QTL architecture of reproductive fitness characters in Brassica rapa

Study system

The recombinant inbred lines (RILs) used in this study have been previously described in detail [48, 54, 55]. Briefly, the RILs were derived from a cross between two inbred B. rapa lines, the seed oil cultivar, yellow sarson (R500), and a rapid-cycling genotype (IMB211). Yellow sarson is a highly inbred, annual B. rapa cultivated in India for over 3,000 years [56]. IMB211 is derived from the Wisconsin Fast Plant™ (WFP) population produced by selection for short generation time [57]. The parental genotypes differ greatly for several reproductive fitness traits. R500 produces the large, heavy, yellow seeds typical of B. rapa cultivars, and IMB211 produces smaller, lighter, brown seeds more similar to naturalized B. rapa populations. IMB211 is phenotypically similar to rapid-developing B. rapa in the United States, and its average flowering day and height are within the range of wild-collected B. rapa [52]. The RILs were developed from self-fertilization of a F1 individual resulting from the R500 × IMB211 cross, and the resulting F2s were propagated by single-seed descent to the S6 generation [48].

Experimental environments

Seeds of 147 RILs and the parental lines were planted in an agricultural field at the University of Minnesota, Saint Paul, MN, April 29–31, 2004. In each of three subplots, four plants of each RIL were planted at 20 cm spacing from each other and the next RIL, resulting in three sets of four adjacent replicates within each RIL. This experiment was part of a larger one examining multiple competitive environments, and further details of the design have been previously described [55]. Bolting date was scored as the date when buds first differentiated from the apical meristem. All flowers were naturally open-pollinated in the field. Plants began to naturally senesce (leaves and stem began to turn brown and dry) in the first week of August and were harvested once 50% of the rosette leaves had senesced on all individuals. At harvest, fruits (siliques) were counted on each plant, and four mature (fully developed) fruits were collected at random from 3–5 replicates of a RIL. Seeds were counted in each fruit, and the length of the longest fruit was measured. Seeds were then pooled within a plant, and seeds-per-fruit, seed mass, seed area, and seed color were determined. Seeds-per-fruit is the mean number of seeds in the 2 longest collected fruit. Seed mass was calculated as the average mass of 20 randomly chosen seeds from the collected fruit. Ten seeds from each plant were then photographed, and their area and color were determined using ImagePro software, which quantifies pixels in digital photographs of the sample (ImagePro Plus V4.5 2001). The trait, seed color, was obtained as the score of the first principal component identified in principal components analysis, using varimax rotation, on pixel densities of three colors: red, green, and blue (PROC FACTOR; SAS, 2001). The first principal component, which was weighted most heavily in red and green, explained greater than 67% of the variation. A lower value for seed color indicates darker, more wild-like seeds.

Fitness traits were also recorded in a greenhouse experiment. Due to space limitations, the greenhouse experiment was conducted over two temporally separated trials, spaced a month apart. In the first trial, on March 14^th, 2007, we planted 8 replicates of each of the parental lines and 137 of the 160 B. rapa RILs into 3 in³ pots, filled with Metromix 200 soil, in greenhouses at the University of Minnesota, Saint Paul, MN. Bolting date was recorded for each individual using the same method as in the field, and these plants were then harvested.

Seeds were planted for the second experimental trial on April 17th, 2007 using the same RILs and in the manner described above. Six replicates per line were planted. At 5 days after bolting, plants began flowering, and the number of open flowers was recorded daily. The majority of flowers were allowed to self-pollinate naturally, except that flowers at the 6^th and 7^th nodes were self-fertilized by hand in order to more closely simulate the effect of pollinators in the field. In early June, the plants began to senesce. At senescence, fruits were counted on each plant, and fruits 6 and 7 (or those fruits arising from hand-pollination) were collected. Fruit length was measured on the longest of the fruits, and the seeds in each collected fruit were counted. Seeds were pooled within a plant, and seed traits were measured as in the field.

Germination characteristics of the seeds collected in the greenhouse experiment were also examined. Seeds collected from the 2 parental lines and RILs with at least 3 replicate plants that set seed were pooled and stored in darkness at room temperature. While excluding RILs with fewer than 3 replicate plants artificially eliminates lines with low viability, this sampling approach helps reduce microenvironmental bias. To further limit microenvironmental maternal effects, the same number of seeds were pooled from each replicate within a RIL combination if possible. Six replicates of 5–10 seeds per RIL were germinated in 3 Conviron E7/2 growth chambers (Controlled Environments Ltd., Winnipeg, Canada) in 2 temporal blocks from February 1–12, 2007 and from March 20–30, 2007. One replicate of each RIL was located in each growth chamber and temporal block, and replicates were randomized within a growth chamber. All seeds were planted onto 0.5% agar in 4.5 cm diameter plastic Petri dishes under a green safe light and then allowed to imbibe water for 24 h in darkness before being moved into the growth chambers. Each growth chamber was fitted with 8 white florescent lights (Silvania Octron, F038/871, 32 W), 2 infrared lights (Industrial Infrared, F32T8/IR-750), and 4 incandescent lights (Sylvania Double Life, Soft White, 120 V/75 W) that generated a R:FR of 1.1 and PAR of 130 – 160 μmol m^-2 s^-1. Germination trials used 12 L:12D lighting and temperatures of 24D:20 N °C. Germination was checked daily for 10 days and scored when the cotyledons appeared green and open to ~90°. Germination percent and day were calculated by dividing the number of seeds germinated by the seed total and by averaging the germination days for all seeds in a replicate Petri dish, respectively.

Quantitative genetic analysis

For all traits, a restricted maximum likelihood (REML) approach (PROC MIXED, SAS 2001) was used to partition variation attributable to RIL (V_L), spatial blocking (V_B), and residual error (V_R) within each environment. The blocking term was subplot for all traits in the field and greenhouse, and temporal block and chamber for the germination traits measured in the growth chamber. Seed color and fruit length met the assumption of normality in both the field and the greenhouse environments, but all remaining traits were transformed using a Box-Cox procedure [58]. All transformations greatly improved normality and homoscedasticity, and results from the transformed data are reported.

Broad-sense heritabilities were calculated as V_L/V_P, where, again,V_L equals the among-RIL variance component and V_P equals the total phenotypic variance for a trait. Least-square means, 95% confidence limits, and best linear unbiased predictor (BLUP) deviations were generated for each RIL using the transformed versions of the traits. BLUP deviations and least-square means were summed, and the resulting values were back-transformed for use in the QTL analysis. The parental genotypes, plants that died before bolting, and RILs with fewer than 3 replicates for a trait were removed from the analyses. Final sample sizes ranged from 730 to 1160 (out of a possible 3600) in the field and 910 to 1400 (out of a possible 2466) in the greenhouse. BLUPs were also used to estimate genetic correlations (r_G) within each environment (PROC CORR, SAS 2001). Differences in bivariate matrix structure across environments was tested using Flury’s hierarchical common principal components (CPC) analysis, using the "jump-up approach" and Fisher’s Z-test [59–61]. Cross-environment correlations (r_GE) for each trait were calculated as cov_1,2/√ (V_1,1 × V_1,2); cov₁₂ is the covariance of a trait across two environments and V_1,1 and V_1,2 are the among-RIL variances within each environment [62, 63].

QTL-analysis. The R500 × IMB211 RILs were genotyped at 227 RFLP and SSR markers across 10 linkage groups, representing the 10 chromosomes in B. rapa. Marker order for the linkage map was estimated from recombination frequencies observed in the entire population of 160 RILs, i.e., from the most complete dataset, in earlier experiments [48]. Centimorgan distances were then re-estimated in the RILs included in the field and greenhouse experiments presented here (those that survived to bolting) to account for slight differences that might exist due to unsampled recombination events. The map distances for each marker locus were calculated from the estimated recombination frequencies using the Kosambi mapping function in RQTL (R Development Core Team 2010). Twelve markers were uninformative, leaving 215 markers at unique positions in the final re-estimated linkage map.

QTL were mapped using the composite interval mapping (CIM) procedure in QTL Cartographer [64]. To control for effects of variation segregating elsewhere in the genome, we identified co-factors using forward-backward regression, and a 5-cM window; a maximum of 5 co-factors was selected for inclusion in the mapping model. We scanned for QTL at 2-cM intervals across the B. rapa genome. The significance threshold of the likelihood-ratio test statistic (LR) was determined for each trait in each treatment independently by randomly permuting the BLUPs 1000 times. QTL with LRs significant at α = 0.05 are reported [65] with 2-LOD support limits [66]. Additive effects and the genetic variance explained by a QTL were calculated in QTL Cartographer, and confirmed using a GLM model with all QTL detected in the genome-wide screen as main effects [67]. For a given locus, a positive additive effect indicates that the IMB211 parent allele conferred a higher value for the trait. Analysis of variance (ANOVA) was used to test for QTL × environment interactions across the field and greenhouse environments; all significant main-effect QTL for the trait, environment, and QTL × environment effects on each genotypic trait mean were included in the model (PROC GLM, SAS 2001).

Descriptive statistics

RILs varied significantly (P < 0.001) in phenotypic expression of all traits in both environments, indicating underlying genetic variation (Additional file 1: Table S1; Figure 1). Broad-sense heritabilities (H ² ) were smaller in the field (range, 0.08-0.79) than the greenhouse (range, 0.26-0.86) for all traits except fruit total (Table 1). We detected 7 QTL affecting 6 traits in the field and 12 QTL affecting 8 traits in the greenhouse (Table 2A and B; Figure 2). For any trait, a single QTL explained 5–56% (greenhouse) or 5–59% (field) of the variance (Table 2). Alleles derived from IMB211 had both positive and negative effects on the majority of traits with > 1 QTL in one or both environments, with the exception that the IMB211 allele always decreased the trait value for days to bolting, seed mass, and seed color.

Genetic architecture between environments

By contrast, genotypic correlations were non-significant for fruit total between environments (Table 3). Three of the five QTL affecting fruit total, including all QTL detected in the field for this trait, were environment-specific, in that the QTL × environment effects were significant (Table 2). QTL × environment effects were also significant for one QTL affecting each of days to bolting, fruit length, seeds-per-fruit, and seed mass, indicating that at least some of the genetic loci affecting each reproductive character, except seed color, were environment-specific or that functional differences between alleles at these QTL were manifested only in one environment.

Relationship among components of fitness

A number of bivariate correlations were conserved between environments. RILs that produced more fruit also produced more seeds-per-fruit, and longer fruits generally had heavier seeds and more seeds (although the fruit length/seeds-per-fruit correlation was significantly stronger in the greenhouse) (Table 3). Plants that produced more seeds-per-fruit also had darker seeds (negative value for seed color) in both environments.

Environmental differences in bivariate correlations were also observed. Plants that produced more fruit and more seeds-per-fruit also had heavier seeds in the field, but these correlations were not significant in the greenhouse (although the difference between environments was not large enough to be statistically significant for the seeds-per-fruit/seed mass correlation). These results suggest that genotypic differences in resource acquisition existed among plants grown in the field, that is, genotypes with high acquisition had higher output for all reproductive components than those with lower acquisition. In the greenhouse, plants did not generally produce both more seeds and larger seeds, suggesting that the genotypic differences in acquisition were smaller in this environment. The relationship between seed mass and color also differed between environments. Although the bivariate correlation between seed mass and color was positive in the greenhouse, the scatter plot showed a sharp increase in seed mass in the darkest colored seeds (Additional file 2: Figure S1), indicating that the lightest and darkest seeds had the largest mass. This relationship was not conserved in the field. Interestingly, significant Z-test differences indicated that the relationship between seed color and seed mass or fruit total reversed between environments, although the relationship between these traits was too weak in each environment for the majority of bivariate correlations to be significant.

The field-specific positive relationship between fruit total, seeds-per-fruit, and seed mass was consistent with FQTL1-1, which decreased the value for all three traits in the field and had a significant QTL × environment interaction effect for fruit total (Table 2). Moreover, GHQTL9-2 had opposing effects on fruit total versus seed mass, and the QTL × environment effects were significant for these traits (Table 2; Figure 2). These results indicate that GHQTL9-2 conferred environment-specific differences in resource allocation to reproductive fitness traits.

Transgenerational fitness

In the greenhouse environment, we examined the relationships between reproductive characters of the maternal parent and germination patterns in the progeny. Interestingly, later bolting maternal plants produced faster germinating seeds, suggesting a phenological relationship across generations (Table 3). This relationship was consistent with GHQTL10-1, which increased time to germination but decreased time to bolting (Table 2; Figure 2), providing evidence of transgenerational antagonistic pleiotropy at this locus. We also observed a positive relationship between both seed mass and seeds-per-fruit and germination percentage, suggesting that more vigorous maternal genotypes partition more resources to offspring and produce many large seeds that also germinate to a higher percentage.