Pair-wise Comparison Analysis for Multiple Pool-seq: an efficient method identified anthocyanin biosynthesis genes in rice pericarp

The complex traits are derived from multiple genes and exhibit a large variety of phenotypes. High-throughput sequencing technologies have become the new strategies for mapping the important traits of crops. However, these methods have their own disadvantages and limitations. Here we introduced Pair-wise Comparison Analysis for Multiple Pool-seq (PCAMP) for mapping the candidate genomic regions involved in anthocyanin biosynthesis in rice pericarp. In this protocol, the second filial generation (F2) populations obtained by crossing two parents with different target traits were divided into n (n>=3) subpopulations according to their phenotypes. Thirty phenotypically identical individuals were selected from each subpopulation and DNA samples were extracted to form a pool for sequencing. Finally, we compared the SNP-index between every two Pool-seqs to map the candidate genomic regions. We applied PCAMP to analyse F2 populations and successfully identified five known genes and five new candidate genomic regions for anthocyanin biosynthesis in rice pericarp. These results demonstrate that PCAMP is an efficient new method for dissecting the complex traits of crops.

The complex traits are derived from multiple genes and exhibit a large variety of phenotypes. High-throughput sequencing technologies have become the new strategies for mapping the important traits of crops. However, these methods have their own disadvantages and limitations. Here we introduced Pair-wise Comparison Analysis for Multiple Pool-seq (PCAMP) for mapping the candidate genomic regions involved in anthocyanin biosynthesis in rice pericarp. In this protocol, the second filial generation (F2) populations obtained by crossing two parents with different target traits were divided into n (n>=3) subpopulations according to their phenotypes.
Thirty phenotypically identical individuals were selected from each subpopulation and DNA samples were extracted to form a pool for sequencing. Finally, we compared the SNP-index between every two Pool-seqs to map the candidate genomic regions. We applied PCAMP to analyse F2 populations and successfully identified five known genes and five new candidate genomic regions for anthocyanin biosynthesis in rice pericarp. These results demonstrate that PCAMP is an efficient new method for dissecting the complex traits of crops.
The world population is likely to grow by 60%-80% within the next 50 years, which requires the food production to nearly double, based on current living standards.
Due to global warming and the increasing reduction of arable land, increasing grain production is a great challenge (Abe et al., 2012). The complex traits of crops are important for adaption to environment and genetic improvement (Mitchell-olds, 2010), which are genetically controlled by multiple genes. Each gene has relatively minor effect on phenotype and is susceptible to environmental influences (Buckler et al., 2009). Therefore, it is quite difficult to dissect the genetic basis underlying the complex traits.
BSA (Bulked Segregant Analysis) is a approach for map inherited traits. Initially, BSA was used to identify molecular markers closely linked to target genes in lettuce (Michelmore et al., 1991) and tomato (Giovannoni et al., 1991). Subsequently, BSA combined with RFLP, RAPD, SSR and other molecular markers was used to reveal target genes in soybean (Darvasi a, 1994;Mansur et al., 1993). With the development of high throughout sequencing technology, Schneeberger et al. (Singh et al., 2017) identified the candidate gene AT4G35090 which leads to slow growth and light green leaves of Arabidopsis thaliana by SHOREmap technology. Austin et al. (Austin et al., 2011) invented a BSA method based on homozygosity mapping, which is mainly applied to map recessive traits. Ehrenreich et al. (Ehrenreich et al., 2010) and Wenger et al. (Wenger et al., 2010)successfully located target genes in yeast and Saccharomyces cerevisiae respectively. Recently, Abe et al. (Abe et al., 2012)used MutMap technology to locate rice green leaf mutant gene OsCAO1. Fekin et al. (Fekih et al., 2013) developed MutMap+ technology, which was applicable to early mutagenesis and lethal or non-inbred mutants. based on MutMap and de novo assembly, Takagi et al. (Takagi et al., 2013A)developed a new mapping approach, MutMap-Gap applicable to the mutation sites which were not found in the reference genome. In 2013, Takagi et al. (Takagi et al., 2013B) first used QTL-seq technology to locate rice blast and seedling vigor related genes. Since then, the BSA has become a common approach for locating genes.
With the application of QTL-seq technology in plant trait mapping, many new methods have been developed. Das et al.(Das et al., 2016) crossed a less pod number variety ILWC 46 with two more pod number varieties Pusa 1103 and Pusa 256 respectively to obtain the F2 population for mQTL-seq, and then narrowed the candidate region through taking the intersection of the two candidate intervals. Kumar et al. (Kumar et al., 2018) identified three genes related to Ascochyta blight resistance of chickpea based on mQTL-seq technology. The application of QTL-seq to some large genome, such as barley, is challenging, so Hisano et al. (Hisano et al., 2017)identified the monogenic Mendelian locus and associated QTLs using Exome QTL-seq. Xue et al. (Xue et al., 2017) modified the QTL-seq method by introducing a | ((SNP-index)| parameter to improve the accuracy of mapping the red skin trait in a group of highly heterozygous Asian pears. Yoshitsu et al.(Yoshitsu et al., 2017) constructed Y-bulk (early-heading), S-bulk (late-heading) and L-bulk (extremely late-heading) from F2 population of two foxtail millet cultivars lines, and then identified two QTLs associated with DTH by genome-wide comparison of SNPs in the Y-bulk versus the S-bulk and the Y-bulk versus the L-bulk. Huang et al. (Huang et al., 2017)identified a Clubroot resistance related gene in Chinese Cabbage using BSR-Seq technology. Dou et al.(Dou et al., 2018) obtained a watermelon shape related candidate region through GWAS analysis, and further validated and narrowed candidate interval based on BSA-seq and genetic linkage analysis. Singh et al. (Singh et al., 2017) constructed two mixed pools using the F2 population of ICPL 20096 (resistant to fusarium wilt and sterility mosaic disease) and ICP 332 (susceptible to FW and SMD), and located FW and SMD resistance related genes through indel-seq approach.
QTL-seq and other derivatives play an important role in plant trait mapping.
However, these methods only select plants with extreme phenotypes of target traits in the segregated population and identify one or two genes/QTLs associated with related traits. In crops, many agronomic traits are complex traits controlled by multiple genes.
The traditional method of quantitative trait analysis is only for single QTL and time-consuming and labor intensive cost (Tuberosa et al., 2005). GWAS analysis can be used to quickly identify complex traits related candidate intervals, but it has limitations (Platt et al., 2010;Farlow, 2013;Yano et al., 2016). In this study, we developed a Pairwise Comparison Analysis for Multiple Pool-seq, which is a powerful tool for cost-effective, rapid and efficient identification of genes related to anthocyanin biosynthesis in rice pericarp.

Phenotypic differences in second filial generation population and construction of bulks
In November 2016, when the rice grain was fully ripe and the moisture content reached 12.5%, grains were randomly selected and hulled, and then the seed coat color of F2 population was identified. The 381 lines developed from Huanghuazhan (HHZ) and Donglanmomi (DLMM) were divided into colored groups(including black, partial black and brown) and white group, and the number ratio was 285:96, which accorded with the separation rule of 3:1 (χ 2 = 0.0079, P > 0.05). In November 2017, the number ratio of colored group to white group was 601:195 for 796 F2 lines, which accorded with the separation comparison of 3:1 (χ 2 = 0.11, P > 0.05). Subsequently, the anthocyanin content of 601 colored pericarps and 30 white pericarps was measured by high performance liquid chromatography (HPLC). The results showed that the seed coats color was mainly caused by cyanidin content, and the content of cyanidin varied from 0.45 ug/g to 1616.03 ug/g. According to the seed coat color and anthocyanin content of 596 lines, the population was divided into four subgroups, and then 30 lines from each subgroup were selected to develop four bulks: white bulk (W), black bulk (B1), partial black bulk (B2) and brown bulk (B3)( Figure.1).

Mapping of reads to the reference genome and identifition of high quality SNPs
HHZ, DLMM, W, B1, B2 and B3 were sequenced using Illumina HiSeq X Ten sequencing platform. After filtration, the total base number of 6 bulks was 161.  (Table S1). According to the coverage depth of 12 chromosomes in rice (Figure. S1), the genomes of 6 samples were evenly covered and the sequencing randomness was good.
GATK (Mckenna et al., 2010A) (Table S2). SnpEff (Cingolani et al., 2012A) was used to annotate and predict the variation effects of these SNPs. The results showed that most of the variation sites were located in upstream region and downstream region of the genes, which may affect the gene function.
Before association analysis, SNP was filtered. The criteria were as follows: firstly, SNPs with multiple genotypes were filtered out; secondly, SNPs with read depth less than 4 were filtered out; thirdly, SNPs with identical genotypes between bulks and SNPs of recessive bulk which were not from recessive parent were filtered out. high-quality SNPs (Table S3).

Anthocyanin biosynthesis related genomic region in rice pericarp
To identify the candidate genomic regions responsible for anthocyanin biosynthesis in rice pericarp, we compared the SNP-index (Takagi et al., 2013C) (Table S9).
The distribution SNP-index and △(SNP-index) ( Figure. 4) of W-B3 showed that the regions showing significant association with the genes involved in anthocyanin biosynthesis in rice pericarp were mapped to 25.15-31.14 Mb interval on chromosome 4 (Table S10) and 7.11-9.80 Mb interval on chromosome 9 (Table S11). Mb intervals on chromosome 6 ( Table S13).
For the candidate genomic regions with overlapping physical positions on the same chromosome, the intersection regions were selected as the final candidate regions.
Therefore, the regions showing significant association with genes involved in anthocyanin biosynthesis in rice pericarp were shown in Table S18.

Candidate genes for anthocyanin biosynthesis in rice pericap
We classified the candidate genomic regions into two groups: (I) the regions that was adjacent to or contained the cloned genes related to anthocyanin biosynthesis genes; and (II) the remaining candidate regions.
(I): The Rd is located at 1.19 Mb upstream of the candidate region on chromosome 1. Furukawa et al. (Furukawa et al., 2007) found that Rd was involved in the proanthocyanidin biosynthesis of rice pericarp. The expression level of Rd between HHZ and DLMM was significantly different ( Figure. 8a). The sequences of HHZ and DLMM were amplified with PCR primer AB003495 (Konishi et al., 2008) and the products were sequenced (Table S19). The 43rd base of the second exon of the Rd of HHZ was changed from C to A. This change belongs to the Rd2 mutant genotype (Konishi et al., 2008) Druka et al. (Druka et al., 2003)  Hu et al. (Hu et al., 1996) identified the Ra gene using the maize homologous sequences within 25.17-31.14 Mb interval on chromosome 4, Ra gene encodes the basic helix-loop-helix (bHLH) transcription factor, which plays a regulatory role in the anthocyanin synthesis pathway. Subsequently, Hu et al. (Hu et al., 2000) indicated that Ra was consisted of Ra1 and Ra2. Recently, Oikawa et al. (Oikawa et al., 2015) successfully cloned Kala4, a key gene responsible for anthocyanin accumulation in rice pericarp, which was found to be the same gene as Ra2. The expression levels of Kala4 between HHZ and DLMM were significantly different ( Figure. 8d). The DNA samples of HHZ and DLMM were amplified by functional primers (Table S19). The results showed that the promoter region of Kala4 in DLMM had a genomic fragment inserted ( Figure. 8e), and this change was the causes of generation of the black rice traits (Oikawa et al., 2015).
At the 0.24 Mb upstream of the candidate region on chromosome 9, Shao et al. (Shao et al., 2012) found that inhibitor for brown furrows1 1 (IBF1) might be involved in the accumulation of anthocyanins in hull, but there was no significant difference in IBF1 expression level between HHZ and DLMM. However, the expression levels of LOC_Os09g15550, LOC_Os09g15570 and LOC_Os09g15590 ( Table S20). The expression levels of LOC_Os02g49140 between HHZ and DLMM were significantly different ( Figure. 8i), and this gene encodes glycosyltransferase.
Previous studies have shown that glycosylation modification of anthocyanin affects its stability in cells (Fukuchi-mizutani et al., 2003).

Discussion
The SHOREmap (Schneeberger et al., 2009) GWAS can be used to map complex plant traits and quickly identify genetic loci controlling target traits. However, GWAS is applicable to natural population with a large sample size leading to high cost and it is also difficult to detect the rare mutations and minor effective genes (Platt et al., 2010;Farlow, 2013;Yano et al., 2016). Family-based QTL mapping plays an important role in the mapping of complex plant traits. This method can identify major and minor loci related to target traits, but the population size is large and multiple generations required to develop pedigrees (Mitchell-olds, 2010).
PCAMP technology divides the segregated population into n subgroups (n ≥ 3) according to the phenotype of the target traits. For every subgroup, the DNA of individual selected from the subgroups was pooled. High throughput sequencing technology is used to sequence the bulk. Then, combination of every two bulks is respectively compared to obtain target traits related QTLs. Using PCAMP technology, we successfully identified 10 candidate intervals related to anthocyanin synthesis in rice seed coat. Among the 10 candidate intervals, Rd (Furukawa et al., 2007), OsCHI (Hong et al., 2012;Druka et al., 2003), and Kala4 (Oikawa et al., 2015) have been cloned, and two regions overlapped with that identified by provious GWAS study (Yang et al., 2018). These results validated the reliability of the method.
The candidate regions on chromosome 4 related to anthocyanin synthesis in rice seed coat could be detected by PCAMP in W-B1, W-B2 and W-B3 combinations..
Finally, we selected three candidate regions that are shared for the real candidate regions, and the candidate gene Kala4 was just located in this region (Oikawa et al., 2015). Likewise, the overlapping regions of candidate regions B1-B2, B2-B3, and B1-B3 on chromosome 3 were used and the candidate region was finally determined to be 17.22-20.86 Mb interval and 20.94-21.02 Mb, which are consistent with our previous study (Yang et al., 2018). The final interval 2.76-5.46 Mb interval on chromosome 12 is only 17.7% of the maximum candidate interval. Therefore, the candidate region of the target gene can be narrowed down by using the PCAMP.
Kala4 can be identified by PCAMP in the combinations of three colored pools and white pool, which plays an important role in anthocyanin synthesis in rice seed coat. However, when the gene kala4 exists, anthocyanin can not be synthesized regardless of the genotype of other genes, which is consistent with the previous report (Maeda et al., 2014) . In the combinations of three colored pools, 17.22-20.86 Mb and 20.94-21.02 Mb of chromosome 3 were identified, which indicated that there were at least two key genes related to anthocyanin synthesis. In the previous study, we used 419 Core Germplasms of rice landraces in Guangxi for GWAS and identified that this region was an important candidate region for anthocyanin synthesis genes in rice pericarpdy (Yang et al., 2018). The Rd was located at 1.19 Mb upstream of the candidate region. What is the reason for this? In this study, we found that the four Both LOC_Os03g18030 and LOC_Os03g32470 encode for leucoanthocyanidin dioxygenase, but their expression levels are significantly differed between HHZ seeds and DLMM seeds, especially LOC_Os03g32470.
After the synthesis of anthocyanin, if it stays in cytoplasm, its high biochemical reactivity will cause toxicity to cells, and anthocyanin itself may also be oxidatively High-throughput sequencing was used to analyze the induced mutant Red-1, and the result showed that the BGIOSGA033874 gene, which was related to flavonoid synthesis pathway in rice, had the same locus as OsCHS1 (Cheng et al., 2014).
LOC_Os12g07690 encodes for Chalcone synthase, which may be involved in flavonoid biosynthesis.

Materials and methods
Plant materials. DLMM (rank No: C133) is the core germplasm of rice landraces in Guangxi (Yang et al., 2018). Its pericarp is black and the anthocyanin content is high.

Determination of pericarp color and anthocyanin content.
After the rice seeds were fully matured and dried to a water content of 12.5%, the whole kernels were randomly selected for shelling, and the pericarp colors of the individuals in the two parents and the F2 population were identified according to the method described by Han & Wei(Han et al., 2006). The content of anthocyanin in brown rice was determined by HPLC method (Beta et al., 2009), which mainly included geranium pigment, morning glory pigment, delphinidin pigment, peony pigment, cyanidin and mallow pigment.  (Murray et al., 1980) to form a DNA pool.

Construction of Genomic
Illumina sequencing. After concentrations of DNA samples of 2 parents and 4 mixed pools were measured, the qualified DNA samples were randomly broken into 350 bp fragments by ultrasonic disruption, and the DNA fragments were terminated by end repair at the 3' end with ploy A, plus sequencing linker, purification, and PCR amplification. A sequencing library was constructed. After being passed the quality check, the library was sequenced by Illumina HiSeq X Ten.

Alignment with reference genome and SNP annotation. The reads obtained by
re-sequencing need to be relocated to the rice reference genome (http://rice.plantbiology.msu.edu/) for subsequent variation analysis. The short sequences obtained by sequencing were aligned with the reference genome using BWA software . After reading the reference genome, reads can count the coverage of the base on the reference genome. SNPs were detected using GATK software (Mckenna et al., 2010B). According to the positioning results of Clean Reads in the reference genome, Picard was used for De-duplication (Mark Duplicates), GATK was used for Local Realignment, Base Recalibration, etc., and then GATK was used for single core. After detection of single nucleotide polymorphism (SNP) and filtering, a final set of SNP sites was obtained. Annotation variation and predictive variation effects were performed using SnpEff software (Cingolani et al., 2012B). Identification of candidate gene. We selected a genomic region above the threshold corresponding to the 95% or 99% confidence interval as a candidate region for the target gene. The same candidate regions appear in the results of different ΔSNP-indexes, and the intersection of these regions was selected as the final candidate region. In the genomic candidate region, it is preferred to use the rice genome annotation site MSU-RGAP (http://rice.plantbiology.msu.edu/) to annotate the genes in the candidate region, and the genes related to plant anthocyanin synthesis (Zhu et al., 2015); secondly, the candidate region containing candidate gene or adjacent to the known gene related to anthocyanin synthesis in rice were taken as a candidate gene; finally, the homozygous variation site with a sequencing depth of not less than 4× and ΔSNP-index=1 was selected as a candidate gene. At the same time, by referring to the method of Yano et al. (Yano et al., 2016), all the polymorphisms in the candidate region were classified into three types. TypeⅠincluded the SNPs that were predicted to induce amino acid exchange or to change splicing junctions; TypeⅡ included SNPs that were located at the 5'-flanking sequences of genes or the up-streams region containing promoter region; and Type Ⅲ included SNPs that were located in a coding region but were not predicted to change an amino acid, an intron or a 3'-noncoding sequence. DNA sequencing. After the PCR product was recovered and homozygous, the ABI3730XL (Applied Biosystems, USA) sequencer was used for bidirectional sequencing with both positive and negative primers. DNA sequences were obtained and analyzed using ClustalX software.
Agarose assay. The PCR product was assayed for the target fragment using a 1.5% agarose gel and visualized using a gel imaging analysis system (Biosens SC 850, Shanghai, China). and Os03g0234200 (Sun et al., 2018) were used as the internal reference genes for normalization of expression levels of candidate genes. The qRT-PCR primers are presented in Table S19. The qPCR reaction system was set as follows: 10 μl of SYBR   Figure 1 The phenotype and anthocyanin content in four pools of F2 population. Black pericarp pool (B1), partial peicarp black pool (B2), brown pericarp pool (B3), white pericarp pool (W). Scale bars, 1.2 mm.

Figure 2
The PCAMP approach for mapping genomic regions controlling anthocyanin biosynthesisbetween W and B1. SNP-index plots of W (a) and B1 (b), △ (SNP-index) plot (c) of chromosome. The y-axis is the name of the chromosome, colored dots represent the calculated SNP-index, and the black line is the fitted SNP-index. The green line, blue line and red line represent the threshold of 90%, 95% and 99% confidence interval, respectively.

Figure 3
The PCAMP approach for mapping genomic regions controlling anthocyanin biosynthesisbetween W and B2. SNP-index plots of W (a) and B2 (b), △ (SNP-index) plot (c) of chromosome. The y-axis is the name of the chromosome, colored dots represent the calculated SNP-index, and the black line is the fitted SNP-index. The green line, blue line and red line represent the threshold of 90%, 95% and 99% confidence interval, respectively.

Figure 4
The PCAMP approach for mapping genomic regions controlling anthocyanin biosynthesisbetween W and B3. SNP-index plots of W (a) and B3 (b), △ (SNP-index) plot (c) of chromosome. The y-axis is the name of the chromosome, colored dots represent the calculated SNP-index, and the black line is the fitted SNP-index. The green line, blue line and red line represent the threshold of 90%, 95% and 99% confidence interval, respectively.

Figure 5
The PCAMP approach for mapping genomic regions controlling anthocyanin biosynthesisbetween B2 and B1. SNP-index plots of B2 (a) and B1 (b), △ (SNP-index) plot (c) of chromosome. The y-axis is the name of the chromosome, colored dots represent the calculated SNP-index, and the black line is the fitted SNP-index. The green line, blue line and red line represent the threshold of 90%, 95% and 99% confidence interval, respectively.

Figure 6
The PCAMP approach for mapping genomic regions controlling anthocyanin biosynthesisbetween B3 and B1. SNP-index plots of B3 (a) and B1 (b), △ (SNP-index) plot (c) of chromosome. The y-axis is the name of the chromosome, colored dots represent the calculated SNP-index, and the black line is the fitted SNP-index. The green line, blue line and red line represent the threshold of 90%, 95% and 99% confidence interval, respectively.

Figure 7
The PCAMP approach for mapping genomic regions controlling anthocyanin biosynthesisbetween B3 and B2. SNP-index plots of B3 (a) and B2 (b), △(SNP-index) plot (c) of chromosome. The y-axis is the name of the chromosome, colored dots represent the calculated SNP-index, and the black line is the fitted SNP-index. The green line, blue line and red line represent the threshold of 90%, 95% and 99% confidence interval, respectively.

Figure 8
The candidate genes of anthocyanin biosynthesis in rice pericarp. The expression level of 13 genes through qPCR (a, c, d, f-q). DNA sequencing of Rd and the 43rd base of the second exon is mutated from C to A (b). Amplification of Kala4 using functional primers (e). The expression level of LOC_Os12g07690 through RT-PCR (p).

Figure 9
The Rd is located within the selective sweep regions on chromosome 1. Figure S1 Genome-wide distribution of coverage depth. (a) Huanghuazhan, (b) Donglanmomi, (c) W, (d) B1, (e) B2, (f) B3. The x-axis is the chromosomal location, and y-axis is the value obtained by taking the logarithm (log2) of the depth of the corresponding position on the chromosome.

Figure S3
Distribution of SNPs on 12 chromosomes of Huanghuazhan and Donglanmomi. Figure S4 The difference in the anthocyanin synthesis between a and b is on the pericarp. (a) Gross morphology of Huanghuazhan and Donglanmomi at the seedling stage, (b) The panicle of Huanghuazhan and Donglanmomi. Scale bars, 2 cm.
Supplementary Tables  Table S1 Summary information for the genome resequencing statistics of six samples.

Table S2
Summary information for the SNPs annotation of six samples.

Table S3
The SNP statistics between any two pools.

Table S4
Rice anthocyanin biosynthesis candidate genomic region on chromosome 2 between W and B1 at the 95% confidence interval.

Table S5
Rice anthocyanin biosynthesis candidate genomic region on chromosome 3 between W and B1 at the 95% confidence interval.

Table S6
Rice anthocyanin biosynthesis candidate genomic region on chromosome 4 between W and B1 at the 99% confidence interval.

Table S7
Rice anthocyanin biosynthesis candidate genomic region on chromosome 3 betweenW and B2 at the 95% confidence interval.

Table S8
Rice anthocyanin biosynthesis candidate genomic region on chromosome 4between W and B2 at the 99% confidence interval.

Table S9
Rice anthocyanin biosynthesis candidate genomic region on chromosome 12 between W and B2 at the 95% confidence interval.

Table S10
Rice anthocyanin biosynthesis candidate genomic region on chromosome 4 betweenW and B3 at the 99% confidence interval.

Table S11
Rice anthocyanin biosynthesis candidate genomic region on chromosome 9 between W and B3 at the 95% confidence interval.

Table S12
Rice anthocyanin biosynthesis candidate genomic region on chromosome 3 between B2 and B1 at the 95% confidence interval.

Table S13
Rice anthocyanin biosynthesis candidate genomic region on chromosome 6 between B2 and B1 at the 95% confidence interval.

Table S14
Rice anthocyanin biosynthesis candidate genomic region on chromosome 1 between B3 and B1 at the 95% confidence interval.

Table S15
Rice anthocyanin biosynthesis candidate genomic region on chromosome 3 between B3 and B1 at the 99% confidence interval.

Table S16
Rice anthocyanin biosynthesis candidate genomic region on chromosome 3 between B3 and B2 at the 99% confidence interval.

Table S17
Rice anthocyanin biosynthesis candidate genomic region on chromosome 12 between B3 and B2 at the 95% confidence interval.

Table S18
The final candidate genomic regions of anthocyanin biosynthesis in rice pericarp.

Table S19
Primers used for this study.