Introduction

The orchid genus Bolusiella Schltr. was established in 1918 and is dedicated to the South African botanist Harry Bolus. It is well delimited both morphologically and geographically. Its six currently recognized taxa (five species and one subspecies) are known only from continental Africa and are easily recognizable by their equitant and fleshy leaves and dense inflorescence. While most of the taxa are widespread in tropical Africa, B. zenkeri (Kraenzl.) Schltr. is only known from the Upper and Lower Guinea domains, and Bolusiella iridifolia subsp. picea P.J.Cribb is confined to the Eastern Afromontane region.

The genus was the subject of preliminary phylogenetic analyses based on molecular data in two studies by Carlsward et al. (2006a, b), which suggested that Bolusiella was monophyletic, but sampling in the genus was limited. These two studies were based on both Bayesian inference and maximum parsimony and yielded identical phylogenetic hypotheses for the evolution of the clade, but due to sampling limitations (only three species: B. iridifolia (Rolfe) Schltr., B. zenkeri, and B. maudiae (Bolus) Schltr. were included), phylogenetic relationships among all the Bolusiella species could not be assessed, nor could the monophyly of its members be tested, even though all the species are morphologically well circumscribed (Verlynde et al. 2013). For example, B. iridifolia has leaves that are deeply sulcate, while those of all other species are not. Similarly, a spur is absent in B. fractiflexa Droissart, Stévart & Verlynde, but present in all other species, where it may either be found in the same plane as the lip (as in B. iridifolia subsp. picea) or completely curled under the lip (as in B. maudiae) (Online Resource 1).

The recent taxonomic revision of Bolusiella (Verlynde et al. 2013) helps to provide greater confidence in the species circumscriptions, allowing us to focus more specifically on the assessment of phylogenetic relationships among these species. The aim of the present study, therefore, was to use various analytical approaches, static homology, direct optimization (using POY version 5.1.2, Varon et al. 2010; Wheeler et al. 2014), and Bayesian inference to confirm the monophyly of Bolusiella and to assess phylogenetic relationships among its species.

Materials and methods

Taxonomic sampling

DNA was obtained from leaf- or floral-tissue samples taken from fertile specimens collected in the wild in Cameroon, Gabon, Guinea-Conakry, and Rwanda. Twenty Bolusiella accessions were sampled for DNA sequencing of one nuclear spacer region (ITS-1) and five plastid sequences (matK, rps16, trnL–trnF, trnC–petN and ycf1) (Online Resource 2). These twenty accessions represent all five species of the genus, but two accessions from the East African B. iridifolia could not be identified to the subspecies level because the voucher specimens lacked flowers.

To assess the monophyly of Bolusiella, sufficient outgroup sampling is necessary (see Darlu and Tassy 1993; Barriel and Tassy 1998). In this study, outgroups included Ancistrorhynchus clandestinus (Lindl.) Schltr., considered to be the sister group of Bolusiella in recent studies (Simo-Droissart et al. 2016), along with specimens of Angraecum bancoense Burg and Angraecum distichum Lindl., which are also included in subtribe Angraecinae, and finally a specimen of Polystachya albescens subsp. imbricata (Rolfe) Summerh., included in the same tribe as the Angraecinae (viz. Vandeae) but in a different subtribe (Polystachyinae). These outgroup taxa were sampled for all six regions.

DNA purification, PCR amplification, and DNA sequencing

The protocol adopted here followed Simo-Droissart et al. (2013), which focused on the phylogenetic study of Angraecum section Pectinaria. The following primers were used for amplification and sequencing of each individual plastid region: Tab-C and Tab-D for the trnL intron and Tab-E and Tab-F for the trnL–trnF intergeneric spacer (Taberlet et al. 1991), rps16-1F and rps16-2R for the rps16 intron (Oxelman et al. 1997), 19F (Molvray et al. 2000), 1326R (Cuénoud et al. 2002), 390F (Cuénoud et al. 2002) and trnK-2R (Johnson and Soltis 1994) for matK, trnC and petN-1R for the trnC–petN intergenic spacer (Lee and Wen 2003), and 3720F, IntR, IntF, and 5500R for ycf1 (Neubig et al. 2009). The nuclear marker ITS-1 was amplified using ITS-A, ITS-B, ITS-C, and ITS-D, designed for angiosperms by Blattner (1999).

Leaf or floral tissue was dried in silica gel for DNA extraction (Chase and Hills 1991). Total DNA was extracted from fresh (1 g) or dried material (0.3 g) using one of two methods. The first method used 1 g of fresh leaves in a modified 2× CTAB protocol (Doyle and Doyle 1987). Proteins were removed with SEVAG (chloroform/isoamyl alcohol 24:1), and DNA was precipitated with ethanol (− 20 °C). At the end of the extraction protocol, turbid or colored DNA extracts were purified further on Macherey–Nagel columns. For some samples, an alternative extraction method used 0.3 g of dried material with the NucleoSpin® plant kit from Macherey–Nagel, following the manufacturer’s protocol.

PCR amplifications were carried out using Biometra TProfessional thermocyclers (PTC–100 or PTC–200; Bio-Rad Laboratories, Inc.) in total volumes of 25 μL, each reaction containing 1–2 μL of template DNA (of unknown concentration), 0.12 μL (5 U/μL) of Taq polymerase (Qiagen), 2.5 μL PCR buffer, 1 μL MgCl2 (25 mM), 0.5 μL dNTPs (10 μM), 0.25 μL of each primer (10 μM), and 18.37–19.37 μL of H2O. The PCR amplification profiles used for the trnL–trnF region, trnC–petN, the rps16 intron, and ITS-1 consisted of an initial denaturation at 94 °C for 3 min followed by 30 cycles of 30 s at 94 °C, 30 s at 52 °C, and 1 min at 72 °C, with a final extension at 72 °C for 10 min. Amplification of matK (19F–1326R and 390F–trnK2R) and ycf1 (3720F–intR and intF–5500R) involved an initial denaturation at 94 °C for 3 min followed by 30 cycles of 30 s at 94 °C, 30 s at 52 °C, and 1 min 30 s at 72 °C, with a final extension at 72 °C for 10 min. PCR products were purified by enzymatic digestion using ExoSAP (Qiagen).

Cycle sequencing was carried out using BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Inc., ABI, Lennik, the Netherlands) with the same primers used for PCR amplification: 1.5 μL of sequencing buffer, 1 μL of BigDye terminator with 0.2 μL of 10 μM primer, 1–3 μL of amplified product (of unknown concentration), and 4.3–6.3 μL of H2O, for a total reaction volume of 10 μL. Cycle sequencing conditions were as follows: a premelt of 1 min (96 °C), 25 cycles each with 10 s of denaturation (96 °C), 5 s of annealing (52 °C), and 4 min of elongation (60 °C). Cycle sequencing products were purified through ethanol precipitation. Sequences were generated on an ABI 3100 following the manufacturer’s protocols (ABI). Both strands were sequenced to assure accurate base calling.

Sequence editing

Complementary and overlapping sequences were assembled using CodonCode Aligner (version 4.2.7, CodonCode Corporation). Each individual base position was examined for agreement between the two to four contigs from both strands. Consensus sequences were edited manually.

Phylogenetic analyses

We used the molecular phylogenetic study of Angraecum section Pectinaria as a template for Bayesian and static homology methodology (Simo-Droissart et al. 2013). For each of the following analytical approaches, phylogenetic analyses were conducted first for each region separately (i.e., ITS-1, matK, rps16, trnC–petN, trnL–trnF, and ycf1). Following this, a single concatenated dataset for the five plastid regions was analyzed (five-marker plastid dataset), and then all six regions (plastid and nuclear) regions were analyzed in a single combined dataset (six-marker combined dataset). We therefore used eight different datasets (six individual markers and two combined datasets).

In addition, given its recent history, Bolusiella was considered as a useful model for testing the utility of various analytical approaches, static homology, direct optimization (using POY version 5.1.2, Varon et al. 2010; Wheeler et al. 2014) as well as Bayesian inference. In Simo-Droissart et al. (2013), the choice was made to code indels as missing data. In this study, indels were coded both as binary presence/absence characters and as missing data in the parsimony analyses. The genus has a relatively small number of species, allowing easy manipulation of datasets, and previous molecular studies provide preliminary evidence of its monophyly.

Finally, concerning the influence of gaps, transitions, and transversions, dynamic homology analyses performed under equal weighting (indel:Tv:Ts=1:1:1) minimized character conflict and yielded the lowest ILD values (Table 1). Therefore, results from this weighting scheme are reported below for parsimony and direct optimization analyses.

Table 1 Total evolutionary steps reported for each optimal tree inferred from the different parameter sets with dynamic homology
Full size table

Dynamic homology approach

Direct optimization (Wheeler 1996; Gladstein and Wheeler 1996; D’Haese 2004) is a method of phylogenetic analysis that does not rely on a priori multiple alignments. Instead, it is based on a dynamic approach (hence, dynamic homology) in which both nucleotide substitutions and insertion–deletion events are considered evolutionary events to be optimized simultaneously to derive the best trees, without reference to a preexisting alignment (Wheeler et al. 2006). Despite the potential for this method, its application has been somewhat limited, but has included several animal phyla, such as Arthropoda (e.g., D’Haese 2002, 2003; Giribet et al. 2001; Arango and Wheeler 2007) and Annelida (Worsaae et al. 2005), and in the orders Squamata and Chiroptera (Frost et al. 2001; Giannini 2003, respectively). Dynamic homology has not been widely used in angiosperms. We only found ten studies, of which three involved dicots (Gottlieb et al. 2005; Weese and Johnson 2005; Pedraza-Peñalosa 2010), five monocots, mainly Poaceae (Lehtonen and Myllys 2008; Souto et al. 2006; Cialdella et al. 2007, 2010; Petersen et al. 2011) and two broader studies including both dicots and monocots (Aagesen 2005; Catalano et al. 2009). To date, dynamic homology has not been used on any dataset concerning the Orchidaceae family.

Phylogenetic analyses using the dynamic homology framework (Wheeler 1996) were performed using the parsimony criterion as implemented in POY version 5.1.2 (Varon et al. 2010; Wheeler et al. 2014). Each analysis was run for nine different transformation-cost regimes (Sankoff matrices for indel, transversion, and transition costs) to test the stability of the results. The influence of gap/transversion and transition/transversion costs was explored through sensitivity analysis (Wheeler 1996) to avoid an arbitrary choice of parameters. Three indel/transversion cost ratios (1, 2 and 4) and three transversion/transition cost ratios (1, 2 and 4) resulted in nine individual analyses for each combination (matK, ycf1, rps16, trnL–trnF, trnC–petN, ITS-1, plastid dataset, and combined dataset). For each of eight datasets combined with nine Sankoff matrices, the analytical procedure followed a three-step strategy: In step 1, a starting pool of 1000 Wagner trees was generated through random addition sequence (RAS). Each replicate was explored by a combination of TBR and SPR branch swapping and then subjected to parsimony ratcheting (Nixon 1999). The resulting topologies were then explored by tree fusing (Goloboff 1999). A final, more thorough branch swapping was performed by retaining trees up to 10% longer than the optimal ones, then retaining only the optimal trees as the result of this step. In step 2, for each of the eight datasets, trees resulting from the nine different transformation-cost regimes were concatenated for a supplementary tree-fusing step. Finally, in step 3, an additional round of tree fusing using iterative pass optimization (Wheeler 2003) for final refinement based on trees obtained for the previous two steps of the analyses.

Character congruence was used as an optimality criterion to choose the parameter set that maximizes congruence among loci. Congruence was measured by the incongruence length difference (ILD) metrics (Mickevich and Farris 1981). This value is calculated by dividing the difference between the overall tree length and the sum of its data components: (length combined–length individual sets)/length combined. The tree from the analysis that minimizes character conflict among all data is taken as the best overall explanation of character variation, and thus the best estimate of the phylogeny.

Approach based on aligned sequences

Consensus sequences were aligned with the CLUSTAL plugin (Larkin et al. 2007) implemented within Geneious (version 4.8.5, Biomatters), with default settings (Online Resource 3).

Static homology—Maximum parsimony (MP) analyses were performed using POY, under the equal weighting scheme (indel:Tv:Ts=1:1:1). Indels were coded in two different ways, as binary characters (present/absent) and as missing data. Heuristic searches were performed using tree bisection-reconnection (TBR) branch swapping, with 1000 replicates of random-taxon addition sequence, holding ten trees at each step. In a second round of analysis, we used all trees found in the tree-limited analysis as starting trees, with a limit of 10,000 trees, which were then swapped to completion.

Bayesian analysis—Bayesian analyses were performed using the MrBayes (Ronquist and Huelsenbeck 2003; Ronquist et al. 2012) module implemented within Geneious version 4.8.5 (Kearse et al. 2012) on the combined matrix, with one partition per gene (six partitions in total). Analyses were run for 2,000,000 generations with four chains (default temperatures) using a model-jumping approach that allows sampling across the entire general time reversible (GTR) model space (i.e., no best-fitting models were defined a priori; Huelsenbeck et al. 2004) and with model parameters unlinked between partitions. We followed the protocol of Simo-Droissart et al. (2013) where trees were sampled every 500 generations, resulting in a total of 4001 trees per run from which the first 500 (12.5%) were discarded as the burn-in phase. In order to avoid any bias, we checked empirically, within the MrBayes module in Geneious, that the plateau was reached at 2,000,000 generations by running the analyses with 10,000,000 generations. In the same fashion, we checked the length of the burn-in phase and adapted the amount of trees to be discarded to avoid discarding too many alternative solutions.

Clade support values—Levels of internal support were estimated for both the dynamic and static homology approaches, using two methods, the bootstrap protocol (Efron 1979; Felsenstein 1985) with 1000 replicates and the decay index (Bremer 1988; Donoghue et al. 1992), both of which were calculated from the aligned data (homology hypothesis implied for the optimal topology) using POY, building, and swapping 1000 trees from the optimal tree.

Results

Monophyly of the genus and species

When markers were analyzed individually, Bolusiella was found to be monophyletic in all of the resulting topologies except for ycf1 (when analyzed under static homology with indels coded as binary characters and dynamic homology approaches, respectively, Online Resource 4e–f), trnL–trnF (when analyzed under dynamic homology approaches, Online Resource 4i), and for rps16 sequences (analyzed using dynamic homology, Online Resource 4q).

Species were also generally found to be monophyletic, except for B. maudiae when ITS-1 sequences were analyzed under static homology approach, for B. iridifolia with ycf1 sequences, and B. talbotii with rps16 sequences analyzed under static homology with indels coded as binary characters and dynamic homology approaches (Table 2; Online Resource 4a–t). However, resolution and branch support among these species are clearly insufficient to recover interspecific relationships within Bolusiella based on any individual plastid dataset, or even when only the five plastid sequences are combined (Fig. 1; Online Resource 4a–c, e–g, i–k, m–o, and q–s). Yet, when all sequences are combined, Bolusiella along with its species is found to be monophyletic, and analyses yielded well-resolved trees with high branch support (Fig. 2).

Table 2 Trees topology specifics
Full size table
Fig. 1
figure 1

Phylogenetic trees resulting from nuclear marker ITS-1 dataset search: a dynamic homology analysis (strict consensus tree with bootstrap percentages shown above or below branches); b static Homology analysis (maximum parsimony), with indels treated as characters (strict consensus tree with bootstrap percentages shown above or below branches); c static homology analysis (maximum parsimony), with indels treated as missing data (strict consensus tree with bootstrap percentages shown above or below branches); d Bayesian analysis (strict consensus tree with posterior probability values shown above or below branches)

Full size image
Fig. 2
figure 2

Phylogenetic trees resulting from six-marker combined dataset (ITS-1, matK, rps16, trnL–F, trnC–petN, and ycf1) search: a dynamic homology analysis (strict consensus tree with bootstrap percentages shown above or below branches); b static homology analysis (maximum parsimony), with indels treated as characters (strict consensus tree with bootstrap percentages shown above or below branches); c static homology analysis (maximum parsimony), with indels treated as missing data (strict consensus tree with bootstrap percentages shown above or below branches); d Bayesian analysis (strict consensus tree with posterior probability values shown above or below branches)

Full size image

Phylogenetic relationships

When individual markers were analyzed separately using dynamic homology, there was insufficient resolution to recover the monophyly of Bolusiella or to assess interspecific relationships. Three of the six analyses of single regions (trnL–trnF, rps16 and ycf1) placed outgroup taxa among Bolusiella (Online Resource 4e–l&q). Analyses for rps16 show insufficient differences between sequences to resolve relationships among all angraecoid specimens (Online Resource 4q). However, in combination, these markers yielded well-resolved trees with high support values (bootstrap (BS) = 81–100%). Within the Bolusiella clade, B. talbotii represented the earliest diverging lineage, sister to a clade comprising all other species (BS = 99%). Inside this latter clade, B. fractiflexa + B. maudiae (BS = 100%) are sister to a clade uniting B. iridifolia and B. zenkeri (BS = 99%) (Fig. 2a).

In the results from the parsimony analyses, differences occur depending on whether indels were coded as missing or binary (presence/absence) characters. In the parsimony trees, B. iridifolia appears as the earliest diverging lineage, sister to the B. talbotii and B. zenkeri group, and to B. maudiae and B. fractiflexa group (Fig. 2c). The topology of the tree resulting from the six-marker combined dataset analysis where gaps are coded as binary characters is not the same as the topology of the tree resulting from the analyses of the six-marker combined dataset where gaps were coded as missing data. However, giving the support levels in the former, the two trees are actually very similar, the latter resolving B. talbotii as the sister group to B. zenkeri while the relationships between B. talbotii, B. zenkeri, and the B. fractiflexa + B. maudiae group are not resolved in the former. While the bootstrap values of the internal nodes of the genus are rather low, support values for the monophyly of individual taxa are still well supported (BS = 99–100%) (Fig. 2b).

The six-marker combined dataset Bayesian analysis yielded a well-resolved tree with high posterior probabilities (PP = 1). Within the genus, B. iridifolia is the earliest diverging lineage, sister to a clade uniting B. talbotii and B. zenkeri, and sister to B. fractiflexa and B. maudiae. The topology of this tree is identical to the one obtained using the nuclear marker (ITS-1) alone (Figs. 1, 2d).

Discussion

For five of the eight datasets used in our study, the implied alignment generated by POY resulted in shorter trees than the analyses based on CLUSTAL alignments (with the exception of ycf1, the five-marker plastid dataset, and the six-marker combined dataset). This result stands in contrast to what Weese and Johnson (2005) described for their study of the genus Saltugilia (V.E. Grant & A.G. Day) L.A. Johnson (Polemoniaceae).

Analyses of ITS-1 resulted in the same topology regardless of the method used. This includes the monophyly of Bolusiella, as well as each of its species (with the exception of B. maudiae using the static homology analyses). The four ITS-1 trees also agree in the placement of B. iridifolia as the earliest diverging lineage in the genus, and the sister-group relationship of two pairs of species, B. talbotii + B. zenkeri and B. maudiae + B. fractiflexa (Fig. 1). Trees resulting from the ITS-1 dataset are also congruent with the trees based on parsimony and Bayesian analyses of the six-marker combined dataset. The agreement between ITS-1 and the six-marker combined dataset suggests that the highest rate of phylogenetically informative characters is derived from ITS-1 and that the plastid markers are much less variable (Álvarez and Wendel 2003). This interpretation agrees with the lack of resolution in the trees based on separate plastid analyses. In the molecular study of continental African species of Angraecum section Pectinaria (Simo-Droissart et al. 2013), the marker with the highest rate of PICs (Phylogenetically Informative Characters, a nucleotide character state shared by two or more taxa) is ITS-1 (19.3%), with nearly twice as many as the plastid markers (an average of 10% for each plastid markers). In our case, the rate of PICs in ITS-1 is 13.56% (99 PICs), whereas the average rate for the plastid markers is 8.16%.

When the ITS-1 dataset is analyzed using dynamic homology, the resulting tree is congruent with the ITS-1 trees obtained using other methods, but not with the trees from the analyses of the six-marker combined dataset (Figs. 1, 2). Moreover, the different analyses of the five-marker plastid dataset do not yield congruent results and do not match the trees obtained using the nuclear marker (Online Resource 4u–x). Despite this, both dynamic and static homologies of the five-marker plastid dataset (coding indels as binary characters) result in identical topologies.

When using the six-marker combined dataset, the earliest diverging lineage of the Bolusiella clade differed in trees based on static homology and Bayesian approaches compared to those derived using dynamic homology. In the Bayesian tree, the lineage leading to B. iridifolia is the earliest diverging one, but in the dynamic homology tree, it is B. talbotii. The topology where B. iridifolia is the earliest diverging lineage was also found in previous molecular phylogenies (Carlsward et al. 2006a, b) based on a much broader sampling of angraecoid orchids (but only three species of Bolusiella). All these observations suggest that the basal placement of the B. iridifolia lineage is the most probable phylogenetic hypothesis.

Indels: valuable or missing data?

Similar to the results of Simo-Droissart et al. (2013), our study shows the same tree topology when analyzing the six-marker combined dataset using both indel treatments, except for the clade uniting Bolusiella talbotii and B. zenkeri, which is not supported when indels were coded as a binary character (Fig. 2b). Furthermore, the lengths of the trees obtained when indels are coded as missing data are logically always shorter than those obtained when indels are coded as discrete characters (Table 3). This can be easily understood because indels are supplementary and “valuable” characters, as measured by levels of homoplasy. However, once the sequences have been aligned, comparisons or homology hypotheses should apply to all positions, some of which may contain bases and gaps (Giribet and Wheeler 1999; Padial et al. 2014). That is, indels have become part of the pattern as much as any other nucleotide or amino acid. The pattern used to code characters for phylogenetic analysis, and consequently the putative recognition of transitions, transversions, and indels in DNA sequences, is the one created by the alignment, not the unaligned pattern that occurs in organisms (Simmons and Ochoterena 2000). This loss of information, when indels are coded as missing data, produces less resolved trees, and interspecific relationships are rarely resolved when plastid markers are analyzed individually (Table 2; Online Resource 4b–c, f–g, j–k, n–o & r–s), a conclusion similar to that of Heath Ogden and Rosenberg (2006). However, in our dataset for Bolusiella, indels did not represent valuable information when analyzing the six-marker combined dataset and therefore should be treated as missing data.

Table 3 Tree statistics for all parsimony analyses with tree lengths, consistency index, and retention index
Full size table

Conclusion

The aims of our study were to test the monophyly of each taxon conisdered and to assess phylogenetic relationships among them using the direct optimization protocol (Wheeler et al. 2014), as well as standard Bayesian and maximum parsimony analyses. While our study confirms the previously published taxonomy of Bolusiella (Verlynde et al. 2013), these results also show a difference in the evolution of plastid and nuclear regions. Results obtained from dynamic homology and those from standard phylogenetic approaches (parsimony and Bayesian approaches using an aligned dataset) allowed us to confirm the monophyly of Bolusiella and its species. The results also confirm that this genus represents five distinct species, notwithstanding the fact that the two subspecies of B. iridifolia were not clearly differentiated. The use of dynamic homology for this dataset was conclusive in that it provided an estimate of the phylogeny, albeit one that is slightly different from the one obtained with probabilistic methods. This method also provided a test of the monophyly of the taxa, but failed to recover interspecific relationships resolved with the other methods, probably because the nodes in question are being supported by a small subset of conflicting characters. Therefore, a broader phylogenetic analysis with more species representing different African angraecoid genera would be helpful to test further the utility of direct optimization in assessing interspecific relationships within the large orchid family.