Alternative Splicing of TaGS3 Differentially Regulates Grain Weight and Size in Bread Wheat

The heterotrimeric G-protein mediates growth and development by perceiving and transmitting signals in multiple organisms. Alternative splicing (AS), a vital process for regulating gene expression at the post-transcriptional level, plays a significant role in plant adaptation and evolution. Here, we identified five splicing variants of Gγ subunit gene TaGS3 (TaGS3.1 to TaGS3.5), which showed expression divergence during wheat polyploidization, and differential function in grain weight and size determination. TaGS3.1 overexpression significantly reduced grain weight by 5.89% and grain length by 5.04%, while TaGS3.2–3.4 overexpression did not significantly alter grain size compared to wild type. Overexpressing TaGS3.5 significantly increased the grain weight by 5.70% and grain length by 4.30%. Biochemical assays revealed that TaGS3 isoforms (TaGS3.1–3.4) with an intact OSR domain interact with WGB1 to form active Gβγ heterodimers that further interact with WGA1 to form inactive Gαβγ heterotrimers. Truncated isoforms TaGS3.2–3.4 , which lack the C-terminal Cys-rich region but have enhanced binding affinity to WGB1, antagonistically compete with TaGS3.1 to bind WGB1, while TaGS3.5 with an incomplete OSR domain does not interact with WGB1. Taking these observations together, we proposed that TaGS3 differentially regulates grain size via AS, providing a strategy by which the grain size is fine-tuned and regulated at the post-transcriptional level.


Introduction
Bread wheat (Triticum aestivum L.) is a global staple crop. High yield, as the prime breeding target in cereals, is determined by the panicles per plant, grain number per spike, and grain weight. Grain weight is the major determinant of yield potential, and it is largely dependent on grain size. Thus, grain size has long been the critical target of selection in wheat breeding [1].
Genetic and molecular analyses have identified numerous genes involved in multiple signaling pathways that regulate grain size [2]. One of the most important pathways is oxidase activity [25], and starch synthase activity [26]. AS is also involved in disease resistance [27] and abiotic stresses tolerance [28][29][30]. Studies related to AS on wheat grain traits are mainly focused on grain quality. AS events and their biological functions in wheat genes related to grain weight and size remain poorly understood.
Here, we report that TaGS3 undergoes AS and produces five splicing variants, corresponding to the constitutive TaGS3.1, the truncated TaGS3.2-3.4 lacking the C-terminal Cys-rich region, and the truncated TaGS3.5 that contains the disrupted OSR domain. Evolutionary analysis showed that AS of GS3 did not emerge during the process of wheat polyploidization, but pre-existed in the ancestor species of Triticeae crops. Given proteinprotein interactions between TaGS3 isoforms and WGB1 (G β ), we biochemically analyzed their nature and found that TaGS3.2, TaGS3.3, and TaGS3.4 interfere with TaGS3.1 function by competitively forming functional G βγ heterodimers, and that TaGS3.5 is unable to form a G βγ heterodimer with WGB1. Correspondingly, overexpression of the five TaGS3 splicing variants in bread wheat exhibited divergent effects on grain development. In particular, TaGS3.5 overexpression produced significantly larger and heavier grains. Taken together, these observations support the proposal that the TaGS3-mediated signal transduction pathway regulates grain weight and size via AS, providing a strategy by which the regulation of grain size is fine-tuned and balanced at the post-transcriptional level.

AS Occurrence of TaGS3 in Wheat
A previous study indicated that TaGS3 negatively regulates grain weight and size in wheat [10]. During TaGS3 cloning, we identified several distinct CDSs of TaGS3.
To decipher the identity of the transcripts, we sequenced the amplicons and confirmed the presence of five unique TaGS3 transcripts ( Figure S1). Except for the constitutive variant TaGS3.1, four splicing variants arise from Alt 3 ss, Alt 5 ss, IR, and ES, leading to the production of TaGS3.2, TaGS3.3, TaGS3.4, and TaGS3.5, respectively ( Figure 1A).
TaGS3.1 contains five exons and four introns, and is predicted to encode the full length γ-subunit protein of 170 amino acid residues, as previously reported [11]. Sequence analysis revealed that AS results in TaGS3.2, TaGS3.3, and TaGS3.4 with varied degrees of retention in the third intron, with TaGS3.2 harboring seven nucleotides (CTTGCAC) at the 3 -end of intron 3, TaGS3.3 harboring 14 nucleotides (GTATGGATTTTCAG) at the 5 -end of intron 3, and TaGS3.4 harboring the full length of intron 3 (82 bp). As a result, TaGS3.2, TaGS3.3, and TaGS3.4 are all frame-shifted, with introduction of a PTC at the 213th nucleotide site of exon 4, 285th nucleotide site of exon 5, and 216th nucleotide site of exon 4, respectively. For TaGS3.5, AS has resulted in skipping the third exon ( Figure 1A and Figure S2).

Expression Patterns and Subcellular Localization of TaGS3 Splicing Variants
The temporal and spatial expression patterns of the five splicing variants of TaGS3 were investigated in KN9204 through qRT-PCR using cDNA-specific primers. TaGS3.1, TaGS3.2, TaGS3.3, and TaGS3.4 were ubiquitously expressed with similar patterns in various wheat tissues at different developmental stages, showing higher expression levels in young spikes and developing grains than in the roots, stems, and leaves. Nevertheless, there were notable differences in expression abundances among the 5 variants; the maximum expression levels of TaGS3.1, TaGS3.2, and TaGS3.3 were observed in the 21 DPA (day post-anthesis) grains, TaGS3.4 expression level peaked in the 28 DPA grains ( Figure 3A), but TaGS3.5 expression was undetectable in all assayed tissues by qRT-PCR.
AS frequently results in alterations in protein structures and subcellular locations [31]. To determine the subcellular localization of TaGS3 isoforms, we expressed the five TaGS3 splicing variants as fusions with GFP under the control of the CaMV 35S promoter in N. benthamiana. The localization of TaGS3.1-GFP and TaGS3.5-GFP fusion protein fluorescence was only observed in the plasma membrane, while TaGS3.2-GFP, TaGS3.3-GFP, and TaGS3.4-GFP fusion protein fluorescence was observed in both the plasma membrane and the nuclei ( Figure 3B). Consistent with these results, Arabidopsis AGG3 was present in the plasma membrane, but AGG3 108-125 , in which the predicted transmembrane domain is deleted, was observed in both the plasma membrane and the nucleus [6].
The stability of TaGS3 isoforms in the corresponding overexpression lines was assayed. Immunoblot analysis revealed trace expression of TaGS3.2-GFP that was barely detected, and the abundance of TaGS3.3-GFP and TaGS3.4-GFP was significantly reduced under the treatment of cycloheximide (CHX), a protein synthesis inhibitor, whereas the accumulation of TaGS3.1-GFP and TaGS3.5-GFP only changed slightly. The accumulation of all TaGS3.2-GFP, TaGS3.3-GFP, and TaGS3.4-GFP but not TaGS3.1-GFP and TaGS3.5-GFP increased under treatment with the proteasome inhibitor MG132 ( Figure S7). These results indicate that TaGS3.1 and TaGS3.5 are more stable than TaGS3.2-3.4, implying the strong influence of the C-terminal Cys-rich region on TaGS3 s stability.

Overexpression of TaGS3 Splicing Variants Confers Different Effects on Wheat Grain Weight and Size
To study the genetic effects of the five TaGS3 splicing variants in bread wheat, we generated TaGS3.1, TaGS3.2, TaGS3.3, TaGS3.4, and TaGS3.5 transgenic lines in T. aestivum cv. KN199 ( Figure 4A). Fold increase in the expression abundances of TaGS3 variants was examined in T3 homozygous overexpression lines ( Figure S6). To investigate the relative abundance of the specific proteins, we tested the transgenic plants expressing GFP taggedfusion of TaGS3.1, TaGS3.2, TaGS3.3, TaGS3.4, and TaGS3.5 in KN199 driven by the   The agronomic attributes of TaGS3.1-3.5 overexpression lines were examined in field studies conducted in 2019. No significant differences were found between TaGS3.1-3.5 overexpression lines and WT in terms of plant height (PH), spike number per plant (SN), spike length (SL), total spikelets per spike (TS), sterile spikelets per spike (SS), and grain number per spike (GN) ( Figure 4A; Table S2). However, significant differences were detected for each TaGS3.1-3.5 overexpression line compared with WT in grain traits such as grain length (GL), grain width (GW), thousand-grain weight (TGW), and grain yield per plant (GY) (Figure 4 C-F). Compared to WT, TaGS3.1 overexpression lines exhibited significantly smaller and lighter grains, with a reduction of 5.04% (0.30 ± 0.04 mm) in GL, 2.49% (0.08 ± 0.02 mm) in GW, 5.89% (2.77 ± 0.54 g) in TGW, and 5.77% (0.79 ± 0.19 g) in GY. There were no significant variations in GL, GW, TGW, and GY between TaGS3.2-3.4 overexpression lines and WT, but a 0.39-0.95% decrease in GL, 0.80-1.52% decrease in TGW and 0.92-1.91% decrease in GY between TaGS3.2-3.4 overexpression lines and WT. Notably, TaGS3.5 overexpression lines produced significantly larger and heavier grains, with an increase of 4.39% (0.26 ± 0.03 mm) in GL, 1.39% (0.05 ± 0.02 mm) in GW, 5.70% (2.68 ± 0.65 g) in TGW, and 5.41% (0.74 ± 0.19 g) in GY ( Figure 4C-F). During three growing seasons, all TaGS3.1-3.5 overexpression lines displayed a recurring phenotype, similar to the results shown above (Table S2). These results suggested that TaGS3.1 and TaGS3.5 make opposing contributions to grain weight and grain size. The G protein consists of G α , G β and G γ subunits, where G α and G β play fundamental roles in the transduction of G protein-mediated growth signals, and G β -mediated signal transduction requires G γ for formation of active G βγ heterodimer [32,33]. To determine which section of TaGS3 is essential for G βγ dimer formation, we generated five TaGS3 isoforms: the full-length GS3 1-170 and the truncated GS3 1-66 , GS3 1-60 , GS3 1-51 , and GS3 67-170 . Different from GS3 1-170 , the GS3 1-66 , GS3 1-60 , and GS3 1-51 isoforms were all devoid of the C-terminal Cys-rich region. GS3 1-66 contains an intact OSR domain, while GS3 1-60 contains an incomplete OSR domain lacking the six amino acid residues of the highly conserved DPFITI motif, and GS3 1-51 contains an incomplete OSR domain that lacks 15 amino acid residues including the conserved DPFITI motif. By contrast, GS3 67-170 only comprises the C-terminal Cys-rich region without the N-terminal OSR domain ( Figure 5A). Yeast two-hybrid and β-galactosidase assays demonstrated that GS3 1-170 , GS3 1-66 , and GS3 1-60 interact with WGB1, in which GS3 1-66 exhibits stronger affinity to WGB1 compared to GS3 1−170 , while GS3 1-60 shows weak affinity to WGB1, indicating that both the conserved DPFITI motif and the deletion of the C-terminal Cys-rich region are required for the enhancement of GS3 binding to WGB1. GS3 1-51 and GS3 67−170 , on the other hand, did not interact with WGB1 ( Figure 5B). These results suggest that GS3 interacts with WGB1 through the OSR domain rather than the C-terminal Cys-rich region, and the section of the 15 amino acid residues in the OSR domain is essential for the binding of GS3 to WGB1.  Subsequently, we addressed whether the five TaGS3 isoforms encoded by TaGS3 splicing variants form a heterodimeric complex with WGB1. Here, the amino acid sequence of TaGS3.1 was the same as GS3 1-170 . Yeast two-hybrid and β-galactosidase assays confirmed that TaGS3.2, TaGS3.3, and TaGS3.4 bind more tightly to WGB1 than TaGS3.1. TaGS3.5, lacking the 15 amino acid residue section in the OSR domain, does not bind to WGB1 ( Figure 5). These results consistently support the necessity of the 15 amino acid residues in the OSR domain for the TaGS3-WGB1 interaction and the enhanced binding to WGB1 by TaGS3 when the C-terminal Cys-rich region is deleted. Furthermore, these results of the yeast two-hybrid assay were confirmed by pull-down and coimmunoprecipitation (Co-IP) assays, where WGB1 interacted with TaGS3.1, TaGS3.

Competitive Interactions of TaGS3 Isoforms with WGB1
In a nonactivated state, the Gα subunit maintains its GDP-bound state, forming an inactive heterotrimer with the Gβγ dimer [34]. To test whether TaGS3 isoforms form a heterotrimer with WGB1 and WGA1, we performed yeast three-hybrid assay and found that

Competitive Interactions of TaGS3 Isoforms with WGB1
In a nonactivated state, the G α subunit maintains its GDP-bound state, forming an inactive heterotrimer with the G βγ dimer [34]. To test whether TaGS3 isoforms form a heterotrimer with WGB1 and WGA1, we performed yeast three-hybrid assay and found that TaGS3.1, TaGS3.2, TaGS3.3, and TaGS3.4, but not TaGS3.5, coupled with WGB1 to interact with WGA1 ( Figure S8). In addition, yeast three-hybrid assay revealed competition between TaGS3

Conservation and Significance of GS3 AS in Gramineae
Recent large-scale genomics studies, especially genome-wide analysis of gene expression profiling, revealed that 30% to 61% of genes underwent AS [35][36][37]. Considering the orthologous gene pairs in Arabidopsis and rice, 58% of the same AS types were found, suggesting a role for the AS event as an evolutionarily conserved mechanism of post-tran- Figure 7. Competition test of the interactions between TaGS3 isoforms and WGB1. Yeast three-hybrid assay showing the competition between TaGS3.1 and the other four truncated TaGS3 isoforms to interact with WGB1 by using fusions with AD (AD-TaGS3 isoforms) and BD (BD-WGB1/BD-WGB1-GS3 isoform) (on the left). AD and BD represent empty pGADT7 and pBridge vectors, respectively. Empty vector was used as the negative control. -LMT, selective medium lacking Leu, Met, and Trp; -LMTH, selective medium lacking Leu, Met, Trp, and His. The specificity of the stringency of the assay was tested by adding 3-aminotriazole (3-AT, 20 mM/L). Quantitative analyses of interactions by β-galactosidase assay were shown (on the right) as mean ± SEM (n = 3).

Conservation and Significance of GS3 AS in Gramineae
Recent large-scale genomics studies, especially genome-wide analysis of gene expression profiling, revealed that 30% to 61% of genes underwent AS [35][36][37]. Considering the orthologous gene pairs in Arabidopsis and rice, 58% of the same AS types were found, suggesting a role for the AS event as an evolutionarily conserved mechanism of posttranscriptional regulation [38]. In this study, five GS3 splicing variants were found to exist in both diploid and hexaploid Triticeae species (Figure 2C), indicative of the functional significance of ancestral candidate AS. The GS3.1, GS3.2, and GS3.3 variants occur in B. distachyon and GS3.1 and GS3.5 in barley ( Figure S9 and Table S3), suggesting that AS of GS3 is conserved in wheat, barley, and B. distachyon. The constitutive variant GS3.1 is a major variant in both diploid ancestor grasses and hexaploid wheat, portraying a similar situation to most AS events in Arabidopsis and cereal crops, in that only one major transcript exists despite the presence of multiple splicing variants [14,39]. The frequency of AS occurrence in wheat tends to decrease after polyploidization, as determined by highthroughput transcriptome sequencing [24]. GS3 AS occurrence tends to decrease along with the increasing degree of wheat polyploidization, since a higher frequency of GS3 splicing variants is found in diploid wheat progenitor species compared with hexaploid wheat ( Figure 2C).
Polyploidization is often accompanied by changes in genomic structure and gene expression [40]. In hexaploid KN9204 and KN199, a single nucleotide difference causes the absence of one canonical splice junction (AG) in the third intron of TaGS3-7A ( Figure S5), which results in the third intron of TaGS3-7A differing from that of homeologs TaGS3-4A and TaGS3-7D. As a consequence, TaGS3.2 could not be isolated from TaGS3-7A in KN9204 and KN199. In our previous reports, TaGS3-4A and TaGS3-7D exhibited increased transcription abundance compared with TaGS3-7A in developing grains, suggesting that TaGS3-4A and TaGS3-7D play more important roles in grain development [11]. Whether the presence of TaGS3.2 contributes to increased TaGS3 expression levels needs further investigation. Our findings here indicate that genomic-structure-triggered AS could contribute to functional diversity among homeologs in bread wheat, which is different from those caused by epigenetic modifications [41][42][43]. Grain size is a complex trait, and different levels of gene function regulation may dominate grain development. Therefore, we inferred the functional significance of TaGS3 AS event as an evolutionarily conserved mechanism in the regulation of wheat grain weight and size.

Differential Functions and Mechanisms of TaGS3 ASs in Wheat
Some splicing variants compete with the constitutive variant and interfere with its function in a dominant negative manner [17,44]. For instance, FLOWERING LOCUS T (FT) gene FT2 generates two splicing variants in B. distachyon, FT2α and FT2β, which play antagonistic roles in regulating the flowering processes [15]. A similar mechanism was identified here. TaGS3.2, TaGS3.3, and TaGS3.4 compete with the constitutive TaGS3.1 variant in binding to WGB1 to form a functional G βγ heterodimer, respectively (Figures 6 and 7), and thereby regulate grain weight and size. However, the overexpression of TaGS3.2-3.4 that contain an intact OSR domain coding region ( Figure 1B) did not obviously affect grain size ( Figure 4C-E), which may be attributed to the low proportion of TaGS3.2-3.4 and their instability ( Figure S7). On the other hand, there is abundant steady-state accumulation of TaGS3.3-GFP and TaGS3.4-GFP in overexpression lines ( Figure 4B, Figure S7). We observed ectopic subcellular localization of TaGS3.3-GFP and TaGS3.4-GFP in the cytoplasm and nucleus ( Figure 3B Furthermore, AS of FLM (FLOWERING LOCUS M) and HAB1 (HYPERSENSITIVE to ABA1) in Arabidopsis exhibited opposite effects by antagonistic splicing variants, respectively [17,44]. In this study, overexpressing TaGS3.5 significantly increased grain weight and size, which was opposite to the negative effect of TaGS3.1 (Figure 4). In the TaGS3.5 transgenic lines, TaGS3.5 expression level was markedly increased, while TaGS3.1 expression level was nearly unchanged (Figure S10), indicating that the positive effect of TaGS3.5 in enlarging grain size is independent of the TaGS3.1 expression level. TaGS3.5 is unable to interact with WGB1 to form active G βγ heterodimer (Figure 6), indicating that the opposing functions of TaGS3.5 and TaGS3.1 do not arise from competition. TaGS3.5 was observed in trace amounts compared to TaGS3.1 ( Figure 2C) but appeared stable in vivo ( Figure S7). Thus, it is reasonable that the constant retention of TaGS3.5 in overexpression lines may obstruct the normal G-protein-mediated signal transduction pathway and lead to the production of large grains.

Functional Diversification of GS3 in Rice and Wheat
OsGS3 overexpression in rice generated shorter plants with smaller grains and variable grain number [45]. TaGS3.1 and TaGS3.5 overexpression here had contrasting and overwhelming effects on morphological grain traits but did not significantly impact plant height and spike characteristics (Figure 4), indicating the functional specialization of TaGS3 in grain size regulation in wheat. Furthermore, the increased TGW resulting from TaGS3.5 overexpression consistently contributed to increased individual grain yield (Table S2), indicating the great value of TaGS3.5 in the genetic improvement of wheat aimed at elevating grain yield.
Nucleotide differences among rice varieties mainly occur in exon regions of OsGS3, which regulate grain traits by producing frameshift mutations that cause premature transcriptional termination [10]. Compared with OsGS3, sequence polymorphism in TaGS3 mainly occurs in intron regions [11], which results in AS rather than allelic variations in TaGS3. Both the truncated OsGS3 and TaGS3 function in the determination of grain size, with the truncated GS3 proteins generated in different ways, namely by allelic variations in rice but by AS in wheat. A nonsense mutation (C to A) in the second exon of OsGS3 in the Minghui 63 allele results in a PTC and, consequently, truncated OsGS3 devoid of 17 amino acid residues of the OSR domain and the entire C-terminal Cys-rich region, with the production of longer grains [9]. Overexpressing TaGS3.5 in wheat also leads to longer grains ( Figure 4). Notably, TaGS3.5, which arises from skipping of the third exon, results in a truncated TaGS3.5 lacking 15 amino acid residues that include the DPFITI motif that is part of the OSR domain ( Figure 1). These observations in both rice and wheat demonstrate that an intact OSR domain is necessary for GS3 to function as a negative regulator of grain size. The role of TaGS3.5 with the whole C-terminal Cys-rich region, which is deleted in the truncated OsGS3 in Minghui 63, illustrates the causal relationship between the deletion of the 15 amino acid residues in the OSR domain and the positive regulation of grain size, showing a new hierarchical regulation mechanism mediated by AS variation in wheat.
The C-terminal Cys-rich region (tail) is characteristic of the noncanonical Gγ subunits unique to rice. Variations in the length of the C-terminal tail of G γ protein distinctly affect the final grain phenotype, in which the long-tailed G γ protein gives long grains in rice while the short-tailed and tailless G γ proteins give short grains [45]. Compared with the wildtype OsGS3, a 1 bp deletion carried by the Chuan 7 allele results in a truncated OsGS3 that retains the OSR and TM domains but with deletion of most sections of the C-terminal tail, resulting in the production of super short grains [10]. Contrary to this, overexpressing the long-tailed TaGS3.1 in wheat results in the production of short grains, but overexpressing the tailless TaGS3.2-3.4 results in medium-sized grains ( Figure 4C-F), suggesting that the length of the C-terminal Cys-rich region in TaGS3 is negatively correlated with grain size in wheat. These observations in both rice and wheat indicate that the C-terminal Cys-rich region of OsGS3 and TaGS3 have divergent functions in regulating grain weight and size.

Isolation of GS3 Splicing Variants
Three sets of primers from 5 UTR (untranslated region) and 3 UTR were used to amplify TaGS3 coding sequences from chromosomes 7A, 4A, and 7D. PCR products were separated by electrophoresis in agarose gels, and the target bands were purified and cloned into the pTOPO-Blunt Simple Vector (Genstar, Beijing, China), which was transformed into DH5α using the heat-shock method (Transgen, Beijing, China).
One set of specific primers corresponding to the CDS (coding sequence) of TaGS3 was used to amplify the specific TaGS3 target DNA, the single colonies of which were identified by PCR to distinguish the corresponding sequences of TaGS3.1 (253 bp), TaGS3.2 (260 bp), TaGS3.3 (267 bp), TaGS3.4 (335 bp), and TaGS3.5 (208 bp). Positive clones were confirmed by sequencing. Each TaGS3 splicing variant was analyzed by alignments with the corresponding coding sequence. The number of positive clones in TMB02, TH02, and Y199 was 609, 484, and 932, respectively. In KN9204, the number of positive clones from chromosomes 7A, 4A, and 7D was 858, 803, and 932, respectively. In KN199, the number of positive clones from chromosomes 7A, 4A, and 7D was 787, 820, and 794, respectively.
The coding regions of HvGS3 and BdGS3 were cloned from H. vulgare cv. Morex and B. distachyon Bd21, respectively, on the basis of the corresponding reference sequences of H. vulgare cv. Lasa Goumang (SDOW01000566) and B. distachyon Beauv (XM_014896470.2), the single colonies of which were identified by PCR. Positive clones were confirmed by sequencing. Each HvGS3 and BdGS3 splicing variant was analyzed by alignments with the corresponding coding sequence. The number of positive clones in HvGS3 and BdGS3 was 47 and 4, respectively. All the primers are listed in Table S4.

Phenotype Assessment
Five to seven T 3 homozygous overexpression lines of each TaGS3 splicing variant and WT were grown in the field at Luancheng Agro-Ecosystem Experimental Station, CAS, China. A random block design with three replications was performed in which each overexpression line was planted in a 2-row plot that was 2 m long with 25 cm row spacing and 40 seeds per row. Prior to harvest, 10 random plants in the central region of each plot in the main tiller were examined to determine their plant height (PH), spike length (SL), and spike number per plant (SN), and 30 spikes were randomly sampled and examined for determination of the total spikelets per spike (TS), sterile spikelets per spike (SS), and grain number per spike (GN). After harvest, the individual grain yield (GY) of 10 representative plants was recorded, and grains were collected for phenotype assessment. Grain agricultural traits, including grain length (GL), grain width (GW), and thousand-grain weight (TGW), were measured using an SC-G multifunctional seed analyzer (Wanshen, Hangzhou, China).

Plasmid Construct
The coding regions of WGA1 (G α ) and WGB1 (G β ) were cloned from KN9204 on the basis of the corresponding reference sequences of WGA1 (G α ) (MG748862.1) and WGB1 (G β ) (XM_037566621) in Chinese Spring (CS). For subcellular localization examination, the coding sequences of five TaGS3 splicing variants were cloned into the pCAMBIA 1300-35S-GFP vector. In the protein-protein interaction experiments, the coding sequences of the five TaGS3 splicing variants and DNA sequences to code the full-length GS3 1-170 and the truncated GS3 1-66 , GS3 1-60 , GS3 1-51 , and GS3 67-170 were cloned into pGADT7; then, the coding sequence of WGB1 was cloned into the pGBKT7 vector for the yeast two-hybrid assay, and the coding sequences of the five TaGS3 splicing variants and WGB1 were cloned into pMAL-C2X and pGEX-4T-1, respectively, for the pull-down assay. For the coimmunoprecipitation assay, the coding sequences of the five TaGS3 splicing variants and WGB1 were cloned into the pCAMBIA1300-35S-GFP and pCAMBIA1300-35S-FLAG vectors, respectively. For the yeast three-hybrid assay, the coding sequences of WGA1 were cloned into pGADT7, and the coding sequences of WGB1 and five TaGS3 splicing variants were cloned into the pBridge vector. All primers for vector construction are listed in Table S4.

RNA Extraction and qRT-PCR
For tissue-specific expression analysis, different leaf tissues from the plant at the seedling, jointing, booting, and heading stages, along with grain samples of 7, 14, 21, and 28 days post-anthesis were collected from the overexpression lines of the five TaGS3 splicing variants grown in the field. Total RNA was isolated using an RNA extraction kit (Tiangen, Beijing, China) and quantified by Nanodrop (Thermo, Waltham, MA, USA). First-strand cDNA was synthesized from DNaseI-treated total RNA using the PrimeScript RT Reagent Kit (TaKaRa, Tykyo, Japan) according to the manufacturer's instructions.
qRT-PCR was carried out in a total volume of 20 µL using the SYBR PCR kit (TaKaRa, Tykyo, Japan) and on a CFX96 Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA) according to the manufacturer's instructions. At least three biological replicates were assayed, for which three technical replications were conducted. The GAPDH locus served as a normalization control to determine the relative expression level of each splicing variant. Specific primer sequences are listed in Table S4.

Yeast Two-and Three-Hybrid Assays
The yeast two-hybrid experiment was performed using the MATCHMAKER GAL4 Two-Hybrid System (Clontech, Mountain View, CA, USA). Briefly, respective prey and bait vectors were cotransformed into yeast strain AH109 using the lithium acetate transformation method. The transformed yeast cells were selected on SD/-Leu-Trp (SD/-LT) dropout medium. Interactions were verified on the SD/-Leu-His-Trp (SD/-LHT) dropout medium with different concentrations of 3-aminotriazole (0 and 2 mM/L).
For liquid β-galactosidase assay with ONPG as substrate, liquid cultures in SD/-LHT with three yeast colonies were inoculated overnight. The calorimetric β-galactosidase assay using the supernatant and the following activity calculation were performed as described in the Clontech Yeast Protocols Handbook.

Pull-Down Assay
TaGS3.1-GST, TaGS3.2-GST, TaGS3.3-GST, TaGS3.4-GST, TaGS3.5-GST, and WGB1-MBP plasmids were transformed into BL21 cells. Protein purification was performed using Glutathione Sepharose TM 4B (GE Healthcare, Uppsala, Sweden) and Amylose Resin (NEB, Beverly, MA, USA) according to manufacturer instructions. GST beads were washed with GST binding buffer four times to remove the ethanol. Purified GST and TaGS3.1-GST, TaGS3.2-GST, TaGS3.3-GST, TaGS3.4-GST, and TaGS3.5-GST proteins were incubated with the same volume of GST beads in binding buffer (20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% Nonidet P40, with 1 mM phenylmethylsulfonyl fluoride (PMSF) and 0.5 mM dithiothreitol) for 2 h at room temperature. These proteins were centrifuged at 800× g for 3 min, and pellets were extensively washed with binding buffer four times to remove redundant proteins. Each purified protein was incubated with WGB1-MBP in binding buffer for another 2 h at room temperature or 4 • C overnight. Next, the mixture was washed four times by binding buffer to remove redundant WGB1-MBP. Samples were then collected and boiled for 5 min in 2× SDS protein loading buffer prior to Western blotting assay. The anti-MBP (TDY Biotech, Beijing, China, 1:5000) and anti-GST antibodies (TDY Biotech, Beijing, China, 1:5000) were used to detect MBP and GST tagged proteins, respectively.

Statistical Analyses
Statistical analyses were based on phenotypic data for grain weight and size. One-way ANOVA was performed in the SPSS System for Windows version 20.0 (IBM Corporation, Armonk, NY, USA) to determine phenotypic differences between overexpression lines and WT, and Tukey tests were conducted to determine the significance of differences.

Conclusions
TaGS3 undergoes AS and produces five splicing variants that show differential functions in the determination of grain weight and size. AS of GS3 did not emerge during the process of wheat polyploidization, but pre-existed in the ancestor species of Triticeae crops. TaGS3.1 overexpression significantly reduced grain weight and length, TaGS3.2-3.4 overexpression did not significantly alter grain weight and size compared to wild type, and TaGS3.5 overexpression significantly increased grain weight and grain length. TaGS3 isoforms with an intact OSR domain (TaGS3.1-3.4) interact with WGB1 to form G βγ heterodimers that further interact with WGA1 to form inactive G αβγ heterotrimers. Truncated isoforms TaGS3.2-3.4 lacking the C-terminal Cys-rich region showed enhanced binding affinity to WGB1 and antagonistically compete with TaGS3.1 to bind WGB1, while TaGS3.5 with the incomplete OSR domain does not interact with WGB1. Taken together, the results indicate that TaGS3 differentially regulates grain size via AS, by which the regulation of grain size is fine-tuned and balanced at the post-transcriptional level.