Polymorphisms and minihaplotypes in the VvNAC26 gene associate with berry size variation in grapevine

Domestication and selection of Vitis vinifera L. for table and wine grapes has led to a large level of berry size diversity in current grapevine cultivars. Identifying the genetic basis for this natural variation is paramount both for breeding programs and for elucidating which genes contributed to crop evolution during domestication and selection processes. The gene VvNAC26, which encodes a NAC domain-containing transcription factor, has been related to the early development of grapevine flowers and berries. It was selected as candidate gene for an association study to elucidate its possible participation in the natural variation of reproductive traits in cultivated grapevine. A grapevine collection of 114 varieties was characterized during three consecutive seasons for different berry and bunch traits. The promoter and coding regions of VvNAC26 gene (VIT_01s0026g02710) were sequenced in all the varieties of the collection, and the existing polymorphisms (SNP and INDEL) were detected. The corresponding haplotypes were inferred and used for a phylogenetic analysis. The possible associations between genotypic and phenotypic data were analyzed independently for each season data, using different models and significance thresholds. A total of 30 non-rare polymorphisms were detected in the VvNAC26 sequence, and 26 different haplotypes were inferred. Phylogenetic analysis revealed their clustering in two major haplogroups with marked phenotypic differences in berry size between varieties harboring haplogroup-specific alleles. After correcting the statistical models for the effect of the population genetic stratification, we found a set of polymorphisms associated with berry size explaining between 8.4 and 21.7 % (R2) of trait variance, including those generating the differentiation between both haplogroups. Haplotypes built from only three polymorphisms (minihaplotypes) were also associated with this trait (R2: 17.5 – 26.6 %), supporting the involvement of this gene in the natural variation for berry size. Our results suggest the participation of VvNAC26 in the determination of the grape berry final size. Different VvNAC26 polymorphisms and their combination showed to be associated with different features of the fruit. The phylogenetic relationships between the VvNAC26 haplotypes and the association results indicate that this nucleotide variation may have contributed to the differentiation between table and wine grapes.


Background
Grapes are one of the most valuable and extensively cultivated fruits, mainly grown for their transformation into wine, juice or raisins, and for direct consumption as fresh fruit [1]. The cultivated grapevine (Vitis vinifera subsp. sativa) derives from its wild ancestor (Vitis vinifera subsp. sylvestris) through several domestication processes [2,3]. Archeological findings suggest that primary domestication events could have taken place between the seventh and fourth millennia BC in the Near East region located between the Black and Caspian seas [4][5][6]. From there, those initial cultivars would had been spread by human civilizations in different directions [4]. Additional secondary domestication events and spontaneous hybridizations among selected individuals and local wild populations likely contributed to the evolution of current cultivars, since the ancestor species was present all around the Mediterranean sea [7,8]. Current cultivated grapevine shows important modifications compared to its wild relative, including the radical change in the sexual form of the plant -from dioecy to hermaphroditism-, and the increase in the number of berries per bunch and their individual size [4,5,[9][10][11].
As for other crops, fruit size is a trait that was preferentially selected during the domestication of grapevine [4,[10][11][12]. Because of the selection to increase yield, berries from cultivated varieties are larger than those from their wild ancestor [2,4]. Moreover, specific berry features have been selected for either wine or table grape production [1,4]. In this light, cultivars with large and fleshy berries are preferred for their use as table grape varieties, whereas cultivars with smaller and juicier berries and a higher skin-to-flesh ratio are preferred for winemaking [2,13]. The existence of divergent selection has likely contributed to the large diversity that can be found nowadays for berry morphology [11,14]. Variation in berry and bunch traits allowed the distinction of three morphotype groups (or proles): the occidentalis, grouping the small-berried wine cultivars of Western Europe, the orientalis, composed by the large-berried table cultivars of Central Asia, and the pontica, with cultivars with an intermediate phenotype and grown around the Black Sea and in Eastern Europe [15]. Relationships between these morphotypes and different nuclear and chloroplast haplotypes have been proposed [7,16], suggesting the use of different genetic pools for the development of wine and table cultivars in different geographical regions. Recently, Bacilieri et al. [2] studied the genetic structure of more than 2000 grapevine accessions, identifying the existence of three main genetic groups in agreement with the morphotypes classification. Additional stratification identified five different genetic groups: a group of wine and table cultivars from the Iberian Peninsula and Maghreb (S-5.1), a group of table cultivars from Far-and Middle-East countries (S-5.2), a group of wine cultivars from West and Central Europe (S-5.3), a group comprising mostly bred table grape cultivars from Italy and Central Europe (S-5.4), and a group of wine cultivars from the Balkans and East Europe (S-5.5) [2]. In a similar approach, Emanuelli et al. [3] identified four genetic groups in 1659 sativa grapevine genotypes by means of a set of SSR markers: a group of Italian/Balkan wine cultivars (VV1), a group of Mediterranean table/wine grapes (VV2), a third group with the Muscats varieties (VV3), and a group of Central European wine grapes (VV4).
To date, several quantitative trait loci (QTL) for berry size have been detected through the analysis of different grapevine progenies from crosses involving either wine or table varieties as parents [17][18][19][20][21][22]. Although this approach has provided useful information for the analysis of the trait, the results are usually restricted to the analyzed progenies [23]. In this sense, association mapping searches for variation in a much broader genetic context, enabling the exploitation of the diversity that is naturally present in a crop as a result of centuries of evolution [24]. Two types of association methods are currently used for the dissection of complex traits: genome-wide association studies (GWAS) and candidate-gene association mapping [24,25]. The last one is a hypothesis-driven approach that requires of a candidate gene selected on the basis of previous results obtained from genetic, functional or physiological studies [24,25]. This approach has been successfully applied in grapevine studies providing evidence for the role of VvMyb genes in the anthocyanin content of berry skin [26,27], VvDXS in Muscat flavour [28], VvPel and VvGaI1 in berry texture [29,30], VvAGL11 in seedlessness [31], and VvTFL1A in flowering time, berry weight and bunch width [32].
NAC domain-containing proteins [from Petunia NO APICAL MERISTEM (NAM) and Arabidopsis TRAN-SCRIPTION ACTIVATION FACTOR (ATAF1,2) and CUP-SHAPED COTYLEDON (CUC)] are one of the largest families of plant-specific transcription factors, being characterized in a wide range of land plants [33]. NAC proteins contain a highly conserved domain at the N terminus (NAC domain) and a highly divergent transcriptional regulatory region in the C-terminal region that determine the specific function of the protein [33,34]. The NAC domain consists of approximately 150-160 amino acids, and is divided into five well-conserved subdomains [34]. This region holds DNA binding activity and can be responsible for protein binding and dimerization [34,35]. This transcriptional factor family has been related to different developmental and morphogenetic processes in Arabidopsis [36][37][38][39][40][41] and other species [42][43][44][45][46][47].
Regarding grapevine, 74 different NAC-like genes (VvNAC) have been identified in the reference genome version 0 [48] and 75 in version 1 [49]. According to their homology to AtNAC genes, some have been predicted to play different roles during grapevine development [48]. In a recent phylogenetic analysis performed between the NAC sequences from V. vinifera, Arabidopsis thaliana, Oryza sativa and Musa acuminata, VvNAC26 showed to be the closest homologue to Arabidopsis NAC-LIKE, ACTIVATED BY AP3/PI (NAP, also known as AtNAP or ANAC029) [50]. AtNAP is a target gene of the flower homeotic transcription factors APETALA3/PISTILLATA (AP3/PI) [38,51], two MADS-box genes required for the determination of petal and stamen identities during flower development in Arabidopsis. In grapevine, Fernandez et al. [52] identified the specific over-expression of a putative AtNAP homolog during the development of flowers and berries of the extreme fleshless berry flb mutant of the cultivar Ugni Blanc, suggesting the involvement of this NAC transcription factor in berry flesh morphogenesis. In fact, VvNAP is also up-regulated in berries of cvs. Ugni Blanc and Cabernet Sauvignon before the onset of ripening [52], suggesting its involvement in normal berry development.
Considering the function of NAP in Arabidopsis cell growth [38] and the likely involvement of its grapevine homolog in berry development and growth [52], VvNAC26 was selected as a candidate gene to analyze its contribution to fruit size natural variation in the cultivated grapevine. VvNAC26 was sequenced in a set of table and wine grapevine varieties that were described over three consecutive years for nine berry and bunch traits. Additional tests to evaluate the linkage disequilibrium (LD) between the polymorphisms detected along the VvNAC26 sequence and the likely stratification of the grapevine varieties used in this work were performed to reduce the presence of false positive marker/trait associations. Moreover, VvNAC26 haplotypes inference and analyses gave us insights of the likely evolution of the gene considering the origin of the varieties used in this study. Lastly, reduced ancestral haplotypes (minihaplotypes) showing association with berry size were identified.

Plant material
A total of 114 grapevine varieties (including 111 V. vinifera cultivars and three inter-specific hybrids) held at the Grapevine Germplasm Collection of the Instituto de Ciencias de la Vid y del Vino (ICVV,FAO Institute Code: ESP-217) were considered (Additional file 1). Most of the cultivars used in this work come from Spain, France, Portugal and Italy. They are maintained under the same agronomical conditions in two separated experimental plots: "Finca Valdegón" (Agoncillo, La Rioja, Spain) and "Finca La Grajera" (Logroño, La Rioja, Spain). Plants at "Finca La Grajera" (5 years old) come from scions taken from "Finca Valdegón" (20-30 years old). This set of varieties was described in three consecutive vintages: 2011 and 2012 (in "Finca Valdegón") and 2013 (in "Finca La Grajera"). Information on the origin, main use and pedigree of the varieties was obtained from the Vitis International Variety Catalogue (VIVC, http://www.vivc.de, accessed: March 2015) (Additional file 1).

Phenotypic data
Due to inter-annual fluctuations, all grapevine varieties could not be described for the three seasons. Thus, 98, 104 and 97 varieties were sampled in 2011, 2012 and 2013 respectively. As a rule, ten mature bunches (at growth stage E-L 38 [53]) were collected per variety and characterized for nine berry and bunch traits (Table 1) as described previously [54,55]. To better fit the assumption of normality in the statistical analyses, the variable "Bunch weight" was square-root transformed, whereas variables "Berry weight" and "Berry volume" were logarithmically transformed. Phenotypic distribution of the traits considered in this study can be found in Additional file 2. Correlations between traits and seasons were performed with SPSS v.22.0 (IBM, Chicago, IL, USA) using the Pearson correlation coefficient.

Genotypic data
Young leaves from the 114 grapevine varieties were sampled and stored at -80°C until DNA extraction. Genomic DNA was isolated using the DNeasy Plant Mini kit (Qiagen, Valencia, CA, USA), following the instructions provided by the manufacturer. DNA was qualitatively and quantitatively evaluated by visual comparison with lambda DNA on ethidium bromide-stained agarose gels (0.8 %), and a NanoDrop 2000 spectrophotometer (Thermo Scientific, Wilmington, DE, USA). Nine nuclear SSR loci (VVS2, VVMD5, VVMD27, VVMD28, ssrVrZAG29, ssrVrZAG62, ssrVrZAG67, ssrVrZAG83 and ssrVrZAG112 [56]) and four chloroplast SSR loci (cpSSR3, cpSSR5, cpSSR10 [57] and cpSSR9 [58]) were analyzed in the 114 varieties. Polymerase chain reaction (PCR), separation of fragments, and data analysis were performed following the procedure detailed in Ibáñez et al. [59]. Pair-wise multilocus comparison with the ICVV nuclear and chloroplast SSR database and The European Vitis database (http://www.eu-vitis.de) was performed for the genetic identification of the variety. Chlorotypes were named according to Arroyo-García et al. [7]. The VvNAC26 gene (VIT_01s0026g02710), including 1000 bp in the promoter region according to grapevine 12X V1 gene predictions (http://genomes.cribi.unipd.it/ gb2/gbrowse/public/vitis_vinifera/), was sequenced together with other set of genes (data not shown). A region of 2184 bp (chr01_12442003:12444186) was targeted for next-generation sequencing (NGS) following a protocol based on the Agilent SureSelect Target Enrichment workflow (http://www.genomics.agilent.com). Paired-end libraries with an insert size of approximately 350 bp were sequenced in an Illumina HiSeq 2000 platform by BGI company (http://www.genomics.cn/en). Target enrichment and sequencing were carried out by BGI. Resulting reads had an average size of 90 nt, and were aligned to the whole 12X V1 Vitis vinifera PN40024 reference genome [60] with Bowtie 2 [61] using the following command line settings: -phred64 -end-to-end -N 0 -L 25 -gbar 2 -np 6 -rdg 6,4 -X 400 -fr -no-unal. The variant caller utility implemented in the SAMtools package [62] was used to detect polymorphisms (SNPs and INDELs) between the reference genome and each of the 114 sequenced varieties. These initially detected polymorphisms were filtered to generate a consensus genotype per variety by means of an ad hoc Perl script in which thresholds of quality score, read depth and frequency of base calls were considered (the source code of the script and a complete description of filtering parameters are available at https://github.com/ ratope/VcfFilter). To verify the consistency of variant calling, polymorphisms were individually checked with the Integrative Genomics Viewer (IGV) software [63]. Polymorphisms are named as suggested by Fernandez et al. [32], using the abbreviation "IND" for the designation of INDELs. Linkage disequilibrium (LD) was estimated considering polymorphisms with a minor allele frequency (MAF) higher than 5 %, by calculating the genotypic correlation coefficient (r 2 ) together with its associated P-value by a built-in function of TASSEL v.3.0 (http:// www.maizegenetics.net/) [64], and LD-blocks were determined considering a critical r 2 value of 0.8.
Prediction of the likely effect of the detected polymorphisms in the encoded protein was carried out with SnpEff v.4.0 [65], and effects of single amino acid substitutions on protein function were predicted in parallel with SNAP [66] and PROVEAN [67] utilities. We also checked for their likely effect on the mRNA secondary structure using two independent web-based applications: RNAsnp [68] and RNAstructure [69].
To predict the likely effect of the polymorphisms located in the promoter, we carried out the detection of the putative regulatory motifs with PlantCARE [70].

VvNAC26 haplotypes and nucleotide diversity analyses
Haplotype inference and diplotype (haplotype pair) estimation were performed with the partition-ligation-expectationmaximization (PLEM) algorithm [71] implemented in PHASE v.2.1, using default settings [72]. Haplotype clustering was carried out by SPSS v. 22.0 (IBM, Chicago, IL) using Ward's hierarchical method. Haplotypes were tested for recombination using the MaxChi, Chimaera and 3Seq algorithms implemented in the Recombination Detection Program v.4.46 (RDP4) [73] with default settings. A median-joining network [74] was constructed for the inferred haplotypes with the software Network v.4.6 (www.fluxus-engineering.com). Molecular diversity was evaluated through the calculation of the nucleotide diversity (π) [75] and the Watterson θ estimate [76] with DnaSP v.5.10 [77]. This software was also employed to obtain insights for testing likely deviations from neutrality, through the computation of Tajima's D [78] and Fu and Li's D* [79] tests. They were calculated for the whole set of haplotypes and separately for the genetic groups detected by STRUCTURE v.2.3, as suggested in Fernandez et al. [32].

Population genetic structure and kinship matrix
The number of genetic groups in the grapevine collection analyzed was estimated by the Bayesian approach implemented in the software package STRUCTURE v.2.3 [80]. It was run on the basis of the nine nuclear SSR markers using an admixture model with uncorrelated allele frequencies.
This model was tested in a number of hypothetical genetic groups ranging from 1 to 15, with 100,000 burn-in iterations followed by 150,000 Markov Chain Monte Carlo (MCMC) iterations for an accurate estimation. Each number of likely genetic groups was performed in 5 independent runs to verify the consistency of the results.
The most probable number of genetic groups was assessed following the criteria proposed by Evanno et al. [81], as implemented in STRUCTURE HARVESTER [82]. Once the optimal number of genetic groups was detected, we used CLUMPP v.1.1 [83] to align the 5 different runs, and the consensus matrix (Q) was used for association analyses. DISTRUCT v.1.1 [84] was used for the graphical visualization and analysis of the population structure. Grapevine varieties were assigned to a genetic group when its membership coefficient was 0.75 or higher; genotypes with no scores over this value were considered as "admixed". As suggested by Ruggieri et al. [85], the effect of the population structure on the variation of the traits considered was evaluated by multiple regression analysis, performed with SPSS v.22.0 (IBM, Chicago, IL, USA).
A kinship matrix (K) was constructed for obtaining the estimators of pairwise relatedness proposed by Wang [86] for our set of varieties, using the related package [87] for R v.3.2.2 (http://www.r-project.org/). They were estimated on the basis of 25 SSR: the mentioned set of 9 SSR markers plus 16 additional SSR markers obtained for 102 varieties from available data previously published by Lacombe et al. [88] and de Andrés et al. [89].

Association analyses
Association analyses between genotypic and phenotypic data were performed separately for 2011, 2012 and 2013 seasons, considering only those polymorphic sites with a MAF ≥ 5 % and the average value obtained for the bunches analyzed of each accession. Four different models were tested using TASSEL v.3.0 [64] to detect the most conservative one, using the P3D (Population Parameters Previously Determined) method and an optimum level of compression as estimation variables. The four methods tested were: Naïve model [a General Linear Model (GLM) without any correction for population structure]; Q model (a GLM model with fixed population structure as covariate); K model [a Mixed Linear Model (MLM) with kinship K as correction factor]; and Q + K model [a MLM model capable to correct for both population structure (Q) and kinship (K) effects [90]]. Association results indicated the last one as the most stringent one (Additional file 3), so only their results are shown and discussed.
To assess significance level, a multiple testing correction based on the number of tests was performed. It was determined considering the number of traits evaluated and the number of independent markers analyzed, which was determined by counting one polymorphism per LD-block plus all interblock polymorphisms [91]. Two thresholds for the P-value were considered: the first one (P-value ≤ 3. 27E -4 ) corresponds to the stringent Bonferroni corrected level for α = 0.05, the second one (P-value ≤ 6.53E -3 ) allows the appearance of one false positive per multiple testing [91].
As suggested by Carter et al. [92], association analyses were also performed between the phenotypic data and a set of reduced haplotypes (minihaplotypes, MH), which were inferred as previously detailed but considering only the most informative polymorphisms. Since nine traits were tested per year, associations showing a P-value lower than 5.55E -3 (the Bonferroni-corrected threshold for nine comparisons for α = 0.05) were considered as significant.

Phenotypic data
A large phenotypic variation was found for the traits evaluated in our set of grapevine varieties (Table 1). Similar levels of variation have been described for these traits in different core collections [11,32], supporting the actual adequateness of the plant material. Variation in fruit size parameters in different years was highly correlated (Additional file 4) what, in addition to high values of broad sense heritability for the studied traits in this set of varieties (data not shown), suggest the existence of a strong genetic component for the observed phenotypic variation in fruit growth-related traits. Interestingly, we found no significant correlation (or it was very low) between the number of seeds per berry and the different berry traits included in this study, in accordance with Houel et al. [11].

Population genetic structure
The existence of population stratification can lead to spurious marker/trait associations given the geographical origin, local adaptation and breeding history of the plant material [24]. STRUCTURE analysis and Evanno's ΔK method suggested the most likely existence of three genetic groups (k1, k2 and k3) (Additional file 5) using 9 SSRs. This set of markers led to a more reliable structure (in base to knowledge on genetic and geographical origin and use of the cultivars) and more conservative association results (lower P-values and R 2 ) than a set of 261 SNP markers (data not shown). Similarly, results using 9 SSRs were compared to those obtained using the set of 25 markers used for kinship estimation (see Material and Methods). Membership coefficients given by the 9 SSR and 25 SSR structures (both obtained by means of CLUMPP) showed a high level of significant correlation (r = 0.9; p < 0.001), and association results were similar (data not shown). Because of the presence of missing values in 12 individuals for 16 SSRs, and the sensitive of STRUCTURE to individuals poorly genotyped [93], the structure based on 9 SSR markers was further considered in this study as correction factor.
Considering a membership coefficient of 0.75 as a critical threshold for the assignation to a genetic group, k1, k2 and k3 include 35, 10 and 25 grapevine varieties respectively, whereas 44 varieties were considered as admixed (Fig. 1). This large proportion of admixed genotypes is in agreement with previous findings [2]. We found that this Q = 3 structure is consistent with both the geographic origin and the main use of the varieties considered in this work (Additional file 1). The genetic group k1 mainly contains Iberian wine or mixed use varieties (e.g.: Airén, Palomino Fino, Tempranillo). Group k2 is primarily composed by varieties mainly grown for producing table grapes, and typically considered part of the orientalis morphotype proposed by Negrul [15]. This group clusters some Muscat and Muscat-derived varieties (like Muscat Hamburg, Alphonse Lavallee and Italia), and other not related varieties (e.g.: Afus Ali, Dominga). k3 mostly includes wine varieties from Western Europe (e.g.: Aligoté, Cabernet Sauvignon, Traminer) and some grown in the Northwest of the Iberian Peninsula (e.g.: Alfrocheiro, Alvarinho). Most of the varieties included in groups k1 and k3 have the morphological features of the occidentalis morphotype [15]. Interestingly, the structure analyses clusters Northwest Iberian wine varieties with European wine varieties, agreeing with recent results that connect those varieties through the parent-offspring relationship existing between Alfrocheiro and Traminer (or Savagnin) [94]. The three genetic groups can be identified as three of the five genetic groups proposed by Bacilieri et al. [2]. In this sense, k1 can be related to the S-5.1 group (Wine and Table/Iberian Peninsula and Maghreb), k2 to S-5.4 (Table/Italian and Central Europe breeds), and k3 to S-5.3 (Wine/West and Central Europe) [2]. Moreover, they show agreement with three of the four groups suggested by Emanuelli et al. [3], with k1 related to the VV2 group (Mediterranean table/ wine grapes), k2 to VV3 (Muscats) and k3 to VV4 (Central European wine grapes).
Chlorotypes have been related with the geographical origin and use of the varieties, and therefore we also considered them in this work (Table 2 and Additional file 1). Chlorotype A was the most common one in the whole set of varieties analyzed (54.4 %), followed by the chlorotypes D (25.4 %) and C (14.0 %); chlorotype B (4.4 %) was only found in varieties attributed to k2 or in admixed varieties. Chlorotype A (characteristic of Western Europe and Northern Africa [7]) was frequently found in the genetic group k1, whereas chlorotype C (commonly found in varieties of Central Europe [7]) was mostly found in varieties of k3. In this genetic group, we also found a high number of varieties with chlorotype A, due to the inclusion of Northwest Iberian varieties, as mentioned above.
Multiple regression analyses were run to evaluate the effect of this stratification on the nine considered traits (Additional file 6). Moderate and significant (P ≤ 0.001) effects were detected for the four berry traits considered, whereas larger effects for bunch length, width and weight were observed, especially for 2013 data, when more than 40 % of phenotypic variance for these bunch traits was explained by the population structure. No significant effect on the number of seeds per berry was observed, whereas the number of berries per bunch was only significantly related in 2011.
Altogether, STRUCTURE results were considered as appropriate and capable to correct for most of spurious associations, so membership coefficients were included in the association tests.

VvNAC26 polymorphisms
A total of 2184 bp of the VvNAC26 gene, including 1000 bp of the promoter region, were sequenced in the 114 grapevine varieties. Sequencing and alignment results showed a 100 % coverage (min 20 reads; 93.8 % of sequence over 80 reads; average coverage depth: 117.5 ± 16.7) in all the grapevine varieties. Data can be accessed Fig. 1 Population structure of the 114 varieties included in this study based on STRUCTURE [80]. The optimal number of genetic groups (K = 3) was set according to Evanno's method [81]. Each variety is represented by a vertical line, divided in colored segments according to the proportion of estimated membership in the three genetic groups: k1 (red), k2 (green), and k3 (blue). Considering that a variety was assigned to a genetic group if its membership is over 0.75, k1, k2 and k3 are composed by 35, 10 and 25 individuals, respectively Frequencies are shown for the global collection (n = 114 varieties) and in the three genetic groups detected by STRUCTURE: k1 (n = 35), k2 (n = 10) and k3 (n = 25) and in the admixed varieties (n = 44). Chlorotype names are given according to Arroyo-García et al. [7] at NCBI's Sequence Read Archive (SRA) under the accession code SRP057099. The locus structure annotated for the PN40024 reference genome [60]

VvNAC26 haplotypes
On the basis of the 69 polymorphisms detected (Additional file 7), the PLEM algorithm [71] implemented in PHASE inferred 26 different haplotypes, including 9 unique haplotypes (present in 1 variety, frequency 0.4 %) ( Table 3). None of the algorithms used in the RDP4 software indicated any evidence of recombination in the 26 haplotypes. Only four haplotypes (H3, H17, H19 and H20) showed a frequency ≥5 %, accounting for 72.8 % of the haplotypes in the grapevine varieties analyzed. H3 was exclusively found in varieties of the k3 genetic group or in admixed varieties; H17 was found in the three groups, with a major presence in k1 and k3; H19 was found only in k1 and k2; and H20 was found in varieties assigned to any of the genetic groups (Table 3). Only four different haplotypes were found in the 10 varieties attributed to the k2 group (H8, H17, H19 and H20) (Table 3), with four table grape varieties (Italia, Cardinal, Paraiso and Afus Ali) being homozygous for the haplotype H20 (Additional file 1).
The diversity parameters and neutrality tests calculated for the VvNAC26 gene sequence in the whole set of varieties and in the three genetic groups are shown in Additional file 9. Nucleotide diversity (π) and Watterson's estimate (θ) released values of 0.00657 and 0.00825 (respectively) for the 26 haplotypes found in the whole The hierarchical clustering of VvNAC26 haplotypes based on Ward's method revealed the presence of two groups of haplotypes (or haplogroups, HG): HGA, comprising 16 haplotypes (accounts for 25.4 % of the haplotype abundance in the set of varieties considered) and HGB, with the remaining 10 haplotypes (Additional file 10A). Accordingly, haplotype network discriminated these two haplogroups (Fig. 3), which differed in ten SNPs (W-962, K-779, W-592, R-160, Y-57, Y-50, S-1, R600, R626 and R780), mostly of the LD-block B (Additional file 8). The other detected LD-blocks are in minor branches of the network (data not shown), so they are not further discussed. Considering the distribution of the haplotypes in the three genetic groups, haplogroup HGA includes haplotypes mainly present in wine varieties of groups k1 and k3; only one variety assigned to the k2 genetic group (Barbera Nera, an Italian wine variety) was found to have a HGA haplotype (H8) (Additional file 1). The haplogroup HGA contains one of the most abundant haplotypes -H3exclusively found in varieties assigned to k3 (Fig. 3 and Table 3). Haplotypes in HGB were well distributed within the varieties assigned to the three genetic groups k1 (35.9 %), k2 (11.2 %) and k3 (15.3 %). This haplogroup contained the other three most abundant haplotypes found in the set of varieties analyzed (H17, H19 and H20, Fig. 3). As mentioned above, H20 was commonly found in the grapevine varieties assigned to the group k2 (Fig. 3).

Association tests
We found eight polymorphisms significantly associated with different berry and bunch traits with a P-value below the established threshold of 6.53E -3 . One of them still showed statistical significance when considering the more stringent threshold (3. 27E -4 ) ( Table 4). Six SNPs located in the LD-block B (W-962, W-596, R-160, Y-57, R600 and R780) showed a significant association with berry length, volume, weight and volume, explaining up to 12.28 % of berry length variation in 2013 ( Table 4). As stated before, the LD-block B was located in the phylogenetic branch differentiating HGA and HGB (Fig. 3).
Y117 -a synonymous SNP located in the first exon of VvNAC26 ( Fig. 2 and Additional file 7) -showed to be significantly associated with berry width, length, weight and volume, as well as with bunch length and weight (P ≤ 6.53E -3 ). P-values obtained for associations with berry length, volume weight and width in 2011 and 2012 were significant even when considering the more stringent threshold (3. 27E -4 ). The strongest association found was between Y117 and berry width in 2012 (P = 2.58E -6 ), and  (H1 -H26). Each haplotype is represented by a circle, which size (see code) is proportional to its frequency in the set of varieties analyzed. Their inner color/s indicate the proportion of varieties assigned to each of the genetic groups detected by STRUCTURE (see color code, Adm.: admixed). Lines connecting haplotypes represent phylogenetic branches, and small transversal lines represent mutational steps (only those polymorphisms significantly associated with berry and/or bunch traits appear named, according to Table 4 Table 5) the marker explained up to 21.7 % of trait variance (Table 4). In the phylogenetic network, this SNP was found in the haplogroup HGB, in the branch separating H17 from H18 (Fig. 3).
Indel IND-649, located in the promoter region, was also significantly associated with berry length, volume, weight and width in 2012 and bunch weight in 2013 (P ≤ 6.53E -3 ) ( Table 4). IND-649 was found in different positions in the network constructed for the 26 VvNAC26 haplotypes (Fig. 3). Specifically, it was found in the phylogenetic branch separating H20 from H18 in haplogroup HGB, as well as in the HGA haplogroup, in the branches separating H13 from H8 and H14 from H12. As stated above, IND-649 involves the insertion/deletion of a unique nucleotide, P-values of associations and variance explained by the marker (R 2 ) are indicated for the MLM models obtained for 2011, 2012 and 2013 *P-value ≤ 6.53E -3 ; **P-value ≤ 3.26E -4 and it was found to be located in a poly-T region, so the variation in this position leads to a (T) 9 or (T) 10 genotype. Alleles found in H13, H14, H20 and H21 are identical in size for this locus [(T) 9 ] but in the network they do not derive from a common ancestor, which may reflect size homoplasy in this site. As commented above, the automatic prediction carried out by means of SnpEff [65] revealed that SNP Y117 does not affect the primary structure of the protein (Additional file 7), and the mRNA structure analyses using two independent tools [68,69] predict that Y117 does not induce any structural change in its secondary structure (Additional file 11). Based on the SnpEff [65] and PlantCARE [70] results, only one SNP (W-962) of the LD-block B would be located in a regulatory region (a CAAT-box). Similar in silico analysis revealed that IND-649 is located in a TATA-box, suggesting the possible regulatory effect of both polymorphisms in VvNAC26 expression.

Associated polymorphisms define minihaplotypes associated with berry size
Single-marker associations and LD suggest that W-962 (representing the associated LD-block B), IND-649 and Y117 contribute particularly to the relationship found between VvNAC26 and berry traits, as well as to the phylogenetic clustering of the inferred haplotypes. In fact, the hierarchical clustering of the 26 haplotypes using only these three polymorphic sites is similar to that obtained when using the 69 polymorphisms, denoting their relevance in the clustering (Additional file 10A and B). To evaluate their joint effect on berry size, we used W-962, IND-649 and Y117 to infer a reduced set of polymorphism combinations (minihaplotypes, MH) for a haplotype-based association analysis, which has been suggested as a more powerful approach since it considers the underlying LD between different polymorphic sites [71,95,96]. Out of the eight possible theoretical combinations, we found five different minihaplotypes in the set of varieties analyzed (Table 5). They have variable frequencies in our set of grapevine varieties, with values ranging from 1.3 % (MH2) to 43.9 % (MH3), and they are unevenly distributed in the three genetic groups established by STRUCTURE: MH3 was the most abundant in the group k1 (50 %), MH5 in k2 (70 %) and MH1 (44 %) and MH3 (38 %) in k3 (Table 5). Minihaplotypes MH1 and MH2 were found in the haplogroup HGA, whereas MH3, MH4 and MH5 were found in HGB (Fig. 3). Thus, minihaplotypes were used for another association analysis, excluding MH2 due to its low frequency. They were also significantly associated with berry dimensions in 2011, 2013 and 2013 ( Table 6). The percentage of variance of the different traits explained by the minihaplotypes is higher than those explained by any of the individual polymorphisms (Table 4), suggesting an additive effect of these three markers in the phenotype of the berry.

Phenotypic values related to associated markers and minihaplotypes
As seen before, Y117 showed to be associated with the size of the berry ( Table 4). The minor allele of this polymorphism (T) was highly frequent in the grapevine collection used (30.7 %) (Additional file 7). Homozygous T:T varieties tend to produce larger berries than the heterozygous C:T and the homozygous C:C genotypes, which have similar berry dimensions in average (Fig. 4). In the same way, Y117 was associated with bunch weight and length, with the grapevine varieties containing two T alleles more prone to produce heavier and longer bunches than the other genotypes (Fig. 4). Similarly, homozygous individuals for the A allele at the SNP W-962 (selected for representing the LD-block B) tend to have bigger berries than those at heterozygous or homozygous states for the minor allele T, which showed a similar phenotype (Fig. 4). This minor allele was highly present in the grapevine collection (25.4 %). Finally, the deletion event at IND-649 (present in 25.4 % of the set of varieties) was associated with larger berries and heavier bunches (data not shown).
Phenotypic effects were also observed when considering the minihaplotypes built through the combination of these three polymorphic sites. Accordingly, varieties carrying in homozygosis the T allele at Y117, the A allele at W-962 and the deletion [(T) 9 ] at IND-649 (so MH5:MH5 varieties) showed the largest berries within the set of varieties evaluated (Fig. 5). As mentioned above, this minihaplotype was the most common one in the group k2 (Table 5),  (Table 5), showed the smallest berries (Fig. 5). This minihaplotype was commonly found in k3 (Table 5), a group mostly composed by occidentalis European wine varieties of small-sized berries (Additional file 1). Heterozygous individuals carrying both minihaplotypes (MH1:MH5) showed a similar phenotype than the homozygous individuals for the MH1 minihaplotype (MH1:MH1) (Fig. 5).

Discussion
Berry size depends on many genetic, developmental and environmental factors, including specific pre-anthesis flower features and multiple post-pollination events [11,97]. In Arabidopsis, the NAC domain containing protein NAP gene has been reported to be involved in multiple developmental processes, from the establishment of flower meristem identity and flower organ formation to fruit ripening and senescence [38,51,98]. A role in flower and berry development has been suggested for VvNAC26 [52], the grapevine NAP homolog [50], on the basis of its gene expression profile. As stated before, several QTL for berry size have been reported [17][18][19][20][21][22], but none of them in the region where VvNAC26 is located. This could be due to the fact that the progenies studied arise from crosses involving only wine or only table cultivars. VvNAC26 was chosen as a candidate gene that has been sequenced in a set of varieties to determine the existing nucleotide variation, and to identify its possible contribution to the natural variation observed for several reproductive traits in grapevine.
A relatively high rate of nucleotide variation was found for VvNAC26 in the grapevine varieties considered, with an average of one polymorphic site every 31 nucleotides. This variation is higher than the reported in other studies that included non-vinifera individuals for the analysis of the nucleotide variation of different grapevine genes [99,100]. Nonetheless, these works do not include the analysis of the promoter region, where we found a high number of polymorphic sites. The analysis of these regulating regions is paramount in association genetics surveys, since different variants in the gene promoter may correlate with different expression level and, ultimately, phenotypic diversity [101]. On the other hand, some of the rare polymorphisms detected in the VvNAC26 sequence were only found in the three interspecific hybrids included in this study, and they are likely attributable to their non-vinifera genetic background. As expected, we found a higher mutation rate in non-coding regions than in coding regions [102], and only twelve polymorphisms were detected in exonic regions. Four of them generated amino acid substitutions, although they are predicted to be neutral in the protein. As a result there is a high degree of conservation of the VvNAC26 protein in the cultivated grapevine. A high level of conservation was also reported for another grapevine NAC protein (VvNAC4), with only one nonsynonymous SNP detected in the gene sequence of 50 wild accessions and 73 cultivars [100]. Average intragenic LD calculated for all pairs of polymorphic sites with frequency over 5 %, was 0.25, similar to the average LD value reported for the VvMybA1 gene [27]. Six blocks of polymorphisms in high LD were identified in the VvNAC26 sequence and, as for other grapevine genes [28,32], some of those polymorphisms were found in high LD despite being largely separated in the nucleotide sequence.
The LD-block B separates the two main haplogroups (HGA and HGB) detected in the sequenced samples, and thus these polymorphisms could be related to ancestral alleles. Considering our set of grapevine varieties and according to the phylogenetic network and the hierarchical clustering of the VvNAC26 haplotypes, HGA and HGB show important differences. HGA includes 16 haplotypes found in low frequency in the global population studied, which are very divergent regarding the high number of polymorphisms found in this group, but very uniform in terms of their use and berry size (wine varieties/ small berries). On the other hand, HGB includes 10 haplotypes, genetically closer (less polymorphisms), and that are found indistinctly in wine and table varieties with diverse berry size.
A positive relationship between haplotype frequency and antiquity has been proposed [99]. Considering that   Table 4). Outliers are indicated as circles  Individual polymorphisms may cause relevant changes in gene expression or in protein function, which may ultimately cause alterations in a certain phenotype. However, polymorphisms are not inherited individually, but in LD with other genetic variants, in which certain alleles of close polymorphisms are found together. Consequently, the combination of some polymorphisms in minihaplotypes may have an stronger biological effect that single markers [95]. Consistent with the association results for the individual markers, the minihaplotype-based association analyses also released significant associations with berry traits. Homozygous individuals for the minihaplotype MH5 showed the biggest berries within the set of analyzed varieties, and all of them are mostly grown for the production of table grapes. Very interestingly, they present different chloroplast haplotypes (Afus Ali: A; Cardinal: B; Italia: C; Paraíso: D), indicating that they have different genetic origins (at least for the maternal lineage), and that this minihaplotype has been selected for table grape production in different genetic backgrounds. In this light, we analyzed the VvNAC26 sequence of cv. Red Globe, a highly appreciated table grape variety characterized by its very big berry size. It has no close relationship with the large-berried varieties studied here, and it is also homozygous for the MH5 genotype (data not shown), supporting the role of this minihaplotype in the berry size, independently of its genetic origin.
Putative functional effects of the three polymorphisms associated with berry size (W-962, IND-649 and Y117) are likely not related to the activity of the encoded protein. SNP W-962 (in LD-block B) and IND-649 are not located in the coding region, but in two common cis-regulatory elements. On the other hand, Y117 is a synonymous mutation, and in silico predictions showed no structural differences in the VvNAC26 mRNAs encoded by both variants in Y117. So, no effect in the stability and conformation of the transcribed VvNAC26 mRNA is expected, which might have affected critical post-transcriptional processes [103]. Considering the long intragenic LD observed for several polymorphic sites within VvNAC26, Y117 could be in LD with an undetected polymorphism responsible for trait variation [104], regulating gene expression and located outside the sequenced region. This situation has been previously suggested to explain the effect of a silent polymorphism of VvGAI1 associated with berry texture [30]. In fact, Clark et al. [105] confirmed the role of a cis-acting enhancer located between 41 and 69 kb upstream from the maize teosinte branched1 (tb1) gene starting site as the main causative factor controlling tb1 expression and tb1-related phenotypes. According to our results, it seems likely a functional effect of the VvNAC26 polymorphisms associated to berry size related to the regulation of gene transcription. Further analyses aimed at evaluating VvNAC26 expression levels in key stages of pistil and berry development in the extreme genotypes found (e.g.: MH1:MH1, MH1:MH5 and MH5:MH5) may yield additional information on the role of this gene and the associated polymorphisms in the final berry size. Consistently with the likely regulatory role of the associated polymorphisms, differential expression of VvNAC26 (=VvNAP) correlated with differential berry development and growth in the grapevine flb somatic variant (bearing fleshless berries), compared to the wild type Fernandez et al. [52]. In this somatic variant, high expression of VvNAP correlated with reduced berry growth. Indeed, Arabidopsis mutants over-expressing NAP showed a reduced size of several floral organs [38]. Altogether, these results suggest that the larger berry size observed for certain VvNAC26 variants might be a consequence of a reduced gene expression.
Analysis of VvNAC26 in the expression atlas developed for cv. Corvina [106] shows that, as seen for Arabidopsis NAP [38], VvNAC26 expression is not only related to VvPI expression (Additional file 12). In this line, a high expression of VvNAC26 is also appreciated in many other tissues, including senescing and mature tissues (Additional file 12) [106], in agreement with the promotion of senescence that have been proposed for NAP-like genes in Arabidopsis and other species [107][108][109]. Recent reports indicate that NAP could function via positive regulation of abscisic acid (ABA) biosynthesis [110][111][112], suggesting that VvNAC26 could mediate its responses by regulating the expression of ABArelated genes. High levels of ABA have been shown to inhibit cell growth in unpollinated tomato (Solanum lycopersicum L.) ovaries, keeping them in a dormant state until pollination [113]. In grapevine, a high level of ABA in flowers at full bloom (coincident with peaks of VvNAC26 expression, Additional file 12) and high levels of its degradation products after pollination have been reported [114,115]. Moreover, expression data reported for cv. Moscatel Rosada shows a high down-expression of VvNECD1 (involved in ABA biosynthesis) in very early pollinated ovaries when compared to the unpollinated ones [116]. These evidences suggest that polymorphisms reducing VvNAC26 expression might result in lower ABA levels, allowing a greater cell growth rate in ovaries and/or berries which ultimately would give place to larger berries. This hypothesis could be confirmed through analyses aimed at determining ABA levels in flowers and berries at several stages of development in different varieties bearing in homozygous state the extreme VvNAC26 minihaplotypes identified.
Association results presented here may have a potential limitation given the number of markers used for structure estimation. Thus, further studies aimed to verify these results are needed, using a different set of varieties. Replication of the genetic association study in additional independent samples is the better approach for verifying (or rejecting) associations [117,118]. Anyway, and considering the suggested role of VvNAC26 in the early development of grapevine flowers and berries [52], VvNAC26 and the polymorphisms and minihaplotypes detected in this work (whether causative or a result of allele selection during domestication and selection processes) are good candidates for their further validation prior their use in marker-assisted selection programs aimed to improve fruit size in grapevine breeding programs.

Conclusions
The analysis of the nucleotide sequence variation at the grapevine VvNAC26 gene and its association with grapevine reproductive traits has allowed the detection of polymorphisms recurrently associated with berry size. The phylogenetic analysis of the observed VvNAC26 haplotypes suggests that some of these polymorphisms could have been selected during the development of table grape varieties, given the key importance of the berry size in their use for fresh consumption. The sequence position and predicted functional effects of two associated polymorphisms suggest that they could affect the expression level of VvNAC26, what could have an effect on cell growth and berry size. Further analyses evaluating the associated VvNAC26 polymorphisms/haplotypes identified in this work are required to confirm this possibility, and also for using the associated polymorphisms for marker-assisted selection to improve fruit size in grapevine breeding programs.