Gesneriaceae is a mostly tropical or subtropical plant family with 150–160 genera and ca. 3200 species (Burtt and Wiehler 1995; Weber 2004; Skog and Boggan 2006; Weber et al. 2013). Pre-phylogenetic or traditional classifications had divided the family into three subfamilies (Gesnerioideae, Cyrtandroideae, and Coronantheroideae) (Wiehler 1983; Burtt and Wiehler 1995). Phylogenetic-based studies support the recognition of two subfamilies (Gesnerioideae and Cyrtandroideae) (Smith et al. 1997; Mayer et al. 2003; Weber 2004; Möller et al. 2009; Woo et al. 2011). More recently (Weber et al. 2013), a third subfamily (Sanangoideae) was added to accommodate the monotypic genus and subfamily that includes Sanango racemosum (Ruiz & Pav.) Barringer. Traditional classifications based on morphological data (e.g. Hanstein 1854; Fritsch 1893–1894; Wiehler 1983) have been shown to be artificial.

Recent molecular-based studies have redefined tribes to reflect monophyletic groups (Smith et al. 1997; Zimmer et al. 2002; Roalson et al. 2005; Woo et al. 2011; Weber et al. 2013). Weber et al. (2013) proposed a new comprehensive formal classification for the family based on molecular phylogenetic studies for New and Old World Gesneriaceae that resulted in a re-circumscription of the traditionally recognized tribes. For the neotropical Gesneriaceae the number of tribes was reduced from nine to the following five tribes: Beslerieae, Coronanthereae, Gesnerieae, Napeantheae, and Titanotricheae (Weber et al. 2013). Gesnerieae, the largest tribe in the family, includes what was previously classified in the following tribes: Gesnerieae, Gloxinieae, Episcieae, Sinningiae, and Sphaerorrhizeae (Weber et al. 2013). Coronanthereae (sensu Weber et al. 2013), previously recognized as subfamily Coronantheroideae by Wiehler (1983) and Burtt and Wiehler (1995), is supported in Gesnerioideae based on recent phylogenetic studies as the sister clade to all other New World Gesneriaceae (Woo et al. 2011; Perret et al. 2012; Weber et al. 2013).

Columneinae (tribe Episcieae sensu Wiehler 1983) is the largest and most diverse subtribe in the Gesnerioideae and in the family with 26 genera and over 700 species (Clark et al. 2006, 2012; Clark, 2009). The Columneinae is distinguished from other subtribes of the tribe Gesnerieae by its nodal anatomy and chromosome number. All members of the Columneinae have a distinctive three-trace trilacunar nodal anatomy with split lateral bundles, in contrast to the conventional one-trace unilacunar nodal anatomy present in members of the other tribes of Gesnerioideae (Metcalfe 1950; Wiehler 1978). Most subtribes in the Gesnerieae have a base chromosome number of x = 13 or 14 (Wiehler 1983; Skog 1984; Burtt and Wiehler 1995; Smith et al. 2004). The Columneinae is unique in the tribe because it has a base chromosome number of x = 9 (or x = 8 in Codonanthe (Mart.) Hanst. and Nematanthus Schrad.) (Wiehler 1978; Smith and Carroll 1997).

All genera in the Columneinae have been extensively evaluated (i.e. inclusion of generic type species and/or broad sampling) in a phylogenetic context except for Paradrymonia Hanst., Nautilocalyx Linden ex Hanst., and Chrysothemis Decne. Recent phylogenetic studies have resulted in major shifts that have redefined most of the genera in the Columneinae so that they represent monophyletic groups (Clark et al. 2006, 2012). Although Paradrymonia, Nautilocalyx, and Chrysothemis form a clade within Columneinae (Clark et al. 2012; Perret et al. 2012), the relationships and generic delimitations within this clade have been problematic because material for generic type species was not available, and taxon sampling was mostly limited to Andean species. For this study we conducted field expeditions to Central and South America to increase taxon sampling and include generic type species that were previously unavailable.

Many genera in the subtribe Columneinae (tribe Episcieae sensu Wiehler 1983) have at least one synapomorphy or distinctive morphological trait. For example, Columnea L. has an indehiscent fleshy berry, Glossoloma Hanst. has resupinate flowers, Drymonia Mart. has poricidal anther dehiscence, and Episcia has stolons.

In contrast, Paradrymonia, Nautilocalyx, and Chrysothemis have traditionally been defined by a combination of symplesiomorphic characters. Paradrymonia has been historically recognized by its epiphytic rosette forming habit, anisophyllous leaves, campanulate corollas, and semi-fleshy bivalved dehiscent capsules; Nautilocalyx by an obligate terrestrial habit and isophyllous leaves; and Chrysothemis by an obligate terrestrial habit and connate calyx lobes.

The species conventionally assigned to the genus Paradrymonia are widely distributed throughout the Neotropics, except southeastern Brazil and the Caribbean. The genus is especially diverse in northwestern South America with centers of diversity in Colombia and Ecuador. Unlike other genera in the Gesneriaceae, which are abundant in mid-elevation montane forests (800–1,500 m) (Perret et al. 2012), most species in Paradrymonia s. l. occur in lowland rainforests (< 800 m). One exception is P. metamorphophylla, which occurs in montane cloud forests up to 2,500 m. Most species of Paradrymonia have narrow geographic ranges and grow in primary wet to moist forests on moss-covered rocks, wet logs, or edges of waterfalls and streambanks.

Paradrymonia is currently recognized with 40 species (Weber 2004; Skog and Boggan 2007). It is the fourth largest genus in Columneinae after Columnea (200 + species.), Drymonia (100 + spp.), and Nautilocalyx (ca. 60 spp.) (Clark 2009). The generic definition of Paradrymonia has been historically confusing because taxonomists have at some time recognized 17 (45%) of the described species as homotypic synonyms in other genera. Paradrymonia was first described by Hanstein (1854) with only one species, P. glabra (Benth.) Hanst. The concept of Paradrymonia as a genus was short-lived, as Hanstein (1865) himself reduced it to a synonym of Episcia Mart. More than 100 yr later, Wiehler (1973) re-established Paradrymonia by transferring four species that were previously recognized in Episcia (section Paradrymonia (Hanst.) Leeuwenb.). Wiehler (1973) justified the re-circumscription and generic status by the presence of a rosette habit and oblong-lanceolate leaves.

The taxonomic challenges and non-monophyly of Paradrymonia are well-established in the literature (Smith and Carroll 1997; Clark and Zimmer 2003; Clark et al. 2006, 2012). Paradrymonia was recovered as monophyletic by Smith (2000); however, this study only included a limited sampling of three species. All other studies (Smith and Carroll 1997; Clark and Zimmer 2003; Clark et al. 2006, 2012) do not support the monophyly of Paradrymonia. The primary objectives of this study were to use additional molecular sequence data from an expanded taxon sampling to: (1) re-circumscribe a monophyletic Paradrymonia; (2) identify well-supported clades among the Paradrymonia segregates; (3) increase taxon sampling relative to previous studies to resolve relationships among Paradrymonia, Chrysothemis, Nautilocalyx, and closely related congeners; (4) to determine the placement of these genera within the subtribe Columneinae; and (5) evaluate diagnostic morphological characters of well-supported clades amongst currently recognized Paradrymonia species.

Materials and Methods

Taxon Sampling—A total of 80 ingroup species belonging to 15 out of 22 genera of the subtribe Columneinae were sampled for the phylogenetic analyses (Appendix 1). Paradrymonia, Nautilocalyx, and Drymonia were represented with 27, 23, and four species respectively. The remaining 13 genera were represented with one to three species each. Sinningia cooperi and S. incarnata were selected as outgroups based on previous phylogenies that strongly support the subtribe Ligeriinae (tribe Sinningieae sensu Wiehler 1983) as the sister group to Columneinae (Perret et al. 2003; Zimmer et al. 2002).

DNA extraction, amplification and sequencing—DNA was isolated from fresh leaves dried in silica gel. Extractions of DNA were carried out using a Qiagen DNeasy plant mini kit following the manufacturer's recommendations (Qiagen, Germantown, Maryland).

The internal (ITS) and external transcribed spacers (ETS) of the nuclear ribosomal cistron, as well as the plastid psbA-trnH intergenic spacer, were sequenced for this study. The ITS region has been shown to be phylogenetically useful for analyses at the species level because it is variable and highly repeated in the plant genome allowing for easy detection, amplification and sequencing (Baldwin et al. 1995). The ETS region was chosen because it shares the same favorable features of ITS and it has been shown to provide more variable and parsimonyinformative characters than ITS (Baldwin and Markos 1998; Linder et al. 2000; Markos and Baldwin 2001, 2002; Vander Stappen et al. 2003). Additionally, the combination of data from both regions has increased resolution and support (Markos and Baldwin 2001; Vander Stappen et al. 2003). The psbA-trnH intergenic spacer is considered among the most variable regions in the angiosperm plastid genome and it has been used in population genetic studies, in species-level phylogenies, and in DNA barcoding (Kress et al. 2005; Štorchová and Olson 2007; Yao et al. 2010).

The following primers were used to amplify these loci: (1) ITS4 and ITS 5 (Baldwin and Markos 1998); (2) 18S-ETS (Roalson et al. 2003) and ETS-B (Beardsley and Olmstead 2002); and (3) psbA-trnH (Sang et al. 1997). Promega Taq DNA polymerase kits were used to amplify DNA in accordance with the manufacturer's instructions (Promega, Madison, Wisconsin).

The PCR reaction conditions for amplification of the ITS and ETS regions were as follows: (1) initial denaturizing at 95°C; (2) 34 cycles of 1 min at 95°C, 1 min at 48°C, 1 min at 72°C; and (3) a 7-min extension at 72°C. The PCR reaction conditions for amplification of the psbA-trnH spacer were as follows: (1) initial denaturizing at 95°C; (2) 30 cycles of 1 min at 95°C, 30 s at 52°C, 30 s at 72°C; and (3) followed by a 7-min extension.

The PCR reactions were cleaned of unincorporated nucleotides and excess PCR primers using a PEG precipitation protocol prior to sequencing. Sequencing reactions were performed with ABI Prism BigDye terminator cycle sequencing ready reaction kits (PE Biosystems, Foster City, California) and were cleaned using standard ethanol/sodium acetate precipitation. Sequencing was carried out at on an ABI 3731 genetic analyzer. Sequence chromatograms were examined and edited using Sequencher version 3.1.1 (Gene Codes Corporation 1998). GenBank accessions with their corresponding voucher information are provided in Appendix 1. Sequence alignments of all the datasets for the phylogenetic analyses are available from TreeBASE (study number 14733).

Sequence alignment—Sequences were initially aligned using ClustalW version 2.0 (Larkin et al. 2007) with default settings and subsequently manually adjusted using Se-Al version 2.0a11 (Rambaut 2002), a multiple sequence editor. Automated sequence alignments for the nuclear regions (ITS and ETS) required few manual adjustments. In contrast, alignments for the plastid marker, psbA-trnH, required extensive manual editing.

Inversions for six species were detected for the psbA-trnH dataset. Each inverted region was replaced with its reverse complement, such that sequence homology was maximized across the alignment (Whitlock et al. 2010). Inversions are known to be frequent for the psbA-trnH region (Sang et al. 1997; Tate and Simpson 2003; Clark et al. 2006; Kårehed et al. 2008; Borsch and Quandt 2009) and they appear to be common in noncoding plastid regions associated with stem-loops in the RNA secondary structures (Kelchner 2000; Štorchová and Olson 2007). The most common explanation for these hairpin-associated inversions is a single mutational event (Kelchner 2000; Quandt et al. 2003; Kim and Lee 2005). Additionally, inversions were scored as a presence/absence character. Three regions with a combined total of 146 characters were especially variable and difficult to align and were therefore excluded from the analyses.

Phylogenetic Analyses—Data were analyzed separately (ITS, ETS, psbA-trnH) for parsimony analyses, and in combination under maximum parsimony (MP), maximum likelihood (ML) (Felsenstein 1973), and Bayesian (Yang and Rannala 1997) methods.

No incongruence was detected from analyses of individual markers based on the evaluation of strongly supported nodes in the tree topologies. Incongruence among data partitions was evaluated by comparison of bootstrap support for clades (>75%) and posterior probability values greater than 0.95. Since there were no strongly supported nodes in conflict the three datasets were combined in a total evidence approach (Kluge 1989) as suggested by Seelanan et al. (1997) and Wiens (1998).

The MP analyses were performed using the heuristic search option in PAUP* (Swofford 2003) with the following settings: 100 random addition replications with tree bisection and reconnection (TBR) branch swapping, and MULTREES option in effect with a maximum of 10,000 trees being retained. Bootstrap analysis (Felsenstein 1985) with 1,000 replicates and 10 random sequenced additions per replicate was conducted to assess support values for the MP analysis.

The best-fitting model of DNA substitution was determined using the Akaike information criterion (Akaike 1974) under ML settings as implemented by jModelTest (Posada, 2008). The best model of sequence evolution was the general time-reversible model (GTR+I+gamma) with some sites assumed invariant and with variable sites assumed to follow a gamma distribution. The following parameter estimates were obtained: (1) base frequencies of 0.2569 (A), 0.2172 (C), 0.2298 (G), 0.2960 (T); (2) substitution rates of 0.8679 (A-C), 2.5181 (A-G), 0.9418 (A-T), 0.7413 (C-G), 4.6177 (C-T), 1 (G-T); and (3) a gamma distribution shape parameter of 0.5030. A ML heuristic bootstrapping search was performed using the default parameters in RAxML 7.2.6 through the CIPRES portal (Stamatakis et al. 2008) for 100 replicates, repeated 10 times to generate 1,000 replicates, starting from random trees.

Bayesian posterior probabilities (PP) for the combined dataset were calculated using the Bayesian Markov Chain Monte Carlo algorithm (MCMC) of BEAST v1.8.0 (Drummond et al. 2012) to further investigate the impact of tree construction method on topology. The species tree prior was set to a Yule Process following author recommendations. We ran two independent runs of 50,000,000 generations and sampled every 5000 generations. Convergence was evaluated by comparing traces and posterior distributions of parameter estimates using Tracer (Rambaut and Drummond 2007), ensuring that the log-likelihood of the model had reached a plateau and the effective sampling size (ESS) of each parameter was at least 200 or greater. The two runs were then combined using LogCombiner after the first 25% of trees were discarded as part of the burn-in phase. The remaining trees were used to calculate posterior probabilities in a maximum credibility tree using TreeAnnotator (Drummond and Rambaut 2007) with a posterior probability threshold of 0.5 and median node. Trees were visualized in FigTree (Rambaut 2009).

Results

Phylogenetic analyses—The total evidence analyses (MP, ML, and Bayesian inference) resulted in trees with similar topologies. The same strongly supported clades were recovered in all analyses, and differences in topology occurred only in clades with low support. Therefore, support values (Bootstrap/posterior probability values) were mapped onto the parsimony strict consensus tree (Fig. 1). The MP, ML, and Bayesian analyses of the combined regions consistently revealed three highly supported main clades in the subtribe Columneinae (Figs. 1, 2 Supplementary Fig. S1): (1) the Guiana Shield clade (MP BS = 92%; ML BS = 97%; PP = 1.0); (2) the Core Columneinae clade (MP BS = 96%; ML BS = 98%; PP = 0.98); and (3) the Paradrymonia alliance clade (MP BS = 98%; ML BS = 98%; PP = 1.0).

Table 1 presents a summary of statistics for the MP analyses of the combined dataset and individual partitions. Among the individual partitions, the percentage of informative characters was highest in the ETS region (51%), while psbA-trnH had the lowest percentage of informative characters (14%, Table 1).

The strict consensus of the MP trees based on the plastid dataset (psbA-trnH) was mostly unresolved (Fig. S2), with only two pairs of species that were well-supported (BS MP > 90%) and congruent with the nrDNA datasets. Those species pairs are: Rhoogeton cyclophyllus + R. viviparus and Nautilocalyx sp. Mora 931 + Nautilocalyx sp. Mora 949 (Fig. S2). In contrast, the individual nuclear markers (e.g. ITS and ETS), as well as the combined analysis (Fig. 1), provided higher resolution and support than the plastid dataset.

The main difference between the topologies of the individual ITS and ETS consensus trees (not shown) is the placement of the Core Columneinae and Guiana Shield clades. The individual analysis of ITS resulted in a poorly supported branch (BS < 50%) that subtends the Guiana Shield and Paradrymonia alliance clades as sister groups. In contrast, the ETS and the total evidence analyses result in a strongly supported sister-group relationship (BS = 100%) of the Guiana Shield clade and the Core Columneinae + Paradrymonia alliance clades (Fig. 1). In addition, a sister-group relationship between the Core Columneinae and the Paradrymonia alliance clades is strongly supported by the MP strict consensus of the ETS marker (BS = 90%); and moderately to strongly supported for the combined analysis (MP BS = 78%; ML BS = 84%; PP = 0.98, Fig. 1).

The traditionally recognized Paradrymonia is strongly supported as polyphyletic (i.e. Paradrymonia s. 1. or Clade II in Fig. 1) based on the phylogenetic analyses of the combined datasets. Currently recognized Paradrymonia species were recovered as belonging to two clades in the Columneinae (Core Columneinae clade and three lineages in the Paradrymonia alliance clade; Figs. 1 & 2). The Paradrymonia alliance clade comprises seven strongly supported clades (herein referred to as clades I–VII in Figs. 1, 2). Among these clades, three contain species of Paradrymonia (clades I, II, and VII in Figs. 1, 2). Clade I (MP BS = 98%; ML BS = 98%; PP = 1) includes the type species (P. glabra), as well as P. ciliosa and P. campostyla. This small clade is strongly supported (MP BS = 98%; ML BS = 98%; PP = 1) as the sister group to the remaining species of the Paradrymonia alliance clade. Clade II is also strongly supported (MP BS = 95%; ML BS = 90%; PP = 1) and includes Nautilocalyx cordatus (= Centrosolenia hirsuta), Nautilocalyx coccineus (= C. coccinea), and Paradrymonia densa (= C. densa). Clade VI (Trichodrymonia clade) contains most of the traditionally recognized species of “Paradrymonia” and is strongly supported as the sister group to clade VI that contains the generic type species for Nautilocalyx (N. bracteatus).

The topology of the Bayesian maximum clade credibility (MCC) tree (Fig. 2) is more resolved than the topology of the MP tree (Fig. 1), but overall there were no strongly supported branches that are incongruent. Likewise, the ML tree (Fig. S1) is congruent with the MCC tree (Fig. 2). One main difference between the MCC (Fig. 2) tree in contrast to the MP tree (Fig. 1) is the placement of Centrosolenia poryphyrotricha (=Nautilocalyx porphyrotrichus). Its phylogenetic placement in the MCC tree is within Centrosolenia (Fig. 2) and it is unresolved near the base of the Paradrymonia alliance in the MP tree (Fig. 1). Relatively long branches (Fig. 2) were found in Cremersia platula (Guiana Shield clade) and Codonanthopsis ulei (Core Columneinae clade).

Phylogenetic utility of ITS, ETS, and psbA-trnH—ITS and ETS were the most informative regions among the three markers. The nuclear regions provided four times as many parsimony-informative characters as the psbA-trnH cpDNA marker (Table 1). Although both nuclear regions provided about the same number of informative characters (ITS = 283; ETS = 285), ETS yielded a greater number of phylogenetically informative characters per sequence length (Table 1). The sequence data from ETS and ITS provided more resolution and more parsimony-informative characters (Fig. S3) compared to psbA-trnH. Of the three markers, psbA-trnH resulted in the least number of variable and parsimony-informative characters and provided little support for clades within the subtribe Columneinae (Fig. S2).

Fig. 1.

Parsimony strict consensus tree from the total evidence analysis of three datasets (nrDNA ITS, nrDNA ETS, and cpDNA trnH-psbA) showing the three main clades within subtribe Columneinae and the principal monophyletic subclades within the Paradrymonia + Nautilocalyx + Chrysothemis clade (Paradrymonia alliance clade). MP bootstrap values are above the branches, and those derived from maximum likelihood (when nodes are shared) are below the branches. Only bootstrap values ≥ 50% are shown. Thickened branches indicate ≥ 0.9 Bayesian posterior probability values. An asterisk (*) following the species name indicates the species is the type species of the genus.

Fig. 2.

Maximum clade credibility (MCC) tree of the combined dataset (ETS, ITS and psbA-trnH) for 82 taxa obtained from the BEAST analysis. Node support is given as Bayesian posterior probability (PP): black circles at nodes denote PP = 1.0, grey circles denote 0.95 ≤ PP ≤0.99, white circles denote 0.90 ≤ PP ≤ 0.94. Unlabelled nodes denote PP < 0.90.

Table 1.

Summary statistics derived from the parsimony analyses of nuclear and plastid markers

Discussion

Comparison with other studies of Columneinae (=Episcieae) phylogeny—Nuclear markers (ITS and ETS) and the combined analyses are congruent in supporting the nonmonophyly of Paradrymonia, as suggested by earlier studies with more limited taxon sampling (Smith and Carroll 1997; Clark and Zimmer 2003; Clark et al. 2006, 2012).

The placement of the Guiana Shield clade as the sister group to all the members of the subtribe Columneinae is moderately supported in our analyses (BS MP = 78%, BS ML = 84%, PP = 0.99; Figs. 1, 2). The sister-group relationship of the Guiana Shield clade with the Core Columneinae + Paradrymonia alliance clades is congruent with Clark et al. (2006), who used a total evidence analysis based on ITS, psbA-trnH, and morphological data to produce a comprehensive phylogeny of the subtribe Columneinae (155 species and 21 out of the 22 genera). However, the support for the placement of the Guiana Shield as sister clade to all of the remaining members of Columneinae was low (BS < 50%). Clark et al. (2006) included more taxa compared to the current study (151 vs. 82), but their focus was on the Core Columneinae clade. In contrast, we sampled 35 (18%) more species representing the Paradrymonia alliance clade and included an additional nrDNA molecular marker (ETS). The addition of ETS and increased taxon sampling resulted in more parsimony-informative characters and increased support at deep branches relative to previous studies (Table 1). The utility of ETS significantly increases support by adding 285 (51%) parsimony-informative characters compared to 283 (37%) for ITS and 165 (31%) for psbA-trnH (Table 1).

In contrast to Clark et al. (2006), the placement of the Paradrymonia alliance and Guiana Shield clades in the present study are not congruent with Clark et al. (2012). The different topology between these two studies could be attributed to the use of different loci. Clark et al. (2012) used four plastid and four nuclear loci for the analysis of the subtribe Columneinae with ITS as the only mutual marker in both studies. Interestingly, many of cpDNA markers in Clark et al. (2012) were limited in parsimony-informative characters. For example, Clark et al. (2012) generated 8,714 base pairs from 14 loci for the Core Columneinae clade and only 383 (4.3%) were parsimony-informative (Table 2 in Clark et al. 2012). In contrast, in the present study the utility of ETS alone accounted for 1300 + base pairs and resulted in 285 (51%) parsimony-informative characters (Table 1). An additional significant difference is taxon sampling where only 20 species are shared between the current study and Clark et al. (2012). It should be noted that the sister-group relationship of the Core Columneinae + Guiana Shield clades in Clark et al. (2012) is weakly supported in the MP and Bayesian analyses (MP BS = 61; PP = 0.77) and strongly supported in the ML analysis (ML BS = 100). In contrast, the sister-group relationship in the present study of the Core Columneinae and the Paradrymonia alliance clades is moderately supported in all the analyses (MP BS = 78; MP BS = 84; PP = 0.99). It is recommended that future studies of the Columneinae include ETS because it provides more parsimony-informative characters relative to previous studies that depended on large quantities of cpDNA markers that resulted in relatively few parsimony-informative characters.

Analyses by Perret et al. (2012) with a broader emphasis in the Gesnerioideae placed the Paradrymonia alliance clade as the sister group to all of the remaining Columneinae. One potential reason for this incongruence could be attributed to the use of plastid markers in Perret et al. (2012) in contrast to the use of nuclear markers here. A number of studies have shown that different genomes (e.g. nuclear vs. plastid markers) could produce incongruent phylogenetic trees due to different historic signals or processes such as hybridization, paralogy, incomplete lineage sorting, and scarcity of nucleotide homologies among others (e.g. Sang et al. 1997; Slovák et al. 2014; Zhang et al. 2015). Likewise, incongruence may also result from errors such as misspecification of models or problems during the sequencing and the alignment that are difficult to discard (Zhang et al. 2015).

Alternatively, discordance in tree topologies between the two studies might be due to differences in taxon sampling. The Perret et al. (2012) study lacked taxa from the Guiana Shield and included only four species from the Paradrymonia alliance clade (Chrysothemis melittifolia, Chrysothemis pulchella, Trichodrymonia macrophylla, and Nautilocalyx aeneus). As a result of limited taxon sampling, comparisons at the level of Columneinae phylogeny in the present study are of limited value with Perret et al. (2012).

Analyses presented support the sister-group relationship between the Paradrymonia alliance clade and the Columneinae core clade (Figs. 1, 2). These two clades are more diverse than the Guiana Shield clade and have centers of diversity in the Amazon, northern Andes, and Central America.

The phylogenetic utility of plastid and nuclear DNA markers—The alignment of the psbA-trnH region was particularly challenging due to the presence of numerous inversions and indels. Three regions with a total of 146 characters were especially variable and difficult to align and were therefore excluded from the analyses. Many regions of indels were similar in unrelated taxa (i.e. species from different genera), indicating a high level of homoplasy. Sang et al. (1997) also noticed considerable homoplasy of indel characters for psbA-trnH within Paeonia (Paeoniaceae). In the present study, the remaining portion of the alignment included in the analysis was rather conserved and therefore of limited value in resolving relationships. Like Shaw et al. (2005), we conclude that psbA-trnH provides limited information relative to other markers and its use in future phylogenetic studies of the Gesneriaceae is not highly recommended.

Likewise, other phylogenetic studies of the Gesneriaceae have shown that the phylogenetic signal of psbA-trnH and other plastid markers are less informative than nuclear noncoding sequences. For example, in Clark et al. (2012), ITS yielded at least five times the percentage of phylogenetically informative characters as any single non-coding plastid region used in the analysis. Some regions like the trnL-F spacer and the trnL intron had 2% or fewer parsimony-informative characters per length of sequence (vs. 20% for ITS). Although Clark et al. (2012) included 14 markers, 10 of which were from cpDNA non-coding regions, the combined regions yielded less than 5% informative characters. In contrast, the present study yielded 35% of parsimony-informative characters using only three markers. The data presented here and in other studies support that when analyzed separately, cpDNA noncoding sequences result in poorly supported phylogenies and limited resolution within the subtribe Columneinae. Although some relationships remain to be resolved, in the present study the combination of ITS and ETS provided sufficient characters for robust results and resolution amongst closely related taxa. Therefore, the use of ETS for future analyses in the Gesneriaceae is encouraged.

Phylogenetic implications for the taxonomy of Columneinae and Paradrymonia s. l.—MP, ML, and Bayesian analyses for the combined ITS, ETS, and psbA-trnH datasets recovered similar phylogenies for subtribe Columneinae. Our results strongly support that Paradrymonia must be substantially re-circumscribed to reflect the phylogenetic results presented here (i.e. monophyletic genera). Formal nomenclatural changes regarding these findings are provided in the taxonomic section. Below we describe the clades presented in Figs. 1 and 2.

Guiana Shield clade—The Guiana Shield clade includes taxa that are endemic to that region in northeastern South America. This region is defined by the underlying geological formation known as the Guiana Shield and extends from western Colombia to eastern Brazil (States of Amapá, Pará, Roraima, and Amazonas), Venezuela (States of Delta Amacuro, Bolívar, and Amazonas), Guyana, Suriname, and French Guiana (Hollowell and Reynolds 2005). Interestingly, this region has a surprisingly large number of endemic genera of Gesneriaceae (Weber 2004). The restricted distribution and endemism of the Columneinae in this clade could be the outcome of geology, topography, climatic factors (Kelloff and Funk 2004; Antonelli et al. 2009; Antonelli and Sanmartín 2011), geographic barriers to dispersal (Avise 2000), and low intrinsic dispersal ability (Croteau 2010), among others. Out of the eight genera endemic to this region, seven belong to the Columneinae: Lembocarpus Leeuwenb., Rhoogeton Leeuwenb., Lampadaria Feuillet & L. E. Skog, Cremersia Feuillet & L. E. Skog (Feuillet and Skog 2003), and the recently described monotypic genera Christopheria and Pagothyra (Smith and Clark, 2013). The results presented here sampled all of the endemic genera except for Lampadaria and Christopheria. While the placement of Lampadaria remains uncertain, Christopheria (as Episcia xantha Leeuwenb. in Clark et al. 2012) was strongly supported as nesting in the Guiana Shield clade. Pagothyra consists of a single species, P. maculata, a climbing vine with isophyllous leaves and bell-shaped yellow flowers mottled with bright red spots. This taxon was first placed in the genus Episcia by Hooker (1890) and then transferred to Paradrymonia by Wiehler (1978). Recently, this taxon was recognized as a monotypic genus, based on the unique apomorphy of the ventral corolla lobe occluding the throat like a trap door (Smith and Clark 2013). A recent analysis (Clark et al. 2012) supports a sister-taxon relationship between Christopheria xantha and Pagothyra maculata.

Core Columneinae clade—The Core Columneinae clade includes Drymonia longifolia Poepp., which had been recognized as a member of Paradrymonia (Wiehler 1973) but was originally described in Drymonia by Poeppig (1840). Wiehler (1973) made the combination in Paradrymonia because he noted the presence of longitudinal slits instead of poricidal anther dehiscence. The presence of poricidal anther dehiscence in Drymonia longifolia was observed in the early stage of anthesis and then developed into longitudinal slits as the flower matured (Clark, pers. obs.). The shift from poricidal to longitudinal anther dehiscence has been observed for other species of Drymonia (Clark et al. 2006) and therefore makes it challenging to assign a definitive character state for anther dehiscence. Furthermore, the presence of isophyllous leaves, leafy calyx, and succulent display capsule are further evidence that this taxon does not belong in Paradrymonia. Results presented here and in previous studies (Zimmer et al. 2002; Clark and Zimmer 2003; Clark et al. 2006, 2012) strongly support the placement of Drymonia longifolia as a member of Drymonia.

Paradrymonia alliance clade—The Paradrymonia alliance clade is strongly supported (Fig. 1) in this study as well as in previous analyses (Zimmer et al. 2002; Clark and Zimmer 2003; Clark et al. 2006, 2012). Our results show that this clade comprises seven moderately to strongly supported lineages (clades I–VII in Fig. 1). Species relationships within these clades varied by analytical approach usually when the nodes were not strongly supported (BS values < 80%, PP < 0.8) or were unresolved.

Paradrymonia s. s. (subclade I)—Paradrymonia s. s. is strongly supported as monophyletic (BS = 98, PP = 1) and includes the generic type species, Paradrymonia glabra (Figs. 3A, B), as well as P. ciliosa (Figs. 3C, D) and P. campostyla. Interestingly, P. glabra was recognized by Wiehler (1973) as a synonym of P. ciliosa. Extensive fieldwork and an ongoing monographic revision of Paradrymonia suggest that these two taxa are not conspecific based on morphological and geographical data. Both species have creeping stems, clustered leaves, elongate and coriaceous leaf blades, fimbriate corolla lobes, and bearded anthers (Fig. 3F). Characters that differentiate Paradrymonia glabra from P. ciliosa include the following: P. glabra has white, straight corollas covered with white trichomes (Fig. 3A), while P. ciliosa has cream-white, angulated corollas covered with red trichomes (Fig. 3C). Additionally, P. glabra is endemic to the Venezuelan Andes while P. ciliosa is a widespread taxon that is distributed from Nicaragua to northeastern Brazil.

As defined here, Paradrymonia s. s. includes species with anisophyllous leaves, succulent, large elliptic-oblanceolate leaf blades with decurrent bases (Fig. 3E), petioles longer than the inflorescence, flowers with calyx lobes linear to lanceolate (Fig. 3), corolla lobes crenate to fimbriate with elongate fimbriations on lower lobe (Figs. 3B, D), and anthers with an apical tuft of trichomes (i.e. “bearded anthers”; Fig. 3F). The presence of bearded anthers in Paradrymonia s. s. is a distinctive synapomorphy that has not been previously recognized at the generic level and bearded anthers are not known to occur in other groups of Gesneriaceae. Although only two species included in the phylogeny have bearded anthers, other species such as Paradrymonia buchtienii, P. lutea, P. yatua, and P. tepui also have this distinctive character. It is predicted that these species belong to the same clade because they present bearded anthers and are vegetatively similar to other members of Paradrymonia s. s.

Fig. 3.

Paradrymonia clade (Paradrymonia s. s. in Paradrymonia clade from Fig. 1). A. Lateral flower of Paradrymonia glabra showing straight (non-angulate corolla). B. Front view of P. glabra showing fimbriations on lower corolla lobe. C. Lateral flower of P. ciliosa showing angulate corolla. D. Flower of P. ciliosa showing fimbriations on lower corolla lobe. E. Leaves of P. glabra. F. Bearded anthers of P. glabra. Images from John L. Clark. A, B, and D from J. L. Clark et al. 11331. B and C from J. L. Clark et al. 12490.

The sister taxon to Paradrymonia glabra and P. ciliosa is Paradrymonia campostyla, which lacks bearded anthers. In addition to lacking bearded anthers it also differs by the presence of elongate climbing stems with isophyllous leaves (vs. anisophyllous leaves and erect shoots), small oblongelliptic to oblong-ovate leaf blades with subcordate bases (vs. oblanceolate with decurrent bases), and petioles shorter than the flowers (vs. petioles longer than the flowers). We refrain from recognizing Paradrymonia campostyla as a different genus even though it is morphologically distinct from the core Paradrymonia clade based on the characters described above.

Centrosolenia (subclade II)—The Centrosolenia clade includes three species (Figs. 1, 2). Centrosolenia is an available name that includes Centrosolenia hirsuta, which is a synonym of Nautilocalyx cordatus and the generic type species for Centrosolenia Benth. The name Centrosolenia Benth. predates Nautilocalyx Linden ex Hanst., but the latter was conserved by Feuillet and Skog (1990).

Morphological characters that are shared among species in this clade include: (1) the presence of decumbent stems that root and branch at the nodes; (2) variegated leaf blades (especially along the midrib) (Fig. 4A) with asymmetrical bases and anastomosed tertiary venation; and (3) elongate narrow-tubular corollas (Fig. 4A). Some immature individuals of Paradrymonia densa have variegated leaves that lose their variegation at maturity. An important diagnostic character for Centrosolenia that distinguishes it from subclade V (Amazonian “Nautilocalyx”) and subclade VI (Nautilocalyx s. s.) is the presence of oblong-lanceolate to oblong-spathulate calyx lobes. The species of Nautilocalyx s. l. in subclades (V and VI) have calyx lobes that range from broadly ovate and leaf-like (Amazonian “Nautilocalyx”, Fig. 4B) to linearlanceolate (Nautilocalyx s. s., Figs. 4C, D). The ML and Bayesian analyses result in placing Nautilocalyx porphyrotrichus into Centrosolenia, but support is moderately low or weak (ML BS = 72; PP = 0.63; Figs. 2, S1). It is interesting to note that Leeuwenberg (1958) recognized six subsections in a classification of the genus Episcia, and included Nautilocalyx porphyrotrichus as Episcia porphyrotricha in section Episcia, subsection Centrosolenia. This species has bullate variegated leaves with white or yellow on the midrib, and red narrow-tubular corollas, characters common in other species in the Centrosolenia clade. For the reasons mentioned above, we propose to transfer Nautilocalyx porpyrothrichus to Centrosolenia. Interestingly, all of the species in the Centrosolenia clade represented in the phylogeny are from the Guiana Shield.

In addition to the species sampled here, it is predicted that other unsampled species also belong to Centrosolenia because they share a suite of characters such as similar habit (terrestrial herb), obovate to oblanceolate leaf shape, membranous leaf texture, crenate margins, oblong calyx lobes, and tubular corollas. These species are from the Venezuelan Guiana Shield (e.g. Nautilocalyx chimantensis, N. crenatus, N. orinocensis, and N. ruber) and the Amazon basin (e.g. N. paujiensis, N. pusillus, N. roseus, and N. vestitus). The present molecular analysis in conjunction with morphological characters and geographic distribution provide strong support for the Centrosolenia clade. The genus Centrosolenia is re-established here with 12 new combinations (see taxonomic treatment below).

Pictus clade (subclade III)—This subclade includes Nautilocalyx pictus and a recently discovered, undescribed species from Peru (M. M. Mora 949). Both species have in common the presence of flowers in axillary fascicles, acuminate linear-lanceolate bracts, elongate calyx lobes, creeping habit, elongate internodes, isophyllous leaves, and bullate leaf blades with crenate to crenulate margins. Taxonomic changes are not recommended for this clade until additional sampling includes more species of Nautilocalyx.

Chrysothemis clade (subclade IV)—This clade includes Chrysothemis pulchella (the generic type species of Chrysothemis), C. friedrichsthaliana, Nautilocalyx melittifolius (=Chrysothemis melittifolia), and N. panamensis (=Chrysothemis panamensis) (Figs. 1, 2). All members of this clade are obligate terrestrial herbs that occasionally form tubers.

Chrysothemis has been traditionally distinguished from other species in the subtribe Columneinae by the presence of connate calyx lobes, which make the calyx look like a cup with irregular or five-pointed rim (Fig. 4E). These calyces have been described as “water calyces” because they often retain water and secrete liquid that allow the immature buds to develop under an aqueous solution that protects the buds from floral herbivores (Burtt and Woods 1975; Endress 1996; Carlson and Harms 2007). Studies of C. friedrichsthaliana by Carlson and Harms (2007) suggest that the water calyx helps protect immature flower buds from a small species of moth (Alucita sp.) that is a floral herbivore.

The present phylogenetic analysis supports the expansion of Chrysothemis to include N. panamensis, and N. melittifolius. We also include new combinations for N. adenosiphon and N. colonensis in Chrysothemis based on the study of herbarium specimens. Although these species have rather relatively free calyx lobes, they share with all other species an obligate terrestrial habit with underground tubers and succulent stems. Other diagnostic characters for the Chrysothemis clade include the following: (1) elliptic leaves with crenate-dentate to dentate-serrate margins; (2) slightly asymmetrical (oblique) leaf bases; (3) cross-venulate to reticulate veins; and (4) semisucculent fully reflexed capsules when opened.

Amazonian “Nautilocalyx” clade (subclade v)—This clade includes most of the species traditionally placed in Nautilocalyx. The species in this clade are obligate terrestrial herbs with similarities to members of the Nautilocalyx s. s. clade such as isophyllous leaves and succulent stems. An important distinction between the Amazon “Nautilocalyx” clade and the Nautilocalyx s. s. clade is that the calyx lobes are overlapping at the base or folded in a conduplicate manner such that each lobe is appressed to an adjacent lobe and folded lengthwise with the margin curved upward (Fig. 4B). The calyx lobes are also leaf-like and cover the basal gibbosity of the corolla tube. Most species in this clade have tubular white or cream corollas. The corolla tube is usually 3 × or longer than the calyx lobes and is typically longitudinally sulcate with blunt ridges on both the dorsal and ventral surfaces (Fig. 4B). The Nautilocalyx s. s. clade has calyx lobes that are nearly free, linear-lanceolate, and the corolla spur is exposed (Figs. 4C, D).

Fig. 4.

Morphological characters in selected species from the Paradrymonia clade (from Fig. 1). A. Centrosolenia hirsuta (= Nautilocalyx cordatus, Centrosolenia clade) showing variegated leaves and elongate tubular corollas. B. Front view of Nautilocalyx ecuadoranus (Amazonian “Nautilocalyx” clade in Fig. 1) showing ovate calyx lobes and corolla tube longitudinally sulcate with blunt ridges on both, the dorsal and ventral sides. C. Nautilocalyx colombianus (Nautilocalyx s. s. clade in Fig. 1) showing trumpet-shaped corolla and lanceolate calyx lobes. D. Nautilocalyx erytranthus showing trumpetshaped corolla with reflexed lobes and lanceolate calyx lobes (Nautilocalyx s. s. clade in Fig. 1). E. Chrysothemis friedrichsthaliana (Chrysothemis clade) showing flower with fused calyx lobes. F. Trichodymonia pedunculata (= Paradrymonia pedunculata, Trichodrymonia clade) showing rosette habit and elongated leaves with decurrent base. G–H. Flowers showing absence of fimbriations on corolla lobes. G. Frontal view of the salverform corolla in Trichodrymonia ulei (= P. ulei). H. Frontal view of the trumpet-shaped corolla in Trichodrymonia aurea (= P. aurea). I. Lateral view of the hypocyrtoid corolla of T hypocyrta (= P. hypocyrta). Images from R. Myrh (A) and J. L. Clark (B–I).

Nautilocalyx s. s. clade (subclade VI)— This study is the first to include Nautilocalyx bracteatus, which is the type species for Nautilocalyx. This species is endemic to Colombia and occurs in the Departments of Antioquia, Chocó, and Santander. The inclusion of this species in the analysis is crucial for the circumscription of Nautilocalyx. The genus was established by Hanstein (1854) in an introduction on genera of Gesneriaceae in the New World, but 11 yr later he reduced it to a synonym of Episcia (Hanstein 1865). Bentham (1876) then divided the genus Episcia into six sections including sect. Nautilocalyx. Sprague (1912) resurrected the genus on the basis of the position of the ovules on one or both surfaces of the placenta. However, Wiehler (1978) found Sprague's character to be inconsistent and variable among some members of Episcieae.

The Nautilocalyx s. s. clade includes N. antioquensis, N. bracteatus, N. colombianus (Fig. 4C), and the recently described species of Nautilocalyx erytranthus (Fig. 4D). This clade is defined by a funnel- or trumpet-shaped corolla with spreading lobes, narrow lanceolate to linear-lanceolate calyx lobes with the dorsal lobe greatly reduced and recurved to accommodate the gibbosity at the base of the corolla, and a corolla tube that is less than 2 × longer than the calyx (vs. 3 × longer than the calyx in the Amazonian “Nautilocalyx” clade).

Most of the species in the Nautilocalyx s. s. clade are distributed in the Chocó biogeographic region in Colombia and Ecuador and Andes and inter-Andean valleys of Colombia. One exception is N. erytranthus (Fig. 4D), which occurs to the east of the Andes in the Amazon basin of Colombia and northern Ecuador.

Trichodrymonia clade—The Trichodrymonia clade is comprised of most of the traditionally recognized species of Paradrymonia and includes the generic type species Trichodrymonia congesta Oerst. (=Paradrymonia congesta (Oerst.) Wiehler). The Trichodrymonia clade is a strongly supported monophyletic group (MP BS = 88; ML BS = 90; PP = 0.94) and is the sister clade to Nautilocalyx s. s. (MP BS = 72; ML BS = 78 PP = 1.0). A major challenge for this project was locating the generic type species of Trichodrymonia because it was only known from the type specimen collected by Liebmann in 1842 in Chinantla, Mexico. After more than a 150-yr hiatus, the first author re-discovered this species on a 2010 collecting expedition to the type locality. The inclusion of Trichodrymonia congesta in the molecular phytogeny is necessary to unambiguously assign a genus name to this clade and to re-establish Trichodrymonia.

To account for the phylogenetic relationships and to accommodate this large segregate of Paradrymonia s. s., here we resurrect the genus Trichodrymonia and make 25 new combinations (see taxonomic treatment below). The species in this clade share many vegetative characters with those belonging to the Paradrymonia s. s. clade such as rosette habit (Fig. 4F), anisophyllous leaf pairs, and elongated oblanceolate leaf blades. An important difference between Trichodrymonia and Paradrymonia is the absence of fimbriations on the corolla lobes in the former (Figs. 4G–I) and presence in the latter (Figs. 4A–F). Additionally, bearded anthers are present in Paradrymonia (Fig. 3) and absent in Trichodrymonia. The following characters define Trichodrymonia: (1) facultative epiphytic herbs; (2) leaves clustered in an apical rosette (Fig. 4F); (3) maroon sulcate petioles; (4) corollas salverform (Fig. 4G) to trumpet-shaped (Fig. 4H), and occasionally hypocyrtoid (Fig. 4I); (5) anthers glabrous with longitudinal dehiscence; (6) leaf pairs usually anisophyllous; and (7) fruits a semi-fleshy bivalved dehiscent capsule.

Conclusions

Analyses of relationships among traditionally recognized Paradrymonia provide the most resolved phylogeny available for the genus. Paradrymonia as previously defined is taxonomically evaluated and recircumscribed. New combinations are provided that are consistent with the recognition of monophyletic genera based on the present phylogeny and inferred morphological synapomorphies. Paradrymonia is reduced to eight species (See taxonomic treatment), while the remaining ex-Paradrymonia species are placed in other genera. Chrysothemis is monophyletic, but the results presented here strongly support a broader circumscription that includes some species previously recognized in Nautilocalyx. Our results also show that Nautilocalyx is clearly paraphyletic, but the recognition of additional new genera is not recommended until more comprehensive taxon sampling is included, as well as additional loci. The Nautilocalyx s. s. clade comprises species mostly from Central America, northwestern South America (the Chocó Biogeographic region) and the northern Andes. Nautilocalyx s. s. is sister to Trichodrymonia and its species mainly differ from the ones comprising the Amazonian “Nautilocalyx” clade by the presence of infundibuliform or trumpet-shaped corollas (vs. tubular) and linear calyces (vs. leafy and ovate). Future studies with increased taxon sampling may support the segregation of the Amazonian “Nautilocalyx” as a genus. Centrosolenia is resurrected based on molecular and morphological evidence. Nautilocalyx needs further taxon sampling before taxonomic changes are proposed.

Taxonomic Treatment

In the present treatment an identification key to major clades and genera in the Paradrymonia alliance is provided. Trichodrymonia and Centrosolenia are resurrected to accommodate species that were previously in Paradrymonia. Descriptions for Chrysothemis, Paradrymonia s. s. and for the resurrected genera Trichodrymonia and Centrosolenia are included. An index of names, including the new combinations, with the accepted names in bold, is provided to facilitate referencing currently recognized species circumscriptions with previous treatments (Appendix 2).