The high polymorphism of ITS sequences observed in I. canariensis is frequently reported for other plant groups [15, 16]. It might have several origins: an incomplete lineage sorting from ancestral polymorphism or an horizontal transfer (introgression) through inter-specific hybridization (alloploidy), both of them not resolved by concerted evolution. Before the discussion on the origin of this polymorphism, the characterization and the fate of these different ITS sequences will be first examined.
The genome of Ilex canariensis contains ITS pseudogenes
High polymorphism of ITS has been explained by the presence of divergent pseudogenes in Gossypium, Nicotiana, Tripsacum, Exospermum, Zygogonum, Zea , Quercus , Leucaena , Adinauclea, Haldina, Mitragyna  and others. Thus, this could also be the case for I. canariensis. Individual criteria are not sufficient to identify pseudogenes unambiguously  and different criteria were chosen: GC content, secondary structure of ITS 2, rate of substitution, pattern of substitution at methylated cytosine sites and substitutions at highly conserved sites of the 5.8S rDNA. ITS sequences with a GC content of 45% are unambiguously pseudogenes and satisfy to all other criteria. Moreover they have a large deletion in the ITS 1 region, which make these sequences certainly non-functional. The deletion allows an easy detection of this pseudogene on agarose gels and it is observed in many individuals of I. canariensis (Figure 2).
Other classes of ITS sequences with a GC content of 53–54% are also suspected to be pseudogenes by one or the other criteria but not by all of them, as expected regarding their relatively high GC content. For instance, some ITS 2 sequences of the GC 53–54% class still have a typical angiosperm secondary structure (for instance can_39_3_53, Figure 3), but have (1) an increased rate of nucleotide substitution, (2) deamination-like substitutions or (3) mutations at normally highly conserved 5.8S rDNA sites. Only the GC 61 % clade contains functional ITS sequences. Thus, it can be considered that the functional ITS GC content is 57 % for I. perado and above 60 %. for I. canariensis.
It is interesting to note that most I. canariensis individuals of the GC 61 % clade never have ITS pseudogenes in the GC 45 % or GC 53–54 % clades. This is probably because these individuals do not contain pseudogenes. For other individuals, a PCR selection for pseudogenes occurred, as reported for Nicotiana , in which ITS sequences with a weak secondary structure (pseudogenes) are preferentially used as templates. The inclusion of dimethylsulfoxide (DMSO) in PCR reactions [10, 19], but see , would allow amplification of functional ITS sequences in these individuals of I. canariensis.
In conclusion, the high divergence found in ITS sequences of I. canariensis with a GC content lower than 60% (clades 1, 2, 3, 4, 5 and 6) could be explained by a release of evolutionary constraint and a subsequent high rate of substitution. Indeed, ITS sequences have functional constraints in relation with the processing of the rRNA precursor producing the functional 18S, 26S and 5.8S subunits.
ITS sequences of Ilex canariensis are recombined
Evidence for recombination in divergent sequences is not obvious. It is difficult to recognize homoplasy generated by recombination from actual homoplasy (parallel history). Statistical methods (based on linkage desequilibrium, neutrality tests and substitution distribution along the locus) are still too rudimentary to precisely describe the recombination events in the set of ITS sequences found in I. canariensis. Moreover, recombinants could result from "jumping" PCR reaction [20–23], where prematurely terminated extension products can act as primer on paralogous templates. This has been shown on nepGS for Oxalis  and on four low-copy genes for Gossypium .
The minimum number of recombination events (RM) calculated with DnaSP  underestimates the total number of recombination events . Thus, there is no doubt that I. canariensis ITS sequences experienced intra-molecular recombinations (Figure 4). The factor RM has been also calculated for PCR products of each individual in order to detect possible jumping PCR artifacts. In few of them (specimens 20, 22, 24, 27, 28 and 38) recombinants have been detected in ITS sequences resulting from a unique PCR reaction (data not shown). This could be the result of jumping PCR. However most of them are multiple recombinants and not simple recombinants as it is expected in jumping PCR . As an example, the alignment represented in Figure 4 shows that specimen 90 comprises two different recombined ITS sequences resulting from the same PCR reaction, that could be the result of jumping PCR. DnaSP did not detect recombination between the four cloned ITS sequences of individual 90 because recombined fragments are paralogous sequences fragments found in other individuals. Moreover, the recombinants result from at least three crossover events and are suggested not PCR artifact. Thus, they represent true organismal intra-molecular recombinations.
The distribution of informative characters shown in Figure 4, as well as the use of programs PIST and LARD based on maximum-likelihood analyses, demonstrate unambiguously that sequences of clade 5 (Figure 4) experienced recombination events. This can not be generalized for other clades. Although DnaSP suggests recombination, an alignment demonstrating recombination, as for clade 5, was not possible for other clades, even with the help of PIST and LARD. This could be explained by the recent origin of the recombination events observed in clade 5 and by the fact that mutations did not yet obscured the recombined orthologous fragments. In this respect it is to be noticed that clade 5 shows much longer branches than other clades. This may indicate that, in clades with relatively shorter branches, mutations (or concerted evolution) did homogenize the recombined fragments, mimicking clonal divergence. Thus it can be considered that most clades also comprise recombined ITS sequences, as DnaSP suggests, but of more ancient origin than those of clade 5, and homogenized by mutation or concerted evolution.
Recombination in highly polymorphic ITS sequences seems a rule in plants. This is not surprising because the mechanisms of concerted evolution in rDNA arrays are based on crossing-over and gene conversion. It has been reported in Begonia , Microseris , Quercus , Amelanchier , Paeonia , Buddia, Gossypium, Nicotiana, Tripsacum , Armeria  and others.
In addition to the high rate of substitution of pseudogenes, at least some ITS sequences experienced recombination. This explains why the divergence between ITS sequences of I. canariensis is much higher than between ITS sequences of all other species investigated, knowing that, according to their GC content (see Figure 1), they all are potentially functional. This also explains the absence of a bootstrap support for a monophyletic clade of I. canariensis ITS sequences because of long branch problems due to accelerated rate of substitution and more certainly to recombination.
The origin of the ITS polymorphism in I. canariensis
Two evolutionary mechanisms could produce the observed ITS polymorphism: an ancestral polymorphism escaping lineage sorting or a past or recent introgression of an alien genotype escaping concerted evolution. Because of the influence of concerted evolution, ancestral polymorphism is not the most likely explanation of ITS polymorphism . On the other hand, a growing number of reports shows that ITS polymorphism is attributable to interspecific hybridization, although the parents are not always identifiable [15, 16].
Assuming that multiple ITS sequences found in I. canariensis are the result of experienced hybridization with another species, or an ancient polymorphism with incomplete sorting, the determination of the identity of the putative hybridizing species or the finding of genetic relationships of the putative polymorphism is not obvious. This is because ITS sequences enclosed in non-functional clusters have dramatically diverged from the putative functional sequences and are recombined. All available ITS sequences of 43 other species of Ilex, representing a good sampling of the genus [2, 3] were incorporated in the phylogenetic analysis, altogether with all ITS sequences found in I. perado and I. canariensis of Tenerife (Figure 1). Most functional (above 60% GC) and non-functional (53–54% GC) ITS clades of I. canariensis group together but with no bootsrap support. They group with an American lineage (I. brevicuspis, I. anomala, I. microdunta, I. integerrima, I. theezans, I. guianensis, I. brasiliensis and I. cassine). Only the GC 45% clade does not group with the bulk of I. canariensis ITS sequences. Its position is not defined and varies in the vicinity of a Eurasian lineage (I. latifolia, I. leucoclada, I. maximocziana, I. rugosa and I. perado). Thus data do not support a particular relationship of most I. canariensis ITS pseudogenes with another Ilex species, except for the pseudogenes with a GC content of 45%, that are frequently observed in I. canariensis.
In the case of hybridization involving the island species I. canariensis, the most probable candidate would be the sympatric species I. perado. It can not be ruled out however that the distribution of I. canariensis was much wider in the past [32, 33] and that this hybridization may have occurred with another unknown or extinct species of the Eurasian lineage represented here by I. latifolia, I. leucoclada, I. maximocziana, I. rugosa and I. perado. Pseudogene sequences (particularly the ITS sequences of clade GC 45%) being too divergent and of different nucleotide composition, the observed relationship of clade GC 45% with the group of species comprising I. perado is questionable because of possible spurious long branch attraction. However, the data of the nuclear encoded plastid glutamine synthetase (a nuclear single copy locus) are not conflicting with an introgression of I. perado in I. canariensis. All the eight polymorphic sites observed in I. canariensis always comprise one allele shared with I. perado. Another possibility is that these ITS pseudogenes represent a relictual ancestral polymorphism in the course of elimination by lineage sorting or concerted evolution. In fact ancestral polymorphism could also be the result of ancient introgressions. The data accumulated here do not allow a definitive conclusion.
If a putative cryptic hybridization between I. perado and I. canariensis is confirmed, the introgression would be unidirectional because ITS sequences of I. perado do not show any polymorphism. This situation is reminiscent of the unilateral hybridization observed between Begonia formosana and B. aptera, where on 60 ITS sequences analysed in natural or artificial hybrids, 58 sequences are clustering with the ovule donor B. formosana, and only 2 are found clustering with the pollen donor B. aptera . Unidirectional interspecific hybridization linked to unilateral incompatibility is frequently described in plants. However, this is not the only mechanism that can explain unidirectional hybridization. The flowering time of I. perado precedes the one of I. canariensis, thus the loading of still living I. perado pollen grains on young effective I. canariensis stigmates is more favored than the contrary. Moreover, there are much more male than female I. perado plants in Tenerife [34, 35]. These evidences could explain the proposed unidirectional introgression.