In particular, VviAGL11 is mainly expressedin seed at fruit set and the expression level declines throughout seed development, well supporting the major role of this MIKC-Type MADS-box factor in the control of ovule morphogenesis and seed coat differentiation (Royo et al., 2018). Nevertheless, VviAGL11 showed a high expression also during berry development with a very low expression around veraison followed by a steady increase during ripening, peaking at the second stage of postharvest berry withering (PHWII) (Figure 9a). Intriguingly, VviAGL11 showed a higher expression in the skin than in the pulp during the last stages of berry ripening. A slight expression of VviAGL11 was also found in flower organs and rachis after veraison, whereas the expression was very low or absent in all the other grapevine samples (Figure 9a).

The VviAGL11 expression trend, retrieved from the berry transcriptomic dataset obtained in ten grapevine varieties (Massonnet et al., 2017), confirmed the general higher expression level during early development and ripening stages, albeit with some differences among genotypes (Figure 9b). Also, thanks to the recent transcriptomic dataset consisting of a weekly collection of Cabernet-Sauvignon and Pinot noir berries from pea size to harvest during three consecutive years (Fasoli et al., 2018), it was possible to ascertain with high precision that VviAGL11 reached its minimum expression level at 10 days after veraison and featured an up-regulation about three weeks after veraison (Figure 9c).

To investigate the transcriptional relationships of VviAGL11 in the seed and the berry during ripening, a co-expression analysis was performed separately on the four stages of seed development and the three ripening stages of the berry of the Corvina expression atlas (Supplementary Dataset S1 online). Genes that were highly co-expressed with VviAGL11 in seed were characterised by very high Pearson correlation coefficients (216 genes; r > 0.9), whereas genes highly co-expressed with VviAGL11 in berries during ripening were characterised by lower Pearson correlation coefficients (only 4 genes > 0.9). Interestingly, no genes were shared between the top-200 genes highly co-expressed with VviAGL11 in seed and the top-200 genes highly co-expressed with VviAGL11 in berry, strongly suggesting that different genes are targeted by VviAGL11 and that a different role is exerted by this factor in the two organs.

The co-expression analysis of VviAGL11 in berry during ripening was further performed using the post-veraison samples of Cabernet-Sauvignon and Pinot noir transcriptomic datasets (Fasoli et al., 2018). The comparison of the list of the top-200 highest co-expressed genes in Corvina, Cabernet-Sauvignon and Pinot noir, revealed some common putative metabolic partners of VviAGL11 (Supplementary Dataset S1 online). By looking at the highest co-expressed genes (Pearson coefficient ≥ 0.8) we found 31 genes in Corvina, 41 in Pinot noir and only three in Cabernet-Sauvignon (Supplementary Dataset S1 online). The potassium-sodium symporter HKT2 was found among the top-ranked genes in all the three lists (Table 3).

Table

titre du tableau
Gene ID	Functional annotation	Corvina	Cabernet-Sauvignon	Pinot noir
VIT_11S0103G00010	Potassium-sodium symporter HKT2	*	*	*
VIT_09S0002G00580	Lipase GDSL 3	*		*
VIT_00S0253G00080	Methyl jasmonate esterase	*		*
VIT_14S0060G00240	Myb domain protein 61		*	*

A methyl-Jasmonate esterase (VIT_00s0253g00080) and a lipase GDSL (VIT_09s0002g00640) were found in the list of Corvina and Pinot noir, and the R2R3MYB transcription factor Myb 61 was highly co-expressed with VviAGL11 commonly in Pinot noir and Cabernet-Sauvignon (Table 3).

Berry sorting by density flotation permits reducing the sample heterogeneity in terms of ripeness in both table and wine grape varieties (Río Segade et al., 2013a; Zouid et al., 2013). In particular, Fournand et al. (2006) reported for wine grapes that the difference in the total sugar content of Shiraz berries belonging to two consecutive density classes (i.e., classes differing from each other by 10 g/L NaCl concentration in the flotation solutions) was ~17 g/L (i.e., 1 % v/v potential alcohol). Similar differences were reported by Torchio et al. (2010) and Rolle et al. (2012a) on Barbera and Nebbiolo grapes, respectively, but only for some density classes. These authors found that the difference in berries sugar content throughout the density classes was not constant, but it varied and decreased by increasing the density of the floating solutions. This different trend, within the same variety, was linked not only to total sugar content but also to vineyard location (Rolle et al., 2012b). The sugar content of berries belonging to different density classes varied similarly also when density flotation was applied to berries with larger volumes (i.e., on table grape varieties), as demonstrated on Italia (Río Segade et al., 2013a), Muscat of Hamburg (Rolle et al., 2015), and Red Globe and Crimson Seedless (Río Segade et al., 2013b)^{grapes. However, when berries of different varieties even though belonging to the same density class were compared, the chemical composition in terms of technological parameters was different. This is in agreement with the results of this study, which clearly showed that density flotation, and subsequent selection of only one density class, separated F1 genotypes and their parents in different quartiles concerning chemical composition (reducing sugar content, pH, titratable acidity, organic acid composition), berry characteristics (weight, size, and seedless character) as well as skin mechanical properties. Therefore, berry sorting by density flotation has shown to be a strategic approach when managing many and highly variable genotypes because, irrespective to the quantification of specific molecules classes, the method introduces a common and more general criterion for berries selection of all genotypes. The wide variability present in our F1 population was confirmed by the transgressive segregation displayed also for highly similar traits in the two parents, like what found by Carreño et al. (2014) and Jiang et al. (2020)studying other F1 progenies.}

Our results point to one major QTL for all texture traits; the confidence interval was difficult to define because this QTL is located at the end of LG18 and only a few SSRs were available per chromosome. Our data provide well-supported information on the co-localisation of average berry weight and all TPA traits on the same QTL on LG 18. The higher impact of whole berry mechanical properties can be due to the different response of each genotype under double compression test, as previously evidenced for grape cultivars (Río Segade et al., 2011b). This study found a link also for berry skin thickness on this major QTL, with a LOD score a little bit higher than the significance threshold. Weak associations with lower percentages of explained variance were found for berry skin elasticity on the same major QTL on LG18 and along LG16.

The suspicion that seeds may affect berry hardness measurements, making the grapes harder, is diluted by our data: the seedless Sultanina variety showed berries harder than Raboso Veronese. This result is in agreement with Migicovsky et al. (2017), that grapes from the Middle East are firmer than those from West Europe, because of human selection for table and wine grapes, respectively. Moreover, TPA data normalisation on berry diameter, surface, and volume reduces the influence of berry dimensions. Berry resilience, springiness, and cohesiveness are highly correlated with each other, as demonstrated by PCA analysis, because they are linked to the ability of the berry to return to its original form after compression. For these parameters, normalised data found stronger associations than raw ones, independently on the type of normalisation adopted. LOD scores were around three-four times higher with normalised TPA parameters and percentages of explained variance were more than double compared to raw data. Consequently, berry dimensions affected raw TPA data in performing the associations with QTLs.

Berry hardness, gumminess, and chewiness assess berry consistency (i.e., firmness) and grouped together in PCA analysis. For these traits, raw TPA data showed two-three times higher LOD values and the percentage of phenotypic explained variance was also 1.5–2.5 times higher than the normalised ones. Even for berry gumminess and chewiness, no marker-trait association was found for TPA parameters normalised on the berry surface. Therefore, berry consistency was independent of berry size. Our results are partially consistent with those obtained by Carreño et al. (2014) and Correa et al. (2016) and agree with Jiang et al. (2020). In particular, the methods developed by Carreño et al. (2014) and used also by Wang et al. (2020), define berry “firmness” as values expressed in N required for a 20 % deformation of the berry grape. This mechanical variable can be associated with berry hardness evaluated in the present study calculated using deformation of the berry equal to 25 %. Compared to previous studies, the novelty of the present one is the different approach to grape phenotyping because we used the berries density as the main factor for selecting the samples for subsequent analyses. Reducing sugars measurement was used in the vineyard only for a rough determination of the harvesting time, but hereafter berries selection was based on flotation and therefore on density, a concept that normalises the berry density characteristics by incorporating more chemical-physical parameters.

Furthermore, flotation is a non-destructive sorting method, which allows the same berries to be subjected to TPA avoiding the need to use an additional sample, presumably similar but not verified. In this way, the effect of different concentrations of the compounds present in the must, especially sugars and acids, is cumulated and balanced. In our F1 the variability in the chemical composition of the berries belonging to the same density class was high.

The average berry weight associates with the SSRs of VviAGL11, because the development of normal seeds stimulates berry growth, probably through growth regulators produced by seeds (Doligez et al., 2013), whereas seedlessness affects the formation of small or very small berries (Royo et al., 2018). Similar results were obtained working on F1 populations with a seedless background (Doligez et al., 2002; Cabezas et al., 2006; Costantini et al., 2008; Mejía et al., 2011). Interestingly, other QTLs, independent from seedlessness and located on LGs different from LG18, were proposed affecting berry dimensions (Ban et al., 2016; Doligez et al., 2013).

In previous studies, doubts about the reliability of some common QTLs identified for different berry traits were explained as possibly related to common ancestors in the parentage lineages analysed (Carreño et al., 2014). This is not the case for our F1, because Raboso Veronese is a local wine grape cultivar of the Veneto region; its parents are Raboso Piave and Marzemina bianca (Crespan et al., 2006, Crespan et al., 2015), two other local varieties having no genetic proximity with Sultanina. Therefore, the main QTL that we found to be related to texture traits is not affected by inbreeding effects and appears to be strong and solid.

All six TPA parameters listed in Table 2 were divided into two statistically different groups (Figure 5). The lowest values for average berry weight and berry firmness (BH, BG, BCh_ratio) and the highest values for berry resilience, springiness and cohesiveness (BR_vol, Bsratio_vol, BCo_vol) showed to be associated to segregation classes ac and bc of AGL11-VMC7F2 marker, then to the genotypes that inherited the Sultanina allele c, linked to seedlessness. While the association between average berry weight and AGL11-VMC7F2 was expected (Costantini et al., 2008), it is a novelty for berry texture features.

Our major QTL is also the only one recurring in three previous works (Carreño et al., 2014; Correa et al., 2016; Jiang et al., 2020), where instrumental quantitative analyses and not sensorial analyses were used for texture measurements. Having found, this time with clearer evidence, a main QTL for berry texture on LG18 on the same chromosomal interval using a different and innovative approach for phenotyping, it is the counter-proof that in that portion of the grapevine genome there is, with high probability, the main key gene to understand the texture characteristics of grapes.

About AGL11, the UNIPROT database (https://www.uniprot.org/uniprot/Q38836) suggests that the related protein, Agamous-like MADS-box protein AGL11, is a probable transcription factor. It is required, together with TT16/AGL32, for the maternal control of endothelium formation, which is essential for female gametophyte development and fertilisation, and seed formation. So, VviAGL11 biological role involves the ovule development, with pleiotropic effects on other traits linked to the poor development of seeds, like berry dimensions. Studies on VviAGL11 monitored its expression at fruit set and pea-size stages in seeds or seed traces (Royo et al., 2018), but information is lacking if this gene plays a role on berry ripening at further fruit developmental stages. By inspecting transcriptomic databases, we could trace the expression of VviAGL11 during the entire berry development and noticed that it raises after veraison in pericarp tissues with a steady up-regulation during ripening. Co-expression analyses highlighted possible transcriptional relationships between VviAGL11 and other genes, different in the seed and the berry, and strongly suggest that different biological functions are exerted by VviAGL11 in the two grapevine organs.

To infer a role of VviAGL11 in the definition of the berry texture characteristics at harvest, we looked at the function of the genes most highly co-expressed with VviAGL11 in berry pericarp after veraison, representing either its potential targets or partners in the same metabolic/developmental process. Intriguingly, transcriptional analyses revealed high co-expression values of VviAGL11 with the VIT_14S0060G00240 gene encoding an R2R3MYB transcription factor functionally annotated as MYB61. This MYB member belongs to the previously defined S13 subgroup of the MYB family (Wong et al., 2016) that, in addition to AtMYB61, includes AtMYB50, AtMYB55, and AtMYB86, all involved in secondary cell wall formation and lignin biosynthesis in Arabidopsis (Huang et al., 2015; MacMillan et al., 2017; Rao and Dixon, 2018; Taylor-Teeples et al., 2015). AtMYB61, the best characterised member of this subgroup, was shown to have a role in the synthesis of the seed coat rhamnogalacturonan or the appropriate deposition of this extracellular polymer (Penfield et al., 2001), and to be a positive regulator of secondary cell wall biosynthesis (Rao and Dixon, 2018). Target genes of AtMYB61, such as a secondary wall repressor, a pectin methylesterase and a caffeoyl CoA 3-O-methyltransferase of monolignol biosynthesis have been identified (Romano et al., 2012).

Consistently, in rice the closest homolog of AtMYB61 targeted a CAD2 (involved in lignin synthesis) and a CESA (involved in cellulose deposition) gene, supporting a role in the coordination of both cellulose and lignin biosynthesis in secondary wall formation (Huang et al., 2015). Overall, these pieces of evidence support the hypothesis that MYB61 could be involved in texture changes during late ripening through the control of cell wall metabolism in berry pericarp.

The identification of the high-affinity K+ potassium transporter 1 (VviHKT1;1), as one of the most highly co-expressed with VviAGL11 in berries, strongly suggests that VviAGL11 could be involved in the maintenance of intracellular K+/Na+ homeostasis. The VviHKT1;1 has been recently characterised by Henderson et al. (2018) as a major gene involved in the control of Na+ exclusion in grapevine roots. We hypothesise that VviHKT1;1 could also have a role in berry in the maintenance of the correct cytosolic ion ratio, that in turn may control aspects like osmoregulation, enzyme activation and cell expansion, all of crucial importance during ripening.

Noteworthy, VviAGL11 resulted highly co-expressed with a MeJA (methyl Jasmonate) esterase, a hydrolase that converts the volatile MeJAs to JAs (Jasmonates), being well-recognised lipid-derived stress phytohormones involved in plant responses to biotic and abiotic stresses (Creelman and Mullet, 1995; Hettenhausen et al., 2013; Huang et al., 2017; Wang and Wu, 2013). As a methyl ester of JA, MeJA has been postulated to be the inactive or storage form of JA (Wasternack and Strnad, 2016; Westfall et al., 2013), and the demethylation process would be, therefore, the first step to activate MeJA. Interestingly, it has been observed that JAs induce leaf senescence activating senescence-associated genes and chlorophyll catabolism-related genes (CCGs), and repressing photosynthesis-related genes (e.g., Chlorophyll A/B Binding Protein1) (Huang et al., 2017; Qi et al., 2015; Zhu et al., 2015). Furthermore, JAs stimulate organ abscission by affecting the metabolism of cell wall polysaccharides (Miyamoto et al., 1997). Thus, in the context of berry ripening, JA could be involved in senescence cellular processes associated with ripening, including cell wall polysaccharides rearrangement.

A strong co-expression between VviAGL11 and a GDSL lipase, an esterase/lipase hydrolytic enzyme with important functions in plants such as the regulation of plant cell wall components (de la Torre et al., 2002), was also found. Interestingly, the involvement of two GDSL-like lipases in the regulation of plant systemic resistance associated with ethylene and auxin signalling was reported (Kim et al., 2014; Lee et al., 2009). Although far to be conclusive, altogether this data strongly suggests that VviAGL11 targets genes involved in lipid and hormone metabolism, transport and gene expression regulation, that in turn may impact cell wall polymer rearrangement and texture features of berry at ripening.

Concluding, previous studies showed how difficult it is to determine the genetic control of berry texture. Our results clearly underline the major role of a QTL located on LG18. This information contributes towards the development of additional tools supporting the early selection of desirable traits in newly bred grapevine varieties. No QTLs strongly associated with berry skin hardness and thickness were found. Large gaps (>20 cM) in some LGs (LG7b, LG8, LG18, LG19) and map length (1028.946 cM) close to the inferior limit of the reference range for Vitis, indicate that the low number of markers used (158) was not enough to obtain a homogeneous saturation for wide QTL detection.

Berry sorting through density flotation showed to be a very advantageous and non-destructive method for highly heterogeneous genotypes population characterisation, in fact, it enables to apply TPA on berries belonging to the same density class, regardless of specific molecules classes quantity. Destructive methods were used in previous studies based on berries ripeness evaluation through crushing and must analysis, which supposes that the other berries used for the following TPA analyses, even though from the same clusters, reached more or less the same ripening stage. Samples heterogeneity, and therefore weak phenotyping data, can be regarded as one of the reasons for the low explained percentage of phenotypic variation associated to the QTLs found in previous studies. Our phenotypic method significantly improved berry texture studies intended be searching for QTLs associated with these traits.

The authors thank Prof. Fabrizio Costa for helpful suggestions in data management, Dr. Graziana Da Rold for help in checking F1 genotypes, Dr. Sara Landolfo for preliminary screening of SSR markers, Dr. Barbara De Nardi for the help in mapping F1 vines, Mrs. Fiorenza Santellani for the management of F1 vines in the vineyard and Valentina Rolle and Roberta Rolle for the collaboration in grape sample preparation for texture analysis. This work was partially supported by ESPLORA project (2011–2013), financed by the Italian Ministry of Agriculture (DM 14658/7303/2010), and by the SIV (Service for Identification of grapevine Varieties) of CREA Research Centre for Viticulture and Enology.