Genetic Dissection of Leaf Development in Brassica rapa Using a Genetical Genomics Approach^,

Genes affecting leaf size and shape are identified by combining gene expression and phenotypic trait data.

Abstract

The paleohexaploid crop Brassica rapa harbors an enormous reservoir of morphological variation, encompassing leafy vegetables, vegetable and fodder turnips (Brassica rapa, ssp. campestris), and oil crops, with different crops having very different leaf morphologies. In the triplicated B. rapa genome, many genes have multiple paralogs that may be regulated differentially and contribute to phenotypic variation. Using a genetical genomics approach, phenotypic data from a segregating doubled haploid population derived from a cross between cultivar Yellow sarson (oil type) and cultivar Pak choi (vegetable type) were used to identify loci controlling leaf development. Twenty-five colocalized phenotypic quantitative trait loci (QTLs) contributing to natural variation for leaf morphological traits, leaf number, plant architecture, and flowering time were identified. Genetic analysis showed that four colocalized phenotypic QTLs colocalized with flowering time and leaf trait candidate genes, with their cis-expression QTLs and cis- or trans-expression QTLs for homologs of genes playing a role in leaf development in Arabidopsis (Arabidopsis thaliana). The leaf gene BRASSICA RAPA KIP-RELATED PROTEIN2_A03 colocalized with QTLs for leaf shape and plant height; BRASSICA RAPA ERECTA_A09 colocalized with QTLs for leaf color and leaf shape; BRASSICA RAPA LONGIFOLIA1_A10 colocalized with QTLs for leaf size, leaf color, plant branching, and flowering time; while the major flowering time gene, BRASSICA RAPA FLOWERING LOCUS C_A02, colocalized with QTLs explaining variation in flowering time, plant architectural traits, and leaf size. Colocalization of these QTLs points to pleiotropic regulation of leaf development and plant architectural traits in B. rapa.

Six Brassica species are cultivated worldwide: three diploid Brassica species, Brassica rapa (A genome; n = 10), Brassica nigra (B genome; n = 8), and Brassica oleracea (C genome; n = 9), and three amphidiploids, Brassica juncea (AB; n = 18), Brassica napus (AC; n = 19), and Brassica carinata (BC; n = 17), derived by spontaneous hybridization among the three diploid species (Nagaharu, 1935). All diploid Brassica species have undergone a whole-genome triplication since their divergence from Arabidopsis (Arabidopsis thaliana; Lysak et al., 2005; Town et al., 2006). The B. rapa genome sequence became available in 2011, which permitted a detailed analysis of gene fate after multiple genome duplications. The 41,174 protein-coding genes in the B. rapa genome are fewer than simple triplication of the 25,498 (125 Mb) genes in the Arabidopsis genome due to gene loss (fractionation) after triploidization (Wang et al., 2011). Gene loss was certainly not random. For example, circadian clock genes and flowering genes were preferentially retained in the evolution in B. rapa (Lou et al., 2012; Tang et al., 2012; Xiao et al., 2013).

B. rapa displays extreme morphological diversity, including leafy vegetables, turnips, and oil types, which is likely the result of both genetic and epigenetic variation, selected by plant breeders (Zhao et al., 2005; Bonnema et al., 2011). Variations in leaf shape, color, size, angle, and numbers affect productivity but also attractiveness, as many B. rapa morphotypes are consumed as vegetables. This variation in leaves is enormous, with, among others, smooth dark green leaves with enlarged white midribs of cv Pak choi, the numerous small dark green smooth round leaves of Wutacai, the knotted leaf surface of the large light green leaves with wide midribs from Chinese cabbage (B. rapa ssp. pekinensis), the many elongated slender leaves or the highly serrated leaves of Mizuna (B. rapa ssp. nipposinica), and the oval or serrated leaves of different turnip types. Dissection of genetic and epigenetic variation will increase our understanding of this phenotypic variation and provide molecular tools to breed for B. rapa vegetable crops with novel leaf characteristics. In Arabidopsis, studies have documented that early leaf development relies on the control of leaf initiation and formation on the flanking regions of the shoot apical meristem (Kim and Cho, 2006; Cha et al., 2007). Leaf size and shape in later development involves the coordination between cell proliferation and polar cell division and expansion (e.g. adaxial-abaxial polarity, proximal-distal polarity, symmetry, and flat morphology; Kim and Cho, 2006; Tsukaya, 2006; Barkoulas et al., 2007; Micol, 2009; Gonzalez et al., 2010). In addition, the plant hormones belonging to strigolactone, auxin, cytokinin, and GA influence leaf development and morphogenesis (Bertoni, 2010; Beveridge and Kyozuka, 2010; Mauriat et al., 2011).

B. rapa has a triplicated genome, and as a result, several genes have multiple paralogs. Duplicated genes are of major importance for evolutionary novelty, since they can contribute to functional innovation by mutation of their coding sequences, expression divergence, and rewiring regulatory networks through variation in interactions among different orthologs (Gaeta et al., 2007; Liu and Adams, 2007; De Smet and Van de Peer, 2012). Several studies have correlated variation in gene expression or the fate of duplicated genes to phenotypic diversity, such as flowering time (Ft) variation, leaf shape, size, and numbers, pest resistance, and stress tolerance in eukaryotes (Gaeta et al., 2007; Hovav et al., 2008; Feng et al., 2009; Whittle and Krochko, 2009; Combes et al., 2012; Costa et al., 2012; Xiao et al., 2013).

Pleiotropy implies a single gene affecting multiple traits, while polygenic control implies that one trait is controlled by multiple genes. The genetic regulation of leaf architecture has been unraveled through quantitative trait locus (QTL) analyses in Brassica species (Lou et al., 2007; Kubo et al., 2010; Li et al., 2012; Yu et al., 2013). Lan and Paterson (2001) located significant QTLs explaining 45% of the phenotypic variation in lamina length (LL), three of which colocalized with QTLs that affected leaf width (LW) in B. oleracea. In another study, Lou et al. (2007) detected 10 QTLs for leaf traits using three different mapping populations. By synteny analysis of QTL regions of B. rapa with the Arabidopsis genome, Li et al. (2009, 2013) identified the leaf lobe depth and leaf hairiness genes, BRASSICA RAPA GIBBERELLIN 20-OXIDASE 3 (BrGA20OX3) AND BRASSICA RAPA GLABRA1, and Zhang et al. (2009) successfully cloned a BRASSICA RAPA TRANSPARENT TESTA GLABRA1 gene controlling leaf hairiness and seed coat color. Genetical genomics or expression quantitative trait locus (eQTL) mapping entails a QTL analysis of high-throughput transcript expression patterns in a genotyped population. Genetical genomics can be used as a tool to identify candidate genes for phenotypic traits, as the cause of variation in many traits is differences in the expression of specific genes (Alonso-Blanco et al., 2009).

In this study, aiming to understand the underlying genetic architecture of leaf development, a genetical genomics approach (Jansen and Nap, 2001) was used in a doubled haploid (DH) population from a cross between cv Yellow sarson and a cv Pak choi accession. A large number of genetic markers based on Arabidopsis leaf development genes (hereafter, leaf genes) were mapped in silico, and a subset was genetically mapped in the DH population, which allowed us to address whether leaf genes are preferentially retained in the B. rapa genome. Using variation in the expression of transcripts and 17 phenotypic traits combined with linkage mapping, we constructed coregulatory networks. Bayesian network reconstruction was used to identify the best potential regulator for the genetic regulation of these complex traits (Spirtes et al., 2000). Using colocalization analysis of phenotypic QTLs (pQTLs) and eQTLs, we identified the cis-regulated genes BRASSICA RAPA FLOWERING LOCUS C_A02 (BrFLC2_A02), BRASSICA RAPA KIP-RELATED PROTEIN2_A03 (BrKRP2_A03), BRASSICA RAPA ERECTA_A09 (BrER_A09), and BRASSICA RAPA LONGIFOLIA1_A10 (BrLNG1_A10) that colocalized with colocalized phenotypic QTLs (copQTLs) and several trans-eQTLs, suggesting pleiotropic regulation of leaf development in B. rapa.

RESULTS

Genetic Map of DH68, Enriched with Leaf Candidate Genes

A total of 176 B. rapa genes were identified in the Chinese cabbage cv Chiifu-401 genome sequence homologous to 91 Arabidopsis genes with roles in leaf development (Supplemental Table S1). Among them, 33% (30 of 91) Arabidopsis orthologous genes were presented by three or more paralogs in B. rapa, 34.1% (31 of 91) by two B. rapa paralogs, 24.2% (22 of 91) by one B. rapa homolog, and for 8.8% (eight of 91) of the Arabidopsis genes, no homologs in B. rapa were identified. For 80 leaf candidate genes, 145 PCR primer pairs were designed and screened for polymorphisms between the parents. The 79 polymorphic PCR products were profiled over the DH68 mapping population, which resulted in genetic map positions for 60 leaf genes distributed over all 10 linkage groups, corresponding to 42 Arabidopsis orthologs (Supplemental Fig. S1A; Supplemental Table S2). The genetic map (whole length, 1,328 centimorgan [cM]) contains 509 markers (average distance = 2.6 cM; Supplemental Fig. S1B).

The 176 B. rapa genes were mapped in silico using the reference genome. The comparison of the genetic map positions of leaf development genes in DH68 with their in silico predicted map positions indicated that the order of these gene markers in the genetic map was almost identical to that of the physical map. Only four had inconsistent order between both maps but mapped to the same linkage groups in all cases (Supplemental Fig. S2). The relationship between the physical and genetic distances in this euchromatic sequenced part of the genome was 1 cM = approximately 210 kb.

Phenotypic Variation

To identify the genetic loci responsible for the variation in leaf development, we conducted experiments in 2008 and 2010 (Table I). The YS-143 parent plant had light leaf color and lobed leaves, with on average 5.5 leaf lobes (LB), while PC-175 had entire dark green leaves with white midribs (Fig. 1; Supplemental Fig. S3). Data from 2008 and 2010 were analyzed separately, as the experimental design differed. For 2008 data, ANOVA showed that, for most traits, there were no significant DH line × block interactions at P < 0.001 (the only exceptions were the traits LB and blade length [BL]; Supplemental Table S3). Every trait value was reported as the average from three blocks in this study. A total of 17 phenotypic traits, including plant architecture, leaf traits, and Ft, were evaluated in this study, and most traits (except for leaf wing depth [LD]) were normally distributed and showed transgressive segregation, indicating that the leaf traits are controlled by multiple genes in the mapping population (Supplemental Fig. S4, A–S). Six traits that were phenotyped in both 2008 and 2010 were analyzed statistically, and significant effects of genotype, year, and genotype × year were identified (Supplemental Table S4). In 2008, LL ranged within the DH population from 66 to 330 mm (5-fold), while LW ranged from 57 to 170 mm (3-fold; Supplemental Table S5). The LB number ranged from 1.5 to 13.7. In addition, the variation of leaf size in time was further characterized in 2010 by measuring leaf size parameters (BL, LL, LW, and leaf index [LI]) every 3 to 5 d during development until the DH lines flowered. As shown in Supplemental Figure S4, T to W, both parents and the DH population showed a similar temporal pattern of growth as measured by LL, LW, and BL. LL increased until stage VIII (50 d), after which growth ceased, while for LW, at stage VI the growth of YS-143 leaves ceased and for PC-175 leaves growth ceased at stage VIII; thus, they became wider than YS-143 leaves (Supplemental Fig. S4, T and U).

Table I. Description of B. rapa morphological traits measured in this study.

SPAD, Special Products Analysis Division.

Trait Name	Trait Code	Trait Description	Stagea	Unit	Replication
LN	LN_2008	Counted the LN	42 d after sowing	No.	Three blocks
	LN_2010	Counted the LN	When first open flower	No.	One block
LC	LC_2008	Chlorophyll level of the leaves, measured with SPAD meter	42 d (fifth leaf)	SPAD	Three blocks
	LC_2010	Chlorophyll level of the leaves, measured with SPAD meter	29 d (second leaf)	SPAD	One block
Pmh	Pmh_2008	Assumed plants do not form extra branches anymore	Measured when seed pods are mature	cm	Three blocks
PB	PB_2008	Assumed plants do not form extra branches anymore	Measured when seed pods are mature	No.	Three blocks
Ft	Ft_2008	Ft	When first open flower	Score	Three blocks
LL	LL_2008	Length from base of petiole to tip of lamina	49 d (third leaf)	mm	Three blocks
	LL_2010	Length from base of petiole to tip of lamina	I to X (third leaf)	mm	One block
LW	LW_2008	Width of leaves at the widest point	49 d (third leaf)	mm	Three blocks
	LW_2010	Width of leaves at the widest point	I to X (third leaf)	mm	One block
BL	BL_2008	Distance from the tip to the first lobe	49 d (third leaf)	mm	Three blocks
	BL_2010	Distance from the tip to the first lobe	II to X (third leaf)	mm	One block
PL	PL_2008	Calculated by subtracting LL from BL	49 d (third leaf)	mm	Three blocks
LA	LA_2008	Total surface of the leaf area, petiole not included	49 d (third leaf)	mm2	Three blocks
LP	LP_2008	Total length of the perimeter	49 d (third leaf)	mm	Three blocks
LI	LI_2008	Index of the leaf, calculated by dividing LL by LW	49 d (third leaf)	Ratio	Three blocks
	LI_2010	Index of the leaf, calculated by dividing LL by LW	I to X (third leaf)	Ratio	One block
LB	LB_2008	No. of LB	49 d (third leaf)	No.	Three blocks
LBb	LBb_2008	No. of LBb	49 d (third leaf)	No.	Three blocks
LBs	LBs_2008	No. of LBs	49 d (third leaf)	No.	Three blocks
LD	LD_2008	Depth of the leaf wings, classified in three classes	49 d (third leaf)	Scale	Three blocks
Lcu	Lcu_2008	Curling of the leaves, classified in seven classes	49 d (third leaf)	Scale	Three blocks

^aNumber of days was counted from sowing to day of measure. Dynamic measure includes 10 stages: I (25 d); II (29 d); III (32 d); IV (36 d); V (39 d); VI (43 d); VII (46 d); VIII (50 d); IX (53 d), and X (57 d).

Figure 1.

Leaf morphology of parental genotypes cv Yellow sarson (YS-143) and cv Pak choi (PC-175) of B. rapa. The different traits related to leaf morphology measured in this study, and parental lines at the 10th stage, are shown. A list of all traits with their descriptions is presented in Table I. In Supplemental Figure S3, both parent genotypes and their DH progeny are shown at the first stage (25 d after sowing) and the 10th stage (57 d after sowing).

Phenotype QTL Analysis

To detect the relationship between genotype and phenotypic traits, pQTL analysis was performed. A total of 167 pQTLs were identified for 17 traits including the temporal phases for LL, LW, BL, and LI in 2010, ranging from 36 pQTLs for LI located on seven linkage groups to only one pQTL for LD and leaf curling (Lcu) on A02 (Fig. 2A; Supplemental Table S6). The 167 pQTLs were distributed over all linkage groups, and 25 or more pQTLs were located on each of the linkage groups A01, A02, A03, A09, and A10. Individual pQTLs explained between 6.2% and 38.9% of the phenotypic variation, with six pQTLs explaining less than 10% and five pQTLs explaining more than 30% (Supplemental Table S6). In Table II, the total phenotypic variation explained by the QTL for each trait is summarized under different growth conditions and stages, and this total explained variation ranged from 12.3% (LL in stage V in 2010) to 95.2% (leaf color [LC] in 2008) of the phenotypic variance. The distribution of pQTLs shows that different loci affected leaf development in the various years and development stages. For example, two pQTLs were detected for LL in 2008 (A01, 89.4 cM, marker BrGRF5P1b; A02, 141.1 cM, marker BrMAF4P1d). Also in 2010, LL was phenotyped 10 times during plant development, resulting in the identification of 25 pQTLs, with a total of one to four pQTLs detected at each developmental stage. pQTLs at A01 were detected both in 2008 and 2010 (stages I–III), while the A02 pQTL was only detected in 2008 (Table II; Supplemental Table S6).

Figure 2.

Profile of pQTLs and eQTLs mapped in DH68 of B. rapa. A, Distribution of pQTLs for leaf morphology-related traits measured in 2008 and 2010. The color on the left indicates the two experiments conducted in 2008 and 2010, whereas the color on the x axis on the top indicates the 10 different linkage groups (A01–A10). The color of the pQTL profile represents the different levels of LOD score. B, Distribution of eQTLs for leaf trait-related candidate genes showing their cis-/trans-regulation. The y axis indicates the physical positions of genes/probes, and the x axis indicates the genetic positions of markers in the 10 linkage groups. The color bars on the left and on the top indicate the 10 linkage groups. The color of each eQTL reflects the different levels of LOD score. Gene names highlighted in the figure indicate colocation with pQTLs, or high-LOD eQTLs, or genes associated with traits in network analyses. C, Location of markers of candidate genes for leaf and Ft traits on the genetic map of the DH68 population of a cross from cv Yellow sarson and cv Pak choi. Marker names in black indicate the genes mapped in DH68, and those in red indicate genes mapped in silico based on the reference genome (Chinese cabbage cv Chiifu-401). The shape and color of the boxes under candidate gene markers indicate cis- or trans-regulation and different functional pathways, respectively. Triangles indicate cis-regulation, squares indicate trans-regulation, and diamonds indicate cis- and trans-regulated genes. For detailed functional pathway descriptions, see Figure 4. The data on the regulation of the Ft genes cis-BRASSICA RAPA MADS AFFECTING FLOWERING4 (BrMAF4) and trans-BRASSICA RAPA PHYTOCHROME A are according to our previous study (Xiao et al., 2013).

Table II. Total percentage of phenotypic variance explained by the additive effects of all detected pQTLs for each trait in B. rapa.

Trait Involved	2008 No. of QTLsa	2010 No. of Stagesb,d	2010 No. of QTLsc,d	Total Variation Explainede
				%
LN	4	1	4	53.0–84.2
LC	6	1	5	88.4–95.2
LL	2	10	1–4	12.3–59.1
LW	3	9	2–5	21.8–70.4
BL	3	7	1–5	19.8–75.7
PL	2	–	–	19.7
LA	2	–	–	23.5
LP	3	–	–	31.3
LI	2	10	2–5	20.4–79.0
LB	3	–	–	30.5
LBb	5	–	–	68.6
LBs	3	–	–	38.4
LD	1	–	–	16
Lcu	1	–	–	12.3
Pmh	3	–	–	48.8
PB	4	–	–	56.6
Ft	4	–	–	80.2

^aIn 2008, each trait was only measured once. The number of pQTLs is listed.  ^bIn 2010, several traits were measured at a number of stages (Table I).  ^cThe range (minimum to maximum) of pQTLs is listed in 2010.  ^d− means that these traits were not measured in this experiment. ^eRange of total variation explained per trait per year and/or measurement at different stages (for details, see Supplemental Table S6).

Identification of copQTL and Colocalization of copQTL with Candidate Genes

We analyzed whether pQTLs for multiple phenotypic traits in different years and growth stages colocalized, and based on that we defined copQTLs (Table III). In total, 25 copQTL regions that integrated 144 initial pQTLs were identified, and 34 Ft and leaf trait genes were inferred on seven chromosomes. These copQTLs were identified for all traits except Lcu (Table III). In total, 20 out of these 34 candidate genes were genetically mapped in DH68, while the genetic map positions of the other 14 genes were estimated based on their physical positions relative to markers with both physical position in the reference genome and genetic map position in DH68 (Supplemental Table S7). copQTL23 combined QTLs for eight traits, leaf size (petiole length [PL], LI, LL, LW, and BL), LC, plant branching (PB), and Ft, integrating 22 initial pQTLs (Supplemental Table S8), and colocalized with BrLNG1 at 73.9 cM on A10, explaining 9.7% to 18.4% of the phenotypic variation for each trait. copQTL6 (21.1–36.9 cM) combined QTLs for seven traits, leaf size (LL, LW, BL, and LI), plant architecture traits (leaf number [LN] and PB), and Ft, integrating 15 initial pQTLs and four peak pQTLs overlapping (conflated 95% confidence interval) with BrFLC2, BrGA20OX3, BRASSICA RAPA ASYMMETRIC LEAVES ENHANCER3 (BrAE3), and BrLNG1 (4.9–44.7 cM), explaining 10.7% to 38.9% phenotypic variation for each trait. Four copQTL combined four traits each: copQTL2 for leaf size (LL, PL, and LI) and plant mature height (Pmh) mapped to BRASSICA RAPA GROWTH-REGULATING FACTOR5 (BrGRF5) at 89.4 cM at A01; copQTL12 for leaf size (LL, LW, and BL) and LN mapped to BrFLC5 at 9.2 cM at A03 and indicated the nearest leaf candidate gene BrAE3; copQTL14 for leaf size (LL, LW, BL, and LI) mapped to BRASSICA RAPA ASYMMETRIC LEAVES1 (BrAS1) at 55.2 cM on A03; and copQTL15 for leaf size (LW and LI), small leaf lobes (LBs), and Pmh mapped to BrKRP2 at 139.5 cM on A03. Similarly, five copQTLs for three traits colocalized with BRASSICA RAPA HASTY1 (BrHST1) on A01, BrGA20OX3 on A02, BRASSICA RAPA LEAFY PETIOLE on A03, BRASSICA RAPA FERREDOXIN-NADP(+)-OXIDOREDUCTASE1 on A09, and BRASSICA RAPA SHOOTMERISTEMLESS, BRASSICA RAPA HUA ENHANCER1 and BRASSICA RAPA PINHEAD (BrPNH) on another locus on A09. In addition, eight copQTLs for two traits and six copQTLs for one trait colocalized within a total of 22 leaf and Ft candidate genes (Table III). Interestingly, some sets of copQTLs colocalized with sets of paralogs: copQTL3 for leaf size (LI) and copQTL17 for leaf size (BL and LI) colocalized with two different paralogs of BRASSICA RAPA SQUAMOSA PROMOTER BINDING PROTEIN-LIKE5 (BrSPL5) on A01 and A05, respectively; while copQTL4 for leaf size (LL, LW, and BL) and copQTL18 for Lcu colocalized with two paralogs of BrHST1 on A01 and A05, respectively.

Table III. Details of copQTLs of the 17 phenotypic traits, with genetic positions of B. rapa genes for leaf traits and Ft that colocate with these copQTLs.

Traits Involved	copQTLa,b	LG	Intervalc	No. of pQTLs	Nearest Candidate Gene	Peak pQTL	Distance between Peak LOD and Candidate Gened	Grounde	Associated Functional Pathway
			cM			cM	cM
LW	copQTL1	A01	67.9–86.6	3	BrKAN2	83.6	2.6	Mapped	Adaxial-abaxial polarity
LL, PL, LI, Pmh	copQTL2	A01	89.4–92.0	9	BrGRF5	89.4	0	Mapped	Leaf shape
LI	copQTL3	A01	97.4–105.9	7	BrSPL5	97.4	0	Mapped	Ft
LL, LW, BL	copQTL4	A01	113.7–123.9	5	BrHST1	122.7	0	Mapped	Adaxial-abaxial polarity
LI	copQTL5*	A02	4.9–9	2	BrBFT	9	0	Mapped	Ft
LN, Ft, PB, LL, LW, BL, LI	copQTL6*	A02	21.1–36.9	15	BrFLC2	24.1	4.7	Mapped	Ft
LL, LW, BL	copQTL7*	A02	14.28–30.82	6	BrGA20OX3	28.8	–	Inferred	Leaf shape
LL, LW	copQTL8*	A02	24.2–48.2	4	BrLNG1	35	–	Inferred	LL
LBb, LD	copQTL9	A02	32–51.6	2	BRASSICA RAPA PISTILLATA	48.2	0	Mapped	Ft
					BrKAN1	48.2	–	Inferred	Adaxial-abaxial polarity
					BrREV	48.2	–	Inferred	Adaxial-abaxial polarity
LL, LA	copQTL10	A02	79.2–85.6	4	BRASSICA RAPA PINFORMED1	82.6	1.6	Mapped	Leaf shape
LL, BL	copQTL11	A02	139.6–148.2	2	BRASSICA RAPA MADS AFFECTING FLOWERING4	141	1.4	Mapped	Ft
					BrKAN2	141	–	Inferred	Adaxial-abaxial polarity
LN, LL, LW, BL	copQTL12	A03	2–14.3	6	BrFLC5	9.2	0	Mapped	Ft
					BrAE3	9.2	–	Inferred	Others
LN, LW, Ft	copQTL13	A03	14.3–29.2	3	BRASSICA RAPA LEAFY PETIOLE	22.4	–	Inferred	Others
LL, LW, BL, LI	copQTL14	A03	49.6–67.5	16	BrAS1	55.2	10.4	Mapped	Symmetry
LBs, Pmh, LW, LI	copQTL15	A03	110.3–137.3	5	BrKRP2	114.2	25.3	Mapped	LL
LW, LP	copQTL16	A04	39.9–63.4	2	BRASSICA RAPA PHYTOCHROME A	58.5	0	Mapped	Ft
BL, LI	copQTL17	A05	111.3–117.8	3	BrSPL5	117.8	0	Mapped	Ft
LC	copQTL18	A05	125.9–175.9	2	BrHST1	158	0	Mapped	Adaxial-abaxial polarity
LN, Pmh, PB	copQTL19	A09	37.2–46.4	3	BRASSICA RAPA FERREDOXIN-NADP(+)-OXIDOREDUCTASE1	41.3	0	Mapped	LW
BL, LI	copQTL20	A09	56.2–65.4	2	BrRON1	61.6	–	Inferred	LL
					BrWUS	61.6	–	Inferred	Meristems
LW, BL, LI	copQTL21	A09	62.9–70.1	15	BRASSICA RAPA SHOOTMERISTEMLESS	70.1	–	Inferred	Meristems
					BRASSICA RAPA HUA ENHANCER1	70.1	–	Inferred	Others
					BrPNH	70.1	–	Inferred	Adaxial-abaxial polarity
LC, LB	copQTL22	A09	114.7–126.5	2	BrER	121.8	2.1	Mapped	Others
PL, LI, LC, LL, LW, BL, PB, Ft	copQTL23	A10	65.9–82.1	22	BrLNG1	73.9	0	Mapped	LL
LI	copQTL24	A10	63.4–96.9	2	BRASSICA RAPA CONSTANS-LIKE1	84.9	0	Mapped	Ft
					BrKAN1	84.9	–	Inferred	Adaxial-abaxial polarity
LI	copQTL25	A10	84.9–111.1	2	BrAE3	96.9	–	Inferred	Others
					BRASSICA RAPA HYPONASTIC LEAVES1	96.9	6.9	Mapped	Others

^aIncluded initial pQTLs (Supplemental Table S6).  ^bAsterisks mean that these copQTLs largely overlap and may represent a single QTL.  ^cA 1.0 LOD 95% confidence interval.  ^d− means that the distance between the candidate gene and the LOD peak cannot be calculated, as the position of the candidate gene is not mapped in DH68, but inferred based on their physical positions and those of neighboring genes with genetic map positions (Supplemental Table S7).  ^eGenes were either mapped in DH68 or their inferred positions were based on their physical positions and those of neighboring genes with genetic map positions (Supplemental Table S7).

eQTL Analysis

To identify the molecular mechanisms underlying the pQTL/copQTL, the expression profiles of leaves from 5-week-old DH68 lines were determined by microarray and quantitative real-time PCR (RT-qPCR). Of all 96,557 probes on the microarray, 96 probes represented 41 Arabidopsis leaf development genes corresponding to 64 B. rapa paralogs (Supplemental Table S9). The 96 probes represented 36 B. rapa genes with eQTLs, while the transcripts of the other 138 genes from the selected 184 B. rapa genes were quantified using RT-qPCR. For seven genes, no PCR product was obtained (due to bad primer quality). Of three genes (BRASSICA RAPA CYCLIN D1;1_A02 [BrCYCLIN D1;1_A02], BrKRP2_A03, and BRASSICA RAPA DEFORMED ROOT AND LEAVES1_A08 [BrDRL1_A08]) with significant cis-eQTLs detected by microarray, their expression was validated by RT-qPCR, illustrating the reproducibility of the microarray results (Supplemental Fig. S5; Supplemental Table S10). Based on both microarray and RT-qPCR transcriptional analyses, a total of 95 B. rapa genes (53.7%, 177) had variation in expression in the DH68 population (fold exchange ranged from 0.3 to 15.3); 47 genes (26.6%, 177) were expressed in both parents and the DH68 population, but their expression in the DH68 population was not variable. Thirty-five genes (19.8%, 177) were not expressed in parents and the DH68 progeny (Supplemental Table S1).

Variation in gene transcript abundance in the segregating population was treated as a quantitative trait and subjected to eQTL analysis against 509 markers of the genetic map. Twelve Ft candidate genes associated with Ft as a positive control were added in the analysis (Xiao et al., 2013). A total of 118 probes/genes representing 110 B. rapa genes orthologous to 72 Arabidopsis genes revealed a total of 173 eQTLs (marker probe associations) against all 509 genetic markers, with at least one to five eQTLs per probe (log of the odds [LOD] ≥ 3; Fig. 2B; Supplemental Table S10). The results showed that 39 (33.1%) probes/genes had cis-eQTLs, 59 (50%) probes/genes had trans-eQTLs, while 20 (17%) probes/genes had cis- and trans-eQTLs (Supplemental Table S10). An almost linear genome-wide relation along the diagonal of the graph was observed, with the effect for the cis-eQTL (average LOD = 7.5, median of 6.3) being stronger than for the trans-eQTL (average LOD = 5.1, median of 4.8; Fig. 2B).

Identification of Traits and Gene Modules

The 17 phenotypic traits measured in 2008 were evaluated on the same DH plants that were used for transcriptional analysis. We next grouped the patterns of trait variation and expression variation of individual genes by Spearman correlation and found seven clusters with strong correlation (Supplemental Fig. S6; Supplemental Table S11). For ease of description, we numbered these clusters from block A through block G. Three out of these seven clusters included both phenotypic traits and expression levels of genes (Fig. 3). In cluster A, an LL leaf gene, cis-BrLNG1_A10, and six Ft genes are positively correlated with leaf size traits (LW, leaf area [LA], leaf perimeter [LP], and BL) and plant architectural traits (PB, LN, and Pmh). In cluster B, the trait LBs is positively correlated with the expression of 11 leaf genes. In cluster C, three Ft genes and four leaf genes were positively correlated with Ft. Notably, phenotypic and expression traits in cluster C were significantly negatively correlated with those in cluster A. The other clusters (D–G) only consisted of gene expression levels with significant positive correlation within clusters (Supplemental Fig. S6).

Figure 3.

Portion of the heat map showing the correlations among phenotyped traits and the expression of B. rapa leaf candidate genes. Spearman correlation was calculated among 118 genes and 17 phenotypes to show coexpression patterns of genes using hierarchical clustering. Regions with absolute Spearman correlation of r > 0.3 and P < 0.05 are shown. Blue indicates negative correlation, and red indicates positive correlation. The traits and genes, and their coexpression, are highlighted in the boxes showing the coexpressed genes based on the grouping structure in hierarchical clustering. The details of whole clusters and their correlation values are presented in Supplemental Figure S6 and Supplemental Table S11.

Genetic Regulatory Network

To further visualize the correlation between traits and genes, we constructed a genetic regulatory network based on LOD values using the Spearman correlation (Fig. 4; Supplemental Table S12). LOD values of 17 phenotypes (2008) and 118 probes/genes that are classified into nine functional pathways were used to construct the coexpression network. The results show that 10 out of 17 traits were associated with 22 (35 links) candidate genes/probes belonging to five functional pathways (adaxial-abaxial polarity, four; Ft, eight; LL, five; leaf shape, four; other, one). Two traits for leaf size (LA and BL) were not associated with the expression profiles of genes. Many traits related to flowering, plant architecture, and leaf shape (Ft, LN, PB, leaf big lobes [LBb], and LD) were positively associated with the Ft genes investigated. A subset of these traits (Ft, LN, and LBb) was also positively associated with four adaxial-abaxial polarity genes (BRASSICA RAPA KANADI2 [BrKAN2_A01], BrPNH_A09, BRASSICA RAPA PHAVOLUTA_A09, and BRASSICA RAPA YABBY2_A09). Three traits defining plant architectural traits (PB) and leaf size (LW and LI) were associated with four LL genes (BrCycB2;4_A07, BRASSICA RAPA AUXIN-REGULATED GENE INVOLVED IN ORGAN SIZE_A07 [BrARGOS_A07], BRASSICA RAPA ARGOS-LIKE_A03 [BrARL_A03], and BrCYCB1_A01), while traits related to leaf shape (LB) and Pmh were mainly negatively associated with genes with function in leaf shape in Arabidopsis. The genetic network showed correlation of the 22 candidate genes to phenotypic traits, which suggests a role for these genes in leaf development, plant architecture, and Ft.

Figure 4.

Network visualizing the correlation between leaf morphological traits and B. rapa candidate genes for leaf development. To show the functional relevance of the genes and leaf morphological traits in a coexpression network, Spearman correlation was calculated based on LOD scores including 12 leaf traits and the expression of 92 candidate genes, and only significant absolute correlation at r > 0.3 and P < 0.05 are shown here. Nodes represent genes or leaf traits, and edges represent correlations. The genes are arranged in nine functional pathways as shown by different node colors. The colors of the edges represent correlation between genes and/or traits: green for edges between genes and phenotypic traits, magenta for edges between phenotypic traits, gray and faint red for edges between interpathway genes, and dark blue for edges between intrapathway genes. The positive correlations are shown by the solid lines, and the negative correlations are shown by the dotted lines. Shapes of the nodes indicate the cis-/trans-regulation of genes (triangles, cis; squares, trans; diamonds, cis and trans), and green round nodes indicate leaf traits.

Bayesian Network Analysis to Identify Genes Regulating Phenotypes

The relationships among the 17 traits were estimated by the PC (for Peter and Clark) algorithm (Spirtes et al., 2000; Supplemental Fig. S7). This clearly showed three groups, one including plant architectural traits (PB, LN, and Pmh), LC, and Ft, one consisting of traits related to leaf size (LL, PL, BL, LP, LA, LW, and LI), and one relating to leaf shape (LB, LBs, LBb, Lcu, and LD), which we call subnetworks 1, 2, and 3. Based on copQTLs and candidate genes, Spearman correlation of trait variation, and the genetic regulatory network described above (see “Materials and Methods”), we prioritized 62 candidate genes/probes associated with the 17 traits. We applied the PC algorithm with a greater significance level (0.05) to further improve the identification of sets of candidate genes that regulate the phenotypes (Fig. 5), with a significance level at 0.01 (Supplemental Fig. S8). Subnetwork 1 included phenotypic traits related to plant architecture, LC, and Ft plus 43 genes (Fig. 5A), resulting in the association of 20 genes (six functional pathways) with the five traits. For example, LC associated with four different genes: BrER_A09 and BrGRF2_A03, both with roles in leaf chlorophyll in tomato (Solanum lycopersicum) and Arabidopsis (Liu et al., 2012; Seo et al., 2012), while BrKRP2_A09 and BrKAN2_A05 have never been implicated in LC variation so far. Subnetwork 2 includes seven traits that define leaf size and 22 genes (Fig. 5B). In this subnetwork, 15 genes (four functional pathways) were associated with the seven traits. For example, LW was associated with three LL pathway genes, BrARGOS_A09, BrARL_A03, and BrKRP2_A03.2. Among of them, BrARGOS_A09 and BrARL_A03 influence cell size and are up-regulated by brassinosteroids (Hu et al., 2006; Feng et al., 2011), while another report describes that auxin acts via the ARGOS protein to regulate leaf size (Hay et al., 2004). BrKRP2_A03.2 was involved in the regulation of cell proliferation and postmitotic cell expansion to control organ size in Arabidopsis (Kawade et al., 2010). The other six traits in this subnetwork are associated with genes from four functional pathways, including five genes involved in LL, three genes involved with adaxial-abaxial polarity, one symmetry gene BRASSICA RAPA GENERAL TRANSCRIPTION FACTOR GROUP E6_A03 (BrGTE6_A03), and two “other” pathway genes (BrAE3_A02 and BRASSICA RAPA CORONA_A07). Subnetwork 3 includes five leaf shape traits and 24 candidate genes (Fig. 5C). Fifteen genes (eight functional pathways) were associated with the five traits. For example, leaf shape (LBb) was associated with the adaxial-abaxial polarity gene BRASSICA RAPA YABBY2_A09, the LW gene BRASSICA RAPA ANGUSITFOLIA3_A09 (BrAN3_A09), the Ft gene BrFLC5_A03, and BRASSICA RAPA RNA-DEPENDENT RNA POLY MERASE6_A01 (BrRDR6_A01) involved in yet another functional pathway (e.g. leaf adaxial polarity and Lcu; Yuan et al., 2010; Kojima et al., 2011). Genes identified through Bayesian network analysis are promising candidates regulating variation in leaf development of B. rapa.

Figure 5.

Three networks for enriched sets of prioritized potential candidate genes for leaf development based on the Bayesian network (PC algorithm with significance level of 0.05) in B. rapa. A, Subnetwork 1 was constructed based on plant architecture traits, LC, and Ft. B, Subnetwork 2 was constructed based on leaf size traits. C, Subnetwork 3 was constructed based on leaf shape traits. For explanations of shape and color descriptions of the nodes, see Figure 4. Red lines are the associations between phenotypic traits and genes; blue lines indicate associations between genes, with at least one gene associated with a phenotype; magenta lines are associations between phenotypic traits. Edge line width indicates the correlation value between corresponding nodes representing the strength of correlation. Solid lines indicate positive correlations, and dotted lines indicate negative correlations.

Identification of Putative Genes Regulating Trait Variation

We combined the results of copQTLs, correlation networks, and genes predicted by Bayesian networks to search for candidate genes for leaf variation. We describe six loci where copQTLs, high LOD cis-eQTLs, and trans-eQTLs colocate; for four of these loci (BrFLC2_A02, BrKRP2_A03, BrER_A09, and BrLNG1_A10), the candidate gene at the peak of the QTL has a high LOD cis-eQTL (Table IV). copQTL6, copQTL15, copQTL22, and copQTL23 mapped to the candidate genes BrFLC2, BrKRP2, BrER, and BrLNG1, respectively. BrFLC2, a MADS domain transcription factor that has a major role in repressing Ft, was cis-regulated at 21.1 to 36.9 cM on A02 (LOD = 24.2). It is important to note that the copQTL6 for Ft, leaf size (LL, LW, BL, and LI), and plant architectural traits (LN and PB) and the cis-eQTL for BrFLC2 colocalize with trans-eQTLs for BRASSICA RAPA WUSCHEL_A06 (BrWUS_A06) and BrKAN2_A09. The BrKRP2 cis-eQTL is located on chromosome A03 at 110.3 to 137.3 cM, with a peak LOD = 25.7, colocalizing with copQTL15 for leaf size (LW and LI), leaf shape (LBs), and plant architectural traits (Pmh). The BrER cis-eQTL is located on chromosome A09 at 99.4 to 133.7 cM (LOD = 12.2), colocalizing with copQTL22-involved LC and leaf shape (LB) traits. At these two loci (BrKRP2 and BrER), additional cis- and trans-eQTLs colocalized, as summarized in Table IV. The BrLNG1 cis-eQTL on chromosome A10 at 65.9 to 84.9 cM (LOD = 15.1) colocalizes with copQTL23 for leaf size (PL, LI, LL, LW, and BL), PB, LC, and Ft and both BRASSICA RAPA CONSTANS_A10 (BrCO_A10) and trans-BrKAN2_A01 eQTLs. Additionally, copQTL14 (which contains 16 initial pQTLs for leaf size) maps on A03 at 49.6 to 67.5 cM, colocalizing with four cis-eQTLs for BrLNG2, BRASSICA RAPA SPLAYED (BrSYD), BrARL, and BRASSICA RAPA ENHANCER OF AG-4 2 (BrHUA2) and five trans-eQTLs for BrAE3_A02, BrHUA2_A02, BrARGOS_A07, BrDRL1_A09, and BrARGOS_A09. A leaf symmetry functional pathway gene, BrAS1, was mapped within the region. The BRASSICA RAPA ROTUNDIFOLIA1 (BrROT1) cis-eQTL mapped to A05 at 17.9 to 66.1 cM (LOD = 7.1), colocalizing with trans-eQTLs for BrRDR6_A01, BrFLC5_A03, and BrAN3_A09 and partly with two leaf shape pQTLs for LBb and LBs.

Table IV. Identification of putative target loci by colocalization of pQTLs and eQTLs in B. rapa.

Peak eQTL Marker (Chromosome: cM)	Candidate Gene Regulated	LOD	Percentage Variance Attributable to eQTLs	cMa	Colocated pQTL Trait	Pb
BrFLC2 (A02: 24.1)	Cis-BrFLC2_A02	24.2	71.4	21.1–36.9	copQTL6	0.011
	Cis-BrTCP11_A02	11.6	45.6
	Trans-BrFT_A07.1	8.2	35.1
	Trans-BrFT_A07.2	3.4	16.6
	Trans-BrSOC1_A05	5.0	23.5
	Trans-BrSOC1_A03	4.9	23.2
	Trans-BrCO_A10	5.5	25.3
	Trans-BrWUS_A06	3.0	15.8
	Trans-BrKAN2_A09	3.0	17.9
BrAS1 (A03: 65.6)	Cis-BrLNG2_A03	4.5	21.4	49.6–67.5	copQTL14	0.002
	Cis-BrSYD_A03	3.2	15.7
	Cis-BrARL_A03	5.0	23.7
	Cis-BrHUA2_A03	4.2	20.1
	Trans-BrAE3_A02	4.5	21.6
	Trans-BrHUA2_A02	6.6	29.8
	Trans-BrARGOS_A07	3.0	15.1
	Trans-BrDRL1_A09	3.4	16.9
	Trans-BrARGOS_A09	3.0	15.1
BrKRP2 (A03: 132.6)	Cis-BrKRP2_A03	25.7	74.8	108.1–139.5	copQTL15	0.008
	Cis-BrGTE6_A03	7.8	32.6
	Trans-BrCDC2_A01.2	3.4	16.2
	Trans-BrPHB_A04	4.3	19.8
	Trans-BrCDC2_A06.2	3.2	16.0
	Trans-BrCycD3;2_A07	3.7	18.1
	Trans-BrGA20OX3_A10	4.0	19.5
BrMS-034 (A05: 41.1)	Cis-BrROT1_A05	7.1	31.7	17.9–66.1	2008-LBb, 2008-LBs	<0.001
	Trans-BrRDR6_A01	3.2	15.8
	Trans-BrFLC5_A03	4.2	19.9
	Trans-BrAN3_A09	6.6	30.2
BrER (A09: 123.9)	Cis-BrER_A09	12.2	47.9	99.4–133.7	copQTL22	0.002
	Cis-BrARF3_A09	4.1	19.6
	Trans-BrKAN2_A05	3.7	18.4
	Trans-BrFT_A07.1	3.2	15.7
	Trans-BrKRP2_A09	9.7	37.6
BrLNG1 (A10: 73.9)	Cis-BrLNG1_A10	15.1	55.5	65.9–84.9	copQTL23	0.002
	Cis-BrCO_A10	4.5	21.2
	Trans-BrKAN2_A01	3.0	14.9

^aTotal length of the overlapped above the LOD > 3.0 threshold region in cM.  ^bLargest overlap P value between pQTL and eQTL traits.

Role of Paralogs of Leaf Genes in Leaf Development

In order to understand the role of duplicated leaf development genes in B. rapa, we investigated whether paralogs of the same Arabidopsis genes colocalized with similar phenotypic QTLs. From the 118 leaf development genes with eQTLs, 67 (56.8%) had two or more paralogs corresponding to 28 Arabidopsis orthologous genes (Supplemental Table S13). From this set of genes, 15 subsets of paralogs colocalized with pQTLs for the same phenotypic QTLs. For example, the cis-BrGA20OX3_A03 eQTL colocalized with pQTLs for leaf size (LL, LW, and BL), plant architecture (Ft and LN), and leaf shape (LBb), integrating 15 initial pQTLs on A03. The trans-BrGA20OX3_A02 had two eQTL regions: one on A01 where no pQTLs mapped, and the other on A03, colocalizing with the mentioned pQTLs and cis-BrGA20OX3_A03. The trans-BrGA20OX3_A10 colocalized with a single pQTL for leaf shape (LBs) on A03. Four paralogs of BrKAN2, distributed on A01, A02, A05, and A09, each colocalized with major pQTLs for leaf size traits (LL and LW). Three paralogs of BrAE3, distributed on A02, A03, and A10, colocalized with QTLs for LI. The fact that paralogs colocalized with pQTLs for the same traits suggests that they maintained similar functions after genome triplication, while the other 13 subsets of paralogs may have diverged.

DISCUSSION

B. rapa is a species that displays extreme morphological variation, from heading Chinese cabbages, nonheading leafy vegetables, turnips, to oil crops. All these morphotypes differ in leaf number, size, shape, and color, and very little is known of the molecular nature of this variation.

Whole-genome duplication coupled with the retention of duplicated genes was proposed to increase morphological complexity in eukaryotic species (Freeling and Thomas, 2006). Compared with Arabidopsis, B. rapa underwent a genome triplication, and as a result, many genes have paralogs in syntenic blocks, and investigation toward their roles in the regulation of leaf development is important. In this study, 61 out of 91 leaf genes from Arabidopsis have two or more paralogs in the B. rapa genome, while 22 have retained a single copy. Similar to this observation, genes related to flowering were maintained in higher numbers than expected when gene loss was random (Xiao et al., 2013; Supplemental Table S1). According to the “gene balance hypothesis” (Birchler and Veitia, 2010), dosage-sensitive genes duplicated by polyploidy are those with products that function in multisubunit complexes. Our work presents 15 sets of duplicated genes/paralogs with eQTLs colocalizing with pQTLs for the same phenotypic traits, while 13 colocalized with pQTLs for different traits (Supplemental Table S13). For the remaining 39 genes, either one or more paralogs did not colocate with pQTLs. A previous study has demonstrated that duplicated genes may lead to new functions of the paralogs (Blanc and Wolfe, 2004).

The utilization of candidate gene markers and genomics platforms (pQTL and eQTL) in combination with segregating populations results in data to discover the complex genetic mechanism of B. rapa leaf architecture variation. The DH lines, derived from a cross between an oil type, cv Yellow sarson 143, and a leafy vegetable, cv Pak choi 175, were evaluated for 15 plant morphological traits, leaf number, and Ft in 2008 and 2010, and in 2010 at consecutive leaf developmental stages. Significant differences were observed between pQTLs for each trait, their additive effect, and LOD size in different years and growth stages. This indicates genotype × environment × developmental stage effects on leaf development. Several other studies also showed the impact of diverse environments on QTL detection for leaf development in Brassica species and maize (Zea mays; Lou et al., 2007; Li et al., 2009; Ku et al., 2012; Raman et al., 2013). Temporal analysis of leaf growth showed both different and common pQTLs at different growth stages, which accumulated to a large number of loci affecting leaf development (Fig. 2A; Supplemental Tables S6 and S10). As shown in this work, the LL increased until measurement stage VIII, after which growth ceased (Supplemental Fig. S4T). For instance, for LL, pQTLs from the first three stages (I–III) in copQTL2 were mapped to BrGRF5 at 89.4 cM on A01. pQTLs for seven stages (I–VII) in copQTL6 were mapped to BrFLC2 on A02 (21.1–36.9 cM). pQTLs from four later stages (VII–X) in copQTL23 mapped to BrLNG1 at 73.9 on A10 (Table III; Supplemental Tables S6 and S8). Hence, a systematic temporal dissection is necessary to decipher the complex genetic bases of quantitative variation in leaf growth. The temporal patterns of pQTLs for LW and BL suggest two phases in leaf development, with early- and late-acting pQTLs (Fig. 2; Supplemental Fig. S4, U and V). In this paper, we do not address heteroblasty, which refers to changes in leaf shape and size (allometry) along stems (Feng et al., 2009; Costa et al., 2012). Leaf shape and size do vary with increasing leaf number; however, we chose for this study the third leaf, which in shape resembles the next leaves, while the first and second leaves are often asymmetric and smaller.

A total of 34 B. rapa candidate genes colocalized with copQTLs, indicating that these genes may affect multiple leaf architecture traits (Table III). copQTL4 for leaf size (LL, LW, and BL) was detected on A01, and copQTL18 for LC was detected on A05, both colocalizing with different paralogs of BrHST1. Mutants of this gene affect several processes, including leaf polarity, reduction in leaf size, sepals, and petals, uprolling of the leaf blade, reduction in leaf number, and cell growth in Arabidopsis development (Serrano-Cartagena et al., 2000; Bollman et al., 2003). copQTL15 colocalizes with BrKRP2 on A03 and includes five initial pQTLs for leaf size (LW and LI), leaf shape (LBs), and a plant architectural trait (Pmh). Interestingly, at this same position, a QTL for leaf size (LW and LI) was also identified in the reciprocal DH38 population (Lou et al., 2007). In this same study, a QTL for LW in an F2:F3 population (rapid cycling 144 × Chinese cabbage 156) was detected on the bottom of A05, corresponding to the copQTL17 colocalizing with the BrSPL5 gene in our study. In the study of Lou et al. (2007) using the reciprocal DH38 population, another copQTL for LW, LA, and LI was detected on A09, which corresponds with copQTL21 for leaf size (LW, BL, and LI) in our research (Fig. 2A; Table III).

Through the integration of gene expression profiling and the colocalization of copQTL and eQTL, four candidate genes with cis-eQTLs (BrFLC2_A02, BrKRP2_A03, BrER_A09, and BrLNG1_A10) were identified that regulate multiple traits (Table III). In a recent publication, QTL colocalizing with the FLC gene contributed to natural variation in Ft and reduced stem branching genes (RSB) in Arabidopsis (Huang et al., 2013). Their genetic analyses showed that the reduced stem branching QTLs RSB6 and RSB7, corresponding to Ft genes FLC and FRIGIDA, regulate stem branching. They also showed that gene FLOWERING LOCUS T (FT), which corresponds to another reduced stem branching QTL, RSB8, caused pleiotropic effects not only on Ft but, in the specific background of active FRIGIDA and FLC alleles, also on the stem branching trait. In our study, we detected a similar phenomenon, as illustrated by copQTL6 for leaf size (LL, LW, BL, and LI), plant architecture traits (LN and PB), and Ft (Table III). BrFLC2 not only regulates Ft but is associated with variation in plant architectural traits (PB, LN, and Pmh) based on correlation analysis (Fig. 4), Bayesian network analysis (Fig. 5A), and colocalization analysis (Fig. 2A; Table III), which point to a pleiotropic regulation of these traits. copQTL15 (LBs, Pmh, LW, and LI) colocalized with a cis-eQTL for BrKRP2_A03, another cis-eQTL for BrGTE6_A03, and five trans-eQTLs for BRASSICA RAPA CELL DIVISION CONTROL2_A01.2 (BrCDC2_A01.2), BrCDC2_A06.2, BRASSICA RAPA PHABULOSA_A04 (BrPHB_A04), BrCycD3;2_A07, and BrGA20OX3_A10 (Table IV). KRP2 and GTE6 play roles in organ size and leaf development in Arabidopsis (Chua et al., 2005; Kawade et al., 2010). The PHB transcription factor causes the transformation of abaxial to adaxial leaf fates (Bao et al., 2004). The other trans-regulated gene, BrGA20OX3_A10, is involved in controlling leaf lobes in B. rapa (Li et al., 2009). CDC2 has a role in the cell cycle (Hemerly et al., 1993), and CycD3;2 has a function in the regulation of cell numbers during apical growth. CycD3;2 was mapped within an important meta-QTL interval involved in leaf angle, leaf orientation value, LL, and LW in maize (Ku et al., 2012). In this study, copQTL22 for LC and leaf lobes colocalized with a cis-eQTL for BrER_A09, which also colocalized with a cis-eQTL for BRASSICA RAPA ADP-RIBOSYLATION FACTOR3_A09 (BrARF3_A09) and trans-eQTLs for BrKAN2_A05, BrFT_A07.1, and BrKRP2_A09. ER encodes a Leu-rich repeat receptor-like kinase, with pleiotropic effects on many traits, including morphological differences, leaf chlorophyll, and tolerance to drought and salt stresses in transgenic tomato plants (Keurentjes et al., 2007; Seo et al., 2012; Villagarcia et al., 2012). In Arabidopsis, Kelley et al. (2012) provided evidence that KAN and ARF3 proteins formed a functional complex active in leaf development.

copQTL23 on A10 (PL, LI, LC, LL, LW, BL, PB, and Ft) at the BrLNG1 locus colocalized with a cis-eQTL for BrLNG1, a trans-eQTL for BrKAN2, and a cis-eQTL for the Ft gene BrCO. Furthermore, both Bayesian network and colocalization analyses showed that BrLNG1_A10 associated with the trait PL (Figs. 2 and 5B; Table III). LNG1 mutants affect leaf polar cell elongation in Arabidopsis (Lee et al., 2006). The previously characterized leaf-adaxialized kan1, kan2 double mutant produces finger-shaped protrusions on the abaxial surface (Pekker et al., 2005). All the above-mentioned copQTL combining traits related to leaf shape, size, flowering time, and plant architecture with eQTL (cis- and trans-regulated) suggest pleiotropic regulation of leaf development and plant architecture traits. To further investigate the set of candidate genes, we sequenced the coding regions and around 1,000-bp promoter regions from a rapid cycling line (RC-144), a Japanese turnip DH line (VT-117), and a cv Pak choi line (PC-001) and compared the sequence with the Chiifu reference sequence (cv Chiifu-401). We identified several amino acid changes in coding regions and several changes in the promoter regions that could affect gene function (Supplemental Fig. S9). Further studies are needed to pinpoint which mutations affect gene function/expression and, thus, phenotypic trait variation.

In conclusion, we combined data analysis of candidate gene expression and phenotypic QTLs for leaf traits, Ft, and plant architecture to increase our understanding of the molecular basis of leaf development. This led to the identification of several candidate genes for these phenotypic traits, with focus on the roles of BrKRP2_A03, BrER_A09, BrLNG1_A10, and BrFLC2_A02 that pleiotropically regulate leaf development, Ft, and plant architecture in B. rapa.

MATERIALS AND METHODS

Plant Materials and Growth Conditions

Brassica rapa ‘Yellow sarson’ YS-143, an early-flowering Indian oilseed plant as female parent, was crossed with B. rapa ‘Pak choi’ PC-175, a late-flowering Chinese leafy vegetable type as male parent, to produce the DH68 population. Seeds of the DH lines and parents were germinated in petri dishes at 18°C in the dark for 36 h in the growth chamber to accelerate and synchronize germination, then transferred to trays with soil and transplanted on day 18 after sowing in pots. Single plants were cultivated in plastic pots (diameter, 17 cm) under 16-h days, 24°C/8-h nights, 10°C in a glasshouse in 2010. The 2008 conditions are described in Xiao et al. (2013).

These DH lines (92) were evaluated for several phenotypic traits under different conditions when grown from October 2008 to January 2009 and from February to May 2010. The 92 genotypes were planted in three blocks in 2008 and in one block in 2010, where genotypes were randomized. A total of 17 traits were recorded in the 2008 experiment, while six phenotypic traits were recorded in 2010 (LN, LC, LL, LW, BL, and LI). In 2010, four traits related to leaf size (LL, LW, BL, and LI) were evaluated at nine or 10 time points spaced in intervals of 3 to 5 d (Table I). Diagrams illustrating the scoring scales of LD and Lcu are shown in Supplemental Figure S10, A and B. LB are classified in two categories, LBb and LBs. The distinction between these two classes is not always clear (Supplemental Fig. S10C). The leaf characteristics of fully expanded third leaves and the phenotypes of the two parents are illustrated in Figure 1 and Supplemental Figure S3.

In 2008, RNA was extracted from the 92 DH lines for eQTL analysis. The third and fourth leaves of three biological replicate were collected 5 weeks after transplanting in the morning (10 am to 12 noon); thereafter, each replicate was ground individually, and equal amounts of powder were mixed and used for transcript profiling.

Development of Genetic Markers for Genes Involved in Leaf Traits

We identified 91 candidate genes belonging to nine functional pathways in the literature that are implicated in leaf development control in Arabidopsis (Arabidopsis thaliana; Supplemental Table S1). Using the Arabidopsis locus name, homologous genes were annotated in the B. rapa reference genome (Chinese cabbage cv Chiifu-401; http://brassicadb.org/brad/). All genes on contigs matching the same Arabidopsis coding sequence generated by BLASTP best hit were considered as B. rapa paralogs (cutoff E-value of e⁻⁵). Gene structures were predicted by sequence comparison with the Arabidopsis coding sequence using DNASTAR Lasergene 9.0 (Lasergene). The PCR primers for genetic markers were designed by Primer 3 (http://frodo.wi.mit.edu/primer3/), with expected sizes of amplified fragments being approximately 200 to 300 bp. The markers based on the candidate genes were named as follows: Br, Arabidopsis gene name, genome locus, ordered primer code.

Linkage Map Construction and pQTL Analysis

Genotyping for polymorphic candidate gene markers was conducted according to the manufacturer’s protocol for the 96-well LightScanner System (ID Technology; Montgomery et al., 2007; Xiao et al., 2013). The primers used for this study are listed in Supplemental Table S2.

Sixty leaf candidate genes, 125 Ft candidate genes, amplified fragment length polymorphism, simple sequence repeat, and insertion/deletion markers were used to construct the genetic map using Joinmap 4.0 (Kyazma; http://www.kyazma.nl/) using a regression approach and the Kosambi map function. Single pQTL analysis was undertaken with interval mapping and restricted and full multiple QTL model mapping using MAPQTL 5.0 (Van Ooijen, 2004). Initially peak markers from a map region with LOD score > 2 were used as cofactors, and a final list of cofactors was selected using the automatic cofactor selection procedure, which uses a backward elimination approach to select a final set of cofactors. The restricted and full multiple QTL model mapping processes were repeated with different sets of cofactors until the QTL profile was stable. For establishing a genome-wide significance threshold for the QTL analyses, a permutation test was done with 1,000 iterations; however, a fixed LOD threshold of 3.0 was used as the final threshold to declare a QTL, because for most of the traits, this was the 95th percentile of the permutation LOD scores. QTLs with a LOD score between 2 and 3 were considered as putative QTLs because, for most of the traits, a LOD score of 2 was the significance threshold for a single specific linkage group. Finally, 1 − LOD support intervals were determined for the assigned QTL.

RNA Isolation and Microarray Design

By using the Trizol reagent (Invitrogen), total RNA was extracted from approximately 300 mg of frozen leaf material (a mixture of three biological replicates per DH line). The first strand of complementary DNA was synthesized from 1 μg of total RNA using the I Amplification Grade kit (Invitrogen) according to the manufacturer’s instructions. Agilent 105K Brassica species oligoarrays (Agilent Technologies), which contain 96,557 features, were used for two-color microarray experiments and implemented in the R package designGG (http://www.rug.nl/research/bioinformatics/). All microarray experiments were performed according to the manufacturer’s manual (Agilent Technologies).

eQTL Mapping Analysis

The eQTL analysis was performed using the basic single marker regression procedure present in R/QTL. The expression of genes represented as 60-mer probes on the array was measured using two-color array technology, and for the mapping, we used the ratio between two genotypes: Y_i = α + βG_i + error, where Y_i = probe intensity and G_i = genetic effect. In this model, the genetic effect was annotated for the expression ratio as described (Fu and Jansen, 2006); β is the effect of the different allele (1 for A > B, 0 for A = B, and −1 for A < B). This model was evaluated at each marker to get an estimate of the allelic effect on the expression probes. This results in a P value, which was transformed into a LOD score. The eQTL with LOD > 3.0 was considered significant. In our study, we used the gene expression and marker information from the 92 DH lines to detect the eQTL for the annotated leaf candidate genes. In addition, the expression of several B. rapa leaf candidate genes that were not represented on the microarray was also profiled using RT-qPCR. The RT-qPCR experiments were performed according to our previous paper (Xiao et al., 2013). The primer sequences used in this study are listed in Supplemental Table S14. cis-eQTL (local eQTL) was defined when the (derived) genetic position of the gene and its eQTL were located within 50 cM. The remaining eQTLs were defined as trans-eQTL (distant eQTL).

Genetic Regulation Network Analyses

To explore the modular association with the traits and gene expression data, we first computed a Spearman correlation among the 17 phenotypes (measured data) and 118 genes/probes (expression data by log₁₀ scale). The modules are shown using the heat-map tool (Spearman product r > 0.3, P < 0.05). Furthermore, to elucidate a coregulation network, the eQTL and pQTL LOD profiles across the genetic map were calculated, and these top covariates (Spearman product r > 0.3, P < 0.05) were used to construct the regulation network.

Additionally, we applied a Bayesian network to evaluate subnetworks of the expression and phenotypic QTL data traits from the DH68 population. The Bayesian network is a probabilistic graphical model of multiple variables that has adequate statistical power (Spirtes et al., 2000; Li et al., 2005; Supplemental Text S1). In order to reduce the computational load, a limited set of candidate genes are selected based on (1) copQTLs and candidate genes; (2) clusters identified by Spearman correlation of trait variation; and (3) a genetic regulatory network based on LOD scores. The selected candidate genes were calculated into three conditionally independent phenotypic structures.

Visualization of the genetic regulation networks was plotted using Cytoscape 2.8.2 (http://www.cytoscape.org/; Shannon et al., 2003). All calculations were done in the open statistical software R 2.13.1 (Ihaka and Gentleman, 1996).

Colocalization of pQTLs, eQTLs, and Candidate Genes

To identify the colocalization of pQTLs, eQTLs, and candidate genes, we computed multiple test P value corrections with false discovery rate (Steibel et al., 2011). Given a particular eQTL region, delimited by a 5-cM interval to each side of the peak, all overlapping pQTL regions were selected (West et al., 2007). The P of two intervals of this length (10 cM) overlapping in a 1,328-cM-long genome (the length of our linkage map) is

Sequence data from this article can be found in the Brassica Database (http://brassicadb.org/brad/).

Supplemental Data

The following materials are available in the online version of this article.

• Supplemental Figure S1. Genetic map of the B. rapa DH68 population.

• Supplemental Figure S2. Mapped leaf trait candidate genes as anchors between the genetic map (YS-143 × PC-175; cM) and the physical map (Mb).

• Supplemental Figure S3. Images of parents and their DH progeny are shown: first stage (25 d after sowing) and 10th stage (57 d after sowing).

• Supplemental Figure S4. Frequency distribution for the morphological traits studied in the B. rapa DH68 population (YS-143 × PC-175).

• Supplemental Figure S5. Validation by RT-qPCR analysis of transcripts with cis-eQTL: BrCYCD1;1_A02, BrKRP2_A03, and BrDRL1_A08.

• Supplemental Figure S6. Correlation heat map of gene expression data using microarray and RT-qPCR data and phenotype traits in B. rapa.

• Supplemental Figure S7. Bayesian network for 17 phenotypic leaf traits of B. rapa.

• Supplemental Figure S8. Bayesian network analysis identified (PC algorithm with significance level of 0.01).

• Supplemental Figure S9. For the genes BrFLC2_A02 (Bra028599), BrKRP2_A03 (Bra012894), BrER_A09 (Bra007759), BrLNG1_A10 (Bra008689), BrLNG1_A02 (Bra023526), BrKAN2_A09 (Bra023254), BrKAN2_A05 (Bra033844), BrHST1_A05 (Bra039468), BrARGOS_A07 (Bra003394), BrARGOS_A09 (Bra007491), BrGRF5_A03 (Bra001532), and BrGRF5_A05 (Bra027384) colocalizing with QTLs for copQTL6, copQTL8, copQTL15, copQTL18, copQTL22, and copQTL23, the coding sequence and 1,000-bp upstream sequence (promoter region) were sequenced to assess allelic variation.

• Supplemental Figure S10. Diagram of LD, Lcu, LBb, and LBs of a B. rapa leaf.

• Supplemental Table S1. List of selected genes involved in different functional pathways of leaf development in B. rapa.

• Supplemental Table S2. Sequence informative markers for the leaf trait genes that are genetically mapped by LightScanner on the 92 B. rapa DH68 lines.

• Supplemental Table S3. Variation explained (%) by genotype, genotype + block in ANOVA test for 17 traits measured in year 2008.

• Supplemental Table S4. Variation explained (%) by different models of ANOVA for the common traits measured in 2008 and 2010 experiments.

• Supplemental Table S5. Summary statistics of leaf morphological traits in the B. rapa DH68 population and parental genotypes in 2008 and 2010.

• Supplemental Table S6. pQTL results of leaf morphological traits in B. rapa.

• Supplemental Table S7. List of all candidate genes for Ft and leaf traits with their physical order and eQTL, either genetically mapped in DH68 or with inferred map position based on physical position and genetically mapped flanking markers.

• Supplemental Table S8. List of copQTLs with initial pQTLs.

• Supplemental Table S9. Annotation of leaf trait genes represented on the microarray.

• Supplemental Table S10. Detailed information for all pQTLs and eQTLs identified in this study with their LOD score, functional pathways, and information on cis-/trans-regulation.

• Supplemental Table S11. Coexpression correlation among all probes and phenotypic traits.

• Supplemental Table S12. Coregulation of all the leaf traits and expression of leaf trait-related candidate genes by calculating the correlation based on LOD score profile.

• Supplemental Table S13. eQTL location of duplicated genes for a set of leaf trait candidate genes (two to four copies).

• Supplemental Table S14. List of primers for leaf development candidate genes designed for RT-qPCR gene expression.

• Supplemental Text S1. Structure learning method for undirected networks.

Glossary

Ft

flowering time

QTL

quantitative trait locus

LL

lamina length

LW

leaf width

eQTL

expression quantitative trait locus

DH

doubled haploid

copQTL

colocalized phenotypic quantitative trait locus

cM

centimorgan

LB

leaf lobe

LD

leaf wing depth

BL

blade length

LI

leaf index

Lcu

leaf curling

LC

leaf color

PL

petiole length

PB

plant branching

Pmh

plant mature height

RT-qPCR

quantitative real-time PCR

LOD

log of the odds

LA

leaf area

LBb

leaf big lobe

LBs

leaf small lobe

LN

leaf number

LP

leaf perimeter