Ruellia nudiflora is a 20-30-cm tall perennial plant that shows few ramifications (3-6) and produces chasmogamous (CH) and cleistogamous flowers (CL) (Long, 1971). Fruit production is similar in both flower types (approximately 70% fruit set success) (Munguía-Rosas et al., 2012). The proportion of CH flowers in this species throughout the season is reported to be 0.43-0.47 (Munguía-Rosas, Campos-Navarrete, & Parra-Tabla, 2013). During the flowering peak in the area under study, individual plants produce 6-12 CH flowers every day, which are visited by at least 6 bee species, such as Apis mellifera Linnaeus (Apidae) and Trigona fulviventris Guérin-Méneville (Apidae), and 5 butterfly species, the most common being Microtia elva Bates (Nymphalidae) (Abdala-Roberts et al., 2012). CH flowers are self-compatible and of tubular shape, the 20-25-mm long corolla opening for only 1 day. Fruits from both flower types are dry and dehiscent, measure 1-2 cm in length, and produce 6-15 seeds each. Seeds from both flower types remain viable for 3-4 months. This plant rarely grows clonally; Abdala-Roberts et al. (2012) found that virtually all recruitment occurs via seeds. Species in the Acanthaceae family disperse their seeds by capsule explosion once they are mature, in both flower types (Witztuma, & Schulgasserb, 1995).

To estimate genetic structure and diversity, and to evaluate demographic history, nine populations were sampled: Conkal (CKL), Inifap (INF), Dzemul (DZL), San José Tzal (SJT), Molas (MOL), San Antonio Tehitz (SAT), Lol-Tum (LTM), Akil (AKL), and Ticum (TCM), as shown in Fig. 1. A total of 76 adult flowering plants (5-23 individuals from each population; Tables 1 and 2) were sampled. Since R. nudiflora is considered a weed, it is frequently removed by people from growing areas, rendering it impossible to obtain many individuals at some sites, and to thereby achieve an equally large sample for all populations.

Figure 1

Table 1

Table 2

To define the climatic gradient along which these R. nudiflora populations that are found, mean annual temperature and mean annual precipitation in these same 9 sites were obtained from WorldClim v1.4 (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005). This database uses historical records from 1950 to 2000 (Table 1) and provides weather surfaces from which global climate data with a spatial resolution of approximately 1 km² can be extracted. Mean temperature and precipitation were selected since these variables have been reported to influence genetic diversity and structure in ruderal plants (Hamasha et al., 2013). For each population, the surrounding vegetation type was characterized according to Duch-Gary (1988) (Table 1).

In 2007 during June and July (the months when the flowering peak typically occurs), the same nine populations were visited and 5-23 plants were selected from each for sampling. Selected plants were similar in size and floral display, free of folivory, and at a minimum distance of 2 m from each other. Three chasmogamous flowers were randomly selected from each of these plants and herkogamy was determined as the height of the lowest stigma lobe minus the height of the highest anther, measured from the apical tip of the ovary using a digital caliper (Fisher; Grass Valley, CA; accuracy ±0.02 mm/±0.001 mm). Measurements were made between 8:00 and 10:00 h.

Evaluation of the outcrossing rate in R. nudiflora requires an array of genotypes of siblings from one maternal plant (Ritland, 2002). Therefore, in June and July 2009, fruits of CH flowers were collected from 22 plants in six of the populations (3-5 fruits per maternal plant); not enough seeds were found in some populations for the outcrossing rate analysis and therefore some sites could not be adequately sampled. Four of these populations (INF, DZL, LTM, TCM) had been sampled previously to estimate genetic diversity and two more were added: Aldana (ALD) and Solorio (SOL). Seeds taken from these fruits were germinated in plastic growing trays. After the seeds had sprouted (approximately 90% germination), seedlings were transplanted to 2-L pots filled with commercial substrate. All seeds from one maternal plant were considered as a genetic family. After 6 months of growth, leaf tissue was collected per genetic family, from a total of 108 individuals and 22 families (see column 2, Table 5).

Table 3

Table 4 *

p < 0.05, ns = not significant.

Table 5

It should be noted that the nine populations selected to estimate herkogamy, genetic diversity, and demographic history adequately represent the bioclimatic gradient and along which R. nudiflora occur in the Yucatán Peninsula (i.e., three populations per each of three bioclimatic regions); this particular bioclimatic gradient in Yucatán has been described previously and was used to explain the distribution of genetic diversity and structure in other plant species in the region (Martínez-Natarén, Parra-Tabla, Ferrer-Ortega, & Calvo-Irabién, 2013 and references therein). Similarly, the six populations selected in order to evaluate the mating system represent the intermediate and two extreme values of variation in herkogamy found in this species (Marrufo, 2011). In particular, herkogamy in R. nudiflora is reported to follow a pattern of reduced anther-stigma separation in the north to greater separation in the south (Marrufo, 2011). The two selected populations therefore represent the "degree" of herkogamy along this gradient: DZL and INF in the north have reduced anther-stigma separation (0.32 ± 0.48 mm and 0.51 ± 0.5 mm, mean ± SD, respectively), the two central region populations an intermediate separation (ALD 1.46 ± 1.00 mm and SOL 1.56 ± 0.51 mm), and the two southern populations a greater separation (TCM 2.99 ± 0.51 mm and LTM 3.69 ± 1.5 mm) (Marrufo, 2011).

Total genomic DNA of adult plants and seedlings was isolated from 200 mg lyophilized leaf material by the mini-prep CTAB method (Doyle, & Doyle, 1990). Isolated DNA was purified using the Wizard^(r) DNA Clean-Up kit (Promega, Madison, WI) according to manufacturer instructions. The quality of the genomic DNA was verified in 1× TBE agarose gels and quantification of each sample was made in an Implen Nanophotometer (Thermo Fisher Scientific; Toronto, Canada).

Amplified fragment length polymorphism (AFLP) analysis was performed using a modified version of the protocol proposed by Vos et al. (1995): 250 ng genomic DNA (1.4-13.4 μL) was digested in a 20-μL reaction with 0.2 and 0.25 μL of the restriction enzymes Mse I and Eco RI respectively, 0.2 μL BSA, 2 μL NE Buffer 2 (both reagents from New England Biolabs; Ipswich, MA), and water to attain the final volume. The reaction was maintained at a constant temperature of 36 °C for 2 h. Adaptor ligation was performed in a 10-μL reaction with 1.08 μL of each of the 100 μM Eco RI and Mse I adapters, 1 μL T4 DNA ligase, 5 μL ligase buffer (both reagents from New England Biolabs), and 3.86 μL water. Ligation occurred at 16 °C for 1 h, and the ligase reaction was inactivated at 65 °C. Samples were then stored at 4 °C.

Pre-amplification was performed in 20-μL reactions using the restriction-ligation products and the primers Eco RI + A and Mse I + C. The temperature profile for amplification was 30 cycles of 94 °C for 30 s, 56 °C for 30 s, and 72 °C for 2 min. A total of 18 selective primer combinations were used to amplify DNA from adult plants and seedlings, based on which the following primer combinations were selected: Eco RI + ACC with Mse I + CTT and Eco RI + ACT with Mse I + CGT, for which a larger number of loci and a clearer band pattern were obtained. Adapters and oligonucleotides used were obtained from MWG Biotech (Huntsville, AL).

Bands of different molecular weight were identified in a transilluminator (Ultra-Violet Products; Upland, CA) using as a reference the band patterns of a marker of known molecular weight (PROMEGA^(r) Lambda DNA/Hin dIII Markers, Promega). A presence/absence (1/0) matrix was constructed for use in these analyses.

To find the scoring error rate and genotype repeatability, eight individuals were selected at random from among all populations, and their band patterns were evaluated twice. The proportion of unshared bands in these two assays averaged 20%. Since this is a high value, bands <100 bp and >1,000 bp (i.e., those with a higher probability of evidencing homoplasy) were eliminated, thus reducing the scoring error rate to 5.54% on average.

To identify outlier loci among the above-mentioned markers, a Bayesian analysis to estimate the posterior probability of selection was performed with BAYESCAN v2.01 (Foll & Gaggiotti, 2008). This program calculates a Bayes factor (BF), which indicates that there is conclusive evidence in favor of selection when BF > 2, while with BF values equal to or close to 1 the null hypothesis of neutrality cannot be rejected (Foll & Gaggiotti, 2008). None of the 104 loci analyzed in R. nudiflora showed conclusive evidence of selection, since all BF values were between 0.8 and 1.2. Therefore, all genetic diversity and structure analyses were run on the full set of 104 loci.

Overall mean proportion of polymorphic loci (as a percentage), effective number of alleles (n_e ), Nei's (1978) genetic diversity index (h ), and Shannon's index of diversity (I ) were estimated in the nine populations, using POPGENE v1.32 (Yeh & Boyle, 1997). To quantify the distribution of rare alleles, we calculated the frequency-down-weighted-marker value (DW ); a high value of this index for a given population indicates a high level of genetic diversity distinctiveness (Schönswetter, & Tribsch, 2005). Furthermore, to complement DW data, the degree of genetic divergence among populations and gene flow (M = Nm ) in each population pair were estimated with ARLEQUIN v3.5.1.2 software (Excoffier, Laval, & Schneider, 2005).

Genetic structure was estimated using a Bayesian approach implemented with HICKORY v1.1 (Holsinger, & Lewis, 2002), which permits evaluation of different models and selection of the one with the best fit (full model, f = 0, θ = 0 and free model), i.e., the one with the lowest deviance information criterion (DIC) (Holsinger, & Lewis, 2002). HICKORY v1.1 was also used to estimate the parameters G_ST-B, θ ^II and θ ^III, which are Bayesian analogs of F _ST, to enable contemporary population differentiation rather than historical differentiation to be tested (Holsinger, & Lewis, 2002). These analyses were performed using the following conditions for all four models: burn-in = 5,000, number of iterations = 100,000, thinnin = 20. Estimates were made using all sampled populations. Analysis of molecular variance (Amova) was also performed with ARLEQUIN v3.5.1.2 software (Excoffier et al., 2005); additionally, paired F _ST values and gene flow were estimated for all possible combinations.

To assess if each plant could be genetically assigned to a given population, STRUCTURE v2.3 (Earl, 2011; Pritchard, Stephens, & Donnelly, 2010) was used. Similarly, to determine the number of homogeneous groups that sampled populations can cluster into, a Bayesian search was performed using this same program with 20 replicates, considering K = 1-9 as the number of possible groups. Additionally, a neighbor-joining (NJ) tree was constructed with the matrix of Nei's (1978) genetic distance, and principal coordinates analysis (PCoA) was carried out using the genetic frequencies of each population; both were done with PAST v.2.15 software (Hammer, Harper, & Ryan, 2001).

To determine if populations were undergoing demographic or spatial expansion, the following were estimated: 1) distribution of paired differences between individuals (mismatch distribution), and 2) number of segregating sites within a population and changes in population size (Rogers, & Harpending, 1992). Populations that have had sustained rapid demographic expansion are expected to have Poisson-type distributions of paired differences while multimodal distributions are expected in populations that are in demographic equilibrium (Rogers, & Harpending, 1992). Distribution was analyzed per population by estimating the sum of squared deviations observed over that expected based on a demographic expansion model (SSD ), as well as by Harpending's raggedness index (spatial expansion model) using the program ARLEQUIN v3.01 (Excoffier et al., 2005). The standard error was estimated based on 1000 resampling replicates. The peak of mismatch distribution provides an estimate of τ , initial time of expansion. Mismatch distribution also provided θ₀ and θ₁ values, which are approximations of the initial and final effective population size, respectively (Excoffier et al., 2005). The standard error and 95% confidence interval were estimated based on 5,000 resampling replicates.

To evaluate potential correlations between neutral genetic differentiation (paired F _ST) and geographic distance, environmental factors and phenotypic divergence, several Mantel and partial Mantel tests were performed (Crispo, Bentzen, Reznicl, Kinnison, & Hendry, 2006). The partial Mantel test evaluates the correlation of matrices X and Y, controlling for similarities given in a third matrix Z (Mantel, & Valand, 1970). The following matrices were compared: (i) geographic distance yielded a spatial genetic pattern (IBD model) when paired F _ST and paired distance in kilometers were compared (mean temperature and mean precipitation were included in the partial Mantel test; Appendix, matrices A and ^B); (ii) environmental variables (mean temperature and precipitation; Appendix, matrices D and ^E) and genetic distance (IBE); geographic distance was included in the partial Mantel, and (iii) phenotypic differences (herkogamy; Appendix, matrix F) and genetic pattern; geographic distance was included in the partial Mantel (Sexton, Hangartner, & Hoffmann, 2013). To obtain matrices of environmental, phenotypic, and geographic variables, pairwise Euclidean distance between populations was estimated using PAST v2.15 (Hammer et al., 2001). The Mantel and partial Mantel tests were carried out with XLSTAT v2014.6.01 (Microsoft Excel); the p -value was estimated by using r (AB) to create a distribution from 10,000 permutations.

AFLP markers have been used to estimate outcrossing rates in numerous plant species regardless of allelic dominance (Gaiotto, Bramucci, & Grattapaglia, 1997; see references in Meudt, & Clarke, 2007). Values in the seedling presence/absence matrix were recoded to classify dominant (1) and recessive (2) genotypes in each band and individual, for use with MLRT v3.4 (Ritland, 2002). Based on the correlated-mating model proposed by Ritland (2002) assuming a mixed mating system model, the following mating system indices were obtained: single-locus outcrossing rate (t_s ), multilocus outcrossing rate (t_m ), biparental inbreeding (t_m -t_s ), multilocus correlation of outcrossed paternity (r_p ), and correlation of selfing (r_s ). The latter 2 indices are based on a sib-pair arrangement, which was used as the observation unit. In this case, r_p estimates the probability that a given pair of plants are full-sibs due to an outcrossing event, and r_s the probability that they are full-sibs due to a selfing event (Ritland, 2002). Hence, a high r_s value has been previously correlated with pollen limitation (Takebayashi, Wolf, & Delph, 2006). Confidence intervals were estimated by bootstrap resampling (1,000 iterations) of progeny within each family.

The overall mean proportion of polymorphic loci was 60.73 ± 13.88% (mean ± SD). At the population level, the percentage of polymorphic loci fluctuated between 40.24% and 78.04% (Table 2). The overall mean effective number of alleles (n_e ) was 1.245 ± 0.029 (mean ± SD). The overall mean for Nei's genetic diversity index (h ) was 0.162 ± 0.02 (mean ± SD), the highest value occurring in SAT and the lowest in AKL. For Shannon's index (I ), the overall mean was 0.261 ± 0.035 (mean ± SD), with the same two populations having the highest (SAT) and lowest (AKL) values. As regards DW , the highest values occurred in MOL, SJT, and INF (Table 2), while the lowest was found in AKL and DZL. Mean values for this index indicate that the latter two populations and SAT may have the longest divergent times while CKL, MOL, TCM, SJT, and LTM have recent divergence times (Table 2). All Nm estimates were >1.0 individuals; CKL and LTM had the lowest gene flow values (Nm = 1.585). In 22 comparisons between populations, Nm had an infinite value while in the remaining 13, Nm = 1.945-60.625 individuals (Appendix, matrix C).

Bayesian analyses indicate differences among populations (Table 3A), as evidenced by the full model (the model with the lowest DIC) in which θ ^II, θ ^III and G_ST - B were very low, albeit significantly different from zero (Table 3). Similarly, the Amova indicates genetic differences among populations (Φ_ST = 0.016, p < 0.05), which account for 1.62% of the total variation. However, within-population differences explain most of the variation (98.38%). The paired F _ST matrix showed low levels of differentiation in almost all comparisons, suggesting a high rate of gene flow in the Yucatán Peninsula, as stated above (Appendix, matrices B and ^C).

The analysis performed with STRUCTURE v2.3.3 obtained an optimum number of clusters at ΔK = 3. Also, as shown in Fig. 2, populations were not observed to be genetically homogeneous groups of individuals; in other words, individuals from the same predefined population had partial membership in multiple populations. NJ and PCoA showed no clear geographic clustering of sampled populations or of individual plants (Fig. 2). This suggests that there are individuals that have ancestral relationships with plants from populations other than the one in which they were collected, which in turn provides indirect evidence of some degree of genetic connectivity.

Figure 2

Mismatch distribution (SSD index) and Harpending's raggedness index showed that all populations have undergone demographic and spatial expansion (Table 4) except DZL (SSD = 0.202, p = 0.04). Also, the peak of the mismatch distribution ranged from 19.2 (IC 10.6-30.6) in AKL to 40.7 (IC 18.9-178.3) in DZL, although the most common value was 27, indicating that population growth is not recent. Overall, these results show a drastic expansion of all populations, each population starting with a few dozen individuals (θ₀ ranged from 0.7 to 11.0; Table 4) and now comprising thousands (θ₁ ranged from 14,456.3 to 91,862.4) as shown in Table 4. These values are approximate, but such gross estimation suggests large demographic expansion along the climatic gradient.

Results of the 3 Mantel and partial Mantel tests were as follows: (i) IBD model: a moderate and significant correlation was found between F _ST and geographic distance; when environmental variables were included in the partial correlation (r = 0.439, p = 0.009), geographic distance explains up to 19% of variance, and when these variables were removed (r = 0.416, p = 0 .012), explained up to 17% of the variance (Fig. 3); (ii) effect of environmental variable matrices: mean temperature (r = 0.010, p = 0.944) and mean precipitation (r = −0.224, p = 0.182) had no significant IBE effect on the genetic matrix both when geographic distance was included in the partial correlation and when it was removed (mean temperature r = 0.015, p = 0.896; and mean precipitation r = 0.162, p = 0.346); (iii) Phenotypic differences (herkogamy) were not correlated with genetic distance when geographic distance was included (r = 0.208, p = 0.234) and a weak, almost significant correlation was found when geographic distance was removed (r = 0.308, p = 0.085).

Figure 3

Elevated single-locus and multilocus outcrossing rates were obtained in R. nudiflora (t_m = 0.998 ± 0.009; t _s = 0.983 ± 0.012; mean ± SE) indicating that nearly 99% of progeny were produced by noninbreeding events. The values obtained for both multilocus correlation of outcrossed paternity (r_p ) and correlation of selfing (r_s ) were low, suggesting that in most cases seeds from one family do not share the same father (r_p = 0.041 ± 0.013; mean ± SE) and that pollen limitation is low in this species (r _s = 0.108 ± 0.003; mean ± SE) (Table 5).

To Mónica Medrano, Luis Abdala, Daniel Piñero, and two anonymous reviewers for valuable comments on the manuscript. We thank Sara V. Good for kindly allowing genetic analyses to be performed in her laboratory at the University of Winnipeg, and Miriam Ferrer for assistance in carrying out these analyses. This study was financially supported by Conacyt as part of a grant provided to VPT (SEP-2004-CO1-4658A/A1).