A Novel Tiller Angle Gene, TAC3, together with TAC1 and D2 Largely Determine the Natural Variation of Tiller Angle in Rice Cultivars

Tiller angle is one of the most important components of the ideal plant architecture that can greatly enhance rice grain yield. Understanding the genetic basis of tiller angle and mining favorable alleles will be helpful for breeding new plant-type varieties. Here, we performed genome-wide association studies (GWAS) to identify genes controlling tiller angle using 529 diverse accessions of Oryza sativa including 295 indica and 156 japonica accessions in two environments. We identified 7 common quantitative trait loci (QTLs), including the previously reported major gene Tiller Angle Control 1 (TAC1), in the two environments, 10 and 13 unique QTLs in Hainan and Wuhan, respectively. More QTLs were identified in indica than in japonica, and three major QTLs (qTA3, qTA1b/DWARF2 (D2) and qTA9c/TAC1) were fixed in japonica but segregating in indica, which explained the wider variation observed in indica compared with that in japonica. No common QTLs were identified between the indica and japonica subpopulations. Mutant analysis for the candidate gene of qTA3 on chromosome 3 indicated a novel gene, Tiller Angle Control 3 (TAC3), encoding a conserved hypothetical protein controlling tiller angle. TAC3 is preferentially expressed in the tiller base. The ebisu dwarf (d2) mutant exhibited a decreased tiller angle, in addition to its previously described abnormal phenotype. A nucleotide diversity analysis revealed that TAC3, D2 and TAC1 have been subjected to selection during japonica domestication. A haplotype analysis identified favorable alleles of TAC3, D2 and TAC1, which may be used for breeding plants with an ideal architecture. In conclusion, there is a diverse genetic basis for tiller angle between the two subpopulations, and it is the novel gene TAC3 together with TAC1, D2, and other newly identified genes in this study that controls tiller angle in rice cultivars.


Introduction
As a key factor in rice plant architecture, tiller angle determines the planting density per unit area and contributes greatly to grain yield. The tiller angle is defined as the angle between the main culm and the side tillers [1]. A favorable rice tiller angle is an important component of the ideal plant architecture that has been selected by humans in the long history of domestication and genetic improvement [2]. Oryza rufipogon, the progenitor of cultivated rice (Oryza sativa), shows a spread-out growth pattern that allows it to escape some diseases induced by high humidity, but this pattern occupies too much space and increases shading and lodging, thus decreasing the photosynthetic efficiency and grain yield per unit area accordingly. By contrast, cultivated rice usually exhibits a better plant architecture, with a smaller tiller angle that leads to high potential yields.
Tiller angle changes dynamically throughout the life cycle of rice to achieve efficient resource use. At tillering stage, tiller angle increases, which enables young plant to occupy a large empty space for subsequent tiller development and simultaneously inhibits weed growth. At later stages, tiller angle decreases, especially at heading stage, when it reaches its minimum, allowing the mature plant to reduce leaf shade and increase its photosynthetic efficiency. Heading stage is a key developmental period in rice, and the planting density per unit area is mainly determined by the tiller angle at this stage.
Due to its importance in rice production, increasing attention has been paid to dissecting the genetic basis of tiller angle over recent decades. Dozens of tiller angle-related quantitative trait loci (QTLs) have been explored in rice via classical bi-parental cross mapping [3][4][5][6][7][8]. However, most of these QTLs are seldom used for marker-aided selection, due mainly to their minor effects. Recently, a few genes controlling tiller angle have been cloned. PROSTRATE GROWTH 1 (PROG1) has been accepted as a domestication-related gene that controls tiller angle and tiller number during both tillering and heading stages in wild rice. PROG1 encodes a zinc finger nuclear transcription factor and is located on chromosome 7. Amino acid changes in the PROG1 protein and regulatory changes during domestication led to loss of the function of this gene, which promoted the transition from the prostrate growth tillers of wild rice to erect growth tiller of domesticated rice [9,10]. Tiller Angle Control 1 (TAC1) is a major QTL located on chromosome 9 that controls tiller angle during heading stage in cultivated rice and encodes an expressed protein without homologous genes in rice. A mutation in the 3'-splicing site of the 1.5-kb intron 'GGGA' , which exists in 88 compact japonica rice accessions, decreases the level of tac1 and leads to a smaller tiller angle; ' AGGA' is present in 21 wild rice and 43 indica rice accessions with spread-out tillers [11].
Rice shoot gravitropism is suggested to be a key factor affecting plant architecture. The rice la mutant, which exhibits a wider tiller angle, has been intensively studied for decades [12][13][14][15][16][17]. However, the LAZY1 (LA1) gene was not identified until 2007. LA1 encodes a novel, grass-specific protein that controls shoot gravitropism by regulating polar auxin transport (PAT) [18,19]. Loose Plant Architecture 1 (LPA1) is an INDETERMINATE DOMAIN protein involved in shoot gravitropism that regulates both tiller angle and leaf angle in rice during the vegetative and reproductive stages [20]. Suppression of OsPIN1 or over-expression of OsPIN2 (two auxin efflux transporters) alters PAT and increases tiller angle [21,22]. Recent research has demonstrated that strigalactones (SLs), a group of newly identified plant hormones, inhibit auxin biosynthesis and attenuate rice shoot gravitropism primarily by decreasing local indoleacetic acid contents. Multiple SOLs (suppressors of lazy1) involved in SL biosynthesis, such as dwarf 17 (d17), d10 and d27, or in SL signaling pathways, such as d14 and d3, can rescue the spreading phenotype of lazy1 [23]. Suppressing the expression of OsLIC1 (Oryza sativa leaf and tiller angle increased controller), a novel CCCH-type zinc finger gene, increases leaf and tiller angles via the BR signaling pathway [24].
Although the identification of these tiller angle-related genes is helpful for understanding the mechanism of tiller angle formation, few genes that can be used for improving rice plant architecture have been isolated based on natural variation. Therefore, mapping additional genes that contribute to the natural variation of tiller angle is required for breeding varieties with an ideal plant architecture resulting in high grain yields. Genome-wide association studies (GWAS) offer a powerful approach to establishing the relationship between DNA markers and phenotypic traits in crops [25]. With the assessment of millions of single nucleotide polymorphisms (SNPs), GWAS can take full advantage of ancient recombination events to identify the genetic loci underlying complex traits at a high resolution using a large number of crop varieties [26]. Many QTLs for agronomic traits have been identified in cultivated rice through GWAS [27][28][29]. Recently, several novel loci and candidate genes for tiller angle at tillering stage have been identified by GWAS and the elite alleles have been explored for plant architecture improvement [30].
Here, we investigated the tiller angle of 529 O. sativa accessions at heading stage and performed GWAS separately in the full population and the indica and japonica subpopulations. We isolated a novel gene, TAC3, as well as several novel QTLs in this study. We also identified distinct genetic regulatory mechanisms for tiller angle between the two subpopulations, providing information on how to improve tiller angle in indica and japonica rice.

Results
Phenotypic variation of the tiller angle in rice during heading stage  (Fig 1a). The variations observed in the two environments showed a similar distribution, skewed towards smaller tiller angles. A significant correlation was observed between the two environments in the whole population (r = 0.66). However, there was a significant difference in tiller angle between the indica and japonica subpopulations (Fig 1b and 1c). On average, indica rice exhibits a larger tiller angle (11.7 ± 5.8°in Hainan; 10.5 ± 5.6°in Wuhan) than japonica rice (8.8 ± 3.6°in Hainan; 9.1 ± 3.6°in Wuhan), and the variation within indica is greater than in japonica. The correlation coefficients between Hainan and Wuhan were 0.71 and 0.49 within the indica and japonica subpopulations, respectively. Two-way analysis of variance (ANOVA) revealed that tiller angle was dominantly controlled by genetic factors but was also influenced by interactions between genotype and environment (Table 1). In the japonica subpopulation in particular, the interaction between genotype and environment accounted for 22.6% of the variation in tiller angle (Table 1). Tiller angle had a high heritability of 0.82.

QTLs commonly identified in the two environments
We performed GWAS separately in the whole population and in the indica and japonica subpopulations for each year. Manhattan plots and quantile-quantile plots of the rice tiller angles among the three populations are illustrated with the results obtained from both the linear mixture method (LMM) (S1 Fig) and linear regression (LR) approaches (S2 Fig). A total of 30 tiller angle-related QTLs were detected (Table 2 and S1 Table and S2 Table). Of them, seven were commonly detected in Wuhan and Hainan. Three QTLs (qTA1b, 3 and 7a) were only identified  through LR, while the remaining was identified through both LMM and LR ( Table 2). Among these seven QTLs, 3 (qTA1b, 7a and 8b) were detected only in the whole population, 2 (qTA3 and 7b) were identified only in the indica subpopulation, and two QTLs (qTA8a and 9c) were commonly detected in the whole population and the indica subpopulation. No significant association signals were commonly detected in the japonica subpopulation in the two environments (Table 2). Two QTLs (qTA1b and 3) on chromosomes 1 and 3 individually explained more than approximately 15% of the variation in the whole population and in the indica subpopulation, respectively. Two QTLs (qTA8a and 8b) on chromosome 8 presented different contributions to the tiller angle in the two environments. The QTL of qTA8a exhibited the major effect in the indica subpopulation.

Associations uniquely identified in one environment
In addition to these 7 QTLs commonly detected in both environments, 10 and 13 QTLs were specifically detected in Hainan (S1 Table) and Wuhan (S2 Table), respectively. In Hainan, 5, 4 and 3 QTLs were detected in the whole population and in the indica and japonica subpopulations, respectively. Two QTLs (qTA7d and qTA12a) were commonly detected in the whole population and in the indica subpopulation. Three QTLs were located on chromosome 7. In Wuhan, 4, 3 and 6 QTLs were detected in the whole population and in the indica and japonica subpopulations, respectively. However, no QTLs were commonly detected in either population. There were 3 QTLs on each of chromosomes 1 and 7.
Co-localization of associated sites with previously reported QTLs for the tiller angle In the past decade,~11 genes were reported to control tiller angle in rice [32]. However, in the present study, only TAC1 was detected in the local LD region via GWAS in both Hainan and Wuhan (Table 2). To evaluate the results of GWAS, we compared the localization of associated sites with those 11 tiller angle QTLs detected in cultivated rice from the gramene web site (http://www.gramene.org) and 14 significant tiller angle loci detected via GWAS in the previous study [30]. A total of 7 associated sites were found to co-localize with 6 previously reported QTLs (S3 Table). Of them, four QTLs were commonly identified in both environments. qTA1b and qTA9c were found in the regions of QTa1 and qTA-9a, respectively; Both qTA8a and qTA8b were located in a large QTL interval covered a 4-Mb genome region [8,30]. qTA4 and qTA7h detected only in Wuhan and qTA9b detected in Hainan were identified in QTL regions. The co-localization of QTLs detected through linkage analysis and GWAS indicated their roles in controlling tiller angle.
TAC3, the candidate gene of qTA3, controls tiller angle The lead SNP sf0329582676 of qTA3 on chromosome 3, which was specifically identified in indica rice, is located in the second intron of Os03g51670, 5 kb upstream of Os03g51660 (Fig  2b). We quantified all 49 SNPs between Os03g51660 and Os03g51670 in 295 indica accessions, and a representation of the obtained pairwise r 2 values showed that some SNPs in both genes were in high linkage disequilibrium (LD) with each other. SNPs in the 5' end of Os03g51660 and Os03g51670 were grouped into one LD block, which made it difficult to determine the candidate gene of qTA3 (Fig 2a). A mutant (05Z11AZ62) was subsequently obtained in which a T-DNA was inserted in the 3'UTR (441 bp from stop codon) of Os03g51670 and the promoter (991 bp from start codon) of Os03g51660 (Fig 2b and 2c); this mutant showed a larger tiller angle compared with the wild type at both tillering and heading stages in 2015 and 2016 of Wuhan (Fig 2g-2i). A segregation analysis of a family of 96 plants (chi square = 3.56) indicated that a single gene controlled the tiller angle. The average tiller angle of homozygous mutants was the largest, while that of heterozygous mutants was intermediate, and wild type plants possessed the smallest tiller angle at both stages in two years (Fig 2g). An expression profiling analysis indicated that Os03g51660 was highly expressed in the tiller base, and the constitutive expression of Os03g51670 (S3d and S3e Fig) implied that Os03g51660 was the candidate gene controlling the tiller angle. We investigated the expression level of Os03g51670 and Os03g51660 between mutants and wild type. The expression level of Os03g51670 was significantly reduced only in the homozygous mutant and not in the heterozygote compared with wild type (Fig 2f). However, the expression level of Os03g51660 was significantly increased in the homozygous mutants and the heterozygotes compared with wild type (Fig 2e). That is, the Os03g51660 expression level co-segregated with the tiller angle. In addition, Os03g51660 was preferentially expressed in the tiller base, where the tiller bud is initiated and outgrown. Therefore, Os03g51660 is the gene underlying qTA3. Hereafter, this gene is referred to as TAC3, and this mutant is named as tac3-1D. To provide strong evidence for our results, another two mutants 1B-24636 (tac3-2D) with a T-DNA insertion in the 5'UTR (146 bp from start codon) of Os03g51660 and 4A-02006 with a T-DNA insertion in the first intron of Os03g51670 ( Fig  2b) were investigated in 2016. The expression level of Os03g51660 in both homozygous and heterozygous tac3-2D mutants significantly increased compared with that in wild type Dongjing (Fig 2j). The average tiller angle was no significant difference between heterozygous tac3-2D and wild type Dongjing at both stages, but the tiller angle of homozygous mutant (12.3°± 2.9°tillering stage and 11.4°±3.8°heading stage) was significantly larger up to 3°than DJ (9.0°±2.9°tillering stage and 8.6°±2.0°heading stage) (Fig 2l-2n). Whereas the expression of Os03g51670 was not changed in tac3-2D as compared to wild type (Fig 2k). Although expression of Os03g51670 was significantly decreased in both homozygous and heterozygous 4A-02006 mutants (S3a and S3b Fig), tiller angle of mutants was not significantly changed (S3c Fig). Therefore Os03g51660 is the identity of TAC3 controlling tiller angle.

Characterization of TAC3
The genomic DNA sequence of TAC3 is 1,717 bp in length, with five exons and four introns (Fig 2b). Its coding sequence is 459 bp in length (S3f Fig), and it encodes a conserved hypothetical protein of 152 amino acids (http://rice.plantbiology.msu.edu/).
We detected 3 major haplotypes based on 10 SNPs (Minor Allele Frequency (MAF) ! 0.05) in TAC3 among the 295 indica accessions ( Table 3). The tiller angle of Hap1 was significantly wider than that of Hap2 and Hap3 in both environments (p 0.01). However, only 1 G/T SNP (sf0329577726) was found in the japonica subpopulation, and all but two accessions carried the 'T' allele, which indicated that a major allele dominated in japonica; by contrast, the 'G' allele was present in all but 10 indica accessions. Nucleotide diversity analysis showed that both TAC3 (π = 1.9e-5) and its surrounding genomic region (π = 1.1e-4) of 100 kb upstream and downstream presented significantly decreased values compared with the average nucleotide diversity across the whole japonica genome (π = 1.45e-3), indicating that TAC3 was selected during japonica domestication and genetic improvement (Table 4).

TAC1, the candidate gene of qTA9c
qTA9c was closely linked to TAC1, a previously reported rice tiller angle-related gene [11], and its lead SNP was located 1.4 kb upstream to TAC1. Assessment of the LD between the lead SNPs and all polymorphic sites in TAC1 showed that the lead SNP sf0920735688 (P LMM = 4.2e-14) was in high linkage disequilibrium with most polymorphic sites in TAC1 containing GGGA/AGGA (sf0920731363) sites, which are known functional nucleotide polymorphic sites (FNPs) of TAC1 (Fig 3). Therefore, TAC1 was assumed to be the candidate gene of qTA9c. We further investigated a total of 13 SNPs (MAF!0.05) throughout TAC1 among 529 O. sativa accessions, and 16 haplotypes were detected. The vast majority (285 indica and 156 japonica accessions) of the accessions belonged to Hap1, Hap2 and Hap3 (Table 5). All but one of the japonica accessions were classified into Hap3 with the 'GGGA' allele. Most of the indica accessions fell into Hap1 and Hap2, which both contain the ' AGGA' allele, whereas 35 indica accessions carried Hap3. There was a highly significant difference (p<0.01) in the tiller angle between Hap1 and Hap3 in the indica subpopulation in both environments (Table 5). These results explained why TAC1 was detected in the whole population and the indica subpopulation, but not in the japonica subpopulation. The nucleotide diversity of TAC1 and its flanking region was subsequently analyzed. Accordingly, a significant reduction in nucleotide diversity was observed at TAC1 in the japonica subpopulation (π = 9.4e-5) compared with both wild rice (π = 1.45e-3) and the average diversity of the whole japonica genome (π = 1.45e-3) ( Table 4). We also observed a lower level of nucleotide diversity (π = 1.7e-4) in the 100-kb region surrounding TAC1 in the japonica subpopulation compared with the whole japonica genome. However, no obvious changes were found in the indica subpopulation (π = 3.0e-3).

D2, the candidate gene of qTA1b
Brassinosteroid, as one of the main plant hormones, is considered to be an important factor in plant development, including the tiller angle. The lead SNP sf0105241388 of qTA1b is located in the last intron of DWARF 2 (D2). A representation of the obtained pairwise r 2 values showed that all SNPs in D2 were in one LD block (Fig 4a). The d2-1 and d2-2 mutant shows a brassinosteroid-deficient phenotype including an erect leaf angle and a short plant height, but the tiller angle of this mutant was not mentioned [33]. Thus, the tiller angle of a mutant we named d2-3 (04Z11MY27) with a T-DNA insertion in the third intron of D2 was investigated (Fig 4b).
To better understand the natural variation of D2, we further analyzed its haplotypes. We constructed the haplotypes of D2 within two subpopulations based on the SNPs with an MAF ! 0.05. In the japonica subpopulation, we obtained four main haplotypes excluding the lead SNP sf0105241388 because all but one accession carried the same allele, ' A' , at this site. Most of the accessions (112) fell into one haplotype. Only 13, 11 and 10 accessions fell into the other three haplotypes. The tiller angle showed no significant differences among these haplotypes ( Table 6). In the indica subpopulation, we also obtained four main haplotypes that were different from those in the japonica subpopulation (Table 7). Hap1 and Hap4 exhibited significantly smaller tiller angles compared with Hap2 and Hap3 (p 0.01). Hap1 and Hap4 carried the allele ' A' at the lead SNP site, whereas Hap2 and Hap3 carried the 'G' allele. Significantly decreased nucleotide diversity (π = 1.6e-4) was observed only in the japonica subpopulation compared with the average nucleotide diversity of its whole genome ( Table 4).

The effect of three-gene combinations on the tiller angle
According to the results described above, it is clear that TAC3, D2 and TAC1 together, along with other genes, contribute to the natural variation in tiller angle in cultivated rice. Thus, the combinations of TAC3, D2 and TAC1 would be expected to result in wide variation of tiller angles in cultivars.
A total of 77 combinations of the genes TAC3, D2 and TAC1 were observed in the 295 indica accessions, but only 9 combinations that were each found in more than 10 accessions were used for the comparative analysis (Table 8). In Wuhan and Hainan, combinations 1 and 2 both pyramided three haplotypes that increased the tiller angle, resulting in the largest tiller angle. Combinations 3-7 with decreased tiller angle haplotypes of D2 and TAC3, independent of TAC1 showed similar compact plant architecture in both environments. Combination 8 pyramided increasing tiller angle haplotypes of D2 and TAC3, but a decreased tiller angle haplotype of TAC1 resulted in a large tiller angle. Combination 9 carried increasing tiller angle haplotypes of TAC1 and TAC3 but a decreasing tiller angle haplotype of D2, and accordingly exhibited a large tiller angle as well. These results indicated that D2 and TAC3 were the major QTLs in the indica cultivars. Among all 295 indica accessions, 5 carried two combinations with decreasing tiller angle haplotypes at three gene loci. Accessions W066 and W155 (TAC3-Hap3/D2-Hap1/TAC1-H ap3) showed more compact tiller angles (W066: 3.8°in Hainan and 3.5°in Wuhan; W155: 6.0°in Hainan and 5.5°in

Diverse genetic basis of the tiller angle between the indica and japonica subpopulations
In general, indica rice exhibits a wide plant type, and japonica rice exhibits a compact plant type. In this study, the indica subpopulation presented wider variation in tiller angle compared with the japonica subpopulation (Fig 1). Although genetic effects explained more of the variation in tiller angle (Table 1), Genotype by environment interactions also significantly affected tiller angle. Moreover, genetic factors contributed more to the variation in tiller angle in indica than in japonica. These results indicated that there are probably different genetic bases underlying the tiller angle in the two subpopulations. Coincidently, GWAS identified more QTLs for the tiller angle in the indica subpopulation than in the japonica subpopulation in both environments. This result indicates that the greater number of QTLs controlling the tiller angle in indica rice would contribute to wider natural variation. Although a dozen QTLs were identified in each subpopulation, there were no common QTLs detected in both subpopulations in this study. Two QTLs, qTA8a and qTA9c, were found in both the whole population and the indica subpopulation, but not in the japonica subpopulation. It is likely that both QTLs were fixed for one major haplotype in the japonica accessions, resulting in failed detection. For example, the locus qTA9c/TAC1 was not identified in japonica because TAC1 is dominantly fixed with Hap3 in this subpopulation ( Table 5). The tiller angle associated with Hap3 was significantly reduced compared with that of Hap1, the major haplotype in indica, which explains the compact plant architecture of japonica rice. There was only one SNP observed in qTA3/TAC3 in japonica, and all but two accessions carried the dominant allele. Thus, the japonica accessions were almost fixed with one haplotype for TAC3. Accordingly, the haplotype analysis of the full population suggested the existence of indica-japonica differentiation for D2: no significant differences in tiller angle were observed among the four main haplotypes within the japonica subpopulation (Table 6), while there was a significant difference in the indica subpopulation (p 0.01) ( Table 7). This suggested that the major QTLs are fixed in japonica. Two QTLs, qTA7a and qTA8b, were only identified in the whole population and were not identified in either the indica or japonica subpopulation. A possible explanation for this finding is that the gene was fixed with different haplotypes in the two subpopulations. When GWAS was performed in each subpopulation, the QTLs were not detected because no polymorphism was present in sequence; however, when GWAS was conducted in the whole population, these QTLs were identified because variation occurred. Hence, it is recommended that GWAS should be performed separately in different populations to discover more QTLs for tiller angle or other traits. According to the above results, we propose that there is diverse genetic basis controlling the tiller angle between the two subpopulations.  The SNP in bold is the lead SNP of qTA1b. The P value of Duncan test is less than 0.01. doi:10.1371/journal.pgen.1006412.t007 TAC1, TAC3 and other genes, regulate tiller angle in rice cultivars rather than PROG1 and LA1 Tiller angle is a domestication-related trait in rice, and selective signatures of domestication include a reduction of nucleotide diversity and altered allele frequencies at domesticationrelated loci [34,35]. The functional SNPs of PROG1, a gene controlling prostrate growth, only exist in wild rice [9,10]. In addition, PROG1 was detected via GWAS in 446 O. rufipogon accessions and was screened for strong selection signals [36]. In this study, PROG1 was not identified in the examined cultivars. LA1 is another major gene isolated from a mutant that controls the tiller angle, and loss of function of LA1 destroys rice shoot gravitropsim through altering the polar transport of auxin [19]. The lazy1 mutant shows a wider tiller angle of more than 60°, especially during the mature stage, associated with a prostrate growth phenotype [18], which was never observed in our collection. Therefore, la1 and PROG1 do not contribute to the variation of tiller angle in our cultivars. In this study, GWAS and a mutant analysis demonstrated that TAC3 regulates tiller angle. The nucleotide diversity of TAC3 in japonica (π = 1.9e-5) was approximately 55-fold lower than that of 111 randomly chosen gene fragments from japonica (π = 1.1e-3) [37], which is equivalent to the japonica whole genome estimation in our global collection (π = 1.45e-3). This result suggested that the low TAC3 nucleotide diversity observed in japonica cannot be explained by a population bottleneck alone and indicated that TAC3 was strongly selected probably due to its function in controlling tiller angle during the domestication and improvement of japonica. Accordingly, GWAS, a mutant analysis and nucleotide diversity analysis demonstrated that D2 regulates the tiller angle. Thus, TAC3 and D2 contributed to the natural variation in the tiller angle and have been subjected to selection in japonica. Interestingly, the major TAC1 haplotype in japonica, Hap3, was carried by only a small subset of indica accessions and probably introgressed from japonica. Our study on the selection of TAC1 was consistent with the previous report [38]. qTA8a and qTA8b were approximately 400 kb away on chromosome 8 and were detected via both the LMM and LR approaches. Interestingly, a locus (marker seq-rs3945) associated with tiller angle at tillering stage in indica rice was fallen into the region of qTA8a and several candidate genes on this QTL were predicted by expression analysis [30]. Thus, the 400-kb region containing qTA8a and qTA8b should be further studied Table 8. The performance of tiller angle for three-gene combinations in indica subpopulation.

Combinations Haplotypes No of accessions Tiller angle (˚)
by developing an indica bi-parental mapping population. Additionally, there were several QTLs that co-localized with previously reported QTLs, while some QTLs were novel. Hence, TAC3, D2, TAC1 and other unknown genes contribute to the tiller angle variation observed in the cultivars.

Favorable haplotypes/alleles for the improvement of tiller angle in rice
Rice tiller angle is a major component of plant architecture that has been subject to selection from both nature and humans over a long time period. Mining of more favorable alleles of tiller angle genes is required to achieve ideal plant architecture in rice. At the single-gene level, we identified favorable haplotypes for three tiller angle genes. Hap2 (represented by Yuexiangzhan) and Hap3 (represented by 9311) of TAC3 decrease tiller angle, as does Hap3 (represented by Zhenshan 97 in indica and Zhonghua 11 in japonica) of TAC1. Hap1 (represented by Minghui 63 and 9311) and Hap4 (represented by IR72) of D2 also decrease tiller angle. A threegene combination analysis of D2, TAC3 and TAC1 in indica showed that these three genes cause wide variation in the tiller angle, ranging from 5.2°to 15.4°on average, and combinations pyramiding decreased tiller angle haplotypes result in a more compact plant architecture. These results indicated that these genes function in the regulation of tiller angle and that their effects could be additive. Therefore, optimized haplotype-combinations among these three genes could serve as targets for designed breeding. More specifically, most of the indica cultivars carried TAC1 haplotypes that increased tiller angle; thus, the TAC1 haplotype Hap3 is the first option for improving plant architecture in indica. As most japonica cultivars carried compact haplotypes for TAC1, TAC3 and D2, selecting for the haplotypes of other unknown genes that decrease the tiller angle would be the first option for improving the tiller angle in japonica.
In summary, the tiller angle of cultivated rice is mainly controlled by genetic factors but is also affected, to some extent, by interactions between genotype and environment. The genetic basis of tiller angle is diverse between indica and japonica rice. TAC3, D2 and TAC1 were found to be the main factors regulating tiller angle. Introgression between two subpopulations would be an efficient means of optimizing the plant architecture through designed molecular breeding. Two-way analysis of variance and heritability Two-way analyses of variance were separately used to test significant difference between environments and genotypes for the whole population and two subpopulations. The analysis was run in the program Statistica 7.0 (StatSoft. Tulsa, OK, USA). Broad-sense heritability (H 2 ) of rice tiller angle in the whole population was calculated based on the experiments using the formula:

Plant materials and field experiments
e and d 2 ge were the estimates of genetic, genotype by environment and error variances derived from the mean square expectations of twoway analysis of variance (ANOVA), respectively; n was the number of environments and r was the number of replicates.

Linkage disequilibrium and haplotype analyses
Linkage disequilibrium (LD) was investigated based on standardized disequilibrium coefficients (D'), and squared allele-frequency correlations (r 2 ) for pairs of SNP loci were determined using the TASSEL5.0 program. The extent of genome-wide and chromosome-wide LD were recently reported [40], and the average distances of LD decay at the genome-wide level in all, indica and japonica populations were 167 kb, 93 kb and 171 kb, respectively. The distances of LD decay in the regions surrounding lead SNPs identified in this study were calculated as below: First, r 2 values were calculated between lead SNP and all SNPs in its upstream and downstream 2 Mb regions. Then averaged r 2 of the top ten percent of r 2 values in the region from 1.5 Mb to 2 Mb away from lead SNP were taken as background r 2 . Finally the LD region was defined a continue region where r 2 was 0.2 larger than background r 2 . LD plots were generated with Haploview4.2, and LD is indicated using r 2 values between pairs of SNPs multiplied by 100; white, r 2 = 0; shades of gray, 0<r 2 <1; black,

Genome-wide association analyses
The whole population was previously demonstrated to present a distinct population structure [31]. The indica and japonica subpopulations and the whole population were subjected to GWAS separately because they presented sample sizes of greater than one hundred. A total of 3,916,415, 2,767,159 and 1,857,845 SNPs (minor allele frequency (MAF) ! 0.05; the number of accessions with minor alleles ! 6) were employed for GWAS using the linear mixture method (LMM) and linear regression (LR) method in the FaST-LMM program [44]. The population structure of Q matrix and kinship (K matrix) was taken into account as cofactor when performing association mapping using the LMM method. The effective numbers of independent SNPs and suggestive thresholds were calculated using a method described by Li et al. [45], and 757,578, 571,843 and 245,348 effective independent SNPs were found in the whole population and the indica and japonica subpopulations, respectively. The suggestive p values used as thresholds for the significance of association signals that were commonly detected in both environments by LMM were 1.3×10 −6 for the whole population, 1.8×10 −6 for indica and 4.1×10 −6 for japonica. However, for significant association signals that were only detected in one environment, we utilized the more stringent p value of 6.0×10 −7 as the threshold. A suggestive p value of 1.0×10 −8 was employed as the threshold for the significance of association signals detected by LR, but only the top 5 loci detected by LR in each environment are presented in the results. The loci that were commonly identified by LR in both environments, but were not in the top 5 are also presented. For loci that were commonly detected by two methods, only the results of LMM are presented. To obtain independent association signals, multiple SNPs exceeding the threshold in a 5-Mb region were clustered based on an r 2 of LD ! 0.25; the SNPs showing the minimum p value in a cluster were considered to be lead SNPs.

Nucleotide diversity analyses
The whole genomic DNA sequences of the 529 cultivar accessions were genotyped with approximately 2.5× coverage, and the genome was sequenced using a bar-coded multiplex sequencing approach on an Illumina Genome Analyzer II [31]. We obtained the genome sequences from RiceVarMap (http://ricevarmap.ncpgr.cn/) [39]. A total of 446 O. rufipogon accessions were used to calculate nucleotide diversity (π). The details of these accessions and their sequencing data have been previously reported [36]. π values were estimated at the whole-genome level, the single gene level, and the 100-kb flanking region level using SAMtools [46].

Genotyping of mutant plants
The d2-3 (04Z11MY27), tac3-1D (05Z11AZ62), tac3-2D (1B-24636) and 4A-02006 mutants were obtained from rice T-DNA insertion libraries from the ZH11 variety [47,48] and DJ variety [49,50]. We identified the genotypes of the mutants via PCR using the genomic primers L and R and the vector primer N (S4 Table). PCR was conducted with an initial incubation step at 95°C for 5 min; a second step of 35 cycles at 95°C for 30 s, 58°C for 30 s, and 72°C for 1 min; and a final extension at 72°C for 7 min.

RNA extraction and expression analysis
Total RNA was extracted from different plant tissues with an RNA extraction kit using TRIzol reagent (Invitrogen) for quantitative real-time reverse transcription-polymerase chain reaction (qRT-PCR). The total RNA (4g) was reverse-transcribed using M-MLV reverse transcriptase (Invitrogen). qRT-PCR was carried out in a total volume of 10 μl containing 2.5 μl of the reverse-transcribed products, 0.25 μM gene-specific primers and 5 μl of Fast Start Universal SYBR Green Master (Rox) superMIX (Roche, Mannheim, Germany) in a QuantStudio (TM) 6 Flex System, according to the manufacturer's introductions. Measurements were obtained using the relative quantification method. Expression levels were normalized against expression of an ubiquitin (UBQ) gene. Error bars indicate standard deviations (n = 3). All primers employed for qRT-PCR are listed in S5 Table. Supporting Information