DNA extraction and AFLP analysis

In each population, a leaf fragment of ca. 2 cm² was collected for 10–20 plants of each of the taxa (see Table 1), and the plant tissue was desiccated using silica gel in individually sealed plastic bags. Genomic DNA was extracted using a slight modification of the CTAB protocol of Doyle & Doyle (1987). Plant leaf material was macerated in 900 µL of standard CTAB buffer, incubated at 60 °C for 30 min, extracted twice with chloroform-isoamyl alcohol, precipitated with isopropanol and washed with 70% ethanol. Precipitated DNA was then resuspended in 30 µL of distilled water. We obtained AFLP fragments using the methods of Vos et al. (1995), with modifications as reported in Moccia, Widmer & Cozzolino (2007) using fluorescent dye-labeled primers. Approximately 250 ng of genomic DNA was digested with EcoRI and MseI restriction endonucleases, and then ligated with the appropriate adaptors. A pre-selective amplification of restriction fragments was conducted using a tem of 1 µL of restriction-ligation product and with EcoRIA + MseIA or MseIC as primers. After a preliminary screening for the variability and reproducibility, five selective combinations were chosen for this study: EcoRIA–MseICGG, EcoRIA–MseIACT, EcoRIA–MseICCAA, EcoRIA–MseICGTA, EcoRIA–MseIACTG.

The selective amplifications were conducted with 1 µL of a 1:10 dilution of pre-amplification product.

Separation and detection took place on a 3130 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). GeneScan-500 LIZ (Applied Biosystems) was used as IS (internal standard). The electrophoregram generated by the sequencer was analysed using the GeneMapper version 3.7 software package (2004; Applied Biosystems). Clear and unambiguous peaks, between 50 and 500 bp, were considered as AFLP markers and scored as present or absent in order to generate a binary data matrix. DNA of both allopatric species was amplified and run in duplicate to validate repeatability. The AFLP analysis was performed considering two data sets: the first, contained the Botton plants group + allopatric (plate-Bt), and the second contained the Bois Niau plants group + allopatric (plate-BN). These two data sets were run and scored independently.

We calculated FST values to estimate the population differentiation using the software AFLP-SURV v. 1.0 (Vekemans, 2002). Genetic structure was explored using Principal Coordinates Analysis (PCoA) in GENALEX (Peakall & Smouse, 2006). We performed a Bayesian clustering analysis that allows to estimate the number of genetic clusters (i.e., populations), allele frequencies within clusters, and the genetic composition of individuals, by assigning the latter to different groups in which deviations from Hardy–Weinberg equilibrium and linkage equilibrium are minimized (Jacquemyn et al., 2012a). Data were analysed in STRUCTURE v. 2.3.1 (Pritchard, Stephens & Donnelly, 2000; Falush, Stephens & Pritchard, 2003) assuming an admixture model and correlated allele frequencies with 50,000 burn-in steps and 100,000 MCMC (Markov chain Monte Carlo) steps and K = 1–10, with ten independent runs per K. The goal was to estimate the K value that best fitted to our data.

The K value was assessed from the likelihood distribution (STRUCTURE output), which is the number of genetic clusters present in the data. K value fitting best with our data was selected using the ΔK statistic (Evanno, Regnaut & Goudet, 2005) produced by STRUCTURE HARVESTER (http://taylor0.biology.ucla.edu/struct_harvest/).

Finally, we used DISTRUCT (Rosenberg, 2004) to graphically display the output obtained with STRUCTURE.

NEWHYBRIDS (Anderson, 2008) was also performed to investigate the genetic profiles of the sympatric zone. We implemented a model that assumed two pure parental species and hybrids. This model assigns posterior probabilities for each individual to belong to one of the possibile six genotypic frequency classes: pure parental species, F1, F2, backcross to each parental species. A burn-in of 100,000 steps followed by run lengths of 1,000,000 steps was used (Jacquemyn et al., 2012a).

Moreover, the Hybrid index was estimated based in order to assess genome-wide admixture (Buerkle, 2005). This method calculated hybrid index (HI) based on a maximum likelihood and ranges between zero and one, corresponding to pure individuals of reference and alternative species, respectively. In our analyses, plants with a HI ranging between 0 and 0.2 were assigned to P. bifolia, whereas individuals with HI between 0.8 and 1 were assigned to P. chlorantha. We used AFLP data obtained from the allopatric P. bifolia and P. chlorantha individuals as parental data, while those obtained from the sympatric area were entered as putatively admixed individuals. This analysis was performed following the same parametric procedure proposed by Jacquemyn et al. (2012a). The plot was produced with the mk.image function in INTROGRESS. The hybrid index was estimated to assess genome-wide admixture using the est.h function ((Jacquemyn et al., 2012a) incorporated in the R program INTROGRESS (Gompert & Buerkle, 2010). Finally, we correlated the molecular hybrid index with morphological index obtained with the discriminant function (described previously) using Spearman’s rho method for non-normally distributed data (Jacquemyn et al., 2012a).