Comparative quantitative trait loci for silique length and seed weight in Brassica napus

Silique length (SL) and seed weight (SW) are important yield-associated traits in rapeseed (Brassica napus). Although many quantitative trait loci (QTL) for SL and SW have been identified in B. napus, comparative analysis for those QTL is seldom performed. In the present study, 20 and 21 QTL for SL and SW were identified in doubled haploid (DH) and DH-derived reconstructed F2 populations in rapeseed, explaining 55.1–74.3% and 24.4–62.9% of the phenotypic variation across three years, respectively. Of which, 17 QTL with partially or completely overlapped confidence interval on chromosome A09, were homologous with two overlapped QTL on chromosome C08 by aligning QTL confidence intervals with the reference genomes of Brassica crops. By high density selective genotyping of DH lines with extreme phenotypes, using a Brassica single-nucleotide polymorphism (SNP) array, the QTL on chromosome A09 was narrowed, and aligned into 1.14-Mb region from 30.84 to 31.98 Mb on chromosome R09 of B. rapa and 1.05-Mb region from 27.21 to 28.26 Mb on chromosome A09 of B. napus. The alignment of QTL with Brassica reference genomes revealed homologous QTL on A09 and C08 for SL. The narrowed QTL region provides clues for gene cloning and breeding cultivars by marker-assisted selection.

via screening the genomic regions that show high differences of reads between 'Highest' and 'Lowest' bulks. In comparison with the traditional QTL mapping, the NGS-aided strategy provides a simple and effective alternative to rapidly identify QTL of interest by genotyping small number of samples from two sets of individuals with distinct or opposite extreme phenotypes 23,24 . By using the NGS-aided strategy, a few QTL of the interested traits have been successfully identified in yeast 23,[25][26][27] , rice 24,28,29 , Arabidopsis thaliana 30 , sunflower 31 , cucumber 32 , wheat 33 , tomato 34 and chickpea 35 .
In this study, the strategies of conventional QTL mapping and high-throughput genotyping were combined to dissect QTL of SL and SW in a doubled haploid (DH) population and its reconstructed F 2 (RC-F 2 ) population of rapeseed. Based on the alignment of SSR markers to the reference genomes of Brassica crops, the genetic region on chromosome A09 where the 17 overlapped QTL of SL and SW on chromosome A09 enriched, was revealed to be homologous with the overlapped QTL for SL on chromosome C08. The major QTL region on chromosome A09 was aligned to a ~1 Mb region on the reference genome of B. rapa and B. napus with high density SNP array.

Results
Variation in silique length and seed weight. The semi-winter parental line 'SWU07' exhibited higher SL and SW values than the winter parental line 'Express' . Wide variation was detected in both the DH and RC-F 2 populations for SL and SW across the years analyzed (Fig. 1). The field performance of the 233 RC-F 2 lines, with an average SL of 6.13 ± 0.71 cm and an average SW of 3.77 ± 0.38 g, was superior to that of 261 DH lines, which had an average SL of 5.67 ± 0.81 cm and an average SW of 3.47 ± 0.46 g. The normal distribution for SL and SW in both populations suggested that SL and SW were controlled by multiple genes (Fig. 1).
The ANOVA results showed significant differences among genotypes, years and genotype-by-year interactions for SL and SW in the two populations (P < 0.01) (Table S1). High broad-sense heritability was detected for SL and SW (average of 83.79% for SL and 79.13% for SW). The significant and positive correlation between SL and SW in both populations (r = 0.49 and 0.34 in the DH and RC-F 2 populations, respectively; P < 0.01) suggested that long silique had the potential to increase SW. QTL analysis. 20 QTL were located on chromosomes A01, A05, A09, and C08 for SL in the DH and RC-F 2 populations across years, totally explaining 55.1-75.7% of the phenotypic variation, whereas 21 QTL were detected for SW on A04, A05, A06, A09, C02, and C05, totally explaining 24.4-62.9% of the phenotypic variation in the DH and RC-F 2 populations across years (Table 1, Fig. 2). Among these QTL, 17 QTL for SL and SW were enriched in a region on chromosome A09 from 80 cM to 107.5 cM with partially or completely overlapped confidence interval. The direction of positive additive effect for these QTL was consistent from parental line 'SWU07' (Fig. 2). No digenic interaction with significant and high effect was found for SL and SW, although nine and ten additive by additive interactions were detected for SL and SW with minor effects, total explaining 2.78% and 4.14% of the phenotypic variation in DH and RC-F 2 populations, respectively. Together the positive correlation between SL and SW and overlapped confidence intervals of QTL, indicated that some common genetic factors might regulate both SL and SW.
Given the high degree of synteny between A09 and C08 (http://genomevolution.org/wiki/index.php/ Brassica_oleracea_v._Brassica_rapa), the microsynteny between QTL of SL on A09 and C08 was compared by aligning QTL intervals with the reference genomes of B. rapa and B. oleracea (Fig. 3). The interval of 17 overlapped QTL for SL and SW on chromosome A09, linked with 9 molecular markers (CNU402 ∼ CNU263), was aligned to the regions from 26.        Fig. 3; Table S2). The co-localization of QTL suggests homoeologous duplicated QTL for SL on chromosome A09 and C08 in rapeseed.
In order to narrow the QTL region in A09, 25 DH lines with extreme performance of SL and SW, were screened by 293 SSR markers to evaluate genetic background. Five and six lines were finally chosen to construct the 'Large' and 'Short' groups, respectively. Thus the two groups shared the same genetic background except for the region of QTL on chromosome A09. The average values for SL and SW in the 'Large' group (SL 7.46 cm and SW 4.18 g) were significantly higher (p < 0.01) than the 'Small' group (SL 3.97 and SW 3.04 g). Each individual from two groups was subjected to high-density selective genotyping using the Brassica 60 K SNP Bead Chip Array. Of the 52,157 SNP markers in the Brassica SNP array, 41,645 (80%) could be detected among 11 DH lines, while only 2751 (6.61%) SNPs were polymorphic between the 'Large' group and 'Small' group, indicating their similar genetic background. The Δ (SNP-index) of each SNP locus was calculated by subtraction of SNP-index of the 'Large' group from the 'Small' group, and the Δ (SNP-index) trends were visualized by means of a sliding window (Fig. 4). A single peak region with a significant (p < 0.01) and high value of Δ (SNP-index), harboring 45 SNP markers (Table S3), was identified (Fig. 4). It corresponded to 1.14-Mb region from 30.84 to 31.98 Mb on chromosome R09 of B. rapa (Fig. 4), and 1.05-Mb region from 27.21 to 28.26 Mb on chromosome A09 of B. napus, which were located into the interval of QTL on A09 detected by conventional QTL mapping. A total of 241 and 225 genes were harbored in these two regions, respectively. Of which, 69.71% (168/241) genes on the chromosome R09 of B. rapa were orthologous to 73.78% (166/225) genes on the chromosome A09 of B. napus, suggesting the homology between two regions.
To confirm the narrowed region harboring QTL, ten region-specific SSR (RS-SSR) markers were developed according to the genomic sequence of the regions in reference genome of B. rapa and B. napus. Of which, two RS-SSR markers (CY-04 and CY-10) that exhibited polymorphisms between two parents were successfully mapped to the confidence interval of the QTL on A09 in DH population (Fig. 4), further indicating the consistency between QTL mapping and NGS-aided studies.

Discussion
QTL mapping is the main approach for genetic dissection of quantitative traits, which provides the start point for map-based cloning of related genes and marker-assisted selection (MAS) in plant breeding. Although QTL mapping for silique traits have been reported [11][12][13][14][15][16][17][18][19] , single population was almost adopted in these studies. In this study, two related populations, DH and its derived RC-F 2 populations were used. The RC-F 2 population holds unique characteristics with normal F 2 population if genotypic selection did not exist in the process of microspore culture, but has advantages over normal F 2 population. For example, it permits the possibility of replicated experiments in multiple years or environments. The major QTL on A09 were repeatedly detected in both populations, indicating the credibility of the QTL on A09.
More than 100 QTL have been detected for SL and SW in rapeseed [11][12][13][14][15][16][17][18][19] , but the detailed comparative or functional analyses of these QTL have not been reported. In this study, we assigned QTL regions onto the reference genomes of Brassica crops through BLAST analysis of markers linked to the QTL. This enabled us to detect homeologous duplicated QTL for SW and SL on B. napus chromosomes A09 and C08. In order to test the power of this approach, we compared our result with the study of Yang et al. (2012), who also identified a major QTL region for SL and SW on chromosome A09 in B. napus using different markers with this study 16 . By aligning markers in the flank of the QTL (Na10-B07 and CNU008) to the reference genome of B. rapa, the confidence intervals of the QTL were aligned to the genomic region on chromosome R09 of B. rapa from 30.1 to 32.2 Mb, which partially overlapped with that of QTL detected in the current study (Fig. 3). These findings supported the credibility of QTL on A09 controlling SL and SW.
The conventional approach to narrow QTL regions is a laborious process that requires the development of DNA markers and the generation of a large number of advanced-generation progenies. These Figure 3. Comparative analysis of QTL for silique length and seed weight on chromosomes A09 and C08 via alignment of the SSR loci linked with the QTL to chromosomes R09 and O08 of the reference genomes of B. rapa and B. oleracea, respectively. ES-DH was derived from the cross between 'Express' and 'SWU07' in this study, and the NS population was from the study of Yang et al. (2012). Lines beside linkage groups represent QTL for silique length and seed weight, respectively. The light-grey highlighted regions on the linkage groups showed the QTL intervals, and the dark-grey highlighted regions on the genomes exhibited the co-location physical regions of the QTL on A09 and C08. requirements limit its use because they are time consuming and costly, particularly in annual crop species like rapeseed. With the release of reference genomes for some species and advances in NGS technology, several novel strategies for QTL mapping have been proposed with the use of high-throughput genotyping, such as microarray-based genotyping or massively parallel sequencing. In the present study, the process of QTL-seq was followed with slight modification, i.e. a high-density Brassica SNP array, instead of resequencing of bulks in QTL-seq, was used to genotype DH lines with the extreme trait values, and the QTL were successfully anchored to a ~1 Mb region on chromosome R09 of B. rapa and A09 of B. napus. Given the high-throughput nature and low cost of the SNP array, this modified approach will dramatically accelerate the process of QTL fine-mapping in a cost-effective manner.

Conclusions
We identified 20 and 21 QTL for SL and SW in DH and RC-F 2 populations of B. napus across three years. A significant and positive correlation between SL and SW, and overlapped confidence intervals among partial QTL for SL and SW detected in this study suggested long silique has the potential to increase SW. The major QTL for SL on chromosome A09 and C08 were aligned to the same region of the reference genomes of Brassica crops, suggesting they are homologous QTL. By high density selective genotyping of DH lines with extreme phenotypes, using a Brassica single-nucleotide polymorphism (SNP) array, the region of major QTL on chromosome A09 was aligned to a ~1 Mb region on the reference genome of B. rapa and B. napus.

Methods
Plant materials and phenotypic evaluation. A B. napus DH population, consisting of 261 lines, was developed from a cross between the European winter cultivar 'Express' (female) and Chinese semi-winter line 'SWU07' (male). Two parental lines showed diversity in SL and SW. Two rounds of random crosses between DH lines were performed to reconstruct 233 RC-F 2 lines. Each DH line was used once each round.
The  Ten well-developed siliques per plant at the base of the inflorescence were collected, and siliques from ten individuals in the center of each plot were pooled to measure silique length and weight at maturity. Statistical analysis. Analysis of variance (ANOVA) was performed using the GLM procedure of SAS , where σ g 2 , σ ge 2 and σ e 2 are estimates of the variances of genotype, genotype × environment interactions, and error, respectively, n is the number of environments, and r is the number of replications per environment 36 . Pearson's correlation coefficients between traits of interest were calculated using the CORR procedure of SAS 37 . QTL analysis. DNA isolation, development of molecular markers and construction of genetic linkage groups were described in previous study 38 , where 293 SSR markers were arranged into 19 B. napus chromosomes, spanning a genetic distance of 1,188 cM with an average distance of 4.05 cM between adjacent markers. Besides, the sequence of target region was downloaded from the reference genomes (http://brassicadb.org/brad/index.ph), and screened for SSR loci using the software "SSR Locator" 39 . Ten region-specific SSR markers were developed according to the genome sequence of the target QTL region. The genotypes for RC-F 2 lines were deduced from the band patterns of their parental lines. A genotype score of '1' was given to a RC-F 2 line if the SSR marker was present in at least one of the parental lines, while the RC-F 2 line was assigned a score of '0' if the marker was absent in both parents. QTL detection was performed with the composite interval mapping (CIM) procedure of the software WinQTL Cartographer 2.5 40 . A 1,000-permutation test was performed to estimate a significance threshold of the test statistic for a QTL based upon a 5% experiment-wise error rate 41 . The genome-wide digenic interactions were estimated using QTL mapper V2.0 software 42 43 . The alignment criteria were set to allow three mismatches and one gap for a given primer pair. When a marker had multiple amplification loci on the same chromosome, an accurate position for a particular locus was determined manually by referring to the physical positions of its upstream and downstream markers.

SNP array analysis.
In order to narrow the confidence intervals of major QTL in A09, two pools of DH lines with extreme performance were constructed for SNP array analysis. Briefly, The DH lines with extreme performance of SL and SW were first evaluated by 293 SSR markers for evaluating genetic background. And the samples which shared the same genetic background except for QTL on chromosome A09, were selected to construct the 'Large' group (extremely long siliques and high seed weight) and 'Small' group (extremely short siliques and low seed weight). Each DH lines of the two groups was genotyped with the Brassica 60 K SNP Bead Chip Array (Illumina Inc., CA, USA), together with the parental line 'SWU07' . The single-nucleotide polymorphism (SNP) loci were aligned to the reference genomes of B. rapa (version 1.5) (http://brassicadb.org/brad/index.ph) and B. napus (http://www.genoscope.cns.fr/ brassicanapus/cgi-bin/gbrowse/colza/) The process of chip array analysis was performed in accordance with the method of Takagi et al. (2013) with slight modification 23 . The Δ (SNP-index) of each locus was calculated by subtraction of SNP-index of 'large' group from 'small' group with the formula, Δ (SNP-index) = k/5 − j/6, where k and j are the number of accessions that exhibit a consistant genotype from 'SWU07' in the 'Large' and 'Small' groups, respectively. A sliding window analysis was applied to generate Δ (SNP-index) plots with a window size of 80 SNP and increment of 1 SNP. 1,000 random resamplings were performed to estimate a significance threshold of the test statistic for a QTL based upon a 1% experiment-wise error. Thus the statistical confidence intervals under the null hypothesis of no QTL were determined at a significance level of p = 0.01