•The present study aims to understand regulation of gene expression in synthetic and natural wheat (Triticum aestivum) allohexaploids, that combines the AB genome of Triticum turgidum and the D genome of Aegilops tauschii; and which we have recently characterized as genetically stable.
•We conducted a comprehensive genome-wide analysis of gene expression that allowed characterization of the effect of variability of the D genome progenitor, the intergenerational stability as well as the comparison with natural wheat allohexaploid. We used the Affymetrix GeneChip Wheat Genome Array, on which 55 049 transcripts are represented.
•Additive expression was shown to represent the majority of expression regulation in the synthetic allohexaploids, where expression for more than c. 93% of transcripts was equal to the mid-parent value measured from a mixture of parental RNA. This leaves c. 2000 (c. 7%) transcripts, in which expression was nonadditive. No global gene expression bias or dominance towards any of the progenitor genomes was observed whereas high intergenerational stability and low effect of the D genome progenitor variability were revealed.
•Our study suggests that gene expression regulation in wheat allohexaploids is established early upon allohexaploidization and highly conserved over generations, as demonstrated by the high similarity of expression with natural wheat allohexaploids.
Wheat species (genera Triticum and Aegilops) in particular have evolved through frequent allopolyploidizations. Durum or pasta allotetraploid wheat Triticum turgidum (2n = 4x = 28, AABB), was formed < 0.5 million yr ago (Feldman et al., 1995; Blake et al., 1999; Huang et al., 2002) while the widely cultivated allohexaploid wheat Triticum aestivum (2n = 6x = 42, AABBDD), also known as common or bread wheat, was formed more recently (7000–12 000 yr ago) (Kihara, 1944; McFadden & Sears, 1946; Feldman et al., 1995; Nesbitt & Samuel, 1996). Concomitantly with the first allotetraploidization event, the Ph1 (Pairing homoeologous 1) locus evolved. This locus enhances the exclusive pairing of homologous bivalents at metaphase I of meiosis and constitutes the main stabilizing factor of these wheat polyploids (Riley & Chapman, 1958; Griffiths et al., 2006). While the relatively stable natural wheat allotetraploid T. turgidum cannot be resynthesized because the diploid progenitor of the B genome is unidentified, the natural wheat allohexaploid can be resynthesized as the progenitors of its AB genome (T. turgidum) as well as the D one (Aegilops tauschii) are available (Shaked et al., 2001; He et al., 2003; Mestiri et al., 2010). Little is yet known about genetic changes and the regulation of gene expression in the wheat allopolyploids carrying the Ph1 locus. In a recent study (Mestiri et al., 2010), we showed that except for variation in homologous pairing, leading to chromosome instability and aneuploidy, no DNA sequence elimination or other rearrangements are observed when analysing euploid plants of newly synthesized allohexaploids.
Characterization of changes in gene expression in different allopolyploids has shown that many genes behave according to an additive model, whereby gene expression is a combination of expression patterns in the parents. However, some genes deviate from this model and novel patterns of expression (nonadditive) are observed. This is the situation with 5.2–5.6% of genes in Arabidopsis synthetic allotetraploids (Wang et al., 2006), 1–6.1% in cotton (Adams et al., 2004; Rapp et al., 2009) and c. 5% in Senecio (Hegarty et al., 2005). Regulation of gene expression in wheat allohexaploids has been the subject of only a few investigations. He et al. (2003) described, using cDNA-amplified fragment length polymorphism (AFLP), that 7.7% of transcripts in a wheat synthetic allohexaploid showed profiles that were different from those of its progenitors. More recently, Pumphrey et al. (2009) reported that expression of c. 15% of 825 genes, 78% of which were differentially expressed between AB and D genome progenitors, were nonadditively expressed.
In Arabidopsis allopolyploids a biased repression was observed for nonadditively expressed genes that were also upregulated in one progenitor (compared with the other). In cotton, a phenomenon of ‘gene expression dominance’ was suggested where one parental genome contributes more to the expression in the allopolyploid than the other (Rapp et al., 2009). Interestingly, reciprocal silencing, translational bias and expression divergences of homoeologous alleles were shown to occur frequently in the cotton natural and synthetic allopolyploids (Hovav et al., 2008; Chaudhary et al., 2009).
The present study aimed to analyse gene expression regulation in wheat synthetic allohexaploids that were recently characterized as genetically stable (Mestiri et al., 2010). We used the available Affymetrix GeneChip Wheat Genome Array (http://www.affymetrix.com/) on which 55 049 different wheat transcripts are represented, allowing a genome-wide analysis of the wheat transcriptome. We report here a comprehensive analysis on: intra- and intergenerational variation and stability of additive and nonadditive changes in gene expression that follow the allohexaploid formation; the effect of variability of the D genome progenitor; as well as comparison with gene expression in the natural wheat allohexaploid.
Materials and Methods
Synthetic allohexaploids and growth conditions
Various wheat allohexaploids were obtained by interspecific hybridization between different genotypes of the tetraploids T. turgidum and several accessions of A. tauschii, followed by spontaneous chromosome doubling (Mestiri et al., 2010). These were characterized as exhibiting different levels of meiotic stability and aneuploidy but no sequence elimination or rearrangements were seen (Mestiri et al., 2010). For analysis of transcriptome changes, we chose in this study the allohexaploids JOY36 and JOY54 having T. turgidum spp. durum cv Joyau as AB genome donor and as D genome donors, Ae. tauschii spp. stangulata accession 36 (AtsD36) and Ae. tauschii spp. tauschii accession 54 (AttD54) respectively (Mestiri et al., 2010). We compared the first-selfed (S0) and second (S1) generations of JOY54, and the first generation (S0) of JOY36.
The same euploid plants characterized by Mestiri et al. (2010) as not displaying structural changes, were analysed for gene expression regulation compared with their progenitors as well as with mid-parent values (MPV).
All plants of progenitors and synthetic allohexaploids as well as the natural wheat allohexaploid (T. aestivum) cv Courtot were grown in growth chambers at 22°C and 16 h daylength.
Shoot samples were harvested at 35 d old (fifth-leaf stage), snap-frozen in liquid nitrogen and stored at −80°C. RNA was extracted from 200 mg starting material by using the Plant RNeasy Mini Kit with on-column DNAse digestion (Qiagen). All RNA samples were checked for their integrity on the Agilent 2100 Bioanalyzer according to the Agilent Technologies (Waldbronn, Germany) specifications.
We used the available Affymetrix GeneChip Wheat Genome Array, on which 55 049 transcripts are represented by probesets.
Total RNA extraction, reverse transcription, synthesis of the double-stranded cDNA, in vitro transcription of cRNA, labelling and hybridization on the Affymetrix GeneChip Wheat microarray were done using standard protocols as described by Krugman et al. (2010). Microarrays were scanned with the GeneChip Scanner 3000 7G piloted by the GeneChip Operating Software (GCOS). All these steps were performed on Affymetrix platform at URGV laboratory, Evry, France.
Gene expression was measured in two sister plants of each of the genotypes analysed: T. turgidum spp. durum cv Joyau, AttD54, AtsD36, the S0 and S1 generations of JOY54, the S0 generation of JOY36 and the natural allohexaploid cv Courtot as well as an in vitro mixture of equal amounts of RNAs from progenitors in order to measure the MPVs, as described by Wang et al. (2006). This resulted in 18 samples that were hybridized and analysed.
Normalization of expression data
The raw ‘.CEL’ files were imported into the R software (R Development Core Team, 2010) for data analysis. The 18 arrays were altogether normalized with gcRMA (Irizarry et al., 2003), available in the bioconductor package. All raw and normalized data are available from the Gene Expression Omnibus (GEO) repository at the National Center for Biotechnology Information (NCBI), accession number GSE20169 (Barrett et al., 2005).
To build reference datasets for comparisons between the different genotypes, transcripts were considered as expressed (detected) in a given genotype when values of hybridization intensity (I), expressed as log-ratio after normalization, were higher than ‘3’ (I > 3), the Affymetrix detection cut-off, in both biological replicates. If one or both replicates were inferior or equal to ‘3’ (I ≤ 3), the transcript was considered as ‘not detected’. The Supporting Information, Table S1, indicates overall number of expressed and unexpressed transcripts for all genotypes and replicates analysed in the present study.
Analysis of variance, comparison between genotype pairs and classification of expressed genes
To evaluate intragenerational variations and consider biological replicates, we performed primary ANOVA using the R software with the normalized intensity values as explained in Notes S1 and Fig. S1, where the intragenerational effect (Rj) was not significant. We considered for further statistical comparisons the two sister plants of a same generation as two biological replicates representing the generation.
After this primary analysis of variance and biological repeat consideration, we conducted further overall comparison between genotype pairs and classified differentially expressed genes using the VarMixt model (Delmar et al., 2005). Instead of the too-stringent homoscedastic assumption of a common variance, leading to an overestimation of differentially expressed genes, or the overparameterized assumption of a specific variance for each gene (per gene variance), leading to an underestimate of differentially expressed genes, the VarMixt model relies on identifying groups of genes having homogeneous variance (Delmar et al., 2005). The variance of each group of genes is then estimated and used, instead of a common or a per gene variance, and a statistical test is applied to identify differentially expressed genes (Delmar et al., 2005). The raw P-values obtained were adjusted for multiple testing effects by the method proposed by Benjamini & Yekutieli (2001) which controls the false discovery rate (FDR). A gene was declared differentially expressed between two conditions if the adjusted P-value is < 5%. To illustrate, Table S2 showed that the assumption of a common variance (homoscedastic) was rejected for several of the pairs of compared genotypes, highlighting its inadequacy. Only the VarMixt model was used for different comparisons in the present study.
Mid-parent value estimations and considerations
In our wheat allohexaploid model, there are theoretically two genomes contributed by the AB genome progenitor (T. turgidum) and only one genome for the D genome progenitor (A. tauschii). Determination of the ratio of AB to D genomes for MPV estimation is a challenge, because of the unbalanced genome composition. Nevertheless, genome-wide comparisons of the progenitor transcriptomes showed that expression level of the majority of the transcripts (c. 83%) was similar in the AB and the D genome progenitor (P-values > 0.05) (Fig. S2). For the remaining c. 17%, there are roughly equal numbers of transcripts that were more or less expressed in each of the progenitors when compared with the other (Fig. S2). These a priori considerations showed no apparent expression bias towards any of the progenitor and favoured AB to D genome ratio as 1 : 1 for MPV determination.
In our study, MPVs were measured by hybridizing on the microarray a mixture of RNAs composed of equal amounts from each of the progenitors, as described by Wang et al. (2006) and Gaeta et al. (2009). In other studies, MPVs were not directly measured on microarray but rather in silico calculated by averaging expression values observed in progenitors (Pumphrey et al., 2009; Rapp et al., 2009). To compare both methods and choose the most appropriate one, we modelled the measured MPV as:
where represents the MPV measured by hybridization of mixture of progenitor RNA; μ is the overall mean effect; is the normalized hybridization intensity for the AB genome progenitor in repetition jAB for probeset k; is the normalized hybridization intensity for the D genome progenitor in repetition jD for probeset k; is the in silico calculated MPV; is the interaction between the AB and D genomes homoeoalleles, for the probeset k; is the error term.
Thus, in case of no probeset effect or no interaction between progenitor homoeoalleles, the measured MPV might be equal to the in silico value, corresponding to the average of progenitor expression , meaning that there is no technical bias. According to the analysis, the in silico calculated MPV accounted for 89.67% and 91.57%, of total measured MPV variance of JOY54 and JOY36, respectively (P < 2.2 × 10−16); while the bias due to interaction between the AB and D genome homoeoalleles, , accounted for 10.25% and 8.33% of total variance (P < 2.2e−16). This bias cannot be neglected and we thus used the MPV that had been measured by hybridization of mixture of progenitor RNAs as a more appropriate control for the additivity hypothesis.
Analysis of S0 generation of the synthetic allohexaploid JOY54 revealed a majority of additively expressed genes
Transcripts differentially expressed between AB and D genome progenitors of JOY54 represent 16.9% (5084) of the total expressed 30 011 transcripts (Table 1). Among these, 57% were more expressed in the AB genome progenitor and 43% in the D one (Table 1).
Table 1. Differences in gene expression between progenitors: additive and nonadditive expression observed in the generation S0 of the wheat synthetic allohexaploid JOY54, having the Triticum turgidum spp. durum cv Joyau as AB genome progenitor and the Aegilops tauschii spp. tauschii accession 54 (AttD54) as D genome progenitor
Comparison of synthetic allohexaploids to MPV
Relationship between progenitors
AB = AttD54
AB > AttD54
AB < AttD54
aData set composed of transcripts detected in at least one of the AB genome progenitor, D genome progenitor, S0 generation and/or S1 generation.
Transcript expression in the S0 generation of the synthetic allohexaploid JOY54 (JOY54_S0) were first classified as additive (P > 5%) or nonadditive (P ≤ 5%), depending on statistical comparison with MPV (Table 1). Non-additively expressed transcripts were further considered as either ‘over-expressed’ if their expression in the synthetic allohexaploid was higher than MPV (> MPV, P-values ≤ 5%); or ‘under-expressed’ if the expression was lower than MPV (< MPV, P-values ≤ 5%). Cross-comparison with classification based on differential expression, or not, between AB and D genome progenitors led to nine major expression categories (Table 1, designated a–i).
Additively expressed transcripts
Additive expression in JOY54_S0 occurred for 93.4% of transcripts (28 034 out of 30 011) (Table 1, expression categories a, b and c).
The majority of the 24 927 transcripts that were equally expressed in progenitors were also additively expressed in the S0 generation (23 792, corresponding to 95.5%) (Table 1, expression category a).
Similarly, 4242 transcripts out of those 5084 that were differentially expressed between progenitors of JOY54 were also additively expressed in JOY54_S0 (Table 1, expression categories b and c). This represents an overall proportion of 83.4%, which is a slightly lower value than that of 95.5% found for expression category a (P < 2.2 × 10−16).
Nonadditively expressed transcripts
JOY54_S0 exhibited 1977 nonadditively expressed transcripts, whose expression values were significantly different (P < 5%) from the MPV (Table 1, expression categories d–i). These transcripts correspond to 6.6% of the 30 011 total detected transcripts.
A total of 1135 transcripts, representing 57.2%, of nonadditively expressed transcripts in the S0 generation, were equally expressed in AB and D genome progenitors (Table 1, expression categories d and g). There was a higher proportion of under-expressed transcripts (S0 < MPVs) than over-expressed transcripts (S0 > MPVs, 677/458) (Table 1, expression categories: d vs g).
The remaining 42.8% (846) of nonadditively expressed transcripts were also differentially expressed between the two progenitors (Table 1, expression categories e, f, h and i). Of these, the 436 transcripts that were expressed more in the AB genome progenitor than in the D progenitor, showed in JOY54_S0 a higher proportion of under-expression (S0 < MPV) (Table 1, expression category e: 316 transcripts) than over-expression (S0 > MPV) (Table 1, expression category h: 120). By contrast, the remaining 405 transcripts, that were less expressed in the AB genome progenitor than in the D one progenitor, showed nearly similar proportions of under-expressed transcripts in JOY54_S0 (Table 1, expression category f: 189) and over-expressed transcripts (Table 1, expression category i: 217).
Comparison of gene expression levels between the S0 generation of JOY54 and its progenitors
We further compared expression levels between JOY54_S0 and its progenitors. As statistical comparisons were done between pairs of genotypes (see Materials and Methods), we selected 19 expression patterns that could be predicted from all three ways of paired comparison (i.e. the AB genome progenitor vs the D progenitor; and each progenitor compared with JOY54_S0). Comparisons, based on statistical confidence intervals (Fig. 1A), revealed seven major expression patterns (Fig. 1A, I–VII) for expression categories a, d and g (Fig. 1B), where expression was equal in both progenitors (AB = D); six patterns (Fig. 1A, VIII–XIII) for expression categories c, e and h (Fig. 1B), where expression in the AB genome progenitor was higher than that of D progenitor (AB > D); and finally, six other patterns (Fig. 1A, XIV–XIX) for the reverse situation (AB < D) of expression categories c, f and i (Fig. 1B).
All genes from the nine expression categories (Fig. 1B, expression categories a–i) were thus dissected according to the 19 expression patterns, revealed from comparison of expression levels between JOY54_S0 and those of its progenitors (Fig. 1A). The main observations and trends are:
•Expression levels of only a few genes, 10 nonadditively and under-expressed transcripts (S0 < MPV), were lower than those of both progenitors (Fig. 1Bg, pattern II, Fig. 1Bh, pattern X). Similarly, expression levels of a few other transcripts, 84 nonadditively and over-expressed transcripts (S0 > MPV), were higher than those of both progenitors (Fig. 1Bd, pattern I, 1Be, pattern IX, 1Bf, pattern XV).
•Expression levels of the majority of transcripts in the S0 generation of JOY54 were equal to those of both progenitors (Fig. 1B, patterns VII, XI, XVII) or those of one of them (Fig. 1B, patterns III, IV, V, VI, XII, XIII, XVIII, XIX), irrespective of whether genes are additively or nonadditively expressed in comparison with MPV.
•For transcripts that were expressed in the AB genome progenitor than in the D progenitor and additively expressed in JOY54_S0 (Fig. 1Bb) a biased detection towards the AB genome progenitor was revealed. There were 22 times more transcripts (1386) for which expression values in JOY54_S0 were equal to those of the AB genome progenitor and higher than those of the D progenitor (Fig. 1Bb, pattern XII) than the reverse expression pattern (Fig. 1Bb, pattern XIII, only 64 transcripts). Conversely, for transcripts that were expressed in the AB genome progenitor than in the D progenitor and additively expressed in JOY54_S0 (Fig. 1Bc), a nearly similar number of transcripts showed expression values that were equal to those of AB genome progenitor and lower than those of the D genome progenitor (Fig. 1Bc, pattern XIX: 378) or the reverse expression pattern (Fig. 1Bc, pattern XVIII: 356).
In order to allow careful comparisons and interpretation of the observed biased detection, we plotted all expression values, measured in JOY54_S0, the MPV and the two progenitors, for all genes from expression patterns XII, XIII, XVIII and XIX (Fig. 2). These four patterns correspond to all cases where expression levels were different between the progenitors whereas in JOY54_S0 they were equal to those of one progenitor and different from the other (Fig. 2A). We noted that when expression is additive in JOY54_S0 (=MPV), biased detection towards one of the progenitors was also observed for the MPV (Fig. 2Bb, Bc, patterns XII, XIII, XVIII and XIX). This is particularly the case for all 1386 transcripts that showed 22 times more biased detection of expression toward the AB genome progenitor (Fig. 2Bb, Pattern XII) than towards the D progenitor (Fig. 2Bb, pattern XIII, 64 genes only). Conversely, the majority of genes that were nonadditively expressed in JOY54_S0 (> or < MPV) showed for their MPV intermediate hybridization values whereas those of other few ones were biased towards one of the progenitors (Fig. 2Be, Bf, Bh, Bi, patterns XII, XIII, XVIII and XIX).
Comparison with S1 generation reveals a high intergenerational stability of gene expression
Direct comparison showed no significant differences in gene expression between S0 and S1 generations of the synthetic allohexaploid JOY54 (P > 5%), indicating a high conservation of gene expression across the two generations.
The high intergenerational stability was also confirmed when we compared sets of additively and nonadditively expressed transcripts revealed from separate comparisons of S0 and S1 generations to MPV (Fig. 3).
A large proportion of additively expressed transcripts were common to both generations (Fig. 3). This includes 22 550 transcripts (representing 95.9%) that were equally expressed and 3769 ones (representing 91%) that were differentially expressed between progenitors (Fig. 3).
However, 269 and 640 nonadditively expressed transcripts were similarly over-expressed and under-expressed (compared with MPV) in both generations (Fig. 3). These 909 transcripts correspond to 46.1% and 34.5% of the total nonadditively detected probes in the S0 and S1 generations, respectively (Fig. 3). About half of them (474, i.e. 50.3%) were differentially expressed between the two progenitors (Fig. 3).
Because the comparison with MPV was done separately for each of the generations, 1083 (representing 53.9%) and 1720 (representing 65.5%) nonadditively expressed transcripts were specific to the S0 and S1 generations, respectively (Fig. 3). Importantly, none of these transcripts, except five and one, was over-expressed (> MPV) in one generation and under-expressed (< MPV) in the other (Fig. 3). Typically, transcripts showed intermediate expression levels that were not statistically different from MPVs or from the other generation (data not shown).
Comparison of gene expression changes between JOY54 and JOY36 synthetic allohexaploids
We analysed transcriptome changes in the S0 generation of the synthetic allohexaploid JOY36 (JOY36_S0) and compared them with those revealed in the allohexaploid JOY54 (already described). Both synthetic allohexaploids share the same tetraploid AB genome progenitor (T. turgidum spp. durum cv Joyau) but another genotype (accession) of A. tauschii (AtsD36) is the D genome progenitor of JOY36.
Gene expression changes in the S0 generation of JOY36
Transcripts differentially expressed between AB and D genome progenitors of the allohexaploid JOY36 represent 17.3% (5194) of the 29 923 transcripts expressed in at least one of the progenitors or the S0 generation (Table 2).
Table 2. Comparison of gene expression between progenitors of the synthetic allohexaploids JOY54 and JOY36 having the Triticum turgidum spp. durum cv Joyau as AB genome progenitor for both and as D genome progenitors, Aegilops tauschii spp. tauschii accession 54 (AttD54) and A. tauschii spp. strangulata accession 36 (AtsD36), respectively
Relationship between progenitors
AB = D
AB > D
AB < D
aData set composed of transcripts detected in at least one of the AB genome progenitor, D genome progenitor (AttD54), S0 generation and/or S1 generation for JOY54.
bExpression categories designation (a–i). Numbers in italics correspond to genes that are common to JOY36 and JOY54 transcript sets.
MPV, mid-parent value.
Gene expression in S0 generation of JOY36 (JOY36_S0)
Additive expression was also the dominant gene expression pattern observed in JOY36_S0 (93%) (Table 2). JOY36_S0 displayed 2104 nonadditively expressed transcripts, which correspond to 7% of the 29 923 transcripts detected (Table 2). Among these, a total of 1202 nonadditively expressed transcripts (57%), were equally expressed in AB and D genome progenitors, whereas the remaining were differentially expressed (Table 2). For nonadditively expressed transcripts, there were similar proportions of under-expression or over-expression, but a tendency for under-expression was observed for those transcripts which were expressed more in the D genome progenitor (Table 2).
Direct comparison of gene expression between both allohexaploids and their respective progenitors
Comparison revealed that 3647 transcripts, differentially expressed between progenitors of JOY36 and JOY54, were shared and exhibited similar patterns of differential expression in both couples of AB and D genome progenitors (Table 2, 71%). This left 1547 and 1437 transcripts that were specifically differentially expressed between progenitors of JOY36 and JOY54, respectively. These cover a large proportion of those 395 transcripts that were differentially expressed between both D genome progenitors (Table 3). The remainder includes those transcripts differentially expressed in one of the progenitors, but expression in the second D genome progenitor was intermediate (data not shown).
Table 3. Direct paired comparisons of gene expression between the S0 generation of the synthetic allohexaploid JOY36 (JOY36_S0) and the S0 and S1 generations of JOY54 (JOY54_S0 and JOY54_S1)
JOY36_S0 vs JOY54_S0
JOY36_S0 vs JOY54_S1
Comparison between D genome progenitors
AttD36 = AtsD54
AttD36 < AtsD54
AttD36 > AtsD54
Both allohexaploids have the same AB genome progenitor (Triticum turgidum cv Joyau) but the D genome progenitor of JOY54 is Aegilops tauschii spp. tauschii accession 54 (AttD54) and that of JOY36 is A. tauschii spp. strangulata accession 36 (AtsD36). Transcripts are also separated according to their expression between the D genome progenitors.
aThe reference set for this analysis consists of all transcripts expressed (I > 3) in at least one of the genotypes analysed (T. turgidum cv Joyau, AttD54, AtsD36, JOY36_S0, JOY54_S0 and JOY54_S1). Each bold value corresponds the total of non-bold numbers that follow of precede.
JOY36_S0 = JOY54_S0
JOY36_S0 = JOY54_S1
JOY36_S0 > JOY54_S1
JOY36_S0 < JOY54_S1
JOY36_S0 > JOY54_S0
JOY36_S0 = JOY54_S1
JOY36_S0 > JOY54_S1
JOY36_S0 < JOY54_S1
JOY36_S0 < JOY54_S0
JOY36_S0 = JOY54_S1
JOY36_S0 > JOY54_S1
JOY36_S0 < JOY54_S1
Direct comparisons between synthetic allohexaploids showed that 138 and 180 transcripts were expressed more in JOY36_S0 than in the S0 and S1 generations of JOY54, respectively; 74 transcripts had similar expression in both (Table 3). Only 18 and 20 transcripts were expressed less in JOY36_S0 than in the S0 and S1 generations of JOY54, respectively; of these 12 had similar expression in both (Table 3).
Approx. 85% of the 395 genes that were differentially expressed between both D genome progenitors were equally expressed when comparing derived synthetic allohexaploids (Table 3). For the remaining transcripts, 32 and 14 transcripts were respectively more and less expressed in JOY36_S0 than in the S0 generation of JOY54, and followed the same trend when comparing their respective D genome progenitors (Table 3).
Comparison of sets of additively and nonadditively expressed transcript
We compared sets of additively and nonadditively expressed transcripts revealed in the S0 generation of JOY36 with those of JOY54.
Approx. 86.4% (24 028) of the 27 819 transcripts that were additively expressed in the allohexaploid JOY36 were also additively expressed in JOY54 (Fig. 4).
However, we found 936 nonadditively expressed transcripts that were shared by generation S0 of JOY36 and at least one of the two generations of JOY54 (Fig. 4). These can be divided into 535 over-expressed transcripts and 401 under-expressed ones and represented 45% of all nonadditively expressed transcripts in the S0 generation of the allohexaploid JOY36. Only 12 nonadditively expressed transcripts would have been expected as common if the nonadditive expression was random and independent in both allohexaploids (P < 2.2 × 10−16).
Finally, it is important to note that the majority of the nonadditively expressed transcripts specific to one of the allohexaploids were additively expressed in the other one (Fig. 4). When compared with MPVs, only 11 and 48 nonadditively expressed transcripts were, respectively, under-expressed and over-expressed in the S0 generation of JOY36 whereas they showed opposite expression patterns in JOY54 (Fig. 4).
Comparison of gene expression between synthetic and natural wheat allohexaploids
Comparisons showed that the majority of transcripts were equally expressed in the natural allohexaploid T. aestivum cv Courtot, S0 (31 182, representing 99.2%) and S1 generations of JOY54 (30 933, representing 98.4%) as well as the S0 generation of JOY36 (30 996, representing 98.6%) (Table 4). Only 183, 303 and 229 genes were expressed more and 78, 207 and 218 genes were less expressed in Courtot than in S0 and S1 generations of JOY54 and S0 generation of JOY36, respectively (Table 4). In comparison, more transcripts (1334, 2486 and 2475) were either over-expressed or under-expressed when comparing T. aestivum cv Courtot with T. turgidum cv Joyau and the two D genome progenitors of JOY54 and JOY36, respectively (Table 4). This comparison indicates that the natural wheat allohexaploid T. aestivum cv Courtot shows an important proportion of transcripts that are differentially expressed when comparing natural wheat allohexaploid T. aestivum cv Courtot with the AB and/or D genome progenitors, whereas they are equally expressed when compared with the derived synthetic allohexaploids (Table 4).
Table 4. Comparison of gene expression between the natural allohexaploid Triticum aestivum cv Courtot, the S0 and S1 generations of JOY54 (JOY54_S0 and JOY54_S1), the S0 generation of JOY36 (JOY36_S0) and the progenitors of these synthetic allohexaploids
T. aestivum cv Courtot vs other genotypes
Genotypes compared with natural allohexaploid Courtot
Progenitors of JOY54 and JOY36
Regulation according to MPV
Both allohexaploids have the same AB genome progenitor (Triticum turgidum spp. durum cv Joyau) but the D genome progenitor of JOY54 is Aegilops tauschii spp. tauschii accession 54 (AttD54) for JOY54 and that of JOY36 and is A. tauschii spp. strangulata accession 36 (AtsD36). Transcripts were classified as equally expressed (P > 5%) or differentially expressed (P ≤ 5%) between T. aestivum cv Courtot and each of the compared genotypes. Expression in the synthetic allohexaploids is detailed according to additive (= MPV, mid-parent values) and nonadditive (< MPV or > MPV) regulation.
aReference set for this table contains the 31 443 transcripts, considered as expressed (I > 3) in at least one of the genotypes analysed (T. turgidum cv Joyau, AtsD54, AttD36, JOY36_S0, JOY54_S0 and JOY54_S1).
Over-expressed (Courtot >)
Equally expressed (Courtot =)
Under-expressed (Courtot <)
By contrast, the majority of transcripts that were nonadditively expressed in S0 and S1 generations of JOY54 and S0 generation of JOY36 (98.3%, 93.0% and 92.0% respectively) were equally expressed when compared with the natural allohexaploid T. aestivum cv Courtot (Table 4). Only a few remaining transcripts nonadditively expressed in the synthetic allohexaploids were either over-expressed or under-expressed in the natural allohexaploid T. aestivum cv Courtot (Table 4).
Our study presents a comprehensive analysis of genome-wide gene expression changes in genetically stable wheat synthetic allohexaploids and compares, for the first time, the effect of variability of D genome progenitor, the intrageneration and intergeneration variation and stability as well as comparison with naturally occurring and domesticated wheat allohexaploids. A total of 5084 and 5194 transcripts were differentially expressed between progenitors of the two wheat allohexaploids studied, representing c. 17% of the transcripts (see the Results section), which is similar to the proportion of 15%, observed in another study between diploid relatives Arabidopsis thaliana and Arabidopsis arenosa species (Wang et al., 2006). Of these transcripts, 42% and 53% were classified as differentially expressed between two pairs of cotton (Gossypium) species (Rapp et al., 2009) and, more surprisingly, 78% of 825 transcripts were classified as differentially expressed between progenitors of another synthetic wheat allohexaploid, which is similar to the material used in our study (Pumphrey et al., 2009). It was suggested that the high number of differentially expressed genes observed in the two studies mentioned can be attributed to ploidy level difference (for wheat progenitor comparison) and evolutionary histories of gene expression regulation, as the cogenic or closely-related species compared have important conservation in their gene sequences (Pumphrey et al., 2009; Rapp et al., 2009). It is important to note that it was not explained by Pumphrey et al. (2009) how the 825 transcripts were selected among the 17k transcripts spotted on their 70-mer oligo-probes feature microarray (Qiagen: http://www.microarrays.com). We believe that some of the discrepancies between studies could be explained by technical as well as statistical considerations and interpretations, as found in our study (Table S2) and in another study (Wang et al., 2006).
Mid-parent value is used as a control, where in the case of no effect of polyploidy, the expression of a given gene in the polyploid might be equal to the value averaged from those of its parents (Wang et al., 2006). By contrast, a gene is nonadditively expressed if its expression value is different from MPV. While in our study MPVs were measured by hybridization of mixtures of equal amounts of RNA from each of the progenitors, as described by Wang et al. (2006) and later by Gaeta et al. (2009), in other studies MPVs were calculated in silico by averaging expression values observed in progenitors (Pumphrey et al., 2009; Rapp et al., 2009). We compared both types of MPV estimations and showed that the in silico MPV explained only 89.67–91.57% of total variance of the in vitro estimation (see the Results section). The remaining variance results from technical bias and homoeoallele interactions that cannot be neglected just by using the in silico MPV. We suggest that such bias, leading to interpretation errors, must be evaluated for each specific gene expression microarray technology, especially in those studies that relied on the in silico MPV in their data analysis (Pumphrey et al., 2009; Rapp et al., 2009).
Additive expression was the dominant gene expression in the synthetic allohexaploids (93%), whereas nonadditively expressed transcripts represented c. 7% only. Similarly, nonadditive gene expression affected 5.2% and 5.6% of genes in Arabidopsis synthetic allotetraploids (Wang et al., 2006), 1–6.1% in cotton (Adams et al., 2004; Rapp et al., 2009) and c. 5% in Senecio (Hegarty et al., 2005). Pumphrey et al. (2009) showed that expression of c. 15% of genes was nonadditive in another synthetic wheat allohexaploid. This higher proportion of nonadditively expressed genes can be explained by the fact that 78% of the 825 genes, on which Pumphrey et al. (2009) based their analysis, were differentially expressed between progenitors, whereas this category represented only c. 17% of the genome-wide transcript array used in our study. When considering separately transcripts that were differentially expressed between progenitors of each of the two allohexaploids analysed in our study, the proportion of nonadditively expressed genes obtained (c. 17%) is comparable to that obtained by Pumphrey et al. (2009).
Finally, as also observed by Wang et al. (2006), it is also important to note that the microarray technology used does not ensure separate detection of each of the progenitor’s gene copies (homoeoalleles) and their respective contribution to the overall gene expression. Therefore, comparison with MPV would underestimate the level of gene expression changes, especially for those cases where a gene from one progenitor is over-expressed, whereas its homoeolog is under-expressed.
Genomic expression bias and progenitor dominance
In Arabidopsis allopolyploids, Wang et al. (2006) observed, for the 65% of nonadditively and under-expressed (< MPV) genes, a biased repression of those that were expressed more in the A. thaliana than in A. arenosa than the reverse. Comparatively, a biased repression of genes that were expressed more in the AB genome progenitor was observed in JOY54 whereas this was the case for those that were expressed more in the D genome progenitor in the case of JOY36.
In a recent study with cotton allopolyploids an important proportion of genes were expressed at the same level of a specific progenitor than the other one (Rapp et al., 2009). Moreover, this ‘genomic expression dominance’ was reversed in another cotton allopolyploid (the same parental genome became expression recessive) (Rapp et al., 2009). It is important to note that Rapp et al. (2009) detailed their analysis irrespective of comparison with MPV and thus without separating gene expression into additive and nonadditive as done by Wang et al. (2006) or in our study (Figs 1 and 2). We believe, however, that their comparison implicated an important proportion of genes that were additively expressed in cotton allopolyploids; as predicted from comparison of proportions of their nonadditively expressed genes (1% and 6.1%) and those differentially expressed between their diploid progenitors (42% and 53%) (Rapp et al., 2009). Biased detection was observed in our study, especially for additively expressed transcripts that were expressed more in the AB than the D genome progenitors (Figs 1 and 2). However, we cannot draw a definitive conclusion on genomic expression dominance towards the AB genome progenitor for this category of additively expressed transcripts, as biased detection was also the case for the MPV, for which values were not statistically different from those observed for the allohexaploid (Fig. 2). [Correction added after online publication 9 July 2010: in the first line of the preceding sentence, the text ‘However, we draw a definitive conclusion’ was corrected to ‘However, we cannot draw a definitive conclusion’.] Thus, our analysis suggests that technical reasons such as homoeoallele interactions and differential affinity to probesets, rather than polyploidy-related regulation mechanisms, are responsible of the observed detection bias in wheat allohexaploids. We suggest that cross-comparisons to progenitors and MPV measured by hybridizing mixture of progenitor RNA, as presented in our study for wheat, are needed as a real control of the additivity hypothesis in order to reveal and determine the significance of ‘genomic expression dominance’ phenomenon in allopolyploids.
Gene expression changes are not caused by apparent gene loss or other DNA rearrangements
Semi-quantitative reverse-transcription polymerase chain reaction (RT-PCR) validation shows that, out of 74 transcripts tested, expression of four nonadditively and over-expressed (> MPV) transcripts, 49 nonadditively and under-expressed (< MPV) transcripts (among which 47 could be considered as silenced – not detectable in allopolyploids), and eight additively and equally expressed ones was consistent with micro-arrays data (Notes S2). Semiquantitative RT-PCR amplifications of the 13 remaining transcripts (two over-expressed and 11 under-expressed) were already at saturation limits in our experimental conditions. We also analysed the 47 silenced transcripts at the DNA level in the allohexaploids JOY54 and/or JOY36 but did not find any evidence for their deletion or rearrangements (data not shown), confirming previous findings in these allohexaploids (Mestiri et al., 2010) as well as other synthetic or natural wheat allohexaploids, none of which show DNA rearrangements or eliminations (He et al., 2003; Bottley et al., 2006).
Gene expression regulation is established in early generations of wheat allohexaploids and highly conserved during their evolution
Direct paired comparison between S0 and S1 generations of the allohexaploid JOY54 as well as with JOY36_S0 indicated a high conservation and a high reproducibility between allopolyploidization events, independently of the variability of the D genome progenitor. This was also confirmed by the high conservation of sets of additively (93%) and nonadditively expressed (34.5–47%) genes, which were separately revealed in S0 and S1 generations of the allohexaploid JOY54 and the S0 generation of JOY36 (see the Results section and Fig. 3). Similarly, c. 41% of nonadditively expressed genes were shown to be shared between two Arabidopsis resynthesized allotetraploids derived from hybridization between same lines of two Arabidopsis species progenitors and characterized as genetically stable (Wang et al., 2006). Gaeta et al. (2009) found much less conservation of nonadditively expressed transcripts when they compared three Brassica napus resynthesized allopolyploids that were derived from hybridization between same progenitor lines of Brassica rapa and Brassica oleracea but characterized as showing different levels of genetic rearrangements. Approx. 10% of their nonadditively expressed genes were shared by all three allopolyploid lines in S0:1 generation and only 3% in the advanced S5:6 generation (Gaeta et al., 2009).
By contrast, our comparisons in our study show that an important proportion of transcripts that were differentially expressed when comparing the natural wheat allohexaploid T. aestivum cv Courtot with the progenitor species of AB and D genomes (c. 5% and c. 8% of total transcripts, respectively), became equally expressed (P > 0.05) in derived synthetic allohexaploids (Table 4). Expression of only c. 0.8–1.6% of transcripts remained divergent between these two types of allohexaploids, the majority of which (59.6–89.3%) were genes that were revealed as additive in the synthetic allohexaploids (Table 4). Together with similar studies with natural and synthetic cotton allopolyploids (Hovav et al., 2008; Chaudhary et al., 2009), our study confirms that allopolyploidization per se has had the greater impact on gene expression regulation in allopolyploid species. Thus, gene expression regulation in wheat allohexaploids are established upon allohexaploidization, the majority of which is maintained over 7000–12 000 yr, as estimated from the date of formation of the natural allohexaploid (Feldman et al., 1995; Blake et al., 1999; Huang et al., 2002).
Although the present investigation was not conducted to identify specific genes or biological processes affected by allopolyploidization, for all nonadditively expressed genes, revealed in both or at least one of the allohexaploids expression patterns as well as best BlastX matches with UniProt and associated gene ontology (GO) terms (biological process, cellular component and molecular function) are provided (Table S3), as performed according to Pavlidis et al. (2004). Primary comparison of frequencies of occurrence of biological process GO terms between additively and non-additively expressed genes showed that those implicated in photosynthesis pathways, transcription and response to stimuli are more enriched (Table S4). These constitute the starting material for further analysis and identification of specific pathways of gene expression regulation in the wheat allohexaploids.
This project was supported by the ANR (Agence Nationale pour la Recherche)-Biodiversité project (ANR-05-BDIV-015). I.M. was funded by a fellowship of the French Research Ministry (MENRT). V.C. was funded by an Évry-Genopole postdoctoral fellowship. We sincerely thank Prof. J. Wendel (Iowa State University, USA) and Prof. A. Leitch (QMUL, UK) for valuable discussions.