A negative MYB regulator of proanthocyanidin accumulation, identified through expression quantitative locus mapping in the grape berry

Authors


Summary

  • Flavonoids are secondary metabolites with multiple functions. In grape (Vitis vinifera), the most abundant flavonoids are proanthocyanidins (PAs), major quality determinants for fruit and wine. However, knowledge about the regulation of PA composition is sparse. Thus, we aimed to identify novel genomic regions involved in this mechanism.
  • Expression quantitative trait locus (eQTL) mapping was performed on the transcript abundance of five downstream PA synthesis genes (dihydroflavonol reductase (VvDFR), leucoanthocyanidin dioxygenase (VvLDOX), leucoanthocyanidin reductase (VvLAR1), VvLAR2 and anthocyanidin reductase (VvANR)) measured by real-time quantitative PCR on a pseudo F1 population in two growing seasons.
  • Twenty-one eQTLs were identified; 17 of them did not overlap with known candidate transcription factors or cis-regulatory sequences. These novel loci and the presence of digenic epistasis support the previous hypothesis of a polygenic regulatory mechanism for PA biosynthesis.
  • In a genomic region co-locating eQTLs for VvDFR, VvLDOX and VvLAR1, gene annotation and a transcriptomic survey suggested that VvMYBC2-L1, a gene coding for an R2R3-MYB protein, is involved in regulating PA synthesis. Phylogenetic analysis showed its high similarity to characterized negative MYB factors. Its spatiotemporal expression profile in grape coincided with PA synthesis. Its functional characterization via overexpression in grapevine hairy roots demonstrated its ability to reduce the amount of PA and to down-regulate expression of PA genes.

Introduction

Flavonoids constitute a diverse family of plant secondary metabolites (Harborne & Baxter, 1999). They are largely present in fruits, vegetables and plant-derived foods and are major determinants of food organoleptic quality with beneficial effects on human health (Bravo, 1998; Harborne & Baxter, 1999; Harborne & Williams, 2000). In grapevine (Vitis vinifera), the three main flavonoids are flavonols, anthocyanins and flavan-3-ols, among which proanthocyanidins (PAs), that is, flavan-3-ol oligomers and polymers, are the most abundant flavonoids in berries (Downey et al., 2004; Mané et al., 2007).

In the grape berry, PAs are synthesized from anthesis to véraison (the midpoint of berry development separating the green stage from ripening). There are four main constitutive subunits for grape PAs: (+)-catechin, (−)-epicatechin, (−)-epigallocatechin and (−)-epicatechin-3-O-gallate, with traces of (+)-gallocatechin. PA composition differs between berry compartments: total content is usually higher in seeds while polymer size is much larger in skin (Prieur et al., 1994; Souquet et al., 1996). Regarding constitutive subunits, (−)-epigallocatechin is a major skin PA subunit (Souquet et al., 1996), while it is undetectable in seeds (Prieur et al., 1994); large quantities of (−)-epicatechin-3-O-gallate are found in seeds, but only traces in skin (Prieur et al., 1994; Souquet et al., 1996).

Grape flavonoid synthesis genes have been isolated through homologous cloning (Fig. 1; Sparvoli et al., 1994; Jeong et al., 2004; Bogs et al., 2005, 2006; Castellarin et al., 2006; Jeong et al., 2006). Anthocyanidin reductase (ANR) and leucoanthocyanidin reductase (LAR) are specific enzymes for flavan-3-ol monomer synthesis (Fig. 1, Tanner et al., 2003; Xie et al., 2003). ANR is responsible for the synthesis of (−)-epi(gallo)catechin, while LAR is responsible for the synthesis of (+)-(gallo)catechin. One and two isogenes have been cloned for VvANR and VvLAR, respectively (Bogs et al., 2005). Despite advances in the isolation of grape PA synthesis genes, the first de novo genetic analysis of grape PA composition via quantitative trait locus (QTL) mapping was published only recently (Huang et al., 2012). QTLs were identified for skin and seed PA composition with variables such as total content, subunit amount and degree of polymerization. Results showed that skin and seed QTLs differed in number, position, involvement of epistasis and allelic effect.

Figure 1.

The grape (Vitis vinifera) proanthocyanidin (PA) synthesis pathway. The PA pathway, and pathways for stilbenes, lignins and anthocyanins, are branch pathways of the general phenylpropanoid pathway, derived from the same initial substrate, phenylalanine. Arrows indicate the enzymatic steps from the entry substrate phenyalanine to the end product, proanthocyanidins. The background colour highlights unknown substrates and mechanisms for PA polymerization. Genes coding for enzymes of the general phenylpropanoid/flavonoid pathway are as follows: PAL, phenylalanine ammonia lyase; C4H, cinnamate-4-hydroxylase; 4-CL, 4-coumarate:coA ligase; CHS, chalcone synthase; CHI, chalcone isomerase; F3H, flavanone 3-hydroxylase; F3′H, flavonoid 3′ hydroxylase; F3′5′H, flavonoid 3′-5′ hydroxylase; DFR, dihydroflavonol reductase; LDOX, leucoanthocyanidin dioxygenase; LAR, leucoanthocyanidin reductase; ANR, anthocyanidin reductase.

The contrasted PA composition and distinct QTL distribution in different compartments of berries suggest a fine-tuned regulatory mechanism for grape PA biosynthesis. However, knowledge about this regulation is still incomplete across plant species, including grapevine. To date, we know that MYB-bHLH-WD40 protein complexes control the expression of flavonoid genes (reviewed by Koes et al., 2005; Ramsay & Glover, 2005). MYB proteins of this complex are most often R2R3 MYB, characterized by two imperfect repeats of the DNA-binding motifs R2 and R3 in their N-terminal side. The C-terminal side of the R2R3 MYB proteins is more divergent and is thought to be responsible for the specific activation of target genes (Martin & Paz-Ares, 1997; Ramsay & Glover, 2005). bHLH proteins are characterized by the basic helix-loop-helix domain, where the basic region is responsible for sequence-specific DNA binding (Massari & Murre, 2000). WD40 facilitates MYB–bHLH interactions without intrinsic transcription factor function (Ramsay & Glover, 2005). The first regulatory complex specifically regulating PA synthesis gene was identified in Arabidopsis (Baudry et al., 2004). This protein complex specifically activates the expression of PA synthesis genes, such as dihydroflavonol reductase (AtDFR), leucoanthocyanidin dioxygenase (AtLDOX) and AtBAN (BANYULS gene) (encoding ANR; Nesi et al., 2001). The specific activation pattern appears to be mainly conferred by the MYB protein AtTT2. Some R2R3 MYB factors were also identified as negative regulators of the general phenylpropanoid pathway (Tamagnone et al., 1998; Jin et al., 2000; Aharoni et al., 2001; Colquhoun et al., 2011; Shen et al., 2012). Some of them were specific to one branch of the pathway: PvMyb4 negatively regulates genes from the lignin pathway but not those of the flavonoid pathway (Shen et al., 2012).

Based on Arabidopsis PA-related genes, several grapevine regulators were isolated through homologous cloning, including two PA-specific MYBs, VvMYBPA1 and VvMYBPA2 (Bogs et al., 2007; Terrier et al., 2009). Two other MYBs, VvMYB5a and VvMYB5b, may indirectly activate PA synthesis through activation of upstream biosynthesis genes (Deluc et al., 2006, 2008). One bHLH protein, VvMYC1, was demonstrated to activate the VvANR promoter in cooperation with VvMYBPA1 (Hichri et al., 2010). However, homologous cloning appears insufficient to obtain a complete picture of grapevine PA regulation because of the greater diversity in PA subunits (four in grapevine versus one in Arabidopsis) and the spatiotemporally diverse PA composition compared with Arabidopsis, where PAs accumulate only in the seed coat.

In this context, expression quantitative trait locus (eQTL) mapping is a possible alternative approach for de novo identification of regulatory loci. Gene expression level is considered as a phenotype (sometimes referred to as an ‘e-trait’) and classical QTL analyses allow mapping of loci involved in expression variation. Several eQTL studies have been reported in plants, including Arabidopsis (DeCook et al., 2006; Kliebenstein et al., 2006; Keurentjes et al., 2007; Sonderby et al., 2007; Wentzell et al., 2007; West et al., 2007), barley (Hordeum vulgare; Potokina et al., 2008a,b), cotton (interspecific cross, Gossypium hirsutum × Gossypium barbadense; Claverie et al., 2012), maize (Zea mays; Swanson-Wagner et al., 2009; Holloway et al., 2011), wheat (Triticum aestivum; Jordan et al., 2007), eucalyptus (interspecific pseudobackcross, Eucalyptus grandis × F1 hybrid (E. grandis × E. globulus); Kirst et al., 2004, 2005), and poplar (interspecific pseudobackcross, Populus trichocarpa × Populus deltoides; Drost et al., 2010). Although most of them surveyed genome-wide gene expression, two of these studies have demonstrated that studying a subset of genes belonging to a same pathway can be valuable (Kliebenstein et al., 2006; Wentzell et al., 2007).

In this study, we aimed to identify novel genomic regions involved in regulation of PA synthesis genes through eQTL mapping. We previously showed that it is feasible to use the eQTL approach with samples collected from the field based on knowledge of berry development and the target pathway (Huang et al., 2013). Following the same logic, transcript levels of five PA synthesis genes (VvDFR, VvLDOX, VvLAR1, VvLAR2 and VvANR) were recorded in two growing seasons on a pseudo-F1 population, S × G, derived from a cross between the Syrah and Grenache cultivars. Transcript levels were mapped using both single-year data and genotypic averages estimated on the 2-yr data set. We then focused on one locus where co-localized both eQTLs and PA QTLs (Huang et al., 2012). By coupling function annotation and transcriptomic data (Terrier et al., 2009), a gene of the MYB family, named VvMYBC2-L1 (Matus et al., 2008), was selected at this locus. Phylogenetic analysis showed its high similarity to other negative MYB regulators. The spatiotemporal expression profile of this MYB and other PA-related genes supported its role in regulating the PA pathway. Functional validation of VvMYBC2-L1 through transformation of grapevine hairy roots indicated that it was a negative regulator of PA synthesis and probably other branches of the phenylpropanoid pathway.

Materials and Methods

Plant material and sample collection

The 191-individual pseudo-F1 population (S × G) was derived from a cross between two grape (Vitis vinifera, ssp. sativa) cultivars, Syrah and Grenache. S × G was previously used in linkage map construction and QTL mapping (Doligez et al., 2006; Fournier-Level et al., 2009; Huang et al., 2012). Details regarding the experimental design and cultural practice have been published previously in Huang et al. (2012).

Grapes were harvested in 2009 and 2010. In order to reduce developmental and environmental differences between genotypes, we paid attention to the following points. Berries were collected at 21–26 d after anthesis according to genotype. This interval corresponded to 35% of the time from the date of anthesis to véraison and was estimated for each genotype in three consecutive years of phenological observations (2005–2007); only berries at the extremity of clusters were harvested; harvests were performed between 07:00 and 10:45 h to avoid expression differences related to circadian variation. Before sampling, the date of anthesis was noted for each genotype when > 50% of flowerhoods had fallen. For each genotype, as many inflorescences as possible were tagged on the same day with plastic ribbons around cluster extremities. Whole berries from three to 14 cluster extremities were harvested and flash-frozen in liquid nitrogen in the vineyard.

Organs (berries, leaves, tendrils, and roots) from the Syrah cultivar were collected at several developmental stages. Plants were grown in the SupAgro-INRA vineyard in Montpellier, France. Young leaves corresponded to leaves explanted from the third node below the shoot tips, with a mean weight of 0.3 g. Old leaves corresponded to fully expanded leaves sampled from the mature shoot part, with a mean weight of 2.8 g. Young and ripe berries for dissection into skin, pulp and seed were collected 18 and 99 d after flowering, respectively.

RNA extraction and real-time quantitative PCR

Total RNA was extracted from whole berries using the RNeasy Plant Mini kit (Qiagen, Courtaboeuf, France) with minor modifications (Gomez, 2009). The quality and quantity of the RNA samples were checked with agarose electrophoresis and Nanodrop 3300 (ThermoScientific, Wilmington, NC, USA).

Reverse transcription, real-time quantitative PCR (RT-qPCR) and expression normalization were performed as described in Huang et al. (2013). Before RT-qPCR, the absence of polymorphism at the priming sites was checked in parental cultivars by Sanger sequencing. The amplification efficiency of primers for RT-qPCR was verified and was comparable for all primer pairs. Primers for RT-qPCR are summarized in Supporting Information Table S1. Primer pairs for stilbene synthases are not isogene specific (Dai et al., 2012).

Statistical analysis

All statistical analyses were performed with R software (R-Development-Core-Team, 2008). When the coefficient of variation of the three technical replicates of RT-qPCR exceeded 0.4, aberrant replicate was suppressed. We performed mixed model fitting in order to extract genotypic averages based on the 2-yr data. Two models were fitted with the lme4 package (Bates et al., 2008) for gene expression data: a model including a random genotypic effect and another with an additional fixed year effect. Model fitting with raw data showed violation of mixed model assumption (i.e. residual and random predictors c. N(0,σ2)). Data were thus log-transformed before mixed model fitting. Based on the Bayesian information criterion (BIC), the best-fit model for expression variation of VvDFR, VvLDOX, VvLAR2 and VvANR was the model based on random genotypic effect and fixed year effect, and for VvLAR1 it was the model with only random genotypic effect. Based on the best fitting model, broad sense heritability (H2) was calculated as inline image, where the variance component was estimated by the restricted maximum likelihood method. Best linear unbiased predictors (BLUPs) were also extracted as genotypic averages for further analysis.

Quantitative trait locus mapping

QTL mapping was performed on single-year data and genotypic averages with the R/qtl package (Broman et al., 2003). Three linkage maps were used in QTL detection, as previously reported (Huang et al., 2012): parental maps, which assessed the marker polymorphisms inside a given parental cultivar, and a consensus map, which assessed the marker polymorphism between parental cultivars. QTL identification was carried out by the regression method (Haley & Knott, 1992) implemented in a multiple QTL model (‘stepwise’ function, ‘hk’ option) at a genome-wide error rate of 0.05. This approach uses forward/backward selection to identify a multiple-QTL model with inclusion of both main effect QTLs and pairwise interactions (Manichaikul et al., 2009). More details regarding multiple QTL mapping parameters have been published in Huang et al. (2012).

Candidate gene identification

To identify candidate genes, cDNA between markers flanking the target eQTL (12X genome sequence project; http://www.genoscope.cns.fr/externe/GenomeBrowser/Vitis/) were selected and manual annotation was conducted by local BLAST (algorithm blastx; threshold E-value of e−25) against the UniProt data base (release 2010_04). Only alignments of > 40% of the total cDNA length and of > 40% of identity with the reference sequence were retained and the best scoring gene was retained for functional annotation. The list of these annotated positional candidate genes was then compared with genes differentially expressed in one of the three transcriptomic studies obtained from a 14K microarray: grapevine wild-type hairy roots versus hairy roots overexpressing VvMYBPA1 or VvMYBPA2 (Terrier et al., 2009).

Gene cloning and function validation through plant transformation

The coding region of the selected candidate gene, VvMYBC2-L1, was amplified from young whole berry cDNA (cv Maccabeu) with high-fidelity Taq polymerase (Advantage®-HF 2 PCR KIT; Clontech, Saint-Quentin-en-Yvelines, France). The PCR reaction contained 20 ng of cDNA and 200 nM of each primer (Table S1) in a final volume of 25 μl. The resulting PCR product was purified (Wizard® SV Gel and PCR Clean-Up System; Promega, Charbonnières, France), directionally cloned into the pENTR/D-TOPO vector (Invitrogen, Courtaboeuf, France) and transferred into the binary vector pH2GW7 through LR recombination (LR clonase II; Invitrogen) to yield the VvMYBC2-L1 over-expressor construct. This construct was electroporated into Agrobacterium rhizogenes strain A4, obtained from Collection Française de Bactéries Phytopathogènes (http://www-intranet.angers.inra.fr/cfbp/). The Maccabeu sequence of VvMYBC2-L1 was deposited in GenBank (accession number JX050227).

Induction and culture of transgenic hairy roots in grapevine (cv Maccabeu) were performed according to Torregrosa & Bouquet (1997) with modifications described in Gomez et al. (2009). Hairy roots were harvested and flash-frozen in liquid nitrogen for DNA and RNA extraction and composition analysis. Hairy roots without the transgene were used as a control. The PA composition of hairy roots was assessed using high-performance liquid chromatography (HPLC) as described by Verries et al. (2008). Stilbenes were analysed on the injections without phloroglucinolysis and detected at 320 nm.

Phylogenetic tree construction

Full-length amino acid sequences of MYB factors from several species were retrieved from public databases. For phylogenetic tree construction, multiple alignment of full-length protein sequences was performed with the ClusalW algorithm implemented in the mega package version 5 (Tamura et al., 2011) with default parameters (except for pairwise and multiple alignment, gap open penalty = 10 and gap extension penalty = 0.5). The phylogenetic tree was then constructed using the neighbour-joining method by the mega program.

Results

Transcript data for PA synthesis genes in the mapping population

The transcript abundance of five grape PA synthesis genes (VvDFR, VvLDOX, VvLAR1, VvLAR2 and VvANR) was recorded by RT-qPCR on whole-berry cDNA of the S × G population and parental cultivars for two consecutive years (2009–2010). Transgressive segregation was observed for all five transcripts (Fig. S1). Large variation of broad-sense heritability (H2) was observed among e-traits, with H2 values of 0.05 for VvLDOX, 0.25 for VvDFR, 0.28 for VvANR, 0.65 for VvLAR1 and 0.75 for VvLAR2 (Fig. S1).

The relationship between e-traits was analysed by pairwise correlation tests. In the same year, strong positive correlations were observed between e-traits, with Pearson correlation coefficients ranging from 0.51 to 0.71 in 2009 and from 0.36 to 0.83 in 2010 (< 0.001; Fig. 2a). VvLAR2 was the only gene showing weak or no correlation with the other genes (0.10–0.21 in 2009 and 0.05–0.31 in 2010; Fig. 2a); the same tendency was also observed for genotypic averages (Fig. 2b). Except for VvLDOX, transcript levels were positively and significantly correlated between 2009 and 2010; Pearson correlation coefficients varied between 0.26 and 0.76 (< 0.001; Fig. 2a).

Figure 2.

Pairwise Pearson correlation matrix of relative expression between proanthocyanidin (PA) synthesis genes in the Vitis vinifera S × G population based on (a) single-year data and (b) genotypic averages. The Pearson correlation coefficient (ρ) is shown with a colour code indicating the level of significance. For single-year data (a), the pairwise correlations were estimated based on log-transformed normalized data and bold grey lines are used to help the reading of the 2-yr data.

eQTL for PA synthesis genes

In order to identify loci involved in the control of PA transcript variation, we mapped the PA transcript abundance on three published genetic maps (Huang et al., 2012). The genomic distributions of all identified loci for 2009, 2010 and genotypic averages are summarized in Fig. 3. In total, 21 eQTLs were identified (Fig. 3; eQTL positions are shown in Fig. S2; Table S3). In addition, two-locus epitasis was detected for expression of VvLAR2 measured in 2009 and for genotypic averages of VvDFR.

Figure 3.

Three panels showing the genome-wide distribution of expression quantitative trait loci (eQTLs) for proanthocyanidin (PA) synthesis genes mapped on (a, b) single-year data for (a) 2009 and (b) 2010, and (c) genotypic averages. The grey vertical lines separate the Vitis vinifera genome into chromosomes 1–19 with cM progressing from left to right along the x-axis. Each row represents the results for a target gene whose position is highlighted by a yellow rectangle with the width proportional to the interval of flanking markers on the consensus map. Horizontal bars represent identified eQTLs at a genome-wide error rate of 0.05; the bar length is proportional to the LOD-1 (logarithmic odds score) confidence interval. In addition to eQTLs detected from consensus mapping (blue horizontal bars), results from parental detections are also indicated by green and red bars for Syrah and Grenache, respectively. Before the drawing, the LOD-1 confidence intervals for eQTLs from parental detection were re-estimated through homeothetic projection on the consensus map based on common loci with biomercator v2.1 (Arcade et al., 2004). Candidate genes for grape PA synthesis are indicated at the top of the whole panel, where the bar size is proportional to the flanking marker interval. Purple rectangles are for genes coding for synthesis enzymes, while pink rectangles are for genes coding for transcription factors. The number above the flanking marker interval indicates the corresponding candidate gene: 1, VvLAR1; 2, VvMYBC2-L1; 3, VvLDOX; 4, VvGTs; 5, VvF3H; 6, VvMYB5b; 7, VvC4H; 8, VvF3′5′Hs; 9, VvMYC1, 10, VvPAL; 11, VvMYB5a; 12, VvMYBPA2; 13, VvCHIs; 14, VvCHS; 15, VvWDR2; 16, VvMYBPA1; 17, VvMYCA1; 18, VvPAL; 19, Vv4CL; 20, VvWDR1; 21, VvLAR2; 22, VvF3′Hs; 23, VvDFR. References for candidate genes are presented in Supporting Information Table S2.

Six eQTLs were identified in 2009 and 2010 and in genotypic averages; we called them ‘stable eQTLs’. Three of them were co-localized with their respective genes for VvDFR, VvLAR1 and VvLAR2. These eQTLs had high LOD (logarithmic odds) scores among loci detected for the same e-trait (4.7–8 for VvDFR, 24–39 for VvLAR1 and 20–48 for VvLAR2; Table S3) and relatively high R2 (8–20% for VvDFR, 43–56% for VvLAR1, and 13–48% for VvLAR2; Fig. 4 and Table S3). Two other stable eQTLs were located in regions distant from target genes: on chromosome 17 for VvLAR1 and on chromosome 3 for VvLAR2. One other stable eQTL was located on chromosome 10 for VvANR, explaining 7.8–13.6% of the trait variation and with a LOD score from 3.5 to 7.4. The other eQTLs were identified either on single year and on genotypic averages, or identified only with genotypic averages (such as for VvLDOX).

Figure 4.

Genotypic variance explained by individual expression quantitative trait loci (eQTLs) and digenic interactions (R2) identified on genotypic averages on the Vitis vinifera consensus map at a genome-wide error rate of 0.05. Loci are indicated on the x-axis with the following nomenclature: chromosome at position of LOD (logarithmic odds score) peak (e-trait). Triangles indicate loci involved in digenic epistasis.

Among all identified eQTLs, only one eQTL for VvDFR on chromosome 6 overlapped with a characterized phenylpropanoid transcription factor, VvMYB5b (Fig. 3). In the other cases, loci overlapped with genes coding for synthesis enzymes, such as target genes themselves, genes coding for upstream enzymes of the phenylpropanoid pathway (C4H on chromosome 6, PAL on chromosomes 8 and 16, and 4CL on chromosome 16) or genes coding for enzymes involved in flavonoid synthesis (e.g. F3′5′H on chromosome 6).

eQTL co-localizations were observed in seven genomic regions; five of them co-localized two eQTLs (loci on chromosomes 3, 6, 8, 16 and 17). In two other loci on chromosomes 1 and 18, eQTLs co-localized for three transcripts: VvDFR-VvLDOX-VvLAR1 and VvDFR-VvLDOX-VvANR, respectively (Fig. 3). In our previous grape PA QTL analysis (Huang et al., 2012), QTLs related to PA composition were also co-localized in these regions: concentration per berry, catechin as terminal unit and degree of polymerization for chromosome 1 and concentration per kilogram of FW, concentration per berry and epigallocatechin as extension unit for chromosome 18. However, the eQTL and PA QTL on chromosome 18 overlapped only at the extremity of confidence interval, while eQTLs and PA QTLs on chromosome 1 had similar LOD profiles and the distance between their QTL peak positions was 8-cM (Additional file 6 in Huang et al., 2012 and Table S3 and Fig. S3). eQTL and PA QTL co-localization suggested that their underlying candidate polymorphisms may be the same. We therefore focused on the region of chromosome 1 for further investigation.

Candidate gene identification

Regulation of the flavonoid pathway is complex (Dixon et al., 2005; Lepiniec et al., 2006). This complexity includes a regulatory loop between regulators (Dubos et al., 2008; Xu et al., 2013). In order to identify putative candidate genes involved in cross-talk under the eQTLs of chromosome 1, we compared annotated transcripts between flanking markers of this locus (Fig. S2, from VMC4F8 toVMC3G9) with a list of genes differentially expressed between wild-type grapevine hairy roots and hairy roots over-expressing VvMYBPA1 or VvMYBPA2, two MYB factors involved in the regulation of PA synthesis (Bogs et al., 2007; Terrier et al., 2009). Among the 605 genes located in the selected region, seven were up-or down-regulated by these MYB factors (Table S4). In this list, the predicted gene GSVIVT01011768001, significantly induced in both transcriptomic screenings, was the only putative transcription factor. GSVIVT01011768001 encodes a protein of 224 amino acids. Its best match in UniProt database is AtMYB4 (with 52.4% identity on 221 amino acid hits), characterized as a MYB-type protein involved in transcriptional repression of the phenylpropanoid pathway (Jin et al., 2000). This gene was previously named VvMYBC2-L1 by Matus et al. (2008) when making an inventory of R2R3-MYB factors of the grape genome.

To understand the structural relationship between VvMYBC2-L1 and other MYB proteins, a phylogenetic tree was constructed using the neighbour-joining method with whole protein sequences of other plant MYB factors, especially those involved in phenylpropanoid-derived pathways (Fig. 5). The phylogenetic tree revealed that this MYB is clustered with subgroup 4 MYB factors, among which AtMYB4, AmMYB308, FaMYB1, ZmMYB31, ZmMYB42, and PhMYB4 were demonstrated to repress phenylpropanoid synthesis, probably via repression of the expression of synthesis genes (Tamagnone et al., 1998; Jin et al., 2000; Aharoni et al., 2001; Sonbol et al., 2009; Fornalé et al., 2010; Colquhoun et al., 2011). MYB factors of subgroup 4 are characterized by two conserved motifs in their C terminal region: the C1 motif (LlsrGIDPxT/SHRxI/L) and the C2 motif (pdLNLD/ELxiG/S; Kranz et al., 1998), which have been shown to be essential for repressing gene expression (Jin et al., 2000). The VvMYBC2-L1 protein sequence showed great sequence homology to other subgroup 4 MYBs and exhibited 63% and 90% similarity, respectively, with their conserved C1 and C2 motifs (Fig. S4). We also noted that three amino acids in the conserved R2R3 domain, important for the interaction with bHLH in dicots (Zimmermann et al., 2004), were conserved in VvMYBC2-L1 (Fig. S4). In addition, we recently showed that one single nucleotide polymorphism (SNP) in an intron is significantly associated with grape berry PA content (Carrier et al., 2013). Given its location under eQTL and PA QTL, its induction by known regulators of the PA pathway, its high sequence similarity to other members of MYB subgroup 4, and its polymorphism association with grape PA content, VvMYBC2-L1 appeared to be a strong candidate for PA pathway regulation. We therefore undertook its functional characterization.

Figure 5.

Phylogenetic tree for MYB transcription factors involved in the phenylpropanoid pathway. The phylogenic tree was constructed from the ClustalW alignment of full-length protein sequences using the neighbour-joining method in the mega5 package. The scale bar represents 0.2 substitutions per site and the numbers next to the nodes are bootstrap values from 1000 replicates. We used MYB factors involved in regulation of plant development as the outgroup. Physiological involvement of MYB proteins is given when information is available. Bold case indicates genes that are functionally validated. Highlighted by the grey box are MYB having conserved domains C1 and C2, classified as ‘subgroup 4’ according to Kranz et al. (1998). All the functionally validated genes of subgroup 4 have repressor activity. The GenBank accession numbers of aligned sequences are as follow: Am308 (P81393.1, previously JQ0960), AmMYB330 (P81395.1), AtMYB4 (AAC83582.1), AtMYB5 (AAC49311.1), AtMYB6 (Q38851.1), AtMYB11 (NP_191820), AtMYB12 (NP_182268), AtMYB32 (O49608.1), AtMYB44 (Q9FDW1.1), AtMYB65 (AEE75047.1), AtMYB75 (PAP1) (NP_176057), AtMYB90 (PAP2) (NP_176813), AtMYB111 (NP_199744), AtMYB117 (AEE30736.1), AtMYB118 (AEE77363.1), AtMYB123 (TT2) (Q9FJA2), FaMYB1 (AAK84064), MdMYB1 (ABK58136), MdMYB10 (ABB84753), MybHv5 (CAA50221.1), OSMYB3 (BAA23339), PhMYB4 (ADX33331.1), PhPH4 (AAY51377), ScMYB12 (CBG76064.1), VvMYB4a (ABL61515.1), VvMYb5a (AAS68190), VvMyb5b (AAX51291), VvMYBA1 (BAD18977.1), VvMYBA2 (BAD18978), VvMYBF1 (ACT88298.1), VvMybPA1 (CAJ90831), VvMYBPA2 (ACK56131), VvMYBC2-L1 (this study), ZmC1 (AAA33482), ZmMYB31 (NP_001105949), ZmMYB42 (NP_001106009), ZmP (P27898), and ZmPL (AAA19821).

Expression and functional characterization of VvMYBC2-L1

The expression profile of VvMYBC2-L1 during Syrah berry development was assessed by RT-qPCR with RNA isolated from the pericarp (skin + pulp; Fig. 6). VvMYBC2-L1 was expressed throughout berry development and exhibited a first expression peak near the end of the green stage (35 d after anthesis). The increase in expression of VvMYBC2-L1 during the green stage correlated negatively with the PA synthesis profile (active synthesis in the early stage of development and a progressive decrease towards véraison) and also with expression of VvMYBPA2 (Terrier et al., 2009), VvDFR, VvLDOX, VvLAR1 and VvANR (Fig. 6a). In contrast, VvMYBPA1 showed a slight increase towards véraison, while expression of VvLAR2 increased markedly during the green stage. The expression of VvMYBC2-L1 rose again towards the end of maturity. VvDFR and VvLDOX, genes involved in both PA and anthocyanin synthesis, exhibited quite constant expression after an initial decrease during the green stage, while expression of PA-specific genes (VvLAR1, VvLAR2 and VvANR) was hardly detectable after véraison, which marks the end-point of PA synthesis.

Figure 6.

Spatiotemporal expression profile of VvMYBC2-L1 and some proanthocyanidin (PA)-related genes in Vitis vinifera cv Syrah: (a) in the pericarp during berry development, (b) in diverse grapevine organs and tissues. In (a), the arrow indicates the time-point of véraison. Gene expression was determined by real-time quantitative PCR and normalized with the expression of VvEF1α (elongation factor). All data points are means of three technical replicates, with error bars indicating + SD.

We also investigated the spatial expression profile of VvMYBC2-L1 and other PA-related genes in different grapevine organs and tissues (Fig. 6b). VvMYBC2-L1 was expressed in the pistil, tendril, leaves, berry skin and pulp and its expression level was low in roots and seeds. All other measured genes were expressed in tissues where PA synthesis is carried out (the pistil, tendril, root, young leaves and skin and seeds of young berries), but their spatial expression profiles were quite different. In leaves, PA-related genes measured in this study were more highly expressed in young leaves than old ones, except for VvMYBC2-L1 and VvLAR2. The range of expression level of VvMYBC2-L1 was similar across all tested tissues, while that of VvMYBPA1 and VvMYBPA2 was smaller in berries than in leaves (Fig. 6b, Terrier et al., 2009). Inside a berry, expression profiles varied between tissues: while VvMYBC2-L1, VvMYBPA1, VvLDOX and VvLAR2 had higher expression in skin than in seeds, VvDFR, VvLAR1 and VvANR, to a lesser extent, exhibited the reverse profile.

To further investigate its function, VvMYBC2-L1 was overexpressed in grapevine hairy roots. Three independent transformants, over-expressing VvMYBC2-L1 66-, 24- and 19-fold compared with the negative control (Fig. 7a), were selected. In those lines, flavan-3-ol total content was drastically reduced from 28- to 6-fold in the transformants (0.12–0.54 mg g−1 FW) compared with the control (3.34 mg g−1 FW; Fig. 7b). A similar reduction was also observed for flavan-3-ol monomers (Fig. 7c). No important change in flavan-3-ol composition was observed after over-expressing VvMYBC2-L1; only a slight change in the relative proportions of PA subunits for one of the transformants (MG5b; P = 0.02; Table S5). The relative content of three major stilbenes, also derived from the phenylpropanoid pathway, was drastically reduced in VvMYBC2-L1 over-expressing lines (Fig. S5).

Figure 7.

Characterization of Vitis vinifera hairy roots over-expressing VvMYBC2-L1. (a) Gene expression of VvMYBC2-L1 relative to VvEF1α (elongation factor), (b) total proanthocyanidin (PA) content, and (c) content of flavan-3-ol monomers. Ctrl, negative control (hairy roots without transgene); MG1C, MG5b, MG4A, three independent transformants. All data points are means of three technical replicates, with error bars indicating ± SD. Light grey bars, epicatechin; dark grey bars, catechin.

To elucidate the mechanisms underlying the phenotypic changes observed in grapevine hairy roots over-expressing VvMYBC2-L1, we measured the expression level of genes involved in this pathway (Fig. 8). Substantial reductions in expression were observed for VvMYBPA1, VvMYBPA2, VvDFR, VvLDOX, VvLAR1 and VvANR, while VvLAR2 showed little transcript reduction and no change in expression was observed for VvMYB5b. We also measured the expression of genes involved in other phenylpropanoid pathways. Over-expression of VvMYBC2-L1 reduced slightly the transcript levels of Vv4-CL1 and VvC4H1 (Fig. S6), two structural genes at the very beginning of the phenylpropanoid pathway (Fig. 1). Among the four tested groups of stilbene synthase isogenes, VvSTS13 and VvSTS16 showed decreases in expression in VvMYBC2-L1 over-expressing lines, while a slight increase or decrease in expression was observed for VvSTS8 and VvSTS27 (Fig. S6). This result suggests the possible involvement of VvMBYC2-L1 in regulating early steps of the general phenylpropanoid pathway and some steps in stilbene synthesis.

Figure 8.

Expression of proanthocyanidin (PA) candidate genes relative to VvEF1α (elongation factor) in control Vitis vinifera hairy roots and VvMYBC2-L1 over-expressor lines. Names of measured candidate genes are indicated at the top right of each panel. All data points are means of three technical replicates, with error bars indicating + SD.

Discussion

eQTL mapping provides new insights into the grape PA pathway

This study presents for the first time grape PA gene expression data in a mapping population as well as grape eQTLs associated with this pathway. Analyses of transcript data obtained from a large number of genotypes across 2 yr provided new insights into PA gene expression behaviour. All five genes showed transgressive segregation (Fig. S1). This indicates that the transcript abundance of these genes may be controlled by alleles of opposite effect originating from the two parents or in a nonadditive manner. It also implies the involvement of multiple loci in gene expression variation.

At another level, as also indicated by the heritability of the traits, transcript correlation across 2 yr was small or nonsignificant for VvDFR, VvLDOX, and VvANR, while the transcript abundance of VvLAR1 and VvLAR2 was highly correlated across 2 yr (Fig. 2b). This indicates that, in our mapping progeny, the relative ranks of expression level for VvDFR, VvLDOX, and VvANR are shuffled between years, that is, individuals react differently between years. This may be the result of random experimental errors and/or environment effects. However, the lack of replication per genotype within year did not allow us to investigate these two possibilities in detail. Nonetheless, seven eQTLs were uniquely detected through single-year data (four for 2009 and three for 2010). They may control the expression profiles of target genes only under specific environmental conditions.

In our previous PA genetic study, we identified a large number of grape PA QTLs: considering only the results for the consensus map, 43 and 103 QTLs were identified for skin and seed PA variables, respectively (Huang et al., 2012). Epistasis has been identified using the multiple-QTL model and, together with additive QTLs, the whole QTL model could explain > 80% of the observed variation. Among QTLs detected for the same variable in both tissues, only five loci out of > 50 had overlapping intervals. This result led to the conclusion that PA synthesis is under complex, tissue-specific regulation. In the present study, because of experimental constraints, we measured expression at the whole-berry level, aiming first to identify major loci detectable despite the background caused by mixing berry compartments. This could explain in part why relatively few eQTLs were identified. Another reason could be that transcript abundance is more subject to environmental and developmental changes, while PA composition remains stable once synthesis has ended.

Among 21 eQTLs identified in this study, six were classified as ‘stable’. Three of them, that is, the locus on chromosome 18 for VvDFR, the locus on chromosome 1 for VvLAR1 and the locus on chromosome 17 for VvLAR2, corresponded to the profile of cis-eQTLs in the literature: positioned in the proximal region of target genes, explaining a large part of variation, and consistently detected across conditions (Brem & Kruglyak, 2005; Keurentjes et al., 2007; West et al., 2007; Potokina et al., 2008b). The locus on chromosome 10 for VvANR has the same feature. In addition, this locus co-localized with a QTL for epicatechin, the direct product of ANR (Huang et al., 2012). Altogether, this information may help to elucidate the genomic location of VvANR, which has not yet been determined in the 12X grapevine genome sequence.

The other two stable eQTLs are also interesting targets for further investigation. However, we favoured the locus where both the eQTL and the PA QTL have been identified, that is, the locus on chromosome 1.

Nature of underlying regulators: the case of VvMYBC2-L1

We combined eQTL, PA QTL (Huang et al., 2012) and transcriptomic data (Terrier et al., 2009) to identify causal candidates. One of the three eQTLs co-localized at this position is classified as a cis-eQTL for VvLAR1. Although this locus can be classified as a cis-eQTL of VvLAR1, the causal polymorphisms may include those from other genes positioned close to VvLAR1. Carrier et al. (2013) noticed that genes up- or down-regulated after over-expression of VvMYBPA1 and VvMYBPA2 were in significantly higher density at PA QTL than in regions where no PA QTL has been detected. Several eQTLs for genes involved in the PA pathway may therefore co-localize at this locus, in the vicinity of VvLAR1. A larger population with a more saturated genetic map or a densely genotyped diversity panel and whole-genome expression profiling will be necessary to determine the causal factors underlying the eQTLs identified in this study and, more generally, all the factors involved in PA pathway regulation. Nonetheless, a candidate gene, VvMYBC2-L1, was identified thanks to several evidences for pursuing its functional validation: it was up-regulated after over-expression of PA-specific MYB proteins (Terrier et al., 2009); it was slightly but significantly associated with grape PA content in a diverse grapevine collection (Carrier et al., 2013); and it is closely related to MYB negative regulators (Fig. 5).

The expression profile of VvMYBC2-L1 during berry development exhibited a double-peak curve with a drop point at véraison. The first peak corresponded to the end of PA accumulation and the second peak corresponded to the end of anthocyanin and stilbene accumulation. These two time-points also coincided with a decrease in the expression of positive regulators and downstream synthesis genes of these metabolites, except for VvLAR2 (Fig. 6a, Bogs et al., 2005, 2007; Castellarin et al., 2007; Terrier et al., 2009). In addition, overexpression of VvMYBC2-L1 in grapevine hairy roots noticeably down-regulated PA synthesis genes, which supported its in planta function as a negative regulator. Moreover, increased expression of VvMYBC2-L1 was observed in hairy roots overexpressing VvMYBPA1, VvMYBPA2 and VvMYBA1, specific regulators of PA and anthocyanin synthesis, respectively (Cutanda-Perez et al., 2009; Terrier et al., 2009). This suggests that VvMYBC2-L1 may act as a counterbalance to finely regulate the accumulation of phenolic compounds in fruit cells, possibly through a permanent regulation loop between activators and repressors of the pathway to fine-tune phenolic content, which is involved in plant self-protection but can also be toxic to the plant itself at high concentrations (Matile, 1984).

In various tissues where PA synthesis occurs, we observed that elevated expression of VvMYBC2-L1 was correlated with low expression of some PA-related genes and vice versa, which supports VvMYBC2-L1 as a negative regulator. Meanwhile, genes exhibiting the opposite expression profile to that of VvMYBC2-L1 in a given tissue may not have this correspondence in another tissue. This is not very surprising, because the same pathway may require different sets of regulators across tissues (Sreenivasulu et al., 2006). Expression profiles in separate berry compartments showed that VvMYBC2-L1, VvMYBPA1, VvLDOX and VvLAR2 have higher expression in skin than in seeds, while VvDFR, VvLAR1 and VvANR exhibited the opposite profile (Fig. 6). This may raise questions when combined with hairy-root data, where a pronounced decrease in expression was observed for almost all assayed PA-related genes in overexpressing lines. Actually, PA-synthesis genes may require different amounts of VVMYBC2-L1 transcripts to be modulated as a result of specific regulatory mechanisms. For example, the low-level transcript in seed may be sufficient to control VvLAR1, while a large amount may be required for VvLDOX. The down-regulation of PA-synthesis genes is probably a result of an amplified upstream effect of the same pathway.

Hairy roots overexpressing VvMYBC2-L1 showed a dramatic reduction in the content of PAs, flavan-3-ol monomers and stilbenes (Figs 7, S5). The expression profiles of the four groups of stilbene synthases that we measured did not perfectly reflect the reductions in the abundances of their products. One possible explanation is that we may not have measured isogenes regulated by VvMYBC2-L1, as at least 48 isogenes coding for stilbene synthase are present in the grape genome (Parage et al., 2012), and that the design of specific RT-qPCR probes for each of them is impossible (Dai et al., 2012). We cannot completely rule out the possibility that overexpression of this transcription factor under the 35S promoter might cause off-target activation. However, similar experiments revealed that the phenotypic response of this plant transformation system was quite specific for one metabolic pathway (Cutanda-Perez et al., 2009; Terrier et al., 2009). Therefore, this provides evidence for the involvement of VvMYBC2-L1 in PA synthesis. Both the decrease in stilbene content and the decrease in the expression of early phenylpropanid synthesis genes (Fig. S6) observed in VvMYBC2-L1 over-expressing lines lead to the hypothesis that VvMYBC2-L1 is involved in other phenylpropanoid branches. However, the phenolic composition in hairy roots does not perfectly reflect that in berries. For example, control hairy roots (without the introduction of a transgene) only contain PA, stilbenes and trace amounts of phenolic esters. No anthocyanin or flavonol was detected in hairy roots (Cutanda-Perez et al., 2009; Terrier et al., 2009). Further investigations in other grape models producing the whole panel of phenolics would be necessary to determine its overall influence on the pathway.

Regarding the regulatory mechanism, VvMYBC2-L1 may act as direct repressor alike its orthologue AtMYB4, as it has the conserved C2 motif in the C-terminal region, which has been shown to be required for direct repression in the case of AtMYB4 (Jin et al., 2000; Zhao et al., 2007). Alternatively, VvMYBC2-L1 may act as an indirect repressor through competition with MYB activator via binding affinity with promoter of bHLH and WD40. This hypothesis is supported by its conserved DNA-binding domain and its conserved three amino acids in the R3 repeat, predicted to be essential for interaction with bHLH (Fig. S4; Zimmermann et al., 2004). For instance, FaMYB1 and AtMYB4, two orthologues of VvMYBC2-L1, were shown to be able to interact with bHLH proteins (Aharoni et al., 2001; Zimmermann et al., 2004). Indirect expression may also be achieved via repression of MYB activators, as suggested by the reduced expression levels of VvMYBPA1 and VvMYBPA2 in VvMYBC2-L1 over-expressors. This property is quite different from that of FaMYB1, which is able to repress PA biosynthesis in Lotus corniculatus, but is ineffective in repressing the accumulation of endogenous R2R3-MYB PA activators (Paolocci et al., 2011).

PA synthesis regulation – a complex phenomenon

Among all the identified eQTLs, only one locus on chromosome 6 for VvDFR co-localized with a known transcription factor, VvMYB5b, described as a general regulator of the phenylpropanoid pathway (Deluc et al., 2008). One may be surprised that no eQTL co-localized with the two PA-specific MYB factors, VvMYBPA1 and VvMYBPA2. Actually, VvMYBPA1 and VvMYBPA2 are almost monomorphic in each parent, except, respectively, for two SNPs (in introns) and three SNPs (two in introns and one in the 3′ untranslated region) identified between Grenache alleles. Lack of allele contrast for these two regulators in our mapping population is probably the reason why no eQTL has been identified at their location. Indeed, no PA QTL was identified in these locations in our previous study (Huang et al., 2012). These findings support the hypothesis that this part of the genome may be quite monomorphic, which does not allow identification of QTLs. By contrast, a previous study of anthocyanin genetics in the same population identified a QTL of large effect co-localizing with a major anthocyanin regulator (Fournier-Level et al., 2009). The same locus has been identified as an eQTL for transcript level of VvUFGT, a specific enzyme of the anthocyanin pathway (Huang et al., 2013) whose variation is controlled by a few loci in our mapping population. This example in the anthocyanin pathway illustrates that the ability to identify QTLs is related to the genetic architecture of the trait in the population used for the study. Nonetheless, our results support the hypothesis of complex PA regulation, as 17 new loci involved in transcript variation of PA synthesis genes were identified. The identification of digenic epistasis also supports the hypothesis of regulatory complexity. As mentioned above, few loci and epistasis identified in the present study compared to the one of PA QTLs (Huang et al., 2012). A large proportion of unexplained phenotypic/genetic variance could thus indicate the presence of undetected small loci and/or variation as a result of environment effects.

Further evidence of the polygenic nature of PA regulation is provided by the particular behaviour of VvLAR2. In either 2009 or 2010, a highly significant correlation was observed between VvDFR, VvLDOX, VvLAR1 and VvANR (Fig. 2), while expression of VvLAR2 was only correlated to that of VvDFR and VvLDOX for 2010. Overexpression of positive (Terrier et al., 2009) or negative (this study) regulators affected the expression of most PA genes, except for VvLAR2. The particularity of VvLAR2 compared with other PA-related genes was also observed during spatiotemporal expression profiling (Fig. 6). It is likely that there is at least one other regulatory mechanism for VvLAR2. Actually, the different regulatory mechanisms for LAR may be a general feature, as the differential transcript profile for isogenes of LAR was also observed in Medicago truncatula, persimmon fruits (Diospyros kaki) and Lotus corniculatus (Pang et al., 2007; Akagi et al., 2009; Paolocci et al., 2011). Furthermore, recent advances in elucidating the regulatory mechanism of glucosinolates, secondary metabolites characteristic of the Brassicale order with important biological functions, showed that similar metabolite profiles may result from independent regulatory mechanisms (Sonderby et al., 2010).

The present study opens doors for further investigations on PA regulation, such as the dissection of the large effect cis-eQTLs or other trans-eQTLs, notably those not co-localized with known transcription factors. In addition, other loci may be involved but did not segregate in our mapping population. Thus, to increase our knowledge of the general genetic architecture of PA transcript variation, it would be necessary to apply the same approach to different mapping populations or to diversity panels in order to assess the genetic architecture of gene expression in an extensive genetic background. The same approach may also be extended to upstream synthesis genes or even known regulators in order to dissect regulation of the whole pathway, and/or may be refined by separating berry compartments to elucidate PA regulatory mechanisms at tissue level, as the PA profile differs between grape skin and seeds. Because many actors in PA synthesis are still missing, such as for PA polymerization, subunit galloylation and intra/intercellular transport, the genome-wide eQTL approach can provide valuable insights regarding these black boxes and the interplay between regulatory networks.

Here, we have carried out the first integrated study involving eQTL mapping of PA synthesis in the grapevine berry and have confirmed the complex regulation of this pathway. The identification and functional characterization of a MYB protein revealed that it is a negative regulator of PA synthesis, and putatively also involved in the regulation of other branches of the phenylpropanoid pathways.

Acknowledgements

This research was supported by INRA (Projet Innovant Blanc of GAP division and Action Nouvelle Soutenue of CEPIA division). Y.F.H. was supported by a PhD grant from INRA and Languedoc-Roussillon region. We gratefully acknowledge all members of research teams in UMR AGAP and UMR SPO for sample collection and E. Meudec for help with stilbene identification. Our gratitude also goes to Dr J. S. Hsieh, Ms M. Ducroquet and Ms F. Cabanlit for their kind assistance with flowering annotation.

Ancillary