Novel stable QTLs identification for berry quality traits based on high-density genetic linkage map construction in table grape

Background Aroma, berry firmness and berry shape are three main quality traits in table grape production, and also the important target traits in grapevine breeding. However, the information about their genetic mechanisms is limited, which results in low accuracy and efficiency of quality breeding in grapevine. Mapping and isolation of quantitative trait locus (QTLs) based on the construction of genetic linkage map is a powerful approach to decipher the genetic determinants of complex quantitative traits. Results In the present work, a final integrated map consisting of 3411 SLAF markers on 19 linkage groups (LGs) with an average distance of 0.98 cM between adjacent markers was generated using the specific length amplified fragment sequencing (SLAF-seq) technique. A total of 9 significant stable QTLs for Muscat flavor, berry firmness and berry shape were identified on two linkage groups among the hybrids analyzed over three consecutive years from 2016 to 2018. Notably, new stable QTLs for berry firmness and berry shape were found on LG 8 respectively for the first time. Based on biological function and expression profiles of candidate genes in the major QTL regions, 3 genes (VIT_08s0007g00440, VIT_08s0040g02740 and VIT_08s0040g02350) related to berry firmness and 3 genes (VIT_08s0032g01110, VIT_08s0032g01150 and VIT_08s0105g00200) linked to berry shape were highlighted. Overexpression of VIT_08s0032g01110 in transgenic Arabidopsis plants caused the change of pod shape. Conclusions A new high-density genetic map with total 3411 markers was constructed with SLAF-seq technique, and thus enabled the detection of narrow interval QTLs for relevant traits in grapevine. VIT_08s0007g00440, VIT_08s0040g02740 and VIT_08s0040g02350 were found to be related to berry firmness, while VIT_08s0032g01110, VIT_08s0032g01150 and VIT_08s0105g00200 were linked to berry shape.

technique. Total 16 significant QTLs were identified on six linkage groups based on phenotypic data of Muscat flavor, berry firmness and berry shape among the hybrids analyzed over three consecutive years from 2016 to 2018. Notably, new QTLs for berry firmness and berry shape were found on LG 8 respectively for the first time, which had the strongest and most stable effects over 2-3 years. Based on biological function and expression profiles of candidate genes in the major QTL regions, 3 genes (VIT_08s0007g00440, VIT_08s0040g02740 and VIT_08s0040g02350) related to berry firmness and 3 genes (VIT_08s0032g01110, VIT_08s0032g01150 and VIT_08s0105g00200) linked to berry shape were highlighted. Overexpression of VIT_08s0032g01110 in transgenic Arabidopsis plants caused the change of pod shape.
Conclusions: A new high-density genetic map with total 6634 markers was constructed with SLAF-seq technique, and thus enabled the detection of narrow interval QTLs for relevant traits in grapevine.
VIT_08s0007g00440, VIT_08s0040g02740 and VIT_08s0040g02350 were found to be related to berry firmness, and VIT_08s0032g01110, VIT_08s0032g01150 and VIT_08s0105g00200 were linked to berry shape.

Background
Marker assisted selection (MAS) technology has been widely used to improve traditional breeding accuracy and efficiency in perennial crops [1]. One of the main objectives of grape breeding now is to develop molecular markers related to traits of interest for genetic selection of target phenotypes [2].
However, it needs to investigate the genetic determinisms for each given trait firstly. Quantitative trait loci (QTLs) mapping is one of the key and efficient approaches for dissecting complex traits in grapevine.
Promoting berry quality traits has been the endless pursuit of grapevine breeders. Muscat flavor, berry firmness and berry shape are three main important quality traits in the breeding of new table grape varieties. The genetic determinants underlying their regulation have attracted extensive attention.
For Muscat flavor, a major QTL on linkage group (LG) 5 has been identified based on the phenotypic variation of the Muscat score analysis and evaluating monoterpenoid contents in three different F1 segregating progenies [16][17][18]. Because its colocalization with the major QTL positioned on chromosome 5, 1-deoxy-d-xylulose-5-phosphate synthase (VvDXS) has been suggested as the candidate gene responsible for Muscat flavor [19]. Guo et al. [20] have also investigated that berry flavor was associated with chromosome 5, while the significant single nucleotide polymorphisms (SNP) associated with berry flavor was identified on VIT_205s0020g03860 (homocysteine Smethyltransferase 2). Besides, some QTLs with smaller effects have been found on LG1, 7, 10 and 12 [16][17][18]21]. Altogether, it can be seen that genetic control of Muscat flavor is complicated; more genetic studies on other hybrid populations grown in different environmental conditions would be needed.
Like most other grapevine agronomic traits, grape berry firmness also follows complex quantitative inheritance. That QTLs for grape berry firmness distributed on the different LGs has been investigated in different mapping populations. Carreño et al. [22] have firstly identified QTLs for berry firmness on LGs 1, 4, 5, 9, 10, 13, and 18 in 'Muscat Hamburg' × 'Sugraone' and 'Ruby Seedless' × 'Moscatuel'. In a progeny of 'Ruby Seedless' × 'Sultanina', the determinants for this trait are located on LG 8 and 18 [23]. While Ban et al. [24] have found two QTLs for firmness located on LGs 3 and 10 in a 'V.
labruscana'× 'V. vinifera' cross. The most recent study has reported three QTLs all located on LG 18 in the progeny of 'Muscat Hamburg' and 'Crimson Seedless' [25]. However, most of the analyses have been performed using genetic linkage maps constructed with SSR markers, which resulted in the relatively large QTL confidence interval, and thus hinders the subsequent candidate genes identification.
For berry shape, to our knowledge, few studies have dealt with it in table grapes, although the diversity in berry shape is great for different grape cultivars. The wine grapes are generally round or nearly round. While, today's cultivated table grapes have diverse shapes, which can be divided into round, nearly round, broad ellipsoid, narrow ellipsoid, ovoid, obovoid, heart-shape, cylindric and so on. But the genetic determinism underlying this diversity is still unknown.
Genetic map construction is essential for the detection of QTLs associated with traits of agronomic interest. So far, a series of parental and consensus genetic maps have been developed by applying amplified fragment length polymorphism (AFLP) [26], sequence related amplified polymorphism (SRAP) [27], and single sequence repeat (SSR) [28][29] markers to different bi-parental segregating populations in grapevine. But majority of the genetic maps consist a limited number of markers with large map spacing and low resolution, thus are not able to provide precise and complete information about the numbers and locations of QTLs controlling the traits [30]. With the rapid development of the next generation sequencing (NGS) technology, genotyping through sequencing becomes the most direct and powerful method for the large-scale detection of SNP,which are considered as the markers of best choice for high-density genetic map construction [31][32]. At present, several high density genetic maps have been constructed using NGS technologies, and some new QTLs have been successfully identified in grapevine [7,11,[33][34]. There is no doubt that these high-density maps improve the efficiency and accuracy of QTL mapping, which can be used more effectively in breeding programs. However, QTL analysis for targeted traits mainly depends on the polymorphism between parents of the mapping populations. Lots of markers have distinct genotypes or linkage relationships in different hybrid populations [35]. So, it is necessary to develop new high density genetic linkage maps of grapevine for further QTL analysis of interest traits in grapevine.
Among all the NGS strategies, whole genome re-sequencing can identify whole genome wide differences between individuals and a large number of SNP markers [36]. But it is still too costly to apply in multiple samples, and generally unnecessary for linkage mapping. Specific length amplified fragment sequencing (SLAF-seq), a reduced representation sequencing approach, has been developed, which exhibits advantages in large-scale de novo SNP discovery and genotyping [37]. In the last 5 years, a series of high density genetic maps have been constructed based on SLAF-seq in diverse plant species [38][39]. Guo et al. [40] and Wang et al. [41] have successfully applied this method to construct high density genetic maps in grapevine too.
On these bases, in the present work, SLAF-seq was used in whole-genome genotyping for grapevine F1 lines; a consensus high-density genetic map was constructed with the developed SLAF markers.
The map will facilitate the further precise identification of QTLs for major agronomic traits, and marker assisted selection in grape breeding programs. Moreover, the berry Muscat flavor, berry firmness and berry shape of F1 progenies and two parents were analyzed over 2-3 successive years.
Quantitative trait loci (QTLs) for these three traits were identified and analyzed based on the consensus map. The candidate genes were predicted and validated by real time PCR. The results will broaden our understanding of the genetic control of these fruit traits, the tightly linked markers may be used to improve the fruit quality of grape.

Plant materials
The mapping population used in this study (line 1002; n = 160) was generated by crossing 'Moldova' (V. vinifera × V. labruscana) and 'Ruidu Xiangyu' (V. vinifera) [42] in 2010. The female parent 'Ruidu Xiangyu' was selected from a cross of 'Jingxiu' × 'Xiangfei' (Both are local varieties in china). 'Ruidu Xiangyu' is a grape with ovoid shape, greenish-yellow color, firm and crisp flesh and an excellent Muscat flavor. While the male parent 'Moldova' with ellipsoid shape, dark blue skin, soft flesh and neutral flavor, was derived from a cross of 'Guzal Kara' and 'Villard blanc'. Moreover, 'Moldova' has high resistance to downy mildew and grey mold [43]. Altogether, there are great genetic differences between two parents, which supply excellent materials for high-density genetic map construction and subsequent QTL mapping. The plants of the two parents and their progeny were grown on their own roots in the experimental vineyard at Beijing academy of Forestry and Pomology (39°58' N and 116°13' E). The plants were spaced 1.0 m apart within the row and 2.5 m apart between rows, and rows were north-south oriented. They were maintained under routine cultivation conditions, including soil management, fertilization, irrigation, pruning and disease control. The average score was used in further analysis.

Phenotypic Measurements
Berry firmness was assessed over 2 years (2017-2018) according to the method of Carreño et al. [22] with universal TA. XTplus testing machine (Stable Micro Systems, Godalming, Surrey, UK). By using the device, firmness values were expressed as force (N) required for a 20% deformation of the berries. The average value of 20 berries per offspring was used in subsequent analysis.
The berry shape index (ShI) was determined as the ratio of the mean berry length (MBL) to the mean berry diameter (MBD). Berry length and diameter were assessed according to the OIV descriptors (OIV, 2009) with little modification. MBL was the mean value of 30 berries taken from the middle part of five representative clusters. MBD was the mean value of 30 berries taken from the middle part of five representative clusters. All measurements were taken with a hand caliper.

Statistical Analysis
The phenotypic data were analyzed using SPSS 13.0 software (SPSS, United States) to generate descriptive statistics, including the mean, minimum, maximum, standard deviation (SD), coefficient of variation (CV), skewness and kurtosis. The frequency distribution of phenotypic data was checked using Sigmaplot 10.0 software. The broad sense inheritability (H b 2 ) was estimated following the methods of Liu et al. [44].

Dna Isolation
For DNA isolation, young and healthy leaves were harvested from each individual F1 plant and the two parents at the beginning of the vegetative period. Sequencing Data Grouping And Genotyping 8 The procedures of sequencing data grouping and genotyping were performed as described previously [37]. Briefly, the Burrows-Wheeler Aligner (BWA) software was used to align clean reads from each sample against the Vitis vinifera reference genome (ftp://ftp.ensemblgenomes.org/pub/plants/release-25/fasta/vitis_vinifera/ ) with default parameters. Sequences mapped to the same position were defined as a single SLAF locus. Locus with two to four SLAF tags was identified as polymorphic SLAF.
The average sequence depths of SLAF markers were greater than 20-fold in parents and greater than 8-fold in progeny. Polymorphic markers were classified into eight segregation patterns (ab × cd, ef × eg, hk × hk, lm × ll, nn × np, aa × bb, ab × cc and cc × ab). Based on the population type of F1, seven segregation patterns (excluding aa × bb) were selected for genetic map construction. In order to ensure the quality of the genetic map, three stringent filtering criteria were considered for the SLAF markers: i) with the average sequence depths of > 40-fold in the parents; ii) The number of SNP is < 8 per SLAF marker; iii) with less than 5% missing data. In addition, the chi-square test was then performed to examine the segregation distortion, and markers with significant segregation distortion (P < 0.01) were initially excluded from map construction.

Genetic Linkage Map Construction
Based on the locations on the grape genome, the filtered SLAF markers were partitioned into 19 linkage groups. The modified logarithm of odds (MLOD) scores between markers was calculated, and Markers with MLOD scores < 5 were filtered out before ordering. Highmap software was used to construct the genetic map of each linkage group as described by Liu et al. [45]. The error correction strategy of SMOOTH [46] was applied to correct genotyping errors, and a k-nearest neighbor algorithm [47] was used to impute genotyping missing. The enhanced algorithm of Gibbs sampling, spatial sampling and simulated annealing (GSS) [48][49] were employed to order markers. Map distances in centi-Morgans (cM) were calculated using the Kosambi mapping function [50]. For the construction of the consensus map, markers mapped in both parental maps and heterozygous markers (ab × ab) were used. Finally, the haplotype map, the heat map, and collinearity between the genetic and physical positions were analyzed by the method of Liu et al [45] to evaluate the quality of the constructed linkage map.

Qtl Mapping
Quantitative trait loci (QTLs) analysis for the berry Muscat flavor, berry firmness and berry shape were carried out on the consensus map using interval mapping with MapQTL 6.0 [51]. The threshold of LOD scores for evaluating the statistical significance of QTL effects was determined using 1,000 permutations. Based on these permutations, a LOD score of 3.0 was used as a minimum to declare the presence of a QTL in a particular genomic region. The genes within QTLs were identified by mapping the associated markers on the physical map. The genes were annotated and analyzed via

Transformation of in Arabidopsis
The full-length cDNA of VIT_08s0032g01110 were synthesized in the Beijing Sunbiotech Company (Beijing, China). The CDS of VIT_08s0032g01110 was then fused to the downstream of CaMV35S promoter at the Bgl II (5' end)/ BstE II (3' end) sites by substitution of the GUS gene in pCAMBIA1301 vector. The pCAMBIA1301:VIT_08s0032g01110 construct was introduced into Agrobacterium tumefaciens strain GV3101, which was used to infect Arabidopsis (ecotype Columbia-0) using the floral dip method. Positive transgenic Arabidopsis lines were screened by Hygromycin (Roche, Germany) and PCR to select T1 plants.

Phenotypic data analysis
The phenotypic variation ranges of Muscat flavor (MF), berry firmness (BF) and berry shape index (ShI) for the two parents and the F1 progenies were presented in Supplementary Table 2 and Table 1.   In terms of berry shape, the berries of 'Ruidu Xiangyu' were nearly rounded; 'Moldova' berries showed elliptical shape. Higher ShI value was observed in 'Moldova'. The values of ShI in F1 population showed continuous variation, and transgressive distribution was observed (Fig. 1C). The F1 population mean value of ShI was 1.19 (2016), 1. 16 (2017), and 1. 15 (2018). The means of ShI in F1 population were over or equal to the mid-parent value in three successive years. Approximately similar normal phenotypic data distributions of ShI were examined for all three years (Fig. 1C).
As phenotypic data shown, all three traits showed quantitative inheritance, suggesting that they were controlled by multiple genes. However, high Broad-sense heritability (H b 2 ) (more than 50%) was investigated for each trait ( were numbered according to the chromosome numbers (Fig. 2). As shown in Supplementary Table 4 Table 5).
The consensus grape map included 6436 markers with a total genetic distance of 3365.41 cM (  Fig. 2 and Supplementary Fig. 2). The average interval distance between markers was   Fig. 3). These analysis results reflected the validity of molecular markers genotyping to a certain extent.
It is believed that haplotype and heat maps can directly reflect the quality of the genetic maps.

Haplotype maps show recombination events in individuals, and heat maps reflect the recombination frequency and mapping location between markers. A haplotype map for
LGs of the consensus map was shown in Supplementary Fig. 4. As the results shown, the occurrence of double crossovers and deletion ratio were low, indicating genotyping and marker-order of the LGs were accurate and reliable.
Heat maps were generated by using pair-wise recombination values for the 6436 mapped SLAF markers. The heat maps for the paternal map were also shown in Supplementary Fig. 5. The linkage between markers decreases with the increase of genetic distance, which indicates that the order of markers in the LGs is correct.
Furthermore, the colinearity between the genetic and physical positions on a linkage map was also analyzed. A relatively high level of genetic collinearity was observed between 19 LGs and the reference genome ( Supplementary Fig. 6). As shown in Supplementary Table 6, the Spearman correlation coefficient ranged from 0.83 to 0.98, and it was higher than 0.90 in most LGs.
In general, from the results of haplotype maps, heat maps and colinearity analysis, the genetic maps constructed were of good performance for further QTL analysis.
Qtl Identification QTL analyses were performed using the consensus genetic map. The QTLs detected for all the three traits are summarized in Table 3. A total of 16 QTLs were mapped on the consensus genetic map using the interval mapping method. Of the 16 QTLs, eight contributed to berry Muscat flavor, three were associated with fruit firmness, and the remaining five were related to berry shape. Three significant QTLs kinked to berry firmness were located on LG1 (qBF-1) and LG8 (qBF-2 and qBF-3) respectively (Table 3). In 2017, two QTLs of berry firmness were identified. qBF-1 was mapped on LG1 explaining 15.50% of PVE with a peak LOD score of 3.8. qBF-2 was mapped on LG8 with a LOD score of 4.14 and an 19.

Candidate Genes Involved In Berry Quality Traits
Because stable genetic intervals were detected for each trait across years, the candidate genes located within these confidence intervals were henceforward being focused on. The linked markers in the confidence intervals were mapped on to the grapevine reference genome sequence. Four good candidate genes related to the berry shape index.

Analysis of expressions of candidate genes during grape berry development
To further evaluate the potential relationship between candidate genes and each specific trait, the relative expression of corresponding candidate genes and berry related traits were analyzed during different grape berry development stages of two parent cultivars, 'Moldova' and 'Ruidu Xiangyu'. As shown in Figure. 4A, berry firmness of 'Moldova' and 'Ruidu Xiangyu' both decreased at veraison.
Thereafter, berry firmness of 'Moldova' still declined at the ripening stage, while that of 'Ruidu Xiangyu' increased significantly at the maturity stage (Fig. 4A). Among all the candidate genes ( Supplementary Fig. 7A), the expression pattern of VIT_08s0040g02350 was consistent with that of parents berry firmness, suggesting that VIT_08s0040g02350 might associate with the variation of berry firmness in grape. There was an obvious increase in the expression of VIT_08s0007g00440 in 'Moldova' at the ripening stage, but a significant decrease in 'Ruidu Xiangyu', which showed a contradictory pattern with the phenotypic variation.
The ShI of 'Moldova' and 'Ruidu Xiangyu' showed different change trends during berry development.
Continuous reduction of ShI was observed in developing 'Ruidu Xiangyu', while ShI of 'Moldova' was firstly decreased but increased at the ripening stage. Among the analyzed genes, the relative expression of VIT_08s0032g01110, VIT_08s0032g01150 and VIT_08s0105g00200 showed a similar or opposite change patterns with that of ShI during berry development ( Figure. 4B and Supplementary   Fig. 7B). In particular, the expression level of VIT_08s0032g01150 was increased gradually during grape berry development stage in 'Ruidu Xiangyu', but in 'Moldova' the relative expression of VIT_08s0032g01150 were reduced at veraison and increased at ripening stage, which presented a completely opposite trends with the changes of ShI (Fig. 4B).
Unfortunately, among all the candidate genes studied, no genes were found consistent with the changes of Muscat flavor in both cultivars during berry development ( Supplementary Fig. 7C).

Overexpression of VIT_08s0032g01110 in Arabidopsis
Transgenic Arabidopsis plants overexpressing VIT_08s0032g01110 were generated to elucidate its functions. As the results shown, differential pod shapes were observed between WT and 35S:VIT_08s0032g01110 seedlings (Fig. 5). The pods of 35S:VIT_08s0032g01110 plants showed curved, and their lengths were shorter than WT plants.

Genetic map
The construction of a genetic map is very essential for mining the genetic basis of relevant traits, in particular of high-density genetic map construction, which will improve the efficiency and accuracy of further QTL analysis [30]. The important step in the construction of high-density maps is highthroughput discovery and genotyping of numerous molecular markers. The advent of NGS-based methods provides good opportunities for SNP markers development. Several high density genetic maps for grapevine have been constructed with NGS techniques [7,11,[33][34][40][41]. In the present work, a high-density genetic map for 'Moldova' × 'Ruidu Xiangyu' was constructed with SLAF-seq technique. The final integrated genetic linkage map consisted of 6433 SLAF markers on 19 LGs spanning a total genetic distance of 3365.41 cM, with an average distance of 0.52 cM between adjacent markers (Table 2 and Supplementary Fig. 2). Comparing to the previous reported maps constructed by the same SLAF-seq method [40][41], the number of markers in this study is lower, but it is still higher than other published high-density maps [7,11,33]. This difference may be related to the different F1 population used and different markers filtering parameters in genetic map construction [7]. In addition, the markers were distributed relative evenly on the whole genome (Fig. 2). The average distance between adjacent markers in all 19 LGs was less than 0.9 cM. The percentage of "Gap < 5 cM" reached up to 98% (Table 2). But it needs to point out that total length of the maps in this study were over 3000 cM, which is similar to that of maps constructed by Zhu et al [52], but longer than most other published maps [7,11,25,[33][34][40][41]. Collard et al. [53] have suggested that the difference in chromosome recombination events occurred during sexual reproduction in each subpopulation can be the reason for the variation in map length. It also has been suggested that the very large map length may probably result from the low quality of markers or difficulty in ordering abundant markers in small populations with low recombination [54]. We were not sure about the reasons for the large length of the maps constructed in this work. However, stringent filtering criteria were considered for the SLAF markers in this study to ensure the marker quality. In the final, only 6436 markers from initial 96416 polymorphic SLAF markers were grouped on the map.
The average sequencing depths of these 6436 markers were up to 68.87-fold for 'Ruidu Xiangyu', 56.12-fold for 'Moldova', and 16.71-fold for each individual progeny ( Supplementary Fig. 3). Moreover, the low occurrence of double crossovers and deletion ratio were observed form haplotype maps ( Supplementary Fig. 4), indicating genotyping and marker-order was reliable in the LG. The heat map of the marker exchange relationship was generated for evaluating the linkage relationship among the markers, suggesting that the recombination frequency and mapping location between markers was basically consistent in each LG ( Supplementary Fig. 5). And a high level of genetic collinearity was observed between 19 LGs and the reference genome ( Supplementary Fig. 6). Anyhow, the construction of this high density map for 'Moldova' × 'Ruidu Xiangyu' provides a key foundation for genetic analysis of many agronomic traits in grapevine.

Qtl Detection
The genetic factors for Muscat flavor have long been a concern of grape breeders. The researchers have identified a major QTL located on linkage group (LG) 5, and 1-deoxy-d-xylulose-5-phosphate synthase (VvDXS) associated with the QTL has been considered as the candidate gene responsible for Muscat flavor [16][17][18][19]. In this work, total eight QTLs related to berry Muscat flavor were mapped on LG5, LG17, LG1, LG7 and LG18 respectively cross 3 successive years based on sensory tasting data (  Table 7), which is corresponding well to the previous finding [17]. While, VIT_05s0020g03860 (a predicted homocysteine S-methyltransferase 3) was detected in the genomic region of 30.802-35.572 cM (Supplementary Table 7). Guo et al. [20] have investigated that one SNP on VIT_205s0020g03860 was significantly associated with berry flavor. In addition, some other unstable QTLs (qMF-3 on LG17, qMF-6 on LG1, qMF-7 on LG7 and qMF-8 on LG18) were identified in 2016 and 2018 (Table 3). Previous studies have reported QTLs on LG1 and LG7 too, but also pointed out the limitation in minor QTLs detection using only Muscat score data [16]. Further QTL studies based on more detailed monoterpenes (the major contributors for Muscat flavor) components and contents will provide more information about genetic determinism for Muscat flavor.
The berry firmness of F1 progenies and parents was evaluated by the compression test, which has been suggested as a reliable method for providing information about the firmness of whole unpeeled berries [55]. Good correlations have been investigated between the quantitative data obtained from the texture analyzer and sensory parameters made by tasting [22]. In our experimental population, the phenotype of firmness showed continuous variation. The fluctuation was observed in firmness across years, suggesting the effect of environmental factors on this trait. But relatively high broadsense heritability (H b 2 ) of berry firmness was estimated reaching a value of 66.98% (Table 1), which were similar to the previous result (87.75% of H b 2 ) obtained from 'Ruby Seedless' × 'Sultanina' progeny [23]. Based on the phenotypic data collected across two years, major QTLs qBF-2 (21.90% of PVE) and qBF-3 (20.10% of PVE) linked to berry firmness were detected on LG 8 (   7). It has been suggested that expansins are involved in reassembly, degradation and expansion of cells and have a function in affecting berry softening in grape [56][57][58]. Pectate lyases have also been investigated to be related to grape berry texture [56,59]. In addition, a minor effect QTL for berry firmness was mapped on LG 1 (Table 3), which is consistent with the previous report.
Carreño et al. [22] have identified a QTL on LG 1 in two segregating progenies. However, we did not identify QTLs on LG18, which have been reported in other populations [22,25]. This may be due to different genetic backgrounds and distinct phenotype evaluation methods [25].
Fruit shape is one of the vital appearance traits for horticulture crops. To date, various QTLs or candidate genes for fruit shape in different horticultural crops with fruits as their edible organs have been genetically confirmed [60][61][62][63][64]. However, few genetic studies have been focused on the identification of QTLs responsible for grape berry shape, although broad ranges of phenotypic variation in berry shape were observed in grape, especially in cultivated  Supplementary Fig. 7C), which has been suggested as the candidate gene responsible for Muscat flavor [19]. But this cannot exclude its role in Muscat flavor formation.
Battilana et al. [68] have discovered the lysine with an asparagine at position 284 of the VvDXS protein would affect Muscat flavor by influencing the enzyme catalytic efficiency. As to berry firmness, besides genes of an expansin-A4 (VIT_08s0007g00440) and a probable pectate lyase 4 (VIT_08s0040g02740), which families have been suggested playing important roles in fruit softening previously [56][57][58][59], the expression of an additional gene (VIT_08s0040g02350) was also consistent with the change of berry firmness, suggesting its role in affecting fruit firmness (Fig. 4A). Three candidate genes were highlighted for berry shape including VIT_08s0032g01110, VIT_08s0032g01150 and VIT_08s0105g00200, the expression of which showed the similar or opposite change patterns with that of ShI during berry development (Fig. 4B). In particular, VIT_08s0032g01110 was predicted as an axial regulator YABBY 5 gene. Its homologous gene SlYABBY2 has been confirmed as one of major genetic factors regulating fruit shape in tomato [60], which VIT_08s0032g01110 might play important regulatory role in grape berry shape formation. The transgenic Arabidopsis overexpressing VIT_08s0032g01150 showed shorted and curved pod shape comparing to WT plants (Fig. 5).

Conclusions
In summary, a new high-density genetic map with total 6634 markers and an average distance of 0.52 cM between adjacent markers for 'Moldova' × 'Ruidu Xiangyu' was constructed with SLAF-seq technique, which provides a foundation for further genetic studies of relevant traits in grapevine. By using this map, 16 reliable QTLs linked to three grape berry quality traits were detected over 2-3 years, including 9 stable major QTLs. These QTL regions were significantly narrowed down compared to previous reports, which facilitated the subsequent candidate genes identification. The subsequent expression data of the candidate genes underlying the QTLs highlighted 3 genes related to berry firmness and 3 genes linked to berry shape respectively. Overexpression of VIT_08s0032g01110 in

Ethics approval and consent to participate
This research was not applicable for ethics approval and consent to participate.

Consent for publication
This research is not applicable to consent for publication.

Competing interests
The authors declare no competing financial interests

Avalilability of data and materials
The Vitis vinifera reference genome referred in this work were downloaded from ftp://ftp.ensemblgenomes.org/pub/plants/release-25/fasta/vitis_vinifera/. The Phenotype data for Muscat flavor, berry firmness and berry shape in this work were presented in Supplementary Table 2.
The information of molecular markers used for map construction in this work was listed in Supplementary    The Changes of berry firmness, berry shape index and expressions of candidate genes during grape berry development in 'Moldova' and 'Ruidu Xiangyu'. (A) Expressions of three filtered candidate genes for berry firmness. (B) Expressions of three candidate genes for berry shape.