Balsaminaceae comprises over 1,000 species, mainly distributed in the Old World tropics and subtropics, with only few species occurring in northern hemisphere temperate regions (Grey-Wilson 1980a; Yuan et al. 2004; Janssens et al. 2009). The family is characterized by spurred zygomorphic flowers with fused stamens surrounding ovary and stigma. Several genera have been described within Balsaminaceae (e.g. Nieuwland and Lunnel 1916; Perrier de la Bâthie 1927; Peter 1928), but currently only Impatiens and the monospecific Hydrocera are recognized as valid genera (Rao et al. 1986; Grey-Wilson 1989a). Impatiens differs from its sister genus in having fused lateral petals and an explosive seed capsule whereas Hydrocera is characterized by five free petals and an indehisent pseudoberry. Impatiens occurs mainly in montane areas of the tropics and subtropics of the Old World (tropical Africa, Madagascar, South India and Sri Lanka, Sino-Himalaya and Southeast Asia) (Fig. 1). In contrast, Hydrocera is native to Indo-Malayan region, where it occurs as a semiaquatic plant in stagnant water around sea level (Grey-Wilson 1980b; Ramadevi and Narayana 1990).

The morphological classification of Balsaminceae is supported by molecular data, which showed that no combinations of previously recognized genera were mutually monophyletic (Yuan et al. 2004; Janssens et al. 2006; Yu et al. 2015). The same molecular systematic studies also showed that the most comprehensive infrageneric treatment of Impatiens from Warburg and Reiche (1895), largely based on phyllotaxy, inflorescence type, and spur length, is artificial (Yuan et al. 2004; Janssens et al. 2006). The discordance between morphological and molecular data has complicated the taxonomic subdivision of the large genus Impatiens. However, by combining molecular and morphological data in a phylogenetic analysis implementing a total-evidence approach, a new global infrageneric classification of Impatiens was proposed (Yu et al. 2015). According to this new classification Impatiens is divided into I. subg. Clavicarpa S. X. Yu ex S. X. Yu & Wei Wang, characterized by a four-locular ovary and one ovule per locule, and I. subg. Impatiens, characterized by a five-locular ovary (rarely four) and many ovules per locule. Within I. subg. Impatiens, seven sections are delimited.

The largest concentration of Impatiens species is found in China and Southeast Asia, including Myanmar, Thailand, Laos, Cambodia, Singapore, Malaysia, Indonesia, the Philippines, Timor-Leste, Brunei, and Papua New guinea (Hooker 1874; Shimizu 1970; Grey-Wilson 1989b; Kress et al. 2003; Chen et al. 2008). Taxonomic treatment of species from this region has been problemetic, mainly due to insufficient studies. One of the sections of Yu et al. (2015), Impatiens section Semeiocardium (Zoll.) S.X. Yu & Wei Wang, has been particularly problematic. The taxonomic status, delimition, and diagnostic characters of I. sect. Semeiocardium have all frequently changed (Bleeker 1858; Backer 1935; Grey-Wilson 1989a; Utami 2009; Yu et al. 2015). This taxon was originally described by Zollinger as a monotypic genus with Semeiocardium arriensii Zoll., from the Indonesian Madura island, as the type species (Bleeker 1858). However, Zollinger produced only a short and ambiguous description with a wrong diagnosis (Bleeker 1858; Backer 1935; Grey-Wilson 1989a), based on an incomplete type specimen. In addition, the type specimen also contained plant parts of a Polygalaceae (Backer 1935). Backer (1935) re-discovered S. arriensii, confirmed the generic status of Semeiocardium, and included a more detailed description based on living material. Grey-Wilson (1989a) summarized the three characters that had been used by Bleeker (1858) and Backer (1935) to distinguish Semeiocardium from Impatiens: 1) the outer lateral sepals enveloping the lower sepal and spur are connate as opposed to free, 2) the lateral united petals are connate as opposed to free, and 3) the ovary is four-locular as opposed to five-locular. He concluded that these character states are not unique to Semeiocardium, but can also be observed in other species such as I. harmandii Hook. f. from Thailand with connate outer lateral sepals, I. kerriae Hook. f. from Thailand with connate lateral united petals, and various Southeast Asian Impatiens species with four-locular ovaries (Grey-Wilson 1989a). As a result, Grey-Wilson synonymized Semeiocardium with Impatiens (Grey-Wilson 1989a). Utami (2009) changed the status of Semeiocardium to subgeneric level as Impatiens subgenus Semeiocardium (Zoll.) N. Utami, based on studies of floral morphology and seed coat ornamentation. She included the type species of the genus Semeiocardium, I. arriensii (Zoll.) T. Shimizu, together with 19 other species distributed from South China to South Indonesia that all possess a four-locular ovary, connate lateral united petals, and a pilose or granulate and finely granular seed coat with long hairs (Utami 2009). In the phylogenetic analysis of Yu et al. (2015) a monophyletic clade with brick-shaped seed as synapomorphies, including eight Impatiens species, was re-defined as I. sect. Semeiocardium within I. subg. Impatiens. In the diagnostic key of Yu et al. (2015) species are classified as belonging to I. sect. Semeiocardium if they possess a four-locular ovary, many ovules per carpel, and obconical capsules.

Fig. 1.

A–G. Representatives of Impatiens from each clade in this study. A. I. damrongii (clade 1). B. I. kerriae (CLUP clade). C. I. psittacina (CLUP clade). D. I. longiloba (clade 2). E. I. racemosa (clade 2). F. I. hians (clade 3). G. I. phuluangensis (clade 3). H. The distribution of Impatiens: the grey area shows the distribution of a small number of widespread species. The other colors represent one of the five Impatiens hotspots. Estimated species number is given. The yellow area represents the distribution of species from the CLUP clade, the callout bullets show the number of species with connate lateral united petals in China and SE Asia respectively.

Although the molecular systematic studies in Impatiens performed to date have included a broad species sampling, in any study less than 3% of representatives from Southeast Asia were included (Yuan et al. 2004; Janssens et al. 2006, 2007, 2009; Yu et al. 2015). Given the evolutionary lability of morphological characters, robust taxonomic classifications require that a large number of species be included to test whether a particular diagnostic character, or combination of characters, can be used to recognize monophyletic taxa. Based on the lack of SE Asian taxa in molecular systematic studies, more taxa from this area need to be included in a phylogenetic analysis. The aim of this study was to: 1) test the monophyly of I. subg. Semeiocardium and I. sect. Semeiocardium using molecular phylogenetics, and 2) reconstruct the evolution of the diagnostic morphological characters, which have been used throughout the taxonomic history of Semeiocardium. We expanded the taxon sampling of previous phylogenetic studies in Impatiens by adding a large number of Asian taxa, including many representatives of the former I. subg. Semeiocardium.

Materials and Methods

Taxon Sampling—In total, 164 accessions of Balsaminaceae (one Hydrocera, the remainder Impatiens) were included in our analysis. We expanded the previous data set of Yuan et al. (2004) and Janssens et al. (2006) with 76 new accessions (Appendix 1). The majority of the new accessions are from Southeast Asia and they include 15 out of 20 recognised species from I. subg. Semeiocardium. Many newly added taxa await formal description as new species, and are here indicated with informal names between quotation marks. Marcgravia umbellata, Souroubea sp., Norantea guianensis (Marcgraviaceae), and Pelliciera rhizophorea (Tetrameristaceae s. l.), are used as outgroup (Bremer et al. 2002; APG II 2003; APG III 2009). Pelliciera rhizophorea was used to root the tree in the phylogenetic analyses.

Molecular Protocols—Genomic DNA was extracted from fresh or silica-dried leaf material and herbarium specimens using a reduced CTAB protocol (Doyle and Doyle 1987). The nuclear ribosomal internal transcribed spacer (“ITS” hereafter) and the plastid atpB-rbcL intergenic spacer (“atpB-rbcL” hereafter) were amplified following Yuan et al. (2004) and Janssens et al. (2006). The total PCR reaction volume was 50 µl and the reaction mix, modified from Techaprasan et al. (2010), contained 1 × PCR buffer (Thermo Scientific, Massachusetts, U. S. A.), 2 mM MgCl₂ 0.2 mM dNTPs, 0.2 mM of each forward and reverse primer (ITS5 and ITS4 derived from White et al. (1990), IMP-atpB and IMP-rbcL derived from Janssens et al. (2006)), 0.05 u/µl Taq DNA polymerase (Thermo Scientific, Massachusetts, U. S. A.), and 4–8 ng/µl genomic DNA. Sequencing of amplicons of both regions, using the original amplification primers as the sequencing primers, was done by Macrogen (the Republic of Korea) and first Base (Singapore), respectively.

Data Matrix Composition and Phylogenetic Analyses—The chromatograms of the ITS forward and reverse sequences showed no evidence of polymorphism. For both loci forward and reverse sequences were therefore combined to create contigs in Geneious 6.1 (Biomatters Ltd, New Zealand). Sequence data matrices were initially aligned with ClustalW (as implemented in BioEdit version 7.0.5.2 (Hall 1999)) using default settings, and then manually adjusted in BioEdit. A partition homogeneity test implemented in PAUP* 4.0b10 (Swofford 2003) was used to determine whether the datasets were incongruent (Farris et al. 1995).

Maximum parsimony analysis (MP) of the separate datasets was conducted using PAUP*. Characters were equally weighted and character states were specified as unordered. Heuristic searches were conducted with tree-bisection-and-reconnection (TBR) branch swapping on 10,000 random addition replicates, with 10 trees held at each step. Support for individual clades was measured by non-parametric bootstrapping (MP-BS) (Felsenstein 1985), using 1,000 pseudoreplicates. For each pseudo-replicate, a heuristic search for the most parsimonious tree was performed by TBR branch swapping on 100 random addition starting trees, with 3 trees held at each step.

For Bayesian analysis, Modeltest 3.06 (Posada and Crandall 1998) was used to determine the best models of sequence evolution for each locus using the Akaike Information Criterion. Bayesian analyses of separate datasets were conducted with MrBayes 3b4 (Huelsenbeck and Ronquist 2001). Two simultaneously independent analyses were run for ten million generations, starting from different random trees. Every 500 generations, a tree was sampled from the chain, resulting in a total of 20,000 sampled trees. The first 25% of sampled trees were discarded as burn-in. Tracer v1.6 was used to check whether stationarity was reached among the post burn-in sample. Support for each clades was measured by posterior probability (PP). Data sets and consensus trees are deposited in TreeBASE (study number 16521).

Testing Monophyly of Taxonomic Classifications—We tested monophyly of taxa by visually inspecting topologies and assess whether taxa that belong to a previous classification of Semeiocardium form a monophyletic clade. To determine membership of previous classifications of Semeiocardium we refered to the list of species supplied by Utami (2009) and the diagnostic key of Yu et al. (2015).

Character Evolution—To test whether morphological characters used to define Semeiocardium are synapomorphies, we reconstructed the evolution of three floral characters across the genus using character optimization: 1) lateral united petals: connate or free, 2) number of ovary locules: four or five, and 3) outer lateral sepals: connate or free. The degree of fusion of the lateral united petals and the number of ovary locules were included because they were used by Utami (2009) to define I. subg. Semeiocardium. Furthermore, the degree of fusion of the outer lateral sepals was included, because this character was originally used to describe Semeiocardium as a distinct genus within Balsaminaceae (Bleeker 1858; Backer 1935; Grey-Wilson 1989a). Seed coat characteristics used by Utami (2009), and pollen characters used by Yu et al. (2015) were not included because of lack of sufficient data. Characters were scored using fresh material or, in case no fresh material was available, herbarium specimens or literature. We used Fitch parsimony (Fitch 1971) as a criterion for character optimization. To account for phylogenetic uncertainty, we traced character histories on 15,000 post burn-in trees from the Bayesian analysis using the ‘Trace Character Over Trees’ command in Mesquite 2.75 (Maddison and Maddison 2011).

Results

Phylogenetic Analyses—Visual inspection of topologies based on nuclear and chloroplast DNA sequences revealed several well-supported differences. The partition homogeneity test confirmed significant incongruence between the ITS and atpB-rbcL datasets (p = 0.001). For the remainder, nuclear and plastid datasets are analyzed separately. Topologies based on ITS sequences were better resolved and showed higher support than topologies based on atpB-rbcL sequences (Table 1; Fig. 2). In all consensus topologies, both Balsaminaceae and Impatiens were monophyletic (Fig. 2). Within Impatiens, the deepest split was between I. omeiana and the remainder of the genus (Fig. 2). According to the nuclear topology, the next split was between clade 1 containing mainly Southeast Asian species and clade A containing species from all distribution areas (Fig. 2). Clade 1 consists of the “connate lateral united petals clade” (CLUP clade) and a small sister clade comprising species with free lateral united petals: I. damrongii, I. aff. augustini, and I. stenosepala (Fig. 2). The CLUP clade contains most representatives of I. subg. Semeiocardium and many other Southeast Asian taxa: 43 out of a total of 75 Southeast Asian taxa sampled (Figs. 2, 3). Clade 2 comprises mainly Chinese taxa. Clade 3 consists of representatives from most Impatiens hotspots, and also included I. santasukii and I. muscicola of I. subg. Semeiocardium (Fig. 2).

In the plastid topology clades 1, 2, 3, and the CLUP clade was also monophyletic, confirming the non-monophyly of I. subg. Semeiocardium. However, clade 1 and 2 together (clade B), and clade 3 was sister to clade B (Fig. 2). Moreover, in clade 1, the CLUP clade was sister to I. damrongii + I. aff. augustini, and subsequently I. stenosepala (Fig. 2).

It was not possible to unambiguously determine whether species belong to I. sect. Semeiocardium. Several species are characterized by combinations of characters, which are typical for membership of several sections. We therefore were unable to test the monophyly of I. sect. Semeiocardium.

Table 1.

Characteristics of the ITS and atpB-rbcL datasets.

Fig. 2.

Summary of the bayesian consensus tree based on nuclear ITS (left) and plastid atpB-rbcL (right) DNA sequence data. Support values for key nodes are indicated (PP/MP-BS). Branches in red lead to the members of I. subg. Semeiocardium. The color of taxon names is indicative of their distribution.

Fig. 3.

Parsimony ancestral character state reconstructions of lateral united petals based on 15,000 post burn-in trees from the Bayesian inference plotted on the Bayesian consensus tree from the ITS (left) and atpB-rbcL (right) data. Taxa in green possess a four-locular ovary, whereas those in black (including all taxa that are part of clade 2 and 3) possess free lateral united petals. Circles at nodes represent pie charts indicating the percentage of Bayesian post burn-in trees with a particular reconstruction, equivocal reconstruction, or the node absent.

Character Evolution—Re-examination of the morphological characters for I. santisukii and I. muscicola revealed that their lateral united petals and outer lateral sepals are not connate but free, and that their ovary is five-locular. In both the nuclear and plastid topologies, connate lateral united petals evolved only once from free lateral united petals, and represent a synapomorphy for the CLUP clade. Reconstruction of the evolution of ovary locule number differs between nuclear and plastid topologies. According to the nuclear topology, 47% of the Bayesian trees sampled suggest a transition from a five-locular ovary in Hydrocera, to a four-locular ovary in Impatiens once and a single reversal (Fig. 4). For the plastid topology, a five-locular ovary is ancestral, with two transitions to a four-locular ovary (Fig. 4).

Connate outer lateral sepals are not a synapomorphy in either the nuclear or plastid topology respectively. Although connate outer lateral sepals are confined to a small group of species of the CLUP clade, shifts between connate and free lateral united petals have occurred six times in both topologies (Fig. 5).

Discussion

Both nuclear and plastid topologies show that I. subg. Semeiocardium is not monophyletic. Thirteen of the 15 included species are part of the CLUP clade together with 30 other representatives from Southeast Asia (Fig. 2). This clade is characterized by connate lateral united petals as a synapomorphy (Fig. 3). The two species of I. subg. Semeiocardium which are not part of this clade also do not possess connate lateral united petals.

Besides connate lateral united petals, species in the CLUP clade also possess a four-locular ovary. Indeed, this character association had already been suggested by Shimizu and Takao (1982) and Shimizu (1987). However, Yu et al. (2010) reported that some species with a four-locular ovary lack connation of the lateral united petals. These species belong to two lineages: the lineage towards I. omeiana, and the lineage(s) towards I. damrongii, I. aff. augustini, and I. stenosepala. The evolutionary pathways of the lateral united petals and the ovary in Impatiens preclude their joint use as diagnostic characters for infrageneric taxa. However, if connate lateral united petals are considered alone, they define a well-supported clade.

Impatiens santisukii and I. muscicola were included in I. subg. Semeiocardium on the basis of the typical diagnostic characters of this subgenus Utami (2009). In our phylogenetic analyses these two species do not form part of the CLUP clade, but instead group together with other pink, flat-flowered taxa distributed from tropical Africa to Southeast Asia. The phylogenetic position of I. santisukii and I. muscicola would imply independent origins of connate lateral united petals and a four-locular ovary. However, on re-examination of these characters for these two species using fresh material, we observed that their lateral united petals are free (cf. Shimizu 1970, 2000), and that they possess a five-locular ovary, similar to the other species of the clade to which they belong.

We detected several cases of well-supported incongruence between nuclear and plastid topologies (cf., Janssens et al. 2006). However, the incongruence did not affect our ability to firmly reject the monophyly of I. subg. Semeiocardium, as in both gene trees the subgenus was not monophyletic. Nevertheless, the incongruence did affect certain inferences of character evolution. Connate lateral united petals were a synapomorphy for the CLUP clade in both gene trees, and the phylogenetic trees confirmed in both cases Grey-Wilson's claim (1980a) that the evolutionary shift occurred from free to connate petals. Connate outer lateral sepals were used as a diagnostic character for the genus Semeiocardium (Bleeker 1858; Backer 1935). Grey-Wilson (1989a) suggested that this character state occurs throughout the genus Impatiens, and it was not used by Utami (2009) as diagnostic for I. subg. Semeiocardium. According to both gene trees connate outer lateral sepals are homoplasious. However, this character state does not occur as widely scattered across the genus as Grey-Wilson (1989a) assumed but it is only present in the CLUP clade, where shifts between free and connate sepals frequently occured. Further research is required to understand the functional significance of connate outer lateral sepals, for instance for interactions with specific pollinators. Inferences on the evolution of the number of ovary locules are most strongly affected by the gene tree incongruence. According to the plastid topology, a four-locular ovary evolved twice independently from a five-locular ovary. Reconstructions according to the set of Bayesian topologies based on nuclear DNA sequences are mostly ambiguous, although in a subset of topologies there is evidence for a scenario where the ancestral condition in Impatiens was a five-locular ovary, with a single shift to a four-locular ovary and a single reversal to a five-locular ovary. The main difference between reconstructions according to the nuclear and plastid topology respectively, is thus that either a four-locular ovary is not homologous (plastid), or a five-locular ovary is not homologous (nuclear). Considering that the floral morphology of I. omeina (four-locular) is very similar to that of some early diverging species of Clade 1 (four-locular), it appears more reasonable that the four-locular ovary is homologous. This would be consistent with the ITS topology. Further sampling of nuclear DNA, to reconstruct multiple independent gene trees, supplemented by a floral ontogenetic study on the four-locular species from the different clades, is required to test this hypothesis.

The recent infragereric classification of Impatiens by Yu et al. (2015) proposed a new I. sect. Semeiocardium within I. subg. Impatiens. Their study included only eight representatives of this section. We expanded the sampling from the Southeast Asian Impatiens hotspot, including many potential members of I. sect. Semeiocardium. To determine whether species are member of I. sect. Semeiocardium, we used the diagnostic key to the subgenera and sections of Impatiens provided by Yu et al. (2015). In several cases species did not key out to a particular section because they contained conflicting character states. For instance, I. opinata and I. mirabilis are characterized by a four-locular ovary and hammerhead-shaped capsule (suggestive of membership in I. sect. Clavicarpa), but also by many ovules per carpel (suggestive of I. sect. Semeiocardium). Moreover, the number of ovules per locule varies from one to many even within the same species for some Impatiens (Shimizu and Takao 1985). We therefore conclude that while the diagnostic key of Yu et al. (2015) may have been sufficient to key out a relatively limited number of species, it does not always work for other Impatiens species. This highlights the need for seeking further characters or character combinations to diagnose monophyletic clades. In our analysis the posession of connate lateral united petals was identified as a synapomorphy for a large clade to which many species belong which posess characters that are largely consistent with a subgroup of I. sect. Semeiocardium (i.e. they have a four-locular ovary and obconic capsules). As opposed to fruit, seed, or pollen characters, this floral morphological character can be readily identified from fresh material and herbarium material and therefore serves as an excellent diagnostic character. Based on the possession of connate lateral united petals, an additional 34 unsampled species are expexted to be part of the CLUP clade. The majority of these additional taxa are also distributed in Southeast Asia (Table 2; Fig. 1). However, connate lateral united petals cannot be used to delimit I. sect. Semeiocardium, as it excludes I. stenosepala and its allies. Further research is required to identify characters that unambiguously identify the clade to which I. stenosepala belongs, before new section delimitations can be applied.

Fig. 4.

Parsimony ancestral character state reconstructions of the number of ovary locules based on 15,000 post burn-in trees from the Bayesian inference plotted on the Bayesian consensus tree from the ITS (left) and atpB-rbcL (right) data. Taxa in green possess connate outer lateral sepals, whereas those in black (including all taxa that are part of clade 2 and 3) possess a five-loecular ovary. Circles at nodes represent pie charts indicating the percentage of Bayesian post burn-in trees with a particular reconstruction, equivocal reconstruction, or the node absent.

Fig. 5.

Parsimony ancestral character state reconstructions of outer lateral sepals based on 15,000 post burn-in trees from the Bayesian inference plotted on the Bayesian consensus tree from the ITS (left) and atpB-rbcL (right) data. Taxa in green possess connate lateral united petals, whereas those in black (including all taxa that are part of clade 2 and 3) possess free outer lateral sepals. Circles at nodes represent pie charts indicating the percentage of Bayesian post burn-in trees with a particular reconstruction, equivocal reconstruction, or the node absent.

Table 2.

List of Impatiens species with connate lateral united petals. Species marked with * were included in molecular analysis. Species marked with ^U were included in I. subg. Semeiocardium by Utami (2009). The distribution is indicated by country (PEN = Peninsular part of Thailand; C = Central)

The combination of molecular phylogenetics and morphological character state reconstruction has facilitated the delimitation of natural infrageneric lineages within the complex genus Impatiens. This study suggests that despite the fact that molecular studies demonstrated the artificial nature of previous Impatiens classifications, new monophyletic and clearly diagnosable lineages can be found via extended character state research. To improve the infrageneric classification of Impatiens, future studies should attempt to increase the taxon sampling, in combination with performing detailed morphological analyses to refine the infrageneric taxonomy of the genus Impatiens.