Allometry models of leaves and petals in MAGIC lines
To detect the shape and size variation of leaves and petals within Arabidopsis thaliana, an allometric method based on a PCA of organ landmarks and outlines was used to quantify this trait. Leaf4, Leaf7 and petals from MAGIC lines were modelled and generated a separate data set. PCA was applied and the resulting principal components (PCs) were ranked according to the proportions of the total variance that each of them described (Figure 1, Supplementary Figure 1).
In Leaf4, the PCA revealed that 90.92% of the variance in organ shape and size was attributed to two PCs (Figure 1A). Leaf4.PC1 accounted for 76.84% of the total variance and affected the leaf size. Higher PC1 values corresponded to larger leaves, whereas lower values yielded smaller leaves. PC2 accounted for 14.08% of the variance and reflected the variations of leaf shape and petiole length. Plants with higher values of PC2 exhibited more elongated leaves with longer petioles, while lower with more rounded leaves and shorter petioles. PC3 explained 3.30% of the variance and displayed the degree of petiole twisting when the leaves were flattened, but its values were not significantly different between genotypes.
In Leaf7, the PCA revealed that 95.25% of the variance in organ shape and size could be attributed to three PCs (Figure 1B). In this model, Leaf7.PC1 caused 80.27% of the total variance, which mostly influenced leaf size, but also had a minor effect on shape. Higher PC1 values corresponded to larger, more elongated leaves, whereas lower values yielded smaller and more rounded leaves. PC2 was responsible for 11.42% of the variance and mostly arranged the steepness of the transition from petiole to blade. Higher PC2 values yielded longer petioles with a steep transition, and lower values yielded shorter petioles with a very gradual transition. PC3 accounted for 3.56% of the variance and affected mainly the shape. Lower values of PC3 yielded more elongated and narrower leaves, whereas higher values of PC3 yielded more rounded and wider leaves.
In petals, the PCA revealed that 92.02% of the variance in organ shape and size could be attributed to two PCs (Figure 1C). The PC1 accounted for 85.43% of the total variance and affected petal size. Higher PC1 values corresponded to larger petals, whereas lower values yielded smaller petals. PC2 accounted for 6.59% of the variance and affected mainly the shape. Low values of PC2 yielded elongated petals with a narrower shape, and high values of PC2 yielded rounded petals with a wider shape. PC3 accounted for 3.63% of the variance and was reflected in petal twisting when the petals were flattened. It was excluded from further analysis, because we could not detect significant differences in different genotypes.
The quantitative leaf and petal variations were captured by allometric models as PC values among the MAGIC lines (Table 1, Supplementary Figure 2). Extensive phenotypic variation was observed for all traits measured among the MAGIC lines, and the relative genetic contribution was estimated by broad-sense heritability (H2). The range of the H2 is from 0.62 to 0.87, which suggested the phenotypic variation among different lines was more attributed to the genetic component.
A correlation analysis between shape and size was also performed, and a number of significant pairwise correlations were observed (Figure 2). Leaf4.PC1 was significantly positively correlated with Leaf7.PC1, which represented the leaf size. Leaf4.PC2 was significantly correlated with Leaf7.PC2 and leaf7.PC3, which represented the leaf shape. Moreover, leaf shape and size showed significant correlations with petals. Petal.PC1 was significantly correlated with Leaf4.PC1 and Leaf7.PC1, which showed a negative size correlation between leaves and petals. Furthermore, both Leaf4.PC2 and Leaf7.PC2 were significantly positively correlated with Petal.PC1 and negatively correlated with Petal.PC2. The correlation between the leaf and petal allometry model indicated the genetic dependency and evolution correlation controlling leaf and petal allometry. Besides, a pairwise correlation analysis was performed between the life history traits and the leaf and petal allometry model (Figure 2). Leaf4.PC1 was correlated with rosette leaf number and stem height; additionally, Leaf4.PC2 was highly correlated with branch number and pod number; Leaf7.PC1 was correlated with days to bolting, days to flower and stem height; Leaf7.PC2 was highly positively correlated with days to bolting, days to flower, rosette leaf number, and branch number; Petal.PC1 was correlated with rosette leaf number and branch number; and Petal.PC2 was correlated with days to bolting and days to flower.
QTLs accounted for leaf and petal allometry
To examine the genetic basis for shape and size variation of leaves and petals along the PCs in the MAGIC lines, we treated each PC as a quantitative trait, whose variation frequency showed a normal distribution (Figure 2, Supplementary Figure 3) for QTL mapping. In QTL mapping of the MAGIC lines, the PCs for the leaf and petal allometry model and 1260 SNP markers among the 19 founder ecotypes were used. We then calculated a series of QTLs associated with the variance of leaf and petal shape and size (Figure 5, Supplementary Table 2, Supplementary Figures 4, 5 and 6). In the leaf model, the QTL analysis for Leaf4.PC1 identified four QTLs located on chromosomes 1 and 3 and one QTL located on chromosome 2 for Leaf4.PC2. For Leaf7.PC1, five QTLs were observed on chromosome 3, one QTL was located on chromosome 2 for Leaf7.PC2, and four QTLs were located on chromosomes 1 and 2 for Leaf7.PC3. In the petal model, three QTLs were identified on chromosomes 1 and 4 in Petal.PC1, and nine QTLs were identified on chromosomes 1, 2, 3, and 5 in Petal.PC2.
After comparing the positions for all the QTLs identified, there was some QTL overlapping in the leaf and petal allometry model. The QTLs for PC2 of the leaf (Leaf4.PC2: LF4_2.1, Leaf7.PC2: LF7_2.1) and petal (Petal.PC2: PE_2.5) on chromosome 2 (~11 Mb) overlapped, and the alleles from the Ler-0 accession formed the most rounded leaves and petals with the widest shape (Supplementary Table 3). This QTL likely stemmed from the mutation of ERECTA, which is known to affect fruit length and is due to the allele from the Ler-0 accession (Abraham et al., 2013). With the exception of the ER locus for leaf and petal shape, the QTLs LF7_1.1, LF7_1.2, LF7_1.3, LF7_1.4, and LF7_1.5 for Leaf7.PC1 on chromosome 3 overlapped with QTL PE_2.6 for Petal.PC2. Moreover, the QTLs LF7_1.3, LF7_1.4, and LF7_1.5 also overlapped with QTL PE_2.7 for Petal. PC2, whereas these QTLs all showed an uncorrelated allelic effects distribution (Supplementary Table 3). For the fourth and seventh leaves, except for the overlapping ER locus (Leaf4.PC2: LF4_2.1, Leaf7.PC2: LF7_2.1) for PC2 described above, the QTLs LF4_1.3 and LF4_1.4 for Leaf4.PC1 overlapped with QTL LF7_1.6 for Leaf7.PC1 on chromosome 3 and showed the same allelic effects distribution with a maximum value in the Mt-0 accession and a minimum value in the Can-0 accession (Supplementary Table 3). The overlapping QTLs might have explained the phenotypic correlation and indicated the correlated genetic modules for leaf and petal allometry in evolution.
Candidate genes for leaf and petal allometry
The genes that explain natural variations in leaf and petal allometry have remained largely unknown. To identify possible candidate genes, we searched for genes containing nonsynonymous SNPs unique to accession according to PC distribution among these accession alleles (Supplementary Table 3). Based on the resequencing and reannotation of the 19 parental accessions (Gan et al., 2011), we identified candidate genes with unique alleles referring to the maximal effects accession in the 95% confidence region (Supplementary Table 4). In the Leaf4 allometry model, the auxin receptor TIR1, brassinolide signalling regulator BSL3, and TIR1, contributing to flowering time repression, had allelic variations in the coding sequence unique to the accession. In the Leaf7 allometry model, hormonal-related genes, such as SUA (a suppressor of abi3-5), ARGOS, serine/threonine-protein kinase PID2, BRI1 suppressor 1 (BSU1)-like 3, and ABI4 genes, had allelic variations in the coding sequence. Moreover, the flower time regulators ELF3 and ELF4, the receptor kinase ERECTA, cell wall modification-related genes and some transcription factors conferred allelic variations unique to the maximal effects accession.
In the petal allometry model, 23 genes were identified with variations unique to the accession. Among these genes, PTL in Petal.PC2 encodes a trihelix transcription factor whose expression is limited to the margins of floral and vegetative organs. It is involved in limiting lateral growth of organs, and recessive mutations have been found to be defective in organ initiation and orientation in the second whorl (Kaplan-Levy et al., 2014). The OFP13 in Petal.PC2 encodes a member of the plant-specific OVATE family of proteins. Members of this family have been shown to bind to KNOX and BELL-like TALE class homeodomain proteins and function as transcriptional repressors that suppress cell elongation (Wang et al., 2011). The SEU in Petal.PC1 encodes a transcriptional coregulator that coordinates with LEUNIG to regulate petal shape by controlling blade cell number and vasculature development within the petal (Franks et al., 2006). Other genes, including the cell cyclin-related protein Cyclin A1;1, the protein kinase, the CYP family protein, the photoperiod-associated ELF6, and the transcription-related genes with nonsynonymous SNPs also contribute to petal PCs. The identified QTLs and candidate genes provided us with a valuable reference for insight into leaf and petal allometry.
The genetic basis for leaf and petal covariation in allometry models
To examine the genetic basis for shape and size covariation between leaves and petals, the leaf and petal modelled data sets obtained above were combined to create Leaf4-Petal and Leaf7-Petal data sets, which allowed overall trends to be identified. To ensure equal weighting of the data from different organs, a constant factor was multiplied to the organ size for all plants as previous study (Feng et al., 2009). The major correlated variations were detected by PCA analysis with both Leaf4-Petal and Leaf7-Petal data sets.
In the Leaf4-Petal model, PC1 accounted for 53.58% of the total variance, representing the negative size covariation between Leaf4 and petals. The higher the PC1 value, the larger the petal size, and the smaller the fourth leaf size. PC2 accounted for 30.26% of the total variance, representing the positive size covariation between the fourth leaves and petals. The higher the PC2 value, the larger the petal and leaf size. PC3 accounted for 5.92% of the total variance representing the positive shape (mainly in width) covariation between the fourth leaves and the petals. The higher the PC3 value, the more rounded the leaves and petals, and the shorter the petioles. PC4 accounted for 3.23% of the total variance representing the negative shape (mainly in width) covariation between the fourth leaves and the petals. The higher the value, the narrower the leaves, the longer the petioles, and the more rounded the petals were. The other PCs represented only one organ shape or size variance, so they were not considered for further analysis (Figure 3).
After QTL mapping in the MAGIC lines for the Leaf4-Petal model, three significant QTLs for PC1, one significant QTL for PC2, two significant QTLs for PC3, and six significant QTLs for PC4 were identified (Figure 5, Supplementary Table 5, Supplementary Figure 7). In each QTL, the candidate genes containing nonsynonymous SNPs unique to the maximal effects accession in the 95% confidence region were identified (Supplementary Tables 6 and 7). In PC1, there were five genes with the unique maximal effects accession allele, including the cell-proliferation-related genes, such as ARGOS, LOM2, and EXPB5. In PC3, which represented the shape (mainly in width) covariation, four genes were identified: ARGOS, FRS3, BSL3, and extensin proline-rich1. In PC4, representing the negative shape (mainly in width) covariation, there were also four genes containing the unique accession allele. Among these genes, the CYCD2;1 gene acting on the G1 phase of the cell cycle to control the cell division rate in both the shoot and root meristems had an allele unique to the Hi-0 accession, and the PRX53 gene influencing cell elongation had an allele unique to the Po-0 accession.
Similar to the Leaf4-Petal model, in the Leaf7-Petal model, PC1 accounted for 68.58% of the total variance, representing the negative size covariation between the seventh leaves and petals, whereas PC2 accounted for 22.51% of the total variance, representing the positive size covariation between the seventh leaves and the petals and the seventh leaf shape variance. PC3 accounted for 2.84% of the total variance, representing the positive shape (mainly in width) covariation, and PC4 accounted for 1.99% of the total variance, representing the negative shape (mainly in width) covariation. The other PCs represented only one organ shape or size variance, so they were not considered for further analysis (Figure 4).
After QTL mapping in the MAGIC lines for the Leaf7-Petal model, two significant QTLs for PC3 and six significant QTLs for PC4 were identified, whereas no significant QTL was identified in PC1 and PC2 (Figure 5, Supplementary Table 5, Supplementary Figure 8). Moreover, candidate genes were also identified (Supplementary Tables 6 and 7). The QTL LF7PE_3.2 in PC3, which represented the positive shape (mainly in width) covariation between the seventh leaves and petals, had the most rounded leaves and petals in Ler-0 and the narrowest leaves and petals in the No-0 accession. In the 95% confidence region, there were 34 genes conferring alleles unique to the Ler-0 or No-0 accession. Among these genes, the GRF gene AT2G22840, pentatricopeptide repeat protein SLOW GROWTH1 (SLO1), ORGAN BOUNDARY1 (OBO1) and OVATE family of protein OFP16 have been reported to affect organ shape or size (Kim et al., 2003; Sung et al., 2010; Cho et al., 2011; Wang et al., 2011). Furthermore, the cyclin-dependent kinase inhibitor KRP4 (Schiessl et al., 2014) and the serine/threonine-protein kinase PINOID (PID) are involved in the regulation of auxin signalling (Saini et al., 2017). Growth-regulating factor 3 (GRF3), which regulates cell expansion in leaf and cotyledon tissues (Kim et al., 2003), as well as other genes associated with cell differentiation, cell expansion, cell wall modification, and flower time control genes, were also identified. The QTL LF7PE_4.4 in PC4, which represented the negative shape (mainly in width) covariation, had the narrowest leaves with the longest petioles and the most rounded petals in the Po-0 accession. There were three genes with alleles unique to the Po-0 accession, including DME, a transcriptional activator involved in gene imprinting; peroxidase 2, which influences cell elongation (Jin et al., 2011); and CYP712A2, a member of CYP712A.