Berry size is an important trait in relation to both yield (table grapes) and quality (wine grapes). Indeed, the flavor in wine results from the ratio of skin to flesh, the former being the source of most aromatic and tannins compounds, the second providing the organic acids and the sugars [1].

The genetic and molecular basis of fleshy fruit size variation have been studied in depth in tomato during the last two decades, using a large panel of diverse resources that made tomato a model species for fleshy fruit crops [2, 3]. Introgression lines between wild and cultivated genotypes [4–6], near isogenic lines (NILs) [7] and artificial or natural mutants [2, 8] have been created and used to study the genetic basis of fruit size variation showing that a large part of it is controlled by less than ten loci. The physiological mechanisms involved have been related to the control of (i) the cell number in the pericarp, as for the fw2.2 locus [9, 2, 10], (ii) the locule number [2, 11], (iii) the late endo-reduplication in pericarp cells [12, 10] and (iv) the cell wall plasticity in relation to the cell expansion [10]. All these advances in tomato are useful to assist the study of similar trait in other crops with fleshy fruits, less amenable to genetic studies, such as perennial fruit trees. Indeed, encouraging results have already shown syntheny within Solanaceae species for Quantitative Trait Loci (QTL) controlling fruit size [2]. However, the degree of transferability of knowledge from tomato to non-Solanaceae species remains an open question.

Like tomato, grapevine (Vitis vinifera) produces fleshy fruits and a large difference in fruit size between wild and cultivated genotypes can be observed [13]. Indeed the wild V. vinifera genotypes produce mature berries weighting less than 1g while berries of some table grape varieties can weigh 10 g and more [14]. The growth of a grapevine berry roughly follows the same pattern as for tomato fruit: the first phase of fruit development is due to both cell multiplication and cell expansion, followed by a lag phase corresponding to a major cell metabolic shift and a second phase of fruit growth, mostly explained by cell expansion but without evidence of endoreduplication [15]. The genetic analysis of grape berry size variation is more difficult than in tomato, due to the long biological cycle of the plant, to the high level of heterozygosity of the genome and to the large field area usually required for plant growth, which makes experiments in controlled environment more costly [16–19]. In addition, berry size studies have often been performed on population segregating for seedlessness, with a strong negative correlation between the two traits: the seedless berries are in average smaller than the seeded berries [16–19]. Up to now, it has not been possible to establish the relationship between QTL for berry size and processes like cell multiplication or cell enlargement.

A natural mutant of V. vinifera cv. Ugni Blanc, which produces fleshless berries similar to those observed in wild genotypes, was identified as an opportunity to get insights into the control of berry development and berry size [20]. It has been shown that the drastic phenotypic changes observed in berry development are controlled by a dominant mutation in the fleshless berry (flb) gene [21]. Like the fw2-2 gene in tomato, the flb gene impairs cell divisions in the developing ovaries [21]. The closest genetic marker linked to the flb mutation defines a 6 cM region located at the top the chromosome 18 that corresponds to a physical distance of 948 kb according to the last version of the grapevine genome assembly http://urgi.versailles.inra.fr/index.php/urgi/Species/Vitis/Resources; in this region, no homolog to the fw2-2 gene has been identified. Considering the importance of berry size for wine quality, a fine mapping of the flb mutation was thus started for its molecular identification.

Here we describe our efforts in reducing the genome interval of the region carrying the flb mutation. We first started by a classical genetic mapping approach. We showed that the mutation is located on a completely homozygous portion of chromosome 18 in Ugni Blanc mutant. No marker could thus be found in coupling with the mutation and the classical approach was abandoned.

We therefore started another approach similar to the one previously proposed for fw2-2 gene in tomato [9]. Since the berries of Ugni Blanc mutant mimic wild V. vinifera berries (both types of berries have little to no flesh and carry round shaped seeds typical of wild genotypes) [13, 20, 22], we hypothesized that flb gene could have been one of the genes selected during the domestication process of grapevine. If so, a signature of selection or selective sweep could be found around this gene. Under this assumption, we performed a preliminary scan of the sequence polymorphism of the flb region in a collection cultivated and wild grapevine genotypes.

With an initial objective to develop markers tightly flanking the flb mutation, 62 genomic regions were scanned for polymorphism along a 1.4 Mb region at the top of chromosome 18, where the mutation was previously located [21]. These regions were either genotyped or sequenced in the genotype carrying the mutation, Ugni Blanc mutant, its wild type (Ugni Blanc) and Chardonnay, which was the other parent of a full sib family segregating for the mutation. The sequenced fragments or markers were completely homozygous in Ugni Blanc and Ugni Blanc mutant, with one marker analyzed each 23 kb in average. Indeed, while analyzing the genome sequence of the heterozygous grapevine cultivar Pinot Noir, Velasco et al [43] showed that, like in other heterozygous species, the frequency of SNPs or INDELs varied along the grapevine genome and found some evidence for scarce quasi-homozygous areas. Here we describe a region of 1 Mb probably completely homozygous that raised two questions. First, as Velasco et al [43] showed that over 65 Mb of sequence are hemizygous in Pinot Noir, we wanted to check if our observations were due to a real homozygosity or to a deletion of a large portion of the top of chromosome 18 in one haplotype of Ugni Blanc. We addressed this issue by two different experiments (Figure 2), a qPCR estimation of the number of copies of a single gene in the homozygous area (FL) compared to genes elsewhere in the genome (three HMGCoA genes). The same number of copies was estimated for FL gene for Ugni Blanc mutant and Chardonnay which is heterozygous in this region. Second, a BAC-FISH hybridization on Ugni Blanc mutant metaphase chromosomes using a BAC clone located in the area was carried out and showed a signal on both homologous chromosomes. We therefore un-ambiguously demonstrated that our observations corresponded to a real homozygosity in Ugni Blanc mutant. This would be consistent with the fact that hemizygous regions identified by Velasco et al [43] would mainly correspond to stretches of repeated sequences, which is not the case of the flb region. These results raised the question whether this high level of homozygosity in Ugni Blanc mutant was restricted to the top of chromosome 18. The scoring of 480 SNPs [44] and 20 SSR ([45], V. Laucou personal communication) regularly spread along the genome showed that whereas this cultivar seems slightly more homozygous in average than for instance Cabernet Sauvignon, Syrah or Chardonnay, the near complete homozygosity observed in the flb region in Ugni Blanc mutant is not the rule on the rest of the genome and may be restricted to this region only. A mechanism which could explain the formation of such large homozygous region in a highly heterozygous out-crosser like grapevine would involve the repair of a DNA double-strand break [46]. When analyzing diversity in the cultivated germplasm, we observed that the cultivar Orbois was also completely homozygous for all fragments re-sequenced at the top of chromosome 18, and confirmed by qPCR assay that it was also due to real homozygosity (data not shown). Whatever its origin, this unexpected result made impossible the fine mapping of the mutation in the available segregating F1 population, which would necessitate the development of a F2 population.

Before having such a population available, we tested another possibility for reducing the interval carrying the flb gene, based on the fact that flb could be a gene selected during grape domestication. Indeed, the berry and seed phenotypes of Ugni Blanc mutant look like the phenotypes of wild V. vinifera seeds and berries [20, 22]. We searched for signatures of selection in the flb region in a set of cultivated genotypes. For this purpose, we sequenced in 33 individuals (26 cultivated and 7 wild genotypes) (i) 69 gene fragments for a total of 34,355 kb along 948 kb in the flb region and (ii) 77 additional, totaling 48,874 kb spread along the 19 grapevine chromosomes. As already observed by Vezzulli et al [41], the nucleotide diversity was lower in average in the set of wild genotypes (π = 0.27) than in the set of cultivated genotypes (π = 0.35). This difference was increased in the flb region probably by the fact that a quarter of the fragments showed no sequence polymorphism at all among the wild genotypes whereas only one was in this case in the domesticated compartment. In the present work (but not in Vezzulli et al [42]) the sample of domesticated genotypes was selected after a comprehensive analysis of the world-wide largest collection of domesticated grapevine accessions [23] with the aim to retain a maximum of diversity. Unlike in Vezzulli et al [42], very few specific SNPs were found in the set of wild genotypes (six SNPs out of 554). All these observations, opposite to what has been observed in many other species [47, 48], could be due to the fact that only seven to ten wild genotypes were sequenced in both studies and that their choice could not be driven by a maximization of the diversity along their complete natural area of growth. However, several surveys including accessions from the wild V. vinifera germplasm also showed this overall lower genetic diversity compared to the cultivated germplasm [49–51]. Indeed, small population sizes [52, 53] as well as dioecy [54] could explain the observed reduced diversity in the wild V. vinifera gene pool, while multiple domestication events [50] and a continuous breeding process the larger diversity in the cultivated gene pool. All these results including ours will have to be confirmed with a larger sample of wild V. vinifera, taking into account all the geographic area where it grows and its overall genetic diversity.

Putative signatures of selection (Tajima's D parameter significantly deviating from neutrality associated with differences between the cultivated and wild pools in sequence diversity) in the cultivated pool of genotypes and SNPs showing a significant association with berry weight variation were found in the flb region (Figure 5). Moreover, LD was found in this region, the genes under selection presenting intragenic LD and intergenic LD with nearby genes (Figure 6), which strengthened the hypothesis that they might be under selection [48]. Like in Fournier-Level et al [55], no LD was however observed between the six out of eight genes possibly under selection. It is well known that the LD extent varies between the organisms and genomic regions [56]. In grapevine, a recent preliminary genome wide study confirmed the low extent of LD in average (less than 3-10 kb), suggesting a large effective size in the grapevine population at the origin of the current domesticated pool [57]. However, in recent history, vegetative propagation and long intervals between generations may have reduced the impact of recombination, maintaining extensive linkage disequilibrium in some regions under selection [58].

The significant genetic associations found between berry weight variation and SNPs in this region were not found in the fragments putatively under selection. It is still possible that one of these genes is involved into the berry weight variation and that the causative sequence polymorphism was not in the exon fragments sequenced. Indeed, trait variation is often due to sequence variation in regulatory regions (see for instance [9, 11, 55]) and such variant sites may be more tightly linked to a neighbor gene. Interestingly, the orthologs in Aradidopsis for the six out of eight gene with signature of selection (At1g76540 for VVC2946A, At1g76400 for VV05791A, At1g42970 for VVC2897A, At1g42540 for VV05785A, At5g43820 for VVC2901 and At1g20696 for VVC2892A) and for the two genes with SNPs associated with berry weight variation (At5g24306 for VV05786A and At1g42440 for VV05775A), are all expressed in flowers and showed a peak of expression at the beginning or during the flowering (http://urgv.evry.inra.fr/projects/FLAGdb++/HTML/index.shtml). In tomato, the genes involved into fruit size variation have been shown to be expressed in very early stages of its development, starting at floral development [2, 11]. Finally, only one gene fragment, VVC2897A, might be under purifying selection in the cultivated pool with a major haplotype in cultivated pool and a LD extended to a neighbor gene which present a SNP significantly associated with berry weight variation (Figure 5 Figure 6). This would fit with the hypothesis that, like in tomato, the selection of a new haplotype by humans would have ensured the transition from berries with little flesh in wild grapevines to berries with more flesh in cultivated grapevines [2, 11]. The real involvement of this gene into berry size variation and in the domestication syndrome remains however to be proven.

While searching for SNP markers in coupling with the fleshless berry mutation, we observed the occurrence of a 1 Mb homozygous region, not associated with repetitive sequences, in the grapevine otherwise highly heterozygous genome. We demonstrated the feasibility to use BAC-FISH on the very small grapevine chromosomes and provided a specific probe for the identification of chromosome 18 on cytogenetic map. Using this method, we showed that the observed homozygosity was not due to a large deletion.

We then searched for signatures of domestication for berry weight along the flb region by re-sequencing 69 gene fragments in 26 domesticated and seven wild V. vinifera genotypes. We found putative signatures of selection associated with significant differences in nucleotide diversity between the cultivated and the wild pool only in one gene (VVC2897A) and also SNPs significantly associated with berry weight variation in three other genes among which one is in DL with VVC2897A. The involvement of these four genes into berry weight variation in grapevine remains to be proved by further functional experiments. In addition, we detected 554 SNPs along the flb region. These polymorphisms could serve to develop a genotyping chip useful for a future fine mapping of the flb gene in a F2 population and for the analysis of genetic diversity in larger sets of wild and cultivated genotypes.