Protein-coding genes are epigenetically regulated in Arabidopsis polyploids

A. suecica and A. thaliana (Landsberg strain) were self-pollinated for at least four generations after they were obtained from original collections (25). C. arenosa (pink flower) was maintained by cross-pollination among a few plants in each generation. Autotetraploid A. thaliana lines (Columbia) and (Landsberg) were obtained as described (22). The plants were grown in vermiculite mixed with 10% soil in an environment-controlled chamber with 14/10 h (day/night) light and 24/20°C (day/night) temperature. Treatment of A. suecica seedlings by using aza-dC was adapted from a published protocol (23).

Total genomic DNA and RNA isolation and DNA blot analysis were performed as described (22). AFLP-cDNA display was modified from a described method (26). Messenger RNA (≈2 μg) was purified by using biotinylated-oligo(dT) and streptavidin-paramagnetic particles (Promega). The mRNA was eluted in water for cDNA synthesis using AMV reverse transcriptase and RNase H and DNA Polymerase I (GIBCO/BRL). AFLP-cDNA display was performed according to the manufacturer's recommendations (GIBCO/BRL). In brief, the double-stranded cDNA (≈500 ng) was digested with EcoRI and MseI and ligated to respective oligo primers (adaptors). The ligated products were subjected to preselective amplification (20 cycles) by using primers with one restrictive base at the 3′ end of the primers and selective amplification by using primers with two restrictive bases at the 3′ end. One of the primers was end-labeled with [γ-³²P]dATP. The generic primers are EcoRI: 5′-GACTGCGTACCAATTCNN-3′ and MseI: 5′-GATGATTCCTGAGTAANNN-3′. The PCR cycles used for selective amplification consist of 12 cycles of 94°C for 30 s, 65°C for 30 s, and 72°C for 1 min and 20 cycles of 94°C for 30 s, 56°C for 30 s, and 72°C for 1 min. The annealing temperature between the first and 13th cycles was linked by a touch-down phase of decreasing 0.7° in each cycle. After amplification, the products were resolved in a 6% polyacrylamide gel as described (23).

To determine the identity of the cDNA fragments, the fragments of interest were excised and eluted in TE (10 mM Tris/1 mM EDTA, pH 8.0) and reamplified by using the same primer pairs as in display analysis. Sequencing reactions were performed by using the purified PCR products as template in the dideoxy chain termination method (27) in an Applied Biosystems 377 sequencer or manually. For duplicate alleles, individual cDNA fragments were cloned in pGEM vector (Promega) and sequenced. About 10–20 inserts from each gene were sequenced from each locus. Sequence alignment was analyzed by using DNASTAR or CLUSTAL W software. Reverse transcriptase (RT)–PCR was performed according to the SuperScript One-Step RT-PCR kit (GIBCO/BRL) with 40 cycles of 94°C for 15 s, 50–65°C for 30 s, and 72°C for 1 min. Four primer pairs for the genes of interest were listed below: (i) RFP, forward 5′-GGCCAGAATGCCACATTATATC-3′ and reverse 5′-CTAGACCAGTGCCTCTGATTTAC-3′; (ii) TCP3, forward 5′-GGTCCACCTTTTCCTAATCAAAC-3′ and reverse 5′-TAGCTTCAAGTGGGGTTAAAGGT-3′; (iii) HYP1, forward 5′-AGAAGATCCGGATTGCTACAGA-3′ and reverse 5′-TAACCGAACTCTCGTCCGTAGAT-3′; and (iv) Mdh, forward 5′-GTTGGTCACATCAACACCAGAT-3′ and reverse 5′-AAAGTCTGAGAGTGGTCCCAAGT-3′. The amplified products were analyzed by using cleaved amplified polymorphic sequence (CAPS; ref. 28).

To discriminate between orthologous transcripts in polyploid genomes, we applied AFLP-cDNA technology (26). This technique employs the use of restriction polymorphism in analyzing cDNA samples. After the cDNA fragments are restricted, they are ligated onto primers with compatible restriction ends. Ligated products can be amplified by PCR and resolved in a sequencing gel. Any fragments that are present in one sample but not the other are isolated and sequenced. The AFLP-cDNA display procedure was optimized to produce constant results in a diploid A. thaliana and its isogenic autotetraploid (2n = 4x = 20) line. Using a given primer combination in four separate experiments, we were able to obtain consistent results between the two lines in each experiment (Fig. 1a). As expected, the majority of genes in diploid and autotetraploid lines were expressed at similar levels (Fig. 1a). However, differential expression patterns of a few genes were observed and verified by RT-PCR (H.-S.L. and Z.J.C., unpublished results). After optimization, we applied the technique to A. suecica and its diploid progenitors, A. thaliana and C. arenosa. A typical display analysis resulted in three expression patterns among the allotetraploid and its progenitors (Fig. 1b). First, no restriction polymorphism was detected between the same-size alleles (open circles). Second, some alleles displayed clear polymorphism and coexpression patterns. For example, sequence analysis of two individual alleles (arrows) confirmed that the orthologous alleles of the gene encoding acylthioesterase (GenBank accession no. Z36911) from A. thaliana and C. arenosa were coexpressed in A. suecica (data not shown). Third, some alleles exhibited polymorphism, but no obvious patterns of coexpression (open squares). Of the 4,428 cDNA fragments assayed, 450 (11%) cDNAs fell into this category. We further sequenced and analyzed individual fragments from these alleles. Sequencing results for three cDNA fragments indicated that they were orthologous alleles encoding TCP3 (29, 30). The single nucleotide substitution (Fig. 1c) allows us to determine that the expressed allele in A. suecica was from A. thaliana. Twenty-five (22.7%) of the 110 cDNA fragments that were sequenced exhibited differential expression patterns of parental genes in the allotetraploid A. suecica. The remaining 85 cDNAs were coexpressed as determined by RT-PCR analysis (data not shown), although they exhibited restriction polymorphism between the orthologous alleles.

We identified and confirmed a set of ten different genes that were differentially expressed in A. suecica and its progenitors. Based on predicted protein sequences, the functions of the ten genes fell into several categories (Table 1), including four transcriptional factors for RNA Polymerase II, a putative transposase, an untranslated RNA, and an unknown gene. The gene was classified as unknown because it is located in an intron region based on current annotation. However, the transcripts were detected by cDNA display and matched ESTs in the databases. We did not observe genes that were expressed only in the allopolyploid but not in the diploids. The data suggest that genes involved in various biological functions are subjected to differential expression in an allotetraploid, contrary to the notion that silenced genes in polyploids contain mainly transposons (17, 18).

We discovered that silencing involves genes encoding several classes of transcriptional factors, including Bell homoeodomain 2, NAC protein, transcription factor IIB, and TCP3 family (29). TCP3 [Teosinte branched 1, Cycloidea 1, and proliferating cell nuclear antigen (PCNA) factors, PCF1 and 2] are a newly discovered family of transcriptional factors that share a common motif in DNA binding domains and bHLH structure, presumably involved in the activation of PCNA during DNA replication and cell division (29). Genetic mutation of Teosinte Branched 1 causes axillary meristem development in maize (31), resulting in a stem structure like that of its ancestor, teosinte. DNA-methylation-dependent epigenetic mutation in Cycloidea 1 (CYC1) is correlated with asymmetrical flower development in Antirrhinum (30), suggesting an important role of epigenetic regulation of gene expression in plant morphology and evolution.

To gain a molecular understanding of chromosomal organization and gene silencing in polyploids, we analyzed genome structure of TCP3. There are two related TCP3 genes in Arabidopsis, one in chromosome 1, another in chromosome 3 (29). Two other ESTs with low homology were also identified in chromosomes 2 and 4, respectively. The cDNA fragment we cloned and sequenced was A. thaliana TCP3 (AtTCP3) in chromosome 1, which has the highest homology to other members in the family (29). AtTCP3 is located in the BAC (F12 M16) that has been sequenced and annotated (32). The BAC has a 130,235-bp insert containing 30 genes (Fig. 2a), 11 of them on the top strand and 19 on the bottom. TCP3 is located on the bottom strand in the middle of the BAC insert. Five genes in the vicinity of TCP3, including three genes on the bottom strand and two genes on the top strand, were further analyzed. Ring finger protein (RFP) is a large family of transcription factors involved in various pathways of gene regulation. One member of this family is COP1, a protein that interacts antagonistically with HY5, which in turn binds to the promoters of light-inducible genes (33). The other three genes in the vicinity of TCP3 encode two hypothetical proteins (HYP1 and HYP2) and a mitochondrial NAD-dependent malate dehydrogenase (MDH), respectively.

The expression profiles of genes in the vicinity of TCP3 are shown in Fig. 2b. By using gene-specific primers in RT-PCR assays, the RFP was amplified in all three lines dependent on reverse transcription (Fig. 2b, compare lanes 2–4 and 5–7). To discriminate between the progenitor transcripts, the amplified cDNA fragments were analyzed by using cleaved amplified polymorphic sequence (CAPS) analysis (28). The A. thaliana RFP was cleaved by XhoI (lane 8), whereas the C. arenosa RFP fragment was uncut (lane 10). A. suecica, like C. arenosa, possessed only the uncut fragment (lane 9), suggesting that the A. thaliana RFP was not expressed in A. suecica. A trivial explanation is the mutation or loss of the A. thaliana RFP in A. suecica. Results obtained from DNA blot analysis (see below) and PCR amplification using genomic template DNA (data not shown) ruled out this possibility.

Our sequencing results indicated that AtTCP3 was expressed in A. suecica, whereas CaTCP3 transcripts were absent (Fig. 1c). This result was confirmed by RT-PCR analysis. Based on KpnI-digested patterns of the cDNA fragments, only A. thaliana-like transcripts were present in A. suecica (Fig. 2b, lanes 14–16). A trivial explanation is heterozygosity of the CaTCP3 alleles in A. suecica, because natural C. arenosa is an out-crossing autotetraploid. However, reactivation of the silenced CaTCP3 in A. suecica (see below) ruled out his possibility.

Our data indicate that the silencing direction of RFP is different from TCP3, even though the two genes are closely located and separated by only two other genes, HYP1 and HYP2. RT-PCR analysis indicated that HYP1 along with HYP2 (data not shown) from both parental origins were expressed in A. suecica (lanes 11 and 12). In addition, MDH, located upstream of TCP3, was coexpressed in A. suecica (lanes 17–19), although the adjacent TCP3 is expressed from A. thaliana origin. The data suggest that expression patterns of parental genes in polyploid genomes are complicated even in a small chromosomal domain.

It has been shown that the silenced rRNA genes in an allotetraploid are maintained by DNA methylation and histone deacetylation (23). Because chromatin modification is a general mechanism for gene regulation, we predict that silenced genes are associated with DNA methylation. To test this hypothesis, we examined the DNA methylation status of the RFP in A. suecica and its two progenitor species. There are 10 HpaII/MspI (CCGG) sites in the promoter and coding sequences of AtRFP (Fig. 3a). The locus-specific methylation status was examined by using a double digestion of the genomic DNA with XbaI and HpaII or MspI. HpaII and MspI are isoschizimers that recognize CCGG sites. However, HpaII will not cut if the inner cytosine is methylated. As predicted, XbaI digestion alone produced a 5.2-kb fragment in A. thaliana (Fig. 3b, lane 1) and a 9.5-kb fragment in C. arenosa (lane 3). Both fragments were present in A. suecica (lane 2), ruling out a possibility of gene loss during polyploid evolution. After the second digestion with MspI, three small fragments were detected in A. thaliana (lane 4). A single fragment detected in C. arenosa (lane 6) indicates fewer MspI sites present in C. arenosa than in A. thaliana. Again, A. suecica-RFP profiles matched a combined pattern of the A. thaliana and C. arenosa genes. A 540-bp, instead of 550-bp, fragment was detected in A. suecica (lane 5), suggesting that one site is accessible to MspI but not HpaII, because of demethylation in the allotetraploid. HpaII-digestion generated a series of fragments in diploids, indicating that RFP is partially methylated in A. thaliana (compare lanes 4 and 7) and C. arenosa (compare lanes 6 and 9). The 4.2-kb fragment detected in A. suecica (lane 8) is derived from A. thaliana (see below), suggesting that, except for the last HpaII site, AtRFP is completely methylated in A. suecica. The C. arenosa-RFP, however, was less methylated (lane 8). Moreover, a 1.1-kb fragment (lane 8) detected in A. suecica is derived from the C. arenosa. A simple explanation is that some CCGG sites of the C. arenosa locus are demethylated in A. suecica compared with the orthologous loci in the diploid progenitors.

To further investigate whether methylation is correlated with RFP activity, we examined RFP methylation patterns in A. suecica treated with aza-dC, a chemical inhibitor for DNA methyltransferases (Fig. 3c). After the treatment, methylation of the HapII sites in the thaliana-RFP locus was eliminated, resulting in several small fragments, as observed after MspI-digestion (compare lanes 3–6).

As a result of demethylation by aza-dC, the silenced A. thaliana-RFP was reactivated in A. suecica (Fig. 4a, lane 12). Moreover, the silenced TCP3 was also reactivated by aza-dC (Fig. 4b, lane 20), and associated with hypermethylation (data not shown). Parental origins of the silenced genes in the two loci are different. In RFP locus, the A. thaliana-like allele is silenced, whereas in TCP3 the C. arenosa allele is silenced. The above data indicate that one of the orthologous alleles of RFP or TCP3 is methylated and silenced in A. suecica allotetraploid, whereas the active genes are hypomethylated and expressed. We conclude that the reactivation of the silenced genes is dependent on demethylation (Fig. 3), but independent of parental origin.

AFLP-cDNA display is a powerful tool to screen for parental genes that are differentially expressed in A. suecica, an allotetraploid derived from A. thaliana and C. arenosa. Compared with DNA microarray, this method is relatively inexpensive and requires no prior sequence information. Using this method, we estimated that about 2.5% of genes (25 of 110 cDNA fragments sequenced that represent 11% of displayed cDNA) were subjected to silencing. The number may be underestimated because orthologous alleles without the EcoRI/MseI polymorphism were not examined.

When two different genomes are combined into a single cell, they must respond to the consequence of genome duplication, especially multiple copies of genes with similar or redundant functions. The majority of orthologous genes are coexpressed, whereas some are silenced. There are two possible models to explain gene silencing in polyploid genomes. First, genes are silenced or lost because of mutations (including deletions, insertions, and rearrangements) of the DNA sequences during evolution (6, 7). Indeed, many isozyme loci are lost or down-regulated during polyploidization (34–37). It was estimated that in the salmonid and cyprinid fish, the loss of duplicate isozyme loci could be as high as 35–65%, suggesting that loss of duplicate gene function is common after polyploidization (10, 13), which occurred 50 million years ago. Second, expression of a progenitor's genes is epigenetically controlled. We hypothesize that during interspecific hybridization or the early process of polyploidization, epigenetic mechanisms reprogram expression of orthologous genes in biological pathways so that a polyploid cell can adjust properly during development. Compared with genetic mutations, epigenetic regulation provides an effective and flexible means for the cell to respond to genome and gene duplication or “genomic shock” (14), because it is established or erased relatively easily (38, 39). For example, one parental set of rRNA genes can be subject to silencing during vegetative growth by chromatin modification and then reactivated during flower development (24). Although the role of developmental regulation in silencing duplicate genes is speculative, this type of regulation has advantages for selection and adaptation of the newly formed polyploid plants. For instance, there is evidence that some protein heterodimers may not function as well as homodimers or vice versa (40, 41). Thus, a silencing strategy could balance the advantage and disadvantage of having multiple copies of orthologous genes or gene products (e.g., transcriptional factors) in a polyploid cell. It is notable that many of the genes we identified encode transcriptional factors important for many cellular processes and morphological development. The epigenetic regulation of heterologous protein production might compete for fitness under various environmental cues or physiological and developmental conditions. It is likely that some of the silenced genes may be needed at a certain time during plant development. The flexibility of epigenetic regulation would allow polyploid genomes to “flip” some genes on and off in response to environmental cues and developmental changes. Indeed, the reactivation of silenced rRNA genes in allotetraploid Brassica napus during flower development (24) may be a signal for rapid cell division and protein synthesis needed for flowering. The epimutation in CYC1 causes heritable mutation for developing dorsoventral asymmetry flowers in Antirrhinum, which is one of the old flower mutations found in Linaria vulgaris described by Linnaeus (30). It is notable that TCP3, a homolog of CYC1, from C. arenosa origin is silenced by hypermethylation in the allotetraploid, suggesting that epigenetic control is involved in the formation of plant form and function.

Silencing of endogenous genes by chromatin modification during polyploid formation is reminiscent of X chromosomes inactivation in mammals, in which almost all of the genes from one of the two X chromosomes are silenced. Silencing is initiated by untranslated RNA at the locus of Xist (42) and Tsix (43) (transcribed in opposite directions) and maintained by DNA methylation and histone modifications. Apparently, the silenced X chromosome can be reactivated during reproductive stages, while every egg carries a viable X chromosome. The choice of which X chromosome is activated is determined by imprinting of the Xist from maternal origin in early embryo proper (43, 44). In the epiblast lineage, however, an X chromosome is randomly silenced. The silencing schemes observed in polyploid genomes are distinct from X chromosome inactivation, however. First, silencing does not occur at every locus in a particular chromosome or even in a small chromosome segment. Second, although many genes are coexpressed in the duplicate genomes, only 2.5% of genes we studied are subjected to differential expression. Third, although the choice of selecting which set of rRNA genes to silence is not random, imprinting was not involved in silencing the uniparental set of rRNA genes (24). Finally, the silencing mechanism itself may be different. If chromatin is involved, the chromatin state of every gene needs to be remodeled. How and why the polyploid genomes choose to “flip” on and off the expression of some parental genes after polyploid formation remains an interesting topic in polyploid biology.