Introduction

The satellite repetitive sequences are major components of noncoding sequence regions of higher eukaryote genomes (Charlesworth et al. 1994; Csink and Henikoff 1998). Satellite DNAs are repeats of 100– to 800–bp unit sequences, usually organized as tandem repeats. They can be highly variable within and between species, and repeats on the same chromosome vary. Satellite DNAs were once considered to be nonfunctional selfish elements that increase and decrease their frequency without any advantage or disadvantage for organisms. However, some satellite families have specific functions and are under natural selection. Centromeric satellite sequences are one such functionally important type of satellite in many eukaryote species (Csink and Henikoff 1998; A.E. Hall et al. 2004).

Centromere regions of various eukaryotes, both plants and animals, consist of highly repetitive satellite sequences (Choo 1997), although many exceptions have been reported (Lo et al. 2001; Saffery et al. 2003; Nagaki et al. 2004). Centromeric satellite sequences in many species have an important role for centromere functions, including chromosome segregation and chromosome positioning during interphase (Choo 1997). Misorientation or missegregation of chromosomes with nonfunctional centromeres causes aneuploidy and is thus severely deleterious for the offspring. Although it is not yet completely clear what determines that a region acts as a centromere, the DNA sequences and chromatin structure, including association with specific proteins, are all considered important. Species-specific centromeric satellite families appear to coevolve rapidly along with adaptive evolution of centromeric proteins (Csink and Henikoff 1998; Henikoff et al. 2001). Mutations in the relevant centromeric proteins could alter their binding to centromere satellite repeat variants, and copy numbers of favored satellite variants will increase over the generations. Variants in centromeric satellite sequences include not only nucleotide substitutions, but also changes caused by other mechanisms such as gene conversion, replication slippage, and unequal crossing-over (Charlesworth et al. 1994). The centromere regions of many organisms have reduced crossover frequencies (Choo 1998; Copenhaver et al. 1998), but exchange of DNA sequences may still occur. Gene conversion has been inferred in heterochromatic genome regions with low crossover frequencies (Jensen et al. 2002), and centromeric satellite repeats sometimes show clear evidence of unequal exchanges (Cabot et al. 1993).

In the present study, we analyzed nucleotide variation in three centromeric satellite families in Arabidopsis halleri and lyrata subspecies, which are the closest relatives of A. thaliana (Miyashita et al. 1998; Koch et al. 2000). A. halleri ssp. gemmifera, and A. lyrata ssp. kawasakiana have three different centromere satellite families (Kawabe and Nasuda 2005), with some chromosome specificity. pAa (similar to the centromere satellite found on all the A. arenosa chromosomes) is the only sequence on four of the eight A. halleri ssp. gemmifera chromosomes, as well as being found on one chromosome together with pAge1; pAge1 is also the sole satellite family on another chromosome, and the other two chromosomes carry exclusively pAge2 (Kawabe and Nasuda 2005). Of the three satellite families, pAa shows the greatest similarity to the A. thaliana pAL1 family, and we show evidence below (see Discussion) that pAa is older than pAge1 and pAge2.

Three related species studied previously, A. thaliana (Martinez-Zapater et al. 1986; Maluszynska and Heslop-Harrison 1991; Murata et al. 1994), A. arenosa (Kamm et al. 1995), and A. griffithiana (Heslop-Harrison et al. 2003), each have only one major satellite family on all the chromosomes. Recent sequencing of long stretches of satellite regions from BAC clones of A. thaliana relatives revealed that species-wide variability is higher than that within BAC clones in all species analyzed (S.E. Hall et al. 2005). Sisymbrium irio satellite sequences are unusually BAC clone-specific (species-wide sequence identity is 83%, versus a mean of about 90% within BACs [S.E. Hall et al. 2005]). FISH analyses using BAC clone probes showed chromosome-specific signal localizations similar to those of the A. halleri ssp. gemmifera satellite families. However, in other species studied, satellite identities are similar at both scales (about 90% either species-wide or within BAC clones [ S. Hall et al. 2005]). This is similar to the findings within satellite families of A. halleri ssp. gemmifera, but the A. halleri ssp. gemmifera satellite families are much more highly diverged (divergence is about 30% [Kawabe and Nasuda 2005]). Thus the variety of satellite families in A. halleri and A. lyrata is unusual, perhaps indicating different selective histories, and/or different functionalities, of the three sequence families.

If these satellite sequences have centromere functions, and if natural selection operates to eliminate or increase variants of these repeat families, previously dominant families that are currently being eliminated from a genome should consist of old satellite sequences and might have high diversity and even non-species-specific variants. In contrast, favored families, which may have increased in abundance very recently, should have lower diversity and their variants should be species-specific and, perhaps, chromosome-specific. The main purpose of this study is to analyze differences in variation in the A. halleri versus A. lyrata satellite families and to investigate the evolutionary mechanisms involved.

Materials and Methods

Plant Materials

A. halleri ssp. gemmifera from Ashibi (Kyoto, Japan), A. halleri ssp. halleri from Pontresina (Switzerland), A. lyrata ssp. lyrata individual Ontario4 (Ontario, Canada), and A. lyrata ssp. petraea plant 99R11-2 (Esja mountain, Iceland) were used for analyses of all three satellite families. In addition to these plants, three plants from the Esja mountain population, two from the other three European populations, and one plant from the Ontario population were also used for pAge1 and pAge2 centromere satellite variation analyses. Total DNAs were isolated from dried leaves using a FastDNA kit (Q-BIOgene) according to the manufacturer’s instruction.

Isolation of Centromeric Satellite Families

Three centromeric satellite families were PCR amplified as described by Kawabe and Nasuda (2005). We did not determine the chromosome locations or copy numbers. After agarose gel electrophoresis to identify dimers, trimers, and tetramers, bands were purified and cloned using the TOPO TA cloning kit (Invitrogen). Sequences of cloned PCR products were determined using the DYEnamic ET sequencing kit (Amersham). In the present study, sequences of at least 10 complete satellite units were newly obtained for each of the three satellite families in each species. DNA sequences of the newly determined centromere satellites were deposited in the GenBank databank with the accession numbers DQ872189–DQ872369.

Data Analyses

Sequences were aligned with the forward primer 5′ end as the first nucleotide. Unless otherwise specified, each unit from dimer, trimer, or tetramer sequences was treated as a separate sequence in the alignment and in our analyses (i.e., a dimer was treated as two monomer units; as described below, a few analyses of dimer and tetramer sequences of the pAge2 family treat dimers as the units, and this will be explicitly mentioned when it is relevant). Table 1 reports the lengths of sequence units excluding alignment gaps. The previously reported sequences, 5 pAa sequences from A. arenosa (Kamm et al. 1995) and 6 pAa, 19 pAge1, and 4 pAge2 sequences from A. halleri ssp. gemmifera (Kawabe and Nasuda 2005) were also included in the analyses, as well as 257 complete units of the pAa family of A. arenosa from S. Hall et al. (2005) whose sequences contained no ambiguous sites (N in the GenBank sequences).

Table 1. Summary of nucleotide variation of the centromeric satellite sequences within species: divergence values between sequences were estimated with Jukes-Cantor correction
Full size table

For the complex pAge2 family described in detail below, trimer sequences were found in A. halleri ssp. halleri and A. lyrata ssp. lyrata, as well as dimers and tetramer sequences (consisting of two dimers), like those previously observed in A. halleri ssp. gemmifera. We first aligned monomer units of the dimer and tetramer sequences from A. halleri ssp. gemmifera and A. lyrata ssp. petraea. Characteristic differences between the sequences of the first and second monomer units of the dimer sequences (see Results) were then used to classify the pAge2 family sequences into “simple” and “complex.” Simple pAge2 dimers are defined as those with distinguishable first and second units (see Fig. 1), first units evidently resembling first units of other dimers of other simple dimers and second units resembling second units of other dimers. Single nucleotide variants are also often present in these sequences. Complex dimer sequences resemble two units of the trimer sequences; in these sequences, at least one unit includes mosaic or chimeric structures that clearly mix parts of first and second units of simple dimers. One dimer sequence of A. halleri ssp. halleri plus one dimer and one tetramer sequence from A. lyrata ssp. lyrata had mosaic structures like those seen in trimers.

Fig. 1.
figure 1

Schematic diagram of pAge2 sequences, showing “simple” and “complex” dimers and two examples of trimer sequences. The two different units that make up simple dimers are represented by black and hatched lines. Trimer sequences are mosaics of first and second units of simple dimer sequences, and complex dimers are portions of trimer sequences; in the case illustrated, the dimer is composed of units 2 and 3 of the upper trimer sequence.

Full size image

Average numbers of nucleotide differences between sequences within individuals, divergence between species or subspecies, and minimum numbers of recombination events (Hudson and Kaplan 1985) were estimated with the DnaSP program (Rozas and Rozas 1999). Numbers of gene conversion events were estimated using GENECONV ver. 1.81 (Sawyer 1999). A few sequence units having deletions longer than 9 bp were excluded from these analyses. Phylogenetic trees were constructed by the neighbor-joining (NJ) method using JC-distances and pairwise deletion, using the MEGA2 program package (Kumar et al. 2002).

Results

Existence of Three Centromeric Satellite Families in A. lyrata and A. halleri Subspecies

We first tested for the presence of each of the three centromeric satellite families in a single A. lyrata ssp. petraea plant from Iceland, an A. lyrata ssp. lyrata plant from Ontario, and an A. halleri ssp. halleri plant from Switzerland, using PCR with the primers described previously (Kawabe and Nasuda 2005). All three families (pAa, pAge1, and pAge2) previously found in A. halleri ssp. gemmifera and A. lyrata ssp. kawasakiana were obtained in ladder patterns typical for tandemly repeated sequence arrays (Fig. 2A). We sequenced multiple units from several plants from each of the taxa studied. The patterns from pAa and pAge1 amplifications are similar for all the taxa.

Fig. 2.
figure 2

PCR results for three centromeric satellite families for A. halleri ssp. halleri, A. lyrata ssp. petraea, and A. lyrata ssp. lyrata. A One plant each from A. halleri ssp. halleri, A. lyrata ssp. petraea, and A. lyrata ssp. lyrata for the pAa, pAge1, and pAge2 families. B pAge1 and pAge2 amplification of A. lyrata populations. The pAge2 family primers specifically amplify single units of the dimer sequence (Kawabe and Nasuda 2005).

Full size image

Complex Structure of pAge2 Family Sequences

For pAge2 the PCR results differ between the taxa studied. The previous study of A. halleri ssp. gemmifera and A. lyrata ssp. kawasakiana found only even numbers of pAge2 repeat units (dimers), and PCR using primers specific for single units confirmed this (Kawabe and Nasuda 2005). In A. lyrata ssp. petraea, PCR with the same primers showed the same strong amplification of even-numbered bands, indicating a predominance of “simple” dimers (see definition in the previous section and Figs. 1 and 2A). PCR amplifications of pAge2 family sequences in more plants from other European A. lyrata populations (at least two plants each from Iceland, Wales, Russia, Sweden, and Germany) showed the same pattern as the initial A. lyrata ssp. petraea plant studied (from Iceland), with only even-number repeat bands (Fig. 2B). In two of these taxa analyzed (A. halleri ssp. gemmifera and A. lyrata ssp. petraea), our new larger numbers of sequences reveal two clear peaks in the numbers of nucleotide differences between monomer sequences (Fig. 3); the peak with the larger number of differences corresponds to differences between the first and the second subunits of dimers, while variation within monomers of these units constitutes the peak with the smaller number of differences. The sequences of these complete dimers include 25 sites at which variants are specific to first or second units (Supplemental Fig. 1).

Fig. 3.
figure 3

Frequency distributions of pairwise nucleotide differences among pAge2 monomer units. Pairwise nucleotide differences per base pair with Jukes and Cantor correction were calculated in MEGA2 (Kumar et al. 2002) with the pairwise deletion option. Frequency distributions are shown with intervals of divergence of 1%.

Full size image

In contrast, PCR with further plants from two different North American populations (A. lyrata ssp. lyrata), the initially studied Ontario population and a population from Indiana, and with A. halleri ssp. halleri, yielded both even- and odd-number repeat bands, with almost the same strength of bands representing even and odd numbers of repeats. We determined sequences of 36 trimers (11 from A. halleri ssp. halleri and 25 from A. lyrata ssp. lyrata), together with 2 pentamer sequences, 7 dimers (5 from A. halleri ssp. halleri and 2 from A. lyrata ssp. lyrata), and 8 tetramers (1 from A. halleri ssp. halleri and 7 from A. lyrata ssp. lyrata) sequences. When multiple sequences were analyzed, plants of these taxa yielded clear peaks only when analysis was restricted to units from “simple” dimer sequences, indicating that these dimers are predominantly complete double units, with minor variants at individual nucleotide sites (Supplemental Fig. 1). Thus the regular simple dimeric form of sequence subunits observed in A. halleri ssp. gemmifera (Kawabe and Nasuda 2005) is still present in these taxa, as well as odd-numbered structures.

The numbers of differences among units from odd-number repeat sequences, or between dimers and sequences with odd repeat numbers (Fig. 3), show no clear peaks and are generally smaller than between the first and the second units of simple dimer sequences, suggesting that these units differ less than do first and second subunits of dimers; these sequences thus have similarity to both first and second subunits of dimer sequences and can be classified as “complex,” following the “complex dimer” terminology defined in Fig. 1.

Among these sequences are some A. lyrata ssp. lyrata trimer sequences with chimeric structures that evidently derive from simple dimer units. Their sequences include long stretches similar to those of dimers from either A. lyrata ssp. lyrata or petraea. They can be classified into three groups, with different combinations of parts resembling portions of simple dimers and different junction regions (Supplemental Fig. 1). One type has high variation (7.60%) but this is based on only two sequenced clones. Within two trimer types, however, we find low variation (mean nucleotide diversity of 1.03%, based on 12 clones sequenced, and 2.31%, based on 11 clones). This is much less diversity than within simple pAge2 dimers (3%–7% depending on the species; see Table 1), and it suggests that these sequences rapidly and recently spread over the genome, particularly in A. lyrata ssp. lyrata.

Trimer sequences from A. halleri ssp. halleri are more complicated, although they cluster with other pAge2 sequences and differ only in the combinations of first and second units of simple dimers. The 11 trimer clones have some minor sequence differences at a few sites, and each clone has a different combination of variants characteristic of first and second units of simple dimer sequences.

Levels of Variation in Centromeric Satellite DNA Families

Including the sequences published previously, we analyzed a total of 308 pAa and 201 pAge1 monomer units (excluding 2 units with large deletions). For the pAa and pAge1 families, the average numbers of pairwise nucleotide differences per site are about 8%–10% for each family species-wide, within each population, or even within single plants (Table 1). For both these families, these values are similar to the mean raw nucleotide divergence values between sequences from different (sub)species (Dxy: Table 2). Thus different units from the same species differ almost as much as units from different species. This indicates that the diversification of these sequences occurred before these species diverged from one another.

Table 2. Levels of nucleotide differences in the centromeric satellite sequences between species: values are raw mean pairwise divergence (D xy)
Full size table

The pAge2 family has the highest level of variation of the three families when monomer units (357 units excluding 7 with large deletions or unsequenced regions) were used for the analyses (not shown), but this is due to differences between first and second monomer units of “simple” dimer sequences, not to differences between different sequences of such dimers. If we analyze simple pAge2 dimer sequences as dimers (112 sequences), the pAge2 family has an average of only 7.20% nucleotide differences, even including all species’ sequences; within each species, these sequences have lower within-species diversity than either of the other two sequence families (Table 1). Divergence between species is also much less for pAge2 simple dimer sequences than for the pAa and pAge1 families (Table 2).

Given its longer divergence time from the other species, A. arenosa should have the most distinct sequences, while the two subspecies of either A. halleri or A. lyrata should differ the least. However, only the pAa family gives this result, having (slightly) greater divergence for A. arenosa than for the sequence comparisons involving the other species and subspecies, which all yield similar estimated values (Table 2). An absence of a relationship to the times when the taxa split suggest that this family is old, and has not recently increased in abundance in the A. halleri and A. lyrata genomes. The pAge1 family shows higher divergence between species than between subspecies (Table 2), but shows little differentiation between populations (Fig. 4), suggesting the absence of rapid homogenization between paralogous sequences within the species.

Fig. 4.
figure 4

Phylogenetic relationships of the pAge1 and pAge2 families. pAge1 monomer (left) and pAge2 dimer (right) sequences were used for tree constructions. The figure shows NJ trees using Jukes and Cantor distances. Branches of units from different species or populations are distinguished by thick black lines (A. halleri subspecies), gray shaded lines (Plech from Germany and Karhumaki from Russia), thin black lines (Esja Mt. from Iceland, Stubbsand from Sweden, and Clogwyn D’ur Arddu, Wales), and dotted lines (Ontario, Canada). Consensus sequences from other families are also included as outgroups. A distance bar is shown at the bottom. A version with each population in a different color is supplied as Supplementary Fig. 2.

Full size image

In the pAge2 family, however, sequences from some populations tend to cluster together, especially those from ssp. petraea (Fig. 4). The pAge2 sequences from the Plech and Karhumaki populations (both A. lyrata ssp. petraea) are similar, and both differ by less than 5% from the A. halleri sequences and, thus, cluster with them (Fig. 4). These sequences differ from those from the three other European populations (from Iceland, Sweden, and Wales), whose sequences are all similar and mostly differ by more than 5% from A. halleri sequences. The scattered phylogenetic positions of sequences of A. lyrata ssp. lyrata pAge2 from the Ontario population probably do not reflect this population’s history but mainly reflect the existence of chimeric sequences (“complex” decayed dimers) and, perhaps, recombination between complex and simple sequences.

Sequence Exchanges in the Three Centromeric Satellite Families

The nucleotide variants within each satellite family suggest the action of crossing-over and/or gene conversion in several species, especially in the pAge2 family (Table 3). All datasets have at least one nucleotide pair for which all four combinations of variants were seen, indicating that some form of exchange has probably occurred, as already suggested above. The minimum estimated numbers of recombination events for the three families varied from 1 to 16 in the species studied. Although these estimates depend on the numbers of sequences in the dataset and numbers of parsimony informative sites, the different satellite families give similar results. It seems unlikely that these exchanges are due to reciprocal recombination, but gene conversion is possible.

Table 3. Estimated numbers of recombination and gene conversion events
Full size table

For the pAge2 family, the presence of “decayed” trimer sequences suggests many sequence exchange events (most likely gene conversions), especially in the A. halleri ssp. halleri and A. lyrata ssp. lyrata dataset. The sequence combinations in A. lyrata ssp. lyrata and A. halleri ssp. halleri pAge2 trimer units suggest that exchange events initially occurred between simple dimer sequences (Supplemental Fig. 1). However, fixed variants are found between the dimer and the trimer sequences, even within the same subspecies, suggesting that sequence exchange between the different categories of sequences may be limited (although the small number of sequences analyzed overestimates the number of fixed differences). The trimer-specific variants are often shared between different subunits (i.e., not confined to the first or any other specific subunit of trimers); this suggests sequence exchanges also between different trimer sequence types, and even between subunits of a single trimer sequence. Although we have not analyzed long stretches of pAge2 units and do not know how the dimeric and other types of units are arranged in the genome, these results suggest that the different types of pAge2 units may represent an early stage in the differentiation of a new satellite sequence family.

Discussion

Multifamily Structure of Centromere Satellite Families in A. halleri and lyrata Subspecies

All A. halleri and lyrata subspecies have three centromeric satellite families, indicating that the time required for fixation of a major centromere satellite family is not shorter than the history of A. halleri and lyrata, which diverged from one another recently (Ramos-Onsins et al. 2004), certainly much more recently than the divergence of their common ancestor from A. thaliana (estimated, based on a molecular clock, to be ∼5 MYA [Koch et al. 2000]).

Fixation of a new satellite sequence in a species requires that a variant must arise in a single unit and then spread throughout its local array on its chromosome of origin and then to all arrays across all chromosomes in the species. Fixation times will therefore be very long unless selection favors variants (Ohta and Dover 1984; Charlesworth et al. 1994). Theoretical studies (reviewed by Charlesworth et al. 1994) show that fixation times of new sequences in multicopy sequence arrays depend on unequal crossover frequencies, gene conversion frequencies, copy numbers, and bias in the gene conversion process (Ohta and Dover 1984; Stephan 1989). Due to the huge numbers of satellite repeats in the A. thaliana centromere, even the first stage of spread within a chromosome will require a long time in the absence of selection. In A. thaliana, the centromere regions include about 2 Mbp consisting of repeats of the 180-bp pAL1 satellite family (Haupt et al. 2001; Hosouchi et al. 2002). There are thus about 10,000 copies in each chromosome, and this is likely to apply also to A. halleri and A. lyrata, whose genomes are larger than that of A. thaliana (Johnston et al. 2005), although recent comparative sequence analyses of pericentromeric region showed a much larger pericentromeric intergenic region in A. thaliana than related species (A.E. Hall et al. 2006). The species studied here were not studied by A.E. Hall et al. (2006), but it is likely that similar results would be found, since it appears clear that the exceptional species is A. thaliana.

The second stage, spread to centromeres of other chromosomes, requires interchromosomal exchanges between units, which are probably rare events (Morgante et al. 1997; Heslop-Harrison et al. 1999; Schindelhauer and Schwarz 2002), although the presence of the same satellite sequence families in the centromeres of different chromosomes indicates some interchromosomal movement for the satellite families studied here (Kawabe and Nasuda 2006). Recombination only slightly increases fixation times, predominantly affecting the variability among members of the multigene family. Since centromere regions have low rates of reciprocal recombination, the largest influences on fixation times of the variants studied here should come from the gene conversion rate and amount of bias in the conversion process, which has effects similar to natural selection (Ohta and Dover 1984).

Assuming that crossing-over does not occur in the centromere region, the expected time to fixation under neutrality can be roughly estimated simply from the effective population size, the repeat number, and the gene conversion rate per repeat unit (Nagylaki and Petes 1982; Ohta 1983). This is of interest, because it can allow one to infer whether selection has been driving replacements of satellite sequences by variant sequences. Using equation (8) of Ohta (1983), the predicted fixation time within one chromosome, without selection or biased gene conversion, is roughly given by the copy number divided by the gene conversion rate (if this value is lower than N e generations) or by 4N e (if the copy number/gene conversion rate ratio exceeds N e generations). N e can be estimated from silent site diversity, which is between 1% and 2% for this species (Wright et al. 2003; Ramos-Onsins et al. 2004); with plausible mutation rates per base pair of 1.5 × 10−8 and 6.5 × 10−9 (Wright et al. 2003), we obtain N e values up to 106. The divergence time between A. thaliana and A. halleri and A. lyrata, whose satellite families are different, is estimated to be about 5 MYA (Koch et al. 2000), which is about 106 generations, assuming one generation per year. Assuming about 104 copies of the satellite sequences in each chromosome, the gene conversion rate must therefore exceed 10−2 for fixation of different satellite families in these species under genetic drift alone. Gene conversion rates have not been estimated for centromere regions in these species but, for noncentromeric regions, are probably as low as the mutation rate, 10−8 to 10−9 per site per generation (Innan 2004). Using this value (i.e., assuming that the rate is not much higher in centromere regions), the fixation of a new satellite sequence within one chromosome is estimated to require about 1010 generations, much larger than the species’ divergence time. Thus, either selection or biased gene conversion is necessary to explain complete substitution of satellite sequences in the available time.

However, pairing of nonhomologous regions or the action of mobile elements could increase sequence exchange rates in centromere regions, making fixation possible in a shorter time than estimated above. Also, in centromere regions, other processes, including sister chromatid exchanges and nonhomologous end joining between physically distant arrays, or even between nonhomologous chromosomes, could cause exchanges similar to those due to gene conversion events. Physical clustering of centromeres in nuclei or ectopic exchange between tranposable elements could potentially cause sequence exchanges between chromosomes. The effects of these processes are not treated separately in the models on which our calculation is based, but they are equivalent to an increased gene conversion rate and will, thus, reduce the fixation time estimated above (Nagylaki and Petes 1982; Ohta 1983). However, this possibility seems implausible, because the gene conversion rate in the centromere regions required for replacement of the satellite sequences of an Arabidopsis centromere under neutrality in the available time would have to be extremely high (10−2 per repeat). Moreover, as just explained, the fixation time of repetitive sequences dispersed among two or more chromosomes will be even longer than that in a single chromosome (Ohta and Dover 1983). All three satellite families are the sole sequences in at least one chromosome of the species studied here (Kawabe and Nasuda 2005). Thus centromere satellite sequences may have been replaced too rapidly to have occurred by genetic drift alone. A biased increase in certain satellite variants is thus necessary to explain the homogenization of sequences throughout the whole A. thaliana genome.

This analysis also suggests that the shared satellite families between A. halleri and A. lyrata probably represent a transient stage. Several other species, including some plants (Harrison and Helop-Harrison 1995; Ananiev et al. 1998; Gindullis et al. 2001), have multifamily centromere satellite sequences, often with some sequences specific to certain chromosomes, as observed in A. halleri and A. lyrata. As in the genus Arabidopsis, relatives of such species usually have just a single, completely different, satellite sequence in all their chromosomes. Coexistence of two or more centromere satellite families within species thus tends to last for evolutionary times shorter than the divergence times of the species studied, consistent with our finding that shared variants are found only in closely related Arabidopsis species. Most cases of satellite families shared between species are in recently diverged animal or plant species (e.g., Mestrovic et al. 1998; Nijman and Lenstra 2001). Cases involving species with estimated divergence times of more than 10 MYA have been reported, but centromere functions of the sequences have not been verified (Vershinin et al. 1996; Robles et al. 2004).

Species and Population Specificity of Different Satellite Sequences

The different levels of similarity of satellite sequences within and between species suggest occasional species-specific amplification of variants. Among the three satellite families of A. halleri and A. lyrata, pAge2 sequences have the greatest species and population specificity, while pAa shows almost no species-specific sequence variants. These differences are partly related to the history of the different satellite families. A. arenosa, a close relative of A. halleri and A. lyrata, has only the pAa family, suggesting that this sequence type evolved in a common ancestor, long enough ago to spread throughout the chromosomes of this species, and is thus probably the oldest of the three families present in A. halleri and A. lyrata. This would imply that pAge1 evolved from a centromere region initially dominated by pAa, in an ancestor of A. halleri and A. lyrata, and that pAge2 emerged and increased subsequently. A recent origin of the pAge2 family is supported by the fact that it shows the lowest levels of PCR amplification (suggesting a low copy number) and the clearest band patterns (Fig. 2). The clear pAge2 bands indicate that length variation is low among sequences in this family, consistent with our sequence analyses and suggesting a recent common ancestor; in contrast, smeared patterns for pAa and pAge1 suggest repeat units of highly variable lengths, i.e., greater age.

The low diversity among A. lyrata ssp. lyrata pAge2 trimers (with two groups of sequences that are both almost-monomorphic combinations of simple pAge2 dimer units) suggests selectively favored variants that are increasing in frequency too rapidly for exchange events to generate variation among sequence copies.

Restriction of Sequence Exchanges to Within Families

If the abundance of one complex dimer sequence variant increases at a given centromere, for any reason, the region involved (presumably initially just one centromere) will initially contain mixtures of sequences with different structures.

Despite the evidence of recombination (probably due to gene conversion), especially for the pAge2 family, we found no clones, even those containing trimer or tetramer sequences, that include partial or complete unit sequences of the other satellite families. If exchanges occur between different chromosomes or distantly localized satellite units, chimeric sequences of two or more families might be observed if larger numbers of sequences were analyzed; our current dataset only includes a small fraction of all satellites. The failure to detect any such chimeric structures suggests that the mechanisms causing exchanges of centromeric satellite sequences are restricted to physically close locations. Only one chromosome of A. halleri ssp. gemmifera has two different satellite families in a single centromere, and so physical separation may explain restriction of exchanges to similar sequence units, consistent with the results suggesting that exchange events between distantly located satellite units are generally infrequent in Arabidopsis species (S.E. Hall et al. 2005).

Exchanges may also be limited by sequence divergence. In many organisms, including plant species, recombination occurs only between similar sequences (Dooner and Martinez-Ferez 1997; Dooner 2004). The A. halleri and lyrata centromeric satellite families’ sequences may mostly be different enough (about 30% nucleotide divergence) to restrict between-family exchange events. However, some exchanges are detected between different pAge2 dimer units, with divergence of about 20%.

Among the A. halleri ssp. halleri pAge2 sequences, many types of trimers were found, and almost every clone is unique (contrasting with the pAge2 results from A. lyrata ssp. lyrata). Double or multiple events have occurred in short stretches of sequences in the centromere regions containing pAge2, which cannot be explained simply by recombination between units.

We detected only a few gene conversion events other than those just described in A. halleri ssp. halleri and A. lyrata ssp. lyrata pAge2. The high variability and short length of unit sequences of pAa and pAge1 make it difficult to detect gene conversion within these families, but the generally conserved dimer structure of pAge2 sequences also suggests limitation of sequence exchanges, even in this satellite family.

Observations of extant sequence variation, or even direct experiments monitoring mutation processes, cannot allow us to fully determine the evolutionary processes that destroy strict dimer units and generate other more complex structures. Nevertheless, the presence of dimer sequences must lead to generation of new satellite variants, allowing for coevolution with centromere proteins. The system in these Arabidopsis species may therefore provide opportunities in the future for testing whether selective events have indeed driven the evolutionary changes observed in their centromere satellite sequences.