Phylogenetic comparative methods improve the selection of characters for generic delimitations in a hyperdiverse Neotropical orchid clade

Taxonomic delimitations are challenging because of the convergent and variable nature of phenotypic traits. This is evident in species-rich lineages, where the ancestral and derived states and their gains and losses are difficult to assess. Phylogenetic comparative methods help to evaluate the convergent evolution of a given morphological character, thus enabling the discovery of traits useful for classifications. In this study, we investigate the evolution of selected traits to test for their suitability for generic delimitations in the clade Lepanthes, one of the Neotropical species-richest groups. We evaluated every generic name proposed in the Lepanthes clade producing densely sampled phylogenies with Maximum Parsimony, Maximum Likelihood, and Bayesian approaches. Using Ancestral State Reconstructions, we then assessed 18 phenotypic characters that have been traditionally employed to diagnose genera. We propose the recognition of 14 genera based on solid morphological delimitations. Among the characters assessed, we identified 16 plesiomorphies, 12 homoplastic characters, and seven synapomorphies, the latter of which are reproductive features mostly related to the pollination by pseudocopulation and possibly correlated with rapid diversifications in Lepanthes. Furthermore, the ancestral states of some reproductive characters suggest that these traits are associated with pollination mechanisms alike promoting homoplasy. Our methodological approach enables the discovery of useful traits for generic delimitations in the Lepanthes clade and offers various other testable hypotheses on trait evolution for future research on Pleurothallidinae orchids because the phenotypic variation of some characters evaluated here also occurs in other diverse genera.

Incongruence between nuclear and plastid datasets. A total of 24 terminals were detected as incongruent with ML and 34 with BI. Of those, 20 terminals were retrieved as incongruent by both inferences (Appendices S1, S6). The topology of the BI, MP and ML trees inferred from the concatenated datasets excluding/ including the plastid conflicting sequences recognized essentially the same generic clades but showed some differences in the topology and support for intergeneric relationships (Appendices S1, S6, S7).  Fernández & Bogarín (clade D) were well supported and the group was sister to Lepanthes + Draconanthes + Pseudolepanthes (clade 2) (LPB = 80% and PP = 0.98). When phylogenetic incongruence was not considered, Pseudolepanthes and Stellamaris clustered in a clade with strong support in the MP tree (PBP = 100). The genus Frondaria (clade E) was found to be related to Lepanthes, Draconanthes, Pseudolepanthes, Stellamaris (clade 3), well supported (PBP = 100 and PP = 0.97) but lacking support in the ML analysis (LPB = 56%). Clade 4 made up by clade 3 + Frondaria and comprised the species more related to the core of Lepanthes whereas Lepanthopsis (clade F) and Gravendeelia (clade G) both clustered in clade 6. The clades 1-4 and 6 are clustered (clade 5) with high support. Most nodes of these clades were well supported (PBP > 100, LPB > 72 and PP > 0.98) with the only exception being clade 6 with low LPB support but well supported by PBP > 100 and PP > 0.98. The genus Opilionanthe was sister to clade 5 + clade 6 with high support for PBP = 100, unsupported by BI (PP = 0.94) and low support for ML (LPB = 58). Topologically, Opilionanthe always clustered apart from the other generic clades discussed here. Related to the groups of clade 7 (members of the core of Lepanthes and Lepanthopsis) was a group consisting of species related to Trichosalpinx s.s. (clade K), Pendusalpinx (clade J) and Lankesteriana (clade I) all highly supported (PBP = 100, LPB ≥94 and PP = 1.0). This topology was retrieved with high to moderate support (PBP = 100, LPB ≥54 and, PP ≥ 0.96) after removing incongruences using the Procrustean Approach to Cophylogeny (PACo) application 27  Character evolution. ASR was based on the one-rate model ER that was consistently better than the SYM and ARD models (Appendices S8, S10). These estimations were obtained using phylograms from MrBayes and ultrametric trees from BEAST calculated under the Birth Death as the best speciation model according to a Bayes factors test (Appendix S11). Estimations based on the reversible-jump Markov Chain Monte Carlo (MCMC) model yielded similar results compared to the rates obtained with SCM (Appendices S12, S14). For the MCMC approach with BayesTraits V3 the best results were obtained with the hyperprior adjusted to the previously obtained ML transition rates (from 0 to 0.03). The ACE, SIMMAP and re-rooting methods yielded identical scaled-likelihoods at the root state and the estimations with MCMC revealed essentially the same results obtained with ACE and SIMMAP with ambiguous estimations for the characters of inflorescence length and synsepal ( www.nature.com/scientificreports www.nature.com/scientificreports/ with non-proliferating, unornamented-papyraceous ramicauls, simultaneously flowering inflorescences, fully opening flowers with concave, ovate-acute dorsal sepals, dissimilar petals, column with a foot, a laminar, motile lip without glenion and a ventral anther with entire stigma ( Table 2). The most common character state transitions are: a caespitose to creeping/pendent habit, ornamented to unornamented-papyraceous bracts, non-proliferating to proliferating ramicauls, simultaneously flowering to successively flowering inflorescences, shortening of inflorescences, fully opening flowers to bud-like flowers, ovate-acute to ovate-acuminate/oblong-acute sepals, concave to flattened dorsal sepals, dissimilar to subsimilar petals, loss of a column foot and synsepal, movable to sessile lip, entire to bilobed stigma, ventral to dorsal anther and pollinarium with naked caudicles to caudicles with a viscidium (Figs 5, 6). Probabilities favoring reversal transitions from proliferating to non-proliferating ramicauls, foliaceous to ornamented/unornamented-papyraceous bracts, creeping to caespitose habit, bud-like to opening flowers, subsimilar to dissimilar petals, oblong-acute to ovate-acuminate/ovate-acute sepals, presence of a glenion to absence, sessile to motile lip, absence of a column foot to presence, dorsal to apical anther, bilobed to entire stigma and pollinarium with caudicles and a viscidium to lack of a viscidium, were found to be unlikely. Lip shape from laminar to bilobed and vice-versa showed a similar probability (Figs 5, 6). Twelve homoplastic characters and seven synapomorphic characters were detected ( Table 2). The combination of a sessile lip, absence of a column foot, dorsal anther and pollinarium with caudicles and viscidium are features only observed in Lepanthes, Draconanthes, Pseudolepanthes and Lepanthopsis, whereas motile lips, a column foot, ventral anther and pollinarium with caudicles are observed in all other genera investigated.

Discussion
phylogenetics of the Lepanthes clade. In this section, we discuss the nomenclatural changes needed to redefine the Lepanthes clade as proposed by Bogarín et al. 23 as well as the relationships among these genera based on the phylogenetic insights and morphological evolution of key characters as presented in this study. The Lepanthes clade comprises four main subclades (clades 7, 9, 12 and 14 as shown in Fig. 3). Zootrophion is plus Epibator, sister to the rest of the subclades. The next successively diverging clade is Anathallis, plus Tubella (see further discussion on Trichosalpinx s.l). Anathallis was initially re-established for the species of Pleurothallis subgenus Acuminatia sect. Alata and Pleurothallis subgenus Specklinia sect. Muscosae 28 . The clade composed of Anathallis plus Panmorphia Luer is confirmed monophyletic, and the exclusion of members of Pleurothallis subgenus Acuminatia sect. Acuminatae, which are related to Stelis s.l. 17 . Also, Karremans 29 established the genus Lankesteriana because its members were closely related to Pendusalpinx rather than to Anathalllis s.s. as suggested by Pridgeon et al. 9 . Trichosalpinx as previously circumscribed 15,30 is confirmed as polyphyletic, and therefore re-circumscribed. The species belonging to Pendusalpinx and Tubella are confirmed to be unrelated to Trichosalpinx and therefore excluded, whereas Gravendeelia, Opilionanthe, Pseudolepanthes and Stellamaris are placed for the first time in a phylogenetic framework and recognized as distinct 23 . The polyphyly of Trichosalpinx was suggested in previous studies but the newly proposed genera were not evaluated or the sampling was too incomplete to allow a redefinition of these groups 11 . The relationships recovered here also suggest that only Pendusalpinx and Lankesteriana are closely related to Trichosalpinx s.s. As suggested by Pridgeon 9 , members of Tubella are isolated from Trichosalpinx and Pendusalpinx but these relationships were not supported 12  www.nature.com/scientificreports www.nature.com/scientificreports/ the inclusion of members of the clade not previously evaluated (Gravendeelia, Opilionanthe, Pseudolepanthes, and Stellamaris), Tubella is now sister to Anathallis.
The most recently diverging clade of the Lepanthes clade consists of Lepanthes and its allied genera: Draconanthes, Gravendeelia, Lepanthopsis, Opilionanthe, Pseudolepanthes and Stellamaris. With the exception of Draconanthes and Lepanthopsis, these genera were formerly all treated under Trichosalpinx s.l. However, we confirm here that they are closely related to Lepanthes and Lepanthopsis rather than to Trichosalpinx s.s. In addition, a clade composed of Lepanthopsis plus Expedicula was found monophyletic. In the next sections, we discuss the morphological characters supporting the new classification for the Lepanthes clade proposed here.

Morphological evolution.
Our character reconstructions improved the understanding of the evolution of phenotypic traits used to classify the genera of the Lepanthes clade. We identified homoplastic characters, that are not suitable for generic circumscriptions, as well as synapomorphies ( Table 2). Plant habit (caespitose or creeping) evolved several times with a higher transition frequency from caespitose to creeping. This was found for other groups within Pleurothallidinae as well, possibly as an adaptation to different environments. Proliferating ramicauls evolved from non-proliferating ones independently in four clades. The lack of ornamentation of the ramicauls confused taxonomists as the close relationship of Zootrophion, Anathallis, and Lankesteriana with Lepanthes, Lepanthopsis and Trichosalpinx s.l. was not recognized previously. In addition, a combination of plesiomorphic and homoplastic characters in Trichosalpinx s.l., such as the ornamentation of the ramicauls, concave dorsal sepals, ovate-acuminate, caudate petals, motile, laminar lips with a column foot and ventral anthers caused misclassifications of the now separated genera Gravendeelia, Pendusalpinx, Opilionanthe, and Stellamaris. Assessment of other potential diagnostic traits was needed for these genera to complement a classification based solely on homoplastic characters. For example, the synapomorphic sub-similar petals in Opilionanthe are a diagnostic feature of the genus, showing a low probability of transition back to the ancestral state, dissimilar petals.
Inflorescence type and length are also variable characters in Pleurothallidinae 13 . Although groups show trends towards the presence of only one of the states, there are always exceptions. For example, all species of Lepanthes studied here have inflorescences shorter than the leaves but some species (not sampled in this study) have inflorescences longer than the leaf. The opposite is observed in Trichosalpinx 30 . The ancestral traits recovered for the anther position, column foot, pollinarium type, and lip motility suggest that these are associated with pollinators that enter the flower using the laminar lip. When trying to depart the flower, the dorsal part of the insect scrapes the anther of the column in the area of the caudicles and removes the pollinarium 26,[31][32][33] . This mechanism predominates in Zootrophion, Tubella, Anathallis, Trichosalpinx, Lankesteriana, Pendusalpinx, Opilionanthe, Gravendeelia, Frondaria and Stellamaris. The recent discovery of biting midges in Forcipomyia (Ceratopogonidae) as pollinators of two species of Trichosalpinx highlights the importance of the motile, papillose, ciliate lip for their pollination 26 . Additional floral micromorphological characters of these three genera, such as the papillose surface of the lip with striated cuticles and secretions of proteins as possible rewards support a hypothesis of floral convergence 26 . The flowers of some species of Anathallis, Tubella and Opilionanthe are similar to other pleurothallids, such as the white-flowered Specklinia calyptrostele, visited by biting midges (Ceratopogonidae) 21 , suggesting that floral similarities are prone to convergent pollination.
The predominance of an ancestral morphology adapted to pollination by biting midges makes these characters unsuitable for generic classification. The combination of a sessile lip, without a column foot, dorsal anther   www.nature.com/scientificreports www.nature.com/scientificreports/ and a sessile lip are key features for pollination by sexual deception 25 . Even though pollination observations are documented only for a handful of species of this genus, the floral synapomorphies indicate that pseudocopulation is likely to be predominant in the group. Lepanthes-like flowers are also found in species of the former Lepanthes subgenus Brachycladium Luer, today known to belong to the distantly related Andinia 34 . Floral convergence is probably due to selective pressure as suggested by Wilson et al. 34 based on pollination observations by Álvarez 35 .
In Lepanthopsis, autapomorphic characters such as a glenion and bilobed stigma suggest an adaptation to different, yet unknown pollinators as compared to Lepanthes and Trichosalpinx 25,26 . Lepanthopsis and Gravendeelia are grouped in the same clade and the need for recognition of Gravendeelia is supported by autapomorphic characters of Lepanthopsis such as the glenion and bilobed stigma. As transitions of these characters to the ancestral state are unlikely, it seems that floral evolution in Lepanthopsis and Gravendeelia took a different path. Floral morphology of Lepanthopsis resembles that of Platystele Schltr. and the autapomorphic characters such as the presence of a glenion and bilobed stigma suggest an adaptation to different, yet unknown pollinators. In contrast, Gravendeelia has a floral morphology oriented towards pollination that likely involves a similar behavior of insects as described in Trichosalpinx s.s 26 . Ambiguous results obtained for inflorescence length, the formation of a synsepal at the root state and, the higher frequency of transitions between different states indicates that these traits evolved independently in several groups within Pleurothallidinae 15 . The synsepal is made up of fused lateral sepals, and this condition varies from unfused to fully fused. A possible correlation between sexual mimicry and successive flowering in Lepanthes suggests that all flowers opening at the same time might not be an optimal strategy to fool male fungus gnats (Sciaridae), because several female-mimicking flowers together may accelerate males not being tricked 36,37 . In contrast, the meagre rewards for female biting midges in Trichosalpinx flowers suggest that several flowers opening at the same time might be more advantageous for attracting pollinators 26 .
Circumscription of the genera in the Lepanthes clade. Lepanthes has been consistently supported as comprising a clade in previous studies 9,12,23 (Fig. 3, Appendix S15). Species of the genus are known for their caespitose habit with lepanthiform sheaths. Among its close relatives, the transversely bilobed petals, bilobed lip with a basal appendix, elongate column with apical anther, and viscidium are diagnostic for our favored circumscription of the genus. Several earlier proposed subgeneric divisions of Lepanthes 38 were not supported by our molecular phylogenetic analyses and will require re-evaluation whenever a broader sampling becomes available. Draconanthes was based on the former Lepanthes subgenus Draconanthes 38 , currently made up of two species known only from high elevations. It is sister to Lepanthes s.s., Draconanthes and Lepanthes are morphologically similar but the former may be distinguished by the rigid sepals, linear elongate, unlobed petals and a fleshy lip with a rudimentary appendix-like structure in contrast with the elaborate appendixes of Lepanthes. As we did not evaluate the evolution of these particular traits, an alternative option is to treat Draconanthes under Lepanthes based on similarities instead of the differences here discussed. Pseudolepanthes is sister to Lepanthes plus Draconanthes, rather than being related to Trichosalpinx as was previously assumed 9 . Pseudolepanthes resembles species of Lepanthes in plant architecture, but its species are immediately set aside by the spreading, linear to narrowly ovate petals, and the laminar, appendix-free lip with a prominent warty callus, which suggest a different pollination strategy as compared to pseudocopulation recorded in Lepanthes 30 . Stellamaris currently includes a single species, Stellamaris pergrata, previously believed to belong to Trichosalpinx 39 . It is sister to Lepanthes plus Draconanthes and Pseudolepanthes instead. With the last, it shares the caespitose habit, but it can be distinguished by a short, few-flowered inflorescence, long-caudate sepals, a callose lip, an elongate column with an incumbent anther and a prominent column foot, and no viscidium 23,30 . Frondaria can be distinguished by the synapomorphic conspicuous foliaceous sheaths along the stems. Contrary to the terminal leaf, the smaller leafy bracts do not have an abscission layer which is consistent with them being overgrown, green bracts rather than true leaves. Frondaria produces elongate inflorescences with simultaneously opening, white flowers with spreading, acuminate sepals that are virtually indistinguishable from those of the distantly related genera Anathallis and Tubella. Lepanthopsis plus Gravendeelia are is sister to Lepanthes, Draconanthes, Pseudolepanthes, Stellamaris and Frondaria. Species of the genus are recognized by the inflorescences with simultaneously opening, flattened flowers, provided with a fleshy, simple lip with a glenion at the base and a short column with a bilobed stigma 40 . A few exceptions to this scheme are found in Lepanthopsis subgen. Microlepanthes Luer 40 . Gravendeelia is monotypic and sister to Lepanthopsis. Gravendeelia chamaelepanthes (Rchb.f.) Bogarín & Karremans, undoubtedly represents a species complex in need of further revision. It differs from Lepanthopsis in the chain-like, pendent habit, the few-flowered inflorescences with tubular flowers with elongate sepals, an elongate lip without a glenion and the elongate column with a distinct foot and unlobed stigma 23,30 . Both plants and flowers of Gravendeelia are different from Lepanthopsis that their close phylogenetic relationship is one of the most unexpected results of this study (Fig. 3, Appendix S15). The flowers resemble those of the unrelated genera Anathallis, Stellamaris and Tubella. Opilionanthe, formerly placed in Trichosalpinx, is sister to Lepanthes, Draconanthes, Pseudolepanthes, Stellamaris, Frondaria, Lepanthopsis and Gravendeelia. The lepanthiform bracts, caespitose habit and more or less tubular white flowers are reminiscent of Tubella, thus the isolated phylogenetic placement of this species was unexpected. However, O. manningii (Luer) Karremans & Bogarín is immediately distinguished from species belonging to other genera by the sub-orbicular leaves and the long-caudate petals, which are subsimilar to the sepals 23 .
Lankesteriana, Pendusalpinx, and Trichosalpinx are florally similar as they share purplish flowers with a motile, ciliate lip, attached to a column foot, and an ventral anther and stigma 23,29,30 . The vegetative morphology, however, is distinct. Species of Lankesteriana can be easily distinguished from Trichosalpinx and Pendusalpinx by the extremely small habit with ramicauls that lack ornamented lepanthiform bracts shorter than the leaves and the successively flowering inflorescences 29 .
Trichosalpinx and Pendusalpinx are vegetatively similar to each other, with a large size long ramicauls and simultaneously flowered inflorescences. Pendusalpinx differs in its pendent habit with large, whitish lepanthiform bracts and glaucous leaves 23 . Based on vegetative morphology alone it is unexpected that Lankesteriana and Pendusalpinx should be sister to each other. However, these findings are congruent with those of previous studies 20,29 . On the other hand, contrary to what was found by those authors 20,29 , Lankesteriana and Pendusalpinx are here found to be sister to Trichosalpinx as previously supported 12 . Due to the contradictory inferences, the relatively long branches of the Lankesteriana accessions, and the highly diverging morphologies, we remain cautious as to the stable phylogenetic relationships between these three genera. It is possible that the similar floral morphology was caused by convergent evolution due to a similar pollination strategy rather than common ancestry 23  www.nature.com/scientificreports www.nature.com/scientificreports/ An alternative hypothesis is to treat these genera under Trichosalpinx based on the high support obtained and floral similarities discussed above (Appendix S15).
Anathallis and Tubella are each well supported but moderately to weakly supported as sister genera in the BI and ML analyses, and therefore their relationship remains weakly supported. Anathallis is distinguished by the non-lepanthiform sheaths, caespitose ramicauls, and the free, star-shaped perianth 16,29 . Some species have purple flowers with motile lips, whereas others share similar micromorphological characters with Lankesteriana, Pendusalpinx and Trichosalpinx s.s. such as the striated cuticles and secretion of proteins 26 . Members of Pleurothallis subgenus Acuminatia sect. Acuminatia are phylogenetically related to Stelis s.l. and should therefore not be considered as part of Anathallis 17 . Allied to Lepanthes, Lepanthopsis, Trichosalpinx and their allies (clade 8) are members of Tubella, a group previously recognized as a subgenus of Trichosalpinx 30,39 . It comprises mostly slender plants with creeping ramicauls, simultaneously flowering inflorescences with whitish flowers and elongate sepals.
Zootrophion plus Epibator are placed as sister to all other members of the Lepanthes clade. They can be distinguished by the partial opening of the flowers due to the apical fusion on the sepals. As a consequence, the flowers have a single opening on each side, giving them a unique appearance. This feature, present in all species of Zootrophion, is not present in the other members of the Lepanthes clade; however, it is present in other unrelated genera of the Pleurothallidinae. The synsepal is thick and verrucose, and the lip is minute. The bracts are large, unornamented-papyraceous and loose.

Conclusions
Generic delimitations of orchids based on morphological traits is made challenging daunting because of extensive homoplasy in characters previously used for circumscriptions. The Lepanthes clade has long challenged systematists and taxonomists due to the floral homoplasy possibly resulting from similar pollination systems. Assessing homoplasy, synapomorphies, and symplesiomorphies on a single major clade within Pleurothallidinae could be regarded as misleading because some characters such as a synsepally, glenion or viscidium are present in other clades of the subtribe. Therefore, we use here the Lepanthes clade and the combination of diagnostic traits coupled with ASRs as an example of an effective strategy for further characterizing its subclades in a novel way. Based on the results of the ASRs, members of these subclades can be either classified as genera, subgenera, or sections under the different views of systematists. Here, we propose the recognition of 14 genera in the Lepanthes clade based on a combination of molecular phylogenetics, and a morphological assessment of characters previously used in the taxonomy of Pleurothallidinae. We acknowledge that our findings can be interpreted differently, producing alternative hypotheses such as, for example, reducing the number of genera recognized here even further or delimiting them using morphological similarities instead of differences. All formerly proposed classifications of the Pleurothallidinae have been published without assessing the evolution of any of the traits. In contrast, and for the first time, we propose a classification based on an ASRs. The strategy followed in our study allows for the detection of those potentially linked to pollination or environmental pressures that can lead to mistaken delimitations. Thus, future research should focus on assessing novel morphological traits not previously used for classifications and including continuous characters that complement the discrete ones. For instance, micro-morphological traits such as cell wall lignification might be promising because they have been overlooked by orchid taxonomists.
Concerning species sampling, future studies should focus on members of Trichosalpinx subgenus Xenia, which are extremely scarce in the wild but need to be phylogenetically evaluated to obtain a complete evolutionary scenario for the Lepanthes clade. Following morphological observations, we suspect that some members might be related to Lepanthopsis and allies but this hypothesis needs further evaluation. Then, it is desirable to increase sampling in other groups such as Lepanthopsis (mainly the Antillean species) and Tubella because of floral similarities. Our phylogenetic framework and methodological approach enables the discovery of useful traits for generic classifications and paves the way for more comprehensive assessments on generic delimitations of similar recalcitrant lineages based on DNA sequences and morphological characters to further improve the systematics of the subtribe.
Amplification, sequencing and alignment. The polymerase chain reaction (PCR) mixture, the primers for the nrITS (17SE and 26SE) 45 and plastid matK (2.1 aF and 5 R) 9 regions and amplification profiles followed. Sanger sequencing of both regions was conducted by BaseClear (https://www.baseclear.com) on an ABI 3730xl genetic analyzer (Applied Biosystems, Foster City, California, U.S.A). Sequences were deposited in NCBI GenBank (Appendix S1). We used Geneious ® R9 (Biomatters Ltd., Auckland, New Zealand 46 ) for the editing of chromatograms and pairwise alignment. Sequences were aligned in the online MAFFT platform (Multiple Alignment using Fast Fourier Transform, http://mafft.cbrc.jp/alignment/server/) using default settings. We adjusted and trimmed the resulting alignment manually. The concatenated dataset (nrITS + matK) was built with Sequence Matrix v100.0 47 . When matK sequences were not available, they were included as missing data in the concatenated matrix.

Phylogenetic analyses. We analyzed the individual and concatenated datasets of nrITS and matK with
Bayesian inference (BI), maximum likelihood (ML) and maximum parsimony (MP) analyses. The model of evolution and the parameters were calculated using the Akaike Information Criterion (AIC) in jModelTest2 v2.1.7 48 . All analyses were run in the CIPRES Science Gateway V. 3.1 (http://www.phylo.org/sub_sections/ portal/) 49 . To evaluate the incongruence between plastid and nuclear datasets we followed the pipeline implemented 12 using the Procrustean Approach to Cophylogeny (PACo) application 27 in R (http://datadryad. org/review?doi=doi:10.5061/dryad.q6s1f). This procedure identifies potential conflicting outliers contributing to incongruent phylogenies. The matK sequences from the retrieved conflicting terminals were removed and replaced by missing data because inferences derived from plastid markers are usually more in conflict with morphological observations as compared with inferences derived from nuclear markers 50 . A new concatenated matrix was re-aligned using the cleaned matK dataset and then analyzed with BI, ML, and MP approaches. These analyses were contrasted with the original inferences from concatenated datasets.
To obtain ultrametric trees for the character evolution assessments we estimated the divergence times in BEAST v.1.8.2 using the CIPRES Science Gateway 49 . The clock-likeness of the data was tested by observing the coefficient of variation (CV) of relaxed clock models. Speciation tree model selection was achieved by executing the Bayes factor test on Yule process (Y), birth death-process (BD) and birth-death-incomplete sampling (BDIS) models under strict and uncorrelated lognormal molecular clock models. For each model, we assigned a normal prior distribution of 16.45 (±2.5 standard deviations) Mya to the root node of the Lepanthes clade and 12.93 (±2.5 standard deviations) Mya to the node of Zootrophion with the remainder of the members of the Lepanthes clade using the values calculated from the fossil-calibrated chronogram of the Pleurothallidinae 12 . We performed two MCMC with 50 × 10 6 generations and sampling every 1,000 generations with a Marginal likelihood estimation (MLE) of 50 path steps, 10 × 10 5 length of chains and log likelihood for every 1,000 generations. We inspected the convergence of independent runs size in Tracer v.1.6 as explained above. To compare the divergence time estimates among the speciation models (Y, BD, and BDIS) we used Bayes factors calculated with marginal likelihood using stepping stone sampling derived from the MLE path sampling.
Ancestral state reconstruction (ASRs). Ancestral state reconstructions were assessed with ML, stochastic character mapping (SCM), and BI using phylograms and ultrametric trees. For the ML approach, we explored the following models: equal rates (ER), symmetrical (SYM) and all rates different (ARD). We relied on the re-rooting method of Yang 62 and the function ACE implemented in the R-package phytools. The best-fitting model was selected by comparing the log-likelihoods among these models using likelihood ratio tests. Scaled likelihoods at the root and nodes were plotted in the time-calibrated consensus phylogenetic tree. For the stochastic mapping analyses based on joint sampling, we performed 100 replicates on 100 randomly selected trees (10,000 mapped trees) from the best fitting time-calibrated BEAST analysis. The trees were randomly selected using the R function samples.trees (http://coleoguy.blogspot.de/2012/09/randomly-sampling-trees.html). Results of transitions and the proportion of time spent in each state were calculated and summarized in phytools with the functions make.simmap and describe.simmap 61,63 . These analyses were performed following the scripts by Portik and Blackburn (2016) 64 . ML and BI analyses were executed in the program BayesTraits V3 [65][66][67] . To account for phylogenetic uncertainty, ancestral character estimates were calculated using a randomly sampled set of 1,000 trees from the post burnin sample of the 50,000 ultrametric trees obtained from the best fitting time-calibrated BEAST analysis as described above. We used the option AddNode for reconstruction of internal nodes of interest comprising every generic group of the Lepanthes clade and the root node. For the ML approach, we used the method Multistate with 10 ML attempts per tree and 20,000 evaluations in order to preliminary assess prior distributions. For the BI, we chose the method Multistate and MCMC parameters of 30,010,000 iterations, sample period of 1,000, burnin of 10,000, auto tune rate deviation and stepping stones 100 10,000. We used the method Reversible-Jump MCMC with hyper-prior exponential to assess the best fitting models in proportion to their posterior probabilities according to the MCMC approach. We chose the hyper-prior approach as recommended by Meade and Pagel 67 in order to reduce the arbitrariness when choosing priors. Therefore, we selected the option reversible jump hyper-prior exponential with prior distribution set according to the transition ranges obtained from a preliminary ML analysis 68 . The input files for BayesTraits V3 were partially constructed with Wrappers to Automate the Reconstruction of Ancestral Character States (WARACS) 69 . The BayesTraits outputs files were analyzed in R with the BayesTraits wrapper (btw) by Randi H Griffin (http://rgriff23.github.io/projects/btw.html) and other functions from btrtools and BTprocessR (https://github.com/hferg). The MCMC stationarity of parameters (ESS values > 200) and convergence of chains were checked in Tracer v1.6.0 and plotted in R with the packages coda 70 and the function mcmcPlots of BTprocessR. We reconstructed the ancestral states for all nodes of the tree and plotted the mean probabilities retrieved at each node with phytools.