Shaping polyploid wheat for success: Origins, domestication, and the genetic improvement of agronomic traits

VRT-A2 facilitated the origin of T. polonicum and T. petropavlovskyi

Polish wheat (Triticum polonicum, AABB, 2n = 4x = 28) is a unique tetraploid wheat species characterized by elongated outer glumes, that is, the basal, sterile, bract-like organs subtending the wheat spikelets (Watanabe, 1999). This long-glume trait is controlled by a single gene locus, P1 (from Polish wheat). Recently, several independent groups, including our own (Liu et al., 2021), have reported the map-based cloning of the P1 causal gene and revealed that the ectopic expression of a MADS-box transcription factor gene, VRT-A2 (VEGETATIVE TO REPRODUCTIVE TRANSITION 2 in genome A), resulted in the long-glume phenotype in T. polonicum (Adamski et al., 2021; Liu et al., 2021). Further evidence suggests that a sequence rearrangement within the first intron of T. polonicum-derived VRT-A2, consisting of a 560 base-pair (bp) deletion coupled with a 157 bp sequence substitution, facilitates the activation of VRT-A2 in floral organs, resulting in the repression of a large set of MADS-box genes and elongated leafy glumes (Li et al., 2020a; Adamski et al., 2021; Liu et al., 2021). A recent study suggested that the deletion/insertion variation in VRT-A2 intron 1 might also cause intron retention and, therefore, a frame-shift mutation (Chai et al., 2021). More interestingly, a sequence rearrangement identical to that of VRT-A2 has been detected in T. petropavlovskyi (also known as “Daosuimai” or rice-head wheat), a hexaploid wheat subspecies that also has elongated glumes (Liu et al., 2021; Xiao et al., 2021). Together, these studies support two major conclusions: (i) the activation of VRT-A2 facilitated the origin of T. polonicum as a unique species; and (ii) the long-glume trait of T. polonicum was likely inherited by T. petropavlovskyi through a genomic hybridization between T. polonicum and an unknown hexaploid landrace, enabling the formation of T. petropavlovskyi as a special subspecies of T. aestivum (Dixon and Boden, 2021; Xiao et al., 2021; Xu et al., 2021).

Tasg-D1 underlies the origin of T. sphaerococcum

The hexaploid T. sphaerococcum (AABBDD, 2n = 6x = 42), also known as Indian dwarf wheat endemic to India and Pakistan, has been classed as a unique wheat species belonging to the Dinkel section (Table 1). T. sphaerococcum shows typical brassinosteroid (BR)-deficient phenotypes, such as hemispherical grains, a semidwarf stature, straight flag leaves, and compact spikes, due to the pleiotropic effect of a single semi-dominant gene locus, S (Sphaerococcum) (Sears, 1947). The S causal gene, designated as Tasg-D1 (Triticum aestivum semispherical grain 1 in genome D), has been isolated by map-based cloning and, consistent with the phenotypes it confers, encodes serine/threonine protein kinase glycogen synthase kinase 3 (GSK3), a key repressor of BR signaling that shows sequence homology to the Arabidopsis BRASSINOSTEROID-INSENSITIVE 2 (BIN2) protein (Cheng et al., 2020; Gupta et al., 2021). Single amino acid substitutions (Arg₂₈₄-to-Gly or Glu₂₈₆-to-Lys) in the highly conserved TREE (Thr₂₈₃-Arg-Glu-Glu₂₈₆) domain enhanced the stability of Tasg-D1 protein, leading to compromised BR signal transduction and consequently the species-forming phenotypes of T. sphaerococcum. Intriguingly, previous studies showed that amino acid substitutions in the TREE domain of GSK proteins in Arabidopsis and rice also caused BR-deficient dwarfism and round grains (Li and Nam, 2002; Liu et al., 2014). These findings suggest a highly conserved biological function of GSK kinases in diverse plant species. Haplotype analyses in different wheat accessions further supported the conclusion that T. sphaerococcum might have originated from T. aestivum after hexaploidization, due to the natural point mutation in Tasg-D1. The discovery of Tasg-D1 as a species-forming gene leads to the notion that a single-gene mutation may cause considerable phenotypic changes and give rise to novel taxonomic species in Triticum L. (Cheng et al., 2020).

Genomic reshaping and the origin of Tibetan semi-wild wheat

The Tibetan semi-wild wheat (T. aestivum ssp. tibetanum Shao, AABBDD, 2n = 6x = 42) phenotypically resembles Tibetan local bread wheat landraces but shows unique brittle rachis characteristics and thus has been identified as an independent subspecies of T. aestivum (Shao et al., 1980). De novo assembly of the draft genome of a semi-wild wheat accession, Zang1817, uncovered extensive reshaping of the genome of Tibetan semi-wild wheat compared with that of normal bread wheat Chinese Spring (CS). This genome reshaping included the addition of many Zang1817-specific genomic segments and the loss of more than 380 Mb of CS-specific genomic segments (Guo et al., 2020). Principal component analysis of these data suggested that Tibetan semi-wild wheats share a closer genetic relationship with Tibetan bread wheat landraces than with other wheat accessions. More intriguingly, a genome-wide association study (GWAS) identified two types of genetic variation associated with the brittle-rachis trait in Tibetan semi-wild wheat accessions: the deletion of a 0.8 Mb fragment on chromosome 3D containing the homologs of the domestication genes Brittle rachis 1 (TaBTR1) and TaBTR2 and the insertion of a 161 bp transposon (TE) in the domestication gene Q (square-head spike) on chromosome 5A (Guo et al., 2020). Indeed, the TE insertion in Q-5A, designated the Q^t allele, is known to contribute to the brittle rachis trait (Jiang et al., 2019). Based on these findings, it is proposed that natural variations in genome structure and specific genetic segments, particularly the TaBTR1-D and Q gene loci, caused the de-domestication of Tibetan semi-wild wheat from Tibetan bread wheat landraces (T. aestivum) to form a novel subspecies (Jiang et al., 2019; Guo et al., 2020).

The C (compactum) locus and the origin of T. compactum

The C gene on chromosome 2D controls the compact spike trait in a species of hexaploid wheat known as Triticum compactum (club wheat). Club wheat cultivars are still grown commercially, but their worldwide distribution is limited to certain agro-climatic regions. For example, club wheat was originally favored due to its drought and shattering resistance, stiff straw, early seeding, and competitive yield in the drylands of the Pacific northwest (USA) (Gul and Allan, 1972). Cytogenetic studies located the locus C on chromosome 2D, and subsequent genetic studies localized C to the centromere (Johnson et al., 2008). Because proximal segments of chromosomes are characterized by highly repetitive DNA fragments with reduced levels of recombination, map-based cloning of the C causal gene will be challenging.

Non-brittle rachis

The wild species of wheat, for instance, the diploid T. urartu and tetraploid wild emmer (Triticum dicoccoides), naturally allow mature grains/spikelets to disperse freely through the disarticulation of the entire spikelet at each spike rachis node. This trait, referred to as “brittle rachis,” helps the wild wheat plants to effectively reproduce themselves in nature (Charmet, 2011). However, brittle rachis is a disadvantage in cultivated wheat species, as it causes grain loss and, thus, low harvest efficiency. Loss of the natural grain-dispersal character was likely the most important step in wheat domestication. About 10, 000 years BP, wild emmer underwent a natural mutation event that led to the formation of cultivated emmer wheat with non-brittle rachis. At almost the same time, a similar mutation event occurred in a relative of T. urartu and conditioned the origin of the non-brittle einkorn wheat. Cultivation of the non-brittle forms of einkorn and emmer wheats enabled easier harvest of the grains.

Three loci controlling rachis brittleness have been referred to as Brittle rachis (Br or BTR). BTR-D, also known as Br₁, has been mapped to the short arm of chromosome 3D and is responsible for the brittle rachis character in Tibetan semi-wild hexaploid wheat (Watanabe et al., 2003; Jiang et al., 2014). The other two loci, BTR-A (Br₂) and BTR-B (Br₃), which control the brittle rachis trait in wild emmer wheat, are located on chromosomes 3AS and 3BS, respectively (Nalam et al., 2006; Watanabe et al., 2006). The first breakthrough in BTR gene cloning was made in barley (Pourkheirandish et al., 2015). The brittle rachis trait of wild barley (Hordeum vulgare subsp. spontaneum [C. Koch] Thell.) is conditioned by two genetically and functionally linked genes, BTR1 and BTR2, on chromosome 3H. Deletions in either BTR1 (a 1 bp deletion in its coding region) or BTR2 (an 11 bp deletion in the coding region) converted the brittle rachis trait of wild barley into the non-brittle trait seen in domesticated barley (H. vulgare L. ssp. vulgare) (Pourkheirandish et al., 2015). BTR1 encodes a membrane-bound-like protein, BTR1, that harbors two lipophilic regions, while BTR2 encodes a small soluble protein with no sequence similarity to BTR1 (Pourkheirandish et al., 2015). The identical non-brittle rachis traits triggered by the deletion of either BTR1 or BTR2 suggest that these two genes may be functionally interdependent, yet their genetic interaction remains to be determined.

Intriguingly, the ortholog of BTR1, but not BTR2, likely underwent parallel selection in crop wheat. In cultivated einkorn accessions, btr1-A harbors a single nucleotide polymorphism (SNP) at position 355 (G₃₅₅-to-A) relative to a wild-type gene in wild einkorn. This SNP led to a non-synonymous amino acid substitution (alanine to threonine at position 119 of BTR1 protein, Ala₁₁₉-to-Thr) and finally a non-brittle rachis. A different btr1-A mutation, a premature stop codon mutation (Gly_97*) caused by a 2 bp deletion (CG at position 291–292 from the start codon) in the BTR1-A coding region, was found in domesticated tetraploid T. durum, T. dicoccum, and hexaploid bread wheat (T. aestivum L.) (Pourkheirandish et al., 2018; Zhao et al., 2019). The different mutations in BTR1-A gene in diploid einkorn and polyploid wheats indicate the divergent origins of their A genomes (Figure 1). In addition to the mutation in BTR1-A, the B-genome-derived BTR1-B in wheat was also disrupted by a 4 kb insertion at position 539 from the start codon in cultivated emmer, suggesting that simultaneous recessive mutations in BTR1-A and BTR1-B are likely minimally required for the non-brittle rachis morphology in polyploid wheat (Avni et al., 2017; Zhao et al., 2019). To date, the contribution of BTR-D to rachis brittleness in hexaploid wheat is completely unknown.

Another domestication gene, Q, located on the long arm of chromosome 5A, is also involved in the control of the brittle rachis trait. The wild wheat 5Aq allele confers fragile rachides, while its semi-dominant mutant form 5AQ gives rise to the non-brittle rachis trait (Zhang et al., 2011). Q/q encodes an APETALA2 (AP2)-like transcription factor that has sequence similarity with Arabidopsis AP2 (Faris et al., 2003; Simons et al., 2006; Zhang et al., 2011). Comparing the coding sequences of the dominant 5AQ allele and the primitive 5Aq allele revealed two conserved nucleotide differences, that is, an A-to-G variation at position 985 (G₉₈₅-to-A) and a C-to-T variation at position 1254 (C₁₂₅₄-to-T) of the coding region, which are considered to be the major causes for the enhanced biological effect of Q relative to q. The A₉₈₅-to-G variation changes an amino acid at position 329 of the Q protein (Val₃₂₉-to-Ile), which might be related to the enhanced protein dimerization activity of Q relative to q (Simons et al., 2006). The C₁₂₅₄-to-T variation causes no amino acid change but attenuates the miR172-mediated cleavage of Q transcripts, leading to greatly elevated Q transcript levels (Debernardi et al., 2017; Greenwood et al., 2017). Based on genotypic analyses of Q/q alleles in wheat accessions with different ploidy levels, it has been speculated that the domestication event giving rise to the Q allele occurred in polyploid wheat only once, but whether it initially arose in tetraploid or hexaploid wheat is still controversial (Simons et al., 2006). Overall, the mutant forms of btr1 and Q shaped the non-brittle rachis trait of the domestication syndrome in wheat species (Figure 1). However, whether btr1 and Q control the non-brittle rachis trait independently or interdependently is not known at present.

Free-threshing grain

A second domestication event was the conversion of hulled wheat (with tough glumes) to a de-hulled (free-threshing) form, and this conversion contributed greatly to the widespread cultivation of crop wheat. The tough glume trait is genetically controlled by numerous minor and major gene loci, including Sog (Soft glume) and Tg (Tenacious glume), on the short arm of the group 2 chromosomes, and 5AQ (Sood et al., 2009). The single recessive gene sog determines the soft glume trait and threshability in diploid einkorn wheat T. sinskajae and has been mapped to the region close to the centromere on 2AS. The semi-dominant Tg, which is thought to confer a non-free-threshing character mainly in tetraploid and hexaploid wheat, has been mapped to the most distal regions on 2BS and 2DS (Taenzler et al., 2002; Jantasuriyarat et al., 2004; Sood et al., 2009). However, the effect of Tg-A1 in the A genomes of T. aestivum or T. urartu has not yet been well defined. Based on the mapping positions, sog from the A genome and Tg from the B and D genomes are probably not orthologs, supporting the speculation that diploid and polyploid (tetraploid and hexaploid) wheat populations experienced independent domestication processes (Sood et al., 2009; Charmet, 2011).

In addition to the recessive mutation in Tg (from Tg to tg), the dominant 5AQ locus also facilitates grain threshability mainly in polyploid wheats. This has been well illustrated by the selection of the naked durum wheat from hulled cultivated tetraploid wheat genotypes (similar to emmer wheat). The dominant mutation from 5Aq to 5AQ conferred this grain threshability conversion (Figure 1) (Oliveira et al., 2012). Intriguingly, the 5Aq allele is associated with not only the hulled grain trait, but also elongated spikes (speltoid), while the 5AQ allele confers soft glumes, de-hulled grains, and subcompact spikes (Debernardi et al., 2017), suggesting a pleiotropic effect of the Q/q gene. Notably, Q has a similar biological effect to q but to a greater degree: five doses of q confer a squareheaded spike, resembling the effect of one copy of Q (Muramatsu, 1963). Notably, the facile grain threshability conferred by Q is likely repressed by the dominant Tg locus, as a synthetic allohexaploid wheat (AABBD^AeD^Ae, genotype 5AQ; tg-2B; Tg-2D^Ae) generated by the combination of the free-threshing tetraploid wheat (AABB, genotype 5AQ; tg-2B) with the non-free-threshing Ae. tauschii (D^AeD^Ae, genotype Tg-2D^Ae) exhibiting the non-free-threshing trait (Kerber and Dyck, 1969). These clues give rise to the notion that in polyploid wheat, the dominant Q locus as well as tg-2B; tg-2D genotypes are simultaneously required to allow the full expression of the free-threshing character (Figure 1) (Jantasuriyarat et al., 2004). The epistatic effect of Tg on Q illustrates a complicated genetic interaction among different domestication genes and might reflect shared downstream signaling pathways.

Timing of heading

Precise timing of flowering/heading is crucial for wheat productivity in a given environment and wheat adaptability to different agro-climatic environments. Relative to their diploid wild progenitors that grow in relatively narrow and limited regions, polyploid wheats, especially the hexaploid bread wheat, exhibit more flexible flowering/heading habits in response to vernalization (long-term exposure to cold temperature) and photoperiod (day length) conditions.

The vernalization response

Vernalization is a process in which a prolonged exposure to low environmental temperatures (between 0°C and 10°C) stimulates a plant to flower (Flood and Halloran, 1984; Zhang et al., 2019b). Depending on whether vernalization is required for flowering/heading, wheat accessions are divided into winter and spring wheat classes. Winter wheats require vernalization to flower at the appropriate time, while spring wheats do not. As a result, spring wheats can be sown in either autumn or spring, but winter wheats can only be sown in autumn. Several genetic factors, including the major vernalization response genes Vernalization 1 (VRN1, also called WAP1), VRN2, and VRN3, have been documented to underlie the requirement for vernalization. In this scenario, VRN2 is a dominant determinant of the winter growth habit, while VRN1 and VRN3 contribute to the spring growth habit (Figure 2).

VRN1, encoding an AP1-like MADS-box transcription factor, is a central regulatory node modulating the vernalization response and initiating the transition of the shoot apical meristem from vegetative to reproductive growth (Law et al., 1976; Quarrie et al., 1995; Dubcovsky et al., 1998; Iwaki et al., 2002; Danyluk et al., 2003; Trevaskis et al., 2003; Yan et al., 2003; Barrett et al., 2008). In non-vernalized winter wheat varieties, VRN1 is expressed at a very low level; after exposure to cold during the winter, VRN1 is dramatically induced in leaves and at the shoot apex, and its expression levels remain high thereafter to promote heading (Loukoianov et al., 2005; Oliver et al., 2009; Deng et al., 2015). In some wheat lines harboring mutations in the proximal promoter or deletions in the first intron of VRN1, the expression of VRN1 is activated even without vernalization, resulting in a conversion from a winter to a spring growth habit (Yan et al., 2003; Fu et al., 2005). In both barley and wheat crops, vernalization triggers decreased histone 3 lysine 27 trimethylation (H3K27me3) levels and increased H3K4me3 levels in the promoter and coding regions of VRN1 (Oliver et al., 2009; Diallo et al., 2012), suggesting that VRN1 transcription in response to vernalization is tightly associated with the chromatin methylation state. A recent study demonstrated that a long non-coding RNA (lncRNA), VAS (TaVRN1 alternative splicing), is expressed from the first exon and part of the first intron of VRN1 in vernalized winter wheat. This lncRNA is recognized by and associates with the transcription factor RF2b to activate the expression of VRN1 (Figure S1), highlighting a feedback regulatory mechanism by which the intron region determines VRN1 activation (Xu et al., 2021). VRN1 transcription is also fine-tuned by several other proteins, such as the RNA-binding protein GRP2 (Glycine-rich RNA-binding protein 2) and the carbohydrate-binding protein VER2 (Vernalization-related 2). Before vernalization, GRP2 directly binds to the first intron of VRN1 to repress its transcription. During vernalization, GRP2 is modified by O-GlcNAcylation, and this modification promotes tight association of GRP2 with VER2 and its translocation from the nucleus to the cytoplasm; this triggers dissociation of GRP2 from the VRN1 intron region and subsequent VRN1 activation (Xiao et al., 2014). Therefore, sequence variations in the first intron of VRN1 that block the binding of GRP2 will lead to higher expression of VRN1, and thus earlier flowering/heading in winter wheat (Xiao et al., 2014; Kippes et al., 2018).

The VRN2 locus includes two tandemly duplicated genes, ZCCT1 (zinc finger-CCT (CONSTANS, CONSTANS-like, and TOC1) domain) and ZCCT2, which encode proteins that have no clear homologs in Arabidopsis but contain a zinc-finger motif mediating DNA binding and a CCT domain observed in the flowering time-controlling protein CO (Yan et al., 2004; Trevaskis et al., 2007). VRN2 is highly expressed in non-vernalized wheat seedlings to inhibit early flowering through the direct or indirect repression of VRN3 and VRN1. Initial map-based cloning of a spring-wheat-derived VRN2 gene identified single point mutations or deletions in the CCT domain of ZCCT1 and ZCCT2, leading to the functional disruption of the ZCCT proteins (Yan et al., 2004; Distelfeld et al., 2009). These mutations contribute to the spring growth habit. Thus, the CCT domains of ZCCT1 and ZCCT2 are likely essential for their full function as flowering repressors. Further evidence revealed that the CCT domains of ZCCT1 and ZCCT2 are also required for their interactions with HAP (Heme activator protein, also known as Nuclear factor-Y or NF-Y) complex. The association of the ZCCTs with HAPs further competes with CO2 for the regulation of VRN3 (Li et al., 2011). After a prolonged exposure to cold temperatures, VRN2 expression is largely repressed, leading to the activation of VRN3 and VRN1 and subsequent initiation of flowering in early spring. In barley, the downregulation of HvVRN2 expression during vernalization is perfectly correlated with the direct association of the HvVRN1 protein with the promoter region of HvVRN2 (Deng et al., 2015). Nevertheless, whether VRN1 also acts as a direct transcriptional repressor of VRN2 in wheat is debated, as it has been reported that the wheat VRN1 is not mandatory for the downregulation of VRN2 during vernalization, but only helps maintain the low transcript levels of VRN2 after vernalization (Chen and Dubcovsky, 2012). These findings suggest that some uncharacterized genes, rather than VRN1, might be employed for the repression of VRN2 in wheat.

VRN3, a homolog of Arabidopsis FLOWERING LOCUS T (FT), is another key floral activation gene in wheat (Yan et al., 2006). FT in Arabidopsis acts as a flowering signal, that is, florigen, moving from leaves to apices through the phloem to promote flowering (Putterill and Varkonyi-Gasic, 2016). These findings shed light on the possible action of VRN3 protein in wheat. Indeed, VRN3 can directly activate the transcription of meristem identity genes, such as VRN1, possibly through physical interaction with the FD-like transcription factor TaFDL2 (Li and Dubcovsky, 2008).

The extensive genetic interactions among VRN genes lead us to propose a model for the vernalization response in wheat (Figure 1). In this model, VRN2 acts as a repressor of the flowering transition through the direct repression of VRN3 and the indirect repression of VRN1; vernalization-triggered upregulation of VRN1 can release the repressive effect of VRN2 on VRN3, which leads to a further upregulation of VRN1 (a feedback regulatory loop) beyond the threshold required for initiation of flowering (Figure 1). Dominant mutations in VRN1 as well as loss-of-function mutations in VRN2 all lead to vernalization insensitivity and conversion from a winter to a spring growth habit.

Adaptation to environmental constraints

As they spread globally to new agricultural areas, polyploid wheats are subjected to diverse stress-related constraints, for instance extremes of heat and cold, high salinity, drought, soil chemical toxicity, and so on. The relatively high ploidy level and complex genome compositions of polyploid wheats give them greater physiological and ecological plasticity than their diploid progenitors, and thus wide adaptability to harsh growth environments. To understand the molecular bases of the improved adaptability of polyploid wheats, researchers have identified many stress-responsive genes through either “omics” techniques or homology cloning. Among these, genes that confer salt and drought tolerance include the histone acetyltransferase gene GCN5/HAG1 (General control non-repressed protein 5), the salt overly sensitive (SOS) pathway genes SOS1, SOS2/TaCIPK (CBL-interacting protein kinase), SOS3/TaCBL4 (Calcineurin B-like 4), and SOS4, the jasmonic acid (JA) synthesis gene TaAOC1 (allene oxide cyclase 1), and the microRNA (miRNA) gene MIR172 (Ramezani et al., 2013; Zhao et al., 2013; Sun et al., 2015; Cheng et al., 2021; Zheng et al., 2021). Genes that confer drought tolerance include the calreticulin gene TaCRT and reactive oxygen species (ROS) scavenging genes including CATs (encoding catalases), SODs (encoding superoxide dismutases), PODs (encoding peroxidases), and APXs (encoding ascorbate peroxidases) (Jia et al., 2008; Dudziak et al., 2019). Finally, the JA synthesis gene TaOPR3 (12-oxophytodienoate-reductase 3), the heat shock transcription factor genes TaHsfA6f and TaHsfC2a, the E3 ubiquitin ligase gene TaSAP5, and the trehalose-6-phosphate synthase gene TaTPS confer thermotolerance (Xie et al., 2015; Xue et al., 2015; Zhang et al., 2017b; Hu et al., 2018; Tian et al., 2020). Based on the annotations of these genes, it is proposed that: (i) metabolism of phytohormones including abscisic acid (ABA), JA, ethylene (ET), and auxin (AUX); (ii) reactive ROS and ionic homeostasis; and (iii) cellular accumulation of organic solutes, are the main processes that together enhance the adaptability of bread wheat to environmental constraints (Budak et al., 2013; Zhang et al., 2019b; Urbanavičiūtė et al., 2021).

Adaptation to high temperature

The diploid progenitors of wheat were initially adapted to the dry, cool-summer Mediterranean climate and thus are relatively sensitive to high temperatures (Shewry, 2009). In particular, exposure of wheat to high temperatures during its reproductive or grain-filling stage can lead to especially severe damage, characterized by reduced grain number, decreased grain weight, and poor grain quality (Parent et al., 2017). Current knowledge of the heat-response signaling networks established in the model plant Arabidopsis suggests that heat shock transcription factors (Hsfs) and stress-related plant hormones (JA, ABA, and ET) are essential for high temperature tolerance (Li et al., 2018a). Indeed, wheat employs multiple Hsfs- and/or phytohormone-related signaling regulons, such as “Hsfa1b–TaOPR3–JA biosynthesis,” “ABA–TaHsfC2a-B–TaHSP70d (Heat shock protein 70d)/TaGalSyn (Galactinol synthase),” “TaHsfA6f–TaHSPs/TaGAAP (Golgi anti-apoptotic protein)/TaRof1 (encodes a co-chaperone),” and “TaHsfA6e–TaMBF1c-7B (Multiprotein bridging factor 1c)–DREB2A (Dehydration-responsive element-binding protein 2A)/HsfB2A/HsfB2B,” to counteract the adverse effects of high temperature (Xue et al., 2015; Hu et al., 2018; Tian et al., 2020, 2021a). Our recent unpublished data also suggest that TaHsfA1 acts as a master regulator of transcriptional reprograming in response to high-temperature stress in wheat: Loss of HsfA1 not only attenuated the activation of large sets of heat-responsive genes, including Hsf and HSP genes, but also abolished tolerance to heat stress (Figure 3).

In addition to Hsfs, several other heat-responsive components, including the phosphoenolpyruvate carboxylase kinase-related kinase TaPERKR2, the cell death suppressor Bax inhibitor-1 (BI-1), ferritin TaFER, trehalose-6-phophate (T6P), and transcriptional coactivators such as TaMBF1c, are also positive regulators of wheat heat tolerance (Qin et al., 2015; Zang et al., 2017, 2018; Lu et al., 2018). More intriguingly, a recent study suggested that stress granules (SGs), cytoplasmic aggregates induced by heat stress, maintain translation efficiency during heat stress by recruiting TaMBF1c through the SG component RNA-binding Ras-GAP SH3 binding protein TaG3BP (Figure 3) (Tian et al., 2021a). However, whether these components and signaling pathways are controlled by or are independent of Hsfs is not well understood. Indeed, a recent study in Arabidopsis revealed that in a hsfa1 null mutant, some sets of heat-responsive genes were still effectively induced by heat stress, suggesting the existence of an HsfA1-independent signaling pathway that responds to temperature fluctuations (Li et al., 2019).

Breeding semi-dwarf stature

The “Green Revolution” of the 1960s led to spectacular increases in cereal crop yields through the introduction of semi-dwarfing traits into cereal crops to increase lodging resistance and achieve maximum yield potential (Hedden, 2003). In wheat, the semi-dwarfing trait was introduced mainly via the Reduced height (Rht) gene loci Rht-B1b (formerly Rht1) and Rht-D1b (Rht2) derived from the Japanese variety “Norin 10” (Peng et al., 1999). Rht-B1b and Rht-D1b, representing semi-dominant mutant alleles of the Rht-B1 and Rht-D1 homoeologs, respectively, can reduce plant height by ~25%, and have been used in more than 70% of commercial wheat cultivars worldwide (Youssefian et al., 1992; Duvick, 1999). To date, more than 25 loci have been designated as Rht loci (McIntosh et al., 2003; Peng et al., 2011; Chen et al., 2015; Mo et al., 2018). Depending on whether the resulting dwarfing phenotype could be restored by the exogenous application of bioactive gibberellin (GA), these Rht genes have been classified into GA-sensitive and GA-insensitive groups. The “Green Revolution” genes Rht-B1b/D1b, as well as their various alleles, including Rht-B1c (Rht3), Rht-B1e (Rht11), Rht-B1p (Rht17), and Rht-D1c (Rht10), are in the GA-insensitive group, while other Rht loci, such as Rht8, Rht9, Rht18, Rht23, and Rht24, are in the GA-sensitive group (Worland et al., 1998; Bazhenov et al., 2015; Chen et al., 2015; Zhao et al., 2018; Tang et al., 2021).

Rht-B1 and Rht-D1 encode DELLA proteins, the key repressors of the GA signaling pathway and plant growth (Peng et al., 1998, 1999; Richards et al., 2001). Although there is limited knowledge available about the Rht-B1/D1 protein functions and their downstream signaling responses in wheat, the functions of DELLA proteins have been extensively analyzed in other plant species (Peng et al., 1998; Velde et al., 2017). The N-terminus of DELLA proteins, which contains DELLA and TVHYNP motifs, is essential for GA signaling; in the presence of bioactive GAs, the N-terminus associates with the GA receptor GIBBERELLIN INSENSITIVE DWARF1 (GID1), triggering degradation of the DELLA proteins (Dill et al., 2001; Velde et al., 2021). In the “Green Revolution” Rht-B1b and Rht-D1b alleles, nucleotide substitutions in the DELLA-motif-coding regions lead to translation re-initiation, producing N-terminally truncated DELLA proteins (DELLA_ΔN) that are more stable (Peng et al., 1999; Bazhenov et al., 2015; Velde et al., 2021). The degradation-resistant DELLA_ΔN proteins inhibit GA signaling without affecting GA catabolism, thereby conferring a GA-insensitive semi-dwarfing phenotype (Figure 4). Rht-B1c, another Rht-B1 mutant allele derived from the wheat variety “Tom Thumb” (McVittie et al., 1978), harbors an in-frame 90 bp insertion in the coding region, leading to the generation of a N-terminally disrupted, more stable Rht-B1 protein. The dwarfism caused by Rht-B1c is far more severe than that of Rht-B1b, making it unsuitable for semi-dwarfing breeding (Pearce et al., 2011; Wu et al., 2011). Rht-D1c, derived from the Chinese variety “Ai-bian 1,” also leads to extreme dwarfism, because it consists of multiple copies of the dominant semi-dwarfing allele Rht-D1b (Börner et al., 1996; Pearce et al., 2011). It is worth noting that although at least five semi-dwarfing alleles in Rht-B1 (Rht-B1b, Rht-B1c, Rht-B1d, Rht-B1e, and Rht-B1f) and three mutant alleles in Rht-D1 (Rht-D1b, Rht-D1c, and Rht-D1d) have been characterized, and some of them have been utilized in wheat breeding, no semi-dwarfing allele of Rht-A1 has been defined, suggesting that Rht-A1 is likely subfunctionalized (Börner et al., 1996; Pearce et al., 2011).

Unexpectedly, the “Green Revolution” alleles Rht-B1b and Rht-D1b also lead to unfavorable agronomic traits, such as shorter coleoptiles and reduced seedling vigor, hampering plant stand establishment especially in dryland regions where deep planting is essential (Richards, 1992; Rebetzke et al., 1999; Ellis et al., 2004). In addition, Rht-B1b and Rht-D1b result in reduced kernel size and loss of grain yield (Zhang et al., 2013; Würschum et al., 2017; Guan et al., 2018). To overcome these pleiotropic effects, the GA-sensitive Rht loci have been used as alternate gene resources for wheat semi-dwarfing breeding. To date, several GA-sensitive semi-dwarfing genes and their effects in wheat breeding have been analyzed (Figure 4).

Rht8 is one of the most important GA-sensitive Rht loci that has been used globally for wheat semi-dwarfing breeding. Rht8 is a recessive locus that shortens wheat plant height by ~10% without affecting coleoptile elongation or grain yield (Rebetzke et al., 1999). Physiological experiments show that the Rht8-triggered semi-dwarfing trait is not affected by exogenous BR, indicating that Rht8 might be involved in BR signaling (Figure 4) (Gasperini et al., 2012). Previous studies mapped Rht8 to a small region on the short arm of chromosome 2D (Gasperini et al., 2012; Chai et al., 2018). Intriguingly, Rht8 and Rht-B1b individually reduced coleoptile length by about 6.75% and 21.64%, respectively, but caused a severe coleoptile phenotype, reducing length by more than 43%, when combined (Grover et al., 2018). These findings suggest a complex genetic interaction between Rht8 and Rht-B1b and, more interestingly, imply that a Rht-B1b-independent signaling pathway conferred by Rht8 regulates plant height.

Rht18, located on chromosome 6A, is another GA-sensitive semi-dwarfing locus, but is dominant. Rht18 was initially identified in the tetraploid durum wheat cultivar “Icaro” (Ford et al., 2018). Recent work identified GA2oxA9, encoding the GA-inactivating enzyme GA 2-oxidase, as the candidate gene for Rht18. GA2oxA9 metabolizes GA₁₂, the GA₁ precursor, into inactive products, resulting in a decreased abundance of bioactive GA₁ and a semi-dwarf phenotype (Figure 4) (Ford et al., 2018). In line with the dominant effect of Rht18 in reducing plant height, GA2oxA9 is more highly expressed in “Icaro” than in rht18 accessions. Genetic experiments revealed that Rht18 behaves identically to Rht-D1b in reducing plant height (by ~25%), but has no effect on coleoptile elongation. However, the combination of Rht18 and Rht-D1b leads to the reduction of plant height by ~35%, less than that of the combination of Rht-B1b and Rht-D1b (Tang et al., 2021). Notably, Rht18 and Rht24 map to overlapping regions and are also linked to the same marker as Rht14. These findings have led some to propose that Rht18, Rht24, and Rht14 are allelic (Tian et al., 2017; Würschum et al., 2017; Ford et al., 2018; Mo et al., 2018). Indeed, a recent study confirmed that the transcriptional upregulation of GA2oxA9 explains the effect of Rht24 (Tian et al., 2021b). More intriguingly, the semi-dwarfing Rht24b allele reduces plant height and increases nitrogen use efficiency, but has no yield penalty, illustrating a favorable semi-dwarfing gene (Tian et al., 2021b).

Another semi-dwarfing gene, Rht12 on chromosome 5A, was identified in bread wheat following γ-radiation mutagenesis. Rht12 shows identical characteristics to Rht18, such as dominant nature and GA-sensitive classification (Sutka and Kovács, 1987; Ellis et al., 2004). Based on either map-based cloning or MutChromSeq approaches in independent studies, GA2oxA13 (also called GA2oxA14 in a different report) has been shown to be the candidate gene for Rht12, functionally similar to Rht18/GA2oxA9 (Sun et al., 2019; Buss et al., 2020). Transcriptional activation of GA2oxA13/A14 in Rht12-harboring accessions reduces the level of bioactive GA₁, causing semi-dwarfism (Buss et al., 2020). Notably, the four GA-sensitive semi-dwarfing alleles found in tetraploid and hexaploid wheats, Rht18, Rht14, Rht24, and Rht12, are all defined as GA inactivation loci and lead to the depletion of bioactive GAs and thus semi-dwarfism (Figure 4). These are completely different from the rice GA-sensitive “Green Revolution” gene sd1, which represents a loss-of-function mutation in the GA biosynthesis gene GA20ox-2 (Sasaki et al., 2002).

Rht23 is a recently reported GA-sensitive Rht locus that was found in an ethyl methane sulfonate (EMS)-induced mutant line NAUH164 in the “Sumai 3” genetic background (Chen et al., 2015). Rht23 is allelic to 5Dq, the D genome homoeolog of the domestication gene 5AQ (Zhao et al., 2018). Because of the pleiotropic effect of q on both plant height and spike morphology, the Rht23-triggered semi-dwarfism is associated with compact spikes. Further sequence analysis revealed a single-nucleotide mutation in 5Dq, that is, a G-to-A substitution at the 3 147 bp position, which potentially attenuates the binding of miR172 to 5Dq transcripts, leading to their excess accumulation (Figure 4). Similarly, different single-nucleotide substitutions in this region of either 5AQ or 5Dq all cause reduced plant height and compact spikes, resembling the effect of Rht23 (Debernardi et al., 2017; Greenwood et al., 2017). Unlike other semi-dwarfing genes relating to GA metabolism or signaling, Rht23 tends to have more complicated genome-wide effects on transcription, including that of genes involved in cell wall biosynthesis, lignin synthesis, and photosynthesis (Zhang et al., 2020).

Breeding strong culms for lodging resistance

Improving lodging resistance by reducing plant height can only go so far, because severe dwarfism leads to inadequate biomass accumulation and, ultimately, lower yield potential (Okuno et al., 2014; Ookawa et al., 2016; Nomura et al., 2021). In addition, neither sd1 in rice nor Rht loci in wheat could completely prevent bending-type stem lodging, especially at high planting densities, because these semi-dwarfing genes reduce culm strength and nitrogen-use efficiency (Li et al., 2018b; Wu et al., 2020). Therefore, the design of novel cereal crop varieties with a strong culm (SCM) phenotype may be a further strategy for enhancing of lodging resistance.

The SCM trait is determined by multiple factors, for example, culm wall thickness, culm diameter, stem solidness, and the lignin/cellulose content in the stem wall (Pinthus, 1967; Berry et al., 2003; Tripathi et al., 2003; Kong et al., 2013; Okuno et al., 2014; Xiao et al., 2015; Xiang et al., 2016). Based on forward genetic strategies, several SCM-related QTLs have been identified on chromosomes 4A, 5A, and 5B for culm stiffness; 3B for solid stem; 1A, 1B, and 2D for culm wall thickness; 2D for tensile breakage force; and 1A, 3A, 2D, and 4D for culm diameter, among others (Keller et al., 1999; Cook et al., 2004; Hai et al., 2005; Verma et al., 2005; Okumoto, 2011; Berry and Berry, 2015; Hyles et al., 2017; Nilsen et al., 2017). However, few of these loci have been cloned by map-based cloning, except for the 3B locus for stem solidness and the 1A locus for culm wall thickness (Hyles et al., 2017; Nilsen et al., 2020).

The 3B stem solidness locus, conferring a solid culm with a thick sclerenchyma layer and the SCM trait, has been designated as SSt1 in durum wheat and Qss.msub-3BL in bread wheat (Cook et al., 2004; Nilsen et al., 2017). A putative Dof zinc finger protein coding gene, TdDof, was cloned as the SSt1 candidate gene (Nilsen et al., 2020). Copy number variation in the TdDof gene region leads to the duplication of this gene, and thus its activation, in solid-stemmed wheat lines. Excess accumulation of TdDof transcripts attenuates programmed cell death (PCD) in pith parenchyma cells, predominantly through the transcriptional repression of NAC (NAM/ATAF1/2/CUC2) and cysteine proteinase (CEP) genes (Nilsen et al., 2020). This finding not only demonstrated the existence of a novel “Dof–NAC/CEP–PCD” signaling pathway for regulation of stem solidness regulation, but also provided a selection strategy for designing solid-stemmed wheat cultivars through transgenic breeding. The tin locus on chromosome 1A, which has pleiotropic effects on tiller number, spike architecture, and grain weight, is also associated with stem thickness (Kebrom et al., 2012). Map-based cloning identified Csl as the candidate gene for the tin locus; this gene encodes a cellulose-synthase-like protein with homology to members of the CslA clade (Hyles et al., 2017). Changing the level of Csl transcripts, and consequently, Csl protein, may alter carbon partitioning in wheat culm tissues, leading to increased lignification and stronger stems, which in turn would improve lodging resistance (Hyles et al., 2017).

The minor effects of most of the other SCM-related QTLs in wheat make them very challenging to characterize. Our current strategy, combining EMS-based mutagenesis with map-based cloning, has enabled the isolation of major-effect SCM-regulating genes. Some of these genes encode key regulators of BR and AUX signaling, and others encode cytoskeleton regulatory factors (unpublished data). More importantly, the SCM candidate genes identified using this strategy, as well as the scm mutant materials obtained, could be readily utilized in wheat breeding and agronomic practices (Figures 5, S2).

In rice, several SCM QTLs have been isolated by map-based cloning, providing gene resources for SCM design in wheat (Minakuchi et al., 2010; Kashiwagi, 2014; Yano et al., 2015; Nomura et al., 2019; Samadi et al., 2019; Chigira et al., 2020; Nomura et al., 2021). Rice SCM2 is allelic to Aberrant panicle organization 1 (APO1), an F-box protein-coding gene controlling meristem fate identity and panicle structure (Ikeda et al., 2007; Ookawa et al., 2010). Notably, gain-of-function alleles of APO1 conferred not only enhanced culm strength and lodging resistance, but also increased spikelet number, suggesting that APO1 would be a favorable target in engineering lodging-resistant and high-yield crop varieties (Ookawa et al., 2010). Another SCM locus, SCM3, is identical to Fine culm 1 (FC1), the rice ortholog of maize Teosinte branched 1 (TB1), which encodes a TCP transcription factor and is involved in the control of tiller number (Takeda et al., 2003; Yano et al., 2015). Activation of OsTB1/FC1 leads to stronger culms with thicker culm walls, increased spikelet number, and decreased tiller number in rice, and similar effects are also observed in wheat (Yano et al., 2015; Dixon et al., 2018). Increased expression of TB1 in wheat reduces plant height by ~10% without decreasing seedling emergence, largely resembling the effect of the semi-dwarfing locus Rht8, thus providing an alternative to Rht alleles in wheat breeding (Dixon et al., 2020). Considering that both APO1 and TB1 enhance culm strength and promote spikelet number, these SCM genes would be valuable genetic factors to balance the trade-off between lodging resistance and yield potential in wheat breeding, resulting in novel SCM wheat varieties with stronger culms, enhanced lodging resistance, and higher grain yield (Yano et al., 2015; Nomura et al., 2019).

Tillering

Tillering, a form of branching that involves the emergence of lateral shoots from axillary buds on the unelongated basal internode (Wang et al., 2018a), is a major factor determining the productivity of cereal crops. An optimal number of tillers per unit area is essential for achieving the highest yield potential for crop wheat. Extensive studies in rice have characterized multiple genetic factors controlling tillering. For example, MONOCULM 1 (MOC1), encoding a GRAS family nuclear protein that promotes the outgrowth of axillary buds (Li et al., 2003), and MOC3, the rice ortholog of WUSCHEL (WUS) (Lu et al., 2015), are the positive regulators of tillering; FC1, together with TAD1 and TE, which separately encode the co-activator and substrate-recognition factor of the anaphase-promoting complex (APC/C) for the degradation of MOC1 (Lin et al., 2012; Xu et al., 2012), are all negative regulators that lead to decreased tiller number. In particular, SPL14 (SQUAMOSA promoter binding protein-like 14), also known as Ideal Plant Architecture 1 (IPA1), is dominant in repressing tillers and shaping ideal plant architecture in rice (Jiao et al., 2010). To date, the wheat orthologs of IPA1 and TB1 have been isolated based on homology cloning, and their contributions to wheat tillering have also been extensively analyzed. The wheat gene SPL17 (TaSPL17) is the ortholog of rice IPA1 and, like that ortholog, is targeted and functionally repressed by miRNA156 (miR156). Interestingly, over-accumulation of miR156 triggers functional attenuation of TaSPL17 and another SPL gene, TaSPL3. As a consequence, there was a dramatic increase in tiller number, converting the wheat plants to a bushy architecture (Liu et al., 2017). These findings suggest an essential role for TaSPL17 and TaSPL3 in maintaining a suitable tiller number. More intriguingly, overexpression of wheat SPL20 and SPL21 significantly promoted grain size without affecting normal tiller formation (Zhang et al., 2017a), suggesting that SPLs are ideal targets for simultaneously manipulating plant architecture and grain size. Tillering is also repressed by TB1, a function genetically validated by characterizing a highly-branched (hb) line derived from a four-parent multiparent advanced generation intercross population. In the hb line, an increased dosage of the TB-D1 gene causes significantly reduced tiller number (Dixon et al., 2018). Notably, in crop wheat, TB1 is also involved in the control of plant height and spike architecture (Dixon et al., 2018, 2020), suggesting a new target for concomitantly engineering spike morphology and plant architecture in wheat.

Spike development

The spike in tribe Triticeae is unbranched, with spikelets attached directly to the inflorescence axis, known as the spike rachis (Koppolu and Schnurbusch, 2019). Normally, a single rachis node bears only one spikelet, and each spikelet includes up to ~12 florets, of which only three to five florets set grains (Sakuma et al., 2019). Therefore, the grain number per spike is determined by two subcomponents: fertile spikelet number per spike (SNS) and fertile florets per spikelet.

Several genetic factors have been characterized as positive or negative regulators of SNS. One is APO1, the role of which in inflorescence determination was initially characterized in rice. The rice apo1 mutant shows abnormal floral organ identity, suggesting that APO1 may repress precocious conversion of inflorescence meristems to spikelet meristems, thus acting as a positive regulator of spikelet formation (Ikeda et al., 2007). Recently, independent genetic mapping studies in wheat have revealed a stable QTL on chromosome 7AL that contributes to increased SNS. Further evidence suggested that several natural variations in both the promoter and coding regions of APO-7A might explain the effect of this QTL (Kuzay et al., 2019; Muqaddasi et al., 2019; Voss-Fels et al., 2019).

The maize barren stalk1 (ba1) mutant fails to produce branches in male and female inflorescences (ears and tassels). Genetic cloning of the causal gene for this phenotype enabled the isolation of the Barren stalk 1 (BA1) gene, which encodes a non-canonical basic helix-loop-helix (bHLH) transcription factor (Gallavotti et al., 2004). The function of BA1 in the initiation of spike lateral meristems and the formation of spikelets is highly conserved in bread wheat. More intriguingly, downregulation of wheat SPL3 and SPL17 considerably attenuates the expression of BA1, further leading to dramatic decrease in SNS, suggesting SPL3 and SPL17 are two essential activators of BA1 expression (Liu et al., 2017).

Some endemic wheat accessions have more than one spikelet at a single rachis node; this unusual spike morphology is defined as the supernumerary spikelet (SS) trait (Zhang and Yuan, 2014; Dobrovolskaya et al., 2015). SS leads to increased SNS, illustrating the potential of this trait for improving wheat grain number and final yield. Paired spikelet, characterized by the emergence of a second spikelet below and adjacent to the typical spikelet, is the most commonly observed SS phenotype in wheat. Several studies have suggested a role for PPD1 and TB1 in the regulation of paired spikelet formation (Boden et al., 2015; Dixon et al., 2018). A high frequency of paired spikelets is observed exclusively in plants with the photoperiod-sensitive wild-type allele PPD-D1b; no paired spikelets are found in lines harboring the gain-of-function photoperiod-insensitive PPD1 allele, suggesting that PPD1 normally acts as an inhibitor of paired spikelet development (Boden et al., 2015). Further evidence suggested that FT is one of the downstream genes of PPD1 that confer this phenotypic variation. TB1, by contrast, is a positive regulator of paired spikelet formation. An increased dosage of TB-D1 triggers abnormal spike architecture with multiple paired spikelets, mainly through interaction of TB-D1 protein with FT and the resulting transcriptional repression of downstream meristem identity genes (Dixon et al., 2018). These findings illustrate a central role of FT in paired spikelet determination, since both PPD1 and TB1 condition paired spikelet formation through FT.

The other SS phenotypes, such as multirow spikes (MRS) and branched spikes (BS), represent more complex wheat spike architectures (Dobrovolskaya et al., 2015). A recent study identified the wheat ortholog of the rice FRIZZY PANICLE (FZP) gene, WFZP, as a causal gene for the triple-spikelet (TRS, a typical form of MRS) trait observed in the Tibetan wheat variety Zang734. A null mutation in WFZP-A coupled with the deletion of WFZP-D shaped this TRS trait (Du et al., 2021). Parallel studies showed that different mutant alleles of WFZP and different genetic backgrounds could lead to largely divergent SS traits, suggesting that the effect of WFZP on the SS trait is complex and depends on genetic background and/or environmental conditions (Dobrovolskaya et al., 2015; Li et al., 2021).

In contrast to the well-characterized genes for the determination of spikelet numbers in wheat, little is known about the genetic determinants of floret fertility (grain setting) in a given spikelet. A recent study opened up a new avenue for research by cloning a key regulator of floret fertility, Grain number increase 1 (GNI1) (Sakuma et al., 2019). A single amino acid substitution (105 Asp-to-Tyr; N105Y) in the A-genome-derived GNI1 protein, GNI-A1, attenuates its activity and leads to increased grain number per spikelet, explaining the effect of a previously characterized QTL for grain number per spikelet on chromosome 2AL (Sakuma et al., 2019). Intriguingly, GNI1 is an ortholog of barley VRS1 (Vulgare row-type spike 1, also known as HvHOX1), which encodes a homeodomain leucine zipper class I (HD-Zip I) transcription factor. It is worth noting that in barley, VRS1 acts as an inhibitor of lateral spikelet fertility and promotes the “two-rowed” state of the barley spike (Komatsuda et al., 2007), while in wheat, GNI1 is likely involved in the regulation of floret fertility, but not spikelet determination. These functional discrepancies between VRS1 and GNI1 might be attributed to differences in spikelet determinacy between barley and wheat (Koppolu and Schnurbusch, 2019). It has also been reported that photoperiod influences floret fertility in wheat, possibly though PPD1, but the underlying mechanism is not known (Prieto et al., 2018).

Grain size

The grain is the final product of wheat and, as such, directly determines the yield of wheat production. Increased grain size during wheat domestication and polyploidization is one of the most important traits that guaranteed the success of wheat as a major crop worldwide (Dubcovsky and Dvorak, 2007). However, current knowledge about the genetic determinants of wheat grain size is limited because grain development is a highly polygenic trait that is influenced not only by numerous minor-effect QTLs, but also by environmental context (Li and Yang, 2017; Brinton and Uauy, 2019). To date, only a few major-effect genes have been functionally characterized as acting in the determination of wheat grain size.

Grain weight 2 (GW2), encoding a RING-type E3 ubiquitin ligase, was initially characterized in rice to be a negative regulator of grain size. Loss of GW2 confers increased cell numbers in the spikelet hull, resulting in larger grain size and increased grain weight (Song et al., 2007). In wheat, recognition of the effect of GW2 on grain size and weight was facilitated by the detection of a QTL for grain weight and yield located in the centromeric region of chromosome 6A, which includes the GW-A2 gene (Su et al., 2011; Simmonds et al., 2014, 2016; Sukumaran et al., 2018; Zhai et al., 2018). The effects of the GW2 homoeologs (from the A, B, and D genomes) as well as their functional interactions were further validated through clustered regularly interspaced palindromic repeats (CRISPR)/CRISPR-associated protein 9 gene editing. Simultaneous knockout of two GW2 homoeologs (GW2-B and GW2-D, for example) produces larger grains than the single mutants, suggesting functional redundancy and additive interactions among GW2 homoeologs (Zhang et al., 2018). However, the loss of GW2 also leads to decreased grain number (Sukumaran et al., 2018; Zhai et al., 2018). This may be due to the well-known trade-off between grain size and number. Thus, it is still challenging to balance the trade-offs not only between grain size and number, but also among other complex yield components in wheat breeding. In addition, a recent GWAS revealed QTLs for grain size in different wheat accessions, among which one, qGW7B.1, was found to include TaSus1-7B, a sucrose synthase gene that has undergone significant selection during wheat breeding (Hou et al., 2014; Pang et al., 2020). These findings indicate the crucial role of endosperm starch synthesis in the determination of wheat grain size.

Notably, some species-forming genes, for example VRT-A2 and Tasg-D1, also affect the development of grains, illustrating their pleiotropic effects. Ectopic expression of VRT-A2 in Polish wheat promotes grain elongation without affecting grain width, resulting in increased grain size and weight (Backhaus et al., 2021; Liu et al., 2021; Xiao et al., 2021). In contrast, gain-of-function of Tasg-D1 in Indian dwarf wheat leads to small, round grains by repressing BR signaling, suggesting that Tasg-D1, in contrast to VRT-A2, restricts grain development (Cheng et al., 2020). Despite its negative effect on grain size, Tasg-D1 is still expected to have potential use in high-yield wheat breeding because it can shape a more compact plant architecture with erect leaves and decreased plant height, which is favorable for dense planting.

Based on the screening of a wheat EMS mutant library, a tgw1 mutant line with considerably smaller grains was recently identified. Molecular cloning further confirmed a premature stop mutation in the KAT-2B (Keto-acyl thiolase 2B) gene, explaining the small-grain phenotype of tgw1 (Chen et al., 2020). KAT-2B affects wheat grain size through the antagonistic manipulation of in vivo JA and ABA levels. More importantly, overexpression of KAT-2B led to increased grain weight and boosted yield, implying a potential application of KAT-2B for future high-yield molecular breeding in wheat (Chen et al., 2020).