Genetic diversity statistics, including alleles (A) revealed, allelic richness (A_R) rarefied to the smallest sample size of 15 diploid individuals per population, observed heterozygosity (H_O), gene diversity within population (H_S), gene diversity in the total population (H_T), inbreeding coefficient (F_IS), and genetic differentiation among populations (F_ST) were calculated at each SSR locus using FSTAT version 2.9.3 (Goudet, 2001). Hardy–Weinberg equilibrium and genotypic disequilibrium were tested with correction for multiple comparisons. Frequencies of null alleles were evaluated with program FREENA (Chapuis and Estoup, 2007), and a refined estimation of population differentiation (F_ST) was obtained after excluding null alleles. F_ST-outlier approach implemented in ARLEQUIN 3.0 (Excoffier et al., 2005) was used to test if microsatellites were affected by selection. Population structure was inferred with microsatellites using discriminant analysis of principal component (DAPC) which do not require the assumptions about Hardy–Weinberg equilibrium and linkage disequilibrium (Jombart et al., 2010). In this analysis, 70% of the total variance of the data was selected to be expressed by the retained axes of PCA. Genetic structure was also investigated using the Bayesian clustering algorithm implemented in the program STRUCTURE version 2.3.4 (Pritchard et al., 2000). The analyses were conducted by assuming an admixture model and correlated allele frequencies among populations. Ten replicated runs were carried out for each possible number of clusters (K) being tested from 1 to 14, the length of burn-in and Markov chain Monte Carlo was set up to 100,000 and 200,000 generations, respectively. The most likely K was selected by analyzing the second order rate of change of L(K) between successive K values (Evanno et al., 2005) using STRUCTURE HARVESTER (Earl and von Holdt, 2012). Additionally, a suboptimal value of K was searched to investigate more detailed structure signal by repeat analysis without the optimal and smaller values of K (Puppo et al., 2016). Individuals were assigned probabilistically to a cluster, or jointly to two or more clusters, if their genotypes were admixed. The genetic variation was partitioned by analysis of molecular variance (AMOVA) using the program ARLEQUIN 3.0 (Excoffier et al., 2005). Chloroplast haplotypes (Supplementary Table S2) were resolved and their relationships were inferred with a TCS network (Clement et al., 2000) implemented in POPART version 1.6 beta (Leigh and Bryant, 2015).

Based on 14 sampled populations, evolutionary scenarios of K. neriifolia on the Canarian archipelago were tested using both nuclear microsatellites and chloroplast sequences with DIYABC V1.0.4.46 beta (Cornuet et al., 2008, 2010). According to haplotype distribution, eight scenarios (Figure 1) were defined by placing each island has the hypothetical ancestral population with exception of La Palma: (1) Lanzarote to west: in this scenario, the oldest populations are in the east and youngest in the west, colonization from Lanzarote to Fuerteventura, Gran Canaria, Tenerife, La Palma, and finally to EI Hierro through La Gomera; (2) EI Hierro to east: this is the reverse scenario to the aforementioned, colonization from EI Hierro to Lanzarote, in this scenario the oldest population are in the west and youngest in the east, and there is a re-colonization of Gran Canaria from Lanzarote; (3) EI Hierro-La Gomera islands as oldest population to east: this is a variant of scenario 2, where EI Hierro and La Gomera are considered as a single ancestral population, and then a colonization eastward; (4) La Gomera as oldest population: La Gomera is the origin place, colonizing toward EI Hierro and La Palma in a westward path, and eastward to Lanzarote; (5) Tenerife as oldest population: in this case, paths of colonization are similar to the previous scenario, but the ancestral population is located in Tenerife; (6) Gran Canaria as oldest population: with a westward path to Tenerife, La Gomera, La Palma and EI Hierro, and eastward path to Lanzarote through Fuerteventura; (7) Fuerteventura as oldest population: in this scenario, one eastward path is to Lanzarote, and other westward to EI Hierro; finally, (8) null scenario: all of populations independently colonized from the outside of archipelago. In all scenarios, there is an ancestral non-sampled population with effective size of NA, we assumed one population effective size for each island NA = N1 = N2 = N3 = N4 = N5 = N6 = N7 and the priors for divergence time t8 > t7 > t6 > t5 > t4 > t3 > t2 > t1 (Supplementary Table S3). For microsatellite markers, generalized stepwise mutation model was assumed for each locus, taken a gamma distribution for individual mutation rate, as no information of mutation rate is available on this species, default values was taken (1 × 10^-5 to 1 × 10^-2; Cornuet et al., 2010). For chloroplast DNA sequences, Hasegawa–Kishino–Yano model (Hasegawa et al., 1985) was selected based on Akaike Information Criterion (AIC) by using the program MEGA v7.0.16 (Kumar et al., 2015) with 10% of invariant sites, taken a default uniform distribution for mean mutation rate (per site per generation) bounded between 1 × 10^-8 and 1 × 10^-7. Summary statistics included mean number of alleles, mean genetic diversity, mean size variance, F_ST, number of haplotypes, number of segregating sites, mean of pairwise differences, and Hudson’s F_ST (Hudson et al., 1992). One million of simulations were performed for each scenario, and the 10% simulated data sets closest to observed data sets were used to estimate scenario probability and posterior distributions of parameters through a local linear regression procedure. The most likely scenario was selected by estimating their posterior probabilities using the direct estimation and logistic regression methods (Cornuet et al., 2008).

The levels of the mean total genetic diversity found in plants of the Canary Islands are higher than those estimated for other oceanic archipelago, which give a hint that repeatedly arrivals of plants in the early phase of inter-island colonization (Francisco-Ortega et al., 2000). Frequent inter-island dispersals could be true for K. neriifolia since widespread distribution of thermophilous habitats on every island makes it possible for the successful colonization of this species. Several chloroplast haplotypes distribute more than one island, such as H1 presents on EI Hierro and La Gomera, H2 on La Palma and Tenerife, H7 and H8 on Gran Canaria, Fuerteventura, and Lanzarote. These common chloroplast haplotypes shared among islands highlight seed dispersal among islands given that the chloroplast genome is transmitted maternally in most angiosperms (Petit and Vendramin, 2007).

Plants on oceanic islands often originate from self-compatible colonizers which could establish a sexually reproducing population and get seeds without specialized pollinators (Baker, 1955; Crawford et al., 2010). The close relatives of Kleinia, groundsels, are apparent self-pollination with occasional outbreeding depending on season of flowering (Haskell, 1953). There is no data available about the model of pollination and breeding system for Kleinia species, however, high levels of inbreeding in some populations (Table 1) may indicate some degree of selfing. Island populations are expected to show increased inbreeding relative to mainland populations due to smaller population size (Frankham, 1998), this may be the reason why some populations in our study deviated from Hardy–Weinberg equilibrium. The fruits of K. neriifolia have conspicuous feathery pappus that enables the seeds to be carried by the wind. Moreover, inter-island dispersals could occur since stochastic and non-standard mechanisms dominate long-distance dispersal in plants (Higgins et al., 2003). The relatively high reproductive capacity, short generation time, and efficient dispersal mechanism of the members in family Asteraceae facilitate their colonization, thus it is not surprised that this family of flowering plants have the highest number of oceanic endemic species (Crawford et al., 2010).

There are no consistent signals of demographical change observed for plant species widespread on the Canary Islands. For non-endemic species, a genetic bottleneck after colonization were revealed in widespread annual Scrophularia arguta (Valtueña et al., 2016), however, Cistus monspeliensis showed higher genetic variation among oceanic island populations than those of the continent (Fernández-Mazuecos and Vargas, 2011). A moderate signal of population expansion in the central islands (Gran Canaria, Tenerife, and La Gomera) was detected in endemic palm species P. canariensis (Saro et al., 2015). In our study, the effective population size show contrast trend in western and eastern island groups. In the western islands, there is a slight decrease in the effective population size toward areas of recent colonization. Despite much older than EI Hierro or La Palma, La Gomera show lower nuclear allelic richness and genetic diversity relative to the other islands. Thus, it is possible that Tenerife between La Gomera and Gran Canaria was necessary for successful plant colonization (Caujape-Castells, 2011), and this explain why populations on Tenerife show some levels of genetic admixture with Gran Canaria (Figure 3). In the eastern islands, the effective population size increase in Lanzarote relative to Gran Canaria and Fuerteventura. Gran Canaria possess highest number of chloroplast haplotypes and highest nuclear allelic richness relative to the other islands, which is consistent with the DIYABC results that Gran Canaria is the origin place of K. neriifolia. Population expansions within Lanzarote could be result from recurrent cycles of extinction and re-colonization (Brown and Pestano, 1998; Navascués et al., 2006). The predominance of westerlies during the Quaternary, potentially enhanced dispersal from Gran Canaria to Fuerteventura and Lanzarote (Caujape-Castells, 2011).

Kleinia neriifolia often grows together with succulent Euphorbia species and constitute the thermophilous and xerophilous vegetation on the Canary Islands. In a recent study, Tenerife-La Gomera are suggested as the origin place of the succulent Euphorbia species (sect. Aphyllis subsect. Macaronesicae) on the Canary Islands (Sun et al., 2016). Three palaeo-islands of Tenerife and La Gomera are rather old and formed a small archipelago until the early Quaternary (Trusty et al., 2005). Considering Gran Canaria as the origin place of K. neriifolia, it thus argue for an extension of the thermophilous and xerophilous vegetation from the “central” islands (Gran Canaria, Tenerife, and La Gomera) to the east and west, this fit the biogeographic prediction that central islands are the predominant diversification and dispersal centers in the Canarian archipelago (Sanmartín et al., 2008).

Exceptionally high diversification rate driven by the onset of the Mediterranean climate has been revealed in some genera of family Asteraceae on the Canarian archipelago, such as Argyranthemum and Cheirolophus (Francisco-Ortega et al., 1996; Fiz-Palacios and Valcárcel, 2013; Vitales et al., 2014). But, K. neriifolia is the only native species of Kleinia on the Canary Islands. The low diversification rate of Kleinia on the Canary Islands could be mainly due to its very young evolutionary times, and differentiation might be impeded by gene flow (Herben et al., 2005) given some levels of genetic admixture revealed in T-Ana, T-Aro, and GC-Anv population.