Gene Pyramiding Strategies for Sink Size and Source Capacity for High-Yield Japonica Rice Breeding

Abstract

In Japan, high-yielding indica rice cultivars such as ‘Habataki’, ‘Takanari’, and ‘Hokuriku 193’ have been bred, and many genes related to the high-yield traits have been isolated from these and other indica cultivars. Many such genes are expected to be effective in increasing the yield of japonica rice, including those that increase sink size. It has been expected that high-yielding japonica rice could be bred by introducing sink-size genes into the genetic background of japonica cultivars such as ‘Koshihikari’, which show strong cold tolerance, have good taste characteristics, and fetch a high price. However, the corresponding near-isogenic lines did not necessarily produce high yields when tested in the field. In this review, we summarize information on the major high-yield-related rice genes and discuss pyramiding strategies to further increase the yield of japonica rice. In parallel with increasing sink size, source capacity needs to be increased by increasing photosynthetic rate per unit leaf area (single leaf photosynthesis), improving canopy structure, and increasing translocation capacity during the ripening stage. To implement these strategies, innovative breeding methodologies that efficiently produce the combinations of desired alleles are required.

Introduction

Rice is broadly classified into two subspecies indica and japonica, the latter is further divided into tropical japonica and temperate japonica. Temperate japonica is grown mainly in Northeast Asia. In China, japonica rice is grown mainly in the Northeast (e.g. Heilongjiang Province) and Jiangsu Province and is traded at high prices as “premium-quality rice” owing to its good taste characteristics. In recent years, the cropping area of japonica rice has been expanding in China (Li et al. 2019). In Japan, mainly japonica rice is grown, because it can satisfy exacting consumer demands for rice taste characteristics and is tolerant to repeated cold stress. Recent climate change, increase in fertilizer prices, and labor shortage on rice farms have led to a need for high-yielding japonica cultivars that can reduce the cost of production per unit yield.

Active efforts in Japan since the 1990s to develop cultivars from indica accessions to significantly increase yield (Takita 2009) have resulted in the “super-high-yielding” cultivars ‘Habataki’ (Kobayashi et al. 1990), ‘Takanari’ (Imbe et al. 2004), and ‘Hokuriku 193’ (Goto et al. 2009). These are grown as livestock feed and have very high brown rice yields under optimal conditions. For example, ‘Hokuriku 193’ holds the record for the highest brown rice yield in Japan (> 12 t/ha, 3-year average; highest yield, 13.16 t/ha) (Okamura et al. 2022). Its yield after hulling is > 15 t/ha (3-year average), and the highest recorded yield is ca. 17 t/ha (Okamura et al. 2022). Many factors affecting yield are considered unstable and difficult to analyze genetically, but the super-high-yielding indica cultivars are stably and significantly higher yielding than conventional Japanese cultivars unless they are exposed to low temperatures, indicating that their high yield characteristics are genetic traits.

A physiological prerequisite for increasing rice yield is an increased sink size, and in recent years, many genes have been isolated that increase the number of grains and the grain size (grain weight) of rice. However, it has been pointed out that increasing grain weight decreases the number of grains, and increasing the number of grains decreases grain filling, a trade-off relationship that does not necessarily increase yield (Ueda et al. 2021). Despite several well-known indica-derived alleles that increase sink size, few have actually been used in Japanese temperate japonica rice breeding except sd1^DGWG (Hori et al. 2021). To achieve a significant increase in yield, it will be necessary not only to increase the sink size, but also to increase the source capacity (photosynthetic capacity) and to promote efficient translocation of photosynthetic products to the sink organ (in the case of rice, the grain). In this review, we propose a strategy for the pyramiding of alleles that increase source capacity and translocation efficiency, together with genes that increase sink size, that would lead to significant yield improvement. We believe that this strategy is the most reasonable way to increase rice yield, but to date, no significant increases in rice yield have yet been reported using this strategy. This review has attempted to summarize the effects and characteristics of the genes and alleles related to sink size, source ability, and translocation that may be useful for this strategy and contribute to future high-yield breeding of temperate japonica rice.

Sink Size–Related Genes

In recent years, many rice genes that are believed to be involved in high yield have been isolated and are widely known among rice researchers and breeders. Many of them affect sink size, such as the number of grains per panicle and the grain size (grain shape and grain weight), and their details have been summarized in recent reviews (Li et al. 2018, 2022; Lu et al. 2022). Here, we overview the isolation, regulation of expression, and/or interaction of the major sink size–related genes shared between japonica and indica. We also describe the phenotypes of NILs that carry sink size-related genes from indica rice in the genetic background of japonica rice. These genes are listed in Table 1.

Table 1 Sink–size–related genes described in this review

• (1) GN1A/OsCKX2

Among the genes involved in rice sink size, Grain Number 1A (GN1A, Os01g0197700) and its regulators are the most well studied (Fig. 1A). This gene was isolated by Ashikari et al. (2005) using a cross between indica ‘Habataki’ (high number of grains per panicle) and japonica ‘Koshihikari’ (standard number of grains per panicle). GN1A encodes a cytokinin oxidase/​dehydrogenase OsCKX2, which degrades active cytokinins, decreases cytokinin accumulation in inflorescence meristems and increases the number of reproductive organs (Ashikari et al. 2005). GN1a expression is regulated by DROUGHT AND SALT TOLERANCE (DST, Os03g0786400), which encodes a zinc-finger transcription factor (Li et al. 2013). Mitogen-activated protein kinase OsMPK6 directly phosphorylates DST and increases its transcriptional activity (Guo et al. 2020). The rice grain size and number 1 (gsn1, Os05g0115800) mutant has a large-grain, sparse-grain panicle phenotype in comparison with indica ‘Fengaizhan-1’ (‘FAZ1’); GSN1 encodes the mitogen-activated protein kinase phosphatase OsMKP1, which interacts with and dephosphorylates OsMPK6 (Guo et al. 2018). Suppression of the individual genes encoding mitogen-activated protein kinases OsMPK6, OsMKK4, and OsMKKK10 results in denser panicles and smaller grains, which rescues the gsn1 phenotype; thus, GSN1 acts as negative regulator of the OsMKKK10–OsMKK4–OsMPK6 cascade by inducing specific dephosphorylation of OsMPK6 to coordinate the trade-off between grain number per panicle and grain size (Guo et al. 2018). The erecta1 (oser1, Os06g0203800) mutant has more grains per panicle than ‘FAZ1’; the receptor-like protein kinase OsER1 functions upstream of the OsMKKK10–OsMKK4–OsMPK6 cascade (Guo et al. 2020). OsMED25 physically interacts with DST on the OsCKX2 promoter, and then recruits RNA polymerase II to activate OsCKX2 transcription (Lin et al. 2022).

Fig. 1

Modified from a figure in Guo et al. (2020) and OsER1 added (Lin et al. 2022). B GW2. Modified from a figure in Li et al. (2022)

Interactions of genes that increase sink size. A GN1A.

GN1a of ‘Habataki’ has a 16-bp deletion in the 5′-UTR, a 6-bp deletion in the first exon, and three SNPs in the first and fourth exons that alter the amino acid sequence compared with ‘Koshihikari’. The Chinese high-yielding cultivar ‘5150’ produces more grains than ‘Habataki’; it has an 11-bp deletion in the third exon that creates a premature stop codon. This cultivar has been suggested to have a null allele (loss-of-function allele) of GN1a (Ashikari et al. 2005).

Introduction of the genomic region containing the GN1a allele from ‘ST-1’ into ‘Koshihikari’ increases the number of grains per panicle by about 30% (Table 1) (Agata et al. 2023). The TASUKE + browser for the NARO Genebank Core Collection (Kumagai et al. 2019; Tanaka et al. 2020; https://ricegenome-corecollection.dna.affrc.go.jpg) indicates that ‘Habataki’-type sequences are widespread in indica, and that differences in GN1a alleles are some of the main factors explaining the difference in grain number per panicle between indica and japonica.

• (2) GW2

Grain Width 2 or Grain Width and Weight 2 (GW2, Os02g0244100) is one of the well-studied genes for grain size (Fig. 1B). The gene was isolated in a cross between japonica ‘WY3’ (very large grain) and elite indica ‘FAZ1’ (small grain). GW2 encodes a RING-type E3 ubiquitin ligase, and the WY3 allele causes larger grains (Song et al. 2007). WIDE GRAIN 1 (WG1, Os02g0512400) encodes the CC-type glutaredoxin OsGRX8, which positively regulates cell proliferation and grain width and weight (Hao et al. 2021). WG1 interacts with the transcription factor OsbZIP47 (Os06g0265400), recruits the transcriptional co-suppressor ASP1 (Os08g0162100), and suppresses the expression of genes involved in cell proliferation, thus inhibiting grain growth (Hao et al. 2021). GW2 ubiquitinates WG1, which is then degraded by the 26S proteasome, and OsbZIP47 triggers gene expression. Grain size is thus controlled by the GW2-WG1-OsbZIP47 regulatory module (Hao et al. 2021). GW2 also ubiquitinates the cell wall protein expansin-like 1 EXPLA1, resulting in its degradation, a looser cell wall, and increased cell growth (Choi et al. 2018).

In comparison with GW2 from ‘FAZ1’, that of ‘WY3’ has a single nucleotide deletion in the fourth exon, creating a premature stop codon (Song et al. 2007). Introduction of the genomic region containing the ‘WY3’-type allele into ‘FAZ1’ increased the yield by 19.7% (Song et al. 2007). According to TASUKE + (Kumagai et al. 2019; Tanaka et al. 2020), the ‘WY3’-type allele was not found among the NARO WRC and JRC core collections, indicating that it is rare.

A newly isolated GW2 allele gamma-radiation-treated indica Ma85, gw2.1, has a mutation (E128K) in the fifth exon and increases grain weight (Huang et al. 2022). The gw2 mutant ‘KEMS39’, identified in an ethyl methanesulfonate-treated ‘Koshihikari’ population has a mutation in the splicing site of the sixth intron, resulting in a deletion of 67 bases in the GW2 mRNA. This mutant has larger grain, a thicker stem, and higher lodging resistance (Yamaguchi et al. 2020). On the other hand, NILs carrying GW2 of ‘BG1’ introduced into ‘Koshihikari’ increased 1000-grain weight but decreased the number of panicles and number of grains per panicle and did not directly lead to high yield (Ueda et al. 2021).

• (3) γ subunits of heterotrimeric G proteins

Heterotrimeric G protein signaling is important and well conserved in plants and animals (Ofoe 2021). Heterotrimeric G proteins consist of three subunits: Gα (encoded in rice by RGA1), Gβ (RGB1), and Gγ (encoded by RGG1, RGG2, GS3, qPE9-1/DEP1, and GGC2. Some heterotrimeric G protein subunits control grain size (Liu et al. 2018; Sun et al. 2018; Cui et al. 2020).

• (3a) GS3

GRAIN SIZE 3 (GS3, Os03g0407400) was isolated in a cross between two indica cultivars ‘Minghui 63’ (large grain) and ‘Chuan 7’ (small grain) as a QTL with significant effects on grain length and grain weight (Fan et al. 2006; Takano-Kai et al. 2009; Mao et al. 2010). The functionally defective gs3 allele makes grains larger and heavier. An SNP in the second exon, resulting in a stop codon, is present in gs3 from ‘Minghui 63’ and ‘H94’ (large grain) but not ‘Chuan 7’ or ‘Zhenshan 97’ (small and medium grain) (Fan et al. 2006). GS3 genetically interacts with GL3.3 (Os03g0841800), which encodes a GSK3/SHAGGY-like kinase; their double mutants have very long grains (Xia et al. 2018). The gain-of-function mutant clg1-1 (Os05g0551000) encodes an E3 ligase containing a CHY zinc-finger domain and a C3H2C3-type RING domain and increases grain size (Kobayashi et al. 2013). Overexpression of CLG1 from ‘Zhonghua 10’ increases grain size, while overexpression of mutated CLG1 with substitutions of three conserved amino acids decreases grain length (Yang et al. 2021). GS3 interacts with CLG1 and is ubiquitinated and degraded in the endosome degradation pathway (Yang et al. 2021).

Nan et al. (2018) achieved a 4.5% increase in yield by introducing the genomic region containing the gs3 of indica GKBR into japonica ‘Kongyu 131’ (Table 1). TASUKE + (Kumagai et al. 2019; Tanaka et al. 2020) shows that gs3-type alleles are present in some indica cultivars in the NARO Genebank Core Collection (WRC).

• (3b) qPE9-1/DEP1

qPE9-1 (QTL for panicle erectness on chromosome 9–1, Os09g0441900)/DEP1 (Dense and erect panicle 1) affects grain size, panicle density and erectness, and nitrogen responsiveness (Huang et al. 2009; Zhou et al. 2009; Sun et al. 2014). Unlike the DEP1 of ‘Nanjing 11’ and ‘Nipponbare’ (non-erect trait), the DEP1 of ‘Shennong 265’ (erect trait) has a 637-bp sequence in the middle of the fifth exon replaced by a 12-bp sequence, resulting in a premature stop codon and the loss of 230 C-terminal amino acid residues (Huang et al. 2009).

GS3 and DEP1 interact with the transcription factor OsMADS1, increase its activity and regulate grain size and shape (Liu et al. 2018). The genomic region containing the DEP1 allele in ‘Shao 314’ increases yield by 40.9% (Table 1; Huang et al. 2009). In China, a japonica rice cultivar with the DEP1 allele from the Italian cultivar ‘Balilla’ is widely grown as a high-yielding cultivar (Xu et al. 2016a). Resequencing of widely cultivated Chinese varieties (69 japonica and 83 indica) revealed a truncation that is present in ‘Balilla’ DEP1 and in all 36 japonica varieties with an erect or semierect panicle, including the super-high-yielding’Liaojing 5’ and’Qianchonglang’, but not in any of the other varieties (Huang et al. 2009).

• (4) TGW6

Auxins regulate grain size. A QTL with a major effect on 1000-grain weight, Thousand-Grain Weight 6 (TGW6, Os06g0623700), encodes IAA-glucose hydrolase, which influences the timing of the transition from the syncytial to the cellular phase by controlling IAA supply (Ishimaru et al. 2013). Loss of TGW6 function increases grain weight (Ishimaru et al. 2013). Compared to TGW6 of the ‘Nipponbare’ allele (small 1000-grain weight type), the ‘Kasalath’ allele (large 1000-grain weight type) has a single-nucleotide deletion that creates a premature stop codon. Ishimaru et al. (2013) found that the introduction of a genomic region containing this TGW6 allele into ‘Koshihikari’ increases yield by about 20%. According to TASUKE + (Kumagai et al. 2019; Tanaka et al. 2020), the ‘Kasalath’-type allele is present only in some indica germplasms in the NARO Genebank Core Collection (WRC), such as ‘Da Hong Gu’, indicating that it is relatively rare.

• (5) Other genes

WFP (Os08g0509600), GW6 (Os06g0266800), GW6a (Os06g0650300) and qGL3 (Os03g0646900) are also known to have large effects on sink size.

• (5a) IPA1/WFP/SPL14

Ideal Plant Architecture 1 (IPA1) / Wealthy Farmer’s Panicle (WFP) is a QTL that controls panicle branching and encodes the SQUAMOSA promoter-binding protein-like (SPL) protein OsSPL14, which is a target of OsmiRNA156 (Miura et al. 2010; Jiao et al. 2010). IPA1 was detected in a cross between ‘Taichung Native 1’ and ‘Shaoniejing’; a point mutation in the OsSPL14 gene of ‘Shaoniejing’ prevents OsSPL14 regulation by OsmiRNA156, resulting in reduced tiller number and increased lodging resistance and yield; a plant carrying this allele was called “ideal” (Jiao et al. 2010). The nucleotide sequences of WFP are identical in ‘Nipponbare’ and ‘ST-12’, but methylation in the upstream region differs and the expression level is higher in ‘ST-12’; it is unclear how the difference in methylation is related to WFP expression (Miura et al. 2010). OsSPL14^WFP is associated with an increase of approximately 40% in primary branch and grain numbers (Miura et al. 2010).

• (5b) GW6

GW6 was isolated in a cross between japonica ‘Nan-Yang-Zhan’ (‘NYZ’; large grain) and indica ‘Hua-Jing-Xian 74’ (‘HJX74’; small grain), which encodes a GAST-family protein regulated by gibberellin (Shi et al. 2020). It is expressed at high levels in young panicles, promotes the expansion of cells within the spikelet hull, and increases grain width. Grain size and weight are reduced in GW6 knockout mutants and increased in overexpressors. The GW6 coding sequence is identical in ‘NYZ’ and ‘HJX74’, but 4 SNPs and a 3-bp (CCT) insertion are found in the GW6 promoter region of NYZ. One of the SNPs is located within the cis-element CAAT-box of the promoter and affects the expression level. Shi et al. (2020) found that the yield per plant was increased by 14.5% by introducing the genomic region containing this GW6 allele into ‘HJX74’ (Table 1).

• (5c) GW6a/OsglHAT1

GW6a was isolated by Song et al. (2015) in a cross between ‘Nipponbare’ and indica ‘Kasalath’; although the ‘Nipponbare’ grains are heavier than those of ‘Kasalath’, the ‘Kasalath’ allele further increases the weight of ‘Nipponbare’ grains. GW6a encodes the GNAT-like protein OsglHAT1, which has histone acetyltransferase activity. In fact, overexpression of either of the parental OsglHAT1 alleles increases grain weight in both parental backgrounds; and increases the seed weight when overexpressed in Arabidopsis, whereas OsglHAT1 antisense expression decreases grain weight. This indicates that GW6a regulates seed and grain weight and size in both monocots and dicots and that its expression level, not a difference in sequence, causes the difference in phenotype. OsglHAT1 affects the cell number of the spikelet hull and is highly expressed in young panicles. Interaction of GW6a with Homolog of Da1 on Rice Chromosome 3 (HDR3), a ubiquitin-interacting motif–containing ubiquitin receptor, increases GW6a stability by delaying its degradation by 26S proteasomes (Gao et al. 2021). Song et al. (2015) reported that introduction of the genomic region containing this OsglHAT1 allele into ‘Nipponbare’ increases the grain yield per plant by 15.8% (Table 1).

• (5d) qGL3/OsPPKL1

Brassinosteroids (BRs) control the grain size. BRs are perceived by the membrane-localized receptor kinase BRASSINOSTEROID-INSENSITIVE1 (OsBRI1, Os01g0718300) (Nakamura et al. 2006) and its partner BRI1-ASSOCIATED RECEPTOR KINASE (OsBAK1, Os02g0283800) (Li et al. 2009), and their loss of function results in BR-insensitive phenotype and small grains (Morinaka et al. 2006; Yuan et al. 2017). GLYCOGEN SYNTHASE KINASEs (OsGSK1, Os01g0205700, OsGSK2, Os05g0207500 and OsGSK3; Os02g0236200) (Gao et al. 2019; Koh et al. 2007; Tong et al. 2012) negatively regulate the levels of BRASSINAZOLE RESISTANT1 protein (OsBZR1, Os07g0580500) and repress BR signaling (Bai et al. 2007).

A major QTL for grain length (qGL3; Os03g0646900) was isolated in a cross between ‘N411’ (extra-large-grain japonica) and ‘93–11’ (high-quality elite indica) (Zhang et al. 2012); qGL3 encodes a Ser/Thr protein phosphatase with a Kelch-like repeat domain (OsPPKL1) (Hu et al. 2012; Qi et al. 2012; Zhang et al. 2012). A rare allele qGL3^N411 leads to a long-grain phenotype; its product has an aspartate-to-glutamate substitution in the conserved AVLTD motif of the second Kelch repeat (Zhang et al. 2012). The rice genome contains other two OsPPLK1 homologs, OsPPLK2 (Os05g0144400) and OsPPLK3 (Os12g0617900) (Zhang et al. 2012). Grain length is regulated negatively by OsPPKL1 and OsPPKL3 and positively by OsPPKL2 (Zhang et al. 2012). Both qGL3^N411 and qGL3⁹³⁻¹¹ physically interact with OsGSK3; qGL3⁹³⁻¹¹ dephosphorylates it and blocks its degradation, but qGL3^N411 lacks this activity (Gao et al. 2019). qGL3 also induces the phosphorylation of a 14-3-3 protein, OsGF14b (Os04g0462500), which modulates its nucleocytoplasmic shuttling and inactivates OsBZR1-dependent transcription, eventually to negatively regulate BR signaling and grain length in rice (Gao et al. 2022). Therefore, grain length is controlled by the qGL3 → OsGSK3 → OsBZR1 ← OsGF14b ← qGL3 module (Gao et al. 2022).

Introduction of the genomic region containing qGL^N411 into ‘93–11’ increases yield by 16.2% (Zhang et al. 2012).

• (6) The use of sink size–related genes in breeding

If each allele that contributes to high yield could increase yield independent of genetic background, it should be very easy to develop breakthrough cultivars that will lead to enormous economic benefits. Moreover, simply pyramiding them should easily provide for the world's food supply for years to come. For example, the combination of GN1a (increase in grain number per panicle: 30%) (Agata et al. 2023) and DEP1 (increase in yield per plant: 40%) (Huang et al. 2009) would lead to a 70% yield increase if their effects were simply additive. However, it is easy for crop breeders and researchers to imagine that this would not be the case. This is because increasing sink size alone will not ensure source capacity. As noted above, the only high-yielding alleles of indica or tropical japonica origin practically used in temperate japonica rice breeding are DEP1 in Chinese japonica cultivars and ‘DGWG’-type sd1, and their yield increasing effects also vary widely depending on genetic background and environment and are much more limited than originally reported (Tomita 2009; Xu et al. 2016a; Ueda et al. 2021). Although among Japanese cultivars, ‘DGWG’-type sd1 is used in recent some cultivars such as ‘Nizinokirameki’, ‘Akidawara’, and ‘Mizuhonokagayaki’ (Hori et al. 2021), it is not present in the five main commercial cultivars grown in Japan (‘Koshihikari’, ‘Hitomebore’, ‘Hinohikari’, ‘Akitakomachi’, and ‘Nanatsuboshi’; Japan Organization for Stable Rice Supply and Security, https://www.komenet.jp/pdf/R03sakutuke.pdf). Practical use of these genes (alleles) in japonica cultivars has been limited and cultivars with GN1a or GW2 introduced, for example, are not expected to spread rapidly. Breeding of cultivars with a single or a few introduced genes has been restricted to improving disease and insect resistance and adjusting heading date (Saka et al. 2010; Matsumoto et al. 2023). The reason for this may be that yield potential is a complex trait that is sensitive to the environment and involves the combinatorial effects of many genes, and therefore DNA marker selection is difficult.

Introduction of Alleles Increasing Sink Size Alone Does not Necessarily Result in High Yields

We have evaluated the yield potential of NILs carrying well-known sink-size–related genes in the background of japonica cultivars under conditions similar to those in productivity tests in breeding and found that, unlike in pot trials, such genes do not necessarily result in high yields in field trials. Large grains also result in a lower ratio of perfect grains harvested (Table 2; Ueda et al. 2021). The perfect grains ratio refers to the proportion of grains remaining after removing damaged, dead, immature and foreign grains, and foreign substances. Trade-off relationships have been observed between panicle weight and number of panicles, and among other yield components (Kuroda et al. 1999; Crowell et al. 2015; Tao et al. 2018; Yano et al. 2019; Ueda et al. 2021). These findings indicate that genetic enhancement of a single yield component does not necessarily increase yield, because such enhancement limits other yield components when source capacity is limited.

Table 2 Performance of NILs for genes that increase sink size in field cultivation (Ueda et al. 2021)

In the case of NILs carrying genes or QTLs related to sink size, Ueda et al. (2021) pointed out that despite sink size and yield increases in pot trials, the competition among individuals for space, solar radiation, and CO₂ may result in failure to ensure source capacity appropriate for the sink size in field cultivation. In fact, high yields of NILs with increased sink size have been reported under high CO₂ concentrations (Nakano et al. 2017). In general, productivity and local adaptability tests are used in the selection of breeding cultivars resulting in a proper balance between sink and source; therefore, increasing sink size by introducing sink-size–related genes into the genetic background of these cultivars not only fails to increase yield, but also decreases the ratio of perfect grains owing to insufficient source capacity. The sink size and source capacity of current leading cultivars are balanced (Fig. 2A). A sink size increase will result in insufficient source capacity and grain filling (Fig. 2B), but if their source capacity can be increased at the same time, an increase in yield can be expected (Fig. 2C). Despite many efforts to isolate genes related to sink size, the use of these genes alone will not lead to high yield of japonica rice unless efforts are also made to increase source capacity (Ueda et al. 2021).

Fig. 2

Conceptual diagram of a high-yield strategy taking into account the sink–source balance. Sugars produced by photosynthesis in the leaves (source) are transferred to the grains (sink). This can be likened to the transfer of water from one bucket to another bucket. A Sink and source are well balanced in conventional cultivars. Lines with well-balanced sink size and source capacity have been selected for breeding (e.g., in productivity tests). B Distorted sink–source balance in cultivars with increased sink size only. Even if sink size increases, the source capacity remains small, resulting in poor ripening. C Sink–source balance in cultivars with both sink size and source capacity increased. Because the sink–source balance is restored, such cultivars will mature

Increasing Source Capacity

Increasing source capacity should proceed in parallel with increasing sink size and include (1) increasing photosynthetic rate per unit leaf area (single leaf photosynthesis), (2) improvement of canopy structure (optimizing the arrangement of leaves as source organs to increase canopy photosynthesis in the field), (3) increasing translocation capacity during the ripening period, and (4) regulating senescence, therefore increasing harvest index, or combinations of the above options. Each option is discussed below.

• (1) Increasing photosynthetic rate per unit leaf area (single leaf photosynthesis).

Photosynthetic rate varies greatly among rice cultivars (Cook and Evance 1983; Sasaki and Ishii 1992; Kanemura et al. 2007; Jahn et al. 2011). QTL analysis has been conducted using crosses between indica ‘Takanari’ or ‘Habataki’ (high photosynthetic rate and high yield) and japonica ‘Koshihikari’ (standard photosynthetic rate and yield). Several QTLs for increased photosynthesis have been detected: Green for Photosynthesis (GPS, Os04g0615000; Takai et al. 2013), a QTL for canopy temperature difference on chromosome 11 (qCTd11; Fukuda et al. 2018), a QTL for high photosynthesis on chromosome 10 (qHP10; Adachi et al. 2019, Yamashita et al. 2022), a QTL for high photosynthesis on chromosome 5 (qCAR5; Yamashita et al. 2022), and Carbon Assimilation Rate 8 (CAR8, Os08g0174500; Adachi et al. 2017); the causative genes have been identified for GPS and CAR8 (Table 3).

Table 3 Source capacity–related genes described in this review

• (1a) GPS

GPS is identical to Narrow Leaf 1 (NAL1; Qi et al. 2008), a gene involved in polar auxin transport (Takai et al. 2013). This pleiotropic gene encodes a serine protease; its natural variation (an A-to-G substitution in the coding region) results in a change of His233 to Arg233 (H233R) in the catalytic center (Wang et al. 2020; Li et al. 2023a); NAL1 in ‘Takanari’ has this mutation (Takai et al. 2013). This allele increases the number of mesophyll cells between vascular bundles, leaf thickness, and photosynthesis rate per unit area, but leads to a narrow-leaf phenotype. Yield trials detected no increase in yield in NILs-NAL1, which carry NAL1 from ‘Takanari’ in the ‘Koshihikari’ background (Takai et al. 2013). The distribution of each of the multiple alleles of Nal1 within rice species and cultivars has been reported (Taguchi-Shiobara et al. 2015; Wang et al. 2020). This gene affects not only source features but also sink features (grain number per panicle and effective panicle number per plant) (Wang et al. 2020).

• (1b) CAR8

CAR8 encodes a putative Heme Activator Protein 3 (OsHAP3) subunit of the CCAAT-box-binding transcription factor OsHAP3H, and is identical to DTH8/Ghd8/LHD1, which controls rice flowering (heading) date (Wei et al. 2010; Yan et al. 2011; Dai et al. 2012). In ‘Habataki’, this gene has a 1-bp deletion at 322 bp from the initiation codon, which causes a frameshift and premature termination of translation (Adachi et al. 2017). NILs carrying the ‘Habataki’-type OsHAP3H region in the ‘Koshihikari’ genetic background have increased nitrogen content, stomatal conductance of the flag leaf, and CO₂ assimilation rate; but decreased grain yield in the paddy field (Adachi et al. 2017). The ‘Habataki’-type OsHAP3H may be responsible for the earlier heading date in NILs than in ‘Koshihikari’. As described above, there are some cases which increasing leaf photosynthetic capacity alone does not directly lead to increased yields. According to TASUKE + (Kumagai et al. 2019; Tanaka et al. 2020), the ‘Habataki’-type allele is relatively rare and is present only in some indica cultivars, such as ‘Davao 1’ in the NARO Genebank Core Collection (WRC).

• (1c) qCTd11, qHP10, and qCAR5

qCTd11 from ‘Takanari’ increases photosynthetic rate and stomatal conductance and decreases leaf temperature (Fukuda et al. 2018). qHP10 from ‘Takanari’ and qCAR5 from ‘Habataki’ increase photosynthetic rate, root hydraulic conductance, and stomatal conductance (Adachi et al. 2019; Yamashita et al. 2022). In both cases, the causative genes await to be identified.

Other approaches to increasing photosynthetic capacity through genetic engineering include modifying rubisco, optimizing the Calvin–Benson cycle, introducing the C₄ photosynthetic pathway into C₃ plants (Evans 2013; Sharwood 2016; Kubis and Bar-Even 2019; Furbank et al. 2023; Rosado-Souza et al. 2023), and introducing or inactivating transcription factors (Ambavaram et al. 2014; Chen et al. 2021; Wei et al. 2022).

• (2) Improving canopy structure (optimizing the arrangement of leaves as source organs to increase canopy photosynthesis in the field)

The loss-of-function or reduced function of the sd1 (‘DGWG’-type sd1) allele causes rice semi-dwarf morphology and improves canopy structure (Ueda et al. 2021). Many modern indica cultivars appear to have improved canopy structure owing to the introduced ‘DGWG’-type sd1 allele. On the other hand, in ‘Tsukuba SD1’, in which ‘DGWG’-type sd1 is introduced into the ‘Koshihikari’ background, the high-yielding effect of sd1 is limited (Tomita 2009; Ueda et al. 2021). The introduction of sd1 into japonica cultivars with a short growing season and small biomass increases their resistance to lodging, but they do not gain sufficient biomass, making it difficult to obtain high yield (Ueda et al. 2021).

There is a long-standing debate about the ideal canopy structure. A common feature of the canopy structure of high-yielding cultivars is that the flag leaf is positioned above the panicle (Setter et al. 1995). Unlike ‘Koshihikari’, the super-high-yielding cultivar ‘Takanari’ has panicles positioned below the upper leaves, resulting in less shading by the panicles (Fig. 3). Takita (2009) pointed out and discussed the “three-layer” canopy structure (the upper leaf layer, middle panicle layer, and lower support layer) as a high-yield factor in the high-yielding feed rice cultivar ‘Bekoaoba’.

Fig. 3

Canopy of typical high-yielding indica ‘Takanari’ and japonica ‘Koshihikari’. Unlike ‘Koshihikari’, the super-high-yielding cultivar ‘Takanari’ has panicles positioned below the upper leaves, resulting in a three-layer structure of the canopy: the upper leaf layer, the panicle layer, and the support layer, with little shading by the panicles

The study of the genes controlling canopy structure has not been sufficiently thorough, because the genetic control of this trait is complex, and it is difficult to detect a single gene with a large effect. Both conventional GWAS and QTL analyses have limited ability to reveal genes related to traits controlled by multiple genes, such as canopy structure. New approaches to genetic analysis are needed to detect and evaluate the combinatorial effects of multiple genomic regions.

• (3) Increasing translocation capacity during ripening

Although not always isolated as alleles that increase japonica yield, genes isolated as related to translocation are presented here. Sucrose synthesized in the source organ (green leaves) accumulates in the rice stem mainly as starch during the vegetative growth period and is mobilized via phloem loading to the sink organ (growing grain) during the reproductive growth period, and accounts for half of the final grain yield (Wang et al. 2017).

There are two phloem loading systems with two sucrose transport pathways, symplastic and apoplastic loading (Braun et al 2014). Symplastic loading depends on plasmodesmata (Wu et al. 2021), while apoplastic loading depends on membrane-localized transporters that play a critical role (Hu et al. 2021). Sugar transport proteins, such as SUTs (sucrose transporters) and SWEETs (sugars will eventually be exported transporters) are carriers for sugar transport in the apoplastic pathway (Hu et al.2021). These genes are listed in Supplemental Table 1.

• (3a) SUT family

SUT proteins contain 12 transmembrane domains that form a pore to transport sucrose across the plasma membrane (Riesmeire et al. 1992). In the rice genome, there are five SUT homologs, four SUT genes, OsSUT1 (Os03g0170900), OsSUT3 (Os10g0404500), OsSUT4 (Os02g0827200) and OsSUT5 (Os02g0576600) are plasma membrane-localized, while OsSUT2 (Os12g0641400) is a tonoplast membrane protein (Aoki et al. 2003; Eom et al. 2011). OsSUT1 is expressed in the endosperm, leaf sheaths, nodes, vascular parenchyma and nucellar projections (Matsukura et al 2000). OsSUT1 mutants (tos17 insertion mutants) cannot produce homozygous progeny, presumably due to defects in pollen germination or tube growth (Hirose et al 2010). Eom et al. (2016) found that only OsSUT1 cDNA driven by the Arabidopsis SUC2 (AtSUC2) promoter complemented the phenotypes of atsuc2 mutants. This suggested that OsSUT1 is a functional ortholog of the AtSUC2 and functions as an apoplasmic phloem loader (Eom et al 2016). OsSUT2 is expressed in lateral roots, inflorescences, stems, seed coats, leaf mesophyll and bundle sheath cells but not in the veins (Aoki et al. 2003; Eom et al. 2011, 2012; Saito et al. 2011). ossut2 T-DNA insertion mutant plants accumulated more sucrose, glucose and fructose in the leaves than controls and the rate of sugar export from their leaves was significantly reduced (Eom et al. 2011). These results suggest that OsSUT2 is involved in sucrose transport across the tonoplast from the vacuole to the cytosol in rice, which plays an essential role in sugar export from the source leaves to sink organs (Eom et al. 2011). OsSUT3 is expressed in pollen, suggesting that OsSUT3 functions in pollen development and maturation rather than phloem loading in source leaves (Aoki et al. 2003; Scofield et al. 2007; Li et al 2020). CRISPR/Cas9-induced OsSUT4 mutants (ossut4) result in a slower net photosynthetic rate and insufficient sugar supply leading to delayed grain filling (Mengzhu et al. 2020). OsSUT5 is most highly expressed in storage leaves (Aoki et al. 2003). Sun et al. (2010) showed that OsSUT1 has a lower affinity for sucrose compared with OsSUT5 in Xenopus laevis oocytes.

• (3b) SWEET family

SWEETs contain seven transmembrane domains and are present in all organisms (Yuan and Wang 2013; Anjali et al. 2020). There are 21 SWEET family members in the rice genome (Yuan et al. 2014). SWEETs play crucial roles in various physiological processes, including seed-filling, pollen nutrition and phloem loading and unloading (Chen et al. 2015; Kim et al. 2021; Eom et al. 2015). Null mutants produced by TALEN-mediated mutagenesis of OsSWEET4 (Os02g0301100) resulted in null mutants with completely empty caryopses, indicating that OsSWEET4 is essential for seed filling (Sosso et al. 2015). CRISPR/CAS9-mediated knockout of OsSWEET11 (Os08g0535200) significantly reduced the sucrose concentration in the mutant embryo sacs and resulted in defective grain filling compared with that of wild type (Ma et al. 2017). In contrast to rice SUT genes, the knockout mutants of other OsSWEET genes did not cause any abnormal phenotype (Hu et al. 2021).

Regulator for Sugar Transporters

• (3c) OsDOF11

Oryza sativa DNA BINDING WITH ONE FINGER 11 (OsDOF11: Os02g0707200) is expressed in the vacuolar cells of photosynthetic organs and in various sink tissues (Wu et al. 2018). A T-DNA insertional mutant of OsDOF11 (osdof11) is semi-dwarf, with fewer tillers and smaller panicles compared with the wild type (Wu et al. 2018). Expression of four SUT genes, including OsSUT1, OsSUT3, OsSUT4 and OsSUT5 and two SWEET genes, OsSWEET11 and OsSWEET14 (Os11g0508600), is altered in various organs of the mutant, including the leaves (Wu et al. 2018). Chromatin immunoprecipitation assays showed that OsDOF11 directly binds to the promoter region of OsSUT1, OsSWEET11 and OsSWEET14, indicating that the expression of these transporters, which are responsible for sucrose transport via apoplastic loading is coordinately regulated by OsDOF11 (Wu et al. 2018). Overexpression of OsDOF11 reduced yield; however, tissue-specific activation of OsDOF11 by fusion with VP16 increased yield (Kim et al. 2021).

• (3d) OsRRM

The rice RNA recognition motif (OsRRM) containing protein gene (Os09g0298700) is expressed in various organs (Liu et al. 2020). The T-DNA insertion osrrm mutant shows dwarfism, late-flowering and smaller seed size phenotypes. The mRNA levels of almost all the sugar transporter genes are severely reduced in the osrrm mutant and this alters sugar metabolism and sugar signaling (Liu et al. 2020). OsRRM binds directly to mRNAs encoding sugar transporter genes such as OsSUT2 and thus may stabilize their transcripts (Liu et al. 2020).

• (3e) Nhd1

The rice transcription factor gene, N-mediated heading date 1(Nhd1: Os08g0157600) simultaneously mediates flowering time and N use efficiency by regulating the florigen gene and several N assimilation and amino acid transporter genes (Zhang et al. 2021a). The CRISPR/CAS9-induced knockout mutant, nhd1, showed a reduced shoot height and biomass accumulation phenotype during the vegetative period; moreover, the Nhd1 mutation also caused an alteration in sucrose distribution between the leaf blade and sheaths due to of the inhibition of OsSUT1 expression (Li et al. 2023b). Nhd1 directly activates OsSUT1 expression by binding to its promoter region (Li et al. 2023b).

• (3f) OsbZIP72

OsbZIP72 (Os09g0456200) is an ABA-responsive transcription factor that confers drought tolerance in rice (Lu et al. 2009). Transactivation analysis showed that OsZIP72 binds directly to the promoter of OsSWEET13 and OsSWEET15 and activates their expression (Mathan et al. 2021). This showed that the higher expression of OsSWEET13 and OsSWEET15 genes potentially modulates sucrose transport and distribution in response to abiotic stresses (Mathan et al. 2021). Through this mechanism, rice could maintain sugar homeostasis under abiotic stresses, such as drought and salinity (Mathan et al. 2021).

• (3g) OsNF-YB1 and OsNF-YC12

Rice Nuclear factor YB1 (OsNF-YB1: Os02g0725900) expression was detected in the aleurone layer (Xu et al. 2016b). A CRISPR/CAS9-induced knockout mutant, osnf-yb1, showed a chalky endosperm and a defective grain filling phenotype (Bai et al. 2016). Electrophoretic mobility shift assays and yeast one-hybrid experiments showed that OsNF-YB1 could directly bind to the promoters of OsSUT1, OsSUT3 and OsSUT4 (Bai et al. 2016).

OsNF-YC12 (Os10g0191900) interacted with OsNF-YB1 in vitro and in vivo and was expressed in both the aleurone layer and the starchy endosperm (Xiong et al. 2019). The CRISPR/CAS9-induced knockout mutant, osnf-y12, showed that functional loss of OsNF-YC12 caused a decrease in grain weight and increased endosperm chalkiness (Xiong et al. 2019). On the other hand, overexpression of OsNF-YC12 increased grain size and grain weight (Xiong et al. 2019). OsSUT1 expression was significantly reduced in osnf-yc12 mutants (Xiong et al. 2019). ChIP-qPCR and yeast one-hybrid assay showed that OsNF-YC12 binds directly to the promoter of OsSUT1 (Xiong et al. 2019). Therefore, OsSUT1 is a common target of both OsNF-YC12 and OsNF-YB1 (Xiong et al. 2019). Furthermore, both osnf-yb1 and osnf-yc12 mutants exhibited the same defective endosperm phenotype (Xiong et al. 2019). Thus, the OsNF-YB1-OsNF-YC12 dimer is likely to regulate the expression of OsSUTs in the aleurone layer for sugar loading in rice endosperm (Xiong et al. 2019).

• (3h) GFD1

By screening a mutant library derived from the chemical mutagenesis of ‘G48B’ (an indica cultivar), the grain filling duration 1 (gfd1) mutant was obtained, which showed a prolonged grain filling period, increased grain size and a reduced number of spikelets (Sun et al. 2023). By a map-based cloning strategy, GFD1 was shown to encode a MATE transporter (Os03g0229500) (Sun et al. 2023). MATE transporters are one of the most prominent families of cation transporter proteins (Omote et al. 2006; Takanashi et al. 2014; Upadhyay et al. 2019; Qin et al. 2021; Zhang et al. 2021b; Zhou et al. 2023). Yeast two-hybrid, firefly luciferase complementation imaging assay and biomolecular fluorescence complementation analysis indicated that GFD1 could interact with OsSWEET4 and OsSUT2 (Sun et al. 2023). In addition, genetic analysis showed that GFD1 might control the grain filling time through OsSWEET4 and regulate grain size through OsSUT2, while controlling the number of grains per spike through OsSUT2 and OsSWEET4 (Sun et al. 2023; Li et al. 2024).

• (3i) APO1/SCM2

ABERRANT PANICLE ORGANIZATION 1 / STRONG CULM (APO1/SCM2; Os06g0665400) encodes an F-box protein that increases the number of spikelets by suppressing the premature transition from the inflorescence meristem to the spikelet meristem (Ikeda et al. 2007). The apo1 mutant with a premature stop codon in the middle of the gene has fewer grains per panicle than the wild type (Ikeda et al. 2005, 2007). The ‘Habataki’-type APO1 allele increases culm strength and lodging resistance (Ookawa et al. 2010). APO1 is expressed predominantly in developing vascular bundles and promotes translocation of photosynthetic products (Terao et al. 2010). The ‘Habataki’ and ‘Koshihikari’ APO1 alleles differ by several SNPs and indels in the promoter and coding regions. Confirmation of the genome sequence with TASUKE + (Kumagai et al. 2019; Tanaka et al. 2020) reveals that the ‘Habataki’-like promoter sequence is present in the genomes of many indica germplasms in the NARO Genebank Core Collection (WRC).

• (3j) GN1A/OsCKX2

GN1A/OsCKX2 increases the number of grains per panicle and was discussed above in detail as a sink size–related gene. Its null allele from the rice line ‘R498’ promotes root development, increases culm diameter and enhances lodging resistance (Tu et al. 2022); such traits may increase translocation capacity. Introduction of GN1A into the background of japonica cultivars does not necessarily result in high yield (Ueda et al. 2021); as discussed above, high yield cannot be achieved without considering the sink–source balance.

• (4) Senescence

Green leaves carry out photosynthesis, but as time passes, chlorophyll content decreases, photosynthetic capacity declines, and the plant eventually dies. In the process, macromolecules such as lipids, proteins, and nucleic acids that make up the leaves are hydrolyzed. The nutrients are translocated from the source organ to the sink organ. In rice, 70–90% of the nitrogen in the panicles is translocated from the vegetative organs (leaves) (Mae 1997). Leaf senescence is a complex and regulated process, controlled by a number of genes. The details are described in a recent paper (Lee and Masclaux-Daubresse 2021). Genes responsible for senescence that may be involved in increasing rice yield are presented here. These genes are listed in Supplemental Table 1.

• (4a) OsSGR

The Stay-green phenotype shows characteristics of slow leaf and stem senescence. Indica rice generally displays early-senescence phenotypes, whereas japonica exhibits late senescence (Shin et al. 2020). Therefore, concerning this trait, the use of japonica alleles to improve indica rice is primarily expected, rather than the use of indica alleles to improve japonica rice.

A QTL for stay-green (OsSGR; Os09g0532000) on chromosome 9 was responsible for the differential senescence phenotype. OsSGR was isolated in a cross between japonica ‘Junam’ (JN; late-senescence) and indica ‘IR72’ (early-senescence), which encodes a chlorophyll-degrading Mg⁺⁺-dechelatase (Shin et al. 2020). To verify that the indica allele of OsSGR is required for its early-senescence phenotype, OsSGR knockout mutants were generated in the indica cultivar ‘Kasalath’ via CRISPR/CAS9 genome editing and knockout plants showed delayed loss of chlorophyll in whole plants, leaves and panicles (Shin et al. 2020). Overexpressors of either the indica or japonica OsSGR allele were found to have a similar early leaf phenotype, suggesting that SNPs in the coding region of either allele are not responsible for indica-type early leaf senescence (Shin et al. 2020). NILs with the indica background harboring the japonica OsSGR allele displayed extended photosynthetic competence with higher chlorophyll content, resulting in higher grain yields (Shin et al. 2020).

Transcription Factors

The NAC (NAM, ATAF1/2 and CUC2) gene family is a group of plant-specific transcription factors, and there are more than 150 NAC genes in the rice genome, many of which are involved in senescence (Singh et al. 2021).

• (4b) OsNAP

The T-DNA insertion mutant, prematurely senile 1 (ps1-D), resembles the wild type at the early growth stage, but when plants reach the tillering stage, leaf senescence is significantly accelerated, and the expression of many senescence-related genes is increased compared to the wild type (Liang et al. 2014). OsNAP (Os03g0327800), isolated as the causal gene for ps1-D, was highly expressed in senescent tissues, and regulated the expression of senescence-related genes in response to ABA (Liang et al. 2014; Zhou et al. 2013; Chen et al. 2014). Overexpressors showed an early senescence phenotype, while RNAi lines showed slower senescence and increased yield compared to controls (Liang et al. 2014).

• (4c) OsNAC2

OsNAC2 (Os04g0460600) was isolated by analysis of the activation-tagging mutant Ostil1 (Mao et al 2007). OsNAC2 is involved in ABA-induced leaf senescence. OsNAC2 overexpressors accelerate leaf senescence, while RNAi lines delay leaf senescence and increase yield (Mao et al. 2017).

• (4d) ONAC096

T-DNA insertion mutants of ONAC096 downregulate ONAC096 (Os07g0138200) expression and maintain green leaf color during natural senescence in the field, thereby extending their photosynthetic capacity. More panicles are present in the mutant compared to the wild type plant, thereby increasing rice yield by 16% (Kang et al 2019).

• (4f) ONAC106

ONAC106 (Os08g0433500) is highly expressed in the yellowing leaf. Leaves of onac106-1D mutants (insertion of the 35S enhancer in the promoter region of the ONAC106 gene) retain their green color under natural senescence and dark-induced senescence conditions and ONAC106 directly modulates the expression of several senescence-related genes (Sakuraba et al. 2015). In addition to delayed senescence, onac106-1D also exhibited a salt stress-tolerant phenotype (Sakuraba et al. 2015). The genes that down-regulate the salt response are rapidly up-regulated in onac106-1D under salt stress. ONAC106 binds to the promoter regions of several senescence-associated genes (Sakuraba et al. 2015). Grain yield per plant was slightly higher in onac106-1D plants than in the wild type (Sakuraba et al. 2015).

Gene Pyramiding Strategies for Sink Size and Source Capacity for High-Yield Japonica Rice Breeding

To achieve high yield in rice, an appropriate balance between sink and source capacity is needed. If it becomes possible to quantify the effect of each allele, design allele combinations and produce a set of NILs with each allele introduced in the background of modern cultivars, and evaluate their effects, sustainable crop improvement based on scientific knowledge can be achieved. Since many allele combinations occur in conventional breeding programs and data on their effects have been accumulated, this information can be used for the evaluation of alleles (Wei et al. 2021). Recently, attempts have been made to combine genomic polymorphism information with trait information for genomic and gene-allele-based traits to predict flowering time and yield potential (Bartholomé et al. 2022). If the effect of each allele could be quantified, an ideal combination of alleles might be possible, but since source capacity is highly dependent on the environment, designing an ideal genotype that takes sink–source balance into account is not simple. Various attempts have also been made to construct yield models based on genotype-by-environment interactions (GxE) (de Leon et al. 2016; Seck et al. 2023), and further progress is expected in this area. However, due to trade-off relationships among yield-related traits, competitive and compensatory relationships can also be assumed for combinations of alleles located at multiple loci for a single trait. It is not easy to predict the ideal combination of indica and tropical japonica derived alleles needed to achieve high yield in temperate japonica cultivars. If the model were a simple one in which there is a bottleneck in the process leading to high yield and the trait involved in the bottleneck should be improved, the logic would be clear. Indica-derived alleles with large effects on sink size and translocation have been identified, and many inter-subspecific crosses have been tried in breeding programs. The fact that these alleles have not been used in practical temperate japonica breeding programs despite these efforts, can be considered instructive: the bottleneck has been the improvement of source ability, for which no alleles with large effects have been found. Additionally, in practice, this is complemented by the need for lodging resistance, which also requires attention to interactions with the environment, especially the close relationship with the wind, making the construction of the ideal type more difficult.

As mentioned above, the only alleles from indica and tropical japonica that have actually been successfully used in temperate japonica rice breeding are sd1^DGWG and DEP1^Balilla. The fact that both of these genes have effects that alter the plant architecture may indicate that improving the source ability by modifying the canopy structure and increasing lodging resistance has been the bottleneck for yield improvement. On the other hand, simply introducing these alleles as they are into a low-biomass temperate japonica background will not result in high yield due to low biomass (Ueda et al. 2021, et al.; Kinoshita et al. 2024). In temperate japonica breeding programs, along with the use of these genes, it can be said that the high-yield effect is obtained by combining genetic backgrounds with multiple genes that increase biomass, i.e. pyramiding multiple genes. In addition to the above, if we can find an allele that can dramatically extend the ripening period, which has not been tried in practical breeding, we may be able to find a new direction for high-yield breeding.

Thus, even though high yield is a genetic trait and many genes for it have been isolated, no model integrating all of them has been found due to its complexity. As described in this review article, several genes related to source ability have been isolated and it will be necessary to start from a simple model of sink-source balance by evaluating a combination of these genes and gene alleles related to sink size. For this purpose, it is necessary to generate NILs introducing alleles related to sink and source balance and to accumulate data verifying the effects of their combinations.

Even if an ideal genotype can be designed that takes sink–source balance and other factors into consideration, it is not easy to achieve it in a conventional breeding program. It is still necessary to use many breeding materials and to repeat crosses and selections many times to pyramid the required alleles. Currently, most conventional rice breeding programs use a bulk-population method that begins with a two-line cross; such crosses are usually made about once every 10 years. With this approach, it would take decades or even centuries to pyramid the desired alleles. The time per rice generation could be shortened through environmental control (Ohnishi et al. 2011; Tanaka et al. 2016), and such methods are beginning to be used to produce NILs. In recent years, speed-breeding technology has been reported in many crops and has been attracting interest (Watson et al. 2018).

Even today, when many high-yield genes have been studied intensively, not all traits and trait combinations required in a cultivar can be explained by known genes and their alleles; therefore, phenotypic selection is necessary for such traits. For effective phenotypic selection, it is necessary to efficiently pyramid the required alleles while preserving genetic diversity in breeding populations. Genomic selection is a method of continuously selecting marked alleles that match breeding goals through allelic reproduction; this method has been used with great success mainly in allogamous crops and livestock. However, it is difficult to efficiently outcross autogamous crops such as rice, wheat, and soybean, so technological breakthroughs are needed to simplify their outcrossing in breeding. For example, Tanaka (2010) has proposed to produce and use male sterility, and Abe et al. (2018) and Akasaka et al. (2018) have attempted to produce transgenic dominant male-sterile rice.

A successful strategy for future high-yield japonica rice breeding will depend on the combination of these techniques: if the isolation of useful genes that enhance source capacity, gene pyramiding considering sink–source balance and other factors (Fig. 2), development of comprehensive prediction models, and efficient generation acceleration technology are integrated, an unprecedented improvement in yield should become achievable.

Conclusions

Although many important genes related to sink size have been identified, achieving high yield using these in isolation is difficult. Not only sink size but also source capacity has to be increased to maintain sink–source balance. To achieve this, it is necessary to isolate and use genes that improve single-leaf photosynthesis and canopy structure and increase translocation during the ripening period, as well as to develop an innovative system to efficiently generate the required allele combinations necessary for these strategies.