Tandem repetitive transgenes and fluorescent chromatin tags alter local interphase chromosome arrangement in Arabidopsis thaliana

In situ localization and direct in vivo visualization of distinct chromosome regions recently became feasible by chromatin-tagging systems. The lac operator/lac repressor system (Robinett et al., 1996; Straight et al., 1996) for instance uses a bacterial DNA binding protein (lac repressor) that, when fused with a green fluorescent protein (GFP) and a nuclear localization signal peptide (NLS), binds to the 256 copies of directly repeated lac operators (∼10 kb) integrated at a specific chromosome locus. Binding at the target locus yields a fluorescent spot of higher intensity than the overall fluorescence of dispersed unbound GFP molecules in the nucleoplasm. The GFP-lac repressor protein is either transiently or stably expressed. The system was applied to various eukaryotes such as yeasts (Aragon-Alcaide and Strunnikov, 2000; Nabeshima et al., 1998; Straight et al., 1996), flies (Gunawardena and Rykowski, 2000; Vazquez et al., 2002), cultured mammalian cells (Robinett et al., 1996; Tsukamoto et al., 2000) and plants (Esch et al., 2003; Kato and Lam, 2001; Matzke et al., 2003). It revealed structural dynamics of chromosomes in interphase as well as mitotic cells by tracing the tagged loci in living cells and contributed to the understanding of chromosome functions (reviewed by Belmont et al., 1999; Gasser, 2002; Lam et al., 2004). However, tagging systems represent artificial chromosome loci and sometimes generate unusual nuclear protein localization. For instance in baby hamster kidney cells in which a lac operator array is amplified about ten times, a nuclear protein complex, the PML (promyelocytic leukemia) body, is formed at the integration locus (Tsukamoto et al., 2000). PML bodies are thought to play a role in regulating transcription. Because PML bodies are not formed at the transgene locus without accumulation of the lac operator binding fusion protein, they are thought to recognize high concentrations of induced foreign proteins (Tsukamoto et al., 2000).

Previously, we used the lac operator/lac repressor system to compare chromosome dynamics in Arabidopsis nuclei of different ploidy level (Kato and Lam, 2003), for example in 2C nuclei of guard cells (stomata) and in nuclei of pavement cells (8C on average). We found that chromosomes in interphase nuclei of Arabidopsis move randomly within a restricted area and that the area size correlates with the nuclear DNA content. Lac operator loci seemed to be stochastically associated in diploid as well as in endopolyploid nuclei according to the GFP spot numbers observed (Kato and Lam, 2003). Contrary to this, Esch and colleagues (Esch et al., 2003) reported that tagged lac operator loci in Arabidopsis are tightly, but not stochastically associated with each other because only one GFP spot was observed in most diploid as well as in polyploid cells. Therefore it was necessary to address the question of whether lower-than-expected numbers of GFP spots are indeed due to associations of homologous loci in Arabidopsis nuclei and, if so, to determine why this is occurring.

Here we determined the number of lac operator FISH (fluorescent in situ hybridization) signals before and after expression of the GFP-lac repressor protein and the proportion of loci exhibiting a GFP spot after induced expression. Furthermore we compared the spatial organization of the tagged chromatin in hemizygous and homozygous conditions. The aim was to test whether the repeat structure and/or the expression of the GFP-lac repressor protein of the transgene have an impact on allelic/ectopic homologous pairing and on association with heterochromatic domains of the regions in question in 2C leaf nuclei. We found that the lac operator arrays are associated with each other more often than predicted by computer model simulation (Pecinka et al., 2004). Expression of the GFP-lac repressor protein further increases the homologous association frequency. Because the chromosome regions adjacent to the tagged loci associated with each other as well as with heterochromatin less often than lac operator arrays, we conclude that homologous tandem repetitive transgenes preferentially associate with each other as well as with endogenous heterochromatin in A. thaliana. In independent experiments homologous pairing and association with heterochromatin were studied for a homozygous multi-copy hygromycin phosphotransferase (HPT) transgene locus of ∼100 kb (Probst et al., 2003) and yielded similar results to those obtained for the homozygous lac operator loci.

Plants of A. thaliana accession Columbia (Col) of hemizygous and homozygous EL702C genotypes (Kato and Lam, 2003), homozygous EL700S genotype (Kato and Lam, 2001; Kato and Lam, 2003) or homozygous line A and the mutant mom1-1 genotype that contains the HPT locus homozygously (Probst et al., 2003) were cultivated as described. To induce expression of the GFP-lac repressor protein, young rosette leaves were detached from the plants and floated on 10 ml of 0.3 μM (homozygous EL702C) or 3 μM (wild type, homozygous EL700S) dexamethasone (Dex) solution in Petri dishes for 6-12 hours. After fixation of leaves in 4% paraformaldehyde, 2C nuclei were flow-sorted as described (Jasencakova et al., 2000). Prior to FISH, Dex-treated 2C nuclei were analyzed for the presence of GFP spots and the positions on slides of nuclei with GFP spots were recorded using a microscopic grid slide. Pachytene chromosomes of the wild type and homozygous EL702C genotypes were prepared as described (Lysak et al., 2001).

The GFP signals were detected as described (Kato and Lam, 2003). Spots were defined as pixel clusters comprising more than the mean number of pixels plus 3.3-fold the value of the s.d. in the analyzed nuclei.

The following DNA clones were used as probes: BAC MGL6 (GenBank accession number AB022217), BAC F18C1 (AC011620), BAC T15P10 containing 45S rDNA sequence (AF167571), 128x lacO-SK (Kato and Lam, 2001), pAL1 (Martinez-Zapater et al., 1986), the BAC contig spanning 4.2 Mb of chromosome 3 top arm from F2O10 to MSL1 (AC013454 and AB012247, respectively) and pGL2 (containing the hygromycin phosphotransferase gene under the control of the CaMV 35S promoter). BAC and plasmid DNA was labeled by nick translation (Ward, 2002) or by PCR using biotin-2'-deoxyuridine 5′-triphosphate (biotin-dUTP) (Probst et al., 2003). For chromosome painting, labeled BACs were precipitated and resuspended in 20 μl hybridization buffer (50% formamide, 10% dextran sulphate, 2× SSC, 50 mM sodium phosphate, pH 7.0) per slide. After mounting the probe, the slides were placed on a heat block at 80°C for 2 minutes and then incubated in a moist chamber at 37°C for ∼12 hours. Post hybridization washes and detection steps were conducted as described (Schubert et al., 2001). Biotin-dUTP was detected by avidin conjugated with Texas Red (1:1000; Vector Laboratories), biotin-conjugated goat anti-avidin (1:200; Vector Laboratories), and again with Texas Red-conjugated avidin. Digoxigenin-dUTP was detected by mouse anti-digoxigenin (1:250; Roche) and Alexa 488-conjugated goat anti-mouse antibody (1:200; Molecular Probes). Cy3-dUTP was observed directly. Nuclei and chromosomes were counterstained with 1 μg/ml DAPI in Vectashield mounting medium (Vector Laboratories).

Fluorescence signals in flow-sorted 2C nuclei and on pachytene chromosomes were analyzed using an Axioplan 2 (Zeiss) epifluorescence microscope with 100×/1.4 Zeiss plan apochromat objective. Images were acquired with MetaVue (Universal Imaging) software and a cooled charge-coupled device camera (Spot 2e, Diagnostic Instruments) separately for each fluorochrome using the appropriate excitation and emission filters. The monochromatic images were pseudocolored, Gauss- or median-filtered to reduce noise and merged using Adobe Photoshop 6.0 (Adobe Systems) software. A spatial overlap of compact spheric FISH signals of homologous and/or heterologous sequences was regarded as homologous pairing and heterologous association, respectively. Allelic versus ectopic pairing of transgenic loci was distinguished on the basis of FISH signals obtained from differently labeled BACs that contain sequences flanking the respective transgene insertion loci.

In our previous studies, we rarely observed four GFP spots in 2C guard cell nuclei in homozygous EL702C plants with two loci on chromosome 3 containing lac operator insertions (Kato and Lam, 2003). Here we compare spot numbers in guard cell nuclei of cotyledons of hemizygous and homozygous EL702C plants and of homozygous transgenic EL700S plants (Fig. 1). As EL700S plants contain the same construct as EL702C plants except for the lac operator array, homogeneously distributed GFP signals but no GFP spots were expected in the nucleoplasm. In hemizygous EL702C plants, we found 5% of 92 nuclei without spots, 66% with one spot, 27% with two and 2% with three spots. In homozygous EL702C plants 12% of 197 nuclei had no spots, 34% showed one spot, 37% two, 11% three and 6% four spots. In homozygous EL700S plants, 55 out of 56 nuclei showed no spots and one nucleus (<2%) showed one spot. The single spot observed in an EL700S nucleus and a third spot in two hemizygous EL702C nuclei are most likely caused by spontaneous aggregation of GFP-lac repressor-NLS molecules. Taking into account our not being able to observe the maximum number of GFP spots in all nuclei and in order to understand the nature of GFP-spot appearance patterns in the nuclei, we compared the observed data with the statistically expected percentages for nuclei with different spot numbers. We considered two statistical models.

We previously observed that lac operator repeats with different lengths of the array may be detected as GFP spots with different probabilities (Kato and Lam, 2001). Therefore, in the second model we assumed that GFP spots appear with different probabilities on each tagged locus, possibly due to the different sizes of lac operator array in these loci (a single array in the distal and two lac operator arrays in the proximal insertion locus in the top arm of chromosome 3 of the EL702C line). We first estimated the probability values p₁ and p₂ for the two loci in hemizygous nuclei. On the basis of the observed percentages of hemizygous nuclei with one spot: p₁(1–p₂) + p₂(1–p₁) =0.66 or with two spots: p₁·p₂=0.27, we calculated: p₁=0.30 and p₂=0.93. For homozygous nuclei, p₁=0.24 and p₂=0.55 were calculated similarly by numerical solution of the corresponding equations. With these estimated probabilities, the observed frequencies of hemizygous nuclei with 0, 1 or 2 spots and of homozygous nuclei with 0, 1, 2, 3 or 4 spots fitted very well with the expected values. However, our result suggests that the probability of spot appearance in hemizygous nuclei is higher than those of homozygous nuclei. Although non-homogeneous GFP accumulation in a nucleus or different levels of accumulation in hemizygous compared to homozygous nuclei might explain the lower probability of GFP spot appearance in homozygous nuclei, we preferred the hypothesis that the tagged loci might have intrachromosomal and/or interchromosomal interactions in the nuclei of Arabidopsis, thereby altering the apparent spot frequencies. To test this hypothesis, we analyzed the positional coincidence of the integrated lac operator loci in isolated 2C leaf nuclei by FISH analyses.

In flow-sorted 2C nuclei of homozygous EL702C plants in which expression of the GFP-lac repressor protein was induced with Dex, GFP spots and FISH signals on the lac operator loci were counted (Fig. 2). Nuclei without clear GFP spots were excluded from evaluation. Out of 63 analyzed nuclei, 30% showed one, 35% two, 25% three and 10% four GFP spots. In contrast, 22% of nuclei showed four FISH signals, 35% showed two or three signals and 8% showed one FISH signal. All GFP spots coincided with a lac operator FISH signal (Fig. 2B) but not vice versa. In total 83% of 252 FISH signals coincided with a GFP signal. Thus, 17% of the transgene loci cannot be detected by a GFP spot in Dex-treated homozygous EL702C nuclei under the applied conditions. Either not all lac operator arrays were accessible to the GFP-lac repressor proteins or some GFP spots could not be discriminated owing to a high overall fluorescence intensity and/or rapid bleaching of signals within a minute of exposure. We hypothesized that less than four FISH signals per nucleus may be due to ectopic or allelic alignment of the lac operator arrays.

The transgenic line EL702C carries three transgenes (each ∼17 kb) inserted at two independent loci on the top arm of chromosome 3, ∼4.2 Mb apart; the proximal locus harbors two transgenes in inverse orientation (Kato and Lam, 2003). FISH analyses with BAC probes containing insert sequences internally flanking the transgene loci were used to quantify ectopic and allelic pairing frequencies of the lac operator arrays in 2C nuclei. To confirm the physical position of the transgene and BAC loci on chromosome 3, we first hybridized these probes to pachytene chromosomes of EL702C. The lac operator probes hybridized to the predicted locations whereas the BAC probes hybridized in reverse order suggesting an inversion of the region between the transgene loci. Also FISH with two differently labeled BAC pools (MBK21 to MSL1 corresponding to the upper region and F2O10 to F28J15 corresponding to the bottom region between the insertion loci) yielded signals of reversed orientation on pachytene chromosomes of homozygous EL702C compared to that in wild-type plants (Fig. 3). FISH signals of the BACs flanking the insertion loci externally appeared in the same order on the wild-type and EL702C bivalents (not shown). These results confirmed the inversion between the transgene loci on the top arm of chromosome 3 in EL702C plants.

Using tri-color FISH with BAC MGL6 (79.5 kb, ∼54 kb downstream of the insertion, red) flanking the distal locus, BAC F18C1 (100.8 kb, ∼55 kb upstream of the insertion, yellow) flanking the proximal locus and lac operator probe (green) (Fig. 4B-D), the ectopic and allelic pairing frequency of the lac operator arrays was assessed in 60 hemizygous untreated, 62 homozygous untreated and 59 homozygous Dex-treated EL702C nuclei. We classified the lac operator array alignments into two different pairing types. If two signals (MGL6, red and F18C1, yellow) colocalized with a lac operator signal (green), we identified the alignment as ectopic pairing. If all signals of either MGL6 or F18C were colocalized with a lac operator signal, we identified the alignment as allelic pairing. In hemizygous nuclei, ectopic pairing was detected for 13% of the lac operator loci without Dex treatment. In homozygous EL702C nuclei without Dex treatment, ectopic pairing was observed for 27% of the lac operator array loci and allelic pairing for 34% of the loci. With Dex treatment, these values increased to 35% (ectopic pairing, P=0.052) and 45% (allelic pairing, P=0.017), respectively.

Using the BACs F18C1 (yellow) and MGL6 (red) that flank the transgene insertions in EL702C nuclei as probes, two-color FISH was conducted to monitor the association of these regions in wild-type nuclei (n=153), hemizygous (n=60) and homozygous (n=62) EL702C nuclei without Dex treatment. We also analyzed 61 wild-type nuclei and 59 homozygous EL702C nuclei after Dex treatment (Fig. 5). Homologous pairing of both regions as well as their heterologous association occurred without significant differences (i.e. 3-6% per locus, see Fig. 5A) in wild-type nuclei and hemizygous EL702C nuclei, irrespective of Dex treatment. In homozygous EL702C nuclei, homologous pairing (10%, P<0.05 for MGL6 and 14%, P<0.001 for F18C1) and ectopic association (9%, P>0.05) occurred more often than in the wild type. A further increase of homologous pairing (20%, P=0.032; and 22%, P=0.177, respectively) as well as of ectopic association (17%, P=0.013; Fig. 5A) was found after induction of GFP-lac repressor protein expression in homozygous EL702C nuclei (all at P<0.001 when compared to levels in the wild type).

During the FISH analysis described above, we noticed a frequent spatial association of lac operator loci with heterochromatic chromocenters that are detected as strongly DAPI-stained regions. Therefore we determined the frequency of positional overlap (colocalization) of FISH signals of F18C1, MGL6 and lac operator probes with strongly DAPI-stained chromocenters in homozygous EL702C nuclei without (n=41) and with (n=31) Dex treatment. For comparison, the overlap of FISH signals of MGL6 and F18C1 probes with those of centromeric repeats and 45S rDNA, the main components of heterochromatin in Arabidopsis (Fransz et al., 2002), were monitored in 62 wild-type nuclei (Fig. 6). Although 8-14% of MGL6 and F18C1 FISH signals colocalized with heterochromatin in all types of nuclei tested, 37% of lac operator signals overlapped with chromocenters in untreated and 44% in Dex-treated homozygous EL702C nuclei (both P<0.001 when compared to the flanking regions). Apparently, the colocalization of lac operators with heterochromatin did not interfere with expression of the GFP-lac repressor protein in homozygous EL702C nuclei although the lac repressor gene is situated close to the lac operator array (Fig. 6). In order to test whether pairing of lac operator loci precedes association with heterochromatin, we counted the number of lac operator loci per overlap with a chromocenter. Within 31 Dex-treated homozygous EL702C nuclei (harboring 124 loci), 54 loci associated with a chromocenter, of which 14 were detected as a single locus, 24 as two, 12 as three and 4 as four paired loci suggesting that transgene pairing is not a prerequisite for association with heterochromatin.

To test whether or not the higher-than-random allelic pairing frequency of the lac operator array is true only for this sequence, the pairing frequency of the homozygous silent transgenic HPT locus (composed of ∼15 rearranged plasmid copies of ∼100 kb in A. thaliana line A) (Mittelsten Scheid et al., 1991; Mittelsten Scheid et al., 1998) was investigated by FISH (Fig. 7). For comparison, the same locus was tested in a mom1-1 mutant background (Amedeo et al., 2000) where HPT silencing is released without alteration of DNA methylation and histone modifications (Probst et al., 2003). In nuclei of line A, 30% of HPT FISH signals were paired. This value is significantly higher (P<0.001) than the ∼5% of pairing observed on average for FISH signals of BACs with inserts from various endogenous euchromatic regions along the Arabidopsis chromosomes (Pecinka et al., 2004) but not significantly different (P>0.05) from the allelic pairing frequency of transgenic lac operator arrays (34% of loci) in homozygous EL702C nuclei. In mom1-1 nuclei where HPT genes are expressed, association of HPT FISH signals (21%) was still significantly higher than the average pairing frequency (P<0.001). Colocalization with heterochromatic chromocenters was found for 50% of HPT signals in line A and for 49% in mom1-1 nuclei. In 269 line A nuclei, 165 out of the 271 HPT loci colocalizing with chromocenters associated as a single locus and 106 as paired loci. In 355 mom1-1 nuclei, 224 out of 346 loci colocalizing with a chromocenter associated as single and 122 as paired loci. Because 60-65% of heterochromatin-associated loci were not paired, homologous pairing seems not to be a prerequisite for spatial association of HPT loci with chromocenters. Hence the association frequency of the HPT locus with heterochromatin is even higher than that observed for the lac operator arrays, independent of the transcriptional status and of previous homologous pairing.

FISH analyses of the transgenic line EL702C with flanking BACs revealed a previously undetected inversion between the two insertion loci of lac operator arrays on the top arm of chromosome 3. Without sequencing the actual breakpoints (∼10-55 kb away from the insertions) we are currently not able to identify the molecular event responsible for that inversion and to decide on one of the models proposed for the generation of inversions during insertion of two transgenes in cis configuration (Laufs et al., 1999).

Homologous pairing of ∼100 kb regions along different chromosomes of A. thaliana accession Columbia occurs on average in about 5% of somatic nuclei (Pecinka et al., 2004). In wild-type nuclei, allelic pairing and ectopic association of the regions that flank the lac operator loci in EL702C occur with a similar frequency (3-6% per locus). These values are within the range predicted for random appearance of `single point' homologous pairing according to simulations based on the `random spatial distribution' model (Pecinka et al., 2004). The homozygous presence of the lac operator arrays results in a four- to tenfold higher frequency of allelic as well as of ectopic pairing of these loci compared to the average values observed for endogenous sequences in wild-type nuclei (compare values for flanking sequences in Fig. 5A with those for lac operator arrays in Fig. 4A). The high allelic pairing frequency of the transgene may exert a `dragging' effect on the flanking regions (Fig. 5A). In hemizygotes, a dragging effect is not obvious because pairing of the transgene is less frequent than in homozygotes and in most cases FISH signals of flanking regions are separated by those of lac operator loci during ectopic transgene pairing. We also found the repeated HPT locus to be paired significantly more often (21-29%) than expected according to a random frequency that was observed for several endogenous euchromatic loci (Pecinka et al., 2004). From these data we speculate that tandem repetitive sequences promote homologous association in Arabidopsis. Such a tendency for homologous association of tandem repeats could also be the reason for association of multiple transgene insertion loci in wheat nuclei (Abranches et al., 2000; Santos et al., 2002). The dispersion of the wheat transgene loci observed after 5-azacytidine or trichostatin A treatment (Santos et al., 2002) might be due to chromatin modifications rather than to transcriptional activity (see below). Expression of the GFP-lac repressor protein in homozygous EL702C nuclei yielded a further increase of allelic and ectopic pairing of the transgene locus by an additional 5-10% (Fig. 4A), with an additional dragging effect on the flanking regions (Fig. 5A). Expression of HPT in the mom1-1 background does not increase homologous pairing of the transgene locus containing the HPT repeat. We speculate that GFP-lac repressor protein binding to the lac operator arrays, rather than just expression of the transgene, enforces allelic and ectopic pairing of the lac operator arrays. Wild-type lac repressor (tetramerizing form) can bind lac operators on different DNA molecules, tethering together loci on different chromosomes (Straight et al., 1996; Weiss and Simpson, 1997). Because we used a dimerizing mutant form of the lac repressor (Kato and Lam, 2001), which can bind only one lac operator site (Robinett et al., 1996), the capability of tethering two chromosomes should be minimized in EL702C. Nevertheless, spontaneous association of GFP-lac repressor protein molecules bound to different lac operator loci might increase the pairing frequency. Previously, we reported that movement of tagged chromatin in Arabidopsis nuclei, in spite of being spatially constrained, may span up to 0.44 μm within 10 minutes (Kato and Lam, 2003). Because homologous chromosome regions of ∼100 kb are either paired or separated by less than 0.2 μm in ∼20% of Arabidopsis nuclei on average (Pecinka et al., 2004), it seems reasonable to assume that during the 12 hours of Dex treatment, random associations of lac operator sites may occur and they then become stabilized due to aggregation of GFP-lac repressor proteins.

In contrast to the FISH signals of the lac operator array flanking regions, of which 8-14% overlapped with heterochromatin, signals of the inserted lac operator arrays colocalized as single or paired loci with chromocenters in 37% of untreated and in 44% of Dex-treated homozygous EL702C nuclei. As lac operator arrays may associate with heterochromatin as single or as paired loci, this colocalization does not depend upon pairing of the repetitive transgene arrays. Probably, tandem repeat loci tend to associate with each other on the basis of sequence homology but also with heterochromatic chromocenters containing other repeat sequences. This would render tandem repeats better candidates for anchoring euchromatin loops to heterochromatin according to the `chromocenter-loop model' (Fransz et al., 2002) than dispersed repeats such as Emi12 elements that colocalize with chromocenters only in 1-7% of nuclei (S. Hudakova, IPK, Gatersleben, Germany and I.S., unpublished results). In total the HPT locus is clearly larger than the lac operator locus and often becomes visible as DAPI-intense chromocenter(s) (Fig. 7) (Probst et al., 2003). Because the HPT locus colocalized more often than the lac operator locus with heterochromatin, the tendency of tandem repeats to associate with heterochromatin in Arabidopsis interphase nuclei may correlate with the size of the entire repeat-containing locus. For the HPT locus the association with heterochromatin was independent of transcription. In the case of the lac operator locus, transcriptional activity of adjacent sequences does not reduce the frequency of its colocalization with endogenous heterochromatin. The mechanism by which repeat sequences are targeted to chromocenters remains to be elucidated.

Finally, our findings suggest that in many nuclei the lac repressor/lac operator chromatin-tagging system does not reflect the spatial organization at the integration loci under wild-type conditions and may lead to invalid conclusions as to single-point homologous pairing frequencies (Esch et al., 2003). This problem could become significant especially when multiple insertions of repetitive arrays are present either in hemizygous or in homozygous conditions. The main reason for the increase in allelic and ectopic association frequency of the lac operator (compared to the flanking sequences in wild-type conditions) is most likely the repetitive nature of the transgene construct. Sequence-specific but more or less location-independent somatic association of multiple inserted arrays of tet operator and lac operator has been reported for budding yeast (Aragon-Alcaide and Strunnikov, 2000), although this was not confirmed by FISH or in the absence of fusion protein. For the same organism, association of tet operator arrays was shown to depend on the expression of the tet repressor fusion protein (Fuchs et al., 2002). In Drosophila, lacO arrays apparently did not reveal a tendency for homologous pairing as it was possible to trace extensive separation of homologues and even of sister chromatids along chromosome arms during pre-meiotic mid-G2 (Vazquez et al., 2002). Our results obtained for the HPT locus further support the idea that in A. thaliana the tandem repetitive nature of a transgene locus might be responsible for an increased allelic and ectopic pairing frequency of homologous transgenic sequences as well as for an increased colocalization frequency with endogenous heterochromatin. GFP-lac repressor proteins tagging such loci may further enhance their tendency for homologous association. Future studies will show whether DNA methylation and histone modifications have an impact on homologous pairing and heterologous association of interstitial tandem repeats and whether such loci represent hot spots for somatic recombination, for example after genotoxin exposure.

We thank Jörg Fuchs, Rigomar Rieger, Jerzy Paszkowski and the anonymous referees for helpful suggestions and critical reading of the manuscript and Joachim Bruder for technical assistance. This work was supported in part by grants of the Deutsche Forschungsgemeinschaft (Schu 951/8-2 and 10-1) to I.S. and by a grant from the Plant Genome Research Program of NSF (#0077617) to N.K. and E.L.