We found inconsistencies between the topology of NJ trees based on IGS sequences and the species phylogenies based on mtDNA sequences. The most plausible explanation for these differences is the introduction of allospecific nuclear DNA through hybridization between D. pulex E and both D. tenebrosa and D. pulicaria. The NJ tree of complete IGS sequences reveals that only the two DpxE3 sequences form a separate branch as predicted by the divergent mtDNA sequence of this individual. The IGS sequences of the other two D. pulex E individuals cluster with those from D. pulicaria or D. tenebrosa. While hybridization between D. pulex E and D. pulicaria or D. tenebrosa has not been documented in the literature, the divergence among their mitochondrial 12S rDNA sequences falls well below the 14% threshold for species which are known to hybridize [ and references within]. Colbourne and Hebert  note that the lack of evidence for hybridization between species with low levels of 12S rDNA sequence divergence involves taxa with allopatric distributions, which is generally thought to be the case for D. pulex E relative to the other species. However, D. pulex NA and D. pulicaria have been found in Europe [16, 20] so opportunities for hybridization do exist.
The phylogenetic relationship among the four Daphnia species in this study, based on mtDNA sequences, is most closely reflected in the relationship among N1 sequences. We observed tight clustering of the three N1 sequences from D. pulex NA, while the N1 sequences of the three D. pulicaria individuals form a looser grouping. However, evidence for introgression is seen in four of the six DpxE N1 sequences. N1 sequences from DpxE1 cluster with the Dten N1 sequence, and N1 sequences from DpxE2 cluster with the Dpc N1 sequences.
The tree topologies of A and B repeats, which are interleaved with one another, are similar. Aside from the repeat sequences from DpxE3, major clusters are formed by orthologous rather than paralogous repeats for both A and B types. The occurrence of this structure in all but the most divergent species suggests that it has persisted for several million years, despite the occurrence of recombination between repeats (discussed below, ). Unfortunately, because all but one of the IGS arrays from D. pulex E appears to have been impacted by hybridization, it is not entirely clear if this position-specific pattern also occurs in this species. However, two observations suggest that it may. First, the only A repeats from different DpxE IGS sequences (DpxE3a-A1 and DpxE3b-A1) that cluster with one another in the NJ tree (Figure 6) are both in the same (first) position. Second, branch lengths between the A repeats in the DpxE3a array are more similar to branch lengths between array positions than within them in the other species (Figure 6).
All but the last of the six B repeats in the DpxE3a array cluster with one another, which is consistent with the pattern observed for A repeats. However, orthologous clustering of terminal F repeats was observed in the IGS of Drosophila melanogaster and Dr. orena . Others have also reported the apparent escape from homogenization experienced by terminal repeats relative to interior paralogs [21–23].
With the exception of the three D. pulex E individuals, for which two complete IGS were sequenced, our data are limited to a single IGS sequence for each individual, and three IGS sequences per species in D. pulex NA and D. pulicaria. This, in combination with the introgression mentioned above, limits the confidence with which we are able to estimate the divergence time necessary for IGS sequences to appear more similar within than between species. However, divergence times between the species in this study, based on the mitochondrial genes , suggest that the threshold for detecting patterns consistent with concerted evolution for the complete IGS must be greater than 4 million years.
Recombination in the IGS
We expect the hierarchically iterative nature of rDNA to facilitate recombination and homogenization at this locus. Indeed, our GARD and GENECONV analyses confirm that recombination occurs at multiple locations across the Daphnia IGS, including the repeats although these analyses do not identify recombination hotspots. The GARD algorithm identifies nonrecombinant segments rather than precise recombination break points and adopts the convention that breakpoints coincide with variable sites because breakpoints can only be resolved to the nearest variable site . In fact, actual breakpoints may be located at invariant sites .
Although it is possible that some of the intraindividual recombination that we observed is due to template switching during PCR amplification, we used a long extension time and a total of 30 cycles. Thus, it seems unlikely that recombination during the PCR reaction is a substantial source of the variation we observed. Indeed, recombination among IGS repeats has been observed in sequences from D. pulex obtained by cloning directly from genomic DNA . However, the frequency with which such artefacts occur could also be tested empirically by combining cloned divergent IGS sequences and amplifying them under our PCR conditions.
The copy number of A repeats, which contain a putative enhancer motif , ranges between one (Dpc1) and five (DpxNA2). Crease  reported that 18 of 21 arrays from seven D. pulex NA individuals contained four repeats, while the remaining three contained either five or six. This length variation is strong evidence that unequal crossing over is occurring between misaligned IGS repeats. Despite this, A and B repeats cluster by position in the array rather than species. This pattern was also observed by Luchetti et al.  in the IGS arrays of Triops cancriformis, which contain three copies of a ~200 nt repeat. In a previous study, we  found that the homogeneity of tandem and interleaved repeats increases as their number increases in arthropod IGS sequences. Thus, the rate of recombination in short arrays may be too low to fully homogenize the repeats. We also observed that duplication and deletion events rarely involve terminal repeats, which is consistent with the results of earlier work in plants. For example, Markos and Baldwin  found that interior repeats evolve in concert in Lessingia spp. (Compositae, Astereae), and Baldwin and Markos  found that sequence similarity of flanking repeats is higher between orthologs than paralogs in Calycadenia (Asteraceae).
Previous studies have suggested that intrachromosomal exchange (between sister chromatids) is more frequent than interchromosomal exchange (between homologues) in rDNA. For example, Crease  showed that intrachromosomal recombination is most likely responsible for patterns of sequence diversity within the IGS repeat arrays of D. pulex NA. Similarly, Schlötterer and Tautz  suggested that intrachromosomal exchange mechanisms are the most parsimonious explanation for the homogenization process in the ITS of Drosophila melanogaster. In contrast, our results suggest that many of the putative gene conversion tracts in the nonrepetitive regions of the IGS occurred between, rather than within, species (i.e. between homologous chromosomes in hybrids). This is consistent with the results of Polanco et al.  who showed that homogenization of the Drosophila IGS is the result of interchromosomal recombination. Our results do not exclude the possibility that intrachromosomal exchange occurs at an equal or even higher frequency than interchromosomal exchange. However, they do suggest that recombination within the IGS occurs during a phase in the cell cycle when homologous chromosomes are in close proximity, either following S phase during meiosis or when actively transcribed rRNA genes come together to form the nucleolus. Recombination can also occur between rDNA arrays on nonhomologous chromosomes, but D. pulex has only a single rDNA array per haploid genome (D. Tsuchiya, unpublished data). The number of rDNA arrays has not been determined for the other species, but they have similar genome sizes  and the same number of chromosomes (n = 12) as D. pulex . Taken together, the above studies corroborate Polanco et al.'s  assertion that different regions within the rDNA unit follow different evolutionary trajectories.
Conserved regions within the IGS
The exceptionally low sequence diversity in the first ~350 nt of N1 suggests that it undergoes homogenization along with the 28S rRNA gene. Liao  also reported that the homogenization of flanking regions in bacterial rRNA genes was the result of hitchhiking, or co-conversion with genic sequences. Moreover, the mean sequence divergence and the topology of NJ trees differs between N1 and N2, and from the repetitive region that connects them. This may be due to differences in the strength of natural selection acting on regulatory regions within the IGS, as well as the frequency with which recombination occurs between paralogous repeat copies whose sequences predispose them to frequent breakage and repair.
Because concerted evolution reduces mean intraspecific p-distance among the members of a MGF despite interspecific divergence, we would expect the ratio of mean intra- and interspecific p-distance (p-distance ratio) to be less than one and decrease with divergence time. On the other hand, if natural selection is constraining sequence divergence, then mean intra- and interspecific p-distance should be low and similar, especially among closely related taxa such as those included in this study. In this case, the p-distance ratio would remain close to one regardless of divergence time.
Although hybridization has blurred the species boundaries between individuals in this study, a comparison of mean p-distances within and between species does suggest that some of regions of the IGS may be under functional constraint. For example, the most conserved of the four N2 segments delimited by GARD breakpoints (N2-3), with a p-distance ratio of 1.0, is located between the putative core promoter and the breakpoint at nt 3900 in the full IGS, which may be the location of an rRNA processing site [35–37]. In contrast, the region that appears to be under the least functional constraint (N2-4, Table 5) is just downstream of this region and upstream of the 18S rRNA coding region, which is highly conserved both within and between species. This increase in both mean intra- and interspecific p-distance is also evident when mean p-distance is calculated after dividing the IGS into sequential 500 nt sections (data not shown).
As previously noted, the lowest overall sequence diversity occurs at the 3' end of the 28S rRNA coding region (N1-1 and N1-2). In contrast, the highest sequence diversity occurs just downstream of this region, in the middle section of N1 (N1-3), which includes a GAn dinucleotide repeat. The p-distance ratio is relatively low in this region (0.76, Table 5), but the lowest ratio (0.5) occurs in region N1-5, which is separated from N1-3 by the only region in the IGS (N1-4) where mean intraspecific divergence actually exceeds mean interspecific divergence (ratio = 1.27). The explanation for this pattern is unclear, but it should be noted that all of the regions in N1 are relatively short (151 - 181 nt). Further examination of this pattern will require analysis of species that diverged from a common ancestor at least 4 million years ago, and between which hybridization does not occur.
The region of the IGS with the highest mean intraspecific sequence divergence is the repeat region, although the p-distance ratio is also high at 0.94 (Table 5). This high level of diversity is primarily driven by differences between repeats in different positions in the array (Figure 3 and Figure 4). As suggested above, one explanation for this is low rates of recombination. However, it has also been suggested that this pattern may be maintained by natural selection despite the occurrence of recombination . Indeed, the A repeats contain an ~27-nt putative TATA motif, which is highly conserved among all A repeats in this and the previous study . This motif is also be found in the IGS repeats of other arthropods  and those containing the motif were found to be significantly more homogeneous than those without it in these taxa. These results suggest that selection is able to maintain homogeneity or diversity among functionally important repeat types regardless of the level of recombination among them .