Unconditional and conditional QTL analyses of seed fatty acid composition in Brassica napus L.

The fatty acid composition of B. napus’ seeds determines the oil’s nutritional and industrial values, and affects seed germination. Many studies have reported correlations among C16:0, C18:0, C18:1, C18:2 and C18:3 based on phenotypic data; however, the genetic basis of the fatty acid composition in B. napus is still not well understood. In this study, unconditional and conditional quantitative trail locus (QTL) mapping analyses were conducted using a recombinant inbred line in six environments. In total, 21 consensus QTLs each for C16:0, C18:0 and C18:2, 16 for C18:1 and 22 for C18:3 were detected by unconditional mapping. The QTLs with overlapping confidence intervals were integrated into 71 pleiotropically unique QTLs by meta-analysis. Two major QTLs, uuqA5–6 and uuqA5–7, simultaneously affected the fatty acids, except C18:0, in most of environments, with the homologous genes fatty acid desaturase 2 (FAD2) and glycerol-3-phosphate sn-2-acyltransferase 5 (GPAT5) occurring in the confidence interval of uuqA5–6, while phosphatidic acid phosphohydrolase 1 (PAH1) was assigned to uuqA5–7. Moreover, 49, 30, 48, 60 and 45 consensus QTLs were detected for C16:0, C18:0, C18:1, C18:2 and C18:3, respectively, by the conditional mapping analysis. In total, 128 unique QTLs were subsequently integrated from the 232 conditional consensus QTLs. A comparative analysis revealed that 63 unique QTLs could be identified by both mapping methodologies, and 65 additional unique QTLs were only identified in conditional mapping. Thus, conditional QTL mapping for fatty acids may uncover numerous additional QTLs that were inhibited by the effects of other traits. These findings provide useful information for better understanding the genetic relationships among fatty acids at the QTL level.


Background
Brassica napus (AACC, 2n = 38) is the second most important oilseed crop worldwide. B. napus' oils have diverse uses, ranging from food to industrial feedstock, and are an environmentally friendly and renewable energy source [1]. Fatty acid (FA) composition significantly affects the function, quality and nutritional properties of vegetable oils. To meet the steadily growing global requirements for rapeseed oil, there is an urgent need to develop desirable cultivars with improved FA compositions.
Although several major QTLs have been identified for seed FA composition in B. napus, few of them could be effectively utilized in breeding programs because most of the studies have been based on low-density genetic maps and applied traditional markers, resulting in QTLs with large confidence intervals. High-density maps could benefit QTL mapping by providing more precise parameter estimates [11]. In B. napus, the Brassica 60 K single nucleotide polymorphism (SNP) BeadChip Array, containing 52,157 SNP loci, was produced [12,13], which has facilitated the construction of a high-density, sequence-based, genome-wide polymorphism screening map. Several high-density SNP maps were constructed to identify agronomically important traits, such as seed fiber [14], boron efficiency [15], apetalous characteristics [16] and seed oil and protein contents [17,18]. Using high-density SNP markers, loci for the FA composition of B. napus were detected in both QTL mapping [19] and genome-wide association studies (GWAS) [20,21].
FA biosynthesis in plants is a very complicated process. In Arabidopsis, more than 600 genes encoding enzymes or regulatory factors are involved in acyl-lipid metabolism [22]. However, only approximately 20% of these genes are represented by defined and characterized mutants [22]. The allotetraploid B. napus has a close evolutionary relationship with Arabidopsis [23,24]. Although the biological pathways of FA biosynthesis and modifications are well documented in Arabidopsis, lipid metabolism and its regulation are less well understood in B. napus. Different FA compositions share the same basic resources and are controlled by the same FA synthesis-related genes in plastids [22]. In most studies, different FAs are correlated with each other based on phenotypic data, and many of the QTLs for different FAs are co-localized [3][4][5][6]9]. When this occurs, it is difficult to distinguish such loci with pleiotropic effects from different tightly linked genes underlying the same locus or the specific genes control multiple traits [25]. A method for the multivariable conditional analysis was proposed for determining the contributions of component traits to a complex trait and for investigating the genetic relationship between two traits at the QTL level [26,27]. The conditional analysis method could exclude the contribution of a causal trait to the variation of the resultant trait [28]. Using the C16:0 and C18:1 content as an example, C18:1 conditioning on C16:0 allows a C18:1 analysis to be conducted independently of variation in C16:0 if C18:1 is genetically correlated with C16:0. The major advantage of this method is that the net contribution of C16:0 to C18:1 could be effectively determined. Based on this methodology, the genetic relationships between putatively interrelated traits in crops, such as plant height with respect to spike and internode lengths in wheat [29] and grain yield and its component traits in rice [30]. In B. napus, Zhao et al. [31] performed an interrelationship analysis between oil and protein contents, and found six QTLs had pleiotropic effects on both traits. However, none of the studies considered the FA composition in B. napus' seeds.
In this paper, a recombinant inbred line (RIL) was used to investigate the genetic relationships among C16:0, C18:0, C18:1, C18:2 and C18:3 in six experiments. The objectives were to: (1) identify QTLs affecting the FA composition of B. napus' seeds using a high-density SNP map; and (2) specify the genetic relationships among FAs at the QTL level by utilizing unconditional and conditional mapping approaches. The research will contribute to a better understanding of the genetic architecture of the FA composition in B. napus' seeds.

Phenotypic variation and correlation analysis for FA compositions
The phenotypic values of C16:0, C18:0, C18:1, C18:2 and C18:3 for the AH population were measured in six experiments. There was a wide segregation range for the five FA compositions, with a continuous normal distribution in all trials (Fig. 1), indicating that the compositions were all quantitative traits controlled by polygenes. Strong transgressive segregations were observed in all experiments ( Fig. 1). Table 1 shows the correlation coefficients between different FA compositions based on means of AH lines in Shannxi and Jiangsu Provinces. C16:0 showed a highly positive correlation with C18:0, C18:2 and C18:3, but had a significant negative relationship with C18:1 in both locations. C18:1 is the most important unsaturated FA in the oil based on potential human health effects, and it had a significant negative relationship with the other FA levels, except C18:0.

Unconditional unique QTL for the five traits
In total, 101 consensus QTLs for the five examined traits were obtained, including 21 each for C16:0, C18:0 and C18:2, 16 for C18:1 and 22 for C18:3. A large proportion of the QTLs formed clusters on several chromosomal regions, indicating that these loci might affect several FA contents. To distinguish genetic explanations of the correlations between the FA concentrations, these consensus QTLs were integrated into unique QTLs. Consequently, 71 unique QTLs distributed throughout 17 chromosomes (excluding A10 and C8) were obtained, with the main QTLs controlling one (49 QTLs) or two (16 QTLs) traits (Additional file 5). Four unique QTLs (uuqA4-2, uuqA5-2, uuqC4-3 and uuqC4-9) simultaneously affected C18:1, C18:2 and C18:3. All four of these unique QTLs had positive additive effects on C18:2 and C18:3, but had negative additive effects on C18:1. Furthermore, two QTLs (uuqA5-6 and uuqA5-7) controlled the FA contents, except C18:0, were scattered over the A5 chromosome, with very close distances, and contributed a large proportion of PV for each FA content in most of the environments. Both QTLs had positive additive effects on C16:0, C18:2 and C18:3, but had significant negative additive effects on C18:1. These findings may explain the high positive correlations between C16:0, C18:2 and C18:3, and their remarkable negative correlations with C18: 1 and weak correlations with C18:0, as shown in Table 1.

Conditional unique QTLs for the five traits
Based on the conditional phenotypic values when C16:0, C18:0, C18:1, C18:2 and C18:3 were conditioned on each other, 49, 30, 48, 60 and 45 conditional consensus QTLs, respectively, were obtained in the six environments. These QTLs were integrated into 128 unique QTLs and distributed across all 19 chromosomes, except for A10 (Fig. 4, Additional files 8 and 18). Of these unique QTLs, 68 affected only one trait, while 60 had effects on two to five traits. Two conditional unique QTLs, cuqA5-8 and cuqA9-8, affected the concentrations of all five FAs, and 11 QTLs influenced four different FA contents (Additional file 18). Moreover, 16 and 31 other QTLs were associated with three and two FA conditional phenotypic values, respectively.

QTL comparison between unconditional and conditional mapping methodologies
In this study, QTLs detected by unconditional and conditional mapping analyses were compared. When QTLs identified by the two methods for the same trait had overlapping CIs, they were assumed to be identical. For C16:0, C18:0, C18:1, C18:2 and C18:3, 21, 21, 16, 21 and 22 consensus QTLs, respectively, were identified by the unconditional QTL mapping (Fig. 4 and Additional file 8).
In total, 71 unconditional unique QTLs and 128 conditional unique QTLs were obtained for the five FAs. A comparative analysis of the unique QTLs detected by the two methods revealed that 88.7% (63/71) of the unconditional unique QTLs were observed co-locating with conditional unique QTLs, and 65 additional unique QTLs were obtained when conditional QTL mapping was performed (Fig. 4, Additional files 8 and 19). The QTLs identified by multiple programs probably contained major genes associated with FA concentrations in B. napus seeds.

Identification of potential candidate genes related to fatty acid synthesis
The 71 unconditional unique QTLs spanned a region of 220 cM, representing 10.85% of the total linkage map length. Further analyses showed that more than 5800 genes in B. napus were located in the QTLs' CIs (data not shown). After a careful comparison with the FA synthesis genes in Arabidopsis [32], 150 of these genes were regarded as potential candidate genes affecting the five FA contents (Additional file 20). These candidate genes have roles in 22 different pathways, including plastidial FA synthesis, triacylglycerol (TAG) synthesis and lipid signaling. QTLs uuqA5-6 and uuqA5-7 were the most important, with large additive effects, that controlled the contents of the FAs, except C18:0, in most of the six environments. Two well-known FA synthesis genes, BnaA05g26900D (homologous gene of FAD2) in the TAG synthesis pathway and BnaA05g27110D (GPAT5) in the aliphatic suberin synthesis pathway, were found in the CI of uuqA5-6. One or more important genes affecting FA concentration may be in the CI of uuqA5-7 based on the QTL mapping results. Four candidate genes, BnaA05g28270D (CYTOCHROME P450) in cutin synthesis, BnaA05g28450D (SUGAR-DEPENDENT 6) in mitochondrial phospholipid synthesis, BnaA05g28620D (AT3G09920) in lipid signaling and BnaA05g28920D (PAH1) in TAG synthesis and eukaryotic phospholipid synthesis (Additional file 20), were found in the CI of uuqA5-7. However, whether these genes or presently unidentified genes exerted great effects on FA concentrations, is still unclear.
In comparison, 65 new unique QTLs were detected by the conditional mapping analysis (Additional file 19). These QTLs covered 205.2 cM, with an equivalent physical region of 37.68 Mb. A total of 4633 B. napus genes were mapped on this region, and 164 of these genes were considered to be potential candidate genes (Additional file 21). A number of genes that had been confirmed to control FA synthesis were also assigned to conditional unique QTLs, such as LPAAT4 (BnaA07g21920D) and KASII (BnaA07g21940D), which were assigned to the QTL cuqA7-3, BC (BnaA09g48250D) to the QTL cuqA9-12, BCCP1 to the QTL cuqC3-1 and GPAT2 (BnaC05g01190D) to cuqC5-1 (Additional file 21). Intriguingly, three regulatory factors underlying the QTLs' CIs were also found: FUS3 (BnaA02g28280D) was located in the CI of the QTL cuqA2-4, LEC1 (BnaC08g20060D) was associated with cuqC8-2 and ASIL1 was in the CI of cuqC6-2. Thus, the combination of the two analytical methods identified promising functional genes that regulate FA biosynthesis.

Discussion
The FA composition of B. napus' seeds determines the oil's nutritional and industrial values, and affects seed germination. Understanding the genetic control is a vital step in improving the oil. The FA levels of seeds are quantitative traits, and a large number of QTLs affecting FAs have been identified in B. napus [3][4][5][6][7][8][9][10]33]. In this study, a high-density SNP map was used to map QTLs associated with five FAs on six environments, which allowed us to identify more reliable QTLs and candidate genes involved in regulating the composition of FAs in B. napus.

Conditional QTLs were divided into four types
The unconditional QTL analysis showed that a number of QTLs affecting multiple traits, which was in accordance with the significant correlations based on phenotypic data. To evaluate possible genetic relationships among the five FAs at the individual QTL level, conditional mapping was performed using data of C16:0, C18: 0, C18:1, C18:2 and C18:3 conditioned on each other, and 232 conditional consensus QTLs were obtained for the five FAs. Compared with the results of the unconditional mapping analysis, these conditional QTLs could be divided into four types: (1) QTLs that were detected only in the unconditional QTL analysis. Taking ucqPA. A5-3 as an example, this QTL was repeatedly detected in all six experiments with a large additive effect for C16:0 (Additional file 1); however, when C16:0 was conditioned on C18:1, ccqPA.A5-6 (co-localized with ucqPA.A5-3) failed to show a significant effect in any of the experiments (Additional file 6). This indicated that ucqPA.A5-3's effect on C16:0 was entirely contributed by C18:1, and the genes underlying this locus simultaneously influenced the C16:0 and C18:1 contents; (2) QTLs that were detected in unconditional and conditional QTL analyses had similar additive effects. This phenomenon can be illustrated using the example of ucqPA.A5-2, which was the major QTL for C16:0 and was expressed in all six experiments (Additional file 1). The conditional QTL ccqPA.A5-5, which co-localized with ucqPA.A5-2, was still repeatedly detected for PA/ST in the six experiments with very similar genetic effect values (Additional file 6). These represent genes in the CI of ucqPA.A5-2 that control the C16:0 content independently from the C18:0 content; (3) Although QTLs could be identified by both unconditional and conditional mapping, the assessment of the additive effects was dramatically changed by the different mapping methodologies. For instance, ucqOL.A5-1 contributed 25.52% to the C18:1 content in the 13NJ environment with an additive effect of − 3.13 (Table 2), while ccqOL.A5-2 (corresponding to ucqOL.A5-1) was still significant when the influence of C16: 0 on C18:1 was excluded, and it explained 19.43% of PV with a reduced additive effect of − 2.55 (Additional file 12). This suggested that the effect of ucqOL.A5-1 on C18:1 was partially due to the genetic effect on the C16:0 content; and (4) Additional QTLs were only detected by the conditional mapping method. These QTLs abounded in the present study, including 31, 14, 34, 41 and 30 for C16:0, C18:0, C18: 1, C18:2 and C18:3 (Additional files 9, 11, 13, 15 and 17), respectively. The expression of these QTLs may have been completely suppressed by conditional traits; thus, their effects could only be detected when the influence of the conditional traits was eliminated. Together, these may better explain the genetic relationships among the five FAs at the individual QTL level compared with the correlations from the phenotypic data. Similar phenomena were also discovered in previous studies relating to oil and protein contents in B. napus [31], plant height, spike and internode lengths in wheat [29], and spike number, kernel number and thousand-kernel weight in wheat [36].
The combination of unconditional and conditional QTL mapping is a powerful tool for dissecting the genetic basis of FA composition The basic pathway of acyl-lipid metabolism is well characterized in Arabidopsis [22]. However, FA biosynthesis, modification and assembly into triacylglycerides are less well understood in B. napus because it has a more complex genomic structure than Arabidopsis. Brassica species and Arabidopsis have high degrees of sequence similarities and chromosomal collinearities [23,24], and the possibility that genes that carry out the core biological processes will be orthologs. In fact, several orthologs encoding major enzymes involved in FA metabolism were mapped in B. napus, such as FAD2 [8,10,37,38], FAD3 [8][9][10] and fatty acid elongase 1 [39,40]. Using a comparative genome analysis, 150 orthologs were obtained underlying the 71 unconditional unique QTLs (Additional file 20). The most important unique QTL uuqA5-6, which simultaneously affected C16:0, C18:1, C18:2 and C18:3, involved two wellknown candidate genes. A candidate for this QTL was FAD2 that encodes the enzyme that catalyzes the desaturation of C18:1 to C18:2, which was in accordance with previous studies [8,10,37,38]. Another candidate was GPAT5, which exhibits a strong preference for sn-2 acylation and produces sn-2 lysophosphatidic acids as the major products of TAG synthesis [41]. Wang et al. [3] reported that GPAT5 was associated with QTLs on A3, C3 and C5 in B. napus. Additionally, uuqA5-7, another major QTL was also detected on A5, explaining 16.48%, 24.42%, 25.98% and 17.60% of the PVs for C16:0, C18:1, C18:2 and C18:3, respectively, in different experiments. Among the genes underlying the CI of uuqA5-7, PAH1 (At3g09560) encodes a phosphatidate phosphohydrolase, which is a key enzyme in the regulation of lipid synthesis and catalyzes the dephosphorylation of PA, yielding DAG and Pi [42]. This gene was the most likely candidate gene for uuqA5-7, but evidence that PAH1 plays an important role in FAs biosynthesis in B. napus is lacking. In addition, FAD3 was associated with QTLs uuqA4-4, uuqC3-6 and uuqC4-9 detected by two, two and three traits, respectively. BCCP2 and beta-CT, subunits of the acetyl-CoA carboxylase complex in plastids, were assigned to uuqA3-2 and uuqC2-1, respectively. Compared with the unconditional QTL analysis, 65 additional unique QTLs were obtained by conditional QTL mapping (Additional file 19), and 164 orthologs were in the CI of the new QTLs, including 6 and 17 genes involved in plastidial FA and TAG synthesis, respectively (Additional file 21). Three critical transcriptional factors, including LEAFY COTYLEDON1 (LEC1, AT1G21970) [43], FUSCA3 (FUS3, AT3G26790) [44] and Arabidopsis 6b-interacting protein 1-like 1 (ASIL1, AT1G54060) [22], regulating the oleosin gene's expression and lipid accumulation were located in the CI of conditional unique QTLs, which could not be found in unconditional QTLs. LEC1 was associated with cuqC8-2, FUS3 was associated with cuqC9-4 and ASIL1 was associated with cuqC6-2. In a previous study, LEC1 was assigned to A3, A8, A9 and C9, while FUS3 was assigned to C7 [3].

Conclusions
In this study, unconditional and conditional QTL mapping analyses were performed to decipher the genetic control of FAs in B. napus. Three pleiotropically unique QTLs (uuqA5-6, uuqA5-7 and uuqC4-9) with important value for MAS were obtained from the unconditional mapping analysis, and uuqA5-7 was a new major QTL for C16:0, C18:1, C18:2 and C18:3. A total of 232 conditional consensus QTLs were detected for the five FAs, and these QTLs were divided into four different types. Compared with the results of the unconditional mapping analysis, 65 new unique QTLs were detected when conditional QTL mapping was performed. The combination of two mapping methodologies provided useful information for MAS and the improvement of the FA composition of B. napus' seeds.

Plant materials
A RIL population containing 189 lines and named the AH population, was used for QTL analyses of seed FA composition in the present study [16]. The two parents (' APL01' and 'Holly') were double low rapeseeds, with traces of erucic acid in the oil, but both had high levels of C18:1 (Fig. 1). The AH population was previously used for developing a high-density SNP map and for detecting QTLs associated with apetalous characteristics [16]. The genetic linkage map covered all 19 B. napus chromosomes of 2027.53 cM, with an average spacing of 0.72 cM between SNP-bins.

Field trials and data collection
The AH population, as well as the two parents, were evaluated in six environments. The materials were  [16]. At maturity, five representational plants were bulk harvested, and the seeds were used for FA measurements. The FAs profiled included C16:0 (Abbreviated as PA), C18:0 (ST), C18:1 (OL), C18:2 (LI) and C18:3 (LN). Bulked seed samples were analyzed by gas liquid chromatography using an Agilent 7890 series gas chromatograph (USA) in 12NJ and 13NJ environments according to Rücker and Röbbelen [45] and were determined by near infrared reflectance spectroscopy in 14DL, 14NJ, 15YL and 15NJ environments using a Foss NIRSystems 5000 according to the WinISI III manual's instructions.

Data analyses
Correlation analyses were implemented using SPSS 18.0 software (SPSS Inc., Chicago, IL, USA). Phenotypic correlation coefficients among the five FA compositions were calculated based on the traits for the two provinces. Unconditional phenotypic values were the mean value of the two replicates for each environment. The conditional values are estimated for the no-variation situation in the secondary trait, a method very similar to the estimation of adjusted values in a covariance analysis. The mixed model method in software QGASta-tion1.0 (http://ibi.zju.edu.cn/software/qga/) of the conditional analysis for quantitative traits was used to predict the conditional phenotypic values y(T1|T2) [27], where T1|T2 indicates trait 1 conditioned on trait 2 [31]. The default parameters of the model were used in the present study. For example, y(OL|LI) is the conditional phenotypic value of OL without the influence of LI. In previous studies, C16:0 showed highly significant correlations with C18:0, C18:1, C18:2 and C18:3 [3,4,21]. To investigate the genetic relationships among C16:0 and other four fatty acids, conditional QTL mapping analysis for C16:0 was also performed, although it is the first fatty acid type comparing to the other four types.

QTL detection and meta-analysis
All five FAs were conditionally analyzed with each other in the six environments. Then, unconditional and the conditional phenotypic values for each trait collected in each environment were employed for QTL mapping analyses and named as unconditional QTLs and conditional QTLs, respectively, by the Windows QTL Cartographer 2.5 using the composite interval mapping model [46]. A stringent LOD threshold (2.8-3.1) of putative QTLs were determined by selecting 1000-fold permutation tests (α = 0.05), and these QTLs were termed 'identified QTLs'. The QTL intervals were established by 2-LOD as approximately 95% QTL confidence intervals (CIs). A 'two-round' strategy of QTL integration was implemented to meta-analyze QTLs with overlapping CIs by the BioMercator V4.2 program [47]. In the first round, identified QTLs consistently expressed in different environments and with overlapping CIs for each trait were integrated into consensus QTLs. If a QTL that explained more than 20% of the phenotypic variation (PV) in at least one environment or more than 10% of the PV in at least two environments, then the QTL was defined as a major QTL [25]. In the second round, overlapping consensus QTLs for the different traits were integrated into pleiotropic unique QTLs [48]. The QTL nomenclature followed the method of Wang et al. [49] with certain modifications. Identified unconditional QTLs, were designated at the beginning with a prefix "uiq" (unconditional identified QTL), follow by the trait abbreviation, experiment code (1 = 12NJ, 2 = 13NJ, 3 = 14NJ, 4 = 14DL, 5 = 15NJ and 6 = 15YL) and linkage group (A1-A10 and C1-C9). If two or more identified QTLs were identified in a linkage group, a serial number was suffixed (e.g., uiqPA6.A1-1). Consensus QTLs were named with the prefix "ucq" (unconditional consensus QTL), trait abbreviation and linkage group (e.g., ucqPA.A5-2). Unique QTLs were named with the designation "uuq" (unconditional unique QTL) followed with the linkage group and the serial number of the QTL (e.g., uuqA5-6). For conditional QTLs, the name of identified QTLs, consensus QTLs and unique QTLs referred to the name of the corresponding unconditional QTLs, a designation beginning with the abbreviation "ciq", "ccq" and "cuq", respectively.