A novel allele of TaGW2-A1 is located in a finely mapped QTL that increases grain weight but decreases grain number in wheat (Triticum aestivum L.)

Key message A novel TaGW2-A1 allele was identified from a stable, robust QTL region, which is pleiotropic for thousand grain weight, grain number per spike, and grain morphometric parameters in wheat. Abstract Thousand grain weight (TGW) and grain number per spike (GNS) are two crucial determinants of wheat spike yield, and genetic dissection of their relationships can help to fine-tune these two components and maximize grain yield. By evaluating 191 recombinant inbred lines in 11 field trials, we identified five genomic regions on chromosomes 1B, 3A, 3B, 5B, or 7A that solely influenced either TGW or GNS, and a further region on chromosome 6A that concurrently affected TGW and GNS. The QTL of interest on chromosome 6A, which was flanked by wsnp_BE490604A_Ta_2_1 and wsnp_RFL_Contig1340_448996 and designated as QTgw/Gns.cau-6A, was finely mapped to a genetic interval shorter than 0.538 cM using near isogenic lines (NILs). The elite NILs of QTgw/Gns.cau-6A increased TGW by 8.33%, but decreased GNS by 3.05% in six field trials. Grain Weight 2 (TaGW2-A1), a well-characterized gene that negatively regulates TGW and grain width in wheat, was located within the finely mapped interval of QTgw/Gns.cau-6A. A novel and rare TaGW2-A1 allele with a 114-bp deletion in the 5′ flanking region was identified in the parent with higher TGW, and it reduced TaGW2-A1 promoter activity and expression. In conclusion, these results expand our knowledge of the genetic and molecular basis of TGW-GNS trade-offs in wheat. The QTLs and the novel TaGW2-A1 allele are likely useful for the development of cultivars with higher TGW and/or higher GNS. Electronic supplementary material The online version of this article (10.1007/s00122-017-3017-y) contains supplementary material, which is available to authorized users.


Introduction
Wheat provides approximately 20% of the calories consumed by humankind (Simmonds et al. 2016). Considering the continued global population growth and the low rates of genetic gain in wheat yield (Zheng et al. 2011;Ray et al. 2013), there is an urgent need to identify, characterize, and incorporate genomic tools that can accelerate wheat yield improvement (Simmonds et al. 2016). Wheat yield is controlled by polygenes and affected by environmental factors, and it mainly relies on three components: thousand grain weight (TGW), grain number per spike (GNS), and spike number per hectare (Simmonds et al. 2014). There are negative correlations among these components (Griffiths et al. 2015), but the genetic basis underlying individual yield traits and their interactions are still largely unknown in wheat.
In rice, it has been shown that grain weight is affected by the genes that functioning in several pathways, i.e., proteasomal degradation (GW2 and GW5/qSW5), phytohormone signaling (GS6, TGW6, and OsCKX2), and G protein-mediated signal transduction (GS3 and RGB1); these genes regulate cell division and/or cell expansion in specific grain tissues (reviewed by Zuo and Li 2014). Using a homology-based approach, several wheat genes have been isolated and suggested to be associated with TGW. These include TaCwi (Jiang et al. 2015b;Ma et al. 2012), TaTGW6 (Hanif et al. 2016;Hu et al. 2016), TaGW2-A1 (Simmonds et al. 2016;Su et al. 2011;Yang et al. 2012), TaGS1a , TaGS5-3A , TaGASR7-A1 (Dong et al. 2014;Ling et al. 2013) and TaCYP78A3 (Ma et al. 2015). Research on these genes has enhanced our understanding on grain weight determination in wheat and has also provided functional markers useful for selecting higher TGW through marker-assisted breeding.
Several major QTLs for grain number have also been isolated and characterized in rice. Some of these QTLs control inflorescence meristem identity (e.g., LAX1, SPA, and FZP), while others influence the rate (e.g., APO1, Gn1a/OsCKX2, LOG, SP1, and DEP1) or duration (e.g., RCN1, RCN2, Ghd7, and Ghd8) of cell division in the inflorescence meristem (reviewed by Xing and Zhang 2010). Although the functions of orthologous genes are generally conserved in rice and wheat, there are several examples of functional divergence between rice grain number genes and their wheat homologs. For instance, OsCKX2, encoding a cytokinin oxidase, is a negative regulator of the number of grains per panicle in rice (Ashikari et al. 2005). On the other hand, the two wheat homologs of OsCKX2, i.e., TaCKX2.1 and TaCKX2.2, have been suggested to positively control GNS . Furthermore, while MOC1 and OsTEF1 are two crucial regulators of rice tillering (Li et al. 2003;Paul et al. 2012), their wheat homologs (TaMOC1-7A and TaTEF-7A) have both been found to be stably associated with spikelet number per spike rather than with tiller number (Zhang et al. 2015;Zheng et al. 2014). Differences such as these may be associated with the contrasting architecture of the inflorescences between wheat (spikes) and rice (panicles).
Here, we report mapping of several stable QTLs for TGW and/or GNS using 191 recombinant inbred lines (RILs) derived from a cross between two wheat lines that differ in both TGW and GNS. Of these QTLs, the one on chromosome 6A displayed a strong TGW-GNS trade-off. This QTL was further validated and precisely mapped using near isogenic lines (NILs). A novel allele of TaGW2-A1 was isolated from the finely mapped interval, which is likely a candidate.

Plant materials
A population of 207 RILs was developed from the cross between two Chinese hexaploid winter wheat cultivars, i.e., Yumai 8679 (Y8679) and Jing 411 (J411), and advanced to the F 9 generation by single seed descent. Of the 207 RILs, 191 were genotyped and evaluated in 11 field trials. An F 9 plant of the RIL line (RIL186) with residual heterozygosity at the marker locus Xbarc118 was selfed and provided an F 9:10 family with 163 progeny. Afterwards, selfing of these plants was carried out to obtain a population of 163 F 10:11 families. Four sets of NIL pairs (NIL1, NIL2, NIL3, and NIL4) were obtained by selfing selected F 11 progeny with overlapping heterozygous fragments. Each NIL set comprised eight to 25 Y8679-type homozygotes and 12-30 J411-type homozygotes.
In total, 1113 wheat accessions with varying ploidy were used to test the allele frequency of a novel TaGW2

Field experiments and phenotyping
The RIL population was grown in three replicates following randomized complete-block design at the following five  Table S4). The field trials were carried out during two crop seasons (2011-12 and 2012-13) at Anhui, four crop seasons (2010-11, 2011-12, 2012-13, and 2014-15) at Beijing, three crop seasons (2011-12, 2012-13 and 2013-14) at Shaanxi, and one crop season (2013-14 or 2014-15) at the other two sites. For the field trials, the seeds were sown in double-row plots (2 m long) at a sowing rate of 30 seeds per row. At maturity, 20 representative spikes (from 20 different plants) were sampled from each plot and threshed together. For the grain samples obtained from the first two crop seasons (2010-11 and 2011-12), GNS, TGW, and grain weight per spike (GWS) were manually assessed. For the samples obtained from the other three seasons (2012-13, 2013-14, and 2014-15), seven traits [i.e., GNS, TGW, GWS, grain surface area (GA), grain circumference (GC), grain length (GL), and grain width (GW)] were recorded using a scaled cameraassisted phenotyping system (Wanshen Detection Technology Co., Ltd., Hangzhou, China). For each RIL line, the mean value of each trait was calculated across three replicates for each individual environment. In addition, best linear unbiased prediction (BLUP) values were predicted for each trait using the PROC MIXED procedure in SAS v9.1.3 (SAS Institute Inc., North Carolina, USA). Pearson's 1 3 correlation coefficient was calculated using SPSS v20.0 (SPSS, Chicago, USA) to assess the correlation among TGW, GNS, GWS, and grain morphometric parameters. The Shapiro-Wilk test was performed using R v3.2.2 to identify departures from normal distribution. Broad sense heritability (h 2 B ) based on a family mean basis was calculated using the PROC GLM procedure in SAS according to the following formula: h 2 B = 2 g ∕( 2 g + 2 ge ∕n + 2 ∕nr), where 2 g is the genotypic effect, 2 ge is the genotype by environmental effect, σ 2 is the residual error, n is the number of environments and r is the number of replicates .
The NIL pairs were evaluated at three sites (Beijing, Hebei, and Shanxi) in crop seasons 2014-15 (NIL1) and 2015-16 (NIL1, NIL2, NIL3, and NIL4). The 163 F 10:11 families were evaluated at Beijing during crop season 2014-15. These NIL pairs and families were grown in single-row plots (2 m long and 30 seeds per row) in randomized complete-block designs with three replicates. At maturity, 20 representative spikes were sampled, and the GWS, GNS, TGW, and grain morphometric parameters (GA, GC, GL, and GW) were assessed. For the members of NIL1 pair, tiller number (TN) and grain weight per plant (GWP) were determined for each of ten representative plants. The significance of phenotypic variations between NIL1-Y8679 and NIL1-J411 was calculated using Student's t test.

Linkage map and QTL analysis
The Y8679/J411 linkage map used here is the one that was described by Zhai et al. (2016). It includes both single-nucleotide polymorphism (SNP) markers and simple sequence repeat (SSR) markers. QTL analysis was conducted using both the within-environment means and the across-environment BLUPs for each trait. The methods used for QTL analysis were the same as those used by Zhai et al. (2016) for other traits in the same population. Briefly, this involved the use of WinQTLCart version 2.5  for composite interval mapping (CIM) using model 6 with forward and backward regression, five markers as cofactors, and a 10-cM scanning window. Empirical threshold LOD scores estimated with 1000 permutations at P ≤ 0.05 were used to declare a significant QTL (Churchill and Doerge 1994). Detected QTLs with overlapping confidence intervals (± 2 LOD away from the peaks of likelihood ratios) were considered equivalent and named as suggested by McIntosh et al. (2011).
For a QTL of interest on chromosome 6A, the 163 F 10:11 families derived from RIL186 were assigned to three genotypic classes (two homozygous families and one segregating families) based on their marker genotypes. This classification was used to estimate the additive effect (a), dominance effect (d), and dominance degree (d/a) (Falconer and Mackay 1996). Multiple comparisons among the mean values were estimated by the LSD method. The inheritance mode of an individual QTL can be classified into four categories, i.e., additive (d/a ≤ 0.20), partial dominance (0.20 < d/a < 0.80), dominance (0.80 ≤ d/a < 1.20), and overdominance (d/a ≥ 1.20), as described in the previous studies (Jiang et al. 2015a;Li et al. 2017).
The SNPs mapped in the QTL regions of interest were positioned onto the newly released reference genome sequence of Chinese Spring by blasting their flanking sequences against the IWGSC RefSeq v1.0 (https://urgi. versailles.inra.fr/blast_iwgsc/blast.php).

SSR marker development
Polymorphic SNP markers flanking the QTL region of interest were used to perform a BLAST search against genomic sequences of Brachypodium, rice, and Aegilops tauschii, to identify orthologous genomic regions (Supplementary Table S5) using the methods described by Zhai et al. (2016). The genes within the corresponding regions were used to search the hexaploid wheat cv. Chinese Spring IWGSC survey sequences (http://www.wheatgenome.org) to find homologous contig sequences for marker development ). These contig sequences were further used to search for SSR motifs (with at least 15 dinucleotide or trinucleotide repeats) and design PCR primers using BatchPrimer3 (http://probes.pw.usda.gov/batchprimer3). For polymorphism detection, PCR products were separated on 8% non-denaturing PAGE gels and visualized with silver staining. The primer pairs of 14 co-dominant SSR markers are listed in Supplementary Table S6.

RNA extraction, cDNA synthesis, and qRT-PCR
For cDNA cloning analysis, the seedling leaves of Y8679 and J411 were collected for total RNA extraction at 7 days after germination (dag). For expression analysis, the immature grains of NIL1-Y8679 and NIL1-J411 were collected for total RNA extraction at 11 days after pollination (dap) using the TransZol Plant Kit (TransGen Biotech). Firststrand cDNA was synthesized using M-MLV Reverse Transcriptase (Promega, WI, USA) according to the manufacturer's instructions.
qRT-PCR was performed to quantify the TaGW2-A1 transcripts in three independent biological repeats using the homoeolog-specific primer pair TaGW2A-Q (Hong et al. 2014) (Supplementary Table S6). qRT-PCR cycling was conducted using SYBR Green PCR master mix (TaKaRa, Japan) on a BioRad CFX96 system (CA, USA) using the following thermal profile: 95 °C for 7 min, 40 cycles at 95 °C for 10 s, 56 °C for 20 s, and 72 °C for 20 s. For each biological sample, three technical replicates were performed 1 3 for TaGW2-A1 and Actin (Supplementary Table S6). After normalizing to the endogenous control, the transcript levels were determined using the 2 −ΔC t method.

Isolation of full-length TaGW2-A1 cDNA
The complete coding sequences (CDS) of TaGW2 genes were obtained from the cDNA of Y8679 and J411 using the primer pair TaGW2-1 (Su et al. 2011) (Supplementary  Table S6). PCR products were gel-purified and cloned into the pGEM-T Easy Vector (Promega). The identities of three homoeologous coding sequences were determined by blasting against the wheat database (http://plants.ensembl.org/ common/Tools/Blast?db=core; Bolser et al. 2015) (Supplementary Fig. S2). The upstream and downstream regions of TaGW2-A1 in J411 were obtained using a rapid amplification of the cDNA ends (RACE) with the GeneRacer Kit (Invitrogen). The gene specific primers for 5′-RACE and 3′-RACE are listed in Supplementary Table S6.

Promoter isolation and activity analysis of TaGW2-A1
Three A-genome-specific primer pairs, including Hap-6A-P1 (Su et al. 2011), TaGW2-A1_InDel, and TaGW2-A1_ProF3 (Supplementary Table S6), were used to amplify a region of approximately 2.0-kb upstream from the initiation codon (ATG) in the TaGW2-A1 gene by PCR. The promoters of 1927 and 2041 bp length were obtained from Y8679 and J411, respectively, with a 114-bp deletion existing in the promoter obtained from Y8679.
To assess the effects of this 114-bp deletion on the promoter activity, promoter fragments from Y8679 (1121 bp) and J411 (1235 bp) were amplified (Supplementary Fig. S3) and fused with the β-glucuronidase (GUS) reporter gene sequence in the pBGWFS7.0 vector using the Gateway ® system following the manufacturer's instructions (Invitrogen, CA, USA). The plasmids were transferred into the Agrobacterium tumefaciens strain GV3101. The leaves of 7-week-old tobacco (Nicotiana benthamiana) plants were infiltrated with the Agrobacterium clones. After 3 days, the total protein was extracted from 1.0-cm leaf discs (without the central veins), and the specific GUS activity of these extracts was determined using a colorimetric assay according to Leborgne-Castel et al. (1999). Briefly, the protein extracts were incubated with p-nitrophenyl β-d-glucuronide (PNPG) substrate at 37 °C and absorbance of the p-nitrophenol (PNP) product was measured at 410 nm (Pawar et al. 2017). Leaves infiltrated with the empty vector were used as controls. Total protein content in the extracts was quantified using the Bradford protein assay.

Phenotypic performance of the RIL population
The population means and ranges of the seven investigated traits are listed in Table 1. Compared with J411, Y8679 had higher TGW and GWS, larger grain size (GA, GC, GL, and GW), and lower GNS ( Supplementary Fig. S4). TGW displayed an obvious shift towards higher values in the Y8679/ J411 RIL population, whereas the other six traits exhibited normal distributions ( Supplementary Fig. S5). All traits had broad sense heritabilities over 0.80. Pearson's correlation coefficients among the seven traits were calculated based on the BLUP values across seven shared environments (i.e., E5, E6, E7, E8, E9, E10, and E11), which showed that GNS was strongly and negatively correlated with TGW, GA, GC, GL, and GW (Supplementary Table S7). Table 1 Parental and population means, ranges, and broad sense heritabilities for TGW, GNS, GWS, and four grain morphometric parameters The traits include thousand grain weight (TGW), grain number per spike (GNS), grain weight per spike (GWS), grain surface area (GA), grain circumference (GC), grain length (GL), and grain width (GW). Delta refers to the difference between Y8679 and J411 phenotypes as a percentage of J411. Broad sense heritability based on a family mean basis was estimated across all evaluated environments for each trait. TGW and GNS were evaluated in 11 environments (from E1 to E11), whereas the four grain morphometric parameters were evaluated in seven environments (from E5 to E11)

Identification of genomic regions harboring stable QTLs for TGW, GNS, GA, GC, GL, and GW
QTLs repeatedly detected in ≥ 3 individual environments and in the analysis of BLUPs were considered to be stable. According to this criterion, 34 stable QTLs for TGW, GNS, GA, GC, GL, and GW were mapped within eight genomic regions (Table 2;  Supplementary Table S8). The corresponding physical intervals of these genomic regions in the Chinese Spring RefSeq v1.0 sequence are listed in Table 2 and Supplementary Table S9. Further 123 putative QTLs for TGW, GNS, GA, GC, GL, and GW that did not meet this criterion are listed in Supplementary  Table S10. Of the 29 QTLs for GWS, two were detected in three (QGws.cau-1B.3) or four (QGws.cau-4D.1) environments, but not in the analysis of BLUPs (Supplementary Table S10). The stable QTL regions on chromosomes 1B, 3A, and 7A were found to have an effect on GNS, but with no significant effect on TGW ( Fig. 1; Table 2). The region on chromosome Table 2 Eight genomic regions harboring stable QTLs for TGW, GNS, GA, GC, GL, and GW in the Y8679/J411 population a Additional details regarding the SNP markers within each QTL region can be found in Zhai et al. (2016) b The corresponding physical intervals (Mb) of the QTL regions on chromosomes 2B, 3A, 3B, 5B, 6A or 7A were obtained by blasting the flanking sequences of SNP markers to the Chinese Spring RefSeq v1.0 sequence (Supplementary Table S9) c The traits include thousand grain weight (TGW), grain number per spike (GNS), grain surface area (GA), grain circumference (GC), grain length (GL) and grain width (GW). The letters within the brackets indicate the origin of the increasing alleles, with 'Y' and 'J' representing Y8679 and J411, respectively d QTLs shown in bold are stable QTLs that were detected in ≥ 3 individual environments and in the analysis of BLUPs  3A and 7A, the alleles from J411 were found to have significant positive effects on GNS. The RILs carrying positive alleles at all three of the stable QTL regions for GNS (YJJ) exhibited significantly higher GNS (by 4.87 grains, P < 0.0001) and GWS (by 0.28 g, P < 0.0001) (Fig. 2a) than those possessing the three opposite alleles (JYY). The two groups did not differ significantly (P = 0.36) in TGW.

Use of NILs to verify and precisely map a QTL on chromosome 6A
RIL186 was selected from the population, because this RIL exhibited residual heterozygosity at locus Xbarc118. Two of its selfed progeny, which possessed alternative haplotypes and were homozygous, were genotyped using the Affymetrix wheat 660 K SNP array (http://wheat.pw.usda.gov/ggpages/ topics/ Wheat660_SNP_array_developed_by_CAAS.pdf). The results showed that they were 99.82% similar, only differing in 1094 SNPs (Supplementary Table S11). Of these 1094 polymorphic SNPs, 895 were located in an interval between 28.22 and 445.78 Mb on chromosome 6A. The flanking sequences of these 895 SNP markers and their physical locations in the Chinese Spring RefSeq v1.0 sequence are listed in Supplementary Table S12. Considering that the corresponding physical interval of the 6A region spanned from 52.40 to 585.43 Mb, we chose to focus on the region from 52.40 to 445.78 Mb (Table 2), i.e., an 8.908 cM interval between wsnp_BE490604A_Ta_2_1 and wsnp_RFL_Contig1340_448996 and designated as QTgw/ Gns.cau-6A, for the development of NILs (Fig. 3a, b).
Fourteen co-dominant polymorphic SSR markers (Supplementary Table S6) were developed using a collinearitybased strategy and used to genotype the F 9:10 family with 163 individuals derived from RIL186. The resultant genetic linkage map spanned 12.18 cM in length, covering an interval from 31.39 to 237.76 Mb in the Chinese Spring RefSeq v1.0 sequence (Supplementary Table S13). Based on genotype of these markers, four sets of NIL pairs (NIL1, NIL2, NIL3, and NIL4) with overlapping recombinant segments were developed (Fig. 3c). The first NIL pair, with no recombination in the concern region, was used to verify the effects of QTgw/Gns.cau-6A.
Across six replicated field trials, the TGW of NIL1-Y8679 (with Y8679 haplotype) was from 6.68 to 11.90% (mean 8.33%, P < 0.001) higher than that of NIL1-J411 ( Fig. 4; Supplementary Table S14). This variation in TGW was associated with average differences of 4.95, 1.87, 1.19, and 3.30% (P < 0.001) in GA, GC, GL, and GW, respectively, with NIL1-Y8679 having the higher values. A difference in GNS (mean 3.05%, P < 0.001) was also observed, with NIL1-Y8679 exhibiting the lower value. This effect varied across six trials with NIL1-Y8679 associated with significant GNS reductions (P < 0.05) from 2.71 to 8.20% in five of them. Overall, NIL1-Y8679 had 3.35% higher GWS (P < 0.001) and 4.29% higher GWP (P < 0.001) than NIL1-J411. In most trials, no significant differences were observed for tiller number (Supplementary Table S14). Collectively, these results demonstrate that the Y8679 haplotype in QTgw/ Gns.cau-6A has a consistent positive effect on GWS and GWP, which is conferred by a large positive effect on TGW, but counterbalanced by a small negative effect on GNS.
Subsequently, we used NIL2, NIL3, and NIL4 to map the high-TGW, low-GNS region into a smaller genetic interval. When these NIL pairs were evaluated for TGW and GNS in three field trials, significant differences in TGW and GNS were found only between the members of the NIL4 pair (Fig. 3d), but not between the members of the NIL2 or NIL3 pairs. Based on these results, QTgw/Gns. cau-6A was narrowed to an interval of 208.80 Mb, which is flanked by marker    (Fig. 3a).
To estimate the additive and dominance effects of QTgw/ Gns.cau-6A, we evaluated the 163 F 10:11 families for TGW, GNS, and GWS in one field trial (Beijing, 2014-15 crop season) in three replicates. Eighty segregating families and 36-41 homozygous families with alternative haplotypes 1 3 across the interval from 6AS-165 to 6AS-1 were used for variation analysis. The Y8679 haplotype of QTgw/Gns.cau-6A showed overdominance (d/a = 1.76) for higher TGW, with the mean values of segregating families (43.51 g) and Y8679-type homozygous families (43.75 g) being significantly higher (P < 0.001) than that of J411-type homozygous families (39.78 g). Likewisely, the J411 haplotype of QTgw/Gns.cau-6A exhibited overdominance (d/a = 2.16) for higher GNS, with the segregating families and J411-type homozygous families bearing 2.80 and 2.69 more grains (P < 0.001) than the Y8679-type homozygous families, respectively. We also observed an overdominance effect (d/a = 11.04) of QTgw/Gns.cau-6A on spike yield, with the mean GWS values of segregating families being 0.11-0.16 g higher (P < 0.01) than those of the two types of homozygous families (Table 3).

Isolation of candidate gene TaGW2-A1
TaGW2-A1, a wheat homolog of rice OsGW2 (Song et al. 2007), was previously reported to negatively affect TGW in wheat (Du et al. 2016;Simmonds et al. 2016;Yang et al. 2012). It was mapped to the short arm of chromosome 6A, near the centromere (Su et al. 2011). As this position corresponds with that of QTgw/Gns.cau-6A, we obtained the complete coding sequence and the promoter sequence (about 2.0 kb upstream from ATG) for TaGW2-A1 from Y8679 to J411, and analyzed their polymorphisms. No differences were observed in the coding sequence, but a 114-bp Insertion/Deletion (InDel) was detected in the promoter region (− 230 to − 117-bp upstream from ATG; Fig. 5a). No other variations were detected in the rest of the promoter sequence. The allele from Y8679 has the 114-bp deletion as  Fig. S6). The primer pair TaGW2-A1_InDel targeting this InDel was used to genotype the F 9:10 family with 163 progeny derived from RIL186. TaGW2-A1 was mapped to chromosome 6A near marker 6AS-1 (Fig. 3c), within the finely mapped interval of QTgw/Gns.cau-6A.
To further explore implications of our findings on wheat breeding, primer pair TaGW2-A1_InDel was used to amplify 848 hexaploid wheat accessions, including 798 Chinese varieties (96 landraces and 702 modern cultivars) and 50 French varieties ( Fig. 5d; Supplementary Table S1). Nearly all accessions (99.6%) amplified the wild-type 336bp J411 fragment, whereas only three (Jinan 13, Jinghua 9 and Gaoyou 0551) amplified the mutant type 222-bp Y8679  1 3 fragment ( Fig. 5e; Supplementary Fig. S7). Thus, at least within the panel evaluated, the favorable allele of TaGW2-A1 is very rare. Moreover, 238 tetraploid and 27 diploid wheat species were also screened for the 114-bp deletion ( Fig. 5d; Supplementary Tables 2 and 3), but none of them carried the 224-bp fragment like Y8679, again suggesting a rare mutation event leading to the 114-bp deletion.

Discussion
Five genomic regions exhibit negligible TGW-GNS trade-offs TGW and GNS are two primary yet negatively correlated determinants of spike yield in wheat (Jia et al. 2013;Shukla et al. 2015). Information about trade-offs between the two traits can help to improve the efficiency of wheat breeding.
Here, we report five genomic regions on chromosomes 1B, 3A, 3B, 5B, or 7A exhibiting negligible TGW-GNS tradeoffs. These represent attractive targets for marker-assisted selection to enhance grain size and/or grain number. The region on chromosome 1B can be attributed to the 1RS/1BL translocation (carried by Y8679). It increased GNS, corresponding to our previous finding that 1RS/1BL translocation enhanced spikelet number per spike across all evaluated trials (Zhai et al. 2016). In the regions on chromosomes 3A and 7A, alleles from J411 increased the grain number without significantly reducing grain weight. The 3A region is probably within the deletion bin 3AL5-0.78-1.00, which has previously been reported to harbor QTLs for grain yield, GNS, spikelet number per spike, test weight, TGW, and heading time (Ali et al. 2011;Cui et al. 2014;Huang et al. 2004;Zhai et al. 2016). Similarly, the region detected on chromosome 7A has previously been reported to affect GNS, spikelet number per spike and grain yield (Kumar et al. 2007;Ma et al. 2007;Reif et al. 2011;Wu et al. 2012). In the regions on chromosomes 3B and 5B, alleles from Y8679 increased TGW without significantly reducing GNS. These two regions coincide with several previously reported QTLs for TGW and grain yield (Cui et al. 2014;Reif et al. 2011;Sun et al. 2009). Notably, the region on chromosome 3B has also been shown to enlarge grain size through enhancing grain filling rate in the Heshangmai/Y8679 population, which shares a common parent with the population used in this study ).

Analysis of the QTLs on chromosome 6A provides new insight into TGW-GNS trade-off
In recent years, evidence is accumulating that robust QTLs for TGW (Cui et al. 2016;Mir et al. 2012;Snape et al. 2007;Zanke et al. 2015) and GNS (Jia et al. 2013;Yuan et al. 2012 From a practical point of view, it is worthy to point out that when pyramiding the QTLs identified here, it seems advisable to select the Y8679 haplotype in the presence of major QTLs for GNS on chromosomes 3A and 7A (Fig. 2c), or the J411 haplotype in the presence of major QTLs for TGW on chromosomes 3B and 7B (Fig. 2d).

A novel mutation of TaGW2-A1 is associated with increased TGW in wheat
In rice, OsGW2 encodes a RING-type protein with E3 ubiquitin ligase activity that negatively regulates the grain weight and width (Song et al. 2007). Its wheat homolog on the short arm of chromosome 6A (TaGW2-A1) locates in a region near the centromere (Su et al. 2011), in which major QTLs for TGW have been mapped (Cui et al. 2016;Simmonds et al. 2014;Zanke et al. 2015). Several reports (Su et al. 2011;Zhang et al. 2013, Jaiswal et al. 2015 have focused on the roles of individual SNPs in the promoter region, but validations of their associations with TGW are still needed. Two mutations in the coding sequence have been reported to be associated with TGW: a frame-shift mutation in exon 8 (Du et al. 2016;Yang et al. 2012) and a splice acceptor site mutation in exon 5 (Simmonds et al. 2016). The effects of these coding sequence mutations have been tested in backcross derived isogenic lines, providing sound evidence that TaGW2-A1, indeed, influences TGW in tetraploid and hexaploid wheat.
In the present study, we located TaGW2-A1 within the finely mapped interval of QTgw/Gns.cau-6A. Furthermore, a novel mutation of TaGW2-A1 with a 114-bp deletion in the promoter was isolated from the high-TGW parent Y8679. Deletion of the wild-type transcription start site reduced 1 3 promoter activity and decreased expression of TaGW2-A1, which might, in turn, enhance TGW according to the previous finding that TaGW2-A1 is a negative regulator of grain weight (Du et al. 2016;Simmonds et al. 2016;Yang et al. 2012). In a germplasm survey of more than 800 hexaploid wheat accessions, the 114-bp deletion was detected only in Y8679 and three other cultivars: Jinan 13, Jinghua 9, and Gaoyou 0551 (Fig. 5e). These four cultivars could be useful as donors for higher TGW in wheat breeding programs. Due to the incomplete pedigree information (Supplementary Table S15), no clear relationship was observed in these accessions. Furthermore, this novel allele was not detected in tetraploid and diploid wheat species that were involved in the formation of hexaploid wheat (IWGSC 2014). Together, these data point to the possibility that the novel TaGW2-A1 allele identified here may arise during or after the hexaploidization event that yielded hexaploid wheat.

QTgw/Gns.cau-6A shows possible overdominance effect on spike yield
Heterosis has been a main contributor to yield increase in many cereal crops (Fu et al. 2014). Despite extensive efforts in hybrid wheat breeding, mechanisms of wheat heterosis are largely unknown (Ni et al. 2013). Here, we provide the first example of overdominance effect on spike yield that conferred by a single locus in wheat. Data from a single year trial (Beijing; 2014-15) suggest that the combination of the Y8679 haplotype and the J411 haplotype of QTgw/Gns. cau-6A in the heterozygotes will most likely lead to an overdominance effect on GWS. Using the NILs developed for QTgw/Gns.cau-6A, the consistency of this overdominance effect across environments will be further investigated in the future.
Previously, Bednarek et al. (2012) found that TaGW2 knockdown through RNAi caused no significant alternation in GNS or in the number of spikelets per spike. Similarly, Hong et al. (2014) also found that TaGW2-RNAi had no significant impact on grain number per plant, and Simmonds et al. (2016) found that a G-to-A transition in the splice acceptor site of exon 5, which leads to mis-splicing in TaGW2-A1, had no significant impact on GNS or on the number of spikelets per spike. Thus, we tend to believe that the overdominance effect of QTgw/Gns.cau-6A on GWS was most likely caused by two or more different dominant genes linked in repulsion phase, with TaGW2-A1 for enlarged grain size, while the other for increased grain number. Dissection of QTgw/Gns.cau-6A into two distinct QTLs for TGW and GNS, respectively, may eliminate linkage drag, but it will be a challenging task, since QTgw/Gns.cau-6A is located near the centromere. Considering the low recombination frequency of the pericentromeric region, utilization of the overdominance effect on GWS in the form of heterozygotes seems to be a feasible way to combine higher TGW and higher GNS phenotypes.
Author contribution statement ZN and QS conceived the project; SX developed the RIL population; HZ, ZF, and JL performed phenotyping of the RIL population under 11 environments; XD, YS, HZ, and LS performed cloning and expressional analyses of TaGW2-A1. HZ and ZQ performed marker development of the QTL region of interest; HZ developed the near isogenic lines; HZ, LL, and JL performed phenotyping of the near isogenic lines; HP, ZH, YY, and MX assisted in revising the manuscript; HZ and ZN analyzed experimental results and wrote the manuscript.