|Home | About | Journals | Submit | Contact Us | Français|
Rhododendron (Ericaceae) is a large woody genus in which hybridization is thought to play an important role in evolution and speciation, particularly in the Sino-Himalaya region where many interfertile species often occur sympatrically. Rhododendron agastum, a putative hybrid species, occurs in China, western Yunnan Province, in mixed populations with R. irroratum and R. delavayi.
Material of these taxa from two sites 400 km apart (ZhuJianYuan, ZJY and HuaDianBa, HDB) was examined using cpDNA and internal transcribed spacer (ITS) sequences, and amplified fragment length polymorphism (AFLP) loci, to test the possibility that R. agastum was in fact a hybrid between two of the other species. Chloroplast trnL-F and trnS-trnG sequences together distinguished R. irroratum, R. delavayi and some material of R. decorum, which is also considered a putative parent of R. agastum.
All 14 R. agastum plants from the HDB site had the delavayi cpDNA haplotype, whereas at the ZJY site 17 R. agastum plants had this haplotype and four had the R. irroratum haplotype. R. irroratum and R. delavayi are distinguished by five unequivocal point mutations in their ITS sequences; every R. agastum accession had an additive pattern (double peaks) at each of these sites. Data from AFLP loci were acquired for between ten and 21 plants of each taxon from each site, and were analysed using a Bayesian approach implemented by the program NewHybrids. The program confirmed the identity of all accessions of R. delavayi, and all R. irroratum except one, which was probably a backcross. All R. agastum from HDB and 19 of 21 from ZJY were classified as F1 hybrids; the other two could not be assigned a class.
Rhododendron agastum represents populations of hybrids between R. irroratum and R. delavayi, which comprise mostly or only F1s, at the two sites examined. The sites differ in that at HDB there was no detected variation in cpDNA type or hybrid class, whereas at ZJY there was variation in both.
The key process in speciation concerns the formation and maintenance of reproductive isolating barriers (Coyne, 1994; Levin, 2000; Wu, 2001). One of the keys to understanding the nature of species is to explain how interfertile species remain distinct in spite of forming populations of fertile hybrids. Species barriers could be formed and maintained via the elimination of intermediate forms or hybrids, which are adapted to the habitat of neither parent (Schluter, 1998; Levin, 2000). However, there is now ample evidence that hybrids may, under certain circumstances, have higher fitness than one or both parents (Arnold and Hodges, 1995; Arnold, 1997; Wang et al., 1997; Rieseberg and Carney, 1998; Campbell, 2004; Rhode and Cruzan, 2005). Hybrid fitness can be habitat-mediated (Johnson et al., 2001; Campbell et al., 2005) and may on occasion permit hybrids to occupy habitats distinct from either parent (Cruzan and Arnold, 1993; Arnold, 2004; Whitney et al., 2006). The presence of hybrids in intermediate habitats provides a potential conduit for germplasm to flow between interfertile species, raising the possibility that some species exist in spite of continuous gene flow between them at loci not subject to selection (Grant, 1981; Wu, 2001). Gene flow can also be prevented, however, if hybridization proceeds only to the F1 stage and no further, which can occur due to low F1 fertility or hybrid breakdown, or occasionally due to apparent habitat-mediated superiority of F1s over other hybrid classes (Milne et al., 2003).
Both these hypotheses (hybrid formation with or without ongoing gene flow) could apply to hybrid zones involving members of Rhododendron subgenus Hymenanthes from south-east Asia. Rhododendron × sochadzeae, which forms F1-dominated hybrid zones, belongs to this subgenus (Milne et al., 2003), so other examples might reasonably be expected within Hymenanthes where habitat circumstances are similar. On the other hand, the south-east Asian members of subgenus Hymenanthes appear to represent a rapidly radiated group that originated <5 Mya (Milne, 2004), and within which natural hybridization occurs commonly (Chamberlain, 1982). They thus might be a ‘Syngameon’, i.e. a complex of species maintained by selection, that are ecologically highly distinctive but capable of exchanging genetic material (Grant, 1957, 1963, 1981; Seehausen, 2004). To understand the mechanisms by which the remarkable diversity of Hymenanthes species in south-east Asia arose and are maintained, it is necessary to examine how species barriers are maintained in this group. However, very few of the natural hybrids that occur in this group have been examined in detail. Furthermore, some described species may in fact be hybrids themselves, particularly as large populations of F1 hybrids can look much like a stabilized species (Milne et al., 2003), and this may obscure the true pattern of species diversity in south-east Asian Hymenanthes.
Rhododendron agastum Balf. f. et W. W. Smith was originally described as a species (Balfour, 1917; Chamberlain, 1982), but more recently has been treated as a hybrid between R. delavayi Franch. (= R. arboreum ssp. delavayi) and R. decorum Franch. based on morphological (Chamberlain et al., 1996) and molecular (Zhang et al., 2007a) evidence. Hybrids between these species closely resemble R. agastum in general appearance (Zha et al., 2008), although they tend to follow R. decorum in having 6–9 corolla lobes and 11–16 stamens (Zhang et al., 2007a). This is in contrast to the type specimen of R. agastum and all other material of this species at the KIB herbarium, which instead resemble R. delavayi in having five corolla lobes and ten stamens (Chamberlain, 1982). Furthermore, some populations of R. agastum occur where R. decorum is rare or absent, but where R. delavayi and R. irroratum Franch. occur together, sometimes in large numbers (H. G. Zha, R. I. Milne, pers. observ.). R. irroratum has the same corolla lobe and stamen numbers as R. agastum, and hence at least some populations of R. agastum may in fact be R. delavayi × R. irroratum. Rhododendron delavayi, R. decorum and R. irroratum are members of subsections Arborea, Fortunea and Irrorata, respectively (Chamberlain, 1982; Chamberlain et al., 1996), so no two of these would appear to be sister species; however, phylogenetic analysis has yet to reveal their precise relationships to one another (R. I. Milne, unpubl. data). All these species have the chromosome number 2n = 26 (Zhang, 2007). In a previous paper (Zha et al., 2008), we examined hybrid populations between R. delavayi and R. decorum; the present study considers R. agastum populations where the putative parents are R. delavayi and R. irroratum, at sites where R. decorum is rare or absent.
Because R. agastum populations do not display the morphological variability normally associated with segregating post-F1 (after first generation) hybrids (H. G. Zha, pers. observ.), these populations might provide a second example of an F1-dominated hybrid zone within Rhododendron subgenus Hymenanthes. As with the parent species of Rhododendron × sochadzeae (Milne et al., 2003), R. delavayi and R. irroratum have distinct but overlapping altitude ranges, i.e. 1500–3000 and 2500–3500 m, respectively, and the known range of R. agastum is intermediate (2200–3350 m) (Chamberlain, 1982). R. delavayi is a remarkably widespread species with five subspecies extending from north-west India to Thailand and GuiZhou in south-west China. In contrast, R. irroratum has a far more limited range mostly within south-west China, occurring from western GuiZhou to northern Yunnan provinces, although it is common within this range (Chamberlain, 1982; Fig. 1). The distribution given for R. agastum in the Flora of China is almost identical to that of R. irroratum (Fang et al., 2005; Fig. 1), although it is much less abundant and appears to only occur sporadically, in the company of R. delavayi and one or both of R. irroratum and R. decorum (T. L. Ming, Kunming Institute of Botany, China, pers. comm.). The distribution of R. decorum is also similar (Chamberlain, 1982; Fig. 1). Ecologically, R. irroratum and R. delavayi occupy similar habitats to one another and often occur mixed together where their altitudinal ranges overlap, whereas R. decorum tends to prefer more shady and humid conditions, at least in Yunnan Province, and rarely tends to mix with the other two species.
In the current study, therefore, two sites where R. agastum occurs together with R. delavayi and R. irroratum were examined in order to determine: (1) whether R. agastum is the hybrid of R. delavayi and R. irroratum at these sites; (2) whether R. agastum populations contain mostly or only F1 hybrids; (3) whether there is variation in which species serves as maternal parent; and (4) whether there is a difference in population structure between the two sites. Of the sites examined, ZhuJiangYuan (north-east Yunnan) was subject to considerable anthropogenic disturbance whereas HuaDianBa (western Yunnan) was almost undisturbed by human activity. Both sites were also surveyed to determine whether R. decorum was present, and where it was, material of this species was also collected for comparison.
To answer the above questions, cpDNA (trnL-F and trnS-trnG) and internal transcribed spacer (ITS) sequences were obtained from the three taxa, and the genotype class of a sample of plants of the three taxa at each site was determined by means of a Bayesian analysis (Anderson and Thompson, 2002), conducted on variation revealed by amplified fragment length polymorphism (AFLP) markers. In addition, the fertility of R. agastum was determined via in vitro germination of naturally set seed gathered from wild material.
The two Rhododendron agastum localities chosen for examination were HuaDianBa (HDB), near Dali, west Yunnan (25°52′N, 99°58′E, approx. 2900 m) and ZhuJianYuan (ZJY), north-east Yunnan (25°54′N, 103°55′E, approx. 2300 m). All collections were made in April, 2006.
Because habitat disturbance has long been known to promote hybridization (Anderson, 1948, 1949; Anderson and Stebbins, 1954; Rieseberg and Carney, 1998), and in certain cases to have a profound effect on which hybrid classes are present (Kyhos et al., 1981; Milne et al., 2003), the population structure of R. agastum might vary according to the level of habitat disturbance, so the two sites chosen for this study had very different levels of habitat disturbance. At HDB, the natural vegetation type is well preserved, with natural forest more or less intact, whereas ZJY has been subject to a long period of habitat disturbance and deforestation, with tree cover now completely removed from most of the area, although the Rhododendrons themselves have been left intact to attract tourists. At HDB, R. agastum was present in a small area on a forested hillside slope, whereas at ZJY it occurred scattered around the top of a large hill.
At HDB, R. agastum could only be found within a small area (about 400 m2) mixed with R. delavayi and R. irroratum. Morphological characters such as corolla colour, ventral leaf surface indumentums and young shoot indumentums were used to distinguish these three taxa in the field. From this site we collected all 14 individuals of R. agastum that were found and examined, plus nine of R. delavayi and ten of R. irroratum, sampled at random. No other Hymenanthes species could be found within several hundred metres of this location, although 20 Rhododendron species including ten Hymenanthes have been recorded from the area. The only plants found of R. decorum were two individuals 1000 m away from the R. agastum population; these were both sampled to determine whether they were involved in the parentage of R. agastum.
At ZJY, the dominant plant species were Castanopsis orthacantha and Pinus yunnanensis, and the Rhododendron species R. delavayi and R. irroratum. Also present were R. aberconwayi (subgenus Hymenanthes), R. racemosum and R. simsii (other subgenera), and R. agastum, which occurred sporadically in the company of R. delavayi and R. irroratum. Despite repeated searches, R. decorum could not be found and appeared therefore to be absent at ZJY. It is possible that this species occurred here in the past, but disappeared relatively recently due the progressive deforestation of this site over recent centuries, which removed the shaded and humid habitats that this species prefers. From this, the possibility had to be considered that a now extinct population or R. decorum contributed to the parentage of hybrids at this site. From this site, 21 individuals of R. agastum, 14 of R. irroratum and ten of R. delavayi were sampled at random.
From each accession, desiccated leaf material (approx. 1 g of fresh leaf mass to approx. 25 g of coarse silica gel) for DNA extraction was collected, and voucher specimens were deposited in the Herbarium at Kunming Institute of Botany, Chinese Academy of Sciences. DNA was extracted from silica-gel-dried leaf tissue using a modified CTAB method (Kobayashi et al., 1998). DNA quality and concentration were assessed by agarose gel electrophoresis with known concentrations of uncut lambda DNA (Takara, Dalian, China).
The ITS region including the intervening 5·8S coding region of the nuclear ribosomal DNA (approx. 700 bp) of all the sampled individuals was amplified using primers ITS1 and ITS4 (White et al., 1990). The chloroplast trnL-F and trnS-G regions were amplified using universal primer pairs (‘C’ and ‘F’ primers from Taberlet et al., 1991, and trnS and trnG from Hamilton, 1999); the former was used in a previous study on natural hybridization in R. delavayi and R. decorum (Zha et al., 2008). Each of the three reactions was carried out in a final volume of 50 µL containing 20 ng template DNA, 10 mm Tris-HCl, pH 8·3, 50 mm KCl, 2 mm MgCl2, 200 µm of each dNTP, 400 pmol of each primer and 1 U of Ex-taq (Takara). All amplifications were performed using a PTC-100 thermocycler (MJ Research, Watertown, MA, USA) with the following conditions: 4 min at 94 °C (one cycle); 1 min at 94 °C, 1 min at 55 °C (53 °C for trnS-G amplification) and 1 min at 72 °C (33 cycles); and 10 min at 72 °C (one cycle).
PCR products were purified using an agarose gel DNA purification kit (Takara) following the manufacturer's instructions. The complete ITS with 5·8S region resolved on agarose as a single sharp band and was sequenced directly from all sampled R. delavayi, R. irroratum and R. agastum individuals. Direct sequencing was performed on both strands on an ABI PRISM 3730 Sequencer using the same primers that were used for the PCR amplifications. The alignment was performed using the program ClustalX v.1·83 (Thompson et al., 1997). If R. agastum were determined to be hybrids of R. delavayi and R. irroratum by rDNA genotyping and AFLP analysis, a χ2 test was conducted to determine if the ratio of R. delavayi to R. irroratum cpDNA (trnL-F or trnS-G sequences) among the hybrids differed significantly from a 1 : 1 ratio.
AFLP markers provide reliable diagnostic loci at varying taxonomical levels, and can be relatively easily generated in sufficient numbers to distinguish between genealogical classes in hybrid populations, especially in the case of weakly differentiated source populations (Miller, 2000; Campbell et al., 2003). Although co-dominant markers are twice as informative per locus as dominant markers, they are much more costly to produce, and it is possible to distinguish between F1, F2 and backcross hybrids with <5 % classification error using fewer than 30 diagnostic AFLP loci (Miller, 2000).
AFLP analysis was performed essentially as described by Vos et al. (1995), with modifications by Gilbert et al. (2002) and Zha et al. (2008). Primers and adapters were synthesized by Sangon Company (Shanghai, China). Enzymes were obtained from Amersham Pharmacia Biotech, unless otherwise stated. Genomic DNA (50 ng) was digested using both EcoRI and MseI enzymes (5 U each in a final volume of 30 µL) and adapters (0·1 µm E-adapter and 1·0 µm M-adapter) were ligated to the resulting fragments. Then, 5 µL of digested DNA from a 1 : 10 dilution with sterile distilled water (SDW) was used for (PCR) preamplification using primers (0·5 µm EcoRI and 0·5 µm MseI primers), complementary to the E- and M-adapters, carrying one selective nucleotide at the 3′-end. In total, 30 cycles were performed at 94 °C for 30 s, 56 °C for 30 s and 72 °C for 60 s in a PTC-100 thermocycler (MJ Research), after an initial cycle of 65 °C for 5 min. The preamplification products were diluted 1 : 10 with SDW and used as template for selective amplification using 0·6 µm EcoRI and 0·1 µm MseI primers, with three selective nucleotides at the 3′-end, with the following thermal cycling conditions: 94 °C for 2 min, followed by one cycle of 94 °C for 30 s, 65 °C for 30 s and 72 °C for 60 s, followed by 12 cycles which were identical except that the annealing temperature was reduced each cycle by 0·7 °C, followed by 23 additional cycles of 94 °C for 30 s, 56 °C for 30 s and 72 °C for 60 s; there was then a final stage of 72 °C for 5 min. In total, seven primer combinations were used: E-AAC/M-ACA, E-AGA/M-CCA, E-AGC/M-ACC, E-AGA/M-AAG, E-AAC/M-CTG, E-AGA/M-CGT and E-ACT/M-ACA. The amplified products were mixed with an equal volume of AFLP loading buffer (98 % formamide, 10 mm EDTA, 0·01 % xylene cyanol and 0·01 % Bromophenol blue) and 5 µL of each sample was electrophoresed on a 6 % denaturing polyacrylamide gel in 1× TBE buffer at 65 W for approximately 2 h. AFLP bands were visualized by silver staining of the gel, as described in Bassam et al. (1991).
AFLP bands were scored manually as 0 for the absence and 1 for the presence of a band. Co-migrating bands within a gel between different individuals were considered to be homologous. Only the polymorphic bands were used in subsequent analyses as the inclusion of monomorphic bands made no difference to the overall relationship between individuals. For each individual, the posterior probability that it belonged to parents or to early generation hybrid classes (F1, F2 or backcross) was estimated using a Bayesian method to analyse the polymorphic AFLP markers. This procedure used a Markov Chain Monte Carlo method and was implemented in the program ‘NewHybrids’ (v1·1 beta3), which allows distinguishing between F1 hybrids, different backcrosses or later generation hybrids and does not require that parental populations are sampled separately (Anderson and Thompson, 2002). The program was used with the input files described by Milne and Abbott (2008), which cause the program to consider 45 possible genotype classes, representing every class that could arise after up to four generations of hybridization. This method has the advantage that it avoids the assumption that only early-generation hybrids are present, and hence greatly reduces the possibility that complex hybrids will be wrongly assigned to a class such as F1. Posterior distributions were evaluated after 105 iterations of the Monte Carlo Markov Chains, following a burn-in of 104 iterations, without using any individual or allele frequency prior information. As in Milne and Abbott (2008), posterior probabilities for the 45 possible classes were pooled into six categories, i.e. parents, F1-type (i.e. F1 or cross between one species and advanced backcross to the other), F2-type (i.e. F2 or complex hybrid not approaching either parent), and backcross-type each way (i.e hybrid derivative with at least 75 % germplasm from one parent, which might be a simple or complex backcross). Individuals were assigned to one of the six genotypic categories if the pooled probability for that category was ≥0·95, or where this was not the case, individuals were assigned to a choice of two or more categories if the combined score for these two categories was ≥0·95.
The presence of at least one copy of each parent-specific marker from both parents in each hybrid genet could indicate that all the individuals are heterozygous for all parent-species-specific markers, and therefore must be F1s. The only other possibility is that each genet by chance has two copies of every such marker surveyed. By calculating that the probability for this latter possibility was extremely small, Milne et al. (2003) demonstrated that all examined individuals of Rhododendron × sochadzeae were F1s. Here, the proportion of such markers expressed in hybrids is very high, but not 100 % (see Table S1 in Supplementary Data, available online), and so a similar approach was followed here to test alternative hypotheses for each accession against the hypothesis that it was an F1, in order to corroborate the results from the NewHybrids analysis.
As with the NewHybrids analysis, the two populations of R. agastum were analysed separately from one another. Direct mathematical analysis was used to provide tests of whether each R. agastum accession might be (1) an F2, or other post-F1 hybrid with 50 % germplasm from each species, or (2) a first-generation backcross to one or other parent species. For this analysis only markers that were exclusive to one or other parent species were used (i.e. markers that were present in all accessions of one parent species, but none of the other). Let Nd and Ni be the number of markers exclusive to R. delavayi and R. irroratum, respectively, and d and i the number of markers specific to each species present in each hybrid accession. For this analysis each marker is assumed to be fixed and single copy within each parent species (but see below).
For any F2 accession, the probability that each marker will be present is 0·75. Hence the probability that it will have d and i markers present is given by
To estimate the probability that an accession is an F2, when the alternative hypothesis is that it is an F1, the probability was calculated, for each accession, that an F2 could have at least d and i markers present. For R. irroratum markers this was done by summing the probabilities for all values of i between the actual value for each accession (here called ai) and the highest possible value, i = ni. Hence the probability that the accession was an F2, based on R. irroratum-exclusive markers alone, was estimated by
Likewise for R. delavayi markers values of d between ad and nd were summed using
giving an estimate of the probability of being an F2 based on R. delavayi-exclusive markers only. The probability that each accession was an F2 was hence estimated using the product of these two probabilities:
A similar approach was used to test whether an accession might be a backcross. A backcross should have all markers present from one parent species, but from the other, the probability that any marker is present will be 0·5g where g is the generation of backcrossing. Hence the probability that a backcross to R. delavayi contains xi markers specific to R. irroratum is given by
and where xi > (1 − xi) (as is the case for all putative hybrid accessions here), the probability drops with increasing generations. For this reason, if an accession can be shown by this method to not be a first-generation backcross, then it follows that it is even less likely to be a backcross of a later generation. For first-generation backcrosses, the formula for g = 1 simplifies to
for backcrosses to R. delavayi and R. irroratum, respectively. As with testing for F2s, the possibility of being a backcross in each generation was tested against the alternative possibility of being an F1. Hence for each accession, the probability of it being a backcross was estimated by summing the probabilities for all values of xi between the actual value for each accession (ai) and the highest possible value, xi = ni. Hence the formula is
for backcrosses to R. delavayi, and
for backcrosses to R. irroratum.
In both these cases, the assumption is made that a marker is fixed in the parent species that it comes from, but it is possible that in some cases the marker-absent allele is present at low frequencies, undetected, in the parent population. If this were so, probabilities for being F2s and backcrosses would be over-estimated by the method described. Equally, however, some markers could be multiple-copy, which would cause these probabilities to be under-estimated. Hence the results from these calculations must be interpreted with caution, and treated as informative rather than as absolute probability values.
Complex hybrid derivatives with 50 % germplasm from each parent would have the same probability scores as F2s using this method, whereas in cases where the proportion of germplasm was unequal, the score would grade towards that for a backcross. The assumption is made that where the combined probabilities of being F2 or a backcross either way does not exceed 5 % then the accession in question is also unlikely to be a complex hybrid derivative.
At both of the sampling sites, Rhododendron individuals in the hybrid swarms could easily be classified into three types, i.e. R. delavayi-like, R. irroratum-like and R. agastum-like, using three morphological characters: corolla colour, ventral leaf surface indumentum and young shoot indumentum. The thin ventral leaf indumentum of R. agastum, which was clearly intermediate between those of R. delavayi and R. irroratum, has been taken as a diagnostic character for R. agastum (Chamberlain, 1982). At both populations examined, each morphological character for each R. agastum-like individual either matched one of R. delavayi or R. irroratum, or was intermediate between these two species (Table 1). This was thus consistent with R. agastum being the hybrid of these two species. By contrast, none of the R. agastum plants examined had the higher numbers of corolla lobes and stamens, nor the leaf shape, characteristic of R. decorum. This, plus the rarity and apparent absence of R. decorum at HDB and ZJY, respectively, made it unlikely that this species was involved in the parentage of R. agastum at these sites.
Within each of the R. delavayi-like, R. irroratum-like and R. agastum-like groups, very little morphological variation was observed. This is consistent either with all three being species, or with R. agastum comprising only hybrids of F1 class. It indicates that if backcrosses or segregating hybrid derivatives are present, they are probably so at low frequency.
The aligned sequence matrix generated a total of 642 characters with no indels or gaps. This comprised the ITS1 (253 bp), 5·8S (164 bp) and ITS2 (225 bp) regions. All individuals in R. delavayi-like groups from ZJY and HDB had identical ITS sequences (GenBank accession numbers EF035043 and DQ295783; ITS-type DD; Table 2) and were polymorphic (C/T) at positions 91 and 203. All R. irroratum individuals from HBD had identical ITS profiles (GenBank accession number EF035044; ITS-type HH; Table 2); at ZJY all R. irroratum accessions except IRR-Z-7 had identical profiles to one another (GenBank accession number EF035045; ITS-type ZZ; Table 2) but differed from those from HDB in that position 211 was polymorphic (T/G) at ZJY but monomorphic (G) at HDB. However, the two accessions of R. decorum examined from HBD had identical ITS profiles to those of R. irroratum at this site (Table 2).
The ITS sequences of R. delavayi and R. irroratum differ at eight positions, six in ITS1 (91, 98, 111, 200, 203, 211) and two in ITS2 (491, 502) (Table 2). Position 91 was polymorphic (C/T) in R. delavayi, whereas it was monomorphic (T) in R. irroratum and R. decorum (Table 2). Hence R. delavayi and the ZJY population of R. irroratum both contain polymorphic loci indicating that at least two slightly differing copies of ITS are present.
Additive ITS profiles were present in all R. agastum accessions examined, and also in one accession (IRR-Z-7) with R. irroratum morphology. ITS profiles of R. agastum at HDB displayed perfect additivity between types DD (R. delavayi) and HH (R. irroratum and R. decorum at HDB) (ITS type DH; Table 2). Likewise, R. agastum at ZJY and accession IRR-Z-7 had additive profiles of types DD (R. delavayi) and ZZ (R. irroratum at ZJY) (ITS type DH; Table 2). As with the profiles of R. irroratum, the profiles of R. agastum at HDB and ZJY therefore differed only at position 211 (Table 2). This additivity strongly supports R. agastum being the hybrid of R. delavayi and R. irroratum, although the ITS data alone cannot reject the possibility that it is R. delavayi × R. decorum.
PCR reactions with the C/F and trnS/trnG primer combination each yielded a single PCR product of the same size for all sampled individuals. Direct sequencing of these PCR products yielded a single haplotype for each amplicon.
There was no variation within R. delavayi-like accessions from HDB or ZJY; all had the same trnS-G haplotypes (GenBank accession number DQ988999), and the same trnL-F sequence as reported for this species in a previous study (Zha et al., 2008, GenBank accession number DQ178247) (haplotype D, Table 3). Regarding R. irroratum, there was no variation within either site, but material from HDB (GenBank accession numbers DQ999963 and DQ989004; haplotype Ih; Table 3) differed from that at ZJY (Genbank accession numbers DQ999964? and DQ989005 haplotype Ih; Table 3) by one trnL-F insertion and one point mutation, and by three trnS-G point mutations. This indicates that material of R. irroratum is quite distinct between these two sites, and/or that one population has captured the cpDNA of another species. cpDNA of R. decorum differed from HDB material of R. irroratum by five mutations, and from ZJY material of R. irroratum by two mutations (Table 3). Both species differed from R. delavayi at 12 or more positions.
At HDB, all R. agastum-like accessions had haplotype D, matching R. delavayi. However, at ZJY, four R. agastum-like accessions had the haplotype Iz, matching local R. irroratum populations, while the other 17 accessions had haplotype D. This strongly indicates that R. agastum at ZJY is a hybrid of R. delavayi and R. irroratum and that either species can be the plastid donor, whereas at HDB R. delavayi is always the plastid donor. Chloroplast DNA is usually inherited maternally in Rhododendron species (Harris and Ingram, 1991), indicating that R. delavayi may always be the maternal parent at HDB, while at ZJY the maternal parent may be either species. The strongly biased ratio of 17 : 4 at ZJY was statistically different from the 1 : 1 expectation for no gender bias (null hypothesis) (χ2 = 8·04, d.f. = 1, P < 0·01), indicating that even here, R. delavayi is more commonly the maternal parent. This significant deviation remains if only individuals shown to be F1s (see below) are taken into account (χ2 = 6·36, d.f. = 1, P < 0·05). The results at ZJY eliminate R. decorum as a possible parent for those accessions with haplotype Iz and, based on the close similarity between these four accessions and the others examined for morphology, ITS and AFLP characters, make it unlikely that R. decorum is a parent to R. agastum at either site.
A total of 119 scorable polymorphic AFLP markers were generated during the analysis. Additional polymorphic markers were present but could not be scored either because of faint, inconsistent amplification or the inability to differentiate two or more fragments of a similar molecular mass.
AFLP analyses were conducted separately and independently for two locations, HDB and JZY. At HDB, AFLP profiles of R. delavayi, R. decorum and R. irroratum were first compared with those of R. agastum to determine whether the profiles of R. agastum accessions were additive for any two of the other species. At HDB, all markers present in R. agastum were also present in either R. delavayi or R. irroratum, including 20 markers that were present only in R. delavayi and R. agastum, and ten that were present only in R. irroratum and R. agastum. Conversely, there were no markers present only in R. decorum and R. agastum. Similarly, at site ZJY, R. agastum contained 20 and 12 markers that were otherwise only present in R. delavayi and R. irroratum, respectively, and none that was otherwise only present in R. decorum (based on material of R. decorum from HDB). At both HDB and ZJY, R. agastum does not contain any markers that are not found in either R. delavayi or R. irroratum. Based on this evidence, R. agastum is a hybrid derivative of R. delavayi and R. irroratum, whereas R. decorum is not or only marginally involved. Therefore, R. decorum, along with markers exclusive to it, was excluded from the following analysis.
The NewHybrids analysis (Fig. 2) indicated that all accessions of R. agastum at HDB, and most accessions of it at ZJY, were F1 hybrids between R. delavayi and R. irroratum. At HDB, all putative parents examined were confirmed to be the pure parent species with >97 % probability, and all hybrids were determined to be F1s with >99 % probability. At ZJY, however, one individual identified as R. irroratum was determined to be a hybrid derivative, although the program could not determine whether it was a backcross to R. irroratum or a complex hybrid. Among 21 putative hybrids examined at ZJY, 19 were determined to be F1s but the remaining two could not be assigned a class with 95 % certainty. Of these two, one had a 91 % probability of being an F1, whereas the other had a 26 % possibility of being an F1 but a 47 % probability of being a backcross to R. delavayi. Therefore, there was no evidence for hybridization proceeding beyond the F1 level at HDB, but at least one, probably two and possibly three accessions examined at ZJY were not F1s, so hybridization does proceed beyond the F1 at this location.
At HDB, there were 20 markers exclusive to R. delavayi and ten markers exclusive to R. irroratum. Accessions of R. agastum expressed between 17 and 20 of the former, and either nine or ten of the latter. Calculated probabilities for any of these accessions being F2s were low, ranging from <0·001 in five accessions to 0·013 in aga-H-12 (Table S1 in Supplementary Data, available online). Probabilities for being backcrosses to R. delavayi were between 0·001 and 0·011, whereas probabilities for being backcrosses to R. irroratum were <0·001 in all accessions except aga-H-10, in which it was 0·001 (Table S1). Based on this, F2s and backcrosses are not present among the sampled material, and hence all R. agastum accessions from this site are F1s (Table S1 in Supplementary Data).
At ZJY, there were 19 markers exclusive to R. delavayi and 11 markers exclusive to R. irroratum. Accessions of R. agastum expressed between 16 and 19 of the former, and between seven and 11 of the latter; however, accession irro-Z-7 expressed only 14 markers specific to R. delavayi. For accession irro-Z-7 the probability for being a backcross to R. irroratum was 0·032, whereas in all R. agastum accessions the probability of being such a backcross was 0·002 or <0·001 (Table S1 in Supplementary Data). The probabilities recorded for being backcrosses to R. delavayi, however, range from <0·001 to 0·274, with four accessions given a probability of >0·05 for being this class (Table S1 in Supplementary Data).
Accession irro-Z-7 was either an F2 or backcross to R. delavayi according to NewHybrids, but the mathematical probabilities calculated for each of these classes were low (0·028 and 0·032, respectively), although these probability values were still higher than those for most other hybrid accessions. Hence this accession could be an F1, F2 or backcross to R. irroratum based on molecular data. This discrepancy between the two analyses suggests that the direct mathematical analysis might under-estimate the probabilities for these classes, and/or that NewHybrids might over-estimate them, relative to the probability for being F1s. As NewHybrids involves fewer untested assumptions, we favour the former possibility; however, the approach here will be to assign F1 class where both NewHybrids and direct mathematical analysis reject all other possibilities, but otherwise to consider any class that is not rejected by both methods as possible for each accession.
Among the 21 R. agastum individuals from ZJY, 15 are indicated to be F1s by both analyses (Table S2 in Supplementary Data). Of the remaining six, four are F1s according to NewHybrids, but of these, two might be F1s and two might be backcrosses to R. delavayi based on direct mathematical analysis, although the probabilities for these in each case are low (0·052, 0·12, 0·113 and 0·113, respectively; Table S1 in Supplementary Data). Of the remaining three accessions, aga-Z-10 is given a relatively high probability (0·274) of being a backcross to R. delavayi, although NewHybrids gave this accession a 0·75 probability of being an F1, so it could belong to either class. Accession aga-Z-5 could be an F1, F2 or backcross to R. delavayi according to NewHybrids, whereas according to the mathematical analysis it could be an F1 or F2 but is unlikely to be a backcross (Table S1 in Supplementary Data).
Rhododendron agastum was described as a species in 1917 (Balfour, 1917) and was still considered as such in 1982 (Chamberlain, 1982) but by 1996 it was listed as a probable hybrid of R. delavayi (Chamberlain et al., 1996), with the second parent suggested to be R. decorum. Indeed, hybrids of R. delavayi and R. decorum do closely resemble R. agastum in some morphological characters (e.g. Zha et al., 2008). However, the populations of R. agastum examined for the current study lack some of the distinctive morphological characters of R. decorum, and instead appeared based on morphology to be hybrids between R. delavayi and R. irroratum. Furthermore, both R. agastum populations examined occurred mixed together with R. delavayi and R. irroratum, and at one of the sites examined R. decorum was rare and found no closer than 1 km to the study population, whereas at the other site R. decorum was absent altogether.
Data from the current study indicated clearly that the R. agastum populations examined are hybrids, with R. delavayi as one parent. Morphological data (especially corolla lobe and stamen number) indicated that the second parent was R. irroratum. Caution is normally required in interpreting morphological data in hybrids (Rieseberg and Ellstrand, 1993). However, in this case artificial hybrids between R. decorum and R. delavayi had the same number of stamens (11–16) and corolla lobes (6–9) as R. decorum (Zhang et al. 2007a), whereas the numbers of these were ten and five, respectively, in the current study. Based on this, parentage from R. decorum seems unlikely, although variation in the dominance of these characters within species cannot be ruled out.
The ITS sequences of R. irroratum and R. decorum are very similar, so these data confirmed only that the R. agastum populations examined are hybrids of one of these two and R. delavayi. However, AFLP data from each site demonstrated that the profiles of R. agastum were additive for R. delavayi plus R. irroratum, but not for R. delavayi plus R. decorum, strongly indicating that R. irroratum, not R. decorum, was the second parent. The only caveat to this was the very small sample size for R. decorum. A third line of evidence was cpDNA: although most R. agastum plants had cpDNA of R. delavayi, four at ZJY had cpDNA of the second parent. As R. irroratum and R. decorum cpDNA data differ at two sites in the trnG-S region it was possible to determine that for the four accessions mentioned, R. irroratum was the maternal parent. Hence parentage from R. irroratum was proved, but only for a minority of plants; however, these four did not differ from other R. agastum in morphology so there was no reason to assume they were unusual in their parentage. Taking morphological, AFLP and cpDNA data together, and considering also the rarity/absence of R. decorum at HDB and ZJY, respectively, it can be concluded with some confidence that R. agastum is R. delavayi × R. irroratum at these two sites.
Plants of R. agastum determined to be R. irroratum × R. delavayi are similar in morphology to those of R. delavayi × R. decorum examined elsewhere, particularly in the ventral leaf surface indumentum, ovary and calyx surface, and corolla colour, which are diagnostic for ‘R. agastum’ (Chamberlain et al., 1996; Zhang et al., 2007a; Zha et al., 2008). Hence although our study indicates that some plants known as R. agastum are R. irroratum × R. delavayi, it is very likely that some of the other populations referred to this species may be R. delavayi × R. decorum, and that R. × agastum, as it should more properly be known, in fact currently represents two distinct types of hybrid that are very similar in appearance. Based on the morphology of the hybrids in this study (Table 1) and those generated by Zhang et al. (2007a), hybrids of the two combinations can be distinguished by characters such as corolla lobe and stamen number, although further work might be necessary to confirm that these characters are always diagnostic for the hybrids.
The type specimen of R. agastum has five corolla lobes and ten stamens, and based on this is very probably R. delavayi × R. irroratum. Hence the other hybrid combination, R. delavayi × R. decorum, requires a new name.
Based on AFLP markers, a Bayesian analysis conducted using the program NewHybrids indicated that all 14 R. × agastum hybrids at HDB were F1s, and as all putative parents were also confirmed to be such, no other hybrid classes were present among the material examined. Direct mathematical analysis of AFLP data gave similar results, indicating that all hybrids at HDB were F1s.
Of 21 putative hybrids at ZJY, however, only 19 were indicated to be F1s by the NewHybrids analysis. The other two were shown to be hybrids, and of these one (aga-Z-10) was more likely an F1 than another class (76 %) while the other (aga-Z-5) was likely to be a complex hybrid, with backcross to R. delavayi the most likely possibility. Direct mathematical analysis confirmed that 15 of the 21 hybrids at this site were F1s, but returned probabilities of 0·052 and 0·12 that two accessions were F2s, probabilities of 0·113 that each of three accessions including aga-Z-5 might be backcrosses to R. delavayi, and a probability of 0·274 that the remaining accession, aga-Z-10, was a backcross to R. delavayi. In addition to these, one putative R. irroratum accession, irro-Z-7, was determined by NewHybrids to be a hybrid of second or later generation, most likely a backcross to this species. That this accession was not pure R. irroratum but a hybrid derivative was also indicated by the ITS data. Direct mathematical analysis did not provide strong support for this accession being a class other than F1; however, this accession was very different in morphology from all others identified as F1s in this study, and so based on this plus the NewHybrids result, the hypothesis that it is an F1 can be rejected. In another Rhododendron hybrid zone where F1s were common but not the only class present, F2s were rare to absent whereas backcrosses comprised up to one-fifth of hybrids present (Milne and Abbott, 2008). Based on this, plus its morphological similarity to R. irroratum, accession irro-Z-7 is most likely to be a backcross, but the hypothesis that it is an F2 cannot be rejected.
Taking the two analyses together, ZJY contained 15 hybrid derivatives that were almost certainly F1s, four more that were probable F1s, and three of uncertain class, of which at least one (irro-z-7) was certainly not an F1 (Table S2 in Supplementary Data). Hence at ZJY backcrossing to both parent species (and hence introgression) might occur but could not be confirmed.
There were no two individuals of R. × agastum detected with identical AFLP profiles, which indicates minimal clonal reproduction in these hybrid populations, and that every hybrid examined was a distinct genet. This is in contrast to Rhododendron × sochadzeae and R. × intermedium, in which a small number of cloned genets were detected (Milne et al., 2003; Milne and Abbott, 2008), and Rhododendron ferrugineum in which clonal reproduction plays a significant role (Escaravage et al., 1998). Hence among hybrid derivatives at HDB, 14 of 14 genets were F1s, whereas between 19 and 21 of 22 detected hybrid genets at ZJY were F1s (Table S2 in Supplementary Data).
F1 hybrids between R. delavayi and R. irroratum are known to be fertile based on field observation and seed germination tests (H. G. Zha, unpubl. res.), and the presence of non-F1 hybrid derivatives at ZJY. Hence populations of R. × agastum at HDB are F1-dominated hybrid zones, where selection appears to remove hybrid derivatives of subsequent generations before they reach adulthood (Kyhos et al., 1981; Milne et al., 2003). At ZJY, however, the situation is more complex, with a minority of second- or later-generation hybrid derivatives reaching adulthood, a situation also observed in Rhododendron × intermedium (Milne and Abbott, 2008) and Phyllodoce (Kameyama et al., 2008). The difference in population structure between ZJY and HDB indicates that there might be some factor affecting the relative fitness of F1 and other hybrid derivatives between the two sites, of which the most likely is habitat-mediated selection.
Asymmetrical hybridization in plants is relatively common (Barton and Hewitt, 1985; Brubaker et al., 1993; Cruzan and Arnold, 1994; Arnold, 1997; Caraway et al., 2001; Wu and Campbell, 2005), and various causes for this have been suggested. Unilateral incompatibility (Harrison and Darby, 1955; Lewis and Crowe, 1958; Gore et al., 1990; Harder et al., 1993) can be ruled out as a cause because occasional hybrids with R. irroratum as maternal parent do occur at ZJY. Of the parents of R. × agastum, R. delavayi is self-incompatible and R. irroratum is self-compatible (Zhang et al., 2007b), so the greater tendency of the former species to outcross could lead to it producing more hybrid seed; however, it might be more common for the self-compatible species in such situations to be the maternal parent (M. Arnold, University of Georgia, USA, pers. comm.). Pollen of short-styled species tends to be less successful in forming complete pollen tubes in long-styled species, relative to the other way around (Briggs, 1964; Williams and Rouse, 1988); however, based on our specimens no significant difference was found between flower size and style length between the mature flowers of the parent species (|t| < t0·2; H. G. Zha, unpubl. res.).
A probable cause for asymmetry in this case is phenology. Rhododendrons are protandrous, and differences in flowering time could lead to an earlier flowering species tending to receive mainly heterospecific pollen onto stigmas towards the end of its flowering period, so favouring it as maternal parent for hybrids. At HDB, R. delavayi flowers about 1 month earlier than R. irroratum (H. G. Zha, pers. observ.), so phenology would make this species the normal maternal parent, as was observed in all R. agastum from this site. However, at ZJY the phenology of the species is different, and they have a longer overlap of flowering time, which could account for the less pronounced asymmetry in parentage at this site.
A third possibility is pollinator behaviour. Some pollinators such as Apis cerana cerana are common to both species (Zhang et al., 2007b), but R. delavayi has bright red flowers, whereas R. irroratum has pale flowers. Hence the possibility exists that pollinators commonly switch from R. irroratum to the more strikingly coloured R. delavayi, but seldom do the reverse; this would account for the asymmetric hybridization seen. The more open habitat at ZJY could affect pollinator behaviour (e.g. perhaps they can see further) and/or which pollinator species are present, either of which could possibly lead to bidirectional crossing at one site but not the other.
The R. agastum populations at HDB and ZJY differed in two respects: first, the HDB population comprised only F1s, whereas other classes were present in small numbers at ZJY; and second, crossing was bidirectional at ZJY, whereas only R. delavayi was maternal parent at HDB. The major difference between the two sites was anthropogenic disturbance at ZJY, at which the forest canopy had been completely removed, but not at HDB, where it was intact. Such disturbance has been shown to extend and alter flowering periods in other species (Lamont et al., 2003) and so could account for the increased overlap at ZJY and hence for the occasional hybrids for which R. irroratum is maternal parent. Habitat disturbance has long been known to promote hybridization (Anderson, 1948, 1949; Anderson and Stebbins, 1954; Semple and Semple, 1977; Rieseberg and Carney, 1998; Bleeker and Hurka, 2001; Lamont et al., 2003) and in some cases can profoundly affect which hybrid classes are present (Kyhos et al., 1981; Milne et al., 2003), possibly because disturbance can favour segregating hybrid derivatives (Abbott, 1992; Arnold, 1997; Rieseberg and Carney, 1998). This provides a possible explanation for why hybrid generations after the F1 are present at ZJY but not at HDB. Therefore, in natural conditions where the habitat is undisturbed, as at HDB, hybridization produces only F1s with R. delavayi mothers, which cannot bring about gene flow between the parent species, and might even reinforce species barriers if F1s out-compete other hybrid generations through superior fitness (Milne et al., 2003). Conversely, at ZJY, hybridization may proceed beyond the F1 and the direction of crossing is variable, making interspecific gene flow or the evolution of new hybrid taxa a possibility. Further work is now required to confirm that these differences in population structure are indeed due to habitat disturbance at ZJY.
This study has shown that at least some populations of R. agastum are the hybrid R. delavayi × R. irroratum, with the former the usual maternal parent and most hybrids being F1s. Hybrids of the combination R. delavayi × R. decorum may also have been referred to as R. agastum, but the type for the name R. agastum matches R. delavayi × R. irroratum. Like R. × sochadzeae from Turkey, R. delavayi × R. irroratum forms populations comprising only F1s in undisturbed habitats, which means that such hybrid populations could have existed indefinitely without any gene flow between the parents having occurred. That a small number of other hybrid classes occur at the heavily disturbed ZJY site indicates the possibility that human disturbance might bring about a small increase in the incidence of backcrossing, and hence introgression; however, the effect of disturbance on this taxon appears far smaller than is seen in R. × sochadzeae (Milne et al., 2003) or Encelia × laciniata (Kyhos et al., 1981).
Future work on Rhododendron hybrid zones, in both these species and others, should concentrate on comparing population structure in disturbed and pristine habitats, and establishing a causal relationship between habitat disturbance and changes in the pattern of hybridization.
Supplementary data are available online at www.aob.oxfordjournals.org and consist of the following tables. Table S1: Probabilities for three hybrid genotype categories calculated using binomial distribution. Table S2: Morphological identification, ITS types, chloroplast haplotypes and identity according to NewHybrids anaylsis of AFLP data, for all accessions examined. Table S3: AFLP data for NewHybrids and direct mathematical analysis for both sites.
We thank Professor Hu Yunqian who placed his laboratory at our disposal for DNA extraction, sequencing and AFLP analysis. We also thank Professors Fang Ruizheng and Ming Tianlu for helpful discussions and specimen identifications. This study was supported by grants from National 973 project (2003CB415103), National Science Foundation of China (40332021, 30625004, 30420120049 to H.S.), Natural Science Foundation of Yunnan, China (2007C086M to H.G.Z.), the Innovation Project of the Chinese Academy of Sciences (KSCX2-1-09 to H. S.) and Yunnan Provincial Government through an Award for Prominent Contributions in Science and Technology to Professor Wu Zheng-Yi in 2001(KIB-WU-02 to H.S.).