Growth form evolution and hybridization in Senecio (Asteraceae) from the high equatorial Andes

Abstract Changes in growth forms frequently accompany plant adaptive radiations, including páramo–a high‐elevation treeless habitat type of the northern Andes. We tested whether diverse group of Senecio inhabiting montane forests and páramo represented such growth form changes. We also investigated the role of Andean geography and environment in structuring genetic variation of this group. We sampled 108 populations and 28 species of Senecio (focusing on species from former genera Lasiocephalus and Culcitium) and analyzed their genetic relationships and patterns of intraspecific variation using DNA fingerprinting (AFLPs) and nuclear DNA sequences (ITS). We partitioned genetic variation into environmental and geographical components. ITS‐based phylogeny supported monophyly of a Lasiocephalus‐Culcitium clade. A grade of herbaceous alpine Senecio species subtended the Lasiocephalus‐Culcitium clade suggesting a change from the herbaceous to the woody growth form. Both ITS sequences and the AFLPs separated a group composed of the majority of páramo subshrubs from other group(s) comprising both forest and páramo species of various growth forms. These morphologically variable group(s) further split into clades encompassing both the páramo subshrubs and forest lianas, indicating independent switches among the growth forms and habitats. The finest AFLP genetic structure corresponded to morphologically delimited species except in two independent cases in which patterns of genetic variation instead reflected geography. Several morphologically variable species were genetically admixed, which suggests possible hybrid origins. Latitude and longitude accounted for 5%–8% of genetic variation in each of three AFLP groups, while the proportion of variation attributed to environment varied between 8% and 31% among them. A change from the herbaceous to the woody growth form is suggested for species of high‐elevation Andean Senecio. Independent switches between habitats and growth forms likely occurred within the group. Hybridization likely played an important role in species diversification.

environment in structuring genetic variation of this group. We sampled 108 populations and 28 species of Senecio (focusing on species from former genera Lasiocephalus and Culcitium) and analyzed their genetic relationships and patterns of intraspecific variation using DNA fingerprinting (AFLPs) and nuclear DNA sequences (ITS). We partitioned genetic variation into environmental and geographical components. ITS-based phylogeny supported monophyly of a Lasiocephalus-Culcitium clade. A grade of herbaceous alpine Senecio species subtended the Lasiocephalus-Culcitium clade suggesting a change from the herbaceous to the woody growth form. Both ITS sequences and the AFLPs separated a group composed of the majority of páramo subshrubs from other group(s) comprising both forest and páramo species of various growth forms. These morphologically variable group(s) further split into clades encompassing both the páramo subshrubs and forest lianas, indicating independent switches among the growth forms and habitats. The finest AFLP genetic structure corresponded to morphologically delimited species except in two independent cases in which patterns of genetic variation instead reflected geography. Several morphologically variable species were genetically admixed, which suggests possible hybrid origins. Latitude and longitude accounted for 5%-8% of genetic variation in each of three AFLP groups, while the proportion of variation attributed to environment varied between 8% and 31% among them. A change from the herbaceous to the woody growth form is suggested for species of high-elevation Andean Senecio. Independent switches between habitats and growth forms likely occurred within the group. Hybridization likely played an important role in species diversification.
Species of the genus Senecio L. (Asteraceae), which were traditionally placed in the genus Lasiocephalus Willd. ex Schltdl. (Cuatrecasas, 1978), comprise a morphologically and ecologically diverse plant group in the northern and central Andes. About 25 species are distributed from Venezuela to Bolivia, with the highest richness in Ecuador (Calvo & Freire, 2016;Cuatrecasas, 1978). Two main growth forms are recognized. Broad-leaved lianas (Figure 1 g,h) inhabit montane forests and secondary thickets usually between 2,800 and 3,800 m, although some species also occur in the forest-páramo shrubby ecotone called subpáramo (usually at 3,800 m). The other growth form is ascending or erect, narrow-leaved subshrub (Figure 1a-c,e) that occur in the páramo dominated by tussock grasses (3,800-4,300 m) and in the uppermost belt of patchy vegetation called superpáramo (up to 4,800-5,000 m). One species, Senecio mojandensis Hieron. (Figure 1d), a basal rosette herb of wet páramo habitats cannot be satisfactorily placed in either of these categories. Most species are morphologically distinct and readily identifiable, although some are variable in leaf size and shape, such as S. otophorus Wedd.
Phylogenetic molecular studies of the tribe Senecioneae suggest that the traditionally recognized Andean genera Lasiocephalus and Culcitium Bonpl. (scapose herbs forming basal leaf rosettes) belong to Senecio (Pelser, Nordenstam, Kadereit, & Watson, 2007;Pelser et al., 2010). Our previous study of 13 Senecio species from the former Lasiocephalus, which all were diploid, based on nuclear DNA sequences (ITS region) and nuclear genome size data (Dušková et al., 2010), identified two major clades that largely correspond to the two habitat types, that is, montane forest and páramo. The results also suggested that Senecio (Culcitium) nivalis (Kunth) Cuatrec. (Figure 1f) was closer to species of former Lasiocephalus than to other taxa of former

Culcitium.
Given its likely origin within ca. the last 2 Myr (Pelser et al., 2010) and occurrence in the montane-alpine habitats, the former Lasiocephalus exemplifies recent plant radiation in the (sub)tropical Andes. Based on extensive population sampling throughout the northern Andes and using an extended sample of ITS sequences complemented with highly variable AFLP (amplified fragment length polymorphism) markers, here we present deeper insights into the relationships among the Andean species of Senecio formerly classified in Lasiocephalus. Specifically, we examine a hypothesis put forward by Dušková et al. (2010) that independent transitions between the montane forest and páramo habitats occurred that were accompanied by growth form changes. We further examine patterns of genetic diversity within the group, and particularly their correlation with environmental factors and Andean geography.

| Plant material
Samples of species from the former Lasiocephalus and former Culcitium, along with co-occurring species of Senecio, were collected during 2006-2010 in Bolivia, Ecuador, Venezuela, and Colombia (Appendix S1). Due to the sampling gap in the central Andes, we lacked the single Peruvian species of former Lasiocephalus, a broadleaved liana S. loeseneri Hieron. This species is, nevertheless, sometimes considered conspecific with S. campanulatus Sch. Bip. ex Klatt from Bolivia (Calvo & Freire, 2016), which was included in our study.
Multiple populations were sampled for most of the species throughout their distribution ranges (Figure 2b). At each locality, geographical coordinates and elevation were recorded. Young, intact leaves were collected and desiccated in silica gel; vouchers were deposited in COL, PRC, QCA, QCNE, and VEN.

| AFLP fingerprinting and DNA sequencing
In total, 356 accessions of 18 Senecio species formerly classified as Lasiocephalus and 18 accessions of Senecio nivalis were genotyped using AFLP fingerprinting (Vos et al., 1995) (see Appendix S2 for details on the protocol). Fragments were manually scored with genemarker version 1.80 (SoftGenetics). Only unambiguous fragments in the range of 60-500 bp. were scored, regardless of their intensity (Tribsch, Schönswetter, & Stuessy, 2002). For 5% of the samples, the whole AFLP protocol was repeated from the isolated DNA onwards to test the reproducibility of the method (Bonin et al., 2004). Internal transcribed spacer (ITS) regions were directly sequenced using the primers ITS4 and ITS5 (White, Bruns, Lee, & Taylor, 1990) for 50 individuals of Andean Senecio (i.e., 44 of former Lasiocephalus, two of former Culcitium, four other members of Senecio). We selected the individuals in order to representatively cover all species of the former Lasiocephalus as well as all clusters and subgroups identified by AFLPs.

| Clustering of AFLP data
Genetic structure was inferred using a Bayesian clustering method implemented in structure 2.2.3 (Falush, Stephens, & Pritchard, 2007) employing a recessive allele model with admixture, assuming independent allele frequencies with 1,100,000 MCMC (Markov chain Monte Carlo) generations, and discarding the first 100,000 generations as burn-in. We limited the number of clusters (K) to 1-10, each K was replicated with ten runs, and we further assessed stability of the results by calculating similarity coefficients between the replicate runs (Nordborg et al., 2005) and delta K (Evanno, Regnaut, & Goudet, 2005), both calculated using the R-script Structure-sum-2009 (Ehrich, 2006). The Ks with consistent results over ten repeats were considered to be plausible and further examined. As the analysis of the entire dataset showed that only runs with K = 3 converged to a consistent solution in ten repeats, subsequent, separate structure analyzes of each of these three partitions (hereafter named clusters A, B, and C) were conducted using the same parameters. Only individuals assigned to a particular cluster with posterior probability >0.9 in the initial analysis were included in these subsequent analyzes. Major trends in the AFLP variation were visualized using principal coordinate analyzes (PCoA) based on Jaccard interindividual distances computed using famd 1.31 (Schlüter & Harris, 2006).
We further investigated the relationships among the major clusters based on a reduced subset of 266 individuals that were identified as nonadmixed (i.e., with posterior probabilities of membership to both major clusters and subgroups >0.9) in the structure analyzes. We reconstructed phylogenetic relationships using a likelihood model for binary restriction site data implemented in mrBayes v3.2.5 (Ronquist & Huelsenbeck, 2003 Figure 2); symbol shape indicates the growth form, that is, square-basal rosette herb, circlenarrow-leaved subshrub, triangle-broadleaved liana F I G U R E 2 (a) Assignment of 374 individuals (entire dataset) of highelevation Andean Senecio into three main AFLP clusters inferred in structure; (b) Geographical locations of populations with growth form and structure cluster assignment indicated; (c) Ordination of AFLP phenotypes by use of principal coordinate analysis (PCoA) based on Jaccard distances. The symbol coloration reflects the assignment of the individuals to the main structure clusters (white-admixed individuals with assignment probability below 0.5); symbol shape indicates the growth form, that is, square-basal rosette herb, circle-narrow-leaved subshrub, triangle-broad-leaved liana fragments by setting a condition that the characters that are absent (i.e., 0) in all individuals cannot be observed. We performed two independent runs of 5,000,000 generations each using the default prior settings, setting the restriction site model (lset nst = 1 coding = noabsencesites) and discarding the first 25% generations as burn-in.

| DNA sequence analyzes
Sequences of the ITS region were aligned by mafft 7 (Katoh & Standley, 2013) and edited using aliView (Larsson, 2014). In addition, we included in the final matrix previously published ITS sequences of: (1) 11 directly sequenced accessions of other Andean Senecio (Pelser et al., 2007) and (2)

| Growth form evolution
Character state reconstructions of the growth forms were performed employing a maximum likelihood approach implemented in the function rayDISC, part of the package corHMM (Beaulieu, O'Meara, & Donoghue, 2013) in R (Ihaka & Gentleman, 1996). This method allows for reconstructions of multistate characters, unresolved nodes, and ambiguities (polymorphic taxa or missing data). Three models of character evolution were evaluated as follows: equal rates (ER), symmetrical (SYM), and all rates different (ARD); an Akaike information criterion corrected for sample size (AICc) was used to select the best fitting model. Association of growth forms and phylogeny was tested by computing Pagel's lambda (Freckleton, Harvey, & Pagel, 2002) using the function fitDiscrete in the package geiger (Pennell et al., 2014) in R (Ihaka & Gentleman, 1996). Statistical significance of estimated lambda was tested by computing likelihood ratio test (LRT) against lambda = 0 model.

| Geographical analyzes of AFLP data
Geographical correlates of the genetic (AFLP) variation were examined after the admixed (i.e., posterior probability <.9), and Bolivian samples were excluded to avoid bias due to unclear cluster assignment and sampling gap, respectively. We tested for a significant correlation between matrices of genetic and geographical distances among populations (isolation by distance) using a Mantel test in adegenet. Among-population genetic chord distances derived from AFLP fragment frequencies were inferred using a Bayesian method with nonuniform priors (Zhivotovsky, 1999) as implemented in famd 1.31 (Schlüter & Harris, 2006).
Climatic data describing mean annual temperature, daily and annual temperature ranges, annual rainfall, and its inter-annual variation expressed as coefficient of variation for each collection site were ex- provides an opportunity to make partial tests to discriminate between pure effects of explanatory variables and their interaction.

| AFLP fingerprinting
AFLP analysis of 374 accessions resulted in 269 reliable fragments, of which 264 (98%) were polymorphic. The overall reproducibility of the dataset was 95%.  Cuatrec.. The Bayesian clustering was also reflected in PCoA ordination, separating the three clusters along the first (cluster A vs. C) and

| Main grouping within the entire dataset
second (cluster B vs. A + C) axes (Figure 2c).

| Finer structure within the main clusters
Separate Bayesian clustering of the accessions assigned to cluster A revealed that K = 2 and K = 3 exhibited high similarity among in-

| ITS and AFLP phylogeny
Bayesian analysis of ITS sequences showed monophyly of a clade comprising all accessions of the former Lasiocephalus and former Culcitium (together with Senecio chionogeton), although it did not support separation of the two former genera (Figure 6a). Instead, along with several unresolved former Culcitium accessions, we identified two clades, corresponding to "páramo" and "forest" clades of Dušková et al. (2010), which with a few exceptions corresponded to major AFLP clusters C and A + B, respectively (Figure 6a,b, Table 1). The "forest clade" further split into several well supported subclades (with uncertain relationships among them) which mostly corresponded to AFLP subgroups (namely B2, B3 + B4, B5, A1 + A3, A2 + A3 subgroups).
There were several remarkable incongruences among ITS and AFLP data. In particular, Senecio nivalis (cluster B1) shared the same ITS haplotypes with S. superparamensis (cluster C4) and both species formed a supported lineage (p1) within the "páramo" clade ( Figure 6a). AFLP cluster C1 was split into both major ITS clades, with the Colombian and Ecuadorian accessions being parts of the "forest" and "páramo" clades, respectively. ITS clones isolated from a single accession of S. aff. quitensis (pop. 88_Pi) fell into both major ITS clades. Finally, S. puracensis from "páramo" cluster C appears nested within the ITS "forest clade." Bayesian phylogenetic analysis of AFLP phenotypes of nonad- for relationships among the AFLP subgroups, except for supported monophyly of cluster A (Figure 6b, Appendix S3D).

| Geographical and environmental analyzes of AFLP data
Senecio populations as a whole and members of cluster A showed a very weak correlation between genetic and geographical distances (isolation by distance, IBD), whereas this relationship was nonsignificant in the other two clusters (Table 2). Subgroups with sufficient F I G U R E 6 (a) Phylogenetic reconstruction of 87 accessions of northern Andean Senecio based on sequences of ITS region of ribosomal DNA. Bayesian 50% majority rule consensus tree with posterior probabilities >0.90 and bootstrap values >50% inferred with maximum parsimony are indicated, respectively, before and after the slash above each supported branch. Supported subclades of the "forest" and "páramo" clades are marked as f1-f6 and p1-p2, respectively. Growth form of each accession is marked by a symbol, membership in the AFLP subgroups (if applicable) is denoted by corresponding letters (A1-C5), accessions with ambiguous structure assignment are marked "MIX." The presence of highly divergent ITS sequences in the same individual of S. aff. quitensis is marked by an arrow. Senecio doryphyllus, S. decipiens, and S. alatopetiolatus, although belonging to the former Lasiocephalus, were not analyzed using the AFLPs. Reconstruction of the growth form evolution according to the equal rates (ER) model has been superimposed onto the ITS tree (see Appendix S3E for original). (b) Relationships among AFLP phenotypes of 266 nonadmixed (see Section 2) individuals of former Lasiocephalus and Senecio nivalis reconstructed in Bayesian framework. Cluster codes correspond with Figures 3-5; branches with posterior probabilities >0.95 are marked with dots numbers of populations were available only in cluster A; here we observed significant correlations in both Ecuadorian-Colombian subgroups A1 and A2 but a lack of correlation in the southern Ecuadorian subgroup A2.
Less than 10% of total variance in the entire AFLP dataset was accounted for by the effects of either environmental (rainfall, temperature, elevation; altogether 6% of variability) geographical (latitude, longitude; 2% of variability) components or their interaction (1.5% of variability) ( Table 2). When the three AFLP clusters were analyzed separately, the geographical component accounted for a similar (5%-8%) proportion of the total variation, whereas the environmental component accounted for almost a third of variation in cluster B but only 8%-9% in clusters A and C. Moreover, there was a distinct interaction (5%) between the two sets of variables in cluster A, whereas the interaction was very low or lacking in the two other clusters. Pelser et al. (2007Pelser et al. ( , 2010 and Dušková et al. (2010) pointed to close relationships between the former genera of Lasiocephalus and

| Lasiocephalus-Culcitium species group
Culcitium. The present study, using ITS sequences and an extended list of species, suggests monophyly of the Lasiocephalus-Culcitium species group but with neither of the two former genera monophyletic.
Although relationships within the group are only partly resolved in both the ITS and AFLP datasets, suggesting a recent diversification (Turner et al., 2013), there is a partial congruence between the two markers, as AFLP cluster C corresponds to the ITS "páramo clade" (except for Senecio nivalis) and AFLP clusters A and B correspond to the ITS "forest clade" (except for S. puracensis) ( Figure 6, Table 1); incongruences will be discussed below.

| Growth form changes and habitat shifts
Species of different growth forms and preferences for páramo or montane forest fell within several different AFLP (sub)groups and ITS (sub) clades, suggesting that independent shifts in ecology were accompanied by changes in morphology. Both AFLP and ITS data indicate that at least two distinct genetic entities occur in the páramo, representatives of which demonstrate convergence in such traits as growth form, size and number of capitula, and leaf morphology (Figure 1). The first entity is the páramo-dwelling AFLP cluster C (largely corresponding to the ITS "páramo clade"), species of which occur throughout most of Ecuador and southern Colombia. The second entity is represented by Venezuelan S. longepenicillatus (AFLP cluster B, f3 subclade within the ITS "forest clade"), which is a narrow-leaved páramo subshrub but sporadically also appears in a broad-leaved form at the tree-line eco- Whereas Lobeliaceae (Knox, Muasya, & Nuchhaala, 2008), Huperzia Bernh. (Wilkström, Kenrick, & Chase, 1999), Chusquea Kunth (Fisher et al., 2009), and Disterigma (Klotzsch) Nied. (Pedraza-Peñalosa, 2009) apparently colonized alpine habitats from the montane forest, for Chaetanthera Ruiz & Pav. and Puya Molina migration was suggested in the opposite or both directions, respectively (Hershkovitz, Arroyo, Bell, & Hinojosa, 2006;Jabaily & Sytsma, 2012). As a grade of herbaceous Senecio species from alpine habitats subtends (although the support is weak) the Lasiocephalus-Culcitium clade (Figure 6a), the ITS phylogeny is consistent with an alpine-to-forest transition for the evolution of the Lasiocephalus-Culcitium group as a whole. If such a relationship is confirmed, a change from the herbaceous (basal leaf rosette) to the woody (liana, ascending subshrub, shrub) state is implied, similar to, for example, Andean Valeriana L., Gentianella Moench, and Loricaria Wedd. (Kolář, Dušková, & Sklenář, 2016;Sklenář et al., 2011). The polytomy consisting of the "páramo clade," "forest clade," and several basal rosette herbs does not permit interpretation of the growth form transitions within the Lasiocephalus-Culcitium clade to evaluate Cuatrecasas' (1978) idea that the páramo growth form of former Lasiocephalus species evolved from the growth form of their montane forest ancestor(s).
However, Cuatrecasas' view could be valid for some páramo subshrubs found within the "forest clade" (e.g., S. longepenicillatus). Moreover, the ITS phylogeny suggests another transition in this clade, that is, to a basal rosette herb in S. mojandensis.
T A B L E 2 Eco-geographical covariates of genetic variation of north-Andean Senecio. Variance partitioning (by means of RDA) of AFLP genetic variation into environmental (rainfall, temperature, elevation) and geographical (latitude, longitude) components and correlation between genetic and geographical distances (by means of Mantel test and quantified by correlation coefficient r M ).
Traces of hybridization are indicated by consistently admixed AFLP profiles across multiple populations of several species (Figure 2a). This is especially apparent for Senecio aff. quitensis, a taxon with morphology varying between subshrubs and lianas and whose accessions variously combine AFLP profiles of clusters A (forest lianas) and C (páramo subshrubs). Furthermore, ITS sequences of this species (including divergent ITS copy types from a single individual) are placed in the divergent "páramo" and "forest" clades, and its genome size is intermediate between them (Dušková et al., 2010).
Incongruence between the AFLP and ITS datasets suggests that hybridization might also have been involved in the origin of other

| Geographical and ecological correlates of genetic variation
Geographical barriers along with ecological differentiation promote species diversification in mountains (Kolář et al., 2016;Luo et al., 2016). As a geographical signal was comparably strong in the three main AFLP clusters, geography may structure genetic variation in a similar way in both the Andean montane forest and the páramo. In support of this, the AFLP data reveal two strikingly similar cases of genetic separation corresponding to geography which are incongruent with morphology-based species limits. Two páramo species, Senecio lingulatus and S. superandinus (cluster C), are readily distinguished morphologically (Figure 1a,e) and are genetically distinct throughout most of Ecuador. Their populations in southern Ecuador, however, merge and form another distinct genetic subgroup (Figure 5a,b).
Similarly, two morphologically distinct species from the montane forest-páramo ecotone, S. involucratus and S. otophorus (cluster A), appear as distinct AFLP entities in northern-central Ecuador and Colombia, but the markers fail to discriminate between them in southern Ecuador (Figure 3a,b). There, the species form a separate subgroup together with another morphologically distinct montane forest liana, S. cuencanus.

As morphologically intermediate plants between S. superandinus
and S. lingulatus occur in southern Ecuador, gene flow due to hybridization might have generated the observed pattern. In contrast, we did not observe any putative hybrids between S. involucratus and S. otophorus, nor did we find any consistent morphological distinction in plants of either species from southern Ecuador. Therefore, we hypothesize that southern Ecuador represents an ancestral area of cluster A where high levels of ancestral polymorphisms have been retained, preventing the genetic discrimination of species by means of the AFLPs. Both species might have independently migrated northwards, leaving a footprint of gradual genetic differentiation which is documented by a significant isolation by distance relationship observed in A1 and A3 subgroups. Such northward migration would be consistent with the biogeographical reconstructions of Andean plant groups such as Azorella, Oreobolus R. Br., and Puya (Andersson, Kocsis, & Eriksson, 2006;Chacón, Madriñán, & Bruhl, 2006;Jabaily & Sytsma, 2012).

The northern and central Ecuadorian Andes experienced a different
Quaternary history from the south of the country, namely in having volcanism and glaciation (Jørgensen & Ulloa, 1994). Glaciation events and volcanism may have structured the genetic patterns of the species through the effects of repeated bottlenecks and founder events (Luo et al., 2016;Vásquez, Balslev, Hansen, Sklenář, & Romoleroux, 2016).
The genetic structure of species from the montane forest-páramo ecotone (cluster A) and páramo (cluster C) showed little association with environmental variables, which we acknowledge might be at least partly due to the lack of precision of extrapolated climatic variables for high mountains (Hijmans et al., 2005;Kirchheimer et al., 2016).
However, in cluster B, the high proportion of genetic variation associated with environmental factors is consistent with the variety of habitats occupied by its species. The ecological differentiation coupled with high AFLP and morphological diversification may suggest a relatively long divergence time and/or efficient isolation (Kolář et al., 2016). The very small or entirely lacking association between environment and geography in clusters C and B, respectively, suggests that two distinct and (largely) independent signals are involved. However, the stronger association in cluster A suggests that migration along the cordilleras was coupled with a shift in species ecology, such as the entry of S. otophorus in the superpáramo in Colombia.