The association mapping panel consisted of 192 tomato accessions including 10 of S. pimpinellifolium, 18 of S. lycopersicum var. cerasiforme or cherry tomato, and 164 of S. lycopersicum, which were obtained from various sources (Table S1). The 164 S. lycopersicum accessions included 23 vintage varieties, 20 Latin American cultivars, 54 fresh-market lines, 59 processing lines, and 8 lines with unknown type. The association panel was grown in a randomized complete block design with three blocks containing each accession in two independent experiments conducted in 2013 and 2014. Plots of each accession consisted of at least four plants.

To identify loci for SSC and FW using the structural population mapping approach, two F₂ populations were developed by crossing processing tomato varieties OH88119 and OH9242, respectively, to a cherry tomato line Black cherry. Both OH88119 and OH9242 have medium-sized fruit (average FW 48.3 g for OH88119 and 73.7 g for OH9242 in 2013 and 2014) with relatively low SSC (average 4.1% for OH88119 and 4.6% for OH9242 in 2013 and 2014), while Black cherry is a small-fruited tomato line (average FW 18.6 g in 2013 and 2014) with a relatively high SSC (average 6.3% in 2013 and 2014). A sub-population of 503 individuals from the F₂ population of OH88119 × Black cherry were grown in the fall season of 2012, and another sub-population consisting of 752 individuals from the same cross as well as 276 individuals from the F₂ population of OH 9242 × Black cherry were grown in the spring season of 2013.

Tomato seeds for all studies were sown in 288 cell flats filled with a mixture of peat and vermiculite (3:1) in a protected greenhouse. Seedlings were transplanted ~50 days after germination to field. All experiments were conducted at Shangzhuang Research Station of China Agricultural University in Beijing, China. Production practices, plant spacing, and row spacing were as recommended for commercial growers (Gao et al., 2010).

Phenotypic data for association analysis were collected on a plot basis. Five to ten ripe fruits were harvested from each plot and subjected to SSC, FW, fruit height, fruit width, and color measurements.

Total weight of fruits collected in each plot was obtained by weighing all fruits using a pair of balances. Mean was obtained by dividing the total FW by the number of fruits and used as FW for individual fruit in each plot. The maximum height and width of a fruit were measured using a vernier caliper (Hangzhou Tool and Measuring Tool Company, Hangzhou, China). The fruit shape index (FSI) was calculated as the ratio of maximum height to maximum width (Brewer et al., 2006). Numeric descriptions of the red, green, yellow, and blue components of tomato color were obtained using the software Tomato Analyzer 3.0 (Brewer et al., 2006) following the description in Darrigues et al. (2008). The software generated a set of L^*, a^*, b^*, hue, and chroma values representing absolute color for each fruit. SSC was measured using a WAY-2S digital ABBE refractometer (Shanghai Precision Scientific Instrument Company, Shanghai, China). Plot means for FSI, values of color parameters and SSC were calculated based on measurements of all fruits in each plot. Pearson correlation coefficients for each trait between 2 years and among traits were obtained using SAS v9.4 (SAS Institute, Cary NC, USA). Analysis of variance was conducted using PROC GLM in SAS with the model show to best fit the data: X_ijb = μ + G_i + Y_j + R_b/j + M_ij+ ε_ijb, where X_ijb is the trait value of the b^th replication of the i^th genotype in the j^th year, G_i is the random effect of the i^th genotype, Y_j represents the fixed effect of the j^th year, R_b/j is the fixed effect of the b^th replication in the j^th year, M_ij the random effect of the genotype by year interaction and ε_ijb is the residual. Broad sense heritability (H²) for each trait was calculated based on the plot level using the equation H² = σ_G²/(σ_G² +σ_GY² +σ_ε²) according to the description in Nyquist and Baker (1991), where σ_G² is genotypic variance, σ_GY² is the variance due to interaction between genotype and year, andσ_ε² is the error variance.

A total of 547 InDel markers (Table S2) evenly distributed across the tomato genome were used to genotype the association panel. These InDel markers were chosen from our previous study and were polymorphic within 10 accessions of S. lycopersicum (Yang et al., 2014). Genomic DNA was isolated from fresh-collected young leaves of each accession using the modified CTAB method (Kabelka et al., 2002). PCR and genotypic data collection were conducted according to the description in Yang et al. (2014).

Nei's genetic distance (Nei, 1972) was calculated for each pair of accessions and marker allele frequency was obtained using the software PowerMarker V3.25 (Liu and Muse, 2005). Polymorphism information content (PIC) was calculated using the formula (Weir, 1990) of PIC = 1-Σp_i², where p_i is the frequency of i^th allele for each marker locus. Markers with a minor allele frequency below 5% were removed from the marker data set to calculate population structure, kinship, and to perform association analysis.

Marker genotypes were used to measure the extent of linkage disequilibrium (LD) within the 164 accessions of S. lycopersicum and 18 accessions of S. lycopersicum var. cerasiforme. The TASSEL 2.1 (Bradbury et al., 2007) software was used to calculate pair-wise r²-values for markers polymorphic within the 164 accessions. P-values for each r² estimate were calculated using 1,000 permutations in TASSEL.

LD decay was calculated by plotting pair-wise r²-values onto genetic distance in base pairs on the same WGS chromosome (SL2.40) of tomato variety Heinz1706 (The Tomato Genome Consortium, 2012). All markers with <25% missing data and a minor allele frequency >5% were used to calculate LD decay. Critical values of r² as an evidence of linkage were derived from the parametric 95th percentile of the distribution of the unlinked markers (Breseghello and Sorrells, 2006).

Population structure (Q matrix) was estimated using Structure 2.3.4 software (Pritchard et al., 2000; Falush et al., 2003, 2007). Number of populations (K) was determined following the instruction in Pritchard et al. (2000) with a burn-in period of 100,000 iterations and Markov Chain Monte Carlo of 100,000. Twenty independent runs were done for K varying from 1 to 10. The most probable K-value was defined according to the method proposed by Evanno et al. (2005).

Unweighted Pair Group Method with Arithmetic Mean (UPGMA) cluster analysis was performed to develop a phylogenetic tree using the software PowerMarker V3.25 and the tree was viewed in MEGA5 (Tamura et al., 2011). Principal coordinate analysis (PCoA) was conducted using the Past 3.13 software (Hammer et al., 2001). The Loiselle kinship coefficients between tomato lines (K matrix) were calculated using the software SpAGeDi (Hardy and Vekemans, 2002).

The software program TASSEL 2.1 was used to conduct association analysis. A mixed linear model (MLM) taking into account both population structure (Q matrix) and the kinship matrix (K matrix), and a general linear model (GLM) using population structure (Q matrix) as a fixed factor were used for association identification of loci conferring fruit traits. Significance of marker-trait association was determined based on p-value at a level of 5% after Bonferroni (1936) multiple test correction. Since it has been popularly proved that MLM+Q+K model is a more effective approach than other models for detecting loci (Yu et al., 2006; Malosetti et al., 2007; Cericola et al., 2014; Pace et al., 2015; Sim et al., 2015), only the data of the MLM+Q+K model was presented in the current study. The phenotypic variation explained by each marker was the R²-value obtained from GLM model.

Since most markers significantly associated with loci for FW or SSC were not polymorphic between Black cherry and OH88119 or OH9242, we selected InDel markers that showing polymorphisms between the parents in Yang et al. (2014). Our goal was to identify two markers per chromosome arm in order to survey the whole genome. Thus, a total of 56 additional markers (Table S3) distributed on 12 chromosomes were used for initial identification of loci conferring FW and SSC in the F₂ population of OH 88119 × Black cherry. ANOVA using SAS v9.4 with a general linear model X_i = μ + M_i+ ε_i (Yang et al., 2005) was performed to determine an association between trait and marker genotype of 126 individuals randomly picked from the F₂ population of OH88119 × Black cherry. Once a marker was identified to be significantly (P < 0.05) associated with a trait, the marker was used to genotype the whole F₂ population of OH88119 × Black cherry and the F₂ population of OH9242 × Black cherry for validation.

The 547 markers generated 1295 alleles in the 192 tomato accession with a range of two to nine alleles (average 2.4 alleles) for individual markers (Table S2). Among the polymorphic markers, ~93% had two or three alleles with the dominance of bi-allele markers (Figure 1). As expected, all markers were polymorphic in the 192 accessions. However, polymorphisms within species was decreased from wild species S. pimpinellifolium to S. lycopersicum var. cerasiforme and then to S. lycopersicum, the cultivated tomato. The proportion of polymorphic markers were 97.4% in 10 accessions of S. pimpinellifolium, 85.2% in 18 accessions of S. lycopersicum var. cerasiforme, 64.7% in 23 vintage accessions, 78.4% in 20 Latin American Cultivars, 67.3% in 54 fresh-market lines, and 73.3% in 59 processing lines (Table S2). Subsequently, the average polymorphic information content (PIC) also decreased from wild species S. pimpinellifolium to S. lycopersicum var. cerasiforme and then to S. lycopersicum (Table S2).

Significant difference for each trait was observed among genotypes, and FSI was the only trait that showed no significant difference between 2 years or three replications (Table S4). However, the phenotypic data for each trait in 2 years were highly correlated with the correlation coefficient (r) of 0.85 for SSC, 0.91 for FW, 0.96 for FSI, 0.86 for L^*, 0.92 for Hue, and 0.83 for Chroma. Thus, the mean of the 2-year data for each trait was used for the following analysis.

A wide range of variation was observed for SSC, FW, FSI, L^*, Hue, and Chroma in the 192 tomato accessions (Table 1, Figure S1). The average of SSC was higher in S. pimpinellifolium than in S. lycopersicum, while the mean of FW was smaller in S. pimpinellifolium than in S. lycopersicum, and the means of SSC and FW in S. lycopersicum var. cerasiforme were between those in S. pimpinellifolium and S. lycopersicum. Significant negative correlation was observed between SSC and FW (r = −0.43, p < 0.0001). No differences of FSI were observed among three species. The means for fruit color parameters L^*, Hue, and Chroma were lower in S. pimpinellifolium than in S. lycopersicum var. cerasiforme and S. lycopersicum. Broad-sense heritability for each trait was obviously different. Chorma was the least heritable trait with heritability of 0.69, while FSI was the most heritable trait with heritability of 0.91 (Table 1).

Although, the 192 tomato accessions were from three species, model without prior population information was used to assign individual accession to a subpopulation using the software package of STRUCTURE2.3.4. In order to define the number of subpopulations within the 192 accessions, a series of independent runs of the data were run at a range of K-values from 1 to 10. The summary plot of membership coefficients (Q) and ΔK analysis (Figure 2) from STRUCTURE software, and the genetic relationships revealed by the phylogenetic analysis (Figure 3) and the PCoA (Figure 4), suggested that the mapping population was sorted into two subpopulations (K = 2). The larger subpopulation was composed of 134 accessions including all fresh-market, vintage, and unknown type accessions. One accession of S. pimpinellifolium LA2183, two-thirds of S. lycopersicum var. cerasiforme or cherry tomato, one-third of processing accessions, and four-fifths of Latin American Cultivars were also assigned into the larger subpopulation. Although, there was no certain trend of clustering in the larger subpopulation, the fresh-market accessions from Florida, USA formed one independent cluster. The smaller subpopulation consisted of 58 accessions including nine accessions of S. pimpinellifolium, two-thirds of processing, one-third of S. lycopersicum var. cerasiforme or cherry tomato, and four Latin American Cultivars. Within the smaller subpopulation, 38 processing accessions were clustered together, while the seven accessions of S. pimpinellifolium formed one cluster.

A mean r² of 0.393 was observed between all pairs of linked loci throughout the genome, with a mean maximum r²-value of 0.571 on chromosome 5 and a mean minimum r²-value of 0.086 on chromosome 10 (Table 2). The rate of LD decay was different across chromosomes (Figure 5). The baseline r²-values varied from 0.283 (chromosome 8) to 0.381 (chromosome 10) estimated by the 95th percentile method. Based on LOESS curves, the baseline r²-values corresponded to physical distances varying from 3.0 to 47.2 Mb. LD decayed within 6.5 Mb on chromosomes 1, 2, 3, 7, 8, 10, and 12, while chromosome 5 showed the lowest decay with a baseline r²-value of 0.335 reached ~47.2 Mb.

Using the MLM K + Q model and Bonferroni correction for p-values, a total of 102 genotype/phenotype associations with phenotypic variation explained >2.0% were detected under the threshold p-values of 1.30 × 10⁻⁴ for SSC, 1.19 × 10⁻⁴ for FW and FSI, 1.22 × 10⁻⁴ for L^* and Chroma, and 1.24 × 10⁻⁴ for Hue. The numbers of associations per trait ranged from three for Hue to 27 for Chroma. Regions carrying the presumed genes/QTLs were identified on 11 of the 12 chromosomes, but none were detected on chromosome 10 (Figure S2, Table 3). In order to match the associations with previously identified QTL, loci linked to one another within 3 Mb, the lowest LD, on the same chromosome were considered as a unit.

Of the 24 genotype/phenotype associations identified for SSC, the phenotypic variation explained by each marker varied from 2.6 to 25.5% (Table 3). Based on the physical distance between markers, at least 19 loci were detected on eight chromosomes and 13 of them had the phenotypic contribution >10%. The 17 markers associated with FW could be assigned to 15 chromosomal regions on seven chromosomes and explained 2.1–22.9% of the phenotypic variation. Five loci on four chromosomes had the phenotypic contribution >10%. Of the 15 associations between FSI and markers detected at 12 chromosomal regions on six chromosomes, the phenotypic variation explained by each marker ranged from 2.0 to 8.1%.

A total of 46 genotype/phenotype associations were detected for three parameters of fruit color, of which 16, 3, and 27 were for L^*, Hue, and Chroma, respectively (Table 3). These associations were on nine of the 12 chromosomes except chromosomes 7, 8, and 10. The phenotypic variation explained by these markers ranged from 2.0 to 23.4% for L^*, 2.2–4.2% for Hue, and 3.5–26.3% for Chroma. One marker Sli685 on chromosome 6 was common for all three parameters, seven markers (Sli2799, Sli2940, Sli477, Sli508, Sli571, Sli1346, and Sli1970) were common for L^* and Chroma, and one marker Sli2792 on chromosome 2 was common for Hue and Chroma.

Co-localization of QTLs was also observed (Table 3). Seven phenotype/genotype associations for SSC also contributed to FW, fruit shape or fruit color. One QTL on chromosome 5 for fruit shape conferred to Chroma as well. For the three parameters for fruit color, it was not surprised that the QTLs for Hue and Chroma were co-localized because the value of Hue was derived from Chroma (Yang et al., 2004). However, eight QTLs for L^* also had impact to Chroma, which was consisted with that the increase of chromaticity makes a color becomes more intense (Yang et al., 2014).

To identify loci for FW and SSC in the F₂ populations, the F₂ population of OH88119 × Black cherry grown in the spring season 2013 was subjected to initial test using 56 InDel markers. Four were significantly (p < 0.05) associated with FW (Table 4). The alleles from Black cherry contributed to small fruit. Two markers Sli2788 and Sli2772 on chromosome 2 contributed 24.2 and 17.9% of total phenotypic variation for FW, respectively. One marker Sli2377 on chromosome 4 explained 6.1% phenotypic variation of FW was close to the marker Sli2388 explaining 6.5% phenotypic variation in the 192 association panel. These four marker-trait associations were validated in the F₂ population of OH88119 × Black cherry grown in fall 2012. However, only the two markers on chromosome 2 could be validated in the F₂ population of OH9242 × Black cherry (Table 4).

A total of 14 markers were identified to be significantly (p < 0.05) associated with SSC in the F₂ population of OH88119 × Black cherry grown in the spring season 2013 (Table 4). These markers were from 10 regions on six chromosomes and contributed 5.3–14.9% of total phenotypic variation. Alleles of most markers from Black cherry contributed to high SSC. However, alleles of the marker Sli762 on chromosome 6 and the marker Sli1926 on chromosome 11 from Black cherry contributed to low SSC. Six markers spanning ~6.7 Mb region on chromosome 6 explained the highest phenotypic variation (11.2–14.9%) was in the same region of the marker Sli761 detecting high association with SSC in the 192 association panel.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.