Introduction

With more than 576 species, Passiflora L. is by far the most important genus in the family Passifloraceae Juss. ex Roussel, both numerically and economically. It is mostly distributed in the Neotropics, from coastal zones up to ca. 4300 m a. s. l. Only 23 species are native to the Old World in Southeast Asia, Australia, and Oceania (Killip 1938; Ulmer and MacDougal 2004). Passifloraceae are very ancient, dating back 65.5 Mya (Paleocene), and seem to follow a biogeographic scenario proposed for several plant groups, originating in Africa, crossing to Europe/Asia and reaching the New World by way of land bridges (Muschner et al. 2012).

Passionflowers are herbaceous or woody vines usually climbing with tendrils, but a few are trees or shrubs. Other typical vegetative traits include alternate leaves, axillary stipules, and petiolar and/or laminar nectary glands. In addition, the genus exhibits several unique floral features such as an androgynophore, a complex corona constituted by one or several concentric rows of filaments, and a limen-operculum system limiting access to the nectary chamber, with impressive interspecific variation in size, shape, and colors (Killip 1938). Their wide morphological variation seems to be the result of their habitat diversity as well as their coevolutionary relationships with many organisms, including a wide range of pollinators as small and large insects, birds, and bats (Ulmer and MacDougal 2004). The most common haploid chromosome numbers are n = 6, n = 9 and n = 12 (De Melo et al. 2001), and the average genome size (2C) is 1.03 pg (Yotoko et al. 2011).

Many species exhibit interesting fruits, pharmacological properties (e.g., sedative effect), and/or ornamental potential (Yockteng et al. 2011). More than 65 species belonging to subgenus Passiflora produce relatively large edible fruits of potential economic interest (Coppens d’Eeckenbrugge 2003). Among the 25 cultivated species, the two botanical forms of P. edulis Sims f. flavicarpa Degener (yellow passion fruit), and f. edulis (purple passion fruit) are by far the most important crops, with a world production estimated at ca. 1.300.000 tons (FAO 2011). Other commercially cultivated passion fruits are P. tripartita var. mollissima (Kunth) Holm-Nielsen & Jørgensen (curuba de Castilla), P. tarminiana Coppens & Barney (curuba India), P. ligularis Juss. (sweet granadilla), P. maliformis L. (granadilla de piedra or stone granadilla), P. quadrangularis L. (giant granadilla), P. alata Curtis (fragrant granadilla), P. laurifolia L. (water lemon), P. popenovii Killip (granadilla de Quijos), and P. setacea DC. (sururuca). The high potential of Passiflora for crop diversification and economic development induced research institutions of different countries to prioritize the characterization and evaluation of wild and cultivated populations (Cerqueira-Silva et al. 2016) and develop strategies for conservation and improvement of these genetic resources.

The high diversity of character combinations related to nectary glands, stipules, corona, operculum, and limen have been heavily used to delimit infrageneric divisions in Passiflora taxonomy, separating subgenera, sections, and series. Although most species appear well delimited, there are many cases in low-level taxa as sections and series, where two or more species are very difficult to distinguish. The largest monograph and most important reference work was that of Killip (1938) who described 355 American species plus 20 species in 1960, placing them in 22 subgenera. Escobar (1988, 1989, 1990, 1994) reviewed subgenera Astrophea (DC.) Mast., Distephana (Juss. ex DC.) Killip, Rathea (Karst.) Killip, and Tacsonia (Juss.) Tr. & Planch., merging subgenera Tacsoniopsis (Tr. & Planch.) Killip and divided into sections and series, proposing series Manicata (Harms) Killip as a subgenus, and publishing one additional subgenus, Porphyropathanthus. Feuillet and MacDougal (2003) proposed a new infrageneric classification of Passiflora, taking into account the Old World species and recognizing only four subgenera, Astrophea (unchanged; lianas, rarely trees and shrubs, associated with the “n = 12” group), Decaloba (DC.) Rchb. (vines with small flowers and fruits, the latter usually black; “n = 6” group), Deidamioides (vines, with two-flowered peduncles, not clearly defined as a morphological group), and Passiflora (vines to lianas, with large flowers and fruits; “n = 9” group). The subgenera were divided into supersections. This classification needs to be revised and completed with the explicit inclusion of the species involved. Krosnick et al. (2009) with DNA molecular data recognized a fifth subgenus, Tetrapathea (vines, inflorescences, small flowers and fruits, yellow-green to red at maturity) from Oceania.

Both classifications of Killip (1938) and Feuillet and MacDougal (2003) are based on the extreme morphological richness and complexity of Passiflora, whose species present numerous particular traits, including a wide variation in leaf shape, even within species and within individuals (heterophylly and heteroblasty are very common), the presence of extrafloral nectaries in different parts (on leaf petiole, lamina or margins, on bracts, on sepals), floral traits showing a high level of coevolution with particular pollinators, particularly in corolla color, the variable development of the hypanthium and that of the corona. However, no clear hierarchy emerges in the relative contributions of these traits to the taxonomy of the genus.

On the other hand, the first molecular studies carried out on significant Passiflora species samples have consistently validated three of the four major subdivisions proposed by Feuillet and MacDougal (2003). The results of Muschner et al. (2003), on nuclear ribosomal internal transcribed spacers (ITS-1 and ITS-2) and plastid trnL-trnF intergenic spacer, Yockteng and Nadot (2004a) on chloroplast matK, Yockteng and Nadot (2004b) on chloroplast-expressed glutamine synthetase (ncpGS), Hansen et al. (2006), on sequences analysis of the chloroplast (rpoC1 and trnL-trnT), and Krosnick et al. (2013), on nuclear (nrITS and ncpGS) and chloroplast (cp trnL-F and ndhF), confirm the clear separation of three clades corresponding to the new contours of subgenera Decaloba, Astrophea, and Passiflora. These three major clades correspond to cytogenetic groups as they appear characterized by chromosome numbers of 2n = 12, 24, and 18, respectively. The results of Hansen et al. (2006) also support the small subgenus Deidamioides (Harms) Killip, whereas the two species that represent it in the study of Yockteng and Nadot (2004b) are split into two widely divergent branches. In addition, the latter study indicated that four other small subgenera, Dysosmia (DC.) Killip, Tryphostemmatoides (Harms) Killip, Polyanthea (DC.) Killip, and Tetrapathea (DC.) Rchb., should also be recognized. Another study carried out by Muschner et al. (2012) on the phylogeny, biogeography, and divergence times in 106 species and four subgenera of Passiflora with plastidial, mitochondrial, and nuclear genomes showed that the genus Passiflora is monophyletic. In addition, subgenus Deidamioides as described by MacDougal and Feuillet (2004) emerged as paraphyletic. They also situated the divergence among the four subgenera in Passiflora from 33 to 38 million years ago, during the Andes uplifting process. Krosnick et al. (2013) studied the phylogenetic relationships of subgenus Decaloba using 148 taxa and four molecular markers (nrITS, ncpGS, cp trnL-F and ndhF). The results showed that subgenus Decaloba is monophyletic and contains seven major lineages that generally correspond to currently recognized supersections. Their results also showed subgenus Deidamioides as circumscribed by Feuillet and MacDougal (2003) to be polyphyletic, supporting Yockteng and Nadot (2004b) and Muschner et al. (2012) in recognizing subgenus Tryphostemmatoides as sister to subgenus Astrophea. That suggests that subgenus Deidamioides be recognized in a strict sense, i.e., the species closely related to P. deidamioides Harms (Nunes 2009; Krosnick et al. 2013). Recently, a dated Passiflora phylogeny with DNA markers (trnL-F, ndhF, ITS and ncpGS) resolved the supersection Tacsonia (Juss.) Feuillet & MacDougal as a monophyletic group, which diverged around 10.7 Mya and underwent radiation at 9–8 Mya (Abrahamczyk et al. 2014). These studies have provided many new insights into the evolution of Passiflora but reveal as many new challenges that need to be addressed. In particular, such phylogenetic molecular studies cannot give us a definitive answer on the relationship between morphologically ill-defined species and species groups. Actually, if two or more species within a series are only distinguished by the exact number of filament whorls in the corona, the color of one of these whorls, or the number of nectary glands on the petiole, we must first question the discontinuity in the variation described, at both morphological and genetic levels, and assess objectively the morphological basis of the classification.

Despite the impressive morphological diversity described among Passiflora species, few studies have compared intra- and intersubgeneric, and intra- and interspecific variation with statistical tools. A first study was conducted by Villacís et al. (1998) on the most common species of supersection Tacsonia (Juss.) Feuillet & MacDougal on Colombian and Ecuadorian accessions. Floral traits were mostly represented in their set of 33 qualitative descriptors, and vegetative traits in their set of 28 quantitative descriptors. The former showed limited intraspecific variation and a consistent picture of interspecific relations, while the latter provided more information on intraspecific variation but a less consistent picture on the differences between species. The descriptor list was corrected to take into account traits specific to subgenus Tacsonia and augmented to 62 qualitative and 67 quantitative descriptors, giving a better balance between floral and vegetative traits. These convergent data contributed to the formal description of the cultigen P. tarminiana as a distinct species (Coppens d’Eeckenbrugge et al. 2001). The same descriptor list was used to study morphological variation in the two most common cultivated curubas (P. tarminiana and P. tripartita var. mollissima) and their hybrids, showing maternal effects in the hybrid phenotypes, and confirmed the spontaneous introgression occurring between P. tripartita var. mollissima and the wild P. mixta L. (Primot et al. 2005). A very detailed descriptor list was also used by Porter-Utley (2014) to study supersection Cieca (Medic.) MacDougal & Feuillet of subgenus Decaloba and particularly the species complexes around P. suberosa L. and P. coriacea Juss. Seventy quantitative traits were measured, finding 33–39 descriptors that could be categorized and gathered with qualitative traits for neighbor joining cluster analyses of the different subsamples. Morphological data appeared consistent with molecular data (Krosnick et al. 2013) in confirming monophyly of supersection Cieca, recognizing P. tridactylites Hook.f. and P. pallida L. as distinct species, and detecting occasional introgression of the latter with its close relative P. suberosa. On the other hand, there was considerable incongruence between molecular (ITS sequences) and morphological phylogenies, which was mostly attributed to a smaller sample size and intraspecific variation in the molecular data. A Brazilian study by De Oliviera Plotze et al. (2005) tested a new morphometric method based on leaf structures, the multiscale Minkowski fractal dimension, and digital images from a sample of ten Passiflora species. The method was very accurate in differentiating among species; however, two of them were not consistently classified, P. foetida L., a problematic species in all classifications, and P. miersii Mast.

In conclusion, despite the interest of completing an objective classification of a plant family with a huge morphological richness and complexity, only few research teams have developed and applied the necessary methodology. However, this methodology tends to be much more laborious than in plain molecular characterizations, due to the need of field germplasm collections including species with variable climatic adaptations in one or very few places where they can develop until flowering. In the case of wild Passiflora accessions, a possible solution is carrying out in situ characterization, which must then be taken into account to avoid or reduce possible environmental bias in the analyses.

The present study benefited from projects on diversity of Colombian Passifloraceae, including a component of collecting and establishing germplasm in field collections. As Colombia is the country with the highest Passiflora species diversity for both wild and cultivated materials (Ocampo et al. 2007, 2010), a wide species sample could be gathered. However, practical limitations allowed describing accessions of only 51 of the 174 reported species, representing the four subgenera of Feuillet and MacDougal (2003), but seven of the 22 Killip’s subgenera plus subgenus Manicata. A few non-native species were added, extending the sample to 61 species and ninth subgenus of the ancient classification. Our goal was twofold: to test the utility of the revised set of descriptors over a wide range of Passiflora species and to study morphological divergence among subgeneric divisions, species, and populations.

Materials and methods

Study area

The morphological study was carried out in three germplasm collections located in different ecological zones in Colombia according to environmental adaptations of species: El Cerrito (Tenerife) (2700 m a. s. l., 3°43′51.49″N; 76°4′36.56″W), El Cerrito (El Moral) in the department of Valle del Cauca (2400 m a. s. l., 3°42′53.80″N; 76°4′37.67″W), and Buenavista (Paraguacito) in the department of Quindío (1200 m a. s. l., 4°23′47.57″N; 75°44′3.54″W).

Plant materials

The total sample was composed of 261 individuals representing 124 populations and 61 species of Passiflora, representing eight Killip’s subgenera and subgenus Manicata (Online Resource 1), and all four Feuillet and MacDougal’s subgenera (Table 1). Geographic distribution was taken into account in the selection of accessions of a species. Narrow endemics such as P. trinervia (Juss.) Poir. are represented only by one population, and widespread species as P. edulis f. flavicarpa by one population per Colombian region. Three plants per population were grown from seeds, at a distance of 3 m between rows and 3–5 m within rows, according to adult plant size. In this study, we followed the two taxonomic treatments of genus Passiflora proposed by Killip (1938, with emends by Escobar 1988, 1989, 1994; and MacDougal 1994) and by Feuillet and MacDougal (2003), as a point of comparison and discussion of our data. Authors of plant names were noted according to Brummitt and Powell (1992).

Table 1 List of accessions of the genus Passiflora L. used in this study
Full size table

Data collection

The descriptor list was developed from the one used by Villacís et al. (1998), with adaptations to describe the wide diversity of our collections. It included 43 quantitative and 84 qualitative descriptors that are presented synthetically in Table 2. These were assessed on three individuals per populations, and five measures were taken for quantitative characters (in millimeters) for each individual plant. Color characters were recorded with the Royal Colour Chart (Royal Horticultural Society 2001). Quantitative fruit traits were not taken into account as they are often submitted to convergent selection processes in both wild and cultivated species.

Table 2 List of 127 descriptors used in the morphological characterization study. Scales for qualitative characteristics: B (binary), O (ordinal), and N (nominal)
Full size table

Analyses of quantitative variation

Shape descriptors were computed as ratios of crude ones. Quantitative data were submitted to an analysis of variance to compare variation among and within subgenera, species, populations, and individuals. To identify characteristics that contributed highly to differentiation among subgenera, we selected traits for which more than half of the variance was caused by variation at this level. When shape descriptors showed similar discriminating power, they were preferred over crude descriptors to avoid giving too much importance to variations in size. The selected descriptor set was submitted to a principal component analysis (PCA) applying the varimax normalized rotation option using the STATISTICA 10.0 software (Hill and Lewicki 2006), retaining those factors with an eigenvalue superior to one, and the individuals were projected onto the first three PCA axes.

Cluster analyses on qualitative data

According to the PCA results, certain quantitative variables were selected on the basis of their particular contribution to the axes, avoiding information redundancies that could be due to allometric relations; they were then categorized according to their distribution patterns and added to the qualitative dataset, provided that the corresponding information was not yet included in a purely qualitative descriptor. The resulting dataset was treated in two steps. A first set of qualitative variables was selected on the basis of their contribution to differentiation among subgenera, discarding those that show frequent variation at lower levels. A second set included all qualitative descriptors. Both sets were submitted to a neighbor joining cluster analysis (Saitou and Nei 1987) using the coefficient of dissimilarity of Sokal & Michener and calculating bootstrap values from 1000 replicates with the DARwin 6.0 software (Perrier et al. 2003). This phenetic approach was preferred because of the relatively poor information on morphological evolution in Passiflora.

Results

Quantitative variation

As expected, a very high variability was observed among the 124 populations. Table 3 gives mean values and coefficients of variation for the whole sample per subgenus. Coefficients of variation are generally higher for subgenus Decaloba compared with subgenus Passiflora that has comparable representation in number of species. This higher relative variation can be seen much clearer in inflorescences (18 out of 20 traits) and shape ratios (10 out of 13 ratios), than for vegetative parts (9 of 21 descriptors), suggesting a higher interspecific differentiation in subgenus Decaloba as compared to the other two.

Table 3 Mean values and coefficients of variation (Mean (CV)) for all the descriptors per subgenera according to Killip’s, Escobar’s and Feuillet and MacDougal’s classification
Full size table

The relative variance components for 57 quantitative descriptors show that all present a residual variance under 25% and then a high repeatability (Fig. 1). Many descriptors appear to be efficient in discriminating among subgenera. Thus, the proportion of variance at this level exceeds 50% for 27 descriptors, including stem diameter, leaf margin serration, leaf length, number of nectary glands on leaf margins and petiole, diameter of peduncle, length of first- and second-order peduncle segments, dimensions and shape of bracts, length of flower, hypanthium, sepals and petals, nectary chamber, crown longest series, androgynophore, stamens and ovary, relative constriction above nectary chamber, and bract/hypanthium length ratio. At the species level, 28 characters are more important and they are related to dimensions of stipule, leaf lobation (angle between lateral nerves, shape of central lobe, length of lateral lobe, distance between leaf sinus and petiole insertion), number of laminar nectary glands, position of petiolar nectary glands, length of peduncle, diameter of hypanthium, length of gynophore, shape of petals and sepals, androgynophore/hypanthium length ratio (defining protrusion of gynoecium and androecium), and pedicel/peduncle ratio. At lower levels, variance among and within populations rarely contributes more than 20% of the total.

Fig. 1
figure 1

Relative variance components for 57 quantitative descriptors. Bold characters are used for traits displaying more than 50% of variance among subgenera

Full size image

Correlations and principal component analysis (PCA)

The Pearson correlation (r) matrix shows high coefficients ranging between 0.71 and 0.97 among dimensions of certain floral components such as petals and sepals, hypanthium, nectary chamber, operculum, androgynophore, stamens, and ovary. The length of the corona (FLFL) is correlated with the upper hypanthium diameter (FLHS, r = 0.75). Bract dimensions are moderately correlated with these traits (BRLE and BRWI, 0.33–0.64, but 0.75 with ovary length—LOV). Stem diameter (STDI) is correlated with peduncle branching length (PDLF, r = 0.73) and central lobe length (LELC, r = 0.72), which is due to the association of these traits in representatives of subgenus Astrophea.

From the 27 quantitative descriptors showing high variation at the subgenus level, 24 were selected for the PCA; the other three were discarded to avoid redundancy between shape ratios and the original traits. Five principal components with an eigenvalue superior to one were retained, representing 84% of total variation (Table 4). The first component (32%) is primarily associated with flower length (hypanthium, nectary chamber, androgynophore) and secondarily with the constriction of the floral cup above the nectary chamber. The second one (27%) is associated with flower width (length of bracts and length of corolla and corona elements) and bract shape. The third one (14%) is associated with peduncle branching, stem width, and leaf length, which relates it clearly to variation between subgenus Astrophea, and secondarily subgenus Deidamioides, on the one hand, and all other subgenera, on the other hand. The fourth one (5%) is associated with the number of nectaries on leaf margins, which essentially relates it to supersection Distephana (DC.) Feuillet & MacDougal. The last one (5%) is only correlated with leaf serration.

Table 4 Factor loadings from the principal component analysis (varimax normalized rotation) carried out on 24 quantitative descriptors
Full size table

Figure 2 presents the accessions in the three first axes, showing a clear grouping by subgenus and supersection. The representatives of supersection Tacsonia are placed on the right along the first axis, in relation to their long and wide flowers. A few accessions, with shorter flowers as P. luzmarina Jørgensen and P. pinnatistipula Cav., appear closer to the origin of this axis, together with individuals of P. manicata (Juss.) Pers. Passiflora trinervia of subgenus Decaloba (supersection Decaloba MacDougal & Feuillet) is placed even further on the right thanks to its very long floral tube; additionally, it is clearly separated on the second axis by its much narrower flowers and minute setaceous bracts. On the left side, subgenera Passiflora and Decaloba are not differentiated by the flower length-related axis 1, but by the second, flower width-related axis 2. On this second axis, we find the large-flowered P. alata Curtis and P. quadrangularis (series Quadrangulares Feuillet & MacDougal of subgenus Passiflora) at one extreme, and those of the small-flowered P. arbelaezii L.Uribe and P. gracillima Killip of subgenus Deidamioides at the other one. As expected, the third axis clearly differentiates subgenera Astrophea and Deidamioides from the five others. In general, subgenera Passiflora, Decaloba, Astrophea, and Deidamioides, as likewise the supersections Tacsonia (subgenus Passiflora) and Decaloba (DC.) MacDougal & Feuillet (subgenus Decaloba), are clearly separated in the main tridimensional space. As expected, P. manicata (section Manicata (Harms) Feuillet & MacDougal) takes an intermediate position between supersection Tacsonia and the other sections of subgenus Passiflora. This species not only combines morphological traits typical of both supersections, but also intermediates eco-climatic requirements, as it may be found at lower elevations than tacsos (Tacsonia), but higher elevations than representatives of subgenus Passiflora (Ocampo et al. 2010). The representatives of the supersection Tacsonia that come closest to P. manicata are P. pinnatistipula and P. luzmarina, two tacsos with relatively shorter floral tubes. The former is also differentiated by a filamentous corona, instead of the typical reduced tacso coronas. Other species taking particular positions are P. foetida L. and P. vesicaria L. of subgenus Passiflora (supersection Stipulata Feuillet & MacDougal/section Dysosmia DC.), placed near both subgenera Passiflora (other supersections) and Decaloba, but closer to the former. Similarly, P. adenopoda DC. (supersection Bryonioides (Harms) MacDougal & Feuillet) is placed between subgenera Decaloba and Passiflora, due to its petiolar nectaries and the intermediate size of its flowers.

Fig. 2
figure 2

Tridimensional plot of the scores of Passiflora accessions for the first three quantitative variation components. Colors refer to subgeneric classification

Full size image

Qualitative variation among and within subgenera

Our first attempt to reduce the number of qualitative descriptors led us to retain 32 of them on the basis of their potential to discriminate among subgenera. The criterion was that the descriptor appears monomorphic or shows a highly dominant condition in at least one subgenus, while polymorphic among other subgenera. Three quantitative descriptors were categorized and added because of their high correlations with the principal components of quantitative variation. Thus, the first component was represented by androgynophore length (FLAL), the second one by sepal length (FLSL), and the third one by stem diameter and leaf length (STDI). The fourth and fifth ones were not included to avoid redundancy with very similar qualitative descriptors. Table 5 synthesizes the observations for these descriptors.

Table 5 Variation for 32 qualitative and four categorized quantitative descriptors in the different subgenera sampled
Full size table

The Colombian species of subgenus Astrophea exhibit the highest number of unique/rare traits including tree habit, wide stems of irregular section, very long leaves, the absence of tendrils, short triangular stipules, dorsal scar-like nectaries (appressed against or very close to petiole), branched peduncles, bright-yellow sickle-sword-shaped corona filaments, and tricostate ovaries. Unique and rare traits of the two species of subgenus Deidamioides include peduncle branching (shared with species of subgenus Astrophea and P. sexflora Juss. of subgenus Decaloba, although this structure is not homologous in these two cases; J.M. MacDougal personal comm.), the presence of tendrils at the axil of the peduncles, and the retuse leaf apex (unique in our sample, although this trait can be observed in individuals of P. emarginata Humb. & Bonpl., subg. Astrophea). Passiflora vitifolia Kunth (subgenus Passiflora/supersection Distephana) is differentiated by its tubular corona formed by the partial fusion of its elements and conspicuous nectary glands on leaf sinus and bracts, P. foetida and P. vesicaria (subgenus Passiflora/section Dysosmia) only by their pinnatisect bracts, and P. trinervia (subgenus Decaloba/supersection Decaloba) by the absence of a limen. Most of these traits are typical for each of these subgenera, ensuring that they will not bias the cluster analysis in terms of subgeneric classification. Subgenera Decaloba (except P. trinervia) and Passiflora do not show unique traits; however, they are clearly segregated by combinations of non-exclusive traits. Thus, in subgenus Decaloba, the presence of a flat hypanthium is only shared with subgenus Tryphostemmatoides, and the relatively small flower size with subgenera Astrophea, Deidamioides and the section Dysosmia of subgenus Passiflora. On the other side, species of subgenus Passiflora share wide flowers and the general presence of petiolar nectaries. Exceptionally, species of subgenus Decaloba (P. trinervia, section Decaloba DC.) may also present large red or pink corolla and long tubular flowers (androgynophore and hypanthium tube more than 11 cm long), typical of species pollinated by the sword-billed hummingbird, Ensifera ensifera Boissonneau. Floral tube length reaches extreme values in P. trinervia and Tacsonia species, with the exceptions of P. pinnatistipula, P. luzmarina and P. manicata related to their pollinators (relatively short-billed hummingbirds). In addition, supersection Tacsonia presents reduced coronas of short filaments or tubercles generally in one row only, while two-row coronas are most common in subgenera Decaloba, Deidamioides, Astrophea, and Passiflora supersection Distephana, and highly complex coronas (more than three rows) are typical in subgenus Passiflora (except supersection Tacsonia). Bracts are foliaceous in subgenus Passiflora. Fruit shape is generally globose to short ovate/obovate in subgenera Astrophea, Decaloba (except P. trinervia), Deidamioides, and Passiflora (supersections Passiflora, Laurifoliae, Distephana, and Stipulata), and oval to fusiform in supersections Tacsonia (with the exception of P. pinnatistipula) and Decaloba (P. trinervia). Fruit color seems also an interesting trait, with a particular frequency of blackish fruits in subgenus Decaloba; however, this descriptor could not be observed in all species.

Certain species show unusual trait combinations in their respective subgenera. This is particularly clear in subgenus Decaloba, where P. adenopoda shows foliaceous bracts, serrate leaf margins with conspicuous nectary glands, orbicular petiolar glands, and a uniseriate corona. Passiflora guatemalensis S.Watson also shows foliaceous bracts and glandless leaves as well as peltate leaves and a yellow uniseriate corona. Passiflora sexflora shows multiple peduncles. Passiflora coriacea, P. suberosa, P. capsularis L., and P. rubra L. lack bracts; in addition, the first two species show petiolar nectaries, while the last two lack such glands in all their organs and produce an elongated fruit with a bright red color.

Cluster analysis on the reduced descriptor list

Figure 3 presents a neighbor joining dendrogram obtained from the observations on the first set of descriptors. The three best represented subgenera, i.e., Passiflora, Astrophea, and Decaloba, are supported by the analysis; however, the former is clearly split between (i) carpenter bee-pollinated species with well-developed coronas (supersections Passiflora, Laurifolia and Stipulata) and (ii) hummingbird-pollinated species with long to very long flower tubes (supersections Distephana and Tacsonia). The bee-pollinated species of section Dysosmia of subgenus Passiflora also forms a distinct cluster. Its representatives, P. foetida and P. vesicaria, appear well differentiated from one another and take a distinct position between subgenus Passiflora and subgenera Astrophea and Decaloba. The consistency of this subclassification of subgenus Passiflora compensates for the low associated bootstrap values.

Fig. 3
figure 3

Dendrogram obtained with a first set of qualitative data. Distances of Sokal & Michener. Images courtesy of PhyloPic (phylopic.org)

Full size image

On the left side of the dendrogram, we find the three subgenera whose species produce small to medium sized flowers and fruits, with relatively simple coronas of generally two rows of filaments (rarely one or three), where petiolar nectaries are rare, with haploid chromosome numbers of 12 for three species (subgenus Astrophea) and 6 for the others. As expected from the number of their rare traits, the three species of subgenus Astrophea appear very uniform and clearly separated in a very distant cluster. Passiflora trinervia (subgenus Decaloba) is placed on another long branch inserted at the same position. A third much larger cluster is constituted by all the species of subgenera Deidamioides and Decaloba, except for P. trinervia and P. adenopoda (subgenus Decaloba). The latter species is placed on a well-separated branch inserted in an intermediate position, very close to the Dysosmia section, between the Passiflora–Distephana–Tacsonia–Manicata clusters and the Astrophea–Decaloba–Deidamioides clusters (Fig. 3). Indeed, this species shows several unusual features and trait combinations when compared to Decaloba as a group. Although the branch bearing the representatives of subgenus Deidamioides (P. gracillima and P. arbelaezii) is relatively long, it is clearly inserted within subgenus Decaloba, suggesting that the qualitative morphological differentiation of subgenus Deidamioides is fragile. This is consistent with the very low number of traits supporting it, but contrasting with the PCA results for quantitative traits.

The first set of qualitative data also allows distinguishing some structures within clusters corresponding to subgenera and supersections. Thus, within the Decaloba–Deidamioides cluster (Fig. 4a) there is a high level of divergence between subclusters whose composition is quite consistent with supersections or sections of the classification. Represented by one or two species only, supersections, Bryonioides (P. adenopoda), Hahniopathanthus (Harms) MacDougal & Feuillet (P. guatemalensis), Auriculata MacDougal & Feuillet (P. auriculata Kunth), and Cieca (P. coriacea and P. suberosa), appear well individualized in particular clusters. In the much larger, and better represented, supersection Decaloba, the morphological structure appears more diverse and complex, as it is split between two distinct subclusters. The one of its section Xerogona (Raf.) Killip, represented by two highly similar species (P. capsularis and P. rubra), is separated from the subcluster of its section Decaloba, including P. alnifolia Kunth and P. bogotensis Benth, P. sexflora, P. misera Kunth, P. trifasciata Lemaire, P. biflora Lam., and P. cuspidifolia Harms. All Decaloba species are consistently separated, except for P. capsularis and P. rubra; despite careful determination, we could not find a combination of traits consistently discriminating them.

Fig. 4
figure 4figure 4

Dendrogram obtained on the complete set of qualitative data. Distances of Sokal & Michener. a Decaloba–Deidamioides cluster, b Passiflora cluster, and c Tacsonia–Manicata cluster. Images courtesy of PhyloPic (phylopic.org) and photographs by John Ocampo

Full size image

Within the Passiflora cluster (Fig. 4b), one main subcluster corresponds to supersection Laurifolia (Cervi) Feuillet & MacDougal, with one branch for series Laurifoliae Killip ex Cervi (P. popenovii and P. nitida Kunth), one for typical representatives of series Quadrangulares (P. quadrangularis and P. alata), one for typical representatives of series Tiliifolia Feuillet & MacDougal (P. tiliifolia L. and P. ligularis), and one for P. maliformis. A second main subcluster is less homogenous as it gathers small clusters of species, mostly from supersections Stipulata and Passiflora. Thus, we observe three couples of closely similar species, respectively, formed by P. gibertii Brown and P. subpeltata Ortega (supersection Stipulata section Granadillastrum Tr. & Planch.), P. lehmannii Mast. and P. smithii Killip (supersection Stipulata section Kermesinae), and P. incarnata L. with the two forms of P. edulis. The placement of other species in this second main subcluster is less easy to understand: P. caerulea L., from section Granadillastrum is not closely associated with P. gibertii and P. subpeltata; similarly, P. cincinnata Mast. is not closely associated with P. incarnata and P. edulis.

Whitin the Tacsonia–Manicata cluster (Fig. 4c) there are three main branches. A first one includes the common and widely dispersed P. mixta, P. tripartita (section Tacsonia (Juss.) Harms), P. tarminiana, P. cumbalensis (Karst.) Harms and Ecuadorian narrow endemic species related to them: P. mathewsii (Mast.) Killip and P. luzmarina. The two cultigens are not clearly separated in the dendrogram, which must be interpreted with caution, given the poor bootstrap values associated with the groups where they are represented. A second one includes all the sampled species of section Colombiana L.K.Escobar (P. tenerifensis L.K.Escobar, P. flexipes Triana & Planch. and P. antioquiensis H.Karst. for series Leptomischae L.K.Escobar, P. linearistipula L.K.Escobar for series Quindiensae L.K.Escobar, P. lanata (Juss.) Poir. and P. adulterina L.f. for series Colombianae L.K.Escobar), plus the similar P. parritae (Mast.) Bailey and P. jardinensis L.K.Escobar of section Parritana L.K.Escobar. The last one includes the relatively short-tubed species P. pinnatistipula (section Insignes (Harms) Feuillet & MacDougal) and its hybrid P. x rosea (H.Karst.) Killip on one branch, and P. manicata (section Manicata (Harms) Feuillet & MacDougal) on the other branch. All species are much better separated in the last two subclusters, with much higher bootstrap values.

Discussion

Morphological diversity and subgeneric classification

At first sight, our factorial and cluster analyses provide good support for the subgeneric classification of Killip (1938). Indeed, eight of its 22 subgenera are represented in our sample, all of them individualized as specific clusters in the principal tridimensional space and in the neighbor joining analysis, except for subgenus Tryphostemmatoides Killip, which is only individualized in the former. Subgenus Manicata, proposed by Escobar (1988), is also individualized only in the factorial analysis. This general consistency of our results with the classifications followed by Killip (1938) and Escobar (1988, 1989) is not surprising as they were essentially based on morphological observations. Representing in Fig. 3 the main pollinators of the different clusters, and keeping in mind the cytogenetic information available, another picture emerges, which points to evolution processes at very different time scales. The three main subgenera of the new classification (Feuillet and MacDougal 2003), associated with divergent cytogenetic groups, occupy clearly different sectors of the morphological dendrogram, except for a few intermediate species of sections Dysosmia (subgenus Passiflora) and Bryonioides (subgenus Decaloba), originating from divergent genetic stocks but sharing much of their pollination syndrome. Other divisions concern supersections whose divergence can be explained by more recent evolutions corresponding to dramatic shifts in plant-pollinator adaptation. Thus, the positions of the Tacsonia and Distephana clusters reflect their coevolution with hummingbirds, with the extreme specialization of the former in its association with a unique long-billed species. Jørgensen and Vásquez (2009) even suggested the possibility that supersection Tacsonia evolved from hillside species of supersection Distephana by adapting to higher altitudes. However, the fact that rare species of subgenus Decaloba, such as P. trinervia, have followed a similar evolution, rather points to the impressive capacity for pollination syndrome shifts in genus Passiflora. Other examples can be found in the isolated shifts to bat pollination in subgenera Passiflora, with P. mucronata Lam. (supersection Stipulata, Sazima and Sazima 1978) and P. unipetala P.Jørg. (supersection Tacsonia, Møller Jørgensen et al. 2012), Decaloba with P. penduliflora Bert. (supersection Decaloba, Kay 2001), and Deidamioides with P. ovalis Vell. ex Roemer (section Tetrastylis (Barb.Rodr.) Harms, Buzato and Franco 1992). Undoubtedly, with the introduction of supersections within a smaller number of subgenera, the new classification allowed separating and interpreting processes of very different time depth in Passiflora evolution and diversification.

Morphological and molecular diversity

To appreciate the organization of Passiflora morphological diversity, we can compare some of the interspecific associations or divergences with results obtained in phenetic studies based on biochemical and molecular markers, on samples including some of the species we studied. A first series of genetic studies were carried out on smaller samples, mostly from Colombia. In the trees obtained with RAPD and cpDNA RFLP markers by Fajardo et al. (1998) and Sánchez et al. (1999), the species of supersection Tacsonia constitute one subcluster within a larger cluster of subgenus Passiflora species. Subgenus Decaloba is represented by P. coriacea and P. adenopoda, both species strongly diverging from this Passiflora–Tacsonia cluster, but also between themselves, which is consistent with our results. In the RAPD study, supersection Distephana and subgenus Astrophea are represented, respectively, by P. vitifolia and P. spinosa (Poepp. & Endl.) Mast., and both species are placed at a considerable distance from the Passiflora–Tacsonia cluster, their divergence being intermediate between that of P. adenopoda and that of P. coriacea. Within the Tacsonia subcluster, the distances between P. tripartita var. mollissima, P. cumbalensis, P. pinnatistipula, and P. antioquiensis Karst. follow the same order as in our morphological cluster. This result is still the same when the comparison is extended to sections Colombiana and Manicata, using the results of Segura et al. (2002, 2003) with AFLP markers (P. tenerifensis and P. parritae) Bailey also included, and with isozymes, although the P. antioquiensis Karst. node is placed closer to the most common tacsos than the P. pinnatistipula node in the isozyme study. The basal position of the P. manicata node in the Tacsonia AFLP study is consistent with its morphological characteristics and climatic adaptations, intermediate between supersections Tacsonia and Passiflora. Another convergence between morphological and AFLP markers is the clear separation of P. maliformis from the typical species of series Tiliifolia and Quadrangulares of subgenus Passiflora (Ocampo et al. 2004).

Genetic relationships between subgenera or supersections and between particular species can also be deduced from subsequent phylogenetic studies carried out on wider species samples representing at least the three main subgenera of Feuillet & MacDougal’s classification by Muschner et al. (2003) with ITS, trnL-trnF and rps4 sequences, by Yockteng and Nadot (2004a) with chloroplastic matK sequences, Yockteng and Nadot (2004b) with sequences of the nuclear chloroplast-expressed glutamine synthetase gene (ncpGS), Hansen et al. (2006) with trnL/trnT sequences, and Krosnick et al. (2013) with nrITS, ncpGS, cp trnL-F, and ndhF sequences. All these studies support the existence of three major monophyletic clades, one corresponding to subgenus Astrophea, a second one formed around subgenus Decaloba, and a third one formed around subgenus Passiflora. When present (i.e., in all studies except that of Muschner et al. 2003), species of supersection Tacsonia form a subclade within the Passiflora clade, except for P. trifoliata Cav. and P. trisecta Mast., which remain isolated from the other tacsos in the study of Hansen et al. (2006). Other clear morphological groups of subgenus Passiflora do not show such consistency between morphological and genetic clusters. For example, none of the cited genetic studies allow identifying a genetic cluster that can be unambiguously related to supersection Distephana or the very uniform series Laurifoliae Killip ex Cervi.

The relative positions of the three major clades differ among studies: Their relative positions are clearly not resolved in the trnL-trnT trees of Muschner et al. (2003) and Hansen et al. (2006), as well as in the consensus tree of the combined ITS + trnL-trnT data of the former group; subgenus Astrophea is sister to a Passiflora–Decaloba clade in the combined maximum likelihood tree of Muschner et al. (2003) and the ncpGS tree of Yockteng and Nadot (2004a), while subgenus Passiflora is sister to Astrophea–Decaloba in the ITS tree of Muschner et al. (2003). Logically, the position of the few morphologically intermediate species is also variable in these studies. Passiflora foetida (section Dysosmia) is clearly placed within the Passiflora clade according to rps4 and trnL-trnF sequences, but more distant, and closer to the basal node of this clade, in the trees obtained with ITS and ncpGS sequences, as well as in the combined tree of Krosnick et al. (2013). Passiflora adenopoda is closer to the basal node of Decaloba clade in the ncpGS tree as well as in the ITS study of Krosnick and Freudestein (2005). Passiflora morifolia Mast. another species of supersection Bryonioides is also closer to the basal node of the Decaloba clade, on a branch between Decaloba and Astrophea in the ITS and trnL-trnF trees. Likewise, the latter two species were resolved as part of supersection Bryonioides by Krosnick et al. (2013) with four molecular markers (nrITS, ncpGS, cp trnL-F and ndhF). In the ncpGS tree of Yockteng and Nadot (2004a) and the ITS study of Krosnick and Freudestein (2005), subgenus Deidamioides, represented by P. tryphostemmatoides and by P. arbelaezii, respectively, is placed close to subgenus Astrophea. The study of Krosnick et al. (2013) indicates that, in fact, subgenus Deidamioides is polyphyletic and that this sister relationship with subgenus Astrophea concerns its section Tryphostemmatoides Harms, which Yockteng and Nadot (2004a) suggested reestablishing at the subgenus rank.

Taking into account that subgenus Deidamioides is only represented by its section Tryphostemmatoides, the general structure presented in Figs. 2 and 3 is consistent with molecular phylogenies. The three main subgenera are clearly separated, and there is no particular association between two of them. Concerning the smaller subgenus Deidamioides, only the few representatives of its section Tryphostemmatoides appear related to subgenus Decaloba in the neighbor joining analysis; however, they are clearly dissociated from the three major clades in the factorial analysis, taking a very distinct position, much closer to subgenus Astrophea, as in the molecular studies. Low-rank taxa that appear intermediate between the three clades in the phylogenetic studies (sections Dysosmia and Bryonioides of subgenera Passiflora and Decaloba) are also intermediate in our phenetic study.

We can also draw elements of comparison between phylogenetic and phenetic studies within the two most diverse subgenera, Decaloba and Passiflora. Within subgenus Decaloba, the phylogeny obtained by Krosnick et al. (2013) supported six of the eight supersections recognized by Feuillet and MacDougal (2003), while supersections Multiflora (Small) MacDougal & Feuillet and Auriculata appeared paraphyletic (Krosnick et al. 2013). Among the five supersections represented in our study, four are strongly individualized: Bryonioides, Hahniopathanthus, Auriculata, and Cieca; however, their branching order is not consistent with phylogenetics: Most obviously, P. adenopoda (supersection Bryonioides) and P. trinervia are not nested in the Decaloba–Deidamioides cluster; within the latter, sections Decaloba and Xerogona of supersection Decaloba diverge at the same level as supersections, so that taking into account the position of P. trinervia, supersection Decaloba is dispersed in three very distinct branches of our tree. Krosnick et al. (2013) also identified problems with species of section Decaloba, such as P. lutea L., P. filipes Benth., and P. pavonis Mast., which are in need of revision in future phylogenetic studies. In any case, the clear differentiation of P. trinervia must not be overinterpreted in phylogenetic terms, as it is obviously related to its very particular floral morphology, adapted to pollination by the sword-billed hummingbird.

With the exception of supersection Tacsonia, the Passiflora clade identified in wide phylogenetic studies generally shows loose relations among species. Other clear morphological groups do not show consistency with genetic clusters. Consequently, the interpretation of the poorly supported subclades is very uneasy, with the partial exception of the ncpGS tree presented by Yockteng and Nadot (2004b), where branches are better defined although not easier to interpret, given, for example, the dispersion of species belonging to series Passiflora and Laurifoliae. Among the close associations documented by our morphological study, we can only recognize those of P. quadrangularis with P. alata and P. incarnata with P. edulis (ITS and ncpGS trees). Part of this lack of correspondence between genetic and morphological data may originate in relatively frequent interspecific gene flow (Ocampo et al. 2016) and consequent reticulate evolution in Passiflora. Paternal and biparental inheritance of cytoplasmic DNA further affects studies based on cpDNA markers, particularly those on subgenus Passiflora (Muschner et al. 2006, Hansen et al. 2007).

Among the short-flowered, insect-pollinated, species of subgenus Passiflora, morphological subclustering is more consistent than phylogenetic grouping, albeit not free from problems either. On the one hand, we can draw three reasonable conclusions: (1) The high divergence of the Dysosmia morphological cluster validates the placement of this section directly under subgenus Passiflora, without any subordination to a supersection (Vanderplank 2013); (2) one clear major group validates supersection Laurifolia, where even closer associations can be recognized among typical representatives of series Laurifoliae, Tiliifolia, and Quadrangulares, contrasting with the fuzzy genetic information; (3) concerning P. maliformis, neither its ancient classification in series Tiliifolia, nor its transfer to series Laurifoliae are validated. In the other major subcluster, only small subgroups gathering the representatives of a couple of very similar species can be recognized, and the relationships between neighboring small groups are very difficult to interpret. On the whole, it seems that only morphologically very uniform groups can be recognized as series or “micro-series” and that few combinations of characters can structure this profuse diversity of short-flowered species in subgenus Passiflora.

Supersection Tacsonia is represented in two of the above-mentioned wide studies. In that of Hansen et al. (2006), they form two subclades within subgenus Passiflora, a small one including P. trifoliata Cav. and P. trisecta Mast., and a main one with six species, whose structure is too weak for useful comparisons with morphological clusters. In the study of Yockteng and Nadot (2004a), they form one poorly resolved subclade, including P. trisecta and several species present in our sample. Passiflora mathewsii is placed close to P. mixta, P. manicata, and P. antioquiensis a little further. This information is poor but consistent with morphology. More recently, Abrahamczyk et al. (2014) used the trnL-F and ndhF plastid markers and the ITS and ncpGS nuclear markers for their phylogenetic study of supersection Tacsonia including 37 species. All our morphological clusters find an equivalent in their study: (1) the close association of P. tripartita var. mollissima, P. mixta, and P. mathewsii from section Tacsonia (Juss.) Harms, together with P. tarminiana, assigned to section Elkea Feuillet & MacDougal; (2) this cluster is also associated with a small cluster of two representatives of section Elkea, P. luzmarina, and P. loxensis Killip & Cuatrec.; (3) another relatively important cluster gathers representatives of section Colombiana, including P. lanata and P. adulterina (series Colombianae), P. leptomischa Harms (series Leptomischae); (iv) the more basal node is that of section Insignes, including P. pinnatistipula. There are only three relative differences: P. formosa Ulmer and P. crispolanata L.Uribe (series Colombianae) do not cluster with the other members of section Colombiana. Passiflora parritae (section Parritana) and P. manicata are placed in intermediate positions between the Tacsonia/Elkea cluster and the Colombiana cluster. In the morphological tree, section Parritana (P. parritae and P. jardinensis) is placed slightly closer to the Colombiana subcluster, while P. manicata is placed closer to the Insignes subcluster.

The subclassification of supersection Tacsonia is clearer than that of short-flowered species of subgenus Passiflora. However, the subdivisions recognized by Escobar (1988) and endorsed by Feuillet and MacDougal (2003) are not clearly supported by the combination of morphological and genetic data discussed here. In all above-mentioned studies, P. tarminiana has shown a very close affinity with the couple formed by P. mixta and P. tripartita. All three species hybridize easily, although each one recovers its own morphology and genetic distinctiveness after very few generations (Segura et al. 2003, 2005; Primot et al. 2005). They should be included in the same series. Similarly, species of the P. cumbalensis group (section Elkea) have regularly shown affinity with those common species of section Tacsonia. They can also hybridize with them, although the high frequency of abnormal phenotypes in the successive segregating hybrid and backcross progenies (Schoenigger 1986) confirms a slightly more distant relationship. They should be considered a neighbor series within the same section. The same kind of comparison may hold for sections Colombiana and Parritana, considering that the differences between their series are limited. Thus, the differentiation of species with “normal” peduncles from those with extremely long peduncles (series Leptomischae of section Colombiana) appears to be a slight one. The two species of series Colombianae, P. lanata and P. adulterina, show a very high morphological similarity, with distances remaining within the range of intraspecific variation. In this respect, it must be noted that our two specimens of P. lanata were typical for all characters, and they exhibited a lanate ovary. Killip (1938) and Mutis and Uribe (1955) mentioned a glabrous ovary for this species, a trait not mentioned in the original description by de Jussieu (1805). Escobar (1988) described the plant as pubescent, except for the leaf upper face, with lanate flowers and fruits, which necessarily implies that the ovary is lanate too.

As a first major point of conclusion on this comparison between morphological and molecular diversity, we can underline that the major morphological divisions observed in our study find support in the genetic studies. The cytological groups are always validated with a clear separation of subgenera Astrophea (n = 12) and Deidamioides and Decaloba (n = 6), both among themselves and from subgenus Passiflora (n = 9). Concerning subgenus Deidamioides, the consistency between morphological and genetic studies is clear only in our factorial analysis, where it is associated with subgenus Astrophea mostly on peduncle traits (third principal component). This trait is also represented in the qualitative descriptors, but its effect is blurred by the high number of traits shared with subgenus Decaloba. While the comparison is difficult for subgenus Deidamioides/Tryphostemmatoides, it is impossible for P. trinervia, which is not represented in molecular studies. Passiflora adenopoda and P. foetida and their close relatives that take an intermediate position in the general morpho-cytological pattern are consistently placed in intermediate positions. In most phylogenetic studies P. adenopoda or P. morifolia (supersection Bryonioides of subgenus Decaloba) appear closer to the root of a general supersection or section Decaloba clade, and P. foetida and P. vesicaria (section Dysosmia) are placed closer to the root of the general Passiflora clade than most species (Krosnick et al. 2013).

The comparison becomes more difficult at lower infrasubgeneric levels, depending on the subgenus considered. Subgenus Decaloba appears better structured than the other subgenera, and its morphological and molecular diversity patterns appear relatively consistent (Killip 1938; Feuillet and MacDougal 2003; Krosnick et al. (2013). In our results, differentiation appears higher in subgenus Decaloba than in subgenera Passiflora and Tacsonia, with clearly distinct morphological clusters. Their order is different from that in genetic studies, which is not surprising in such a wide and diverse clade, as strong morphological changes may be related to minor changes in gene sequences. More problematic is the fact that the largest supersection Decaloba is split among several distant clusters. This divergence between morphological and genetic data should be further investigated, considering also the polyphyly of several supersections of subgenus Decaloba in the phylogeny of Krosnick et al. (2013). The placement of P. adenopoda on a well-separated branch may look surprising, as it is not consistent with either classification Decaloba section Pseudodysosmia (Harms) Killip in Killip (1938), or supersection Bryonioides in Feuillet and MacDougal (2003); however, this species shows several unusual features (e.g., sturdy hooked trichomes) as compared to subgenus Decaloba as a group. Passiflora foetida (section Dysosmia) takes a very similar position, but the convergence is purely morphological, as has been shown in molecular studies.

Interestingly, these two problematic species materialize the separation between the two cytogenetic groups in our tree. Indeed, chromosome counts for P. adenopoda give 2n = 12 (MacDougal 1994) as in most species of subgenus Decaloba, while those for P. foetida are 2n = 18, 20, or 22 (Snow and MacDougal 1993; De Melo et al. 2001; Souza et al. 2008). According to De Melo et al. (2001) and De Melo and Guerra (2003), P. foetida appears cytologically quite isolated but closer to the n = 9 group, its smaller chromosomes and articulate interphase nuclei being similar to species with n = 6, while its chromosome number, higher karyotype symmetry, CMA staining properties, and the number of 45S rDNA sites make it similar to species of subgenus Passiflora. In any case, more species should be gathered in a same morphological study before revising objectively the morphological classification.

Within the n = 9 group, phylogenetic studies place supersections Tacsonia and Distephana within a Passiflora clade. However, only the former constitutes a distinct and consistent group, while our morphological analysis supports both groups and the group of insect-pollinated species at the same level of differentiation, as in the classification of Killip (1938). This strong morphological differentiation is obviously related to the fact that species of supersections Distephana and Tacsonia have developed ornithophily. It is likely that they coevolved independently with hummingbirds. Indeed, we see no arguments, either morphological or genetic, supporting an underlying monophyly that could explain this relative convergence. Instead, the difference in floral tube length only explains a small part of their morphological divergence, as can be easily deduced by the proximity between relatively short-flowered species as P. mathewsii and P. luzmarina and their close long-flowered relatives, respectively, P. mixta and P. cumbalensis, in our morphological tree. Another indication comes from the parallelism between our results and those of Abrahamczyk et al. (2014): If tube length had a very significant impact on our morphological classification, section Insignes would be better separated in the morphological tree, as compared to the molecular phylogeny tree, which is not the case. Thus, the considerable distance between P. vitifolia and the Tacsonia cluster is not simply related to a difference in the bill of their pollinator. Many more combinations of morphological characters are involved. A last argument is that supersections Distephana and Coccinea Feuillet & MacDougal have separated from supersection Passiflora much later (4.93 My ago according to Abrahamczyk et al. 2014) than supersection Tacsonia (10.7 My).

Whether the probable evolution of Tacsonia and Distephana from a “Passiflora-like” common ancestor justifies their inclusion in the bee-pollinated Passiflora subgenus as proposed in the new classification, it is just the same classical question about paraphyly and considering birds as dinosaurs. In the end, it seems an issue of giving more emphasis to the adaptative forces commanding evolution or to the genetic structure that subtends them. Here the argument of paraphyly is quite strong, as the evolution of supersection Tacsonia and Distephana is much more recent and their separation from subgenus Passiflora has not been established. It is reinforced by the lability of the pollination syndrome, as evidenced by the many cases of Passiflora species having shifted from insect pollination to bat pollination, or even from bird pollination to bat pollination (Sazima and Sazima 1978; Buzato and Franco 1992; Kay 2001; Møller Jørgensen et al. 2012; Abrahamczyk et al. 2014).

What is important in the case of supersection Tacsonia is that the shift to ornithophily is at the same time ancient (10.7 My) but recent when compared to the evolution of the n = 9 cytological group (around 24 My; Abrahamczyk et al. 2014) and that its radiation is closely related to the conquest of the high elevation habitats to which they are very specifically adapted (Ocampo et al. 2010), as well as their main pollinator. We have underlined above that the introduction of the supersection rank in the new classification allowed to take into account different levels of evolutionary history.

Concerning subgenus Passiflora sensu Killip, no clear structure appears at the interspecific level that could result in clear subdivisions into series. The study of sequence variation for the ncpGS gene (Yockteng and Nadot 2004b) provides the only tree with reasonably well-supported structure at this level. However, several obvious abnormalities question the robustness of the information. Our morphological observations only confirm closer associations between the most typical representatives of some series, although the lack of a clear hierarchy in the branch structure point to the difficulty of the work and the risk of under or overclassification, which will lead to choose between a limited number of poorly supported series or a great number of poorly represented series.

Similarly, the structure of the Tacsonia–Manicata branch of our dendrogram does not clearly support the recognition of many sections and series in supersection Tacsonia, but it allows differentiation between four groups of tacsos in our sample: one corresponding to common species that probably have their center of diversity in Ecuador as is obvious for P. cumbalensis, P. luzmarina and P. mathewsii, and very likely for P. mixta, P. tripartita, and P. tarminiana (Segura et al. 2005); another cluster including only species endemic to Colombia with a slight but clear differentiation related to extreme variation in peduncle length; one for section Insignes with species of variable flower tube length, having maintained a minimal filamentose corona; and one for section Manicata, intermediate between the long-flowered species of supersection Tacsonia and the shorter-flowered species of other supersections of subgenus Passiflora.

Relationship between Passiflora capsularis and Passiflora rubra

In our study, supersection Cieca of subgenus Decaloba is represented by six accessions of P. capsularis and four accessions of P. rubra. Our results (Fig. 4) show that the distinction between these species should be reconsidered, as the use of so many descriptors did not allow any consistent discrimination, indicating that the very few diagnostic characters used for their identification (color of the base of the corona and ovary pubescence) may in fact vary at the infraspecific level, allowing different combinations. Another argument against the distinction of two species comes from the fact that our samples could be collected in the same perturbed habitat. The presence of two so similar species growing spontaneously in the same ecological niche seems in contradiction with the principle of competitive exclusion.

Conclusions

A shorter list of 32 qualitative traits selected after analyzing variation among Killip’s and Feuillet & MacDougal’s subgenera allowed us to consistently classify the 61 species sampled in our study, using a phenetic approach. Most discriminant characters include size of stems and leaves, the presence of tendrils, number and distribution of extrafloral nectaries, dimensions and general shape of bracts, width and length of flowers, corona complexity, and, although they were not systematically analyzed, fruit size and color. Furthermore, the smaller number of descriptors providing valuable information used in this approach will also allow a reduction in the labor, time and resources spent in the characterization of genetic resources of genus Passiflora.

Concerning lower levels of Passiflora classification, our morphological study shows different situations in the main subgenera and supersections. It provides general support to supersections within subgenus Decaloba, albeit pointing to some inconsistencies in its highly diverse supersection Decaloba. Within subgenus Passiflora, section Dysosmia is validated both morphologically and genetically. Supersection Tacsonia seems to be overclassified. For example, sections Tacsonia (5 species) and Elkea (15 species) could be downgraded and considered series within a same section. Similarly, the integration of section Parritana (2 species) in section Colombiana (19 species) should be considered. The other supersections of subgenus Passiflora are not supported by genetic studies, not even supersection Laurifolia which is so clearly supported by the morphological analysis. In this context, morphological data appear much more consistent and useful for classification. Even so, only morphologically uniform groups can be clearly recognized as series or “micro-series,” but few combinations of characters appear to structure the profuse diversity of short-flowered species in subgenus Passiflora. Thus, the lack of a clear hierarchy points to the difficulty of the work and the risk of under or overclassification, i.e., the recognition of a limited number of poorly supported series or a great number of poorly represented series.

At the species level, our study has shown the fragility of the distinction between P. capsularis and P. rubra, as trait combinations have allowed distinguishing among populations, not between species. Variation between them should be considered at the intraspecific level.