Grape accessions, including 50 wild Chinese grapevine species and 45 cultivars from the European species V. vinifera or the American species V. labrusca (Table 1), were grown under natural field conditions at the National Grape Germplasm Resources Repository of Zhengzhou Fruit Research Institute, Chinese Academy of Agricultural Sciences. Warm temperate continental climate of Zhengzhou has clear four seasons. The average annual precipitation is about 630 mm and mean temperature is 14.4°C. The details of climatic data were shown in Supplementary Table S1. The experiment vines were planted 9 or 10 years ago in sandy fluvo-aquic soil. And no special cultural practices were taken. All of the vines were in good condition. Grape berries were collected from June to September and leaves were picked at the end of June in 2013 and 2014. Samples were harvested from three grape vines for each accession. For the berries, three grape clusters on each plant were picked, one from the top, middle, and bottom of the canopy, respectively. To ensure that all berries were harvested at their full ripeness, we checked the seeds in the berries every 2 days from June till September. When the seeds completely ripened, the size of berries was no longer increasing, and the red grapes were fully colored, the berries were sampled from that accession. For the leaves, the second or third leaves (depending on healthiness) from the bottom of three different branches with more than 10 leaves were picked in the end of June. Unhealthy berries (cracking, smaller and other underdeveloped fruits) were removed before the samples were quickly frozen in liquid nitrogen and held at -80°C until use.

Trans-resveratrol levels in berry skins and leaves were measured using HPLC as described by Li et al. (2006) with some modifications, in 95 grapevine accessions in 2013 and 2014. The standard for trans-resveratrol was purchased from Sigma–Aldrich (USA). Fruits were peeled and juice was soaked up using filter paper.

Three gram samples were ground to powder using a porcelain mortar and pestle in liquid nitrogen, extracted by 15 mL ethyl acetate in the dark at 25°C for 48 h, and centrifuged at 10,000 r⋅min^-1 for 10 min. The supernatants were transferred into a tube containing 5 mL ethyl acetate, followed by centrifugation at 10,000 r⋅min^-1 for 10 min. All supernatants were evaporated to dryness by Nitrogen blowing instrument (DCY-12S, Qingdao Haike, China) at 40°C. Dried samples were then dissolved in 2 mL of methanol and stored at -80°C. The samples were filtered through a 0.22 μm PTFE membrane filter before resveratrol analysis. Extractable amounts of resveratrol were analyzed using a Waters e2695 HPLC system (USA). Elution was carried out with a mobile phase delivered using a Waters C18 HPLC pump at a flow rate 0.8 mL⋅min^-1. A Waters 2996 UV detector was used at 306 nm. Mean values and standard deviations were obtained from three biological replicates. An HPLC chromatogram of resveratrol was made with a standard solution. The resveratrol content was analyzed by Excel 2003 (Microsoft, USA) and SPSS 17.0 software (IBM, USA).

Genomic DNA was extracted using Ezup Column Plant Genomic DNA Purification Kit following the manufacturer’s protocol (Sangon Biotech, Shanghai, China). The concentration of the extracted DNA was assessed using a Thermo ND 2000 spectrophotometer (ThermoFisher, USA). Genomic DNA was adjusted to a final concentration 50 ng/μL and was used for PCR amplification.PCR reactions were carried out in a final volume of 20 μL. Amplification reactions were carried out on a ABI Veriti thermal cycler (USA) using the following cycling profile: 95°C for 5 min, followed by 35 cycles at 95°C for 45 s, 48–56°C for 45 s, and 72°C for 1 min, and a final extension step at 72°C for 10 min. The amplification products were separated through polyacrylamide gel electrophoresis.

Based on predicted STS gene sequences in the 12x grapevine PN40024 genome¹ and the gene positions of these 44 STS genes (Shi et al., 2014), a total of 79 pairs of SSR primers on chromosomes 10 and 16 were designed using GRAMENE ssrtool². Parameter settings were as follows: tetramer for the maximum motif-length group, and 4 for the minimum number of repeats.

Allelic variation was analyzed by calculating the number of alleles (Na), effective number of alleles (Ne), observed heterozygosity (Ho), and expected heterozygosity (He) using Popgene software. Polymorphism information content (PIC) was calculated using PIC-CALC.

Genetic distance matrices were obtained using SSR data in DPS software³. A phylogenetic tree was constructed by the unweighted pair-group method with arithmetic averages (UPGMA) with MEGA 6.0 software⁴.

Using 40 STS-gene-associated SSR markers, the genetic population structure of the 95 accessions was determined by Structure 2.1⁵. A burn-in phase of 10,000 iterations was followed by 100,000 Monte Carlo Markov Chain iterations. The optimal population number k (from 1 to 10 assumed in this study) was estimated (Evanno et al., 2005). Ten replicates were performed for each cluster, k. When an inflection emerged in the LnP (D) curve, the corresponding k value was adopted as the optimal group number. The values of the estimated membership probability (Q) were calculated to serve as covariates in the association analysis with general linear model (GLM) in Tassel 2.1⁶. Phenotypic effect values of some marker alleles were evaluated according to null allele as suggested by Breseghello and Mark (2006).

The trans-resveratrol levels in skin and in leaf collected from all accessions were determined by HPLC (Figure 1). The trans-resveratrol content in berry skins ranged from 0.05 to 67.82 μg⋅g^-1 FW in 2013 and from 0.03 to 68.44 μg⋅g^-1 FW in 2014. For both seasons, the highest levels were from the wild Chinese grapevine V. adenoclada accession Shuangxi 03. In leaves, the trans-resveratrol content ranged from 0.04 to 10.27 μg⋅g^-1 FW in 2013 and from 0.09 to 11.69 μg⋅g^-1 FW in 2014. The highest levels for both years were in leaves from wild Chinese grapevine V. amurensis accession Gaoshan No.2. Resveratrol contents were higher in berry skins than in leaves for each genotype.

Between the 2 years, the variation of resveratrol content was more stable in wild grapevine accessions than that of cultivated ones. More of the cultivated accessions (51%) showed year-to-year variations of resveratrol content in skin greater than 50%, compared to only 8% of wild grapevine ones, showing such large variations. Similarly, in leaves, 22% of wild accessions and 67% of cultivated ones showed resveratrol content variations greater than 50% (Supplementary Table S2). The results suggested that wild ones retained stable resveratrol biosynthetic capacity.

Based on the predicted STS gene sequences of the 12x grapevine PN40024 genome, 79 SSR primers were designed. These 79 markers were analyzed in the 95 grapevine accessions. Forty SSR markers showed higher polymorphism, and 123 alleles were identified. The PICs of the SSR loci ranged from 0.0206 to 0.6712, with an average of 0.2877 (Supplementary Table S3).

When the STRUCTURE software was run using all 95 grapevine accessions, the delta k showed a significant peak when k = 2; thus the grapevine accessions were divided into two populations, termed P1 and P2 (Figure 2). This division of the population was supported by statistical probability and could ensure the accuracy of association analysis with minimum false association. P1 included 45 grapevine cultivars, both table and wine grapes, whereas P2 included 50 accessions, all of which were wild Chinese grapevine accessions (Figure 2). A phylogenetic tree was constructed by UPGMA analysis based on genetic distances calculated from the SSR data of the 95 accessions (Figure 3). Due to sufficient variability, all selected accessions were discriminated. The accessions clustered into two main groups, with six accessions (Nos. 50, 54, 89, 90, 93, and 95) forming a third, distinct cluster (black square). All accessions formed a branch with other accessions and cultivars, except two, namely V. davidii accession Dongxiangjiao (No. 55, black circle), which did fall in close to another branch, and V. yeshanensis accession Yanshan (No. 94), which did not sort into near wild grapevines. This corresponded to the evaluated populations with STRUCTURE software, with a few exceptions. The above SSR analysis generally agreed with the geographic origins and pedigree of the grapevine accessions.

Linkage disequilibrium (LD) among genes was the basis of the association analysis. Distribution of LD among the 40 SSR loci in the two groups (according to Figure 2) was shown as Figure 4A. Loci with high LD values (D′ > 0.7; upper right corner) were Sh13, Sh16, Sh22, Sh31, Sh37, Sh68, and Sh78.The LD among the wild Chinese grapevines (Figure 4B) was significantly higher than those of the grapevine cultivars (Figure 4C, including table grapes and wine grapes). The mean frequency distribution of the D′ value (P < 0.001) was 0.5329 for all experimental samples (Table 2A), 0.6046 for the V. vinifera cultivars, and 0.7037 for the wild Chinese accessions (Table 2B). The higher D′ in the wild population indicates more variation. In addition, the number of LD loci among the grapevine cultivars was fewer than in the wild Chinese accessions (Table 2B).

Based on LD analysis and the current suitable population, association analysis was performed with candidate markers using Tassel 2.1 software. Eight SSR loci, namely Sh5, Sh9, Sh21, Sh28, Sh56, Sh63, Sh76, and Sh77, were significantly (P < 0.001) associated with resveratrol content and their explained phenotypic variation (EPV) were all higher than 10% (Table 3). Loci Sh5, Sh21, Sh28, Sh63, and Sh76 were associated with high resveratrol content in the leaves, whereas loci Sh5, Sh9, Sh56, and Sh77 were associated with high resveratrol in berry skins (Table 4). Moreover, these associations were independent of the year. Locus Sh5 was associated with high resveratrol content in both tissues in both seasons.

The phenotypic effects of the different alleles of the eight loci significantly associated with resveratrol content were evaluated (Table 4). Allele 236 at locus Sh5 produced negative effects four times. On the other hand, allele 232 produced positive effects three times, including in leaves in the 2 years. Loci Sh9 and Sh56, through seven alleles, exhibited only positive effects in berry skins, whereas one allele of locus Sh77 created negative effects in berry skins in the 2 years. The rest of the loci showed negative effects at least once.

The eight loci significantly associated with resveratrol content were mapped to the 12x grapevine PN40024 genome. This revealed that locus Sh56 (location 16506665–16506789 on Chromosome 16) was closely linked to VvSTS27 (16507444-16503155) and that locus Sh77 (16:366055-16:366171) was closely linked to VvSTS17 (16372414-16366426) (Table 5). The other six loci were not very closed to known STS genes. However, future investigation of predicted genes at these loci may reveal their functions in secondary metabolism.