WHAT DOES MORPHOLOGY TELL US ABOUT ORCHID

Transcription

1 American Journal of Botany 86(2): WHAT DOES MORPHOLOGY TELL US ABOUT ORCHID RELATIONSHIPS? A CLADISTIC ANALYSIS 1 JOHN V. FREUDENSTEIN 2 AND FINN N. RASMUSSEN Department of Biological Sciences, Kent State University, Kent, Ohio 44242; and Botanical Laboratory, University of Copenhagen, Gothersgade 140,1123 Copenhagen K, Denmark A cladistic analysis of Orchidaceae was undertaken for 98 genera using 71 morphological apomorphies based on a reconsideration of previous character analyses and newly discovered variation. The equally weighted analysis found most parsimonious trees with low consistency (CI 0.29) but high retention (RI 0.83). The strict consensus reveals a significant amount of structure, and most traditionally recognized subfamilies are supported as monophyletic, including the Apostasioideae, Cypripedioideae, Spiranthoideae, and Epidendroideae. Orchidoideae in the broad sense are paraphyletic, giving rise to spiranthoids. Vanilloids are sister to epidendroids, although exhibiting several states otherwise found only in clearly basal groups, such as Apostasioideae. The nonvandoid epidendroids are poorly resolved, due to a high degree of homoplasy. The vandoids appear to be monophyletic, contrary to recent molecular evidence, possibly due to repeated parallel development of the vandoid character suite. The importance of vegetative characters as evidence putatively independent from floral features is demonstrated in the placement of Tropidia. Implied weighting analysis of these data resulted in similar patterns at high levels, although the Orchidoideae and Spiranthoideae may each be monophyletic and the nonvandoid epidendroids are more resolved. The high degree of structure implied in previous orchid classifications must be reconsidered, given the poor resolution at lower levels in the present trees. Key words: cladistics; classification; morphology; Orchidaceae; systematics. The Orchidaceae, at an estimated species (Atwood, 1986), are perhaps the largest family of angiosperms. They are known for their diversity of specialized pollination and ecological strategies and provide a rich system in which to study evolutionary patterns. Most systematic work on the family has been concentrated at the species level, describing the seemingly boundless variation that occurs particularly in tropical regions. This is not to say that no attention has been given to relationships at higher levels, however; for two centuries systems of classification have attempted to trace a pattern in variation above the generic level. Swartz (1800) was the first to produce such a treatment, and he was followed by Brown (1810), Lindley (1840), Pfitzer (1887), Schlechter (1926), and others. Recent systems of classification have been framed in the context of phylogenetic patterns, utilizing as much character information as possible. Such systems include those of Dressler and Dodson (1960), Garay (1960, 1972a), Vermeulen (1966), Brieger, Butzin, and Senghas ( ), Dressler (1981, 1993), Rasmussen (1985), Burns-Balogh and Funk (1986), and Szlachetko (1995). The development of the cladistic approach to data analysis has marked a shift toward a more explicit methodology in systematics. Explicitly cladistic studies of or- 1 Manuscript received 18 December 1997; revision accepted 11 June The authors thank Gustavo Romero, curator of AMES, in particular, as well as the curators of C and K, and the Botanical Garden of the University of Copenhagen for access to specimens, Robert L. Dressler and W. Mark Whitten for useful comments on the manuscript, and Kevin C. Nixon for providing CLADOS, with which one of the figures was printed. JVF was supported by an Arnold Arboretum Mercer Fellowship from Harvard University for part of this work and by NSF grant DEB FNR received support from grants and from the Danish Natural Science Research Council. 2 Author for correspondence. 225 chids were done as early as the beginning of the 1980s (Linder, 1981; Rasmussen, 1982), but involved only portions of the family. The first cladistic study of the family as a whole was that of Burns-Balogh and Funk (1986), which suffered several shortcomings. First, it was not explicit in the methods used to generate the tree shown; neither was it clear whether the tree presented was optimal, and if so, whether it was the only tree at that level of optimality. Second, the tree shown does not wholly correspond to the data matrix presented. Third, vegetative characters were excluded from the analysis. Fourth, there are significant questions about character-state recognition and interpretation, many of which were outlined by Dressler (1987). In spite of these problems, the study does serve as a starting point for further work. The most recent work of Dressler (1990a c, 1993) does not attempt an explicit analysis of the family as a whole, or even of smaller portions, but does employ phylogenetic thinking in suggesting relationships; it brings together much information on character variation and presents clear hypotheses of relationship to be tested further. Szlachetko s (1995) system is explicitly polythetic and therefore essentially intuitive, rejecting a cladistic approach as being too restrictive in determining the composition of groups. Hence, until now there has been no explicit morphological cladistic analysis of Orchidaceae. We here report the results of such an analysis and believe that it is important for several reasons. First, it is a useful tool to evaluate previous systems of classification. It is relatively easy to suggest relationships of organisms based on intuition; with other alternatives having been discarded, such hypotheses tend to take on an authoritative stature that can be difficult to evaluate. An explicit analysis reveals alternative hypotheses of pattern that might not otherwise be considered. Second, it provides a concrete, explicit set of character data and codings that form a solid

2 226 AMERICAN JOURNAL OF BOTANY [Vol. 86 basis for continued systematic work, as well as allowing the evaluation of traditionally emphasized characters. As new characters are developed, they can be added to the data set and subjected to further analysis. Third, it reveals where further efforts should be focused, in two ways. Instances of missing data in the matrix reveal an obvious need for additional study. These gaps occur primarily in poorly known and rare taxa, for which focused field work might be appropriate. Regions of the tree that show poor resolution also can direct future study, since they indicate a need for additional characters or possibly more intensive character analysis and improved hypotheses of homology. Finally, such an analysis provides a morphological character set that can be combined with other (i.e., molecular) data sets. Synthesis of the varied perspectives of these data sets would be expected to yield a maximally informative pattern. MATERIALS AND METHODS One difficulty with cladistic analysis of large groups is the choice of terminals for the analysis. In this step there are two goals: (1) trying to represent all of the variation that is present for the characters and (2) scoring only those character combinations that are present in real taxa. To accomplish the first goal, we chose terminals from most of the subtribes recognized by Dressler (1993) and included additional taxa that we knew had unique character combinations. With respect to the second goal, scoring all characters for a selected set of species would have been optimal, but is impractical, because we often relied on the literature for character scoring and different species have been examined for different characters. As a compromise, we chose a generic exemplar approach, meaning that if a character state was observed in at least one species it was scored for the genus, unless it was clear that a polymorphism was present, in which case the genus was divided into more than one terminal. Hence, it is possible that there is no single species in which all of the states that we scored for a genus occur together, but at least we have limited the error to the genus level. Characters were selected by reviewing those used in previous analyses and searching for variation that had not previously been analyzed. Our only criteria for character inclusion were that the states were mutually exclusive and that they did not vary within terminals. Characters were scored to the extent possible from living and preserved specimens, based primarily on collections at AMES, C, and K. All scoring information is available on request from the authors. This information was supplemented with published observations from the literature. Details concerning character scoring and important literature sources are given in the Results (Character Analysis section). The data matrix is shown in Table 1. We chose Hypoxis as the outgroup because the genus has been one of the traditional candidates for a member of the sister group to orchids (e.g., Hutchinson, 1959) and recent molecular analyses have shown it to be closely related to the orchids (Chase et al., 1995). Additional outgroups would have been important if we had intended to test the monophyly of the Orchidaceae, but this was not a goal of the analysis. The data were analyzed in two ways. The primary analysis used Wagner parsimony as implemented in NONA (Goloboff, 1993b). The heuristic search strategy was as follows: for each of 3000 random taxon addition sequences an initial set of trees was found by one pass through the data; these trees were then subjected to TBR (tree bisection reconnection) branch swapping, keeping 20 trees whenever the current shortest length or shorter was found. This collection of trees was then subjected to further TBR branch swapping, and all shortest trees up to the set maximum of trees were saved and a strict consensus tree was produced. The sequence of commands used was: nona; out xxx.out; h*;p xxx.dat; pack r rs 0; h/20; na wa h* mult*3000; h 60000; max* sv xxxtr.out g500 sv; nels in; s/ zzz. Several additional runs were conducted using the same procedure, although sometimes keeping fewer trees, in order to more thoroughly explore the tree space. The second analysis employed the implied weighting function of Goloboff (1993a), as implemented in PIWE (Goloboff, 1993b). This weighting procedure employs a weighting or fitness function that is based on the number of steps required for each character on a particular tree. Optimal or fittest trees are those that maximize the fitness function when summed over all characters. The details of this analysis were the same as for the Wagner procedure, except for using 2000 random addition sequences, saving 15 shortest trees for each, and then swapping up to fittest trees. The default constant of concavity was used (k 3). Jackknifing was implemented using the parsimony jackknifing program of Farris (1995). The data set was first modified by recoding all ordered multistate characters as binary characters. Ten thousand replications were performed. RESULTS All multistate characters were coded as unordered unless specified otherwise. Character analysis 0. root tubers 0 absent, 1 present Root tubers are thickened roots that serve as organs of perennation, producing shoots in subsequent seasons. These structures occur primarily in the Diseae, Orchideae, and Diurideae and have been studied intensively (e.g., Irmisch, 1850; Germain de St.-Piérre, 1855; Prillieux, 1865; White, 1907; Ogura, 1953; Pridgeon and Chase, 1995). Triphora was scored as having a true root tuber, but Pogonia and Isotria were not, even though their unthickened roots are known to produce shoots (Ames, 1922). Nodular root thickenings are known from Apostasia (Stern and Warcup, 1994), Tropidia (Dressler, 1981), and possibly others, but these evidently do not produce new shoots. 1. root epidermis 0 rhizodermis, 1 velamen 2. exodermis 0 unthickened, 1 uniformly thickened, 2 outer walls thickened 3. exodermal cell shape 0 isodiametric, 1 elongate 4. velamen cell thickenings 0 absent, 1 linear, 2 circular These characters derive primarily from the work of Porembski and Barthlott (1988); velamen and exodermal features were also discussed thoroughly by Pridgeon (1987). Instead of coding the ten Porembski and Barthlott velamen types as states for the terminals, which would necessarily have to be coded as unordered states of a single character, we chose to code component features of the variation that they describe. 5. spiranthosomes 0 absent, 1 present Spiranthosomes are specialized amyloplasts known only from the spiranthoid orchids and were first described by Stern et al. (1993a). Most of the occurrences of spiranthosomes are taken from Stern et al. (1993a, b). Many of the coded absences, outside of spiranthoid orchids, are assumed, since they have not been described before from other groups.

3 February 1999] FREUDENSTEIN AND RASMUSSEN CLADISTICS OF ORCHIDACEAE growth pattern 0 sympodial, 1 monopodial Most orchids follow the sympodial growth pattern common in monocots (Holttum, 1955). Notable exceptions are Vanilla, which is a monopodial vine with elongate internodes, and members of the Vandeae, which have monopodial growth. Pfitzer (1882) discussed this character at length. 7. thickened stem 0 absent, 1 present Stem thickening may be in the form of swelling in aerial stems, as in some species of Dendrobium, corms (e.g., Aplectrum), or pseudobulbs. If the stem is notably but uniformly thickened, as in Vanilla, the taxon is not scored as having a thickened stem. Pfitzer (1882) coded essentially the same information in his character stems homoblastic or heteroblastic, depending on whether the stem internodes are uniform or not, but there are some differences between our scoring and his. For example, Pfitzer called the stem of Catasetum typically homoblastic, while we recognize it as a thickened stem (the upper internodes are thickened more or less uniformly). 8. number of thickened internodes 0 several, 1 one This character codes variation in the number of internodes that comprise a pseudobulb. In many taxa, it is clearly one internode (e.g., Bulbophyllum), while in others (e.g., Catasetum), it can be several. Those taxa with no thickened stem are coded as unknown. Pfitzer (1887) also used this character. 9. phyllotaxy 0 spiral, 1 distichous Phyllotaxy in orchids is usually described as either spiral or distichous, with the latter supposed to be the more advanced state (Dressler and Dodson, 1960; Dressler, 1993). We found that in all species examined, the condition at the base of a stem, represented by the first bracts surrounding a bud, is distichous. In some taxa (e.g., Apostasia, Neuwiedia, many spiranthoids and diurids), the phyllotaxy soon shifts to more or less spiral, so that most of the leaves appear spirally arranged. In many of these, the arrangement is not strictly spiral, as was shown by Fuchs and Ziegenspeck (e.g., 1927a, b, c). In other taxa (e.g., species examined of Dactylorhiza, Orchis, many epidendroids), the majority of leaves show a distichous arrangement. Even in those species with distichous leaves, the inflorescence is often spiral in floral arrangement. In some cases, even the inflorescence is distichous (e.g., Tropidia effusa, Liparis gibbosa, Phalaenopsis cornu-cervi). Hence the basic model for orchid phyllotaxy appears to be a distichous basal condition on the shoot, with a possible shift to a spiral or pseudospiral condition at some point (before inflorescence or not). We scored states according to whether a taxon had spiral (or pseudospiral) or distichous phyllotaxy on the leaf-bearing portion of the shoot. 10. leaf morphology 0 flat non-plicate, 1 plicate, 2 conduplicate Variation among orchid leaves is more complex than it appears at first glance. Phyllotaxy plays a part in determining mature morphology, and since this was scored as a separate character, it is important not to duplicate this information. Leaves with convolute vernation usually are spirally arranged, while conduplicate leaves have duplicate vernation. The state found in many of the terrestrial spiranthoids and orchidoids has been termed by Dressler (1990c) nonplicate herbaceous, and is referred to here as flat nonplicate. It describes those leaves that are essentially flat (not conduplicate), without prominent midrib, usually fleshy in texture, and that are the result of convolute vernation. Plicate leaves may result from either convolute or duplicate vernation, and are characterized by the accordion-like pleating of the lamina. Conduplicate leaves, as coded here, are those with a single fold of the lamina at the midrib, and without any plication. There are some conduplicate plicate leaves (e.g., Sobralia), but, again, in order to avoid redundancy, these were simply scored as plicate. 11. winter leaf 0 absent, 1 present Some genera of terrestrial orchids have a leaf that is produced at the end of the growing season, overwinters, is photosynthetically active when conditions are favorable, then withers at or before flowering. These are primarily members of Corallorhizinae (e.g., Aplectrum, Calypso; cf. Freudenstein, 1994), but occasionally other species show a similar modification (e.g., some Spiranthes have overwintering leaves [F. Rasmussen, unpublished data]). 12. leaf articulation 0 absent, 1 present Pfitzer (1887) may have been the first to use this character. It has been discussed and the state noted for each orchid group by Dressler (1981) and denotes the presence or absence of an abscission zone at the base of a leaf. As a general rule, most epiphytic orchids are articulate, while terrestrial species are not. 13. stegmata 0 conical, 1 spherical, 2 absent Most of the information on this character is from Møller and Rasmussen (1984). Additional scoring was derived from sources quoted in Solereder and Meyer (1930) and from Stern et al. (1993b) for Spiranthoideae. The character seems to be largely uniform within genera, although Dressler and Cook (1988) found conical stegmata in Eria javanica while other members of the genus are known to have spherical stegmata. Since the outgroup (Hypoxis) does not have stegmata, the plesiomorphic state in the analysis is absence, although Apostasia and Neuwiedia, considered to be members of a basal lineage, do have conical stegmata. Because stegmata are always found in association with fibrous support tissue, absence may be due either to loss of the stegmata themselves (where sclerenchyma is present), or to the loss of sclerenchyma. Since presence of fibrous support tissue in the leaf is scored as a separate character, those taxa that have no leaf sclerenchyma were scored as unknown for the presence of stegmata. 14. leaf fiber bundles 0 present, 1 absent Many orchid leaves have sclerenchyma associated with the vascular bundles, as well as independent fiber bundles, while other leaves have no sclerenchyma. Stern et al. (1993b) used this character in their analysis of Spi-

4 228 AMERICAN JOURNAL OF BOTANY [Vol. 86 TABLE 1. Data matrix for the cladistic analysis. Taxa Hypoxis Apostasia Neuwiedia Cypripedium Paphiopedilum Epipactis Listera Cephalanthera Tropidia Goodyera Ludisia Zeuxine Spiranthes Manniella Gomphicis Altensteinia Ponthieva Cryptostylis Chloraea Thelymitra 1 Thelymitra 2 Diuris Prasophyllum 1 Prasophyllum 2 Caladenia 1 Caladenia 2 Acianthus Pterostylis 1 Pterostylis 2 Dactylorhiza 1 Dactylorhiza 2 Platanthera Disa Satyrium Disperis Arethusa Bletilla Calanthe Calopogon Eriodes Chysis Plocoglottis Sobralia Arpophyllum Vanilla Epistephium Lecanorchis Palmorchis Pogonia Triphora Nervilia Gastrodia elata Gastrodia sesamoides Epipogium Coelogyne 1 Coelogyne 2 Sirhookera Liparis nervosa Liparis viridiflora Cryptarrhena Calypso Eria javanica Eria Ceratostylis Appendicula Agrostophyllum Thelasis Glomera Characters ????000?0100??001?10001?0000????? ?00?????? 00???000? ? ????000? ? ???000? ? ?1200? ???000? ???000?1000? ? ???000? ? ? ? ???100?0000? ? ?0000? ????100?0000? ? ? ?0000? ????100?0000? ? ?20?????? 0?????00?0000?10? ?20?????? ?0000? ? ?0000? ???00? ?0000? ? ???000?0000? ? ???000??000? ? ???000??000? ? ???000??000? ? ???000??000? ? ???000??000? ? ???000??000?1?10? ? ???000??000?1?10? ? ???000??000? ?20?????? 10???000?0000? ?? ???000?0000? ?? ???000?1000?10? ???000?1000?10?0? ???000?1000? ?2001?001 10???000?0000? ? ???000??000? ? ????000??000?1010? ?? ? ? ? ???? ? ??000?? ?2??????? 01??? ? ? ? ?? ?21?????? 0120?000? ? ? ?21?????? ? ? ? 0????000? ? ?? ????000?1???????010100?10110?? ? ????000? ? ??? ?1000? ??0? ???000?1000? ?? ??? ? ? ????00?????????? ? ? ??? ????0??????????? ?? ??100????? 0????00??????????0? ? ?? ?? ? ???? ? ?? ? ???? ? ?? ? ? ? ?? ?21?????? ?1001? ? ? ?10??1??? ? ? ? ? ? ???? ?? ?1?? ? ?10? ? ?120100?? ?? ?00011?

5 February 1999] FREUDENSTEIN AND RASMUSSEN CLADISTICS OF ORCHIDACEAE 229 TABLE 1. Continued. Taxa Isochilus Epidendrum Cattleya Schomburgkia Meiracyllium Pleurothallis Dendrobium 1 Dendrobium 2 Bulbophyllum Sunipia Aplectrum Dactylostalix Govenia Tipularia Zygopetalum Xylobium Lycaste Maxillaria Dichaea Ornithocephalus Eulophia Telipogon Bromheadia Jensoa Cymbidium Genyorchis Thecostele Acriopsis Catasetum Stanhopea Oncidium Lockhartia Polystachya Neobenthamia Phalaenopsis Acampe Aerangis Angraecum Restrepia Characters ? ? ? ? ? ?? ? ? ? ? ???? ?? ?1?? ? ? ????000?1?00?1000? ?2? ? ? ? ? ?? ? ???000?1201? ?? ? ?120120? ?? ? ?120100?? ??001?? ? ???? ? ??001? ? ?10?111000?001?1??0101? ?20?????? ? ?210220? ? ? ? ? ? ????000? ? ? ?120110? ? ? ?1?? ?? ? ?? ?000? ? ?? ranthoideae. Presence of sclerenchyma is often, but not exclusively, associated with large leaf size. 15. leaf abaxial epidermal cells 0 straight, 1 wavy Lavarack (1971) used this character in his phenetic study of Australian Orchidaceae (Diurideae), and Stern et al. (1993b) surveyed it in the Spiranthoideae. The states are usually quite distinct, with either clearly sinuous or straight-polygonal anticlinal walls when viewed in paradermal section, (cf. Dressler, 1993, figs. 2 6). Adaxial surface cells may also exhibit the variation, but it is most pronounced on the abaxial surface. 16. subsidiary cells 0 present, 1 not distinguishable Some detailed studies of subsidiary cells in orchids have been done (Williams, 1975, 1979; Rasmussen, 1981, 1987), but not on a broad enough scale to make it possible to score the different developmental patterns for many of our terminals. Hence, we have simply scored the presence/absence of subsidiary cells that are morphologically distinct from surrounding epidermal cells, without regard to their ontogeny. 17. inflorescence position 0 terminal, 1 lateral In most orchids this appears to be a relatively straightforward character, although, as discussed by Bentham (1881) and shown by Andersen et al. (1988) for Eria, there may be variation within a genus and the true state can be difficult to ascertain. Pfitzer (1887) emphasized this character in his classification, delimiting the major groups Acranthae (those with terminal inflorescences) and Pleuranthae (those with lateral inflorescences). 18. floral abscission 0 absent, 1 present, ovary not stalked, 2 present, ovary stalked Many orchids have an abscission layer at the base of the pedicel; if the flower is not pollinated, it falls from the inflorescence. As opposed to perianth abscission, floral abscission seems to occur only in a more advanced group, the epidendroids. Within the Pleurothallidinae there is a further specialization, in that the abscission zone is located between the pedicel and ovary, so that

6 230 AMERICAN JOURNAL OF BOTANY [Vol. 86 after the flower falls a distinct stalk remains (Dressler, 1993; Neyland and Urbatsch, 1993). This character was coded as ordered (equivalent to coding the stalked ovary as a distinct character). 19. perianth abscission 0 present, 1 absent The perianth may abscise at the summit of the ovary. Dressler (1983) described this feature and noted its occurrence in the basal orchid groups. When it does not abscise, the perianth simply decays in place. 20. calyculus 0 absent, 1 vanilloid, 2 polystachyoid The calyculus is a series of small bract-like structures outside of the normal perianth. Because of its resemblance to a small perianth whorl, significant evolutionary implications sometimes have been ascribed to the structure such as Lindley s (1847) homologizing the calyculus with an additional floral whorl and consequent reinterpretation of the inner whorls. Soon thereafter, Crüger (1849) concluded that the calyculus is not an additional floral whorl, but Vermeulen (1966) again saw significance in the structure. Kurzweil (1987a) examined the development of these structures in Neobenthamia and found no relation to the perianth whorls. In this study two types of calyculus are recognized. Epistephium and Lecanorchis, and to a lesser extent, Vanilla, have a distinct collar below the perianth (Hashimoto, 1990; Dressler, 1993). The type known from Neobenthamia, and also observed here (and by Kurzweil, 1987a) in Polystachya, is different, being essentially a series of swellings on the ovary valves. A bract-like calyculus has also been reported from some species of Bulbophyllum (e.g., Lindley, 1838; Seidenfaden, 1979; Dressler, 1981). 21. slipper-shaped labellum 0 absent, 1 present This distinctive labellum form has the middle portion greatly expanded and the apex and distal margin pulled back toward the column, forming a hollow, shoe-like structure. This morphology is characteristic of the Cypripedioideae. Superficially similar structures occur occasionally in the Epidendroideae (e.g., Calypso), but do not include the inrolled margin that is present in cypripedioids. 22. apiculate perianth 0 absent, 1 present The perianth apices of Apostasia and Neuwiedia are prolonged into distinctive tips, the result of extension of the midrib. To our knowledge this feature has not been discussed before in a phylogenetic context, although it is visible in illustrations of the flowers (e.g., de Vogel, 1969). 23. carinate petals 0 present, 1 absent Rasmussen (1982; fig. 61d) illustrated an unusual feature in Vanilla interlocking sepals and petals. This condition results in a keel on the abaxial surface of the petal along the midrib that occurs in several putatively basal orchid groups. 24. lip-column marginal adnation 0 absent, 1 present In some vanilloid orchids the labellum is fused to the column marginally to varying degree. Other orchids have labellum column fusion of other types, but this is not included here. 25. dorsal median stamen 0 present, 1 absent 26. lateral inner stamens 0 present, 1 absent The homology of the functional stamens present in orchids has been discussed by Brown (1833), Lindley (1853), Darwin (1862), Swamy (1948), and Rao (1974). Further investigation was not undertaken here; we assume that all monandrous orchids have A 1, all diandrous orchids have a 1 and a 2, and triandrous orchids have A 1, a 1, and a 2. No evidence has ever been presented to substantiate Garay s (1960) claim that two lateral stamens of the outer whorl are present in Satyrium. Occasional unusual situations, such as the presence of a third functional stamen in Phragmipedium lindenii (Atwood, 1984) and the putatively peloric forms observed by Chen (1982) were not included. 27. anther orientation 0 erect, 1 bending late, 2 bending early Anther bending (incumbency) during ontogeny has been discussed by Hirmer (1920) and Dressler (1981, 1986b, 1993). Although there has been some question of whether advanced epidendroids bend at all (Dressler, 1981), more recent study suggests that these taxa bend very early (Dressler, 1986b; Kurzweil, 1987a). We distinguish between early and late bending following Kurzweil s (1987a, p. 438) stage 2 3 distinction, relative to the time of column elongation. This character is ordered based on ontogenetic information the erect state is more general than the incumbent state (cf. Kurzweil, 1987a). Degree of anther bending (illustrated by Hirmer, 1920) could also be scored, but it is essentially a continuous character, and so was not included here. 28. operculate anther 0 absent, 1 present Typical lilioid anthers dehisce by slits, releasing pollen without also shedding parts of the anther wall. Some orchid anthers (Apostasioideae, Cypripedioideae, Orchidoideae, Spiranthoideae) also have essentially this type of dehiscence. Incumbent anthers usually need to be physically disturbed to allow the pollinia to be removed. In some taxa with incumbent anthers, the anther develops in tight proximity to the clinandrium; in order for pollinia to be released, the anther (sometimes called the anther cap ) must be removed. This removable type of anther is called an operculate anther and has been used as a character since the beginning of orchid taxonomy by Swartz (1800), Lindley (1840), Pfitzer (1887), and Reichenbach (1852), as well as later authors such as Dressler (1981). 29. Endothecial thickenings 1 0 other, 1 intermediate, 2 type II 30. Endothecial thickenings 2 0 other, 1 type III/IV Endothecial thickening morphology in orchids was surveyed by Freudenstein (1991). Here the distinctive type II thickening is coded as one character, and the types III/IV and their intermediates are coded as a second char-

7 February 1999] FREUDENSTEIN AND RASMUSSEN CLADISTICS OF ORCHIDACEAE 231 acter. State 1 in character 29 is used for those thickenings that are intermediate in morphology between types I and II. Because it is a essentially a morphocline, the character is coded as ordered. 31. basal caudicles 0 absent, 1 present Caudicles are pollinium stalks that are composed of pollen and/or pollen-derived substances, as opposed to rostellar tissue (Richard, 1817; Mansfeld, 1934; Rasmussen, 1986a). In epidendroid and spiranthoid orchids they are produced apically in the anther, due to the bending of the anther or the apical position of the rostellum, respectively. In orchidoids the caudicles are basal extensions of the pollinia that are held in an erect anther. Pfitzer (1887) termed the condition where the rostellum is near the basal part of the anther basitonic (as opposed to acrotonic, in which the rostellum is near the anther apex), and considered it an important character. The diurids show intermediate states, sometimes termed pleurotonic (Mansfeld, 1954) or mesotonic (Dressler, 1981), but there is a continuum from acrotonic to basitonic, so this additional variation was not used here. 32. hammer stipe 0 absent, 1 present A distinctive stipe with a hammer-like morphology was identified by Rasmussen (1986a) in Sunipia. We found this type also in Genyorchis. 33. tegula 0 absent, 1 present Rasmussen (1982) distinguished between the tegula, a pollinium strap formed from the epidermis of the rostellum, and a hamulus, the apex of the rostellum itself. The tegula may be a multilayered epidermis or may consist solely of the rostellar cuticle, as in Doritis (Rasmussen, 1986a). 34. pollen unit 0 monad, 1 tetrad The variation in pollen unit at anther dehiscence has been a subject of study since Reichenbach (1852). It was described by Schill and Pfeiffer (1977) for a large number of species; others were reported in Newton and Williams (1978), Ackerman and Williams (1980, 1981), and Hesse, Burns-Balogh, and Wolff (1989). Ackerman and Williams (1981) described cases of some diurids (e.g., Caladenia) in which both states occur. Wolter and Schill (1986) suggested that the occurrence of tetrads in orchid pollen may be a paedomorphic transformation. 35. pollen tectum 0 reticulate, 1 smooth Pollen structure has been described by Williams and Broome (1976), Schill and Pfeiffer (1977), Burns-Balogh (1983), Hesse, Burns-Balogh, and Wolff (1989), Zavada (1990), and Schlag and Hesse (1993). Rather than focus on details of the tectal structure, which has not been studied in enough genera, we simply scored the appearance of the pollen grains, whether reticulate or smooth. This variation has been known since the time of Reichenbach (1852). 36. pollen apertures 0 colpate/sulcate, 1 porate, 2 inaperturate, 3 polyporate The aperture state of orchid pollen was taken from Erdtman (1944, 1952), Schill and Pfeiffer (1977), Newton and Williams (1978), Ackerman and Williams (1980), and Hesse, Burns-Balogh, and Wolff (1989). The majority of orchid pollen is inaperturate (Schill and Pfeiffer, 1977), but putatively basal groups have colpate/sulcate or porate pollen (Newton and Williams, 1978; Hesse, Burns- Balogh, and Wolff, 1989). Some of the vanilloids (Vanilla, Epistephium, Lecanorchis) have polyaperturate pollen (Erdtman, 1944, 1952; Schill and Pfeiffer, 1977; Ackerman and Williams, 1980). 37. operculate colpus 0 absent, 1 present Burns-Balogh and Funk (1986) used this character in their analysis. It was described and illustrated by Schill (1978) and Newton and Williams (1978) and only appears in Apostasia and Neuwiedia. 38. massulae 0 absent, 1 orchidoid, 2 epidendroid, 3 arethusoid Sectile pollinia (those with pollen subpackaged in massulae) were recognized very early in orchid taxonomy; the term massula was coined by Richard (1817). More recently, they have been shown to vary with respect to layering and regularity of massulae (Vermeulen, 1965; Freudenstein and Rasmussen, 1997). Orchidoid sectile pollinia have a single layer of uniform massulae, while epidendroid pollinia have variable numbers of layers of irregular massulae. Pollinia of Arethusa and Calopogon are hollow at maturity (cf. Pace, 1909; Freudenstein and Rasmussen, 1997), while most other orchids have solid pollinia. 39. pollinium texture 0 granular, 1 solid Pollinium texture is largely a continuous character (Dressler, 1986a), but it is possible to distinguish those pollinia that are truly coherent from those that are soft enough to be crushed easily when touched, and the character has been used to distinguish groups since Swartz (1800) and Richard (1817). The structural basis for this difference has been elucidated by ultrastructural studies of pollinia (Chardard, 1958, 1962, 1963, 1969; Cocucci and Jensen, 1969; Schill and Pfeiffer, 1977; Wolter and Schill, 1986; Yeung, 1987b; Hu and Yang, 1989; Zavada, 1990; Pandolfi, Pacini, and Calder, 1993). The most important difference is whether or not exine is deposited on internal pollen grains; if not, a more cohesive, calymmate pollinium results, while those that do have exine on all grains are much more friable and are termed acalymmate (van Campo and Guinet, 1961). 40. pollinium number: 2 0 absent, 1 present 41. pollinium number: 8 0 absent, 1 longitudinal, 2 transverse The primary pollinium numbers in orchids are 2, 4, and 8. Other numbers are sometimes reported (e.g., in Laeliinae), depending upon whether additional small masses of pollen that sometimes are found along the caudicles are interpreted as pollinia. Four is the predominant number. It is found in the putatively basal orchid groups, as well as in outgroups (where there are four anther locules). Freudenstein and Rasmussen (1996) found that there are at least two ways to produce eight pollinia by longitudinal or transverse division of embryonic pollen masses.

8 232 AMERICAN JOURNAL OF BOTANY [Vol pollinium orientation 0 juxtaposed, 1 superposed This character refers as much to the structure of the anther as it does to the pollinia. Richard (1817) used the Latin term superposita to describe the anthers of Calypso and Corallorhiza. Bentham (1881) also recognized the variation, although he did not use the term superposed. More recently, Dressler and Dodson (1960) recognized the superposed state, where, when four, the pairs of pollinia are stacked one on another, as opposed to the juxtaposed (Freudenstein and Rasmussen, 1996) condition, in which the pollinia are arranged side by side. We have detected two distinct ways in which the superposed state can occur either by inward or outward rotation of anther thecae (Freudenstein and Rasmussen, unpublished data), but as we have not yet completed developmental study on a sufficient number of taxa, we have here coded all superposed pollinia as the same state. 43. ovary locule number 0 one, 1 three Most orchids have a single ovary locule, while a very few putatively basal groups have three locules, as does the outgroup. Transverse sections of each type are shown in Atwood (1984) and, in diagrammatic form, in Garay (1960). 44. stigma 0 protruded, 1 sunken Variation in stigma morphology was discussed by Dressler (1981, 1993), Rasmussen (1982), and Dannenbaum, Wolter, and Schill (1989). In many taxa portions of one or more stigma lobes protrude, form a raised triangular or circular mass, or are of another shape (cf. Rasmussen, 1982, fig. 73). These morphologies are also sometimes known as convex. A sunken stigma is usually a circular depression ( concave ), which as Dressler (1993) has suggested, appears to be specialized to receive hard pollinia (cf. Rasmussen, 1982, fig. 73:2.1.2); this type is usually encountered in Epidendroideae. 45. stigma receptive cells 0 various other, 1 finger, 2 prosenchymatic This character derives from the work of Dannenbaum, Wolter, and Schill (1989). It describes the shape of the receptive cells of the stigma that are usually hidden under stigmatic secretions in living plants. 46. viscidium 0 none, 1 diffuse, 2 detachable Most orchids have some sort of adhesive associated with insect transfer of pollen masses. This may be either in the form of a glue that is transferred to the insect before it contacts pollen (diffuse), or a more elaborate cellular structure composed of rostellar tissue that is attached either directly, or via a stalk, to the pollinia (detachable). Some controversy over terminology involving the viscidium exists, with Dressler (1986a) and Dressler and Salazar (1991) arguing for restricting use of the term to a detachable structure, while Rasmussen (1982) used the term more broadly to include also any secreted adhesive. Here it is used in the broad sense simply for convenience, with the two senses of the term being the states. The character is coded as ordered because all detachable viscidia also have adhesive secretion. 47. endocarpic trichomes 0 absent, 1 present Endocarpic trichomes, or Schleuderhaare, were described in the 19th century and have been generally overlooked since. Blume (1848), Beer (1857), Prillieux (1857), and Horowitz ( ) illustrated and discussed them at some length, showing some of the variation in shape (filiform or flattened) that occurs. Pfitzer (1882) also mentioned them. Their function has always been assumed to be in seed dispersal, and according to Malguth (1901) their presence is correlated with epiphytism, although not all epiphytes have them and some terrestrial orchids do. More recently, Hallé (1986) illustrated them in transverse sections of ovaries of New Caledonian taxa. A scanning electron micrograph of a hair from Pteroceras appears in Pedersen (1993). The only genus in which we found them outside of Epidendroideae is Prasophyllum. 48. seed laterally compressed walls 0 absent, 1 present 49. seed testa cell shape 0 all isodiametric, 1 end isodiametric, middle elongate, 2 all elongate 50. seed striations 0 absent, 1 transverse/reticulate, 2 longitudinal 51. seed intercellular spaces 0 absent, 1 present 52. seed wax caps 0 absent, 1 present 53. seed covered cell border 0 absent, 1 present Seed morphology has provided a promising new set of characters for orchids (Barthlott, 1976; Dressler, 1986b, 1990a, 1993; Molvray and Kores, 1995). Most of these characters derive from the work of Ziegler (1981). Much of the scoring was done from the plates in Ziegler (1981), Tohda (1983, 1985, 1986), Chase and Pippen (1988, 1990), and Kurzweil (1993). Rather than coding seed morphology as types, such as those recognized by Ziegler (1981) and Dressler (1993), we coded component features to the extent possible. Laterally compressed testa walls refer to the extremely narrow cell lumen seen in some taxa (e.g., Vandeae), resulting from the close positioning of lateral anticlinal testa walls. Some seeds have distinct variation in testa cell size depending on location with either all cells isodiametric, the cells at both ends isodiametric and those in the center elongate, or all cells elongate. When present, striations on the sunken testa cell lumina may be either transverse/reticulate or longitudinal/parallel. Distinct spaces occur among the cells in some taxa. Wax caps are present at the ends of testa cell protrusions in some members of Cymbidieae (Ziegler, 1981; Chase and Pippen, 1990). In some seeds the abutment of adjacent testa cell walls is clearly evident, while in others the line of demarcation between them is covered by tissue from one or the other cell (a covered cell border). Excluded characters A number of additional characters have been used in previous analyses, are commonly used in keys to separate major groups, or at least have been suggested to be of phylogenetic significance. Here we describe those characters and the reasons for their exclusion. In addition to those listed here, there are many characters that essentially reflect shape differences that are clearly continuous or problematic in scoring. This cat-

9 February 1999] FREUDENSTEIN AND RASMUSSEN CLADISTICS OF ORCHIDACEAE 233 egory includes some of the characters employed by Burns-Balogh and Funk (1986), such as degree of style fusion and rostellum shape. Vessel perforation plate type Cheadle and Kosakai (1982), Stern, Cheadle, and Thorsch (1993), and Thorsch and Stern (1997) described the variation in perforation plates in the Orchidaceae. While most taxa have scalariform plates, Apostasia and Neuwiedia have a significant proportion of simple perforation plates. The problem is with state delimitation. An individual plant may have both simple and scalariform plates, meaning that simple perforation plates are a tendency in apostasiads, rather than a true defining feature. Pollen tetrad shape Reichenbach (1852) illustrated several tetrad cell arrangements. Konta and Tsuji (1982) and Konta and Hayakawa (1982) recognized up to six shapes of pollen tetrads, but found that all species had more than one type of tetrad, with some having all of the recognized types. Yeung (1987a) also discussed this variation. The rampant intra-individual polymorphism precluded use of the character. Resting nucleus type Tanaka (1971) has recognized five morphological classes of interphase chromatin. He found the states to be largely distinct, with few intermediates. Additional information has been provided by Okada (1988), who recognized only three states, and Sera (1990). The character was not included here because there are still a significant number of missing data. Sclerotic seed coat Significance has been attributed to the fact that both Apostasia and Vanilla (as well as Selenipedium, which was not a terminal in this analysis) have a hard, black sclerotic seed (Swamy, 1947). Garay (1986) interpreted this state as a symplesiomorphy. Rübsamen (1986) and Nishimura and Tamura (1993) showed that in Apostasia it is the inner layer of the outer integument that becomes sclerotic, while in Vanilla it is the outer layer of outer integument (Swamy, 1947; Krupko, Israelstam, and Martinovic, 1954), indicating that the condition is not homologous in these taxa. Number of vascular traces in the pedicel Swamy (1948) and Lavarack (1971) discussed this character, indicating that cypripedioids and apostasioids have six vascular traces that enter the flower, while the monandrous orchids have three. Working with fresh material we found it difficult to ascertain the number. Observations made from serial sections showed three traces in Habenaria tridactyle, but six in Thelasis pygmaea, and apparently six in Appendicula hexandra. Without a careful study of this character, we chose to omit it. Resupination Twisting of the flower, resulting in the labellum being lowermost, is characteristic of many orchid groups. It may be achieved in either of two ways. The pedicel may be curved (but not twisted) and directed to the opposite side of the raceme, so that the subtending bract is actually at the top of the flower rather than below. This is very easily achieved in inflorescences with only one flower (many Paphiopedilum, various others), but can also be accomplished in racemes such as those of Spiranthes. Alternatively, the pedicel may be twisted 180. In this case the subtending bract is lowermost and the flower faces the same side of the raceme. No solitaryflowered species observed here showed this type of resupination. Some taxa (e.g., Malaxis monophylla var. monophylla) have the pedicel twisted 360, which gives a nonresupinate orientation. Arching inflorescences provide another complication in this feature since the inflorescence axis is no longer vertical; in order to maintain an orientation with the labellum lowermost, 180 of twisting may not be necessary. Dressler (1983) imagined a flower twisting away from its subtending bract as a precursor to resupination. This feature has not been used traditionally in classification, but is a useful character in keys. It is often polymorphic in genera and can be in species (e.g., Malaxis monophylla). Column foot This structure is a thickening at the base of the column where the labellum is attached. Whether it is actually column or labellum tissue that is thickened is often not clear. Bentham (1881) and Schlechter (1926) used the character particularly within Vandeae. Garay (1972b) discussed the ambiguity of this character and avoided it in his scheme for Vandeae. It is polymorphic within genera and difficult to homologize. Embryogeny The form of the suspensor in orchid embryos is a promising systematic character. Its variation has been described in most detail by Treub (1879), Swamy (1949), and Veyret (1965), and it is possible to describe character states (e.g., the suspensor types of Swamy [1949]). Unfortunately, there remains a majority of taxa included here for which no information on embryogeny is available. Auricles A number of protrusions associated with the anther and column of monandrous orchids are known. These are often considered to be staminodes, but it is difficult to compare them without developmental information. The type seen in Orchideae, termed auricles, have been considered a defining character for the group (e.g., Dressler, 1981). Vermeulen (1966) believed that they are nonstaminodal. Kurzweil (1987b) examined these structures developmentally and also concluded that they do not represent staminodes, but that they are appendages of the fertile anther. He observed other structures, which he called basal bulges, that may represent staminodes. The auricles and basal bulges fuse late in development with the lateral margins of the median stigma lobe, making them indistinguishable in the mature flower (Kurzweil, 1987b). Further complexity was encountered in the South African Orchideae (Kurzweil, 1991). Because of the difficulty in distinguishing these structures and establishing their homology, we did not include the character. Cohesion strands Burns-Balogh and Funk (1986) employed a character called cohesion strands, which they found only in neottioid and spiranthoid groups. They described them as acetolysis-resistant strands that appear only when pollinia are pulled apart, suggesting that they are different from elastoviscin, which is found in many