Combining metabolomic and transcriptomic approaches to assess and improve crop quality traits

Abstract

Plant quality trait improvement has become a global necessity due to the world overpopulation. In particular, producing crop species with enhanced nutrients and health-promoting compounds is one of the main aims of current breeding programs. However, breeders traditionally focused on characteristics such as yield or pest resistance, while breeding for crop quality, which largely depends on the presence and accumulation of highly valuable metabolites in the plant edible parts, was left out due to the complexity of plant metabolome and the impossibility to properly phenotype it. Recent technical advances in high throughput metabolomic, transcriptomic and genomic platforms have provided efficient approaches to identify new genes and pathways responsible for the extremely diverse plant metabolome. In addition, they allow to establish correlation between genotype and metabolite composition, and to clarify the genetic architecture of complex biochemical pathways, such as the accumulation of secondary metabolites in plants, many of them being highly valuable for the human diet. In this review, we focus on how the combination of metabolomic, transcriptomic and genomic approaches is a useful tool for the selection of crop varieties with improved nutritional value and quality traits.

Background

Understanding the genes involved in metabolism and dissection of the metabolic pathways are essential to improve plant adaptation to biotic and abiotic stresses, to improve food quality, and to increase crop yield, all being crucial factors to secure global human nutrition (Gong et al. 2013). Crop domestication and intensive breeding along history contributed to decrease genetic diversity though successive bottlenecks (Bai and Lindhout 2007; Sauvage et al. 2014). While breeders have traditionally focused on improving agronomical important traits, such as yield and pest resistance, crop quality traits, such as flavour or nutritional value, have long been omitted. In fact, deterioration in flavour quality of important commercial crops, such as tomato, is a major cause of consumer complaint (Klee and Tieman 2018). To reverse this decline, identification and quantification of the compounds responsible for consumer liking must be first undergone, followed by the understanding of what has been lost during domestication and intensive breeding (Tieman et al. 2017). In addition, improving seed quality of main crops is a major target of breeding programs, to ensure human nutrition, as they provide a significant percentage (around 20%) of the consumed energy, protein and dietary fiber (Peng et al. 2018).

Most crop agronomic and quality characteristics are being determined by multiple quantitative traits (QTL) and influenced by the environmental and culture conditions. Loci underlying these economically-important characteristics can be located by linkage mapping and genome-wide association (GWAS). Association between genetic variation and traits of interest can further be used for marker-assisted selection in the breeding programs, in order to obtain high quality crops.

As a result of plant metabolism, fruits and cereals accumulate a myriad of metabolites, which main functions are to protect them against pathogens and herbivores and to attract seed dispersers (Pott et al. 2019). Primary metabolism leads to the synthesis of sugars, amino and organic acids, which are used as building components by the cells and serve as precursors for secondary metabolism. This specialized metabolism, crucial for the plant interaction with its entourage, is extremely diverse and results in the accumulation of a plethora of compounds, i.e. polyphenols, terpenoids or nitrogen/sulfur-containing metabolites (Aharoni and Galili 2011; Pott et al. 2019). Many of these secondary metabolites have been associated with health-promoting characteristics (Pott et al. 2019; Sun et al. 2018).

In this sense, recent advances in metabolite profiling technologies, which now allow the simultaneous detection and quantification of thousands of metabolites, combined with genomic and transcriptomic platforms, are fundamental in dissecting crop composition and in identifying the genetic variants underlying metabolic content (de Abreu e Lima et al. 2018; Ballester et al. 2016; Garbowicz et al. 2018; Labadie et al. 2020; Li et al. 2019; Osorio et al. 2011, 2019; Rambla et al. 2016; Vallarino et al. 2019). While the integration of genetic and metabolic information is a powerful strategy to dissect the bases of plant metabolism and to associate complex traits with genotype, it is also a key strategy for the breeding of high-yielding and nutritionally rich crops (Luo 2015; Wen et al. 2018). In this review, we focus on recent studies, published during the last 6 years, which combined metabolomic approaches with genomic or transcriptomic advances to dissect plant biochemistry for crop quality improvement. Web of Science and Scopus databases were screened using the keywords metabolomics, transcriptomics, genomics, proteomics, GWAS, fruit OR crop, quality. Recent studies (2014–2020) were selected when combining metabolomic approach with any of the other above-mentioned omic strategies. As a result of this initial screen, we decided to focus our review on three important and diverse groups of metabolitesin different crops associated with fruit quality, i.e. (i) essential amino acid content in cereals, (ii) health-promoting and colourful secondary metabolites (anthocyanins and carotenoids) in fruits and (iii) taste- and aroma-related metabolites and the recovery of the associated alleles as a result of tomato domestication and intensive breeding. In this sense, the review will allow us to cover a wide range of major crop species, including cereals, fruits and tubers, focusing mainly on strategies to improve plant nutritional value. In parallel, the unprecedented example of tomato genetic dissection for fruit quality improvement will enable us to present omic strategies which could be translated to other crop species. In addition, methods and techniques used for metabolomic, genomic and transcriptomic analyses in the studies reviewed here are summarized in Table 1.

Table 1 Summary of the different metabolomic, transcriptomic and genomic methods and softwares (when information available) used in the cited studies, to improve crop quality

Main text

Combining metabolomic and genomic tools to increase amino acid content in cereals and soybean

Primary metabolites, mainly sugars, acids and their derivatives, serve both as cell building blocks and as essential nutrients for animal and human consumption, as they accumulate in sink organs, such as fruits, seeds or tubers (Wen et al. 2018).

The rapid development of next generation sequencing platforms has allowed high-density genotyping with SNPs, favouring the use of GWAS to fine-map loci involved in plant metabolic complexity. In particular, combining GWAS with metabolomic platforms (mGWAS) has permitted the simultaneous screening of an enormous number of accessions for primary metabolic content first in the model plant Arabidopsis (Chan et al. 2010) and then in a series of economically important crops, such as maize, rice or tomato (Chen et al. 2014; Sauvage et al. 2014; Wen et al. 2014). Levels of most primary metabolites are found to be normally controlled by multiple loci, each of them explaining a low to moderate percentage of the metabolic variability (Chen et al. 2014; Deng et al. 2017; Wen et al. 2015). Identified QTLs are normally not randomly distributed on the chromosomes, but QTL ‘hotspots’ are commonly observed, suggesting the combined action of closely linked genes or pleiotropic effect (Chen et al. 2014, 2016; Deng et al. 2017; Gong et al. 2013; Peng et al. 2018; Vallarino et al. 2019). It is also important to outline that the overlapping between different populations or varieties in the identified loci is quite limited, as shown by Chen et al. (2014) or Deng et al. (2017). Indeed, combining metabolic profiling of rice (Oryza sativa) leaves from indica and japonica subspecies with GWAS, only 155 out of 514 identified loci were common in both groups, indicating heterogeneous genetic control (Chen et al. 2014). Furthermore, most of the loci identified by GWAS appear to be tissue-specific, implying separate genetic and biochemical regulation (Chen et al. 2016; Gong et al. 2013; Wen et al. 2015, 2018). This tissue-specific metabolic variation is possibly determined by differential allelic expression (Chen et al. 2016). However, common strategies for the genetic control of certain metabolites may be conserved between plant organs (Wen et al. 2018), subspecies (Chen et al. 2014) or even species, as demonstrated by a comparative GWAS between the two major crops maize and rice (Chen et al. 2016). In this sense, integrative approaches, combining metabolomics with functional genomics, may help crop rapid improvement. In the next paragraphs, we will focus on loci and gene identification for amino acid content in major crops, an essential nutritional trait to secure global food.

Indeed, there is an increasing demand for major crops with enhanced nutritional quality, to both sustain a growing population and improve human health (Galili et al. 2002). In particular, improving the content of minerals, vitamins and proteins of major crops, which are the main nutrient sources in developing countries, is a key strategy to solve severe widespread health problems associated with nutritional deficiencies (Deng et al. 2017; Galili et al. 2002; Galili and Amir 2013). For example, the seed protein content of most consumed cereals, such as maize or wheat, is directly associated with their nutritional quality, and, as animal proteins are more expensive, part of the humanity depends on it to fulfil this dietary requirement (Mandal and Mandal 2000). While protein content depends on the availability of free amino acids, breeding improvement has been hindered due to the tight regulation of their synthesis and to a highly efficient catabolic rate (Galili et al. 2016). An additional difficulty is the limited availability of some essential amino acids, which cannot be synthesized by humans, in most cereals and legumes. (Galili and Amir 2013). In particular, the limited availability of lysine, methionine, threonine or tryptophan in most crops made necessary the identification of the genes regulating these metabolites, and is a key step for breeding improvement and crop biofortification as a critical strategy to struggle against global health nutritional deficiencies (Deng et al. 2017; Galili et al. 2016; Galili and Amir 2013; Yang et al. 2020).

GWAS and linkage mapping studies were combined to examine the genetic architecture of amino acids in the maize (Zea mays) kernel, with a special focus on essential amino acids, using a panel of diverse inbred lines and different recombinant inbred line populations (Deng et al. 2017; Wen et al. 2015, 2018). While Deng et al. (2017) were able to outline 308 candidate genes, that could be associated to 528 identified loci, Wen et al. (2018) found 153 significant loci related to the content of 61 primary metabolites, including 23 amino acids. Furthermore, it appeared that epistatic interactions may play an important role in controlling free amino acid content, as shown by Loudet et al. (2003) and Wen et al. (2015). Epistasis can be explained by physical or functional interactions between genes or gene products, such as protein–protein interactions or transcription factors (TF) controlling the expression of structural genes (Wen et al. 2015).

Interestingly, only 27% of the candidate genes identified by Deng et al. (2017) encodes enzymes or proteins which affect amino acid metabolism, and the function of up to 35% was unknown. Similarly, only a small amount of the 153 loci were known or well characterized previously to the study published by Wen et al. (2018), confirming the necessity of mGWAS and association studies for the identification of natural variation for metabolic traits. Once loci are identified, with underlying putative causal genes, further validation of these candidates is fundamental, before marker development for breeding programs. Due to the usual high number of identified loci, this downstream analysis can be complex, for which it is highly important to prioritise the selection of the putative candidates, according to strong evidence.

These evidences may include expression QTL analysis (eQTL), as undergone by Deng et al. (2017), which took advantage of previously published RNA-sequencing data of young maize kernels (Fu et al. 2013). Interestingly, 16.2% of the 308 candidate genes were found to possibly affect amino acid variation via transcriptional regulation, as significant correlations between gene expression, identified eQTL and metabolite content were established (Deng et al. 2017). Further evidences may consist of gene functional annotation matching corresponding metabolite of the direct association between gene expression levels and metabolite content (Wen et al. 2018). Wen et al. (2015) noticed that most candidate genes they found were only putatively annotated; however, their genetic mapping fell together with gene predicted functions in many cases, including candidate for essential amino acid content. In particular, six genes involved in the aspartate-derived amino acid pathway were proposed as candidate genes, as they co-localized with QTL for lysine, methionine and homoserine.

Among the promising candidate genes found to be associated with essential amino acid content by GWAS studies in maize kernels were an arogenate dehydratase, responsible for phenylalanine accumulation, and Opaque2 (O2), a basine leucine zipper (bZIP) TF involved in the regulation of the expression of several genes during kernel development, and associated with lysine/total amino acid ratio (Table 2) (Deng et al. 2017; Wen et al. 2015, 2018). A 811-bp insertion was found in the 5′ untranslated region of the arogenate dehydratase with an allelic frequency of 46% and explaining both the decrease in gene expression and in phenylalanine content (Wen et al. 2018). The Lys/total amino acid ratio was significantly correlated with O₂ expression, for which a very strong cis-eQTL was also found. Furthermore, a duplication present in chromosome 7 was also seen to affect lysine content and its ratio to total amino acids and was identified both by GWAS in the diverse inbred lines and by QTL mapping (Deng et al. 2017; Liu et al. 2017). The expression of five genes, located within the duplicated region, were found to be significantly correlated with lysine to total amino acid ratio. Additionally, haplotype analysis for one of the genes showed significant differences both for Lys/total amino acid and expression. Combination of duplication and haplotype variants had greater effect on phenotype than single variation (Deng et al. 2017). Another candidate, an acetolactate synthase 1, involved in branched-chain amino acid metabolism, was found to be associated with leucine/total amino acid (Table 2). Interestingly, two eQTL were detected for this gene, including a cis- and trans-eQTL, which results to be O₂ (Deng et al. 2017). Taken together, the natural variation underlying the above mentioned loci and genes can be used for the improvement of essential amino acid composition of cereal crops.

Table 2 Candidate genes involved in nutritional and organoleptic crop quality traits

GWAS for amino acid content was also performed in bread wheat, rice or soybean (Chen et al. 2016; Lee et al. 2019; Peng et al. 2018; Qin et al. 2019; Sun et al. 2020). A highly diverse panel of 182 accessions of Triticum aestivum was used to map 328 significant quantitative trait nucleotides with six different multi-locus models. Based on chemical structure and previous knowledge of the synthetic pathways, 15 candidate genes could be tentatively associated to free amino acids levels, in particular six genes annotated as amino acid transporters and permeases co-localized with identified loci. Furthermore, the function of a tryptophan descarboxylase, found in a locus associated with tryptamine levels, was validated in vitro as a proof of concept (Table 2) (Peng et al. 2018). Two genes annotated as tryptophan decarboxylases have also been associated with tryptophan content in maize kernel (Wen et al. 2014). Very recently, a combination of GWAS and functional analysis in rice leaves of 520 accessions validated a bZIP TF, OsZIP18, as being the main genetic determinant for branched-chain amino acid (BCAA) content in rice (Sun et al. 2020). As humans and animals are not able to synthesize BCAA, OsZIP18 may be a promising candidate for increasing rice nutritional value; however, OsZIP18 is mainly express in leaf tissues, for which its impact on grain BCAA content needs further validation. In soybean (Glycine max), GWAS for seed composition in 321 accessions, including amino acid content, pinpointed a major-effect QTL on chromosome 8, harbouring an aspartokinase-homoserine dehydrogenase, a key enzyme involved in the synthesis of aspartate-family amino acids (Zhang et al. 2018). Interestingly, the same study suggests that free amino acid profiles in soybean seeds may be under a different genetic control from total protein levels, implying that the former may be improved without affecting protein levels. Indeed, as 92 amino acid-associated QTL were detected, only four were common with both dry weight- and protein-based amino acid content (Zhang et al. 2018). (Qin et al. (2019) detected 15 amino acid-associated SNPs located near 14 candidate genes in a GWAS analysis comprising 249 soybean accessions, which will be further validated with the objective to develop molecular markers for breeding purposes.

Transcriptomic and metabolomic platforms for improved crop colour

Anthocyanin pigments

Another group of plant metabolites which are particularly attractive for breeding programs are pigments, responsible for the appealing aspect of many ripe crops. In particular, anthocyanins have widely attracted research interest, for conferring red, blue, pink or purple colours to a large number of fruits and vegetables (Pott et al. 2019). Also, they are important for their health-promoting role due to their high antioxidant capacity and their ability to modulate mammalian cell signalling pathway (Butelli et al. 2008; Giampieri et al. 2015; Petrussa et al. 2013; Pott et al. 2019). In fact, engineering methodologies for increased anthocyanin content have been proposed in crops with suboptimal concentration of these metabolites, such as tomato fruits (Butelli et al. 2008; Zhang et al. 2013, 2015). Anthocyanin synthesis through the phenylpropanoid and flavonoid pathways has been widely studied (Vogt 2010; Fraser and Chapple 2011; Hassan and Mathesius 2012). However, due to the complexity and diversity of plant secondary metabolism, i.e. more than 630 anthocyanins have been identified (He and Giusti 2010), OMIC data set integration may be an optimal strategy to identify potential genes involved in pigment formation regulation. In particular, metabolomic platforms, using ultrahigh-performance liquid chromatography coupled with mass spectrometry is the technique of choice for anthocyanin and other polyphenol metabolite detection in plant tissues (Cho et al. 2016; Liu et al. 2020; Pott et al. 2020; Vallarino et al. 2018). Metabolite profiling combined with transcriptomic and genomic data and multivariate statistical models can be key in revealing the molecular mechanisms associated with crop pigment accumulation, as highlighted in the next section and in Fig. 1.

Fig. 1

Main factors controlling anthocyanin accumulation and colour formation in crops, based on recent multiomic-approach studies (Cho et al. 2016; Fang et al. 2016; Li et al. 2018b; Liu et al. 2020; Xu et al. 2013; Zhang et al. 2020) Transcription factors including basic helix-loop-helix (bHLH), WD40-repeat protein (WD40) and R2R3-MYB are shown to control anthocyanin synthesis in a wide range of crops, such as potato, plum, kiwifruit, pear or pepper. Moreover, reactions catalysed by leucoanthocyanidin dioxygenase (LDOX) and UDP-glucose:flavonoid O-glycosyltransferase (UFGT) seem controlling carbon flux towards the accumulation of coloured anthocyanins (Cho et al. 2016; Wang et al. 2017a). Hormonal control (mainly via auxins, ethylene and jasmonic acid) and anthocyanin transport (via GST or MATE) to the vacuole are also key processes for crop pigment accumulation. ERF: ethylene response factor, JA-Ile: jasmonoyl-L-isoleucine, JAZ: JASMONATE ZIM DOMAIN transcriptional repressor proteins, JIH: jasmonoyl-L-isoleucine hydrolase, GST: glutathione-S-transferase, MATE: multidrug and toxin extrusion, GST: glutathione-S-transferase

Combining metabolite profiling with high-throughput RNA-sequencing analysis, Cho et al. (2016) established a correlation network between metabolites and genes involved in the pigmentation of three potato (Solanum tuberosum) cultivars, ‘Hongyoung’ (light-red skin and flesh), ‘Jayoung’ (dark-purple skin and flesh) and ‘Atlantic’ (white-coloured). In particular, they focused on compounds belonging to the flavonoid class, including anthocyanins, i.e. delphinidin, petunidin and malvidin-derived metabolites, responsible for purple and dark hues, and derivatives of cyanidin and pelargonidin, mainly associated with bright-red-coloured tones (Jaakola 2013), and on differentially-expressed transcripts between cultivars, belonging to flavonoid metabolism, hormone metabolism, cell signalling or regulation of transcription (Cho et al. 2016). Interaction networks divided genes and metabolites into five clusters (I to V) and four groups (A to D), respectively, and correlation between cluster and group may indicate functional connection between a particular set of genes and flavonoid metabolites, shedding light on the regulation controlling the specific synthesis of anthocyanins in potato. Common mechanisms leading to the synthesis of anthocyanin may encompass the upregulation of cluster III genes, which expression was enhanced in both coloured cultivars, and which include TT8 and WD40-repeat protein (WD40). Interestingly, TT8 is a bHLH-type regulation factor, which forms a complex with WD40 and R2R3-MYB TF, responsible for the regulation of the flavonoid pathway, and precisely anthocyanin biosynthesis in Arabidopsis and other crops (Fig. 1) (Cho et al. 2016; Schaart et al. 2013; Xu et al. 2013). This observation suggests that TT8-mediated pathway also regulates pigmentation in coloured potato cultivars. Similarly, an RNA-Seq-based comparative transcriptome analysis allowed to identify TF which were coexpressed with anthocyanin biosynthetic genes during the ripening of the red-fleshed plum ‘Furongli’ cultivar (Prunus salicina). A bHLH gene, homologous to AtTT8, was found to accumulate to higher levels in late ripening stages and to show significant correlation with anthocyanin biosynthetic genes (Fang et al. 2016). In pepper (Capsicum spp.), some varieties show purple or black tones during the immature period due to the accumulation of anthocyanins (Lightbourn et al. 2008). By combining metabolite profiling with RNA-seq analysis, Liu et al. (2020) built a weighted gene co-expression network, highlighting the strongly connected genes. This way, a forin-like protein 11, an unknown gene (Capana02g003118), WDR68 (a WD40 gene), a solute carrier family 40 member 1 and MYB113 may be key genes in flavonoid synthesis in pepper, suggesting a key role of TF and transport in anthocyanin accumulation (Liu et al. 2020).

Hormonal roles in anthocyanin accumulation were also outlined by several transcriptomic and metabolomic studies (Cho et al. 2016; Wang et al. 2017b, a; Zhang et al. 2020). Indeed, most auxin-related genes were negatively correlated to the anthocyanins found in the both coloured potato cultivars, suggesting that their reduced expression may enhance pigment synthesis by removing the inhibition upon the flavonoid pathway. On the opposite, as ethylene-related genes were upregulated in the two pigmented potatoes in comparison with the colourless cultivar, it can be hypothesized that this phytohormone is associated with anthocyanin synthesis (Cho et al. 2016). However, NAC2 was down-regulated in both coloured cultivars. NAC genes are plant-specific TF, and are involved in a myriad of physiological processes, inclusive of fruit ripening and leaf formation and senescence, and stress responses (Shan et al. 2012; Zhang and Gan 2012). Some of them have been described to take part in hormone cascade signalling, including ethylene (He et al. 2005). Indeed, while NAC2 promotor was shown to be activated by ethylene in banana, NAC1 and NAC2 was shown to interact with ethylene insensitive (EIN) protein, a downstream component of ethylene signalling (Shan et al. 2012). EIN2-NAC2 pathway induces the expression of genes involved in senescence, for which anthocyanin accumulation in potato tubers may not be senescence-dependent and EIN2-NAC2 ethylene transduction pathway should be discarded (Cho et al. 2016; Woo et al. 2013). The positive role of ethylene-responsive genes (ERF) upon anthocyanins accumulation has also been outlined in pear and fig (Wang et al. 2017a, b; Zhang et al. 2020).

Specific to light-red potato cultivar compared to the dark-purple one, up-regulation of jasmonic acid (JA) signalling genes was observed by Cho et al. (2016), including JIH, JAZ8 and JAZ10 genes. JA is known to increase anthocyanin synthesis by favouring the degradation of negative regulators of JA (JAZ genes). Up-regulation of JIH in ‘Hongyoung’ cultivar, which catalyses the cleavage of JA-isoleucine, the JA conjugate which promotes JAZ degradation, combined with enhanced transcripts of JAZ8 and JAZ10 suggests that JAZ degradation-mediated anthocyanin biosynthesis is possibly inactive in the light-red cultivar compared to ‘Jayoung’, and may explain the lower colouration in the first one (Cho et al. 2016; Thines et al. 2007). Furthermore, the increased expression of several enzymes involved in flavonoid first-step reactions, a UDP-glucose:flavonoid O-glycosyltransferase (UFGT) and a leucoanthocyanidin dioxygenase (LDOX) in ‘Jayoung’ may explain how metabolic flux is favoured towards the synthesis of delphinidin- and petunidin-derived anthocyanins, conferring its purple tone (Cho et al. 2016). LDOX and UFGT genes have also been pointed out as being predominant structural genes for red-colour formation of kiwifruit (Actinidia arguta) flesh, together with AaF3H, a flavanone 3-hydroxylase (Li et al. 2018b). A model was proposed in which their expression was controlled mainly by AaMYB, AabHLH and AaHB2 TF (Li et al. 2018b).

In pears (Pyris communis), bud mutation is an important method for the selection of new varieties, including the natural occurrence of red-skin fruits, accumulating anthocyanins and proanthocyanidins (condensed tannins) in the peel (Zhang et al. 2020). A comparative metabolic and transcriptomic analysis of the young fruit of ‘CF’ cultivar and its red mutant ‘RCF’, whose colouration appears as early as five days after full bloom, may facilitate the understanding of the mechanisms underlying pigment accumulation due to the high similarity between the two fruits. Twenty candidate genes associated with anthocyanins and proanthocyanidins, which shared common flavonoid precursors, were outlined, including six structural flavonoid genes and four R2R3-MYB genes. Interestingly, while one the MYB candidate, PcMYB114, has been previously known to be involved in anthocyanin synthesis in pears (Yao et al. 2017), this approach allowed to outline upon which specific anthocyanins it acts and to newly identify three MYB genes associated with this process (Zhang et al. 2020). Furthermore, other TF found to be associated with anthocyanin and proanthocyanidins accumulation were PcKNAT1, PcbZIP1 and WRKY28 (negatively correlated). The latter two have been previously described to be involved in anthocyanin accumulation in other species, though WRKY28 putative function as a repressor was found to be the opposite in apple (An et al. 2019). PcKNAT1 belongs to the class I KNOX HD gene family, known to be involved in meristem development and leaf morphogenesis (Lincoln et al. 1994). Although a role in anthocyanin synthesis has been described for TF belonging to other HD subfamilies, the identification of PcKNAT1 as a candidate gene for this process is novel (Zhang et al. 2020). Finally, a glutathione-S-transferase, PcGST12, was found to be correlated with seven anthocyanins and seven proanthocyanidins and is functionally characterized as an anthocyanin and proanthocyanidin transporter from cytosol to vacuole. Zhang et al. (2020) found that PcGST12 expression is positively regulated by PcMYB114 and stable overexpression of PcGST12 in Arabidopsis pointed out that PcGST12 may be involved not only in transport, which is also a key process in determining anthocyanin levels, but in other steps of anthocyanin and proanthocyanidin biosynthesis.

The role of anthocyanin transport has also been highlighted as a key factor controlling pigment accumulation in other crops, such as carrots (Meng et al. 2020), or in two different red-flesh pears varieties, ‘Red Bartlett’, which shows a decrease in anthocyanin content along ripening concomitant with colour fading, and ‘Starkrimson’, which maintain purple-red colour throughout maturation (Wang et al. 2017b). Indeed, six GST and two multidrug and toxic compound extrusion transporters, also involved as flavonoid carrier, were downregulated at the later stage of ‘Red Bartlett’ ripening and could explain at least partially colour fading mechanism. In addition, Wang et al. (2017b) highlighted the possible role of two LDOX genes and a flavonol synthase (FLS) in this differential behaviour between the two varieties. While LDOX genes were downregulated during ‘Red Bartlett’ ripening, FLS was upregulated and its expression was negatively associated with anthocyanin content. In addition, several TF were also outlined as possible candidates in the differential colouration pattern between the two varieties, including MYB, bHLH, NAC, WRKY and ERF TF, however, no clear candidate could be highlighted. Finally, two peroxidases and an intracellular laccase, with a possible function in anthocyanin degradation, were differentially expressed between the two cultivars. Interestingly, comparing the transcriptome and metabolome of three varieties of a wild peach species (Prunus mira), with different flesh pigmentations, Ying et al. (2019) also highlighted a peroxidase, which expression was strongly enhanced in the white-fleshed pear. This enzyme has been associated with anthocyanin degradation in several fruits, and this result suggests that besides TF and structural gene downregulation, anthocyanin enhanced degradation favoured white-flesh colouration in wild peach (Ying et al. 2019).

Apart from pears, colour mutation is also common in other important crop trees, such as fig (Ficus carica). Four cyanidin glucosides, identified in ‘Purple Peel’ fig cultivar, were responsible for the colour difference between ‘Green Peel’ cultivar and its mutant. Interestingly, accumulation of other colourless flavonoids was also observed in the mutant, suggesting increased nutraceutical properties (Wang et al. 2017a). The authors of the study also outlined that even ‘Purple Peel’ is considered the mutant of ‘Green Peel’, it should probably be considered the first one as the wild type condition, based on the fact that the seeds inside fig syconia are bird-dispersed, and fruits with this characteristic are generally black or red (Willson and Whelan 1990). In this sense, green-fruit phenotypes can be understood by a functional loss of key enzymes in the flavonoid or anthocyanin pathway or by mutations, such as retrotransposon insertion, in key regulatory genes, such as MYB TF, as it has been described for white grapes (Kobayashi et al. 2004; Wang et al. 2017a). Interestingly, differential expression of transposons and retrotransposons were observed between ‘Green Peel’ and ‘Purple Peel’ cultivars, and in the first one a general upregulation of reverse transcriptases, integrases and gag sequences could be noticed (Wang et al. 2017a).

Another regulating mechanism underlying anthocyanin accumulation during fruit ripening may involve sugar signalling cascade (Dai et al. 2014; Jia et al. 2016; Luo et al. 2019). In particular, carbon starvation may affect pigment accumulation in red-fleshed kiwifruits (Nardozza et al. 2020). Indeed, metabolite profiling combined with RNA-Seq analysis in fruits grown under low (one leaf per fruit) or high (four leaves per fruit) carbon supply showed a general downregulation of genes and metabolites from both carbohydrate and anthocyanin biosynthetic pathways. In particular, the sugar trehalose-6-phosphate, a sucrose sensor and regulator, was proposed as a signalling molecule reflecting the levels of imported sugars to heterotrophic tissues (Figueroa and Lunn 2016). Trehalose-6-phosphate may act upon MYB27, a repressor of anthocyanin synthesis, negatively controlling its expression via a cis-acting sugar-repressive element present in MYB27 promotor (Nardozza et al. 2020). In this sense, it was proposed that under carbon starvation, decreased levels of trehalose-6-phosphate mirroring carbon status favoured MYB27-mediated repression upon anthocyanin synthesis (Nardozza et al. 2020).

Carotenoid pigments

Carotenoid are the second largest group of pigments, after anthocyanins, contributing to the red, orange or yellow tones of many fruits and crops. They play important roles in planta, such as photoprotection, and serves as precursors of a series of physiological-important molecules, both for plants (strigolactones, abscisic acid or apocarotenoid volatiles, among others) and humans (provitamin A) (Ashokkumar et al. 2020; Pott et al. 2019; Sun et al. 2018). While QTL mapping for carotenoid genes and underlying candidate genes has been reviewed in cereal crops such as durum wheat (Colasuonno et al. 2019) and will not be discussed here, GWAS and linkage analysis have recently shed light on the genetic control of carotenoid metabolism in important fruit and crop species, and will be summarized in this section. In addition, current QTL mapping and GWAS in important crops, such as maize or chickpea, remains a valid approach to unravel carotenogenesis genetic architecture (Azmach et al. 2018; Esuma et al. 2016; Owens et al. 2014; Rezaei et al. 2019; Xu et al. 2019; Zhai et al. 2018). However, most genes underlying pinpointed QTL are still unknown, and further studies are needed.

Rind and flesh melon (Cucumis melo) colours are important quality traits, linked to consumer liking, and which depends on chlorophyll/carotenoid synthesis and accumulation. Surprisingly, GWAS performed on 120 melon accessions with contrasting rind colours yielded no significant association between pigment content and genetic markers (Oren et al. 2019). However, whole-genome linkage analysis in two segregating bi-parental populations led to the identification of unique highly significant locus on chromosome 4 (Oren et al. 2019). Sequencing of an Arabidopsis pseudo-response regulator2-like (CmAPPR2) gene, found within the locus confidence interval, outlined five major polymorphisms, including two independent base substitutions leading to premature stop codons in the two light-rind parents (Table 2). Of particular interest, carotenoid profiles in the melon flesh were also associated with CmAPPR2 haplotypes. Furthermore, APRR2 overexpression in tomato resulted in higher levels of both chlorophyll in unripe fruits and carotenoid pigments in ripe fruits (Pan et al. 2013); in addition, white immature fruit colour in cucumber (Cucumis sativus) was consequence of a frameshift mutation leading to a premature stop codon in APRR2 open reading frame (Liu et al. 2016). These multiple and independent polymorphisms found in CmAPPR2, and their low frequency, could explain the lack of GWAS success in pinpointing any associated locus. While this example outlines the need to improve GWAS resolution for low-frequency causative variants, alternative layers of information obtained from OMIC data integration can be used as a valuable alternative approach to identify important loci (Lee and Lee 2018). Interestingly, the watermelon (Citrullus lanatus) homolog ClAPRR2 was also mapped by whole-genome linkage analysis and a putative causative SNP, leading to a 16-pb deletion in the mRNA of the light rind parental line, was identified (Oren et al. 2019). Watermelon red flesh intensity, which depends on the accumulation of the red carotenoid lycopene, is also considered an important quality trait. Whole genome variation analysis in populations derived from scarlet- and coral-red fleshed accessions mapped a 40 kbp region on chromosome 6 responsible for lycopene variation. Strikingly, three unknown and four glycine-rich cell wall structural genes were found in the confidence interval, and showed different expression along fruit ripening (Li et al. 2020). While two InDel markers were able to explain the observed phenotypic variance and could be useful for watermelon breeding programs, the underlying gene and the molecular mechanism responsible for differential lycopene accumulation in the flesh still remains elusive (Li et al. 2020). In another QTL mapping study, a major locus was identified on chromosome 4, with a lycopene β-cyclase being the most probable candidate gene underlying the QTL (Wang et al. 2019). A SNP variation leading to a nonsynonymous substitution in the candidate amino acid sequence may explain differences in protein levels or functionality between red- and yellow-flesh lines (Wang et al. 2019). Out of curiosity, some watermelon accessions accumulate mainly the orange carotenoid β-carotene, the main source of provitamin A, instead of lycopene (Tadmor et al. 2005). Genome-wide mapping in a population segregating for flesh colour led to the identification of a single major QTL on chromosome 1 (Branham et al. 2017). No clear candidate gene could be identified, however, several TF were located within the confidence interval. Furthermore, a second still unidentified locus acting epistatically with the major QTL may explain flesh colour segregation in watermelon (Branham et al. 2017).

Sweet potato (Ipomoea batatas) orange-flesh varieties are able to accumulate high levels of β-carotene. However, a negative correlation between this important source of provitamin A and starch is observed, due to the linkage between phytoene and sucrose synthase enzymes, which catalyse the rate-limiting reactions of carotenoid and starch biosynthetic pathways, respectively. Indeed, QTL mapping in a biparental F₁ progeny identified two loci on linkage groups 3 and 12 affecting both β-carotene and starch content (Gemenet et al. 2020). Interestingly, both phytoene and sucrose synthases were found within the locus interval on chromosome 3, suggesting that physical linkage, instead of pleiotropic effects, was responsible for the negative correlation between starch and carotenoid accumulation. In addition, fine mapping on chromosome 12 allowed to pinpoint the Orange (IbOr) gene as the candidate underlying the identified QTL (Gemenet et al. 2020). Or gene is involved in chromoplast biogenesis and, additionally, positively affects phytoene synthase protein levels (Park et al. 2016; Sun et al. 2018; Zhou et al. 2015). As no correlation between IbOr expression and β-carotene content was observed in sweet potato, it can be hypothesized that nucleotide polymorphisms leading to protein changes may impact chromoplast biogenesis and phytoene synthase levels (Gemenet et al. 2020). In fact, melon Or gene, which co-segregate with flesh colour in F₂ and backcross populations, present two major haplotypes, with a single SNP inducing an amino acid change from a highly conserve arginine to a histidine and responsible for β-carotene accumulation (Table 2) (Tzuri et al. 2015). Carrot (Daucus carota) Or gene was also identified by GWAS analysis in 674 domesticated and wild carrot accessions. A nonsynonymous mutation was found in Or exon 5, causing a serine to leucine change in the amino acid sequence. Interestingly, Or sequence fell within a domestication sweeps, suggesting that alleles favouring carotene accumulation have been fixed by artificial selection (Ellison et al. 2018).

Another omic approach, combining metabolite profiling with proteomic analysis to decipher carotenoid accumulation in banana (Musa spp.) pulp, is worth mentioning (Heng et al. 2019). Both proteome and metabolome analyses outlined a relation between enhanced glycolysis activity and carotenoid accumulation, suggesting that carotenogenesis may be supported by carbohydrate metabolism. Together, these results imply that multi-target approach may be indispensable to improve carotenoid content in crops (Heng et al. 2019).

Transcriptomic, GWAS and metabolomic approaches to understand flavour quality changes during domestication and intensive breeding: tomato as an example

Tomato (Solanum lycopersicum) is an important crop for the human diet, being a source of fiber and micronutrients, in addition to be a study model for fruit development and ripening (Giovannoni 2001; Sauvage et al. 2014). Due to intensive breeding in the last centuries, a reduction of the genetic diversity through two successive bottlenecks has led to the loss of important fruit quality parameters, such as flavour and nutritional value (Bai and Lindhout 2007; Bauchet et al. 2017). As a consequence, consumer complaint in the last decades about flavourless tomatoes has become a major issue for breeding programs, and the need to understand the genetic mechanisms underlying this quality loss is shared by both academic research and food industry. In addition, the combination of metabolomics, transcriptomics and mGWAS is a key approach to understand the fruit metabolic and quality changes which have occurred along domestication and improvement (Zhu et al. 2018).

QTL approach in biparental populations has shed light on the genetic architecture of a series of valuable traits for fruit quality improvement (Causse 2004; Goulet et al. 2012; Mageroy et al. 2012; Tieman et al. 2006; Tikunov et al. 2013; Zanor et al. 2009b); however, as QTL mapping does not take into account the entire genetic diversity present in germplasm collections, GWAS strategy, which allows the screening of a wide range of accessions, has been more recently used to decipher the inheritance of important fruit metabolic traits (Bauchet et al. 2017; Ruggieri et al. 2014; Sauvage et al. 2014; Tieman et al. 2017; Ye et al. 2019; Zhu et al. 2018). Additionally, combination of transcriptomic and metabolomic profiling also depicted some important genetic regulation of quality traits (D’Angelo et al. 2019; D’Esposito et al. 2017).

GWAS for primary metabolites, in particular, sugars and organic acids, have been performed in several hundreds of tomato genotypes, including domesticated S. lycopersicum, wild S. lycopersicum variant cerasiforme and the most closely related wild species S. pimpinellifolium accessions (Bauchet et al. 2017; Sauvage et al. 2014; Tieman et al. 2017; Ye et al. 2019). The identification of candidate genes underlying the associated loci can be crucial for modern tomato flavour improvement, the main consumer complaint, as tomato flavour is defined by sugars, acids and a specific set of volatile compounds (Baldwin et al. 2008; Tieman et al. 2012).

Sauvage et al. (2014) identified 44 loci for 19 primary metabolic traits, some of the peak SNPs targeting previously described or putative candidate genes. Indeed, in one of the detected loci associated to soluble solid content was found lin5 gene, which encodes a cell wall invertase, a key enzyme for fruit sugar uptake and a major locus controlling sugar content in this organ (Table 2) (Fridman et al. 2004). Lin5 was previously identified in a QTL study for fruit sugar content and further validated by stable silencing in tomato plants (Baxter et al. 2005; Fridman et al. 2004; Vallarino et al. 2017; Zanor et al. 2009a). Bauchet et al. (2017), which also implemented GWAS for both primary and secondary metabolites in 300 tomato accessions and detected 79 loci for 33 metabolic traits, were able to validate lin5 gene as being responsible for fructose, glucose and soluble solid content. Furthermore, Tieman et al. (2017) identified by GWAS performed in 398 tomato accessions two loci, on chromosomes 9 and 11, associated with glucose and fructose levels. Interestingly, they detected the same locus on chromosome 9, and which corresponds to lin5 gene by QTL mapping in a F₂ population, derived from a cross between a variety producing small but flavourful fruits and a modern large-fruit inbred line (Tieman et al. 2017). This strong candidate for tomato sugar content was validated at molecular level, as the lin5 SNP resulted in an amino acid change (Asn to Asp) in the protein sequence. Stable transgenic tomato plants overexpressing one or the other lin5 variants allowed to confirm that significant differences in sugar content were observed based on the amino acid change in lin5 sequence (Tieman et al. 2017). Furthermore, an explanation about flavour loss along domestication could be provided, as the two hexose-associated loci on chromosomes 9 and 11 were located within domestication and improvement sweeps. By continuously selecting for fruit size, breeders have set the combination of alleles at the two loci which result in the lowest sugar content, and which are found in almost all the modern cultivars included in the GWAS analysis (Tieman et al. 2017).

Apart from main sugars present in ripe tomatoes, the predominant organic acids, malate and citrate, are essential for fruit taste perception (Etienne et al. 2013). Interestingly, Sauvage et al. (2014) identified two significant SNPs located on chromosomes 2 and 6 linked to malate content, explaining 39% of the metabolic variation. Interestingly, the SNP on chromosome 6 was found within the coding sequence of an unknown protein, but two aluminium-activated malate transporters-like (likely a unique gene) were detected in the vicinity (Bauchet et al. 2017; Sauvage et al. 2014). Deeper analysis of chromosome 6 candidate region indicated that the aluminium-activated malate transporter-like expression correlated with malate content in eight tested accessions, and that a nonsynonymous mutation in a transmembrane domain may explain this correlation (Table 2). While a polymorphism in the gene encoding the unknown protein cannot be discarded in impacting malate levels in the assessed accessions (Bauchet et al. 2017), another GWAS analysis in 272 tomato accessions, combined with linkage mapping and transgenic complementation approaches, allowed to outline that expression of aluminium-activated malate transporter9 on chromosome 6is the main determinant for tomato fruit malate levels in cultivated accessions (Ye et al. 2017). Indeed, a 3pb deletion in the promotor sequence resulted in higher aluminium-activated malate transporter9 expression, increasing fruit malate content (Ye et al. 2017, Table 2).

Another important acid, for its essential function in human health and which has to be obtained through fruit and vegetable ingestion is ascorbate (vitamin C). Ascorbate synthesis takes place mainly through the D-Mannose/L-Galactose pathway, also known as the Smirnoff-Wheeler (SW) pathway, although alternative routes have also been proposed, through myo-inositol and D-galacturonate. In fruits, both the SW and the galacturonate pathways have been described to be functional, depending on the ripening stage (Fenech et al. 2019). Interestingly, in tomato, ascorbate levels showed the most significant differences between S. lycopersicum, S. lycopersicum variant cerasiforme and S. pimpinellifolium accessions, being higher in the wild species genotypes (Sauvage et al. 2014; Ye et al. 2019) and suggesting the possibility of lost alleles during domestication, which could be used by breeding programs to improve this important trait. In fact, a SNP associated with fruit ascorbate content was found 423 kb upstream of a monodehydroascorbate reductase (NADH)-like protein on chromosome 9 (Sauvage et al. 2014). More recently, Ye et al. (2019) identified a SNP associated with ascorbate content on chromosome 9 by GWAS on 302 tomato accessions, and explaining 15.9% of the variation in this metabolic trait. By measuring the expression of possible candidate genes in accessions showing contrasting ascorbate levels, a bHLH TF, SlbHLH59, could be highlighted as the likely candidate underlying this locus. Furthermore, a deeper analysis of the gene sequence outlined four different haplotypes, with haplotype 4, found in most S. pimpinellifolium tested accessions, harbouring nine consensus polymorphisms. In particular, an 8 pb-insertion in the promotor region resulted in the formation of a 5′UTR Py-rich stretch motif associated with elevated expression of downstream genes. The candidate was then functionally validated by stable RNAi silencing in tomato plants (Table 2) (Ye et al. 2019). In addition, expression analysis in SlbHLH59-silenced and overexpressed lines and yeast two-hybrid allowed to establish that SlbHLH59 protein can directly bind to the promotor region of enzymes of the D-Man/L-Gal ascorbate biosynthetic pathway and modulate their expression during tomato fruit development (Ye et al. 2019). Decreased ascorbate fruit content was selected during domestication, as suggested by the analysis of the InDel present in SlbHLH59 promotor sequence in more than 500 accessions. Indeed, the insertion frequency was high in S. pimpinellifolium genotypes (83.3%), and sharply decreased in S. lycopersicum cerasiforme variants (16.3%) and modern varieties (1.8%). The low nucleotide diversity in the latter ones further confirmed that SlbHLH59 was located in a domestication and improvement sweep (Lin et al. 2014).

Ruggieri et al. (2014) performing a GWAS on landraces, vintage and modern S. lycopersicum varieties, found three associations for ascorbate content, different from the ones identified by Sauvage et al. (2014) and Ye et al. (2019). In the genomic region in linkage disequilibrium with one of the SNP (on chromosome 3) a cluster of pectinesterases, pectate lyase and polygalacturonase genes were identified. Furthermore, another SNP on chromosome 5 was found to be associated with ascorbate content and was located in the sequence of an ethylene-responsive transcription factor 1 (Ruggieri et al. 2014). Interestingly, a previous study using introgression lines from a cross between S. lycopersicum ‘M82′ and S. pennellii, outlined a pectinesterase and two polygalacturonases to be upregulated in a line with higher levels of ascorbate (Di Matteo et al. 2010). Furthermore, several ethylene-related transcripts showed increased expression in this line, suggesting that ascorbate accumulation in tomato fruit may occur through pectin degradation and is triggered by ethylene signalling, which corroborate the GWAS analysis by (Ruggieri et al. 2014). Taken together, these data imply that both the predominant SW pathway and the alternative D-galacturonate pathway are important for the regulation of tomato fruit ascorbate content.

A combined metabolomic and transcriptomic study in Andean landrace accessions and the wild specie S. pimpinellifolium shed light upon the domestication of important quality traits, such as fruit aroma. In particular, significant differences between the wild ancestor and the Andean varieties were observed in the content of branched-chain amino acids and threonine, important precursors of tomato aroma volatiles. In particular, isoleucine and its substrate, threonine, was increased in five Andean accessions, and correlated with a lower expression of threonine aldolase gene, favouring both essential amino acid content and aroma formation (D’Angelo et al. 2019). Furthermore, one of the accessions showed a remarkable flavour when compared to the other genotypes; interestingly, a gene involved in volatile synthesis, TomLoxC was highly upregulated in this cultivar (D’Angelo et al. 2019, 2018). In fact, TomLoxC was outlined in a QTL study in a recombinant inbred line population, as controlling lipid- and carotenoid-derived volatile content (Table 2) (Gao et al. 2019). A nonreference allele for TomLoxC was identified in a pan-genome analysis, present in 91.2% of S. pimpinellifolium accessions and only in 2.2% of S. lycopersicum heirlooms, implying a strong negative selection during domestication. Interestingly, modern tomato varieties showed increased frequency of the rare allele (7.2%) compared to heirlooms, possibly as a consequence of recent introgressions from wild species into breeding cultivars (Gao et al. 2019).

Among volatiles impacting tomato aroma, the phenylpropanoid phenylethanol and its precursor phenylacetaldehyde have a positive effect on the fruit flavour, conferring sweet and fruity aromas (Baldwin et al. 2008). A highly significant association was found for phenylethanol and phenylacetaldehyde on chromosome 4. In this region, a glucosyltransferase was identified as the most probable candidate gene (Bauchet et al. 2017). Indeed, an allelic variation on the third exon of the gene appeared to be nonsynonymous, and significant variation of glycosides were observed according to the different alleles. Taken together, this result suggests that the release of volatiles may be regulated by glycosylation. Interestingly, glycosyltransferases have been previously described to play an important role in volatile conjugation (Bauchet et al. 2017; Tikunov et al. 2013; Yauk et al. 2014). However, further validation of the candidate is needed to confirm its impact on the two volatiles.

A recent study also mapped a QTL for phenylpropanoid-derived volatiles (2-phenylethanol, phenylacetaldehyde and 1-nitro-2-phenylethane) on chromosome 4, using a diversity panel of 94 cultivars and several F₂ and F₆ populations of a half diallel cross (Tikunov et al. 2020). Near isogenic lines, harbouring the marker alleles associated to the genomic region on chromosome 4 for either low or high phenylpropanoid volatiles, allowed to confirm the correlation of this locus with fruity and rose-hip aroma (Tikunov et al. 2020). Fine-mapping the locus on chromosome 4 narrowed down to 11 genes in the associated region, and one of them showed high similarity to a 3-methyl-2-oxobutanoate dehydrogenase, an enzyme with decarboxylase activity, and which may be involved in the decarboxylation of phenylalanine, to produce the common precursor of the three volatiles (Table 2). Gene editing of the candidate gene, named FLORAL4, by CRISPR led to an important reduction in phenylalanine-derived volatiles and to a less extent in 3-methylbutanol levels. On the other hand, 3-methylbutanol precursor, the branched-chain amino acid leucine, was significantly increased, while no differences were observed in phenylalanine content. Even if gene editing validated the effect of FLORAL4 deleterious mutations in planta, the decarboxylase activity of the gene could not be verified in vivo (Tikunov et al. 2020).

Other phenylpropanoid-derived volatiles, guaiacol and methylsalicylate, have a negative impact on consumer liking, being associated with ‘smoky’ or ‘medicinal’ aroma (Zanor et al. 2009b). Tieman et al. (2017) detected a locus on chromosome 9 linked to the two metabolically-related volatiles, and suggested that E8, a specific highly-expressed ripening-related gene, could be the candidate based on the analysis of silenced tomato lines, even if the molecular mechanism remained unclear. On the other hand, using a combination of omics approaches, Tikunov et al. (2013, 2020) highlighted the role of a non-smoky glycosyltransferase (NSGT1), also located on chromosome 9, and responsible for the volatile release. Very recently, and taking advantage of the new long-read technology sequencing, a collection of 100 tomato accessions was sequenced, shedding light upon structural variants underlying important domestication traits, including the smoky locus (Alonge et al. 2020). Indeed, this new assembly allowed a better resolution of the locus sequence and to fill the gaps flanking E8 with NSGT1 present in the previous reference genome. Furthermore, haplotype analysis seemed to confirm that multiple mutant alleles of nsgt1 underlie the natural variation in guaiacol and methylsalicylate volatile content (Alonge et al. 2020). GWAS analysis and the generation of an F₂ population between two accessions segregating for haplotype V (23-kpb deletion removing both E8 and NSGT1) and functional NSGT1 analysis validated that the deletion of both E8 and NSGT1 results in higher guaiacol and methylsalicylate content (Table 2) (Alonge et al. 2020).

Another group of flavour-important volatiles, such as geranylacetone and 6-methyl-5-hepten-2-one (MHO) are carotenoid-derived compounds (apocarotenoids). Tieman et al. (2017) identified one associated locus for MHO, four for geranylacetone and two for both compounds. Interestingly, by comparing the frequency of alleles for both volatile content in wild, heirloom and modern accessions, they found that allelic combinations were progressively lost along domestication. While breeders have selected the two alleles responsible for the highest MHO content, the two most frequent alleles for geranlyacetone were not the most advantageous ones for geranylacetone levels (Tieman et al. 2017). Indeed, breeders have indirectly favoured increased MHO levels, by enhancing lycopene content, the main carotenoid pigment responsible for the bright red ripe fruit colour and MHO precursor. On the opposite, they could not easily select for geranylacetone accumulation, as it is derived from the oxidative cleavage of carotenoids present in small amounts and which do not contribute to the fruit colour (Tieman et al. 2017). In addition, Tieman et al. (2017) could identify a phytoene synthase, the enzyme catalysing the first step of carotenoid pathway, in a region where a strong association with both apocarotenoids was detected.

Conclusions

Current challenges in modern breeding include increasing yield and quality under climatic changing conditions to secure global food production. From one hand, there is an urging need to improve cereal and legume nutrient composition, as a key strategy to fight against health issues in developing countries. On the other hand, recent advances in multiomic platforms is allowing to understand better the complexity of crop and fruit metabolism and genetics, in order to increase their organoleptic and nutraceutics quality, as requested by consumers. While metabolomic and/or proteomic approaches enables to outline genotypes with quality-related characteristics, we can expect that deep sequencing combined with other omics platforms will increased our knowledge about the genetic architecture of quality traits. In a foreseeable future, this information will be easily translated to breeding programs for crop improvement, after trait (phenotype)-genotype validation, as it is already occurring for tomato fruit (Li et al. 2018a; Rodríguez-Leal et al. 2017; Rothan et al. 2019). In the sense, it is worth mentioning that the rapid progresses made in multiomic technologies are matched with advances in gene validation and editing techniques. Indeed, recent development of gene editing applications, and in particular CRISPR/Cas9 system, will allow base editing or targeted gene replacement to facilitate allele substitution, where conventional marker-assisted selection may not be achievable (Gaston et al. 2020; Rothan et al. 2019).

Furthermore, as breeding targets can be shared between crop species, mining tomato data can be a fundamental approach to help the exploitation of less studied and genetically more complex species, as suggested for the octoploid cultivated strawberry (F. x ananassa) by a recent review (Gaston et al. 2020). Other future perspectives may consist of GWAS meta-analysis, as recently performed in tomato by Zhao et al. (2019), which identified 37 promising candidate genes for flavour improvement and in particular sugar content. This strategy, together with marker-assisted selection or gene editing, may be implemented for other crop species, and will allow to substitute undesirable alleles in modern cultivars by favourable alleles for fruit nutrition and flavour traits.