Unravelling plant diversification: Intraspecific genetic differentiation in hybridizing Anacyclus species in the western Mediterranean Basin
Abstract
Premise
The interfertile species Anacyclus clavatus, A. homogamos, and A. valentinus represent a plant complex coexisting in large anthropic areas of the western Mediterranean Basin with phenotypically mixed populations exhibiting a great floral variation. The goal of this study was to estimate the genetic identity of each species, to infer the role of hybridization in the observed phenotypic diversity, and to explore the effect of climate on the geographic distribution of species and genetic clusters.
Methods
We used eight nuclear microsatellites to genotype 585 individuals from 31 populations of three Anacyclus species for population genetic analyses by using clustering algorithms based on Bayesian models and ordination methods. In addition, we used ecological niche models and niche overlap analyses for both the species and genetic clusters. We used an expanded data set, including 721 individuals from 129 populations for ecological niche models of the genetic clusters.
Results
We found a clear correspondence between species and genetic clusters, except for A. clavatus that included up to three genetic clusters. We detected individuals with admixed genetic ancestry in A. clavatus and in mixed populations. Ecological niche models predicted similar distributions for species and genetic clusters. For the two specific genetic clusters of A. clavatus, ecological niche models predicted remarkably different areas.
Conclusions
Gene flow between Anacyclus species likely explains phenotypic diversity in contact areas. In addition, we suggest that introgression could be involved in the origin of one of the two A. clavatus genetic clusters, which also showed ecological differentiation.
During the diversification process, the ecological niche of a lineage (i.e., the population component of the niche corresponding to the distribution of individuals across environments within a region; Ricklefs, 2010) is expected to become increasingly occupied because of progressive differentiation and specialization to new environmental conditions. This complex phenomenon is molded by niche-based processes normally acting in concert, such as environmental filtering, biotic interactions, and trade-offs (Chase and Myers, 2011), differentially affecting populations across varying spatial and temporal scales. During a lineage's diversification, gene flow may occur between locally adapted populations of either the same species or closely related ones with different degrees of incomplete reproductive isolation (Rieseberg, 1997; Arnold, 1997, 2006; Abbott, 2017). A continuum of phenotypic variation often results, hindering the delimitation of species (Eckhart et al., 2004; Streisfeld and Kohn, 2005; Brennan et al., 2009; Wachowiak et al., 2015; Sung et al., 2018; Bresadola et al., 2019) and adding complexity to studies to disentangle the diversification process. In these cases, however, the study of phenotypic variation of reproductive characters in plants is particularly interesting because of the direct implications on their fitness (Endler, 1986; Holsinger, 2000; Fenster et al., 2004; Harder and Barrett, 2006; Armbruster et al., 2009; Broz et al., 2017). Besides affecting fitness, the suit of reproductive traits, such as flower shape, phenology, sexual and breeding system, or reward allocation will altogether shape mating patterns and thus determining the probability of gene flow between new evolving linages.
The genus Anacyclus (Anthemideae, Asteraceae) encompasses eight species of herbs, mostly growing in open anthropic environments, with their center of diversification in western Mediterranean Basin (Humphries, 1979; Vitales et al., 2018). From an evolutionary viewpoint, Anacyclus has been an object of study due to its diversity of floral phenotypes at the species (Humphries, 1981; Bello et al., 2013, 2017; Álvarez et al., 2020) and population levels (Agudo et al., 2019; Cerca et al., 2019), including various sexual systems and female flower shapes, colors, and sizes. Interspecific hybridization is likely responsible of such a high diversity of floral phenotypes because most of the Anacyclus species are interfertile (Humphries, 1981; Álvarez et al., 2020) and show some degree of spatial overlap in their distributions. Although the existence of current gene flow between these species has never been addressed, interspecific hybridization, which would be mediated by the active pollinator community visiting Anacyclus populations (Cerca et al., 2019) was supported by analyses on genome size (Agudo et al., 2019) and ribosomal sites variation (Rosato et al., 2017) of A. clavatus and A. valentinus populations. In our recent investigations including intra- and interspecific experimental crosses between A. clavatus, A. homogamos, and A. valentinus, Álvarez et al. (2020) found that parents, second-generation (F2) hybrids, and backcrosses may show similar phenotypes. This highlights the relevance that interspecific hybridization may have on the phenotypic variation in this species complex. Hence, the identification of species in sympatry based solely on morphological characters may become troublesome and probably incorrect (Manzanilla et al., 2021).
In the present study, we characterized the climatic niche, estimated genetic structure, and inferred the impact of gene flow among populations of A. clavatus, A. homogamos, and A. valentinus across the western Mediterranean Basin. We expected to find well-differentiated climatic niches and genetic clusters for each species, except in sympatric populations, which might be represented by a combination of genetic backgrounds from the co-occurring species. We implemented a combination of two well-established methodological approaches to unravel the complexity of the lineage diversification in Anacyclus. On the one hand, we estimated the genetic structure based on nuclear microsatellites previously developed for several species of the genus (Agudo et al., 2013) using Bayesian model-based clustering algorithms and complemented with ordination methods. This multispecies approach is rare in the literature because the difficulty of its implementation in cases of high genetic divergence among species (Barbará et al., 2007; Fu et al., 2016; He et al., 2017). On the other hand, based on a set of climatic factors summarizing species’ niche and spatial properties, we applied ecological niche models (ENM; Guisan et al., 2017) and niche overlap analysis (Broennimann et al., 2012) to species and genetic clusters to estimate their environmental suitability and their degree of ecological niche similarity (Hespanhol et al., 2022). The climatic variables used in ENM represent putative selective pressures, which jointly with their demographic history would partially account for the geographic distribution of Anacyclus species in western Mediterranean Basin.
MATERIALS AND METHODS
Study system
The three study species—A. clavatus, A. homogamos, and A. valentinus—mainly differ in the type of peripheral florets of the capitulum (Humphries, 1979; Álvarez, 2019). Anacyclus clavatus presents heterogamous capitula (Figure 1), with 8–15 peripheral female flowers that display a showy white ligule (0.8–1.5 cm long). In contrast, the peripheral female flowers of A. valentinus may be white or yellow with a smaller ligule (0.3 cm long) usually hidden by the involucral bracts (Figure 1). Finally, in A. homogamos, all flowers are tubular and bisexual (Figure 1). Therefore, both A. homogamos and A. valentinus show discoid capitula, whereas A. clavatus displays rayed ones. These three species occur in the western Mediterranean Basin (Figure 1), inhabiting anthropogenic open places but in ecologically different areas. For example, Anacyclus clavatus occurs in both coastal and inland areas across the region, indicating that the species can cope with a gradient of Mediterranean subclimates mainly characterized by the severity of winter conditions from mild coastal areas to cold inland areas. In contrast, A. valentinus and A. homogamos are distributed in areas with particular environmental conditions: A. valentinus mainly occupies areas along the coast or with a strong coastal influence, whereas A. homogamos is mostly restricted to the Atlas region in Morocco (Álvarez et al., 2020).
Phenotypically mixed populations (hereinafter “mixed populations”) of A. clavatus and A. valentinus occur throughout the distribution of A. valentinus. These mixed populations may be composed by individuals with clavatus-like and valentinus-like phenotypes (Figure 1), and/or individuals exhibiting intermediate phenotypes between the two species (Figure 1). In mixed populations from the Middle Atlas in Morocco, A. homogamos may also be present (Humphries, 1979; Agudo et al., 2019).
Study area and sampling
After reviewing 1562 Anacyclus specimens from B, BC, BCN, G, LISE, LISI, LISU, MA, MGC, SEV, and VAL herbaria (according to Thiers, 2016) and several field campaigns between 2010 and 2013, we eventually used 964 specimens to represent the geographic distribution of the Anacyclus species of study across western Mediterranean Basin (Figure 1) and to model their potential distributions (see below). To study the among- and within-population genetic structure and the extent of gene flow, we used 12–36 individuals from each of 31 populations representing the study species, totaling 585 individuals (Appendix S1). Ten populations were of A. clavatus, nine of A. valentinus, seven of A. homogamos, and five represented mixed populations. In three of these five mixed populations, individual phenotypes were categorized as clavatus-like, valentinus-like, or as intermediate (Appendix S1). For the two other mixed populations, we were not able to categorize each individual, but we considered the ratios of different phenotypes observed in the field. Individuals from each population were haphazardly collected and separated by at least 10 m from each other. Leaves from each individual were dried and stored in silica gel.
To cover the distribution range more densely and to increase ENM performance of the resulting genetic clusters (see below), we estimated genetic structure of a larger data set that included all individuals of the 31 populations used for the intrapopulation genetic analysis plus 1–3 individuals from other 98 populations (i.e., 69 populations of A. clavatus, 22 of A. valentinus, three of A. homogamos, and four of mixed populations), totaling 721 individuals (Appendix S2).
Microsatellite genotyping
We extracted total genomic DNA from silica-dried leaves using the DNeasy Plant Minikit (QIAGEN, Hilden, Germany). We used eight microsatellite markers (i.e., locus 9, 15, 17, 19, 20, 21, 24, and D3) previously developed for A. clavatus (Agudo et al., 2013). AllGenetics & Biology SL (A Coruña, Spain) amplified fragments using the protocols described by Agudo et al. (2013). We scored and manually checked the fragments with the software Geneious v.7.1.2 (Kearse et al., 2012). We tested for homozygote excess with micro-checker v.2.2.3 (Van Oosterhout et al., 2004) to evaluate the presence of null alleles per locus at the species and population levels. The observed frequency of homozygote classes was analyzed using Monte Carlo simulations (×1000) and a Bonferroni (Dunn–Šidák correction)-adjusted 95% confidence interval. To avoid bias, we discarded individuals with amplification success below 50% across loci, ending up with a data set of 585 individuals from 31 populations (Appendix S1). The same criterion was applied to the extended data set (Appendix S2).
Effect of homoplasy and suitability of microsatellites
Scoring errors produced by the presence of insertions and deletions (indels) in microsatellite flanking regions may produce homoplasy, which tends to increase with phylogenetic distance between species (Germain-Aubrey et al., 2016). For determining the presence of such indels, all loci in 43 individuals (25 of A. clavatus, six of A. homogamos, seven of A. valentinus and five of phenotypically mixed individuals) were sequenced by AllGenetics & Biology SL (A Coruña, Spain) using a 3730XL DNA Analyzer (Applied Biosystems, Foster City, California, USA). Each reaction contained 12.5 μL of Supreme NZYTaq Green PCR Master Mix (NZYTech, Lisbon, Portugal), 0.5 μM of each primer, 25 ng of template DNA, and PCR-grade water up to 25 μL. The thermal cycling conditions were 95°C for 5 min; 35 cycles at 95°C for 1 min, 56°C for 1 min, and 72°C for 45 s; and a final extension step at 72°C for 5 min. PCR products were verified on 1% agarose gels stained with GreenSafe Premium (NZYTech, Lisbon, Portugal). We aligned the sequences of the flanking regions and repeat motifs with Geneious using the known sequences of each locus as a reference (Agudo et al., 2013). Overall, we considered loci with a poor performance when they had low amplification rates and/or high frequencies of null alleles. To assess the effect of homoplasy and the presence of unsuitable loci, we compared the results of genetic structure analyses with and without loci with poor performance.
Genetic diversity and structure
We used the software GenAlex v.6.501 (Peakall and Smouse, 2012) to estimate percentage of polymorphic loci, number of alleles, observed (Ho) and expected heterozygosity (He), inbreeding coefficient (FIS), and pairwise fixation index (FST) for each species and population. We conducted analyses of molecular variance (AMOVA) to decompose the genetic variance among species, among populations within species and among individuals within populations. We excluded the within-individual level of variation, interpolated missing data, and used 999 permutations to test the significance of FST statistics. We also conducted Mantel tests between pairwise Euclidian geographic and genetic distances (i.e., the proportion of pairwise allelic differences among individuals) with 999 permutations to test for significance of isolation-by-distance.
We estimated genetic structure among all Anacyclus individuals with two complementary methods. First, we performed a Bayesian model-based analysis to estimate the number of genetic clusters using structure v.2.3.4 (Pritchard et al., 2000; Falush et al., 2003) under an admixture ancestry model with the default parameters. Burn-in period was set to 50,000 runs and MCMC repetitions after burn-in to 100,000. For each simulation, we ran 2–10 genetic populations (K) and 20 iterations per K. Analysis and visualization of the population structure was performed with R package pophelper v.1.1.9 (Francis, 2017). All R packages used here were run with R v4.1.1 (R Core Team 2021). We inferred the best K using the ΔK method (Earl and vonHoldt, 2012). The average matrix of multiple runs per K that was used for subsequent calculations, as well as the highest value of the symmetric similarity coefficient (H′) across runs per K were obtained with Clumpak (Kopelman et al., 2015). A posterior probability of Q ≥ 0.90 was used as a threshold to assign full ancestry to a single genetic cluster, and samples below this threshold were considered as individuals with admixed ancestry. Second, we used R package adegenet v.2.1.7 (Jombart, 2008) to perform a principal coordinate analysis (PCoA) based on synthetic variables summarizing the genetic distance matrix among individuals.
Ecological niche and spatial metrics
We estimated niche metrics with a recently described hypervolume-based approach and spatial suitability metrics with ecological niche model (ENMs) projections for current timeframe (Hespanhol et al., 2022). We only used macroclimatic variables to estimate niche and spatial suitability metrics because all Anacyclus species studied grow in similar open habitats under various human disturbances (e.g., mostly road margins). We used the bioclimatic variables available in WorldClim v.2.1 at a 1-km resolution (Fick and Hijmans, 2017) as predictors in the ENMs. As background, we randomly selected 10,000 points over the entire study area. To avoid multicollinearity, we ran a correlation analysis on the background points and eliminated one of the variables in each pair with a Pearson correlation value >0.8. The variables ultimately included in the models were annual mean temperature (BIO1), isothermality (BIO3), mean temperature of the wettest quarter (BIO8), mean temperature of the driest quarter (BIO9), precipitation seasonality (BIO15), and precipitation of the coldest quarter (BIO19).
We predicted the climatic suitability for A. clavatus, A. valentinus, and A. homogamos using presence data confirmed by our own observations during field campaigns and herbarium revisions. To avoid sampling bias (Syfert et al., 2013), we only retained points for each species that were separated by at least 1 km from each other and thus matching the resolution of the climatic data, resulting in 964 occurrences. For phenotypically mixed populations with individuals showing clavatus-like and valentinus-like phenotypes, we considered occurrences for both species.
The ENMs were based on an ensemble approach (Araújo and New, 2007) using three modeling techniques: generalized linear models (GLM; McCullagh and Nelder, 1989), gradient boosting machine (GBM; Friedman, 2001) and random forest (RF; Breiman, 2001). We performed models with R package BIOMOD v.2.0 (Thuiller et al., 2009) using the default parameters. We calibrated and evaluated all models with 70% and 30% of the data using the area under the ROC curve (AUC) and true skill statistics (TSS). For each modeling technique, we replicated the procedure 10 times with random training and evaluation data sets. To remove spurious models, we generated the ensembles using models with AUC > 0.8 and TSS > 0.7. The contribution of each model to the final ensemble model was proportional to their goodness-of-fit statistics. Climatic suitability was considered as a consensus across statistical techniques and their contribution to the ensemble was proportional to their AUC values. We converted the consensus model into a binary model (presence/absence) applying a threshold that allows a maximum 5% omission error (i.e., the percentage of the real presence predicted as absences in the model; false negative; Fielding and Bell, 1997).
We estimated the spatial overlap between species, between species and genetic clusters, and between genetic clusters for the selected final ensemble models with the Schoener's D (Schoener, 1970; Warren et al., 2008). We plotted and transformed observations of each species and genetic cluster into densities using a gridded environmental space of 100 × 100 cells. We also plotted and calculated overlaps between species pairs and between genetic cluster pairs by using binary models.
RESULTS
Homoplasy and marker performance
There were no indels in flanking regions of 51 sequences from loci D3, 17, 19, and 20. In contrast, loci 21 and 24 presented indels of 14 or 15 nucleotides in 18 and 21 sequences. We could not assess homoplasy in six sequences from locus 9 and four from locus 15 because of low sequence quality. In addition, locus 15 was not amplified in any of the A. homogamos populations and in most of A. valentinus. To assess the effect of low performance of locus 15 and indels in loci 21 and 24, we ran two analyses of genetic structure, one including all loci and the other excluding loci 15, 21, and 24.
Our results showed that the most probable number of genetic clusters was four (K = 4) when considering all loci (see below), and five (K = 5) when excluding loci 15, 21, and 24. However, just a few samples (5.5%) scattered across different A. clavatus and A. valentinus populations represented the fifth genetic cluster, which did not provide useful information to interpret our results. Furthermore, in the K = 5 scenario, half of the A. valentinus samples showed Q values < 0.90. In contrast, when all loci were included (K = 4), most of these A. valentinus samples were assigned to one of the four genetic clusters. Finally, the assignments to genetic clusters for all samples with Q ≥ 0.90 were identical when including or excluding loci 15, 21, and 24. Therefore, we used all loci in all analyses to enhance the resolution of the genetic structure.
Genetic diversity
We identified 521 different genotypes across all 585 Anacyclus individuals sampled (89.1% of different genotypes), ranging from a low of 71.7% of different genotypes in A. valentinus populations to a high of 100% in A. clavatus. No genotype sharing was observed between different species, except in mixed populations in which 1–9 genotypes were shared among several other populations of A. valentinus and A. clavatus. Within species, the pattern of genotypes shared among populations was variable. For example, 1–3 A. homogamos genotypes were shared by 2–4 populations, 1–10 A. valentinus genotypes were shared by 2–9 populations, and no genotype sharing was observed among A. clavatus populations.
Anacyclus clavatus, A. homogamos, and mixed populations exhibited high genetic diversity parameters, whereas A. valentinus populations showed the lowest values (Table 1). In short, genetic diversity was higher in mixed populations and in A. clavatus (He = 0.38–0.40), intermediate in A. homogamos (He = 0.31) and lower in A. valentinus (He = 0.18). On average, the inbreeding coefficient (FIS) varied from a low of 0.229 in A. clavatus to a high of 0.328 in mixed populations. The genetic differentiation among populations was similar in A. homogamos (FST = 0.378), A. clavatus (FST = 0.352) and A. valentinus (FST = 0.342), whereas mixed populations exhibited the lowest estimate (FST = 0.176).
| Species/group | NPOP | NIND | Na | Ho | He | FIS | FST | PL |
|---|---|---|---|---|---|---|---|---|
| A. clavatus | 10 | 225 | ||||||
| Mean | 18.46 | 3.113 | 0.297 | 0.380 | 0.229 | 0.352 | 78.75 | |
| SD | 0.85 | 0.192 | 0.032 | 0.029 | 0.147 | 0.074 | 5.61 | |
| A. homogamos | 7 | 106 | ||||||
| Mean | 9.79 | 2.625 | 0.212 | 0.309 | 0.273 | 0.378 | 67.86 | |
| SD | 0.91 | 0.281 | 0.029 | 0.037 | 0.108 | 0.147 | 3.72 | |
| A. valentinus | 9 | 159 | ||||||
| Mean | 12.81 | 1.819 | 0.147 | 0.180 | 0.261 | 0.342 | 41.67 | |
| SD | 0.89 | 0.192 | 0.026 | 0.030 | 0.139 | 0.150 | 5.10 | |
| Phenotypically mixed | 5 | 95 | ||||||
| Mean | 15.35 | 3.100 | 0.286 | 0.399 | 0.328 | 0.176 | 87.50 | |
| SD | 0.84 | 0.270 | 0.042 | 0.041 | 0.152 | 0.036 | 6.85 |
The analysis of molecular variance (AMOVA) indicated that 79% of the genetic variance was found within populations, whereas the remaining 14% and 7% of the genetic variance was found among species and among populations within species, respectively. For each species separately, the highest genetic variance was also found within populations: 71% for A. clavatus, 89% for A. homogamos, 90% for A. valentinus, and 85% for mixed populations.
Mantel tests indicated a significant correlation between geographic and genetic distances when all samples were analyzed together (r = 0.300, P = 0.001). In contrast, isolation by distance slightly fluctuated around significance (P = 0.05) in A. homogamos populations (r = 0.693, P = 0.040) and A. clavatus populations (r = 0.200, P = 0.068), and it was not significant in A. valentinus populations (r = 0.326, P = 0.129) and mixed populations (r = −0.352, P = 0.175).
Genetic structure
The Bayesian analysis of genetic structure estimated four (K = 4) as the most probable number of genetic clusters (hereafter named GC1, 2, 3, and 4) for this data set (Figure 2A), with an average symmetric similarity coefficient H′ = 0.99. Principal coordinate analysis (PCoA) supported quite well this structure made of four genetic clusters (Figure 2B), including the assignment of populations to specific clusters of each species, and depicting the genetic complexity of mixed populations (see below).
Considering the maximum mean proportions of genetic cluster membership among individuals within populations (Figure 2A; Appendix S3), we found that populations of A. valentinus and A. homogamos were very homogenous falling within their respective genetic clusters (GC3 and GC4, respectively). Anacyclus valentinus and A. homogamos populations exhibited mean membership proportions of 0.95 (9 populations; range = 0.94–0.96) and 0.90 (7 populations; range = 0.70–0.90), respectively. In contrast, populations of A. clavatus fell within up to three different genetic clusters. Two clusters specific to A. clavatus (GC1 and GC2) with mean membership proportions of 0.87 (5 populations; range = 0.74–0.96) and 0.91 (3 populations; range = 0.85–0.95), respectively, and the GC3—the one specific to A. valentinus—with a mean membership proportion of 0.63 (2 populations; range = 0.60–0.66). Finally, the mixed populations were more erratic: one population clustered in GC2—one of the specific A. clavatus group—with a mean membership proportion of 0.60, three populations fell within the other A. clavatus cluster (GC1) with a mean membership proportion of 0.81 (range = 0.73–0.95), and another population clustered with the A valentinus group (GC3) with a mean membership proportion of 0.63.
Individuals with admixed ancestry (i.e., with a low cluster assignment; Q < 0.90) were mostly present in A. clavatus (10 populations; mean proportion of individuals with Q < 0.90 per population = 0.35; range = 0.00–0.83) and in mixed populations (5 populations; mean = 0.27; range = 0.00–0.47). In contrast, few individuals with Q < 0.90 were present in A. homogamos (7 populations; mean = 0.18; 0.05–0.65) and A. valentinus (9 populations; mean = 0.07; 0.00–0.11). Interestingly, four populations of A. clavatus (populations 71, 72, 75, and 148) and one of A. homogamos (population 18) had high admixed ancestry (range Q < 0.90 = 0.59–0.83). These populations were mostly located in contact zones between A. clavatus and A. valentinus (Figure 2C).
Ecological niche and spatial metrics
The climatic suitable distribution of species predicted by the ENMs mostly included all presence data and extended to larger areas where no presence was recorded (Figure 3A–C). The genetic cluster assignments of the samples used for the intrapopulation genetic analysis and the ENMs were mostly consistent (Appendices S3 and S4), indicating that the larger data set with populations represented by a lower number of individuals was not biasing the general picture of genetic structure in this system. The climatic suitable distribution of the genetic clusters mostly coincided with the suitable distribution of the species, showing even a better fit than those of the species that they represent (Figure 4A–D). The final ensemble models across all species and genetic clusters were statistically reliable, showing AUC values >0.9.
The most important climatic variables accounting for the distribution of each species were precipitation of the coldest quarter (BIO19) and mean temperature of the driest quarter (BIO9) for A. clavatus, precipitation seasonality (BIO15) and annual mean temperature (BIO1) for A. homogamos, and precipitation of the coldest quarter (BIO19) and annual mean temperature (BIO1) for A. valentinus (Table 2). The suitable climatic distributions predicted by the ENMs for each of the species partially overlapped (Figure 3D–F). However, overlap was higher between A. clavatus and A. valentinus (Figure 3E; Schoener's D = 0.73, overlap with binary models = 0.42) than between A. clavatus and A. homogamos (Figure 3D; D = 0.60, binary overlap = 0.05) and between A. valentinus and A. homogamos (Figure 3F; D = 0.64, binary overlap = 0.06).
| Variable | A. clavatus | A. homogamos | A. valentinus | GC 1 | GC 2 | GC 3 | GC 4 |
|---|---|---|---|---|---|---|---|
| BIO1 | 0.12 ± 0.02 | 0.58 ± 0.21 | 0.35 ± 0.05 | 0.38 ± 0.19 | 0.55 ± 0.21 | 0.53 ± 0.17 | 0.64 ± 0.20 |
| BIO3 | 0.27 ± 0.03 | 0.29 ± 0.15 | 0.23 ± 0.02 | 0.34 ± 0.10 | 0.55 ± 0.20 | 0.48 ± 0.20 | 0.36 ± 0.17 |
| BIO8 | 0.10 ± 0.01 | 0.22 ± 0.10 | 0.24 ± 0.03 | 0.13 ± 0.04 | 0.24 ± 0.06 | 0.22 ± 0.05 | 0.26 ± 0.13 |
| BIO9 | 0.28 ± 0.03 | 0.37 ± 0.12 | 0.47 ± 0.05 | 0.34 ± 0.19 | 0.33 ± 0.05 | 0.42 ± 0.17 | 0.43 ± 0.16 |
| BIO15 | 0.18 ± 0.01 | 0.78 ± 0.06 | 0.12 ± 0.03 | 0.25 ± 0.12 | 0.27 ± 0.10 | 0.17 ± 0.02 | 0.80 ± 0.14 |
| BIO19 | 0.30 ± 0.03 | 0.39 ± 0.07 | 0.24 ± 0.04 | 0.26 ± 0.07 | 0.31 ± 0.15 | 0.29 ± 0.14 | 0.31 ± 0.07 |
For genetic clusters, annual mean temperature (BIO1) and isothermality (BIO3) emerged as the most important climatic variables accounting for their suitable distribution (Table 2), except for GC4, in which precipitation seasonality (BIO15) was the most important predictor (Table 2). The highest spatial overlap in suitable distribution was observed between GC2 and GC3 (Figure 4E; D = 0.79, binary overlap = 0.14), and the lowest between GC1 and GC2—the two specific clusters of A. clavatus (Figure 4E; D = 0.60, binary overlap = 0.006). The rest of the genetic cluster pairs exhibited intermediate overlaps in their suitable distributions (Figure 4E; range D = 0.70–0.76, binary overlap = 0.002–0.09).
Finally, we also explored the overlaps between species and genetic clusters. The suitable area of GC1 had a high overlap with that of A. clavatus (D = 0.77) and to a lesser extent with that of A. valentinus and A. homogamos (D = 0.68 in both cases). The suitable distribution for GC2 had the highest overlap with that of A. homogamos (D = 0.72) and the lowest with A. clavatus (D = 0.55). The GC3 had similar overlap with that of A. valentinus (D = 0.73) and with that of A. homogamos (D = 0.75). Finally, GC4 had the highest overlap with that of A. homogamos (D = 0.85) and much lower with those of the other two species (D = 0.61 for A. clavatus and 0.62 for A. valentinus).
DISCUSSION
Our study on the multispecies genetic structure for Anacyclus in the western Mediterranean Basin provided important insights into species identity and the likely role of gene flow and climatic variation for the genetic diversification in this species complex.
Species identity
In previous phylogenetic analyses of the genus Anacyclus based on nrDNA and cpDNA sequences (Oberprieler, 2004; Vitales et al., 2018), none of the three study Anayclus species were sisters, although in both cases, A. clavatus and A. valentinus were more closely related to each other than to A. homogamos. Recent evidence also supported the divergence between these three species, such as the significant differences in their genome sizes (Agudo et al., 2019) and the partial post-zygotic reproductive barriers between them (Álvarez et al., 2020). This study provided additional support to the recognition of these three Anacyclus species, but also brought to light the complexity posed by gene flow among them and the necessity of a thorough review of their diagnostic characters. In addition, the climatic variables accounting for their suitability distributions differed among species (Table 2), suggesting an ecological differentiation.
However, there was not a correspondence between genetic clusters and phenotypes in contact areas between species, suggesting the existence of gene flow between them. Gene flow is supported by the geographic distribution of A. clavatus samples assigned to GC2 and GC3, which only occur in areas where A. valentinus is present (i.e., mainly in Mediterranean coastal areas; Figure 2C). We believe that samples with a clavatus-like phenotype assigned to GC3 could represent hybrids between A. clavatus and A. valentinus. In addition, samples assigned to GC2—one of the specific genetic clusters of A. clavatus—form mixed populations, showed clavatus-like, valentinus-like, and intermediate phenotypes, suggesting extensive gene flow in these populations. On the contrary, in areas where A. valentinus was absent, all A. clavatus populations fell within GC1.
Although there is no data on the geographic–morphologic relationship in A. clavatus, there exists a geographic pattern of genome size variation (Agudo et al., 2019) matching the genetic structure of A. clavatus found in this study. In particular, A. clavatus populations from areas where A. valentinus is present had significantly smaller genomes compared with those in A. clavatus populations from other areas. The presence of two clearly differentiated genetic clusters in A. clavatus with a different genome size and occupying different geographic areas, suggests the existence of two distinct lineages within A. clavatus. The presence of clearly divergent intraspecific genetic clusters with a marked geographic pattern is a common result in regional-scale population genetic studies (Brennan et al., 2014; Mandák et al., 2016; Castilla et al., 2020), illustrating the evolutionary and demographic imprint on plants’ genetic structure across their distributions.
In contrast, A. valentinus and A. homogamos had no intraspecific genetic structure, and each species had a clearer correspondence with its own genetic cluster. Although it is difficult to determine the causes of such a relationship among species and genetic clusters in the Anacyclus complex, the association among distribution, genetic cluster membership and genetic diversity are more complicated than previously anticipated. For example, A. valentinus had the lowest genetic diversity among all studied species, despite occurring across a large area along the Iberian and North African Mediterranean coast. In A. valentinus, inbreeding and isolation seem unlikely because it is self-incompatible (Álvarez et al., 2020) and mostly occurs along roadsides. Moreover, the inbreeding coefficient (FIS) for A. valentinus did not indicate an excess of homozygosis, and the AMOVA showed a structure congruent with connected populations. In contrast, the genetic diversity of A. homogamos was similar to that of A. clavatus and to the group of phenotypically mixed populations, despite its more restricted distribution.
Gene flow, introgression, and phenotype variation
Our study provided the first genetic evidence of gene flow among Anacyclus species (Humphries, 1979, 1981; Rosato et al., 2017; Agudo et al., 2019; Álvarez, 2019; Vitales et al., 2020). As in other systems (e.g., Emanuelli et al., 2013; Ortego et al., 2017; Castilla et al., 2020), the genetic admixture within populations, including individuals with varying admixed genetic compositions, indicates the existence of current gene flow among different genetic clusters and species. Genetic admixture was very clear in the group of phenotypically mixed populations, where hybrid individuals with admixed ancestry were frequent. These hybrids showed intermediate phenotypes, but they also presented clavatus-like phenotypes and, to a lesser extent, valentinus-like phenotypes, which might be the result of backcrossing or introgression between A. clavatus and A. valentinus.
Gene flow also seems to occur in populations from areas off contact zones. This was the case of population 148 of A. clavatus (Figure 2B, C), in which some individuals showed admixed genetic compositions. This population is located outside of what it can be considered as contact areas between species (Figures 1, 2C). Dispersal from those areas seems unlikely because seed dispersal in all annual Anacyclus species mostly occurs by means of rain (i.e., ombrohydrochory; Bastida et al., 2010; Torices et al., 2013). However, since Anacyclus species grow mainly as roadside weeds (Humphries, 1979; Álvarez, 2019), we cannot rule out occasional dispersal events facilitated by human activity. Alternatively, we might have underestimated the size of the contact areas between species. In any case, these results stress the need for further studies using phylogenomic approaches and more intense samplings, particularly within populations, for the delimitation of contact areas between species.
Species and genetic cluster distribution
Niche modeling for within-species genetic clusters is increasingly frequent in the literature (Piñeiro et al., 2007; Gotelli and Stanton-Geddes, 2015; Marcer et al., 2016; Milanesi et al., 2018; Martínez-Minaya et al., 2019). Here, ENMs largely predicted similar distributions for species and genetic clusters, although the climatic factors driving such distributions were different for species and their genetic clusters (Table 2). On top of methodological limitations to compare comprehensively distributions between species and genetic clusters (e.g., sample sizes for species and genetic clusters were very different), other factors may account for some discrepancies between observed and predicted distributions. For example, in the wide area across southwestern Iberia predicted for A. clavatus, where we know that the species is absent, the competition with other dominant congeneric species, such A. radiatus (Álvarez, 2020), might explain such inconsistency.
As expected, A. clavatus had the largest geographic area predicted by the environmental niche models. The fact that the predicted areas for the two genetic clusters represented by this species (GC1 and GC2) were mutually exclusive (Figure 4A, B) partly explains the wide climatic optimum for A. clavatus. The lack of geographic barriers between the predicted areas for the two A. clavatus genetic clusters suggests that climate might be a selective pressure explaining their exclusionary distributions. The analysis of niche overlap in the environmental space also supports this hypothesis, as the overlap between GC1 and GC2, was the lowest of all comparisons between genetic clusters. In contrast, the optimal area predicted for GC2 almost completely coincided with part of that of GC3—the genetic cluster corresponding to A. valentinus. Finally, the genetic cluster represented mainly by A. homogamos (GC4) showed the most exclusive predicted area (western Morocco and southwestern Iberia) whose climatic optimum exhibited low overlap with that of the rest of genetic clusters.
Recent studies on the genetic structure and genomic differentiation of the annual mustard Arabidopsis thaliana across the same region (Iberia and Morocco) showed some similarities with the results presented here for the Anacyclus complex that are worth mentioning. For example, A. thaliana in Morocco is made of a specific genetic cluster with a strong presence in the Atlas Mountains (Brennan et al., 2014; Durvasula et al., 2017). In the case of Anacyclus, this specific cluster corresponds to A. homogamos. Furthermore, there is a strong differentiation between southern and northern A. thaliana populations in Morocco (Brennan et al., 2014). In the case of Anacyclus, such a differentiation corresponds to the predominant occurrence of A. homogamos and A. valentinus in southern and northern Morocco, respectively. Finally, Iberian A. thaliana is made of four genetic clusters, one of them closely related to the Moroccan cluster (Brennan et al., 2014; Durvasula et al., 2017). In the case of Anacyclus, the predominance of A. clavatus in Iberia, followed by the presence of A. valentinus and traces of A. homogamos, also indicates a shared history between Moroccan and Iberian Anacyclus. Overall, we hypothesize that the shared geologic (e.g., the isolation of the two continents after the Messinian Crisis about 5.5 Ma) and climatic history of the region (e.g., repeated pluvial periods shifting from arid to moist conditions) might have shaped the intra- and interspecific genetic structure of multiple organisms in a similar manner.
We conclude by stressing the value of the multispecies approach and the simultaneous modeling of species and genetic clusters to provide a snapshot on the evolutionary dynamics of complex plant groups under a diversification process, such as the Anacyclus complex in western Mediterranean Basin. Although the origin of species and genetic clusters is beyond the scope of this study, our results suggest that introgression in contact zones might have played an important role in shaping genetic structure and possibly species delimitation. Further research should focus on the origin of these intraspecific genetic clusters with the incorporation of whole-genome sequence data for the genomic characterization of admixed linages. In addition, we need to increase sampling efforts for each Anacyclus species of the complex with a particular emphasis on the within-population level in contact zones harboring phenotypically mixed individuals.
AUTHOR CONTRIBUTIONS
Conceptualization: I.A. and R.T.; data curation: A.B.A., A.M., I.A., R.G.M., and F.X.P.; formal analysis: A.B.A., A.M., I.A., R.G.M., and F.X.P.; funding acquisition: A.B.A., A.M., I.A., R.G.M., R.T., and F.X.P.; investigation: A.B.A., I.A., R.T., and F.X.P.; methodology: A.M., I.A., R.G.M., R.T., and F.X.P.; project administration: I.A. and R.T.; resources: A.M., I.A., R.G.M., R.T., and F.X.P.; supervision: I.A., R.T., and F.X.P.; visualization: A.M., I.A., and R.G.M.; writing original draft: A.B.A. and I.A.; writing review and editing: A.M., I.A, R.G.M., R.T., and F.X.P. All authors read and approved the final version of the manuscript.
ACKNOWLEDGMENTS
We are grateful to Joel Calvo, Raúl Gonzalo, Leopoldo Medina, Alejandro Quintanar, María Santos, and Salvatore Tomasello for collection of some specimens. We thank Alberto Herrero for technical assistance. Thanks also to the two anonymous reviewers and to the Associate Editor for their suggestions and thorough work on improving this article. The European Regional Development Fund and the Spanish Ministry of Science and Innovation (CGL2010-18039) funded this study to I. Álvarez. A.B.A. and R.T. received support from grants of the Spanish Ministry of Economy and Competitiveness (BES-2011-048197 and BVA 2010-0375). A.M. and F.X.P. were supported by PID2019-104135GB-I00 from the Agencia Estatal de Investigación (AEI) of Spain and the European Regional Development Fund (FEDER, UE). The European Regional Development Fund and the Spanish Ministry of Science and Innovation supported R.G.M. through the NexTdive project (PID2021-124187NB-I00).
Open Research
DATA AVAILABILITY STATEMENT
Data sets for this study can be found in the Supporting Information section.
