Genomewide Nonadditive Gene Regulation in Arabidopsis Allotetraploids
Jianlin Wang, Lu Tian, Hyeon-Se Lee, Ning E. Wei, Hongmei Jiang, Brian Watson, Andreas Madlung, Thomas C. Osborn, R. W. Doerge, Luca Comai, Z. Jeffrey Chen


Polyploidy has occurred throughout the evolutionary history of all eukaryotes and is extremely common in plants. Reunification of the evolutionarily divergent genomes in allopolyploids creates regulatory incompatibilities that must be reconciled. Here we report genomewide gene expression analysis of Arabidopsis synthetic allotetraploids, using spotted 70-mer oligo-gene microarrays. We detected >15% transcriptome divergence between the progenitors, and 2105 and 1818 genes were highly expressed in Arabidopsis thaliana and A. arenosa, respectively. Approximately 5.2% (1362) and 5.6% (1469) genes displayed expression divergence from the midparent value (MPV) in two independently derived synthetic allotetraploids, suggesting nonadditive gene regulation following interspecific hybridization. Remarkably, the majority of nonadditively expressed genes in the allotetraploids also display expression changes between the parents, indicating that transcriptome divergence is reconciled during allopolyploid formation. Moreover, >65% of the nonadditively expressed genes in the allotetraploids are repressed, and >94% of the repressed genes in the allotetraploids match the genes that are expressed at higher levels in A. thaliana than in A. arenosa, consistent with the silencing of A. thaliana rRNA genes subjected to nucleolar dominance and with overall suppression of the A. thaliana phenotype in the synthetic allotetraploids and natural A. suecica. The nonadditive gene regulation is involved in various biological pathways, and the changes in gene expression are developmentally regulated. In contrast to the small effects of genome doubling on gene regulation in autotetraploids, the combination of two divergent genomes in allotetraploids by interspecific hybridization induces genomewide nonadditive gene regulation, providing a molecular basis for de novo variation and allopolyploid evolution.

WHOLE-genome duplication may occur via autopolyploidization by multiplying a single genome or via allopolyploidization by combining two or more divergent genomes (Grant 1971; Stebbins 1971). The common occurrence of allopolyploidy in many plant (Stebbins 1971; Masterson 1994) and some animal (Becak and Kobashi 2004) species in nature suggests an evolutionary advantage of allopolyploids over their progenitors and implicates allopolyploidy as a rapid speciation process (Soltis and Soltis 2000; Wendel 2000). The combination of homeologous chromosomes from divergent species not only promotes functional divergence of duplicate genes (Adams et al. 2003; Blanc and Wolfe 2004), but also generates heterozygosity and novel interactions leading to genetic and phenotypic variability and heterosis (Ramsey and Schemske 1998; Soltis and Soltis 2000; Wendel 2000; Osborn et al. 2003) that are stably maintained in the disomic allopolyploids. The data document rapid changes, such as de novo phenotypic variation, transposon activation, nucleolar dominance, gene loss and silencing, and subfunctionalization (Song et al. 1995; Chen and Pikaard 1997a; Pikaard 1999; Comai et al. 2000; Wendel 2000; Ozkan et al. 2001; Kashkush et al. 2002, 2003; Adams et al. 2003, 2004; He et al. 2003; Osborn et al. 2003; Wang et al. 2004) in allopolyploids, which are caused by mechanisms involving dosage compensation, regulatory incompatibility, genetic alteration, and epigenetic modifications (Osborn et al. 2003). Evidently, polyploidy is a prominent and pervasive force in plant evolution (Soltis and Soltis 2000; Wendel 2000), in contrast to the notion that polyploidy has contributed little to progressive evolution (Stebbins 1971).

Despite the general importance and increased interest in understanding the mechanisms and evolution of polyploidy (Soltis and Soltis 2000; Wendel 2000; Wolfe 2001; Osborn et al. 2003), little is known about genomewide effects on the expression of progenitors' genes between the diverged genomes in nascent allopolyploids. We have produced Arabidopsis allotetraploids using interspecific hybridization between two tetraploid species, Arabidopsis thaliana (Ler) and A. arenosa (Comai et al. 2000; Chen et al. 2004; Wang et al. 2004), and tested the consequences of interspecific hybridization on gene expression during early stages of allotetraploid formation. We report the first comprehensive analysis of transcriptome divergence between the progenitors and their allotetraploid lineages. Approximately 3900 genes (∼15%) were differentially expressed between A. thaliana and A. arenosa. The majority of nonadditively expressed genes in the synthetic allotetraploids displayed expression divergence between the parents and were involved in various biological pathways, which may provide a molecular basis of de novo variation for the selection and adaptation of new allopolyploid species.


Plant materials and RNA samples:

Plant materials included A. thaliana isogenic autotetraploids (At4, accession no. CS3900) and diploid (At2, Ler), A. arenosa (Aa, accession no. CS3901), and synthetic allotetraploid lines (Allo733 and 738) (accession no. CS3895–3896). These plant materials were generated as previously described (Comai et al. 2000; Wang et al. 2004). All plants were grown in a growth chamber at 22° and under 16 hr of light per day at the University of Washington with two biological replications. Leaves were collected from 20 plants in each biological replication prior to bolting (with seven to eight rosette leaves) in each line to minimize developmental variation among species and bulked for DNA and RNA analyses (Madlung et al. 2002; Wang et al. 2004). Another 20 plants were grown until flowering, and flower buds were harvested when the first flower bloomed in individual plants.

Total RNA was isolated using Trizol reagent (Invitrogen, San Diego) according to the manufacturer's recommendations. Each RNA sample was quantified by measuring 260/280 ratios using a UV-spectrometer (GeneQuant pro; Amersham Biosciences, Arlington Heights, IL) and by agarose–formaldehyde gel electrophoresis. Total RNA was subjected to mRNA isolation, using a Micro-FastTrack 2.0 mRNA isolation kit (Invitrogen). Equal amounts of mRNA from A. thaliana and A. arenosa were mixed as a midparent value to detect nonadditive gene expression in the allotetraploids.

Fluorescence in situ hybridization:

Fluorescence in situ hybridization (FISH) in anther meiocytes was performed using A. thaliana- or A. arenosa-specific 180-bp centromeric repeats as probes (Comai et al. 2003). The chromosomal images in meiotic cells were analyzed using a Zeiss Axiovert microscope.

Analysis of spotted oligo-gene microarrays:

Spotted Arabidopsis 70-mer oligo-gene microarrays (microarray data are available at using 26,090 annotated genes were cooperatively developed with QIAGEN (Valencia, CA) and Operon (Alameda, CA) (Lee et al. 2004; Tian et al. 2005; Wang et al. 2005). The 70-mer oligo was designed from the 3′-end of each annotated gene. Every feature was spotted once on each slide. We used 500 ng of mRNA in each labeling reaction using Cy3- or Cy5-dCTP (Amersham Biosciences). The Cy3-dCTP reaction was mixed with the Cy5-dCTP reaction for one hybridization, and then an equal amount of RNA samples was reversely labeled for another hybridization (supplemental Figure 1 at Therefore, two “identical” samples each containing an equal amount of Cy3- and Cy5-labeled cDNAs were hybridized with two slides, which constitutes one dye swap. The dye-swap experiment was replicated using an independently isolated RNA sample. Each experiment contains four dye swaps (eight slide hybridizations) or two dye swaps per biological replication (supplemental Tables 1 and 2 at (Chen et al. 2004; Tian et al. 2005).

A total of 48 slide hybridizations were performed for six experimental comparisons to determine changes in expression between the progenitors (At4 and Aa), Allo733 and midparent value (MPV) (leaves), Allo738 and MPV (leaves), Allo733 and MPV (flower buds), Allo738 and MPV (flower buds), and At2 and At4 (leaves) (supplemental Table 2 at Probe labeling, slide hybridization, and washing were performed as previously described (Tian et al. 2005). Raw data were collected using Genepix Pro4.1 after the slides were scanned using Genepix 4000B. The data were processed using a lowess function to remove nonlinear components and analyzed using a linear model (Lee et al. 2004). This linear model was employed to partition variation in the observed data relative to technical and biological variation. Given that each feature is represented once on an array, the linear model isMathwhere μ represents the overall mean effect; A, D, T, and G represent main fixed effects from the array, dye, treatment (e.g., RNA from two species), and gene, respectively; and i = 1, …, 8, k = 1, 2, j = 1, 2, and l = 1, …, 27,648 (including 26,090 70-mer Arabidopsis oligos plus controls). The interaction terms AG, DG, TG, and TDG represent array-by-gene, dye-by-gene, treatment-by-gene, and treatment-by-dye-by-gene interactions, and εijklm denotes the random error and is used to test for significance of main and interaction effects in the model. Due to confounding and/or aliasing issues involving the array, dye, and treatment terms, not all two-way interactions are included in the model. The model residuals are assumed to be normally distributed with a common variance [i.e., εijklm iid N(0, σ2)], unless evidence of variance nonconstancy is observed. In such a case, a per gene variance is assumed [i.e., εijklm independent N(0, σl2)].

We tested differential expression using significant differences in T + TG terms for a particular gene (Black 2002) because we are interested in changes in expression beyond the average treatment effect. The hypotheses that reflect whether a gene, g, has undergone differential expression between treatments, t and t′ (e.g., A. thaliana and A. arenosa) areMathMath

A standard t-test statistic is used for this comparison, based on the normality assumption for the residuals. To control for multiple testing errors the false discovery rate (FDR) of Benjamini and Hochberg (1995) was employed as it provides weak control of the familywise error rate (FWER) and controls the FDR below level α. The FDR is defined as the expected proportion of incorrect rejections of H0, relative to the total number of rejections. The significance level α = 0.05 was chosen for these investigations. All analyses of variance models were fit using standard statistical packages (SAS, R, and Matlab) (Moser et al. 1988; Ihaka and Gentleman 1996).

As mentioned previously, the common variance assumption was used for all genes and per-gene variances for individual genes to estimate the significant changes of gene expression between the two treatments. The genes that were expressed differently at a statistically significant level (FDR, α = 0.05) using a common variance had largefold changes, some of which also had high standard deviations, whereas the genes that were expressed significantly differently using a per-gene variance included those with smallfold changes, which may be difficult to verify. We used a conservative approach and selected the genes that were expressed significantly differently under both statistical tests.

Functional categories of up- and downregulated genes were classified using PENDANT ( and compared using Venn diagrams. The nonadditively expressed genes (identities) were mapped to oligonucleotide sequences using Perl scripts. The oligos were mapped to genomic coordinates using high (red) and low (blue) gradients corresponding to gene densities. Vertical lines above and below the chromosomes showed up- and downregulation, respectively, and the length was proportional to the logarithmfold changes in differential gene expression.

RT–PCR, qRT–PCR, and SSCP analyses:

Approximately 5 μg of total RNA was treated with DNase I, and first-strand cDNA was synthesized using reverse transcriptase (RT) Superscript II (Invitrogen) according to the manufacturer's recommendations. An aliquot (1/100) of cDNA was used as template in quantitative (or real-time) RT–PCR (qRT–PCR), single-strand conformation polymorphism (SSCP) (Adams et al. 2003), and cleaved amplified polymorphism sequence (CAPS) analyses (Wang et al. 2004). qRT–PCR was performed in an ABI7500 machine (ABI Biosystems, Columbia, MD), using the primers (supplemental Table 5 at and SYBER green dye method as previously described (Lee et al. 2004), except that ACT2 was used as a control to estimate the relative expression levels of the genes tested. The expression levels were converted to log-ratios (supplemental Table 4 at in comparison with the microarray data. For SSCP and CAPS analyses, the primers were from A. thaliana loci and used to amplify both A. thaliana and A. arenosa loci. The PCR reactions were performed using one cycle of 94° for 2 min followed by 25–30 cycles of amplification at 94° for 30 sec, 53° for 30 sec, and 72° for 90 sec. The amplified products were digested by a restriction enzyme and subjected to agarose gel electrophoresis (CAPS analysis) or denatured in a loading buffer and resolved in a 0.5× mutation detection enhancement (MDE) gel (SSCP analysis). The images were captured, and band intensities were quantified using a Fujifilm Phosphorimager.


Genetically stable allotetraploids resembled the A. arenosa parent:

Allotetraploids can be formed through a combination of unreduced gametes or interspecific hybridization between diploid species followed by chromosome doubling (Grant 1971; Stebbins 1971). To study early events of gene regulation in synthetic allopolyploids, we created independent allotetraploid lineages (Allo733 and Allo738) by pollinating A. thaliana autotetraploids (At4) with A. arenosa tetraploids (Aa) (Comai et al. 2000; Wang et al. 2004) (Figure 1), which appear to have the same number of genes at the same ploidy levels (Comai et al. 2000). Heterozygosity in the allotetraploid progeny was minimized by self-pollination for five generations. The chromosome numbers and parental origins were verified in the first and fourth generations, using FISH (Comai et al. 2003) and informative microsatellite markers (data not shown). Without exception, each line possessed five pairs of chromosomes from A. thaliana and eight pairs from A. arenosa (Figure 1) (Comai et al. 2000, 2003). The morphology of these plants varied between lineages, coincident with rapid genetic and epigenetic changes observed in new allopolyploids (Comai et al. 2000; Madlung et al. 2002; Wang et al. 2004). Many allotetraploid lineages resembled the A. arenosa parent and A. suecica, a natural allotetraploid (Pikaard 1999; Comai et al. 2000; Madlung et al. 2002). These morphological characteristics include long leaves, tall stature, many branches, deeply serrated rosette leaves, and large rosettes and flowers. The data indicate that A. arenosa appear to be morphologically dominant over A. thaliana in the allotetraploids. The flower colors varied from pink (like A. arenosa) in the early generation (S1), to a mixture of pink and white flowers in the intermediate generations (S2–4), to white colors in the late generation (S5), suggesting the notion of stochastic and rapid changes in gene expression (Comai et al. 2000; Wang et al. 2004).

Figure 1.—

(A) Production of stable synthetic allotetraploids (Allo733 and 738). A self-fertile A. thaliana autotetraploid (Ler, At4) was pollinated with a natural A. arenosa tetraploid (Aa). Multiple independent allotetraploids in S1 were self-pollinated by single-seed descent to the S5 generation. Allo733 and Allo738 resembled A. arenosa and natural A. suecica (Comai et al. 2000). Fluorescence in situ hybridization (FISH) analysis indicates that two sets of centromeres in Allo733 are derived from A. thaliana (At4) and A. arenosa (Aa), respectively. Bar, 5 mm. Allo, allotetraploid.

Transcriptome divergence between the progenitors:

To determine the molecular basis of phenotypic differences, we analyzed transcriptome changes in the progenitors, using spotted oligo-gene microarrays designed from A. thaliana annotated genes (Tian et al. 2005). Microarray data from four dye-swap experiments (i.e., two dye swaps per biological replication) (supplemental Figure 1 at were analyzed using a linear model (Lee et al. 2004) and the results were adjusted for multiple comparisons (Tian et al. 2005). Unless otherwise noted, we selected the differentially expressed genes that were statistically significant under both common and per-gene variances (Figure 2, Table 1).

Figure 2.—

Logarithm fold change vs. per-gene standard deviation in a replicated experiment containing four dye swaps. The hybridization probes were cDNAs from Allo733 and two parents, A. thaliana tetraploid and A. arenosa. The data were analyzed using a linear model as previously described (Tian et al. 2005). Green, black, and red dots indicate the pools of significant genes selected by multiple comparison tests (false discovery rate, FDR, α = 0.05) using a per-gene variance, a common variance, and the intersection of the two, respectively. Statistical significance for extremely smallfold changes was detected for two features replicated 6 and 49 times within each microarray slide, indicating the power of replication in microarray experiments.

View this table:

The number of differentially expressed genes detected using a common variance and/or a per-gene variance

We characterized transcriptome differences between A. thaliana and A. arenosa that diverged ∼5.8 MYA (Koch et al. 2000). We found that 3923 (∼15%) genes were differentially expressed between the progenitors, of which 2105 (∼8%) and 1818 (∼7%) were expressed at significantly higher levels in A. thaliana and A. arenosa, respectively (Figure 3A, Table 1). The differentially expressed genes represented as much as ∼43% of the transcriptome, using a per-gene variance analysis (Table 1), indicating a wide range of gene expression differences between the two species, which is reminiscent of the >50% of transcriptome changes in Drosophila species that diverged ∼2.5 MYA (Ranz et al. 2003). Among 11,199 differentially expressed genes, 5232 (47%) genes were expressed at a higher level in A. thaliana than in A. arenosa, whereas 5967 (53%) genes were expressed at a higher level in A. arenosa than in A. thaliana. In a separate study using Affymetrix chips, Schmid et al. detected several hundred genes that were expressed more than twofold differently between A. thaliana Col and Ler ecotypes (Schmid et al. 2003). Although the two arrays employ different analytical tools, it appears that the gene expression differences detected between species are much greater than those between ecotypes.

Figure 3.—

Transcriptome divergence and nonadditive gene expression between allotetraploids and their progenitors. (A) The proportion of transcriptome that was highly expressed in A. thaliana (At4), A. arenosa (Aa), or equally expressed (both). (B) Venn diagram showing the number of genes with expression divergence between the progenitors (blue) and between Allo733 (red) or Allo738 (green) and the midparent value (MPV, supplemental Figure 1 at (C) Chromosomal distribution of the 820 genes displaying nonadditive expression in both allotetraploids (see text).

Genomewide nonadditive gene regulation in the allotetraploids:

To determine how transcriptome divergence contributes to genetic and morphological variation in allotetraploids, we compared mRNA abundance in an allotetraploid with the MPV (an equal mixture of RNAs from two parents) (supplemental Figure 1 at Violating the null hypothesis for no gene expression difference between the allotetraploid and midparent value suggests that a gene(s) is nonadditively expressed; however, we cannot detect the situation where silencing of a locus is compensated by increased expression of its homeologous locus. Thus, microarray analysis may underestimate the number of genes that are differentially expressed between an allotetraploid line and the parents. We discovered that 1362 (∼5.2%) and 1469 (∼5.6%) genes were expressed nonadditively in Allo733 and Allo738, respectively (Table 1). When a per-gene variance was used, the nonadditively expressed genes accounted for ∼32% (8377, Allo733) and ∼38% (9875, Allo738) of the transcriptome. The data suggest that orthologous genes in allopolyploids are frequently expressed in a nonadditive fashion.

If the regulatory changes inherited from the parents determine species divergence, the genes displaying species-specific expression patterns may be modulated in the allotetraploids. Indeed, among the 2011 genes that were nonadditively regulated in two allotetraploids, 1377 (∼68%) genes were included in those that were differentially expressed between the parents (Figure 3B), which are significantly different from a random distribution of nonadditively expressed genes (∼15%, χ2 = 1180.5, P ≤ 0.00001). Among them, 820 (∼41%) genes were common to both allotetraploids (Allo733 and Allo738), whereas 649 (∼32%) and 542 (∼27%) genes were unique to All733 and Allo738, respectively, indicating general and specific effects of allopolyploid formation on gene regulation in the independently derived allotetraploids. The 820 nonadditively expressed genes in both allotetraploids (Allos) were randomly distributed across the genome and displayed no obvious chromosomal regions susceptible to allopolyploidy-dependent gene regulation (Figure 3C).

Progenitor-biased gene repression in the allotetraploids:

We analyzed direction of change and parental origin of nonadditively expressed genes. Among them, 1038 (∼76%) and 952 (∼65%) genes were downregulated in Allo733 and Allo738, respectively (Figure 4A), suggesting that repression is a mode of nonadditive gene regulation in synthetic allotetraploids. We divided the repressed genes into three categories on the basis of their expression patterns in the parents. First, 838 (∼99%) and 611 (∼94%) genes that showed higher levels of expression in A. thaliana than in A. arenosa were repressed in Allo733 and Allo738, respectively (Figure 4B), which coincides with the silencing of A. thaliana but not of A. arenosa rRNA genes (Chen et al. 1998; Pikaard 1999) and the overall suppression of A. thaliana phenotype in new allotetraploids and in natural A. suecica. Second, 90 (∼35%) and 159 (∼50%) genes that were expressed at higher levels in A. arenosa than in A. thaliana were downregulated in Allo733 and All738, respectively (Figure 4C). Third, 110 (∼42%) and 182 (36%) genes that were equally expressed in A. thaliana autotetraploid and in A. arenosa were repressed in Allo733 and Allo738, respectively (Figure 4D). There was no bias toward gene repression in the last two categories. The data demonstrate that the genes more highly expressed in A. thaliana autotetraploids than in A. arenosa are subject to orchestrated repression in the synthetic allotetraploids.

Figure 4.—

Downregulation of A. thaliana genes in the synthetic allotetraploids. (A) Distribution of nonadditively expressed genes detected in each allotetraploid (Allo733 or Allo738) or both allotetraploids (Allos). (B) The nonadditively expressed genes in each allotetraploid matched the genes that were highly expressed in the A. thaliana autotetraploid. (C) The nonadditively expressed genes in each allotetraploid matched the genes that were highly expressed in A. arenosa. (D) The nonadditively expressed genes matched the genes that were equally expressed in both parents. The percentages of downregulated genes are indicated above the columns in each histogram.

Nonadditive gene regulation in various biological pathways:

According to 15 functional classifications of 820 nonadditively expressed genes detected in both allotetraploids (Figure 5A), the percentages of genes in the hormonal regulation and cell defense and aging categories were 150–175% of those in the same categories classified using all annotated genes in Arabidopsis (Figure 5B), suggesting that these genes are particularly susceptible to expression changes in response to the perturbation resulting from intergenomic interactions in the allotetraploids. Many genes involved in the ethylene biosynthesis pathway were repressed in one or two allotetraploids (Figure 6A and supplemental Table 3 at, which may induce expression changes in ethylene-responsive genes involved in a wide range of developmental processes and fitness responses, including seed germination, leaf and flower senescence, fruit ripening, programmed cell death, and biotic and abiotic stress responses (Guo and Ecker 2004). Of the 97 HSPs in Arabidopsis (Arabidopsis Genome Initiative 2000), 33 displayed expression differences from the midparent value (Figure 6B). Thirty-one HSPs that were highly expressed in A. thaliana were repressed, which may reflect “buffering” effects (Queitsch et al. 2002) on pathway redundancy. Notably, fewer than expected transposons altered expression in the allotetraploids (Figure 5B), although some might be included in the unclassified category. This appears to be inconsistent with B. McClintock's notion of “genomic shock” (McClintock 1984) but we note that these allopolyploid lineages represent the survivors among the original F1 products (Comai et al. 2000) and in the late generation (S5).

Figure 5.—

Classification of nonadditively expressed genes detected in synthetic Arabidopsis allotetraploids. (A) The 820 genes detected in both Allo733 and Allo738 lines (Allos) were classified into 15 functional categories using the PEDANT analysis system ( (Arabidopsis Genome Initiative 2000). (B) The percentages of the genes in each functional category detected in Allo733, Allo738, or both (Allos). The relative ratios in the y-axis were estimated using the percentage of genes detected in each functional category in an Allo line divided by the percentage of all ∼26,000 annotated genes in the Arabidopsis genome (Arabidopsis Genome Initiative 2000). The percentage of the genes detected in the Allo line equal to that of all genes in the whole genome is shown as 100% (dashed line).

Figure 6.—

Nonadditive gene regulation occurs in various pathways. (A) Progenitor-dependent repression of the genes involved in the ethylene biosynthesis pathway in Arabidopsis allotetraploids. Each number in parentheses below an enzyme or molecule in the pathways indicates the fold change for the expression of a gene, homolog (h), or putative (p) homolog detected by microarray analysis. Red, green, blue, and purple colors indicate that gene expression differences are detected in both allotetraploids (Allo733 and -738), in Allo738, in Allo733, and between the two parents, respectively. (B) Repression of heat-shock protein (HSP) genes in Arabidopsis allotetraploids. Thirty-one out of 33 HSPs were repressed in each allotetraploid (Allo733 or -738). The length and directions of vertical bars indicate logarithm fold changes in up- (above the line) or downregulation (below the line) of the HSPs relative to the midparent in the allotetraploid lines.

Developmental and parental contributions to nonadditive gene regulation:

We tested whether nonadditive gene regulation in synthetic allotetraploids is sensitive to developmental changes by comparing the gene expression divergence detected in leaves and flowers. Allo733 displayed 1355 genes nonadditively expressed in flower buds, of which 175 (∼7%) were also detected in the leaves (Figure 7A), and 1180 were nonadditively expressed only in flowers. Little overlap of the genes detected between leaves and flowers suggests a developmental role in nonadditive gene regulation in the allopolyploids in a manner reminiscent of developmental derepression of silenced rRNA genes (Chen and Pikaard 1997b) and subfunctionalization of some duplicate genes (Adams et al. 2003). It is notable that gene expression changes may occur during the transition from vegetative to reproductive development, but appear to be consistent within a tissue type (e.g., rosette leaves) (Chen and Pikaard 1997b). Compared to Allo738, fewer nonadditively expressed genes were detected in the flower buds in Allo733 (supplemental Table 1 at, which may reflect developmental variation among allotetraploid lineages (Comai et al. 2000; Madlung et al. 2002).

Figure 7.—

Developmental and parental contributions to nonadditive gene expression. (A) Venn diagrams of the genes that displayed nonadditive expression in leaves and flower buds in Allo733. Only 175 of 2542 genes showed overlap between leaves and flower buds. (B) Verification of 11 genes detected in microarrays by qRT–PCR. The gene expression levels in the parents were higher in A. thaliana (At > Aa), higher in A. arenosa (Aa > At4), and equal (At4 = Aa). (C) SSCP and CAPS analyses showing parental contributions to nonadditive gene regulation in the allotetraploids. The genes studied in B and C are in the functional classifications of stress (HSP or HSP90 and HSP17.6b); cell cycle, defense, and aging (CYC, CHI, LRR, PDF, and WRKY); metabolism and energy (BCB, PORa, PORb, and SPP); and flower development (COL2 and FLC). The restriction enzymes used in CAPS analysis are indicated in parentheses.

We verified expression patterns of 11 nonadditively expressed genes using qRT–PCR analysis (Figure 7B). Six were repressed and 5 were upregulated in the allotetraploids, consistent with the microarray data (supplemental Table 5 at Five genes (WRKY, BCB, HSP90, PDF, and LRR) that were expressed at higher levels in A. thaliana than in A. arenosa (At4 > Aa) were repressed in the allotetraploids. Three of four genes (FLC, PORa, and PORb) that displayed higher expression levels in A. arenosa than in A. thaliana (Aa > At4) were upregulated, and one (CYC) was repressed in the allotetraploids. Two genes (CHI and SPP) that were equally expressed in the parents (At4 = Aa) were upregulated in the allotetraploids. Using locus-specific SSCP or CAPS assays, we analyzed the contribution of A. thaliana and A. arenosa loci to the nonadditive gene regulation (Figure 7C). For WRKY, BCB, and COL2, both A. thaliana and A. arenosa loci were repressed in the allotetraploids, whereas HSP17.6b repression was due to the A. thaliana locus. The repression of several COLs and upregulation of FLC may correlate with late flowering in the allotetraploids. Similarly, upregulation of PORb and SPP in the allotetraploids was related to A. thaliana loci, whereas upregulation of CHI was caused by the A. arenosa locus. The data suggest both cis-regulatory and trans-acting effects (Wittkopp et al. 2004) on nonadditive gene regulation in the allotetraploids. Furthermore, upregulation of SPP encoding starch phosphorylase and of PORa and PORb encoding protochlorophyllide oxidoreductases in the photosynthetic pathway may lead to vigorous growth in the allotetraploids.

Autopolyploidization does not induce genomewide nonadditive gene regulation:

To determine whether nonadditive gene regulation is affected by genome dosage, we analyzed transcriptome differences between the A. thaliana diploid (At2) and the isogenic autotetraploid (At4) (supplemental Table 1 at Only 88 genes were expressed significantly differently between the diploid and the autotetraploid, which is reminiscent of the dosage-dependent regulation of a dozen genes as observed in yeast autoploids (Galitski et al. 1999). The results suggest that doubling the same genome in autopolyploids has much smaller effects on gene regulation than combining the divergent genomes in allopolyploids. However, allopolyploidy effects may not be as simple as the sum of “hybridization” and “genome doubling” (see discussion).


Effects of autopolyploidization and allopolyploidization on gene regulation:

Polyploidy effects on gene regulation may be caused by genome doubling and/or intergenomic interactions. Autopolyploidization induces gene expression changes in response to the increase in genome dosage (Birchler 2001). Only 12 and 88 genes, respectively, respond to autoploidy changes in yeast (Galitski et al. 1999) and Arabidopsis, suggesting that increasing genome dosage affects a small subset of genes. During autopolyploidization, mechanisms such as dosage compensation (Birchler 2001) are responsible for maintaining expression patterns of the genes except those associated with the large size of polyploid cells (Galitski et al. 1999). For the majority of genes studied in maize, their expression levels are dependent on the dosage of chromosomes or chromosome arms (Guo et al. 1996; Auger et al. 2005).

The dramatic changes in nonadditive gene regulation observed in the allotetraploids may be induced by interspecific hybridization. Our data suggest that 15% of the transcriptome diverged between A. thaliana and A. arenosa, which accounted for 68% of nonadditively expressed genes (2011) in the synthetic allotetraploids. In addition to 820 genes that changed expression in both allotetraploids, 649 and 542 genes were unique to Allo733 and Allo738, respectively. These genes may correlate with specific changes in individual allotetraploids and facilitate selection and adaptation of new allopolyploid species in response to environmental cues and developmental changes. Indeed, nonadditive gene regulation is developmentally regulated, which may lead to subfunctionalization of duplicate genes (Lynch and Force 2000; Adams et al. 2003) in different organs or tissues (Chen and Pikaard 1997b). Transposons are underrepresented in the genes that display expression changes in the allotetraploids. It is likely that many transposons are not included in the annotated genes for microarray analysis. Alternatively, the effects of genomic shock (McClintock 1984) may be “settled” in the selfing progeny (S5). Finally, dosage-dependent gene regulation (Birchler 2001; Auger et al. 2005) may account for part of the gene expression changes in the allotetraploids. Indeed, 51% (45/88) and 32% (28/88) of the genes that display expression divergence between A. thaliana diploids and isogenic autotetraploids were also expressed nonadditively in two allotetraploids.

There is a possibility that 70-mer oligos designed from A. thaliana may not hybridize well to the A. arenosa genes, although we have shown that 192 A. thaliana oligos hybridized equally well to A. arenosa and Brassica genes (Lee et al. 2004), probably because of >95% genic sequence identity between A. thaliana and A. arenosa (Lee and Chen 2001) and >85% of it between A. thaliana and B. oleracea (Cavell et al. 1998). As a result, the sequence divergence may also contribute to the difference in gene expression detected between A. thaliana and A. arenosa.

Insights into nonadditive gene regulation in the synthetic allotetraploids:

In Arabidopsis allotetraploids, the progenitor-dependent gene regulation is not restricted to rDNA loci subjected to nucleolar dominance (Pikaard 1999) but occurs at a genomewide scale in various biological pathways. The available data suggest that the expression of orthologous genes during evolution and speciation is not purely neutral. Selection and adaptation over evolutionary time may promote divergence of regulatory elements and/or transcription factors and regulatory proteins. The competition between the diverged regulatory pathways may determine nonadditive gene regulation in allopolyploids of Arabidopsis (Wang et al. 2004), cotton (Adams et al. 2004), Senecio (Hegarty et al. 2005), and wheat (Kashkush et al. 2002; He et al. 2003), in interspecific hybrids (Wittkopp et al. 2004) in Drosophila and intraspecific hybrids in maize (Guo et al. 2004; Auger et al. 2005), and in sex-dependent gene regulation in Drosophila (Ranz et al. 2003; Gibson et al. 2004). It is notable that outcrossing in A. arenosa and inbreeding in A. thaliana may accelerate their divergence during evolution. Each progenitor might have evolved specific regulatory systems affecting rDNA and other loci, perhaps via concerted evolution (Coen et al. 1982), and the interactions between these diverged regulatory systems in allopolyploids may trigger repression of the A. thaliana-“specific” genes and of the rDNA loci (Chen and Pikaard 1997a; Pikaard 1999; Wang et al. 2004). Although the underlying mechanisms for preferential repression of A. thaliana genes are yet to be determined, sudden reunification of divergent genomes may induce genome instability (Madlung et al. 2002; Wang et al. 2004) and changes in chromatin structure and RNA-mediated processes (Osborn et al. 2003; Chen et al. 2004). Hybrid- or allopolyploidy-induced incompatibilities may be overcome by gene expression modulation through chromatin modifications, transcription factors such as Myb (Barbash et al. 2003), and/or RNA interference. Interestingly, nonadditive gene regulation in the allotetraploids depends largely on expression divergence between the parents. Thus, hybrids derived from distantly related species may induce a high level of gene expression changes in a nonadditive fashion, providing molecular bases of hybrid vigor (Birchler et al. 2003) and of novel variation in the allotetraploid progeny (Comai et al. 2000; Wang et al. 2004). Furthermore, the stochastic establishment of nonadditive gene regulation in newly synthesized allotetraploids (Wang et al. 2004) may increase the potential for fitness and selective adaptation. In contrast to the lethality and sterility observed in interspecific hybrids (Barbash et al. 2003), nonadditive gene expression changes may be maintained and transmitted in meiotically stable allopolyploids, providing a mechanism for de novo variation and evolutionary opportunities for selection and adaptation of new allopolyploid species.


We thank James A. Birchler, Gary E. Hart, Robert A. Martienssen, J. Chris Pires, Douglas E. Soltis, Jennifer Tate, and Jonathan F. Wendel for critical suggestions. This work was supported by a grant from the National Science Foundation Plant Genome Research Program (DBI0077774). Work in the Chen lab is supported in part by a grant from the National Institutes of Health (GM067015). The authors declare that they have no financial conflict of interest.


  • 1 Present address: Department of Biology, University of Puget Sound, Tacoma, WA 98416.

  • Communicating editor: G. Gibson

  • Received July 7, 2005.
  • Accepted September 19, 2005.


View Abstract