Molecular Ecology (2010) 19, 1675–1690 doi: 10.1111/j.1365-294X.2010.04585.x Hybridization and polyploidy as drivers of continuing evolution and speciation in Sorbus ASHLEY ROBERTSON,* TIMOTHY C. G. RICH,† ALEXANDRA M. ALLEN,* LIBBY HOUSTON,* C A T R O B E R T S , * J O N R . B R I D L E , * S T E P H E N A . H A R R I S ‡ and S I M O N J . H I S C O C K * *School of Biological Sciences, University of Bristol, Woodland Road, Bristol, BS8 1UG, UK, †Department of Biodiversity & Systematic Biology, National Museum of Wales, Cathays Park, Cardiff, CF10 3NP, UK, ‡Department of Plant Sciences, University of Oxford, South Parks Road, Oxford, OX1 3RB, UK Abstract Interspecific hybridization and polyploidy are pivotal processes in plant evolution and speciation. The fate of new hybrid and polyploid taxa is determined by their ability to reproduce either sexually or asexually. Hybrids and allopolyploids with odd chromosome numbers are frequently sterile but some establish themselves through asexual reproduction (vegetative or apomixis). This allows novel genotypes to become established by isolating them from gene flow and leads to complex patterns of variation. The genus Sorbus is a good example of taxonomic complexity arising from the combined effects of hybridization, polyploidy and apomixis. The Avon Gorge in South-west Britain contains the greatest diversity of Sorbus in Europe, with three endemic species and four putative endemic novel hybrids among its 15 native Sorbus taxa. We used a combination of nuclear microsatellite and chloroplast DNA markers to investigate the evolutionary relationships among these Sorbus taxa within the Avon Gorge. We confirm the genetic identity of putative novel taxa and show that hybridization involving sexual diploid species, primarily S. aria and S. torminalis and polyploid facultative apomictic species from subgenus Aria, has been responsible for generating this biodiversity. Importantly our data show that this creative evolutionary process is ongoing within the Avon Gorge. Conservation strategies for the rare endemic Sorbus taxa should therefore consider all Sorbus taxa within the Gorge and must strive to preserve this evolutionary process rather than simply the individual rare taxa that it produces. Keywords: apomixis, biodiversity, conservation, hybridization, polyploidy, self-incompatibility, Sorbus Received 9 July 2009; revision received 25 January 2010; accepted 29 January 2010 Introduction Interspecific hybridization and genome multiplication (polyploidy) are now accepted as pivotal processes in plant diversification and speciation, both separately and more commonly in combination, through the formation of allopolyploids (Stebbins 1950; Grant 1981; Arnold 1997; Rieseberg & Willis 2007). Indeed, hybridization and polyploidy offer the clearest routes to sympatric speciation in plants. Many examples of plant speciation through hybridization and ⁄ or polyploidy have now Correspondence: Simon J. Hiscock, Fax: +44 (0) 117 3317985; E-mail: simon.hiscock@bristol.ac.uk  2010 Blackwell Publishing Ltd been described (see: Arnold 1997; Rieseberg 1997; Soltis & Soltis 1999; Abbott 2003; Rieseberg & Willis 2007). Some (e.g. Helianthus, Arabidopsis, Brassica, Tragopogon, Spartina and Senecio) have proved to be valuable model systems for investigating genetic and genomic processes involved in plant speciation (Adams & Wendel 2005; Hegarty & Hiscock 2005, 2008; Salmon et al. 2005; Chen 2007). The breeding systems associated with plants involved in hybridization and polyploidy play a critical role in determining the likelihood of hybrid and ⁄ or polyploid formation and in determining the likelihood of their long-term survival (Grant 1981; Rieseberg 1997; Hörandl 2006). In sexually reproducing plants, outbreeding often 1676 A . R O B E R T S O N E T A L . encourages hybridization and the formation of new homoploid and allopolyploid offspring, while inbreeding (selfing) will perpetuate new homoploids and allopolyploids and reinforce reproductive isolation between them and their parental taxa (Grant 1981; Rieseberg 1997; Coyne & Orr 2004). Asexual reproduction, through apomixis (hereafter synonymous with agamospermy; the production of ‘maternal’ clonal seeds), offers a more secure means of perpetuating new hybrid and allopolyploid genotypes than selfing and affords immediate, often total, reproductive isolation from parental taxa. Not surprisingly therefore, apomixis is frequently associated with hybridization and polyploidy where it has been critical for the establishment of many novel taxa arising by these processes (Grant 1981; Briggs & Walters 1997; Hörandl 2006; Hörandl & Paun 2007; Whitton et al. 2008). The association between polyploidy and apomixis has been widely noted and it has been suggested that polyploidy either acts as a stimulus for apomixis or a prerequisite for its maintenance or both (Grimanelli et al. 2001; Whitton et al. 2008). Because apomixis essentially ‘freezes’ the new hybrid or polyploid genotype, it can facilitate the emergence of highly complex patterns of phenotypic diversity when hybridization and polyploidy occur recurrently among predominantly sexual parental taxa (Hörandl 2006; Hörandl & Paun 2007; Whitton et al. 2008). The result is taxonomic complexity in genera where apomixis is associated with hybridization and polyploidy, for example, Alchemilla (Sepp et al. 2000), Hieracium (Bicknell et al. 2000; Fehrer et al. 2007), Arabis (Koch et al. 2003), Rosa (Nybom et al. 2004), Craetegus (Lo et al. 2009), and Sorbus (Nelson-Jones et al. 2002; Robertson et al. 2004a, b). Recent reviews have considered the role of hybridization and polyploidy in divergence and speciation within sexually reproducing plants (Rieseberg & Willis 2007; Hegarty & Hiscock 2008), yet these processes in asexual (agamic) complexes, where hybrids and ⁄ or polyploids reproduce via obligate or facultative apomixis, have been less well studied (but see: Richards 2003; Hörandl 2006; Hörandl & Paun 2007). In the genus Sorbus (whitebeams, service trees, and rowans), hybridization, allopolyploidy, autopolyploidy and apomixis (both obligate and facultative) have contributed to the complex patterns of variation seen among the ‘intermediate’ European taxa (Liljefors 1953, 1955; Warburg 1962; Challice & Kovanda 1978; Proctor & Groenhof 1992; Rich & Jermy 1998; Nelson-Jones et al. 2002; Robertson et al. 2004a, b; Robertson & Sydes 2006; Chester et al. 2007). Embryological studies have shown that apomixis in Sorbus is gametophytic apospory with development of the embryo dependent upon pollination to initiate an endosperm (pseudogamy) (Liljefors 1953, 1955; Robertson et al. 2004b). Interest in the genus Sorbus lies with both the vulnerability of the taxa per se and the evolutionary processes that have generated these taxa (Hollingsworth 2003; Ennos et al. 2005). Taxonomically complex groups, such as Sorbus, pose important evolutionary questions about the interactions between hybridization, genome multiplication and breeding systems among parental and derivative taxa. Furthermore, generation of taxonomic novelty has direct consequences for taxon conservation. In Europe, there are three major centres (‘hot-spots’) of on-going Sorbus diversification and speciation: Scandinavia, South-east Europe and Britain (Hedlund 1901; Kovanda 1961; Warburg & Kárpáti 1968). In Britain, Sorbus is represented by 39 named taxa and at least five unnamed taxa (Stace 1997; Rich & Jermy 1998; Rich & Proctor 2009; Rich et al. 2009). Of the 19 species (i.e. excluding hybrids and recently described taxa) listed in the British Red List (Cheffings & Farrell 2005), three are recognized as ‘critically endangered’, four as ‘endangered’ and four as ‘vulnerable’ according to IUCN (2001) criteria, making Sorbus a priority genus for conservation in the UK. The Avon Gorge in South-west Britain (Bristol) is the richest and most important site for Sorbus diversity in Europe. There are four sexual diploid species [Sorbus aucuparia (rowan), S. aria (whitebeam), S. torminalis (wild service-tree) and S. domestica (service tree); the latter is not involved in hybridization and is not discussed further], one primary diploid hybrid (S. · thuringiaca) and seven polyploid taxa (S. anglica, S. bristoliensis, S. eminens, S. leighensis, S. porrigentiformis, S. whiteana, and S. wilmottiana). The sexual species are outcrossing and self-incompatible (Raspé & Kohn 2007; Robertson et al. unpublished) whereas the polyploids are believed to be apomictic. In addition to these native Sorbus taxa, there are five introduced species (S. croceocarpa, S. decipiens, S. intermedia, S. latifolia and S. glabruiscula) all of which are polyploid and believed to be apomictic. There are also four taxa that show unusual ‘intermediate’ morphologies which have been described recently as hybrids (Rich et al. 2009); these are S. · avonensis (= S. aria · S. porrigentiformis), S. · houstoniae (= S. aria · S. bristoliensis), S. · proctoriana (= S. aucuparia · S. scalaris) and S. · robertsonii (= S. aria · S. eminens). Existing hypotheses to account for the origin of the Avon Gorge polyploid Sorbus taxa are based primarily on morphological, cytological and peroxidase isozyme studies (Warburg 1962; Richards 1975; McAllister 1986; Proctor et al. 1989; Proctor & Groenhof 1992; Rich & Jermy 1998; Fig. 1). Molecular evidence supporting these hypotheses has come from two studies investigating relationships among all UK Sorbus taxa. Nelson-Jones et al. (2002) used restriction fragment length polymorphism (RFLP) variation in nuclear and mitochondrial DNA to ascribe the majority of the polyploid ‘intermediate’ Sorbus taxa to one of three major aggregates:  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1677 Nothosubgenus Tormaria Subgenus Aria S. aria Subgenus Torminaria S. bristoliensis 2n = 3x = 51 AAT S. decipiens S. torminalis 2n = 4x = 68 2n = 2x = 34 TT S. croceocarpa 2n = 4x = 68 S. latifolia 2n = 4x = 68 AATT 2n = 2x = 34 AA Nothosubgenus Soraria S. porrigentiformis Subgenus Sorbus 2n = 4x = 68 AAAA S. anglica S. eminens 2n = 4x = 68 AABB S. aucuparia 2n = 4x = 68 AAAA S. x thuringiaca 2n = 2x = 34 BB S. whiteana 2n = 2x = 34 AB 2n = 3x = 51 AAA S. wilmottiana 2n = 3x = 51 AAA S. intermedia 2n = 4x = 68 AABT Fig. 1 The current hypotheses for the origins of Sorbus taxa in the Avon Gorge, according to Rich & Jeremy (1998), and Nelson-Jones et al. (2002) excluding recently described taxa. Members of the nothosubgenus Tormaria and nothosubgenus Soraria are thought to represent allopolyploids, with Tormaria taxa having genomes inherited from both S. torminalis and a member of the subgenus Aria and Soraria taxa having genomes inherited from both S. aucuparia and a member of subgenus Aria. Each member of subgenus Aria is thought to comprise solely of genomes inherited directly or indirectly from S. aria. The proposed genome composition (Liljefors 1955; Richards 1975) is indicated with capital letters: A: subgenus Aria genome, B: S. aucuparia genome, T: S. torminalis genome. For chromosome count references see Bailey et al. (2008). subgenus Aria (formerly ‘Aria group’), nothosubgenus Soraria (formerly ‘Anglica group’), and nothosubgenus Tormaria (formerly ‘Latifolia group’), derived from the diploid female parents S. aria, S. aucuparia and S. torminalis, respectively. More recently, Chester et al. (2007) used plastid DNA microsatellite markers to determine the female parentage of endemic UK Sorbus taxa and produced results consistent with those of Nelson-Jones et al. (2002). Neither of these studies included all of the Avon Gorge Sorbus taxa and neither could ascertain the paternal parents of hybrid taxa. Here we describe an extensive molecular analysis of 227 individuals from 18 Sorbus taxa in the Avon Gorge and from two S. rupicola individuals using seven nuclear microsatellite markers and a chloroplast (cp)DNA marker. For each of the 15 intermediate (polyploid ⁄ putative apomictic) Sorbus taxa co-existing in the Avon Gorge we have assessed: (i) genetic diversity and whether reproduction is sexual or apomictic (obligate or facultative); (ii) the number of independent origins; (iii) maternal and paternal affinity; and (iv) evidence for backcrossing. This has allowed us to formulate an extensive picture of the evolutionary relationships between the Avon Gorge Sorbus taxa and to assess the extent of ongoing evolutionary processes occurring within and among the Sorbus taxa that comprise this unique agamic complex. material. Ten native species, four introduced species and four intermediate putative hybrids were sampled (see Table 2). Species identifications were based on morphological characteristics of leaves and berries, previously identified as robust diagnostic characters defining the different Sorbus species (Rich & Jermy 1998). Where possible leaf buds were collected from between 20 and 30 individuals for each taxon. The only material collected from outside of the Avon Gorge was from two S. rupicola trees, one from the Wye Valley and the other from South Wales. Sorbus rupicola was included in this study as it is the only polyploid Sorbus that is widespread in Northern Europe and it has been previously demonstrated to be involved in the origin of other intermediate Sorbus taxa (Robertson et al. 2004a, b). Even though S. rupicola is not currently present in the Avon Gorge, it may have been present post-glacially because populations can be found within 20 km of the Avon Gorge. Herbarium voucher material for each sampled individual was deposited at the National Museum of Wales (NMW), Cardiff. Samples were collected in a non-random method (selectively sampled from each patch in the Gorge) in order to maximize the representation of potential genotypes (Bayer 1990). GPS readings of tree locations and herbarium voucher details are included in Table S1 (Supporting Information). Materials and methods DNA extraction Plant material Field surveys of the Avon Gorge were undertaken during autumn 2004 to identify and collect Sorbus  2010 Blackwell Publishing Ltd DNA was extracted from frozen leaf buds using the Qiagen DNAeasy plant kit following the manufacturer’s instructions. 1678 A . R O B E R T S O N E T A L . Chloroplast DNA markers A single non-coding region of the Sorbus chloroplast genome, trnH–psbA was amplified using primers designed by Hamilton (1998). Polymerase chain reaction (PCR) mixtures (20 lL total volume) contained: one unit of Hotstart Taq DNA polymerase (Qiagen), 2 lL of 10· reaction buffer (Qiagen), 2 mM of each dNTP (Bioline), with 0.2 lM of each forward and reverse primer and 20–50 ng of genomic DNA. PCR reactions were performed in an MJ-Research PTC-200 Peltier Thermal Cycler using the following programme: 1 step at 95 C for 15 min followed by 35 cycles of 1 min at 94 C, 1 min at 55 C, 1 min at 72 C and a 10 min extension step of 72 C. PCR products from 12 of the Avon Gorge taxa were purified using QIAquick spin columns (Qiagen). These purified PCR products were sequenced using the Big Dye terminator cycle sequence kit (ABI) and an ABI 3730xl sequencer (Perkin-Elmer). Sequences were analysed using ChromasPro sequence analysis software (Technelysium Pty Ltd). For RFLP analysis of PCR products, digestions were conducted with the restriction enzyme Tsp5901 (New England Biolabs), according to the manufacturer’s instructions. Digestion products were separated and analysed by horizontal gel electrophoresis using 3.0% Metaphor agarose gels. Gels were prepared in 1 · TBE buffer. Running times were varied to ensure that the DNA products migrated three-quarters of the way down the gel. Gels were stained with ethidium bromide for 15 min and de-stained for 15 min before visualization under UV light. Gels were photographed with the Bio Image system to retain a permanent record. Nuclear microsatellite markers Seven nuclear microsatellite loci were amplified using three sets of primers developed for Malus · domestica (Gianfranceschi et al. 1998) and four pairs developed for Sorbus torminalis (Oddou-Muratorio et al. 2001). Four of these primer-pairs were redesigned, from sequences deposited in Genbank, to facilitate amplification of a wide range of Sorbus taxa (Table 1). Five of the forward primers were labelled with fluorescent dyes. Reverse primers were tailed with a non-target-specific sequence to overcome allele sizing problems associated with the primer sequence-dependant adenylation of PCR products by most Taq polymerases (Brownstein et al. 1996). PCR reaction mixtures (20 lL total volume) contained the following components ⁄ concentrations: one unit of Hotstart Taq DNA polymerase (Qiagen), 2 lL of 10· reaction buffer (components here) (Qiagen), 2 mM of each dNTP (Bioline), with 0.2 lM of each forward and reverse primer and 20–50 ng of genomic DNA. PCR reactions were performed in a MJ-Research PTC-200 Peltier Thermal Cycler. The following programme was used: 1 step at 95 C for 15 min followed by 35 cycles of 1 min at 94 C, 1 min at 55 C, 1 min and 30 s at Table 1 Nucleotide sequences of nuclear and cpDNA primers used in this study Locus Dye Primers Repeat Na Ha *CH01F02† 6-Fam (GA)22 11 *CH02D11† Vic MSS5‡ 6-Fam MSS13‡ 6-Fam MSS16‡ Vic *CH01F09† No dye MSS6‡ No dye trnH§ psbA§ No dye Forward CCACATTAGAGCAGTTGAGGATGA Reverse ATAGGGTAGCAGCAGATGGTTGT Forward AAATAAGCGTCCAGAGCAACAG Reverse GGGACAAAATCTCACAAACAGA Forward CCCCAACAACATTTTTCTCC Reverse CCTCTCGCTCTTTGCCTCT Forward GAAAATTCCTTCCCGAACTTCAT Reverse AACTCACTCGGATTTTGGAACCT Forward ATGTCACATCTCTCCCCTTGTGT Reverse TTTTGCCCTCAAAGAATGCCTTA Forward ATGTACATCAAAGTGTGGATTG Reverse GGCGCTTTCCAACACATC Forward CGAAACTCAAAAACGAAATCAA Reverse ACGGGAGAGAAACTCAAGACC ACTGCCTTGATCCACTTGGC CGAAGCTCCATCTACAAATGG (AG)21 26 (GA)19 15 (GA)12 11 (GA)28 23 (AG)22 (CA)14 4 *CH01F02, CH02D11, and CH01F09 are also referred to in this paper by their abbreviated names F02, D11, and F09 respectively. †Microsatellite primers derived from Malus · domestica DNA (Gianfranceschi et al. 1998). ‡Microsatellite primers derived from S. torminalis (Oddou-Muratorio et al. 2001). §Chloroplast primers from Hamilton (1998). Na = number of alleles amplified across all Avon gorge taxa. Ha = number of chloroplast haplotypes detected.  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1679 72 C and a 10 min extension step of 72 C. PCR amplification products from the labelled primers were separated on a laser-based capillary electrophoresis instrument, the ABI 3730xl. Alleles were sized relative to an internal size standard and resulting electropherograms were analysed using GeneMarker genotyping software. Each individual electropherogram was scored manually. The PCR amplification products from the two unlabelled primers were separated by horizontal agarose gel electrophoresis (2%). Gels were stained with ethidium bromide and visualized under UV light. Data analysis Models used to detect genetic diversity within and among species and those that determine population structure and hybrid origins, rely on the assumptions that allele frequencies can be readily determined and that source populations are in Hardy–Weinberg equilibrium. Apomictic taxa are generally polyploid and contain genomes where individual loci are not free to recombine, thus violating the principal assumption of most genetic models. To complicate matters further, allele frequencies are difficult to determine for polyploid taxa. The number of different alleles detected in a polyploid individual is often lower than its ploidy level. For example, a diploid individual carrying alleles A and B at a locus must have one copy of each, while a triploid individual carrying alleles A and B may have two copies of either A or B and one of two different genotypes, AAB or ABB. The most appropriate measure to study genetic diversity for apomictic taxa is to compile and compare multilocus genetic phenotypes. Therefore, for each individual analysed in this study, microsatellite allele phenotypes were determined for each of the seven amplified loci. These data were combined to construct the multilocus phenotypes of each sample. The numbers of individuals belonging to each multilocus phenotype were determined for each species. To estimate clonal diversity the following parameters were calculated: 1 the proportion of distinguishable genotypes (Ellstrand & Roose 1987) was measured as Ng ⁄ Ni, where Ng is the number of genotypes and Ni is the total number of individuals sampled. 2 Simpson’s diversity index (D) modified for finite sample size by Pielou (1969) measures the probability that two individuals selected at random from a population of N plants will have different genotypes. D can range from 0 to 1, with 1 being the maximum diversity. P Each index value was calculated as: D = 1 ) {ni(ni ) 1) ⁄ N(N ) 1)} where ni is the number of individuals of genotype i and N is the sample size.  2010 Blackwell Publishing Ltd 3 Genotypic evenness (Fager 1972) was measured as: E = D ) Dmin ⁄ Dmax ) Dmin, where Dmin = (G ) 1)(2N ) G) ⁄ N(N ) 1) and Dmax = (G ) 1)(N) ⁄ G(N ) 1) where G is the number of clones and N is the sample size. Index values can range from 0 for a population dominated by one genotype, to 1 for a population in which all genotypes are represented equally. Two simple measures of allele diversity, the total number of different alleles seen across all loci and the mean number of different alleles carried by each individual, averaged across loci, were calculated using F-Dash (Obbard et al. 2007). Ploidy level estimates were based on the maximum number of displayed alleles at a single locus. For example, at two loci, MSS5 and MSS16, three alleles were displayed for S. bristoliensis suggesting a triploid genome. Each multilocus phenotype was transformed into binary code and metric distances were computed between them using Dice similarity coefficients (Dice 1945) with NTSYSPC version 2.11a software (Rohlf 2002). The relationships among the multilocus phenotypes were visualised using principal coordinate analysis (PCO) with NTSYSPC software (Rohlf 2002) and neighbour-joining analysis (NJ) using the NEIGHBOUR programme in PHYLIP 3.6 (Felsenstein 2004). To determine the origins of the putative apomictic taxa simulated matings (crosses and selfs) were performed among and within the Sorbus taxa to determine whether any of the various mating possibilities (self-fertilization, interspecific cross-fertilization, or intraspecific cross-fertilization) could generate a specified polyploid multilocus phenotype. A table of results for each polyploid was compiled listing the number of allele mutations required for any mating to generate a phenotypic match. For any mating where the number of allele mutations required was zero, the taxa were considered potential progenitors. For phenotypes that could not be matched, the next best mating and the number of allele mutations required to make the match were recorded. All putative apomictic multilocus phenotypes were considered as potential parents of each other; there were no prior assumptions regarding taxonomic hierarchy used in this study. For the sexual taxa, all possible multilocus phenotypes that could be created from the observed population genepool were considered potential parents. Table S5a and S5b (Supporting Information) shows how this process was carried out for S. eminens. The only cross that matches the S. eminens multilocus phenotype is S. aria · S. porrigentiformis (Table S5b). It is also possible to derive from this table the number of alleles that are missing from each cross in order to match the multilocus phenotype of S. eminens. For example, five specific alleles are missing from the S. aria genepool for S. aria to be the sole parent of S. eminens (Table S5b) and three alleles are missing 1680 A . R O B E R T S O N E T A L . from the combined genepools of S. aria and S. rupicola for the origin of S. eminens to be S. aria · S. rupicola. Character compatibility analysis (Mes 1998) was used to determine if any of the observed differences seen among polyploid multilocus phenotypes was best explained by the accumulation of mutations or whether the character patterns observed are more parsimoniously accounted for by genetic recombination through any of the various mating possibilities outlined above. An incompatibility is illustrated by a comparison of potential patterns observed in pair wise, binary characters; if mutation is the sole mechanism for the generation of variation, only three of the four possible combinations of characters can be achieved in a lineage, excluding the presence of back mutations. If all four possible character combinations are present, this is deemed an incompatibility and is most parsimoniously explained by genetic exchange. Matrix incompatibilities were calculated, using the Jactax.exe function in PICA version 4.0 (Wilkinson 2001), for three taxonomic groups based on their cpDNA haplotypes, subgenus Aria, nothosubgenus Tormaria and S. torminalis and nothosubgenus Soraria and S. aucuparia. Multilocus phenotypes responsible for the greatest number of incompatibilities were successively deleted from the data set until all incompatibility was resolved or the number of genotypes exhausted. If removal of virtually all the genotypes is required to achieve matrix compatibility, it can be concluded that the relationships among these genotypes are essentially the result of recombination. The binary coded matrix, derived from the multilocus phenotypes, was used as the input for the Jactax programme, with the modification that the interdependence of markers was specified. This prevents the calculation of character incompatibility between alleles of the same microsatellite locus. Results Chloroplast (cp)DNA analysis Sorbus aria, S. aucuparia and S. torminalis each had unique trnH–psbA sequences. Two insertion ⁄ deletion polymorphisms (indels) were found that distinguished S. aria from S. torminalis and three nucleotide substitutions were found that distinguished S. aucuparia from both S. aria and S. torminalis (Fig. 2). Sorbus eminens, S. porrigentiformis, S. wilmottiana, S. whiteana and Fig. 2 Chloroplast DNA sequences for the intergenic region trnH–psbA. Nucleotide differences are shaded and the three nucleotide substitutions that distinguish S. aucuparia from S. aria ⁄ S. torminalis are highlighted (*). Regions A and B represent sequence deletions ⁄ insertions that distinguish S. aria from S. torminalis. NNN: region not clearly sequenced due to polymerase slippage. GenBank accession numbers are given in brackets.  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1681 S. latifolia (with the exception of a single nucleotide substitution) each had sequences that matched S. aria (Fig. 2). Two taxa, S. bristoliensis and S. croceocarpa, had sequences identical to S. torminalis (Fig. 2). Sorbus aucuparia-type sequences were found in S. anglica, S. · thuringiaca and S. intermedia (Fig. 2). One of the nucleotide substitutions that differentiated both S. aria and S. torminalis from S. aucuparia was situated within a Tsp5901 restriction digestion site (AATT). By digesting the PCR products with Tsp5901, S. aria, S. aucuparia and S. torminalis each had a unique RFLP pattern (data not shown). This diagnostic PCR-RFLP assay provided a convenient and efficient method for determining the chloroplast haplotypes of large numbers of Avon Gorge Sorbus taxa. When applied in this manner, the test revealed that there was no variation in restriction site pattern within any of the sexual taxa, S. aria, S. torminalis or S. aucuparia or, as expected, within any of the polyploid taxa. The S. aria type restriction pattern was also found in S. · robertsonii, S. leighensis and S. · avonensis, and the S. torminalis type was also found in S. decipiens, and S. · houstoniae. These haplotypes are consistent with those reported by Lemche (1999) and Chester et al. (2007). Nuclear microsatellite analysis In total, 86 alleles from five labelled microsatellite loci were amplified from 227 adult Sorbus individuals. The number of alleles amplified at each locus across all taxa ranged from 11 at locus F02 to 26 at locus D11 (Table 1). The total number of different alleles displayed by each taxon ranged from 10 for S. leighensis, and S. · avonensis, to 38 for S. torminalis (Tables 2, 3 and Table S2, Supporting Information). The mean number of different alleles carried by each individual averaged across all loci ranged from 1.6 per locus in S. aria to 3.4 per locus in S. porrigentiformis (Table 2). Amplification products from the two unlabelled microsatellite loci, CH01F09 and MSS6, proved to be genome-specific, with CH01F09 only being amplified in S. aria and associated hybrids and MSS6 only amplified in S. torminalis and associated hybrids. These two loci were scored on a presence-absence basis for each of the 227 adult individuals (Table 3, Table S2). When data from all seven loci were combined, the total number of detectable multilocus phenotypes scored across all taxa was 86. With the exception of S. eminens (with two multilocus Table 2 Genetic diversity measures for the Avon Gorge Sorbus taxa Taxon Native sexual taxa S. aria S. aucuparia S. torminalis S. · thuringiaca Native polyploids S. eminens S. · robertsonii S. porrigentiformis S. leighensis S. · avonensis S. wilmottiana S. whiteana S. bristoliensis S. · houstonii S. anglica Introduced polyploids S. croceocarpa S. decipiens S. latifolia S. intermedia Taxon outside of Avon Gorge S. rupicola N Ni Xo Xe A’ H’ Ng Ng ⁄ Ni D E >1000 20 50 2 34 14 20 2 2 2 2 2 2 2 2 2 26 29 38 14 1.6 1.6 1.7 1.7 34 14 19 2 — — — — — — — — — — — — 80 1 80 80 2 40 20 300 1 40 27 1 22 4 2 26 5 32 1 20 4 3 4 3 3 3 3 3 4 3 4 na 4 na na 3 3 3 na 3, 4 15 13 17 10 10 11 12 11 15 15 3 2.6 3.4 2 2 2.2 2.4 2.2 3 3 2 1 1 1 1 1 1 1 1 1 0.07 — 0.04 — — 0.03 0.2 0.03 — 0.05 0.07 — — — — — — — — — 0.00 — — — — — — — — — 6 3 10 6 4 1 6 4 4 3 4 4 4 4 4 4 16 12 16 16 3.2 2.4 3.2 3.2 1 1 1 1 — — 0.16 — na 2 4 4 16 3.2 1 — — — — — — — — — — — — — N = estimated population size in the Avon Gorge; Ni = number of individuals sampled; Xo = observed ploidy, based on the maximum number of displayed alleles at a single locus; Xe = expected ploidy, based on published cytological work; A¢ = total number of different alleles seen across all loci; H¢ = mean number of different alleles carried by each individual averaged across all loci; Ng = number of multilocus phenotypes detected; Ng ⁄ Ni = proportion of clones detected; D = estimate of multilocus phenotype diversity (multiclonal spp. only); E = multilocus phenotype diversity (multiclonal spp. only).  2010 Blackwell Publishing Ltd 1682 A . R O B E R T S O N E T A L . Table 3 Multilocus phenotypes for the polyploid Sorbus taxa Alleles at locus: Taxon F02 Subgenus Aria S. eminens 196,202,206 S. eminens 154 196,202,206 S. x robertsonii 196,200,202 S. porrigentiformis 196,202,206,208 S. leighensis 196,202 S. x avonensis 202,206 S. wilmottiana 200,208 S. whiteana 200,208 S. rupicola 196,204,206,214 Nothosubgenus Tormaria S. bristoliensis 192,202 S. x houstonii 192,196,202 S. croceocarpa 192,198,202 S. decipiens 194,198,208 S. latifolia 192,194,200,206 Nothosubgenus Soraria S. anglica 192,202,206 S. intermedia 194,196,206 MSS5 D11 MSS13 120,126,134,142 120,126 120,126,142 120,132,136,142 120,140,142 120,126,140 126,132,140 126,134,140 124,132,136,180 159,163,179 163,179 163,179 159,205 159,193 159,179 159,171 159,189,193 159,169,175 199,201,205 199,201,205 203,205 199,201,203,209 201,205 199,205 199,205 199,205 199,201,203 120,134,144 120,134,140,144 126,132,140 126,130,134 128,132,142 159 159,179 159,179,196 159,163 159,169 126,136,142 124,128,132 phenotypes), each of putative apomictic taxa displayed just one multilocus phenotype (Table 3). In contrast, every S. aria and every S. aucuparia individual and 19 out of 20 S. torminalis individuals had a unique multilocus genotype. All individuals genotyped for the sexual taxa displayed either one or two alleles at each locus, confirming them to be diploid. In contrast, each individual genotyped among the putative apomictic taxa displayed more than two alleles, usually three or four, at a minimum of one locus, confirming that they were all polyploid (Table 3). Of the 86 microsatellite alleles, distributed among the 19 Sorbus taxa investigated, 43 were shared between the sexual taxa and the polyploids, 28 were found only among the sexual taxa and 15 were restricted to the polyploids. Excluding apomicts the sexual taxa had high numbers of characteristic alleles, 13 for S. aria, 18 for S. aucuparia and 22 for S. torminalis (Table S2). Genetic similarity and dissimilarity, among the 86 multilocus phenotypes was calculated and the subsequent matrices were used as the input for principal coordinate analysis (PCO) and neighbour-joining (NJ) analysis respectively. The PCO showed complete differentiation of the three sexual taxa S. aria, S. torminalis and S. aucuparia (Fig. 3). Members of subgenus Aria formed a distinct group but with a clear subdivision, one sub-group consisting solely of S. aria individuals and the other sub-group consisting of a mix of S. aria and the subgenus Aria polyploids. All members of nothosubgenus Tormaria, with the exception of S. decipiens, formed a MSS16 F09 MSS6 166,178 166 166,172,178 166,170,178 166 166 166,172 166,172 166,170 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 199,201 199,201,205 187,199,201,205 195 187,199,207 166,178,198 166,178,198 166,168,198 166,192,200 166,168,196,234 1 1 1 1 1 1 1 1 0 1 157,159,205 199,201,203 162,166,170 1 0 153,159,165 199,201,203 162,166,170,182 1 1 single group lying between subgenus Aria and S. torminalis. The slightly anomalous position of the central European S. decipiens may represent further hybridization between a member of the nothosubgenus Tormaria and a member of the subgenus Aria or be a consequence of its parent taxa not being included in the genetic analysis. Sorbus · thuringiaca individuals fell between the S. aria subgroup, of the subgenus Aria, and S. aucuparia. Sorbus anglica sat between the polyploid subgroup of the subgenus Aria and S. aucuparia. The position of S. intermedia was central among the three main groups. This result is consistent with the findings of Nelson-Jones et al. (2002) who found that S. intermedia had inherited a proportion of its genome from each of S. aria sensu lato, S. torminalis and S. aucuparia. The results of the NJ analysis concur with those of the PCO analysis with both showing a clear differentiation of the three sexual taxa S. aria, S. torminalis and S. aucuparia and also the intermediate polyploid taxa forming clusters near or between the proposed progenitor taxa (Fig. 4). What the NJ tree does show, which is not apparent in the PCO, is the polyphyletic nature of the polyploids within both the subgenus Aria and the nothosubgenus Tormaria. Where clearly individuals within each microspecies are monophyletic, as they are genetically identical (with the exception of S. eminens), each polyploid group as a whole consists of a number of monophyletic clusters, four within the subgenus Aria and three within the nothosubgenus Tormaria (Fig. 4). It must be noted however that without statistical support  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1683 Au 0.64 Au Au Au Au Au Au Au Th3 Au Au 0.38 Au Au Au Au Th5 Dim–2 Aria group 1 0.11 Ar Ar Ar Ar ArArArAr Ar Ar Ar Ar Ar Ar Ar ArAr ArAr Ar Ar Ar Ar Ar Ebc Ar ArWil ArAr Ar Ar ArAr Whi EmB PoB ArPoC S. aucuparia Dec Fig. 3 Principal coordinate analysis of the similarity relationships among the 86 Avon Gorge Sorbus multilocus phenotypes, showing the first and second axes. The major groupings are circled and labelled. Subgenus Aria is divided into two; Aria group 1 consists solely of S. aria and Aria group 2 consists of both S. aria and subgenus Aria polyploids. Taxon abbreviations as per Table 4. The proportion of the variation accounted for by these two axes is 31%. Int Ang Rup S. torminalis Por Em –0.15 Aria group 2 Cro Bbc Nothosubgenus Tormaria –0.42 –0.54 –0.26 0.02 To Br Lat To ToTo To To ToTo To To To To To ToTo To To To To 0.30 0.58 Dim–1 values phylogenetic conclusions from NJ trees must be interpreted with caution. Mating simulations Simulated matings (crosses and selfs) were performed among and within the 19 Sorbus taxa to determine whether any of the various mating possibilities (self-fertilization, interspecific cross-fertilization or intraspecific cross-fertilization) could generate a specified polyploid multilocus phenotype. A table of results for each polyploid was compiled listing the number of allele mutations required for any mating to generate a phenotypic match (Table S3a–17b, Supporting Information). For any mating where the number of allele mutations required was zero, the taxa were considered potential progenitors (Table 4). For phenotypes that could not be matched, the next best mating and the number of allele mutations required to make the match were recorded (Table 4). Within the subgenus Aria single genotype matches were found for S. eminens (S. aria · S. porrigentiformis) and for S. · robertsonii (S. aria · S. eminens). Multiple genotype matches were found for S. wilmottiana (two), S. leighensis (two) and S. · avonensis (12). The genotypes of S. porrigentiformis, S. rupicola and S. whiteana could not be matched (Table 4). No single genotype matches were found for any of the nothosubgenus Tormaria taxa. Two taxa, S. bristoliensis (eight) and S. ·  2010 Blackwell Publishing Ltd houstoniae (three), had multiple phenotype matches (Table 4). The genotypes of the remaining three taxa, S. croceocarpa, S. decipiens and S. latifolia could not be matched, which is consistent with their treatment as introduced species in the British flora. Single genotype matches were found for S. anglica (S. aucuparia · S. porrigentiformis) and for S. · thuringiaca genotype 252 (S. aucuparia · S. aria). The genotype of S. intermedia and S. · thuringiaca genotype 232 could not be matched (Table 4), the former consistent with it being an introduced species. Incompatibility analysis Based on cpDNA haplotypes each of the three taxonomic groups had high incompatibility counts, subgenus Aria 266, nothosubgenus Tormaria and S. torminalis 285 and nothosubgenus Soraria and S. aucuparia 165 (Fig. 5). Within each of these groups the majority of genotypes had to be removed to allow complete compatibility of the data set, subgenus Aria 89%, nothosubgenus Tormaria and S. torminalis 76% and nothosubgenus Soraria and S. aucuparia 66%. The only compatibility found among the polyploid taxa was between the two multilocus phenotypes of S. eminens. Each of the remaining polyploids, including S. eminens, (the subgenus Aria was reanalysed without the single S. eminens phenotype 154) was removed from the 1684 A . R O B E R T S O N E T A L . 0.1 To19 To18 To14 To11 To3 To7 To10To12 To16 To15 To13 To5 To2 To9 To8 To6 To20 To4 To1 To17 Au6 Au1 Au5 Au2 Au4 Au3 Au8 Au10Au11 Au14 Au13 Au7 Au12 Lat Au9 Th3 Br Cro Bbc Rup Int Th5 Por Ang Whi Wil Dec Ar7 Ar10 PoB Ar23 Ar9 Ar6 Ar33 Ar4 Ar15 Ar27 Ar21 Ar1Ar26 Ar24 Ar20 Ar22 Ar17 Ar5 Ar14 Ar28 Ar11 Ar16 Ar34 Ar2 Ar32 Ar12 Ar19 Ar8 Ar3 Ar31 Ar29 Ar25 EmB Ar30 Ar13 PoC Em Ar18 Ebc Fig. 4 Neighbour-joining tree of 86 Avon Gorge Sorbus multilocus phenotypes. The tree was constructed using a dissimilarity matrix, 1-Dice similarity estimates, based on microsatellite analysis. Taxon abbreviations are as Table 4. analysis before compatibility had been reached. It can be concluded that relationships among the majority of the multilocus phenotypes within each group are essentially the result of recombination, as opposed to mutation. The origin of each of the polyploids, with the exception of the single S. eminens 154 genotype, was therefore found to be incompatible with clonal reproduction. Discussion This study has shown conclusively that the generation of genetic novelty within the agamic Sorbus complex of the Avon Gorge has been driven primarily by a series of interspecific hybridisations and backcrosses among closely related taxa, with each new genotype being fixed and propagated via apomictic reproduction. The result is a series of new closely related taxa, microspecies and hybrids (as defined by Rich et al. 2009), that are reproductively isolated from each other, but which occasionally participate in further sexual hybridisation events leading to a complex pattern of ongoing reticulate evolution of Sorbus within the Avon Gorge. The distribution of nuclear microsatellite alleles within and among the various Sorbus taxa identified those which reproduce sexually and those which reproduce primarily by apomixis. For each of the sexually reproducing taxa (S. aria, S. aucuparia and S. torminalis) each individual (with the exception of two S. torminalis individuals) had a unique multilocus genotype, as would be predicted for obligate (self-incompatible) outcrossing taxa and, with the exception of S. eminens, each of the putative apomictic taxa had a single multilocus genotype, as would be expected for asexually reproducing, apomictic, taxa. The microsatellite allele data also allowed confirmation of ploidy levels among the different taxa. Sexual taxa were diploid and apomictic taxa polyploid, which is largely consistent with other  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1685 Table 4 Origins of the Avon Gorge Sorbus taxa compiled from the results of the cpDNA and nuclear DNA analysis Mating simulation matches Taxa (Key) CpDNA type PCO position S. eminens (Em) S. · robertsonii (EbC) S. porrigentiformis (Po) S. aria S. aria S. aria Aria group Aria group Aria group S. aria (Ar) · Po Ar · Em S. leighensis (PoB) S. · avonensis (PoC) S. aria S. aria Aria group Aria group S. wilmottiana (Wi) S. whiteana (Wh) S. aria S. aria Aria group Aria group Ar · Em, Ar · Po 12 possible crosses see supplementary data table Ar · Po, Ar · Wh S. rupicola (Ru) S. bristoliensis (Br) S. aria S. torminalis Aria group Nothosubgenus Tormaria S. croceocarpa (Cr) S. torminalis Nothosubgenus Tormaria S. decipiens (De) S. torminalis S. latifolia (La) S. aria Groups with S. intermedia and S. anglica Nothosubgenus Tormaria S. · houstonii (BrB) S. torminalis Nothosubgenus Tormaria S. anglica (An) S. aucuparia S. intermedia (In) S. aucuparia S. thuringiaca 232 (Th3) S. aucuparia Intermediate between S. aria, S. aucuparia Intermediate between S. aria, S. aucuparia and S. torminalis Intermediate between S. aria and S. aucuparia S. thuringiaca 252 (Th5) S. aucuparia Intermediate between S. aria and S. aucuparia estimates made from cytological studies (Bailey et al. 2008). The total number of alleles recorded across all loci ranged from 10 to 17 for the polyploid taxa and from 26 to 38 for the sexual taxa. The lower values of A¢ (the total number of different alleles scored across loci, Table 2) for the polyploid taxa suggest monophyletic origins (single hybridization events) and subsequent reproductive isolation through apomixis. In general sexual taxa have higher levels of A¢ due to their  2010 Blackwell Publishing Ltd Next best mating simulation match, (n) = number of allele mutations required to make a match Em · Ru (3) EbC · Ru (3) BrB · An (3), Em · An (3), Int · Em (3) Ar · An, Ar · Br, Ar · BrB, Ar · Cr, Ar · De, Ar · Em, Ar · Po, Ar · PoC, Ar · Wi, all (2) Em · S. torminalis (Tor), Po · Tor, BrB · BrB, BrB · Em, BrB · Po, BrB · Pob, BrB · PoC, BrB · Tor Ar · Tor (2), Ar · Brb (2) PoC · Tor (2), Cr · PoC (2), Cr · Wi(2) Ar · Br, Em · Tor, EmC · Tor S. aucuparia (Au) ·Po Au · Po (2), Au · Ru (2) Ar · Au (3), Au · Au (3), Au · In (3) Ar · Au ability to recombine genotypes among individuals. The average number of alleles at each locus, carried by each individual, ranged from 2.0 to 3.4 for the polyploid taxa and from 1.6 to 1.7 for the diploid taxa. High values of H¢ (the average number of alleles at each locus carried by each individual, Table 2) for polyploid taxa are consistent with these taxa being the progeny of genetically diverse parents with this genetic diversity subsequently being fixed in them by apomictic reproduction. Matrix incompatibility 1686 A . R O B E R T S O N E T A L . 300 250 T A S 200 150 100 50 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 Number of phenotypes deleted Fig. 5 Reduction of matrix incompatibility upon successive deletion of multilocus phenotypes in each of the three taxonomic groups based on cpDNA haplotypes. Subgenus Aria = A, Nothosubgenus Tormaria and S. torminalis = T and Nothosubgenus Soraria and S. aucuparia = S. Hybrid origins of the Avon Gorge Sorbus microspecies Comparisons of cpDNA type and multilocus microsatellite genotypes among the 19 Sorbus taxa of the Avon Gorge allowed us to make predictions about the origins of the polyploid apomictic taxa. The cpDNA type of S. eminens, S. porrigentiformis, S. wilmottiana and S. whiteana matched that of S. aria and the PCO and NJ analyses placed these four taxa within the subgenus Aria cluster. These results are consistent with previous morphological, cytological and molecular studies that proposed S. aria was at least one of the parents (directly or indirectly) of these taxa and the other parent, if not S. aria, was a member of the subgenus Aria (Warburg 1952; Proctor & Groenhof 1992; Nelson-Jones et al. 2002; Rich & Houston 2006; Chester et al. 2007). Nuclear microsatellite markers used in the present study were variable in their ability to determine taxon origins so we were unable to determine the precise origins of all the taxa. However, they were able to show, through character compatibility analysis, that the origin of each polyploid (with the exception of the single S. eminens genotype 154) was incompatible with clonal reproduction and so was explained most parsimoniously by hybridization and genetic recombination. The origin of S. eminens was conclusive: only by crossing S. aria with S. porrigentiformis could the multilocus phenotype of S. eminens be matched. All other parental combinations (intraspecific and interspecific) and selfing scenarios were excluded. These results are consistent with the findings of Nelson-Jones et al. (2002) which placed S. eminens and S. porrigentiformis in the same subgenus and with those of Proctor & Groenhof (1992), who found that S. porrigentiformis, type D, found on Seven Sisters Rocks in the Wye Valley, shared a similar peroxidase phenotype with S. eminens. Nelson-Jones et al. (2002) went on to propose an autopolyploid origin for S. eminens from S. aria. However, our data show conclusively that S. eminens is the product of an interspecific cross (S. aria · S. porrigentiformis), albeit from two species within the S. aria group. The precise origin of S. porrigentiformis could not be determined because its multilocus phenotype could not be matched by any parental recombination of microsatellite alleles from within the Avon Gorge, even though S. aria is clearly one of its parents on morphological grounds. The most likely explanation for this result is that S. porrigentiformis originated outside the Avon Gorge. Previous studies of S. porrigentiformis revealed varying degrees of genetic variation over wide geographical scales and concluded that this is best explained by inferring numerous independent origins (Proctor & Groenhof 1992; Nelson-Jones et al. 2002; Chester et al. 2007). Such a conclusion however must be treated with caution because there are other equally likely sources of genetic variation in apomictic species, such as mutation and facultative sexuality (reviewed in Hörandl & Paun 2007). Indeed the results of our study suggest that genetic variation among S. porrigentiformislike clones could be the consequence of S. porrigentiformis backcrossing with one of its parents. Two S. porrigentiformis-like clones were found in the Avon Gorge, S. leighensis and S. · avonensis, and it was shown that the most likely origin of these taxa was from the cross S. aria · S. porrigentiformis. The PCO analysis placed S. leighensis and S. · avonensis in an intermediate position between S. porrigentiformis and S. aria and the results of the mating simulations supported these placements. Two potential crosses matched the multilocus phenotype of S. leighensis: S. eminens · S. aria and S. aria · S. porrigentiformis. As S. eminens has been shown to be the progeny of S. porrigentiformis, it is clear that S. leighensis originated from a S. porrigentiformis backcross, either directly or indirectly, with S. aria. The mating simulation results for S. · avonensis were not so conclusive as there were 12 possible cross matches. The crosses S. eminens · S. aria and S. aria · S. porrigentiformis are among the possibilities and S. porrigentiformis was involved either directly or indirectly in the remaining crosses. These results clearly show that S. leighensis and S. · avonensis had taxonomically different parents to S. porrigentiformis so can be ascribed independent taxonomic status (Rich et al. 2009). For similar reasons, S. · robertsonii (Rich et al. 2009) was not considered a clone of S. eminens since the PCO and mating simulations showed conclusively that it originated from the cross S. aria · S. eminens. Two potential crosses matched the multilocus phenotype of S. wilmottiana: S. aria · S. porrigentiformis and S. aria · S. whiteana. Support for the cross S. aria · S. porrigentiformis being responsible for the origin of  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1687 S. wilmottiana was also found by Chester et al. (2007). Morphologically similar to S. wilmotianna, S. whiteana is a recently described polyploid (Rich & Houston 2006) also placed in subgenus Aria by its authors and this designation is supported by our data. The cpDNA type of S. whiteana matches that of S. aria and the PCO analysis places it clearly within the subgenus Aria cluster. The results of the mating simulations were however inconclusive as no match was found for the multilocus phenotype of S. whiteana. It is likely that S. whiteana originated outside the Avon Gorge, presumably in the Wye Valley, which is the only other area where it is found (Rich & Houston 2006). The cpDNA type of S. bristoliensis matched that of S. torminalis and the PCO and NJ analyses placed S. bristoliensis in an intermediate position between subgenus Aria and S. torminalis. These results are consistent with previous morphological, cytological and molecular studies (Wilmott 1934; Warburg 1952; Proctor & Groenhof 1992; Nelson-Jones et al. 2002; Chester et al. 2007), which suggested that S. torminalis was the female parent of S. bristoliensis and the paternal parent was from subgenus Aria. The results of our mating simulations for S. bristoliensis were not conclusive and it is possible that either S. eminens or S. porrigentiformis (from subgenus Aria) is the paternal parent. Nevertheless, six other simulated crosses also gave the multilocus phenotype of S. bristoliensis, but all of these included progeny of either S. porrigentiformis or S. bristoliensis, so were not considered potential paternal parents of S. bristoliensis. Closely related to S. bristoliensis was the new taxon S. · houstoniae, which closely resembled (phenotypically) S. bristoliensis and could easily be mistaken for another S. bristoliensis clone. However, S. · houstoniae was found to be a tetraploid, rather than a triploid, and upon closer inspection appeared morphologically closer to S. aria than S. bristoliensis (Rich et al. 2009). The cpDNA type of S. · houstoniae matched that of S. torminalis and the PCO analysis placed it in an intermediate position between S. bristoliensis and subgenus Aria. The mating simulations were also consistent with these results and gave three possible crosses that matched the origin of S. · houstoniae: S. aria · S. bristoliensis, S. eminens · S. torminalis and S. · robertsonii · S. torminalis. The most likely cross was concluded to be S. aria · S. bristoliensis, which would require that an unreduced triploid S. bristoliensis egg be fertilized by a reduced haploid S. aria sperm. It is not uncommon for an unreduced triploid egg to be fertilized by a reduced haploid sperm in the genus Sorbus (Liljefors 1953; Robertson et al. 2004b), making this seemingly unlikely sexual scenario the likely source of S. · houstoniae B. The S. eminens · S. torminalis cross would require either S. torminalis to produce an unreduced egg in a cross with a reduced  2010 Blackwell Publishing Ltd diploid S. eminens male gamete or S. torminalis producing a normal haploid egg crossed with an unbalanced triploid S. eminens male gamete. The S. · robertsonii · S. torminalis cross would require similar gamete formations to S. aria · S. bristoliensis except that on this occasion the subgenus Aria member, S. eminens, would need to produce an unbalanced triploid male gamete that crossed with a reduced haploid S. torminalis egg. Unreduced S. torminalis gametes and unbalanced subgenus Aria male gametes have not yet been recorded in Sorbus (Liljefors 1953; Robertson et al. 2004b). The cpDNA type of S. anglica matched that of S. aucuparia and the PCO analysis placed S. anglica in an intermediate position between subgenus Aria and S. aucuparia, consistent with previous studies (Warburg 1952; Proctor & Groenhof 1992; Nelson-Jones et al. 2002; Chester et al. 2007). The results of the mating simulations were conclusive with only one simulation matching the multilocus phenotype of S. anglica: S. aucuparia · S. porrigentiformis. This result is consistent with the findings of Proctor & Groenhof (1992), who suggested, based on isozyme studies, that S. porrigentiformis was likely to be the subgenus Aria member that crossed with S. aucuparia. Continuing evolution of Sorbus in the Avon Gorge Hybridization and backcrossing appear largely responsible for ongoing evolution of new apomictic polyploid Sorbus taxa within the Avon Gorge. Indeed, similar evolutionary dynamics have been described for Sorbus in other UK and Scandinavian populations (Liljefors 1955; Bolstad & Salvesen 1999; Nelson-Jones et al. 2002; Robertson et al. 2004a, b; Houston et al. 2009) and such processes appear common among agamic complexes as mechanisms for generating genetic and phenotypic novelty that is then maintained by natural selection (Hörandl et al. 2001; Hörandl 2006; Hörandl & Paun 2007; Whitton et al. 2008). The frozen niche variation model (Vrijenhoek 1984) proposes that such recurrently formed apomictic lineages ‘freeze’ subsets of parental genetic variation that allows for niche partitioning among the apomictic lineages. Such apomictic lineages have narrower ecological niches than sexual taxa but collectively they use the resource space of their environment more effectively than sexuals (Vrijenhoek 1984; Hörandl 2006). The model predicts that apomicts will be better able to colonize spatially heterogeneous environments than sexuals (Hörandl 2006). The Avon Gorge represents a particularly heterogeneous environment with extreme local niche variation associated with, for instance: altitude, gradient, soil depth, shade and moisture and anecdotal observation suggests that apomictic polyploids are more likely to be found in the more 1688 A . R O B E R T S O N E T A L . ‘extreme’ niches than their sexual counterparts (Hiscock et al. unpublished observations). These observations are therefore consistent with the frozen niche variation model even though the sexual diploid taxa S. aria, S. torminalis and S. aucuparia coexist with the polyploid apomicts within the Gorge, albeit perhaps within distinct niches. Further ecological investigations will be needed to shed light on the niche preferences of sexual and apomictic Sorbus taxa within the Avon Gorge. The general lack of genetic diversity found within the Avon Gorge Sorbus polyploids was interesting given that, with the possible exception of S. porrigentiformis and S. whiteana, both parents of each taxon are still found in the Gorge and so could provide opportunities for additional hybridisations and backcrosses. This lack of genetic variation within the microspecies could indicate relatively recent evolutionary origins of the apomictic polyploids, probably since the last ice-age when the Avon Gorge is presumed to have acted as an important refugium for native plants during the forest maxima (Pigott & Walters 1954; Pring 1961). Alternatively, sexual reproduction may be an extremely rare event among Sorbus apomicts in the Avon Gorge, because of the need for generation of unreduced or unbalanced gametes. For instance, in the origin of S. eminens, and probably many other Sorbus polyploids including S. porrigentiformis and S. rupicola, one or both of their parents will have needed to produce unreduced or unbalanced gametes to attain their formation. Sorbus eminens has inherited two copies of its genome from S. aria and two from S. porrigentiformis. The two copies inherited from S. porrigentiformis should not have posed any significant problems as it represents a normal meiotic reduction for a tetraploid. The two copies inherited from S. aria require the production of an unreduced gamete. Unreduced gamete formation has been shown to occur in Sorbus by Liljefors (1953) and Aas et al. (1994) where it has been proposed to be involved in the origin of a number of taxa such as S. bohemica in central Europe (Jankun & Kovanda 1988). If the production of such unreduced gametes is rare, then the lack of multiple origins for S. eminens and possibly other polyploid taxa, is clear. This molecular marker analysis of the origins of the Avon Gorge Sorbus complex has shown that, as with the Sorbus rupicola–S. arranensis–S. pseudofennica-S. pseudomeinichii sequence on Arran (Robertson et al. 2004a, b; Robertson & Sydes 2006) and its equivalent in Scandinavia, the S. meinichii complex (Bolstad & Salvesen 1999), repeated backcrossing of apomictic derivatives with parental taxa is responsible for the generation of new apomictic taxa. As with Arran, evolution of Sorbus in the Avon Gorge is therefore ongoing and the dynamic nature of this evolutionary process indicates that conservation strategies for the rare endemic Sorbus species in the Avon Gorge should not be simply species-based, but rather should aim to conserve the evolutionary processes that have been responsible for their formation by conserving all Sorbus taxa within the Avon Gorge equally (see Ennos et al. 2005). At a time of unprecedented speeds of climate change the identification and preservation of such dynamic evolutionary processes will be essential for maintaining biodiversity. Acknowledgements We thank Matthew Hegarty and Christopher Thorogood for helpful comments on earlier versions of this manuscript and Christian Lexer and two anonymous referees for constructive suggestions on the final improvements. This work was funded by a research grant from the Leverhulme Trust. References Aas G, Maier J, Baltisberger M, Metzger S (1994) Morphology, isozyme variation, cytology and reproduction of hybrids between Sorbus aria (L.) Crantz and S. torminalis (L.) Crantz. Botanica Helvetica, 104, 195–214. Abbott RJ (2003) Sex, sunflowers, and speciation. Science, 301, 1189–1190. Adams KL, Wendel JF (2005) Polyploidy and genome evolution in plants. Current Opinion in Plant Biology, 8, 135– 141. Arnold ML (1997) Natural Hybridization and Evolution. Oxford University Press, New York. Bailey JP, Kay QON, McAllister H, Rich TCG (2008) Chromosome numbers in Sorbus L. (Rosaceae) in the British Isles. Watsonia, 27, 69–72. Bayer RJ (1990) Patterns of clonal diversity in the Antennaria rosea (Asteraceae) polyploid agamic complex. American Journal of Botany, 77, 1313–1319. Bicknell RA, Borst NK, Koltunow AM (2000) Monogenic inheritance of apomixis in two Hieracium species with distinct developmental mechanisms. Heredity, 84, 228–237. Bolstad AM, Salvesen PH (1999) Biosystematic studies of Sorbus meinichii (Rosaceae) at Moster, S. Norway. Nordic Journal of Botany, 19, 547–559. Briggs D, Walters SM (1997) Plant Variation and Evolution, 3rd edn. Cambridge University Press, Cambridge. Brownstein MJ, Carpten JD, Smith JR (1996) Modulation of non-templated nucleotide addition by Taq DNA polymerase: primer modifications that facilitate genotyping. BioTechniques, 20, 1004–1010. Challice J, Kovanda M (1978) Flavonoids as markers of taxonomic relationships in the genus Sorbus in Europe. Preslia, 50, 305–320. Cheffings CM, Farrell L (2005) Species Status No. 7. The vascular plant Red Data List for Great Britain, JNCC, Peterborough. Chen ZJ (2007) Genetic and epigenetic mechanisms for gene expression and phenotypic variation in plant polyploids. Annual Review of Plant Biology, 58, 377–406. Chester M, Cowan RS, Fay MF, Rich TCG (2007) Parentage of endemic Sorbus L. (Rosaceae) species in the British  2010 Blackwell Publishing Ltd EVOLUTION AND SPECIATION IN SORBUS 1689 Isles—evidence from plastid DNA. Botanical Journal of the Linnean Society, 154, 291–304. Coyne JA, Orr HA (2004) Speciation. Sinauer Associates, Massachusetts. Dice LR (1945) Measures of the amount of ecologic association between species. Ecology, 26, 297–302. Ellstrand NC, Roose ML (1987) Patterns of genotypic diversity in clonal plant species. American Journal of Botany, 74, 123–131. Ennos RA, French GC, Hollingsworth PM (2005) Conserving taxonomic complexity. Trends in Ecology and Evolution, 20, 164–168. Fager EW (1972) Diversity: a sampling study. American Naturalist, 106, 293–310. Fehrer J, Gemeinholzer B, Chrtek J, Bräutigam S. (2007) Incongruent plastid and nuclear DNA phylogenies reveal ancient intergeneric hybridization in Pilosella hawkweeds (Hieracium, Cichorieae, Asteraceae). Molecular Phylogenetics and Evolution, 42, 347–361. Felsenstein J (2004) Inferring Phylogenies. Sinauer Associates, Sunderland, Massachusetts. Gianfranceschi L, Seglias N, Tarchini R, Komjanc M, Gessler C (1998) Simple sequence repeats for the genetic analysis of apple. Theoretical and Applied Genetics, 96, 1069–1076. Grant V (1981) Plant Speciation, 2nd edn. Columbia University Press, New York. Grimanelli D, Leblanc O, Perotti E, Grossniklaus U (2001) Developmental genetics of gametophytic apomixis. Trends in Genetics, 17, 597–604. Hamilton MB (1999) Four primer pairs for the amplification of chloroplast intergene regions with intraspecific variation. Molecular Ecology, 8, 521–523. Hedlund T (1901) Monographie der Gattung Sorbus. Kungliga Svenska Vetenskaps-Akademiens Handlingar, 35, 1–147. Stockholm. Hegarty MJ, Hiscock SJ (2005) Hybrid speciation in plants: new insights from molecular studies. New Phytologist, 165, 411–423. Hegarty MJ, Hiscock SJ (2008) Genomic clues to the evolutionary success of polyploid plants. Current Biology, 18, R435–R444. Hollingsworth PM (2003) Taxonomic complexity, population genetics and plant conservation in Scotland. Botanical Journal of Scotland, 55, 55–63. Hörandl E (2006) The complex causality of geographical parthenogenesis. New Phytologist, 171, 525–538. Hörandl E, Paun O (2007) Patterns and sources of genetic diversity in apomictic plants: implications for evolutionary potentials and ecology. In: Apomixis: Evolution, Mechanisms and Perspective (eds Hörandl E, Grossniklaus U, Van Dijk PJ, Sharbel T), pp. 169–194. ARG-Gantner, Ruggell, Liechtenstein. Hörandl E, Jakubowsky G, Dobeš Ch (2001) Isozyme and morphological diversity within apomictic and sexual taxa of the Ranunculus auricomus complex. Plant Systematics and Evolution, 226, 165–185. Houston L, Robertson A, Rich TCG (2008) The distribution, population size and growth of the rare English endemic Sorbus bristoliensis A. J. Wilmott, Bristol Whitebeam (Rosaceae). Watsonia, 27, 37–49. Houston L, Robertson A, Jones K, Smith SCC, Hiscock SJ, Rich TCG (2009) An account of the Whitebeams (Sorbus L.,  2010 Blackwell Publishing Ltd Rosaceae) of Cheddar Gorge, England, with description of three new species. Watsonia, 27, 283–300. IUCN (2001) IUCN Red List Categories. Version 3.1, The World Conservation Union, Gland. Jankun A, Kovanda M (1988) Apomixis and origin of Sorbus bohemica (Embryological studies in Sorbus no. 2). Preslia, 59, 97–116. Koch MA, Dobes C, Mitchell-Olds T (2003) Multiple hybrid formation in natural populations: concerted evolution of the Internal Transcribed Spacer of nuclear ribosomal DNA (ITS) in North American Arabis divaricarpa (Brassicaceae). Molecular Biology and Evolution, 20, 338–350. Lemche EB (1999) The Origins and Interactions of British Sorbus Species, PhD thesis. Darwin College, Cambridge. Liljefors A (1953) Studies on propagation, embryology and pollination in Sorbus. Acta Horti Bergiani, 16, 277–329. Liljefors A (1955) Cytological studies in Sorbus. Acta Horti Bergiani, 17, 47–113. Lo EY, Stefanović S, Dickinson TA (2009) Population genetic structure of diploid sexual and polyploid apomictic hawthorns (Crataegus; Rosaceae) in the Pacific Northwest. Molecular Ecology, 18, 1145–1160. McAllister HA (1986) The Rowan and Its Relatives (Sorbus spp.). Ness Series 1. Ness Botanic Gardens, Liverpool. Mes T (1998) Character compatibility of molecular markers to distinguish asexual and sexual reproduction. Molecular Ecology, 7, 1719–1727. Nelson-Jones EB, Briggs D, Smith AG (2002) The origin of intermediate species of the genus Sorbus. Theoretical and Applied Genetics, 105, 953–963. Nybom H, Esselink GD, Werlemark G, Vosman B (2004) Microsatellite DNA marker inheritance indicates preferential pairing between two highly homologous genomes in polyploid and hemisexual dog-roses, Rosa L. Sect. Caninae DC. Heredity, 92, 139–150. Obbard DJ, Harris SA, Pannell JR (2007) Simple allelicphenotype diversity and differentiation statistics for allopolyploids. Heredity, 97, 296–303. Oddou-Muratorio S, Aligon C, Decroocq S, Plomion C, Lamant T, Mush-Demesure B (2001) Microsatellite primers for Sorbus torminalis and related species. Molecular Ecology Notes, 1, 297– 299. Pielou EC (1969) An Introduction to Mathematical Ecology. WileyInterscience, New York. Pigott CD, Walters SM (1954) On the interpretation of discontinuous distributions shown by certain British species of open habitat. Journal of Ecology, 42, 95–99. Pring ME (1961) Biological Flora of the British Isles no. 78. Arabis stricta Huds. Journal of Ecology, 49, 431–437. Proctor MCF, Groenhof AC (1992) Peroxidase isoenzyme and morphological variation in Sorbus L. in South Wales and adjacent areas, with particular reference to S. porrigentiformis E. F. Warb. Watsonia, 19, 21–37. Proctor MCF, Proctor ME, Groenhof AC (1989) Evidence from peroxidase polymorphism on the taxonomy and reproduction of some Sorbus populations in south-west England. New Phytologist, 112, 569–575. Raspé O, Kohn JR (2007) Population structure at the S-locus of Sorbus aucuparia L. (Rosaceae: Maloideae). Molecular Ecology, 16, 1315–1325. 1690 A . R O B E R T S O N E T A L . Rich TCG, Houston L (2006) Sorbus whiteana (Rosaceae), a new endemic tree from Britain. Watsonia, 26, 1–7. Rich TCG, Jermy AC (1998) Plant Crib 1998. BSBI, London. Rich TCG, Proctor MCF (2009) Some new British and Irish Sorbus (Rosaceae) taxa. Watsonia, 27, 207–216. Rich TCG, Harris SA, Hiscock SJ (2009) Five new Sorbus (Rosaceae) taxa from the Avon Gorge, England. Watsonia, 27, 217–228. Richards AJ (1975) Sorbus L. In: Hybridisation and the Flora of the British Isle (ed Stace CA), pp. 233–238. Academic Press, London. Richards AJ (2003) Apomixis in flowering plants: an overview. Philosophical Transactions of the Royal Society Series B, 358, 1085–1093. Rieseberg LH (1997) Hybrid origins of plant species. Annual Review of Ecology and Systematics, 28, 359–389. Rieseberg LH, Willis JH (2007) Plant speciation. Science, 317, 910–914. Robertson A, Sydes C (2006) Sorbus pseudomeinichii, a new endemic Sorbus (Rosaceae) microspecies from Arran, Scotland. Watsonia, 26, 9–14. Robertson A, Newton AC, Ennos RA (2004a) Multiple hybrid origins, genetic diversity and population genetic structure of two endemic Sorbus taxa on the Isle of Arran, Scotland. Molecular Ecology, 13, 123–143. Robertson A, Newton AC, Ennos RA (2004b) Breeding systems and continuing evolution in the endemic Sorbus taxa on Arran. Heredity, 93, 487–495. Rohlf FJ (2002) NtSYSpc, Numerical Taxonomy and Multivariate Analysis System. Version 2.11a, User guide. Exeter Software, New York. Salmon A, Ainouche ML, Wendel JF (2005) Genetic and epigenetic consequences of recent hybridisation and polyploidy in Spartina (Poaceae). Molecular Ecology, 14, 1163– 1175. Sepp B, Bobrova VK, Troitsky AK, Glazunova KP (2000) Genetic polymorphism detected with RAPD analysis and morphological variability in some microspecies of apomictic Alchemilla. Annales Botanici Fennici, 37, 105–123. Soltis DE, Soltis PS (1999) Polyploidy: recurrent formation and genome evolution. Trends in Ecology and Evolution, 14, 348–352. Stace CA (1997) New Flora of the British Isles, 2nd edn. Cambridge University Press, Cambridge. Stebbins GL (1950) Variation and Evolution in Plants. Columbia University Press, New York. Vrijenhoek RC (1984) Ecological differentiation among clones: the frozen niche variation model. In: Population Biology and Evolutio (eds Woermann K, Loeschcke V), pp. 217–231, Springer, Berlin. Warburg EF (1952) Sorbus L. In: Flora of the British Isle (eds Clapham AR, Tutin TG, Warburg EF), pp. 539–556, Cambridge University Press, Cambridge. Warburg EF (1962) Sorbus L. In: Flora of the British Isle (eds Clapham AR, Tutin TG, Warburg EF), 2nd edn. Cambridge University Press, Cambridge. Warburg EF, Kárpáti ZE (1968) Sorbus L. In: Flora Europaea, Vol. 2 (eds Tutin TG et al.), pp. 67–71, Cambridge University Press, Cambridge. Whitton J, Sears CJ, Baack EJ, Otto SP (2008) The dynamic nature of apomixis in the angiosperms. International Journal of Plant Sciences, 169, 169–182. Wilkinson M (2001) PICA 4.0: Software and Documentation, Department of Zoology, The Natural History Museum, London. Wilmott AJ (1934) Some interesting British Sorbus. Proceedings of the Linnean Society of London, 146, 73–79. Ashley Robertson is a population geneticist who uses molecular markers to study genetic diversity and speciation in plants. Timothy Rich is a plant taxonomist specializing in the taxonomy and evolution of critical plant groups in Britain. Alexandra Allen is a plant molecular geneticist working on plant mating systems, particularly self-incompatibility. Libby Houston is a plant ecologist and all-round naturalist with expert knowledge of the flora of the Avon Gorge. Cat Roberts is an undergraduate student with interests in ecology and genetics. Jon Bridle is a population geneticist and evolutionary biologist with an interest in how organisms adapt and evolve in response to ecological change. Stephen Harris is a population geneticist working on many aspects of plant systematics. Simon Hiscock is a plant geneticist with various interests in plant evolutionary genomics, mating system evolution, hybrid speciation, and polyploidy. Supporting Information Additional supporting information may be found in the online version of this article. Table S1–S17b Supporting information tables referenced in the text (Word document) Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.  2010 Blackwell Publishing Ltd