Phenotypic and genetic evidence for ecological speciation of Aquilegia japonica and A. oxysepala



  • Natural selection is thought to be a driving force that can cause the evolution of reproductive isolation. The genus Aquilegia is a model system to address how natural selection promotes the process of speciation.
  • Morphological differences between A. oxysepala, A. japonica and their hybrids were quantified for two vegetative (plant height and leaf area) and three floral morphological (sepal area, corolla length and diameter) traits. We also evaluated the genetic variability of the two species and their hybrids based on two chloroplast (1225 bp), four nuclear (5811 bp) genes and 15 microsatellites.
  • Our results revealed that differentiation of A. japonica and A. oxysepala at the ecological and morphological levels also involved divergence at the genetic level. In addition, the analysis of nucleotide variation patterns showed that the two species possessed numerous fixation sites at nuclear genes gAA4, gA7 and gAA12. Furthermore, we found that all of the phenotypic hybrids also showed a genetically admixed ancestry.
  • These findings suggest that natural selection has indeed facilitated the formation of distinct genetic variation patterns in the two Aquilegia species and habitat adaptation has been driving the ecologically based evolution of reproductive isolation.


Ecological speciation is the process whereby reproductive isolation evolves between populations or incipient species as a consequence of natural divergent selection on resource or habitat use (Schluter, 2001; Rundle & Nosil, 2005; Via & West, 2008). Recent empirical studies have demonstrated that natural divergent selection can arise from environmental differences, sexual selection or ecological interactions, and which can eventually promote the evolution of reproductive isolation in diverse ways (Rundle & Nosil, 2005; Schluter, 2009; Schluter & Conte, 2009).

The prominent examples of ecological speciation in plants have been reported from the studies of monkeyflower (Bradshaw et al., 1995, 1998; Schemske & Bradshaw, 1999a,b; Bradshaw & Schemske, 2003; Ramsey et al., 2003; Hall & Willis, 2006; Martin & Willis, 2007; Wu et al., 2007). For example, genetic analyses of floral traits differentiating Mimulus species revealed that genes with a large effect on pollinator preference could play an important role in the evolution of plant reproductive isolation and speciation (Bradshaw et al., 1998; Schemske & Bradshaw, 1999a,b; Bradshaw & Schemske, 2003). In addition, evidence of ecological speciation has been documented in a series of organisms including fish (Hatfield & Schluter, 1999; Rocha et al., 2005), finches (Ryan et al., 2007), mosquitoes (Turner et al., 2005; Turner & Hahn, 2007; Hahn et al., 2012), sunflowers (Rieseberg, 2000) and fungi (Douhan et al., 2008). These empirical studies have demonstrated that natural divergent selection could drive the evolution of reproductive isolation both directly, by acting on morphological and physiological traits, and indirectly, through the effects of pleiotropy and linkage (Sobel et al., 2009).

In this study, we take advantage of the model system Aquilegia L. (columbine) to investigate how natural selection drives the evolution of reproductive isolation and ultimately causes the occurrence of speciation. The genus Aquilegia includes c. 70 species and is widely distributed in temperate regions of northern hemisphere (Munz, 1946). It has been shown that species within Aquilegia show obvious differences in vegetative and floral traits with specializations for different ecological niches and pollinators (Hodges & Arnold, 1994; Nold, 2003; Whittall & Hodges, 2007; Kramer, 2009; Kramer & Hodges, 2010; Sharma et al., 2014). For example, the diversification of columbines in North America is associated mainly with the adaptation to a number of different pollinators (Grant, 1952; Hodges & Arnold, 1995; Hodges, 1997; Fulton & Hodges, 1999). By contrast, geographic isolation and shifts in habitat use have played more important roles in the evolution of reproductive isolation between European columbines (Bastida et al., 2010; Fior et al., 2013). In Asia, the number of columbine species is nearly the same as those of Europe and North America (Munz, 1946), yet relatively few investigations have focused on adaptive speciation, leaving the factors underlying their diversification unclear to date. Recently, several studies have investigated the mating system, floral traits and habitats of three Aquilegia species of central China and demonstrated that mechanical isolation through differential pollinators and habitats has contributed to the reproductive isolation between these species (Huang et al., 2004; Yu & Huang, 2006; Tang et al., 2007).

Here, we investigated two Aquilegia species of northeastern China, namely A. japonica Nakai and A. oxysepala Trautv. et Mey. Aquilegia japonica is a typical alpine species with ground-hugging habit and usually shows a tendency toward reduction in plant height and leaf area (Fig. 1 and Supporting Information Fig. S1). By contrast, A. oxysepala is broadly found in low elevation habitats, such as open land, roadside and forest margins (Figs 1,S1). Notably, several putative hybrids with mixed phenotypic characters were found in a contact zone of A. japonica and A. oxysepala. We noted that the two Aquilegia species were patchily distributed within the contact zone and that the hybrids are distributed with their parents respectively. To assess the morphological divergence between A. oxysepala, A. japonica and their hybrids, we measured two vegetative (plant height and leaf area) and three floral (sepal area, corolla length and diameter) traits. In addition, previous studies have demonstrated that adaptive radiation of this genus has led to a large number of species with obvious morphological and ecological differences but low genetic divergence (Hodges & Arnold, 1994; Ro & McPheron, 1997; Whittall et al., 2006; Cooper et al., 2010; Martinell et al., 2010). For instance, although A. formosa and A. pubescens show obvious morphological and ecological differences, low levels of genetic differentiation were observed at nine nuclear genes (Cooper et al., 2010). To this end, we evaluated the genetic diversity and differentiation of A. japonica and A. oxysepala based on 15 nuclear microsatellites, two chloroplast (rps16 and psbH-psbB) and four nuclear (gA7, gAA4, gAA12 and gEST8) genes. Our aim for this study was to evaluate the morphological differences of floral and vegetative traits between the two Aquilegia species and their hybrids, and determine their intra- and interspecific genetic variability. We expect findings of this study to provide new insights into the underlying factors driving the diversification of Asian columbines.

Figure 1.

Flowers and leaves of Aquilegia japonica (a, d, g), hybrid (b, e, h) and A. oxysepala (c, f, i).

Materials and Methods

Material sampling and morphological analysis

All of the samples used in this study were collected from 2010 to 2012 and dried with silica gel. The localities and other information about these specimens are given in Table 1 and Fig. 2. In brief, 145 A. oxysepala individuals were collected from a total of 13 populations across its range in China, whereas 130 individuals of A. japonica were sampled from three populations. In addition, 12 hybrids of A. oxysepala and A. japonica were obtained from the hybrid zone of their parental localities. Furthermore, 20 individuals of A. yabeana Kitag and A. parviflora Ledeb were collected from two allopatric localities, respectively.

Table 1. Population sampled and number of individuals used in each analysis
TaxaPop.aElevation (m a.s.l.)Latitude (N)Longitude (E)No. indiv. cpDNANo. indiv. microsatelliteNo. indiv. gA7No. indiv. gAA4No. indiv. gAA12No. indiv. gEST8
  1. a

    The population names in this table are the same as those used in Figs 2 and 4.

Aquilegia japonica JCB180442.053128.048586016161621
A. oxysepala OBY23039.594122.57810102224
A. yabeana YBZ17041.364121.4222102422
A. parviflora PAL69950.422121.4762102422
Figure 2.

Sampling locations of 19 populations. Different species are indicated by distinct symbols (square with black shadow, Aquilegia japonica; circle, A. oxysepala; square with spot, hybrid; hexagon, A. yabeana; star, A. parviflora). The dotted line represents the distribution range of A. oxysepala in northeastern China.

In order to evaluate the morphological divergence among A. oxysepala, A. japonica and their hybrids, two vegetative traits of plant height (height of the whole plant) and leaf area (a single leaf of each leaflet) were quantified due to the crucial roles these features play in the diversification of European columbines (Medrano et al., 2006; Castellanos et al., 2011). In addition, the sepal area, corolla length and diameter (diameter from one petal to the opposite petal) were also measured because previous studies have demonstrated that floral morphologies have profound effects on the evolution of reproductive isolation (Hodges & Arnold, 1995; Hodges, 1997; Fulton & Hodges, 1999; Hodges & Derieg, 2009; Kramer, 2009; Sharma et al., 2014). As the individuals from the same location of a population tend to show similar phenotypic characters, we therefore randomly sampled only one individual from each location. Accordingly, specimens examined for morphology included 15 A. japonica individuals from population JPD and JCB. Likewise, 15 individuals of A. oxysepala were also collected from OPD and OCB, respectively. In addition, nine hybrids of A. japonica and A. oxysepala were sampled from the population HPD. Level of morphological differences among the two Aquilegia species and their hybrids were assessed with t-test using SPSS software (SPSS Inc., Chicago, IL, USA). In addition, we also performed a Principal Coordinate Analysis (PCA) (Anderson, 2003) based on both vegetative and floral variables of the two species and their hybrids.

DNA extraction, chloroplast genome scanning and sequencing

Total genomic DNA was extracted from mature leaves of each individual using a Plant Genomic DNA kit (TianGen, Beijing, China). A preliminary universal primer scanning of the chloroplast genome was conducted on four individuals sampled from different populations of A. oxysepala. Two of these primer pairs (rps16 and psbH-psbB) showed different haplotypes within the four examined individuals of A. oxysepala were thereafter used for the large-scale survey of nucleotide diversity. PCR was performed using an ABI 2720 Thermocycler (Applied Biosystems, Foster City, CA, USA) in 30-μl volumes each containing the following components: 20–50 ng template DNA, 1× PCR buffer (Mg2+ free), 0.2 mM each dNTP, 2.5 mM Mg2+, 0.5 μM for forward and reviser primer, 1 unit of rTaq DNA polymerase (Takara, Dalian, China). The PCRs were carried out under the following conditions: 5 min at 94°C, then 35 cycles of 30 s at 94°C, 30 s at the annealing temperature (Supporting Information Table S1), and 90 s at 72°C, with an 8 min extension of 72°C treatment in the final cycle. These PCR products were visualized following agarose gel electrophoresis and sequenced on an ABI 3730 DNA analyzer (Applied Biosystems). All sequences were deposited in GenBank under the accession numbers KF757475KF758265, KF768356KF768360 and KJ678976KJ679271.

Microsatellite genotyping

Nuclear DNA variation was characterized at 15 nuclear microsatellites (Table 2) for all samples. These microsatellites were retrieved from Li et al. (2011) based on the reliability of amplification, the ease of scoring and transferability. The sequences of these primer pairs and PCR conditions were described in detail in Li et al. (2011). In this study, these microsatellites were amplified and scored with two different conditions: using 0.3 μM M13-tailed forward primer, 0.1 μM dye-labelled (FAM, HEX and TAMRA) M13 primer (5′CACGACGTTGTAAAACGAG3′) and 0.4 μM reverse primer; PCR products were resolved in an ABI PRISM 3100 Genetic Analyser (Applied Biosystems).

Table 2. Genetic diversity and differentiation for each microsatellite at the species level
LocusScaffoldaDescription F ST Aquilegia japonica A. oxysepala
N A b H E c H O d N A H E H O
  1. aThe locations of each microsatellite were determined according to the scafford accession of Phytozome. bNumber of alleles. cExpected number of heterozygotes. dObserved number of heterozygotes. eThis microsatellite is located in the exon region. fThis microsatellite is located in the intron region. gThis microsatellite is located within 5 kb upstream or downstream of this gene. hNo gene is located in the downstream of this microsatellite in Phytozome. We therefore annotated the function according to BLASTX of GenBank.

EST1 3Galactinol synthasee0.11200.870.6750.690.42
AA9 5Peroxidaseg0.1830.470.3450.230.22
A9 7Cyclin Bg0.0050.330.2320.050.04
A41 7Serine/threonine protein kinaseg0.1940.200.2050.500.70
EST6 7WD-repeat proteine0.00260.900.61210.870.55
EST3 8Guanine nucleotide binding proteine0.00110.760.6670.580.32
AA12 9YABBY proteinf0.50120.730.6650.190.11
A7 13Histidinol dehydrogenasef0.7040.230.2240.260.17
EST8 13Vesicle associated member proteine0.06160.910.77160.850.48
A6 25AIR synthase-related proteing0.0030.190.0630.300.10
A26 26WD-40 repeat family proteinf0.2960.570.5840.270.18
B8 45ABC transportere0.0080.660.4750.280.23
EST7 58Phage chock proteine0.3290.760.5580.460.26
AA4 64Integrase core domain containing proteing,h0.0820.040.0020.010.00
EST2 76eIF-2B family proteine0.0060.630.5550.720.49

Nuclear gene isolation, identification and sequencing

In order to identify the physical location of each microsatellite, sequences of the 15 microsatellites were searched against the Phytozome v9.1 nucleotide database using BLAST ( Accordingly, all of the 15 microsatellites exhibited significant homology to the known genomic scaffold of A. coerulea Goldsmith (Table 2). To further investigate the nucleotide variation patterns of A. oxysepala and A. japonica, two microsatellites (A7 and AA12) that showed high interspecific genetic differentiation and located in the intron region were chosen for subsequently analyses (Table 2). In detail, locus AA12 is located in the intron of a member of the YABBY gene family (Table 2). Previous studies have documented that this gene family played important roles in the development of all lateral organs including cotyledons, leaves, floral meristems and organs (Golz & Hudson, 1999; Bowman, 2000; Huang et al., 2013). Likewise, locus A7 is located in the histidinol dehydrogenase gene which plays a critical role in plant growth and development (Muralla et al., 2007; Gao et al., 2008). By contrast, two microsatellites (AA4 and EST8) were also selected because they exhibited low interspecific genetic differentiation. For instance, gEST8 is located in the vesicle-associated membrane gene and is mostly involved in vesicle docking (Levine et al., 2001). The nuclear gene gAA4 is located within 5 kb upstream of an integrase core domain containing protein (Table 2). It has been demonstrated that this gene functions in mediating the final step of retrotransposon propagation (Suoniemi et al., 1998). To obtain the upstream and downstream sequences of the four microsatellites from A. oxysepala and A. japonica, PCR primers were designed according to the reference of A. coerulea using Primer Premier 5 ( One to 21 individuals were selected from the 19 populations and then surveyed for nucleotide polymorphisms of the four nuclear genes (Table 1). Standard PCR reactions were performed in 20-μl volumes containing 20–50 ng genomic DNA, 1× PCR buffer (Mg2+ free), 2.5 mM Mg2+, 0.2 mM of each dNTP, 0.5 μM of forward and reverse primers and 1 unit of LA Taq (Takara). The amplification profile was 94°C for 5 min, followed by 35 cycles at 94°C for 30 s, annealing temperature (Table S1) for 30 s, 72°C for 3 min, and a final extension at 72°C for 8 min. These amplified products were purified using a quick gel extraction kit (Takara) and cloned using the pMD-18 vector (Takara). For the individuals of A. oxysepala and A. japonica, we sequenced only one clone for each gene. By contrast, in order to gain the entire haplotypes of each hybrid, two to five clones were sequenced for each gene. The clones with positive insert were sequenced using M13 forward and reverse primers on an ABI 3730 DNA sequencer (Applied Biosystems).

Phylogeny, divergence time and interspecific gene flow

In order to gain the systematic relationship of A. oxysepala and A. japonica, we performed phylogenetic analyses of genus Aquilegia based on internal transcribed spacer (ITS) and 10 chloroplast genes (Table S1), respectively. In brief, one individual was randomly selected from each population of the four investigated Aquilegia species. A total of 17 individuals covering the distribution range of A. oxysepala and A. japonica and two individuals of A. yabeana and A. parviflora were sequenced. All other homologous sequences of ITS and chloroplast genes of Aquilegia were retrieved from GenBank according to Bastida et al. (2010) and Fior et al. (2013). To further confirm the phylogenetic relationship of A. oxysepala and A. japonica, we also sequenced the single copy nuclear gene glyceraldehyde 3-phosphate dehydrogenase (G3PDH) of the four Aquilegia species and retrieved homologous sequences of other five Aquilegia species from GenBank according to Whittall et al. (2006). Based on the maximum-likelihood tree (Fig. 3) of Fior et al. (2013), the nine species represented six distinct groups of Aquilegia and are distributed across North America, Asia and Europe. Therefore, systematic analysis based on G3PDH can also partly reflect the phylogenetic relationship of A. oxysepala and A. japonica.

Figure 3.

Measurement of morphological differences (a–e) and PCA analysis (f) between Aquilegia japonica, A. oxysepala and their hybrids. Pa, Pb and Pc are P-value of the comparisons between A. japonica and hybrid, A. japonica and A. oxysepala, and A. oxysepala and hybrid, respectively. The box plots are the 95% confidence intervals and the horizontal dashed lines in each box plot are the range of value. The circles in (d) are the abnormal value because the corolla length of some samples deviates from the mean value.

Initial sequence editing and alignment were performed using ContigExpress (Informax Inc. 2000, North Bethesda, MD, USA). These raw sequences were aligned in ClustalX 1.83 (Thompson et al., 1997) and if necessary edited manually in BioEdit 7 (Hall, 1999). The combined chloroplast data matrix was obtained by integrating the 10 chloroplast genes. Bayesian inferences (BI) of the combined chloroplast and G3PDH dataset of these Aquilegia species were performed using MrBayes (Ronquist & Huelsenbeck, 2003). The best-fit models of nucleotide substitutions were estimated by jModelTest (Posada, 2008) under the Akaike information criterion (Posada & Buckley, 2004) and the substitution models are TVM+G and GTR for combined chloroplast and G3PDH, respectively. In addition, a haplotype network was inferred by constructing the statistical parsimony network for matK and ITS using the software TCS 1.21 (Clement et al., 2000) due to the small number of informative characters found at the two genes. We also applied the software IMa2 (Hey, 2010) to estimate divergence time of A. oxysepala and A. japonica based on the chloroplast gene rps16 and psbH-psbB. To infer the migration parameters between the A. oxysepala and A. japonica, MIGRATE-N (Beerli & Felsenstein, 1999, 2001) analyses were performed for the microsatellites, chloroplast and nuclear genes separately. This coalescent-based program uses a Markov chain Monte Carlo approach to explore possible genealogies with branch lengths and with migration events (Beerli & Felsenstein, 1999, 2001). The input data were defined as DNA/RNA model for DNA sequences, and microsatellite model for microsatellite data. Each model was run using random genealogy and values of the parameters θ and M generated by FST estimation as start condition. An adaptive heating scheme with four different temperatures (1.00, 1.50, 3.00 and 1000 000) was employed and a constant mutation rate was assumed for all loci. The migration parameter M was reported by combining five independent replicates.

Nucleotide diversity and statistical parsimony networks

Sequences from each of the chloroplast (rps16 and psbH-psbB) and nuclear (gA7, gAA12, gAA4 and gEST8) genes were aligned separately in ClustalX 1.83 (Thompson et al., 1997). The two chloroplast loci rps16 and psbH-psbB were combined into a single contiguous dataset for further analysis. The genetic analyses of the sequence polymorphism were performed using DnaSP v5 (Librado & Rozas, 2009), including number of segregating sites (S), number of haplotypes (h), genetic differentiation (FST), haplotype diversity (Hd), nucleotide diversity (π) (Tajima, 1983) and (θ) (Watterson, 1975). In addition, the neutrality test statistic Tajima's D (Tajima, 1989) was also estimated for cpDNA and nuclear genes using DnaSP (Librado & Rozas, 2009).

In order to assay the possible intra- and interspecific relationships of chloroplast and nuclear haplotypes, statistical parsimony networks were estimated for the combined chloroplast and each nuclear gene independently using TCS 1.21 (Clement et al., 2000). For the combined cpDNA data, the INDELs were treated as potential information sites with a 95% connection limit. By contrast, abundant INDELs were found in the nuclear genes gA7, gAA12, gAA4 and gEST8, and therefore we treated the gap as missing and attempted to connect the haplotypes by increasing the connection threshold up to a maximum of 50 steps.

Population structure analyses of microsatellite loci

For the microsatellite data, the program Micro-checker (Van Oosterhout et al., 2004) was applied to test the null alleles. In addition, we employed ARLEQUIN 3.5 (Excoffier & Lischer, 2010) to survey the expected heterozygosity (HE), observed heterozygosity (HO) and number of alleles (NA) for each locus at both the population and species levels. Allelic frequencies of each microsatellite locus were calculated to evaluate the pattern of genetic variation between the two species. The estimator FST (Weir & Cockerham, 1984) was also calculated for each locus using ARLEQUIN 3.5 (Excoffier & Lischer, 2010). Significant differences found during the comparisons between FST of each locus were evaluated with t-test. Furthermore, we also tested for isolation-by-distance with IBDWS (Jensen et al., 2005) for all samples of A. oxysepala, A. japonica and their hybrids.

Additionally, we explored the admixture of the four investigated Aquilegia species using STRUCTURE (Pritchard et al., 2000). The admixture model with the option of correlated allelic frequencies between populations was chosen because it assumes that allelic frequencies between populations are prone to be similar owing to migration or ancestral polymorphism (Falush et al., 2003). Thereafter, the numbers of ancestral K clusters were calculated by comparing the likelihood rations in 10 replicates runs for K values from 1 to 10. Each analysis consisted of 105 steps following a 105 step burn-in period.


Divergence time, morphological and phylogenetic analyses

Molecular phylogenetic analyses of the genus Aquilegia were performed using the obtained sequences of ITS, G3PDH and the combined chloroplast dataset. A total of 534, 779 and 4896 characters were included in the ITS, G3PDH and chloroplast data matrix, respectively, of which 23, 56 and 117 characters (respectively) were variable sites. Parsimony network analysis based on ITS revealed that all haplotypes of A. oxysepala and A. japonica were grouped together as a distinct clade and three haplotypes were shared between them (Fig. S2). Similarly, the Bayesian tree of G3PDH (Fig. S3) also demonstrated that A. oxysepala and A. japonica were the closest species within the nine Aquilegia species. This conclusion was further supported by the Bayesian tree of combined chloroplasts (Fig. S4) which showed that A. oxysepala and A. japonica are the most closely related species within Aquilegia. Notably, we found that although the species A. oxysepala, A. yabeana and A. parviflora used in this study were consistent in their phylogenetic placement with that found in Fior et al. (2013), the phylogenetic position recovered herein for A. japonica (Fig. S4) differed greatly from Fior et al.'s (2013). However, we noted that the voucher of A. japonica used in Fior et al. (2013) was collected from Japan. According to the Flora of Japan, the species A. japonica was also known as A. flabellata var. pumila, and we therefore suspect that the vouchers of A. japonica used in our study and Fior et al. (2013) belonged to different subspecies or varieties. This possibility was further illustrated by the parsimony network of the matK gene (Fig. S5, haplotype A) wherein the species A. flabellata used in Bastida et al. (2010) shared the same haplotype with the species A. japonica of Fior et al. (2013). Taken together, our findings suggest that A. oxysepala and A. japonica from northeastern China originated from the most recent common ancestor.

In addition, the analysis of divergence time based on rps16 and psbH-psbB revealed a recent divergence of A. oxysepala and A. japonica (0.05 Myr ago). Interestingly, PCA analysis based on the vegetative and flora traits suggested that the two Aquilegia species and their hybrids showed obvious morphological differentiation (Fig. 3f). Specifically, comparisons from morphological analyses revealed that the two species exhibited significant differences in plant height, leaf area and sepal area (Fig. 3a,b,e).

Genetic diversity and population differentiation

Summary statistics for microsatellites revealed that a total of 159 alleles were detected across 15 microsatellite loci in all samples. The number of alleles (NA) for each population ranges from 28 in OBY to 95 in JCB (Table S2). Genetic diversity in A. japonica was relatively higher than that of A. oxysepala at both the species and population levels (Tables 2, S2). For the analyzed chloroplast and nuclear genes, genetic analyses of gA7, gAA12 and gEST8 revealed that A. japonica also harbored relatively higher nucleotide diversities than those of A. oxysepala (Table 3), akin to the pattern observed in the microsatellite dataset. By contrast, locus gAA4 exhibited slightly lower nucleotide diversity in A. japonica than in A. oxysepala (Table 3). This result was consistent with the combined chloroplast dataset where A. oxysepala showed relatively higher haplotype diversity than that of A. japonica (Table 3). Taken together, these findings suggest that A. japonica harbors higher nuclear genetic diversity than that of A. oxysepala. In addition, significant negative values of Tajima's D were observed in most of the nuclear genes of A. japonica and A. oxysepala, suggesting a relative excess of rare variants in these genes.

Table 3. Information of nucleotide diversity for chloroplast and nuclear genes
LocusSpeciesLength (bp)θaπbSchdHdeDf
  1. All values are significant: *, P value < 0.05. aThe Watterson estimator of per base pair (Watterson, 1975). bAverage number of nucleotide differences per site between two sequences (Nei, 1987). cNumber of segregating sites. dThe number of haplotypes.

  2. eThe haplotype diversity. fTajima's D (Tajima, 1983). gThis gene is located in chloroplast genome. hThis gene is located in nuclear genome.

psbH-psbB g Aquilegia japonica 4060.00100.0005120.4201.394
A. oxysepala  0.00060.0005120.2310.298
rps16 g A. japonica 8190.00100.0005230.4331.790
A. oxysepala  0.00090.0007340.6100.578
cpDNA A. japonica 12250.00100.0005350.4452.126*
A. oxysepala  0.00080.00064100.6220.610
gA7 h A. japonica 16930.00700.013790321.000−1.831*
A. oxysepala  0.00200.007046230.909−2.618*
gAA12 h A. japonica 19520.01190.0194117311.000−1.470
A. oxysepala  0.00380.011873270.985−2.538*
gAA4 h A. japonica 17090.00240.008658300.994−2.665*
A. oxysepala  0.00280.008859300.994−2.544*
gEST8 h A. japonica 4570.00670.00619130.8850.279
A. oxysepala  0.00100.0036760.300−1.954*

Measures of population genetic structure and differentiation of these Aquilegia species were performed for the 15 microsatellites. Bayesian estimation of the number of distinct clusters is presented in Fig. 4 (= 3–6). Accordingly, the population assignment analysis showed a clear genetic break between A. oxysepala and A. japonica (= 3). Within A. japonica, the genetic variation patterns of the three populations remained intimately associated with their geographical distribution in the STRUCTURE analysis (= 6). However, low levels of population genetic differentiation were observed among the three A. japonica populations (Table S3). Likewise, Bayesian analysis also revealed the geographic pattern of A. oxysepala. The majority of individuals from southern populations (OLB, ODH, OCB, OHN, OJA, OTD, OKD and OBY) were assigned to one genetic group and nearly all individuals from the northern populations (OLC, OPD, OLK, OJX and OWC) were assigned to another genetic group (= 6).

Figure 4.

Bayesian population assignment test for the four Aquilegia species inferred by STRUCTURE based on different microsatellites (= 3–6). The bars on the figure represent these individuals were retrieved from the same species. Bar plots showing Bayesian assignment probabilities. Each vertical bar corresponds to one individual. Populations are separated by black bars and identified at the bottom.

Haplotype network based on cpDNA and nuclear genes

Across the entire cpDNA dataset, a total of 285 Aquilegia sequences were generated in this study, which corresponded to 13 unique haplotypes. Accordingly, 11 haplotypes were found in A. oxysepala and A. japonica (Fig. 5a). Among the 11 haplotypes, four were shared between A. japonica and A. oxysepala, six were specific to A. oxysepala and one was unique to A. japonica (Fig. 5a). In addition to the observed patterns of chloroplast haplotype sharing, the geographic distribution of the cpDNA haplotypes was also found for the two species. Within A. japonica, for example, only one haplotype (HapH) was identified in population JCB and all individuals of JLB were HapC except one individual which had a haplotype (HapA) alone. A similar phenomenon was also found in A. oxysepala in which HapC, HapD, HapE, HapG, HapI and HapK were mainly found in southern populations. The geographic patterning of the two species found in the cpDNA network was broadly in agreement with those depicted in the analysis of STRUCTURE based on microsatellites (Fig. 4).

Figure 5.

Parsimony networks of the cpDNA (a), gEST8 (b), gA7 (c), gAA12 (d) and gAA4 (e) in the Aquilegia species. Each circle represents one haplotype and circle area is proportional to the frequency of each haplotype and dots on branches indicate mutational steps.

By contrast, there was striking genetic distinction between A. japonica and A. oxysepala at nuclear genes gA7, gAA12 and gAA4. At locus gAA4, for instance, all of the 32 examined A. oxysepala individuals fell within a lineage that was 20 mutational steps apart from the cluster containing all the haplotypes of A. japonica (Fig. 5e). Similar results were also found in gA7 and gAA12 where haplotypes of the two species were clearly separated and haplotype sharing was nonexistent (Fig. 5c,d). The observed haplotype patterns of the three nuclear genes supported the conclusion that the two Aquilegia species harbored substantial genetic divergence. Nonetheless, results of gEST8 (Fig. 5b) showed that two haplotype were shared between A. japonica and A. oxysepala.

Gene flow and heterogeneous genomic differentiation

Empirical studies have illustrated that recently diverged species usually share some of their alleles owing to their common ancestry and historical or recent gene flow. As expected, sharing alleles or haplotypes between A. oxysepala and A. japonica were observed at ITS (Fig. S2), cpDNA (Fig. 5a), gEST8 (Fig. 5b) and microsatellites (Fig. S6). Although population genetic analyses based on the 15 microsatellites and three nuclear genes (gA7, gAA12 and gAA4) revealed that the two species exhibited distinct patterns of genetic clusters (Figs 4, 5c–e), results from MIGRATE-N suggested that gene flow might have occurred in the process of ecological speciation (Fig. 6). In addition, the IBD analysis based on all populations of the two species and their hybrids revealed a positive correlation between genetic and geographical distance (= 0.025) (Fig. S7), suggesting increased gene flow with geographic proximity both within and between species. Notably, we found a reduction of gene flow at the three nuclear genes gA7, gAA12 and gAA4 (Fig. 6a).

Figure 6.

Migration rate between Aquilegia japonica and A. oxysepala based on gA7, gAA12 and gAA4 (a), 15 microsatellites (b), gEST8 (c) and two chloroplast genes (d) and. The numbers on the y-axis represent the mutation-scaled effective immigration rate M. MJ → O indicates the gene flow from A. japonica to A. oxysepala; MO → J indicates the gene flow from A. oxysepala to A. japonica. The box plots are the 95% confidence intervals and horizontal lines are the range of value. Dashed lines separate the different molecular markers.

Patterns of genetic differentiation were analyzed for each microsatellite and nuclear gene separately. The results showed that values of FST between the two species were highly variable across the 15 microsatellites (t-test, < 0.01) (Table 2). This phenomenon was further documented by the results of nuclear genes in which the interspecific FST values for gA7, gAA12, gAA4 and gEST8 were 0.40, 0.46, 0.82 and 0.19, respectively. In addition, several microsatellites showed differences in allelic frequencies between A. japonica and A. oxysepala (Fig. S6). Through examining the nucleotide variation patterns of the four nuclear genes, we found several fixation sites (five sites for gA7, nine sites for gAA12 and 20 sites for gAA4, respectively) that distinguish between A. japonica and A. oxysepala.

Phenotypic and genealogical patterns of hybrids

Twelve hybrids of A. japonica and A. oxysepala were collected from the hybrid zone between population OPD and JPD (Fig. 2). Notably, we found that the two Aquilegia species are distributed patchily in the contact zone and their hybrids cohabit with their parental species. As shown in Figs 1 and 3, the hybrids possessed intermediate morphological characteristics (excepting the corolla length or diameter) between A. japonica and A. oxysepala, providing the phenotypic evidence for their hybrid nature. The genetic analyses also revealed the admixed ancestry of these hybrids. For the cpDNA dataset, all of the 12 examined hybrids fell into two haplotypes (HapB and HapH) which were also found in sympatric populations of their parents (JPD for A. japonica and OPD for A. oxysepala, respectively). Similarly, the result of STRUCTURE based on microsatellites also revealed that the 12 hybrids were characterized by two parental structured genetic clusters (Fig. 4, K = 3). These phenomena were further confirmed by the network analyses of gA7, gAA12 and gAA4 which demonstrated that each investigated hybrid contained two distinct haplotypes and each of them was tightly clustered with their parental species (Fig. 5c–e). Specifically, we noted that these hybrids comprised a mosaic of different parental genotype frequencies (Fig. 4). Taken together, these findings indicated that the 12 hybrids were indeed derived from recent interspecific hybridizations between A. japonica and A. oxysepala.


Genetic and phenotypic divergence associated with adaptation to distinct ecological habitats

It has been documented that the genus Aquilegia underwent recent adaptive radiations with specialization to different pollinators and habitats (Hodges & Arnold, 1995; Hodges, 1997; Whittall & Hodges, 2007; Bastida et al., 2010). Here, we analyzed the morphological divergences of two vegetative and three floral traits between A. japonica and A. oxysepala, and evaluated the genetic diversity and differentiation of the two Aquilegia species. Previous studies have demonstrated that floral traits of Aquilegia including flower color and length of nectar spurs have profound effects on pollinator visitation (Hodges & Arnold, 1995; Hodges, 1997; Fulton & Hodges, 1999; Whittall & Hodges, 2007; Kramer & Hodges, 2010; Sharma et al., 2014). For example, reproductive isolation between A. formosa and A. pubescens is influenced by differences in their flowers through their effects on pollinator visitation and pollen transfer (Hodges et al., 2002). In addition, it has been reported in European columbines that the morphological divergences of vegetative traits including plant height and leaf length were associated with ecological specialization (Medrano et al., 2006; Martinell et al., 2010; Castellanos et al., 2011). In this study, morphological analyses of floral traits between A. japonica and A. oxysepala demonstrated that although no significant differences were observed in corolla diameter and length, the two species showed obvious differences in flower color and petal area (Figs 1, 3). Similarly, comparisons of the vegetative traits also revealed that the two species showed significant differences in plant height and leaf area (Figs 1, 3). The observed morphological distinctions between the two species suggested that both floral and vegetative traits might be the potent force in the ecological speciation process. As shown in Fig. S1, however, A. japonica is a typical alpine species with substantially reduced plant size and shows a clear pattern of ecological differentiation from the low-elevation congener A. oxysepala. It indicated that habitat specialization might play more important roles in keeping the reproductive isolation. Another similar example of ecological speciation has been documented in A. formosa and A. pubescens where both floral and ecological isolation were observed between the two species, yet the differences in floral morphologies are the ‘primary factors’ that mechanistically cause the evolution of reproductive isolation (Hodges & Arnold, 1994; Fulton & Hodges, 1999; Hodges et al., 2002).

Closely related species that have undergone recent and rapid adaptive radiation are usually expected to share most of their alleles and show low genetic divergence. Indeed, this hypothesis has been documented by some previous studies which showed that low genetic divergence existed between North American and European columbine species (Hodges & Arnold, 1994; Cooper et al., 2010; Martinell et al., 2010). Here, our findings showed that although A. japonica and A. oxysepala possessed high genetic diversity and diverged recently (0.05 Myr ago), strong interspecific genetic differentiation was observed between them. For example, genetic analyses based on the 15 microsatellites and three nuclear genes (gA7, gAA4 and gAA12) revealed that the two species showed substantial genetic discontinuities. Specifically, a total of 34 sites reveal fixed differences between the two species in genes gAA4, gAA7 and gAA12. In addition, our results also revealed geographical patterning of both A. oxysepala and A. japonica. Garrido et al. (2012) also reported the spatial genetic structures of three Sardinian columbine species, but the geographical pattern they found is not fully compatible with the current taxonomic affiliations. By contrast, our observations demonstrated that the two species showed obvious genetic dissection, and even the geographically adjacent populations of the two species (e.g. JPD vs OPD) also exhibited a strong genetic break. Together, our findings suggest that divergence of A. japonica and A. oxysepala at the ecological and phenotypic levels also involved differentiation at the genetic level.

Ecological speciation with gene flow and heterogeneous genomic divergence

Empirical investigations have demonstrated that reproductive isolation can evolve in the presence of gene flow because in such cases, natural divergent selection could overcome the homogenizing effects of gene flow, and then generate distinct and isolated gene pools (Turelli et al., 2001; Coyne & Orr, 2004; Gavrilets, 2004; Fitzpatrick et al., 2009). Here, we present another case of divergence with gene flow, likely due to the divergent selection in contrasting habitats, between A. japonica and A. oxysepala by integrating morphological and genetic analyses. As shown in our results, MIGRATE simulations (Fig. 6) and IBD (Fig. S7) demonstrated that gene flows occurred between the two species. Notably, migration rates of the three nuclear genes gA7, gAA12 and gAA4 showed a reduction of gene flow (Fig. 6a), suggesting that natural selection might limit the gene exchange between the two species at these loci. These attributes, coupled with the observed morphological divergences, support the hypothesis that ecological adaptation in response to divergent selection can generate phenotypic diversity in the presence of gene flow (Smith et al., 1997; Freedman et al., 2010; Kirschel et al., 2011; Muñoz et al., 2013).

We noted that heterogeneous genomic divergence existed between A. japonica and A. oxysepala. The possible mechanism is that divergent selection itself can promote genetic differentiation through increasing genetic differentiation of regions affected by selection (Barton, 2000; Nosil et al., 2009). Taking the microsatellite AA12 as an example, the allelic frequency diagram showed that allele 123 tends to be fixed in A. oxysepala, while A. japonica shows a more diverse pattern in this locus (Fig. S6). This observation is consistent with the hypothesis that divergent selection on one locus can strongly affect the frequency of alleles at selected loci (Kaplan et al., 1989; Andolfatto, 2001; Nosil et al., 2009). In addition, the analyses of the nucleotide variation pattern revealed that nine fixed sites were identified at the nuclear gene gAA12 between A. oxysepala and A. japonica. Nevertheless, it should be noted that although the nuclear gene gAA4 is located upstream of the microsatellite AA4, a high level of interspecific genetic differentiation was observed at this gene. A possible explanation might be that gAA4 is physically linked with selected genes. These findings, associated with the observed morphological divergences between A. oxysepala and A. japonica, yielded the conclusion that natural divergent selection might have directly or indirectly shaped the genetic variation patterns in the two species and eventually resulted in heterogeneous genomic divergence.

Hybridization between A. oxysepala and A. japonica

Hybridization between closely related species has been well documented in diverse plant species (Ellstrand et al., 1996; Campbell, 2003; Arias et al., 2008, 2012; Abbott et al., 2013; Lagache et al., 2013). In this study, 12 individuals with hybrid phenotype and mixed ancestry were identified in the contact zone of A. oxysepala and A. japonica. In general, the alpine species A. japonica occupies high-elevation habitats and A. oxysepala usually occurs in open land habitats of low-altitude regions, such as the margins of road, farmland and forest (Fig. S1). Interestingly, several high-elevation individuals of A. oxysepala were distributed along the newly built road of population OPD and show an overlapping distribution with A. japonica. Similarly, we also found that some A. japonica individuals showed an overlapping distribution with A. oxysepala at low altitudes. It seems possible that A. japonica and A. oxysepala recently spread into the low- and high-altitude regions, respectively, likely by migrating along the margins of the new road. The formation of a contact zone putatively wrecked the ecologically based reproductive isolation between A. oxysepala and A. japonica.

Three hybrid zone models have been proposed according to how selection acts on hybrids and parent species (Arnold, 1999; Abbott & Brennan, 2014). The tension zone model assures that hybrids are of low fitness relative to their parents and selection against hybrids is environment-independent (Barton & Hewitt, 1985). By contrast, the hybrids of the bounded hybrid superiority zone model have higher fitness than either of the parent species in the hybrid zone and selection on hybrids is environment-dependent (Moore, 1977). In the mosaic hybrid zone model, hybrids comprise a mosaic of different parental genotype frequencies and selection acts against hybrids in both environment-independent and –dependent manners (Harrison, 1986, 1990). In this study, our results from genetic and morphological analyses have tended to support the mosaic hybrid zone model better than the other two models. First, the two Aquilegia species are patchily distributed in the contact zone and the hybrids cohabit in both ecological environments with their parents. Second, the morphological analyses demonstrate that these hybrid individuals showed diverse phenotypic variations (Fig. 3). This attributes were also documented by the population genetic analysis where distinct hybrids possessed different parental genotype frequencies (Fig. 4). Together, the findings presented herein provide direct evidence for the formation of a mosaic hybrid zone and ecological adaptation has contributed to the speciation of the two Aquilegia species.


We thank Richard Abbott, Dolph Schluter and Ren-Chao Zhou for valuable suggestions for a previous version of this manuscript. We are also grateful to Elena Kramer and the three anonymous reviewers for their constructive comments. This work was financially supported by the Program for Introducing Talents to Universities (B07017) and National Natural Science Foundation of China (31100157).