Geographical distance and barriers explain population genetic patterns in an endangered island perennial

Oceanic islands are of particular importance for the conservation of global diversity. Rare island plants with small population sizes and unique genetic patterns resulting from prolonged evolutionary isolation are usually extremely fragile. Therefore, research dedicated to conservation of these plants should consider threat factors, population structure and genetic diversity to enable the development of scientifically supported management programs. For Lactuca watsoniana, Azorean lettuce, we performed a comprehensive population genetic study, unravelling genetic diversity patterns intrinsically related to geographical distances and barriers. Our study showed that genetic diversity levels in island plants can be relatively high even for rare plants.


Introduction
Oceanic islands have been important study systems for ecologists, evolutionary and conservation biologists and are widely recognized as natural laboratories, for studying evolution due to their discrete geographical nature and diversity of species and habitats (Emerson 2002). Oceanic islands generally have lower overall species numbers per unit area (Whittaker and Fern andez-Palacios 2007) but show higher percentages of endemism than mainland areas (Kier et al. 2009).
Geological, geographical and ecological conditions, such as island age and area, geographical isolation by oceanic barriers, climatic stability, environmental heterogeneity, high habitat diversity and the absence of competitors, have been pointed out as key determinants of the speciation patterns observed within volcanic archipelagos (Gillespie and Baldwin 2009;Garc ıa-Verdugo et al. 2014;Mairal et al. 2015).
Biodiversity of remote islands has arisen through evolution and adaptation of the few initial colonists (Gillespie 2007), with high levels of endemism displaying unique genetic patterns in comparison with continental relatives (Crawford et al. 1987;Stuessy and Ono 1998;Emerson 2002;Fernandez-Palacios 2007, Bramwell andCaujapé-Castells 2011;Silva et al. 2015;Takayama et al. 2015). Differentiation of island populations is further enhanced by intrinsic traits, such as reduced dispersal capabilities, evolutionary innovations, size and reproductive changes (Price and Wagner 2004;Kaiser-Bunbury et al. 2010;Hutsemékers et al. 2011) which, in addition to ecological opportunity and chance, play a key role in successful island colonization, and often result in adaptive radiation (Gillespie 2007).
Several classical studies have been directed to estimate the genetic diversity of island endemic plants, including the Hawaiian silversword alliance (Baldwin 2009), the Juan Fernandez Islands groups (Crawford et al. 2001) and several Canary Island taxa (Francisco-Ortega et al. 2000;Caujapé-Castells et al. 2010). Recent studies from Macaronesia Garcia-Verdugo et al. 2015) rejected the hypothesis that island populations always have a low genetic diversity due to bottleneck effects, small population size and adaptation to specific ecological conditions (Bouzat 2010;L opez-Pujol et al. 2013;Meloni et al. 2015). Similar results had already been obtained in other parts of the world (e.g., Abies nebrodensis- Conte et al. 2004;Nothofagus alessandrii-Torres-D ıaz et al. 2007; Cedrus brevifolia -Eliades et al. 2011). These high levels of genetic diversity may have been the result of genetic drift (Dias et al. 2014) or might have resulted from changes in ploidy levels (Crawford et al. 2009. In the Azores, the endemic plant species studied so far showed a considerable range of population genetic patterns: (i) high genetic diversity but low population differentiation-Picconia azorica ); (ii) relatively low genetic diversity and low population differentiation-Prunus azorica and (iii) high genetic diversity (with the exception of very small populations) and high level of differentiation-Juniperus brevifolia . Some of this supports the hypothesis of a possible Linnean shortfall in the Azorean flora (Schaefer et al. 2011;Moura et al. 2015a), with previously overlooked diversity in several taxa (Leontodon, Moura et al. 2015b;Platanthera, Bateman et al. 2013;Viburnum, Moura et al. 2013).
Among the endemic Asteraceae, the genera Leontodon and Tolpis have already been studied and showed high population genetic diversity and a complex genetic structure, with clear geographical-linked patterns (Dias et al. 2014;Moura et al. 2015b;Silva et al. 2016).
The Azores lettuce (alfacinha), L. watsoniana, is a perennial herb, endemic to the Azores, today restricted to four of the nine islands of the archipelago (Faial, Pico, São Miguel and Terceira). It is probably extinct in São Jorge Island, from where two specimens exist in the Lisbon University herbarium (LISU), collected between the valleys of 'Ribeira do Salto' and 'Ribeira de S. João'. Lactuca watsoniana is today restricted to the steep slopes of craters, ravines, and temperate juniper rain forest, between 600 and 800 m above sea level (Schaefer 2005;Silva et al. 2009;ED, pers. obs.).
Estimates for its total population size range from 500 to 2000 individuals (Schaefer 2005;Silva et al. 2009Silva et al. , 2011, but recent field observations indicate lower numbers (probably fewer than 500 individuals). Like many other endemic plants from oceanic islands (Francisco-Ortega et al. 2000;Caujapé-Castells et al. 2010) it is considered a priority species for conservation and was listed as endangered [EN B2ab(i,ii,iii); C2a(i)] on the IUCN Red List 2013. Furthermore, it was included in Habitats Directive as a Priority species (Annex B-II) and also in Bern Convention (Appendix 1, Annex 1).
The species is threatened by habitat loss and degradation resulting from changes in land use, namely expansion of pastureland, invasion by exotic species, introduced herbivores and disturbance of sensitive areas by tourists and locals ). Lactuca watsoniana propagules are dispersed by wind and water (Schaefer 2003;Silva et al. 2009) but it is unknown, how efficient these strategies are under the specific conditions in the Azores archipelago with predominantly westerly winds and individual islands separated by distances of 6 km (Faial to Pico) to 600 km (Santa Maria to Corvo) of open ocean.
As consequence, the remaining often small and geographically isolated populations on the different islands are likely to have low genetic diversity and could suffer from inbreeding depression (Lowe et al. 2005).
Besides the scientific interest related to the study of evolutionary processes in islands and particularly in the Azores, where many scientific gaps still exist (Carine and Schaefer 2010;Schaefer et al. 2011;Moura et al. 2015b, c), practical issues are also involved when it comes to the conservation of endangered endemic plants. Recently, Silva et al. (2015) proposed that population genetic studies should be included in more holistic approaches to research devoted to rare island plants, since different views exist about crucial aspects such as propagule provenance (Edmands 2007;Silva et al. 2011;Weeks et al. 2011;Dias et al. 2014;Silva et al. 2016).
In the context of a holistic view, and considering that a previous germination study was already conducted for L. watsoniana, we additionally performed a comprehensive population genetic study of L. watsoniana populations using newly developed microsatellite (Simple Sequence Repeats, SSR) markers. Based on the relative depauperation of the extant populations, we expected to find relatively low levels of genetic diversity.
Our objectives were (1) to determine the population genetic structure of L. watsoniana; (2) to estimate intra and inter-population genetic variation patterns; (3) to identify populations with low genetic variability and inbreeding; and (4) to verify the occurrence of other putative operational taxonomic units.
Beyond the conservation of L. watsoniana, we also aim to use the results of the present study to better understand the evolution of island plants in the Azores as well as in other relatively remote and geologically young island systems.

Study site
The Azores Archipelago includes nine volcanic islands, located in the NE of the Atlantic Ocean, between 36 and 43 N and 25 and 31 W, extending for more than 600 km and lying in a northwest-southeast direction. The closest mainland is the Iberian Peninsula, approximately 1363 km East, Newfoundland is 2272 km Northwest, Bermuda 3455 km southwest and Madeira 340 km southeast. There are nine major Azorean islands in three main groups: Flores and Corvo, to the west; Graciosa, Terceira, São Jorge, Pico and Faial in the centre; and São Miguel, Santa Maria and the Formigas Reef to the east. The archipelago surface is c. 2334 km 2 . However, the islands reveal very uneven dimensions: the larger, São Miguel (745.8 km 2 ), Pico (448.4 km 2 ) and Terceira (403.4 km 2 ), represent 70 % of the total surface; São Jorge (245.9 km 2 ), Faial (173.8 km 2 ) and Flores (141.6 km 2 ) have an intermediate size; Santa Maria (97.1 km 2 ), Graciosa (61.2 km 2 ) and Corvo (17.2 km 2 ) are the smallest of the archipelago. Santa Maria is the oldest island of the archipelago (6.3 M years) and Pico being the youngest island of the Azorean Archipelago (0.27 M years; Avila et al. 2016). The Azorean climate is temperate oceanic with a mean annual temperature of 17 C at sea level, low thermal amplitude, high mean relative humidity, persistent wind and rainfall ranging from 800 to 3000 mm/m 2 , increasing with altitude (Azevedo 1996).
The vascular plant flora is currently thought to comprise c. 1110 taxa, including 73 endemic taxa (Silva et al. 2010). However, these numbers likely underestimate the true diversity, since recent molecular studies have repeatedly revealed new endemic taxa (Schaefer and Schönfelder 2009;Bateman et al. 2013;Moura et al. 2015b, c;Schaefer 2015). Many species introductions and land use changes led to the replacement of natural plant communities, with more than 60 % of the surface today covered by pasture land (Schaefer 2003;Lourenço et al. 2011;Costa et al. 2012;Marcelino et al. 2013).

Plant material and sampling
A total of 13 different populations and 135 individuals of L. watsoniana were sampled, along with a subsample of five individuals of Lactuca palmensis, one or two individuals from selected North American species and one individual of L. sativa (Table 1). All the places from which L. watsoniana populations have been mentioned in historic or recent records, or which were known to local botanists or to the Environmental Services were visited, but only 13 could be confirmed. We searched without success for the only population known from São Jorge and for one population in São Miguel (Lagoa do Fogo, see Table 1).
In 2012 and 2013, trips to Terceira and Pico islands were carried out in order to complement the samples of L. watsoniana already available at the DNA bank collection of the AZB herbarium (Biology Department, Azores University). Samples from every single individual were taken in populations with up to 30 plants and in larger populations, 20 samples were collected in populations. Lactuca palmensis from La Palma Island (Canary Islands) and North American species-L. biennis, L. canadensis, L. floridana, L. graminifolia, L. hirsuta, L. ludoviciana and L. tenerrima-were obtained from herbarium specimens of the United States National Herbarium, Smithsonian Institute; L. tenerrima samples were also obtained from specimens at the BM herbarium of The Natural History Museum, London; and a L. sativa sample was obtained in the local market. Depending on leaf size, we collected one or two leaves per individual and immediately stored them in a plastic bag with silica gel. After drying, the leaves were vacuum sealed in plastic bags and stored in folders. The locations of all populations were geo-referenced and mapped using ArcMapV R 10.2.2 ( Fig. 1). DNA extraction and microsatellites development DNA was extracted from dry leaves using a modified Doyle and Dickson CTAB protocol (Doyle and Doyle 1987). Due to the difficulties encountered in obtaining high quality DNA, modifications proposed by Borges et al. (2009). DNA quality and quantity were measured using a Nanodrop 2000 (Thermo Fisher Scientific) spectrophotometer. Samples were conserved at À20 C until use. Total DNA from fresh leaves of two individuals of L. watsoniana was sent to the Savannah River Ecology Laboratory (University of Georgia) where an Illumina paired-end (IPE) shotgun library was prepared for microsatellite sequencing (Castoe et al. 2012). The resulting sequences were analysed to identify microsatellite repeats for which primers were designed with PAL_FINDER_v0.02.03.
Out of the 48 sequences of primer pairs provided by the Savannah River Ecology Lab we selected 24 primer pairs with expected PCR products ranging between 100 and 400 bp to allow multiloading of PCR products (Oetting et al. 1995). All the primer pairs (with the tag sequence included) were selected on criteria of noncomplementarities within and between primers, low secondary structures and 3 0 -end instability (Rychlik 1995).
One primer from each pair was extended on the 5 0 -end with an engineered sequence (M13R tag 5 0 -GGAAACAGCTATGACCAT-3 0 ) to enable the use of a third primer identical to the M13R, which allows for an inexpensive fluorescent labelling of the PCR product  Table 1. Geographical distribution Lactuca watsoniana populations sampled in the Azores archipelago, and different herbarium specimens included in this study (lactuca palmensis and several Lactuca species from North America). A total of 150 individuals were sampled; (N), total number of individuals per island/location, (n) number of individuals per population. * Population excluded from the analyses (less than 6 individuals).

Microsatellite selection
In the first phase of the test, all 24 primer pairs were tested on four samples of L. watsoniana using a final volume of 25 ll, including 25 ng of DNA, 1Â NH 4 Buffer, 1.5 mM MgCl 2 , 0.4 lM of FAM-M13R, 0.08 lM of tagged primer, 0.36 lM of unlabelled tag primer, 200 lM of dNTPs, 1U of Polymerase (Immolase, Bioline) and using a Biometra TGradient thermocycler. Touchdown thermal cycling programmes (Don et al. 1991) encompassing a 10 C span of annealing temperatures, ranging between 64 and 54 C, were used for all loci. The PCR program consisted of the following steps: 95 C for 7 min (hot start); 95 C for 3 min; 20 cycles at 95 C for 30 s, the highest annealing temperature of 65 C (decreased by 0.5 C per cycle) for 30 s, and 72 C for 30 s; 20 cycles of 95 C for 30 s, 55 C for 30 s and 70 C for 30 s; and finally, 72 C for 10 min for the final extension of the PCR products. Five microliters of PCR products were run on a 3.5 % agarose gel, stained with SafeView TM Nucleic Acid Stain (abm) and visualized under UV light to check for amplification, polymorphism and scorability of the bands. Eleven primer pairs exhibited scorable amplified products of the expected length range and with at least two alleles. In the second phase, the variability of the 11 polymorphic loci was assessed in 13 samples, one from each population. After analysis of the quality of the PCR products obtained with the universal primer M13R, eight primer pairs with acceptable to high scorability were selected to run the complete study (Table 2).

Full-scale genotyping
After optimization, all samples were amplified following Dias et al. (2014). The M13R was labelled either with PET, FAM, NED or VIC. The Taq polymerase used with all the markers was Immolase (Bioline). Amplification products were diluted, multiloaded, run on an ABI-3130xl Genetic Analyser and sized with LIZ500 size standard. The genotypes obtained were scored using the software Geneious R8 (http://www.geneious.com; Kearse et al. 2012). The population matrix is available at DEMIURGE (http://www. demiurge-project.org/) with digest code D-CPMIC-101.

Data analyses
SSR preliminary analyses showed a pattern compatible with tetraploidy since all individuals displayed four alleles per locus. Cytological analysis of seedling root tip material later confirmed the tetraploid nature of L. watsoniana (manuscript in prep). This information was used to define the statistical analysis methods, since polyploid organisms raise analytical limitations, namely, problems associated with inferring allele frequencies and assumptions regarding inheritance (Obbard et al. 2006;Dufresne et al. 2014).
Genetic diversity. Considering the tetraploid nature of L. watsoniana, we used specific software to generate the genetic dataset. Therefore, genetic diversity for each population was calculated using a codified co-dominant matrix with SPAGeDi (Hardy and Vekemans 2002), as a multilocus average for the eight loci, including the number of alleles (NA), the allelic richness (AR), Nei's gene diversity (1978) (He), global inbreeding coefficient (F Is ), computed for all loci and individual inbreeding coefficient (F i ), computed as kinship coefficient between homologous gene copies within individuals. The total genetic diversity (H T ) was determined using the unbiased estimate method of Nei (1987). The percentage of private alleles was calculated using Microsoft Excel V R based on SPAGeDi allelic frequency data.
Population structure. Taking into consideration the ploidy level, we first analysed the population structure with a Principal Coordinate Analysis (PCoA), based on the matrix of Bruvo distances between individuals, using the R package PolySat (Clark and Jasieniuk 2011). Then we used a Bayesian approach estimate the number of genetic clusters excluding populations with fewer than six plants (PIRB, SMCC, TEGA and TENR) which resulted in a final number of 129 individuals in the analyses. The data were exported from PolySat to a dominant matrix (presence/absence of alleles), which was run with the program STRUCTURE version 2.3.4 (Pritchard et al. 2000;Kumar et al. 2009). We followed the admixture model along with the assumption of correlated allele frequencies between groups (Falush et al. 2003), with 500 000 MCMC repetitions, exclusion of a burn-in period of 50 000 and 10 replicates for each K value ranging from 1 to 10. Estimation of best K value was conducted with STRUCTURE Harvester (Earl and von Holdt 2012) following the approach of Evanno et al. (2005). The optimal K repetitions were permuted in Clumpp version 1.1.2 (Jakobsson and Rosenberg 2007) using the Greedy algorithm and results were graphically represented using Distruct version 1.1 (Rosenberg 2004 Geographical barriers and isolation by distance. To determine the order of barriers to gene flow possibly occurring within the archipelago, the Nei's genetic distance matrix corresponding to the Azorean populations was analysed with the Monmonier algorithm, using BARRIER version 2.2 (Manni et al. 2004) and allowing a maximum number of nine barriers (the total number of populations analysed after excluding four populations with fewer than six individuals). Nei's genetic distance (Nei 1972(Nei , 1978 was calculated using a dominant matrix with GenAlEx 6.5 (Peakall and Smouse 2012).
We followed the approach of Meirmans (2012) to discern between a Hierarchical Island model and a Stepping Stone model using standard, stratified and partial Mantel tests using the package VEGAN (Oksanen et al. 2009) in R (http://www.r-project.org/). We also modelled Nei's genetic distance as a function of geographical distance and of the presence of geographical barriers, using beta regression with the BETAREG package in R (Cribari-Neto and Zeileis 2010). We compared the model based on geographical distance alone with models including an increasing number of geographical barriers. We defined the geographical barriers in two ways: (i) considering the existence of barriers between the different clusters identified with STRUCTURE; and (ii) considering the existence of barriers between populations as determined with BARRIER. Model fit was compared using AIC and pseudo-R 2 .

Genetic diversity
The eight developed primers showed acceptable to high scorability results, with exception for the L. sativa sample (all loci) and L. biennis (loci Lw-01.08037 and Lw-01.08435). A total of 129 alleles were found for L. watsoniana, with a total genetic diversity (H T ) of 0.85, an overall excess of heterozygotes (Multilocus F is ¼ À0.1912, P < 0.001; Average Multilocus F i values ¼ À0.074, P < 0.001), and with a total multilocus average proportion of private alleles of 26.5 %. The total number of alleles per locus varied from 10 to 31 ( Table 2).
The 'Lagoa do Can ario' population of São Miguel Island displayed the highest percentage of private alleles (7.0 %) and the highest multilocus number of alleles (9), while the 'Miradouro do Can ario' population in São Miguel Island and 'Subida da Montanha' in Pico Island did not show any private alleles (Table 3). The mean allelic richness ranged from 4.58 alleles in the 'Subida da Montanha' population up to 6.18 alleles in the 'Lagoa do Can ario' population. The populations showed similar values for gene diversity (0.7 and 0.8, Table 3).

Population genetic structure
The initial analysis of 148 samples (all except L. sativa e L. biennis samples) from 13 populations, based on eight microsatellite loci, revealed the existence of a close relationship between the North American species and L. watsoniana populations and a clear separation from L. palmensis, as shown by PCoA (Fig. 2). Lactuca watsoniana individuals were also separated according to their geographical location, but with one cluster combining the populations of the neighbouring islands Faial and Pico (Fig. 2). A Bayesian analyses, showed a peak of DK for K ¼ 5. Taking into consideration, the variation across runs, and the partitioning of individuals, we assumed K ¼ 5 as the most likely value. The results obtained with STRUCTURE (Fig. 3) show a non-admixture pattern with the definition of five genetic clusters corresponding to each of the islands of origin (Faial, Terceira and São Miguel), and to the separation of the four Pico Island populations into two genetic clusters (PIAP and PICA, more to the west versus PIGM and PISM, more to the east). This result is in accordance with those obtained with PolySat (Fig. 2) and GenAlEx 6.5 (data not shown).  Table 3. Genetic diversity parameters based on eight SSR loci within nine populations of Lactuca watsoniana, with multilocus average. Abbreviations: number of samples (N), number of alleles (NA), allelic richness (AR), gene diversity (He) and inbreeding coefficient (F i ), ignificance levels (Pv: *P < 0.01; **P < 0.05), and average percentage of private alleles (Pa).

Island
Populations The AMOVA analyses applied to the groups defined above showed that the largest proportion of genetic variation was found within populations (55.0 %) but with relatively high levels of genetic variation both among clusters (27.3 %) and (17.68 %) among populations within clusters (Table 4). Global F st value calculated (0.45) denoted a considerable overall differentiation among populations, with F st values ranging from 0.13 between São Miguel populations (SMLC and SMMC) up to 0.99 between PICA and TERC populations (Table 5). Gene flow values calculated between populations indicated a relatively high number of migrants between some of the geographically close populations within São Miguel and Pico islands (Table 5), but were much lower for the remaining cases.

Geographical barriers and isolation by distance
The first-order barrier to gene flow was computed between São Miguel Island and the islands of the Central Group, while the last barrier was located between the two populations of São Miguel (Fig. 4).
Regarding isolation by distance (IBD) models, L. watsoniana seems to correspond to an intermediate case between hierarchical island and stepping stone models, with no clear discrimination based on the results obtained from standard, stratified and partial Mantel tests (Table 6). Globally, the Mantel test indicated that there was a significant correlation between genetic distance and geographical distance (r ¼ 0.77, R 2 ¼ 0.66, Fig. 5). However, model fit increased from the model including geographical distance alone (AIC ¼ À140.40, pseudo-R 2 ¼ 0.58) to the models including an increasing number of geographical barriers: (i) considering the existence of barriers between different populations clusters as determined with STRUCTURE (best model including all the barriers, AIC ¼ À168.5, pseudo-R 2 ¼ 0.90); (ii) considering the existence of barriers between populations as determined with BARRIER (best model including the first to the third order barriers, AIC ¼ À161.5, pseudo-R 2 ¼ 0.84). The graphical representation of the best fitting model shows how the inclusion of barriers increases the fit of the geographical model (Fig. 5). Also, on average, Pico populations seem to be genetically closer to all the other populations (Fig. 5).

Implications of tetraploidy
The SSR profile obtained in this study supported that L. watsoniana is tetraploid. To the best of our knowledge, only two of the nine endemic Azorean Asteraceae are polyploid, namely, Pericallis malvifolia (Jeffrey 1992;Jones et al. 2014) and L. watsoniana while seven are diploids: Bellis azorica, Leontodon filii, L. hochstetteri, L.rigens, Solidago azorica, Tolpis azorica and T. succulenta (Lack 1981;Fiz et al. 2002). In the genus Lactuca, Fer akov a (1977) distinguished three main groups based on chromosome number: the first group with n ¼ 8 includes perennial species from Europe and the Himalayas; the second group with n ¼ 9 comprises the majority of the taxa from Europe, Mediterranean, Middle East, Africa and India; and the third group includes North American species with n ¼ 17, considered to be of amphidiploid origin (Dole zalov a et al. 2002). The ploidy level of L. watsoniana was previously unknown (Fer akov a 1977;Lack 1981). Our results suggested by the SSR pattern observed, point to a closer phylogenetic relationship with the North American taxa.
In the Canary Islands, species-rich lineages like the Asteraceae originated from diploid colonizers (Suda et al. 2003), while in other archipelagos (Galapagos Islands, Hawaiian Islands, Robinson Crusoe Islands and St. Helena) the endemic diversity apparently originated  Table 1). Red ellipse limits the individuals from Sao Miguel, individuals from Terceira are limited with a blue ellipse and L. palmensis in the lower right part of the chart are limited by a green ellipse. from polyploid colonists (Harbaugh 2008;Crawford et al. 2009). In the Azores Islands, ploidy levels for most endemic colonizers have not been studied yet, although many taxa are now diploid (Bateman et al. 2013;Dias et al. 2014). However, tetraploids have been recently found for Prunus lusitanica ssp. azorica . It was hypothesized that diploid colonizers might be more common in archipelagos closer to continental source areas while successful polyploid colonizers could be more frequent at more remote islands (Carine et al. 2004, Crawford et al. 2009. However, at this stage, the number of detected polyploids in the Azorean flora does not support such a claim. Two possible ancestral colonization events may have occurred in the case of L. watsoniana: (a) an allotetraploid colonizer from North America or (b) an ancestral diploid population leading to an autotetraploid form, with later extinction of the ancestral diploid. Considering the absence of other endemic or native Lactuca species in the Azores, and the existence of allotetraploid North American Lactuca species (Dole zalov a et al. 2002) closely related with L. watsoniana (Fig. 2), an early colonization from North America seems a more plausible hypothesis. This is a very rare colonization pathway for vascular plants in the Azores, but was confirmed for another endemic Asteraceae, the Azores goldenrod Solidago azorica (Schaefer 2015). Therefore, L. watsoniana is not likely to be a neopolyploid, although formation of new cytotypes and their demographic establishment have been recorded for ocean islands (Crawford et al. 2009;Stuessy et al. 2014). More comprehensive morphological and molecular phylogenetic analyses are currently underway to verify this hypothesis.

Genetic diversity
The genetic variability found in the L. watsoniana populations was comparably high to that found for other Azorean herbaceous endemic taxa (Dias et al. 2014;Silva et al. 2015), and we found no evidence for inbreeding, despite the population decline that most likely took place in the past five centuries. It might also be a direct expression of polyploidy which causes an increase in gene dosage at the genome. Polyploid taxa often contain higher levels of genetic variation to their diploid relatives (Obbard et al. 2006;Garcia-Verdugo et al. 2015).
This high diversity of tetraploid colonizers together with self-compatibility would have been an advantage in the establishment of a sexually reproducing population Stuessy 1997, Crawford et al. 2009), providing favourable conditions for establishment in one island with subsequent dispersal to other parts of the  archipelago. However, studies addressing Azorean endemic plant reproductive strategies are very limited (see Crawford et al. 2015 for Tolpis spp.). Meanwhile, field and laboratory observations indicate that isolated L. watsoniana individuals flower regularly but mostly produce infertile seeds whereas in relatively large populations, such as those found in Pico Island, a very high number of viable seeds are produced, annually. In this sense, according to Ferrer and Good-Avila (2007), the basal members of the tribe Cichorieae are dominated by selfcompatibility or partial self-incompatibility exemplified by members of the genera Cichorium and Lactuca. Further studies should be undertaken to enlighten this matter, namely experiments of forced self-pollination and cross-pollination (Zhang et al. 2015).

Population genetic structure
We found five genetic groups in L. watsoniana, supported by AMOVA, PCoA and Bayesian model-based clustering. This genetic structure coincides spatially with the studied islands, with the exception of Pico Island, which is home to two genetic groups. The average F st value (0.45) was higher than those found for many other plant species (Bottin et al. 2005;Lucio 2011;Chaves 2013). We also observed that genetic variation was mainly found within populations (55.0 %), which is in accordance with the results obtained for Leontodon filii and L. rigens (Dias et al. 2014) and in contrast with Canary Islands endemic species, where most genetic variation was found between populations (G st ¼ 77.3 %) (Francisco-Ortega et al. 2000). However, a relatively high genetic variation was found among populations within islands (17.7 %), only surpassed in the Azores by the dioecious Juniperus brevifolia ). An even larger proportion of variation was found among islands (27.3 %), an extreme case in the Azores so far. Estimated gene flow values were always below 1, contrary to the much higher values obtained for Leontodon (Dias et al. 2014), but similar to the values obtained for Viburnum treleasei , although the latter study was conducted with dominant markers. The highest gene flow value (1.94) between two São Miguel populations (Table 5)   absence of relevant geographical barriers. The results of our F i analysis did not reveal any evidence for inbreeding. A high proportion of private alleles were found in São Miguel (7.04 %), Pico (5.88 %) and Terceira (3.64 %), indicating that some degree of differentiation occurred between those islands.

Geographical distance and barriers
Lactuca watsoniana show a strong geographic structure. Isolation by distance (IBD) and the existence of effective barriers to gene flow, the Atlantic Ocean and Pico Mountain, were found to be related with genetic distance. Therefore, L. watsoniana seems to be an intermediate case between hierarchical island and stepping stone models (Table 6) in what could be seen as an approximation to a two or three dimensions Kimura and Weiss (1964) stepping stone model, complicated by inexistent stones or by insurmountable barriers, i.e. the sea. This may be due to the existence of regional structuring (Morand et al. 2002), since oceanic barriers are, in general, more effective than topographic barriers at promoting isolation in insular systems (Gillespie and Clague 2009).
Lactuca watsoniana diaspores disperse by anemochory and hydrochory (Schaefer 2003;Silva et al. 2009), which are more efficient in small distance than in long distance dispersal. In this case, plants in spatially closer populations will be more similar genetically than those from more distant populations (Meirmans 2012). A limitation in dispersal ability is expected to be a key factor in geographical isolation (Kisel and Barraclough 2010), with a significant impact on relatedness within plant taxa (Papadopulos et al. 2011).

Disturbance from volcanism
After birth and growth of volcanic islands, destructive events such as secondary eruptions, landslides and thick volcanic deposits on the islands surface, might reshape the islands flora due to a massive population extirpation by soil sterilization (Mairal et al. 2015;Borges Silva et al. 2016), and lead to habitat fragmentation and genetic isolation of populations, which in due course could lead to differentiation and speciation within or among-islands (Gillespie and Roderick 2014;Mairal et al. 2015).
Considering our results, differentiation within Pico populations, grouped into two clusters, might be a consequence of the existence of Pico Mountain, acting as a natural geological barrier.
On the other hand, besides geographical distance and the existence of barriers, population genetic structure might be the result of the populations sampled being relicts, resulting from geological events, such as volcanic eruptions, as hypothesized for Juniperus brevifolia and Tolpis azorica , Borges Silva et al. 2016 as an explanation of the population genetic patterns found in Terceira Island. 'Sete Cidades' and 'Lagoa do Fogo' have been among the most active volcanic systems in São Miguel Island during the past 5000 years (Booth et al. 1978;Gaspar et al. 2015), where the most recent explosive eruption occurred just before human colonization in the 15th century (Ferreira et al. 2015). In Pico, Macedo (1981a, b) and Machado (1962) documented four major volcanic eruptions in the 16th and 18th centuries, in 'Prainha' (1562), 'Santa Luzia, São João' (1718) and 'Silveira' (1720). These episodes, with deposits of thick pumice, might have exterminated nearby populations of L. watsoniana, contributing to the unique observed patterns of genetic variation. Implications of island age on the observed genetic patterns are difficult to determine at this stage, due to the possibility of the later occurrence of volcanic disturbance, as illustrated above, and to a considerable effect of human disturbance (see below). Moreover, during a preliminary analysis, we included island age and differences in island age between populations, in our models to

Anthropogenic disturbance
Human colonization has had a strong impact on oceanic islands endemic plants since pristine ecosystems have evolved or persisted in the absence of novel or exotic disturbances (Connor et al. 2012). Human population settlement on previously pristine ecosystems, with large scale changes in land use, originated habitat loss and fragmentation both of which contribute to a decline in biological diversity (Wilcox and Murphy 1985;Kuussaari et al. 2009;Krauss et al. 2010;Triantis et al. 2010 (Espinosa and Noor 2002). Therefore, although the impact of human disturbance was observed in most populations, at this stage L. watsoniana populations managed to maintain a high degree of genetic diversity and low levels of inbreeding. However, the continuation of this level of disturbance might still lead to more pronounced population decline with unpredictable consequences on future diversity levels.

Implications in conservation
Our results, of a high genetic diversity, with substantial population differentiation among islands, strongly advocate that the concept of propagule provenance should be taken into consideration when developing augmentation or reintroduction strategies Weeks et al. 2011;Dias et al. 2014;Borges Silva et al. 2016). For example, the heterogeneity between the western and eastern populations in Pico means that seeds or individuals from the eastern part of this island should not be used in reintroduction programs on the western part of the island or even on the other islands. Since the species shows a considerable degree of genetic diversity, high seed production and high germination rate   Dias et al. 2014;Borges Silva et al. 2016). The uniqueness of some of the studied populations and the existence of populations with extremely low number of individuals, indicate that the populations of Pico (PIRB, PIAP, PICA, PIGM and PISM), 'Caminho da Criação' on São Miguel (SMCC) and Terceira (TEGA and TENR), should be considered priority for conservation.
The awareness of the threats by habitat loss and degradation resulting from changes in land use indicates that extreme care should be taken regarding the implementation of augmentation conservation strategies. Some measures have been implemented to locally control invasive species (Costa et al. 2013). However, L. watsoniana populations in São Miguel island are severely threatened by herbivores (feral goats and rabbits), which in Terceira have already alarmingly obliterated most of the Serra do Galhardo population. These animals should be controlled or eradicated from these protected areas.

Conclusion
Our research underscores the importance of considering a broad array of intrinsic traits and external factors when studying the evolution and management of plants in oceanic islands.
Regarding intrinsic traits, our results support previous claims suggesting that high levels of genetic variation are often associated with polyploid taxa, therefore highlighting the potential importance of this evolutionary mechanism in island systems (Obbard et al. 2006;Garcia-Verdugo et al. 2015). Indeed, while showing signs of population reduction, fragmentation, and local extinction, we found an example of a rare species that still maintains high levels of genetic diversity. External factors such as geographical distance and geographical barriers improved our ability to model genetic distance. Therefore, the classical models should be tested with newer and more integrated versions, including not only geographical distances but also relevant geographical barriers. While addressing oceanic island floras, discrepancies between the hierarchical and stepping stone classical models could be unravelled (Meirmans 2012), contributing to the eventual development of a more unified modelling approach for genetic distance, including also other dimensions such as temporal aspects (Anderson et al. 2010).
Finally, this study revealed that L. watsoniana populations show considerable genetic diversity, also representing another case in the Azores of pronounced genetic differentiation among populations in different islands. This information on the remaining Lactuca populations in the archipelago should be taken into consideration in the design of a recovery plan.

Sources of Funding
This study was funded by Fundo Regional da Ciência (M3.1.2/F/032/2011) and FEDER funds through the Operational Programme for Competitiveness Factors-COMPETE and by National Funds through FCT-Foundation for Science and Technology under the UID/ BIA/50027/2013 and POCI-01-0145-FEDER-006821.

Contributions by the Authors
E.F.D., participated in sampling, performed genetic analyses, and wrote the first draft; E.F.D and L.S. conceived the idea for the paper; L. S. was involved in the statistical analysis; M.M. was involved in the genetic analysis; H.S. participated in field work. All authors helped revise the manuscript.

Conflict of Interest Statement
None declared.