A transcriptomic analysis of superhybrid rice LYP9 and its parents

Abstract

By using a whole-genome oligonucleotide microarray, designed based on known and predicted indica rice genes, we investigated transcriptome profiles in developing leaves and panicles of superhybrid rice LYP9 and its parental cultivars 93-11 and PA64s. We detected 22,266 expressed genes out of 36,926 total genes set collectively from 7 tissues, including leaves at seedling and tillering stages, flag leaves at booting, heading, flowering, and filling stages, and panicles at filling stage. Clustering results showed that the F1 hybrid's expression profiles resembled those of its parental lines more than that which lies between the 2 parental lines. Out of the total gene set, 7,078 genes are shared by all sampled tissues and 3,926 genes (10.6% of the total gene set) are differentially expressed genes (DG). As we divided DG into those between the parents (DG_PP) and between the hybrid and its parents (DG_HP), the comparative results showed that genes in the categories of energy metabolism and transport are enriched in DG_HP rather than in DG_PP. In addition, we correlated the concurrence of DG and yield-related quantitative trait loci, providing a potential group of heterosis-related genes.

Extensive sequence diversity at the microstructural level has been demonstrated in a number of plant species (1), and such diversity can extend even to allelic regions (2). These intraspecific allelic variations should have impacts on gene expressions that lead to phenotypic variation, perhaps including hybrid vigor as a beneficial trait used in crop breeding. In a hybrid, in which 2 different alleles of a gene are often brought together, the combined allelic expression may deviate from that of either parent or the midparent predictions (3). In maize, both allelic diversity and expression variation were found between inbred parents and their hybrid (4). In maize hybrids, not only the allelic variation in gene expression but also different responses to extrinsic stimuli supported the presence of allelic expression variation in the same genetic context (5). Large-scale transcriptome profiling has been used for heterosis studies in maize (6), Arabidopsis (7), and wheat (8). In rice, an investigation of a yield-related quantitative trait locus (QTL) resulted in a discovery of allelic variation that affected the expression of a leucine-rich repeat receptor kinase gene cluster (9). Another survey with a cDNA microarray concerning 9,188 expressed sequence tags on expression polymorphism between an elite rice hybrid and its parental varieties revealed significant heterotic expression for 141 expressed sequences (10).

We have recently focused our heterosis research on Liang-You-Pei-Jiu (LYP9), a superhybrid rice strain from a cross between the maternal inbred PA64s, a photothermosensitive male sterile line, and the paternal inbred 93-11, an elite indica variety, after we sequenced the 2 parental genomes (11, 12). Two-dimensional electrophoresis analysis among 93-11, PA64s, and LYP9 revealed significant numbers of different embryo protein spots, many of which were shown to display mirrored relationships between parents and the first filial generations (13). Further analysis on mature embryos of this hybrid triad identified 54 differentially expressed proteins involved in major biological processes including nutrient reservoir, response to stress, and metabolism. Among these embryos, most of the storage proteins exhibit overdominance and stress-induced proteins display additivity (14). We also carried out transcriptome profiling for the hybrid and its parents using both sequencing-based (15–17) and hybridization-based methods (18). We now report a rather large-scale comparative transcriptome analysis of the triad, concerning 7 tissues sampled across developmental times and different tissues. We expect this genome-wide transcriptome comparison to be an initial step forward in understanding the causative mechanism of the altered gene expression in the hybrid and the molecular mechanism underlying heterosis.

Results

The Rice Whole-Genome Microarrays Are of Satisfactory Quality.

Our 70-mer oligonucleotide microarray, with 36,926 unique features identified, was designed based on known and predicted gene models of the indica rice 93-11 genome (18). We calibrated our microarray by doing 4 preliminary tests. First, a self-hybridization experiment was conducted, detecting only 9 false differentially expressed genes (DG) with marginal intensity above the background. Second, we conducted hybridizations between the seedling shoot and the filling panicle and discovered >5,000 DG with correlation coefficients of 0.85 between duplication and correlation coefficients of 0.81 in dye-swapping experiments. Third, to better define the background and fold changes we introduced a polyubiquitin gene as positive control, the fold changes of which are both consistent and always below the threshold (Fig. S1). We acquired at least 3 independent replicates for each sample pair in general and a total of 48 raw datasets (96 slides) for 7 tissues from the triad (collective correlation coefficient among all replicates >0.8). Finally, we validated our microarray results with semiquantitative RT-PCR, and out of 25 primer pairs with amplification products, 20 (80%) DG showed consistent results compared with those obtained from the microarray data (Fig. S2). Collectively, these results demonstrated the satisfactory quality of our experimental procedures and data.

Transcriptome Profiles of LYP9 and Its Parents Revealed Consistent Trends with Phenotypic Observations.

Our data were derived from 7 tissues of the LYP9 hybrid triads, including seedling shoot, leaf at tillering stage, flag leaf at booting stage, flag leaf at heading stage, flag leaf at flowering stage, flag leaf at filling stage, and panicle at filling stage, out of which we identified 11,448–14,592 genes expressed in each pairwise comparison (Table S1). Our analysis revealed 7,078 genes expressed in all studies tissues and 22,266 genes expressed collectively.

We used a cluster analysis method to investigate correlations among transcriptome profiles. The results revealed that tissues from different cultivars at the same developmental stage always formed the primary groups (Fig. 1). In a broader spectrum, the transcriptome profiles of LYP9 are similar to PA64s (maternal) at the early developmental stages but closer to 93-11 (paternal) at the later stages. Both are consistent with the morphological appearances or characteristics of the hybrid plant at corresponding stages, observed empirically in the field as either 93-11-like or PA64s-like. A distinct result was found in the cluster of the panicle at filling stage, where the profile of LYP9 is more similar to that of 93-11 because PA64s is a photothermosensitive male sterile rice line (19), and many of its genes may not express appropriately or at levels comparable to those of 93-11 and LYP9.

Fig. 1.

Hierarchical clustering analysis of all gene models based on expression data. Normalized expression values for the microarray (37K) clustered with Genespring (Silicon Genetics). Each horizontal line refers to a gene. The color represents the logarithmic intensity of the expressed genes. N, L, and P stand for 93-11, LYP9, and PA64s, respectively. Numbers 1–7 denote samples from the following tissues in order: seedling shoot, leaf at tillering stage, flag leaf at booting stage, flag leaf at heading stage, flag leaf at flowering stage, flag leaf at filling stage, and panicle at filling stage.

When we looked at universally expressed genes, some are undoubtedly housekeeping genes whereas the molecular category of structure was found overrepresented (Fig. S3). We also noticed that among structure molecules, genes encoding cytoplasmic (60S/40S) protein and plastid ribosomal (50S/30S) protein have a synergistic expression profile except in the filling stage panicle where the former are up-regulated and the latter are down-regulated (Fig. S4) as compared with those in other tissues. This result is consistent with the fact that the number of chloroplasts in panicles is significantly lower than that found in leaf tissues.

DG and Their Functional Analysis.

We defined DG between the parental lines as DG_PP and those between the hybrid and its parents as DG_HP. DG_PP only denote the differences between 2 inbred lines, but DG_HP may underlie heterosis because differences in expression between hybrid and parents should underlie their phenotypic differences. DG_HP can be divided into 2 classes—i.e., those shared by DG_PP and DG_HP (DG_O) and those uniquely belonging to DG_HP (DG_HPU). We found 3,926 (10.6%) DG observed at least once among all sample pairs (Dataset S1), and the numbers of DG_HPU are larger than DG_O in all 7 tissues investigated (Table 1). By comparing DG between the hybrid and its parents, we found that the great majority of DG are close to either their maternal or their paternal parent and that a minority of them are close to neither parent. To further understand the function of DG, we classified these genes according to their functional categories and relatedness. For instance, DG_PP are enriched in 16 out of 161 categories as compared with DG_HP, which are enriched in 25 function categories (Table S2 and S3). Since DG_HP are composed of DG_O and DG_HPU, we expected that heterosis-related genes may be enriched in DG_HPU rather than in DG_O. Indeed, the DG_HPU identified in this study are enriched mostly in the categories of energy metabolism and transport (Table 2).

Table 1.

Number and classification of DG

Sample	DGPP	DGHP
		L/N	L/P	DGHPU	DGO	H2P	CHP	B2P	CLP	L2P
S1	305	243	167	215	161	19	190	21	142	4
S2	312	309	266	328	201	17	247	46	208	11
S3	472	323	412	424	272	14	465	42	174	1
S4	389	345	313	447	180	36	324	17	235	15
S5	342	337	333	401	208	57	315	40	182	15
S6	331	313	323	347	203	36	222	53	199	40
S7	383	405	451	505	289	24	321	11	400	38
Total	2132	1913	1898	2260	1280	196	1851	198	1316	108

N, L, and P refer to 93-11, LYP9, and PA64s, respectively. DG_PP refers to DG between both parents, DG_HP refers to DG between the hybrid and parent. DG_HPU denotes the unique portion of DG_HP, and DG_O denotes the overlap between DG_PP and DG_HP. H2P, CHP, B2P, CLP, and L2P represent higher than both parents, close to higher parent, between both parents, close to lower parent, and lower than both parents, respectively.

Table 2.

Functional classification of unique portion of DG_HPU

Functional categories	S1	S2	S3	S4	S5	S6	S7
Metabolism
Amino acid metabolism	7	14*	14	10	16	19**	29**
Biosynthesis of polyketides and nonribosomal peptides	0	1	0	1	0	1	1
Biosynthesis of secondary metabolites	7	15	23	15	19	19	20
Carbohydrate metabolism	19	28	39	37	47**	37*	40
Energy metabolism	9**	14**	17**	11	11	29**	39**
Glycan biosynthesis and metabolism	0	0	6	3	3	6	9*
Lipid metabolism	5	7	8	7	8	12*	6
Metabolism of cofactors and vitamins	10	19	21	14	24	17	16
Metabolism of other amino acids	4	8*	4	4	8*	3	7
Nucleotide metabolism	6	5	5	8	8	2	10
Xenobiotics biodegradation and metabolism	9	19	24	20	21	19	12
Genetic Information Processing
DNA metabolism	6	1	3	3	9	2	0
RNA metabolism	4	15	26	26	26	13	24
Cellular protein metabolism	17	34	61**	50	44	22	41
Environmental Information Processing
Signal transduction	1	6	12	17*	17**	5	3
Transport	11	27	43**	39*	42**	33**	38
Cellular Processes
Cell motility	0	1	1	1	1	0	0
Cell cycle	2	6	4	2	4	2	7
Cell–cell signaling	0	0	2	1	1	1	1
Cell death	5	4	1	2	5	5	3
Cell growth	0	1	1	1	0	0	0
Other	28*	32	59**	59**	52**	50**	75**
Unknown	123	181	193	229	155	152	249
Total	215	328	424	447	401	347	505

* and ** denote significant enrichment of DG among function category with P < 0.05 and P < 0.01, respectively.

Since the most important trait of hybrid rice is grain yield, we analyzed the genes involved in carbohydrate biosynthesis (20, 21)—such as starch biosynthesis—and noticed that genes involved in starch synthesis have much higher expression in the panicle of LYP9 than PA64s at filling stage, including the key enzymes in starch biosynthesis such as sucrose synthase, ADP-glucose pyrophosphorylase, and starch synthase. The result is in agreement with the fact that starch biosynthesis cannot take place in the panicles of PA64s. In addition, rubisco, a key protein in the pathway, showed a lower expression level in LYP9 than in PA64s, thus supporting the fact that the panicle of PA64s remained green long after flowering was observed in the field. It is interesting that the genes taking part in sucrose and starch metabolism, such as ADP-glucose pyrophosphorylase, sucrose-P synthase, invertase, and branching enzyme, tend to be highly expressed in the hybrid (Fig. 2).

Fig. 2.

Expression profiles of DG between LYP9 and its parents in carbohydrate biosynthesis pathway. Genes involved in carbohydrate metabolism were identified according to their Enzyme Commission annotation, and those genes that differentially expressed at least once were shown. The log2-transformed ratio between the hybrid and either parent was used (L, LYP9; N, 9311; P, PA64s). Each row represents a single gene, and the number indicates a group of isoenzymes in the pathway according to its position in the path and order. Red and green colors denote up- and down-regulated genes, respectively. The genes are listed as follows: (1) ribulose-bisphosphate carboxylase, (3) glyceraldehyde-3-phosphate dehydrogenase, (5) fructose-bisphosphate aldolase, (6) fructose-bisphosphatase, (7) glucose-6-phosphate isomerase, (9) transketolase, (10) sedoheptulose-1,7-bisphosphatase (SBPase), (12) phosphoribulokinase, (14) ADP-glucose pyrophosphorylase, (15) UDP-glucose pyrophosphorylase, (16) sucrose-P synthase, (18) sucrose synthase, (19) invertase, (21) starch synthase, and (22) branching enzyme.

Nonadditive-Expressed Genes.

Concerning the relative level of gene expression among a hybrid–parent triad, we often expect 2 scenarios to come into play. In the first scenario, gene expression in the hybrid exhibits a cumulative mode, contributed by each allele from the respective parents. In the other scenario, the expression deviates from the midparental level. The former scenario is additive, indicating that alleles from both parents may contribute to gene expression in the hybrid, attributable mostly to a cis-regulation mechanism. The latter scenario is nonadditive, in which other regulators probably contribute to an altered expression of the corresponding alleles in the hybrid, attributable mostly to trans-regulation (3). In comparison with gene expression among the LYP9 triad, we detected 860 up-regulated and 1,095 down-regulated nonadditive genes (NAG). The number of NAG in each sampling triad ranged from 195 to 497 (Table 3); they composed 0.5–1.4% of the total gene set and 29.6–53.7% of DG_HP identified at 7 tissues.

Table 3.

Nonadditive-expressed genes in LYP9

Sample	Number of NAG				Number of NAG in DGHP
	Up	Down	Total	a%	DGHP	b%	DGHPU	c%	DGO	d%
S1	80	115	195	0.5	144	38.3	108	50.2	36	22.4
S2	97	180	277	0.8	184	34.8	148	45.1	36	17.9
S3	163	147	310	0.8	206	29.6	168	39.6	38	14.0
S4	182	220	402	1.1	261	41.6	222	49.7	39	21.7
S5	140	126	266	0.7	239	39.2	209	52.1	30	14.4
S6	103	177	280	0.8	218	39.6	189	55.4	29	14.3
S7	158	339	497	1.4	426	53.7	264	52.3	162	56.1
Total	860	1095	1846	5.0	1481	46.5	1245	55.1	488	38.1

a% denotes the percentage of NAG in the total gene set (36,926), b%, c%, and d% denote the percentage of NAG in total numbers of DG_HP, DG_HPU and DG_O, respectively.

DG_HP Are Enriched in Known QTLs.

We were able to map 2,673 DG_HP to 3,128 QTLs classified into 9 categories and 209 traits in the rice genome (www.gramene.org). One important piece of evidence supporting the correlation between the 2 types of data is the fact that the fraction of DG_HP in the transcriptome profiles (36,926 expressed genes) is 8.6% as compared with the fractions of DG_HP mapped to QTLs—10.1% and 11.8% in the QTL intervals that harbor less than 50 and 10 genes, respectively (Table S4). Among DG_HP-related QTLs, many are well characterized, including 1000-seed weigh (e.g., AQCY015, CQAS23, AQAI076, and CQAS23), filled grain number (e.g., AQCY010, AQCY059, AQAK009, and AQAK011), grain number (CQB22, AQDR015, AQDR059, and AQED038), and grain yield per panicle (AQDR091, AQDR103, and AQDR104). The potential association between DG_HP and QTLs were also suggested within many QTL regions, such as Starch synthase III (Os055024_01) to AQCY010 for filled grain number, putative sugar transporter (Os055048_01) to AQAI076 and AQEY022 for 1000-seed weight, and auxin response factor (Os016758_01) to CQK15 for panicle number. To help portray this DG_HP-QTL correlation, we aligned DG_HP over yield-related QTL regions covering less than 100 genes on rice chromosomes (Fig. 3).

Fig. 3.

Distribution of DG located in yield-category QTL of small intervals. Yield-category QTL of small intervals (number of genes ≤100) that harbor DG_HP were aligned to TIGR's rice pseudochromosome version 5. The long horizontal lines represent 12 rice chromosomes, the short horizontal lines represent QTL intervals, and the short vertical lines represent DG_HP.

Discussion

Complex Regulatory Mechanisms Probably Underlie Gene Expression Changes in Hybrid.

Transcriptomes are not only always specific to cell types but also are regulated at different levels, such as transcription and splicing, and through genetic or epigenetic mechanisms. Although in this current report we are unable to show detailed sequence comparisons and validations for different alleles of annotated DG, allelic sequence variation—especially those in the regulatory sequence/element—is undoubtedly one of the causes of gene-expression change in hybrids. We will certainly proceed in identifying these allelic differences of all DG in our dataset. Another class of gene regulators is trans-regulators, such as transcription factors (TFs). The dosage effect of such regulatory genes had been proposed to affect phenotypes in hybrids (22). We indeed found that 187 TFs exhibited differential expression in the hybrid compared with either parent. It is quite a coincidence that a recent study using seedling tissue of 2 hybrid triads, based on the genomic sequence of 93-11 and nipponbare, also suggested that altered gene expression caused by interactions between transcription factors and the allelic promoter region in the hybrids was one plausible mechanism underlying heterosis in rice (23).

Moreover, we noticed that among those differentially expressed TFs, the AP2-EREBP family—potential targets of miRNA (24)—is overrepresented. Noncoding RNAs are involved in epigenetic regulations, and other epigenetic mechanisms including DNA methylation, acetylation and deacetylation of histones, and chromatin remodeling. It had been reported that the degree of methylation in hybrids is different from that in inbred lines in Arabidopsis and rice (25, 26). A recent study reported that epigenetic regulation of a few regulatory genes (CCA1 and LHY in this case) induced cascade changes both in downstream genes (TOC1, GI, etc.) and in physiological pathways, and ultimately induced growth and development, which also indicates the presence of a general mechanism for the growth vigor and increased biomass commonly observed in hybrids (27). In the present survey, we also found that among DG_HP there were many epigenesis-related genes, including methyltransferase, hydroxymethyltransferase, serine O-acetyltransferase, histone acetyltransferase, acetyl-CoA acyltransferases, and chromodomain helicase-DNA-binding protein 3. The expression of these genes is being verified experimentally as is their involvement in related biological pathways.

Gene Expression Variations in Hybrid Suggests Correlation to Genetic Mechanisms Responsible for Heterosis.

The dominance and overdominance hypotheses (28) were proposed to explain heterosis before the molecular concepts of genetics were formulated, and these hypotheses are not closely allied with molecular principles. We categorized DG between hybrid and parents (DG_HP) into 5 basic categories: overdominance (H2P), underdominance (L2P), dominance (CHP and CLP), and midparent (B2P). We found that dominant expression was the most prevalent among DG_HP (81.6–91.8%). Additive and nonadditive expression represent another possible genetic model for gene expression in hybrids (3). Whether or not a transcript shows nonadditive expression is most likely to be influenced by the contributions of cis- and trans-acting factors of this gene (29–32). In our data, the majority of genes in the hybrid showed additive expression, and the phenomenon suggests that cis-acting elements usually play a major role in the control of general gene expression. Nonadditively expressed genes in our entire dataset constituted only 0.5–1.4% of the total discovered in each sampled tissue but accounted for 29.6–53.7% of DG_HP. A similar result was observed in a study of maize heterosis, in which the nonadditive-expressed genes were found to contain 2.2% of the total genes and 22% of DG (6). It should be noted that we were unable to detect those genes where the silencing of 1 allele was compensated by overexpression of the other, which might cause underestimation of nonadditive genes in hybrids, as mentioned previously (31). The analysis of nonadditive gene expression indicates that allelic expression in hybrids may not just be a combination of alleles from the 2 parents but is rather regulated by other genes or epistatic mechanisms. Nonadditive gene expression was also considered as midparent heterosis or heterotic expression (10, 23).

A study in gene expression in maize endosperm revealed heterochronic expression of 3 allele pairs (33). In the present study, 85% of DG were detected only once in 7 tested tissues. For those DG that appeared more than once, 63% (LYP9/PA64s) to 75% (LYP9/93-11) differed in the same direction; i.e., either up- or down-regulated. This trend indicates that their corresponding regulatory mechanisms may function in the same way in different tissues and under different conditions. In contrast, 25–39% of those genes follow a different trend; they differ in the opposite direction, so that a gene in the hybrid may be under a different control mechanism or the regulatory factors may function in a different way under variable conditions.

DG Are Candidates for Genes That Play an Important Role in Heterosis.

Microarray-based expression studies allowed us to identify genes that are differentially expressed between a hybrid and its parents, and these DG are often found to be expressed in a biased pattern in comparison with regular transcriptomes. For example, we found that the DG_HP involved in the carbohydrate–metabolism pathway had a larger fraction of up-regulated genes than down-regulated genes, similar to the recent studies (23, 27). However, of the genes taking part in oxidative phosphorylation, there were more down-regulated genes identified in the hybrid than in the parental lines. In addition, heading stage is an important period for panicle development and grain-yield formation, and our previous serial analysis of gene expression (SAGE) analysis showed that genes related to protein biosynthesis and peptide transport were up-regulated in the panicle of the hybrid LYP9 (16). Based on our current data, a similar conclusion was reached in the analysis of gene expression in flag leaves of heading stage and flowering stage. It was interesting to find that sucrose-transport genes are up-regulated in LYP9 panicles as compared with those in 93-11 panicles, suggesting that the transportation of carbohydrate from the source to the sink in LYP9 is more efficient than in at least one of the parents.

An altered expression of the maize domestic gene tb1 was characterized as the cause of observed quantitative phenotypic changes by a fine-mapping approach (34), and a transcription activator was demonstrated to be responsible for the significant plant-height changes in an Arabidopsis hybrid (35). Recently, a major quantitative gene in rice, Ghd7, isolated by map-base cloning and encoding a CCT domain protein, was considered as a crucial factor for increasing productivity and adaptability of an elite hybrid cultivar, Shanyou 63, and some other indica varieties (36). In our current study, not only have we found many TFs in our DG collections, but we also mapped a high fraction of DG to the intervals of grain-yield-related QTLs. These results led us to believe that DG between the hybrid and parents may contribute in a significant way to heterosis. We also have constructed databases integrating heterosis-related genes among major crops and experimental plants, identifying altered sequences among differentially expressed alleles (37), and mapping relative DG to QTLs discovered in this study.

Materials and Methods

Rice Whole-Genome Oligonucleotide Array.

The whole-genome array was developed based on annotated and predicted genes from the genome assembly of indica rice 93-11 (11, 12). Oligonucleotides were arrayed onto 2 poly-L-lysine-coated microscope slides as a set with a SpotArray72 microarrayer (Perkin–Elmer) in the microarray laboratory at Beijing Genomic Institute, and the slides were processed according to a standard procedure (38).

Plant Materials and Data Processing.

LYP9 and its parental lines (93-11 and PA64s) were grown in a greenhouse for the seedling samples and in the rice field for all other samples. The plant tissues were collected and stored at −80 °C. RNA samples were isolated (39), quantitated by using a NanoDrop1000 spectrophotometer (Nanodrop Technologies), and labeled (40, 41). Each sample had at least 3 biological replications to minimize systematic errors. Separate tiff images of Cy5 and Cy3 channels were obtained by ScanArray Lite scanner (Perkin–Elmer), and spot intensities were quantified by using the Axon GenePix Pro 5.1 image analysis software.

We categorized our raw data with 3 simple criteria. First, features were flagged as “bad” either by using Genepix or by manual investigation, second, a false positive rate ≤5% in reference to the controls was found, and third, legitimate features were found in at least 2 of the 3 replicate sets or 3 of the 4 replicate sets. The processed data were normalized based on the mean of all expressed genes. The normalization of the 2-channel data for each array was done by using the intensity-based Loess method with R language. DG were defined by a log-scale ratio between paired samples with a P value <0.01 (Z test).

Functional Annotation.

For each gene identified, we performed detailed functional annotations by using standard tools, such as BLAST (42, 43) and HMMer (44), against public data, including (i) the The Institute for Genomic Research (TIGR) Rice Pseudomolecules and Genome Annotation database (release 5.0, http://rice.plantbiology.msu.edu); (ii) the knowledge-based Oryza Molecular Biological Encyclopedia (http://cdna01.dna.affrc.go.jp/cDNA/); (iii) the TIGR Rice Gene Index (http://compbio.dfci.harvard.edu/tgi/); and (iv) the UniProtKB/Swiss-Prot (www.ebi.uniprot.org). We also used the Kyoto Encyclopedia of Genes and Genomes (http://www.genome.jp/kegg/) and Gene Ontology databases (45) for protein annotation (E value <10⁻⁷). The HMMpfam program (http://hmmer.janelia.org/) was used to search Pfam hidden Markov models retrieved from Pfam release 18 (46) for structural domains E value (<0.001).

Because some categories are larger (i.e., involve more genes) than others, they tend to show more frequently in any set of genes; thus, it is essential to identify the statistically significant categories in a set of genes. We took the whole set of genes as the default background distribution and used the reported method (47) to decide the significance of DG in each category, with P value cutoff of 0.05 as the significance threshold.

Mapping DG to QTL.

Rice QTL data with physical positions on the TIGR release 5 genome were acquired from Gramene (www.gramene.org) and 2,685 DG were mapped to 2,729 rice QTL, covering 9 QTL categories and 211 QTL traits. For better demonstration of the relationship between DG and QTL, we classified yield-related QTL according to the number of genes in each chromosome region and performed an enrichment test according to the method described in ref. 47.