Phenotypic expression of spike-branching in the mapping population
Spike morphologies of the parents of the mapping population, i.e., cv. Bellaroi and TRI 19165 (‘Miracle wheat’), are shown in Fig. 1a, c, respectively. The sessile spikelets in a standard spike of Bellaroi are attached directly onto the rachis node in a distichous arrangement (Fig. 1a, b). However, most of the bottom nodes in ‘Miracle wheat’ spikes do not carry sessile spikelets but instead form ‘mini-spikes’ like branches carrying sessile spikelets by their own (Fig. 1c, d). Previously, we have identified and characterized the ‘Miracle wheat’ allele (bht-A1) from spike-branching tetraploid wheat (Poursarebani et al. 2015; Wolde et al. 2019a). Because spike-branching appeared to be a quantitative trait, we conducted the current study to identify other loci controlling spike-branching in tetraploid wheat.
As expected, the RILs used in this study generally showed three different types of spike phenotypes (Fig. 2). The first one was the appearance of SS (Fig. 2a, b). The second type was the appearance of ‘mini-spikes’ from the bottom half of the spike resembling the ‘Miracle wheat’ phenotype (Fig. 2c, d). The third type was the extension of the rachilla, and/or partial conversion of the rachilla to rachis leading to the appearance of alternated spikelets and florets (Fig. 2e, f). However, out of the 146-mapping population only one RIL, i.e., 7769-4-78, showed the rachilla extension phenotype. Although the three phenotypic classes were detected, all led to the appearance of additional spikelets per spike. Therefore, the number of supernumerary spikelets (SS) per spike, i.e., additional spikelets per spike (addSPS), was used for mapping spike-branching. The phenotypic data are summarized in Supplementary Tables 1 and 2.
Genome-wide identification of loci controlling spike-branching in tetraploid wheat
To identify loci controlling spike-branching, we mapped SS formation per spike (Fig. 2a, c, e). Following composite interval mapping (CIM) analysis, three QTL, i.e., QSS.ipk-1AS, QSS.ipk-2AS, and QSS.ipk-2BS, were identified as controlling spike-branching in this mapping population (Fig. 3a). The QTL on chromosome group 2 consistently appeared in all four environments, i.e., IPK14, IPK15, HAL15, and GH15. Since the ‘Miracle wheat’ allele (bht-A1) had already been identified from chromosome 2A short arm (Poursarebani et al. 2015), we genotyped the RILs using diagnostic CAPS markers derived from the bht-A1 allele, as well as from the homoeo-allele from chromosome 2B, i.e., TtBH-B1. Then, we incorporated the two new CAPS markers to the GBS markers and re-run the mapping analysis. Our results showed that the two QTL identified earlier, i.e., QSS.ipk-2AS and QSS.ipk-2BS, were closely linked to the corresponding CAPS markers (Fig. 3c, e). Based on our previous results (Poursarebani et al. 2015) and the complete linkage of QSS.ipk-2AS with the bht-A1 CAPS marker (Fig. 3c), we concluded that QSS.ipk-2AS was indeed the bht-A1 allele. Tight linkage of the TtBH-B1-derived CAPS marker with QSS.ipk-2BS also suggests that TtBH-B1 or a very closely linked gene was responsible for the phenotypic variance at QSS.ipk-2BS, further implying that QSS.ipk-2AS and QSS.ipk-2BS are likely to be the homoeoloci, bht-A1 and bht-B1, respectively. The QTL on chromosome 1AS, QSS.ipk-1AS, had a significant effect on the spike-branching phenotype in one environment (HAL15); but remained under the significance level in the other three environments. We, therefore, decided to term this new locus as bht-A2. The phenotypic variance explained by these three QTL is shown in Supplementary Table 3.
Furthermore, we performed a GWAS analysis based on 302 tetraploid wheat including 27 ‘Miracle wheat’ accessions. First, we looked at the diversity of different wheat species within the dataset which enabled us to carefully analyze population structure. The principal component analysis (PCA) revealed that the spike-branching wheats are distinctly grouped from all other accessions (Supplementary Figure 1). Despite the small number of spike-branching accessions in the data set, the distinct position of spike-branching accessions within the data set allowed us to conduct GWAS using the linear model with kinship information.
The GWAS analysis revealed two QTL on chromosome 2A and 2B (Fig. 3g). However, the slight deviation of the lead marker (black dot) from the bht-B1 derived CAPS marker (green) could partly be due to the linkage disequilibrium (LD) of the lead marker as a result of different factors including population structure and genetic linkage. Taken together, the biparental QTL mapping and GWAS analysis suggest that the allelic variation close to bht-B1 is highly linked with QSS.ipk-2BS.
To further test the effect of the different homoeo-allelic combinations on spike-branching, we divided the F7-RILs into nine genotypic groups using the CAPS markers derived from bht-A1 and bht-B1. The genotypes were designated as AABB (those RILs carrying both homoeo-alleles from Bellaroi), AABb (those RILs carrying the A copy from Bellaroi, but heterozygous for the B copy), AaBB (those RILs heterozygous for the A copy, but carrying the B copy from Bellaroi), AaBb (those RILs heterozygous for both copies), Aabb (those RILs heterozygous for the A copy, but carrying the B copy from TRI 19165), aaBB (those RILs carrying the A copy from TRI 19165, and the B copy from Bellaroi), aaBb (those RILs carrying the A copy from TRI 19165, but heterozygous for the B copy), AAbb (those RILs carrying the A copy from Bellaroi and the B copy from TRI 19165), and aabb (those RILs carrying both homoeo-alleles from TRI 19165). The summary of the phenotypic comparison among the nine groups across the four different environments is shown in Fig. 4. Due to the functional allele from Bellaroi for the 2AS QTL (AA), the first four groups of RILs, i.e., AABB, AAbb, AABb, and AaBb, did not show any form of the spike-branching phenotype across all environments. This suggests that the wild type allele from Bellaroi was sufficient for maintaining the canonical spike form. On the other hand, RILs carrying both homoeo-alleles from ‘Miracle wheat,’ i.e., aabb, showed more SS and/or genuine spike-branching as compared to those RILs carrying only the recessive A allele (aaBB) that only showed SS formation (Fig. 4, boxed). This also suggests a branch modifying role of the BB genome allele from TRI 19165, i.e., bht-B1. Because the suppressive effect of the wild-type (i.e., the BB allele from Bellaroi) did not completely abolish SS formation (Fig. 4), the wild-type allele from the AA genome plays a major role in maintaining the canonical spike form in tetraploid wheat (Poursarebani et al. 2015). Genome-wide epistatic interaction mapping also revealed a significant additive-by-additive epistatic effect between bht-A1 and bht-B1 (Supplementary Table 4).
Next, we plotted the two-dimensional graphs of the phenotypic response (the reaction norm) of RILs with genotype AABB, AAbb, aaBB, and aabb across three different environments (Fig. 5). Besides the suppression effect of an allele from Bellaroi (BB) on the SS and/or spike branch formation, clearer epistatic interaction was also deduced from the differences observed in the slopes of the lines connecting each group (orange vs blue). Taken together, these results indicate that phenotypic variation for spike-branching in the RILs was mainly controlled by the genetic loci containing the homoeo-alleles bht-A1 and bht-B1.
Effects of QSS.ipk-2A, QSS.ipk-2B, and Rht-B1b on spikelet fertility, grain number, and grain weight
Compared to tall wheat varieties, semi-dwarfing of the modern wheat varieties with Reduced height (Rht) genes partition more dry matter to the developing spike, resulting in increased grain number per spike (Brooking and Kirby 1981; Flintham et al. 1997; Miralles et al. 1998; Youssefian et al. 1992). Bellaroi is a semi-dwarf modern durum wheat variety with a reduced height gene, i.e., Rht-B1b allele. The semi-dwarf RILs were selected based on the plant height QTL that was mapped to a region harboring Rht-B1b (Supplementary Figure 2). Thus, we compared 13 semi-dwarf RILs carrying QSS.ipk-2A (bht-A1) and QSS.ipk-2B (bht-B1) to those without positive alleles at QSS.ipk-2A and QSS.ipk-2B (n = 16) to better analyze the effect of the Rht-B1b allele on spikelet fertility and grain number as the spikelet number increases. Despite increased spikelet number due to SS formation and/or mini branches, which resulted in a significantly higher number of grains per spike (Fig. 6b), spikelet fertility or grain number per spikelet was decreased (Fig. 6e). This suggests that increasing sink or spike size alone is not sufficient without solving the problem of spikelet infertility, which is also termed as floret abortion (Sakuma and Schnurbusch 2020). Importantly, we did not find significant differences in kernel weight between the two groups in all three locations (Fig. 6f), which suggests the possibility of increasing wheat yields by increasing spikelet number without significantly affecting the required kernel weight in a semi-dwarfed background.
Both homoeo-genes were expressed during early stages of spike development
To complement mapping results, transcript levels of TtBH-A1 and TtBH-B1 were measured at three spike developmental stages using four RILs selected from the mapping population based on different allelic combinations of the homoeo-alleles (See Materials and Methods). Consistent with the observed QTL effects (Fig. 3a) and group phenotypic effects of the RILs (Fig. 4), our expression analysis also indicated that TtBH-A1 was the higher expressed homoeo-allele, followed by TtBH-B1 (Fig. 7). Taken together, these results further demonstrated that TtBH-A1 is the gene majorly controlling SM identity and maintenance of the canonical spike form in tetraploid wheat.
Sequence analysis of homoeologous TtBH-A1 and TtBH-B1 genes
To investigate the allelic variation between the two homoeo-genes, we sequenced the homoeologous genes from 116 tetraploid wheat species (T. turgidum and T. durum; TtBH-A1, TtBH-B1). The AA copy, i.e., TtBH-A1, of the gene from these species encodes for 299 amino acids, while the BB copy, TtBH-B1, encodes for 307 amino acids. Both homoeo-genes have a highly conserved AP2/ERF DNA-binding region as well as a conserved motif in the C-terminus region (Fig. 3b, d). Mutations in the conserved regions led to SS formation and/or spike-branching (Dobrovolskaya et al. 2015; Poursarebani et al. 2015). Similarly, the bht-A1 or the ‘Miracle wheat’ allele (Fig. 3b and Table 2, tHAP_6) arose due to a non-synonymous substitution (T287C; L96P) in the AP2/ERF coding region of TtBH-A1 (Poursarebani et al. 2015). From the sequenced 116 tetraploid wheat species (T. turgidum and T. durum), we did not find any other causative mutation in TtBH-A1 other than the T287C substitution (Table 2). In TtBH-B1 (taking Chinese Spring as the reference genome), three non-synonymous substitutions A491G, A619T, and T785C were identified. Interestingly, all of the spike-branching accessions carry the T785C substitution, suggesting that T785C is most likely linked with an increased phenotypic expression of SS formation and/or spike-branching in RILs, which have combined both alleles, i.e., aabb, from TRI19165 (Fig. 4). Of course, the causative phenotypic effect of the mutation from TtBH-B1 cannot be easily deduced because of the masking effect, especially when the TtBH-A1 is functional. This was exactly the case in the tHAP_5 haplotype, where the spike was normal and did not show any SS, even though the TtBH-B1 protein is non-functional as a result of early stop codon due to frameshift mutation incurred as a result of six base pair deletion (Table 2). This result further confirms that TtBH-A1 is sufficient to maintain the canonical spike architecture regardless of the allelic status at TtBH-B1. Hence, all accessions carrying the tHAP_1, 2, 3, and 5 showed a non-branching spike due to a functional TtBH-A1 allele (Table 2, Fig. 4). Bellaroi, which was used as a non-branching parent for the mapping population, was grouped in tHAP_3, while TRI 19165 was grouped in tHAP_6.
The single accession in tHAP_7, PI 67339 carries a non-synonymous substitution outside of the AP2/ERF region (A19G) of TtBH-A1 and shows the sham ramification (Shr) phenotype. Interestingly, the locus, Shr2, controlling the sham ramification phenotype from this accession has been mapped on the long arm of chromosome 2A (Amagai et al. 2017). The sham ramification phenotype was also seen in tHAP_4 carrying the functional TtBH-A1 allele. This indicates that the A19G substitution in TtBH-A1 was not the causative mutation for the observed sham ramification phenotype in these two accessions.
Insertion of ‘Miniature Inverted-repeat Transposable Elements’ (MITE) near TtBH-B1
To further investigate allelic variation close to the two homoeo-genes, we performed sequence analysis in the putative promoter regions of TtBH-A1 (1.3 kb) and TtBH-B1 (2.1 kb) from 39 canonical and 28 spike-branching accessions. The results are summarized in Fig. 8. For TtBH-A1, the only sequence variation identified was the SNP located at position -387 (A/G) bp from the start codon (Fig. 8a). From the sequenced accessions (total n = 67), only six non-spike-branching accessions carry the adenine (A), while the remaining accessions, including all spike-branching mutants, carry the guanine (G). This result clearly shows that the putative promoter SNP is not linked with the appearance of spike-branching in ‘Miracle wheats.’
For TtBH-B1, all the sequenced accessions carry the MITE at position − 1079 to − 777 bp (Fig. 8b). Furthermore, we also detected several sequence variations, including insertion and deletions (indels), within TtBH-B1 (Supplementary File 1). Interestingly, the SNP located at position − 54 (G), located in the putative 5′-UTR region (Fig. 3d), was present in all the spike-branching tetraploid accessions (n = 49). From 48 canonical accessions, 14 of them also carry the G, suggesting that the SNP is not directly linked with spike-branching phenotype. However, it is important to note that all the 14 accessions carry the functional TtBH-A1 allele that can easily mask the effect of the mutation. We thus hypothesize that the inserted MITE and/or the − 54G close to TtBH-B1 might have contributed to its lowered gene expression in the spike (Fig. 7), or differential tissue expression patterns such as between root and spike. For example, in hexaploid wheat, TaBH-B1 is also expressed in roots where in some cases the expression exceeds that of TaBH-A1 (Supplementary Figure 3).
Alleles at the Q locus do not directly contribute to spike-branching in tetraploid wheat
Almost all ‘Miracle wheats’ carry the qdel-5A allele, making them double mutants for the two important genes involved in the SM identity (bht-A1) and SM determinacy (qdel-5A) (Poursarebani et al. 2015; Wolde et al. 2019b). Since the Q protein is involved in spike development through rachis and rachilla morphogenesis, we hypothesized that the qdel-5A allele might also be involved in spike-branching. We therefore re-mapped spike-branching after phenotyping the same mapping population (i.e., F9 RILs) grown in the greenhouse. This time, phenotyping was performed by using a qualitative scale from 0 (non-branching) to 3 (spike-branching with mini-spike branches), thereby trying to quantify genuine spike-branching instead of additional spikelet per spike alone. However, spike-branching was not mapped to the region harboring qdel-5A allele.
Notably, previous studies also failed to map spike-branching to this chromosomal arm (Dobrovolskaya et al. 2009; Echeverry-Solarte et al. 2014), indicating that the Q protein might not directly contribute to spike-branching. However, consistent with previous studies (Simons et al. 2006), spike or rachis length was mapped to chromosome 5AL at the q locus, thereby linking spike morphometric traits to the qdel-5A allele (Wolde et al. 2019b). We further looked at whether an epistatic interaction exists between bht-A1 and qdel-5A. Consistent with our mapping result, we did not detect an epistatic interaction between bht-A1 and qdel-5A for spikelet number per spike (Fig. 9a–c). This was not surprising as Q does not affect spikelet number but rather floret number per spikelet via rachilla elongation (Debernardi et al. 2017; Greenwood et al. 2017). Interestingly, we detected an epistatic interaction between bht-A1 and qdel-5A for the spike (rachis) length (Fig. 9d–f). This might suggest an indirect effect of the qdel-5A allele on the morphogenesis of the ‘mini-spikes’-like branches of ‘Miracle wheat’ regardless of the spikelet number on the ‘mini-spikes.’