Introduction

With 28,349 recognized taxa (The Plant List 2013), Orchidaceae are one of the largest and most diverse angiosperm families. Historically a great deal of controversy has surrounded the phylogenetic relationships in the family and its higher-level classification (Fay and Chase 2009). Starting with Chase et al. (1994), numerous molecular studies have provided an increasingly detailed phylogenetic framework for the orchids (e.g., Cameron and Chase 1999; Cameron et al. 1999; Cameron 2004; Carlsward et al. 2003, 2006b; Goldman et al. 2001; Górniak et al. 2010; Martos et al. 2014; Micheneau et al. 2008; Rakotoarivelo et al. 2012; Russell et al. 2010; van den Berg et al. 2000), and some authors have also incorporated morphological data to aid in the circumscription of subordinate taxa (see Mytnik-Ejsmont et al. 2013; Simo-Droissart et al. 2013).

With the publication of the most recent classification of Orchidaceae (see Chase et al. 2015), a great deal of progress has been made in understanding the phylogenetics of Epidendroideae, its largest subfamily, which currently comprises 16 tribes, of which Vandeae is the third most speciose. Within Vandeae, the angraecoid orchids are of particular interest because of their spectacular white, nectariferous, long-spurred flowers, which make them especially prized by hobbyists. However, they remain one of the most taxonomically problematic groups (Pridgeon et al. 2014). Historically, angraecoid species were divided into two subtribes: Aerangidinae (ca. 350 spp.) and Angraecinae (ca. 410 spp.) (see Summerhayes 1966). These two subtribes were distinguished from other groups based on rostellum shape and chromosome number. While both subtribes, as traditionally circumscribed, have since proven to be polyphyletic based on molecular analyses, together they form a well-supported clade now recognized as a more broadly defined subtribe Angraecinae (Carlsward et al. 2006b; Chase et al. 2015). The generic name Angraecum Bory (1804), on which the terms “Angraecinae” and by extension “angraecoid” are based, is the oldest within this group. Angraecum is both species-rich and morphologically diverse, with 221 currently recognized species (The Plant List 2013), about 75 % of which are endemic to Madagascar and the Mascarene Islands of La Réunion, Mauritius, and Rodrigues, while most of the remaining taxa occur in continental Africa. Recent studies have shown that Angraecum is polyphyletic (Carlsward et al. 2006a, b; Micheneau et al. 2008), with several species scattered across other African angraecoid genera or clades. The phylogenetic relationships of species from the Western Indian Ocean Islands have been examined in detail (see Micheneau et al. 2008), but this is not the case for members of the genus from continental Africa and the adjacent islands in the Gulf of Guinea.

Among the 19 sections recognized within Angraecum by Garay (1973), the last author who treated the group as a whole, section Dolabrifolia occurs exclusively in continental Africa and in the Gulf of Guinea Islands. Its members are easily identifiable by their laterally compressed and densely imbricate leaves with a groove on the upper surface, the lateral compression being a unique feature within the genus. This prompted Szlachetko and Romowicz (2007) to recognize Dolabrifolia as a distinct genus, an interpretation followed by Szlachetko et al. (2013) based on molecular studies (which included three of the five currently described species in the group) and morphological data. However, the relationships between the Dolabrifolia group and other groups generally regarded as belonging to Angraecum remain unclear, and a broader study is needed before any firm taxonomic decisions can be made.

In the study of Micheneau et al. (2008), the Dolabrifolia group appeared to be monophyletic, although only two of the five currently described species were included. They were placed as sister to the continental African group Angraecum sect. Pectinaria, a result supported by the phylogenetic study of Simo-Droissart et al. (2013). Despite sampling limitations, these recent studies lend support to the recognition of the Dolabrifolia group at the sectional level within Angraecum. Among its five published species, only four are currently recognized; the status of the recently described A. poppendickianum Szlachetko and Olszewski (2001) has been controversial because of its morphological similarity to A. distichum Lindl., under which it was regarded as a synonym by Govaerts in 2003 according to Govaerts et al. (2016) and by Droissart et al. (2006).

During the last two decades, many additional specimens of plants belonging to Angraecum sect. Dolabrifolia have been collected, and as new material has accumulated the morphological distinction between most species (in particular A. aporoides Summerh., A. distichum, and A. bancoense Burg) has become increasingly blurred, primarily reflecting significant overlap in the morphological variability of leaf and floral characters within and among species. Previously, Arends et al. (1980) had pointed out that A. bancoense and A. distichum share the same characteristic foliage and could not be distinguished based on vegetative traits. In turn, the lack of clarity in species delimitation presents problems for molecular phylogenetic studies that aim to clarify relationships among members of the section and to test the delimitation of morphologically defined species. In order to rectify this situation, we have taken a three-step approach in this study. First, we investigated species delimitations in A. sect. Dolabrifolia by visual inspection to define coherent morphological groups. For that, we used the available collections and included nomenclatural types, applying the appropriate species name to each distinct group (when available) or otherwise treating it as a potentially new entity. Second, we used a morphometric approach to identify the most informative quantitative and qualitative traits for defining and differentiating these species, especially A. distichum and A. bancoense. Finally, we investigated the delimitation of each species using DNA sequences from six markers (obtained from one nuclear and five plastid regions).

Materials and methods

Data matrix construction for morphometric analyses

Dried and spirit-preserved specimens representing all described species of Angraecum sect. Dolabrifolia were examined from the following herbaria: BM, BR, BRLU, K, MA, MO, NY, P, WAG, and YA (acronyms according to Thiers (continuously updated)). Specimens were first grouped based on overall morphology. The groups obtained were assigned when possible to one of the four currently accepted species after comparison with the diagnoses provided in the protologue and with nomenclatural types. Herbarium specimens were used only for the initial recognition of species. Following this step, data for the morphometric analyses were collected only from alcohol-preserved material (from BR, BRLU, K, MO, P, and WAG; see Online Resource 1) because flowers on herbarium specimens are typically flattened and/or deteriorated, which obscures key features and precludes accurate measurements for morphometric analyses.

After grouping based on morphology, it was found that four of the five distinct groups corresponded to the following four accepted species: A. aporoides, A. bancoense, A. distichum, and A. podochiloides Schltr. In addition, we identified a fifth group that was vegetatively similar to A. aporoides (these two entities have the longest leaves in the section) but differs in by flower length and the shape of the leaf apex. This fifth group is hereafter referred to as A. aff. aporoides. While the type specimens of all four accepted species were used to help circumscribe the initial species, three of them could not be used in the morphometric analysis. The type of A. distichum is represented only by a drawing, while those of A. bancoense and A. podochiloides are preserved only on herbarium sheets.

After excluding spirit-preserved specimens for which one or more key vegetative or floral characters were missing as well as those that were too brittle to dissect, morphological measurements were taken from 172 high-quality fertile specimens preserved in alcohol (including the type material of A. aporoides) to perform multivariate analyses (Online Resource 1). Most of the fertile specimens (148 out of 172) were recently collected by our team either in the field or in a shadehouse cultivation system operated in Central Africa since 1997 (see Droissart 2009; Simo 2014; Stévart 2003). Thirteen quantitative characters (Table 1) were measured with graph paper (1 mm2 grid) and standardized by subtracting the character mean and dividing by the standard deviation to eliminate the distorting effects of different scales of measurement on the output results (Cupido 2003; Marcysiak et al. 2007; Poulsen and Nordal 2005). Two qualitative characters (Table 1) were also recorded as factors (i.e., with each discrete character state).

Table 1 List of variables assessed for the study of Angraecum sect. Dolabrifolia
Full size table

Data analysis

All morphometric analyses were performed using the R 3.0.1 software package (R Core Team 2013). To investigate the morphological variation within and among species, an extended principal component analysis (PCA) (Hill and Smith 1976) that included multistate discrete characters (i.e., a multivariate analysis allowing mixed quantitative variables and factors) was performed using the function dudi.hillsmith of the library ade4 (Chessel et al. 2004; Dray and Dufour 2007; Dray et al. 2007). The principal components of this analysis are centered and normalized vectors maximizing the sum of squared correlation coefficients with quantitative variables and correlation ratios with factors. Using the function dist of the package stats, with the “Euclidean” method, we obtained a distance matrix to compute the distances between individual objects (samples) of our data matrix (Borg and Groenen 1997). We used the function hclust of the package stats to perform a hierarchical cluster analysis using a set of dissimilarities for the 172 individuals being analyzed. Initially, each individual was assigned to its own cluster and then the algorithm proceeded iteratively, at each stage joining the two most similar clusters, continuing until there was just a single cluster (Sneath and Sokal 1973). The clustering method used was the Ward’s minimum variance, where the criterion for choosing the pair of clusters to merge at each step is based on the optimal value of an objective function, such as the error sum of squares. The Ward’s minimum variance method is designed to find compact, spherical clusters.

Prior to performing statistical tests among groups obtained after the extended PCA, the distribution of each quantitative variable was examined using the Shapiro–Wilk test of normality (Royston 1982) through the function shapiro.test available in the package stats (R Core Team 2013). None of the 13 quantitative variables followed a normal distribution. We therefore performed nonparametric Kruskal–Wallis tests (Hollander and Wolfe 1973) to assess significant differences between the five groups obtained after the extended PCA for each variable, using the kruskal.test function available in the package stats. When a character differed significantly among groups, we performed multiple comparison tests between groups using the function kruskalmc (Siegel and Castellan 1988) available in the package pgirmess (Giraudoux 2013), with pairwise comparisons adjusted appropriately. The function multcompLetters (Piepho 2004) of the package multcompView (Graves et al. 2012) was then used to convert a logical vector into a character-based display in which common characters identify levels or groups that are not significantly different.

Plant material and DNA purification

DNA was obtained from leaf tissues taken from fertile specimens collected in Guinea-Bissau, Ivory Coast, São Tomé and Príncipe, Cameroon, Gabon, Democratic Republic of the Congo and Rwanda (Online Resource 2). Plants that were not fertile at the time of collection in Cameroon and Gabon were cultivated and monitored in shadehouses until they produce flowers, enabling accurate identification. Additional leaf and flower material was provided from the Gabonese orchid collection initially established at the Wageningen University Greenhouse (Netherlands) and now housed in the greenhouse of the Botanic Garden Meise (Belgium). A total of 37 accessions were used in the study: ten for Angraecum aporoides, two for the potential novelty A. aff. aporoides, four for A. bancoense, and five for each of the two remaining species in section Dolabrifolia, viz., A. distichum and A. podochiloides; one from each of the five currently recognized species of A. sect. Pectinaria from continental Africa; one from the genus Diaphananthe (continental Africa); and two from the genus Tridactyle (continental Africa). Unlike for the morphometric study, we did not attempt to include the type collection of A. aporoides (Cooper 82/3) in the phylogenetic analysis because it comprises only a liquid-preserved specimen, and previous attempts to obtain DNA extracts from such material proved unsuccessful, presumably due to poor DNA quality. Three taxa of Polystacha (P. albescens subsp. imbricata, P. calluniflora, and P. pyramidalis) were also included as out-groups because subtribe Polystachyinae, to which they belong, has been identified as the sister clade to the angraecoids (Chase et al. 2015; Górniak et al. 2010). Vouchers for each accession are deposited either at BR or BRLU (Online Resource 2).

Leaf and flower tissue were dried in silica gel for DNA extraction (Chase and Hills 1991). Total DNA was extracted from fresh (1 g) or silica-gel-dried material (0.3 g) using one of the two methods detailed in Simo-Droissart et al. (2013).

PCR amplification and DNA sequencing

The following primers were used for amplification and sequencing of each individual plastid region: (1) Tab-E and Tab-F for the trnL-F intergenic spacer (Taberlet et al. 1991); (2) rps16-1F and rps16-2R for the rps16 intron (Oxelman et al. 1997); (3) 19F (Molvray et al. 2000), 1326R (Cuenoud et al. 2002), 390F (Cuenoud et al. 2002), and trnK-2R (Johnson and Soltis 1994) for matK; (4) trnC and petN-1R for the trnC-petN intergenic spacer (Lee and Wen 2003); and (5) 3720F, IntR, IntF, and 5500R for ycf1 (Neubig et al. 2009). The nuclear marker ITS-1 was amplified using ITS-A and ITS-C designed for angiosperms (Blattner 1999).

PCR amplifications were carried out in one of three thermocyclers (Biometra TProfessional thermocycler, PTC-100 or PTC-200, Bio-Rad Laboratories, Inc.) in a total volume of 25 μL, with 1–2 μL of template DNA extract (of unquantified concentration), 0.125 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 18.375–19.375 μL of H2O. The PCR amplification profiles used for the six DNA regions are detailed in Simo-Droissart et al. (2013). PCR products were then 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, 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 (unquantified concentration), and 4.3–6.3 µL of H2O for a total reaction volume of 10 µL. Cycle sequencing conditions used are detailed in Simo-Droissart et al. (2013). Sequencing products were cleaned by ethanol precipitation and then separated on an ABI 3100 automated capillary DNA sequencer following the manufacturer’s protocols (ABI). Both strands were sequenced to ensure accurate base calling. Sequence chromatograms were imported into Geneious Pro v.6 (Drummond et al. 2005). They were automatically trimmed at both ends using 5 % chance of error per base, after which the sense and antisense chromatograms were assembled to generate a consensus sequence. All incongruities and ambiguities were manually checked and edited. Consensus sequences were then aligned with the plugin MAFFT (Katoh et al. 2002) implemented within Geneious. Alignments were visually checked and edited wherever necessary. For coding regions, such as matK and ycf1, nucleotides were translated into amino acids to verify that the sequences corresponded to a protein, using a reference sequence from NCBI.

The number of accessions included in each of the six individual matrices was as follows: 37 accessions for ITS-1 (Online Resource 3), the rps16 intron (Online Resource 4), and trnL-F (Online Resource 5); 37 accessions for the combined matK regions (Online Resource 6), i.e., 19F–1326R and 390F–trnK-2R (although amplification of the region 390F–trnK-2R failed in Angraecum gabonense, two accessions of A. aporoides and two accessions of A. distichum); 36 accessions for the combined ycf1 regions (Online Resource 7), i.e., 3270F–IntR and IntF–5500R (Tridactyle bicaudata did not amplify while A. gabonense and one accession of A. aporoides had incomplete sequences, represented by the fragments IntF–5500R and 3270F–IntR, respectively); and 34 accessions for trnC-petN (Online Resource 8) (Polystachya calluniflora, one accession of A. aporoides and one accession of A. distichum failed to amplify).

Parsimony and Bayesian analyses

Cladistic analyses using Fitch parsimony (Fitch 1971) were performed using PAUP* 4.0 beta 10 (Swofford 2003). All characters were unordered with equal weight; gaps were coded as missing data. Heuristic searches were performed using tree bisection–reconnection (TBR) branch swapping with 1000 replicates and random taxon addition, holding ten trees at each step, and saving 20 trees per replicate to reduce time spent in swapping on large islands of trees. 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. Levels of internal support were estimated using the bootstrap method (Efron 1979; Felsenstein 1985) with 1,000 bootstrap replicates with random taxon addition and TBR branch swapping, retaining ten trees at each step, and saving ten trees per replicate. Parsimony analyses were first run separately for each region (i.e., ITS-1, matK, rps16, trnC-petN, trnL-F, and ycf1). Consensus trees and bootstrap values generated from each region were then compared visually for congruence. As there were no conflicts involving any of the well-supported clades, we combined data for all plastid regions into an initial alignment matrix (hereafter referred to as the plastid matrix, Online Resource 9) and then all plastid and ITS-1 regions into a second alignment matrix (hereafter the combined matrix, Online Resource 10). For taxa that had missing sequence regions, those partitions were coded as missing data (4 %) in the plastid and combined matrices.

Bayesian analyses were performed using MrBayes 3.2.1 (Ronquist and Huelsenbeck 2003; Ronquist et al. 2012) on the combined matrix, with one partition per gene (six partitions in total). Two independent analyses were run for 2 million 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. The separate runs were analyzed and compared using TRACER v1.5 (Drummond and Rambaut 2007) to assess stationarity and convergence, and to verify that the effective sample size for all parameters was sufficiently high (ESS > 200). Convergence of runs was also assessed by a graphical exploration of the posterior split probabilities (hereafter PP) using the online version of AWTY (Nylander et al. 2008). Trees were sampled every 500 generations, resulting in a total of 4,001 trees per run from which the first 1000 (25 %) were discarded as the burn-in phase. The majority-rule consensus tree was constructed using the function sumt in MrBayes.

The species were defined using a combination of the morphological (Mayr 1969) and phylogenetic (Purvis et al. 2005) species concepts. In this study, we thus circumscribed species to comprise morphologically coherent entities that included, whenever possible, accessions belonging to a single clade.

Results

Preliminary species delimitation within Angraecum sect. Dolabrifolia based on morphometric analyses

The first three axes of the extended PCA (the Hill-Smith ordination) explained 93 % of the total variance among the 172 specimens included in the multivariate analysis (Online Resource 11, Fig. 1a, b). The variation explained by the first axis (68 % of the total) largely correlates with the highest negative loadings for lateral sepal length (LSL), dorsal sepal length (DSL), lateral petal length (LPL), and flower length (FLL). The second axis (16 % of total variance) has the highest negative loadings for the white-yellowish color of the flower (FLC.whiteyellow) and the acute apex of the leaves (LFA.acute). The variation along the third axis (9 % of total variance) correlates with the highest negative loading for the white-yellowish color of the flower (FLC.whiteyellow) and the highest positive loadings for the width and the length of the leaves (LFW and LFL, respectively).

Fig. 1
figure 1

Scatter plots of the first three axes of the extended principal component analysis based on 15 characters scored from 172 specimens of Angraecum sect. Dolabrifolia. a Axes 1 and 2; b axes 1 and 3. Variables showing the highest loadings are represented (see also Online Resource 11). Codes used for variables are detailed in Table 1. Circles summarize specimens from the same species

Full size image

The projection using the first two axes reveals four groups of specimens (Fig. 1a). The first group corresponds to a mix of all specimens assigned to Angraecum aporoides and A. distichum, whereas the three other groups correspond, respectively, to A. aff. aporoides, A. bancoense, and A. podochiloides. The projection using the first and the third axes reveals four groups of specimens, with A. aporoides and A. distichum forming separate albeit neighboring groups (Fig. 1b). Again, three of these four groups correspond to species, albeit not the same set as above (A. aporoides, A. aff. aporoides and A. bancoense), and the fourth group comprises specimens assigned to A. distichum and A. podochiloides. The ordination on the three PC axes is thus consistent with the delimitation of five species. The main discriminant variables for A. aff. aporoides are the dimensions of the leaf along with its acute apex, whereas those for A. bancoense are the dimensions of the flower (i.e., small lengths of the sepals and lateral petals, see also Table 2).

Table 2 Median, standard deviation, and range of the 13 numeric characters measured in this study
Full size table

The clustering dendrogram based on distance measures between specimens likewise reveals five groups that correspond to the five species (Fig. 2). These are cluster 1: Angraecum aff. aporoides, with eight specimens; cluster 2: A. podochiloides, with ten specimens; cluster 3: A. aporoides, with 25 specimens; cluster 4: A. distichum, with 63 specimens; and cluster 5: A. bancoense, comprising 66 specimens. Boxplots (Fig. 3) for each of the six numerical variables showing the highest loadings in the Hill-Smith ordination (based on the first three axes, see Online Resource 11) provide a graphical depiction of variation among the five species.

Fig. 2
figure 2

Clustering dendrogram obtained by computing the distance matrix measured between the 172 specimens of Angraecum section Dolabrifolia. In bold, specimens with representatives in molecular analyses. Members of the two subclusters of A. aporoides are represented by light gray and dark gray boxes, respectively

Full size image
Fig. 3
figure 3

Box and whisker plots depicting the character variation ranges among the five species in section Dolabrifolia. a Leaf length, b leaf width, c flower length, d dorsal sepal length, e lateral sepal length, f lateral petal length

Full size image

The multiple comparisons of medians obtained using the Kruskal–Wallis test showed that all 13 quantitative variables studied differed significantly among the five groups (Table 2). Of these 13 variables, leaf width (LFW) and flower length (FLL) provide clear-cut separation among them. Indeed, Angraecum aporoides and A. aff. aporoides have the widest leaves, while A. bancoense has the smallest flowers. Considering those species possessing the longest leaves, specimens assigned to A. aporoides differ from those of A. aff. aporoides by eight of the 13 numeric variables used in our study (Table 2). Of these eight variables, the most informative, providing a clear-cut separation between these two taxa, are the lengths of the lateral petals (LPL), the lip (LIL) and the spur (SPL). Among the species possessing the smallest leaves, A. podochiloides is clearly distinct from the two other species (i.e., A. bancoense and A. distichum) by having the narrowest leaves, which are lanceolate and have an acute or subacute apex. Angraecum podochiloides is also the only species of section Dolabrifolia to possess yellow or orange tips on its perianth parts.

The detailed examination of herbarium material showed that around 70 % of specimens of Angraecum distichum—described by Lindley (1836)—have been confused with the recently described A. bancoense—described by Burg (Arends et al. 1980), although it was first collected in 1860. Angraecum distichum differs from A. bancoense by all 13 of the numeric variables used in our study (p < 0.0001; Table 2). The seven most important variables that provide a clear-cut separation of these two species are flower, dorsal sepal, lateral sepal, lip and spur lengths (FLL, DSL, LSL, LIL and SPL, respectively), and lateral sepal and petal widths (LSW and LPW, respectively). The most distinctive feature differentiating the two species is FLL. Indeed, A. distichum has longer flowers than A. bancoense (11–17 vs. 6–9 mm) and is also easily distinguishable by its curved lip and the slender apex of its spur.

Phylogenetic analyses based on DNA sequences

The number of aligned characters in the matrices of each of the six markers examined (i.e., ITS-1, matK, rps16, trnC-petN, trnL-F, and ycf1) is detailed in Table 3. The plastid matrix contained 37 accessions, and 6224 aligned characters of which 931 (14.9 %) were potentially parsimony-informative. The combined matrix also contained 37 accessions, with an aligned length of 6591 characters, of which 998 (15.1 %) were potentially parsimony-informative (Table 3).

Table 3 Matrix values and statistics of parsimony analyses
Full size table

When analyzed separately, each marker produced a consensus tree (Fig. 4a–f) with insufficient resolution to evaluate the circumscription of the individual species recognized in Angraecum sect. Dolabrifolia, although the results did provide strong support for the monophyly of the section, with moderate to high bootstrap values (hereafter BS) of 78–100 %. Indeed, the five accessions of A. podochiloides formed a clade with weak support (BS = 64 %) in the analysis with the trnL-F matrix and with strong support using the ITS-1, matK, trnC-petN, and ycf1 matrices (BS = 85–100 %). The four accessions of A. bancoense formed a clade with weak support (BS = 61–62 %) in the analyses using the ycf1 and trnL-F matrices, respectively, and with strong support using ITS-1 (BS = 95 %). The five accessions of A. distichum formed a weakly supported clade (BS = 62 %) only in the analyses using the ycf1 matrix. The two accessions of A. aff. aporoides formed a clade with weak support (BS = 62 %) in the analysis using the trnC-petN matrix and with strong support using ycf1 and rps16 (BS = 75 % and 86 %, respectively). The ten accessions of A. aporoides did not form a clade in any of the six single-marker analyses.

Fig. 4
figure 4

Parsimony analysis (strict consensus tree with bootstrap percentages shown above or below branches) of ITS-1 (a), matK (b), rps16 (c), trnC-petN (d), trnL-F (e), ycf1 (f), plastid matrix (g), and combined matrix (h). Taxa: Angraecum sect. Dolabrifolia: A. apo = A. aporoides; A. aff. = A. aff. aporoides; A. ban = A. bancoense; A. dis = A. distichum; A. pod = A. podochiloides; Angraecum sect. Pectinaria: A. atl = A. atlanticum; A. dor = A. doratophyllum; A. gab = A. gabonense; A. pun = A. pungens; A. sub = A. subulatum; Diaphananthe: D. odo = D. odoratissima; Tridactyle: T. aur = T. aurantiopunctata; T. bic = T. bicaudata; out-groups (genus Polystachya): P. alb = P. albescens subsp. imbricata; P. cal = P. calluniflora; P. pyr = P. pyramidalis. Details of each analysis are given in Table 3. The arrow indicates the section Dolabrifolia. Nodes with 100 % bootstrap values are indicated by a solid black circle

Full size image

When analyzed together, the markers yielded more fully resolved trees that generally had higher support values (Figs. 4g–h, 5). Parsimony analysis of the combined matrix yielded 36 most parsimonious trees of 1424 steps each, with a consistency index (CI) of 0.91 and a retention index (RI) of 0.96 (Fig. 4h; Table 3). In the Bayesian analyses, a run length of 2 million generations appeared to be sufficient to obtain a satisfactory sampling of the posterior distribution (average standard deviation of split frequencies <0.001; ESS > 200 for all parameters). AWTY plots of the posterior split probabilities showed that the two independent runs were close in parameter (tree) space and confirmed the convergence diagnostic. The Bayesian analysis provided a well-resolved tree (Fig. 5), which showed the same topology as the consensus tree obtained from the parsimony analysis.

Fig. 5
figure 5

Consensus tree obtained from Bayesian analysis of the combined molecular data set. Posterior probabilities (PP) are given above or below the branches. Scaled phylogram obtained from Bayesian analysis is shown in the upper left corner, demonstrating the relative branch lengths. A. = Angraecum; D. = Diaphananthe; T. = Tridactyle; P. = Polystachya

Full size image

Phylogenetic analyses conducted on the combined matrix (Figs. 4h, 5) confirmed that Angraecum sect. Dolabrifolia is monophyletic (BS = 100 %; PP = 1). Similarly, strong support was provided for the delimitation of four of the five entities recognized, viz., A. aff. aporoides (BS = 94 %; PP = 1; clade C, Fig. 5), A. bancoense (BS = 99 %; PP = 1; clade D, Fig. 5), A. distichum (BS = 74 %; PP = 0.99; clade E, Fig. 5), and A. podochiloides (BS = 100 %; PP = 1; clade A, Fig. 5), but the accessions of A. aporoides were placed into two distinct and well-supported clades (each with BS = 100 % and PP = 1; clades B and F, respectively, Fig. 5). The first clade of A. aporoides (clade B), comprising three accessions from Príncipe Island and Rabi (Gabon), is sister (BS = 97 %; PP = 1) to the clade of A. aff. aporoides (clade C), known only from Gabon. The second clade of A. aporoides (clade F), comprising seven accessions from Bifa and Bipindi (southern Cameroon) and Gabon, is sister (BS = 97 %; PP = 1) to the subclade uniting A. bancoense and A. distichum. All three accessions of clade B were represented in the morphometric analyses (see cluster 3A in Fig. 2) while only four of the seven accessions of clade F were represented (see cluster 3B in Fig. 2).

Within Angraecum sect. Dolabrifolia (Figs. 4h, 5), A. podochiloides (clade A, Fig. 5) is sister (BS = 100 %; PP = 1) to a clade comprising the four other members of the group, namely A. aporoides, A. aff. aporoides, A. bancoense, and A. distichum (clades B–F, Fig. 5). Clade B (Fig. 5) of A. aporoides is sister to A. aff. aporoides (clade C, Fig. 5), and the subclade formed by clades B and C is in turn sister (BS = 100 %; PP = 1) to a subclade uniting the three other clades (clades D–F, Fig. 5). Finally, the subclade comprising A. bancoense and A. distichum (clades D and E, respectively, Fig. 5) is sister (BS = 97 %; PP = 1) to the clade F of A. aporoides.

Comparison between morphometric and phylogenetic analyses of Angraecum aporoides

Accessions of Angraecum aporoides do not appear to form a monophyletic group in our phylogenetic analyses. The type of this species could not be included in the molecular study, but in the morphometric analyses, it was placed in cluster 3B of the dendrogram (Fig. 2), which comprises 15 specimens, four of which are represented in the phylogenetic analyses. In an attempt to elucidate why the ten accessions identified as A. aporoides form two clades that are not sister to one another, multiple comparisons of medians using the Kruskal–Wallis test were performed on the two subclusters of A. aporoides observed in the clustering dendrogram (see light and gray boxes in Fig. 2). These comparisons indicated that nine of the 13 numeric variables are significantly different between the two subgroups, namely the dimensions of dorsal (DSL, DSW) and lateral (LSL, LSW) sepals, lateral petal (LPL, LPW), lip (LIL, LIW) and the spur length (SPL) (Table 4). However, none of these nine variables showed a clear-cut separation between the two subclusters of A. aporoides. Using box and whisker plots (Fig. 6), these nine numeric variables were used to graphically depict the variation among the two subclusters (cluster 3A with 10 specimens and cluster 3B with 15 specimens) of A. aporoides. Since none of the nine variables with significant differences was informative in distinguishing the two subclusters, we thus performed multiple comparisons of medians using the Kruskal–Wallis test, but restricted to the seven specimens represented in the molecular analyses, i.e., the three accessions of clade B and four (out of seven) of clade F (Fig. 5). The new multiple comparisons based on these collections showed that only three variables were significantly different between the two clades, lateral sepal width (LSW, 1.8–2 mm ≠ 2.2–3.2 mm), lip width (LIW, 3.3–3.5 mm ≠ 4.5–5.3 mm), and spur length (SPL, 6–6.5 mm ≠ 7.3–7.6 mm) for clades B and F, respectively.

Table 4 Median, standard deviation, and range of the 13 numeric characters measured in this study for the two clusters of A. aporoides
Full size table
Fig. 6
figure 6

Box and whisker plots depicting the character variation ranges among the two subclusters of Angraecum aporoides. a Dorsal sepal length, b dorsal sepal width, c lateral sepal length, d lateral sepal width, e lateral petal length, f lateral petal width, g lip length, h lip width, i spur length

Full size image

Discussion

Monophyly of Angraecum sect. Dolabrifolia

Based on a phylogenetic analysis using four plastid DNA regions, Micheneau et al. (2008) found that Angraecum sect. Dolabrifolia, as defined by Garay (1973), was monophyletic, although their sampling included only a single accession from each of just two taxa belonging to the group. Our study confirms their finding, but with much more comprehensive sampling from sect. Dolabrifolia. Each of our analyses (i.e., those based on the six individual markers as well as the combined data set) placed all members of the five recognized species of section Dolabrifolia in a well-supported clade that is sister to the continental African group comprising A. sect. Pectinaria. The results of the morphometric and phylogenetic approaches used in this study to circumscribe taxa and test their delimitation provide a robust foundation for conducting a taxonomic revision of section Dolabrifolia (M. Simo-Droissart et al. submitted), as was recently done for A. sect. Pectinaria (Simo-Droissart et al. 2014).

Szlachetko and Romowicz (2007) raised Angraecum sect. Dolabrifolia to the rank of genus based on the unique foliar characters of the group (viz., the laterally compressed and densely imbricate leaves), and this treatment was followed by Szlachetko et al. (2013) based on molecular and morphological data. The results presented in our study do not support this interpretation, but a final decision on how best to treat this group must await the resolution of phylogenetic relationships within the entire genus Angraecum and allies, in the context of our efforts to develop a robust generic/infrageneric classification system that takes into account all members of the broader angraecoid orchid clade.

Species hypotheses in Angraecum sect. Dolabrifolia

When considered together with the clustering dendrogram constructed using the distance matrix of the 172 specimens of Angraecum sect. Dolabrifolia, the combination of the first three axes of the extended PCA yielded clear delimitations of each of the five species-level entities identified prior to conducting the morphometric analyses. The results of the molecular phylogenetic analyses support the delimitation of four of these species, while accessions of the fifth, A. aporoides, were placed in two separate, well-supported clades. Below, we discuss the circumscription of each of these taxa.

Angraecum podochiloides

This species, native to the Guineo-Congolian Region (see White 1979), possesses the narrowest leaves in Angraecum sect. Dolabrifolia, which are lanceolate and have an acute or subacute apex. It also has the narrowest lateral sepals of any species in the group. While the other members of the section have entirely white flowers, A. podochiloides is easily recognized by the yellow or orange tips of its perianth parts (see Simo et al. 2010). In the morphometric study, the projections based on the extended PCA placed A. podochiloides and A. aff. aporoides close to one another, primarily because the leaves of both species have an acute apex (Figs. 1a, 2). The trees resulting from the phylogenetic analyses based on molecular sequence data placed the five accessions of A. podochiloides in a well-supported clade sister to the remaining clades of section Dolabrifolia, with the exception of the rps16 tree, which lacked resolution (placing all Dolabrifolia species in a polytomy; Fig. 4c).

The Angraecum aporoides complex

The ten accessions of Angraecum aporoides included in our molecular phylogenetic analyses form two clades that are not sisters (clades B and F, Fig. 5). These two clades correspond precisely to the two subclusters of A. aporoides in the dendrogram resulting from the extended PCA performed on morphological data (clusters 3A and 3B in Fig. 2), cluster 3A comprising 10 specimens and cluster 3B comprising 15 specimens. Although nine of the 13 numeric variables are significantly different between the two subclusters, none of them shows any clear-cut separation. Moreover, there is no apparent difference in geographic distribution among the members of clades B and F. Indeed, accessions of clade B (and their representative specimens) were collected on Príncipe Island and in Gabon (with other specimens of cluster 3A coming from Cameroon and São Tomé Island) while those belonging to clade F came from Cameroon and Gabon (with other specimens of cluster 3B also coming from Cameroon and Gabon, as well as Nigeria, where the type specimen was collected). Multiple comparisons of morphometric characters from the seven specimens of A. aporoides represented in the molecular analyses (clades B and F) revealed that three of the 13 variables (LSW, LIW, and SPL) are significantly different, with a clear-cut separation between the two clades. However, these morphological differences are observed only when considering the reduced sample size of A. aporoides (seven out of 25), and not with the broader sampling of 25 specimens used in the dendrogram. As such, these features alone are thus not sufficient to distinguish between the two subclusters of this species. The type specimen of A. aporoides (specimen A. aporoides 9) is placed in cluster 3B in the clustering dendrogram. The distinction between clades B and F appears based solely on plastid data, since this portion of the ITS tree is not resolved. Indeed, all five plastid sequences are derived from a single, non-recombining genome, and therefore represent a single marker. Another possible explanation of the non-sister placement of clades B and F (Fig. 5) could be that hybridization has taken place and that the tree obtained from the combined data set reflects lateral transfer of the plastid genome. Further studies including additional specimens for both morphometric and DNA-based phylogenetic analyses (especially using nuclear markers) will be needed to clarify whether the two clades of A. aporoides recovered in our molecular phylogenetic analyses warrant taxonomic recognition.

Angraecum aporoides versus A. aff. aporoides

As indicated above, while examining material of Angraecum sect. Dolabrifolia in order to define species, some specimens were found to be similar to A. aporoides in general appearance but to have leaves that clearly differed in shape and by the presence of an acute apex (a unique feature among the members of the section with entirely white flowers). All of these atypical specimens were collected in Gabon, while material of typical A. aporoides was obtained from São Tomé and Príncipe, Gabon, and Cameroon. The material of A. aff. aporoides thus appears to represent a new entity best recognized at the species level, which is being described in a separate paper (M. Simo-Droissart et al. submitted).

The specimens assigned to Angraecum aff. aporoides clearly differ from those of A. aporoides by eight of the 13 numeric variables, of which the length of the lateral petals (LPL), the lip (LIL), and the spur (SPL) are the most informative, providing a clear-cut separation between these two entities. With their relatively small flowers, plants of A. aff. aporoides resemble A. bancoense, which is widely distributed in the Guineo-Congolian Region, but they differ clearly by leaf length (LFL) and width (LFW), and the length of the flower (FLL), the lateral sepals (LSL), and the lateral petals (LPL) (see Table 2). None of the trees resulting from the phylogenetic analyses placed the accessions of A. aff. aporoides (clade C) and those of the A. aporoides complex (clades B and F) as sisters.

Angraecum bancoense versus A. distichum

These two species are clearly distinct from one another on the basis of flower size, but are difficult to distinguish vegetatively. In their original description of Angraecum bancoense, Arends et al. (1980) pointed out this difficulty, noting the similarities in leaf shape and concluding that the two species were vegetatively indistinguishable. For this reason, we used the largest possible sample size to observe and quantify variation in leaf shape. Taken together, the specimens of A. bancoense (66 specimens) and A. distichum (63 specimens) represent 75 % (129 out of 172) of the specimens used in our morphometric analyses. This is particularly important because variation in leaf shape between specimens of these two species often leads to misidentifications in the absence of floral parts, as Arends et al. (1980) had suggested. In fact, most fruiting collections that retained remnant floral parts, notably the spur (whose length is informative in distinguishing the two species), were erroneously identified as A. distichum, long regarded as the most widespread and abundant species in section Dolabrifolia. These numerous misidentifications may also have resulted from the fact that these two species are sometimes found growing together, which has confused collectors and orchid taxonomists alike.

The numerous misidentifications involving material of Angraecum distichum and A. bancoense prompted us to question the taxonomic status of these two species and to consider whether other morphological features, together with geographic distribution, could be used to differentiate them. According to Arends et al. (1980), A. distichum differs from A. bancoense by its larger flowers, which are longer than the leaves. However, we found several specimens clearly referable to A. bancoense that also possess flowers longer than the leaves. All of the accessions used in the molecular analyses were unambiguously assigned to either A. bancoense or A. distichum and were placed in the corresponding clades, which were well supported as sister groups (Figs. 4, 5), confirming that they represent divergent lineages, but also share a most recent common ancestor.

As currently circumscribed, Angraecum bancoense and A. distichum are widely distributed in West and Central Africa. Preliminary tests of autonomous self-pollination in these two species have shown that they both require a pollinator to set fruits, and self-pollination carried out by hand showed that they are self-compatible. While A. bancoense and A. distichum occur in sympatry and flower throughout the year, no intermediate forms have been observed, suggesting the presence of one or more reproductive barriers between them. This interpretation is further supported by the fact that preliminary cross-pollination tests between the two species failed to yield any fruit.

Angraecum poppendickianum

The status of the recently described Angraecum poppendickianum Szlachetko and Olszewski (2001) has been controversial since its publication in the Flore du Cameroun, particularly because it was not accepted in the world checklist (Govaerts et al. 2016). The description suggests that A. poppendickianum may be an intermediate form between A. distichum (which it resembles vegetatively) and A. aporoides (which has similar flowers), and Szlachetko and Olszewski (2001) commented that the material on which they based their novelty might indeed be of hybrid origin. However, examination of these specimens (N. Hallé 872, P 00259997!, the holotype, and Merle 32, P 00259998!, a paratype) clearly shows that they belong to A. distichum because the sizes of their leaves, flowers, and floral parts fall within the range of material belonging to that species. In light of this information, we therefore regard A. poppendickianum as a synonym of A. distichum.

Conclusions

Comparison of the results from the morphometric investigations of the members of Angraecum section Dolabrifolia with those from phylogenetic analyses based on DNA sequence data has helped to clarify species circumscriptions within this group, an indispensable prerequisite for undertaking a taxonomic revision. Our results confirm the monophyly of the section and indicate that it comprises at least five species, namely A. aporoides, A. bancoense, A. distichum, A. podochiloides, and a new taxon, provisionally referred to as A. aff. aporoides. The delimitation of four of these five species is well supported, but the multiple accessions of the fifth, A. aporoides, appear to form a paraphyletic group, comprising two well-supported clades. The members of these two clades differ morphologically by the size of the flowers (lateral sepal width, lip width, and spur length), but these differences do not show a clear-cut separation between the two subclusters obtained in the morphometric analyses. Moreover, no geographic pattern was detected in either the two clades or the two subclusters. In an attempt to assess whether the two clades of A. aporoides can be differentiated morphologically, a broader morphometric analysis may be needed that would include expanded sampling from throughout the geographic range of this species. Also, molecular phylogenetic analyses including more nuclear markers should be performed.

Results from the present study have also helped to differentiate two frequently confused species, Angraecum bancoense and A. distichum, which can be easily distinguished based on the length of their flowers and sepals, as well as the length of their lip and spur, along with the width of their lateral sepals and petals. The new species, currently referred to as A. aff. aporoides, is closely related to A. aporoides but differs based on both morphometric and molecular data. This new taxon is only known from Gabon and is being described as part of a broader taxonomic revision of the entire section (M. Simo-Droissart et al. submitted).