Phylogenetic relationships within the Gentianales based on NDHF and RBCL sequences, with particular reference to the Loganiaceae†
The authors thank Anders Backlund, Kåre Bremer, Pablo Goloboff, and John Parnell for advice and valuable comments on the manuscript; the two reviewers Mark Chase and Joachim Kadereit for helpful suggestions; David L. Swofford and James S. Farris for permission to use unpublished versions of their computer programs PAUP* 4.0d64 and xac 1.2, respectively; Victor A. Albert, Lennart Andersson, Jacquelyn Kallunki, Jette Knudsen, Johan Rova, Bengt Sennblad, Jim Solomon, Lena Struwe, Bertil Ståhl, Gerhard Thijsse, Barry Unwin, and the curators at GB, MO, S, and UPS for their assistance with obtaining plant material or DNA; and Nahid Heidari and Jessica Rönnholm for assistance with the laboratory work. This study was supported by a grant from the Swedish Natural Science Research Council to B. B.
Abstract
Phylogenetic relationships in the Gentianales with focus on Loganiaceae sensu lato are evaluated using parsimony analyses of nucleotide sequence data from the plastid genes rbcL and ndhF. Inter- and intrafamilial relationships in the Gentianales, which consist of the families Apocynaceae (including Asclepiadaceae), Gelsemiaceae, Gentianaceae, Loganiaceae, and Rubiaceae, are studied and receive increased support from the combination of rbcL and ndhF data, which indicate that the family Rubiaceae forms the sister group to the successively nested Gentianaceae, Apocynaceae, and Loganiaceae, all of which are well supported. The family Gelsemiaceae forms a distinct, supported group sister to Apocynaceae. The Loganiaceae sensu stricto form a strongly supported group consisting of 13 genera: Antonia, Bonyunia, Gardneria, Geniostoma, Labordia, Logania, Mitrasacme, Mitreola, Neuburgia, Norrisia, Spigelia, Strychnos, and Usteria. These genera form two well-supported lineages. Several members of Loganiaceae sensu Leeuwenberg and Leenhouts, i.e., Androya, Peltanthera, Plocosperma, Polypremum, and Sanango are clearly not members of the Gentianales. The earlier exclusion of Buddlejaceae (including Buddleja, Emorya, Gomphostigma, and Nicodemia) as well as the reclassification of the genera Nuxia and Retzia to Stilbaceae of the Lamiales are all well supported.
Members of the Gentianales (APG, 1998) share several vegetative, floral, and phytochemical traits and range from small alpine herbs to large rain forest trees, including many ornamentals and economically important plants (e.g., Catharanthus, Cinchona, Coffea, and Strychnos). In general agreement with most recent classifications, Gentianales consist of the families Apocynaceae, Asclepiadaceae, Gentianaceae, Loganiaceae, and Rubiaceae (e.g., Wagenitz, 1959). A majority of the plants in the order are woody with opposite, entire leaves, often with stipules and colleters. The latter are a special type of multicellular glandular hair, located on the stipules, at the base of the leaves, or inside the calyx. Generally the flowers are regular and pentamerous, and endosperm formation is nuclear (Schumann, 1891, 1895; Gilg, 1895; Hakki, 1980); this in contrast to other euasterid II orders, e.g., Solanales and Lamiales (APG, 1998). The most significant feature of their wood anatomy is internal phloem, the presence of which led to the inclusion of the monotypic family Saccifoliaceae, with its only member Saccifolium bandeirae from the Guyana Highlands, in the order (Cronquist, 1981). Since then this species has been considered as part of or closely allied to Gentianaceae (Metcalfe and Chalk, 1983; Takhtajan, 1997), an opinion mainly based on general morphological similarities in wood anatomy and flower morphology (Maguire and Pires, 1978). The absence of internal phloem in the Rubiaceae has occasionally been used as an argument to exclude the latter from the Gentianales (Cronquist, 1981). Furthermore, several chemical constituents, of which indole alkaloids are the most well known, unite Loganiaceae (excluding Retziaceae and Buddlejaceae) with Rubiaceae and Apocynaceae according to Kisakürek and Hesse (1980).
The circumscription of Gentianales has long been debated (de Candolle, 1824–1873; Bartling, 1830; Lindley, 1833; Meisner, 1836–1843; Endlicher, 1841; Bentham and Hooker, 1862–1883; Baillon, 1888, 1889; Engler, 1898; Solereder, 1899; Wilhelm, 1910; Hallier, 1912; Bessey, 1915; Wettstein, 1924; Rendle, 1952; Tournay and Lawalrée, 1952; Hutchinson, 1959, 1973; Wagenitz, 1959, 1964; Cronquist, 1968, 1981, 1983, 1988; Soó, 1975; Thorne, 1976, 1983, 1992a, b; Benson, 1979; Dahlgren, 1980a, b, 1983; Dahlgren, Jensen, and Nielsen, 1981; Takhtajan, 1987, 1997; Dahlgren, 1992; Nicholas and Baijnath, 1994; Struwe and Albert in Struwe, Albert, and Bremer, 1994). Taxa with contorted flower aestivation were grouped by Bartling (1830) in Contortae, later renamed as Gentianales by Lindley (1833). Recognized in Genera Plantarum (Bentham and Hooker, 1862–1883), the order consisted of the six families Apocynaceae, Asclepiadaceae, Gentianaceae, Loganiaceae, Oleaceae, and Salvadoraceae. Nearly 100 years later Wagenitz (1959) was the first author to include Rubiaceae, earlier associated with Caprifoliaceae (Jussieu, 1789; Baillon, 1880) in the order. Debate continued, and at the “angiosperm meeting” in Sydney in 1983, several new different circumscriptions of the Gentianales were proposed, some including Rubiaceae (Dahlgren, 1983; Thorne, 1983) and some not (Cronquist, 1983). In later phylogenetic studies, Gentianales were conceived as a monophyletic group including Apocynaceae, Asclepiadaceae, Gentianaceae, Rubiaceae, and parts of Loganiaceae (Bremer and Struwe, 1992; Downie and Palmer, 1992; Olmstead et al., 1993). Asclepiadaceae have been shown to belong to Apocynaceae according to several recent studies (Endress et al., 1996; Sennblad and Bremer, 1996; Sennblad, 1997). In contrast to the order, Loganiaceae (sensu Leeuwenberg and Leenhouts, 1980) in all these studies was suggested to be polyphyletic, eventually leading to the erection of two new families by Struwe and Albert (in Struwe, Albert, and Bremer, 1994), Gelsemiaceae and Geniostomataceae. In his latest comprehensive classification Takhtajan (1997) recognized nine families in Gentianales: Antoniaceae, Gelsemiaceae, Geniostomataceae, Gentianaceae, Loganiaceae, Plocospermataceae, Saccifoliaceae, Spigeliaceae, and Strychnaceae.
Leenhouts (1962) and Leeuwenberg and Leenhouts (1980) emphasized Loganiaceae as a core family in Gentianales and believed it to constitute a link between the other families (Leeuwenberg and Leenhouts, 1980, p. 19). A number of studies (Bremer and Struwe, 1992; Downie and Palmer, 1992; Chase et al., 1993; Olmstead et al., 1993; Struwe, Albert, and Bremer, 1994) have indicated polyphyly of Loganiaceae sensu Leeuwenberg and Leenhouts. Several taxa have been demonstrated not to be part of Gentianales, but instead related to other groups, e.g., Retzia (Bremer et al., 1994) and Buddlejaceae (Oxelman, Backlund, and Bremer, 1999) to Lamiales and Desfontainia to Dipsacales (Bremer et al., 1994; Backlund and Bremer, 1997). No recent previous study, however, has presented a complete analysis of all suggested elements of Loganiaceae. An overview of some previous classifications of Loganiaceae is presented in Table 1.
As sequencing targets for this study the two plastid genes rbcL and ndhF were selected. The rbcL gene codes for the large subunit of the ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) and has been widely used in plant systematic studies (Ritland and Clegg, 1987; Kim et al., 1992; Olmstead et al., 1992; Chase et al., 1993; Morgan and Soltis, 1993; Olmstead et al., 1993). The ndhF gene is (on the basis of observed homologies) presumed to code for subunit 6 of NADH-dehydrogenase (Sugiura, 1992) and due to its different substitution rate has been advocated as an alternative to rbcL (Kim and Jansen, 1995).
The combined sequence data from rbcL and ndhF were analyzed using parsimony methods. This approach of combining data sets has provided important contributions in studies of several other plant groups (Olmstead and Sweere, 1994; Olmstead and Reeves, 1995; Scotland et al., 1995; Chase and Cox, 1998; Soltis et al., 1998; Savolainen et al., 2000).
The aims of this study were threefold: (1) to evaluate the monophyly of Gentianales, (2) to study the inter- and intrafamilial relationships of the order, and (3) to determine in further detail the relationships and systematic positions of the members of Loganiaceae sensu Leeuwenberg and Leenhouts (1980).
MATERIAL AND METHODS
Taxon sampling
Attempts were made to obtain a representative sampling of Gentianales. Additionally, several taxa from other closely related orders (Olmstead et al., 1992, 1993; Chase et al., 1993; Bremer et al., 1994; Gustafsson, Backlund, and Bremer, 1996; Backlund and Bremer, 1997; Oxelman, Backlund, and Bremer, 1999) were included. This was done to permit an extra-ordinal evaluation of the systematic positions of the former “loganiaceous” taxa (Leeuwenberg and Leenhouts, 1980). Taxon sampling in Loganiaceae included all genera recognized by Leeuwenberg and Leenhouts (1980), with the exception of Norrisia (due to lack of material). The tribe Antonieae, in which Norrisia was placed by Leeuwenberg and Leenhouts, is, however, represented by all the other genera (Antonia, Bonyunia, and Usteria).
A matrix consisting of rbcL and ndhF sequences of 62 taxa was compiled, in total 124 sequences. Of these, eight rbcL and 22 ndhF sequences were previously unpublished. All sequences analyzed are listed in Table 2 with their familial affinities according to Takhtajan (1997), EMBL accession numbers, and voucher information or references to their original publication.
Methods
Total DNA was extracted from fresh or silica-gel dried leaves (Chase and Hills, 1991) according to the methods of Doyle and Doyle (1987) or from herbarium material (Oxelman, Backlund, and Bremer, 1999). Double-stranded DNA of the two genes were amplified by the polymerase chain reaction (PCR) using two primers for the rbcL gene (Olmstead et al., 1992) and four for the ndhF gene (Kim and Jansen, 1995). The positions and sequences of the primers are listed in Table 3. For the manually sequenced DNA, a second run with asymmetric amplification was performed to obtain single-stranded DNA (Kaltenboeck et al., 1992). Amplification products were purified with the QIAquick™ Gel Extraction Kit according to the manufacturer's instructions (QIAGEN®). DNA was sequenced using internal primers designed by G. Zurawski at the DNAX Research Institute (for rbcL) and K.-J. Kim, R. Jansen, and B. Oxelman (for ndhF, Table 3, including a schematical drawing of the primer positions and directions). Sequencing reactions were performed according to the method originally devised by Sanger, Nicklen, and Coulson (1977). Sequences were produced either manually with S35 radioactively labelled dideoxy nucleotides resulting in autoradiograms or automated with the PRISM™ Ready Reaction Dye Deoxy Terminator FS kit (Applied Biosystems) and analyzed on a ABI PRISM 377 automated sequencer (PE corporation).
Sequence alignment and matrix compilation
To investigate the systematic position of the taxa studied, the new sequences were analyzed together with other relevant sequences already published. These were obtained from the National Center for Biotechnology Information (NCBI, “GenBank”) database or from the European Bioinformatic Institute (EBI, “EMBL Nucleotide Sequence Database”). The sequences were manually aligned to the reading frame of the corresponding genes in the complete Nicotiana plastid genome sequence (Shinozaki et al., 1986; GenBank Z00044). Positions 114168–112096 (reverse complement) were used for the ndhF and 57612–59020 for the rbcL gene. The rbcL sequences have no nucleotide insertions or deletions, whereas the ndhF sequences exhibit variation in length. The complete aligned matrices can be obtained from http://www.botany.org/bsa/ajbsupp/v86/s01-01.html, or upon request.
Phylogenetic analysis
The combined rbcL and ndhF matrix was analyzed using parsimony methods with PAUP* 4.0d64 (D. Swofford, personal communication) and xac 1.2 (S. Farris, personal communication) both under the assumptions of Fitch parsimony (Fitch, 1971).
The analyses were made on a Power Macintosh 8600/250 using 85 Mb of RAM. As a starting point 100 random stepwise additions were used for a thorough branch swapping. This was made by the tree bisection-reconnection algorithm of PAUP (Swofford, 1993). To estimate stability of the obtained result, Bremer support values (b) were determined (Bremer, 1988, 1994; Källersjö et al., 1992) using the generalized reversed constraint approach (Eernisse and Kluge, 1993). A batch processing file for calculating the Bremer support values was constructed using the computer program AutoDecay 2.9.5 (Eriksson, 1995). The constrained searches for each of the internal nodes were performed in the same way as the initial analysis but with ten random stepwise additions. PAUP* was also used to perform a bootstrap analysis (Felsenstein, 1985). Here 100 replicate matrices were produced and each of these was analyzed using five random stepwise additions followed by branch swapping as described above. Finally parsimony-jackknife analyses (Farris et al., 1996) were made both with PAUP* and with xac. Equal probability deletion of 37% of the characters were made after recommendation by Farris et al. (1996) and the resulting 100 (PAUP*) or 1000 (xac) replicate matrices were analyzed analogously to the bootstrap analysis.
Furthermore, parsimony and jackknife analyses were performed in a similar manner as described above for each of the two genes separately.
Character weighting
A series of arguments have been brought forward both promoting and criticizing posteriori character weighting (Farris, 1983; Goloboff, 1993, 1995; Turner and Zandee, 1995; Allard and Carpenter, 1996; Nixon and Carpenter, 1996). In this study we decided to use successive approximation weighting (Farris, 1969). With this method characters are weighted according to their behavior on a (or a set of) tree(s). Different indices such as the consistency index (CI; Farris, 1969), retention index (RI; Farris, 1989), or the rescaled consistency index (RC; Farris, 1989) can be used as basis for the weighting. According to recommendations by Farris (1989) and Swofford (defaults in PAUP) we have used RC. The analysis using Farris' successive approximations weighting procedure were performed with analogous settings and options as described for the unit-weighted analysis.
RESULTS
Separate vs. combined analyses
Results from the separate as well as combined analyses of nucleotide sequence data from the genes rbcL and ndhF are largely congruent. Apart from minor differences in relationships of taxa outside Gentianales, none of the well-supported groups were in conflict. Hence, only results from more thoroughly analyzed combined matrix will be discussed.
The unit-weighted analysis
The initial analysis using Fitch parsimony yielded 12 equally parsimonious trees with a length of 5941 steps, a consistency index (CI; Kluge and Farris, 1969) of 0.36 and a retention index (RI; Farris, 1989) of 0.63. The strict consensus tree is shown in Fig. 1.
The successively weighted analysis
The analysis using the successive approximation weighting (Farris, 1969) produced one most parsimonious tree. The topology of this was identical to one of the 12 equally parsimonious trees obtained from the unit-weighted analysis and is selected for further discussions (Fig. 2). Branch lengths and support indices are given in Table 4.
Tree topology and implied relationships
The trees were oriented using an outgroup consisting of five taxa belonging to the euasterid II clade (Chase et al., 1993; Bremer et al., 1994; Backlund and Bremer, 1997; Oxelman, Backlund, and Bremer, 1999; APG, 1998).
Euasterid I
The ingroup corresponds to the euasterid I (Chase et al., 1993; APG, 1998) and is divided into two major lineages, Gentianales and Lamiales. The sole representative for the order Solanales, Nicotiana, is found sister to this major dichotomy. The support for euasterid I is high (branch length [bl] = 83, Bremer support [b] = 32, jackknife value according to PAUP* [j] and xac [x] as well as bootstrap value [bt] all = 100%). This indicates that none of the members of Loganiaceae, sensu Leeuwenberg and Leenhouts, has a relationship to euasterid II with the exception of Desfontainia, which has previously been placed as a member of Dipsacales (Bremer et al., 1994; Gustafsson, Backlund, and Bremer, 1996; Backlund and Bremer, 1997).
Lamiales
The support for this group is high (bl = 52, b = 17, j = x = bt = 100%). In this lineage, a number of genera sometimes previously placed in the Loganiaceae are found, in agreement with the more detailed recent works (e.g., Oxelman, Backlund, and Bremer, 1999). The topology in this part of the tree differs slightly from the topologies obtained solely from rbcL data (Chase et al., 1993; Olmstead et al., 1993). The consensus tree presented by Olmstead and Reeves (1995) is, however, congruent with the results by Oxelman, Backlund, and Bremer (1999) as well as the results from this study. It is shown, both in this and in the study by Oxelman, Backlund, and Bremer (1999), that several of the genera previously included in Loganiaceae sensu Leeuwenberg and Leenhouts belong instead in Lamiales sensu Takhtajan.
Gentianales
The other major lineage within euasterid I (node 28) corresponds to Gentianales, which have high support indices (bl = 58, b = 27, j = x = bt = 100%). Within Gentianales, we find a basal dichotomy between a well-supported Rubiaceae (node 29, bl = 39, b = 18, j = x = bt = 100%) and a lineage (node 37, bl = 39, b = 6, j = 98%, x = 96%, bt = 95%) composed of the Gentianaceae, Apocynaceae, Gelsemiaceae, and a large portion of Loganiaceae. Gentianaceae (bl = 145, b = 96, j = x = bt = 100%) include the three former loganiaceous genera Anthocleista, Fagraea, and Potalia.
Loganiaceae
The core Loganiaceae is well supported (bl = 40, b = 23, j = x = bt = 100%), and this node (49) is used to define which genera are included in Table 5. At the base Loganiaceae divides into two well-supported lineages, one including Usteria, Antonia, and Bonyunia, and the other (node 52) further divided into two groups. One of these (node 53), including Gardneria, Neuburgia, Spigelia, and Strychnos, is weakly supported. The remaining group (node 56), on the other hand, is well supported and consists of Mitreola, Logania, Mitrasacme, Geniostoma, and Labordia.
DISCUSSION
Monophyly of Gentianales
We find strong support for Gentianales (node 28 in Fig. 2). The inclusion of the families Apocynaceae, Gelsemiaceae, Gentianaceae, Loganiaceae, and Rubiaceae is congruent with several previous molecular studies of the group (Downie and Palmer, 1992; Chase et al., 1993; Olmstead et al., 1993; Bremer et al., 1994), as well as morphological (Struwe, Albert, and Bremer, 1994) and phytochemical (Jensen, 1991, 1992) patterns of variability. Notable among the morphological traits is the combination of opposite and entire leaves with stipules (with some exceptions), colleters, and nuclear endosperm formation (typically cellular in the Lamiales and Solanales; cf. Johri, Ambegaokar, and Srivastava, 1992). The phytochemical distinction is most evident in the widespread and prominent occurrence of complex indole alkaloids and seco-iridoids synthesized solely along the iridoid biosynthesis route I, as compared to route II compounds, which are encountered in Solanales and Lamiales (Jensen, 1991, 1992).
Inter- and intrafamilial relationships of Gentianales
Recent phylogenetic studies indicate that within the order there are two major evolutionary lineages, one comprising the families Gentianaceae, Apocynaceae, Gelsemiaceae, and Loganiaceae sensu stricto and the other consisting entirely of the Rubiaceae, which is the sister group to the rest of the order (Bremer, 1996b). In this study, the support for both of these lineages is robust. This dichotomy is also supported by the occurrence of superior ovaries and internal phloem in the Gentianaceae-Apocynaceae-Gelsemiaceae-Loganiaceae lineage as well as by differences in biosynthesis of iridoid and indole alkaloid compounds (Jensen, 1991, 1992).
Rubiaceae
The mainly tropical Rubiaceae, with 10 200 species (Mabberley, 1997), is usually considered as a monophyletic group, but the systematic position of the Rubiaceae has been discussed (Verdcourt, 1958; Bremekamp, 1966; Robbrecht, 1988; Bremer, 1996b). Lack of internal phloem (otherwise ubiquitous in the order) and presence of an inferior ovary (otherwise superior in Gentianales with few exceptions), obturator, Casparian thickenings, and early sympetaly (Erbar, 1991) distinguish the family from the rest of the order. Bentham and Hooker (1862–1883) separated Rubiaceae due to its inferior ovary into a monofamilial Rubiales. There it was retained by Cronquist (1981, 1988) together with the small monogeneric family Theligonaceae (usually included in Rubiaceae; Wunderlich, 1971; Robbrecht, 1988, 1993; Bremer, Andreasen, and Olsson, 1995; Natali, Manen, and Ehrendorfer, 1995), as well as by Takhtajan (1997), who also appended the families Dialypetalanthaceae and Carlemanniaceae. Utzschneider (1947, 1951), on the other hand, was the first to indicate that Rubiaceae are part of the Gentianales. This was later supported by Wagenitz (1959, 1964), but the proposal did not gain wide acceptance until the 1980s after which it was included in Gentianales in most systematic schemes (Dahlgren, 1980a, b, 1983; Thorne, 1983, 1992a, b; Takhtajan, 1987). Many recent works based on both molecular and morphological data indicate that Rubiaceae are part of Gentianales (Bisset, 1980; Bremer and Struwe, 1992; Struwe, Albert, and Bremer, 1994; Bremer, 1996a, b). The topology within Rubiaceae retrieved from this analysis is largely congruent with previous studies, which have more thorough sampling (Bremer, Andreasen, and Olsson, 1995; Bremer, 1996b) and so will not be discussed further here.
Gentianaceae
The cosmopolitan family Gentianaceae mostly consist of herbs, rarely shrubs or trees and comprise 1200 species (Mabberley, 1997). The inclusion of the woody genera Potalia, Fagraea, and Anthocleista (former tribe Potalieae of Loganiaceae, sometimes recognized as the family Potaliaceae; Watson and Dallwitz, 1992) in Gentianaceae, has long been argued (Bureau, 1856; Fosberg and Sachet, 1980; Bremer and Struwe, 1992; Struwe, Albert, and Bremer, 1994; Mészáros, Laet, and Smets, 1996). Morphological features that have been interpreted as synapomorphies and support of this placement are the monadelphous filaments (not seen in Fagraea), dextral-contorted buds, septate parenchyma, and several phytochemical features such as presence of gentianine (Bisset, 1980). These conclusions are also supported from molecular data (Downie and Palmer, 1992; Olmstead et al., 1993) and appear here to be well supported. The phylogenetically derived position of the woody genera Potalia, Fagraea, and Anthocleista, compared to the mainly herbaceous genera Exacum and Gentiana, is also well supported.
The family Menyanthaceae, sometimes included as the tribe Menyantheae in the family Gentianaceae (Bentham and Hooker, 1862–1883) or considered as a close relative (Leeuwenberg and Leenhouts, 1980) have been used as one of the outgroups here. The placement of the Menyanthaceae in Asterales has been indicated by rbcL analyses (Downie and Palmer, 1992; Chase et al., 1993; Olmstead et al., 1993; Gustafsson, Backlund, and Bremer, 1996; Backlund and Bremer, 1997) and corroborated by studies of morphological data (Gustafsson and Bremer, 1995).
Apocynaceae
Most members in the Apocynaceae are tropical or subtropical, and the family consists of 4800 species (including Asclepiadaceae; Mabberley, 1997). Many systematists have argued that Asclepiadoideae, sensu Jussieu, should be recognized as the family Asclepiadaceae (Brown, 1810; Schumann, 1895; Cronquist, 1981; Takhtajan, 1987; Rosatti, 1989a, b; Nicholas and Baijnath, 1994), but it has been shown that this group of taxa is clearly nested within Apocynaceae, as one of its morphologically most derived groups. This placement is supported by several recent studies, which have included data from morphology, palynology, phytochemistry, and DNA sequences (Sennblad and Bremer, 1996; Sennblad, 1997). In this combined analysis of rbcL and ndhF data, we find a strongly supported Apocynaceae (bl = 44, b = 27, j = x = bt = 100%) with a well-supported internal structure congruent with the results from the studies mentioned above.
Gelsemiaceae
The two genera Gelsemium and Mostuea, with three and eight species, respectively, are tropical and subtropical shrubs and lianas (Mabberley, 1997). They have long been regarded as closely related (Bentham and Hooker, 1862–1883; Solereder, 1892–1895; Leeuwenberg and Leenhouts, 1980). The placement of this group has varied, in Apocynaceae (Persoon, 1805), as a separate tribe within Apocynaceae (Don, 1837–1838; Endlicher, 1841; Bureau, 1856), in Antonieae of Loganiaceae (Bentham, 1856), or later in a tribe of its own as Gelsemieae in Loganiaceae (Bentham and Hooker, 1862–1883). As such they remained until the classification of Struwe and Albert (in Struwe, Albert, and Bremer, 1994), when the former tribe Gelsemieae was elevated to familial rank.
The conflicts among systematists regarding the affinities of Gelsemiaceae are due to several morphological features that have appeared to be contradictory and indicative of different systematic positions. Features indicating a close relationship to Apocynaceae include anther anatomy (number of tapetum layers; Johri, Ambegaokar, and Srivastava, 1992), phytochemistry (a special form of C-17 indole alkaloids; Jensen, Nielsen, and Dahlgren, 1975; Jensen, 1992) and cytotaxonomy (Moore, 1947).
Other features appear instead to unite Gelsemiaceae with Loganiaceae (especially the tribe Antonieae). Among these characters we find imbricate corollas (always contorted in Apocynaceae; Leeuwenberg and Leenhouts, 1980), late sympetaly (instead of an “intermediate” stage in Apocynaceae; Erbar, 1991), absence of lacticifers (ubiquitous in the Apocynaceae; Sennblad and Bremer, 1996), features of seed anatomy (horny endosperm; Leeuwenberg and Leenhouts, 1980), and thin thyloses and wide multiserate rays (giving a wood anatomy particularly resembling Antonieae; Mennega, 1980) as well as phytochemistry (presence of scopoletin, kaempferol, and quercetin; Bisset, 1980). Furthermore, a large number of easily detectable autapomorphic features such as heterostylous flowers, twice-dichotomously divided stigmas, and latrorse anthers tend to set Gelsemiaceae apart from both groups of potential closest relatives.
This analysis indicates a sister-group relationship to Apocynaceae. This relationship is not strongly supported, but there is strong support for Gelsemiaceae as a monophyletic group. Because Gelsemiaceae already exist (retaining stability), form an easily distinguishable and well-supported lineage (ease of identification), and constitute more than one genus, this avoids redundancy. Albeit with weakly supported sister-group relationship (phylogenetic information) we find that all guidelines suggested by Backlund and Bremer (1998) are fulfilled by retaining the family as it presently stands.
Loganiaceae
In the circumscription of Leeuwenberg and Leenhouts (1980) Loganiaceae consisted of 600 species in 30 genera (Mabberley, 1997) and included predominantly tropical, woody plants. Since the description by Martius (1827), both the circumscription of the family and the intrafamilial relationships have been a matter of debate. Some authors accept one large (600 species) family (Leeuwenberg and Leenhouts, 1980), whereas others prefer Loganiaceae split into 12 different families allocated to several distantly related orders (Takhtajan, 1997).
Bentham and Hooker (1862–1883) divided the family into seven tribes (Antonieae, Buddlejeae, Desfontainieae, Euloganieae including Strychnos and relatives, Gelsemieae, Potalieae, and Spigelieae). Solereder (1882–1895) recognized two subfamilies (Loganioideae with six tribes, and Buddlejoideae with one tribe) within the family, but he also excluded the two genera Plocosperma (of the Gelsemieae sensu Bentham and Hooker) and Desfontainia (of the Desfontainieae sensu Bentham and Hooker). Almost 100 years later Hutchinson (1973) further divided the Loganiaceae sensu Bentham and Hooker into seven distinct families and established a new order (Loganiales) for these families together with the family Oleaceae. The morphologically enigmatic genus Plocosperma was simultaneously placed as a monotypic family in the new order Apocynales. Nevertheless, Gentianales with a variously circumscribed Loganiaceae as part of its core was retained by most systematists. Among these, Loganiaceae was considered to occupy a central evolutionarily position (Bisset, 1980; Leeuwenberg and Leenhouts, 1980).
According to many recent studies, Loganiaceae sensu Leeuwenberg and Leenhouts are polyphyletic (Downie and Palmer, 1992; Olmstead et al., 1993; Struwe, Albert, and Bremer, 1994; Takhtajan, 1997; Oxelman, Backlund, and Bremer, 1999). This heterogeneity is also reflected in the classification by Leeuwenberg and Leenhouts (1980) in which the Loganiaceae were divided into ten tribes, some of which consisted of one or only a few species.
In this study, we find a strongly supported group corresponding to the tribe Antonieae (node 50) sensu Leeuwenberg and Leenhouts. These genera share several characters such as valvate aestivation, coriaceous leaves, ability to accumulate aluminium (Leeuwenberg and Leenhouts, 1980), and a homogeneous wood anatomy featuring absence of continuous rays, interxylary phloem of foraminate type, large cavities in rays, and vessels in tangential pairs or small clusters (Mennega, 1980). Also the internal node (51) in this group is strongly supported, and the taxa (Antonia and Bonyunia) share a venation pattern, as well as the same geographical distribution (South America).
The branch supporting the remaining taxa (52) is also well supported. This relationship, however, has not been encountered among previous studies using morphological data (Bremer and Struwe, 1992; Struwe, Albert, and Bremer, 1994). As would be expected, few or no obvious morphological features are easily identified as diagnostic of this large group. This group further divides into two lineages. The first (53) comprises the two heterogeneous genera Spigelia and Strychnos together with Gardneria and Neuburgia. This group, apart from Spigelia, corresponds to the tribe Strychneae of Leeuwenberg and Leenhouts (1980). Neither this lineage, nor its internal structure, is particularly well supported by the molecular data. Additional features that could be interpreted as supporting this hypothesis are valvate aestivation (also found in Mitrasacme), some indole alkaloid derivatives, aluminium accumulation (also in the Antonia group), and loganine-type iridoids. The former Strychneae were united on the basis of their anther appendages and indehiscent fruits, features not reported for Spigelia by Leeuwenberg and Leenhouts (1980). The other lineage (node 56), by contrast, is well supported both on morphological and molecular grounds. Characters include partly apocarpous carpels (or possibly homologous, early from apex-splitting fruits), ochrea instead of stipules (Leeuwenberg and Leenhouts, 1980), a general change from the presumed plesiomorphic basal chromosome number of x = 11 to x = 10 (Gadella, 1980), and a general absence of alkaloids (Bisset, 1980). Here we encounter the type genus Logania, together with Mitrasacme and Mitreola (two members of Spigelieae sensu Leeuwenberg and Leenhouts) as well as Geniostoma and Labordia (Geniostomataceae sensu Struwe and Albert in Struwe, Albert, and Bremer, 1994).
We conclude that neither Geniostomataceae nor Strychnaceae (sensu Struwe and Albert in Struwe, Albert, and Bremer, 1994) can be retained. Geniostomataceae with very strong support form a clade with Loganiaceae (node 56, and sensu Struwe and Albert in Struwe, Albert, and Bremer, 1994). The Geniostomataceae are thus deeply nested within Loganiaceae with high support. Spigeliaceae (sensu Struwe and Albert in Struwe, Albert, and Breme, 1994) are by the combined rbcL and ndhF data indicated to consist of at least two lineages. One of these corresponds, as discussed above, to the former Antonieae of Leeuwenberg and Leenhouts and is strongly supported in our study. The remaining taxa of Strychnaceae (sensu Struwe and Albert in Struwe, Albert, and Bremer, 1994) correspond to the former Strychneae of Leeuwenberg and Leenhouts with the addition of Spigelia. This lineage is also present in our results and with high support forms a monophyletic group with the Loganiaceae–Geniostomataceae clade.
In summary, Loganiaceae s.s. form a strongly supported monophyletic group consisting of 13 genera: Antonia, Bonyunia, Gardneria, Geniostoma, Labordia, Logania, Mitrasacme, Mitreola, Neuburgia, Norrisia, Spigelia, Strychnos, and Usteria.
Systematic positions of genera formerly included in Loganiaceae
The majority of loganiaceous species belong in Gentianales in the three closely related families Gelsemiaceae, Gentianaceae, and Loganiaceae s.s. However, a number of genera previously regarded as members of the Loganiaceae s.l. obviously belong in Lamiales. The positions of these genera here are congruent with the results of Olmstead and Reeves (1995) and Oxelman, Backlund, and Bremer (1999).
Buddlejaceae
Buddleja was first placed in the Scrophulariaceae by Jussieu (1789) but later moved to Loganiaceae by Bentham (1856) and recently regarded as a tribe within Loganiaceae by Leeuwenberg and Leenhouts (1980). Buddlejaceae were separated by Dahlgren (1975) from the rest of Gentianales and placed in Lamiales because of their seco-iridoids (biosynthesis route II; Jensen, 1991). This issue has attracted attention, and several features from wood anatomy, including lack of borders on pits of imperforate tracheary elements, and embryological features such as cellular endosperm, are characteristic of Lamiales (Bendre, 1973, 1975; Mennega, 1980; Carlquist, 1986, 1992; Maldonado de Magnano, 1986; Engell, 1987; Hegnauer, 1989; Dahlgren, 1992). For a further treatment of Buddleja and Buddlejaceae see Oxelman, Backlund, and Bremer (1999).
Retzia and Nuxia
Several recent studies indicate that South African Retzia of the monotypic tribe Retzieae, sometimes recognized as the distinct family Retziaceae (Bartling, 1830), differs in many characters from Loganiaceae (Dahlgren et al., 1979; Bremer et al., 1994; Struwe, Albert, and Bremer, 1994). Retzia, on phytochemical and anatomical grounds, had been associated with the genus Stilbe (Dahlgren et al., 1979; Dahlgren, 1980a; Carlquist, 1986), originally placed in the Verbenaceae.
The genus Nuxia, by Leeuwenberg and Leenhouts (1980) included in the tribe Buddlejeae, has been here and in a previous molecular study (Oxelman, Backlund, and Bremer, 1999) strongly supported as a close relative of Retzia and Stilbe. These genera are now considered to form the family Stilbaceae of Lamiales (Thorne, 1992a, b; Bremer et al., 1994; Oxelman, Backlund, and Bremer, 1999).
Plocosperma, Polypremum, Peltanthera, and Sanango
Apart from the taxa discussed above, these four genera should also be excluded from Loganiaceae s.s. as well as from Gentianales (Jensen, 1992; Oxelman, Backlund, and Bremer, 1999). Plocosperma is placed next to Borago, and the other three genera are with high support placed in Lamiales in our results.
The monotypic Central American genus Plocosperma was earlier the sole member of the tribe Plocospermeae of Loganiaceae (Leeuwenberg and Leenhouts, 1980). In some classifications, Plocosperma has been suggested to show a close relationship to Apocynaceae (Hutchinson, 1973; Cronquist, 1981; Takhtajan, 1987). Both here and in other recent phylogenetic studies, both these placements have been refuted (Struwe, Albert, and Bremer, 1994; Endress and Albert, 1995; Oxelman, Backlund, and Bremer, 1999). According to this study, as well as in Oxelman, Backlund, and Bremer (1999), Plocosperma is supported as belonging to euasterid I. The sampling in this part of the tree is sparse, and the support indices for this placement are low. The position of Plocosperma therefore may still be regarded as uncertain.
The likewise monotypic American genus Polypremum, former member of the tribe Spigelieae of Loganiaceae, is in this analysis indicated to occupy a relatively early branch within Lamiales. This placement is completely congruent with the strongly supported placement together with the genus Tetrachondra in Oxelman, Backlund, and Bremer (1999).
Peltanthera and Sanango, two other monotypic American genera, were earlier placed in the tribe Buddlejeae of the Loganiaceae (Leeuwenberg and Leenhouts, 1980). Our study indicates a well supported and close relationship between both Streptocarpus and Sanango to Gesneriaceae, as well as between these and Peltanthera.
Desfontainieae with the single genus Desfontainia have previously been placed in the order Dipsacales (Bremer et al., 1994; Backlund and Bremer, 1997) and more specifically in the family Columelliaceae (Backlund, 1996).
Conclusions
Combined sequence data from the plastid genes rbcL and ndhF resolve (with minor exceptions) inter- and intrafamilial relationships within the Gentianales, in most cases with high levels of internal support. Gentianales, as here defined, consist of the families Apocynaceae, Gelsemiaceae, Gentianaceae, Loganiaceae, and Rubiaceae. As previously indicated in several studies (Sennblad and Bremer, 1996; Sennblad, 1997) Asclepiadaceae must be reduced to Apocynaceae. The recently proposed family Geniostomataceae cannot according to the monophyly criterion be retained without describing several additional small families. It together with Loganiaceae (sensu Struwe and Albert in Struwe, Albert, and Bremer, 1994) forms a well-supported group for which the name Loganiaceae has priority (Reveal, 1993). Nor can Strychnaceae (sensu Struwe and Albert, in Struwe, Albert, and Bremer, 1994) be retained as they become paraphyletic due to the successive nesting of Loganiaceae and Geniostomataceae within them. The other, in our opinion less appealing, possibility would be to recognize several additional small families. Of the 30 genera included in Loganiaceae by Leeuwenberg and Leenhouts (1980), 17 are, according to these results, excluded from the remaining Loganiaceae s.s. Twelve of these genera are indicated to belong to other orders, in one case even to another subclass than Loganiaceae sensu stricto and with affinities to several different large families (e.g., Boraginaceae, Gesneriaceae, Scrophulariaceae s.l.). The remaining five genera excluded from Loganiaceae are placed in Gentianaceae (Anthocleista, Fagraea, and Potalia) and Gelsemiaceae (Gelsemium and Mostuea). Loganiaceae therefore consists of 13 genera: Antonia, Bonyunia, Gardneria, Geniostoma, Labordia, Logania, Mitrasacme, Mitreola, Neuburgia, Norrisia, Spigelia, Strychnos, and Usteria.
The formerly recognized tribes Spigelieae and Strychneae as delimitated by Leeuwenberg and Leenhouts (1980) are all invalidated by the monophyly criterion applied to our results. Retzieae, Desfontainieae, and Plocospermeae as suggested by Leeuwenberg and Leenhouts (1980) are monotypic and may therefore remain as tribes, albeit not in Loganiaceae according to these, as well as earlier results (Bremer et al., 1994; Backlund and Bremer, 1997; Oxelman, Backlund, and Bremer, 1999). Groups corresponding to the tribes Antonieae, Gelsemieae, and Potalieae, sensu Leeuwenberg and Leenhouts (1980) are retrieved from molecular data, but their interrelationships are different than previously suggested. One of these groups remains as a distinct entity of Loganiaceae (Antonieae), one forms a family of its own (Gelsemiaceae), and one is nested within Gentianaceae (Potalieae). To retain the latter as a family, as suggested by Watson and Dallwitz (1992), thus becomes impossible.
The molecular data at hand, in combination with results from previous studies, provide sufficient information for a preliminary synopsis and realignments, summarized in Table 5. We refrain at this point from attempting a tribal classification of Loganiaceae in its new circumscription.
Consensus tree of 12 equally parsimonious trees resulting from the unit-weighted analysis. Familial classification follows results from this and previous studies (Olmstead and Reeves, 1995; Oxelman, Backlund, and Bremer, 1999). Taxa indicated by dots (•) have previously been assigned to Loganiaceae by Leeuwenberg and Leenhouts (1980)
One of the most parsimonious trees from the unit-weighted analysis with the topology identical to the single most parsimonious tree from the successively reweighted analysis. Familial classification according to results from this and previous studies (Olmstead and Reeves, 1995; Oxelman, Backlund, and Bremer, 1999). Nodes with jackknife and bootstrap values >90% are indicated with thick lines. Taxa indicated by dots (•) have previously been assigned to Loganiaceae by Leeuwenberg and Leenhouts (1980)
