Genomic changes in synthetic Arabidopsis polyploids


(fax +1 206 685 1728; e-mail


Polyploids are common and arise frequently by genome duplication (autopolyploids) or interspecific hybridization (allopolyploids). Neoallopolyploids display sterility, lethality, phenotypic instability, gene silencing and epigenetic changes. Little is known about the molecular basis of these phenomena, and how much genomic remodeling happens upon allopolyploidization. Extensive genomic remodeling has been documented in wheat, but little remodeling occurs in cotton. Newly synthesized Arabidopsis allopolyploids, which display phenotypic instability and low fertility, displayed several, possibly related mechanisms that can remodel genomes. We detected transcriptional activity of several transposons although their transposition was limited. One represents a new family of conditionally active En-Spm-like transposons of Arabidopsis thaliana, which underwent remodeling of CG methylation upon allopolyploidization. A random amplified fragment length polymorphism survey suggested remodeling at few, specific loci. Meiotic analyses revealed the appearance of chromosomal fragments in a substantial fraction of anther meiocytes. In several individuals produced by hybrids between the synthetic and a natural allopolyploid pollen viability inversely correlated with meiotic instability. Activity of selected DNA transposons and the possibly related chromosomal breaks could cause changes by inducing translocations and rearrangements.


Allopolyploids combine and maintain diploid sets of chromosomes from two or more parental species. These special types of polyploids are frequent in nature, fertile, well adapted and genetically stable, demonstrating the success of this evolutionary strategy. Allopolyploids of recent origin, however, commonly display phenotypic instability, low fertility and low embryonic viability (Comai et al., 2000; Pikaard, 1999; Schranz and Osborn, 2000; Soltis and Soltis, 1995). Three kinds of genomic changes might explain the observed instability: first, epigenetic alterations leading to activation or repression of gene expression; second, activation of transposable elements; and third, large-scale chromosomal rearrangements. Together, these related phenomena could have profound effects. Indeed, McClintock (1984) postulated that interspecific hybridization might activate transposons, initiate genomic remodeling (meant here as structural chromosomal changes) and lead to novel phenotypic variation.

We use allopolyploids of Arabidopsis to explore potential sources of instability. Arabidopsis suecica (2n=2x=26) is a natural allopolyploid whose maternal ancestor is A. thaliana (2n=2x=10) and whose paternal ancestor is A. arenosa (2n=4x=32) (Hylander, 1957; Loeve, 1961; Mummenhoff and Hurka, 1995; O'Kane et al., 1996). We produced synthetic allopolyploids using the same parental species (Comai et al., 2000). These neoallopolyploids display similarity to the natural A. suecica in genome constitution and in remodeling of DNA methylation and epigenetic gene silencing (Comai et al., 2000, 2003; Lee and Chen, 2001; Madlung et al., 2002). Fluorescent in situ hybridization (FISH) with centromeric (CEN) probes indicated that the neoallopolyploids have the expected chromosome complement and that their chromosomes undergo mostly homologous pairing (thaliana with thaliana, arenosa with arenosa) (Comai et al., 2003).

Consistent with the finding in A. thaliana, changes in gene expression have been observed in neoallopolyploids of Brassica, cotton and wheat (Adams et al., 2003; Comai et al., 2000; Kashkush et al., 2003; Lee and Chen, 2001; Osborn et al., 2003; Pikaard, 1999; Schranz and Osborn, 2000; Soltis and Soltis, 1995). However, almost no genomic changes were found in neoallopolyploids of cotton (Liu et al., 2001), while neoallopolyploids of wheat displayed extensive loss of DNA sequences interspersed in most genomic regions and representing a substantial fraction of the genome (Ozkan et al., 2001; Shaked et al., 2001). In these wheat lines, activation of read-through transcripts from activated transposon long terminal repeats (LTRs) might be responsible for the silencing of flanking genes (Kashkush et al., 2003). Therefore it is possible that transposon activation may also underlie genomic remodeling. These observations raise several questions: why are the genomic responses to allopolyploidization so different between cotton and wheat? Which of the two is more representative of other plant responses? Do characteristics specific to each genome affect the level of the response? And, is genome remodeling adaptive or deleterious?

Characterization of genomic responses in A. thaliana should be informative because the genome of A. thaliana is sequenced and the genetic resources available in this system facilitate the functional evaluation of candidate mechanisms. It is known, for example, that demethylation and epigenetic restructuring of the genome can lead to transcriptional activation and mobilization of transposons (Miura et al., 2001; Singer et al., 2001). Because epigenetic restructuring takes place during allopolyploidization of A. thaliana, its effect on transposons could account for phenotypic instability as well as for the appearance of a rearranged genome.

To analyze the genomic changes that occur after allopolyploidization we present here a molecular and cytogenetic analysis of synthetic allopolyploids. We show that although increased activity of several transposons can be found in most lineages of newly formed allopolyploids, this phenomenon does not lead to remodeling on the scale seen in wheat. On the contrary, changes involving chromosomal segments could be detected in meiotic cells. Together, these data support the hypothesis that polyploidization can lead to activation of heterochromatic elements and to genomic changes, and that these changes may facilitate rearrangements of large chromosomal blocks.

Results and discussion

Transposon activity in newly formed polyploids

To investigate transposon instability in synthetic auto- and allotetraploids (Figure 1), we examined the global expression pattern of a genomic region that has a high density of assorted transposons. We employed a recently described genomic microarray that interrogates a heterochromatic region of chromosome 4 containing multiple transposons (Lippman et al., 2004). By comparing the hybridization of probes prepared from parental and allopolypoid mRNA, we found that microarray features corresponding to certain transposons displayed specific activation in the allopolyploid hybrid compared with its autotetraploid parental lines. Among the differentially transcribed features on the microarray (data not shown) two genes with similarity to a Petunia transposon (At4g03910 and At4g03900) (Napoli et al., 1990) displayed dramatic transcriptional activation in two different 1 kb microarray features. These genes are part of an En-Spm-like transposon belonging to a novel small family here named Sunfish (Suf, Figure 2a). Suf1-1 is located in the knob region of chromosome 4 and is delineated by perfect terminal-inverted repeats (TIRs) and a typical 3-base pair (bp) target site duplication signature (Figure 2a).

Figure 1.

Pedigree of allopolyploids. A. thaliana ecotype Landsberg erecta underwent spontaneous tetraploidization during tissue culture. Its autotetraploid offspring was pollinated by pollen from autotetraploid A. arenosa and four independent F1 lines were generated (Comai et al., 2000). Three of these lines were selfed and propagated. Offspring of line 605A crossed with pollen from A. suecica, and then propagated by selfing.

Figure 2.

Analysis of Sunfish in allopolyploids.
(a) Gene model of Sunfish (Suf1-1) in the knob of chromosome 4. Numbers at the terminal-inverted repeats (TIRs; arrows) refer to nucleotide coordinates within BAC T25H8 ( Grey and black shaded boxes represent, respectively, predicted exons and introns of two putative genes. Opposing arrows indicate the primer sites used for RT-PCR in (b). H3: HindIII sites.
(b) RT-PCR of Suf 1-1 genes and loading control (cyclophilin).
(c) Southern blot using DNA digested with methyl-sensitive HpaII or MspI, and methyl-insensitive SspI. Differential digestion indicates DNA hypomethylation in allopolyploids.
(d) Southern blot using HindIII-digested DNA. 2x At (first lane): diploid A. thaliana; 4x At (second lane): autotetraploid A. thaliana; Aa: two individuals of A. arenosa; Allos: inbred lines 733, 738, 747, and hybrid line 745H. Rearrangements consistent with transposition (arrows in ‘d’) are displayed by tetraploid A. thaliana and two allopolyploids. The differential bands (bars) in 745H are alleles inherited from A. suecica (data not shown).

We designed PCR primers that amplify three different regions of At4g03910 to confirm the transcriptional activation observed in the microarray analysis of the allopolyploid hybrid. We detected the strongest transcript signal in the first predicted intron of At4g03910 in one neoallopolyploid line, which had not been tested on the original microarray (Figure 2b). Gene expression levels varied between different individuals. For the other two segments of At4g03910 tested by RT-PCR (both in the predicted exon 1) different cDNA preparations of the same genotype showed different expression, but consistently displayed strong transcriptional activity in the same allopolyploid line that also showed the strongest transcription within the first intron (data not shown). The microarray features showing upregulation covered both predicted exon and intron regions of the gene. We also assayed transcription of the other predicted Suf1-1 gene, At4g03900. RT-PCR analysis showed activation in two of the three independent neoallopolyploid lines, as well as strong activity in the allopolyploid hybrid but not in the parents (Figure 2b).

To assess cytosine methylation at CG and CCG sites within the At4g03900 gene of the transposon and in other highly related genes, we performed methylation-sensitive Southern blot analysis on the autotetraploid parents A. thaliana and A. arenosa, an F5 individual of the same allopolyploid line that had shown transposition of Mu, and an F5 individual of the hybrid allopolyploid (Figure 2c). DNA of the paternal A. arenosa did not hybridize to the transposon probe. In the A. thaliana parent, MspI caused the appearance of a new 0.7 kb fragment indicating CCmGG methylation of that site, but most other fragments were resistant to its action indicating widespread methylation of the CmC?GG type. When the allopolyploids are considered relative to the parent, the relatively light digestion by HpaII, but proportionally heavier digestion by MspI indicate that methylation was lost completely only at few sites (cut by HpaII). More predominantly, CmCmGG sites (sites resistant to HpaII and MspI) were converted to CCmGG sites (resistant to HpaII, sensitive to MspI). Therefore, loss of DNA methylation from CmG to CG, and more predominantly from CmCmG to CCmG was associated with allopolyploidy and the concomitant transcriptional activation of the element.

To determine whether transcriptional activation of the element correlated with transpositional activation we performed methyl-insensitive Southern blot analysis (Figure 2d). DNA of the maternal tetraploid A. thaliana Ler, hybridized to the probe at multiple identical sites, including one that was not present in diploid A. thaliana Ler, suggesting that transposition occurred during autopolyploidization perhaps in conjunction with tissue culture. In the subsequent generations this transposition event was fixed and transpositional activity subdued as tested by Southern blotting in 20 different individuals (data not shown). Two of the three independent allopolyploid lines displayed a loss of a common genomic fragment carrying the Sunfish element, consistent with possible transposon excision and loss. Contribution of a previously active transposase from the paternal genome seems unlikely as neither the Southern probe nor the RT-PCR detected Sunfish sequences in A. arenosa.

To define the Sunfish family in A. thaliana, BLAST searches were performed with the two genes, the TIRs and with the entire element. Family members are dispersed throughout the genome and share considerable nucleotide similarity (Figure 3a,b). To further verify that a copy of Sunfish had transposed in the autotetraploid A. thaliana, we performed inverse PCR. We cloned and sequenced three copies of Sunfish that were not present in the sequenced Col genome. Performing BLAST searches with both TIRs against the incomplete and fragmented Ler sequence ( we noticed that at least three copies of Sunfish were located in different locations throughout the Ler genome compared with their locations in the Col genome (Figure 3a) but we did not determine which, if any, of these elements was activated in the autotetraploid or the hybrid allopolyploid.

Figure 3.

Map, relationship, and excision analysis of the En-Spm transposon Sunfish family members.
(a) Suf1-2 through Suf1-7 were located in the Col genome by BLAST searches using the terminal-inverted repeat of Suf1-1, and both genes At4g03900 and At4g03910. The first gene of each element (tnpD homolog) as annotated ( is: Suf1-1: At4g03900, Suf1-2: At1g35590, Suf1-3: At1g36460, Suf1-4: At39190, Suf1-5: At2g10630, Suf1-6: At3g29730, Suf1-7: At3g30670. The Suf elements in the Ler genome are those that could be identified unequivocally from the Ler sequence ( Suf1-8 was first located by inverse PCR (iPCR) and later confirmed from the Ler sequence.
(b) Phylogenetic tree of Suf elements was made in TreeView ( using a Clustal W ( alignment of 1300 bp of the tnpD gene.
(c) Empty-site PCR was performed to detect excision of Suf1-9 in tetraploid A. thaliana. PCR primers complementary to the chr. 5 sequence flanking Suf1-9 allow amplification of an approximately 250-bp product if excision has occurred. Actin primers used as internal control produce a 900-bp product. Col2x: A. thaliana (Col) diploid, Ler2x: A. thaliana (Ler) diploid, Ler4x: A. thaliana (Ler) autotetraploid, Aa: A. arenosa.

To identify active Sunfish in the autopolyploid we designed flanking primers for several members to detect excision events (Figure 3c). Empty-site analysis for Suf1-9 on chromosome 5 in Ler (absent in Col sequence;, displayed its excision in 2 Ler autotetraploids out of 20 tested (Figure 3c), further supporting transposon activation in this lineage. This could be explained by a single excision event, perhaps as early as the original tissue culture treatment followed by segregation in the tetraploid line (Singh, 1993). Transposition of Sunfish was also observed in diploid A. thaliana chromatin maintenance mutants (T. Kagochi and A. Madlung, unpublished data).

In conclusion, Sunfish is an En-Spm-like transposon whose transpositional activity has not been observed before. Transposition of Sunfish occurred in the autotetraploid parent of the allopolyploids most likely early after polyploidization, and likely again in two of three allopolyploid lines (Figure 2d). While silent in the 4x A. thaliana parent, the element became transcriptionally active in allotetraploids.

In addition to activation of Suf, microarray analysis suggested activation of other transposons (data not shown). Using RT-PCR we verified the activity of three more elements: two En-Spm-like elements, At4g04270 and At4g04100, and a Ty-1 copia-like retrotransposon, At4g04410 (Figure 4). Primers for At4g04270 were designed to span an annotated intron. One of the three allopolyploids showed an abundant transcript spanning the intron (Figure 4). The other two inbred allopolyploids, the hybrid allopolyploid, and the diploid Ler were all silent at the tested locus. In contrast, the autotetraploid Ler expressed the spliced mRNA. Interestingly, A. suecica, which was used in the hybridization with the allotetraploid line to yield the hybrid, showed expression of both the spliced and intron-spanning RNAs, while the hybrid was silent. At4g04270 may be unstable and readily activated, possibly by small methylation changes. Retrotransposon At4g04410 showed the strongest expression in two of the allopolyploids and weak expression in the allopolyploid hybrid, while En-Spm-like transposon At4g04100 only displayed transcriptional activation in the allopolyploid hybrid but not in the parents nor in A. suecica. The retrotransposon At4g04410 was activated in two of the three allotetraploids while silent in both parents, in one allotetraploid, the hybrid allotetraploid and A. suecica. Both En-Spm elements (Suf, and At4g04270) showed unexpected gene expression, either within sequences annotated as introns (At4g03910, Figure 2) or as unspliced transcripts (At4g04270, Figure 4). The prototypical element Suppressor-mutator (Spm) from maize produces alternative splice forms of its transcript, which can contain all or parts of gene 1 (tnpD), gene 2 (tnpA), and a large intron (Masson et al., 1989). It is thus possible that alternative splice variants are epigenetically activated in the allopolyploids.

Figure 4.

RT-PCR analysis of transposon expression. The En-Spm-like transposon At4g04270, the retroelement At4g04100, and At4g04410 show variable transcriptional activation in inbred F4 allopolyploids (733, 738, 747), and the hybrid allopolyploid (745H). Primers for At4g04270 span an intron. An intron-spanning RNA is seen in allopolyploid 738, while A. thaliana (At4x) shows transcription of the spliced product. Arabidopsis suecica (As) displays both products. Transcripts for At4g04410 are seen in two of the three allopolyploids and are absent in parents, hybrid, and A. suecica. At4g04100 is detected only in the allopolyploid hybrid. Primers for At4g04270 did not hybridize with genomic DNA from A. arenosa (Aa). Actin primers were used as loading control. RT-PCR of the same cDNA with actin primers spanning an intron showed no contamination with genomic DNA (data not shown).

Several of the transposons represented as features of the heterochromatic array of chromosome 4 displayed additive transcript levels, i.e. consistent with the parental levels (data not shown). An example of a transposon subject to additive regulation is AtMu1, a member of a class of transposons named after maize Mu. Singer et al. (2001) reported activation of AtMu1, a three-member family of elements, in the methylation-deficient ddm1 mutant of A. thaliana, detecting background transposition activity (1/122 individuals) in wild-type Ler, but not in Col. Primers for RT-PCR analysis that amplify all AtMu1 elements also amplified a cDNA and a genomic DNA product in A. arenosa, the paternal parent of the allopolyploids, indicating that an active element was present in this genome. No expression was found in autotetraploid A. thaliana, the maternal parent (Figure 5a). Two lineages of synthetic allopolyploids showed additive amounts of transcripts, while one displayed partial silencing (Figure 5a). Consistent with the hypothesis of stochastic silencing, expression of Mu1 also varied among siblings within an allopolyploid line (data not shown), a type of observation previously reported for these lines (Comai, 2000; Madlung et al., 2002; Wang et al., 2004). Southern analysis showed additional transposon bands in autotetraploid Ler and one of the three allopolyploid lines (Figure 5b). No change in pattern was seen in seven A. arenosa individuals (data not shown). These observations are consistent with AtMu1 activity, possibly during tissue culture or as a consequence of autotetraploidization. A closely related element was already active in the father of the allotetraploids, A. arenosa. Thus, parental contributions could account for transposon activity, although it is unclear whether the paternal Mu-like element is transpositionally active.

Figure 5.

Expression and genomic analysis of AtMu1 in allopolyploids and their parents.
(a) RT-PCR displays expression of Mu1 in A. arenosa, two of three inbred allopolyploids, and the allopolyploid hybrid 745H, but not in autotetraploid A. thaliana. Cyclophilin served as loading control.
(b) Southern blot of HindIII digested DNA probed with AtMu1 probe 2 from Singer et al. (2001). Rearrangements consistent with transposition (arrow) occurred between A. thaliana 2x and A. thaliana 4x, and between the F1 and F4 of allotetraploid 733.
(c) Nine siblings of a late generation of tetraploid A. thaliana were digested with HindIII and probed with the same probe as in (b). One of the siblings displayed rearrangements consistent with transposition.

In conclusion, we show release of transcriptional repression of selected transposons, such as Sunfish, in synthetic allopolyploids. This is consistent with the activation of transposons seen in wheat (Kashkush et al., 2003). Unlike in wheat, in Arabidopsis allopolyploidization consisted of two steps, autopolyploidization of A. thaliana and, separated by multiple generations, hybridization of autopolyploid A. thaliana and A. arenosa. Sunfish, transposed during the transition from diploidy to autotetraploidy (which also involved tissue culture). Its transcription and that of other transposons, however, was silenced in the advanced autotetraploid generations, but clearly active in the allotetraploids. Sunfish was most methylated in the autotetraploid. The observed activity in the allopolyploids was coincident with demethylation and suggests the possibility that loss of methylation may have caused the release of the silenced element. Notwithstanding transcriptional activation, there was only minor transpositional activity in the allopolyploids. This finding, while curious, might be explained by selection. Most gametes and zygotes formed by these plants fail (Comai et al., 2000). It is possible that individuals in which extensive transposition has taken place might experience strong negative selection because of deleterious consequences, such as chromosomal breaks.

Random survey of genomic rearrangements

Our survey of active transposons in the F4 of allopolyploids could have missed elements because of sampling or because they activated as a burst in the very early stages of allopolyploidization and later subsided. To identify changes caused by unsampled or previously undetected transposons, we undertook an amplified fragment length polymorphism (AFLP) analysis. AFLP detect changes in random restriction fragments, predominantly in single-copy DNA and therefore provide a survey of euchromatic genome regions. AFLP analysis was used to provide an estimate of genomic changes in wheat and cotton (Liu et al., 2001; Shaked et al., 2001). In our analysis we included the three independent synthetic allopolyploid lines (733, 738, 747), two F1 lines corresponding to two of the three inbred allopolyploid lines, the hybrid line (745H), and the parents. Approximately 3680 genomic loci were displayed on the AFLP gels (Figure 6). Because of dominance, bands that were common to both parents were not informative of product loss. Additionally, heterozygosity of A. arenosa causes segregation in synthetic allopolyploids and for that reason arenosa-specific AFLP products were not scored for changes. We scored 250–337 thaliana-specific products. Among the three inbred allopolyploid lines, two displayed a loss of the same three AFLPs (Table 1), which were present in the corresponding F1s. Novel products were about one-tenth as frequent as losses. Together, these results indicate low to moderate variation within the allopolyploid genome between parents and allopolyploids. The probability of random identical changes is extremely low (Table 1), suggesting susceptibility of specific loci to rearrangements.

Figure 6.

AFLP analysis of genomic changes. Portion of a gel image displaying AFLPs. The arrow points to an example of an A. thaliana-specific product present in all F1s, but lost in the F4s of two of three inbred synthetic lines, and in the hybrid allopolyploid. Two independent experiments are displayed side by side for each genotype. See Experimental Procedures for scoring details. At: A. thaliana, Aa: A. arenosa, Mix: 1:1 control mixture of the two parental DNAs. The numbers correspond to the line designation of the allopolyploids in Figure 1. The image was scaled vertically to reduce its size.

Table 1.  Survey for genomic changes at AFLP-displayed random loci
Genomic changesaAllopolyploid
  1. aGenomic changes were scored from a total of approximately 3680 AFLPs. The lower number of AFLPs for line 738 reflects a smaller sampling.

  2. barenosa-specific AFLPs were not scored because changes could be due to allele polymorphism.

  3. cShared by two or more lines.

  4. dFrequency = novel products/3680.

  5. eP of the null hypothesis for losses. See Experimental procedures for calculations.

  6. fNot available.

thaliana-specific AFLPb337250337
Unique product loss000
Common product lossc033
Change already in F1na00
Total loss changes0 (0%)3 (1.2%)3 (0.9%)
Unique novel product500
Common novel productc342
Change already in F1na00
Total novel productsd8 (0.2%)4 (0.1%)2 (0.05%)
P of coincident lossesenaf1 × 10−54 × 10−6

While wheat shows very frequent genomic changes (10–15%) during allopolyploidization (Feldman et al., 1997; Shaked et al., 2001) allopolyploid formation in cotton could not be correlated to any appreciable change in genome structure (Liu et al., 2001). These two as well as the present study were based on AFLP surveys and are thus directly comparable: allopolyploidization in A. thaliana led to some genomic alterations at selected loci (overall approximately 1%). Quantitatively, the frequency of this response is closer to that of cotton. Qualitatively, however, it is closer to that of wheat. If, for example, the DNA elements undergoing rearrangements were related to repeats, their lower abundance in the small genome of A. thaliana compared with their abundance in the genomes of cotton and wheat, would suggest that they may be subject to similar instability in both A. thaliana and wheat.

Chromosomal breaks, bridges, and rearrangements

Even limited transposon activity can have profound effects on genome structure if it leads to chromosome breaks, fusions and translocations. We investigated the potential for large-scale rearrangements by examining anther meiotic squashes from 4x A. thaliana, A. arenosa, synthetic allopolyploids, and the hybrid allopolyploid 745H. Neither tetraploid A. thaliana nor A. arenosa displayed abnormalities. All allopolyploids, however, displayed about 30% abnormal meioses (see Experimental procedures for details). In previous studies we observed, by simple DAPI staining, meiotic abnormalities in one of two A. arenosa individuals examined, albeit at lower frequency (10%) than seen here for the allopolyploids (Comai et al., 2000). Thus, instability can occur in some but not all A. arenosa individuals. Instability was not observed in autotetraploid A. thaliana. Thus, allopolyploids displayed increased meiotic instability when compared with the parents.

The type of abnormalities we observed suggests the occurrence of chromosomal breaks. We observed CEN signals, some quite large, that were either present in higher numbers than expected or abnormally placed, and linked to DAPI signals of variable intensity and consistent with chromosomal sizes from normal to hardly detectable (Figure 7a–d,g,h). These signals are very different from the punctate supernumerary CEN signals associated with little or no visible DNA that can be observed in otherwise normal meioses of both diploid and autotetraploid A. thaliana (Comai et al., 2003). These small CEN signals are consistent with the occurrence of repeat-rich extra chromosomal circular DNA, a common feature of many eukaryotes thought to result from intrachromosomal recombination (Cohen et al., 2003). The aberrant CEN signals seen in the allotetraploids are larger and in some cases appeared connected to the spindle apparatus. In other cases CEN DNA of two types (thaliana and arenosa) appeared associated with the same DAPI signals (Figure 7a, bottom), as if two independent fragments had been aggregated or ligated.

Figure 7.

Formation of chromosomal fragments and bridges during meioses of synthetic allopolyploids. Anther meiocytes were labeled with centromeric probes for A. thaliana (AtCEN, red), A. arenosa (AaCEN, green) and DNA (DAPI, blue).
(a) Metaphase I cell displaying four normal bivalents of A. thaliana chromosomes. The fifth bivalent has an abnormally stretched single CEN signal (see arrow in ‘b’) instead of two punctate ones. At least three additional AtCEN signals exceed the expected number of 10 and likely represent chromosome segments.
(b, d) the DAPI channel of (a) was inverted to show DNA in black and better display the AtCEN and AaCEN signals (respectively), demonstrating that both types of CEN DNAs are present in higher number than expected (16 arenosa AaCEN). In some cases, both AtCEN and AaCEN appear present on the same chromatin fragment, presumably due to spurious joining events.
(c) A drawing summarizing the arrangement of chromosomes with AtCEN in the meiosis shown in (a), (b) and (d).
(e) Symmetric bridges during anaphase II are labeled by AtCEN. Inside a single meiocyte – there is no cytokinesis during meiosis of Arabidopsis – two spindles and associated chromosomes are engaged in the equational division of meiosis. Five pairs of thaliana chromatids and eight pairs of arenosa chromatids are separating from each metaphase II plate. At each plate a stretched supernumerary AtCEN is visibly associated with chromatin and is interpreted as a misdivided chromosome (f, j).
(g, h) Metaphase I displaying four normal bivalents marked by the red AtCEN probe. Two abnormal bipartite AtCEN signals are positioned toward the middle of the metaphase plate and likely represent abnormal thaliana chromosomes.
(h) Same as in (g) but showing only the AtCEN signal.
(i) Drawing interpreting the events in (g) and (h).
(j) Hypothetical steps illustrating a chromosome undergoing breakage in its centromeric region (I), followed by duplication (II), and by repair (III). During the last step the two sister chromatid right arms fuse forming an isochromosome, while the left chromatid arms become telocentric and remain independent. At anaphase II, the isochromosome forms a bridge while the chromatid arms on the left move to the poles. White scale bar: 10 μm.

Chromosomal bridges occurring at anaphase II commonly displayed an unusual pattern of CEN labeling (Figure 7e,f). Typically, a bridge is made of a continuous chromosomal strand tensioned between two kinetochores and should be visible as a DAPI-stained thread between two CEN-painted objects. Here, many bridges appeared as CEN-painted threads tensioned between the two groups of oppositely migrating anaphase II chromosomes. Next to the CEN DNA and close to the metaphase plate were larger DAPI-staining chromatin bodies. Although these structures could be interpreted as laggard chromosomes (left behind the normally moving chromosomes), the apparent opposing tension indicates that they are bridges. Remarkably, these bridges appeared symmetrically on the two halves of the anaphase II cell (6E, there is no cytokinesis between meiosis I and II in A. thaliana) and in all sampled cases the central bridge region hybridized to the same CEN probe: either from A. thaliana or from A. arenosa. Figure 7(j) illustrates a mechanism that could explain these abnormalities: breakage of a chromosome in its CEN region followed by repair and formation of telocentric chromosomes and of an isochromosome (Darlington, 1940). Chromosome breaks can result from meiotic misdivision (Kaszas and Birchler, 1996) or from genetic factors (Endo, 1990). Their frequency in synthetic allotetraploids suggests that they could play a role in genome remodeling: breaks in heterochromatin followed by fusion could relocate chromosomal segments, arms and nucleolar organizer regions. Another type of abnormality consisted of unbalanced meiotic products (data not shown), displaying a 2–8 or 3–7 segregation ratio of A. thaliana chromosomes. We have also observed these abnormal ratios in telophase II tetrads (data not shown). These unbalanced products, and the formation of univalents that are then subject to misdivision may result from mispairing, such as pairing between homeologous partners. Although homeologous pairing might occur, we could not document it at metaphase I in the allopolyploids (Comai et al., 2003). The syntenic relationships between the five chromosomes of A. thaliana and the eight chromosomes of A. arenosa remain to be ascertained. Taking the eight chromosome genome of Capsella rubella as a possible model (Boivin et al., 2004) one would expect that homeologous recombination between thaliana and arenosa chromosomes should form at least some dicentric chromosomes, forming classical bridges at anaphase I in which chromatin is tensioned between two CEN signals. We did not, however, observe such bridges. Thus, although homeologous recombination seems a plausible cause of instability, we lack direct evidence for it.

We can conclude that there is an increase in meiotic instability in neoallopolyploids. This instability should generate considerable chromosomal diversity. The frequency with which these events lead to heritable rearrangements, however, is expected to be much lower than the frequency of the observed meiotic abnormalities. Most rearrangements are likely deleterious, leading to defective meiosis whose products have aneuploid genomes. Defective gametes and zygotes are subject to strong viability selection. Indeed, the synthetic allopolyploids have relatively low fitness displaying high lethality of gametes and zygotes (Comai et al., 2000).

As a start toward ascertaining the relationship of meiotic abnormalities to gametic lethality, we compared the meiotic behavior and pollen viability in eight individuals whose progenitor had hybridized to A. suecica, and in two pure synthetic allopolyploids. In a double-blind experiment, we scored the frequency of abnormal meioses and of pollen viability immediately before anthesis (see Experimental procedures). Figure 8 shows that the 10 individuals clustered in two distinct groups: one with poor pollen (10–30% live) and high meiotic abnormalities (30–36%), the other with much better pollen (58–80% live) and lower meiotic abnormalities (20–21%). Thus, meiotic instability and pollen viability were inversely correlated. Interestingly, the difference between 20 and 35% meiotic abnormalities appeared critical for pollen survival, suggesting synergism, or threshold effect, or the action of two types of meiotic abnormalities, one more damaging than the other.

Figure 8.

Relation of pollen viability to meiotic abnormalities in allopolyploids. The dotted line shows the best linear fit. ‘Mixed’ refers to individuals segregating A. suecica genes, in contrast to ‘pure synthetic’ (see Experimental procedures).

The low fitness phenotype of the synthetic allopolyploids seems incongruent with the success of the corresponding natural allopolyploid A. suecica. This discrepancy might be explained by survival of an unfit neopolyploid for a sufficient number of generations to gain fitness, perhaps in an ecological niche with low competition. Another explanation is that variation might exist among A. thaliana ecotypes for fitness of the allopolyploids. We have observed that Ler-derived allopolyploids have higher fitness (viable seed/plant) than Col-derived ones (C. Josefsson and L. Comai, unpublished observations). Thus, only favorable genotypes might produce allopolyploids with sufficient fitness. Once fertile allopolyploids are present, they might rescue newly formed ones through hybridization, as exemplified by line 745H, which segregated individuals that are more fertile and meiotically stable than the pure synthetic allopolyploids (Figure 8). The genetic and molecular basis of fertility in these allopolypoids should be of interest.


We have recreated in the laboratory allopolyploidization events similar to those that had occurred in nature in prehistorical times and resulted in the establishment of the natural allotetraploid A. suecica. We investigated the impact of this event on genome structure as a possible mechanism in the establishment of allopolyploids. The synthetic allotetraploids displayed instability of selected transposons, genomic rearrangements, and chromosomal abnormalities involving the formation of bridges and chromosomal breaks. While these changes fall short of the frequency and scale suggested in synthetic wheat, where up to 20% of one parental genome was reported to be rapidly lost in the form of interspersed deletions, our observations are consistent with more instability than that observed in synthetic cotton allopolyploids. Thus, remodeling of the transcriptome and of the epigenetic landscape during allopolyploidization (Comai et al., 2000; Madlung et al., 2002) is accompanied by genomic instability that could cause changes in chromosomal structure. These changes may result from instability of few, specific loci. Instability may affect predominantly heterochromatic regions, perhaps through the activation of selected transposons that produce chromosomal breaks and facilitate rearrangements. Indeed, rearrangements affecting the nucleolar organizing regions have been observed in allopolyploids derived from the same parents (Pontes et al., in press). Thus, our survey of transposons, random changes, and meiotic stability indicates that early generations of Arabidopsis allopolyploids display mechanisms favoring genomic changes.

Experimental procedures

Plant material

The autotetraploid A. thaliana Ler (line 612, CS3900) and A. arenosa (accession Care-1 or CS3901) parents were used to produce three independently synthesized lines (F4s) of allotetraploids. In addition, we used the natural allotetraploid (A. suecica, Sue-1, accession LC1 or CS22505), and one F2 hybrid, referred to as 745H or ‘hybrid allotetraploid’, derived from a cross of a fourth synthetic allotetraploid line to Sue-1 (confirmed by microsatellite analysis; data not shown) (Comai et al., 2000; Lee and Chen, 2001; Madlung et al., 2002) (Figure 1). The neoallotetraploids had 10 chromosomes from A. thaliana and 16 from A. arenosa, a karyotype corresponding to that of the natural allotetraploid A. suecica (Comai et al., 2000, 2003).

Transposon activation survey and amplified fragment length polymorphism analysis

To survey the status of multiple transposons in a heterochromatic region we employed a microarray tool recently developed (Lippman et al., 2004). In this genomic microarray, about 2000 1-kb-long fragments span the heterochromatic knob region on the short arm of chromosome 4 in a tiling pattern. Therefore, genomic regions representing both transcribed and intergenic regions can be probed with cDNA probes to determine their transcriptional activity. cDNA from the parents and the hybrid allopolyploid was hybridized to the array and analysis of the collected data was performed as reported (Lippman et al., 2004).

Transcriptional activity of selected transposons was verified with RT-PCR and performed as reported earlier (Madlung et al., 2002). Briefly, individuals representing the tested lines were planted at the same time and location. RNA was extracted from different individuals and reverse transcribed for RT-PCR (PCR primer sequences are available upon request). Genomic DNA isolation and analysis was conducted as described (Comai et al., 2000) except that for gel blots 5–10 μg of DNA was digested overnight with the respective enzymes. Inverse PCR (iPCR) was performed as described (Comai et al., 2000). Briefly, DNA was digested with EcoRI, and circularized by ligation. Primers pointing outwards from the 5′ end of the Sunfish element were used in long-distance PCR together with 0.5 ng of ligated DNA. The iPCR primers used were: 5′-ACGGGGAATTTGTCAGCCGAATA-3′ and 5′-AACCGCCACTACCGCCATAATTG-3′. PCR fragments were purified over Sephadex columns, cloned using the TOPO vector system (Invitrogen, Carlsbad, CA, USA), plasmid DNA from selected colonies was purified and sequenced according to standard procedures. For empty-site analysis of Suf1-9 the following primers flanking the transposon were used: upstream: 5′-CAGCAACTCGACTCCTCTCA-3′; downstream: 5′-GCCGTTGATTCTTACATGGTG-3′. To analyze activity of the AtMu1 family, we designed PCR primers to the flanking regions of all three AtMu1 elements and found that element F20A21 had excised in an ancestor of Ler. All individuals tested, including the parent of the allopolyploids, lacked MULE F20A21 at the expected site as evidenced by genomic empty-site PCR (data not shown; PCR primers for F20A21: upstream of Mu element: 5′-CAAAATCATGAAAATTCAAATCCA-3′; downstream of Mu element: 5′-AATCTAAATCGTAAATCGCACAA-3′). AFLP analysis was performed as described (Madlung et al., 2002) except that genomic DNA was used instead of cDNA. Two independent experiments were performed for each genotype. The probability of the observed coincident losses (N) was calculated from the following formula: PN = N! × pexp(N), where p = losses observed/scored products.

Analysis of meiosis by fluorescent in situ hybridization

To examine meiotic abnormalities we conducted FISH using 180-bp CEN repeat probes essentially as described earlier (Comai et al., 2003). Briefly, buds of plants were fixed in Carnoy, spread on glass slides and the DNA of cells in various stages of meiosis was hybridized with rhodamine (red) labeled A. thaliana CEN probes and FITC (green) labeled A. arenosa CEN probes employing stringent conditions that ensure specific hybridization. Abnormalities were assessed by the following criteria: (i) presence of centromeres that at early anaphase I appeared in the wrong position or exceeded the expected number of 10 for AtCEN and 16 for AaCEN; (ii) presence of bridges or laggards at anaphase I or II; (iii) presence of micronuclei or bridges in telophase II. The experiment was repeated with multiple genotypes. In one experiment the following numbers of abnormal meioses were scored out of the total given: A. thaliana 2x and 4x, 0/multiple (not counted); A. arenosa, 0/31; Allopolyploid 733: 79/232 (F4, same as below); Allopolyploid 738 (F5, same as below): 26/73; Allopolyploid 745H: 41/136 (same below).

To correlate meiotic abnormalities to pollen viability, eight individuals derived from hybridization of synthetic allopolyploids to A. suecica were used, five from line 745H and three from other similar hybridizations. A ninth individual was an F5 from allopolyploid 738, and a tenth was an F4 from allopolyploid 733. For these 10 individuals we screened, respectively, 112, 103, 57, 97, 89, 73, 104, 111, 232, and 136 meioses as described above, and 466, 498, 286, 329, 593, 354, 233, 549, 326, and 178 pollen grains after vital staining as described in Comai et al. (2000). The data were analyzed with the statistical package JMP 5.1 of SAS (


We thank our colleagues of the polyploid genome consortium for helpful discussions, Zach Lippman and Vincent Colot for help with the genomic microarray procedure. This work was funded by NSF Plant Genome grant DBI0077774 (Functional genomics of polyploids) to L.C., R.W.D, and R.M., by UPS Research Awards to T.K., and A.M. and a grant from the Murdock Charitable Trust (Murdock grant number 2003192) College Research Program for the Life Sciences to A.M.

Accession numbers of Arabidopsis stocks used: CS3900, CS3901, CS22505.