The long-term relationship between a transposable element and its host involves exceedingly complex interactions that are challenging to address 
. Indeed, the best understood elements are the DNA transposons that remain active in a species only temporarily and depend upon horizontal transfers to propagate over long periods of time 
. R2 elements, on the other hand, appear to have co-evolved with their hosts since the origin of most animal taxa 
. The absolute specificity of R2 for a unique site in the 28S rRNA genes has greatly aided attempts to define three key parameters needed to build a population genetic model for the stability of a mobile element: i) the regulation of element expression, ii) the number of elements and their rates of turnover, and iii) the potential affects of each insertion on the host. In the case of the latter parameter, each insertion blocks the production of intact 28S rRNA from one rDNA unit. R2 element evolution suggests they are simply selfish elements, and are not preserved by the host to aide the regulation or the evolution of the rDNA locus. Many species contain multiple subfamilies of R2, as well as members of other mobile element families inserting into the rRNA genes (e.g. R1 elements). This proliferation of element families and subfamilies is consistent with the selfish propagation of parasites to fill a niche, rather than their maintenance for a useful function 
. Even the arguments that mobile elements might provide useful genetic diversity 
have little application to R2. The sequence of the 28S gene around the insertion site is nearly identical in all eukaryotes, and the expression of the rRNA genes follows a similar pattern in all animals.
In this report we have attempted to integrate into a simple population genetic model all previous findings concerning the structure, regulation and turnover of both R2 elements and the rRNA genes. The critical findings were that D. simulans
populations could be divided into R2-active and R2-inactive individuals 
, that genetic control over R2 activity mapped to the structure of the rDNA locus itself 
, that the key regulatory step in R2 activity was at the level of transcription 
, and that R2 transcripts are generated by self-cleavage from a 28S rRNA co-transcript 
. Furthermore, studies to determine why some rDNA loci in natural populations supported R2 transcription while others did not suggested that the size of the rDNA locus and the number of R2 insertions only weakly correlated with R2 activity [20, Figure 2]
. Instead, the property of the rDNA locus that best predicted R2 activity was the size of the largest contiguous block of rDNA units free of R2 insertions 
. These studies of R2 expression could be readily integrated with two previous findings concerning the expression of the rRNA genes themselves. First, that only a small fraction of the rRNA genes are transcribed, an estimated 35 to 50 rDNA units in Drosophila
. Second, that the rDNA units activated for transcription are contiguous, not individual units distributed throughout the locus 
. These findings concerning R2 and the rRNA genes gave rise to the transcription domain model, which we have incorporated into computer simulations to model the long-term stability of R2 elements.
Critical to reproducing the population data was the assumption that most crossover events within the rDNA loci were localized within or near the transcription domain. Only then did most R2-inserted units remain single copy (i.e. not duplicated by recombination) consistent with the population data (, right panel). This clustering also generated populations with all animals containing a relative narrow range in numbers of R2 (, middle panel). The clustering of crossover events in the region with the lowest number of R2-inserted units (the transcription domain) predicts that over many generations the number of uninserted rDNA units within a locus would change more rapidly than the number of R2-inserted units. This is precisely what was observed in our study of the Harwich mutation accumulation lines of D. melanogaster
. The Harwich lines, originally derived from one inbred stock 
, had been maintained as separate sublines for over 400 generations. During the 400 generations, the size of the rDNA locus on the X chromosomes of the 19 lines changed dramatically shrinking to a low of 140 units in some lines and expanding to over 300 units in others 
. The vast majority of this variation in number of rDNA units was associated with the uninserted units, with less than 1% of the variation associated with the number of R2-inserted units. There were also no instances within the 400 generations in which an R2-inserted unit was duplicated by recombination 
. These findings provide strong empirical data to support the model that most crossovers within the rDNA locus occur in regions free of R2 insertions.
The clustering of crossovers within the transcription domain of the rDNA locus could be a result of two non-mutually exclusive mechanisms. First, active RNA transcription may be inducing the crossover events. RNA transcription, or more broadly chromatin structure, has long been associated with various types of genome instability including recombination 
. Indeed, one of the first experiments suggesting this connection involved transcription of the rRNA genes 
. Second, clustering of recombination in the transcription domain may result from the more efficient pairing of this region between chromosomes. The presence of an R2 insertion within a rDNA unit will disrupt its ability to completely align with an uninserted unit. Therefore, if crossovers involve the precise alignment of DNA sequences spanning multiple rDNA units, then those regions most likely to undergo a crossover would be the regions free of insertions. We have conducted computer simulations in which the locations of the crossover events were influenced by the composition of the surrounding rDNA units 
. When crossovers were permitted only when four contiguous units matched between chromosomes or chromatids (i.e. uninserted matched uninserted unit, and inserted units with inserted units), regions of the locus free of insertions were quickly generated.
It should also be noted that the localization of crossovers to the transcription domain, and thus to regions typically near the center of the locus (Figure S5
) does not prevent the concerted evolution of the rDNA genes. Simulations of the rDNA locus involving millions of generations and the addition of low rates of nucleotide substitutions demonstrated that concerted evolution of rDNA units were efficient whether the crossovers were distributed throughout the locus or restricted to the middle of the locus, (Eickbush, M. and Eickbush, T., unpublished). The only aspect of the concerted evolution process that differs between the two models was that only mutations in units near the center of the rDNA locus became fixed when recombinations were restricted to the center of the locus, while variants in units from throughout the locus could become fixed under conditions of uniform crossovers.
An unusual aspect of our population genetic model for the propagation of R2 elements is that R2 activity does not depend upon periodic failure in the host regulatory systems or on low levels of R2 activity that escape host control. The driving force that maintains R2 elements within a population is the recombinations that result in the concerted evolution of the rDNA locus. Because of the stochastic nature of these crossovers, rDNA loci that contain a large region free of R2 insertions and thus have not supported R2 activity for many generations will occasionally undergo crossover events that reduce the R2-free regions to the point that R2-inserted units are transcribed. Loci with active R2 elements will subsequently increase the number of R2 insertions within the same X chromosome, and in females also on the paired X chromosome. Over time, loci with the most active R2 elements will be eliminated from the population by selection. This build-up of R2 elements has been detected in laboratory stocks of D. simulans
with active R2 elements 
. While the fitness of stocks with low levels of R2 transcription are similar to that of stocks with no R2 transcription, the fecundity of lab stocks with very high levels of R2 transcription are significantly reduced [D. Eickbush, unpublished data]. The activation of R2 activity by stochastic recombinational forces within the rDNA locus, instead of a reliance on overcoming the host regulatory machinery, may explain why R2 elements are so stable in most lineages of Drosophila, while most mobile elements that insert throughout the genome show a patchy species distribution and extensive variation in abundance 
Another possible factor contributing to the long-term stability of R2 is that these elements may be less susceptible to control by the small RNA pathways, notably the piwi
pathway, which are known to regulate mobile elements 
. Properties of R2 elements which may contribute to their greater resistance to piwi
regulation include: a) their strict specificity to the 28S rRNA target means they are unlikely to become part of the piRNA clusters of Drosophila 
, b) their strict orientation within the rDNA unit means antisense RNAs are unlikely to be generated, and c) they are co-transcribed with the highly abundant 28S rRNA, a transcript that simply can not be shut down in any cell type. However, even given these unusual properties of the R2 element, it is interesting to speculate that small RNA pathways are still playing a critical role in R2 regulation by determining what region of the rDNA loci is transcribed. A likely model is that heterochromatin formation induced by small RNA occurs initially on R2 sequences, spreads to the entire rDNA unit, and then into flanking units. As a result the largest region of the rDNA locus free of R2 insertions would be the most likely to remain active for transcription.
Finally, additional support for our model of R2 propagation as well as estimates of the variable parameters used in our simulation could be derived from two sources. First, electron microscopic observations of actively transcribing rDNA loci are needed for D. simulans
. Previous studies have only been conducted on D. melanogaster
and more distant species 
. Direct examination in D. simulans
would provide better estimates of the number of units within the transcription domain, and in particular whether there is flexibility in the size of this domain in flies where R2 transcripts are detected. Second, more data is needed concerning the frequency of sister chromatid exchanges and the size of the offset. Experiments to estimate these critical parameters are currently underway using pulsed field gels to monitor changes in the spacing of R2 insertion in rDNA locus over time. Finally, more data is needed on the fitness consequences of low levels of R2 expression.