*Solyc12g006930*) correlates positively with nine TAGs with an average *r*-value of 0.43. The enzyme is essential in the process of chain termination during de novo fatty-acid synthesis [75]. Another example is the 1-acyl-sn-glycerol-3-phosphate acyltransferase gene (GPAT) (*Solyc11g065890*), which correlated with 18 unsaturated TAGs. The *A. thaliana* ortholog (*At3g57650*) was shown to be involved in phospholipid and TAG biosynthesis [76,77]. In tomato, GPAT catalyzes acylation at the *sn*-1 position of glycerol-3-phosphate to produce lysophosphatidic acid (LPA) with subsequent TAG synthesis [78]. For the same gene we found a correlation between level of expression from leaf dataset and MGDG 32:6. Differences in connections depending on the tissue type could suggest altered function of the same gene in fruits and leaves.

Interestingly, our data revealed mainly positive correlations between expression of lipid-related genes and levels of lipophilic compounds in tomato fruit, whereas in tomato leaves these were mostly negative (Supplementary material Table S3). This may suggest that the genes involved in biosynthesis and regulation of lipid metabolism are generally different between fruits and leaves. These results are supported by a higher number of identified mQTLs in fruit compared to leaves [44]. This may indicate that lipid-related genes were less affected in leaves than in fruits in the context of the domestication process [79]. This could further mean that tomato fruits as sink tissues, which are dependent on carbon supply from source tissues, might need a tighter regulation of lipid production, compared to tissues like leaves where carbon is assimilated, making a flux to lipid metabolism "shorter" and more flexible [80].

In our leaf dataset, a high number of lipid-related genes were found to be correlated mainly with phospholipids and galactolipids compared to other subclasses. Galactolipids, for instance, represent the most abundant lipid class in thylakoid membranes, organelles specifically in leaves [81]. For instance, in our study we identified correlations between phospholipids and phospholipid-translocating flippase (*Solyc01g011100*), diacylglycerol kinase (*Solyc01g096500*), and lipid transfer proteins (*Solyc03g119210*, *Solyc10g075070*). Moreover, *Solyc03g119210* correlates with galactolipids. Another lipid-related gene candidate expressed in tomato leaves—3-ketoacyl CoA thiolase 1 (*Solyc09g061840*)—exposes correlations with DAGs and phospholipids [82]. The gene ortholog in *A. thaliana* was suggested to be involved in fatty-acid beta-oxidation [83]. Several other lipid-related genes such as particle serine esterase (*Solyc04g077180*) [84], cyclopropane-fatty-acyl-phospholipid synthase (*Solyc04g056450*) [85], acyl-CoA-binding protein (*Solyc08g075690*) [86], and long-chain fatty alcohol dehydrogenase (*Solyc09g090350*) [87] showed correlations with various phospholipids, galactolipids, and glycerolipids (Supplementary material Table S3).

It has been observed that lipid metabolism could be genetically regulated on intra-class and inter-class levels [44]. We have identified several examples of genes following the pattern of intra-class level regulation such as GPAT (*Solyc11g065890*), which correlates with 18 TAGs. In contrast, the pattern of inter-class regulation was followed by lipid transfer protein gene (*Solyc03g079880*) or 3-ketoacyl CoA thiolase 1 (*Solyc09g061840*), which contributes to regulation of two different lipid subclasses simultaneously.

In this study, we evaluated metabolic trait correlations and performed global analysis of trait associations across a *S. pennellii* IL population. This is by no means the first time that network analysis has been used for evaluation of the relationships between traits in wide genetic populations, with many previous examples in Arabidopsis, potato, and maize [88–90]. Besides, the analysis has been applied for a range of different traits and tissue types in tomato populations [30,31,45]. However, here we included for the first time three major different lipid classes and revealed several insights concerning

the interrelation of traits from yield-associated traits, primary, and secondary metabolism, volatiles with lipids.

Our network using correlation between gene expression and metabolite levels combined with DNA sequence analysis highlighted several candidate genes putatively involved in lipid biosynthesis or regulation. The presented results complement previous studies regarding metabolic traits in a *S. pennellii* IL population [10,22,30,44,45] and can be used for expanding the knowledge of lipid metabolism in tomato.

#### **4. Materials and Methods**

#### *4.1. Plant Material*

Data used in this study were based on *S. pennellii* ILs. The *S. pennellii* IL population was created by replacement of marker-defined genetic regions of the wild species *S. pennellii* with homologous fragments of the cultivated tomato *S. lycopersicum* (M82), representing whole wild-genome coverage of *S. pennellii* [16].

We used already available data for primary and secondary metabolites, lipids, and phenotypic traits. The data were obtained using a population grown in the Western Galilee Experimental Station in Akko, Israel, in a completely randomized design with one plant per m2. The field was irrigated with 320 m<sup>3</sup> of water per 1000 m2 of field area throughout the season. The harvest of fruit was done when 80%–100% of tomatoes were red [16]. All the data were obtained from peeled-off fruit pericarp. Primary and secondary metabolites were available for three independent seasons: 2001, 2003, and 2004 [30,45]. Lipids data were available for seasons 2001 and 2003 [44]. Yield-related traits were available for seasons 2001 and 2004 [30], and flower-, seed- and fruit-related phenotypic traits—for season 2004 [30,91].

The data for volatile compounds were obtained from an *S. pennellii* IL population [10]. All lines were grown in randomized, replicated plots in three different sites (Gainesville, Citra, and Live Oak, Florida) over the seasons of 2002 to 2004. Volatile data we focused on in our study were consistently available only for season 2003. We also used available data for volatiles of interest for season 2004. Plants were grown using standard commercial practices in raised plastic mulched beds. Fruits from all plants for each line were combined and analyzed as they reached the red ripe stage [10].

Transcriptomic data across *S. pennellii* ILs RNA-seq of 1431 lipid-related genes from young leaf [47] and fruit peeled-off pericarp were obtained under http://ted.bti.cornell.edu [46]. We selected all expressed lipid-related genes across the tomato genome (based on GO). We extracted 647 and 786 lipid-related genes from leaf and fruit datasets, respectively.

#### *4.2. Correlation Analysis*

All metabolite and transcript values used for correlation analysis correspond to the standard scores of the log transformed data. Spearman correlation matrices were calculated in R (R Development Core Team, 2010) using the *cor* function of the *stats* package (https://www.rdocumentation.org/packages/ stats). For the trait/trait correlations we used a critical *p*-value of 0.05, since it is a commonly used threshold for statistical analysis. Correlation *p*-values were obtained by performing a permutation test based on Spearman correlations using the using the perm.cor.test function of *jmuOutlier* package (https://www.rdocumentation.org/packages/jmuOutlier/versions/2.2/topics/perm.cor.test), set for 20,000 permutations. To select the most meaningful correlations for the network analysis, arbitrary cut-offs were set to an absolute correlation coefficient higher than 0.3 and 0.4 for trait/trait and trait/transcript networks, respectively. We set up the cut-off for trait/trait correlations so that known correlations would be incorporated in our network. Approved significant correlation between brix and sugars were shown before [65]. In our network the average correlation between brix and sugars were 0.378. Additionally, we used a relatively relaxed correlation coefficient threshold of 0.3, because we were integrating data from different platforms. For trait/transcript cut-off 0.3 we reported in total 1537

correlations with lipid classes. To get more insight on some of these correlations we decided to use a stronger cut-off of correlation coefficient (≥0.40).

Depending on the measurement or dataset, we had different amounts of replicates for the different traits in the introgression lines. For GC-MS replicate number was between 1 and 11, for LC-MS between 1 and 12, for lipids between 3 and 4, for phenotypic traits between 1 and 12 and for volatiles always 1. The number of replicates for M82 wild type was usually much higher.

The metabolite/metabolite network plot was produced using Cytoscape version 3.6.1 with nodes representing different metabolites and phenotypic traits and edges representing pairwise correlation above the set threshold. All metabolites and transcripts exhibiting at least one pairwise correlation above the metabolite/transcript correlation network cut-off were selected to be represented in the heatmap produced using the *pheatmap* (https://CRAN.R-project.org/package=pheatmap) package in R.

#### *4.3. Promoter Analysis*

Promoter analysis of *Solyc05g050710* and *Solyc09g091050* was performed on the accessions *S. lycopersicum* (M82) and *S. pennellii* (LA0716). Alignment of the promoter region of *Solyc05g050710* and *Solyc09g091050* was done with CLUSTALW (http://www.genome.jp/tools/clustalw/).

#### *4.4. Trait Classes Used for Correlations*

Phenotypic traits were divided to flower traits (anther length, anther width, anther length/width ratio, ovary length, ovary width, ovary length/width ratio, style length, style width, style length/width ratio), seed traits (seed length, seed width, seed length/width ratio, seed weight, seed number per fruit, seed weight per fruit, seed number per plant, seed weight per plant, seed number per fruit unit, inflorescence, flowers per inflorescence, flowers per plant), fruit-related (fruit length, fruit width, fruit length/width ratio, fruit pericarp thickness, fruit length/pericarp thickness ratio, fruit width/pericarp thickness ratio, fruit locule number) and yield-related ((brix (BX), brix yield (BY), plant weight (PW), total yield (TY), harvest index (HI), biomass (BM), fruit number (FN), red fruit weight (RED), earliness (EA), mean fruit weight (FW)) subgroups. Specialized metabolites were divided to flavonoids, glycoalkaloids, phenolics, N-containing compounds, hydroxycinnamate derivatives, acyl sugars, polyamines, and unspecified compounds subgroups, lipophilic compounds to trialycglycerols, diacylglycerols, phospholipids, digalactoyldiacylglycerols, monogalactoyldiacylglycerols subgroups, and volatile compounds to carotenoid-, lipid-, and amino acid-derived subgroups. Volatile compounds were divided to carotenoid-, lipid-, and amino acid-derived subgroups.

**Supplementary Materials:** The following are available online http://www.mdpi.com/2218-1989/10/4/152/s1; Figure S1: Percentage of positive and negative correlations in overall metabolic network, Figure S2: Percentage of positive and negative correlations between lipophilic compounds and secondary metabolites, Figure S3: Percentage of positive and negative correlations within the class of lipophilic compounds. Table S1: classes and subclasses of metabolic and phenotypic traits used for the network analysis, Table S2: All identified correlations between metabolic and phenotypic traits used for the network analysis. Table S3: correlation between lipid-related genes in fruits and leaves.

**Author Contributions:** Conceptualization, S.A.;Y.B.; methodology, L.P.d.S., A.K., M.W.A.; formal analysis, A.K.M., W.A.; writing—original draft preparation, S.A., A.K., M.W.A.; writing—review and editing, S.A., Y.B., L.P.d.S., J.V., A.K., M.W.A.; visualization, A.K., M.W.A.; supervision, S.A., Y.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was in part supported by the PlantaSYST project by the European Union's Horizon 2020 research and innovation programme (SGA-CSA No 664621 and No 739582 under FPA No. 664620), and ISRAEL SCIENCE FOUNDATION (grant No. 859/19).

**Acknowledgments:** We thank Dr. Hezi Tenenboim for his kind editorial assistance.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Metabolomics to Exploit the Primed Immune System of Tomato Fruit**

**Estrella Luna 1, Amélie Flandin 2,3, Cédric Cassan 2,3, Sylvain Prigent 2,3, Chloé Chevanne 2, Camélia Feyrouse Kadiri 2, Yves Gibon 2,3 and Pierre Pétriacq 2,3,\***


Received: 14 February 2020; Accepted: 4 March 2020; Published: 6 March 2020

**Abstract:** Tomato is a major crop suffering substantial yield losses from diseases, as fruit decay at a postharvest level can claim up to 50% of the total production worldwide. Due to the environmental risks of fungicides, there is an increasing interest in exploiting plant immunity through priming, which is an adaptive strategy that improves plant defensive capacity by stimulating induced mechanisms. Broad-spectrum defence priming can be triggered by the compound ß-aminobutyric acid (BABA). In tomato plants, BABA induces resistance against various fungal and bacterial pathogens and different methods of application result in durable protection. Here, we demonstrate that the treatment of tomato plants with BABA resulted in a durable induced resistance in tomato fruit against *Botrytis cinerea*, *Phytophthora infestans* and *Pseudomonas syringae*. Targeted and untargeted metabolomics were used to investigate the metabolic regulations that underpin the priming of tomato fruit against pathogenic microbes that present different infection strategies. Metabolomic analyses revealed major changes after BABA treatment and after inoculation. Remarkably, primed responses seemed specific to the type of infection, rather than showing a common fingerprint of BABA-induced priming. Furthermore, top-down modelling from the detected metabolic markers allowed for the accurate prediction of the measured resistance to fruit pathogens and demonstrated that soluble sugars are essential to predict resistance to fruit pathogens. Altogether, our results demonstrate that metabolomics is particularly insightful for a better understanding of defence priming in fruit. Further experiments are underway in order to identify key metabolites that mediate broad-spectrum BABA-induced priming in tomato fruit.

**Keywords:** tomato; metabolomics; biochemical phenotyping; priming; BABA; *Botrytis cinerea*; *Phytophthora infestans*; *Pseudomonas syringae*

#### **1. Introduction**

The increase in world food demand and the indiscriminate use of chemical fertilisation highlight the need to adopt sustainable crop production strategies. Given the major threat of phytopathogenic microbes to food production [1] and ecosystem stability worldwide [2], novel practices are needed to combat these threats. Tomatoes are a highly consumed fruit that represent the eleventh largest commodity, with nearly 183 million tons produced in 4 million hectares in 2017 [3]. Crop yields are strongly affected by filamentous and bacterial pathogens, including the fungus *Botrytis cinerea*, the oomycete *Phytophthora infestans* and the bacterium *Pseudomonas syringae.* These pathogens can claim the complete loss of the crop within days of exposure [4–6]. Currently, strategies of control against these biological threats are based on the use of chemical pesticides applied at a pre-harvest stage. However, up to 50% of tomato losses occur at a post-harvest stage [7] due to, among many reasons, the inability to use chemicals at this stage due to residue toxicity. Therefore, new methods of disease control are needed in order to control infections by pathogenic microbes. Exploiting the plant immune system can represent an effective strategy to provide sustainable disease protection [8,9].

Plants are able to defend themselves against pathogens thanks to their innate immune system [10]. In addition, plants are able to sensitise their immune system to protect themselves against biotic threats. This is known as the priming of defence, which is commonly referred as the adaptive part of the plant immune system [11]. Priming occurs after the perception of stimuli that lead to an enhanced responsiveness of defence mechanisms upon subsequent attack [12]. Among stimuli, the chemical agent β-aminobutyric acid (BABA) has been widely studied for its capacity to result in broad-spectrum-induced resistance (IR) in a broad range of plant species [13]. BABA is a non-protein amino acid that has been demonstrated to be a plant product [14]. The work done in the model plant *Arabidopsis* (*A.*) *thaliana* revealed that this outstanding performance is due to the priming activity of multiple signalling pathways [13]. BABA primes salicylic acid (SA)-dependent defences and the deposition of callose, which result in effective protection against biotrophic and necrotrophic pathogens, respectively [15,16]. Importantly, it has been reported that, in many plant species, treatments with BABA result in a stress phenotype that manifests as changes in plant development (e.g., growth, yield, seed production) [17–19]. The discovery of the molecular receptor of BABA in *A. thaliana* sheds light into the reasons behind the stress response associated with this chemical: BABA binds to an aspartyl-tRNA synthetase and blocks the enzyme, consequently triggering the accumulation of its canonical substrate, uncharged tRNA, which leads to the activation of the stress response associated with amino acid imbalance in the plant [20]. Moreover, high concentrations of BABA or in specific plant species such as potatoes, can also lead to stress, as the chemical directly activates defence mechanisms, which is a costly trade-off in terms of energy resources to the plant [19,21].

In the tomato system, BABA is known to be able to induce resistance against at least 10 different pests and pathogens, including *B. cinerea*, *P. infestans* and *P. syringae* [13]. Similarly to what has been described in *A. thaliana*, priming of SA-dependent mechanisms has been reported [13]. However, it is likely that further priming mechanisms are responsible for its capacity to induce broad-spectrum resistance. Moreover, studies have demonstrated that priming of BABA is long-lasting [22]. For instance, it has been reported that, after the treatment of tomato seedlings, BABA-IR against *B. cinerea* is maintained for weeks in leaves [18]. Moreover, analysis of the resistance phenotypes in tomato fruit demonstrated that BABA-IR reaches the fruiting stages, as tomatoes from plants that had been treated with BABA at the seedling stage were more resistant to *B. cinerea* than the control [23].

Treatments of plants with BABA did not impact yield or fruit size but resulted in delayed fruit production and ripening [23]. In fruit, durable induced resistance has been linked to the accumulation of specific metabolites. For example, it was reported that, after treatment of seedlings with BABA, there was an accumulation of metabolites associated with alkaloid, terpenoid or jasmonate pathways [23]. It was therefore speculated that these metabolites could be responsible for the enhanced resistance, and therefore could mark the priming fingerprint in tomato fruit [24]. Importantly, however, the metabolites responsible for expression of priming of defence mechanisms after infection still remain unknown. Here, we aimed to determine the metabolic shifts that underpin the BABA priming of immune responses against pathogens of a different nature that infect fruit, including necrotrophic and biotrophic microbes. A combination of targeted biochemical phenotyping for major plant compounds involved in central metabolism and untargeted metabolomics could unveil discriminant metabolic biomarkers that respond to BABA priming and inoculations. To the best of our knowledge, this is the first metabolomic study with multiple fruit pathosystems in relation to the priming of immune responses.

#### **2. Results**

#### *2.1. E*ff*ect of BABA on Broad-Spectrum Resistance in Fruit*

We determined whether the treatment of tomato plants with BABA resulted in a durable induced resistance in tomato fruit against the fungal necrotrophic pathogen *B. cinerea* (*Bot*), the biotrophic oomycete *P. infestans* (*Phy*) and the hemibiotrophic bacteria *P. syringae* pv. *tomato* (*Pst*). Box-plots showing sample datapoints and mean of the scored symptoms (*n* = 11) revealed that BABA-treated plants produced fruit that were statistically more resistant to *Bot*, *Phy* and *Pst* (*P* < 0.01) than the controls (Figure 1). This indicates that BABA is able to induce resistance in tomato plants for a durable fruit protection against pathogens that have different infection strategies.

**Figure 1.** ß-aminobutyric acid (BABA) primes tomato fruit for a durable disease resistance against three different pathogenic microbes. (**A**) photographs showing tomato fruit 2 days after infection with *Botrytis cinerea* (*Bot*, left), *Phytophthora infestans* (*Phy*, middle) or *Pseudomonas syringae* pv. *tomato* (*Pst*, right). (**B**) box-plots of disease symptoms from fruit that originated from plants treated with water or BABA (500 μM) then inoculated with water as a mock control or with *Bot* (left), *Phy* (middle) and *Pst* (right). Symptoms were scored 2 days after inoculation from 11 biologically replicated fruit (*n* = 11). Asterisks indicate statistically significant differences between water- and BABA-treated plants *(t-*test, *P* < 0.01).

#### *2.2. E*ff*ect of BABA on Fruit Yield and Development*

We investigated whether the treatment of tomato seedlings with BABA could impact fruit development and yield. As described previously [23], no differences were found in fruit size (Figure 2A), but delayed fruit production (Figure 2B) and fruit ripening (Figure 2C,D) were reported. However, it was observed that, at 13 weeks of growth, the proportion of green fruit was statistically significantly higher in BABA-treated plants (40%) compared to control plants (33%) (Figure 2D), indicating that fruit production did not slowdown in BABA-treated plants after ripening processes had begun (Figure 2D). This might suggest a positive trade-off for BABA treatment on fruit development.

**Figure 2.** Effect of BABA on fruit yield and development. (**A**) fruit size (cm) from plants treated with water (blue) or 500 μM BABA (yellow). NS: not statistically significant (*t*-test, *P* > 0.05). (**B**) number of green fruit produced from water (blue)- or BABA (yellow)-treated plants after weeks of growth. Asterisks indicate statistically significant differences (*t*-test, P > 0.05). (**C**) number of ripped fruit produced from water (blue)- or BABA (yellow)-treated plants after weeks of growth. Asterisks indicate statistically significant differences (*t*-test, *P* < 0.05). (**D**) Proportion of fruit at different stages of fruit ripening in control and BABA-treated plants, expressed as percentage of occurrence per treatment at different weeks of growth. Asterisks indicate statistically significant differences between distributions at specific timepoints (Chi-square test, *P* < 0.05).

#### *2.3. Global Metabolomics after BABA Treatment and After Inoculation*

In order to substantiate the disease resistance after BABA treatment (Figure 1), we investigated fruit metabolism from ethanol extracts of freeze-dried tomato pericarps (*n* = 4) using *i*) targeted biochemical profiling of several major compounds involved in the central metabolism [25,26], and *ii*) untargeted metabolomics of semi-polar metabolites, including specialised compounds, via ultra-high-performance liquid chromatography coupled to electrospray ionisation orbitrap high-resolution mass spectrometry (thereafter referred to as LCMS). For this, the first, second and third fruit developed in the plants were used for pathoassays with *B. cinerea (Bot)*, *P. infestans (Phy)* and *P. syringae (Pst)*, respectively. Unbiased processing of LCMS data, followed by filtering of the most reliable variables, generated 6887 metabolomic features (see Materials and Methods). A global overview of metabolic profiles was visualised by an unsupervised multivariate statistical method, Principal Component Analysis (PCA), for all combinations of priming and pathosystems (Figure 3).

**Figure 3.** Global metabolomic changes after BABA treatment and after pathogen inoculation. Principle component analysis (PCA) score plots (*n* = 4) of 6887 LCMS-based metabolomics features (**A**) and 11 major compounds analysed by targeted biochemical phenotyping (**B**). Maximal variance explained by each PC is given in brackets.

Firstly, for the 6887 metabolomic signals (Figure 3A), PCA explained 35% of the maximal variance of the dataset and resulted in a clear differentiation of the water and BABA treatments, thus suggesting a greater impact on metabolomic profiles for direct BABA application as compared to other conditions. This was confirmed by a univariate statistical method through a two-factor ANOVA (*P* < 0.05), which

quantitatively resulted in more statistically significant markers for the BABA factor (3052; 44%) than for the inoculation factor (2309; 33%) or the interaction (401; 6%) (Table 1).


**Table 1.** Univariate statistical analysis of the metabolomic features and major compounds.

Secondly, we tested 10 major compounds involved in central metabolism (sucrose, fructose, glucose, starch, fructose-6-P, glucose-6-P, glutamate, malate, fumarate and total proteins), as well as total polyphenols. PCA explained 83% of the maximal variance in the dataset and resulted in a separation of fruit by developmental characteristics (i.e., the first and second fruit versus the third fruit) rather than by pathosystems (Figure 3B). Hence, this multivariate differentiation indicates that the profiles of primary metabolites mostly respond to the developmental stage of the fruit, which supports the idea that central metabolism is tuned to fruit growth [27–29]. Complementarily, two-factor ANOVA (*P* < 0.05) only generated significant markers for the inoculation factor (4; 36%), including sucrose, fructose, glutamate and fumarate (Table 1 and Figure S1). Interestingly, such markers dropped upon *Bot* and *Phy* infections, while they were not drastically affected by *Pst* infection (Figure S1). Besides, fructose pools remained low across all treatments within the *Pst* pathoassay (i.e., third fruit). Altogether, this indicates that fruit infection affects the pools of central metabolites and those changes depend on the pathosystem.

Furthermore, a partial segregation of pathogen inoculations was observed on the PCA score plots obtained for each pathosystem from a dataset combining LCMS and targeted analyses (Figure 4). This was further exemplified by a supervised Partial Least Square Discriminant Analysis (PLS-DA) allowing a better differentiation of pathosystems and priming treatments (Figure S2). Two-factor ANOVA (*P* < 0.05) for each pathosystem not only confirmed that the BABA factor quantitatively outweighed the inoculation factor and the interaction, but also showed that all these factors were substantial (Table 2). Hence, this indicates that microbial challenges elicit distinct metabolic profiles. Overall, these results reveal metabolic shifts in fruit upon BABA priming and after pathogen inoculation, notably towards semi-polar biochemicals potentially involved in plant stress mitigation (i.e., specialised metabolites).


**Table 2.** Univariate statistical analysis for each pathosystem.

**Figure 4.** Partial segregation of pathogen inoculations for each fruit pathosystem. PCA score plots (*n* = 4) of 6898 features (6887 electrospray ionisation orbitrap high-resolution mass spectrometry (LCMS) variables + 11 major compounds) showing metabolomics overview between the three different pathosystems. Maximal variance explained by each PC are given in brackets.

#### *2.4. Primed Responses to Specific Pathogenic Microbes*

To gain more insight into BABA priming upon different fruit infections, we next performed quantitative binary comparisons of metabolic markers for each pathosystem by comparing water-treated, mock-inoculated fruit versus *i*) BABA-treated, mock-inoculated fruit, *ii*) water-treated, pathogen-inoculated fruit, and *iii*) BABA-treated, pathogen-inoculated fruit. The resulting statistically significant metabolic markers (*t*-test, *P* < 0.01) were used to construct Venn diagrams showing common and specific markers (Figure 5A). Very few overlaps were observed between BABA (red) and pathogen (blue) conditions (2, 2 and 0 for *Bot*, *Phy* and *Pst*, respectively) and between pathogen and BABA priming (green) conditions (7, 4 and 4 for *Bot*, *Phy* and *Pst*, respectively). Instead, several markers overlapped between BABA treatment and BABA priming (118, 12 and 158 for *Bot*, *Phy*, and *Pst* respectively), and most markers were found either for BABA treatment (*Phy* and *Pst*) or for BABA priming (*Bot*) (Figure 5A). This suggests that BABA results largely in the accumulation of metabolites that could be used during the expression of priming. In addition, metabolic markers that specifically responded to BABA priming in the different fruit pathosystems were compared through a Venn diagram in order to reveal the common metabolic signatures of BABA priming against the three different pathogens (Figure 5B). Strikingly, no common markers were found upon the three infections, although very few markers were observed between *Pst* and *Bot* (10), *Pst* and *Phy* (3), and *Bot* and *Phy* (1). Hence, the primed responses are likely tailored to the encountered pathogenic microbes.

**Figure 5.** The primed responses are tailored to the encountered pathogenic microbes. (**A**) Venn diagrams showing quantitative binary comparisons were performed for each pathosystem (*t*-test, *n* = 4, *P* < 0.01) between water-treated, mock-inoculated fruit versus BABA-treated, mock-inoculated fruit (BABA, red); water-treated, pathogen-inoculated fruit (pathogen, blue); BABA-treated, pathogen-inoculated fruit (priming, green). (**B**) Venn diagrams showing the resulting priming clusters for each fruit pathosystem.

#### *2.5. Putative Annotation of Metabolic Markers*

We then conducted a tentative annotation of the 14 metabolic markers that were common to *Pst* and *Bot* (10), *Pst* and *Phy* (3), and *Bot* and *Phy* (1) based on their detected *m*/*z* by high-resolution orbitrap-MS (Table 3). A Kruskal–Wallis test with correction for false rate discovery (Benjamini–Hochberg, *P* < 0.05) confirmed that these 14 metabolic markers showed statistically significant variations that were

visualised by bar charts (Figure S3). Putative prediction of compounds and pathways indicated several markers that belonged to the plant defence metabolism, including stress hormones and flavonoids. Interestingly, fungal pathogens (*Bot*, *Phy*) were associated with the induction of the putative marker jasmonoyl–isoleucine. Besides this, (hemi)biotrophic microbes (*Pst* and *Phy*) triggered the accumulation of putative salicylic derivatives and flavonoids (Figure S3). The *Pst*-related primed response further correlated with the depletion of a putative cytokinine. Hence, our results suggest that BABA priming against three different fruit pathogens rely on the induction of pathways involved in the defence hormonal metabolism. Further analytical studies are required to confirm the putative annotation of these priming markers.

#### *2.6. Modelling of Resistance to Multiple Fruit Pathogens*

Using a predictive biology approach based on generalised linear models [30], we aimed to determine whether resistance to fruit pathogens could be predicted by the detected metabolic markers (Figure 6). Based on those models (Figure 6A), good correlations were observed between measured and predicted values (mean = 0.87), and were statistically different from correlations based on randomly generated resistance (*t*-test, *P* < 2.2 <sup>×</sup> 10−16), which indicated the robustness of the predictions. Furthermore, according to the occurrence of metabolic markers in the models (Figure 6B), fructose appeared to be the best positively correlated predictor (appearing in 99% of the models), as well as sucrose, to a lesser extent (appearing in 41% of the models) (Table S1). Most predictors (32 out of 34) also showed a high statistical significance from a Kruskal–Wallis test with correction for false rate discovery (Benjamini–Hochberg, *P* < 0.05, Table S1). This corroborates the outcome from the two-way ANOVA method (Tables 1 and 2, and Figure S1). Hence, this indicates that soluble sugars involved in the central metabolism are essential to predict resistance to fruit pathogens. The analysis of such compounds is therefore critical for studies involving fruit–pathogen interactions. In addition to sugars, other metabolic predictors appearing in more than 25% of the models showed positive (19 markers) and negative correlations (15 markers) (Figure 6B and Table S1). Further analytical studies are required to annotate and/or identify such markers. Nonetheless, a tentative annotation of the top 15 predictors based on their detected *m*/*z* by high-resolution orbitrap-MS is presented in Table S1. Unsurprisingly, the resulting putative metabolites belonged to defence pathways (i.e., phenolics, flavonoids, terpenes, amino acid conjugates) and lipids. This suggests that immune perception and signalling seem pivotal in predicting resistance to fruit pathogens.

*Metabolites* **2020**, *10*, 96


**Table 3.** Putative annotation of the primed response markers.

194

Benjamini–Hochberg

 method.

**Figure 6.** Prediction of biotic resistance from metabolic markers. (**A**) correlation between predicted and measured resistance based on generalised linear models. (**B**) occurrence (%) of the metabolic markers in the models that showed a positive or negative correlation with the resistance to fruit pathogens. Details of untargeted markers are presented in Table S1.

#### **3. Discussion**

In the present study, we evaluated the metabolic composition of tomato fruit in relation to the BABA-priming of young tomato plants and the infection of three different pathogens at the fruit stage. To the best of our current knowledge, this is the first metabolomic study on three different fruit pathosytems interacting with BABA priming.

Firstly, untargeted metabolomic profiling indicated a great impact of BABA treatment on metabolic profiles (Figure 3A and Table 1). Hence, the treatment of young tomato plants with BABA metabolically primes fruit tissues, and this stimulation was likely more critical than it was for the pathogen inoculations. This might result from the hormonal nature of BABA, which deeply affects plant metabolism, or from stress-related responses that are activated by the chemical itself, as it has been reported previously for high concentrations of BABA or in another *Solanum* species (i.e., potato) [21].

Secondly, targeted analyses of compounds involved in central metabolism demonstrated that the primary metabolic pools responded to the pathosystem inoculations, which reflected the developmental stage of the fruit, as exemplified by the multivariate distinction between the first/second fruit and the third fruit (Figure 3B). In complement, fructose, sucrose, fumarate and glutamate showed statistically significant variations upon inoculation (Table 1). Since fruit of slightly different ages harbour different profiles of primary compounds, we could assume that central metabolism is tuned to fruit growth, more specifically soluble sugars, amino and organic acids. This agrees with previous phenotyping and modelling studies on tomato that demonstrate metabolic shifts in carbon metabolism in the growing fruit [25,28,29,31]. Furthermore, it has been recently confirmed through transcriptomics and proteomics that the developing fruit not only undergoes metabolic shifts in central pathways, but also redox metabolism, such as for pyridine nucleotides that are detrimental to energy homeostasis [27,32,33]. However, major questions remain regarding the nature and dynamics of shifts in central metabolism upon pathogen inoculation. It is reasonable to expect that further investigations involving a more comprehensive view of fruit primary metabolism and how microbial challenges dynamically affect such pathways might significantly improve our understanding of the relationships between central metabolism and fruit–pathogen interactions. In turn, this should provide novel strategies to obtain fruit of better quality and stress resilience [32].

Upon pathogen challenge, while BABA is effective in leaf tissue, very little is known about its contribution in fruit. According to our fruit pathoassays (Figure 1), the treatment of tomato seedlings with BABA resulted in a broad-spectrum resistance against microbes that have different infection strategies, including necrotrophic or biotrophic, and fungal, oomycete or bacterial pathogens. Further, the primed responses are tailored to the encountered pathogen, as exemplified by the little overlap between the different primed states of the three pathosystems (Figure 5). This implies that the induced resistance state is very specific, which strongly suggests that BABA primes multiple signalling pathways through which such different microbes are resisted in the fruit. Among those metabolic responses, hormonal regulations appear detrimental to BABA-induced immunity [23,24,34]. Accordingly, putative annotation of metabolomic markers indicates that hormone conjugates, including salicylic and jasmonic derivatives, and other defence compounds (i.e., flavonoids), are induced upon infection and BABA treatment (Table 3). Given the diverse set of immune responses that the fruit deploys against different microbial stresses, our study highlights the adaptability of priming as a "stimulus-dependent plasticity of response traits" [35]. For this reason, the exact underlying molecular mechanisms of priming are difficult to describe precisely and their description requires further research [36].

Whilst BABA treatment in many plant species results in a stress phenotype that manifests through developmental alterations (e.g., growth, yield, seed production) [17–19], we found no differences upon BABA application in fruit size, but observed delayed fruit production or fruit ripening (Figure 2), as described previously [23]. However, after the number of ripened fruit had equalised between both treatments, BABA-treated plants continued producing fruit at a much faster rate than the water-treated plants (Figure 2). Seemingly, through its induction of immune responses, BABA thus provides a positive fitness element for tomato plants. This trade-off might emerge, in part, from the stimulation of various signalling pathways, more specifically the ones that link to the central metabolism, such as amino acids or carbohydrates [34]. As a result, BABA-treated plants would perform particularly well. This agrees with what we know about plant perception of BABA in *A. thaliana*. The binding of BABA by an aspartyl-tRNA synthetase blocks the enzyme, consequently triggering the accumulation of its canonical substrate, uncharged tRNA, which leads to changes in amino acid pools in the plant, therefore affecting primary metabolism [20]. Subsequent signalling modulations might result from an alteration in amino acid precursors (e.g., ethylene, auxin) that would alter fruit production [37].

Despite its economic importance, the molecular mechanisms underlying the pathogenicity of *B. cinerea, P. infestans* and *P. syringae* are poorly understood in fruit. From a computational systems biology perspective, the study of plant–pathogen interactions involved structural and comparative genomics, transcriptomics, and protein–protein interactions [38]. Further, high-resolution metabolome data and sufficient datapoints over time are essential to calculate metabolite coefficients and thus predict metabolic fluxes [39]. Recently, genome-scale metabolic models of *Solanum* species (i.e., potato, tomato) and *Phy* have been integrated to simulate the metabolic fluxes that occur during infection [40,41]. These studies yield insights into the molecular aspects of photosynthesis suppression by *Phy* via the flux of carboxylation to oxygenation reactions, or the nutrient intakes by *Phy* during different phases of the infection cycle. Interestingly, stage-specific profiles embedded in the joint metabolism of the host and pathogen could potentially be refined by integrating the high-resolution metabolome data of tomato infection [41]. Such elegant works involve leaf tissues. Here, we show that fruit metabolomics and modelling can assist in addressing fruit–pathogen interactions. Using top-down modelling based on the construction of generalised linear models [30], we demonstrate that metabolomics data can be used to accurately predict the measured resistance to various fruit pathogens (Figure 6A). Besides, through the evaluation of the occurrence of best predictors, our data indicate that soluble sugars, more specifically fructose [42], and defence metabolites are pivotal to predict the resistance to fruit pathogens (Table S1). Clearly, a more global systems biology approach based on a higher level of variation in the conditions (e.g., multiple genotypes or priming treatments, various growth stages of the fruit, several infection points) will shed some light on the underlying mechanisms of fruit–pathogen interactions.

Overall, our study validates the value of metabolomics and modelling approaches in the field of phytopathological investigations. This work provides a great perspective for the structural elucidation of the key metabolites involved in broad-spectrum BABA-induced priming in tomato fruit. Although it was only possible to tentatively annotate metabolic biomarkers on the basis of detected HR-accurate *m*/*z* (Table 3 and Table S1), our analytical and statistical approach can be further optimised for, e.g., metabolite identification through structural elucidation by NMR or targeted MS/MS analyses. A combination of LCMS with purification steps (e.g., SPE cartridge, fractionation) could prove useful for de novo identification.

#### **4. Materials and Methods**

#### *4.1. Tomato Cultivation*

Tomato (*Solanum lycopersicum*) Micro-Tom was used for all experiments described in this publication. Seeds were incubated for 4 days in wet paper at 28 ◦C to promote homogeneous germination. Germinates were then planted in individual 80 mL pots containing M3 soil. Plants were grown in a controlled-environment greenhouse chamber with 16h of light, at 26 ◦C, and 8 h of darkness at 21 ◦C, and 200 μM.m<sup>−</sup>1.s−<sup>1</sup> light intensity. Experiments were performed from November 2017 until May 2018 in the United Kingdom.

#### *4.2. Biochemicals, Reagents and Treatments*

All solvents and reagents used in this study were of analytical or MS grades. B-aminobutyric acid (BABA) was obtained from Sigma-Aldrich (A4420-7). Treatments with BABA were performed entirely as described in [23]. Briefly, 2 week-old tomato seedlings were soil-drenched with 8 mL per pot of either water or 5 mM BABA solution, to generate a final concentration of 0.5 mm in the soil. One week post-treatment, roots were carefully washed under running tap water and then plants were transplanted into individual 2.2 L pots containing untreated M3 soil. Plants were allowed to grow for between 9 and 12 weeks until the fruit turned red, at which point they were harvested and infected with the different pathogens.

#### *4.3. Fitness Parameters of Tomato Fruit*

Growth and yield were assessed entirely as described in [23]. Assessment of fruit ripening was done as described in [29], by classifying fruit in different levels of maturity by colour.

#### *4.4. Pathogens and Inoculations*

Cultivations of *Botrytis cinerea* strain R16 [43], *Phytophthora infestans* 88,069 [44] and *Pseudomonas syringae* pv. *tomato DC3000* (*Pst DC3000*) [45] were done as described in the corresponding publications. For inoculations with *B. cinerea,* the first fruit were used. Inoculations were performed entirely as described in [23]. For infections with *P. infestans,* the second fruit were used. Inoculations were performed by placing 10 <sup>μ</sup>l drops of a spore concentration of 5×10<sup>4</sup> spores/mL onto the needle-wounded tip of the tomato fruit. After infection, fruit were kept at 20 ºC in the dark. For *P. syringae* infections, the third fruit were used. Inoculations were done by spraying bacteria onto the fruit in a concentration of 10<sup>8</sup> cells/mL in 10 mM MgSO4 and 0.05% (v/v) Silwet L-77. Infected fruit were kept in the dark at 25 ◦C. Mock inoculations were performed by following the exact same protocols but without pathogens in the solutions. Fruit were 56 days post-anthesis (dpa), 63 dpa and 70 dpa for the first, second and third fruit, respectively.

Scoring of *B. cinerea* symptoms were performed entirely as described in [23]. Scoring of *P. infestans* disease was done by classifying lesions into different categories of fruit colonization: Class 0; healthy, Class I; necrosis associated with the lesion, Class II; necrosis and mycelium associated with the lesion, Class III; necrosis and mycelium spread in the fruit. Scoring of *Pst DC3000* disease was done by classifying lesions into different categories of fruit damage: Class 0; healthy, Class I; turgent but cracking fruit, Class II; cracked fruit, Class III; fruit tissue collapse. Disease severity rates were calculated from the nominal lesion categories of four fruit per plant (*n* = 11), as described in [46]. Statistical analysis of disease phenotypes was performed as described in [23].

#### *4.5. Metabolite Extraction*

For metabolome analysis, the first, second and third fruit developed by plants were used for pathoassays with *B. cinerea (Bot)*, *P. infestans (Phy)* and *P. syringae (Pst)*, respectively. Experiments on each type of fruit were separated by one week. Infections were performed as described above when the corresponding fruit were fully ripened. Two days after inoculation with the different pathogens, fresh pericarps were rapidly collected into 2 mL-microtubes, then flash-frozen in liquid nitrogen and freeze-dried for 72 h (Pilote Compact, SARL CRYOTEC, Saint-Gély-du-Fesc, France). Fine grinding of dried material was subsequently performed using a ball mixer for 2 min at 30 Hz (Retsch Mill MM400, fisher scientific, Bordeaux, France) after adding two metal beads (Beads inox AISI 400C 5 mm, CIMAP, Caen, France) to each tube (Micro-tube, 2 mL PP, Sarstedt, Germany). Ten milligrams of each replicated sample were weighed into 1.1 mL-micronic tubes (MP32033L, Micronic, Lelystad, Netherlands), randomised onto a 96-micronic rack (MPW51001BC6, Micronic, Lelystad, Netherlands) then capped using a robotised capper–decapper (Decapper 193000/00, Hamilton, Bienne, Switzerland). Each rack also contained an empty tube corresponding to the extraction blank. The resulting micronics were then stored at −80 ◦C. Extraction of metabolites was conducted on four biologically replicated pericarp samples (*n* = 4) using a robotised extraction method developed at *Bordeaux Metabolome Facility* (https://metabolome.cgfb.u-bordeaux.fr/en, Villenave d'Ornon, France). The robot was a bespoken piece of equipment that allowed for pipetting solvents, mixing, cooling and centrifuging racks of micronics. After decapping the micronics, the extraction began by adding 300 μL of solvent A containing 80% ethanol and 0.1% formic acid (v/v) with 250 μg/mL methyl vanillate as the internal standard. Racks were agitated on the robot (30 sec, 500 rpm) then placed for 15 min into a sonicator containing ice-cold water (Elmasonic S300, Elma, Singen, Germany). Racks were put back on the robot and centrifuged (5 min, 1350 *g*). The first round of extraction stopped by pipetting 300 μL of the resulting supernatant into new 1.1 mL-micronic tubes. A second round of extraction was performed with 300 μl of solvent A, and the resulting pellet was finally washed with solvent B (50% ethanol (v/v)). The micronics-containing supernatants were kept for filtration and the micronics with the pellets were kept for further starch and total protein analysis.

Filtration was also robotised (Microlab STARlet, Hamilton, Bienne, Switzerland) and allowed for the transfer of the supernatants onto a filtration 96-well sterile clear plate (MSGVS2210, 0.22 μM, Hydrophil. Low Protein Binding Durapore, Millipore, Molsheim, France) according to the supplier's instruction. Filtrates were subsequently collected into a new micronic tube. Finally, quality control (QC) samples were prepared by robotically pipetting 15 μL of each sample into a single tube that was mixed afterwards (Microlab STARlet, Hamilton, Bienne, Switzerland). Each rack was supplemented with a micronic tube containing the QC mix. The QC sample was replicated six times along the project run.

#### *4.6. Targeted Biochemical Phenotyping*

Targeted analyses of sucrose, fructose, glucose, starch, fructose-6-P, glucose-6-P, glutamate, malate, fumarate, total soluble proteins and total polyphenols were conducted on the HiTMe plateau at *Bordeaux Metabolome Facility*. Measurements were based on coupled enzyme assays as described previously [25,26], except for total soluble proteins that were evaluated via Bradford assay [47], and total phenols that were measured colorimetrically using a redox reaction with Folin–Ciocalteu reagent and gallic acid as the standard [48].

#### *4.7. Untargeted Metabolic Profiling*

Untargeted metabolic profiling by UHPLC-LTQ-Orbitrap mass spectrometry (LCMS) was performed using an Ultimate 3000 ultra-high-pressure liquid chromatography (UHPLC) system coupled to an LTQ-Orbitrap Elite mass spectrometer interfaced with an electrospray (ESI) ionisation source (ThermoScientific, Bremen, Germany). The system was controlled by Thermo XCalibur v.3.0.63 software. Chromatographic separation was achieved at a flow rate of 350 μL/min using a GEMINI UHPLC C18 column (150 × 2 mm, 3 μm, Le Pecq, Phenomenex, France) coupled to a C18 SecurityGuard GEMINI pre-column (4 × 2 mm, 3 μm, Le Pecq, Phenomenex, France). The column was maintained at 35 ◦C and the injection volume was 5 μL. The mobile phase consisted of solvent A (0.05 % (v/v) formic acid in water) and solvent B (acetonitrile) with the following gradient: 0–0.5 min 3% B, 0.5–1 min 3% B, 1–9 min 50% B, 9–13 min 100% B, 13–14 min 100% B, 14–14.5 min 3% B, 14.5–18 min 3% B. Ionisation of samples was performed in both negative and positive mode with the following parameters: ESI- (Heater temp: 300 ◦C, Sheath Gas Flow Rate: 45 (arb), Aux Gas Flow Rate: 15 (arb), Sweep Gas Flow Rate: 10 (arb), I Spray Voltage: 2.5 kV, Capillary Temp: 300 ◦C, S-Lens RF Level: 60%), and ESI<sup>+</sup> (Heater temp: 300 ◦C, Sheath Gas Flow Rate: 60 (arb), Aux Gas Flow Rate: 20 (arb), Sweep Gas Flow Rate: 10 (arb), I Spray Voltage: 3.2 kV, Capillary Temp: 300 ◦C, S-Lens RF Level: 55%). MS full scan detection of ions was operated by FTMS (50–1500 Da) at a resolution of 240,000. Prior to analyses, the LTQ-Orbitrap was calibrated by infusing a solution of the calibration dependent of the ionisation mode (Pierce© ESI Negative Ion Calibration Solution (ref: 88324); Pierce LTQ Velos ESI Positive Ion Calibration solution (ref: 88323). The injection sequence started with three blank extracts, then three QC samples, then one blank extract, and each group of samples was subsequently injected, followed by a blank extract. Another two QC samples were injected throughout the analysis. In total, six QC samples and 16 blank extracts were injected to correct for mass spectrometer signal drift, and to filter out variables detected in blanks, respectively.

#### *4.8. Processing and Statistical Analysis of Metabolomic Datasets*

Processing of raw LCMS data using XCMS in R (v 3.6.1) [49] yielded 10,875 detected RT-*m*/*z* pairs for ESI<sup>+</sup> and 5,796 for ESI- . After data-cleaning (blank check, ΔRT < 60 s, Δ*m*/*<sup>z</sup>* < 0.015 Da, CV QC < 30%), 6887 variables were retained for further chemiometrics. Both untargeted and targeted metabolomic data were first normalised by median normalisation, cube-root transformation and Pareto scaling using MetaboAnalyst v.3 [50] before applying multivariate and univariate statistical analyses [51]. The normalised dataset is available as Supplemental Material 1. PCA and PLS-DA were performed with MetaboAnalyst v.3 providing satisfactory validation parameters of the multivariate models (R2 > 0.87 and Q2 > 0.35). PC coordinates for metabolomic features that are responsible for PC1 and PC2 are presented in Supplemental Material 2. Univariate statistical methods were performed using MeV v.4.9.0. [52] at *P* < 0.05 for two-factor ANOVA and *P* < 0.01 for binary comparisons by *t*-tests. In

addition, MarVis v 1 was used to confirm the statistically significant variation om the priming markers through a Kruskal–Wallis test at *P* < 0.05 with correction for false discovery rate [53,54]. Putative annotation of such markers was performed by screening the detected exact *m*/*z* against multiple online databases, including METLIN chemical database (https://metlin.scripps.edu/) [55] and KNApSAcK (http://kanaya.naist.jp/KNApSAcK/) [56]. The resulting predicted pathways were checked using the PubChem database (https://pubchem.ncbi.nlm.nih.gov/).

#### *4.9. Top-down Modelling Approach*

Generalised linear models were constructed in R (v 3.6.1) using the *glmnet* package (v 3.0-2) [30] in order to identify potential links between detected metabolic markers and resistance to the fruit pathogens. Those models were used to predict resistance values based on the detected metabolic markers. Cross-validation was applied by randomly dividing the datasets into two parts: 80 % of the individuals were used to construct the models and 20 % to check for the quality of the prediction. The quality of the models was assessed based on the mean square error between real and predicted values. To cope with this randomisation, 500 models were constructed for each measurement. Generalised linear models contain a penalisation value, allowing less informative variables to be discarded as this value increases (1000 values were tested for each of the 500 models), hence variables occurring the most in the models can be seen as the most stable predictors of resistance to biotic challenges. Given the high number of metabolic variables and the relatively small set of plants, 500 randomly generated resistance datasets were created to estimate the chances of predicting random values. A Student's *t*-test was used to compare the quality of predictions of real and random resistances.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2218-1989/10/3/96/s1, Figure S1. Inoculation-responsive central metabolites. Major compounds involved in central metabolism statistically responded to the inoculation factor of a two-factor ANOVA (P given into brackets). Bar charts indicate means of normalised intensities of four independent bioreplicates (n = 4; ± SEM). Bot: *Botrytis cinerea*, Phy: *Phytophthora infestans*, Pst: *Pseudomonas syringae* pv. *tomato*, B: BABA-treated plants, W: water-treated plants, M: mock-inoculated fruit, P: pathogen-inoculated fruit. Figure S2. Partial least square discriminant analysis for each pathosystem. PLS-DA score plots (n = 4) of 6898 features (6887 LCMS variables + 11 major compounds) between the three different pathosystems. Validation parameters of the PLS model are given in red for each plot. Figure S3. Metabolic markers for BABA primed responses against fruit pathogens. LCMS significant markers that overlap between *Bot* and *Phy* (A), *Bot* and *Pst* (B) and *Phy* and *Pst* (C) in response to BABA priming and after infection (see Table 3). Markers are labelled according to their high-resolution detected *m*/*z*. Bar charts indicate means of normalised intensities (n = 4; ± SEM). See Figure S1 for sample labels. Table S1. Putative annotation of the top 15 predictors. Supplemental Material 1: metabolomic parameters detected y LCMS. 2: P indicating the statistical significance from a Kruskal–Wallis test followed by correction for false discovery rate using the Benjamini–Hochberg method. Supplemental Material 1. Normalised metabolomics dataset combining 6887 *m*/*z* features and 11 major compounds. Supplemental Material 2. Coordinates for Principal Component Analyses ranking the important metabolomic variables.

**Author Contributions:** Conceptualization, E.L. and P.P.; Data curation, E.L., A.F., C.C. (Cédric Cassan) and P.P.; Formal analysis, E.L., A.F., C.C. (Cédric Cassan), C.C. (Chloé Chevanne), C.F.K., S.P. and P.P.; Funding acquisition, E.L., Y.G. and P.P.; Investigation, E.L., A.F., C.C. (Cédric Cassan), C.C. (Chloé Chevanne), C.F.K., S.P. and P.P.; Methodology, E.L., A.F., C.C. (Cédric Cassan), S.P. and P.P.; Project administration, E.L. and P.P.; Resources, E.L., S.P., Y.G. and P.P.; Software, S.P.; Supervision, E.L. and P.P.; Validation, E.L., S.P., Y.G. and P.P.; Visualization, E.L., S.P. and P.P.; Writing—original draft, E.L. and P.P.; Writing—review & editing, E.L., S.P., Y.G. and P.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** The authors are grateful for financial support from MetaboHUB (ANR-11-INBS-0010) and PHENOME (ANR-11-INBS-0012) projects to INRAE, and for the BBSRC Future Leader Fellowship BB/P00556X/1 and BB/P00556X/2 to EL.

**Acknowledgments:** The authors thank Sam Wilkinson for his comments on the project. *P. infestans* isolate was provided by Steve Whisson (The James Hutton Institute).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

1. Donatelli, M.; Magarey, R.D.; Bregaglio, S.; Willocquet, L.; Whish, J.P.M.; Savary, S. Modelling the impacts of pests and diseases on agricultural systems. *Agric. Syst.* **2017**, *155*, 213–224. [CrossRef] [PubMed]


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Identification of Bioactive Phytochemicals in Mulberries**

#### **Gilda D'Urso 1, Jurriaan J. Mes 2, Paola Montoro 1, Robert D. Hall 3,4 and Ric C.H. de Vos 3,\***


Received: 29 November 2019; Accepted: 18 December 2019; Published: 20 December 2019

**Abstract:** Mulberries are consumed either freshly or as processed fruits and are traditionally used to tackle several diseases, especially type II diabetes. Here, we investigated the metabolite compositions of ripe fruits of both white (*Morus alba*) and black (*Morus nigra*) mulberries, using reversed-phase HPLC coupled to high resolution mass spectrometry (LC-MS), and related these to their in vitro antioxidant and α-glucosidase inhibitory activities. Based on accurate masses, fragmentation data, UV/Vis light absorbance spectra and retention times, 35 metabolites, mainly comprising phenolic compounds and amino sugar acids, were identified. While the antioxidant activity was highest in *M. nigra*, the α-glucosidase inhibitory activities were similar between species. Both bioactivities were mostly resistant to in vitro gastrointestinal digestion. To identify the bioactive compounds, we combined LC-MS with 96-well-format fractionation followed by testing the individual fractions for α-glucosidase inhibition, while compounds responsible for the antioxidant activity were identified using HPLC with an online antioxidant detection system. We thus determined iminosugars and phenolic compounds in both *M. alba* and *M. nigra*, and anthocyanins in *M. nigra* as being the key α-glucosidase inhibitors, while anthocyanins in *M. nigra* and both phenylpropanoids and flavonols in *M. alba* were identified as key antioxidants in their ripe berries.

**Keywords:** mulberry; high resolution mass spectrometry; antioxidant activity; in vitro gastrointestinal digestion; α-glucosidase inhibitory activity

#### **1. Introduction**

Mulberry belongs to the genus *Morus*, plant family Moraceae, which comprises 24 different species and one subspecies, with at least 100 varieties [1]. The most commonly known *Morus* species are *Morus alba* (white mulberry), *Morus nigra* (black mulberry) and *Morus rubra* (red mulberry.) [2]. They are deciduous trees originating from China and Japan, and have spread into America and Europe for silkworm breeding. Globally, the main use of mulberry trees is to produce leaves as feed for cultivating silkworms, but in various regions, they are also much appreciated for their fruit, which can be consumed both fresh and as an ingredient in processed food products [3].

Mulberry species have been used in Traditional Chinese Medicines (TCM) for the treatment of several diseases, especially diabetes mellitus type II; they contain specific molecules, i.e., iminosugars or iminocyclitols, which are low molecular weight carbohydrates in which the endocyclic oxygen atom has been replaced by a nitrogen atom. These compounds are known to be able to inhibit the enzyme α-glucosidase which is present in the brush border of the human intestine [4–6]. Inhibition of α-glucosidase leads to a decreased rate of glucose absorption thus resulting in a lower postprandial blood glucose level. Inhibitors of α-glucosidase can prevent the development of diabetes in individuals with impaired glucose tolerance and/or impaired fasting of blood glucose [7].

Mulberries are also a good nutritional source of a variety of phenolic compounds, like flavonols and phenolic acids, as well as coloured anthocyanins in the case of black and red mulberry fruits [8–10]. Phenolic compounds are the subject of increasing scientific interest; they are natural antioxidants in plant-derived foods and food products and their intake is frequently related to human health. Many of the bioactivities ascribed to mulberries, such as antioxidant action, hypolipidemic effect and macrophage activating effect, have also been linked to their phenolic compound composition [11–14].

The present study aimed to determine differences in chemical composition between the ripe fruits of *M. alba* and *M. nigra,* which are the major species growing in Italy, by metabolomic analysis and to identify the bioactive compounds responsible for their antioxidant and α-glucosidase inhibitory bioactivities. To this purpose, we used I) HPLC-PDA with an online, post-column antioxidant detection system, and II) HPLC-PDA- HR Orbitrap FTMS with on-line fractionation into 96-wells plates followed by off line in vitro α-glucosidase inhibitory activity testing of the contents of the individual wells. A series of compounds present in mulberry fruits as well as major bioactives were further characterized using MSn fragmentation. In addition, we subjected mulberries to an in vitro gastrointestinal digestion system in order to investigate potential effects of digestion on the observed bioactivities upon consumption of these berries.

#### **2. Results**

#### *2.1. Identification of Phytochemicals in Morus using LC-PDA-Orbitrap FTMS*

To identify the metabolites present in white and black mulberry fruits, HPLC-PDA-Orbitrap FTMS analysis was performed on their aqueous methanol extracts, thus generating both an LC-PDA and an LC-MS profile per extract (Figures S1 and S2, respectively). Based on the exact mass of their molecular [M+H]<sup>+</sup> ion masses, their MS<sup>n</sup> fragmentation patterns and their UV/Vis absorbance spectra, we putatively identified 35 compounds in the fruits (Table 1).


**Table 1.** Metabolites manually identified in *Morus alba* and *Morus nigra* using accurate mass LC-MSn in positive ESI mode. L. I.: level of identification according

 to

In the fruits of *M. nigra*, four anthocyanins with a characteristic absorbance maximum at around 500–520 nm were identified. Compound **6**, cyanidin hexoside, showed a pseudomolecular ion at *m*/*z* 449.1084, corresponding to the molecular formula C21H21O11<sup>+</sup> that, upon fragmentation, gave one principal product ion at *m*/*z* 287.0546; Compound **8**, cyanidin hexose-deoxyhexose, showed a pseudomolecular ion at *m*/*z* 595.1662, corresponding to the molecular formula C27H31O15<sup>+</sup>, that fragmented into two principal product ions at 449.1058 (C21H21O11<sup>+</sup>: loss of deoxyhexose) and 287.0546 (loss of hexose+deoxyhexose); Compound **9**, pelargonidin hexoside, showed a pseudomolecular ion at *m*/*z* 433.1135, corresponding to the molecular formula C21H21O10<sup>+</sup>, and gave one principal product ion at 271.0596; Compound **11**, pelargonidin hexose-deoxyhexose showed a pseudomolecular ion at *m*/*z* 579.1714 corresponding to molecular formula C27H31O14<sup>+</sup> that, upon fragmentation, gave two principal product ions of *m*/*z* 433.1115 and 271.0596. Thus, anthocyanins detected represented both pelargonidin and cyanidin conjugated with one, two or three C6-sugars. These anthocyanin compounds have previously been identified in *Morus alba* fruits [17] and are responsible for the dark color of black mulberry fruits. These anthocyanins were present in berries of *M. nigra* and were not detectable in those of *M. alba* (Supplemental Table S1), as was expected from their differential colors.

A series of flavonols with different substituents were present in both white and black mulberry fruits (**12**, **14–17**, **19–24**, **26**, **28–30**, **32**, **35**). All these compounds showed the characteristic flavonol absorbance peaks at around 260 nm (resulting from the A-ring) and at around 350 nm (due to B-ring) and producing daughter ions of *m*/*z* 303.0496 (quercetin) or 287.0547 (kaempferol); many of the compounds detected have not previously been described for mulberry fruits. For instance, compounds **14**, **19**, **22** show the same pseudomolecular ion [M+H]<sup>+</sup> at *m*/*z* 773.2135 with the same fragments of *m*/*z* 303.0496, 465.0995 and 611.1565, corresponding to the loss of two hexose and one deoxyhexose moiety, but with different RTs. These compounds were thus identified as different isomers of quercetin-hexose-hexose-deoxyhexose. A similar fragmentation pattern has been reported for a quercetin-trisaccharide in tomato fruit [19]. Specifically, compound **14** was confirmed as quercetin-3-O-rutinoside-7-O-glucoside based on the retention time of the reference compound reported in tomato fruit database [19].

Compounds **17**, **23**, and **24** likewise showed a similar pseudomolecular ion [M+H]<sup>+</sup> at *m*/*z* 757.2192 with the same fragments of *m*/*z* 287.0547, 449.1065 and 611.1576, corresponding to the loss of two hexoses and one deoxyhexose moiety, but with different RTs. These were thus identified as kaempferol-hexose-hexose-deoxyhexose isomers. The fragmentation pattern of these compounds agrees with known kaempferol glycosides in tomato [19]. Moreover, compound **17** was confirmed as kaempferol-3-O-rutinoside-7-O-glucoside based on the retention time of the reference compound reported in tomato fruit database [19].

Compound **16** showed a pseudomolecular ion at *m*/*z* 713.1544 that upon fragmentation, gave three principal product ions at *m*/*z* 303.0496, 463.1021 and 551.1015 corresponding to the loss of two hexose moieties and one malonyl moiety; this compound was thus tentatively identified as quercetin hexose-malonyl-hexoside. This fragmentation pattern is consistent with that reported for a quercetin malonyl glucoside in lettuce [21]. Compound **21** showed a pseudomolecular ion at *m*/*z* 697.1597 which gave three principal product ions at *m*/*z* 287.0545, 449.1065 and 535.1076, corresponding to the loss of two hexose and one malonyl moiety; this compound was identified as kaempferol hexose-malonyl-hexoside. This compound showed a similar fragmentation pattern as reported in Cycorium intibus [24]. Thus, similar flavonol conjugates consisting of both quercetin and kaempferol esterified with one to three C6-sugars, or one or two sugars with one malonyl group, were present in both white and black mulberries.

Compound **20** showed a pseudomolecular ion at *m*/*z* 451.1235 that gave one principal product ion at *m*/*z* 289.0703, corresponding to the loss of a hexose moiety; this compound was tentatively identified as dihydrokaempferol-hexoside. A similar fragmentation pattern was reported for dihydrokaempferolhexoside in raspberry [23].

Three N-containing sugars, i.e., compounds (**1**) 1-deoxynojirimicin, (**2**) N-nonil deoxynojirimicin and (**3**) fagomine, were found in fruits of both mulberry species. These compounds have previously been reported for leaves of *M. alba* [16] and are well known for inhibiting the enzyme α-glucosidase and consequently, might contribute to an antihyperglycemic effect [4–6].

Four piperidine alkaloids, morusimic acids B, C and E (compounds **7**, **25**, **27**) were also identified based on their exact molecular mass and fragmentation; these compounds have previously been reported in fruits of *M. alba* from Turkey [18].

Both *M. alba* and *M. nigra* fruits also contained caffeoylquinic acids monomers (**5**, **10**, **13**) as well as dimers (**31**, **33**, **34**). All three caffeoylquinic acid isomers (**5**, **10**, **13**) showed a pseudomolecular ion at 355.1024 that, upon fragmentation, gave the same daughter ion at *m*/*z* 163.0386, corresponding to the loss of their quinic acid moiety. These compounds have also been reported in *M. alba* fruits from Serbia [17]. The three dicaffeoylquinic acid isomers (**31**, **33**, **34**) showed a pseudomolecular ion at 517.1341 that produced the same MS/MS base peak at 163.0387. These compounds have previously been reported in leaves of *M. alba* [27].

Compound **4** showed a pseudomolecular ion at *m*/*z* 182.0817, that fragmented into two principal product ions at 165.0544 and 136.0755. It was putatively identified as the alkaloid 2-formyl-1Hpyrrole-1-butanoic acid, previously reported in *M. alba* fruits by Kim et al. [11].

#### *2.2. Global Metabolome Di*ff*erences between Morus Alba and Morus Nigra Fruits*

Ripe fruits of *Morus alba* and *Morus nigra* were collected from trees growing at various locations in the Campania Region (Italy) and subjected to untargeted LCMS-based metabolite profiling. Subsequent unbiased data processing generated a dataset with the relative intensities and in-source mass spectra of 361 putative metabolites in the samples (Supplemental Table S1; note that this metabolite list misses some of the manually identified compounds described in Table 1, indicating that one or more parameter settings in the untargeted data processing workflow appears suboptimal for these specific compounds). An unsupervised multivariate statistical method, Principal Components Analysis (PCA), was subsequently applied to the entire metabolite dataset resulting in a clear differentiation of *M. alba* and *M. nigra* fruit samples (Figure S3). Among the most significantly (*p* < 0.05) differing metabolites were anthocyanins (Supplemental Table S1), as was expected from the differential fruit colours of both species. In addition, it was possible to identify two other flavonoids only detectable in *M. nigra*, namely dihydroquercetin hexoside and dihydrokaempferol hexoside. In fact, an important step for the biosynthesis of anthocyanidins is the reduction of dihydroflavonols catalysed by the enzyme DFR (dihydroflavonol 4-reductase) converting dihydroquercetin and dihydrokaempferol into colorless leucoanthocyanidins, which are further converted by the enzyme anthocyanin synthase (ANS) into cyanidin and pelargonidin, respectively [29], thereby providing the fruit colour in *M. nigra*. Several flavonol conjugates, including quercetin glycosides **19** and **26** and kaempferol glycoside **29** (Table 1), were also significantly (*p* < 0.05) higher in *M. nigra* fruit, while the mono- and di-caffeoyl quinic acids (phenylpropanoids) were not differential between both fruit species (Supplemental Table S1). These data suggest that *M. nigra* fruits exhibit a higher activity of the general flavonoid pathway than *M. alba* fruit. The alkaloids identified did not significantly differ between the *M. nigra* and *M. alba* fruit samples analyzed (Supplemental Table S1).

#### *2.3.* α*-Glucosidase Inhibitory Activity and E*ff*ect of In Vitro Gastrointestinal Digestion*

The α-glucosidase inhibitory activity of the black and white mulberries was firstly evaluated using the crude aqueous-methanol extracts of the fresh fruits. The extraction solvent was evaporated by freeze-drying and the metabolites re-dissolved in MQ-water. These water extracts were then tested for inhibiting α-glucosidase enzyme activity, monitored through the increase in the pNP product, detected at 412 nm, using 96-wells plates kept at 30 ◦C; the α-glucosidase inhibitor acarbose was used as a positive control (Figure 1b). Both mulberry extracts showed a marked and similar α-glucosidase inhibitory activity as compared to the water blank (Figure 1a).

**Figure 1.** α-glucosidase inhibitory activity of mulberry methanol extracts. The Y axis represents the α-glucosidase activity (increase in 415 nm absorbance per minute) and the X axis the sample type tested. (**a**) inhibitory activity of water extracts of *Morus alba* and *Morus nigra* compared to the negative control (water). (**b**) enzyme activity inhibition by acarbose (positive control) at increasing concentrations (mM) in the assay. Data represent means and standard deviations (*n* = 3 assays).

The *M. nigra* fruit showed an IC50 value of 0.75 ± 0.004 mg/g DW (*n* = 3), while that for *M. alba* fruit was 0.93 ± 0.003 mg/g DW (*n* = 3). In comparison, the IC50 value of acarbose was 13.83 ± 0.02 mg/g.

A simulated gastrointestinal digestion was then applied to estimate the effect of consumption and digestion on the α-glucosidase inhibitory activity of mulberry fruits (Figure 2). For both fruit types, the bioactivity measured in the original fruit extracts was partially lost upon this in vitro digestion (GI samples compared to MN and MA samples). Gastric digestion (PG) resulted in a slight decrease in bioactivity in *M. nigra* only. The control incubation consisting of water instead of fruit extract in the digestion test (DC samples) showed a slight inhibition of the α-glucosidase activity as compared to the negative control (NC of undigested fruits: water instead of both fruit extract and digestion enzymes and buffer): a decrease of 0.04 enzyme units. Taking this inhibiting effect of the digestion conditions on α-glucosidase into account, it was calculated that the simulated gastrointestinal digestion resulted in an overall reduction in α-glucosidase inhibitory activity of about 50% (a decrease of about 0.055 units from 0.13 in DC to 0.075–0.08 in GI samples, compared to a decrease of about 0.115, i.e., from 0.17 units in NC to about 0.055 in original MN and MA extracts).

**Figure 2.** α-glucosidase inhibitory activity after in vitro gastrointestinal digestion. Inhibition activity of original *Morus nigra* (MN) and *Morus alba* (MA) fruit extracts, and after theirin vitro stomach (Post Gastric, PG) digestion and in vitro gastrointestinal (GI) digestion. DC: digestion control, representing the digestion process, including all enzymes, without plant material; NC: negative control (NC), representing only water. Data represent means values and standard deviations (*n* = 3 measurements).

#### *2.4. LCMS Combined with 96-Well Format Fractionation*

In order to pinpoint those compounds in *Morus* fruits that are responsible for the observed α-glucosidase inhibitory activity, we subsequently used HPLC separation combined with both 96-well plate fractionation and Orbitrap FTMS detection. Injection, fractionation and FTMS analyses of the *M. alba* and *M. nigra* crude extracts, as used in the α-glucosidase inhibition assay, were performed in triplicate; a water blank was injected as a control. The fractionation plates were subsequently dried under a gentle N2 flow at 30 ◦C, the dried well contents re-dissolved in water and tested for α-glucosidase inhibitory activity. Compounds present in bioactive wells were then further characterized from their corresponding UV/Vis spectra and FT-MS data.

The results of the α-glucosidase inhibitory activity of individual wells are shown in Figure 3A,B for *M. alba* and *M. nigra*, respectively. Based on the average enzyme activity measured in the wells of the water control sample, we set a threshold value at 0.17 absorbance units per minute, below which we considered a sample well to possess α-glucosidase inhibitory bioactivity.

The compounds identified in the active wells of both fruit types were the amino sugar acids **1–3** and the flavonoids **15**–**17**, **19**–**20** and **32**. Two anthocyanins, **6** and **9**, only present in *M. nigra* (see Table 1), also showed bioactivity. In addition, other fractions clearly showing α-glucosidase inhibitory were detected e.g., between 34.7 and 35.2 min in *M. alba*, although we have yet been unable to pinpoint and identify the specific bioactive compound(s) (Table 2).

**Figure 3.** α-glucosidase inhibitory activity of 96-well LC-MS fractions of (**A**) *M. alba* and (**B**) *M. nigra* extracts. The Y axis shows the enzyme activity and the X axis the retention time corresponding to the LC-MS fraction. The vertical line at an enzyme activity of 0.17 indicates the average value in the water control. The wells considered bioactive are the ones below an enzyme activity value of 0.15.



*2.5. Total Antioxidant Activity and HPLC with Online Antioxidant Detection*

The total antioxidant activity of the mulberry fruits was compared between other fruits well known for their antioxidant activity: cultivated strawberry (*Fragaria* × *ananassa*) and wild strawberry (*Fragaria vesca*). This antioxidant assay (Table 3) indicated that the aqueous-methanol extract of *M. nigra* is slightly more active than that of *M. alba*; in fact the mulberry fruits showed about the same activity as strawberry, which is among the fruit species with the highest antioxidant capacity [30].



Subsequently, a HPLC-PDA system coupled to online ABTS·<sup>+</sup> cation radical reaction and detection [31] was used to determine the relative contribution of each individual component to the total antioxidant activity (Figure 4). Several antioxidant components could be identified by comparison of their retention times and absorption spectra with those of the LC-PDA-FTMS/MS analysis using the same chromatographic conditions. According to this online antioxidant assay, the key antioxidants in *M. nigra* corresponded to anthocyanins, in particular, compounds **6** and **9**. The other compounds responsible for antioxidant activity in both *M. alba* and *M. nigra* were caffeoylquinic acids, like compounds **10** and **13**, and flavonols like compounds **15**, **26**, **28**, **32** (see Table 1).

**Figure 4.** Antioxidant activity. Overlay of representative antioxidant chromatograms of fruit of *Morus alba* (in blue) and *Morus nigra* (in black). Antioxidant profiles of fruit extracts were determined online, by a post column reaction with ABTS·<sup>+</sup> cation radicals after HPLC separation and PDA detection of compounds. The ABTS-radicals remaining after post-column reaction were recorded at 600 nm: negative peaks thus indicate antioxidant activity. The numbers refer to the main peaks identified (see Table 1): **6** cyanidin hexoside, **9** pelargonidin hexoside, **10** and **13** caffeoylquinic acid isomers, **15** dihydroquercetin hexoside, **26** quercetin hexose deoxyhexose, **28** quercetin hexoside, and **32** kaempferol hexoside.

#### **3. Discussion**

In the present study, we compared ripe fruits of *Morus alba* and *Morus nigra* for their metabolite composition in relation to their potential relevant bioactivities upon consumption, i.e., α-glucosidase inhibiting and antioxidative activities. Using HPLC-PDA-Orbitrap FTMS analysis of aqueous-methanol extracts, we were able to detect a large series of compounds and identified a number of metabolites, previously reported for mulberry or other fruit species, as well as new compounds being present in either

or both *M. alba* and *M. nigra*. Fruit of both species exhibited a marked α-glucosidase inhibiting activity in vitro, an indication of their potential beneficial effect with regard to type II diabetes. Moreover, we showed that this α-glucosidase inhibiting activity was partially resistant to simulated gastric and intestinal digestion. Anthocyanins appear among the potential bioactive compounds in *M. nigra* fruit (Figure 3) and the general instability of anthocyanins at the alkaline conditions of gastrointestinal digestion [32] may at least partly explain the loss of α-glucosidase inhibitory activity in *M. nigra* fruits. When calculating the α-glucosidase inhibiting activity of mulberries in units of acarbose, a well known type II diabetes drug based on its α-glucosidase inhibiting activity (https://www.drugs.com/pro/precose. html), our data suggest that consumption of about 20–25 g of fresh mulberry fruit corresponds to 50 mg of acarbose, taking into account a 50% bioactivity loss upon digestion. It has been shown that an intake of 100 mg acarbose 3 times a day can significantly reduce type II diabetes risk factors [33]. Thus, a daily consumption of 100–150 g fresh mulberries may exert relevant pharmacological effects with regard to type II diabetes. Using analytical LC-based extract fractionation, it was possible to pinpoint three known iminosugar acids, i.e., [1-deoxynojirimycin (**1**), N-nonil-deoxynojirimycin (**2**) and fagomine (**3**)], and 7 phenolic compounds, including five flavonols [dihydroquercetin hexoside, (**15**) quercetin hexoside malonyl hexoside (**16**), kaempferol-3-O-rutinoside-7-O-glucoside (**17**), quercetin hexose (**28**) and kaempferol hexoside (**32**)] present in both *M. alba* and *M. nigra*, and 2 anthocyanins [cyanidin hexoside (**6**), pelargonidin hexoside (**9**)] only present in *M. nigra*, as the key α-glucosidase inhibitors in mulberry fruits. While compounds **1**, **2**, **3**, **6**, **28** and **32** have already been reported to exert this bioactivity [5,16,34,35], in our study, we were able to detect novel α-glucosidase inhibitory compounds in mulberries. A similar approach, using accurate mass LCMS coupled to 96-well fractionation and bioactivity testing, has recently been used to identify novel compounds in pepper fruits interacting with the human hot-taste receptor [36].

Although it was not yet possible to identify the novel α-glucosidase inhibitory compounds in mulberry fruits on the basis of the observed accurate mass only, this method can well be optimized and adapted for further structural characterization of these bioactives, e.g., by using so-called multistage mass spectrometry at high mass resolution [19], if needed, combined with NMR experiments. For the latter approach, the same bioactive wells from replicate plates may be pooled to get sufficient NMR signals for the de novo identification. Alternatively, bioactive extracts can be re-injected in a LC-MS-SPE set up to collect and concentrate individual LC-MS peaks upon repeated injections; the SPE cartridges containing the active compounds (based on their known accurate mass and LC-retention time) can then be subjected to NMR for structural elucidation [37].

In addition to the α-glucosidase inhibitory activity, the ABTS+-radical based total antioxidant assay indicated significant antioxidant activity present in the same mulberries, comparable to that of strawberries, which are among the fruit species with the highest antioxidant capacity [32]. The higher activity in *M. nigra* compared to *M. alba* fruits is likely due to the presence of anthocyanins, which both provide fruits with their dark color and contribute to antioxidant activity [38]. Indeed, using HPLC with online antioxidant detection [32], we were able to pinpoint anthocyanins as the main phenolic antioxidants in *M. nigra*, while both phenylpropanoids and flavonols were the key phenolic antioxidants in *M. alba*.

This work shows that it is well possible, using analytical scale techniques, to pinpoint the compounds that are key to the well described bioactivities of mulberry fruits, and to validate the value of metabolomics technologies in the phytochemical and bioactivity evaluation of functional foods. However, further studies towards, for example growth conditions, genotypic variation, fruit development and ripening are needed to obtain the best material for preparation of such functional foods with optimal composition of bioactive ingredients or for purification of the bioactive compounds.

#### **4. Materials and Methods**

#### *4.1. Mulberry Materials*

Fruits of *M. alba* and *M. nigra* were manually picked at ripe stage in May 2014 in different areas of the Campania region in Italy, in particular, the geographical locations Solofra (GPS coordinates latitude: 40.8291: longitude: 14.8456), Roccadaspide (GPS coordinates latitude: 40.4253: longitude: 15.1917), Fisciano (GPS coordinates latitude: 40.7728: longitude: 14.7994), San Sossio Baronia (GPS coordinates latitude: 410712: longitude: 15.2005). *Morus alba* fruits (MAF) were collected from 4 locations and coded as MAF-S (collected in Solofra) MAF-big (collected in San Sossio Baronia), MAF-wt (collected in San Sossio Baronia), MAF-R (collected in Roccadaspide); *M. nigra* fruits (MNF) were collected at five locations and coded as MNF-R (collected in Roccadaspide), MNF-U14 (collected in Fisciano), MNF-U13 (collected in Fisciano), MNF-S (collected in Solofra) and MNF-U13 (collected in Fisciano). All samples were botanically identified by Prof. V. De Feo (Department of Pharmacy University of Salerno) and compared with reference materials, then were freeze-dried before being transported to The Netherlands. They were then ground to a fine powder and stored at −80 ◦C until analysis.

For comparing antioxidant activities, fresh frozen mulberries were compared with fresh fruit of strawberry and wild strawberry collected in June 2014 in Campania (Italy).

#### *4.2. Extract Preparation*

The sample extracts used for LC-MS analysis were prepared essentially as described in De Vos et al. [31]: 30 mg of freeze dried samples were extracted with 1200 μL of 75% methanol in MQ water containing 0.1% of formic acid. Mixtures were then sonicated for 15 min, centrifuged at 12,500 g for 10 min and filtered over a 0.45 μm filter (Minisart SRP4, Biotech GmbH, Germany).

For α-glucosidase inhibitory activity testing, for both NanoMate fractionation and HPLC with online antioxidant analysis, 1 mL of supernatant was dried in a speedvac and taken up in 250 μL of water, sonicated and filtered through a 0.45 μm filter (Minisart SRP4, Biotech GmbH, Germany). These concentrated extracts were prepared in 3 independent replicates.

#### *4.3. LC-PDA-Orbitrap FTMS Analysis*

A metabolite analysis was performed using an HPLC (Waters Aquity) coupled to both a photodiode array detector (PDA; Waters) and an LTQ Ion trap-Orbitrap Fourier transformed Mass spectrometer (FTMS; Thermo) hybrid system. A Luna 3 μm C18 150 × 2 mm column (Phenomenex, USA) at 40 ◦C was used to separate the extracted metabolites, with MQ water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid (B) as solvents. A linear gradient from 5% to 35% B in 45 min, at a flow rate of 0.19 mL/min, was used [33]. In order to prevent possible order and batch effects in the LCMS analysis, all samples were analysed in a single series and in random order. Three quality control samples (QCs) prepared from a mix of all samples were included and equally distributed over the study samples, in order to check system stability and estimate overall technical variation.

Electrospray ionization (ESI) in positive mode was used to generate ions from eluting compounds. ESI source parameters were as follows: capillary voltage 43 V; tube lens voltage 120 V; capillary temperature 295 ◦C; Sheath and Auxiliary Gas flow at 40 and 3 (arbitrary units), respectively, Sweep gas 0 n, Spray voltage 5 V. MS spectra were acquired by full range acquisition covering *m*/*z* 104–1350 at a resolution of 60,000 FWHM.

#### *4.4. LCMS data Processing and Multivariate Analysis*

The raw LCMS data files were processed using Metalign software [34] for baseline correction, noise estimation, and ion-wise mass spectral alignment. MSClust software [35] was then used to assemble redundant mass signals derived from the same metabolite, including natural isotopes, adducts and in-source fragments, based on their corresponding retention times and relative abundance patterns across samples. This resulted in the relative intensities of 361 mass peak clusters, each representing a

(reconstructed) putative metabolite, present in at least two samples. These metabolite intensity data were then subjected to a multivariate analysis using GeneMaths XT software version 2.12 (Applied Maths, Belgium). Metabolite intensities were firstly log2-transformed and then mean-centred across samples.

Using multiple online databases, including KNApSAcK (http://kanaya.naist.jp/KNApSAcK/), Dictionary of Natural Products (http://dnp.chemnetbase.com), Metlin (https://metlin.scripps.edu/), HMD (http://www.hmdb.ca), in-house libraries based on standards, as well as the mass spectra information within the clustered mass peaks and from additional LC-MS<sup>n</sup> runs generating accurate mass spectral trees from the top 3 intensity ions every 30 s [19]. Selected metabolites (Table 1) were manually annotated as far as was possible using the mass data and the UV/Vis-absorbance spectral data available.

#### *4.5. In Vitro Simulated Gastrointestinal Digestion*

In vitro digestion was carried out on freeze-dried fruit samples of both *M. alba* and *M. nigra*, following the protocol described by McDougall et al. [36] with slight modification. Release of phytochemicals from fruit was checked by LC-MS at different stages of digestion, i.e., after gastric digestion (post gastric, PG) and gastrointestinal digestion (GI). Both PG and GI samples were stored at −80 ◦C until further analysis. During the process, three different controls were used: (1) plant material without digestion solutions and enzymes, diluted in water using the same ratio used for the samples coming from the digestion process (2) the solutions with all the ingredients for digestion but without plant material (DC: Digestion control), (3) plant material with all the ingredients for digestion but without active enzymes (the enzymes where added at the end of the digestion process, to the cold extract). Both α-glucosidase inhibitory activity and antioxidant activity were investigated for each of these PG and GI samples using the methods described below.

#### *4.6.* α*-Glucosidase Inhibition Assay*

The α-glucosidase assay uses the synthetic substrate p-nitrophenyl-α-D-glucopyranoside (pNPG), which is hydrolyzed by α-glucosidase to release p-nitrophenol (pNP), a color agent that can be monitored at 415 nm. Briefly, 10 μL of extract was combined with 40 μL of 100 mM phosphate buffer (pH 6.8) and 20 μL of α-glucosidase (0.6 units per mL buffer). After mixing and incubation for 5 min at 37 ◦C, 20 μL of a 20 mM pNPG solution in buffer was added to start the reaction. The reaction was monitored in time at 415 nm by a TECAN SpectraFluor microplate reader. Acarbose was used as a positive control, while water was used as a negative control for enzyme inhibition. The enzyme activities were evaluated as increase in the absorbance at 415 nm per minute and the percentage of enzyme inhibition was calculated. Three dilution series of extracts were used for IC50 determination. Dose−response curves and IC50 values were obtained by use of GraphPad Prism (version 6.00.283). The assay was performed with 3 replicates.

#### *4.7. NanoMate LC-Fractionation of Extracts*

The HPLC–PDA–FTMS system was adapted with a chip-based nano-electrospray ionization source/fractionation robot (NanoMate Triversa, Advion BioSciences) coupled between the PDA and the inlet of the Ion Trap/Orbitrap hybrid instrument [19]. In this system, the compounds separated and eluting from the analytical column firstly passed the PDA detector for determining their UV/Vis absorbance spectra and then the eluent was automatically split by a NanoMate LC-fraction collector/injection robot (Advion) into a nanoflow for chip-based ESI nanospray Orbitrap FTMS analysis and the rest for fractionation into microwells with a collection time of 28 sec per well. The sample injection volume was 5 μL. The gradient and flow conditions were the same as described above, with an additional 30 μL/min 100% isopropanol added into the LC flow via a T-junction between the PDA and the NanoMate, in order to improve the solvent composition for generating a stable nano-electrospray. The eluent flow was split by the NanoMate at a ratio of 219.5 μL/min to the fraction collector and 0.5 μL/min to the nano-electrospray source. LC-fractions were collected every 28.2 s (i.e., 100 μL

solvent) into 96-well plates (Twin tec, Eppendorf). After collection, the plates were dried at 30 ◦C under a gentle N2 flow, and then tested for α-glucosidase inhibitory activity as described above (performed in 3 replicates).

#### *4.8. Antioxidant Activity and HPLC Analysis with Online Antioxidant Detection*

The total antioxidant capacity of fruits was analyzed using the ABTS·<sup>+</sup> radical scavenging method, essentially according to Capanoglu et al. [37] with slight modifications. The fruits were collected in June 2014 and the antioxidant activity was tested on basis of fresh weight. 0.5 g of samples (fresh fruit) were extracted in 1.5 mL of methanol (final MeOH concentration about 77%, taking into account a fruit water content of 95%) containing 0.05% of formic acid, sonicated for 15 min and centrifuged at 12,500 g for 15 min, filtered through 0.45 μm (Minisart SRP4, Biotech GmbH, Germany) and then 10 μL of extract was used to test the antioxidant activity. Trolox was used as a reference.

To determine total antioxidant capacity, 10 μL of sample extracts or standard solution was mixed with 90 μL of ABTS-radical working solution (pH 7.4) and after 40 s, the remaining ABTS·<sup>+</sup> radicals were measured at 415 nm using 96-well microplates (Nunc, Roskilde, Denmark) and an Infinite® M200 micro plate reader (Tecan, Gröding, Austria). The analyses were done using 3 replicates and the results were expressed in terms of mg Trolox Equivalent Antioxidant Capacity (TEAC) per g fruit FW. In addition, the contribution of individual antioxidants to the total antioxidant capacity of the crude mulberry extracts was determined using an HPLC-PDA system coupled to post-column on-line antioxidant detection [37,38]. For this, the extracts of *M. alba* and *M. nigra* fruits, also used for the LC-MS analysis, were analyzed using a W600 Waters HPLC system coupled to a Waters 996 PDA detector (240–600 nm) [37,38]. Eluted compounds were allowed to react online for 30 s at 40 ◦C in a buffered solution of ABTS· <sup>+</sup> cation radicals (pH 7.4). Then, the absorption of the remaining ABTS·<sup>+</sup> radicals was monitored at 412 nm by a second detector (Waters 2487, dual-wavelength UV–vis detector). Peak identification was done by comparing PDA-absorbance spectra and retention times of eluting peaks with data taken from the literature and annotations were confirmed by HPLC-FTMS and MS/MS analyses, as described above.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2218-1989/10/1/7/s1, Figure S1: Representative LC-MS chromatograms, recorded in ESI positive ionization mode, of *M. nigra* (upper trace) and *M. alba* (lower trace) fruit extracts; Y-axes are on the same scale (1.5\*E7; base peak intensity in ion counts/sec). Values above peaks indicate retention times (in minutes) and detected *m*/*z* value, Figure S2: Representative LC-PDA chromatograms of aqueous-methanol extracts from ripe fruits of *M. nigra* (A and B) and *M. alba* (C and D). Figures show absorbance at 520 nm (A and C) representing elution profile of anthocyanins, and at 355 nm (B and D) representing mainly flavonoids and phenylpropanoids, Values above peaks indicate retention times (in minutes). Note: intensity scales (Y-axes) are similar for all traces, Figure S3: 3 dimensional PCA plot of 5 *Morus alba* and 4 *M. nigra* fruit samples, harvested from trees spread over region Campania, Italy, based on their variation in 371 metabolites detected by the untargeted LCMS approach. The 3 quality control samples are technical replicates of a mix of samples. The X-axis (PC1) explains 33.2% of the total metabolites variation, the Y-axis (PC2) 18.6% and the Z-axis (PC3) 14.2%, Table S1: Relative intensity of all putative metabolite features (clusterID's) for each of the analyzed mulberry samples, Table S2: Description of column heads, Table S3: MSI Identification level.

**Author Contributions:** Conceptualization, G.D., P.M. and R.C.H.d.V.; methodology, G.D., J.J.M. and R.C.H.d.V.; software, G.D. and R.C.H.d.V.; formal analysis, G.D.; investigation, G.D., P.M. and R.C.H.d.V.; resources, P.M., J.J.M. and R.D.H.; data curation, G.D. and R.C.H.d.V.; writing—original draft preparation, G.D.; writing—review and editing, G.D., J.J.M., P.M., R.D.H. and R.C.H.d.V.; visualization G.D'U and R.C.H.d.V.; supervision, P.M., J.J.M., and R.C.H.d.V.; project administration, G.D., P.M. and R.C.H.d.V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** This research was carried out at the Bioscience department of Wageningen Plant Research of Wageningen UR. Special thanks to Harry Jonker and Bert Schipper for their technical help in HPLC-antioxidant and LC-PDA-FTMS analysis, and to Monic Tomassen (Fresh Food and Chains, Food & Biobased Research, Wageningen UR) for helping with the in vitro gastrointestinal digestions.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


*Metabolites* **2020**, *10*, 7

38. Beekwilder, J.; Jonker, H.; Meesters, P.; Hall, R.D.; Van Der Meer, I.M.; De Vos, C.H.R. Antioxidants in raspberry: On-line analysis links antioxidant activity to a diversity of individual metabolites. *J. Agric. Food Chem.* **2005**, *53*, 3313–3320. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Untargeted and Targeted Metabolomic Profiling of Australian Indigenous Fruits**

#### **Vuanghao Lim 1,2,\*, Sara Ghorbani Gorji 1, Venea Dara Daygon <sup>1</sup> and Melissa Fitzgerald 1,\***


Received: 18 February 2020; Accepted: 11 March 2020; Published: 19 March 2020

**Abstract:** Selected Australian native fruits such as Davidson's plum, finger lime and native pepperberry have been reported to demonstrate potent antioxidant activity. However, comprehensive metabolite profiling of these fruits is limited, therefore the compounds responsible are unknown, and further, the compounds of nutritional value in these native fruits are yet to be described. In this study, untargeted and targeted metabolomics were conducted using the three fruits, together with assays to determine their antioxidant activities. The results demonstrate that targeted free and hydrolysed protein amino acids exhibited high amounts of essential amino acids. Similarly, important minerals like potassium were detected in the fruit samples. In antioxidant activity, Davidson's plum reported the highest activity in ferric reducing power (FRAP), finger lime in antioxidant capacity (ABTS), and native pepperberry in free radical scavenging (DPPH) and phosphomolybdenum assay. The compounds responsible for the antioxidant activity were tentatively identified using untargeted GC×GC-TOFMS and UHPLC-QqQ-TOF-MS/MS metabolomics. A clear discrimination into three clusters of fruits was observed using principal component analysis (PCA) and partial least squares (PLS) analysis. The correlation study identified a number of compounds that provide the antioxidant activities. GC×GC-TOFMS detected potent aroma compounds of limonene, furfural, and 1-R-α-pinene. Based on the untargeted and targeted metabolomics, and antioxidant assays, the nutritional potential of these Australian bush fruits is considerable and supports these indigenous fruits in the nutraceutical industry as well as functional ingredients for the food industry, with such outcomes benefiting Indigenous Australian communities.

**Keywords:** Davidson's plum; finger lime; native pepperberry; antioxidant; amino acids; metabolomics; GC×GC-TOFMS; UHPLC-QqQ-TOF-MS/MS; bush fruit

#### **1. Introduction**

Australia is famous for its rich diversity of native plant foods, which are also known as bush tucker, or bush food. There are about 6500 types of bush foods, and only a handful have been commercialised, and are considered to be worth about \$18–25 million to the Australian economy [1,2]. Among these, there are about 2400 native fruits found in Queensland alone. In the stocktake published by the Australian native food industry, [joint collaboration of the Australian Native Food Industry Limited (ANFIL) and Rural Industries Research and Development Corporation (RIRDC) now AgriFutures Australia], emphasis has focussed on twelve key crops for further development. In the list, several native fruits have been identified, such as Davidson's plum, desert limes, quandong, lemon aspen, riberry, muntries, finger lime, kakadu plum, and native pepperberry [3]. These native edible fruits possess health benefits and can be used in applications such as functional foods and nutraceuticals, contributing to the emerging commercialisation in pharmaceutical industries. In the present work, three fruits were selected for in depth metabolomic profiling based on future potential: Davidson's plum, native pepperberry, and finger lime.

Davidson's plum (*Davidsonia pruriens* F. Muell) belongs to the Davidsoniaceae family, and it grows mostly in north-east Queensland in areas like coastal and upland rainforests (Djirbalngan, Yidinjdji, Djabuganjdji, Kuku-yalanji Nations). There are two other varieties; *D. johnsonii,* which is found in the south-east Queensland and New South Wales (Bundjalung Nation) while *D. jerseyana,* cultivates in the northern New South Wales (Bundjalung and Gumbainggir Nations) [4]. The tree can reach up to 20 m high [5]. The fruit is a purple plum described as tasting intensely sour [6], due to a high amount of acid with very little sugar to counteract it [6]. The fruit is rich in flavonoids, vitamins, minerals and several other important secondary metabolites, such as anthocyanins as well as proanthocyanidins [4]. Few preliminary studies have proclaimed that the extracts of Davidson's plum fruit inhibited in vitro cancer cells, metabolic syndrome enzymes, and contained high antioxidant activity [7]. Belonging to the Rutaceae family, finger lime, *Citrus australasica* var. *sanguinea* is also called Rainforest Pearl, found in the rainforests of Queensland (Bundjalung Nation) and northern part of New South Wales (Gumbainggir Nation) [8,9]. The fruits come in various shapes and sizes, and a range of colours including purple, green, yellow and pink. The native finger lime cultivated in Australia is one of the seven citrus species with 'caviar like' appearance of the fruit pulp. In general, finger lime is rich in vitamins, minerals and terpenes, such as limonene [10]. The bioactivities and phytoconstituents of finger lime are not well-established with reported properties limited to antioxidant and anti-inflammatory activity [9,11]. Native pepperberry [Tasmannia lanceolata (Poir.) A.C. Smith], which belongs to the family of Winteraceae grows at highland areas in Tasmania (Palawa Nation) and southeastern Australia (Boonwurrung, Woiworung, Jaitmatang, Bidwell, Yuin, Ngarigo Nations). As a native shrub tree, the fruit is black (dark purple) in colour and contains many tiny black seeds [12]. Native pepperberry has been ethnopharmacologically used as an ailment to treat stomach discomfort, and as an antifungal for skin diseases by Indigenous people, and it has scientifically proven to inhibit in vitro platelet aggregation, microbial activity, as well as antioxidant capacity [13–15]. Sesquiterpene polygodial, a major phytoconstituent which is mostly found in the oil of native pepperberry makes the spicy and pungent flavour [14]. Additionally, other major secondary metabolites like guaiol, calamenene, hexacosanal, drimenol and linalool were also reported to possess antimicrobial properties [12,16].

In order to support the medical, nutritional and food significance of these bushfoods, it is important to identify the functional compounds and understand their activities. In this regard, the identification of phytoconstituents from various analyses is emphasised, considering the synergistic or antagonistic activity of the metabolite-metabolite interactions for certain bioactivities. Plant metabolomics provides the tools necessary to analyse and potentially identify all the metabolites that possess bioactive properties. Analytical platforms, such as ultra-high performance liquid chromatography mass spectrometry (UHPLC-MS), have been widely used in plant science for metabolomics applications to identify and quantify compounds [17,18]. The comprehensive profiling and metabolomics studies of bush fruits are important in reaping better insight into commercial viability of these fruits. Antioxidant activity has been reported for the fruits [7,11,14,15,19], however the correlation of the activity with bioactive compounds through metabolomics approaches has not been reported. Therefore, this study was conducted to bridge the gap for the discovery of antioxidant-based active metabolites together with comprehensive profiling of Davidson's plum, finger lime and native pepperberry. In the current study, profiling of the fruits was conducted using targeted applications such as mineral analysis and amino acid analysis, and untargeted applications for semi-polar and aromatic compounds. Multivariate data analysis (MVDA), both supervised and un-supervised was applied to assess the association and discrimination of the compounds in the fruit samples. Data comparison of antioxidant scavenging activities was carried out together with total phenolic, flavonoids and flavonols content. The correlation between identified compounds and antioxidant activity was then conducted to identify potential bioactive markers in the fruit samples, as a mean of comprehensive findings for the fruits to be used in the nutraceutical industry.

#### **2. Results**

#### *2.1. Untargeted Metabolic Profiling Using GC*×*GC-TOFMS and UHPLC-QqQ-TOF-MS*/*MS*

The aromatic compounds in the powdered fruits were analysed by two-dimensional gas chromatography time of fight mass spectrometry (GC × GC-TOFMS), and approximately 616 peaks were found from each sample. The samples were screened using ChromaTOF software for the presence of common components. A total of 604 compounds were tentatively identified in the samples of fruits based on the library match searching data in NIST 11 v 2.0 and our in-house library. The metabolite profiles were compared in the PCA scores plots by submitting the combined Davidson's plum, finger lime and native pepperberry triplicate samples as shown in Figure 1a. Clustering of the scores was observed in three groups based on the three types of fruits. Two PCs were found with the greatest eigenvalues recorded at 55.7% and 25.7% of the total variance. The discrimination of the samples in three clusters indicates the aroma components in these fruits were different.

**Figure 1.** The principal component analysis of Davidson's plum (DP), finger lime (FL) and native pepperberry (NP) using (**a**) GC×GX-TOFMS; (**b**) UHPLC-QqQ-TOF-MS/MS; and (**c**) loading score plot representing compounds using UHPLC-QqQ-TOF-MS/MS. Compounds are coloured to indicate the relative density of peak areas.

The untargeted metabolic profile of the fruit samples obtained from the negative mode of ultrahigh-performance liquid chromatography triple quadrupole-time-of-flight mass spectrometry/mass spectrometry (UHPLC-QqQ-TOF-MS/MS) provided 1166 compounds (after data processing), with 542 annotated compounds tentatively identified by MS1 and MS2 matchings; 478 by MS1 and 64 by MS2 matching only. The PCA analysis dataset of the 3 types of fruits provided distinct differences of clustering patterns as shown in Figure 1b. Noticeably, PC1 indicated 56.1% deviation between the 3 fruits with finger lime clustering on the right side, whereas native pepperberry was observed towards the negative quadrant of PC1. The PCA was further analysed in loading plot to observe the discrimination of the compounds (Figure 1c).

#### *2.2. Antioxidant Activity Using DPPH, ABTS, FRAP and Phosphomolybdenum Assays*

The fruit samples were subjected to in vitro antioxidant activity using different spectrophotometric assays as shown in Table 1. From the table, finger lime exhibited the strongest capability in ABTS, however the lowest in DPPH assay despite the same radical scavenging activity. In reducing power ability, Davidson's plum reported the highest, while finger lime and native pepperberry did not show any significant difference. Phosphomolybdate method was later investigated for total antioxidant activity and expressed as gallic acid equivalent (μmol/gDW). The activity was found to decrease in the order of native pepperberry > finger lime > Davidson's plum.

#### *2.3. Total Phenolic Content (TPC), Total Flavonoid Content (TFC) and Total Flavonol Content (TFlC)*

Further investigation was conducted for TPC, TFC and TFlC. Native pepperberry exhibited a significantly higher amount of TPC, TFC and TFlC compared to finger lime, as shown in Table 1. Finger lime showed the lowest gallic acid and quercetin equivalent for all the assays. These assays showed a similar trend for the samples, with native pepperberry the highest, followed by Davidson's plum, and then finger lime.

#### *2.4. Correlation Between Antioxidant Activity And Compounds in The GCMS Dataset*

A supervised multivariate data, partial least squares (PLS) was applied to fathom the relationship between antioxidant activity and the fruit extracts. Based on Figure S1 in Supplementary Materials, the model showed cumulative R2X = 0.864 and Q<sup>2</sup> = 0.886, indicating good fitness and high predictability (>0.5). In this analysis, the X-variables denote the aroma compounds and Y-variables are the antioxdant activity. Both PC1 and PC2 explained 86.4% of the variation in the aroma compounds showing discrimination in the compounds for antioxidant activity, with Y-variables recorded at 98%. In order to determine the aroma compounds that may contribute to the antioxidant activity, variable importance in projection (VIP) was carried out. The potential bioactive aroma compounds were chosen from the variables with VIP values of greater than 2.5. A total of 21 aroma compounds were sorted and the details are tabulated in Table 2. The identified aroma compounds were categorised into 6 different groups, namely, terpenes, aldehydes, terpenoids, furans, isoprenoids, and alkanes.

**Table 1.** In vitro antioxidant activity of the 3 fruit samples. Data is expressed in dry weight, DW and presented as mean values ± standard deviation (*n* = 3). Different superscript letters within each column indicate significant (*p* < 0.01) difference between samples.


ABTS: 2,2 -azino-bis(3-ethylbenzothiazoline-6-sulfonic acid); DPPH: 2,2-diphenyl-1-picrylhydrazyl; FRAP: Ferric Reducing of Antioxidant Power Assay; TPC: Total Phenolic Content; TFC: Total Flavonoid Content; TFlC: Total Flavonol Content; DP: Davidson's plum; FL: Finger lime; NP: Native pepperberry; GA: Gallic acid equivalent; QTE: Quercetin equivalent.



<sup>a</sup> 1tR: First dimension retention time, 2tR: Second dimension retention time; <sup>b</sup> Mf: Molecular formula; <sup>c</sup> Level: Level of identification based on the guidelines [20]. L1—level 1 identified through authentic chemical standards; L2—putatively identified compounds through library matching.

#### *2.5. Correlation Between Antioxidant Activity And Compounds in The LCMS Dataset*

With regard to the clear variance defined by the PCA (Figure 1b) and antioxidant activity (Table 2), a supervised multivariate data analysis, i.e., partial least squares (PLS) was utilised to organise and distinguish the samples according to their MS dataset. The correlation of the antioxidant activity and identified compounds was conducted by setting ABTS, DPPH, FRAP and phosphomolybdenum assays as Y variables, while the identified compounds were assigned as the X variables (Figure 2). In this experiment, a clear discrimination was achieved between Davidson's plum, finger lime and native pepperberry with good discriminant model indicators, R2Y at 0.968 with the goodness-of-prediction

value, Q<sup>2</sup> at 0.965. The robustness of the PLS model on the antioxidant activity for discriminating the identified compounds was confirmed by the results of permutation tests (Figure S2 in Supplementary Materials). The correlation of identified compounds with antioxidant acitivty showed discrimination of the fruits with scattered Y variables as shown in Figure 2b. Y variables of DPPH and FRAP are seen in the right quadrant of PC1 (X = 47.5%; Y = 55.1%), whereas the ABTS and phosphomolybdenum assays are towards the left. PC2 accounts for 49.5% of X variables and 43.3% in Y variables.

**Figure 2.** (**a**) Partial least square (PLS) score plot derived from UHPLC-QqQ-TOF-MS/MS on 3 fruit sample. (**b**) PLS biplot plots showing correlation between identified compounds with antioxidant activity. X = compounds, Y = antioxidant activity. (**c**) The variable importance in the projection (VIP) values (>1.5) represented by red coloured bars, (1.0–1.5) in green while blue coloured bars are VIP values (<1.0). The coloured compounds in (**b**) are VIP values >1.5.

Compounds that are responsible for the separation in Figure 2b were observed through variable importance in projection (VIP) selection approach. In this model, we set the VIP values at greater than 1.5 and loadings correlation coefficient [p(corr)] values above 0.5 to be significant for sample separation in the PLS model. From the analysis, a total of 44 compounds displayed good VIP values as listed in Table 3, with 7 unknown compounds. From the VIP scores, sugars, flavonoids and terpenes were the most important classes of compounds for antioxidant activity, followed by other phenolic compounds. Next, in order to better understand the distinctive incidence in antioxidant activity, we provide a fold-change analysis on the compounds (LogFC, Table 3). This analysis aims to correlate the significant value changes between the two group means. In this analysis, the fold-change was set at two, and any numbers that surpassed the threshold were considered significant. In particular, finger lime and native pepperberry showed the highest series of fold change with the fold-change distributions ranging from 3 to 10 times higher compared to Davidson's plum and finger lime. Nevertheless, Davidson's plum and native pepperberry only exhibited 5 significant fold-change of putatively identified compounds, ie. isovitexin, quercetagetin, racemosic acid, quercetin 3-[rhamnosyl-(1->2)-alpha-l-arabinopyranoside], and {3-[2-(3-hydroxy-5-methoxyphenyl)ethyl]phenyl}oxidanesulfonic acid, with the last compound also exhibited the highest fold-change (17.55).

#### *2.6. Targeted Free And Protein Amino Acid Profilling Using UHPLC-MS*

In total, 11 free amino acids were found in Davidson's plum, 19 in finger lime and 14 in native pepperberry (Table 4). Interestingly, essential amino acid, lysine, exhibited the highest free amino acid for all the three fruit samples with Davidson's plum (74.11%), finger lime (53.74%), and native pepperberry (67.88%). Other amino acids such as isoleucine, cystine and histidine are relatively high among the samples. Most of the free amino acids were detected in finger lime, and included aromatic amino acids, phenylalanine, tyrosine and tryptophan; sulfur-containing amino acids, cysteic acid, taurine and cystine; and non-essential amino acids, arginine, aspartic acid, glutamic acid, glycine, proline, and serine, but not alanine. Trace amounts of some free amino acids (<0.7%) were detected in native pepperberry compared to lysine, and isoleucine.

Analysis of the protein amino acids showed that Davidson's plum exhibited 10 amino acids, and both finger lime and native pepperberry reported all the protein amino acids (Table 4). Similar to the profile of free amino acids, the three fruit samples demonstrated that lysine was the highest amount of hydrolysed protein amino acid, followed by isoleucine indicating a significant difference between the samples for isoleucine. lysine is significantly higher in Davidson's plum compared to both finger lime and native pepperberry, in which the content is insignificant. Different trends of amino acids were observed for the three samples, where the percentage of amino acids in Davidson's plum ranged from 55.53% (lysine) to 0.08% (valine), finger lime, from 45.185% (lysine) to 0.235% (norleucine), and native pepperberry, from 37.07% (lysine) to 0.18% (cysteic acid).

Un-supervised, PCA of the free and hydrolyed protein amino acids was later conducted to observe the association of the amino acids with the fruits (Figure S3 in Supplementary Materials). From the biplots, finger lime (positive PC1) is discriminated from Davidson's plum and native pepperberry (negative PC1) and most of the free amino acids are associated with finger lime. In contrast, hydrolysed protein amino acids showed clustering of finger lime with native pepperberry and separating from Davidson's plum with negative PC1. Almost all the amino acids were associated with finger lime and native pepperberry except lysine, cysteine and isoleucine.


*Metabolites* **2020**

, *10*, 114


**Table 3.** *Cont.*

*Metabolites* **2020** , *10*, 114


*Metabolites* **2020** , *10*, 114

**Table 4.** Amino acid composition

 of Davidson's

 plum (DP), finger lime (FL) and native pepperberry

 (NP) for (**a**) free amino acids; and (**b**) hydrolysed

 protein amino

#### *2.7. Targeted Minerals and Heavy Metals Profilling using ICP-OES*

Here, we targeted 18 minerals and heavy metals for the fruit samples and the analysed data are shown in Figure S4 in Supplementary Materials. The heat map shows the relative concentration of all the minerals and heavy metals in each sample. Minerals (Mn, Na, Fe, Zn) are abundantly distributed in native pepperberry compared to Davidson's plum and finger lime. However, Davidson's plum contained elevated levels of Mg (816.1 ± 7.4 mg/kg) and Al (114.2 ± 1.2 mg/kg) while finger lime with higher levels of Ca (1390.1 ± 35.8 mg/kg) and P (870.63 ± 24.2 mg/kg) than any other elements analysed in the study (Table S1 in Supplementary Materials). From the statistical analysis, 9 elements varied significantly (*p* < 0.05), i.e., Al, Ca, Fe, K, Mg, Mn, Na, P, and Zn as shown in Figure 3 and represented in box and whiskers plot. Interestingly, the relative amount of Al, K, and Mg were highest in Davidson's plum compared to native pepperberry and finger lime. Al, Ca, Mg and P were moderately high in native pepperberry except for potassium. Finger lime exhibited a relatively low amount of most elements, but did show an amount of Ca and P compared to the other fruits.

**Figure 3.** Box and whisker plots of nine elements varied significantly. DP: Davidson's plum, FL: Finger lime, NP: Native pepperberry. The Y-axis of box and whisker plots indicates the amount in mg/kg.

#### **3. Discussion**

Davidson's plum, finger lime and native pepperberry are among Australian native foods that are important to be classified as functional foods due to vast biologically active primary and secondary metabolites, especially terpenes, and flavonoids. Recently, with the inclusion of Indigenous foods in the food industry, natural antioxidants from bush fruits have gained expanding attention due to the growing demand for novel flavours, new functional compounds, and clean labelling. These fruits are rich in polyphenolic compounds, which are prominent natural antioxidants. The identification of aroma compounds using GC×GC-TOFMS provides an important factor in discriminating the clusters of the fruit samples. Correlation between antioxidant activity and aroma compounds exhibited abundance in terpenoid and terpene groups such as limonene, which generally contributes to the fruity smell due to its low odour threshold. Similar with previous studies, limonene was relatively high in finger lime, showing one of the major volatile compounds compared to other terpene groups [2,9,10]. The abundance of terpenes and terpenoids in the fruit samples suggest that they are suitable for processing of the fruits into jam and chutneys or even other food products with pleasant aroma and colour [9].

In the current study, we also conducted a correlation between bioactive compounds with antioxidant activity of the fruits using LCMS metabolomics. Through the comprehensive metabolite profiling, the results of various scavenging activities along with TPC, TFC and TFlC are presented in Table 2. In relation to the reported studies, Davidson's plum exhibited proportionate amount of total phenolic content with damson plum, *Prunus domestica* subsp. *Insititia* L. (124.32 GAE μmol/gDW) cultivated in France [21], but three times higher (38.20 GAE μmol/gDW) compared with the same plum planted in Serbia [22]. In general, bush fruits are rich in phenolics, terpenes and flavonoids, forming the vital class of compounds for scavenging activities in human body [14]. Each antioxidant assay provides different data, therefore four types of antioxidant assays were conducted in order to completely evaluate the efficacy of the powdered fruit extracts. Generally, different fruits exhibited different strengths in their respective antioxidant assays when compared with gallic acid. The antioxidant capacity changes with in vitro assays depending on the affinities of the active compounds present in the fruits [23]. The outcome of this study is also synergistic with the TPC, TFC and TFlC activities with their respective antioxidant activity. Thus, these findings support the antioxidant function of the fruits, which is in good agreement with previous results [2,11,14,24,25].

The relative variability of the compounds in the fruits was measured using MVDA to characterise the tentative identification of peaks that are bio-actively related to antioxidant activity. PCA and PLS were employed with clear discrimination of the score plots among the fruits. The correlation of the compounds with antioxidant activity (DPPH, ABTS, FRAP and phosphomolybdenum) was conducted using a PLS biplot to reveal the dominant compounds that may be responsible for the activities. The search of the compounds that attributed to antioxidant activity was conducted through the VIP values greater than 1.5 (Table 3). Therefore, from the findings of both GC and LC chromatographic analyses, the identified notable compounds were highly possible for the antioxidant activity. Some aroma compounds such as limonene, furfural, α-pinene, terpinen-4-ol, and γ-terpinene and polar compounds like dicaffeoylquinic acid, quercetagetin, and 2-O-acetylrutin have been reported to possess potent antioxidant activity [22–32].

In addition to the secondary compounds identified, the fruits, especially finger lime and native pepperberry, are rich in amino acids. Nevertheless, most of the essential amino acids were present in the samples and can therefore act as a functional food in dietary supplements [33]. Though lysine is well represented in some vegetable species [34,35], it is absent in many plants compared to other amino acids and this is a disadvantage for vegans. According to the Academy of Nutrition and Dietetics (AND), vegetarians are recommended to consume an array of plant-based food that are rich in protein in order to meet nutritional and health requirements [36]. Interestingly, lysine exhibited the highest amount in the fruits, which is suitable for vegetarians as well as those who are allergic to beans and legumes [37].

Minerals and trace elements in fruits are acquired and transported naturally for biological processes in the plants. However, some of these minerals this could be toxic to humans when they occur at particular levels. Instead of posing a threat, mineral nutrients under the normal level assist in mitigating toxicity caused by heavy metals. In our mineral study, the ANOVA statistical analysis highlighted nine important minerals from the heatmap (Figure S4 in Supplementary Materials) of the fruit samples. As the major mineral in human dietary intake, potassium contributed the highest percentage for all the samples ranging from 52–68% showing a good source of potassium, especially for Davidson's plum (68%). The requirement of potassium by humans is greater than 100 mg/day and with this fact, it is anticipated that the contribution of these fruits to dietary intake will grow in future. The potassium levels of Davidson's plum (6877 ± 138 mg/kg), and finger lime (6697 ± 98 mg/kg) found in this study are higher than potassium levels in any Colombian fruits [38], and subtropical fruits grown in Spain, such as custard apple, avocado, mango, banana, papaya, persimmon and starfruit [39]. The amount of phosphorus (835 ± 49 mg/kg), calcium (788 ± 37 mg/kg) and magnesium (723 ± 23 mg/kg) were relatively high in native pepperberry, compared to jujube fruits grown in China [40]. Bush fruits have been reported for rich sources of calcium and magnesium with an added advantage because both minerals are important in forming DNA and are especially important in repairing damaged DNA. They are equally crucial in segregating the chromosomes during DNA synthesis [19].

Taken together, all the data indicate that these three native fruits are excellent sources of important compounds, that offer nutritional value to consumers. For the food industry, these fruits offer potential to extend shelf-life naturally, and increase the functional value of the food, enabling the industry to make health claims about the food. This information could also assist with the development of Indigenous businesses to supply these high value foods to the food industry. In addition, this study illustrates that the reason for different antioxidants values from the different methods is based on the nature and classes of the secondary compounds much more than any of the other compounds.

#### **4. Materials and Methods**

#### *4.1. Chemicals and Reagents*

All organic solvents were purchased from Fisher Scientific (Pittsburgh, PA, USA) unless stated otherwise. Ammonium ferrous sulfate, sulfuric acid, aluminium chloride, sodium nitrate, sodium hydroxide, sodium acetate, aluminium hexahydrate, ammonium molybdate, sodium phosphate, sodium nitrite, quercetin, gallic acid, glacial acetic acid, 2,4,6-Tris(2-pyridyl)-s-triazine (TPTZ), 2,2-Di(4-tert-octylphenyl)-1-picrylhydrazyl (DPPH), potassium persulfate, formic acid, 2,2-azinobis (3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt (ABTS), were purchased from Sigma-Aldrich (St. Louis, MO, USA).

#### *4.2. Plant Materials*

Davidson's plum, native pepperberry, and finger lime and were purchased from Taste Australia Bush Food Shop (Queensland, Australia). The samples were pulverised into fine powder using TissueLyser II (Qiagen, Tokyo, Japan). The pulverised samples were extracted with solvent or water (specified in respective method) in triplicates for chemical assays.

#### *4.3. GC*×*GC-TOFMS Analysis*

The aroma in fruits was comprehensively analysed using static headspace extraction coupled with separation by two-dimensional gas chromatography time-of-flight mass spectrometry (GC × GC-TOFMS, LECO Pegasus 4D, Castle Hill, Australia). Briefly, sample (500 mg) was weighed into 20 mL silicon capped GC headspace vials (Restek, Germany) and kept at −80 ◦C until further analysis. 2.5 mL headspace syringe was used to collect 1.5 mL of sample headspace after sample agitation of 10 min at 80 ◦C. An empty vial was used as blank and quality assurance (QA) standard was prepared by mixing all the samples. The program settings, conditions and parameters are provided

in Table S2 in Supplementary Materials [41]. LECO ChromaTOF 4.50 software was used to process the GC×GC-TOFMS data for pre-processing baseline correction and identification was conducted by library matching (NIST 11 v2.0) and from authentic reference standards created in an in-house library. The similarity of ≥80% with the NIST library was defined as putative identification (when standards were not available) [42–44].

#### *4.4. UHPLC-QqQ-TOF-MS*/*MS Analysis*

The pulverised fruits powder (10 mg) from 3 plant species were dissolved in methanol (600 μL) and vortexed for 15 min at 30 ◦C. The mixture was then centrifuged at 28,500× *g* for 15 min and an aliquot of each supernatant (100 μL) was transferred into new Eppendorf tubes. Millipore water was added in ratio 3:1 to the supernatant. The mixture was vortexed and filtered through a 0.2 μm PTFE filter into an autosampler vial. For pooled biological quality control (PBQC) samples, 5 μL of each sample was taken and vortexed for 1 min. For the blank, 100 μL of Millipore water was used. The UHPLC-QqQ-TOF-MS/MS analysis was performed using Shimadzu Nexera UHPLC system (Kyoto, Japan; LC-30AD pump, SIL-30AC autosampler and CTO-30A column oven) equipped with Shimadzu Q-TOFMS-9030 detector. Separation of the sample analytes was conducted on a Shimadzu Velox C18 (2.1 × 100 mm, 1.8 μm, part number 227-32007-03, Shimadzu, Kyoto, Japan). The mobile phase consisted of A (0.1% [*v*/*v*] formic acid in acetonitrile) and B (0.1% [*v*/*v*] formic acid in water). The flow of the solvent gradient flow was set as follows: 97% A for 0–0.75 min, 5% A for 0.75–13 min, 97% A for 13–16 min. The sample injection volume was 1 μL with consistent flow rate at 0.4 mL/min. The column temperature and auto-sampler were set at 40 ◦C. Negative ionisation mode was operated for mass spectrometry analysis equipped with electrospray ionisation (ESI) source with collision energy set at 70 eV. The MS data were collected from *m*/*z* 70–700 Da; nebulisation gas at 3 L/min; source temperature, 120 ◦C; and desolvation temperature, 200 ◦C. The MS data were centroided and acquired with Lab Solutions software version 5.80. The raw data files were exported to Lab Solutions Insight software in LCD (\*.lcd) format for pre-processing, correction of retention time, and baseline. The raw files were also converted to mzML format for peak discrimination, filtering and alignment using MS-DIAL [45,46].

#### *4.5. Targeted Analysis of Free Amino Acids*

#### 4.5.1. Extraction of Free Amino Acids from Fruit Samples

The samples were weighed (20 mg) in Eppendorf tube. Methanol (500 μL) was added and vortexed. The mixture was centrifuged at 30,000× *g* for 5 min, then the supernatant was transferred into a 2 mL tube. The precipitate was resuspended in 500 μL of Millipore water and vortexed. Next, the mixture was centrifuged at the same speed for 5 min. The methanol and water extract were combined and filtered through a 0.2 μm PTFE filter.

#### 4.5.2. Free Amino Acid Derivatisation

The free amino acids were derivatised using AccQ.FluorTM reagent according to the Waters AccQ.TagTM pre-column derivatisation procedure [47]. Briefly, 35 μL of borate buffer was put into a tube (2 mL). 5 μL of sample was then mixed and vortexed for several seconds. Ten μL of reconstituted derivatisation reagent were then admixed to the buffered samples and immediately vortexed. Next, the mixture was left at room temperature for 1 min. After that, the samples were heated in a heating block for 10 min at 55 ◦C to finalise the derivatisation. The derivatised free amino acids were then transferred to autosampler vial for analysis. PBQC samples were prepared by pooling 5 μL from each sample, vortexed for 1 min and derivatised using the same method as mentioned previously.

#### 4.5.3. Standards

Standard amino acid mix solution (Batch. No SLBS2232V; Sigma-Aldrich, St. Louis, MO, USA) was used for the identification and quantification of amino acids. The standards were prepared in serial dilutions of 40% stock solutions to 20%, 10%, 5%, 2.5% and 1.25%. Derivatisation of the standard amino acids were conducted according to the methods explained in Section 4.5.2. As for blank, 5 μL of Millipore water was used to replace sample at the beginning step, then followed by the derivatisation steps mentioned in Section 4.5.2.

#### 4.5.4. UHPLC-MS Conditions

The analysis of amino acid derivatives was analysed on a Shimadzu Nexera UHPLC system (Kyoto, Japan; LC-30AD pump, SIL-30AC autosampler and CTO-30A column oven) equipped with Shimadzu MS-2020 detector. Waters Acquity UPLCTM BEH C18 column (2.1 <sup>×</sup> 100 mm, 1.7 <sup>μ</sup>m, part number 186003837, Waters, Milford, MA, USA) was used for the chromatographic separation at consistent temperature of 55 ◦C. The MS parameters were set as follows: Acquisition mode, SIM (refer Table S3 in Supplementary Materials); detector voltage, 0.1 V; interface voltage, 2.5 V; ionisation mode, positive; heat block temperature, 500 ◦C; interface temperature, 350 ◦C; nebulising gas flow, 1.5 L/min; injection volume, 10 μL. The mobile phase consists of A: 0.1% formic acid (*v*/*v*) in Millipore water, and B: 0.1% formic acid (*v*/*v*) in acetonitrile. The flow rate was set at 0.7 mL/min based on the gradient profile: initial-0.54 min (0–0.1% B); 0.54–5.74 min (0.1–15% B); 5.74–8.74 min (15–21.2% B); 8.74–10.50 min (21.2–59.6% B); 10.50–11.50 min (59.6% B); 11.50–12.00 min (59.6–0.1% B) and finally at 0.1% B until 13 min. The interconnected cleaning purge was set within 1 min (rinsing speed 35 μL/sec), and equilibrium was repeated for 5 min at initial conditions. The whole cycle time took 13 min to complete before the next injection.

#### *4.6. Targeted Analysis of Protein Amino Acids*

#### 4.6.1. Sample Digestion

Fruit samples (20 mg each pulverised) were digested in a glass vessel containing 2 mL of 6 N HCl and 0.1% of phenol. The glass vessels were flushed with N2 before sealing. Samples were hydrolysed at 110 ◦C for 20 h. Next, the samples were filtered using 0.2-μm filter, and the filtrate was neutralised with freshly prepared 6 N NaOH solution [48].

#### 4.6.2. Protein Amino Acid Derivatisation

The protein amino acid derivatisation, PBQC and standards were prepared according to the methods mentioned in Sections 4.5.2 and 4.5.3.

#### 4.6.3. UHPLC-MS Analysis

The standards and derivatised protein amino acid were injected and analysed according to the UHPLC-MS conditions mentioned in Section 4.5.4.

#### *4.7. Targeted Analysis of Minerals and Heavy Metals*

#### 4.7.1. Sample Preparation

The fruits samples were weighed (100 mg) into new and clean 15 mL digestion tubes. To each tube, concentrated nitric acid (2.0 mL) was added and samples were pre-digested overnight. They were then placed into a digestion rack of Hotblock® Digestor SC100 Digestion System (Environmental Express, Vernon Hills, IL, USA). The samples were digested for 1 h at 100 ◦C. Upon completion, the samples were cooled, and then water added to bring the volume to 15 mL. The samples were shaken and centrifuged for 5 min [49].

#### 4.7.2. ICP-OES Analysis

The operating conditions for inductively coupled plasma optical emission spectrometer (ICP-OES) were performed according to the parameters: RF incident power, 1000 W; plasma argon flow rate, 15.0 mL/min; auxillary argon flow rate, 1.50 mL/min; nebulizer argon flow rate, 0.75 mL/min; mist chamber, tracey and nebulizer SeaSpray, 2 mL/min flow rate. The wavelength measured for each element is as follows: Al 396.152 nm, As 188.980 nm, Ca 422.673 nm, Cd 214.439 nm, Co 238.892 nm, Cr 267.716 nm, Cu 324.754 nm, Fe 238.204 nm, K 766.491 nm, Mg 279.553 nm, Mn 257.610 nm, Mo 202.032 nm, Na 588.995 nm, Ni 231.604 nm, P 213.618 nm, Pb 220.353 nm, S 181.972 nm, and Zn 213.857 nm.

#### *4.8. Determination of in Vitro Antioxidant Activity*

The DPPH, FRAP, ABTS and phosphomolybdenum activities of the fruit samples were evaluated by the methods reported by [44,50,51], respectively. The ABTS, DPPH and phosphomolybdenum assays were reported as gallic acid equivalents (GAE μmol/gDW), whereas FRAP as Fe2<sup>+</sup> equivalents (Fe2<sup>+</sup> μmol/gDW).

#### *4.9. Total Phenolic, Flavonoid and Flavonol Contents*

Spectrophotometric technique was used for total phenolic, flavonoid and flavonol contents and were quantified following the protocol of [52,53]. Total phenolic contents were expressed as gallic acid equivalents (GAE μmol/gDW), whereas total flavonoid and flavonol contents were presented as quercetin equivalents (QTE μmol/gDW).

#### *4.10. Data Processing and Analysis*

Data for bioactivities were presented as mean ± SD for all triplicate analysis and a one-way analysis of the variance (ANOVA) was carried out using SPSS Statistics version 25 (IBM Corp, Armonk, NY, USA). The mean comparisons were conducted using the post-hoc Tukey's (HSD) multiple comparison test. Values with *p* < 0.05 were considered statistically significant.

The processed data from GC×GC-TOFMS were analysed using SIMCA-P software version 15 (Umeå, Sweden) for multivariate data analysis (MVDA). In order to access the clustering and trends of the comprehensive depiction of the fruit samples, principal component analysis (PCA) was used. Partial least squares (PLS) chemometric method was conducted to further analyse the correlation of antioxidant activity with volatile compounds in the fruit samples [41,44]. The variables were pareto scaled for PCA and PLS analyses. Afterwards, the variables selection namely VIP was used to select those notable volatile compounds (VIP > 2.5).

For UHPLC-QqQ-TOF-MS/MS data, the variables were pareto scaled and log transformed for PCA and PLS to lower heteroscedasticity and asymmetry in the statistical distribution [44]. In order to tentatively identify the compounds, the acquired raw data in mzML format were processed using MS-DIAL by comparing the MS/MS spectra with those in the spectral library. VIP was applied to identify the most significant compounds contributing to the antioxidant activity [54]. Compounds with VIP score > 1.5 were maintained for further elaboration using methods from MetaboAnalyst 4.0, an open source web-based tools for metabolomics data analysis [55].

For targeted MVDA of amino acids, both targeted free and hydrolysed amino acids were pareto scaled and log transformed. The heatmap and box and whiskers were generated using MetaboAnalyst 4.0 [55].

#### **5. Conclusions**

We have successfully reported that the Australian bush fruits, Davidson's plum, finger lime and native pepperberry are rich in terpenes, phenolics, flavonoids, flavonols, minerals, essential and non-essential free and hydrolysed protein amino acids, and functional bioactive compounds, which are promising fruits to be commercialised in the nutraceutical industry and food industry. We found that the fruits are an abundant source of antioxidant compounds (sugars, terpenes, flavonoids etc.) and that may serve as the source of natural antioxidants in food products or as new medicines. The use of metabolomics in correlation to the primary and secondary metabolites as well as the minerals would render an opportunity to obtain more potential bioactivities of the Australian bush fruits as functional food in nutraceutical industry. We also show that when reporting antioxidant activity, it is important to use more than one method to obtain a true understanding of the activity.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2218-1989/10/3/114/s1, Table S1: Targeted analysis of 18 minerals and heavy metals from Davidson's plum (DP), finger lime (FL) and native pepperberry (NP). Results are expressed as mean values (mg/DWkg) ± SD (*n* = 3). Superscript letters within each column indicate statistically significant (*p* < 0.05; Tukey test), Table S2: GC×GC-TOF-MS parameters for comprehensive profiling of fruit samples, Table S3: Amino acids' single ion monitoring for mass spectrometry detector, Figure S1: (**a**) Partial least square (PLS) score plot based on GC×GX-TOFMS data. (**b**) PLS biplot plots showing correlation between identified aroma compounds with antioxidant activity. X = compounds, Y = antioxidant activity, Figure S2: Permutation test of the PLS model based on UHPLC-QqQ-TOF-MS/MS data (**a**) ABTS; (**b**) DPPH; (**c**) FRAP; (**d**) Phosphomolybdenum assays, Figure S3: Biplot association of (**a**) free amino acid; (**b**) hydrolysed protein amino acid in Davidson's plum (DP), finger lime (FL) and native pepperberry (NP), Figure S4: Heat map of 18 mineral nutrients and heavy metals found in the fruit samples.

**Author Contributions:** Conceptualization, M.F. and V.L.; methodology, V.L. and S.G.G.; software, V.L. and S.G.G.; validation, V.L., S.G.G., V.D.D. and M.F.; formal analysis, V.L., S.G.G.; investigation, V.L., S.G.G., V.D.D.; resources, M.F.; data curation, V.L.; writing—original draft preparation, V.L.; writing—review and editing, V.L., S.G.G., V.D.D., M.F.; visualization, M.F.; supervision, M.F.; project administration, M.F.; funding acquisition, M.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** V.L. would like to thank Universiti Sains Malaysia and Ministry of Higher Education, Malaysia for the SLAB Scholarship under the Post-Doctoral program. We also thank Shimadzu Australia for access to some equipment and assistance with data analysis.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
