Background

Grain number per spike (GNS) is one of three component traits that contribute to the grain yield of wheat (Triticum aestivum L.) [1]. Spikelet number per spike (SNS), which includes the total number of spikelet as well as fertile and sterile spikelet (referred to as TSS, FSS, and SSS, respectively), and grain number of spikelet (GNSL) are the two direct components that show a positive correlation with GNS [2,3,4,5]. Selecting wheat varieties with larger spike styles, containing more spikelets, has proven to be an efficient strategy for increasing yield potential [6, 8,9,10]. Until now, QTLs for GNS and its related traits have been detected through linkage map** in many studies and distributed across nearly all of the 21 chromosomes of wheat [11,12,14]. Stable-expressed QTLs for GNS and TSS, which have shown effectiveness across various environments and genetic backgrounds have been identified on chromosomes 1A, 1B, 2D, 3A, 4B, 5A and 7A [4, 8, 12, 15,16,17,18,19,20].

Several genes related to FSS or GNSL have been reported. For instance, WAPO1, an ortholog of the rice gene APO1, is located on chromosome 7AL and can affect spikelet number [15, 17]. The gene GNI1 encods a HD-Zip I transcription factor and the reduced-function allele of GNI-A1 can increase the GNSL [21]. Additionally, TaMOC1 [22] and TaDEP1 [8]. Subsequent remap** efforts led to the identification of tightly-linked SNPs associated with this locus, utilizing a high-density genetic map based on the Wheat 660 K SNP array. By aligning the SNPs from various studies to the wheat reference genome, we identified an overlap** physical interval.

Previous studies have demonstrated that conditional QTL map** can effectively reflect the genetic effects of complex traits. In wheat, both unconditional and conditional QTL analyses have been conducted to uncover the dynamic genetic factors that influence target traits. For example, Liu et al. [27] conducted a conditional QTL analysis, revealing that the stable QTLs for grain weight were primarily influenced by grain width. Zhang et al. [28] performed unconditional and conditional QTL map**, examining the temporal and spatial expression patterns of the stable QTL related to plant height. These findings indicate that conditional analysis is valuable for investigating the developmental behavior of quantitative traits at the QTL level.

The wheat germplasm “Pubing 3228 (PB3228)” was derived from the F7 progeny of the cross between Triticum aestivum cv. “Fukohokomugi” and Agropyron cristatum “Z559” with 21 pairs of chromosomes. It exhibits superior characteristics in yield-related traits, particularly in spike morphological traits, such as large spike and higher grain number [3, 29]. In the present study, a recombinant inbred line (RIL) population derived from the cross between “PB3228” and “Gao 8901 (G8901)” (PG-RIL) was utilized to achieve the following objectives: (i) construct a high-density genetic linkage map; (ii) identify stable QTLs for GNS and elucidate the genetic contribution of the individual component traits to GNS at the specific QTL level; (iii) dissect the genetic basis of the superior grain number characteristic exhibited by PB3228, and predict potential candidate genes associated with the key QTLs; and (iv) develop user-friendly markers for application in marker-assisted selection (MAS) during the breeding process. The results of this study could potentially provide beneficial QTLs for use in MAS and fine-map**, thereby aiding in the enhancement of yield in wheat breeding and offering further insights into the underlying genetic mechanisms governing superior grain number in wheat.

Results

Phenotypic variation and correlations among traits

The broad-sense heritability (hB2) of the spikelet number traits (67.9% and 59.8% for TSS and FSS, respectively) were higher than grain number-related traits (50.9% and 53.8% for GNS and GNSL, respectively. Table S1). Significant genetic and environmental effects were observed for all the investigated traits (Table S1). These findings underscore the significance of genetic background in explaining the overall phenotypic variation of the four traits, while also indicating that environmental factors exert a greater influence on GNS and GNSL as opposed to TSS and FSS. PB3228 demonstrated significantly higher phenotypic values for all the traits and environments, with the exception of TSS in 2016LC (Table S1). The phenotypic values of the four traits exhibited broad and continuous variation among the 176 RILs, displaying significant transgressive segregation for both the parental lines (Table S1, Fig. S2). This phenomenon could potentially be attributed to the distinct genetic background of the two parents and the complex polygenic inheritance patterns governing these traits. The correlation analysis indicated a significant relationship between GNS and all three examined traits. Specifically, GNSL exhibited a stronger correlation with GNS (r = 0.79, P < 0.001), while TSS (r = 0.55, P < 0.001) and FSS (r = 0.68, P < 0.001) displayed comparatively lower correlations GNS (Table 1). Additionally, GNSL displayed positive correlations with both TSS and FSS. These findings suggest that within the PG-RIL genetic background, GNSL potentially exerts a more significant influence on GNS compared to the influence of TSS/FSS.

Table 1 Correlation coefficients between the investigated traits of the RILs and their parents
Full size table

Genoty** and map construction

The 176 RILs of PG-RIL population along with their parents were genotyped using the SNP probes derived from the Wheat 660 K SNP array. The RILs were determined to have Call Rates ranged from 97.0% to 98.8%, Heterozygote Rate ranged from 2.6% to 7.9%; and all were passed the quality control. Among the 630,517 SNP probes, 115,836 that had Call Rate > 97%, Heterozygote Rate < 20%, and Minor Allele Frequency > 5% were selected for further analysis. Among them, 101,151 that had Conversion Type “Poly High Resolution (PHR)” and showed polymorphism between the two parents PB3228 and G8901 were selected for map construction. The 101,151 functional SNPs were binned to 4477 bins. The representative SNPs of the 4,477 bins along with 45 other markers were used for linkage analysis and map construction.

In total, a high-density genetic map with 101,136 loci spanning 4357.3 cM was released based on 101,091 SNP markers and 45 SSR, kompetitive allele-specific PCR (KASP), or morphological markers (Table 2). Most markers were mapped to the A genome (49.0%) and B genome (41.8%), whereas only 9.2% markers were mapped to D genome. For the map lengths, A, B, and D genomes covered 32.9%, 30.2%, and 36.9% of the total map length, respectively. The chromosome sizes ranged from 134.25 cM (chromosome 6A) to 311.75 cM (chromosome 7D), with an average of 207.49 cM per chromosome. The number of markers on each chromosome ranged from 261 (chromosome 4D) to 10,879 (chromosome 1A), with a mean of 4,816 markers per chromosome. The average distance between adjacent bin markers ranged from 0.67 cM for 3A to 1.64 cM for 6D, with an average of 0.96 cM. The maximum distance between adjacent bin markers ranged from 6.25 cM for 3A to 29.57 cM for 3B. Gaps exceeding 20 cM were present in chromosomes 3A (24.9 cM), 2B (21.98 cM), 3D (28.65 cM), 6B (22.69 cM) and 7B (29.57 cM) (Fig. 1).

Table 2 Summary information of the recombinant inbred line population derived from PB3228/Gao8901 (PG-RIL) high-resolution genetic map
Full size table
Fig. 1
figure 1

The high-density genetic linkage map of recombinant inbred line (RIL) derived from PB3228/Gao8901 (PG-RIL) population. The ruler on the left represents the position of the marker on the chromosome. The markers with red and green color are from the short and long arms, respectively, of the wheat whole genome assembly IWGSC RefSeq v1.0; while the markers with black color have no physical position information

Full size image

Collinearity analysis indicated that the genetic map showed high collinearity with the wheat whole genome assembly IWGSC RefSeq v1.0, with the exception of chromosome 5BS, in which a segment inversion was detected (Fig. 2, Fig. S3). A total of 95.6% of the markers were located on the identical chromosomes between the PG-RIL linkage map and the physical map of IWGSC RefSeq v1.0 (Fig. S3).

Fig. 2
figure 2

Collinearity between the genetic and physical positions (IWGSC RefSeq v1.0) for mapped SNPs in recombinant inbred line (RIL) population derived from PB3228/Gao8901 (PG-RIL) genetic map

Full size image

QTL analysis

A total of 79 QTLs (135 traits × environments) were detected, distributed on 18 chromosomes (Fig. S4, Table S2). The QTLs explained 0.5–33.0% of the phenotypic variation. Five environmentally stable QTLs were identified in at least three environments (Table S2). For GNS, 26 unconditional QTLs were detected (Fig. S4, Table S3). The PB3228- and G8901-derived alleles contributed to the additive effect for 15 and 11 QTLs, respectively. Two environmentally stable QTLs, QGns-1A.1 and QGns-4A.2, were detected (Table 3). The QTL QGns-4A.2 was significant in all the seven environments, with phenotypic variation explained (PVE) ranged from 9.6% to 13.1%; the RILs carrying the PB3228-allele had 3.21–4.39 more grains than that carrying the G8901-allele across the seven environments. The QTL QGns-1A.1 was significant in three environments, with PVE ranged from 2.6% to 14.6%; the PB3228-allele can produce 2.89–7.26 more grains in the three environments (Table 3).

Table 3 Information of important QTLs related to grain number per spike (GNS) and their co-located QTLs for GNS-component traits
Full size table

Ten QTLs were identified for GNSL (Fig. S4, Table S2). The G8901-derived alleles conferred to the QTLs on chromosomes 2A, 3B, 4B and 7D, whereas the PB3228-derived alleles conferred to the other six QTLs on 1A, 3D, 4A, 5A, 5B and 7B. The QTLs QGnsl-3D and QGnsl-4A were detected in both the two investigated environments; and were correlated with QGns-3D.2 and QGns-4A.2, respectively (Table 3). Six more QTLs QGnsl-1A, QGnsl-3B, QGnsl-4B, QGnsl-5A, QGnsl-5B and QGnsl-7B were also co-located with GNS-QTLs.

For the two SNS related traits TSS and FSS, 17 and 13 QTLs were detected, respectively (Fig. S4, Table S2). Four QTLs for TSS (QTss-3D, QTss-4A, QTss-6A.2 and QTss-7B.2) and three for FSS (QFss-3D.1, QFss-6A and QFss-7B.2) were found to be significant in two or more environments (Table 3). Co-located environmentally stable QTLs for TSS and FSS were detected on chromosomes 3D (also co-located with QGns-3D.1), 6A, 6B and 7B (Table S2). The QTL QTss-3D was significant in five environments, with PVE ranged from 10.4% to 21.5%; while QFss-3D.1 was significant in four environments, with PVE 3.4–19.7%. At this locus, the PB3228-derived allele improved TSS and FSS by 0.44–0.69 and 0.36–0.55, respectively. Co-locating of TSS/FSS-QTLs with GNS-QTLs was also found in 11 cases (Table S2).

The correlation coefficients between GNS|GNSL, GNS|TSS, GNS|FSS (conditional traits) and GNSL, TSS, FSS (unconditional traits) were shown in Table S4. A total of 22 conditional QTLs comprising 42 QTL × environments were detected for GNS after conditioned on GNSL, TSS or FSS (Fig. S4, Table S3). Among them, nine loci were also identified in unconditional analysis; five were co-located with unconditional QTLs for GNSL and one for both GNSL and TSS (Table 4). Thirteen QTLs were newly detected, among which five were co-located with unconditional QTLs for TSS or FSS. Different situations of QTL detection were found when using the conditional trait values of GNS conditioned on each of its components (Table S3). After conditioned on GNSL, six QTLs were identified for GNS, among which three were newly detected. Thirteen and twelve QTLs were detected by conditional values based on TSS and FSS, respectively, among which four and six were newly detected.

Table 4 The contribution information of the component traits to grain number per spike (GNS) at single QTL level
Full size table

KASP marker and QTL validation

The SNPs closely linked to the two major QTLs QGns-4A.2 (AX-108966946 (chr4A:679986545) and AX-109857541 (chr4A:681651241)) and QGns-1A.1 (AX-109901702 (chr1A:4072201) and AX-109466381 (chr1A:8088981)) were converted to KASP markers. The markers were used to screen the PG-RIL and a diversity panel comprising of 141 accessions (Fig. S5). The genoty** result of the KASP markers for PG-RIL were almost the same as the Wheat 660 K SNP array; which mutually verified the accuracy of the two genoty** methods.

Two-tailed t-test was conducted based on the genoty** result of the KASP markers for both populations to verify the effects of the QTLs. Significant marker-trait correlations were found for all the four KASP markers. The two markers for QGns-4A.2, AX-108966946 and AX-109857541, were correlated with GNS in all the seven environments (P < 0.001); and with GNSL and TSS/FSS in 20/24 of the trait × environments for PG-RIL (Table S5). They were also correlated with GNS and TSS/FSS in 6/24 of the trait × environments for the diversity panel (Fig. 3, Table S6). As for QGns-1A.1, AX-109901702 and AX-109466381 were correlated with GNS and GNSL in 3/14 and 2/4 of the trait × environments, respectively, but not with TSS/FSS in any environment for PG-RIL (Table S5). However, the two markers were correlated with GNS and TSS/FSS in 8/12 and 4/12 of the trait × environments, respectively, for the diversity panel (Fig. 3, Table S6). In the diversity panel, the PB3228 haplotype showed 1.24–3.39 and 0.62–3.30 more grains than G8901 haplotype for QGns-4A.2 markers (AX-108966946 and AX-109857541) and QGns-1A.1 markers (AX-109901702 and AX-109466381), respectively (Table S6).

Fig. 3
figure 3

Phenotypic performance of different haplotypes for Grain number per spike (GNS), total- and fertile spikelet number per spike (TSS and FSS) in the diversity panel based on the kompetitive allele-specific PCR (KASP) markers AX-109857541 (left) for QGns-4A.2 and AX-109901702 (right) for QGns-1A.1. * and ** represent the phenotype of opposite alleles was significantly different at P < 0.05 and P < 0.01 levels, respectively. Error bar stands for standard deviation

Full size image

Prediction of candidate genes in the two key loci for GNS

In the present study, two important QTLs were detected for GNS. Among these, the QTL QGns-4A.2 was identified in seven environments and the RILs carrying the PB3228-allele had 3.21–4.39 more grains than that carrying the G8901-allele. The physical confidence interval of QGns-4A.2 was 2.6 Mb (679,072,862- 681,651,241 bp, RefSeq v1.0) and contained 28 high-confidence annotated genes (Table S8). Among these genes, TraesCS4A02G407000 showed higher expression level in spike and grain [30] and may be the candidate gene for QGns-4A.2. The QTL QGns-1A.1 was identified in three environments and the PB3228-allele can produce 2.89–7.26 more grains than the G8901-allele. The physical confidence interval of QGns-1A.1 was 2.1 Mb (5,469,646- 7,536,121 bp, RefSeq v1.0) and contained 43 high-confidence annotated genes (Table S9). Among these genes, TraesCS1A02G011300 showed much higher expression level in spike [30] and may be the candidate gene for QGns-1A.1. Further research is needed to firmly conclude the candidate genes.

Discussion

Comparison of the major QTLs with previous observations

One of the objectives of the present study was to identify constitutively-detected major QTLs for GNS and related traits. A total of five stable QTLs that were significant in at least three environments were detected. For GNS, the QTL QGns-1A.1 was significant in three environments, with PVE ranged from 2.6% to 14.6%; the PB3228-allele can produce 2.89–7.26 more grains in the three environments. We located the QTL in the interval 5.5–7.5 Mb of chromosome 1A by BLAST-searching against the Chinese Spring reference genome sequence (RefSeq v1.0). The superior grain number character of wheat germplasm PB3228 was previously studied using a population of 237 F2:3 families from 3228/**g 4839 [29] and a QTL for GNS and floret number per spikelet (FNS) was found in the interval Xpsp3151-Xwmc24 on chromosome 1A. We located the two flanking markers at the positions 21.1 Mb and 27.3 Mb of 1A (RefSeq v1.0). Chen et al. [31] identified a QTL related to GNS near the position 458.4 Mb of 1A in the population derived from Pubing 3504/**g 4839. Heidari et al. [20] detected a QTL for GNS near the position 4.1 Mb of 1A. Yuan et al. [32] found a QTL related to GNS near the position 344.7 Mb of 1A using a double haploid (DH) populations derived from Huapei 3/Yumai 57. Using the same population, Deng et al. [33] identified a QTL related to GNS near the position 159.5 Mb of 1A. Xu et al. [8] detected a QTL for GNS near the position 280.6 Mb of 1A in the population derived from ** physical interval (4A: 680398739–683638403) as our result [37]. The results indicated that QGns-4A.2 detected in the present study was a stable QTL that could be identified in different environments and different trails, which represented a valuable target for marker-assisted selection to enhance GNS in wheat breeding.

Besides, the environmentally stable QTL QTss-3D was significant for TSS in five environments, explaining 10.4–21.5% of the phenotypic variations. The QTL was also detected for FSS in four environments (QFss-3D.1), explaining 3.4–19.7% of the phenotypic variations. The QTL was located in the interval AX-109499958-AX-109516184-AX-108807742-AX-111064903 and mapped to mapped to the physical interval 65.8–124.0 Mb based on the IWGSC RefSeq v1.0. The gene WAG2-3D was reported to be involved in wheat spikelet development and located at the position 100.9 Mb on chromosome 3D [38]. This gene may be one of the candidate genes of QTss-3D/QFss-3D.1 and need further analysis.

How do PB3228 alleles contribute to GNS at individual QTL level?

Wild relatives of wheat are a reservoir of genetic diversity for improving wheat productivity. Introgression from wild relatives have a great potential to broaden the availability of beneficial allelic diversity for wheat improvement in breeding programs. Agropyron cristatum (2n = 4x = 28, PPPP) is a wild relative of common wheat and confers several desirable agronomic traits to wheat, such as high grain number per spike [39]. Unfortunately, many translocation lines have large and complex chromosome segments, leading to questionable value in wheat breeding program. In this study, the wheat germplasm PB3228 was obtained through a cross between the wheat cultivar “Fukuhokomugi” and the tetraploid A. cristatum accession Z559. Accession Z559 possesses several favorable traits, including a superior spike with a high grain number [40]. During the 1990s, accession Z559 was crossed with Fukuhokomugi and line 4844 was chosen based on the significant spike size observed in the progeny lines [40]. The characterization of chromosome addition and substitution lines derived from line 4844 through in situ hybridization, SSR, and gliadin analyses revealed that the increased numbers of florets and kernels within a spike of line 4844 were genetically controlled by gene(s) located on chromosome 6P of A. cristatum [41]. After several rounds of selection, a genetically stable derivative exhibiting elite characteristics was obtained from the descendants of the wheat-A. cristatum chromosome addition line 4844–12 (2n = 44). This derivative, named PB3228, showed exceptional yield traits, including a high grain number and an impressive number of superior florets per spike. We developed a RIL population derived from the cross of PB3228 / Gao 8901 and constructed a high-density genetic linkage map to identify stable QTLs for grain number per spike (GNS) from PB3228, which will contribute favorable QTLs for MAS and fine-map** and facilitate yield improvement in wheat breeding.

Furthermore, we carried out conditional analysis to identify QTLs for GNS based on trait values conditioned on each of its component traits. The results indicated that GNSL and FSS/TSS were both important for GNS; GNSL contributed a little more than FSS/TSS at the genetic background of PG-RIL population, which was consistent with the results of correlation analysis.

The unconditional QTL QGns-1A.1 was identified in three environments and the PB3228-allele can produce 2.89–7.26 more grains than the G8901-allele. The QTL was also found when conditioned on TSS or FSS but failed to be detected when conditioned on GNSL (Table 4). In the same interval, a QTL for GNSL was detected, but none for TSS/FSS. Therefore, the PB3228-derived allele contributed to QGns-1A.1 by improving GNSL. The situation of QGns-5B and QGns-7B.2 were the same as QGns-1A.1. QGns-5A.2 was a little different. A conditional QTL conferred by G8901 was also detected after conditioned on GNSL. Besides the influence of GNSL conferred by PB3228, G8901 may also had some effect on this QTL.

The unconditional QTL QGns-4A.2 was identified in seven environments and the RILs carrying the PB3228-allele had 3.21–4.39 more grains than that carrying the G8901-allele. The QTL was also detected after conditioned on either GNSL (with sharply dropped PVE) or TSS/FSS (with slightly dropped PVE). In the same interval, QTLs for GNSL (QGnsl-4A) and TSS (QTss-4A) were also detected (Table 4 and Table S3). This locus may be controlled by both components, and GNSL had a larger effect than TSS/FSS. The PB3228-derived allele contributed to GNS by improving both GNSL and TSS/FSS. QGns-7D.2 was similar to QGns-4A.2, but no QTL was found for GNSL or TSS/FSS. Considering the PVE reduction for conditional analysis, the PB3228-derived allele had positive effect on GNS by improving GNSL and TSS simultaneously, and TSS had a greater effect than GNSL.

The QTLs QGns-3B.1 and QGns-4B were also detected when GNS was conditioned on TSS or FSS and failed on GNSL, but with negative additive effect (Table 4). QTLs for GNSL with negative additive effect were also identified, indicated that G8901-derived alleles for QGns-3B.1 and QGns-4B have positive effect on GNS by improving GNSL.

The relationships between GNS and its components were also predicted by QTL clustering. The three QTLs QGns-3D.1, QGns-6B.3 and QGns-7B.1 were co-located with TSS/FSS QTLs with positive additive effect. PB3228-derived alleles may contribute to GNS by improving TSS/FSS (Table 4). Besides, there were also three QTLs (QGns-2A, QGns-5A.1 and QGns-6A.2) that were influenced by TSS/FSS and conferred by G8901-derived alleles. QGns-3D.2 was clustered with QTLs for both GNSL and FSS with positive additive effect. The PB3228-derived allele may have positive effect on GNS by improving both GNSL and FSS (Table 4).

Using a high-density genetic linkage map, unconditional and conditional QTL analysis dissected the contribution of GNSL and SNS to GNS at individual QTL level. In PG-RIL, PB3228 contributed to most of the GNS-QTLs by improving either or both of GNSL and SNS. The environmental-stable QTLs such as QGns-4A.2, QGns-1A.1, QGns-3D.1 and QGns-7B.2 were more preferable for further study. The KASP markers for QGns-4A.2 and QGns-1A.1 can be used in MAS for wheat high-yield improvement.

The advantage of SNP markers and high-density genetic map

Over the past two decades, a number of linkage maps of wheat have been reported and successfully applied to identify QTLs for various wheat traits [8, 42,43,44,45,46]. The genetic maps were mainly composed of PCR-based molecular markers such as SSR, EST-SSR, RAPD and AFLP, etc. [2, 47,48,49]. However, most of the linkage maps were with restricted marker density and the mapped QTLs were within relatively large confidence intervals [4, 8,9,10, 50]. However, most of the genetic maps differed greatly in molecular markers; and the PCR-based markers such as SSR were short of sequence information and difficult to be located precisely in the reference genome sequence. Therefore, identifying loci or genes with the same type of marker is a promising way for researchers to share their achievement more efficiently.

The SNP marker with accurate physical position which is generally more abundant, stable and efficient is the best alternate choice. SNP arrays are flexible in terms of sample and data point number customization, which contributes to its high-density scanning and high robust call rates compared to PCR-based markers. Using the high-throughput SNP arrays, consensus genetic maps have been released [46, 51,52,53]. High-density linkage maps based on bi-parental populations were also reported using the SNP arrays, such as the wheat 9 K, 90 K, 55 K and 660 K arrays [11, 37, 54, 55]. The maps were used in QTL map** for GNS, grain size, spike-related and processing quality traits in bread wheat. The information of tightly-linked SNPs facilitated the comparative analysis and candidate gene prediction of the QTLs [37]. The high-density linkage map also allowed the possibility of comparing its collinearity with the physical map. The high collinearity of our genetic map with the physical map indicated no obvious structural variation between PB3228 and Chinese Spring. Sun et al. [56] evaluated seven types of wheat SNP arrays and found that the Wheat 660 K SNP array may be the best choice for genetic analysis in wheat.

Materials and methods

Plant materials

The population used for constructing the map and performing QTL map** consisted of 176 F2:9 RILs, which were derived through single seed descent from a cross between PB3228 and G8901. PB3228 was a germplasm with superior features in spike morphological traits, such as large spike size and superior grain number (Fig. S1) [3, 29]. On the other hand, G8901 was a strong gluten line, with a grain number lower than PB3228, but higher than that of most cultivated varieties. To validate the effects of SNPs linked to major QTLs, a diversity panel consisting of 141 wheat accessions was utilized.

Field trials

During the growing seasons from 2011–2012 (2011LC) to 2016–2017 (2016LC), six field trials were conducted at the Luancheng Agro-ecosystem Station, the Chinese Academy of Sciences. The coordinates of the Station are 37° 53′ 15″ N, 114° 40′ 47″ E, with an elevation of 50 m. During the 2015–2016 (2015SJZ) growing season, one trail was conducted at the greenhouse of the Center for Agricultural Resources Research, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences located at Shijiazhuang. Hereafter, “2011LC”, “2012LC”, “2013LC”, “2014LC”, “2015LC”, “2015SJZ” and “2016LC” represent the seven environments, respectively.

The four trials conducted in the environments of 2013LC, 2014LC, 2015LC, and 2016LC followed a completely randomized design (CRD) with three replications. The three replications were designated as the main plots, while the 176 RILs and their parents served as subplots.

Seeds were planted in early October, and plants were harvested in mid-June once they reached physiological maturity. Each subplot covered an area of 1.5 m2 and contained four rows, with each row measuring 1.5 m in length and spaced 0.25 m apart. Twenty seeds were planted in each row. Adequate watering was provided for all fields, and local standard protocols for field trial management were followed. In the case of 2011LC, 2012LC and 2015SJZ, only a single row measuring 1.5 m in length was planted with 20 seeds for each line.

The diversity panel was cultivated across four environments 2012LC, 2013LC, 2014LC, and 2015LC. The experimental design and phenoty** methods used in these trails were described as in Ma et al. [https://wheat.pw.usda.gov/ggpages/topics/Wheat660_SNP_array_developed_by_CAAS.pdf) at China Golden Marker Corporation (Bei**g, China; http://cgmb.com.cn/). In addition to this, the RILs were screened using 40 SSR markers, two KASP markers, and two markers identifying glutenin subunits (Table S7). A morphological marker for awn length was also used during the construction of genetic map. SNPs that met the criteria of having a Call Rate > 97%, Heterozygote Rate < 20%, and Minor Allele Frequency > 5% were selected for map construction. To eliminate redundant markers, the BIN function in QTL IciMap** V4.1 (http://www.isbreeding.net/) was employed for binning purposes [57]. For each bin, only one marker with the smallest missing rate was retained. The MAP function was used to construct linkage map for the PG-RIL population. SNPs were grouped together with a LOD score of 3.5 and then ordered using the nnTwoOpt function. Rippling function was carried out with a SARF criterion and a window size of 5. The confirmation of the 21 chromosomes and marker order was done based on the physical position of most SNPs in the IWGSC RefSeq v1.0 [26]. The genetic map was drawn using MapChart 2.2 [58].

Data analysis and QTL map**

The data was analyzed using the AOV function in QTL ICIMap** V4.1 to perform Analysis of Variance (ANOVA). The broad-sense heritability (hB2) was determined by calculating VG/VP; where VG represents the genetic variance and VP represents the phenotypic variance. The environments were regarded as replications, and the genotype × environment interaction was considered as the error term [8, 50].

QTL map** was carried out using the constructed linkage map of PG-RIL. The BIP function, based on inclusive composite interval map** (ICIM), in QTL IciMap** v4.1 was utilized for this purpose. Additionally, conditional analysis was conducted to investigate the effects of the component traits on the expression of QTL for GNS. The conditional phenotypic values (y(GNS|GNSL), y(GNS|TSS) and y(GNS|FSS)) represent the net genetic variation of trait values for GNS that is independent of GNSL, TSS or FSS. These values were evaluated using QGAStation 1.0 (http://ibi.zju.edu.cn/software/qga/). The raw data from each environment were organized as follows: the first column indicated the block (replications), the second column represented the genotype (176 RILs), and the subsequent columns contained the trait data. Conditional genetic analysis was performed by considering “Conditional on Final Stage” for traits in each individal environment. Both the observed and conditional phenotypic values were utilized for QTL analysis, with the identified QTLs categorized as unconditional and conditional QTLs, respectively. A LOD threshold of 2.5 was set to declare a significant QTL. In cases where QTLs had overlap** confidence intervals, they were considered equivalent. Finally, all QTLs were named according to Fan et al. [59].

Conversion of SNPs to KASP markers

Conversion of SNPs to KASP markers was conducted for the SNPs closely linked to the two major QTLs QGNS-4A.2 and QGNS-1A.1 (Table S7). These KASP markers were subsequently utilized in the screening of the PG-RIL population and the diversity panel. The PCR were performed in a reaction volume of 6 μl, consisting of 0.0825 μl KASP primer mix, 3 μl 2 × KASP Master Mix (LGC Genomics, UK), and 3 μl genomic DNA at a concentration of 20 ng/μl. The fluorescence emitted during the reaction was detected using the BIO-RAD CFX real-time PCR system, and the obtained results were analyzed with Bio-Rad CFX Manager 3.1 software.