Of particular interest to evolutionary biologists is the pattern of divergence of the whole rDNA array, which is often influenced by sequence homogenization that functions to replace existing genic units with variants of that unit over time, a process known as concerted evolution (Dover et al., 1982; Eickbush and Eickbush, 2007). Concerted evolution itself may be important to maintain sufficient numbers of active rDNA units, and Ohta (1989) suggested from computer models that homogenization acts to reduce mutational load and is favoured by selection. Patterns of ribosomal RNA (rRNA) gene expression are strongly influenced by epigenetic events such as cytosine methylation and histone acetylation, silencing potentially initiated by small interfering RNAs (Preuss and Pikaard, 2007). In hybrid organisms and allopolyploid species (interspecific hybrids with chromosome number duplication) silencing of entire loci is common, often leading to one parental rDNA array being expressed in preference to another, a phenomenon known as nucleolar dominance (Cermeno et al., 1985; Preuss and Pikaard, 2007).

Ribosomal DNA and the nuclear domain of its activity, the nucleolus, have long intrigued biologists; Alberts et al. (2002) stated in a 1898 review that there were already 700 references. Despite a huge volume of ongoing research (web-of-science search on 2 July, 2007, for ‘rDNA or ‘ribosomal DNA’ or ‘nucleolus’ revealed 25 437 hits), there are still critical and fundamental unanswered questions. These include: (a) why do many eukaryotes have more rRNA genes than required for ribosome biosynthesis; (b) why are large numbers of rDNA units often, but not always, epigenetically silenced, frequently leading to nucleolar dominance; (c) why are rDNA evolution and divergence often, but not always, associated with sequence homogenization, leading to concerted evolution; and (d) are any or all of these fundamental processes or phenomena interrelated?

To determine the influence of time on the divergence of rDNA and distribution of loci carrying rDNA (nucleolus organizer regions), it is necessary to estimate the time elapsed since the allopolyploids were formed. To do this, Clarkson et al. (2004) dated the nodes of a phylogenetic tree that included all diploid taxa, allotetraploid N. tabacum and four allotetraploid species of Nicotiana section Repandae. By combining data sets, they were able to generate trees with good bootstrap support. They determined likely maximum ages of all nodes on the tree (cf. Richardson et al., 2001) using calibrations based on ages of oceanic volcanic islands on which there are endemic species of Nicotiana. Section Repandae includes four species, two of which (N. nesophila and N. stocktonii) occur only on the Revillagigedos Islands in the Pacific Ocean (San Benedicto, Socorro and Clarión) far off the coast of Mexico. The oldest of these islands is Clarión, dated at approx. 1·2 million years. Assuming that N. nesophila and N. stocktonii diverged from mainland taxa subsequent to formation of this island, then the maximum age of the node that separates these species from N. repanda on the mainland is approx. 1·2 million years (cf. Clarkson et al., 2004). Likewise, if it is assumed that N. cordifolia formed on the island of Masafuera (Juan Fernández Islands off the coast of Chile in the Pacific Ocean), then dating enables predictions of the maximum likely age of the node that separates N. cordifolia from its sister taxon on the South American mainland, N. solanifolia, at 2·4 million years old (Stuessy et al., 1984). Using these as calibration points, it is possible to estimate the maximum age of allopolyploid groups (cf. Table 1; for more details on the dating procedure, see Clarkson et al., 2005).

The first reports of genetic change occurring to rDNA in association with hybrids and allopolyploids in Nicotiana were in N. tabacum × Atropa belladonna hybrids, in which novel rDNA units were observed (Borisjuk et al., 1988). Similarly in an established (1904) horticultural hybrid N. × sanderae (N. alata × N. forgetiana; Goodspeed, 1954), new IGS polymorphisms were observed (Lim et al., 2006a). In the synthetic tobacco line Th37, only three out of 20 plants analysed showed an additive pattern of rDNA RFLPs observed in the parents. In 75 % of plants, the IGS from N. tomentosiformis, accounting for approx. 2000 rDNA units, had been entirely replaced by a novel hybrid-specific rDNA cluster of comparable abundance (Skalicka et al., 2003). All novel rDNA variants were of N. tomentosiformis origin but 1–3 kb longer (Fig. 1A), primarily due to amplifications of the SR II and SR VI repetitive subregions of the IGS (subregions first identified by Volkov et al., 1999). Such structural changes in rDNA units have also been encountered in intergeneric somatic hybrids of N. tabacum × A. belladona (Borisjuk et al., 1988) and interspecific somatic hybrids of Medicago (Fabaceae) (Cluster et al., 1996).

We selected three Th37 synthetic tobacco plants with different 35S rDNA RFLP profiles for cytogenetic analysis. (1) Plant Th37·9 had RFLP bands that were additive for those observed in the parents (Fig. 1A) and had an additive number of 35S rDNA loci (i.e. three loci from the maternal parent N. sylvestris on chromosomes S10, S11 and S12, and one from the paternal parent N. tomentosiformis on chromosome T3; Fig. 1B). (2) Plant Th37·1 had a hybrid-specific cluster of rDNA units (Fig. 1A) and a new rDNA locus on one of the two homologues of chromosome T4 of N. tomentosiformis-origin (Fig. 1B). (3) Plant Th37·8 had lost the paternal N. tomentosiformis-derived rDNA units, but the rDNA sites on both T3 homologues were still present; the units at these sites had been replaced by the new hybrid-specific cluster of rDNA units (Fig. 1A). In addition both homologues of chromosome T4 now carried a new site (Lim et al., 2000a).

Analysis of Th37 synthetic N. tabacum (Skalicka et al., 2003) revealed that: (a) in most plants there was a rapid amplification of 35S rDNA units (in four generations) involving many thousands of genes that were either new or in sub-threshold copy numbers for detection in N. tomentosiformis; (b) amplification of 35S rDNA units first occurred at a new chromosomal locus (on chromosome T4; similar amplification of units at a new locus has also been observed in synthetic allopolyploids of Arabidopsis suecica; Pontes et al., 2004); and (c) new 35S rDNA units in most plants had been transferred to chromosome T3. The mechanism for this last observation could have been (a) sequence homogenization, an ill-defined process involving recombination machinery and perhaps unequal recombination between homologous or non-homologous sequences, saltatory replication and rDNA unit amplification or (b) reciprocal translocations between the rDNA loci on chromosomes T3 and T4, followed by random segregation of chromosomes at meiosis that could lead to the loss of all chromosome T3-derived rDNA units in subsequent generations.

RFLP analysis of the structure of the 35S rDNA units in N. tabacum revealed that the pattern of bands does not correspond to that found in extant N. tomentosiformis and N. sylvestris (Kovarik et al., 1997; Lim et al., 2000a), suggesting that N. tabacum has evolved its own distinct gene family. Sequence analysis revealed that the tobacco-specific units arose by reorganization of the parental N. tomentosiformis-inherited units followed by their subsequent amplification and homogenization between rDNA loci (Volkov et al., 1999). The sequence changes mainly involved amplification and reduction of subrepeats upstream and downstream of transcription start sites whereas point mutations were found to be rare.

Similarly, N. rustica and N. arentsii have also lost parental units (Fig. 3), although retaining the expected number of rDNA loci. However, the extent of parental gene replacement varied significantly between species. The extent of rDNA sequence homogenization decreases in the order N. arentsii > N. tabacum > N. rustica. Nicotiana rustica still maintains a considerable number of unconverted N. paniculata-origin units. In both N. tabacum and N. rustica upstream IGS subrepeats were fully homogenized, whereas downstream subrepeats were only partly homogenized, enabling IGS to be resolved into several distinctive families (Matyasek et al., 2003; Kovarik et al., 2004). IGS subrepeats may have a role in homogenization, perhaps by promoting or facilitating recombination between units. In Nicotiana, shorter upstream SRII subrepeats (10–23 bp) could undergo recombination more rapidly than the longer SRVI subrepeats (approx. 135 bp), or perhaps they diverge more rapidly. In Drosophila melanogaster the IGS contains topoisomerase II sites that are thought to promote recombination and homogenization. The putative AT-rich topoisomerase II binding element has been cloned from N. tomentosiformis IGS and shown to promote amplification of a linked reporter transgene (Borisjuk et al., 2000).

All N. tabacum cultivars show relatively uniform RFLP profiles, with >90 % of novel hybrid-specific units and <10 % of unconverted N. sylvestris-types of unit (Skalicka et al., 2003). However, a feral tobacco collected in Bolivia had a higher proportion of N. sylvestris-type unit (20 %), indicating that cultivation and breeding may accelerate the action of homogenization (Kovarik et al., 2004) or that cultivated strains were derived from a narrow range of genotypes compared with variation exhibited by plants in South America (truly wild plants of tobacco are unknown – they are always associated with human habitation); certainly we can expect inbreeding to promote a reduction in heterozygosity and fixation of alleles. Dadejova et al. (2007) examined seven varieties of cultivated N. rustica, which contained rDNA units designated as P- and U-units depending on the ancestry from the N. paniculata or N. undulata progenitor (Fig. 3B). All varieties displayed a reduced number of N. paniculata-origin units relative to prediction. However, the proportion of N. paniculata-origin rDNA varied (10–40 %) between, but not within, varieties (Fig. 3D). In addition, analysis of the IGS subregion downstream from the 26S gene showed significant length polymorphism between varieties, probably reflecting IGS rearrangements and further homogenization in the course of variety diversification. Such IGS polymorphisms in crop plants can be potentially exploited for genotyping purposes in breeding programmes.

It may be surprising that molecular reorganization of rDNA units and their amplification within and between rDNA loci are more rapid genetic events than changes in numbers of rDNA loci. Nicotiana is not exceptional; allopolyploids of Gossypium that are several million years old have also maintained the expected rDNA locus number, whereas the parental units at these loci have been overwritten with new, variant sequences (Wendel et al., 1995). However, over millions of years of allopolyploid divergence in Nicotiana, we observe a reduction in rDNA locus number. Thus in cotton, locus number may be a more conserved genetic trait than in Nicotiana (see below). Variation in locus number also seems to accompany evolution of many diploid (Dubcovsky and Dvorak, 1995; Lim et al., 2000b, 2007; Siroky et al., 2001; Dobigny et al., 2003) and even autopolyploid species (Weiss-Schneeweiss et al., 2007). No apparent reduction of locus number was observed in Brassica (Hasterok et al., 2006), recent Nicotiana (Kenton et al., 1993; Moscone et al., 1996; Lim et al., 2000a; Matyasek et al., 2003) and Tragopogon (Kovarik et al., 2005) allopolyploids. These studies indicated that locus number changes are not necessarily associated with the early evolution of allopolyploids.

An analysis of the number of 5S and 35S rDNA loci in species of sections Polydicliae (Fig. 4) and Repandae (Fig. 5) revealed a marked reduction in number from expectation. This reduction results in the total number of sites typical of Nicotiana diploids (i.e. two or three 35S rDNA sites and one 5S rDNA site). These allopolyploid species in sections Polydicliae and Repandae also show unique rDNA RFLPs that differ from diploid parents, indicating that each species has evolved unique rDNA family/ies. Nicotiana nesophila, N. repanda and N. stocktonii possess similar ITS sequences, suggesting their recent diversification from a common allopolyploid ancestor (Chase et al., 2003). However, each species has unique IGSs (Clarkson et al., 2005), indicating either that (a) homogenization of IGS occurs more rapidly than that of ITS, or (b) homogenization is equally rapid, but changes to ITS are selected against more strongly than those to the IGS.

We propose that nucleolar dominance, established early in allopolyploid formation including F₁ hybrids, plays a significant role in the evolution of rDNA. We suggest that epigenetic silencing of rDNA loci makes them less vulnerable to homogenization and more likely to be lost, perhaps thousands or millions of years later. The data from Nicotiana point to a link between epigenetic (nucleolar dominance) factors and genetic outcome (homogenization).

In early allopolyploids there may be over-expression of genes or imbalanced gene-product dosages arising from genome duplication. If this applies to rDNA, there are several ways that gene expression patterns can be tuned to meet cellular/organism physiology: (a) physical elimination of genes; (b) epigenetic inactivation of units (epigenetic rDNA dominance); (c) overwriting of one rDNA variant with another – perhaps by a version that will be silenced (genetic rDNA dominance); or (4) a combination of these. Active rRNA genes organized as decondensed chromatin may be most vulnerable to recombination events that can promote both homogenization and altered copy numbers (Kobayashi and Ganley, 2005; Ganley and Kobayashi, 2007). As a consequence, hybrid-specific units may evolve relatively rapidly, colonizing nucleolus organizer regions at all active loci, potentially from both subgenomes of an allopolyploid (depending on whether both are active). Epigenetic inactivation will increase levels of chromatin condensation and heterochromatin, which might act to protect rDNA from genetic recombination. Under this scenario, inactive, non-homogenized rDNA parental units may be transmitted over many generations even in high copy numbers. Recent findings indicate that rDNA may become dispersed in yeast mutants with abnormal methylation of histone proteins, production of siRNAs and factors involved in chromatin condensation (Peng and Karpen, 2007) showing linkage between genetic stability of repeats and epigenetic pathways. In the future it will be worth analysing epigenetic markers of chromatin (e.g. histone H3K9 methylation) that have survived unaltered during homogenization in allopolyploid species.

Homogenization is a powerful process that keeps the tandem arrays uniform, acting to remove non-functional units from the cluster (Ohta, 1989). We can anticipate that silenced rDNA units that are not homogenized would acquire changes leading to loss of function and could potentially be lost over evolutionary time. Certainly no alteration of inactive units would have any selective consequence. Thus, we can predict that the consequences of epigenetic silencing of rDNA loci are: (a) reduced or eliminated interlocus homogenization (and perhaps also intralocus homogenization); (b) divergence of units in the silenced loci, including substitutions that lead to loss of function, indels and loss of array uniformity; and (c) deletion of silenced genes (Fig. 7), perhaps through accumulation of indels, as for ancient retroelements (Ma et al., 2004), or, in those species in which rDNA arrays occupy a subterminal position, via deletions of chromosome termini. If this cascade of events does apply, then we might envisage that the lost rDNA loci in sections Repandae and Polydicliae were epigenetically silenced early in evolution of allopolyploid sections.