Systematics and leaf architecture of the Gunneraceae.
I. Abstract II. Introduction A. Overview of Gunnera Taxonomy B. Generic Circumscription and Subgenera III. Materials and Methods A. Cleared Leaf Preparation B. Leaf Architectural Analysis C. Phylogenetic Analysis IV. Results A. Systematic Leaf Architecture of Gunneraceae B. The Search for Sister Groups: Comparative Leaf Architecture C. Myrothamnaceae D. Saxifragaceae E. Cladistic Analyses F. Phylogenetic Analysis within the Gunneraceae V. Discussion A. Evolutionary Trends B. Implications for Historical Biogeography C. Ecological Trends D. Implications for Macrosystematics: A Herbaceous Radiation of Eudicots? VI. Conclusions VII. Acknowledgments VIII. Literature Cited IX. Appendix: Characters and Character States A. Foliar Morphology: Form B. Venation Characters C. Marginal Teeth, Glands, and Sinuses D. Epidermal Characters E. Pollen F. Miscellaneous, Reproductive, and Anatomical Characters
A. OVERVIEW OF GUNNERA TAXONOMY
Gunnera is a dicotyledonous genus of 40-60 species, ranging in habit from small, stoloniferous herbs to fleshy-stemmed, rhizomatous herbs with enormous leaves (Schindler, 1905; Mora-Osejo, 1984; Bergman et al., 1992) (Fig. 1). Species of Gunnera are successful colonizers of disturbed sites and poor soils in subtropical or wet temperate regions in the Southern Hemisphere. This is due in part to a unique intracellular mutualism with nitrogen-fixing cyanobacteria (Silvester & Smith, 1969; Osborn et al., 1991). Research in recent years has elucidated details of the incorporation, metabolic transport, specificity, and interdependence of Gunnera and its symbionts (Towata, 1985; Zimmerman & Bergman, 1990; Bergman et al., 1992; Johansson & Bergman, 1992, 1994; Osborn et al., 1992; Stock & Silvester, 1994).
[FIGURE 1 OMITTED]
Gunnera is one of the oldest living angiosperm genera, with a fossil record of characteristic palynomorphs that appear as early as the Turonian (Late Cretaceous), ca. 93 Ma (Brenner, 1968; Jarzen & Dettmann, 1989; ages follow Geological Society of America, 1996) and a current distribution that can be related to the breakup of Gondwanaland (Mora-Osejo, 1984; Fuller, 1995a; Wanntorp & Wanntorp, 2003). The macrosystematics of this genus have been problematic (Doyle & Scogin, 1988a, 1988b; Bergman et al., 1992; Baum, 1994), although recent work has suggested that Gunnera is part of an early radiation of tricolpate (eudicot) angiosperms (Fuller 1995a, 1995b; Wilkinson, 1998; Angiosperm Phylogeny Group, 2003; Hilu et al., 2003; Soltis et al., 2003). Genetic sequence data of several Gunnera species have been subjected to cladistic analyses and have provided the first systematic assessment of phylogenetic relationships of subgenera within Gunnera (Wanntorp et al., 2001, 2002). This provides a basis for discussing trends in morphological evolution within Gunnera (Wanntorp et al., 2003, 2004) and its historical biogeography (Wanntorp & Wanntorp, 2003). The present article contributes additional evidence for phylogenetic relationships on the basis of detailed leaf morphological studies but differs in some of its conclusions from recent genetic studies.
Numerous affinities have been suggested for Gunnera (Table I). Linnaeus defined Gunnera on the basis of the African species G perpensa and associated it with genera of the Orchidaceae and with Forstera (Stylidiaceae) (Linne 1767, 1787). Other early workers placed it in the Araliaceae (Lindley, 1846) or Urticales (Jussieu, 1789; Bartling, 1830; Endlicher, 1837), an affinity suggested by Airy Shaw in Willis (1966). Meissner (1836-1843) and Endlicher (1837) placed Gunnera in its own family, a circumscription followed by several modern taxonomists (Takhtajan, 1980, 1983; Cronquist, 1981, 1988; Thorne, 1992) and this article. However, Bentham and Hooker (1865), de Candolle (1868), Engler and Prantl (Peterson, 1893), and Schindler (1905) supported assignment within the family Haloragaceae, an assignment that remains the consensus in recent taxonomic and floristic sources (e.g., Macbride, 1959; Allan, 1961; Standley & Williams, 1963; Davis, 1966; Meijden & Caspers, 1971; Hutchinson, 1973, Moore, 1983; Mora-Osejo, 1984; Osborn et al., 1991; Heywood, 1993). An important character in allying them is the distinctive pseudo-polystelic vascular system of Gunnera, which has been claimed to indicate a monostelic, aquatic ancestry, such as from the Haloragaceae (Scott, 1891; Arber, 1920; Batham, 1943; Osborn et al., 1991).
However, striking differences between Gunnera and haloragaceous taxa have been noted by numerous researchers (e.g., Praglowski, 1970; Orchard, 1975; Behnke, 1986; Wilkinson, 1998). Orchard (1975: 27), based on the floral anatomy of Haloragaceae, considered the floral vasculature of Gunnera "so reduced as to be useless for comparison with Haloragaceae." Nevertheless, Gunneraceae has remained associated with Haloragaceae in much of the taxonomic literature by placement in the same order (Takhtajan, 1969; Hutchinson, 1973; Cronquist, 1981, 1988), despite a long list of character differences between the two families. The highly autapomorphic nature of Gunnera was highlighted by its "wild-card" status in genetic studies (cf. Chase et al., 1993; Baum, 1994; Hoot et al., 1999; Soltis et al., 2000). Gunnera was elevated to its own order by some (Dahlgren, 1983; Takhtajan, 1997; Soltis et al., 2003), but, although this emphasizes its unique nature, such a placement does little to reveal its evolutionary affinities.
Several nonhaloragaceous assignments also have been proposed. Assignment to the Saxifragales (Huber, 1963; Takhtaj an, 1980, 1983; Dahlgren, 1983) received support from analyses of leaf phenolics and other phytochemicals (Doyle & Scogin, 1988a, 1988b; Doyle, 1990) and is accepted in this article. As part of Haloragaceae, Gunnera has often been placed in the Myrtales, alongside Onagraceae (Emberger, 1960; Melchior, 1964; Soo, 1975; Moore, 1993). Although there are phytochemical similarities between Gunnera and Onagraceae (Gibbs, 1974; Doyle & Scogin, 1988b; Doyle, 1990), Haloragaceae and Gunnera differ from Myrtales in numerous morphological and anatomical characters (Corner, 1976; Doyle & Scogin, 1988b; Conti et al., 1996). Cladistic analyses of rbcL, 18S, and atpB sequence data also argue against this placement (Chase et al., 1993; Conti et al., 1996; Hoot et al., 1999; Soltis et al., 2000). On the basis of chemical similarities, Gibbs (1974) suggested placement in the Umbellales, although additional data countered this (Doyle & Scogin, 1988b). Thorne (1989, 1992) put Gunneraceae near Vitaceae, Haloragaceae, and Cornaceae within the Cornales. Orchard (1975) supported the affinity of Haloragaceae and Cornaceae on the basis of floral and wood anatomy but excludes Gunnera. On the basis of sieve-tube plastids, Behnke (1981) associated Gunnera with Vitaceae and Leeceae, and later suggested a close affinity to Connaraceae and Eucryphiaceae (Behnke, 1986). A relationship has also been suggested between Balanophoraceae, Gunnera and Haloragaceae on the basis of an epigynous, valvate perianth with opposite stamens; a single, pendulous ovule; and seeds with a strongly adhering, thin testa, and cellular endosperm (Hooker, 1856; cf. Hansen, 1980; Mabberly, 1993). Most older classifications agree in placing Gunnera in the subclass Rosidae (or Thorne's Rosanae), an idea supported by the presence of mutualistic Mycorrhizae in Gunnera petaloidea (Koske et al., 1992). However, the rosid status of Gunnera now seems unlikely in light of recent genetic studies that fail to support the placement of Gunnera near Myriophyllum (Haloragaceae), nor within clearly rosid groups (Chase et al., 1993; Morgan & Soltis, 1993; Qiu et al., 1993; Soltis et al., 1993, 2000, 2003; Sytsma & Baum, 1996; Soltis & Soltis, 1997; Hoot et al., 1999; Angiosperm Phylogeny Group, 2003; Hilu et al., 2003). The other taxa suggested above were also distant.
As molecular-based phylogenies have multiplied, Gunnera has found a prominent place in the phylogeny of the lower angiosperms with tricolpate pollen (i.e., the eudicots). Early rbcL analyses put Gunnera as a sister group to a caryophyllid-rosid-asterid clade derived from a paraphyletic basal hamamelid grade. Among the closest taxa to Gunnera, according to the rbcL study, are Trochodendron, Tetracentron, Nelumbo, Platanus, and Hamamelis (Chase et al., 1993). On this basis our leaf architectural studies included basal hamamelid groups for detailed morphological studies (Fuller 1995a, 1995b; this article). Combining rbcL and ITS sequences, Soltis and Soltis (1997) placed Gunnera with some magnoliid taxa, on a branch outside a broadly saxifragalean-hamamelidalean group.
Recent analyses, including additional molecular data sets, have strengthened the likelihood of evolution early within the tricolpate angiosperms and have proposed the hypothesis of a sister-group relationship with Myrothamnaceae (Hoot et al., 1999; Soltis et al., 2000, 2003; Hilu et al., 2003). A placement near "Berberidopsidales" or Buxaceae is also suggested by recent genetic analyses. Caution is warranted here because long terminal branch length often makes cladistic placement unstable (Baum, 1994) and leads to a more basal placement of such terminal taxa (Sytsma & Baum, 1996). Because the fossil record suggests that Gunnera has been a distinct lineage for at least 93 million years, considerable anagenesis and apomorphy acquisition is likely.
The implications of tricolpate pollen in Gunnera have never been seriously considered. Recent phylogenetic studies suggest that the tricolpate and tricolpate-derived pollen of higher dicotyledons are monophyletic, forming a "eudicot clade" (Doyle & Hotton, 1991; Chase et al., 1993; Albert et al., 1994; Doyle et al., 1994; Crane et al., 1995; Sytsma & Baum, 1996; Hoot et al., 1999; Soltis et al., 2000, 2003; Hilu et al., 2003), if the tricolpate pollen of Illiciales, which shows radial arrangement of colpi within the pollen tetrad, is excluded as being an independent, convergent origin (Huynh, 1976; Donoghue & Doyle, 1989; Doyle & Hotton, 1991). Tricolpate pollen is largely absent from the traditional Rosidae, with the tricolporate state considered primitive in the subclass (Dickison, 1989). Notable exceptions are some Saxifragaceae, namely some Saxifraga s.l., especially section Micranthes, and Chrysosplenium (Ferguson & Webb, 1970; Heusser, 1971; Gupta & Sharma, 1986), some Podostemaceae, and Gunnera (Erdtman, 1966).
The fossil record indicates that tricolpate pollen originated after, and evolved from, monosuleate forms, and preceded polycolpates and all porate forms (Doyle, 1969; Doyle & Hickey, 1976; Hickey & Doyle, 1977; Traverse, 1988; Doyle & Hotton, 1991). Doyle and Hickey (1976) suggested that tricolpate pollen represents an advance, allowing more efficient release of recognition proteins and pollen tubes. The multiple origins of tricolpatederived lineages suggest that there are adaptive reasons for advancing beyond the tricolpate toward the porate condition. This directional evolution is strongly supported by the fossil record in which, after the early Aptian, tricolpates quickly spread from the equatorial zone to all latitudes (Brenner, 1976; Hickey & Doyle, 1977; Lidgaard & Crane, 1988), and stratigraphically higher, tricolpate-derived forms increase in relative proportion to tricolpate types (Doyle, 1969; Hickey & Doyle, 1977; Traverse, 1988). There are no documented reversals of this trend, so on this basis alone it seems highly unlikely that Gunnera belongs with the Haloragaceae (s.str.) with its polycolpate and porate pollen forms resembling the fossil formgenus Normapolles (Praglowski, 1970).
Pollen of this level of advancement originates in the late Cenomanian, considerably after tricolpate pollen (Doyle, 1969; Hickey & Doyle, 1977; Traverse, 1988; Kedves, 1989). Thus the derivation of Gunnera from monostelic Haloragaceous ancestors (Scott, 1891; Arber, 1920) would imply major reversals in pollen morphological evolution, since nonmonostelic, terrestrial species like Haloragodendron or Glischrocaryon (compared with Gunnera by Meijden & Caspers, 1971) with hexacolpate, pertectate pollen and secondary growth (Orchard, 1975), would have to be considered ancestral in such a scheme. In fact, putatively primitive species of Gunnera do have a limited vascular cambium in their rhizome but none in their stolons or petioles (Batham, 1943), making a fully monostelic ancestry difficult to accept. Similarly, the spherical, triporate pollen of Myrothamnus suggests a later evolutionary development than do the pollen characters of Gunnera.
Recent genetic studies linking Gunnera and Myrothamnus as sister groups (Soltis et al., 2000, 2003; Hilu et al., 2003) raise interesting evolutionary questions but fail to offer insight into the evolutionary origins of the Gunneraceae and its phylogenetic relationship to largescale trends in dicot evolution. Although this pairing has arisen in recent studies, caution is warranted due to concerns over long branch length attraction as well as anatomical and morphological characters. As noted by Wilkinson (2000), there is little anatomical or morphological evidence for a close relationship, which is confirmed by leaf architectural, pollen, and other characters discussed below. Were the hypothesis of a Gunnera-Myrothamnus clade to be accepted, their highly divergent morphologies and the specialization of Myrothamnus would mean that the ancestral morphological character states and the evolutionary origins of either group were not clear.
B. GENERIC CIRCUMSCRIPTION AND SUBGENERA
Schindler (1905) defined Gunnera largely on the basis of fruit and floral morphology. Gunnera has panicles of "reduced" flowers, sometimes unisexual, consisting of two bithecal anthers with subtending sepals and/or a single unilocular pistil with two long, feathery stigmas. Its fruits are single-seeded drupes with small, cordate embryos. When a new species, G. herteri, which lacked long stigmas, was added to the genus (Osten, 1932; Mattfeld, 1933), it became clear that the presence of Nostoc in specialized glands is also a defining character of the whole family, although this was not shown to be a symbiotic adaptation until later (Silvester & Smith, 1969).
Gunnera is traditionally divided into six subgenera that are largely geographically distinct (Fig. 2). Schindler (1905) identified five subgenera on the basis of size, biogeography, and general floral habit, to which Mattfeld (1933) added the monotypic subgenus Ostenigunnera. Milligania incorporates 6 (to 11) species of creeping, stoloniferous herbs from New Zealand and one species from Tasmania, which are often dioecious, although the species G. cordifolia, G. strigosa, and G. monoica are monoecious (Schindler, 1905; Cheeseman, 1925; Allan, 1961; Webb et al., 1988; Table II). Misandra includes three species of prostrate, stoloniferous herbs, all dioecious, found in South America. Pseudo-Gunnera is erect and stoloniferous, producing panicles of unisexual female flowers basally and perfect, proterandrous flowers apically. The sole species of this subgenus is G. macrophylla, found in New Guinea, the Philippines and scattered on some volcanic islands of Melanesia (Meijden, 1975). Perpensum contains a single nonstoloniferous, erect species, Gunnera perpensa, found in South Africa, Madagascar, and East Africa (Lowry & Robinson, 1988). Infloresences in Perpensum are like those in Pseudo-Gunnera. Panke consists of more than 19 species in South America and the Hawaiian and Juan Fernandez Islands (Mora-Osejo, 1984; Doyle, 1990). These species produce laminae up to 2 meters across, supported by erect petioles that arise near the apex of a pachycaul. Infloresences are large (up to 2 meters long) panicles of hermaphroditic, but occasionally, unisexual, flowers. The subgenus Panke lacks stolons. The number of species in Panke and Milligania is problematic because of a high degree of gross morphological variability within and between populations, as well as interspecific hybridization (Palkovic, 1974, 1978; Webb et al., 1988; Doyle, 1990; Pacheco et al., 1991).
[FIGURE 2 OMITTED]
Several additional subgeneric classifications have been suggested (Table II). To Schindler's treatment, given above, Mattfeld (1933) added the monotypic subgenus Ostenigunnera, incorporating the newly discovered species Gunnera herteri (Osten 1932), distinctive for its minuscule size, small (but otherwise stenopalynous) pollen (Praglowski, 1970; Jarzen, 1980), cauline growth without stolons, and axillary spikes with proterandrous male flowers apically and clusters of female (occasionally bisexual flowers) basally. Maclaughy (1917) published a slightly different classification of Gunnera, subsuming G. perpensa into the subgenus Pseudo-Gunnera. More recently, it has been suggested that only two subgenera should be recognized, the mono-specific Ostenigunnera and Gunnera comprising all other species within three sections, Panke, Misandra, and Gunnera (i.e., Perpensum, Pseudo-Gunnera, and Milligania) (Meijden & Caspers, 1971; Meijden, 1975). However, this article will use the subgeneric divisions of Schindler (1905) as emended by Mattfeld (1933).
None of these earlier taxonomic works was explicit in proposing a phylogeny within the genus Gunnera. Most authors have pointed to Milligania as the most basal subgenus (Schindler, 1905; Batham, 1943; Mora-Osejo, 1984; Bergman et al., 1992), whereas the possible basal status of Ostenigunnera is implied in the work of others (Bader, 1961; Meijden & Caspers 1971). More recent morphological evidence (Fuller, 1995a, 1995b; Wilkinson, 1998) and genetic data (Wanntorp et al., 2001, 2002) strongly indicate the basal status of G. herteri. The cladistic analysis of ITS, rbcL, and rps 16 data sequences of 22 species of Gunnera by Wanntorp et al. (2002) led to their proposed phylogeny, in which the large-leafed G. macrophylla is sister group to small-leafed Milligania and Panke is descended from small-leafed Misandra. These analyses were based on the assumption of Myrothamnus as the sole outgroup. Given the importance of outgroup selection for polarizing character states within Gunnera, the present study considered a wide range of potential outgroups in order to find the best morphological sister group for analyzing phylogenetics within Gunnera.
III. Materials and Methods
A. CLEARED LEAF PREPARATION
In order to study foliar morphology, leaf clearings were made following the procedure of Foster (1952) with modifications from Hickey (1973). Most specimens were obtained from herbarium sheets, although Gunnera insignis came from material preserved in FAA solution and fresh leaves of G hamiltonii were collected from a botanical garden (Strybing Botanical Garden, San Francisco). Clearing was done with 5% NaOH solution followed by acetolyzation and further clearing with 5% sodium hypochlorite. Chloral hydrate 250% solution was used as mordant, and staining was done with an ethanol solution of 1% safranin O, produced by EM Scientific, Cherry Hill, NJ. Specimens were mounted in a 50-60% concentration of Canada balsam in xylene, from Anchemia Scientific, Montreal. Specimens were flattened with a photographic roller and mounted with an identification label between slides of optical glass. All specimens were added to the National Cleared Leaf Collection (NCLC), currently housed at Yale University (Table III).
The process differed to some degree among specimens. Some specimens (Gunnera killipiania, G. herteri) were clear enough after NaOH treatment that the acetic acid and sodium hypochlorite steps were skipped. Bleaching was then followed by an additional rinse. Specimens of the subgenus Milligania species consistently retained opaque blotches that were removed by soaking in 5% chromium trioxide solution for a period of 10 minutes to an hour.
B. LEAF ARCHITECTURAL ANALYSIS
The systematic use of leaf morphology has been demonstrated in paleobotanical and recent systematic studies (e.g., Dickison, 1973; Hickey & Wolfe, 1975; Wolfe, 1989; Hickey & Taylor, 1991 ; Taylor & Hickey, 1992; Gornall et al., 1998). Leaf architecture seemed a particularly useful approach to Gunnera because this genus encompasses a wide range of leaf size, form, and venation pattern. Meanwhile, variation in fertile structures is rather limited, as the minute, single-seeded drupes and paniculate infloresences are conservative throughout the genus. We examined cleared leaf material from 18 Gunnera species for leaf architectural characters. In addition to the cleared leaves, herbarium specimens were consulted for confirmation of many characters. For the larger-leafed species only small portions could be cleared, and these had to be supplemented by observation of herbarium specimens. Owing to the large size of leaves in subgenus Panke species, herbarium specimens are often only partial leaves or have been folded repeatedly. That made it difficult to assess characters of overall leaf form and the course of its primary venation. We therefore consulted as many specimens, published drawings, and photographs as possible (e.g., St. John, 1946, 1957; Mora-Osejo, 1984). In these largest leaves, the highest order of venation (sixth or seventh) proved difficult to observe because they are poorly lignified and do not stain well.
The basic system used to describe venation character states derives from Hickey (Hickey, 1973, 1977, 1979; Hickey & Wolfe, 1975) and conforms to the codification of the Leaf Architecture Working Group (Ash et al., 1999). Terms and concepts that were not in the original system (Hickey, 1979) are discussed here. In addition, a few concepts developed during the course of this study are discussed in the description of Gunnera leaf architecture (see "Results" and "Discussion"; also Fig. 3). Basal lateral primary veins and basal secondary veins form a spectrum and can all be considered agrophic veins' (approximately equivalent to pectinal veins sensu Spicer, 1986), which are characterized by giving rise to a succession of higher-order, excurrent, adaxial veins. (These are equivalent in their course and behavior to secondary veins.)
[FIGURE 3 OMITTED]
Often the basal-most excurrent branch of an agrophic vein behaves in a similar fashion, giving rise to a series of lateral veins. In this case, different orders of agrophics can be distinguished by Greek-letter prefixes, as alpha (the first), beta (the next), and so forth (Fig. 3). Tertiary veins can be divided on the basis of the position of their insertion and the orientation of their course (following Pole, 1991; see Fig. 3C). One set of tertiary veins runs to the leaf margin. Among these tertiary veins, the externals originate excurrently on the abaxial (basal) side of secondary veins, and the counterexternals originate on the adaxial (apical) side of the secondaries. These are distinguished from reticulate or ramifying tertiaries, and joining tertiaries (cf. Pole, 1991), which include percurrent veins and interangular veins (sensu Pole, 1991). These terms can also be usefully applied to some of the second- and third-order venation in Gunnera.
Venation within marginal teeth was described with terms adopted from Hickey and Taylor (1991), distinguishing the principal vein from the conjunctals, which converge with the principal vein toward the tooth apex. The admedial is a vein that originates just beneath the tooth and runs beneath or toward the sinus of the tooth but does not become the principal vein of another tooth. Accessories are higher-order veins within a tooth framed by the principal, admedial, and conjunctal veins. In addition to describing venation, leaf rank was useful for determining the level of leaf organization and evolutionary sequences (Hickey, 1977; cf. Doyle & Hickey, 1976; Hickey & Doyle, 1977; Hickey & Taylor, 1991). Trichomes were also examined following the guidelines and terminology of Theobold et al. (1979). Crystalline inclusions visible in cleared leaf tissue were classified on the basis of the terminology of Radford et al. (1974). Epidermal structure was described following the terminology of Dilcher (1974). However, epidermal characters could not be assessed for several specimens in which the epidermis appeared to have been damaged by the clearing process.
C. PHYLOGENETIC ANALYSIS
The characters obtained from study of cleared leaves formed the core for a cladistic analysis of the phylogeny within Gunnera. In addition to 39 foliar characters derived from this study, a literature survey produced 11 pollen characters as well as 15 other characters relating to growth habit, inflorescence morphology, reproductive anatomy, and sieve tube anatomy (see Appendix). Although most of the second two groups of characters are considered common to the entire genus (cf. Cronquist, 1981; Wilkinson, 1998, 2000), most have only been studied in a limited number of species, especially in subgenera Panke, Pseudo-Gunnera, and Milligania. Even though we have coded these on the basis of the available data, there remains the possibility that more than one state of the character exists within the genus.
Certain difficulties were encountered in coding the characters. Some characters intergrade along a spectrum that made dividing them into distinct character states somewhat arbitrary. For example, the common leaf margin in Gunnera consists of convex-convex teeth, forming a single order of crenations. Some species have a fully developed second order of teeth, but many others have only an occasional secondary tooth. Coding the latter as either singly or doubly crenate unnecessarily biases cladistic analyses toward one group or another. Another solution would be to establish a separate state for intermediate conditions. This worked for some characters, but in other cases, as with marginal configuration, this would have unnecessarily divided species with essentially the same character state. In such cases arbitrary criteria had to be established. For example, leaves were only considered to have two orders of dentition if the majority of teeth had secondary teeth associated with them.
Several of the characters considered were dependent on the presence of another character (e.g., the characters of primary venation that were applicable only to palmately veined leaves, or characters describing venation within marginal teeth were necessarily absent from entire-margined species). Following the reasoning of Hickey and Taylor (1991), such characters were assigned separate, not-applicable (N/A) states for each taxon. This makes each character a unique autapomorphy for the terminal taxa and thus will not link taxa on the basis of absences, treated as shared-derived characters by the cladistic algorithm.
The resulting data matrix (Table IV) was the basis for several analyses using PAUP 3.1 (Swofford, 1993). Trees were further explored with MacClade (Maddison & Maddison, 1992). In light of more recent phylogenetic work on Gunnera and the proposed sister-group relationship with Myrothamnus, leaf architectural characters were examined on Myrothamnus, and this taxon was added to the matrix. Identical analyses were performed (now with PAUP 4.0) with the addition of Myrothamnus to test its effects on topology. Although the matrix includes 36 terminal taxa, a cladistic analysis of the entire data set was problematic. Several of the outgroups considered are distant, often derived groups (cf. Cronquist, 1981; Takhtajan, 1983; Thorne, 1992; Heywood, 1993). Necessarily, many intermediate taxa were excluded, as well as significant apomorphies of many outgroup taxa. Because the analysis concentrated on characters of Gunnera and then looked for comparable characters in the outgroups, the data are not necessarily appropriate for resolving higher-level relationships between outgroups.
In order to mitigate these problems the analysis was broken down into several steps. First, the least likely outgroups were removed on the basis of absolute and mean patristic distances from representative Gunnera species (Table V). The more similar outgroup taxa, including a basal Hamamelid outgroup and a rosid outgroup, were run using two outgroup rooting strategies. The data matrix was further reduced by using six Gunnera species to represent the ingroup. Both of these analyses were rerun with the addition of Myrothamnus. The resulting data sets could be searched using the branch-and-bound algorithm, which finds the absolutely most parsimonious trees. On the basis of the above analyses the most likely sister groups of Saxifragacaeae and Platanaceae were used as outgroups to root an analysis of all Gunnera species.
A. SYSTEMATIC LEAF ARCHITECTURE OF GUNNERACEAE
Gunnera (Linne, 1767) is a genus of herbaceous plants producing simple leaves ranging in size from less than 1 cm across in G. herteri (Fig. 1A) to more than 2 meters across in G magnifica (St. John, 1957). The wide range in size is accompanied by a range in texture from membranaceous in G herteri to coriaceous in the subgenera Pseudo-Gunnera and Panke, as well as in the species G hamiltonii (Fig. 1B-1D), with all other species being basically chartaceous. There is also a correlated increase in leaf rank (sensu Hickey, 1977; Hickey & Doyle, 1977); i.e., the regularity in vein organization through the hierarchy of vein orders. The veins in G herteri are highly irregular; i.e., low first rank, where the course of any given secondary vein is not predictable from that of another (Fig. 5B). Other small-leafed species, such as in the subgenera Milligania and Misandra, are low second rank, where the course of secondary veins is similar from one vein to the next, but higher-order veins are highly irregular (Figs. 3D, 4G). The subgenera Perpensum (Fig. 6A) and Pseudo-Gunnera (Figs. 3C, 8A) are high second rank, having tertiary veins that are fairly consistent in angle of origin and course. The subgenus Panke (Figs. 6B, 8B, 8C) has third-rank leaves, in which regularity in arrangement can be found up to fourth- or fifth-order veins. The fossil record of early angiosperms indicates a general evolutionary trend toward increasing leaf rank (Hickey & Doyle, 1977). The spectrum of leaf rank in Gunnera therefore suggests a polarity for the venation characters of the genus, with the smaller-leafed, low-rank subgenera being more basal.
[FIGURES 4-6 & 8 OMITTED]
The thick, coriaceous leaves of the large-leafed subgenera (Pseudo-Gunnera and Panke) also have an alveolar texture, in which veins on the lower (abaxial) surface are highly prominent, forming alveolae between them (Fig. 1D). These same veins are highly impressed on the upper (adaxial) surface, forming a colliculate ("hilly") texture (Fig. 1E, 1F). In these alveolar species four or more vein orders show high relief. In Perpensum only the first and second orders, and sometimes the third, have such relief and can be considered "subalveolar." Finally, in Misandra lower surface primary veins are highly prominent, whereas secondary veins are sometimes slightly so.
Most Gunnera species are heterophyllous, producing petiolate leaves as well as bracteose, cauline leaves that are sessile on the rhizome, although these cauline leaves have sometimes been confused with stipules (cf. Cronquist, 1988). They have usually been considered discrete leaves, which are often improperly called "ligules" or "squamae rhizomatis" (Schindler, 1905; Palkovic, 1974). An ontogenetic relationship of these structures with leaves is argued by Mora-Osejo (1984) and is clearly supported by the descriptive study of Wanntorp et al. (2004). In the subgenus Panke the venation in these "ligules" appears to be a highly reduced form of that in the normal, petiolate leaves. The species G. herteri and G. perpensa are isophyllous and lack these cauline leaves (cf. Osten, 1932; Humbert, 1950; Wanntorp et al., 2004). The leaf architecture described below is that of petiolate leaves.
Lamina are basically ovate (Fig. 1A-1E), to reniform-ovate and in many species more properly termed "reniform" (Figs. 1D, 1E, 3C), to essentially orbicular in many Panke (Figs. 6, 9). The leaf apex is rounded or acute (subgenus Milligania, and sometimes Pseudo-Gunnera), with the basically rounded apex in Panke becoming emarginate or deeply embayed (Figs. 1D, 1E, 3A, 3B). The leaf base is decurrent or lobate (Fig. 1F).
[FIGURE 9 OMITTED]
Although there is a wide range in marginal form within Gunnera, it is possible to suggest an underlying order that unites the variations in a hypothetical developmental spectrum (Fig. 3). In Panke species with orbicular leaves, there are marginal indentations between primary veins. Although the projections thus formed might be termed "lobes," we will refer to one of these projections as an akroterion (from the Greek, meaning a small peninsula). This new term is necessary to allow a comparison between lobes and teeth. Lobes have been defined on the basis of depth of indentation toward the midvein (Hickey, 1979; Ash et al., 1999), whereas teeth refer to much smaller and strictly marginal portions of the leaf. An akroterion allows both kinds of structures to be considered in relation to each other and to venation.
An akroterion is the smallest lobe or marginal projection in which a primary vein or agrophic secondary vein terminates and which is set apart laterally from other primary veins by embayments of the margin. An akroterion cohort consists of all the akroteria that are produced by marginal bifurcation of one of the basal primary veins of the leaf. A primary vein may end in a tooth, but an akroterion is the full lobe with all of the secondary ramifications of a single primary branch. An akroterion cohort may consist of only a single akroterion (Fig. 3D), or a pair of them if the primary bifurcates on its way to the margin (Fig. 3A). Additional primary branches produce cohorts with more akroteria (Fig. 3B). For example, in the subgenus Panke multiple akroteria can be grouped into cohorts. In Panke species with a pedate margin (Gunnera pilosa, G. talamancana), cohorts may include only pairs of akroteria (Fig. 3A). In orbicular-leafed Panke (Fig. 3B) akroterion cohorts usually consist of numerous akroteria. The basic number of akroterion cohorts appears to be three. In nonlobed, but large-leafed, species--i.e., G. macrophylla and G. perpensa--akroterion cohorts are still present (Fig. 3C). In such species the leaf can often be divided into three zones, each of which is served by a cohort of primary veins sharing a common basal origin. The boundaries of these akroterion cohorts are marked by only the slightest indentation or notch in the margin. In this example akroteria are merely the marginal teeth in which the primary veins terminate. Thus, although these leaves are not lobed, they may still be compared with those of Panke, and it appears that the akroteria of Panke are homologous to primary marginal teeth of Perpensum and Pseudo-Gunnera.
The small leaves of the subgenera Milligania and Misandra can also be described using this concept. Most of these species lack lobes, although akroteria can be distinguished. In Gunnera monoica each primary vein is contained within a shallow lobe, which is therefore an akroterion (Fig. 3D). In other species of Milligania the akroteria are less clearly distinguishable although the teeth terminating agrophic veins tend to be larger than those supplied by lower-order teeth. These akroteria can still be grouped into akroterion cohorts, of which there are three in the unlobed leaves of G. strigosa, G. dentata, and G. prorepens (Figs. 3D, 4D, 4E). Since these akroteria are formed around single primary veins (or strong basal secondary veins), they are also equivalent to the akroterion cohorts of larger species. The marginal "lobes" in G. lobata are enlarged crenations. Sometimes these are akroteria in their own fight (Fig. 4A, apical tooth), whereas on other specimens there are additional secondary teeth (as in the basal akroteria of Fig. 4A). Although G. lobata has five akroteria, these can be grouped into three akroterion cohorts because there are three distinct veins at the leaf base. In G. magellaniea (Fig. 4B, 4G), G. herteri (Fig. 5B), and G. cordifolia akroteria are not distinct from teeth; primary veins are terminated by marginal crenations, but these are not set apart from other portions of the margin.
Thus by relating margin patterns with the constituent venation it is possible to identify related patterns of venation development and leaf form that may be elaborated or simplified in the course of phylogeny. The akroterion cohort present in the subgenus Milligania species is equivalent to an akroterion cohort in the subgenus Perpensum, Pseudo-Gunnera, or Panke, except that in the latter taxa this cohort has been expanded through the reiteration of akroteria and filled out with more teeth (Fig. 3). The akroteria in Panke, which in some cases are clearly demarcated lobes, are the elaborated equivalents of the primary marginal teeth in Perpensum, Pseudo-Gunnera, or small-leafed species. This suggests a developmental spectrum in which higher-order veins ending in small marginal teeth are promoted to lower-order (primary) veins forming large lobes, resulting in a form of peramorphosis (cf. Alberch et al., 1979; Niklas, 1994). These relationships of leaf form and venation could be the result of reduction if the polarity is assumed to be from larger-leafed species toward the small; recapitulation (peramorphosis) is congruent with the directionality of evolution implied by leaf ranking and outgroup comparison (see "Phylogenetic Analysis within the Gunneraceae" below).
Primary venation in Gunnera is basically palinactinodromous, except in some species of the subgenus Milligania. A number of pinnately veined species occur in Milligania (Fig. 4C, 4E), although their basal, agrophic secondaries are considerably thicker than the more apical secondaries. These agrophic secondaries are clearly homologous with the lateral primaries of actinodromous species, such as G. monoica and G. strigosa (Figs. 3D, 4D). Gunnera hamiltonii is intermediate between the palinactinodromous and the pinnate condition, because it has strong basal lateral veins that arise from distinct lateral petiole veins, although these are somewhat weaker than the midvein; nonbasal venation is like that in pinnate species (Fig. 4F). In addition, Milligania includes G. cordifolia, which is palinactinodromous. In the monotypic subgenera Perpensum and Pseudo-Gunnera, the primary veins fork numerous times to form a reticulum of primaries (Figs. 3C, 6A).
The course of the primary veins shows characteristic trends that distinguish Gunnera from most of the other taxa examined in this study. The basic primary venation appears to consist of three primary veins at the leaf base. In G. herteri these arise sub-basally by two rapid bifurcations (Figs. 5B, 7A), whereas in G. hamiltonii they derive from separate, lateral petiolar veins (Fig. 4F). In Panke and some species of Milligania, the two bifurcations have moved nearly opposite one another, giving the appearance of a basal trichotomy, although the sympodial origin of this trichotomy is often still detectable (e.g., Fig. 4C, 4D). The lateral primaries bifurcate again above the base, giving rise either to the five basal primary veins of palinactinodromous species or to strong secondaries (in Milligania, with the exception of G. dentata, and G. hamiltonii, this bifurcation has been greatly reduced or lost). These basal, lateral primaries (or basal secondaries in pinnate species) are agrophic veins; i.e., they produce a comblike series of lower-order veins toward the leaf margin (Fig. 3). Thus the [alpha]-agrophic sometimes bifurcates to produce an abmedial [beta]-agrophic. This bifurcation process can be reiterative for as many as five orders of agrophics (Perpensum), as successive orders of external veins (sensu Pole, 1991) develop to form a basally directed spiral of veins (Fig. 4H). It is this reiterative venation that forms the earlike flap of laminar tissue at the lobate leaf base (Figs. 1D, IF, 3A, 3B), which is most pronounced in the larger-leafed species. Some species of Panke have peltate leaves or include peltate populations (e.g., G. peltata, G. insignis, G. kuaiensis Rock (see Palkovic, 1974; Doyle, 1990). This study did not include any peltate leaves.
[FIGURE 7 OMITTED]
The midvein is often dichotomous in some sections of the genus. In dichotomous species (in subgenus Panke and Gunnera herteri) the midvein bifurcates between one-quarter and one-half of the distance from the lamina base to the apex. Often each of the resulting primaries bifurcates again (Figs. 1C, 3B). The midvein bifurcates once in the deeply lobed, pedate species (e.g., G. pilosa, G. talamancana, G. brephogea, Figs. 1D, 3A), and twice in species with generally larger, orbicular leaves (e.g., G. insignis, G. killipania, G. manicata). Bifurcation of the midvein reaches an extreme in G. magellanica, in which there is no midrib but rather two medial primaries that bifurcate at the leaf base (Fig. 4B, 4G). Other species (e.g., G. perpensa, G. macrophylla) show a simple midvein with two pronounced subopposite lateral veins (sensu Pole, 1991) of the same, or nearly the same, thickness that arise from the midvein and then curve toward the leaf apex (Fig. 6A). This suggests that the bifurcating midvein of Panke is derived from a simple midvein by splitting the midvein and strengthening the lateral veins. This surmise is supported by venation in leaves of young plants of Panke that have actinodromous-pinnate venation with similarities in the pattern of their midveins to G. macrophylla, G. perpensa, and subgenus Milligania (Fig. 6B-6D).
Secondary veins are reticulate or craspedodromous and often curved, although some are slightly recurved (Gunnera dentata, G. hamiltonii, Fig. 4E, 4F). Secondaries usually originate at angles ranging from 35[degrees] to 55[degrees], but more obtuse angles occur in some species of the subgenus Panke. Members of Panke have a succession of interangular veins between primary veins (four in G. pilosa; six or seven in G. insignis (Fig. 6B) and G. killipiania), whereas other subgenera have significantly fewer. Intersecondaries are composite (Panke and G. macrophylla) or lacking. Tertiary and higher-order veins are reticulate and usually orthogonal. In G. perpensa and G. macrophylla (Fig. 8A) secondaries from adjacent primaries merge into a single vein running to the leaf margin that equals the strength of its two source veins (Figs. 6A, 8A). Third- (and sometimes fourth-) order veins follow a similar pattern. Due to repeated reticulations in G. perpensa, veins near the margin are extremely attenuated, and different vein orders are almost indistinguishable on the basis of size (Fig. 9B). In the subgenus Pseudo-Gunnera the veins are reticulate and converge in the same pattern as do those in G. perpensa (Figs. 8A, 9A). However, primary veins are stouter toward the margin. These thicker primaries may be necessary to support the larger leaves of G. macrophylla. In Panke some opposite tertiary veins join, but secondaries do not (Fig. 8C). Instead, adjacent secondaries run toward each other and are joined by a series of third-order veins that ramify between them (Fig. 8B).
Areoles range from imperfect to well developed and are predominately quadrangular. In smaller-leafed subgenera (Ostenigunnera, Milligania, Misandra) areoles tend to be elongated radially, as though radiating away from the leaf base (Fig. 5B, 5G). Areoles generally lack freely ending veinlets. However, some veinlets occur in a minority of the areoles of large leaves (subgenus Panke, also in Gunneraperpensa and G. macrophylla); these are simple and usually curved. Marginal venation is generally looped, although some freely ending, hooked veins occur in G. perpensa (Fig. 9B).
[FIGURE 9 OMITTED]
The leaf margin is always toothed, with generally crenate or dentate projections. The basic tooth shape is convex-convex (Type A1, Hickey, 1979) although some acuminate teeth are found. In Milligania the margin is dentate in two orders, with Gunnera dentata (Figs. 4E, 5A) and G. hamiltonii (Fig. 4F) tending toward serrate, and the second-order teeth of these species are generally convex-concave. Each tooth has a glandular, tylate apex. Ostenigunnera, Milligania, and Misandra have dark-staining apical processes (Fig. 5A, 5B) that are distinct from the deciduous papillae on many rosid teeth (e.g., Haloragidaceae). Teeth of Milligania also have a small apical foramen.
The teeth of Gunnera are basically of the chloranthoid type, with conjunctal veins merging with the principal vein of the tooth oppositely or nearly so--i.e., in the manner of Chloranthus (Figs. 5A, 5B, 9B-9D)--but in some cases alternately--i.e., like those of Ascarina (Figs. 5C, 9E, 9F). In the putatively advanced subgenus Panke, tooth venation sometimes parallels the rosoid configuration, as the conjunctal veins remain distinct (i.e., connivent) and splay before joining, although some vascular strands always join the principal (e.g., G. mexicana, G. insignis, Figs. 5C, 9F).
The principal vein of the teeth has a direct, central course that terminates abruptly (truncate) or splays. Associated with the principal vein is an admedial vein, as well as conjunctals and sometimes reticulate accessories. The admedial vein is generally of a lower vein order than the principal, although in Gunnera herteri, G. strigosa, and G. prorepens the admedial and principal veins are of the same order. In some teeth G. perpensa has two successive pairs of conjunctals (Fig. 9B). The epidermal tissue on the teeth often has suberized, open stomata, which appear black in cleared specimens. These are rarely found elsewhere on the leaf. The sinuses of Gunnera are round and unbraced, except in the subgenus Misandra, which has glandular sinuses into which admedial and accessory veins converge and give rise to conjunctals (Fig. 9D). Otherwise, venation of the sinus consists of the conjunctals, admedials, and sometimes branches from them. In G. perpensa the sinus is subtended by a branch from a conjunctal vein. Sometimes this vein produces freely ending veinlets in the direction of the margin, as though to an incipient gland in the sinus (Fig. 9B).
In species of the subgenus Panke the margin usually has four or five orders of teeth. In some species "teeth" do not protrude far enough from the margin to have sinuses (sensu Hickey & Taylor, 1991). The resultant uneven margin is neither truly dentate nor entire but can be termed "repand" (Figs. 1D, 8C, 8D). These processes are clearly teeth, on the basis of their glandularity, venation, and homology with teeth on orbicular-leafed Panke (Fig. 8B). In some species the teeth become acuminate-acuminate (e.g., Gunnera killipiania, G. insignis, G. manicata, and G. mexicana).
Except for Gunnera herteri, Gunnera has unicellular, glandular hairs. These trichomes are ensiform with acuminate tips, but a few notable exceptions include the long, slender hairs of G. dentata and G. lobata. An occasional long, slender trichome is interspersed among the wider ones on G. manicata leaves. Of special note are the trichomes of G. lobata, which appear to have long, thin, black inclusions within the hair. The placement of trichomes is variable, but they are always found along the leaf margin and on veins of the abaxial surface. Often trichomes also occur along veins of the adaxial surface and in adaxial areoles, but less densely than on the lower surface. Only in G. dentata and G. perpensa were trichomes found in the areoles of the lower surface. The base of trichomes is socketed in a ring of epidermal cells that otherwise appear unspecialized. Some species of subgenus Panke also possess multiseriate and capitate trichomes, although it is unclear how widespread these are because they have only been reported for G. kauaiensis Rock (Wilkinson, 1998, 2000).
In addition to trichomes, Gunnera macrophylla and Panke species possess multicellular laminar processes often called "colleters" (Soloreder, 1908; Palkovic, 1974; Mora-Osejo, 1984). These have not been noted on any of the other subgenera or outgroups examined. These glandular processes are multicellular, round growths of tissue at the junction of high-order veins (fourth or fifth order). Soloreder (1908: 337) describes them as "more or less distinctly hemispherical warts, which consist of a considerable number of epidermal cells, arranged in longitudinal section, in the form of a fan, and adjoined internally by a few isodiametric parenchymatous cells." Colleters stain more darkly than does the surrounding leaf tissue. A second kind of process, which is nonglandular, is also found on the upper surface of Gunnera leaves. These "pseudo-colleters" are conical growths that arise in areoles and are subtended by sixth- or seventh-order veins. Although these two kinds of processes have both been considered colleters (Mora-Osejo, 1984), they are not functionally or evolutionarily homologous because no transitional states were observed. Wilkinson (1998, 2000) outlines a wide range of variation in epidermal processes, but the clearest division is between glandular "colleters" and nonglandular "pseudo-colleters."
Epidermal characters could not be observed in cleared specimens of all species. Nevertheless, the epidermis usually consists of five- or six-sided cells, which are elongate with undulate walls. In larger species (subgenera Pseudo-Gunnera, Panke) these cells are interspersed with polygonal cells that are prevalent along major veins. In a number of species stomata occur on either surface, often in clusters, especially on marginal teeth. Stomates are anomocytic, possibly the plesiomorphic state for angiosperms (Doyle et al., 1994) and therefore not likely to be phylogenetically informative for the placement of Gunnera. Some leaves have crystalline inclusions. Druses were noted in the subgenera Perpensum and Pseudo-Gunnera and in G. monoica. Gunnera dentata has sandy inclusions. Druses were reported for G. herteri by Mattfeld (1933).
Ostenigunnera is the most distinct subgenus of Gunnera. Gunnera herteri is a cauline, rhizomatous perennial with isophyllous, opposite leaves. Vegetative axes and infloresences are produced in the leaf axils. In addition, axes sometimes bifurcate (Fig. 1A). This species also has very low rank, membranous, widely ovate leaves with a decurrent bases, rounded apex, and marginal crenations (Fig. 5B). Although loosely palinactinodromous, the weak and reticulating primaries approach a flabellate condition. The midvein usually bifurcates twice before reaching the leaf apices. Secondary and tertiary veins of G. herteri branch at random angles to form a reticulum of irregular, imperfect areoles. Higher-order veins are lacking, as are trichomes. In the petiole, two thin but distinct veins run alongside the stout medial vein. These enter the base of the lamina and curve outward between the basal-most primary and the basal leaf margin (Fig. 7A). Here each soon bifurcates, with one branch joining the basal primary and the other tapering out into the laminar tissue. The course, branching, and position of this vein resemble the lateral primaries in herbaceous Saxifragaceae (see below), which are persistent in the petiole as distinct veins (Fig. 7B), as well as in the fossil taxon from the Aptian of Koonwarra, Australia described by Taylor and Hickey (1990). This vein provides negligible vasculature to G. herteri and probably represents a vestigial or incipient form of the three-veined petiole of an ancestral taxon. The homologous status of this character is further supported by its occurrence in G. hamiltonii (Fig. 4F), as well as by vestiges in G. strigosa (Fig. 7C).
B. THE SEARCH FOR SISTER GROUPS: COMPARATIVE LEAF ARCHITECTURE
Despite older taxonomic opinion (see Table I), Gunnera shows little affinity with Haloragaceae in its leaf architecture. In Proserpinaca (Haloragaceae), leaves are narrowly ovate with an acute apex, a decurrent base, and a serrate margin (Fig. 10A). The basic venation of this genus is pinnate with a massive primary vein. This suggests that leaf width might have been reduced in response to an aquatic habit while the midrib remained stout. This inference would seem to argue against a relationship to small-leafed Gunnera in which the midvein is more moderately proportioned. Secondaries in Proserpinaca are craspedodromous, sinuous, and marginal, arising at narrowly acute angles and becoming more obtuse apically. The tertiaries and quaternaries form a reticulum of irregular areoles lacking freely ending veinlets. Marginal venation is looped. Although Proserpinaca can be compared to G. dentata and aspects of the subgenus Milligania or G. herteri generally, the pinnate species in Gunnera suggest an actinodromous ancestry, whereas Proserpinaca leaves are strongly pinnate and suggest reduction from a true pinnate ancestry. In terms of the angles of origination and curvature of secondaries and intersecondaries, Proserpinaca shows similarities with the venation of Penthoraceae, which has been allied with Haloragaceae based on genetic data (Morgan & Soltis, 1993; Soltis & Soltis, 1997). Proserpinaca trichomes are multiseriate, whereas those in Haloragis are uniseriate and multicellular, three cells long and with a socket ring of particularly small cells. Teeth of Proserpinaca are strikingly different from those of Gunnera in their straight-straight margins, lack of accessory veins, deciduous papillate apices, and clearly rosoid venation syndrome of nonjoining, connivent conjunctals (Fig. 10A).
[FIGURE 10 OMITTED]
Onagraceae show some possible affinities with Haloragaceae but none with Gunnera. The basic venation is pinnate and craspedodromous in narrowly ovate leaves. Lopezia has simple, linear, freely ending veinlets within well-developed, quadrangular areoles. Marginal teeth are of a modified rosoid type, characterized by convergent, splaying veins but in which the course of the principal vein is deflected by a thickened apical conjunctal (Fig. 10B). However, Lopezia does have epidermal cells with undulate walls and unicellular trichomes like those in Gunnera.
Vitaceae was examined through both Vitis (Fig. 10C) and Ampelopsis. The two genera share ovate leaves with acute apices, cordate bases, and basally originating agrophic veins (primaries in Vitis) that give rise to a second pair of agrophics (for Ampelopsis see Hickey, 1977, Pls. 39.4, 40.2). Areoles are imperfect and irregular and contain simple, freely ending veinlets. Leaves are often alveolar. Ampelopsis has angular, glandular sinuses. Secondaries are craspedodromous; tertiaries, percurrent. Teeth have a characteristic venation that may be termed "vitoid" and appears to be chloranthoid derived (Fig. 10C). Alternate or subopposite conjunctals bifurcate just before joining the principal vein, which splays out into an epithemous glandular tooth apex. Admedial veins are of a higher order than is the principal vein of the tooth. There are two orders of accessory veins. Epidermal cells are hexagonal, straight walled, and subisodiametric. Leaves of the Vitaceae are most similar to the large-leafed Gunnera, a polarity that, if accepted, would imply the evolution of small Gunnera by reduction.
Pachysandra of the Buxaceae shows similarities to G. herteri and some species of the subgenus Milligania, with its ovate, crenate, decurrent-based leaf, its unicellular trichomes, and its teeth with alternately joining conjunctals (Fig. 11A). In addition, epidermal cells are isodiametric and straight walled. However, its principal veins originate suprabasally. In addition, incomplete areoles with numerous freely ending veinlets and spiked marginal venation clearly set the leaves of Pachysandra apart. However, marginal and tooth venation is similar to that of some Hamamelidaceae (Fig. 10D).
[FIGURE 11 OMITTED]
Aucuba and Griselinia were chosen as cornalean representatives because of their marginal teeth, although the comaceous affinity of Aucuba is problematic (Eyde, 1988; Chase et al., 1993; Morgan & Soltis, 1993; Angiosperm Phylogeny Group, 2003). Its teeth are clearly rosoid with connivent, splaying conjunctals. The teeth in Griselinia, when present (most species in the genus are entire), are reduced to a simple, spinose principal vein with nonbracing, unsplayed conjunctal veins. Such teeth have been suggested to be a derived rosid type (Hickey & Wolfe, 1975). Leaves of both genera are pinnate, semicraspedodromous with incomplete areolation, and lack agrophic veins. Epidermal cells are isodiametric and straight walled.
The leaf architecture of the "basal Hamamelids" has been more extensively studied and illustrated than have many groups due to their paleobotanical interest (e.g., Hickey, 1977; Crane, 1989; Wolfe, 1989). In general, families in this "grade taxon" share some features with Gunneraceae, but most of these features are widespread and seem more likely to be symplesio-morphies. Leaves are often ovate, with round or truncate bases and sharply acute leaf apices (acuminate in Trochodendron).
The pinnate-brochidodromous venation of Trochodendron shows affinities with acrodromous venation in that its secondaries are concentrated near the leaf base. The acrodromous venation of Tetracentron consists of five primaries that originate from the petiole (Fig. 11B). This is notably different from Gunnera, in which the petiole gives rise to three basic primary veins, of which the lateral two provide the fourth and fifth primaries at the lamina base. The marginal teeth have chloranthoid venation (Fig. 10F), with oppositely joining conjunctals, but lack a gland (cf. Hickey & Wolfe, 1975). Unlike Gunnera, the sinus venation consists of the conjunctal and its branch. Areolation is imperfect or incomplete (Tetracentron), irregular in shape, and encloses once-branched, freely ending veinlets. Trochodendron has stellate hairs, whereas Tetracentron lacks hairs. The leaves of Trochodendron and Tetracentron contain stellate idioblasts. Epidermal cells are hexagonal and isodiametric with straight cell walls. Stomata, confined to the underside of the leaves, are laterocytic (Endress, 1993c), thus of a more derived form than in Gunneraceae.
Cercidiphyllum shows few affinities with Gunnera. Its ovate leaf, with a truncate-cordate base, acute apex, and crenate margin, has acrodromous venation. Secondaries are brochidodromous. Areolation is incomplete, with multiply branched veinlets. In modern leaves the common condition is for the tooth to have a large apical foramen beneath which "conjunctal" veins converge but do not connive with the principal vein (Fig. 10G). However, in long-shoot and sucker-shoot leaves of living Cercidiphyllum, as well as in many fossil leaves, oppositely joining conjunctals are present. Fossil evidence indicates that well-developed opposite conjunctals fusing with the principal are common in leaves that appear to be related to Cercidiphyllumm (cf. Chandrasekharam, 1974) and that the modern condition of short-shoot leaves evolved after the Paleocene. Trichomes are lacking. Anomocytic stomata are found on the underside of leaves (Endress, 1993b), whereas the remaining epidermal cells are like those in Tetracentron and Troehodendron.
To encompass the full range of leaf variation in Hamamelidaceae, Liquidambar, Disanthus, and Hamamelis were examined, either of the latter two suggested to be at the primitive end of the family (Cronquist, 1968; Scwarzwalder & Dilcher, 1991; Endress, 1993a). The venation of Disanthus consists of acrodromous primaries with brochidodromous secondaries, suggesting affinities with Cercidiphyllum and Tetracentron. Unfortunately, Disanthus lacks marginal teeth for comparison. Hamamelis has teeth with only a single, high-order conjunctal vein joining the splaying principal vein on its basal side (Fig. 10D), not unlike those in Pachysandra. Hamamelis has pinnate venation with strong basal secondaries (agrophic veins), which suggests an actinodromous ancestry (cf. Wolfe, 1989). Stomata are paracytic (Endress, 1993a), whereas epidermal cells are undulateelongate. Areoles are incomplete, with twice-branched, freely ending veinlets.
Platanus shows general similarities with Gunnera, especially with subgenus Panke, in having actinodromous primaries and lobes that are homologous with teeth (akroteria). However, the primaries in Platanus originate suprabasally. Like some species of Panke. Platanus has craspedodromous secondaries, percurrent tertiaries, and an orthogonal reticulum of higher orders including imperfect, quadrangular areoles, although Platanus is of higher rank than even the most regular Gunnera. Platanus is also somewhat alveolar with three orders of adaxially impressed, abaxially prominent veins. However, it differs in the four- to six-vein orders of alveolarity in Gunnera. Platanus also differs in that its freely ending veinlets are twice branched. Epidermal cells are isodiametric and straight walled with five or six sides. The marginal teeth show affinities with the rosoid tooth because the conjunctal veins connive with the principal vein and splay out at the tooth apex without merging with the principal vein (Fig. 10E; cf. Wolfe, 1989). Some authors have indeed suggested a relationship between Platanaceae and the ancestors of the rosid subclass (Hickey & Wolfe, 1975; Crane, 1989; Endress, 1993a); others derive Platanus within the Hamamelidaceae (Schwarzwalder & Dilcher, 1991) or near basal Hamamelids such as the Trochodendrales (Thorne, 1992; Chase et al., 1993; Endress, 1993a; Kubitzki, 1993a; Loconte, 1996). A hypothesis of a close relationship with Gunnera would imply that small, low-rank Gunnera groups were derived through reduction from larger, more regularly organized leaves.
A number of other suggested taxa can easily be removed from consideration. The long, ovate, entire, pinnately veined leaflets of Eucryphiaceae, with straight, freely ending veinlets and no marginal teeth, offer few features that compare to those of Gunnera. A similar situation holds for Connaraceae (cf. Dickison, 1973). The Urticales were discounted because of their third-rank, pinnate venation with percurrent tertiaries and their derived pollen.
The Berberidopsidales of recent genetic phylogenies (e.g., Soltis et al., 2000; Angiosperm Phylogeny Group, 2003) were considered on the basis of Berberidopsis and Aextoxicon, which are highly divergent from one another. Berberidopsis has ovate leaves, with pinnate, semicraspedodromous venation. The basal two pairs of secondaries represent agrophic veins that originate at narrowly acute angles, whereas secondaries farther along the midrib originate at a wide-acute angle. The single, multistranded petiole vein is flanked by two separately sheathed strands, indicating derivation from a three-veined petiole. Berberidopsis leaves are of high second rank, with reticulate-random tertiaries, imperfect areolation, and multiply branched, freely ending veinlets. Marginal crenations are nonglandular with tapering principal veins and alternately joining conjunctals. These features are compatible with a basal eudicot derivation and with Takhtajan's (1997) placement of them at the base of the Violales, or near the Dilleniales.
Aextoxicon, by contrast, has toothless, elliptical leaves with slightly emarginate (retuse) apices and strongly decurrent bases. Venation is pinnate and brochidodromous. Its leaves are highly ordered and of low fourth rank, with regularly orthogonal quaternary veins, well-developed areolation, and occasional freely ending veinlets. These characters are more in keeping with placement in Celastrales (Cronquist, 1981) and show no affinities with Berberidopsis, Gunnera, or other "basal Hamamlids" examined.
Genetic analyses have recently suggested a relationship between Gunnera and Myrothamnaceae (Hoot et al., 1999; Soltis et al., 2000, 2003; Angiosperm Phylogeny Group, 2003; Hilu et al., 2003). This hypothesis was accepted without further consideration in the analyses of Wanntorp et al. (2001, 2002), and Myrothamnus genetic sequences only were used to polarize the character states in their Gunnera cladograms. Cleared leaves of Myrothamnus moschatus Baill., as well as published morphological and anatomical studies (Zavada & Dilcher, 1986; Endress, 1989; Carlquist, 1990; Kubitzki, 1993b), were examined to code equivalent characters for cladistic analysis and identify possible homologies.
Few characters show similarities between Myrothamnus and Gunnera. Although Myrothamnus has three petiole veins, in common with G. herteri and some species of the subgenus Milligania (Fig. 7A), this trait is likely to be a plesiomorphy of many early eudicots and early angiosperms, as indicated by its occurrence in the fossil record since the Barremian, 135 Ma (Taylor & Hickey, 1990; also Takhtajan, 1969: 51). The venation of Myrothamnus is of higher rank than that of Gunnera (low- to mid-second rank). Its areoles have simple, freely ending veinlets. The marginal teeth are rosoid with conniving, but not joining, conjunctals. The leaf tissue contains very fine, sandy, crystalline inclusions. As discussed above, the pollen of the Myrothamnaceae is triaperturate, has a clavate exine deriving from reduced columellae and a foot layer, and has no thickening near the apertures (Zavada & Dilcher, 1986).
Saxifragaceae s.str., delimited as an essentially herbaceous family (Hideux & Ferguson 1976; Takhtajan, 1983; Thorne, 1992; Morgan & Soltis, 1993; Soltis et al., 1993; Soltis & Soltis, 1997), provides the most likely candidates for comparison to Gunnera. We thus made an attempt to examine some of the more basal taxa in this family for leaf architecture. A wide range of additional genera was examined as herbarium material to establish widespread leaf architectural characters, including probable symplesiomorphies. Initially, taxa were selected on the basis of pollen, assuming that the tricolpate condition was more basal, for reasons argued in the introduction, as well as for their low leaf rank, for which the fossil record also suggests directionality (Doyle & Hickey 1976; Hickey & Doyle, 1977). Tricolpate pollen with reticulate exines occurs in Saxifraga, section Micranthes, and in the genus Chrysosplenium (Ferguson & Webb, 1970; Heusser, 1971; Hideux & Ferguson, 1976; Gupta & Sharma, 1986). These taxa also have lower-rank leaves than have most of the family.
Astilbe, often considered to be basal within the family (Savile, 1975; Soltis et al., 1993; but not in the more recent analyses of Soltis et al., 2001a), has tricolporate, reticulate pollen (of. Hideux & Ferguson, 1976) and higher-rank leaves (high second rank) than does Gunnera. Astilbe teeth differ from those of Gunnera in having alternately conniving conjunctals that splay at the tooth apex alongside the principal vein to produce a typical rosoid tooth. In some teeth a few vascular strands in the conjunctals join the principal vein, thus suggesting an evolutionary relationship (derivation) with chloranthoid teeth. In addition, Astilbe has a compound leaf with pinnately veined leaflets. Several recent genetic studies, which have elucidated well-supported clades within the Saxifragaceae, have failed to offer consistent resolution of basal relationships within the family (Soltis et al., 1993, 1996; Johnson & Soltis, 1994, 1995; Soltis et al., 2001a).
In some analyses Chrysosplenium is an early branch within the family or within the "heuchroid" clade of the family; Saxifraga, section Micranthes, also sometimes falls low within this group. The position of Chrysosplenium and a proposed sister relationship with Peltoboykinia is more weakly supported in a comparative analysis of multiple gene data than are most of the clades in this family (Soltis et al., 2001a). Genetic data strongly indicate that Micranthes is distinct from the rest of Saxifraga, which implies that shared morphological features of the two groups are symplesiomorphies, perhaps of the entire family (Soltis et al., 1996). Within Saxifraga s.str., the section Irregulares and S. mertensiana are primitive (Soltis et al., 1993: fig. 1; 1996, 2001a; Johnson & Soltis 1995: fig. 3) and were thus chosen to represent basal leaf architecture for Saxifraga s.str.. Thus, cleared leaves of Micranthes, Irregulares, S. mertensiana, and Chrysosplenium were examined in detail. The large number of similar venation characters suggests that these taxa retain numerous leaf architectural plesiomorphies, many of which are also shared with Gunneraceae.
Members of the Saxifragaceae examined have ovate (to reniform) leaves that are palinactinodromous with three distinct veins entering the lamina base from the petiole (Fig. 11C). The primaries form a loose reticulum with the secondaries and tertiaries. The lateral primaries dichotomize as they near the margin, not unlike those of Gunnera herteri. Areoles are imperfect and irregular in shape, but usually radially elongated. They lack freely ending veinlets. Teeth are convex-convex, with an outline intermediate between crenate and serrate.
The venation of saxifragaceous teeth includes conjunctals, admedials, and a single order of reticulating accessories. The tooth apices have round tylate processes and principal veins with truncate to slightly bulbous terminations, similar to those in the subgenera Ostenigunnera and Milligania; marginal veins are looped. An unusual feature of these teeth is that the conjunctals can be formed from extensions or branches of subadjacent secondary veins. Micranthes has multicellular, uniseriate hairs along its margin, as do many Saxifraga spp. (Gornall, 1986). Irregulares and S. mertensiana both have multiseriate trichomes; Chrysosplenium lacks trichomes. Chrysosplenium and Saxifraga have chloranthoid teeth with oppositely converging and joining conjunctal veins (Fig. 7D-7F). Some Chrysosplenium species show a tendency toward conjunctals that splay before joining (Fig. 7F), which is typically the case in section Micranthes (Fig. 11C) and Gunnera. In the saxifragaceous taxa surveyed, the principal vein of some teeth is joined by two pairs of conjunctal veins. Epidermal cells in Chrysosplenium and Micranthes are elongate with undulate cell walls, like those found in Gunnera. Stomata are anomocytic (Moreau, 1984). This survey is sufficient only to indicate a strong general affinity of Saxifragaceae with Gunneraceae. A broader study of the leaf architecture of the Saxifragaceae is necessary in order to fully elucidate its affinities with Gunnera.
A conspicuous feature across several genera of Saxifragaceae is the tendency to develop three lobes or three leaf areas that can be understood as akroteria, as described in the previous section on the leaf architecture of Gunnera. These can be observed widely across representatives of many saxifragaceous genera, including Saxifraga s.str., Sullivantia, Jepsonia, Heuchera, Peltiphyllurn, Leptarrhena, Mitella, and Tolmiea. This suggests that the basic ontogenetic pathway of these leaf features is shared across Saxifragaceae and Gunneraceae and that lobes have evolved multiple times by this pathway.
Two relictual species in southern South America are putatively basal within Chrysosplenium (Hara, 1957). Both have opposite leaves with a bifurcating axis, producing additional axes and infloresences in the axils of leaves. In this habit they (and other Chrysosplenium) resemble Gunnera herteri (Fig. 1A). They are isophyllous; i.e., lacking sessile cauline leaves, which are common among most Saxifragaceae and Gunneraceae but are also lacking in G. herteri and G. perpensa. Another possible plesiomorphy of Chrysosplenium and Gunnera is the growth of axillary infloresences that do not overtop the foliage. Hara (1957) considered this feature as peculiar to C. micrantha, but the same habit is found in G. herteri and G. monoica. Although recent genetic work has placed the South American C. valdivicum Hook as a derivative of a terminal East Asian clade (Soltis et al., 2001b), morphological features point to a very ancient, relictual status for the other South American species, C. micrantha.
E. CLADISTIC ANALYSES
In order to streamline the cladistic analysis presented here, some of the 15 outgroup taxa were eliminated on the basis of their mean patristic distances from Gunnera (Table V). The nearest sister group of Gunnera was determined by searching among outgroup taxa with averaged patristic distances from Gunnera of under 0.6 when rooted by more distant taxa. These analyses included a single representative of each of the Gunnera subgenera. This approach allowed Gunnera to be rooted in any of its subgenera and also adduced cladistic support for the closest sister group. Because the polarization of characters is crucial, analyses were performed using both basal hamamelid and rosid outgroups. The most parsimonious trees were then explored by including other taxa and minimizing the additional tree length. In response to the recent hypothesis of a sister relationship with Myrothamnus, these analyses were run again with the addition of Myrothamnus.
Using a basal Hamamelid outgroup, represented by Hamamelis, Tetracentron, and Cercidiphyllum, eight equally parsimonious trees were found in a branch-and-bound search. Their consistency indices ranged from 0.593 to 0.62. The tree with the highest consistency index is reproduced as Figure 12, which also shows the frequency of occurrence of each branch among all eight trees. In these, Gunnera forms a strong clade with the Saxifragaceae. When two trees in which Gunnera is not monophyletic and which have the lowest consistency indices are disregarded, all other trees support this Gunneraceae-Saxifragaceae clade. Gunnera herteri is the basal taxon in the Gunnera clade, with the rest of Gunnera (corresponding to the subgenus Gunnera sensu Meijden & Caspers 1971) forming a well-supported clade.
[FIGURE 12 OMITTED]
Among the outgroups, Pachysandra is the next-closest clade, followed by Platanus, although the placement of Pachysandra is the most variable aspect of these trees. With the addition of Myrothamnus, the Saxifragaceae-Gunneraceae clade remains strongly supported in the most parsimonious tree, and Myrothamnus groups with Hamamelidaceae. However, the varied polarization of characters in this set of trees made the ingroup topology of Gunneraceae differ among them, and Saxifragaceae became paraphyletic (Fig. 13).
[FIGURE 13 OMITTED]
The Saxifragaceae-Gunneraceae clade was also supported in analyses using rosid outgroups. This search included several traditional sister groups, as well as Pachysandra. Platanus was defined as the outgroup, because it was the closest taxon to the Saxifragaceae-Gunneraceae clade in the first analysis. In the initial branch-and-bound search, a monophyletic saxifragaceous clade was the sister group to Gunnera supported by 16 apomorphies. Pachysandra formed the next branch. When Griselinia and Aucuba were added, the shortest tree separated the Pachysandra-Saxifragaceae Gunneraceae clade from a rosid clade including well-supported Comaceae and Vitaceae clades (Fig. 14). With the addition of Myrothamnus, all of four equally parsimonious tress in branch-and-bound search grouped Saxifragaceae and Gunnera with Pachysandra as the next branch, whereas Myrothamnus was distant. All of the above analyses strongly support Saxifragaceae as the nearest sister group to a monophyletic Gunneraceae, whereas the Comaceae, Haloragaceae, and Vitaceae are all very distant and form a separate rosid group.
[FIGURE 14 OMITTED]
The cladistic relationship between Saxifragaceae and Gunneraceae is supported by a consistent set of apomorphies from the above analyses as well as by other shared traits between the two groups. The most striking difference between Gunneraceae and Saxifragaceae is in floral morphology. Whereas the latter usually has perfect, pentamerous flowers, Gunnera flowers have two or four parts and are all highly simplified. However, Saxifragaceae share with Gunnera the presence of two styles. In Chrysosplenium these are extremely short and arise from connate locules. In G. herteri the styles are also extremely small. However, in Gunnera they arise from a single, uniovulate locule. The two styles suggest the evolution of the Gunnera flower partially through reduction, as does the unitegmic ovule. As with about half the genera of Saxifragaceae, including Chrysosplenium (Savile 1975; Soltis et al., 1993), Gunnera has parietal placentation. Savile (1975) has pointed out that flowers within the Saxifragaceae are often morphologically specialized for particular dispersal mechanisms.
Gunnera infloresences would seem to be specialized for wind pollination and the dispersal of visibly exposed, red drupes. Gunnera flowers are reduced to essentially two parts, although they sometimes have four. The tetramerous, apetalous flowers of Chrysosplenium are likely to be plesiomorphic, as opposed to the perfect pentamerous flowers of the rest of the family. This is in line with recent recognition that tetramerous or dimerous flowers are basic to the basal eudicots (Soltis et al., 2003; Zanis et al., 2003). As suggested for Proteaceae, Buxaceae, and Papaveraceae (Soltis et al., 2003: 467), two decussately opposite dimerous whorls form a superficially tetramerous flower. This process of doubling meroisity through the merging of sequential whorls can be understood through the anthion concept developed to describe the homologies of early angiosperm inflorescence evolution (Hickey & Taylor, 1996); indeed, the earliest fossil flowers are essentially dimerous, consisting of spikes of decussately opposite, single ovules subtended by bracts (Taylor & Hickey, 1990, 1992).
F. PHYLOGENETIC ANALYSIS WITHIN THE GUNNERACEAE
In order to examine systematic relationships below the genus level, 18 Gunnera species were analyzed, with Saxifragaceae as a monophyletic outgroup, through a branch-and-bound search (Fig. 15). Four equally parsimonious trees resulted, with CI = 0.694, RI = 0.834, and a length of 164. The only differences between the trees were in the topology within Saxifragaceae and the placement of G brephogea within the Panke lineage, either at the base or in a position between a pedate-repand lineage (G pilosa and G talamancana) and larger, orbicular-leafed lineage. Rerunning our data matrix with Myrothamnus as outgroup to all Gunnera species yielded 12 equally parsimonious trees, including four with higher-consistency indices that are identical to our analysis in subgenus placement. The remainder of this discussion refers to our analysis that places Saxifragaceae as the outgroup, because this provides a more coherent polarization of characters for considering morphological evolution.
[FIGURE 15 OMITTED]
Although traditional taxonomy places the subgenus Milligania at the base of Gunnera (cf. Schindler 1905; Meijden & Caspers, 1971), all of the above analyses support G. herteri as the most basal extant species, as suggested also by genetic research (Wanntorp et al., 2001). All other species share a monophyletic ancestry, characterized by 13 apomorphies, such as unicellular trichomes, straight "thichotomous" midveins, the presence of agrophic veins, orthogonal tertiaries, pollen more than 25 [micro]m, and the presence of ligules (lost in subgenus Perpensum).
Above the node of G herteri, the genus can be divided into two lineages, one with unisexual flowers (the "Prorepens" clade in Fig. 16), the other often hermaphroditic with enlarged leaves (the "megaphyll" clade in Fig. 16). The grouping of the subgenera Misandra and Milligania has also been suggested by Wilkinson (2000) on the basis of an anatomical review. Nevertheless, Milligania shows some distinct apomorphies, such as glandular sinuses, in leaf-venation patterns. Milligania appears to have additional phytochemical apomorphies, as suggested by the fact that only leaves from this subgenus of Gunnera proved difficult to clear (see "Materials and Methods" above). By contrast, the phylogeny proposed by Wanntorp et al. (2001, 2002) raises problems. When we forced the cladogram into the subgeneric relationships of Milligania-Pseudo-Gunnera and Misandra-Panke, tree length was 180, 16 steps less parsimonious than the scheme proposed here. In addition to numerous homoplasies in leaf architecture, this topology requires two convergent origins of tmisexual/dioceous flowers (Fig. 17).
[FIGURES 16-17 OMITTED]
Milligania is further divisible into a pinnately veined lineage (G. dentata, G. prorepens, G. hamiltonii) and an actinodromously veined lineage (G. monoica, G. strigosa, and probably G. cordifolia). The two groups within Milligania were also recognized by Schindler (1905) in his diagnosis on the basis of fruit shape; clavate fruits are restricted to the actinodromous species.
Gunnera macrophylla (subgenus Pseudo-Gunnera) is the sister group of the subgenus Panke, sharing with it alveolarity, colleters, orthogonal fifth-order veins, and quadrangular areoles but differing in the course of its primary venation, marginal akroteria, and apparently independent development of the stoloniferous habit. Stolons in the subgenera Milligania and Misandra are axillary and presumably derived from axes and have adventitious root primordia below their apices (Wanntrop et al., 2004). The lack of root primordia (Wanntorp et al., 2004) in G. macrophylla may represent a distinctive evolutionary character. Whereas the stolons in Milligania and Misandra are subtended by foliose leaves, those in Pseudo-Gunnera are not and appear to substitute for leaves (authors' observation). In this regard, Meijden and Caspers (1971) made the intriguing observation that adventitious roots are produced on the undersides of G macrophylla leaves. Thus the stolons in Pseudo-Gunnera may be derived from leaves!
The subgenus Panke can tentatively be divided into two main groupings, based on whether their leaves are orbicular or pedate. However, resolution of evolutionary relationships within Panke is likely to be complicated by reticulate evolution because hybridization between these leaf groups appears to be frequent (Palkovic, 1978; Doyle, 1990; Pacheco et al., 1991). In the pedate species, the margin is sinuous (repand) and has incipient or poorly developed teeth, whereas the orbicular group has leaves with four or five orders of teeth. The groups may also be distinguished by the presence of deciduous sepals in the orbicular assemblage, as opposed to persistent sepals in the pedate species (cf. Schindler, 1905). Gunnera mexicana, G killipania, and G insignis have round, glandular colleters only and lack pseudo-colleters, which supports the suggested closeness of these species (Palkovic, 1974). Gunnera manicata and G chilensis, a close paraphyletic pair in this analysis, both have thick, succulent infructescence axes (Schindler, 1905) and striate cuticles (Wilkinson, 1998, 2000). More refined systematic studies of the largest-leafed, orbicular species of Panke are surely required, but it is likely that hybridization and reticulate evolution will make conventional parsimony analysis ineffective.
Our cladistic analysis supports aspects of previous taxonomic treatments of Gunnera while providing a firm basis for an improved infrageneric taxonomy (Table II). All five subgenera of Schindler (1905) and the sixth proposed by Mattfeld (1933) are monophyletic. The division of Meijden and Caspers (Meijden & Caspers, 1971; Meijden, 1975), in which G herteri is distinguished from the rest of the genus as its own subgenus, is clearly supported, because G. herteri represents a distinct lineage sharing an ancestry with a lineage containing all other extant species (also supported by the genetic studies of Wanntorp et al., 2001, 2002). However, the sectional divisions within the subgenus Gunnera of Meijden and Caspers are problematic, because their section Gunnera is clearly polyphyletic. The combination of G. perpensa and G macrophylla by Maclaughy (1917), though a paraphyletic grouping, was perceptive of their close affinity and shared ancestry, particularly evident from leaf architecture. As regarded by previous taxonomists, the subgenus Panke is a monophyletic clade with many apomorphies in leaf architecture, growth habit, and inflorescences.
|Printer friendly Cite/link Email Feedback|
|Author:||Fuller, Dorian Q.; Hickey, Leo J.|
|Publication:||The Botanical Review|
|Date:||Jul 1, 2005|
|Previous Article:||Index nominum familiarum plantarum vascularium.|
|Next Article:||Systematics and leaf architecture of the Gunneraceae.|