1. Introduction
Gametophytic self-incompatibility (GSI) is a genetic mechanism present in flowering plants that is controlled by the multi-allelic S-locus and that prevents self-fertilization, by enabling the pistil to reject pollen from genetically related individuals (de Nettancourt, Reference de Nettancourt1977). Recognition is a ‘lock-and-key’ system, and each ‘lock-and-key’ is defined as a specificity (Charlesworth et al., Reference Charlesworth, Vekemans, Castric and Glemin2005). The S-locus contains two separate genes: one that determines pistil specificity and another that determines pollen specificity. GSI has been extensively studied in Solanaceae, Plantaginaceae and Rosaceae species. In these families, the pistil component of GSI has been shown to be an S-RNase (see review by Wang et al., Reference Wang, Wang, Skirpan and Kao2003). S-RNase-based GSI has been proposed to have evolved only once, before the separation of Asteridae and Rosidae (Igic & Kohn, Reference Igic and Kohn2001; Steinbachs & Holsinger, Reference Steinbachs and Holsinger2002; Vieira et al., Reference Vieira, Fonseca and Vieira2007c).
In GSI, many allele specificities are maintained by frequency-dependent selection for very long evolutionary times (Wright, Reference Wright1939; Takahata, Reference Takahata1990). In Solanaceae, for instance, alleles from species that diverged 30 million years ago are found mingled in the phylogenetic tree (Richman et al., Reference Richman, Uyenoyama and Kohn1996; Charlesworth & Guttman, Reference Charlesworth and Guttman1997). Therefore, an allele from one species may be more closely related to an allele from another species than to other alleles from the same species, a pattern named trans-specific evolution (Richman et al., Reference Richman, Uyenoyama and Kohn1996). Trans-specific evolution has been described also in Prunus (see for instance, Ortega et al., Reference Ortega, Boskovic, Sargent and Tobutt2006). Nevertheless, it is unclear whether trans-specific evolution can be taken as evidence for the very old age of S-alleles, as argued by Richman et al. (Reference Richman, Uyenoyama and Kohn1996), in genera where no estimation of the species age is available. For the genus Prunus, the relationship of the different species has been determined (Bortiri et al., Reference Bortiri, Oh, Gao and Potter2002), but no estimate is given for the speciation times. In this work, the first age estimate for the genus Prunus is provided, using the published chloroplast sequence data and the estimated divergence time between the Maloideae and Prunoideae (32 million years; Wikstrom et al., Reference Wikstrom, Savolainen and Chase2001).
Levels of synonymous variability are, on average, 3·7 times lower in the Prunus S-RNases than in Solanaceae S-RNases (Vieira et al., Reference Vieira, Morales-Hojas, Santos and Vieira2007b). This observation suggests that Prunus S-RNase alleles are considerably younger than the Solanaceae S-RNase alleles, or that the neutral mutation rate is different in Prunus and Solanaceae. Furthermore, S-alleles from Prunus species never cluster with Maloideae S-alleles (Ushijima et al., Reference Ushijima, Sassa, Tao, Yamane, Dandekar, Gradziel and Hirano1998; Ma & Oliveira, Reference Ma and Oliveira2002). Therefore, Prunus S-alleles must be younger than 32 million years. In this work, for the first time, the age of the oldest Prunus S-alleles is estimated. This is feasible since more than 150 different Prunus S-RNase sequences have been reported (although most are partial); thus, it is likely that at least some of the oldest S-alleles are present in the sample. Furthermore, the pollen component of Prunus GSI has been identified as being an F-box gene (named SFB; Entani et al., Reference Entani, Iwano, Shiba, Che, Isogai and Takayama2003; Ushijima et al., Reference Ushijima, Sassa, Dandekar, Gradziel, Tao and Hirano2003). Levels of variability at the S-RNase and SFB genes are similar, the evolutionary histories of the two genes are correlated (although not completely), and amino acid sites under frequency-dependent selection (those likely to be responsible for specificity differences) are found in both genes (Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a, Reference Vieira, Morales-Hojas, Santos and Vieirab). More than 75 SFB allele sequences have been reported; thus, it is likely that at least some of the oldest S-alleles are present in the sample. Therefore, it is feasible to calculate the age of the oldest Prunus S-alleles based on the SFB data, as well.
In finite populations, the number of specificities that can be maintained at equilibrium is dependent on selection, mutation and drift (Wright, Reference Wright1939). For the same strength of selection and mutation rate, large populations will harbour more specificities than small populations (Wright, Reference Wright1939). For most of the species exhibiting GSI (16 of 19 species; Lawrence, Reference Lawrence2000) the estimated number of alleles in natural populations is below 45. In Physalis crassifolia (Solanaceae), which is estimated to have 44 specificities, the implied effective population size is 6000 to 10 000 individuals (Richman et al., Reference Richman, Uyenoyama and Kohn1996). In Prunus, most species sampled exhibit fewer than 34 specificities (Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a). The screening of a large number of individuals never revealed more than 33 specificities. By screening 145 P. lannesiana individuals Kato & Mukai (Reference Kato and Mukai2004) found 22 specificities; in 65 P. avium individuals studied by de Cuyper et al. (Reference de Cuyper, Sonneveld and Tobutt2005) 18 specificities were found, and later by screening 164 P. avium individuals Schueler et al. (Reference Schueler, Tusch and Scholz2006) found 15 specificities; Ortega et al. (Reference Ortega, Boskovic, Sargent and Tobutt2006) found 25 specificities to be present in 104 P. dulcis cultivars. Thus, the effective population size of most Prunus species exhibiting GSI is thought to be below 10 000 individuals. Historical changes in population size may lead to S-allele loss. Therefore, another goal of this work was to determine the number of specificities found in the common ancestor to all Prunus species. This number can be compared with that observed in extant Prunus species in order to infer historical changes in population size. Such changes in population size also have been inferred in Solanaceae species (Richman et al., Reference Richman, Uyenoyama and Kohn1996; Richman & Kohn, Reference Richman and Kohn2000; Lu, Reference Lu2006).
Theoretical models predict that a novel specificity is expected to replace the one that gave origin to it in the local population where it arose, but migration from another population may prevent the loss of the original specificity (Uyenoyama et al., Reference Uyenoyama, Zhang and Newbigin2001). Nevertheless, when two closely related specificities are present in one population, it is possible that in a fraction of all attempted fertilizations one of the specificities is misrecognized as being the closely related one (Newbigin & Uyenoyama, Reference Newbigin and Uyenoyama2005). Therefore, closely related specificities should be rare in natural populations, but could be frequent when comparing closely related species. Prunus S-RNase and SFB amino acid sites under frequency-dependent selection have been identified (Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a, Reference Vieira, Morales-Hojas, Santos and Vieirab). Therefore, it is possible to test whether, in Prunus, closely related specificities are found at the expected frequency, under the assumptions that they are not preferentially lost during evolution, and that a single amino acid change at a site identified as being under frequency-dependent selection results in a different specificity. Although the latter assumption is debatable, in two P. spinosa populations two alleles differing at a single such site have been found in the same individual, implying that they probably represent two different specificities (Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a). Here, we use for the first time a simulation approach to investigate the expected shape of the relationship between specificities, under the above assumptions, using only information at sites identified as under frequency-dependent selection.
2. Materials and methods
(i) Data sets used and computational methods
To estimate the age of the different Prunus subgenera, neutral genes are needed, since the S-locus is under frequency-dependent selection. Three genes for which nucleotide sequence data are available for at least one species from each of the Prunus subgenera (Prunus, Amygdalus and Cerasus) and one species of Maloideae were used (the non-coding chloroplast regions trnL-trnF, trnS-trnG intergenic spacers, and the rpl16 intron; accession numbers are shown in Supplementary Table 1 at http://evolution.ibmc.up.pt/data/prunus_GR). The nucleotide sequences were aligned using ClustalX (Thompson et al., Reference Thompson, Gibson, Plewniak, Jeanmougin and Higgins1997). The number of silent changes per silent site (Ks values with Jukes–Cantor correction) was computed using the DNasp software (Rozas et al., Reference Rozas, Sanchez-DelBarrio, Messeguer and Rozas2003). Based on two plastid genes (rbcL and atpB) and one nuclear gene (18S rDNA), and using as a calibration point two fossils that imply the split between the Fagales and Cucurbitales occurred 84 million years ago, Wikstrom et al. (Reference Wikstrom, Savolainen and Chase2001) provide a mean estimate of 32 million years ago for the split between Prunoideae and Maloideae species.
Prunus S-RNase and SFB protein sequences were retrieved from GenBank by searching blastp with one complete S-RNase (BAC65203) and SFB (BAC65204) protein sequence, respectively. Sequences labelled as non-functional and those from Prunus self-compatible species were included in the analyses, since they are closely related to functional S-alleles (see, for instance, Hauck et al., Reference Hauck, Ikeda, Tao and Iezzoni2006; Tsukamoto et al., Reference Tsukamoto, Hauck, Tão, Jiang and Iezzoni2006). Sequences labelled as non S-RNases and non-SFB that were retrieved in this way were discarded from the data set except for two S-like RNases from P. dulcis (PD1 and PD2) and one SFB-like sequence from P. armeniaca. These sequences were used to root the linearized minimum evolution trees. Since, using this procedure, S-RNase and SFB paralogous genes were also retrieved, it is likely that all Prunus S-RNase and SFB sequences available in GenBank, at that time, were retrieved. For the same species, when two or more identical protein sequences were retrieved only the longest one was used. Furthermore, when similar proteins (less than three amino acid differences) from the same species with identical specificity names were found only the longest one was used. P. dulcis S-allele synonyms (i.e. identical sequences with different specificity names) reported in Ortega et al. (Reference Ortega, Boskovic, Sargent and Tobutt2006) were also considered. The final data set contains 158 and 78 S-RNase and SFB sequences, respectively (see Supplementary Tables 2 and 3 for accession numbers at http://evolution.ibmc.up.pt/data/prunus_GR). Protein sequences were aligned using ClustalX (Thompson et al., Reference Thompson, Gibson, Plewniak, Jeanmougin and Higgins1997). The minimum evolution trees were obtained using pairwise deletion as implemented in the MEGA software (Kumar et al., Reference Kumar, Tamura and Nei2004), since most sequences are partial. Maximum likelihood and Bayesian methods for phylogeny reconstruction, using the corresponding nucleotide sequences, could not be used, since they use a complete-deletion approach. Since most sequences are partial, all positions have alignment gaps (data not shown). Furthermore, there is nucleotide substitution saturation when sequences from the Maloideae and Prunoideae are compared (data not shown). In order to estimate the age of the oldest Prunus S-allele, as a calibration point we use the age of the estimated split between Prunoideae and Maloideae species (32 million years; Wikstrom et al., Reference Wikstrom, Savolainen and Chase2001).
We used our estimate of the age of extant Prunus species, and our calibrated amino acid minimum evolution tree, to estimate the number of S-alleles in the ancestor to extant Prunus species. For pairs of S-RNase and SFB amino acid sequences, which could represent instances of two copies of the same ancestral specificity inherited by two different species (see Section 3), the corresponding nucleotide sequences were retrieved from GenBank (see Supplementary Tables 4 and 5 at http://evolution.ibmc.up.pt/data/prunus_GR for accession numbers). The nucleotide sequences were aligned using ClustalX (Thompson et al., Reference Thompson, Gibson, Plewniak, Jeanmougin and Higgins1997), and the number of silent changes per silent site (Ks values with Jukes–Cantor correction) was computed using the DNasp software (Rozas et al., Reference Rozas, Sanchez-DelBarrio, Messeguer and Rozas2003). For the SFB gene, it is not possible to estimate the per site silent site nucleotide substitution rate, since the Maloideae orthologous gene has not been identified yet (Sassa et al., Reference Sassa, Kakui, Miyamoto, Suzuki, Hanada, Ushijima, Kusaba, Hirano and Koba2007). For the S-RNase, the average Ks for comparisons involving Maloideae and Prunoideae sequences is larger than 4·93 (data not shown), and it can be shown that there is nucleotide substitution saturation (data not shown). Thus, it is not possible to obtain an accurate estimate for the silent site nucleotide substitution rate, as well.
In order to estimate the degree of ancestral specificities shared between Prunus species A and B (where species B is always the one for which more specificities have been described) we calculated the percentage of ancestral specificities in species A that are represented in species B.
For the purpose of evaluating whether there is a lack of closely related Prunus specificities, only amino acids located at sites identified as being under frequency-dependent selection (those likely to be responsible for defining specificities) at the S-RNase (Vieira et al., Reference Vieira, Morales-Hojas, Santos and Vieira2007b) and SFB (Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a) genes were used to construct two independent sets of words, named S-RNase-w and SFB-w. Only sequences encompassing all sites identified as being under frequency-dependent selection were used. We have used only amino acid sites that were identified with a high degree of confidence (those amino acid sites that have a posterior probability of selection higher than 50% when using two different methodologies; Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a, Reference Vieira, Morales-Hojas, Santos and Vieirab).
Random word networks were then generated using the following computational approach (program available on request from the authors): for the S-RNase-w and the SFB-w data sets, containing n different words (92 and 68 for the S-RNase-w and SFB-w, respectively), we first calculate the number and set of amino acid variants at each position; then a set of words is generated by performing the following steps: (i) Randomly select one word from the data set being considered (either the S-RNase-w or the SFB-w data set), and add it to a bag of words. (ii) Choose the distance x (number of differences) between the selected word and a new word to be created. This value is randomly taken from a Poisson distribution with a mean equal to the mean of the mean distance between a given word and all other words present in the data set being considered (either the S-RNase-w or the SFB-w data set). (iii) Create a new word by incrementally making x mutations; thus, by doing so x new words will be created and added to the bag of words. In order to perform this step, a random position in the word is first selected and then one amino acid is randomly selected from the set of amino acids observed in the data set being used, at that particular position. Repeat this process x times using the most recently created word as a template. (iv) For every word position check whether it uses all amino acids observed at that same position in the data set being considered; if not, then randomly select one word from the simulated bag of words and repeat steps (ii) and (iii). This procedure will generate a bag of simulated words much larger than the number of words in the data set being considered, thus suggesting that the empirical data set (either the S-RNase-w or the SFB-w) contains only a fraction of all words that could have arisen during evolution. The procedure described above is repeated 100 times, to generate 100 independent simulated bags of words for the S-RNase-w and the SFB-w data sets. From each bag of words, we take 100 random samples with size n. Thus, in the end we have 10 000 simulated samples from which we compute the appropriate summary statistics. The graphics and summary statistics were calculated using standard functions in the R language and environment (http://www.r-project.org/).
3. Results
(i) Age of the subgenera Amygdalus, Prunus and Cerasus
Bortiri et al. (Reference Bortiri, Oh, Gao and Potter2002) determined the relationship of the three Prunus subgenera (Amygdalus, Prunus and Cerasus). Nevertheless, there is no estimate for the age of the subgenera or speciation times. In Prunus, trans-specific evolution has been described (see, for instance, Ortega et al., Reference Ortega, Boskovic, Sargent and Tobutt2006). Nevertheless, without an estimation of Prunus speciation times it is not possible to know whether this observation can be used as evidence for the very old age of S-alleles, as in Solanaceae (Richman et al., Reference Richman, Uyenoyama and Kohn1996).
The split between the Prunoideae and Maloideae lineages has been estimated by Wikstrom et al. (Reference Wikstrom, Savolainen and Chase2001) to have occurred between 29 and 35 million years ago (the average of the three estimates given by these authors is 32 million years). The relative synonymous divergences, based on published chloroplast data, of species from two different Prunus subgenera to the synonymous divergence of those species and Maloideae species are shown in Table 1. In this table, estimated ages (in millions of years) for the separation of the different Prunus subgenera are also shown. The common ancestor to the living species of the subgenera Amygdalus, Prunus and Cerasus, based on neutral genes, lived no more than 5 million years ago (Table 1).
Average silent site divergence and estimated age in million of years (within parentheses) between pairs of species from the three Prunus subgenera and from the Maloideae estimated using three chloroplast gene regions
a The average values between Prunus subgenera and Maloideae species were used in the calculations to obtain divergence ages in million of years.
b The split between the Prunoideae and Maloideae lineages has been estimated to have occurred between 29 to 35 million years ago (Wikstrom et al., Reference Wikstrom, Savolainen and Chase2001); thus we use the average of these values (32).
(ii) Age of the oldest Prunus S-alleles
Patterns of variability at Prunus S-RNase and SFB genes suggest that Prunus S-alleles are, on average, much younger than S-alleles in Solanaceae (Vieira et al., Reference Vieira, Morales-Hojas, Santos and Vieira2007b). At the S-RNase gene, the average amino acid divergence between Prunoideae and Maloideae sequences is 32% (data not shown). This value corresponds to 32 million years (Wikstrom et al., Reference Wikstrom, Savolainen and Chase2001). A linearized minimum evolution tree, using all available Prunus S-RNase amino acid sequences, is shown in Fig. 1a. In this tree, the oldest specificities show an estimated 15–20% amino acid divergence relative to most other specificities. Therefore, the oldest specificities are about 15–20 million years old.
Linearized rooted minimum evolution tree showing the relationship of S-RNase (a) and SFB (b) sequences from Prunus species. The tree shown in (a) has been rooted by using the PD1 and PD2 sequences from P. dulcis, two S-RNase-like sequences. The tree shown in (b) has been rooted by using an SFB-like sequence from P. armeniaca. Squares delineate sets of alleles that are believed to be derived from the same ancestral specificity (see text for details). Dotted squares indicate uncertainty. Five short amino acid sequences could not be incorporated (P. armeniaca S13, P. salicina S12, P. salicina S13, P. armeniaca Sh and P. webbii S5) since they did not overlap other short sequences already in the tree. Arrowheads point to sequences that show similarity with one of those sequences. Black dots point to pairs of closely related alleles from the same species that are unlikely to represent the same specificity because they show amino acid differences at sites previously identified as being positively selected.
A linearized minimum evolution tree using all available SFB amino acid sequences is presented in Fig. 1b. It is not possible to calibrate this tree in the same way as the S-RNase tree since the Maloideae orthologous gene has not been identified yet (Sassa et al., Reference Sassa, Kakui, Miyamoto, Suzuki, Hanada, Ushijima, Kusaba, Hirano and Koba2007). If we use the rate of change estimated for the S-RNase gene, then the oldest specificity also seems to be 15–20 million years old.
(iii) Specificity number in the ancestor to extant Prunus species
Historical changes in population size may lead to S-allele loss, since the number of alleles that can be maintained in a population depends on the effective population size (Wright, Reference Wright1939). Therefore, by comparing the number of specificities in the ancestor to extant Prunus species with that observed in extant Prunus species (fewer than 34 specificities), historical changes in population size can be inferred.
The common ancestor to the living species of the subgenera Amygdalus, Prunus and Cerasus lived no more than 5 million years ago, and, in this time period, S-RNase amino acid sequences are expected to have accumulated about 5% amino acid divergence. Therefore, pairs of Prunus S-RNase alleles that represent the capturing of the same ancestral specificity by two different Prunus species are expected to show less than 5% amino acid divergence. The same approach cannot be performed for the SFB gene, since a clear orthologue has not yet been identified in Maloideae species (Sassa et al., Reference Sassa, Kakui, Miyamoto, Suzuki, Hanada, Ushijima, Kusaba, Hirano and Koba2007). Nevertheless, the depth of the Prunus S-RNase and SFB trees is similar (see Fig. 1). Thus, it seems reasonable to use the 5% value for the SFB gene as well.
In Fig. 1a (the S-RNase tree), there are 100 groups of alleles that differ by more than 5% amino acid divergence from any others. Therefore, the most recent common ancestor of extant Prunus species is assumed to have harboured at least 100 specificities. Of the five short amino acid sequences that could not be incorporated in the tree (see Fig. 1a legend; P. armeniaca S 13, P. salicina S 12, P. salicina S 13, P. armeniaca S h and P. webbii S 5), two (P. armeniaca S 13 and P. salicina S 13) differ by more than 5% amino acid divergence from any others, thus increasing to 102 the number of specificities in the most recent common ancestor of extant Prunus species. In Fig. 1b (the SFB tree), there are 64 groups of alleles that differ by more than 5% amino acid divergence from any others. Therefore, the most recent common ancestor of extant Prunus species is assumed to have harboured at least 64 specificities.
A pairwise deletion approach was used to build the minimum evolution trees shown in Fig. 1 due to the different lengths of the amino acid sequences used. Thus, we are assuming that the different regions of these proteins are evolving at the same rate, which is known not to be true since there are conserved and hypervariable regions along these proteins (Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Vieira et al., Reference Vieira, Morales-Hojas, Santos and Vieira2007b). Therefore, these trees should be interpreted with caution and used only as a rough guide for the relationship of the different sequences. Most confidence can be placed on the relationship of the closely related amino acid sequences that are supported by high bootstrap values. Only two pairs of alleles (X1 and X2, Fig. 1a) that seem to have been diverging for less than 5 million years do not present high bootstrap values. Silent (synonymous plus intron sites; Ks values with Jukes–Cantor correction) divergence values were thus obtained for all pairs of S-RNase and SFB alleles with less than 5% amino acid differences (see Supplementary Tables 6 a and 6 b at http://evolution.ibmc.up.pt/data/prunus_GR). Only four of 69 such allele pairs produce Ks values above 10%. For the S-RNase and SFB gene 62% and 57%, respectively, of the closely related alleles from two different species differ in at least one amino acid site identified as being under positive selection by Vieira et al. (Reference Vieira, Santos, Ferreira and Vieira2007a, Reference Vieira, Morales-Hojas, Santos and Vieirab; see Supplementary Tables 6 a and 6 b at http://evolution.ibmc.up.pt/data/prunus_GR).
The percentage of ancestral specificities shared between Prunus species pairs ranges from 0 to 0·45, and from 0 to 0·23 for comparisons involving Prunus species from the same subgenus or from two different subgenera, respectively (see Supplementary Table 7 at http://evolution.ibmc.up.pt/data/prunus_GR). Despite the large variances associated with these estimates, Prunus species from the same subgenus share on average a higher percentage of ancestral specificities (the average is 0·18) than Prunus species from different subgenera (the average is 0·07; non-parametric Mann–Whitney test; P<0·025).
(iv) Relatedness of putative Prunus specificities
Theoretical models predict that closely related specificities could be rare in natural populations, but could be frequent when comparing closely related species (Uyenoyama et al., Reference Uyenoyama, Zhang and Newbigin2001; Newbigin & Uyenoyama, Reference Newbigin and Uyenoyama2005). It is conceivable that distantly related S-RNase or SFB sequences could represent closely related specificities. Thus, it is better to use information on amino acid sites shown to be under frequency-dependent selection than the full S-RNase or SFB sequences, as was done here, for the first time. Identifying all amino sites under diversifying selection (those likely to be responsible for specificity differences) at the S-RNase and SFB genes is, nevertheless, a difficult task since the different methodologies (a phylogenetic one (Yang, Reference Yang1997), and a population genetics method that uses an approximation to the coalescent with recombination (Wilson & McVean, Reference Wilson and McVean2006)), typically yield overlapping but not fully compatible amino acid sites (Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a, Reference Vieira, Morales-Hojas, Santos and Vieirab). Furthermore, it is unknown whether the methodologies used are able to detect all amino acid sites under diversifying selection. It is assumed that a single amino acid change at an amino acid site identified as being under frequency-dependent selection results in a different specificity. Although this assumption is debatable, there is some evidence to support it. For the larger data set (the S-RNase data set), the frequency of closely related specificities (those that are less than 5 million years old) is not statistically different (P>0·05; two-sided Fisher exact test) within species (0·26% of all 1948 possible comparisons) and between species (0·20% of all 14 165 possible comparisons).
A simulation approach was, nevertheless, used, for the first time, to investigate the expected shape of the relationship between specificities under a simple model that assumes that closely related specificities are not preferentially lost during evolution. In order to do so, words must be assembled into word clusters. If a word can be connected by fewer than five changes to another word then we consider that both words belong to the same cluster. Five changes is an arbitrary choice, and using a different cluster definition could affect the results for a given summary statistic (the factors considered here are: number of clusters, average cluster size, maximum cluster size and average distance within clusters). Nevertheless, the simulations we performed using different cluster definitions show that there is always at least one feature in the empirical data that is unexpected (data not shown). Thus, the conclusion that the empirical data are incompatible with the simple model being tested is not dependent on the cluster definition used.
For instance, using the above cluster definition, the most striking feature of the S-RNase-w data set is the presence of a word cluster of size 50. Obtaining such a large cluster is unexpected (P<0·001; see Supplementary Fig. 1a at http://evolution.ibmc.up.pt/data/prunus_GR). Other aspects of the S-RNase-w data set, such as the average size of the clusters, and the total number of clusters, occur with a probability of 0·05 to 0·1 (see Supplementary Fig. 1a at http://evolution.ibmc.up.pt/data/prunus_GR). The within-cluster average word distance is slightly higher than in the simulated data (2·57 vs 2·38, respectively; P>0·05; see Supplementary Fig. 1a at http://evolution.ibmc.up.pt/data/prunus_GR). For the SFB-w data set the within-cluster average word distance is significantly lower than the average of the simulated data (1·46 vs 2·10, respectively; P<0·005; Supplementary Fig. 1b at http://evolution.ibmc.up.pt/data/prunus_GR). The average cluster size is also significantly lower in the SFB-w data set than in the simulated data (1·33 vs 1·94, respectively; P<0·0005; Supplementary Fig. 1b at http://evolution.ibmc.up.pt/data/prunus_GR). More clusters are found in the SFB-w data set than in the simulated data and this difference is significant (51 vs 37 respectively; P<0·0001; see Supplementary Fig. 2 b at http://evolution.ibmc.up.pt/data/prunus_GR). Nevertheless, the observed maximum cluster size is compatible with that obtained in the simulations (P>0·05; see Supplementary Fig. 1b at http://evolution.ibmc.up.pt/data/prunus_GR).
4. Discussion
In Solanaceae, trans-specific evolution has been taken as evidence for the very old age of specificities, since alleles from species that diverged 30 million years ago cluster in the phylogenetic tree (Richman et al., Reference Richman, Uyenoyama and Kohn1996; Charlesworth & Guttman, Reference Charlesworth and Guttman1997). Here we show, for the first time, that extant Prunus are a group of closely related species, since all speciation events happened in the last 5 million years. There is a Prunoideae fossil that is estimated to be 44 million years old (Magallón et al., Reference Magallón, Crane and Herendeen1999). The observation that a fossil shares similarities with living species of a given family does not indicate that the fossil taxon is part of the crown group of living species (Wikstrom et al., Reference Wikstrom, Savolainen and Chase2001). It is thus concluded that, in Prunus, trans-specific evolution cannot be taken as evidence for the very old age of alleles, as in Solanaceae.
Given the large sample size, it is likely that very old Prunus specificities are present in our sample. Based on the available S-RNase and SFB amino acid sequence data, the oldest specificities have an estimated age that ranges from 15 to 20 million years. Therefore, a much younger age for Prunus than for Solanaceae specificities is the likely cause of the difference in synonymous variability levels at the S-RNase, when the two groups are compared (Vieira et al., Reference Vieira, Morales-Hojas, Santos and Vieira2007b).
Present-day Prunus species harbour fewer than 34 specificities (Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a). Although it could be argued that the relatively small number of specificities found in Prunus could reflect in many cases the continuing process of domestication, a similar number of specificities has been found by Raspé and Kohn (Reference Raspé and Kohn2002) for two wild Rosaceae (Maloideae) species (Sorbus aucuparia and Crataegus monogyna presenting 24 and 27 inferred specificities, respectively). Recently Kato et al. (Reference Kato, Iwata, Tsumura and Mukai2007) reported 75 P. lannesiana putative specificities. However, P. lannesiana forms interspecific hybrids with P. jamasakura and P. incisa and it is very difficult to distinguish precisely between these hybrids and P. lannesiana (Kato et al., Reference Kato, Iwata, Tsumura and Mukai2007). Thus, depending on the number of hybrids sampled, the number of specificities in P. lannesiana could be inflated. In contrast, it is inferred that the ancestral Prunus population harboured at least 102 specificities, implying a larger effective population size. The genus Prunus is geographically widely distributed, and therefore it is conceivable that the common ancestor to extant Prunus also was geographically widely distributed (Bortiri et al., Reference Bortiri, Oh, Gao and Potter2002).
The relationship of Prunus S-RNase and SFB allele pairs older than 5 million years is poorly resolved (see Fig. 1). This lack of resolution can be due to the pairwise deletion option used since it implies that different regions of the protein are evolving at the same rate and this is not a realistic assumption. It could also simply reflect a problem of amino acid saturation that it is difficult to correct for. A maximum-likelihood approach using long Prunus S-RNase and SFB nucleotide sequences also shows poor resolution for old specificities (see, for instance, Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006). Therefore, the lack of phylogenetic resolution observed in this work is not entirely attributable to using amino acid sequences, a pairwise deletion approach and minimum evolution for tree reconstruction.
Features compatible with recombination have been observed at the S-RNase and SFB genes, although the evidence is still not unequivocal (Vieira et al., Reference Vieira, Charlesworth and Vieira2003, Reference Vieira, Morales-Hojas, Santos and Vieira2007a; Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Ortega et al., Reference Ortega, Boskovic, Sargent and Tobutt2006). In principle, recombination could also significantly affect the shape and resolution of the phylogenies, although it remains to be demonstrated that the amount of recombination implied by the SFB and S-RNase data sets is enough to cause the observed pattern. It should be noted that rare recombination does not, in principle, greatly affect our estimate of the number of specificities present in the common ancestor. Only recombination events occurring after Prunus speciation may create the illusion that the recombinant allele already existed in the common ancestor to all living Prunus, thus inflating our estimate. Since the common ancestor to present-day Prunus species lived about 5 million years ago, and recombination is assumed to be rare at the S-locus, it is unlikely that our estimate is greatly inflated.
Our conclusion on the number of specificities in the ancestor to extant Prunus is also dependent on the assumption that at that time there was a single ancestral species. Alternatively, there could have been multiple species (all except one of the lineages going extinct), with restricted gene flow among them. Extant Prunus species from the same subgenus hybridize easily (Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Kato et al., Reference Kato, Iwata, Tsumura and Mukai2007; http://www.rjb.csic.es/floraiberica/PHP/cientificos.php), but not Prunus species from different subgenera (Surbanovski et al., Reference Surbanovski, Tobutt, Konstantinović, Maksimović, Sargent, Stevanović and Bosković2007). Under the scenario of multiple species with frequent gene flow among them, the entire nuclear genome is expected to be affected. Therefore, this hypothesis can be tested by looking at variability patterns of nuclear genes in extant Prunus species, but such data are not yet available. Such data for multiple extant Prunus species are needed to achieve accurate estimates of the effective population size in the current and ancestral species, and thus to estimate the magnitude of the reduction in population size. A reduction in population size can lead to loss of alleles and, thus, to a variability loss at the S-locus.
Even closely related Prunus species do not share more than 50% of the ancestral specificities (see Supplementary Table 7 at http://evolution.ibmc.up.pt/data/prunus_GR) and, on average, species from the same Prunus subgenus share more ancestral specificities than species from different subgenera. Therefore, it is possible that in Prunus speciation is often associated with population bottlenecks or that the process of domestication of Prunus species resulted in the loss of specificities.
Closely related specificities are expected to be rare in natural populations, but could be frequent when comparing closely related species (Uyenoyama et al., Reference Uyenoyama, Zhang and Newbigin2001; Newbigin & Uyenoyama, Reference Newbigin and Uyenoyama2005). Many closely related S-alleles (those that are less than 5 million years old) from different Prunus species present differences at amino acid sites previously identified as being positively selected (see Supplementary Tables 6 a and 6 b at http://evolution.ibmc.up.pt/data/prunus_GR). Therefore, many of the closely related S-alleles could represent different specificities, although the lack of information on the minimum number of changes needed to create a new specificity precludes a firm conclusion. Ideally, crosses should be made between individuals harbouring these closely related alleles, but often this is not possible since individuals are not marked in the field, and/or because it is not possible to cross different species (Surbanovski et al., Reference Surbanovski, Tobutt, Konstantinović, Maksimović, Sargent, Stevanović and Bosković2007). Using the S-RNase data set (the larger one), the frequency of closely related putative specificities (those that are less than 5 million years old) is not statistically different (P>0·05) within species (0·26% of all possible comparisons) and between species (0·20% of all possible comparisons). Therefore, theoretical expectations seem not to be fulfilled.
The simulation approach shows, on the other hand, that the empirical data are incompatible with a scenario where closely related specificities are not preferentially lost during evolution, and this conclusion is not dependent on the cluster definition used. The conclusion that closely related specificities are preferentially retained (not predicted by current theoretical models) or lost (as predicted by Uyenoyama et al., Reference Uyenoyama, Zhang and Newbigin2001; Newbigin & Uyenoyama, Reference Newbigin and Uyenoyama2005) is, however, dependent on the cluster definition used, and it is not obvious which one should be used. Under the assumption that two words that can be connected by fewer than five changes belong to the same cluster, the S-RNase data set seems to be biased towards an excess of closely related putative specificities, and this is unexpected according to current theoretical models. An apparent excess of putative specificities may be caused by wrongly assuming that the alleles considered represent different specificities. Although little polymorphism exists within specificities (Nunes et al., Reference Nunes, Santos, Ferreira, Vieira and Vieira2006; Vieira et al., Reference Vieira, Santos, Ferreira and Vieira2007a), the S-RNase sequences used come from different Prunus species, and thus it is likely that the same specificity in two different species is not represented by the same amino acid sequence. Nevertheless, this feature is not observed in the SFB data set, where sequences also come from closely related Prunus species. The SFB data set seems to be biased towards an excess of distantly related putative specificities, in agreement with theoretical expectations. It should be noted, however, that this is a smaller data set than the S-RNase. More SFB sequences from many Prunus species are needed in order to solve this issue.
Barbara Mable made helpful comments on an earlier version of this work. This work has been partially funded by Fundação para a Ciência e Tecnologia (FCT; research projects POCTI/AGG/44800/2002 and POCI/BIA-BDE/59887/2004 funded by POCI 2010, co-funded by FEDER funds). N. A. F. is the recipient of a Postdoctoral grant SFRH/BPD/26737/2006 from FCT.
