Genetic loci regulating the concentrations of anthocyanins and proanthocyanidins in the pericarps of purple and red rice

The pigmented flavonoids, anthocyanins and proanthocyanidins, have health promoting properties. Previous work determined that the genes Pb and Rc turn on and off the biosynthesis of anthocyanins (purple) and proanthocyanidins (red), respectively. Not yet known is how the concentrations of these pigmented flavonoids are regulated in grain pericarps. Quantitative trait locus (QTL) analysis in a population of rice (Oryza sativa L.) F5 recombinant inbred lines from white pericarp “IR36ae” x red+purple pericarp “242” revealed three QTLs associated with grain concentrations of anthocyanins (TAC) or proanthocyanidins (PA). Both TAC and PA independently mapped to a 1.5 Mb QTL region on chromosome 3 between RM3400 (at 15.8 Mb) and RM15123 (17.3 Mb), named qPR3. Across 2 years, qPR3 explained 36.3% of variance in TAC and 35.8% in PA variance not attributable to Pb or Rc. The qPR3 region encompasses Kala3, a MYB transcription factor previously known to regulate purple grain characteristics. Study of PbPbRcrc progeny showed that TAC of RcRc near isogenic lines (NILs) was 2.1–4.5x that of rcrc. Similarly, study of PbPbRcRc NILs, which had 70% higher PA than pbpbRcRc NILs, revealed a mutual enhancement, not a trade‐off between these compounds that share precursors. This suggests that Pb and Rc upregulate genes in a shared pathway as they activate TAC and PA synthesis, respectively. This study provides molecular markers for facilitating marker‐assisted selection of qPR3, qPR5, and qPR7 to enhance grain concentrations of pigmented flavonoids and documented that stacking Rc and Pb genes further increases both flavonoid compounds.


INTRODUCTION
With increasing global health challenges in the incidence of chronic diseases (e.g., heart disease, type II diabetes, obesity, and cancers), nutrient dense whole grain cereals, including rice (Oryza sativa L.), have the potential to contribute to their prevention (Dipti et al., 2012;Jones & Engleson, 2010). Rice contains health beneficial compounds, such as simple phenolics, flavonoids, and lipophilic antioxidants, specifically tocopherols, tocotrienols, and γ-oryzanols (Chen et al., 2006(Chen et al., , 2017Min et al., 2011). These compounds are primarily deposited in the bran layer (pericarp, aleurone, and germ) of the whole grain rice (Dipti et al., 2012). Among these phytonutrients, the pigmented flavonoids anthocyanins and proanthocyanidins (PAs) in purple and red pigmented rices, respectively, have been associated with several health benefits including anti-inflammatory and anti-cancer activities, cardiovascular disease prevention, diabetes alleviation, and obesity control, as well as having antioxidant capacity (Boue et al., 2016;Chen et al., 2023;González-Abuín et al., 2015;He & Giusti, 2010). Development of nutrient dense rice cultivars with higher contents of these health promoting flavonoids requires better knowledge of the genes regulating their synthesis.
Anthocyanins and PAs are two of the six flavonoid subclasses synthesized late in the biosynthetic pathway along with flavonols, flavones, flavanones, and flavan-3-ols (monomeric units for PAs) subclasses ( Figure S1). These flavonoids are secondary metabolites, synthesized by the flavonoid biosynthetic genes that are grouped into early (EBGs) and late biosynthetic genes (LBGs). The EBGs, which include chalcone synthase, chalcone isomerase, flavanone 3-hydroxylase, and flavanone 3′-hydroxylase, are transcriptionally regulated by MYB-type regulatory proteins (Li, 2014). The LBGs include dihydroflavonol 4-reductase (DFR), leucoanthocyanidin dioxygenase, anthocyanidin reductase, and TT12, most of which are reported to be activated by a complex known as MBW, formed from three different types of regulatory proteins-a MYB, a beta helix-loop-helix (bHLH), and a WD40 (Hichri et al., 2011;Koes et al., 2005;Li et al., 2014). The function of MYB transcription factors (TFs) is to direct the MBW complexes for their differential regulation of the synthesis of flavonoids across plant organs and tissues, while the bHLH TFs regulate the expression of flavonoid biosynthetic genes, sometimes in a partially overlapping pattern involving one or more branches of the biosynthetic pathway (Gonzalez et al., 2008;Kiferle et al., 2015;Montefiori et al., 2015;Petroni & Tonelli, 2011). The WD40 protein stabilizes the MBW complex without a direct regulatory function (Hichri et al., 2011;Ramsay & Glover, 2005).

Core Ideas
• Anthocyanins and proanthocyanidins are pigmented flavonoids with desirable health-promoting properties.
• Pb and Rc are transcription factors that activate biosynthesis of anthocyanins and proanthocyanidins, respectively. • Mutual enhancement was discovered, with Rc increasing anthocyanin content and Pb increasing proanthocyanidins. • Three QTLs were newly identified as regulating concentrations of anthocyanins and/or proanthocyanidins. • Pb, Rc, and these molecularly tagged QTLs can be used to enhance the nutritional quality of rice varieties The transcriptional regulation of anthocyanins has been well studied in many plant species. In maize, the C1/Pl and R/B genes encoding R2R3-MYB and bHLH transcription factors, respectively, coordinately regulate anthocyanin biosynthesis in a tissue-specific manner (Dooner et al., 1991). The gene OsC1 in rice, homologous to maize C1, has been identified to be a R2R3-MYB group and is associated with the apiculus anthocyanin pigmentation phenotype (Reddy et al., 1998;Saitoh et al., 2004). Sun et al. (2018) further demonstrated that anthocyanin synthesis in rice hulls was regulated by both the C1, a color-producing gene on chromosome 6 and S1, a gene encoding a bHLH protein for tissue specificity for purple phenotype. Conversely, the C1 MYB TF was not required for pericarp pigmentation (Sun et al., 2018). The mapping of Kala3 on chromosome 3 and subsequent molecular and genetic analyses demonstrated that it is a MYB TF and a key gene regulating the black (also called purple) rice pericarp trait and the biosynthesis of anthocyanins (Kim et al., 2021;Maeda et al., 2014).
Two bHLH TF genes located on chromosome 4 that regulate rice purple pericarp characteristics were cloned using a cDNA probe of maize B gene (Sakamoto et al., 2001). These genes, OsB1 and OsB2, were located 10 kb apart at a purple leaf (Pl) locus, in which three alleles (Pl i , Pl j and Pl w ) were previously identified and shown to control tissuespecific anthocyanin deposition (Kinoshita & Maekawa, 1986;Sakamoto et al., 2001). Both OsB1 and OsB2 contain conserved N-terminal acidic domains and bHLH domains, while the C-terminal domain in OsB1 is completely deleted in OsB2. Either one of them along with maize C1, in a transient complementation assay, activate anthocyanin biosynthesis in rice aleurone cells, a cell type that does not normally synthesize anthocyanins (Sakamoto et al., 2001). A purple bran (Pb) gene controlling the purple pericarp character was mapped to chromosome 4 and sequence analysis revealed that a 2-bp (GT) deletion in the exon 7 expressed purple rice phenotype while the non-purple rice lines contained no GT deletion (Wang & Shu, 2007). Today, OsB1, OsPb, and OsRa (a gene homologous to maize R gene), all of which were mapped to similar location on chromosome 4 and were found to have striking sequence similarity and exon/intron boundaries, are considered to represent the same gene (Hu et al., 1996(Hu et al., , 2000Sakamoto et al., 2001;Wang & Shu, 2007). Similarly, per Oikawa et al. (2015), OsB2 is considered the same gene as OsKala4 reported by Maeda et al. (2014).
Relative to anthocyanins, the factors regulating synthesis of PAs are less investigated. As reported in Arabidopsis, PA synthesis is controlled by regulatory proteins forming a MBW complex, similar to anthocyanin regulation (Li et al., 2014). While some of the regulatory proteins are shared between anthocyanins and PAs, such as the WD40 of AtTTG1 and the bHLH of AtTT8, others are specific for PA, such as the MYB of AtTTG2, AtTT1, and AtTT16 (Koes et al., 2005). In wheat (Triticum asetivum), however, only the single Red grain (R) gene was needed to confer the red pericarp trait (Himi et al., 2005). Currently in rice, PA synthesis has been reported to be regulated by a single gene, OsRc, on chromosome 7 (Furukawa et al., 2007;Septiningsih et al., 2003;Sweeney et al., 2006). The OsRc is a bHLH TF gene and sequence analysis showed the dominant red allele to be 14 bp longer than the recessive white allele, in which the 14-bp deletion results in two premature stop codons and a truncated protein before the bHLH domain (Sweeney et al., 2006). Genome-wide association mapping among pigmented and non-pigmented rice lines further verified that the OsRc and OsRa genes were the main effect loci for colored pigments in rice bran (Shao et al., 2011;Xu et al., 2016).
Despite recent advances in knowledge of transcription factors regulating pigmented phenotypes in rice, there is still a large gap in understanding the genetics especially regarding regulation of the variable concentrations of these flavonoids (Poulev et al., 2018). The wide range of concentrations in anthocyanins and PAs of pigmented rice genotypes suggests that concentration differences might be due to genetics as well as environmental effects (Chen et al., , 2017. Tricin is a flavone flavonoid synthesized by the EBGs. In a study of genotypic diversity of tricin content in rice grains, Poulev et al. (2018) demonstrated that purple genotypes (containing anthocyanins) had higher tricin than red and light-pigmented rice genotypes, indicating that anthocyanin synthesis did not reduce flux through the flavone pathway or vice versa. On the other hand, over expression of anthocyanidin synthase (ANS), also known as leucoanthocyanidin dioxygenase (LDOX), increased the accumulation of a mixture of flavonoids and anthocyanins but lowered the PA content (Reddy et al., 2007). Competition between the flavone and anthocyanin pathways was also demonstrated in T65-Plw seedlings, a purple-leaf rice line, relative to T65 (Shih et al., 2008).
We generated a mapping population by crossing a white bran genotype with a pigmented bran genotype known to contain both anthocyanins and PAs to identify QTLs contributing to the concentrations of these pigment-associated flavonoids. We hypothesized that both structural genes and those encoding regulatory factors might be involved in determining not just the presence/absence of anthocyanins and PAs in rice pericarp, but also their concentrations. Examination of TAC and PA in these segregating progeny lines also allowed us to determine if there is a trade-off between these two flavonoids that share much of their biosynthetic pathways.

Mapping population of F 5 recombinant inbred lines (RILs)
The mapping population for the study of genetics of anthocyanins and PAs was from a cross between IR36ae x "242." IR36ae is a white grain rice (pbpbrcrc) (Juliano et al., 1990) and "242" (PI 245710) has both anthocyanin (PbPb) and proanthocyanidins (RcRc) in its pericarp. Molecular markers tagging the Pb (anthocyanin synthesis gene, chromosome 4) and Rc (PA synthesis, chromosome 7) genes, described in more detail in Section 2.5, were used to verify two F 1 s as true hybrids. F 2 seed harvested from the two F 1 plants were seeded into peat pots (6.5 cm × 6.5 cm × 7 cm tall) filled with commercial potting soil (Bacto Premium Potting Soil 0.085-0.10-0.05, Michigan Peat Co.) and wetted on May 11, 2016 with fertilizer solution (4 g L −1 Jack's Professional 20-20-20 N-P-K [J.R. Peter], plus 0.17 g L −1 iron chelate [Sequestrene 330 Fe, 10% Fe, Becker Underwood]) to initiate germination. Plants grew in the greenhouse for 4 weeks, during which time the peat pots were watered regularly (pots were placed on a 2-cm layer of soil within water-holding flats) and fertilized twice weekly with the same fertilizer solution. Leaf tissues were analyzed for rice SSR marker RM190 (chromosome 6) in support of research on a trait unrelated to bran color, and on June 8, 2016, all F 2 plants that proved homozygous for RM190 were transplanted as single plants to the field, of which 428 F 2 plants survived to produce naturally self-pollinated seed. For the F 3 generation, two progeny per F 2 were grown (total 856 F 3 s), which were advanced by natural self-pollination and single seed descent to 856 F 2:4 plants. Grains from F 4 plants were dehusked and rated visually for having either colored or standard tan (white) bran, and leaf tissues were collected and analyzed per plant for molecular markers tagging Pb and The Plant Genome Rc (described in Section 2.5) that were based on polymerase chain reaction (PCR) amplification.
The F 5 recombinant inbred lines (RILs) for QTL mapping were enriched for colored-bran progeny by including in 2019 progeny from just the 418 F 4 plants having colored bran and confirmed using molecular markers to be homozygous for Pb, Rc, or both. These F 5 RILs were germinated as described for the F 2 generation on April 3, 2019 and transplanted on May 1, 2019 in single-plant plots using a randomized complete block design with two replications, augmented with eight plots of each parental line per replication. Also transplanted in 2019 were several sets of Pb-NILs created from F 3 and F 4 plants homozygous for Rc, but segregating yet for Pb. Similarly created Rc-NILs differing for Rc or rc were also transplanted, as were F 5 progeny from F 4 plants heterozygous at both Pb and Rc. Grains from F 6 NILs grown in 2020 were used to study potential trade-off effects between TAC and PA of combining Pb and Rc. Leaf tissues were collected from all field-grown F 5 and F 6 plants and analyzed for molecular markers tagging Pb, Rc, and a newly identified QTL for the purpose of identifying different subsets of F 5 plants for the following studies. Characterization of the phenotyping methods for TAC and PA is presented in Sections 2.7 and 2.8, respectively.

Initial mapping of QTL for TAC in 31 PbPbrcrc progeny
QTLs associated with differences in anthocyanin concentration were mapped in progeny determined from molecular analysis (Section 2.5) to be fixed for PbPb to ensure they were producing anthocyanins (Lim & Ha, 2013;Wang & Shu, 2007), and fixed for rcrc because TAC is more efficiently measured in the absence of PAs (detailed in Section 2.7). Grains from F 5 plants determined molecularly to be PbPbrcrc were dehusked and rated visually for having light purple or dark purple grains. This was based on visual perception of the color intensity and percentage of the grain area having purple coloration, as can be seen in Figure 1; the two PbPbrcrc photos are on the left, with the upper photo showing light purple grain, the lower photo dark purple grain. Visual ratings of the two F 5 replications were used to identify 15 F 5 with consistently dark purple bran and 16 having consistently light purple bran. These same two replications of grains were scanned and grain images evaluated for %purpleness as described in Section 2.7. The same grains were also chemically analyzed for TAC and PA, using laboratory methods detailed Sections 2.7 and 2.8, respectively. Genotypic characterization for approximately 1000 single-nucleotide polymorphism (SNPs) loci is described in Section 2.6.

2.3
Additional TAC QTLs sought using PbPb progeny homozygous also for RcRc and the QTL identified from initial mapping effort The initial QTL mapping among 31 F 5 RILs (Section 2.2) identified a QTL for light/dark purple on chromosome 3, molecularly tagged by RM3400. However, when the PbPbrcrc F 5 progeny were further classified for containing or not (+, the "242" allele; -, the IR36ae allele) the newly identified QTL on chromosome 3, a wide range of TAC was observed among F 5 RILs fixed for the "242" allele at RM3400, suggesting the possibility of additional genes. Thus 29 F 5 RILs now fixed for RcRc with PbPb, and homozygous also for the "242" allele at RM3400 were genotyped with approximately 1400 SNPs as described in Section 2.6.

QTL mapping of PA in RcRc progeny
To assess the regulation of PA, a subset of 76 F 5 RILs that were molecularly determined to be RcRc homozygous was selected for phenotyping and genotyping. Because the presence of anthocyanin does not interfere with measurement of PA, it was not a concern if these F 5 RILs were homozygous or heterozygous for Pb or pb. The RcRc RILs were genotyped as described in Section 2.6.

PCR-based molecular markers tagging Pb, Rc, and qPR3
To detect the Pb purple pericarp anthocyanin regulatory gene (Rahman et al., 2013), a dinucleotide functional marker was developed based on a 2-bp (GT) deletion in exon 7 of LOC_Os04g47080. The primer sequences F: ATGCCAGTGGAAGGAATTGC and R:TCCAAACAGACGTATAGATACAGAG amplified a 212 bp fragment in purple pericarp lines and a 214 bp fragment in white lines. The Rid12 marker for the Rc gene (LOC_Os07g11020) developed by Sweeney et al. (2006) was used to detect the Rc/rc functional nucleotide polymorphism for red pericarp. To fill a SNP marker gap on chromosome 3 in the 1k-RiCa V1 marker map used to identify the QTL for TAC, all F 5 and F 6 RILs were genotyped for six molecular markers that included one marker tagging an insertion/deletion (InDel) site, MJInDel2 (Venu et al., 2014) located at 13,221,636 bp, and five simple sequence repeat (SSR) markers: RM232, RM282, RM15123, RM3400, and RM5626 on chromosome 3 at 9, 755,622 bp, 12,408,664 bp, 15,764,172 bp, 17,266,134 bp, and 24,866,271 bp, respectively. The SSR markers were selected using information publicly available at www.gramene.org, where SSR markers and their primers developed by McCouch et al. (2002) are listed by their chromosomal locations along the Os-Nipponbare-Reference-IRGSP-1.0 assembly (Kawahara et al., 2013). These PCR-based markers were amplified and analyzed as described in Costanzo et al. (2011).

Genome-wide SNP characterizations
The initial set of 31 PbPbrcrc F 5 plants were genotyped for some 1000 SNPs using freeze-dried leaves from the first of the two 2019 field replications and the DNA extraction and genotyping services provided by AgriPlex Genomics (Cleveland, OH) to characterize the plants using the 1K-Rice Custom Amplicon (1k-RiCA) (https://doi.org/10.1186/ s12284-019-0311-0). A second set of 29 PbPbrcrc F 5 RILs and 76 RcRc RILs were characterized by AgriPlex for some 1400 SNPs by using both the IRRI 1k-RiCA V4 Panel (https:// www.agriplexgenomics.com/irri-v4) (Arbelaez et al., 2019) and the LSU500 panel (Cerioli et al., 2022), which contain some common SNPs. The SNP calls returned by AgriPlex were manually filtered to remove nonpolymorphic SNPs and SNPs with excessively high failure rates and excessive heterozygosity. Imputation of missing data was accomplished as described in Barnaby et al. (2022) using neighboring marker information and was implemented using the Perl script simple_impute_CSSL.pl (https://github.com/jeremyde/examples_and_templates/ blob/master/simple_impute_CSSL.pl). Heterozygous allele calls were converted to missing data to conform to a RIL population QTL model.

Analysis of grains for TAC and purpleness
Previous study of 42 purple pericarp cultivars showed that TAC was associated with color intensity determined by a colorimeter (Chen et al., 2017), suggesting that visual ratings and measurement of purple intensity using image analysis would be non-destructive methods for mapping TAC. Grains of both 2019 replications of the 31 RILs selected as per Figure 1 PbPbrcrc photos for their dark and light purple extremes were scanned with an imaging analysis system (WinSeedle™ Pro version 2013b, Regent Instruments, Canada Inc., Sainte-Foy Quebec, Canada) and quantified for their percentage of grain surface area comprised of dark (purple) pixels, a trait now designated as "%purpleness." For analysis of TAC and PA, grains were first ground to flour for 2 min using Mini-Beadbeater-96 with stainless tube and beads (BioSpec Products, Inc., Bartlesville, OK, USA). Then, for RILs not containing PAs (molecularly rcrc), a subsample of 100 mg whole grain flour was extracted with 0.8 mL acidified methanol (85% methanol in 0.23 N HCl) for 1 h at room temperature with shaking; the mixture was centrifuged for 10 min at 12.4 K × g and the supernatant was saved. The pellet was extracted one more time with 0.8 mL acidified methanol. The pooled supernatant was used for anthocyanins analysis (Min et al., 2011).
The concentration of anthocyanins in the extract was determined using a microplate reader by measuring the absorbance at wavelength 530 nm. The absorbance of the sample extract was then corrected for its turbidity at wavelength 700 nm. This was followed by background noise subtraction at wavelength 530 nm using extract of IR36ae after subtracting its absorbance at 700 nm for turbidity. After the corrections for turbidity and background noise, TAC in the extract was calculated against a linear curve of kuromanin (cyanidin-3glucoside chloride, Sigma-Aldrich, St. Louis, MO., USA). For RILs known to contain PA (molecularly RcRc or Rcrc), the extract was first adsorbed with Sephadex LH-20 (Lipophilic Sephadex, Sigma-Aldrich, St. Louis, MO., USA) to remove interfering compounds near 450 nm prior to the microplate spectrophotometric analysis. Briefly, 100 mg LH-20 was hydrated in 0.4 mL 50% methanol in a 2-mL tube, and then a 1-mL sample extract, which was first equilibrated to be in 75% methanol using 100% MilliQ water, was added. After 30 min of gentle shaking, the tube was centrifuged for 1 min and the supernatant was measured at 520 nm for anthocyanins. After the subtraction of the turbidity and background noise, the extract absorbance was multiplied by a conversion factor of 1.9, which was then used to calculate the anthocyanin concentration against the linear curve of kuromanin. The LH-20 adsorption method was validated using six sets of extracts not containing PAs, and the TAC value of the extract adsorbed with LH-20 was 104 ± 7% of that without LH-20 adsorption. For the Rc-NIL study, which segregated PbP-bRcRc and PbPbrcrc, all the sample extracts were adsorbed with LH-20 gel prior to the spectrophotometric analysis and quantification.

Analysis of grains for PA
PA was determined using the 4-dimethylaminocinamaldeyde (DMAC) method of Prior et al. (2010) with some modification. Briefly, 100 mg of whole grain flour was extracted with 0.8 mL acidified methanol (85% methanol in 0.23N HCl) for 1 h at room temperature with shaking; the mixture was centrifuged for 10 min at 12.4 K × g and the supernatant was saved. The extraction was repeated one more time. The pooled supernatant was used for PA analysis. For the DMAC color reaction, an aliquot of 50 μL extract and 150 μL of 0.1% DMAC was added to a 96-well microplate and the color reaction was read every 1.5 min for 45 min, at 25˚C. The maximum absorbance of the sample extract and the standards (procyanidin B2, Sigma-Aldrich, St. Louis, MO., USA) was used to calculate the PA in the extract. Presence of anthocyanins does not interfere with the DMAC method for determining PA.

Grain trait statistics
Summary statistics, frequency distribution, analysis of variance (ANOVA), and Fisher's least significant difference (LSD) mean comparison for the TAC and PA were performed in Minitab® statistical software (Minitab Ltd., Coventry, UK). JMP 14 (SAS Institute Inc. 2018) was used to calcu-late best linear unbiased predictors (BLUPs) from multiple replications and years of trait data for use in QTL analyses.

QTL analyses
Identification of QTLs based on non-random distribution of SNP alleles between the qualitative light purple versus dark purple classifications from the 31 PbPbrcrc plants, genotyped with the 1k-RiCA V1 SNP panel, used the R/qtl2 package (Broman et al., 2019). The grain color classification data were coded as 0 for light purple and 1 for dark purple. Log-of-odds (LOD) thresholds were established for α = 0.1 and α = 0.05 using 1000 permutations. The locations of significant QTL peaks were established using the LOD thresholds and 95% Bayesian credible intervals. QTL analyses for %purpleness and TAC in these 31 RILs, and QTL mapping of TAC and PA in the additional populations genotyped with the 1k-RiCA V4 and LSU500 SNP datasets were conducted using the single marker regression (SMR), composite interval mapping (CIM LS), and multiple interval mapping (MIM) modules in QGene V4.4.0 (Joehanes & Nelson, 2008). Heterozygous allele calls were converted to missing data to conform to the RIL population QTL model in QGene. LOD thresholds at alpha = 0.1 and alpha = 0.05 were determined per trait using 1000 permutations. The left and right edges of QTLs were defined as regions having a LOD score greater than peak LOD minus 1, or the LOD threshold, whichever was higher. Tables, figures, and discussion text will focus on CIM results and include MIM when considering multiple peaks per trait. QTLs were named according to the nomenclature guidelines by McCouch (2008), with a "q," followed by trait acronym and number indicating the chromosome onto which the QTL mapped.

Summary of TAC and PA of RILs
The quantitative phenotypic traits of the subset of 31 F 5 RILs visually selected for having extreme dark (n = 15) and light purple (n = 16) grains verified using molecular markers (Pb and Rid12) to be PbPbrcrc are presented in Table 1, with example samples presented in Figure 1. The average TAC for "Dark" and "Light" purple classes were 352.8 and 16.8 μg g −1 , respectively. The average "%purpleness" of pixel purple area per kernel for "Dark" and "Light" were 51.5% and 12.3%, respectively. The average PA in RcRc lines (n = 76) was 1147.2 and 1025.0 μg g −1 in F 5 and F 6 , respectively. The ranges of PA were more than 7x in F 5 and > 9x in F 6 . The purple grains of 147 RILs (Pb-RILs) of F 5 (2019) and F 6 (2020) having PbPbrcrc and PbPbRcRc were used for validating and molecularly tagging a QTL for TAC using PCR-based molecular markers. The TAC in grains of F 5 and F 6 RILs having PbPb were on average 127.2 and 184.2 μg g −1 , respectively. The TAC in grains from some F 5 Pb-RILs was below the minimum detection level (expressed as 0 μg g −1 ), while the minimum F 6 TAC was 4.7 μg g −1 . The maximum TAC for F 5 and F 6 RILs' grains were 730.5 and 1545.4 μg g −1 , respectively. The Pb-RIL extremes were lower and higher than the TAC of "242" pigmented grains (1144.1 and 974.0 μg g −1 ) in 2019 and 2020, respectively. The averaged TAC for the Pb-RILs was slightly lower than the averaged TAC (184.8 μg g −1 ) from the subset of extreme dark/light RILs plus "242" parent used for QTL mapping. The grains of 176 RILs (Rc-RILs) having pbpbRcRc and PbPbRcRc were used for validating and molecularly tagging a QTL for PA using PCR-based molecular markers. The averaged PA in seeds of F 5 and F 6 were 987.9 and 865.1 μg g −1 , respectively. The minimum PA in F 5 and F 6 were 310.5 and 466.6 μg g −1 , respectively, while the maximum PA in F 5 and F 6 were 2061.5 and 3146.8 μg g −1 , respectively, which were lower and higher than the PA of "242" pigmented grains (2223.8 μg g −1 in 2019 and 1796.6 μg g −1 in 2020). Similar ranges of PA (6.6×) were found between F 5 176 Rc-RILs and the QTL mapping subset. However, a wider range of PA in F 6 Rc-RILs was shown (17×) relative to the randomly selected subset of RILs used for PA QTL mapping.
Analyses of variances of TAC and PA indicated that the genotypic factor of "RILs" accounted for most of the variances of TAC (R 2 = 0.76) and PA (R 2 = 0.84), while "Year" (F 5 and F 6 generations in 2019 and 2020, respectively) explained significant but small proportions of variation that were less than 1/10 th and 1/20 th the TAC and PA variations explained by RILs in each year.

QTLs for TAC and color among 31 PbPbrcrc RILs with extreme color phenotypes
Among the 1k-RiCA V1 SNPs, 446 were polymorphic between IR36ae and "242," but with the small population size (n = 31) restricting recombination rates, several closely linked SNPs mapped to the same cM location in the genetic map, making them redundant. A total of 259 non-redundant polymorphic SNPs from the 1k-RiCA V1 Panel were used for our marker-trait analyses. The average cM distance between nonredundant marker loci in the genetic map was 4.7 cM, with two large gaps: a 47.2 cM gap on chromosome 2 and a gap greater than 50 cM on chromosome 3 ( Figure S2). The marker gap on chromosome 3, from 03_14933451 to 03_31427789, proved especially problematic because the QTL analyses of both color class and %purpleness identified a single QTL on chromosome 3 with its peak at 03_14933451 (Table 2; Figure S2). These were large QTLs that spanned the marker gap but showed stronger linkage to 03_14933451 than to 03_31427789. On the other hand, analysis of the TAC data from grains harvested in 2019 indicated two QTLs, one on chr3 having the same QTL peak and span as the color class and %purpleness QTLs, which we now name as qPR3, and a second QTL on chr7 (qPR7) between 07_19408599 and 07_23535052.
To confirm and more precisely map qPR3, molecular markers surrounding this chr3 locus were used to genotype F 5 and F 6 Pb-RILs (n = 147). This analysis used TAC data from grains harvested in two years (2019 F 4:5 and 2020 F 4:6 ), two replications per year. The five SSRs and the InDel ranged from chr3 9,755,622 to 24,866,271 bp (Table 3). The markers found to be most closely associated with TAC were RM3400 (03_17266134 bp) (R 2 = 0.419 and 0.308 in F 5 and F 6 , respectively) and RM15123 (03_15764172) (R 2 = 0.360 and 0.274 in F 5 and F 6 , respectively). This verified qPR3 as a TAC QTL, mapped it more precisely between RM15123 (03_15764172) and RM5626 (03_24866271), and showed that it could be molecularly tagged in segregating progeny using RM3400. The "242" allele at RM3400 was associated with dark purple, while the IR36ae allele was associated with lighter purple. Throughout this manuscript we will distinguish the dark and light alleles as RM3400D and RM3400L, and consider these markers to indicate presence of the qPR3D dark or qPR3L light alleles, respectively. The chromosome 7 QTL peak mapped between SNPs at 07_19408599 and 07_23535052, within a 38 cM marker gap. This QTL region contains just one gene known T A B L E 2 QTLs identified by interval mapping as associated with rice pigmented pericarp traits in segregating progeny from a cross between IR36ae (white pericarp) x "242" (purple and red pericarp) QTL  T A B L E 3 Percentage of variance explained (from ANOVA R 2 ) a by polymerase chain reaction (PCR)-based molecular markers for total anthocyanin content (TAC) in Pb-RILs or for total proanthocyanidin content (PA) in Rc-RILs to be associated with color pigmentation, anthocyanidin 3-O-glucosyltransferase (AGT) (LOC_Os07g32620) (Cohen 2019). Evidence of qPR7 was less confident than that for qPR3. Interval mapping placed both QTL peaks in marker gaps, but while qPR3 had significantly associated markers on both sides of the marker interval, for qPR7 only one of the flanking markers (07_19408599) was significantly associated with TAC, and it was the allele from unpigmented IR36ae that was associated with increased TAC. Furthermore, analysis of the TAC data using MIM retained a notable QTL peak for qPR3, but not for qPR7 ( Figure S2) suggesting it might be a false positive from pseudolinkage within our small population (n = 31) of extreme phenotypes. The putative qPR7 QTL was examined in an additional study (Section 3.3).

Additional anthocyanin QTLs sought using RILs fixed for Pb, Rc, and qPR3D and a denser genetic map
Segregation for loci of large effect can confound ability to identify other QTLs. Therefore, a second effort to identify TAC QTLs involved 29 RILs that were fixed for PbPbRcRc, and fixed as well for qPR3D, the "242" allele. To increase reliability of TAC data for QTL mapping, these RILs were evaluated using two years (F 4:5 2019, F 5:6 2020) of grains, two replications per year, and QTL analyses were run using a 2-year BLUP calculated across years, as well as with singleyear BLUPs. RILs were genotyped by AgriPlex using both the 1K RiCA V4 and the LSU500 panel. This generated 1574 SNP datapoints per genotype, but due to overlap between SNP of the two panels, there were approximately 1400 unique SNPs represented, 528 of which were polymorphic between IR36ae and "242." When analyzed among the total 92 F 5 RILs selected for TAC and PA analysis, they yielded a genetic map containing 401 discreet SNP loci separated by an average 2.7cM, with only three marker-gaps greater than 20 cM (on chr 1, 2, 2), and no marker gaps exceeding 50cM . Interval analysis of the 2019, 2020, and 2year TAC BLUPs were in consensus finding no additional QTLs. Hence, this analysis of 29 RILs fixed as PbPbRcR-cqPR3DqPR3D was unable to provide further evidence for qPR7.

QTL of PA
A total of 75 F 5 RILs having RcRc were randomly selected for mapping PA QTLs. Two QTLs were identified, one on chromosome 3 and another on chromosome 5 (Table 2; Figure S3). The peaks of the PA QTLs differed slightly between 2019, 2020, and the 2-year BLUP data, but there was consensus in placing a QTL on chr3 with its peak between 15.4 and 21.5 Mb, making it co-located with the qPR3 QTL identified as impacting dark/light purple classification, %purpleness, and TAC. This PA QTL had LODs > 8 in all datasets, explained from 40% to 45% of the variance in PA, with an additive effect of 295 units or more. The same six PCR-based markers on chr3 used to validate qPR3 for TAC among a larger set of PbPb RILs (Section 3.2) also validated association with PA within a larger set (n = 176) of RcRc RILs (Table 3). Of the six PCR-based markers, RM3400 (R 2 = 0.382 and 0.333 in F 5 and F 6 , respectively) and RM15123 (R 2 = 0.305 and 0.291 in F 5 and F 6 , respectively) were the markers having the highest R 2 values explaining PA (Table 3). The SNPs flanking the PA QTL peak on chromosome 3 in both the 2020 and 2-year BLUP analyses were 03_21003799 and 03_21502680, while the 2019 data placed the QTL peak between 03_16734121 and 03_17794287. These peaks encompass or are near RM3400 at 03_17266134. Altogether, the data indicate that the PA QTL on chromosome 3 is the same as qPR3, first identified for association with TAC and purple colorations. In fact, this locus was named qPR3 to indicate its association with both purple and red coloration.
The second PA QTL, qPR5, achieved a significant LOD in the F 6 (2020) and 2-yr BLUP data analyses, but a subthreshold LOD peak in analysis of F 5 (2019) data (Table 2; Figure S3). There was consensus among all three datasets in placing qPR5 between the SNPs 05_18005952 and 05_18405433. qPR5 had smaller genetic effect than qPR3 all three datasets, with additive effects and percentages of variance explained approximately half those for qPR3. None of the genes between 05_18005952 and 05_18405433 had annotated gene functions associating them with flavonoid concentration, preventing identification of a candidate gene within qPR5. This QTL region is, however, flanked by a MYB TF at 19.0 Mb (LOC_Os05g04210) and a WRKY gene at 17.4 Mb (LOC_Os05g03900). WRKY TFs were shown to affect MYB-regulation of anthocyanin and/or proanthocyanidin biosynthesis and vacuolar transport in apples Mao et al., 2021), pears (Alabd et al., 2022;Li et al., 2020), and grapes (Amato et al., 2019).

Means and ranges of TAC in Pb-RILs having qPR3D versus qPR3L
Frequency distributions and descriptive statistics of TAC in F 5 and F 6 Pb-RILs divided by containment of qPR3D and qPR3L alleles are presented (Figure 2). The Pb-RILs with qPR3D (n = 46) had a mean TAC of 261.6 and 404.0 μg g −1 in F 5 and F 6 , and were > 4.2x and > 5.1x of those of RILs having qPR3L (61.8 and 77.8 μg g −1 ), respectively. The ranges were also very different with the range of TAC in Pb-RILs having qPR3D being 3.1x and 2.4x of that of qPR3L Pb-RILs in F 5 and F 6 , respectively. However, the results clearly showed that Pb-RILs with qPR3L deposited low amounts of anthocyanins in the pericarp of pigmented grains, indicating reduced but not entire inactivation of anthocyanin synthesis.

Means and ranges of PA in Rc-RILs having qPR3D versus qPR3L
Frequency distributions and descriptive statistics of PA in Rc-RILs divided by qPR3D and qPR3L are presented (Figure 3). The Rc-RILs with qPR3D had wider ranges, higher means, and higher extreme PA phenotypes than Rc-RILs with qPR3L. Rc-RILs with qPR3D (n = 71) had average PA of 1279.8 and 1183 μg g −1 in F 5 and F 6 , respectively. Rc-RILs with qPR3L (n = 100) had average PA of 781.8 μg g −1 in F 5 and 636.5 μg g −1 in F 6 . The PA in Rc-RILs with qPR3D was 64% and 86% higher than that in Rc-RILs with qPR3L in the F 5 and F 6 generations, respectively.

Effect of Rc and qPR3 on TAC
The synthesis of anthocyanins and PAs shares the entire portion of the flavonoid biosynthetic pathway that is controlled by EBGs (CHI, CHS, F3H, F3'H) and DFR, a LBG, and these genes are expressed in grains having both red and purple pericarps (Li et al., 2014;Lim & Ha, 2013;Oikawa et al., 2015).
Here we assessed if active versus inactive Rc TF, known to turn on or off production of PAs, would have effect on TAC. This was not addressed during the TAC QTL mapping efforts because the first mapping population was fixed as PbPbrcrc and the second fixed as PbPbRcRc, with neither population segregating for Rc/rc as needed to address the question. Using Rid12 to distinguish RILs as having active (Rc) or inactive (rc) Rc, and using RM3400D versus RM3400L to distinguish RILs having the qPR3 allele for darker or lighter colors, we evaluated the effects of +/− Rc and +/− qPR3 within the 147 Pb-RILs. ANOVA of F 5 data showed that +/− qPR3 significantly impacted TAC (p < 0.001) while the effect of Rid12 was significant at p = 0.058; for F 6 TAC data, only qPR3 was significant (p < 0.001) while Rc was not (p = 0.77). Dividing the Pb-RILs based on +/− Rc and +/− qPR3 creates four genetic classes. In agreement with the QTL mapping, Pb-RILs homozygous for qPR3D contained more TAC than those with qPR3L ( Figure S4). Therefore, it was important to look at +/− Rc effects in Pb-RIL classes also fixed for qPR3. Within Pb-RILs homozygous for qPR3D, the TAC did not differ for +/− Rc. In contrast, among RILs having qPR3L, the rcrc RILs had lower average TAC than RcRC RILs in both F 5 (2019) and F 6 (2020) data, though the means were significantly different only in the F 5. Visible in Figure S4, and also in Figure 2a,c, the Pb-RILs containing qPR3D had especially wide ranges in TAC. Along with the high standard deviations seen for all four genetic classes in Figure S4, the data suggest that unidentified factors are causing wide variance for TAC among the 18 to 51 individuals per genetic class, and this wide variance could be masking any smaller variance attributable to +/− Rc. To minimize the possible confounding effects from segregation of other genes, we then used residual heterozygosity for Rc found within some of the RILs to create NILs +/− Rc (hereafter called Rc-NILs) with which to evaluate effect of Rc in pairs of more uniform genetic backgrounds (Figure 4).
All four sets of Rc-NILs are F 6 generation and are homozygous for the Pb and qPR3D (Figure 4a-c). The Rc-NIL1 set contains RcRc and rcrc progeny from an Rcrc F 5 line that was fixed for both Pb and qPR3D. The TAC averaged over multiple RcRc F 6 plants (TAC = 72.7 μg g −1 ) was 2.9x of TAC (μg g -1 grains) qPR3D qPR3L qPR3D qPR3L that of the rcrc sibling NILs (TAC = 24.7 μg g −1 ). Rc-NIL2 was a set of homozygous and heterozygous F 6 progeny lines descendent from a heterozygous Rcrc F 4 plant, and the averaged TAC of F 6 Rc-NIL2 plants having RcRc (183.2 μg g −1 ) and Rcrc (129.7 μg g −1 ) were 2.1x and 1.5x of the TAC in rcrc NIL progeny (85.7 μg g −1 ), respectively. The Rc-NIL3 were F 6 progeny descendent from divergence for RcRc versus rcrc detected in the F 3 generation (traced to same F 2 parent), and the averaged TAC of RcRc (191.8 μg g −1 ) was 4.5x of that of rcrc (42.8 μg g −1 ). These three sets of Rc-NILs all indicated that synthesis of PA from Rc caused an increase in TAC. Rc-NIL4 consisted of F 6 progeny from an F 4 fixed for Pb but heterozygous for both Rc and qPR3 (Figure 4d). The TAC of rcrc F 6 -NILs homozygous also for qPR3D (88.3 μg g −1 ) were not different from those heterozygous at qPR3 (97.3 μg g −1 ), but both had significantly higher TAC than rcrc NILs having qPR3L (20.4 μg g −1 ), indicating that qPR3D has complete dominance over qPR3L. For evaluation of Rc effects, progeny in this NIL set that were RcRc and heterozygous for qPR3 had the highest averaged TAC (168.8 μg g −1 ) among the four progeny genotypes, higher than any of the rcrc progeny, providing further evidence that Rc positively impacts TAC.

Effect of Pb and qPR3 on PA
The QTL mapping among Rc-RILs detected an effect of qPR3 on PA but did not detect an effect from Pb, though it also segregated in this relatively small population (n = 75). The effects of Pb and qPR3 on PA were investigated further using 170 Rc-RILs evaluated for PA over 2 years, the F 5 and F 6 generations. The ANOVA detected both Pb and qPR3 as significant factors in the model, explaining PA at R 2 = 0.014 (p = 0.052) and R 2 = 0.38 (p < 0.001), respectively, in F 5 Rc-RILs and at R 2 = 0.041 (p = 0.001) and R 2 = 0.33 (p < 0.001), respectively, in F 6 Rc-RILs. Thus, though Pb effects were found significant, they explained 1/10 th the variance in PA as that explained by qPR3. Use of means comparison to compare the average PA among the four genetic classes within the Rc-RILs The Plant Genome  F I G U R E 3 Frequency distributions of total proanthocyanidin content (PA) (μg g −1 grain) of F 5 and F 6 Rc-RILs homozygous for the qPR3D allele (a,c) and those homozygous for the qPR3L allele (b,d) are shown. Summary statistics are presented in (e). RIL, recombinant inbred line.
( Figure S5) showed that Rc-RILs containing qPR3D had more PA than those with qPR3L, and this was true in both the F 5 and F 6 generations, regardless of whether the Rc-RILs contained Pb or pb. In contrast, effects of Pb on PA were seen only among F 6 Rc-RILs homozygous for qPR3D, where Rc-RILs homozygous for Pb had more PA than those homozygous for pb. The effect of Pb was not significant among grains harvested from F 5 Rc-RILs in 2019, nor was it significant among the F 6 Rc-RILs homozygous for qPR3L. However, the wide variability for PA within each genotypic class of the 76 Rc-RILs ( Figure S5) can make it difficult to detect a small effect of Pb on PA. To reduce the background noise, PA was further studied using Pb-NILs, paired F 6 progeny differing only for Pb/pb because they were fixed in a previous generation for Rc and qPR3 ( Figure 5). Pb-NIL sets 1 through 3 (Figure 5a-c) diverged for Pb and pb in the F 5 generation, Pb-NIL4 (Figure 5d) diverged in the F 4 generation, and the remaining three Pb-NIL sets (Figure 5d-f) each traced to a different F 2 parent, and F 3 divergence for Pb/pb. For all seven Pb-NIL paired sets, the multiple PbPb progeny had more PA than the pbpb progeny, though this difference was not significant for two of the seven sets (Pb-NIL2 and Pb-NIL4). The PA averages in the PbPb versus pbpb progeny were, from Pb-NIL 1 through 7 (Figure 5a-g), 1670 μg g −1 versus 990 μg g 1 , 462 μg g −1 versus 344 μg g −1 , 339 μg g −1 versus 207 μg g −1 , 698 μg g −1 versus 557 μg g −1 , 1272 μg g −1 versus 861 μg g −1 , 474 μg g −1 versus 229 μg g −1 , and 1794 μg g −1 versus 1062 μg g −1 , respectively. These seven PbPb versus pbpb comparisons among related progeny lines demonstrated that the Pb TF positively modulates PA.

DISCUSSION
In this study, we identified three factors affecting TAC after activation of anthocyanin biosynthesis by Pb, namely qPR3, qPR7, and the Rc TF, and identified three factors affecting PA after activation of PA biosynthesis by Rc, specifically qPR3, qPR5, and the Pb TF. These factors regulating the concentrations of pigmented flavonoids (anthocyanins and PAs) were (a-c) F 6 Rc-NILs, progeny lines from individual PbPbRcrc-RILs of F 5 , F 4 , and F 2 plants, respectively, now segregating for rc and Rc. (d) Rc-NIL4 are progeny lines from an F 4 that was heterozygous for both Rc and qPR3. Different letters for the means indicate significant difference among the genetic variants at 95% confidence level using Fisher LSD method. LSD, least significant difference; NIL, near isogenic line; RIL, recombinant inbred line.
identified using RILs with pigmented grains from an IR36ae x "242" mapping population. IR36ae has white pericarp, and "242" is a pigmented rice cultivar containing both anthocyanins and PAs and was ranked among the highest of 25 purple rice genotypes for its total flavonoid and phenolic contents as well as antioxidant capacity (Chen et al., 2017;Wan et al., 2021). Effect of qPR3 on TAC was verified with independent identification in two population subsets (PbPbrcrc, n = 29, and PbPbRcRc, n = 31), documented also in a larger set of 147 Pb-RILs genotyped using SSRs tagging qPR3, and further validated in a set of NILs (Rc-NIL4). Effect of qPR3 on PA was also verified with identification in two populations, being detected initially in 76 SNP-genotyped Rc-RILs, then validated in a set of 176 SSR-genotyped Rc-RILs. The effect of Pb on PA is smaller than that of qPR3 and was not detectable via QTL analysis but was detected using the more uniform genetic backgrounds provided by NILs. Effect of Rc on TAC was likewise detected using NILs. Though we were unable to validate the second TAC QTL, qPR7, neither by MIM nor in a subset of 31 PbPbRcRc RILs, lack of validation does not prove nonexistence. The notably small size of the present mapping populations limited the ability to identify additional QTLs of small effect. For example, effect of Pb on PA was detected in NIL analysis, but was not detected through QTL mapping in 76 Rc-RILs. Validation of qPR7 in this second population fixed for Rc would be especially difficult if its effects are reduced in the presence of Rc. Epistasis like this was found to impact effects of Pb on PA, which were larger in NILs with qPR3D than in NILs having qPR3L. Lack of residual heterozygosity for qPR7 and qPR5 among our RILs fixed at the Rc, Pb, and qPR3 loci prevented us from analyzing qPR7 and qPR5 in NILs.
In addition to QTL discovery, this study demonstrated the ability to use a non-destructive quantitative measure of bran color purpleness to identify TAC QTLs. Previous study of 25 purple pericarp cultivars showed that the color parameter b* of the CIE L*a*b* color space was negatively correlated with TAC across purple genotypes (Table 4 in Chen et al., 2017); b* measures yellowness-blueness with negative b* indicating blueness. In this study, we demonstrated that by specifically selecting purple pixel area, defined by the color space of hue/saturation/intensity, of the image of anthocyanin containing grains to quantify %purpleness, we obtained nondestructive quantitative data that successfully identified qPR3 as affecting variance in TAC among PbPb rices. In addition to time efficiency compared to either wet chemistry or the use of the HunterLab Miniscan XE Plus colorimeter previously used (Chen et al., 2017), which requires at least 40 g grains for the analysis, the currently used flatbed imaging analysis did not require a minimum number of grains. We did not use image analysis or visual selection for the redness of Rc-RIL grains because previous analysis of 32 red pericarp cultivars  demonstrated that while the color parameter a*, which measure redness-greenness, was positively correlated with cell-wall associated PAs, which accounted for only 4.4% (g) Pb-NIL7, F 2:6 F I G U R E 5 The effects of Pb and qPR3 (as tagged by RM3400) on total proanthocyanin content (PA, μg g −1 grain) using Pb-NILs having RcRc and fixed also for either the qPR3D or qPR3L alleles from the "242" or IR36ae parents, respectively. Pb-NILs 1 through 3 (a-c) are F 6 Pb-NILs, progeny lines from a Pbpb F 5 RIL; PB-NIL4 (d) is a set of F 6 progeny differing for PbPb or pbpb and derived from a single F 4 RIL. (e-g) F 6 Pb-NILs differing for PbPb or pbpb due to divergence that occurred in the F 2:3 generation. Different letters for the means indicate significant difference among the genetic variants at 95% confidence level using Fisher LSD method. LSD, least significant difference; NIL, near isogenic line; RIL, recombinant inbred line. of total PAs, it was not indicative of the extractable PAs (95% of the total PA). The present study identified PA QTLs using data on extractable PAs. Furthermore, PAs can be colorless or colored, developing a red/brown color only after oxidation by flavonoid oxidase (Pourcel et al., 2005). The two pbpbRcRc samples in Figure 1 demonstrate differences between externally visible and extractable PA, with the grains in the upper image (qPR3LqPR3L) containing less PA but being visually redder than the than the qPR3DqPR3D grains in the image below.
Both TAC and PA mapped independently to qPR3 between RM15123 (03_15764172 bp) and RM3400 (03_17266134 bp).Two genes in this QTL region have annotated gene functions associated with biosynthesis of anthocyanin. One is a R2R3 MYB transcription factor, Kala3 (LOC_Os03g29614) located at chromosome 3 16,879,442 bp and previously shown to be associated with purple/black rice bran color (Maeda et al., 2014;Kim et al., 2021). The other is a leucoanthocyanidin dioxygenase (LDOX) gene (LOC_Os03g32470), located at chr3 18,570,651 bp. The predicted placement of qPR3 between RM15123 and RM3400 suggests that the underlying gene is more likely to be Kala3 than the nearby LDOX. Confirmation would require further mapping with additional recombinant progeny and biochemical analyses. Wang et al. (2022) identified 11 LDOX genes in rice using BLASTP searches and deduced protein homology, two of which, LOC_Os01g27490 and LOC_Os06g42130, are also known as OsANS1 and OsANS2, respectively, and have been documented to affect purple coloration and/or TAC in rice pericarps (Jung et al., 2019;Lepiniec et al., 2006;Reddy et al., 2007). Anthocyanin is synthesized from anthocyanidins, and ANS/LDOX synthesizes anthocyanidins from leucoanthocyanidins ( Figure S1); thus, the LOC_Os03g32470 ANS/LDOX locus could be a candidate for regulation of anthocyanin. However, we also mapped PA to qPR3, requiring consideration of ANS/LDOX effects on PA as well as TAC. There are two branches of flavonoid biosynthetic pathway that synthesize flavan-3-ols, monomers of PAs, from leucoanthocyanidins; one is catalyzed by leucoanthocyanidin reductase (LAR) and the other is via ANS/LDOX to anthocyanidins and then by anthocyanidin reductase (ANR) to flavon-3-ols (Shih et al., 2008). The expression of ANS/LDOX was found upregulated in pericarp of black rice, but it was barely expressed, if at all, in red pericarp, while LAR was upregulated in red pericarp, not in purple pericarp (Lim & Ha, 2013;Oikawa et al., 2015). Therefore, with an inactive ANS/LDOX, that is, pbpbRcRc genotype with non-purple red grains, PAs can be synthesized. However, in that case, PA would not be associated with ANS/LDOX. Our results showed that qPR3, tagged by RM3400, associates with PA in RILs having pbpbRcRc (red pericarp) ( Figure S5); that is, pbpbRcRc RILs having qPR3D had significantly higher PA than those having qPR3L. Biochemically, ANS/LDOX does not fit all of our observations, making the Kala3 MYB transcription factor a more likely candidate gene for qPR3.
A TAC QTL was also identified on chromosome 7, near an anthocyanidin 3-O-glucosyltransferase gene (AGT, LOC_Os07g32620). Kim et al. (2021) demonstrated that OsKala3 and OsKala4 (a.k.a., OsB2) mediate the activation of anthocyanin pathway synthesis genes. Further, the promoter of Kala3 correlates strongly with the induction of its own expression and forms a positive feedback loop with its own promoter (Kim et al., 2021). A similar functional gene located on rice chromosome 6, UDP-glucosyl transferase (UGT) (Os06g0192100), which along with ANS were specifically upregulated in black pericarp of F 1 plants (Oikawa et al., 2015). Thus, it is possible that qPR3 (i.e., Kala3) is upregulating itself in addition to increasing qPR7 gene expression, leading to association with TAC. Though our study did not validate qPR7, it remains worthy of further investigation.
Like qPR7 for TAC, the second PA QTL, qPR5, was not validated within this study. Use of additional recombination dissecting this QTL region to fine-map this QTL would be required to clarify relationships between qPR5 and the two flanking genes, MYB and WRKY, whose functions potentially impact accumulation of pigmented flavonoids.
The general model across all species analyzed to date for the regulation of flavonoid biosynthetic pathway is that the MYB transcription factor activates the EBGs while the regulation of the LBGs for anthocyanins and PAs requires the MYB-bHLH-WD40 (MBW) complex (Hichri et al., 2011;Koes et al., 2005;Li et al., 2014). In Arabidopsis thaliana, there are three MYBs activating EBGs and these are different from those activating LBGs. For the LBGs, the MYBs of Arabidopsis that activated anthocyanin pathway genes were different from those that activated PA biosynthesis. On the other hand, a common bHLH AtTT8 protein, a 'B' component of MBW, regulates both anthocyanins and PAs. Other bHLH TFs that also control both anthocyanin and PA pathways are morning glory bHLH2 and IVS (Park et al., 2004(Park et al., , 2007. In rice, two bHLH genes (Pb and Kala4) regulate purple grain phenotype and one bHLH gene (Rc) controls red grain. OsKala3, also known as OsMYB3, is a MYB protein that interacts with OsKala4 to form a complex that regulates purple pericarp (Maeda et al., 2014;Zheng et al., 2021;Kim et al., 2021). Transfecting rice protoplasts with OsKala3 and OsKala4 together but not alone, activates anthocyanin biosynthetic genes, both EBGs and LBGs (OsCHS, OsCHI, OsF3H, OsDFR, and OsANS1) (Zheng et al., 2021). However, the involvement of OsKala3 in regulating genes specifically for biosynthesis of PAs (for example, LAR) was not previously reported. Our study showed that OsKala3 is the candidate gene underlying the qPR3 QTL associated with PA as well as TAC. Our findings, along with previous studies on anthocyanin regulation, suggest that Kala3 is most likely the MYB component of the MBW complex for regulating pigmented flavonoid biosynthesis in rice pericarp.
Prior studies using genome-wide association of a diverse population including white and colored-pericarp genotypes identified QTLs for total flavonoid content on multiple chromosomes and OsRa (Shao et al., 2011;Xu et al., 2016). Among the previously reported QTLs for total flavonoid, only two, OsRc and OsRa/OsPb were co-located with loci identified in the present study as affecting TAC or PA in Pb-RILs or Rc-RILs, respectively. The contents and classes of flavonoids in pigmented and non-pigmented rices are complex, comprising multiple classes of anthocyanins, proanthocyanidins (monomers, oligomers, and polymers), flavones, flavanones, and flavonols (Chen et al., 2012;Mbanjo et al., 2020;Poulev et al., 2017). Kim et al. (2021) reported that pericarp color was determined by the number of repeat units (RUs) in the OsKala3 promoter region rather than differences in its coding region. Each of the RUs contained five MYB-binding cis-elements and five seed-specific cis-elements (Kim et al., 2021). In studying 40 rice cultivars (13 white, 16 black/purple, and 11 red), all black pericarp cultivars contained two RUs and all white and red pericarp cultivars contained one RU. It was concluded that the expression of OsKala3 harboring a single RU in the promoter region was not sufficient to activate its own promoter via a positive feedback loop (Kim et al., 2021). Nonetheless, all the white grain cultivars had inactive rc and kala4 and all the red grain cultivars had inactive kala4 (Kim et al., 2021), and rice cultivars having inactive kala4 do not accumulate anthocyanins. Therefore, red grain cultivars having inactive kala4 and kala3 cannot be used to prove that the one RU in the promoter of kala3 is the cause of nonpurple grain phenotype. Zheng et al. (2021) reported similar transcript levels of OsKala3 for red and purple pericarp, and both have higher OsKala3 transcripts than the white pericarp. Therefore, there might be an additional mechanism involved besides the number of RUs in the promoter of OsKala3.
IR36ae is pbpbrcrc and "242" has PbPbRcRc. We selected Pb-RILs and Rc-RILs from this IR36ae x "242" mapping population using the dinucleotide functional Pb marker and the Rid12 marker to tag Pb and Rc, respectively. Further study of the Pb-RILs and Rc-RILs led to the discovery that qPR3 regulates the TAC and PA in purple and red rice grains, and does not function as an on/off switch for these pigmented flavonoids as Pb and Rc do. However, the magnitudes of the qPR3 allele effects on TAC and PA appeared to be different. Dividing each of these subsets by RM3400-indicated qPR3 alleles, the TAC of Pb-RILs having qPR3D averaged across both years/generations was 4.7x of that RILs having qPR3L ( Figure S4), while averaged PA of Rc-RILs having qPR3D was nearly 75% higher than RILs having qPR3L ( Figure S5). On top of that, Rc-RILs with qPR3L accumulated significant PA in grains ( Figure S5). Rc was identified as a transcription factor regulating red pericarp phenotype in two mapping populations and transgenic lines (Furukawa et al., 2007;Septiningsih et al., 2003;Sweeney et al., 2006), and no other genes or QTLs were identified in those studies, suggesting that Rc, a bHLH protein, might activate PA biosynthetic pathway genes alone. The bHLH transcription factors can individually bind to structural genes without requiring association with a MYB. For example, the petunia Spinacia oleracea bHLH JAF13 and Arabidopsis TT8 can individually bind the SoANS and AtDFR promoters alone (Shimada et al., 2006).
Both the black and red rice pericarps expressed the genes in the shared portion of the biosynthetic pathway, including the EBGs (CHS, CHI, F3H, F3'H) and DFR, a LBG (Li et al., 2014;Hichri et al., 2011;Koes et al., 2005). In this study, we assessed whether stacking Rc and Pb would cause a trade-off between anthocyanins and proanthocyanidins due to competition for precursors. The Rc-NILs having qPR3D and PbPb showed that NILs having RcRc had 2x more TAC than the rcrc NILs. As for the Pb-NIL comparisons, where progeny contrasted for Pb/pb but were homozygous for RcRc and for either qPR3D or qPR3L, the PAs of active PbPb NILs were significantly higher than in their paired pbpb NILs. We demonstrated that Rc, a gene identified as a regulator for synthesis of PAs, elevated anthocyanin content, and conversely showed that Pb, a gene known as a bHLH transcription factor for anthocyanin synthesis, increased PA in pigmented rice. In addition to the specificity of Pb and Rc in regulating individual respective pathway LBGs, the Rc and Pb proteins might also be acting as a modulator by upregulating the expression of the shared genes, both EBGs and DFR, thereby accumulating more precursors and leading to the enhancement of both anthocyanins and PAs deposited in pericarps. Modulators that play a positive or negative role on the activity of MBW complex for anthocyanin synthesis have been reported (Li et al., 2014).
Among them, two mechanisms that positively regulate pigmented flavonoid synthesis might explain effects of Pb and Rc on both TAC and PA. The TCP3, as a modulator, reportedly interacts with all three Arabidopsis R2R3-MYBs to stimulate the transcription of EBGs, thereby increasing flavonol production (Li & Zachgo, 2013). The Arabidopsis TCP3 and TT1 organized the formation of MBW complexes by synergistically associating with R2R3-MYBs, resulting in increased TAC and PA (Appelhagen et al., 2011;Li & Zachgo, 2013). In addition, bHLH proteins could directly bind DNA, that is, the G-box of the target genes (Hichri et al., 2011).
A previous study showed that over expressing ANS/LDOX channeled the PA precursors, leucoanthocyanidins, toward the synthesis of anthocyanidins and resulted in the increase of anthocyanins and the decrease of PAs (Reddy et al., 2007). In contrast, our data show a mutual enhancement of TAC and PA from Pb and Rc rather than a trade-off. If these genes are modulating the expression of EBGs, then the supply of precursors for the biosynthesis of PAs and TAC would not be limited, and thus would enhance both TAC and PA rather than one at the expense of the other (Oikawa et al., 2015). Our findings indicate that rice cultivars with increased content of health beneficial flavonoids can be developed by combining both Pb and Rc along with three antioxidant-enhancing QTLs (qPR3, qPR5, qPR7) for which genetic markers to facilitate marker-assisted breeding are now provided.

CONCLUSION
Genetic mapping of RILs from IR36ae x "242" revealed three QTLs (qPR3, qPR5, qPR7) that regulate the concentrations of anthocyanins and proanthocyanidins, the purple and red pigmented flavonoids in pericarp of rice grains whose synthesis is turned on by the Pb and Rc transcription factors, respectively. The candidate gene for the QTL with largest effect, qPR3, is a MYB transcription factor for purple/black grains gene variably called OsMYB3. We demonstrated via QTL mapping and again in NILs that the qPR3 allele from "242," the parent with colored bran increased the concentrations of anthocyanins that were deposited in the grain pericarp, and was dominant, while the allele for reduced pigmentation from IR36ae was recessive. Our study further showed that this candidate MYB transcription factor at qPR3 regulated the concentrations of proanthocyanidins in rice grains containing active Rc, a bHLH transcription factor gene previously known for turning on and off the proanthocyanidin biosynthesis. Additionally, the Rc and Pb genes, which activate the synthesis of proanthocyanidins and anthocyanins, respectively, were found to positively modulate the depositions of both anthocyanins and proanthocyanidins. The mutual enhancement of both anthocyanin and proanthocyanidin concentrations by both Pb and Rc further demonstrates that rice breeders can maximize antioxidants in rice grains by combining these genes, without concern about limited precursor production causing a trade-off between anthocyanin and anthocyanidin contents.

AU T H O R C O N T R I B U T I O N S
Ming-Hsuan Chen: conceptualization; supervision; investigation; methodology; resources; data curation; formal analysis; and writing-original draft. Shannon R. M. Pinson: conceptualization; supervision; investigation; methodology; resources; and writing-original draft. Aaron K. Jackson: formal analysis and writing-review. Jeremy D. Edwards: methodology; formal analysis; and writing-review.

A C K N O W L E D G M E N T S
This work was supported by USDA-ARS CRIS-6028-21220-005-00D. The authors thank Dr. Lee Tarpley for advice and assistance in data analyses. We also acknowledge the technical support of Jace Everette for photography skills, harvesting, seed preparation, and analysis of TAC and PA; Eric Grunden for generation advancement as well as field support; Alex Humphries, Matthew Schuckmann, and Laduska Sells for field support; and Melissa Jia and Brenda Lawrence for molecular genetic characterizations.

C O N F L I C T O F I N T E R E S T S T A T E M E N T
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Mention of a trademark or proprietary product does not constitute a guarantee or warranty of the product by the USDA or Louisiana State University and does not imply its approval to the exclusion of other products that also can be suitable. The USDA is an equal opportunity provider and employer. All experiments complied with the current laws of the United States, the country in which they were performed.