The phenotype of de novo synthesized hexaploid wheat, its allotriploid form and its progenitors

The F1 plants (allotriploid; 3x = 21, ABD) were generated from a wide cross between either T. turgidum (2n = 4x = 28, AABB) ssp. turgidum accession AS2255 or ssp. durum LDN (cv. Langdon) and A. tauschii accession AS60 (2n = 2x = 14, DD). The chromosomes in the allotriploid metaphase I meiocytes typically formed 21 univalents (Fig. 1a and Additional file 1: Figure S1a), but a significant number of immature microspores formed a dyad, diagnostic of the presence of an unreduced gamete (Fig. 1b and Additional file 1: Figure S1b). Zygotes derived from the union of two unreduced gametes represent a WGD event, producing an allohexaploid (2n = 6x = 42, AABBDD) embryo (Fig. 1c, d and Additional file 1: Figure S1c, d).

The allotriploid plants were less vigorous than their T. turgidum progenitor (Fig. 1e and Additional file 1: Figure S1e). Their leaf length and width were intermediate between the two parents’ during the early stage of the plants’ development (Fig. 1f). In contrast the leaf length of the allohexaploid plants in the first generation (S1) was greater than their T. turgidum progenitor’s in both combinations (Fig. 1f and Additional file 1: Figure S1f; Student’s T-test, P < 0.01). As allotriploid and allohexaploid S1 plants were most distinguishable at this stage (Fig. 1e) and there were no noticeable differences in morphology among S1-S4 euploid plants (Additional file 1: Figure S1e), we collected leaves at this stage from S1 plants to isolate RNA for sequencing. For all S1 plants used, root tip cells were checked to make sure they contained a complete set of chromosomes without structure variation.

Homoeologue discrimination and transcript abundance in the allotriploid, allohexaploid and parental plants

RNA-Seq analysis was applied to RNA extracted from allotriploid, allohexaploid (S1) and the progenitor plants (Additional file 2: Dataset S1). The number of clean reads ranged from 46.7 to 89.4 million, and a mean of 77.3% of these were mappable to the cv. Chinese Spring (CS) draft genome sequence. After a stringent filtering step using the method by Pfeifer et al. [33], an average of 36% of the mapped reads could be allocated to one of the homoeologues. The transcript abundance of each homoeologue was then quantified using HTSeq-count [34], based on the high confidence gene (HC1-HC4) dataset [32], comprising 99,386 gene models. Counts were expressed as fragments per kilobase of exon model per million mapped base pairs (FPKM) [35]. Only homoeologues/genes showing an FPKM greater than unity in at least one sample within a lineage were retained. Abundances of biological duplicates proved to be strongly correlated (R² = 0.87–0.96 in the AS2255 × AS60 combination, in Additional file 1: Figure S2a). A correlation dendrogram showed that the allotriploid was more closely related to the allohexaploid plant than to its tetraploid progenitor, with the diploid progenitors appearing as outliers (Additional file 1: Figure S2b, c), as would be expected from the various plants’ genome constitutions.

Although the transcripts of the two T. turgidum accessions mapped for the most part to the A and B genome chromosomes, a small number were associated with a D genome location; similarly, a few of the A. tauschii transcripts apparently mapped to an A or B genome chromosome location (Additional file 1: Figure S2d, e). These locations are likely artefacts, deriving from polymorphism between the parental genome sequences and corresponding subgenome sequences of CS. These reads were discarded. The median relative abundance of the ~20,000 genes for which a read of each homoeologue was obtained was close to unity in both hybridization lineages, while the peak value for FPKM was >100 (Additional file 1: Figure S2f). Applying a threshold of FPKM >1, 15,418 (AS60) and 37,321 (LDN × AS60 allohexaploid) genes were identified for each species (Additional file 1: Table S1). Based on the criteria 90% sequence identity and 90% alignment, the CS gene models could be classified as 8,339 triplets (each of the three homoeologues represented), 8,338 duplets (two of the three homoeologues represented) and 47,622 singletons (only one homoeologue represented). An average of 60% triplets, 50% duplets and 30% singletons were identified in the transcriptomes of hybridization combination (Additional file 1: Table S2). For the most part the two hybrid lineages behaved similarly. What follows relates mostly to the AS2255 × AS60 combination, unless indicated otherwise. Although the identity of the T. turgidum parent had little influence over the pattern of homoeologue transcription, there were some differences in the estimated overall gene number. Data relating to the LDN × AS60 combination are given in Additional file 3: Dataset S2, Additional file 4: Dataset S3, Additional file 5: Dataset S4 and Additional file 1: Figures S3–S5 .

WGD restored homoeologue transcript abundances disturbed by hybridization

In all, 32.6% of the set of D-subgenome homoeologues (5,162 out of 15,837) in the allotriploid plants compared to diploid parents AS60 of AS2255 × AS60 combination were down-regulated. The equivalent frequency for the A and B genome homoeologues compared to tetraploid parents AS2255 was, respectively, only 1.9% and 1.7% (Fig. 2a; Additional file 6: Dataset S5). The frequency of up-regulated homoeologues was small for all three genomes (1.1% for A, 1.7% for B and 1.9% for D). An even smaller proportion (0.7% for each genome) was altered in the move from ABD to AABBDD (Fig. 2a; Additional file 7: Dataset S6). However, 3,787 D-subgenome homoeologues appeared to be down-regulated when the contrast was made between the allohexaploid and AS60, a number which was ~25% less than in the contrast between the allotriploid and AS60. The proportion of A and B genome homoeologues down-regulated in the allohexaploid was also lower (0.4% vs ~1.8% in the allotriploid vs AS60 contrast). The implication is that WGD effectively reduced the number of differentially transcribed homoeologues, probably by minor changes but non-significant difference on statistical analyzing using FDR-adjusted p-value below 0.05 that potentially cause inflated comparisons of expression level differences. In order to confirm this suggestion we analyzed theses differential expressed genes only appeared between F1 and parents by boxplot analysis using R packages. This corrective effect was supported by the contrasting effects of hybridization and WGD on homoeologue transcript abundances: for example, 1,946 (38%) D-subgenome homoeologues were significantly down-regulated in the allotriploid plants (Fig. 2b, panel 3, right side), and their transcript abundances were marginally increased in allohexaploid (S1-D), as demonstrated by the upward shift in the median abundance. A similar contrasting effect was observed for homoeologues up-regulated in the allotriploid plants, even for the A and B genome homoeologues. At the same time, 3,176 D genome homoeologues remained down-regulated in the allohexaploid (Fig. 2b, panel 4). The large number of genes for which the level of transcription was corrected in this manner suggests that this is an important feature of the polyploid wheat transcriptome.

WGD corrected the non-additive down-regulation of genes induced by hybridization

Based on the CS gene models, it was possible to determine the transcript abundance of each gene in the parental lines and their allotriploid and allohexaploid derivatives (Additional file 8: Dataset S7). When mid-parent values were compared with those recorded in the allotriploid and allohexaploid plants, non-additivity was assignable to 354 genes (3.6%) of the genes in the allotriploids, the abundance of most of which (318/354) was below the mid-parent value (Fig. 3a). GO analysis showed that genes related to cellular component terms “plastid” and “thylakoid” were well represented (Fig. 3b). The D-subgenome homoeologues were especially affected: 243 (68.6%) were less abundant in the allotriploids than in AS60 (Fig. 3c). However, the transcript abundance of only 40 of these genes remained below the mid-parent value in the allohexaploid plants.

ELD was frequent in both the allotriploid and allohexaploid plants

There was a five fold greater number of ELD-ab genes (those showing a similar level of transcription in AS2255 and the allotriploid) than ELD-d genes (showing a similar level of transcription in AS60 and the allotriploid) (1,435 vs. 272, see Additional file 1: Table S3). In each case, the transcription level in the allotriploid was more similar to that of the parent showing the higher transcription level (ELD-ab: 984/1,435, ELD-d: 246/272; Fig. 4a). The pattern was largely retained at the allohexaploid: more than 72% of the ELD-ab genes behaved equivalently in the allotriploid and allohexaploid. Of the set of 1,245 ELD-ab genes identified in the allohexaploid, only 205 were not classed as ELD in the allotriploid (Fig. 4b). A GO analysis of the ELD genes shared by F1 and S1 revealed an enrichment for genes assigned to the cellular component “plastid”. Genes homologous to components of the RNA-dependent DNA methylation (RdDM) pathway, in particular those encoding ARGONAUTE 4 (AGO4), DEFECTIVE IN MERISTEM SILENCING 3 (DMS3), and the RNA-binding protein INVOLVED IN DE NOVO2 (IDN2), were among the ELD-ab genes identified in both the allotriploid and allohexaploid (Fig. 4c). Meanwhile, almost all (426/451) of ELD-ab genes identified in the allotriploid plant achieved this status thanks to the down-regulation of their D genome homoeologue (Additional file 1: Table S3).

Complementation of genome-specific singletons in allotriploid and allohexaploid

As referred to earlier, 53.3% of the 89,315 annotated CS genes were classified as singletons (Additional file 1: Table S2). Of the 13,505 singletons assigned to high confidence group 1 (the HC1 group listed in Additional file 9: Dataset S8) [32], over a quarter were associated with an FPKM greater than unity, and most were consistently transcribed in the AS2255 × AS60 allotriploid and allohexaploid (Fig. 5a). Among the 968 D-subgenome singletons transcribed in the allotriploid, 886 were also transcribed in the allohexaploid. A GO analysis revealed some enrichment in the processes “cellular macromolecule biosynthesis” and “cellular biosynthesis” (Fig. 5a). Some additional processes were enriched for the B genome singletons, including “secondary metabolism” and “response to abiotic stimulus”; and similarly for the A genome singletons with respect to “cellular biosynthesis”. A number of functions were represented among the set of singletons (Fig. 5b). The singletons identified in the allotriploid and allohexaploid of the other lineage (LDN × AS60) covered a similar set of functions (Additional file 1: Figure S5). Members of the set of D-subgenome singletons were more frequently down-regulated in both the allotriploid and allohexaploid (Additional file 1: Table S4).

The re-programming of D genome homoeologue transcription is a prominent characteristic of the allotriploid transcriptome

Hybridization appears to relax the regulation of transcription, thereby inducing differences between a hybrid and its progenitors with respect to the level of transcription of many genes. In diploids such as Arabidopsis thaliana [37] and rice [22], the sequence-based tagging of alleles is straightforward, but in polyploids such as Brassica napus [6], cotton [28] and wheat [29], the situation is complicated by the presence of homoeologues. Bread wheat has evolved from two independent and temporally well separated polyploidization events. The first of these, which occurred ca. 0.5 M years ago, involved the hybridization of T. urartu with an Aegilops species belonging to the Sitopsis section to form T. turgidum; the second was between a cultivated form of T. turgidum and A. tauschii and dates to some 0.01 M years ago [25]. Both events are likely to have a profound effect on the transcriptome. The asymmetry involved in combining a tetraploid with a diploid may account for the observation that it was the D genome homoeologues which suffered the greatest extent of down-regulation in the de novo allotriploid. One possibility is that the silencing involved is activated by the small RNA-mediated DNA methylation system, given that there appears to be a concentration of 24 nt RNAs around a number of genic loci in the D genome [27, 29]. Here, the data imply that D genome homoeologue regulation was the major contributor to the altered transcription patterns exhibited by the allotriploid plants. A major proportion of the non-additive transcribed genes and the ELD-ab genes were also associated with the D-subgenome homoeologues. Curiously, non-additive transcription in the allotriploid plants was for the most part reversed in the allohexaploids, whereas a large proportion of ELD genes maintained their status following the WGD. It is conceivable therefore that the former genes are in some way responsible for the inferior growth of the allotriploid plants, while the latter help to restore vigour to the allohexaploids.

The restoration by WGD of parental homoeologue transcript abundance

WGD is required to restore fertility via the provision of pairs of homologous chromosomes. It also has the effect, as shown here, of reversing some of the disturbance to transcription induced by the hybridization; a similar effect has been noted in Senecio × baxteri when hybrids are created between tetraploid and diploid forms [20, 21]. In wheat, WGD appears to have a relatively small, non-genome specific effect on homoeologue transcription. It reduced the number of UTHs present, an effect which may contribute to the observed reduction in non-additively transcribed genes and hence to the restoration of hybrid vigour. This contrasts to the role of WGD in polyploid rice, in which WGD has been shown to disrupt transcription [22]. The high rate of maintenance of ELD status in the transition from allotriploid to allohexaploid (especially for the ELD-ab genes) may go some way to explaining the morphological similarity between the allohexaploid and tetraploid plants. Several of the ELD-ab genes were homologues of genes encoding RNA-dependent DNA methylation (DMS3, AGO4, and IDN2), which could imply that the maternal parent’s epigenetic modification system was maintained during the process of allohexaploidization.

Singletons may represent a source of novel functionality

The bioinformatics-based analysis of the CS transcriptome suggested that over 50% of the gene content comprised singletons. This number may well represent an overestimate, since the current wheat gene model set was derived from an assembly of rather short reads [32]. A large number of the supposed singletons were found in the transcriptomes of both the allotriploid and allohexaploid, so they may well have a prominent effect on the phenotype of both hybrid forms. The combination of singletons from the three genomes may provide extra functions in hexaploid wheat. In maize, genes exhibiting so called “single parent transcription” - that is those which are silent in one of the parents of a hybrid, but transcribed both in the other parent and in the hybrid - have been shown to contribute materially to heterosis [38, 39]. In bread wheat, a number of genes underlying agronomic traits (notably disease resistances) have no known homoeologues [40], which confirms the notion that singletons in allopolyploids can be very important determinants of biological function.

Plant materials

Two independent T. turgidum (AABB, 2n = 4x = 28) accessions were used as the female parent in the wide cross: these were ssp. durum cv. Langdon (LDN) and ssp. turgidum accession AS2255. The male parent in each case was A. tauschii accession AS60 (DD, 2n = 2x = 14). Both the LDN × AS60 and AS2255 × AS60 allotriploid were produced, along with spontaneously doubled allohexaploid individuals. All three parents were shown to be highly homozygous on the basis of genotyping at 160 microsatellite loci [43]. The crosses required neither embryo rescue nor any hormone treatment. A total of, respectively, 4,035 and 810 allohexaploid grains from 8748 florets and 4802 florets were obtained following the generation of unreduced gametes from 11 LDN × AS60 and six AS2255 × AS60 allotriploid plants. They were grown as the first generation (S1) of allohexaploid plants. The S1 plants were self-pollinated to produce S2, S3, and S4 generations.

RNA was extracted from pot-grown plants raised outside under the local conditions used to grow winter wheat. Before potting, root-tips of individual allohexaploid plants were checked for euploidy by sequential multi-color genomic in situ hybridization (GISH) and fluorescence in situ hybridization (FISH) analyses, as described previously [3]. Out of the analyzed ~100 S2-S4 seeds in each combination, about 10% in LDN × AS60 and 20% in AS2255 × AS60 were aneuploid. No chromosome structural variation was observed except a chromosome fragment in a seed from AS2255 × AS60. Only euploid plants with a complete set of wheat chromosomes were selected for further analysis. The authenticity of triploid F1 hybrids was confirmed by meiotic observation of pollen-mother-cells, as described previously [3]. Measurements were taken of the length (from the base to the tip of a leaf) and width (at the widest part of the leaf) of the first to fourth leaves of seven AS2255 × AS60 allotriploid plants and 12 plants from each of the other lines.

RNA sample preparation and transcriptome sequencing

The youngest fully expanded leaf at development stage 5 [44] was collected from three-five individuals of the parental, allotriploid and allohexaploid plants, snap-frozen in liquid nitrogen, and stored at −80 °C. Total RNA was extracted from the leaf tissue using an RNAprep Pure Plant kit (TIANGEN, Beijing, China), according to the manufacturer’s protocol. The integrity of the extracted RNA was validated using a 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). The RNA was used to construct and sequence 13 RNA-Seq libraries: two each from the AS2255 × AS60 and LDN × AS60 allotriploids, the AS2255 × AS60 and LDN × AS60 allohexaploids, AS2255 and AS60, and one from LDN. Paired end sequencing libraries (average insert size: 200 bp) prepared using a NEBNext® Ultra™ RNA Library Prep kit (New England Biolabs) and sequenced using a HiSeq2000 device (Illumina, San Diego, CA, USA) according to the manufacturer’s standard protocols. Raw RNA reads were de-multiplexed using bcl2fastq v1.8.4 (support.illumina.com/downloads/bcl2fastq_conversion_software_184.html). All contaminants and low quality reads were removed by enforcing a Q30 threshold of 80% and a maximum of 0.2% ambiguous base calls.

Mapping of RNA-Seq reads to the reference genomes

TopHat v2.0.9 software (http://ccb.jhu.edu/software/tophat/index.shtml) was used to align the set of filtered reads from the two AS60 samples against the draft genome sequences of both A. tauschii (ftp://climb.genomics.cn/pub/10.5524/100001_101000/100054/D/Assembly/) and bread wheat (urgi.versailles.inra.fr/download/iwgsc/), allowing a maximum of two mismatches per alignment (parameters: −-bowtie1 -N 2 -r 40 --library-type fr-unstranded). An average of 77.6% of the reads was mappable to the former genome sequence (Additional file 1: Table S5), and of 74.2% to the latter one, indicating a similar mapping efficiency. As genic sequence proved to be very highly conserved (>99% identity) [32], the bread wheat sequence was selected as the reference genome. Thus the reads obtained from all 13 libraries (Additional file 2: Dataset S1) were aligned against the IWGSC draft sequence using the same set of TopHat parameters as given above. The resulting alignments were filtered using the method described by Pfeifer et al. [33].

Quantification of transcript abundance and the recognition of differential transcription

Transcript abundances were obtained from a set of 99,386 high confidence genes (HC1-HC4) represented in the CS gene model (urgi.versailles.inra.fr/download/iwgsc/Gene_models/). Read numbers were normalized by expressing as read fragments per kilobase of exon model per million mapped base pairs (FPKM) [35], using HTSeq-count v0.6.0 software [34], with the parameters -s and no. Correlations between pairs of samples were based on Pearson’s correlation coefficient, calculated from the “cor” function implemented in the R based on log2 (FPKM + 1) transformed data [33]. A correlation dendrogram of genotypes was generated using the function flashClust. Genes for which the transcript abundance was below unity in both parents and their derived hybrids were considered to be transcribed at too low a level and were removed from the dataset. To identify differentially transcribed genes, the R packages DESeq (for comparisons between samples represented by two replicates) [29, 45] or Ebseq (for comparisons involving LDN) [46] were applied. Genes for which the FDR-adjusted p value was <0.05 were considered to be transcribed at a statistically significant different level.

Triplet, duplet and singleton genes

The 99,386 high-confidence CS gene models (HC1-HC4) were compared against each other with BLASTP (E-value threshold of 1e⁻⁵) by considering only alignments with a minimum of 90% sequence similarity as homoeologous genes among the A-, B- and D-subgenome [33, 47]. This produced 28,828 A genome, 30,707 B genome and 29,780 D genome transcripts, which resolved into 8,339 triplets (8,339 × 3 = 25,017 genes/homoeologs), 8,338 duplets (2,370 AB, 3,410 AD, and 2,558 BD) and 47,622 singletons (14,709 A, 17,440 B and 15,473 D) (Additional file 9: Dataset S8). Duplet and triplet genes were considered when searching for genes showing non-additive transcription or ELD, while only the triplet genes were considered in the context of the differential transcription of homoeologues. To compare the transcript abundances between pairs of homoeologues (A/B, A/D, and B/D), read numbers were normalized by dividing by the length of the corresponding gene model before calculating FPKM. To analyze the behaviour of singletons in the allotriploid and allohexaploid plants, a set of 13,505 HC1 genes was considered (of these, 3,992 originated from A genome, 5,614 from B genome and 3,899 from D genome loci, see Additional file 9: Dataset S8).

Gene Ontology (GO) and MapMan enrichment

The genes were annotated using the best-matched rice gene models (RICE MSU 7.0) [48]. AgriGO (bioinfo.cau.edu.cn/agriGO/) was used for GO analysis [49]. MapMan (mapman.gabipd.org/) was used to identify biological processes and pathways of individual genes and homoeologue sets [50]. GO terms showing a corrected FDR of below 0.05 were considered as being significantly enriched.