In this study, we analyzed eight plant architecture traits in two environments based on a large-scale phenotypic screen comprised of 306 worldwide wheat accessions, originating from more than 70 countries (Additional file 2: Table S1). The 306 accessions included 179 landraces and 38 traditional cultivars (Additional file 2: Table S1). To investigate the population differentiation of wheat accessions across the globe, we performed population structure analyses. The PCA result revealed that most varieties from Middle East, Europe, and Asia could be distinguished by the first principal component (PC1), with an overall gradient from the Middle East to both Europe and Asia (Additional file 1: Fig. S1a), which is consistent with the history of wheat dispersal [15, 16]. The neighbor-joining tree suggested that the 306 worldwide accessions could classify into some clades, which were associated with their geographical distribution (Additional file 1: Fig. S1b). For instance, Asian accessions and some accessions from Middle East and Africa grouped into the same clade (Additional file 1: Fig. S1b). Most European accessions clustered into one single clade (Additional file 1: Fig. S1b). The eight traits analyzed were the lengths of each of the shortest tiller length (STL), the longest tiller length (LTL), the main shoot length (ML), the peduncle length (first internode below spike) (PL), the second internode length (SIL), the third internode length (TIL), the fourth internode length (FIL) and the difference in the length between the shortest and the longest tiller length (LSTL) (Fig. 1a).

Phenotypic analyses of wheat accessions for eight plant architecture traits. a Overview of the eight traits in wheat. The shortest and longest tillers were selected from fertile tillers with spikes. The length of the main shoot as the length of the strongest tiller. The difference between the length of the shortest and longest tillers (longest-shortest tiller) was calculated for each plant. b Distribution of phenotypic values and broad sense heritability of the eight plant architecture traits. The broad sense heritability was estimated from the repeatability between raw phenotypes. c Pearson’s correlation coefficients were calculated using the phenotypic values for the 306 worldwide wheat accessions

All eight traits displayed extensive variation across the 306 wheat accessions (Fig. 1b). These eight traits exhibited close association and obvious differences between the two environments (Additional file 2: Table S2). We assessed broad sense heritability by calculating repeatability between raw phenotypes (Fig. 1b). Most of the traits demonstrated relatively high heritability (0.71–0.88), while the difference in length between the longest and the shortest tiller exhibited a lower heritability (0.27) (Fig. 1b).

The high-resolution dissection of traits related to plant architecture captured several novel relationships (Fig. 1c). For instance, the lengths of the longest tiller, the shortest tiller, and the main shoot were strongly and positively correlated (R ≥ 0.76). This observation suggests the consistency of the longest tiller, the shortest tiller, and the main shoot within individual plants. Peduncle length was strongly and positively associated with the length of the longest tiller (R = 0.79) and the shortest tiller (R = 0.76), but showed a relatively weak correlation with the length of the other three internodes (R = 0.50–0.54). However, the length of the third internode strongly correlated with the length of the fourth internode (R = 0.82). The main shoot is the strongest tiller, which is the same to the longest tiller in some cases. This may partially explain the strong correlation (R = 0.79) between the length of main shoot and longest tiller.

In this study, we used genotypic data of wheat accessions from the whole-genome genetic variation map of wheat (VMap [7, 14]). The latest version of VMap (VMap 2.0) [17] consists of 1062 wheat accessions with multiple ploidy levels, from which we selected 306 hexaploid wheat accessions worldwide for this study. The high coverage whole-genome sequencing (~ 10 ×) enabled the identification of 40,710,923 filtered SNPs with minor allele frequency (MAF) > 0.05 across the 306 wheat accessions. Based on these 40,710,923 SNPs, we performed GWAS for the phenotypic values of the eight plant architecture traits in each of the two environments (Y1, 2) and their BLUE values and identified 56,096 (Y1:20,877; Y2:23,315; BLUE:11,904) significant marker–trait associations (− Log10 (P-value) > 5.0) (Fig. 2a, Additional file 2: Table S3). We used linkage disequilibrium (LD) and connections between markers to delineate about 330 significantly associated loci, when at least five nearby SNPs were above the significance threshold, associated with at least one of the eight plant architecture traits (Additional file 2: Table S4). Of the 330 loci, 83 were associated with a single trait, while the remaining 247 showed pleotropic effects on more than one plant architecture traits (Additional file 2: Table S4). Notably, we observed that some significant loci were specially associated with the length of the four internodes, suggesting the relative independence of the genetic control of internode length in this study.

GWAS and network analysis of tiller height across a panel of wheat accessions. a Manhattan plots showing the SNP marker-trait associations for the length of the peduncle, the second internode, the third internode, and the fourth internodes. Orange dots indicate SNPs above the significance threshold (− Log10[P-value] = 5.0). b Association networks across different traits in wheat. The nodes represent plant architecture traits and their associated loci. Longest tiller length, LTL; shortest tiller length, STL; length difference between longest and shortest tiller; longest tiller-shortest tiller length LSTL; main shoot length, MSL; peduncle length, PL; second internode length, SIL; third internode length, TIL; fourth internode length, FIL. The eight traits are indicated by different colors. The edges between the loci from different traits are linked by their LD. Only the edges with an average LD ≥ 0.5 are displayed. The orange solid circles indicate the four loci that are specifically associated with the length of the two internodes. The overlapping loci covering Rht-D1, TaPIN1-6D, Ppd-D1, and TaTB1-4D are indicated by the orange dashed circles. c Distribution of XP-CLR scores (Chinese landraces versus cultivars) for 21 wheat chromosomes. The selected regions and candidate genes detected based on their p ratio are shown. The genome-wide threshold was defined by the top 5% of values

Pleiotropy and LD play important roles in validating phenotypic correlations [18]. Of the 247 loci with pleotropic effects, 182 were associated with more than four traits, and 64 loci had associations with two or three traits (Fig. 2b, Additional file 2: Table S4). These results suggest that these traits might be genetically co-regulated. We took TEOSINTE BRANCHED1 (TB1) as a control, because recent work reported that TB1 regulates height and stem internode length in bread wheat [19]. The genomic region (Chr4D: 15,938,883-19,054,796) including TB1 was associated with the length of second internode, third internode, shortest tiller, longest tiller, and main shoot (Additional file 2: Table S5), suggesting its potential connection with internode length, which is consistent with previous finding [19]. Notably, the major loci for the length of the main shoot (Y1-MSL-1A-1; 2,891 associated SNPs; B-MSL-1A-1:391 associated SNPs), the longest tiller (Y1-LTL-1A-1; 6,079 associated SNPs; B-LTL-1A-1: 27 associated SNPs), the shortest tiller (Y1-STL-1A-1: 1002 associated SNPs; B-STL-1A-1: 5 associated SNPSs), and the peduncle (Y1-PL-1A-1: 5,768 associated SNPs; B-PL-1A-1: 396 associated SNPS; Y2-PL-1A-1: 4 associated SNPs) located in chromosome 1A: 45,791,400-49,172,693 (Additional file 2: Table S4). Most loci for the length of the longest and the shortest tillers overlapped, thus explaining the strong phenotypic correlations between peduncle length and tiller length as well as the consistency of the length among the tillers of individual plants.

To confirm the selected loci in the wheat genome during the gradual improvement of grain yield in China, we selected 59 Chinese wheat accessions (17 cultivars versus 42 landraces) from the 306 wheat accessions for further analysis. These 59 Chinese accessions were selected to display the genomic selection during Chinese wheat breeding process, which will be further used to examine the selection of the associated peaks identified by GWAS. The 59 Chinese accessions are important varieties during Chinese wheat breeding process. We combined the results of whole-genome differentiation of the cross-population composite likelihood ratio (XP-CLR) (Fig. 2c, Additional file 2: Table S6). In total, we determined that 49% of the detected loci (163 of 330) during this study were significantly selected in wheat improvement for higher yield (Fig. 2c, Additional file 2: Table S4). The loci associated with the length of the main shoot, the second internode length, and the longest and shortest tiller in this study (e.g., Y1-MSL-2A-7, Y2-MSL-6B-1, Y2-SIL-2A-1, Y1-SIL-2A-2,, Y2-STL-2A-1, and Y1-LTL-2A-1) as well as known plant height genes (e.g., Rht-D1, and Rht8) appeared to have experienced a strong selective sweep (Additional file 2: Tables S4 and S6). Moreover, the loci that are specifically associated with the length of the peduncle, and the third and fourth internodes (Y1-PL-1A-1 B-PL-1A-1, chr1A:45791400-49094709; Y1-TIL-3D-1, chr3D:35786271-39382820; Y1-FIL-2A-3, chr2A:613558313-619417320) were among the most selected genomic regions (Fig. 2c, Additional file 2: Table S4). Strong selection for these loci reflected the modifications of plant height traits over the past several decades in wheat breeding. In addition, several candidate genes (e.g., TaPIN1-6D, Ppd-D1, SEP1/2-4A, and Rht-D1) associated with spike development, and grain number and size were under strong selection (Additional file 2: Table S6). This result reflected the improvement of adaptation to local environments and grain yield in wheat breeding.

The internodes within individual stems play different roles in determining grain yield in crops. To search for loci that can separately regulate the length of the four internodes and determine the genomic basis of geographical differentiation and breeding selection for these traits, we examined the distribution of internode length as a function of the geographical provenance of each accession. We then examined the proportion of each haplotype in all geographical regions based on the four major loci specifically associated with the length of each of the four internodes (Fig. 2a, Additional file 2: Table S1) in the 306 worldwide wheat accessions. In addition, we characterized the extent and direction of the changes of haplotype composition in the 831 wheat accessions (most of the 831 Chinese accessions are modern cultivars) that were introduced from other countries or developed in China since 1900 (Additional file 2: Table S7). The genetic control of agronomical traits might be best understood by comparing a progenitor organism with its derivatives. Genetic crosses between progenitor and derivatives would identify the genetic factors that accounted for their different phenotypes. Thus, we examined the haplotype distribution in the wheat accessions in the two pedigrees between the 1920s and 1970s in China.

Peduncle (the first internode below spike) length showed a distinct distribution pattern across the seven continents/regions (Fig. 3a, Additional file 2: Table S8). European accessions had the longest peduncles (52.88 cm), whereas Asian accessions had the shortest peduncles (46.83 cm) (Fig. 3a, Additional file 2: Table S8). The major locus (Y1-PL-1A-1(5768 associated SNPs); B-PL-1A-1(396 associated SNPs); Y2-PL-1A-1 (four associated SNPs) associated with peduncle length mapped to chromosome 1A:45,791,400-49,094,709. We identified three haplotypes of this locus (peduncle length-long, PL-L; peduncle length-medium, PL-M; peduncle length-short, PL-S) in the 306 worldwide wheat accessions: HapPL−S = 49.27 cm (48.37% of accessions), HapPL−M = 48.83 cm (20.26% of accessions), HapPL−L = 53.84 cm (31.37% of accessions) (Fig. 3b, c, Additional file 2: Tables S9 and S10). European accessions exhibited a higher proportion of HapPL−L, whereas HapPL−S and HapPL−M was more heavily represented in Asian accessions (Fig. 3b, c, Additional file 2: Table S10). In addition, within each of the three haplotypes, the European accessions always had a longer peduncle than African accessions (Additional file 2: Table S10). These results revealed the genetic basis of the differences in peduncle length between geographical areas.

Geographical distribution and breeding selection of the haplotype blocks associated with peduncle length on chromosome 1A. a Peduncle length across wheat accessions originating from the seven continents/regions. Data are means ± standard deviation (SD, n = 5). Significant differences were determined by ANOVA. Different lowercase letters indicate significant differences (P