OsTPR boosts the superior grains through increase in upper secondary rachis branches without incurring a grain quality penalty

Summary

To address the future food security in Asia, we need to improve the genetic gain of grain yield while ensuring the consumer acceptance. This study aimed to identify novel genes influencing the number of upper secondary rachis branches (USRB) to elevate superior grains without compromising grain quality by studying the genetic variance of 310 diverse O. sativa var. indica panel using single‐ and multi‐locus genome‐wide association studies (GWAS), gene set analyses and gene regulatory network analysis. GWAS of USRB identified 230 significant (q‐value < 0.05) SNPs from chromosomes 1 and 2. GWAS targets narrowed down using gene set analyses identified large effect association on an important locus LOC_Os02g50790/LOC_Os02g50799 encoding a nuclear‐pore anchor protein (OsTPR). The superior haplotype derived from non‐synonymous SNPs identified in OsTPR was specifically associated with increase in USRB with superior grains being low chalk. Through haplotype mining, we further demonstrated the synergy of offering added yield advantage due to superior allele of OsTPR in elite materials with low glycaemic index (GI) property. We further validated the importance of OsTPR using recombinant inbred lines (RILs) population by introgressing a superior allele of OsTPR into elite materials resulted in raise in productivity in high amylose background. This confirmed a critical role for OsTPR in influencing yield while maintaining grain and nutritional quality.

Introduction

One of the major achievements of the Green Revolution was to alter the architecture of rice panicle from lax to more dense types (Khush, 2001). Rachis branching pattern in rice mainly depends on the timings of inflorescence meristem activity abortion, rachis branch meristem conversion to terminal spikelet meristem, and next order rachis branch formation to lateral spikelet formation (Itoh et al., 2005). A delay in the shift of lateral meristem identity to spikelet meristem identity leads to increased number of secondary rachis branches and spikelets (Yoshida et al., 2013). Many genes and quantitative trait loci (QTLs) associated with rice panicle architecture regulation have been cloned, which includes MOC1 (Li et al., 2003), LAX (Komatsu et al., 2003a), APO1 (Ikeda et al., 2007), FZP (Bai et al., 2016; Komatsu et al., 2003b), Gn1a (Ashikari et al., 2005), LOG (Kurakawa et al., 2007), DEP1 (Huang et al., 2009), WFP1 (Miura et al., 2010), LP (Li et al., 2011), FUWA (Chen et al., 2015) and CPB1 (Wu et al., 2016) among others. A major QTL, qSrn7/FZP, was recently identified to increase the number of secondary rachis branches in the three upper primary rachis branches of the panicle and also grain yield in chromosome segment substitution lines resulting from a cross between Koshihikari and Kasalath (Fujishiro et al., 2018). Several key candidate genes encoding phytohormones and transcription factors corroborated the potential role in governing panicle architectural traits. These includes, interruption of binding of an auxin response factor OsARF6 due to deletion in upstream regulatory region of FZP, a gene known to control TSRB, and grain yield (Huang et al., 2018). Similarly, increased expression of OsCKX2, a gene playing a role in cytokinin signalling, results in higher number of both primary and secondary rachis branches (Ashikari et al., 2005). Another transcription factor, DST, which directly activates OsCKX2 also regulates the number of primary and secondary rachis branches, leading to 63.8% increase in grain number per main panicle (Li et al., 2013).

Most of these studies focused on increasing yield by modifying the overall panicle architecture in terms of branching pattern and spikelet density. The frequency of secondary rachis branches within a panicle showed a major impact on spikelet number variability and panicle structural plasticity (Adriani et al., 2016). However, yield increases achieved by boosting secondary rachis branching are generally offset by the poorer end‐use quality of the grains which form in the lower sections of a highly branched panicle (Cheng et al., 2007; Matsue et al., 1994). Therefore, highly dense or compact panicles suffer grain filling problems (Mohapatra et al., 2011), subsequently failing to reach the expected yield potential (Panda et al., 2009; Sekhar et al., 2015) and thus widening the uniformity of quality traits among grains (Panda et al., 2015). This negative correlation between grain yield and grain quality is an important consideration in the context of the adoption of modern high‐yielding varieties (Panigrahi et al., 2019). Unlike the number of primary rachis branches, factors regulating the number of secondary rachis branches at the upper portion of the panicle remain to be elucidated. Thus, a priority for rice improvement is to identify genes which act to increase secondary rachis branching without compromising end‐use quality.

The trade‐off between yield increases for reduced grain quality is unacceptable in rice as value proposition defined through milling, appearance and cooking qualities remain important to value chain stake holders (Park et al., 2019). Since most high‐quality grains are found in the upper panicle region (Cheng et al., 2007; Matsue et al., 1994), it is perhaps more advantageous to focus on the upper part of the panicle to improve yield without compromising grain quality (Mohapatra et al., 1993). In the light of increased incidences of obesity (2 billion people) with rapid raise in type II diabetes (460 million people), it is important to improve the starch quality and lower the glycaemic index (GI) of rice varieties while ensuring sustained yield of rice staple. The GI of carbohydrate‐rich food depends mainly on starch. Most of the rice varieties are classified as high GI in nature due to low/intermediate amylose content and lack of resistant starch (Anacleto et al., 2019; Parween et al., 2020). To lower the GI, several high amylose mutants have been identified in rice; however, most of the high amylose mutants exhibit severe yield penalty with impairment in the grain quality (Jukanti et al., 2020).

Although studies showed that the number of secondary rachis branches is more important to potential yield than the number of primary rachis branches (Mei et al., 2006), no studies have been conducted to identify genes influencing the number of USRB with superior grains that are not compromised for quality traits. Hence, this study aimed to identify candidate genes associated with USRB in indica diversity panel using genome‐wide association studies (GWAS) and gene set analysis. The integrative genomics approach taken to identify central hubs of genetic variants in candidate genes through association (single‐locus, multi‐locus and targeted gene association) affirms to link it to gene regulatory networks of transcriptome data to identify novel regulators influencing USRB without compromising end‐use quality. Furthermore, we have explored the role of OsTPR in overcoming the yield barrier of high amylose breeding population and low glycaemic index advanced breeding lines which could be targeted for reducing hyperglycaemia while addressing market acceptability.

Results

Correlations between key panicle traits and spikelet number

The 310 members of the indica gene pool germplasm panel were phenotyped with a suite of 41 panicle architecture traits. The trait ‘total number of secondary rachis branches’ (TSRB) with wide range of phenotype plasticity exhibited a heritability (H ²) of 0.69 (Figure S1A). TSRB was highly positively correlated with the trait ‘total number of spikelets’ (TSP: r ² = 0.97), while ‘total number of primary rachis branches’ (TPRB) was moderately correlated (r ² = 0.64) with the TSP (Figure S1B). When the regions of the panicle (upper, middle and bottom) were considered separately, the r ² values between TSP and the number of upper (USRB), middle (MSRB) and bottom (BSRB) secondary rachis branches were estimated as, respectively, 0.66, 0.76 and 0.82, respectively (P < 0.05). The r ² values between TSP and the number of upper (UPRB), middle (MPRB) and bottom (BPRB) primary rachis branches were observed somewhat lower, with value of 0.40, 0.35 and 0.55, respectively, at 0.05 level of significance (Figure S1B).

The genetic control of USRB

A single‐locus GWAS was used to identify the genomic regions harbouring genes influencing various panicle architecture traits using a set of 747 087 high‐quality biallelic single nucleotide polymorphism (SNP) markers. The most prominent GWAS peaks identified for USRB were present on chromosomes 1 and 2 (Figure 1a) conferring the phenotypic variance explained (PVE) of 49% and for TSP trait this genetic region explained PVE of 37%. Out of the 230 associated SNPs, 33 mapped within the 592.6 kbp interval of chromosome 1 flanked by LOC_Os01g10680 and LOC_Os01g11660 (Table S1). Remaining 197 SNP loci were mapped to chromosome 2 in a 126.4 kbp genomic interval covering five distinct linkage disequilibrium (LD) blocks. These SNPs exhibited equal value of β‐coefficients reflecting same allele effects to the phenotype of interest (Figure S2A). Haplotypes within LD block 4 distinguished three haplotypes exhibiting significantly different USRB values suggesting underlying higher phenotypic variation of this region (Figure S2B).

Figure 1.

Association of OsTPR non‐synonymous SNPs with some traits related to yield and quality. (a) A Manhattan plot depicting genome‐wide associations, revealing multiple SNPs (highlighted in light green) mapping to chromosomes 1 and 2 at q < 0.05 with P < 1.53E^‐05. Within the gene model of the candidate gene OsTPR, six nonsynonymous SNPs and a SNP at promoter region affects a potential binding site for SPL3/SPL14 transcription factor/s (marked with red triangle); TPR domain marked in blue colour; CDS‐coding sequence; UTR‐untranslated region; kb‐kilo basepairs. (b) The presence of haplotype AGGATCA assembled from allelic variants in the OsTPR promoter region and coding sequence was associated with enhanced panicle branching and a higher TSP but a reduced level of grain chalkiness. Plants harbouring the haplotype TATTCCG produced grains with low amylose and higher per cent grain chalkiness. Asterisks in the boxplots indicate significant differences based on a two‐tailed Students’ t‐test (*, **, ***, ****: P < 0.05, <0.01, <0.001, <0.0001). (c) Appearance of panicles in selected germplasm lines carrying the superior haplotype AGGATCA showing high upper secondary rachis branching (USRB) while lines with inferior haplotype TATTCGG depict low USRB; an allele highlighted in red indicates its source from promoter region.

To further narrow down, a gene‐level analysis was conducted to identify the ten genes most strongly associated with variation in USRB (Table S2). Gene set analysis enabled us to narrow down to a few candidate genes LOC_Os02g50700, LOC_Os02g50790 and LOC_Os02g50799, which were prominently associated to USRB with a lower p‐value of 0.013 (Table S3). The two latter genes share homology with genes encoding a nuclear pore anchor protein, and both encode products involved in floral development based on the suggested gene ontology in Rice Genomes Project (Kawahara et al., 2013). The ‘AUGUSTUS’ software, following its optimization based on an indica type genepool training set, predicted these two genes to represent a single gene, OsTPR, of length 24 444 bp, encoding a 2050 residue polypeptide (Figure S2C). OsTPR sequence was identified as homologous to that of AtTPR, an A. thaliana translocated promoter region gene (Figure S3). Resequencing of OsTPR in the indica genepool germplasm panel revealed a total of 60 SNPs, of which six were identified as non‐synonymous SNP, responsible for changes in the polypeptide sequence. The two most likely functional alleles comprised a Gln1160His variant within the OsTPR domain (Figure S4) and a polymorphism in the promoter region affecting a potential SPL3/SPL14 transcription factor binding site (Figure 1a).

The correlation between OsTPR haplotype and panicle branching

A haplotype analysis based on seven most informative SNPs encompassing 6 non‐synonymous and 1 present in regulatory region in the OsTPR sequence revealed that haplotype1 (AGGATCA) was present in panel entries exhibiting a high USRB, while those exhibiting a low USRB harboured either haplotype2 (TATTCGG), haplotype3 (TATTCCG) or haplotype4 (TATACCG) (Figure 1b,c). The OsTPR promoter region present in all entries included binding motifs for both SPL3 (HUSRB: p = 0.0013, E = 0.6591; LUSRB: p = 0.0010, E = 0.5027) and SPL14 (HUSRB: p = 0.0062, E = 3.0470; LUSRB: p = 0.0045, E = 2.1844), although the associated p‐ and E‐values were lower for SPL3 (Table S4). Importantly, plants harbouring haplotype1 produced grains of intermediate chalk content (mean of 28%), in contrast to those harbouring haplotype3 of high chalk (mean content = 60%; Figure 1b). Interestingly, lines possessing haplotype 3 depicted lower amylose content with lower viscosity properties (Figure S5). Furthermore, a survey of the 3000 Rice Genomes database revealed that within indica types, haplotype1 was the most strongly represented while haplotype3 the least (Figure 2a). Conversely, haplotype1 was completely absent from the japonica type genepool. Underneath SNPs of identified haplotype in indica types are observed with more nucleotide diversity (π) than in other major sub‐populations (Figure 2b). Besides, phylogenetic analysis conducted in this study revealed OsTPR of Oryza sativa subsp. indica diverged from Oryza nivara, whereas Oryza sativa subsp. japonica diverged from Oryza rufipogon (Figure 2c).

Figure 2.

Haplotype distribution and nucleotide diversity exist in the OsTPR sequence. (a) The distribution of haplotypes of OsTPR in the sub species of rice. The AGGATCA haplotype was represented mostly within the indica genepool, while the TATACCG haplotype was revealed abundant in the japonica and other genepools. (b) Nucleotide diversity was identified based on the seven SNPs in OsTPR, which was assessed higher in the indica genepool than in either of the japonica or aromatic genepools. (c) The phylogenetic analysis of OsTPR shows that closest relative to the copy present in the indica genepool is the one harboured by O. nivara, while that for the japonica genepool copy is the one harboured by O. rufipogon.

Genetic variants associated with rachis branching

To precisely identify the genomic regions furthermore with small to medium effect associated loci with panicle branching and TSP, a multi‐locus GWAS analysis was then applied resulting to the identification of an additional 208 loci distributed across the whole genome. Among these loci, 37 genes either encoding a transcription factor or responsive to phytohormone(s) were associated with a variation in primary or secondary rachis branching with small to an intermediate effect QTLs (Table S5). These include, the upstream region of HOX4 was particularly associated with variation in TSP (LOD score 7.84), a non‐synonymous SNP in a FAR1‐related gene exhibited variation for TSRB (LOD score 4.1), a SNP downstream of TAWAWA1 tags variation for TPRB (LOD score 7.79), a number of polymorphisms in the MYB4P and MYB/SANT sequences showing variation for USRB, SNPs in UBC6 and an AP2 for variation in MSRB, and FBOX75 and FBOX252 for variation in BSRB (Table S5). Only one of the genes identified (TAWAWA1) by the multi‐locus GWAS analysis has been previously shown to influence panicle architecture. In addition, the key candidate genes previously identified for panicle branching through transgenic and mutant work were validated to confer minor effect association with TSP, TPRB, and various secondary rachis branching traits (USRB, MSRB, BSRB, TSRB) (Table S6).

Furthermore, when association networks for the set of genetic variants were assembled, it was apparent that OsTPR together with DWARF11 and MOC1 superior haplotypes contributed to an increase in TSP, as a result of their effect on rachis branching (Figure 3). In order to assess the total effect of each of candidate genes OsTPR, DWARF11, MOC1 and MYB4P on USRB and TSP, we evaluated the per cent phenotypic variation explained (PVE) for them. For OsTPR, the highest PVE of 11.69 for USRB was noted, while DWARF11, MOC1 and MYB4P showed the PVE of 0.06, 0.10, and 3.69, respectively. Similarly, OsTPR showed the relatively highest PVE for TSP with value of 7.26, over other three DWARF11, MOC1 and MYB4P, which showed PVE of 2.57, 1.87, and 1.55, respectively. Plants harbouring the optimal haplotype for each of DWARF11, MOC1, MYB4P, and OsTPR exhibited additional total spikelet increase (mean value of 250) due to their enhanced rachis branching phenotype (Figure 4); on the other hand, they developed slightly fewer panicles per plant. These pyramided lines with higher TSP exhibited the trend of increased weight of filled spikelets per plant contributing as yield component trait, along with relatively superior values of key grain quality parameters, namely peak viscosity, breakdown and gelatinization temperature (Figures 4 and S5).

Figure 3.

Association network of genes regulating panicle branching and total spikelets (TSP) in the indica genepool based on single‐ and multi‐locus GWAS and targeted association analyses. Multiple SNPs were significantly (P < 0.05) associated with secondary rachis branching at each of the bottom (BSRB), middle (MSRB), and upper (USRB) sections of the panicle, as well as over the entire panicle (TSRB); with total primary rachis branching (TPRB) and with TSP; based on multi‐locus GWAS (blue nodes) and targeted association analysis (red nodes). SNPs associated with multiple components of yield were identified in the OsTPR (pink edge), DWARF11 and MOC1 sequences. Edge associations were based on the respective value of β‐effects.

Figure 4.

Variation in panicle architecture, grain yield and grain quality traits in the indica genepool associated with haplotype variation at OsTPR, D11, MOC1 and MYB4P. The TSP of entries classified as superior (shown in green) is 1.5‐fold higher than the global mean of 147.6, while that of entries classified as inferior (orange) is 1.5‐fold lower. With only few exceptions, the entries with higher TSP harbour the superior haplotypes AGGATCA _TPR AAT _D11 , CG _MOC1 , A _MYB4P . The superior entries exhibit a significantly higher TSP, more extensive panicle branching and a higher grain weight, without any penalty with respect to per cent head rice yield, grain chalkiness, the size of the cooked grain, grain amylose content, gel consistency, trough viscosity, final viscosity or retrogradation. The number of panicles per plant was somewhat reduced in the superior entries. Asterisks in the boxplots indicate significant differences based on a two‐tailed Students’ t‐test (*, **, ***, ****:P < 0.05, <0.01, <0.001, <0.0001).

Integrating allelic variants with co‐expression networks governing panicle development

An attempt to rule out spurious associations arising from the GWAS procedure was made by adding an analysis of the co‐expression networks governing panicle development stage, which lead to the identification of distinct modules (Figure 5a). Results of co‐expression analysis showed MYB4P and OsTPR displayed a high degree of connectivity (Figure 5b), with allelic variation on both genes exhibits a strong association with variation in the USRB phenotype (Figure 3). Further analysis of the closest neighbours of OsTPR identified 243 associated nodes (Figure 5b). The predominant functional categories of the set of genes co‐expressed in the OsTPR sub‐module were ‘cell cycle organization’, ‘RNA regulation’, ‘development’, ‘protein synthesis’ and genes encoding a wide array of transcription factors associated with OsTPR (Figure 5c, Tables S6 and S7). Several genes controlling panicle branching associated with OsTPR displayed a particularly high degree of connectivity (Figure 5): these included GRF4 (LOC_Os02g47280) which is associated with variation in USRB; SPL7 (LOC_Os05g45410) with variation in MSRB; and APO2/RFL (LOC_Os04g51000), and LAX1 (LOC_Os01g61480) with variation in BSRB. These genes also featured in the targeted association analysis (Table S5). The same module featured several members of the SPL gene family, including SPL14/IPA1/WFP (LOC_Os08g39890), SPL5 (LOC_Os07g10390), SPL4 (LOC_Os02g07780) and SPL (LOC_Os01g01080). Other key genes linked with panicle architecture which were strongly co‐expressed with OsTPR included GAD1 (LOC_Os08g36320), SPPL1 (LOC_Os10g25360), DWARF11 (LOC_Os04g39430), MOC2/FBP1 (LOC_Os01g64660), FON1 (LOC_Os06g50340) and MFT1 (LOC_Os06g30370). Furthermore, notably FLO11 (LOC_Os12g14070) exhibiting role in amyloplast biosynthesis in rice endosperm during seed development was found strongly co‐expressed with OsTPR, while FLO16 (LOC_Os10g33800) was observed in same blue module where OsTPR expressed (Table S7). Besides, genes involved in the hormone metabolism, such as gibberelin synthesis (LOC_Os01g11150), auxin regulation (LOC_Os07g38890), ethylene responsive transcription factors (LOC_Os03g12950) and brassinosteroid synthesis (LOC_Os06g39880) were revealed to be strongly co‐expressed.

Figure 5.

Regulatory networks derived from an analysis of co‐expression and protein–protein interactions. (a) Co‐expressed modules (shown in different colours) derived from the hierarchical clustering of transcriptome data involved in panicle development, sampled at five developmental stages (panicle length > 1 mm, 3–5 mm, 10–15 mm, 40–50 mm and at heading). (b) Network of genes co‐expressed in the blue module containing OsTPR and interactions at an edge weight of >0.30. Node sizes represent the degree of connectivity, with the central hub gene marked by a thick black border. OsTPR was strongly co‐expressed with a number of genes known to influence panicle development, including DWARF11, MOC2, LAX, SPL14 and APO2. (c) Based on MapMan annotation, the first neighbours of OsTPR within the blue module are involved, inter alia, in protein metabolism and RNA processing, including SPL14, D11, and MOC2 which regulate panicle architecture. (d) OsTPR’s predicted first‐shell protein interactors and molecular actions emphasize the importance of its interactions with proteins involved in SUMOylation, ubiquitination and the determination of plant architecture.

In addition, strong co‐expression of Hd18 (LOC_Os08g04780) controlling the flowering time was also observed with OsTPR (Table S8). Besides, co‐expression of other genes regulating flowering time namely OsMADS15 (LOC_Os07g01820) and OsMADS56 (LOC_Os10g39130) was also observed in the same blue module (Table S7).

Further experimental evidence for the importance of the OsTPR product was obtained from a characterization of relevant protein–protein interactions. The nuclear pore protein (encoded by LOC_Os11g42420) bound with LIC (LOC_Os06g49080), a transcriptional activator involved in the regulation of plant architecture, and also interacted with SAC3/GANP proteins (LOC_Os07g45160, LOC_Os03g22870) potentially involved in nuclear mRNA export. Moreover, this nuclear pore protein activates OsTPR as well as an mRNA‐splicing factor ATP‐dependent RNA helicase (LOC_Os02g19860) which binds with the MYB4P transcription factor, further highlighting the strong correlation of OsTPR and MYB4P found in the co‐expression network analysis. According to gene ontology, two Ulp1 proteases (LOC_Os03g29630 and LOC_Os03g22400) and also a small ubiquitin‐like modifier (SUMO) protease (LOC_Os01g25370) are associated with floral development (GO:0009908), hydrolase activity (GO:0016787) and protein modification (GO:0006464). These three proteases all localize to the nucleus and potentially interact with the products of LOC_Os01g25370 and LOC_Os03g22400. The prediction is that OsTPR is able to interact with other proteins involved in either ubiquitination or SUMOylation, thereby influencing both plant architecture and floral development (Figure 5d).

Further meta‐analysis was performed with 3116 rice samples in the Genevestigator database which consists of a comprehensive collection of public microarray and RNA‐Seq study results (Hruz et al., 2008). OsTPR (represented by LOC_Os02g50790) was shown to be highly expressed at panicle initiation during stem elongation stage (mean = 15.06, SD = 0.69, n = 89) and also during booting stage (mean = 14.83, SD = 0.84, n = 89; Figure S7A). Across 54 anatomical parts, highest mean levels of OsTPR expression were found in panicle cell types and shoot apex while the expression of MYB4P was found in the panicle branching (Figure S7B). Two‐gene plot analysis showed that OsTPR is expressed at different panicle stages and showed particularly high expression during early stages of panicle development (0–4 cm) while MYB4P might be expressed at latter stages of panicle development (6–16 cm) (Figure S7D). A survey on nine developmental stages showed that OsTPR is highly correlated (r ² > 0.917) with multiple genes encoding different proteins related to flower development such as homeobox protein, frigida, MYB family transcription factor, among others (Figure S7E). Similarly, MYB4P is highly correlated (r ² > 0.96) with multiple genes encoding proteins related to flower development (Figure S7F).

Role of OsTPR in overcoming the yield barrier of high amylose and low glycaemic index breeding population and advanced breeding lines

Among the progenies of F₅‐derived RILs, substantial phenotypic variation was observed for TSP (66–403), TPRB (5–15) and USRB (1–17). Lines with the superior OsTPR haplotype not only exhibited significantly higher USRB compared with the lines carrying inferior OsTPR haplotype, but also showed yield advantage over the lines containing inferior haplotype (Figure 6). Path correlation analysis also showed a direct negative effect of USRB (standardized parameter estimate (std) = −0.53) and MPRB (std = −0.57) to TSP in samples with inferior OsTPR haplotype (Figure S6A), while lines with superior OsTPR haplotype have a positive effect of TSP to USRB (std = 0.38; Figure S6B). Interestingly, the lines with both superior haplotypes combination from OsTPR (AGGATCA) and SBEIIb (A) showed USRB to be a non‐limiting factor to increase TSP with increase in yield while combining superior haplotypes for high amylose content with a low GI property (Figures 6 and S8; Table S9). The significant differences were also observed among RIL groups combining OsTPR and SBEIIb haplotypes in terms of other grain quality traits such as viscosity, trough, breakdown, setback, peak viscosity time, pasting temperature and retrogradation (Figure S8). In general, RILs having the superior SBEIIb haplotype have higher amylose content, pasting temperature and peak viscosity time, and lower viscosity, retrogradation, setback, trough and breakdown compared to RILs with inferior SBEIIb haplotype.

Figure 6.

Recombinant inbred lines with OsTPR superior haplotype showed increased USRB and yield component traits. Breeding scheme followed with Samba Mahsuri and IR36ae (high amylose), as two parental cultivars, which were crossed, and advanced until F6 generation through single seed descent using rapid generation advancement (RGA) method. The F5:6 progenies derived from this cross combination carrying the superior OsTPR haplotype showed higher values for USRB and yield components traits in the boxplots; OsTPR + SBEIIb haplotypes with amylose and GI phenotypes in the box plots; representative panicles images of progenies carrying OsTPR superior haplotype. Asterisks in the boxplots indicate significant differences based on a two‐tailed Students’ t‐test (*,**,***,****: P < 0.05, <0.01, <0.001, <0.0001). Selected lines with higher yield, high amylose and low GI properties along with its haplotype information are listed in the table.

Previously identified three low GI gene bank lines (Anacleto et al., 2019) were further validated using in vivo method, and the modern breeding lines IRRI 162 and IRRI 163 in vivo GI values were in the range of 57 and 64, respectively. The 5 low GI germplasm were scrutinized for OsTPR haplotypes and subjected to large plot based yield estimation and phenotyping of panicle architectural traits (Figure 7). Two IRRI breeding lines IRRI 162 and IRRI 163, which possessed the superior OsTPR haplotype, produced a total yield of 6.40 tons/ha and 6.9 tons/ha, respectively. Similarly, these two breeding lines also represented with 2–5 more USRB and about 50 more spikelets than other lines with inferior haplotypes. The low GI gene bank line GQ02522 which possessed an inferior OsTPR haplotype (haplotype4: TATACCG) produced the lowest USRB and yield (3.84 tons/ha).

Figure 7.

Haplotype information, yield and panicle traits of low GI lines. A set of five low GI lines identified using in vitro and in vivo GI estimation were scrutinized for the presence of different haplotypes of OsTPR and its yield has been estimated using large plots and different panicle traits were assessed. Letters within bars indicate significant differences among cultivars/varieties based on the Tukey honest significant differences at 95% confidence level.

Discussion

In the past, a number of attempts have been made to genetically dissect variation in architecture of the rice panicle by focusing on the identification of the genetic basis of the number of primary rachis branches, the length of the panicle and/or TSP. However, the power of these analyses has been largely constrained by the use of low or at best mid‐resolution genotypic data (Rebolledo et al., 2016; Ta et al., 2018). In the present study, high‐density genotypic data were employed to dissect the genetic basis of panicle architectural traits using resequencing resources. Furthermore, the present experiments have revealed that it is a higher number of secondary rather than of primary rachis branches which is more strongly correlated with an increase in TSP (Figure S1B). Nevertheless, the potential limitation of focusing on grains located on bottom portion of secondary rachis branches is that they are believed to be of inferior quality (Das et al., 2018); hence, we need to target upper secondary rachis branching to increase yield potential without affecting grain quality.

OsTPR participates with known flowering and panicle architecture genes to influence USRB

The candidate gene, OsTPR was not found homologous to any rice family genes. Its orthologue, in A. thaliana, protein AtTPR has been shown to participate in SUMOylation and RNA export (Xu et al., 2007) and is thought to be involved in nuclear export (Jacob et al., 2007; Krull et al., 2004). Mutation of AtTPR results in acceleration in flowering, a reduction in the number of terminal flowers formed, and loss in fertility and apical dominance (Jacob et al., 2007). The function of AtTPR is believed to be similar to that of the related proteins SAR3 and HST, which are associated with auxin sensitivity and flowering time, respectively (Jacob et al., 2007). Here, the rice OsTPR protein was shown to interact with SAC3/GANP proteins – LOC_Os07g45160 and LOC_Os03g22870, which are homologs of, respectively, the A. thaliana proteins AtSAC3B and AtSAC3C. AtSAC3B, a component of a transcription and mRNA export complex, regulates DNA methylation, epigenetic gene silencing and production of small interfering RNAs (Yang et al., 2017). Most of rice OsTPR protein interactors were SUMO proteases predicted to be involved in floral development, hydrolase activity and protein modification. The predicted SUMO protease interactor was orthologous with AtESD4 which when mutated results in early flowering and dwarfism (Murtas et al., 2003). A further direct interactor with rice OsTPR was LIC, a protein known to negatively regulate brassinosteroid signalling and also to exert a positive influence on the number of rachis branches and total number of grains set per panicle (Wang et al., 2008).

The observation that OsTPR was co‐expressed along with a number of genes regulating rachis branching and other panicle traits (namely SPL4, SPL5, SPL14, GRF4, GAD1, APO2, LAX, FON1, MFT1 and SPPL1) implies that OsTPR too participates in panicle development (Figure 5). Remarkably, various genes acting upon both primary and secondary rachis branching were identified here by GWAS and shown to be members of a co‐expression network; the suggestion is that they are regulated similarly and that their products share a common function (Wolfe et al., 2005). LAX is identified as co‐expressed candidate and controls the initiation of axillary meristem in rice, which decides the panicle branches and lateral spikelets in the panicle (Komatsu et al., 2003a). The application of multi‐locus GWAS and a targeted association approach, followed by the derivation of co‐expression networks have identified the importance of both OsTPR and MYB4P to the determination of USRB. In addition, DWARF11 and MOC1 have been identified as central hub proteins influencing TSP via their influence over TPRB, MSRB and BSRB. The GWAS results pointed to the evidence that OsTPR explains higher PVE to USRB, and thus, a major QTL contributing factor, while MYB4P, DWARF11 and MOC1 confer lower PVE. DWARF11 encodes a P450 superfamily protein involved in brassinosteroid synthesis; the His360Leu substitution results in the formation of panicles characterized by a cluster of primary branches (Wu et al., 2016). A point deletion in the DWARF11 promoter reduces grain size, rachis length, TSP and secondary rachis branching, while overexpressors of DWARF11 set a higher number of grains, thereby enhancing yield (Zhou et al., 2017). In addition, a haplotype of MOC1, a gene which encodes a GRAS family transcription factor affecting tillering (Li et al., 2003), was associated with an improved performance with respect to both BSRB and TSP. Plants carrying superior alleles at each of DWARF11, MOC1, OsTPR and MYB4P enjoyed a distinct advantage with respect to higher TSP due to increased secondary branching at upper and lower part of the panicle, with trend of impairing grain quality. A number of the indica type germplasm entries harboured the optimal OsTPR, DWARF11, MOC1 and/or MYB4P alleles identified in the diversity panel (Figure 4).

The upstream sequence of OsTPR includes a number of SNPs predicted to interfere with the binding of SPL3 and SPL14 and the transcription factors which include an SBP domain, comprising 76 highly conserved residues, which influence both flowering time and panicle branching (Preston and Hileman, 2013). One of the SNPs in the OsTPR promoter sequence was associated with variation in USRB and included a matching with cis motif of the SPL3 or SPL14 binding site; in addition, several non‐synonymous SNPs in the OsTPR coding sequence contributed to haplotype1, the presence of which was notably correlated with a higher TSP (achieved by an increase in USRB), without any compromise to grain quality. Future experiments targeting the functional validation of OsTPR through CRISPR‐Cas9 will be valuable. The product of SPL3 directly activates other flower‐related transcription factors and also regulates floral meristem identity (Yamaguchi et al., 2009).

Allelic variation at OsTPR influences USRB and grain quality in the indica genepool

The indica germplasm entries harboured the superior OsTPR haplotype exhibited the greatest yield potential and produced the highest quality grain. This finding corroborates with the PVE reflected by each of the genes, where OsTPR turned out to be a major gene for USRB, and showing strongest effect on USRB and TSP in our germplasm panel. The gene was preferentially not selected in japonica subspecies and moderately represented in indica genepool for higher yield without compromising grain quality during the rice breeding history. This is supported by the observation that the OsTPR has higher nucleotide diversity within cultivated indica genepool. On the contrary, this gene did not display appreciable nucleotide diversity in japonica genepool suggesting lack of intense selection pressure (Wang et al., 2018). Among aromatic germplasm, the level of nucleotide diversity in OsTPR was low, implying that strong selection for aromatic grain has resulted in the selection of only a few of the possible haplotypes.

The inferior haplotype (TATTCCG) represented at a lower percentage in indica germplasm is also being associated with lowering amylose content and high percentage of chalk (Figure S5). Notably, FLO11 encode the heat shock protein 70 involved in regulation of grain amyloplast development in developing endosperm, found to affect grain chalkiness (Zhu et al., 2018) especially under high temperature (Tabassum et al., 2020) and was found to be strongly co‐expressed with OsTPR. Similarly, co‐expression of another candidate FLO16 was found in same blue module with OsTPR, which plays a pivotal role in redox homeostasis central for forming compound starch grain and regulating starch biosynthesis in rice endosperm (Teng et al., 2019). These finding suggest the likely potential participation of OsTPR in controlling starch biosynthesis and thereby rice grain quality.

Role of OsTPR to raise the yield potential of rice in high amylose and low GI lines

The concept of GI is an important parameter which estimates the rate at which carbohydrates break down and absorbed into the bloodstream upon digestion, such that foods with low GI (≤55) can be digested, absorbed and metabolized more slowly and they elicit lower blood glucose levels as compared to foods with higher GI values (>70) (Jukanti et al., 2020; Fuentes‐Saragoza et al.,2011). In general, rice with low GI values have high amylose content rendering more resistance to enzymes due to its helical chain structures that tend to form compact matrices (Lu et al., 2013). Amylose content affects resistant starch which is the proportion of starch not converted into glucose by amylase after two hours of digestion, giving improved control in blood glucose levels (Parween et al., 2020; Sajilata et al., 2006). Although large number of high amylose mutants identified in rice which is demonstrated to have significant human health benefits with increased resistant starch content and lowered GI, these are not recruited to the breeding programmes due to lower yield potential, increase in chalkiness and impaired cooking quality (Jukanti et al., 2020). Silencing SBEIIb through artificial microRNA resulted in elevating amylose and substantially lower GI to 46 (Butardo et al., 2011). The presence of superior OsTPR haplotype in F₅‐derived F₆ RIL mapping population confer to increase in USRB and overcome the yield barrier by efficiently combining haplotypes for OsTPR (Samba Mahsuri parent) and SBEIIb (IR36 amylose extender) mutant with high amylose. As a result, advanced RILs with elevated amylose (more than 30%) with low GI (median of 38) exhibit good yield potential were identified. Likewise, the advantage of combining OsTPR was also evidenced in key low/intermediate GI varieties. IRRI breeding lines possessing OsTPR superior haplotype showed marked advantage for yield over other gene bank line carrying corresponding inferior haplotypes among the low/intermediate GI lines.

In summary, a survey of a panel of 310 indica type rice accessions resulted in the identification of a number of genes putatively involved in enhancing secondary rachis branching. One of these genes was OsTPR, a homolog of the Arabidopsis thaliana gene AtTPR, which encodes a nuclear pore anchor protein, a likely homolog of the vertebrate translocated promoter region protein. The performance of plants harbouring the superior haplotype of the OsTPR (AGGATCA), in combination with high amylose alleles of SBEIIb (A), was notable with respect to the increased number of panicle branches and spikelets formed, with the number of superior grains set per plant with increased amylose of 30–34%, confirming a low GI property. Recruiting high throughput in vitro GI estimation of diversity germplasm and breeding lines and scanning for the presence of superior haplotypes of OsTPR was found to be a useful strategy to identify low GI lines with high yield potential. In conclusion, a rational breeding strategy aiming to increase the yield potential of rice without degrading the quality of the grain would be next to introgress the superior allele of OsTPR into japonica germplasm to enhance yield potential.

Methods

Experimental design

A set of 310 indica gene pool entries with days to maturity of at most 140 was selected from 3000 resequenced rice genomes (The 3000 rice genomes project, 2014) and grown in completely randomized block design with four replications at experimental station of the International Rice Research Institute (IRRI), Laguna, Philippines (14°10’N, 121°15’E) during the 2015 dry season. Standard field and crop management practices of IRRI were applied uniformly for all replicates.

Phenotyping and transformation

Panicles from the main culms of two plants for each accession were manually harvested prior to full grain ripening and phenotyped using the Panicle Traits Phenotyping Tool (AL‐Tam et al., 2013). Phenotypic means were transformed using WarpedLMM to ensure the Gaussian distribution of residuals (Fusi et al., 2014). Correlation analysis was performed using Pearson’s correlation method at α = 0.05.

Genotypic data

SNP data of 1 680 146 biallelic loci extracted for the 310 selected indica genepool entries of the genomic sequences acquired by the 3000 Genomes Rice Project (The 3000 rice genomes project, 2014) using PLINK 1.9 software (Chang et al., 2015). Loci for which the missing call rate was >0.20%, along with those for which the minor allele frequency was <10%, were ignored in order to minimize the effect of false positives (Tabangin et al., 2009). The end dataset comprised 747 087 high‐quality SNPs scored across 309 of the original 310 entries.

Single‐locus GWAS

Using Efficient Mixed Model Association eXpedited (EMMAX) ver. Beta (Kang et al., 2010), a variance component model was implemented to estimate fixed effects and test the association of each SNP to various panicle architectural traits while accounting for the effect of genetic relatedness from the computed marker‐based Balding‐Nichols kinship matrix to address population structure (Balding and Nichols, 1995). Genomic associations were tested based on a generalized least square F‐test (Kang et al., 2010; Kariya and Kurata, 2004). False discovery rate represented by q‐values was calculated using Bioconductor’s q‐value package (Storey and Bass, 2019). The Manhattan and quantile–quantile plots were created using R v1.0.153 software (R Core Team, 2018) in order to visualize significant genomic regions and assess the reliability of the associations. The GWAS significance threshold (red line) was set based on the Bonferroni‐corrected α′ = 6.69266096e‐08, calculated as α′ = 0.05/s, where s was the total number of SNPs used. The suggestive significance threshold (blue line) was set to P < 1.0000e‐5 and SNPs associated with a q‐value of <0.05 were highlighted, following Lin and Lee (2015). Haplotype blocks were determined using the blocks function implemented in PLINK 1.9, and the HaploView program (Barrett et al., 2005) was used to determine tag SNPs based on a threshold of the LD coefficient (D′) set to 0.80.

Gene and gene set analyses

Gene and gene set analyses were performed using the MAGMA v1.06 program (de Leeuw et al., 2015). SNPs identified as being associated with trait variation through the use of GWAS were mapped to a genomic location using the Rice Genome Annotation Project (Kawahara et al., 2013). The SNP‐wise mean model was used in MAGMA for the gene‐level analysis, incorporating the genotypic data and the P‐values of SNPs computed from EMMAX, and also to estimate the extent of LD between SNPs. After calculating correlations of neighbouring genes and other gene‐level metrics such as Z‐score and P‐value per gene, various sets of genes were created by removing one gene from the set of the top ten genes based on the P‐value for every gene set. A competitive gene set analysis was then conducted based on a linear regression model from gene‐level analysis in order to test whether or not the mean association of the genes in the gene set was greater than that of the genes not in the gene set (de Leeuw et al., 2015).

Multi‐locus GWAS analysis

Multi‐locus GWAS was performed using the multi‐locus random‐SNP‐effect mixed linear model (mrMLM) function implemented in R software as described elsewhere (Misra et al., 2018). The inputs were the transformed phenotypic data, the filtered genotypic data and a kinship matrix for each of the panicle traits TSP, BRSB, MRSB, USRB, TSRB and TPRB. The SNP data used for the single‐locus GWAS were recoded into ped and map files using the recode function provided in PLINK 1.9, and then converted into a diploid hapmap file using TASSEL5 software (Bradbury et al., 2007). The mrMLM and FAST multi‐locus random‐SNP‐effect EMMA (FASTrEMMA) methods were implemented, applying a critical LOD threshold of 3 to ascribe significance to a quantitative trait nucleotide (QTN). A search radius of 20 was chosen to identify potentially associated QTNs. The parameter applied for the FASTrEMMA method was restricted maximum likelihood. The resulting sets of identified QTNs were validated by reference to the single‐locus GWAS outcomes.

Targeted association analysis

Biallelic SNPs lying within 2 kbp upstream and 1 kbp downstream of each gene associated with panicle architecture were identified by aligning the resequencing data acquired from the set of 309 accessions against the cultivar Nipponbare genome sequence (Kawahara et al., 2013). These were then tested for association with each trait of interest (TSP, TPRB, TSRB, USRB, MSRB, and BSRB) using mixed linear model. SNPs associated with a P‐value of <0.05 were considered significant, and were assembled to construct the haplotypes. Pairwise comparisons between haplotypes were based on Students’ t‐test and P‐values were adjusted using Holm’s method (Holm, 1979). Results were visualized using boxplots created using the R‐based ggplot2 package (Wickham, 2016).

Association networks

Cytoscape software (Shannon et al., 2003) was used to visualize the inter‐relationship of the SNPs associated with USRB, MSRB, BSRB, TSRB, TPRB and TSP. SNPs associated with a LOD score >3.0 according to the multi‐locus GWAS, those associated with a significant (P < 0.05) β value according to the single‐locus GWAS and those associated with a significant (P < 0.05) β value identified through the targeted association within known panicle architecture‐related genes, were merged and then transformed into a network input file (source node: traits; target nodes: SNPs; interaction: β values). The source and target nodes of the network are the panicle traits and SNPs, respectively. The connections among these nodes are the edges coding the significant β‐values. Each trait (source) is represented as big solid circle and highlighted as lemon green for USRB, light green for MSRB, brick red for BSRB, pink for TSRB, blue for TPRB and aquamarine for TSP. The SNPs (target) associated with each of trait are shown as blue (based on multi‐locus GWAS) or red (based on targeted association) solid circles with sizes varying according to their degree of connectivity. The edges that link the source and target nodes are highlighted in grey and pink colour lines. The pink edges are the ones connecting to more than two panicle traits.

Co‐expression network analysis

Public domain microarray data (GSE19024 and GSE41556) were extracted as CEL files from six panicle developmental stages (Panicle1 (length < 1 mm), Panicle2 (3–5 mm), Panicle3 (10–15 mm) (three biological replicates of each stage), Panicle4 (40–50 mm), Panicle5 (heading stage) (two biological replicates of each stage) involving the cultivars Zhenshan 97 and Minghui 63; heading stage data obtained from cultivar Nipponbare was represented by three biological replicates. A GCRMA‐based normalization of these microarray data was performed using the justGCRMA function (Wu et al., 2019) implemented in R software. A set of 47 panicle tissue samples and 5,041 genes transcribed in the panicle were selected from the KnetMiner database (Hassani‐Pak, 2017), and differentially expressed genes already reported across the developing panicle tissues in rice (Kudo et al., 2013; Wang et al., 2010). The WGCNA package (Langfelder and Horvath, 2008) implemented in R software was used to identify clusters of correlated genes. Using the network construction algorithim described by (Zhang and Horvath, 2005), first the correlation matrix (coefficient < 0.75) was transformed into an adjacency matrix which was then converted into a topological overlap (TO) matrix. Subsequently, hierarchical clustering was performed on TO similarity using cutreeDynamic function to identify clusters (Zhang and Horvath, 2005). Eigenvectors for each cluster were calculated and those having a similar eigenvalue (threshold 0.25) were merged using the merge Close Modules function (Zhang and Horvath, 2005). Nodes having an adjacency value of zero were removed, and edges with a weight >0.30 were selected. Cytoscape software was used to visualize the final co‐expression network: genes were represented as nodes and connection strengths as weighted edges. The degree of connectivity was calculated to identify hubs. Genetic variants strongly associated with various panicle architecture traits were traced in the gene networks and interactions were highlighted.

Identification of cis motifs

Potential binding sites of transcription factors affecting the expression of candidate genes were determined using TomTom software (Gupta et al., 2007). Probe sequences of length 31 nt overlapping each of the significant SNPs detected via the targeted association method were reverse complemented and then used as input, applying an E threshold of <10. The motifs were compared with entries in the JASPAR Core plant non‐redundant DNA database (Khan et al., 2018).

Protein–protein interactions

Predicted protein–protein interactions involving OsTPR were determined using STRING‐DB v10.5 software (Szklarczyk et al., 2017), applying a confidence level of 0.40 for the minimum required interaction score. Supporting evidence for the interactions include experimental evidence, text‐mining, co‐expression data, and curated database.

Gene prediction

According to MSU 7.0 annotation, LOC_Os02g50790 and LOC_Os02g50799 represent two adjacent genes of identical biological function, sharing extensive homology with heterologous genes encoding OsTPR proteins. The relevant genome sequence was inputted into FGENESH software (Solovyev et al., 2006) and default parameters applied in order to derive a gene model. Sequences acquired from maize, rice, wheat and barley were used as a training set. The resulting gene model was validated using the AUGUSTUS program (Stanke and Waack, 2003). Gene structures were visualized using GSDS v2.0 (Hu et al., 2015).

Multiple sequence alignment and phylogenetic analysis

A multiple sequence alignment was performed using ClustalW software (Thompson et al., 1994), involving sequences present in 13 species, including representatives of the japonica and indica genepools. The japonica type OsTPR sequence was used to identify orthologs/homologs in other taxa (Zhang et al., 2016). Protein sequences represented in the Ensemble database (plants.ensembl.org/index.html) were used to identify the orthologs/homologs via a BLAST search (Altschul et al., 1990). The MEGA X program (Kumar et al., 2018) was used to create a phylogenetic tree, applying 1,000 bootstrap replicates to quantify confidence in each node. Evolutionary distances were calculated using the Poisson correction method (Zuckerlandl and Pauling, 1965). A further multiple sequence alignment of predicted protein sequences encoded by eight contrasting rice entries (four samples producing panicles with a high and four with a low USRB) was included.

Quantification of nucleotide diversity

Seven SNP positions within the OsTPR coding sequence and 15 within the local LD block were obtained based on the SNP‐Seek database (Alexandrov et al., 2015). The full population was separated into sub‐populations based on the classification provided by the SNP‐Seek database. Minor allele frequencies were calculated for each SNP at the sub‐population level. The level of nucleotide diversity was quantified using the VCF program (Danecek et al., 2011).

Development of RILs with OsTPR and SBEIIb superior haplotypes

To verify the significance of candidate OsTPR, an advanced recombinant inbred lines (RILs) of F₅‐derived F₆ mapping population derived from the two parents carrying contrasting superior haplotypes for OsTPR (Samba Mahsuri parent) and SBEIIb (IR36 amylose extender) was generated. The F5‐derived F6 recombinant Inbred lines (RILs) mapping population were developed by employing the hybridization between Samba Mahsuri—a widely preferred variety in India with excellent cooking and eating qualities and possesses superior haplotype of OsTPR and inferior haplotype of OsSBEIIb, and IR36ae—a high amylose line due to mutation in SBEIIb (de Guzman et al., 2017) but poor yielding and possesses inferior haplotype of OsTPR and superior haplotype of OsSBEIIb. In F_2:3 generation, a total of ~5500 progenies of IR36ae x Samba Mahsuri were raised. Upon examining the SNP variations exist in SBEIIB (LOC_Os02g32660) gene between both parents, one SNP (snp_02_19362520, A→G) previously reported by Butardo et al 2012, located at the exon‐intron junction (Exon 11 coordinates 19362521–19362640) was genotyped using Kompetitive allele specific PCR (KASP) marker assay. After assessing their fit to the Mendelian segregation ratio, a total of 369 F₃ progenies were selected and eventually forwarded to advanced generation until F₅, utilizing single seed descent (SSD) method through rapid generation advancement (RGA). During the mid‐2019WS season, a set of selected 298 F₅ progenies were grown in greenhouse using randomized complete block design (RCBD) with 2 replications and KASP genotyping assay was performed for OsTPR haplotypes using the KASP markers (designed based on allelic variations in OsTPR haplotype). Furthermore, a total of 298 progenies (298*2 = 596) from the F_5:6 RILs population were phenotyped for panicle architectural traits using GIMP and P‐TRAP software and scored for different agronomic traits and yield potential.

To analyse the KASP genotyping outputs, samples bearing at least one missing call for the significant alleles in each candidate gene were removed. In addition, lines containing heterozygotes were also removed when forming the haplotypes. Lines possessing superior and inferior haplotypes with homozygous alleles for OsTPR and OsSBEIIb were identified and phenotypic box plots were derived where their significance was tested using the t‐test. Boxplots were constructed using ggplot package in R after removing heterozygotes. Path correlation analysis was performed using the model below:

'TSP ~ RXL + USRB + UPRB + USRB + UPRBL + USRBL + BPRB + BSRB + BPRBL + BSRBL + MPRB + MSRB + MSRBINT + BSRBINT + USRBINT.

USRB ~ TSP + UPRB + UPRBL + USRBL + USRBINT + RXL'.

The total number of spikelets per panicle (TSP) was mainly used as the effect variable to understand contributing panicle features to TSP, while the number of top secondary rachis branches per panicle (USRB) was also used as an effect variable to check its effect to TSP given different OsTPR and/or MOC backgrounds of samples.

To further identify lines with low glycaemic index (GI) properties the milled grain samples of F₅‐derived F₆ mapping populations were subjected to in vitro GI estimation by following the methods described in Anacleto et al. (2019). In the present study, we also screened 92 breeding lines using in vitro GI method and identified two additional low GI lines (≤55) and further tested using in vivo GI method by recruiting human subjects by following the methods described in Anacleto et al. (2019). For five selected cultivars identified for slow digestible attributes (three low GI lines identified in Anacleto et al., 2019, and two breeding lines with intermediate GI from the present study), replicated yield trials were established in total bigger plot size (>600 m²) in 2019 dry season by employing standard crop management practices across all plots at Zeigler Experimental Station, IRRI, Laguna, Philippines (14°10’N, 121°15’E). For panicle architecture phenotyping, panicles from the main culms of three plants for each cultivar were manually harvested prior to full grain ripening and phenotyped using the Panicle Traits Phenotyping Tool (AL‐Tam et al., 2013). Yield (in ton/hectare) was calculated using standard formula to maximize the yield observed in respective plot area (>600 m²) for each cultivar.

Conflicts of interest

The authors declare that results connected to the manuscript have been submitted for US provisional patent application.

Author contributions

The research was conceived by E. A. P. and N. S., and supervised by N. S.; E. A. P. analysed panicle architecture, generated the GWAS data and was responsible for targeted association, annotation, protein–protein interaction prediction, gene set analysis, cis motif searching and haplotype analyses; R. S. A. performed the GWAS and post hoc analyses; G. M. was responsible for gene model prediction, multiple sequence alignment, nucleotide diversity calculation and phylogenetic analysis; S. P analysed co‐expression data and constructed association networks; S. B. performed breeding of RILs and collected yield and genotype data. The manuscript was written jointly by N. S., E. A. P. and S. B., and reviewed by A. K.

Supporting information

Figure S1. Diversity in the 310 member panel of resequenced indica entries.

Figure S2. Genetic variants associated with USRB identified through single‐locus GWAS.

Figure S3. Amino acid sequence alignment of AtTPR and OsTPR.

Figure S4. Comparison of OsTPR amino acid sequences of USRB contrasting lines.

Figure S5. Variation in panicle architecture, grain yield and grain quality displayed by the panel of resequenced indica type germplasm as explained by OsTPR haplotype.

Figure S6. Path correlation analysis of recombinant inbred line population.

Figure S7. A survey on the expression levels of OsTPR (represented by LOC_Os02g50790) and MYB4P across nine developmental stages and anatomical parts.

Figure S8. Yield, USRB, amylose content and other cooking properties of the RILs from the cross between Samba Mahsuri x IR36ae possessing different haplotypes.

Table S1. List of genes identified to be associated with USRB in the GWAS.

Table S2. Top 10 genes identified through gene‐level analysis for USRB.

Table S3. Sets of genes possibly associated to USRB based on gene set analysis.

Table S4. Results of motif searching for upstream significant SNPs in OsTPR and LOC_Os02g50700 based on TomTom Software and JASPAR Core (2018) Plants Non‐Redundant DNA Database.

Table S5. SNPs associated with different panicle traits of 309 diverse indica accessions based on multi‐locus GWAS.

Table S6. Different panicle traits of 309 diverse indica accessions associated with genic SNPs including 2 kb upstream and 1 kb downstream of cloned panicle genes at 95% confidence level using targeted association analysis.

Table S7. Genes co‐expressed in the blue module where OsTPR was found in the panicle development transcriptome atlas.

Table S8. First neighbour genes of OsTPR identified through gene regulatory networks.

Table S9. Amylose content, yield, and USRB traits of Samba Mahsuri x IR36ae RILs.

Pasion, E. A. , Badoni, S. , Misra, G. , Anacleto, R. , Parween, S. , Kohli, A. and Sreenivasulu, N. (2021) OsTPR boosts the superior grains through increase in upper secondary rachis branches without incurring a grain quality penalty. Plant Biotechnol. J., 10.1111/pbi.13560