HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by Krosnick, S. E. Articles by Freudenstein, J. V. Search for Related Content PubMed Articles by Krosnick, S. E. Articles by Freudenstein, J. V. Agricola Articles by Krosnick, S. E. Articles by Freudenstein, J. V.

Patterns of anomalous floral development in the Asian Passiflora (subgenus Decaloba: supersection Disemma)¹

Department of Evolution, Ecology, and Organismal Biology, Ohio State University Herbarium, 1315 Kinnear Road, Columbus, Ohio 43212 USA

Received for publication June 3, 2005. Accepted for publication January 10, 2006.

ABSTRACT

Approximately 22 species of Passiflora are native to the Old World. All of these species are placed in subgenus Decaloba, supersection Disemma. Within Disemma, three species vary in stamen and carpel number ( eight stamens and five carpels). The mode of development was determined for two of the anomalous species, P. moluccana var. glaberrima and P. siamica. Ontogenetic patterns were compared to normal development in P. perakensis and P. holosericea. Passiflora siamica develops additional stamens through dédoublement of a single widened stamen primordium, while P. moluccana var. glaberrima exhibits congenital dédoublement where stamens emerge already doubled. Phylogenetic analysis using ITS and the trnL-F intron and spacer resolve the anomalous species as monophyletic and sister to P. perakensis. This signifies a single loss of genetic regulation in stamen and carpel number within Disemma. Floral whorls were examined across the Passifloraceae, Malesherbiaceae, Turneraceae, and Flacourtiaceae s.l.. Similar doubling in these families suggests that this Eurosid lineage may have a genetic propensity for variability in floral whorl number.

Key Words: anomalous • bifurcation • dédoublement • Disemma • floral development • Passiflora

Assessing homology of morphological features in families that contain large genera is often difficult for several reasons. Such families are frequently comprised of one or two large genera and several smaller genera. In the larger genera, rapid species radiation may have occurred, often resulting in diverse morphologies that are difficult to interpret across the rest of the family. The situation may be complicated when a well-supported familial phylogeny is unavailable, possibly due to limited sampling in the smaller genera or from a lack of phylogenetic signal in rapidly radiating lineages. A primary example of this situation is the Passifloraceae. This family consists of approximately 17 pantropical genera, the largest of which are the primarily New World Passiflora L. and Old World Adenia Forsk. Both of these large genera have floral and vegetative morphologies that are difficult to relate to the smaller, often monotypic genera in the rest of the family. Passifloraceae has been traditionally divided into two tribes, Passifloreae and Paropsieae. Passifloreae is comprised of approximately 12 genera that are primarily climbing plants with tendrils (MacDougal, 1994 ). Paropsieae primarily consists of shrubs and small trees and has been viewed as transitional between Malesherbiaceae, Turneraceae, and Flacourtiaceae (De Wilde, 1971 ; MacDougal, 1994 ; Bernhard, 1999 ).

The taxonomic circumscription of Paropsieae has been difficult because many of its characters suggest conflicting familial affinities. While stamen number does not exceed 10 in the tribe Passifloreae, it may range from five to 20 or more in Paropsieae, especially in the genera Viridivia and Smeathmannia (De Wilde, 1971 ). Very little is known about the primitive floral condition in Passifloraceae. Bernhard (1999) examined floral development in members of both Passifloreae and Paropsieae to determine the primitive androecial condition in the family and to address questions of homology among the corona and disk structures observed across the genera. The rarity of many of the smaller genera in Passifloraceae restricted sampling in that analysis, thus limiting conclusions about ancestral floral character states. Recent molecular phylogenetic analyses by Alford (2005) on the Flacourtiaceae sensu lato suggest that the monophyletic tribe Paropsieae is sister to the remainder of the Passifloraceae. However, a well-supported phylogeny at the familial level is still unavailable, leaving many unanswered questions about the evolution of this group.

Passiflora, the largest genus in the family, contains approximately 490 species of vines, lianas, and small trees native to Central and South America, with approximately 22 additional species endemic to Southeast Asia and the Austral Pacific. The genus is well-known for the complex and diverse morphologies observed in both its vegetative and floral organs, but few of these features have been studied in an evolutionary context. All species have axillary tendrils, and many species are distinguished by the presence and type of laminar and petiolar nectaries. Passiflora flowers are characterized by an androgynophore, one to several series of coronal filaments, five stamens, and three carpels (Ulmer and MacDougal, 2004 ). There is a high degree of variation in floral morphology across the genus, even within small groups of closely related species (MacDougal, 1994 ; Kay, 2003 ; Porter-Utley, 2003 ). The morphological details of the hypanthium, coronal filaments, operculum, and limen are extremely variable in Passiflora. For this reason, it has been difficult to assess the homology of these organs within the genus and even more difficult to compare these features to other genera in Passifloraceae and allied families.

Relatively few studies have addressed the developmental patterns leading to the complex floral morphology of Passiflora.Payer (1857) was the first to examine the development of the flower in the hybrid Passiflora x loudonii.Masters (1871) provided the most extensive treatment of development and homology in the genus. This work was the first of its kind to identify and define the morphological features unique to Passiflora. There has since been little investigation into the nature of these features. The ontogenetic sequence of floral development in the economically important Passiflora edulis Sims. and P. quadrangularis L. was characterized by Moncur (1988) , but these are the only complete ontogenetic sequences available for the genus to date. Bernhard (1994) investigated the developmental pattern of the corona and operculum in several species of Passiflora and Adenia. Most recently, Bernhard (1999) examined the development of the androecium, corona, operculum, and floral disk in P. racemosa to assess homologies of these structures among the genera of Passifloraceae.

Within Passiflora, the number of reproductive structures is generally uniform, with most of the ca. 520 species consistently exhibiting five stamens and three carpels united on an androgynophore. However, three Asian taxa, Passiflora tonkinensis W.J. De Wilde, P. moluccana var. glaberrima (Gagnep.) W.J. De Wilde, and P. siamica Craib have significant variability in these features. These taxa have between five and eight stamens and three to five carpels. In P. siamica, the number of stamens ranges between five and seven, and the number of carpels may be either three or four. In P. moluccana var. glaberrima, the number of stamens is either seven or eight, and the number of carpels either four or five. Passiflora tonkinensis has been documented with seven stamens and between three and four carpels (De Wilde, 1972 ). All other morphological features in these taxa are similar to the general conditions observed across the rest of the genus.

In previous hypotheses of relationships in Passiflora, the 17 species endemic to Asia were thought to be sister to the remainder of the genus (Cusset, 1968 ; Tillett, 1988 ). This was supported by two characters observed in some of the Asian taxa: the presence of branched inflorescences and the insertion of additional stamens and carpels in certain species. Branched inflorescences are common in Passifloraceae, but this feature has been greatly reduced in Passiflora, suggesting that this might be a plesiomorphic state in those species retaining this feature. The second hypothesized ancestral feature, the presence of additional stamens and carpels, is similar to the situation in the dioecious genus Adenia Forssk. and in the hermaphroditic Mitostemma Mast., Dilkea Mast., and Ancistrothyrsus Harms. Based on these similarities, increased stamen number was thought to be a plesiomorphic character state in the Asian species.

Krosnick and Freudenstein (2005) showed that the 22 species of Old World Passiflora are a monophyletic group and supported their current placement in subgenus Decaloba (DC.) Rchb. supersection Disemma (Labill.) J.M. MacDougal & Feuillet. Disemma was shown to be nested within subgenus Decaloba, refuting previous hypotheses that placed Disemma as sister to the rest of the genus. Additionally, the morphologically similar supersection Multiflora (Small) J.M. MacDougal & Feuillet was shown to be sister to supersection Disemma. Within supersection Disemma, two monophyletic groups were resolved: an Asian clade spanning India, China, and Southeast Asia, and an Austral-Pacific clade. While it is now clear that the Old World Passiflora are not sister to the rest of the genus, their anomalous morphology raises several new issues. De Wilde (1974) hypothesized that the ancestor of Passifloraceae was diplostemonous or possibly even triplostemonous. According to his hypothesis, the five stamens would represent the innermost whorl, while the outer whorls have been reduced and transformed into the limen that protects the nectary at the base of the flower. De Wilde proposed that the Asian species retained some members of these outer whorls, resulting in six to eight functional stamens. Bernhard (1999) refuted this hypothesis based on both morphological and histological differences between the secretory cells characteristic of the limen and the true staminodial tissue in the androgynophore. Moreover, the delayed initiation of the limen until long after the differentiation of the stamens suggests that these structures are not likely to be homologous.

Thus, at present there is no convincing explanation for the presence of additional stamens and carpels in these anomalous species. It is possible that the condition observed in the Asian species is a synapomorphy for the anomalous taxa, resulting from a unique ontogenetic pathway unrelated to the polyandry characteristic of other genera in Passifloraceae. Alternatively, the Asian species may represent a reversal to the plesiomorphic polyandrous condition for the family. Therefore, the aims of this study were to (1) elucidate the generalized mode of floral development in Passiflora by examining supersection Multiflora, the putative sister clade to the Old World species, (2) compare the generalized pattern for Passiflora to normal Asian species that have five stamens and three carpels, (3) characterize the pattern of floral development in the anomalous Asian species, and (4) examine the implications of development in the Asian species for our understanding of the plesiomorphic developmental sequence for Passifloraceae and related families.

MATERIALS AND METHODS

Phylogenetic analysis
Taxon sampling and outgroup selection
Representatives of eight supersections of subgenus Decaloba were included (see Appendix). In order to address relationships within the anomalous Asian taxa more thoroughly, multiple accessions of the same species were utilized where possible. Two accessions of P. moluccana var. teysmanniana from Hainan, China, and two different collections of P. siamica from Yunnan, China, were used in this analysis. Passiflora cupiformis (GuangXi Province, China), P. henryi (Yunnan Province), and P. perakensis (Thailand, Malaysia) were included in the phylogeny as Asian representatives with normal floral development. Voucher specimens, their taxonomic placement, and GenBank accession numbers are listed in the Appendix. The combined data set has been deposited in TreeBASE (study accession number: SN2352, matrix accession number: M8692).

DNA extraction and purification
Fresh or silica-gel-preserved leaf material was used for DNA extraction. Total genomic DNA was extracted using the CTAB method of Doyle and Doyle (1987) . For some species, it was necessary to further purify and clean the DNA using Elu-Quik (Schleicher and Schuell, Keene, New Hampshire, USA) to achieve strong amplification.

Amplification and sequencing
The entire ITS region, including ITS-1, the 5.8S gene, and ITS-2 was amplified using primers 5 and 4 of White et al. (1990) . Amplification reactions of 50 µL contained 1 µL DMSO, 40 µL HPLC water, 1 µL of each 10 µM primer, 0.5 µL Taq, 0.5 µL of 0.2 µM dNTPs, 5 µL10x buffer (100 mM Tris-HCl pH 8.8, 35 mM MgCl₂, 250 mM KCl), and 1 µL DNA. The amplification program for the ITS region was a single initial denaturation of 95°C for 5 min, followed by 35 cycles of 95°C for 1 min, 50°C for 2 min and 72°C for 2 min, followed by a final 7 min extension at 72°C. The trnL-F region of chloroplast DNA was amplified with primers c and f from Taberlet et al. (1991) . The trnL-F amplification reactions contained 40 µL HPLC water, 1 µL of each of 10 µM primer, 5 µL 10x buffer (100 mM Tris-HCl pH 8.8, 35 mM MgCl₂, 250 mM KCl), 0.5 µL of 0.20 µM dNTPs, 0.5 µL Taq polymerase, and 0.5 µL of 10 µg/µL of bovine serum albumen. The amplification program for trnL-F was a single initial cycle of 96°C for 3 min, followed by 30 cycles of 96°C for 45 s, 64°C for 45 s, and 72°C for two min, followed by a final 5 min extension at 72°C. Amplifications were purified using 50 µL of 20% polyethylene glycol–2.5 M NaCl and sequenced directly. Dideoxy cycle sequencing reactions were performed using BigDye Terminator version 3.1 chemistry (Applied Biosystems, Foster City, California, USA) scaled down to quarter reaction volume. Following ethanol purification, reactions were run on an ABI Prism 3100 automated sequencer (PE Biosystems, Foster City, California, USA). Contigs were assembled using Sequencher 3.1.1 (Gene Codes, Ann Arbor, Michigan, USA).

Phylogenetic analyses
All ITS and trnL-F sequences were initially aligned using Clustal W (Thompson et al., 1994 ) and then adjusted manually using Se-Al (Rambaut, 2000). Matrices were analyzed using PAUP* version 4.0b.10 (Swofford, 2002 ). Initial heuristic searches on the ITS, trnL-F, and combined data sets were performed with the following parameters: 2000 random addition sequences, holding two trees at each step during stepwise addition, followed by tree-bisection-reconnection (TBR) branch swapping with Multrees in effect, each TBR replicate limited to saving two trees, and swapping on best trees only. Trees saved from the initial search were then used as the starting topologies for a second round of extensive branch swapping with Multrees in effect, swapping on the best trees only, but with no limits on the number of trees saved. A total evidence approach (Kluge, 1989 ; Baum, 1992 ; Kluge and Wolf, 1993 ; Nixon and Carpenter, 1996 ) was utilized to permit full interaction of all characters, to allow secondary signal hidden within the data sets to be revealed, and to provide a maximally parsimonious solution.

Branch support
Branch support was assessed using 10 000 jackknife replicates in PAUP*, resampling at 37%, and using the "emulate Jac" command. The heuristic searches utilized two random addition sequences per replication, saving only two trees per random addition sequence, with Multrees in effect, and swapping on best trees only. Only clades with a frequency of 50% or higher were retained in the jackknife consensus tree.

Floral development
Floral buds at various developmental stages of P. siamica, P. moluccana var. glaberrima, P. perakensis Hallier f., and P. holosericea L. were preserved in 70% ethanol if collected in the field or in FAA (1 part formalin, 1 part glacial acetic acid, 18 parts 70% ethanol) if collected in the greenhouse (see Appendix). Buds were taken through a dehydration series to 95% ethanol, stained with safranin, and dissected in the laboratory with a Wild EpiMakroskop M450 compound microscope (Leica Microsystems, Illinois, USA). Dissected buds were dehydrated further in absolute ethanol, critical-point-dried with CO₂ in a Pelco CPD2 critical point drier (T. Pella, Redding, California, USA), and mounted on aluminum stubs with carbon conductive adhesive tabs (T. Pella, Redding, California, USA). Specimens were then coated with gold-palladium using a Cressington Model 108 sputter coater (Cressington Scientific Instruments Inc., Pennsylvania, USA), and micrographs were taken at 15 kV on a Phillips XL30 scanning electron microscope (SEM; FEI Company, Oregon, USA) in the Campus Microscopy and Imaging Facility, Ohio State University, Columbus, Ohio, USA.

RESULTS

Phylogenetic analyses
Sequence characteristics
The ITS alignment comprised 748 positions, of which 219 were parsimony informative characters. Sequence divergence was 29.5% (adjusted for missing data) between Disemma (using P. cupiformis as a representative) and the closest outgroup species, P. multiflora. Among the species of Disemma, sequence divergence ranged from 1.4–37.3%. The ITS sequences of the two accessions of P. moluccana var. teysmanniana (nos. 198, 326) were identical to one another, as were the two accessions of P. siamica (nos. 5, 346). The trnL-F alignment comprised 912 nucleotides, of which 129 were parsimony informative. Sequence divergence was 25.9% (adjusted for missing data) between Disemma (P. cupiformis) and P. multiflora. Sequence divergence ranged from 13.0–36.2% between species of Disemma. Sequence divergence between the two accessions of P. moluccana var. teysmanniana was 15.6% and was 12.0% between the accessions of P. siamica.

Relationships
For the ITS data set, tree searches based on 219 informative characters produced six equally parsimonious trees of 632 steps (CI = 0.52, RI = 0.66). The strict consensus for the ITS data set retains full resolution of relationships in the anomalous Asian taxa (Fig. 1A). For the trnL-F data set, tree searches based on 129 informative characters produced 133 equally parsimonious trees of 386 steps (CI = 0.45, RI = 0.50; tree not shown). The strict consensus of the trnL-F data set does not provide resolution of the relationships within Disemma. The combined analysis consisted of 348 informative characters and produced four equally parsimonious trees of 1049 steps (CI = 0.48, RI = 0.59; Fig. 1B). The combined data support the monophyly of supersection Disemma with a jackknife value of 60%. Passiflora multiflora is sister to supersection Disemma, although this relationship is not strongly supported in the jackknife analysis. Within subgenus Decaloba, supersections Hahniopathanthus, Bryonoides, Cieca, and Decaloba are highly supported as monophyletic with jackknife values of 100. However, supersection Multiflora is paraphyletic; P. monadelpha is resolved as sister to P. auriculata (supersection Auriculata) instead of P. multiflora, though the support for this clade is not very high (75%).

View larger version (29K): [in this window] [in a new window]   Fig. 1. Phylogenetic analysis of Passiflora supersection Disemma using ITS and trnL-F sequence data. (A) Partial representation of strict consensus of six most parsimonious trees using only ITS data, focusing on the relationships within the Asian species of supersection Disemma. (B) Strict consensus of four most parsimonious trees based on combined ITS and trnL-F sequence data. Numbers above the branches indicate jackknife consensus values of 50% or higher for 10 000 replicates

Developmental analyses
Passiflora holosericea (Figs. 2–10)
The calyx originates in a spiral sequence (Fig. 2), ultimately resulting in imbricate aestivation. When the two smallest sepal primordia have elongated to enclose the floral apex, the petal and stamen primordia arise almost simultaneously, with the five stamen primordia alternating with the petals (Fig. 3). The petal and stamen primordia elongate further; at first, the petal primordia slightly surpass the stamen primordia (Fig. 4), and then become equal in length (Fig. 5). At this point, the stamen primordial zone is evident in lateral view (Fig. 6, arrows). Here the stamen primordia are basally united to an equal degree for all members. When the stamen primordia almost enclose the center of the floral apex, three carpel primordia begin to emerge (Fig. 7). The carpel primordia initiate spirally, resulting in a size gradient among them (Figs. 7, 8). In later development, zonate growth of the gynoecium below the carpel apices results in the formation of a single, congenitally fused style (Fig. 9). At the latest stage observed, the style has become thickened and differentiated into a distinct appendage from the ovary, the carpel apices have broadened into recognizable stigmatic surfaces, and the coronal filaments are beginning to emerge (coronal filaments indicated by arrow; Fig. 10).

View larger version (189K): [in this window] [in a new window]   Figs. 2–10. Passiflora holosericea.2. Floral apex with early petal formation indicated by arrows. Three sepals removed. 3. Floral apex with stamens emerging. 4. Apex with larger stamen primordia, petals are elongating. 5. Stamens surpassing the petals in length. 6. Arrows indicate the similar degree of lobing between stamen primordia. 7. Three carpel primordia begin to emerge from the center of the apex; stamen primordia begin to differentiate into anthers and filaments. 8. Further development of the carpel primordia, where the spiral initiation is evident in the different height of each carpel. 9. Early ovary formation coincides with anther maturation, carpels are still of different heights. 10. Style primordia begin to mature, anthers are well developed. Arrow indicates coronal filaments beginning to emerge. Bars = 100 µm. Figure abbreviations: a, anther; c, carpel; f, stamen filament; p, petal; s, sepal; st, stamen; st*, doubled stamen; sg, stigma

View larger version (187K): [in this window] [in a new window]   Figs. 11–19. Passiflora perakensis.Figs. 11–13, polar view of floral apex; Figs. 15–-19, lateral view of floral apex. 11. Floral apex with early development of petals and stamen primordia. 12. Stamen primordia emerging from the apex, with two petals removed. 13. Stamen primordia enlarging further. 14. Stamen primordia surpassing the petals in length, one petal removed. 15. View of developing stamen primordia. Arrow indicates the depth of the division between the individual stamen primordia. One petal removed. 16. Spiral emergence of the carpels from the floral apex, corresponding to the start of differentiation of anther and filament in stamen primordia. Two petals and three stamen primordia removed. 17. Carpel primordia, one anther removed. 18. The beginning of differentiation of the carpel primordia into an ovary, coinciding with differentiation of the anther and stamen filaments. Arrow indicates coronal filaments beginning to emerge. 19. Style primordia widening and elongating from the apices of the carpels. Bars: Figs. 10–13, 15–16 = 100 µm; Figs. 14, 17 = 200 µm

View larger version (128K): [in this window] [in a new window]   Figs. 20–25. Passiflora siamica in polar view, development in normal floral apices. 20. Early stage of the floral apex with only petal primordia present. 21. Stamen primordia beginning to emerge from the floral apex while the petal primordia enlarge. 22. Enlargement of stamen and petal primordia. 23. The petal and stamen primordia are of approximately the same size. One petal removed. 24. The carpel primordia begin to emerge from the center of the apex, while the stamens are longer than the petals. Two petals removed. 25. Further development of the carpel primordia, coinciding with the differentiation of the anther and filament from the stamen primordium. Four stamens removed. Bars: 100 µm

View larger version (200K): [in this window] [in a new window]   Figs. 26–37. Passiflora siamica, development in anomalous floral apices.Figs. 26–28, 31, 33, 35–36, polar views of; Figs. 29–30, 32, 34, 37, lateral views. Figs. 26–33 are six-stamen and three-carpel forms; Figs. 33–34 are six-stamen and four-carpel forms. Figs. 35–37 are seven-stamen forms. Fig. 36 is a seven-stamen and three-carpel form. 26. Early floral apex with anomalous stamen development. Two stamen primordia have widened (st*), while three stamen primordia remain the normal size. This is an equivalent stage to the normal P. siamica in Figs. 22–23. One petal removed. 27. A later stage with the dédoublement of the stamen primordium into two distinct units. Two petals removed. 28. Further development of the doubled stamens. Two petals and two stamens removed. 29. Side view of the doubled stamen primordia in Fig. 28. 30. The depth of the division between the doubled units is much less than in normal forms. 31. Carpel primordia development where two carpels are alternate with the stamen primordia and one is opposite. Three stamens removed. 32. In later stages, it is difficult to determine which stamens have been doubled. Anthers begin to differentiate on the stamen primordia. 33. In the six-stamen form, two carpel primordia are alternate and two are opposite with the stamen primordia. Three stamens removed. 34. Later ontogenetic stage of the six stamen form, with fully differentiated anthers and staminal filaments. Stylar primordia are beginning to thicken and elongate from the ovary. 35. An early stage of the seven-stamen form where it is difficult to determine which stamens have been doubled. One petal removed. 36. Later stage in the seven-stamen form where three carpels develop. 37. A lateral view of a similar stage in the seven stamen form. Arrows point to the difference in the depth of the lobing between the doubled primordium and the normal stamen side by side. Bars: 25–33 = 100 µm, 34–36 = 200 µm

View larger version (37K): [in this window] [in a new window]   Fig. 56. Generalized ontogenetic pathways for Passiflora siamica and P. moluccana var. glaberrima. Doubled stamen primordia are illustrated as occurring in the interspace created by the first sepal primordium; if more than one stamen was doubled, the primordia were oriented in the gaps created by sepals one, two, and three, respectively. (A) P. siamica. The undifferentiated floral apex may follow one of three pathways: normal development (left), a single stamen primordium doubles (center), or two primordia double (right). In the six stamen form (center), three or four carpels may develop. In the seven stamen form, only three carpels were observed. (B) P. moluccana var. glaberrima. The undifferentiated floral apex may follow either of two pathways: three primordia may double and two normal stamens emerge (left), or two primordia may double and three normal stamens emerge (right). Carpel number ranges between four and five and may be observed in all combinations with stamen number

View larger version (130K): [in this window] [in a new window]   Figs. 38–43. Passiflora moluccana var. glaberrima at early stages of development in the floral apex. Figs. 38–40, 42–43 polar views; Fig. 41, tangential view. Figs. 39, 42–43, eight-stamen forms, Figs. 40 and 41, seven-stamen form. 38. Earliest stage observed where the petal primordia are emerging but the stamen primordia have not yet emerged. 39. Doubled stamens are in pairs, separated by two nondoubled stamens. The doubled stamens are on either side of and opposite to the isolated petal. Two petals removed. 40. Two stamens are doubled, located opposite of the isolated petal. Four petals removed. 41. Tangential view of Fig. 40, where arrows indicate differences in depth of division between doubled (left) and nondoubled (right) stamen primordia. 42. Later stage of the form in Fig. 39; the doubled stamens are located on both sides of and opposite of the isolated petal. 43. Later stage of the eight-stamen form; the doubled primordia can no longer be detected. The zone at the center of the apex becomes convex, leading to the emergence of carpel primordia. Two petals removed. Bars = 100 µm

View larger version (192K): [in this window] [in a new window]   Figs. 44–55. Passiflora moluccana var. glaberrima.Figs. 44–47, 49, 54 represent polar views of anomalous growth; Figs. 48, 50–53, 55, lateral views. Figs. 44, 46, 48, eight-stamen and five-carpel forms; Figs. 45, 52, 55, eight-stamen and four carpel forms; Figs. 47, 53–54, seven-stamen and five-carpel forms; Fig. 49, seven-stamen and four-carpel form. 44. Five emerging carpel primordia, four stamen primordia removed. 45. Four carpel primordia at the center of the floral apex. 46. Further developed carpels emerging from the floral apex, five stamen primordia removed. 47. Carpel primordia enclosed by seven stamen primordia, two stamens removed. 48. A later stage of Fig. 45, with four carpel primordia emerging from the floral apex in an eight-stamen form, four stamens removed. 49. Four carpels emerging at the center of a seven-stamen form, the carpels are neither fully alternate or opposite with the stamens, four stamens removed. 50. Lateral view of the stamen primordia; the depth of the division between each primordium is approximately equal. 51. Closer view of Fig. 50; arrows point to the equivalent depth of the divisions between the stamen primordia. 52. Maturing gynoecium; ovary is beginning to differentiate, stylar primordia are lengthening from the carpel apices. All anthers removed. 53. Lateral view of later stage of the seven-stamen and five-carpel form. The five stylar primordia are beginning to differentiate from the carpels, with two paired and one lone style primordium. The anther is beginning to differentiate from the filament. Two stamens removed. 54. Polar view of Fig. 53. All members of the androecium appear normal in size and shape. The two pairs of style primordia form two flattened rows of styles. Two stamens removed. 55. Mature eight-stamen and four-carpel form. The anthers are fully differentiated from the staminal filaments. The ovary is rounded and the stylar primordia are elongated and becoming thickened. Three anthers removed. Bars: Fig. 44 = 100 µm; Figs. 45–55 = 200 µm

Phylogenetic relationships within the Asian Passiflora
In the most recent treatment of the Old World taxa, De Wilde (1972) defined Passiflora moluccana with three varieties: P. moluccana var. moluccana, native to the Moluccan Islands of Indonesia, P. moluccana var. teysmanniana, found in Guangdong and Hainan Provinces of China, and P. moluccana var. glaberrima, endemic to Yunnan Province, China. These three varieties are easily distinguished from one another based on highly divergent floral and vegetative morphologies and noncontiguous geographical distributions. Passiflora moluccana var. moluccana and P. moluccana var. teysmanniana both have normal numbers of stamens and carpels, and neither has fused staminal filaments. Additionally, P. moluccana var. teysmanniana is unique within Passifloraceae in having a subopposite to completely opposite leaf arrangement. The three varieties of P. moluccana have different leaf shapes, petiolar and laminar nectary glands, and floral morphology.

The combined ITS and trnL-F data sets resolved the two accessions of P. moluccana var. teysmanniana together as sisters, but these accessions were not resolved as sister to P. moluccana var. glaberrima. This indicates that the P. moluccana complex is not a monophyletic group as currently defined. The two accessions of P. siamica grouped together, supporting the current delimitation of this species. The Chinese species P. cupiformis and P. henryi are similar morphologically and form a clade in the combined analysis presented here. The combined molecular data resolve a single clade that contains P. perakensis, P. moluccana var. glaberrima, P. siamica, and P. tonkinensis. This grouping is supported by a key morphological synapomorphy, which is the presence of partial or complete fusion of the staminal filaments around the ovary. No other species in supersection Disemma have any fusion of the filaments on the androgynophore beyond the base of the ovary. In Passiflora moluccana var. glaberrima, the stamen filaments are basally united around the gynoecium, thus partially enclosing the ovary. Similarly, P. siamica and P. tonkinensis have partial fusion of the staminal filaments around the ovary. In P. perakensis, however, the fusion of the stamens is more extensive, resulting in the ovary being entirely enclosed by the filaments. Leaf shape in the four taxa is very similar, ranging from lanceolate to linear, with disc-shaped petiolar nectaries and between 2–16 laminar nectaries on the abaxial surface of the leaf.

The clade containing P. moluccana var. glaberrima, P. tonkinensis, P. siamica, and P. perakensis is fully resolved in the strict consensus of the six most parsimonious trees produced by the ITS data set alone (Fig. 1A). According to this hypothesis, P. perakensis is sister to all of the anomalous taxa. Within the anomalous taxa, P. moluccana var. glaberrima is sister to P. tonkinensis, and this clade is then sister to P. siamica. This topology implies that a single evolutionary event is responsible for the anomalous growth pattern observed in P. moluccana var. glaberrima, P. tonkinensis, and P. siamica. Additionally, the topology predicts that the developmental pattern of P. tonkinensis should be more similar to P. moluccana var. glaberrima than to P. siamica. Flowers from a single plant of P. moluccana var. glaberrima may have between six and eight stamens and four to five carpels, while P. tonkinensis has seven stamens and three to four carpels. However, P. siamica has five to seven stamens and three to four carpels, so a distinct affinity of P. tonkinensis to either species is unclear based on stamen or carpel number alone.

The ITS data are not entirely congruent with the trnL-F data. When analyzed alone, the trnL-F data set produced 133 equally parsimonious trees, resulting in poor resolution of relationships within Disemma. However, when the trnL-F data were combined with the ITS data set, four trees with two distinct topologies resulted. Two of the trees resolved P. perakensis as sister to the rest of the anomalous taxa and within that clade, P. moluccana var. glaberrima and P. tonkinensis are sisters. In the other two topologies produced by the combined data, P. tonkinensis and P. perakensis are sisters, and P. moluccana var. glaberrima is sister to P. siamica. This second hypothesis suggests two independent derivations of the anomalous condition in these closely related taxa. Additionally, it suggests that the similarity in numbers of stamens and carpels in P. moluccana var. glaberrima and P. tonkinensis is not significant. Regardless of the conflict between the ITS and trnL-F data in the Asian taxa, the combined data set clearly supports the hypothesis that the change from normal to anomalous floral development occurred within a single lineage rather than multiple times independently. The question remains as to whether the anomalous condition was derived one or two times within these species. Therefore, a synapomorphy for this group might be defined as a partial or entire loss of ontogenetic regulation of stamen and carpel number, which might allow for single or multiple doubling events within the floral whorls.

General mode of development in Passiflora
The observations presented here for Passiflora holosericea and P. perakensis represent a generalized developmental pathway for the genus in which flowers develop five stamens and three carpels. The patterns noted here are congruent with those observed in P. edulis, P. quadrangularis (Moncur, 1988 ), and P. racemosa (Bernhard, 1999 ). The calyx is initiated spirally (Fig. 2) and alternately with the petals. The petal primordia emerge shortly after the sepal primordia (Figs. 3, 12). Throughout development of the androecium, primordial height increases at the same rate among all individuals within the whorl (Figs. 6, 15). The carpels are initiated sequentially (Figs. 8, 16). The formation of both the androecium and gynoecium is complete just as the first coronal filaments appear (Figs. 10, 18). The emergence and elongation of the androgynophore occurs later in the differentiation of the flowers.

Masters (1871) noted several cases of teratology within Passiflora, particularly in members of subgenus Passiflora. He cited cases of flowers that had increased numbers of sepals, petals, and stamens from the normal five to six. Additionally, he noted a case in P. alata Curtis where the petals increased from five to six and were arranged spirally such that the sixth petal appeared to occupy a new whorl within the flower. While these may be cases of genetic mutation due to hybridization or polyploidy, they may also indicate a genetic predisposition to an increase in number of floral parts within the genus.

Anomalous development in the Asian taxa
Androecium
The two species of Passiflora investigated have anomalous patterns of floral ontogeny (Figs. 20–55). The developmental series presented for P. siamica and P. moluccana var. glaberrima represent the variation observed within inflorescences on individual plants. This increased level of variation suggests that regulation of stamen and carpel number has been relaxed in these species. Though floral material was unavailable for P. tonkinensis, the variability in stamen and carpel number observed in herbarium specimens indicates that the same is true for this species. Both P. siamica (Figs. 20–22, 26) and P. moluccana var. glaberrima (Fig. 38) retain the normal pattern of initiation for sepals and petals, but after this stage their development is less canalized. The five stamen positions are strictly maintained whether the position is filled with a single larger stamen primordium or a bifurcated pair of smaller stamen primordia (Figs. 26–28, 40–43). Stamens may initiate in a spiral sequence in P. siamica, but if doubling of stamen primordia results in seven stamens (Fig. 35), the positions of the primordia become ambiguous with regard to the petals. In the case of six stamens and four carpels in P. siamica (Fig. 33), the stamens are not entirely opposite or alternate with the petals, and two of the carpels are opposite the stamens, while two are alternate. In P. moluccana var. glaberrima, the doubling event occurs earlier, and while some measure of symmetry may be retained in buds developing eight stamens (Fig. 42), those that develop seven stamen primordia are also not strictly alternate or opposite the petals (Fig. 40). Almost all possible combinations of stamen and carpel number are observed in P. siamica, and all combinations are observed in P. moluccana var. glaberrima.

The lability of stamen number observed in both P. siamica and P. moluccana var. glaberrima is uncommon within the genus Passiflora, but the insertion of additional members into the staminal whorl is observed throughout the angiosperms (Endress, 1994 ). The first organs that initiate on the floral apex, usually the sepals, establish a particular symmetry that influences development of the remaining floral whorls (Endress, 1990 ). Additionally, organs that develop spirally are characterized by having unequal interspaces between them (Endress, 1997 ). In the case of the genus Dillenia, for example, there are five large sepals that develop spirally with long plastochrons. This results in three broad and two narrow interspaces on the apex where additional stamens are frequently inserted (Endress, 1997 ). In Passiflora, the sepals also develop spirally, and as a consequence, they vary greatly in size and width in the early stages of development. The sepals in most species of Passiflora have 2/5 phyllotaxy (Fig. 56A, B), resulting in three large and two small primordia. Because the petal primordia arise almost simultaneously and alternately with the sepals, their positions are affected by the initial orientation of the sepals on the floral apex. This results in three larger and two smaller interspaces between the petals, similar to the case in Dillenia.

Endress (1987) noted that increases in stamen number within the androecial whorl are facilitated by the small insertion area of each stamen. Because they require less circumferential space within the floral apex, there may be more flexibility in their arrangement. It is likely that the extra space available between petal primordia on the floral apex in the Asian Passiflora has aided in the doubling of stamen primordia in both P. siamica and P. moluccana var. glaberrima. The non-anomalous species P. holosericea (Fig. 2) and P. perakensis (Fig. 11), as well as the anomalous P. siamica (not shown) and P. moluccana var. glaberrima (not shown) have regular pentamerous spacing of the sepal primordia. In illustrating the generalized patterns of development in P. siamica and P. moluccana var. glaberrima, we assumed that doubling of primordia would occur preferentially in the largest space created by the first-initiated sepal primordium. Doubling of other initials would follow in the interspaces created by the second and third sepals (Fig. 56A, B). However, due to limited availability of flowering material for P. siamica and P. moluccana var. glaberrima, we could not confirm the association of the doubled stamen primordia with a particular interspace created by the developing sepal primordia. In other words, it is unclear if doubling occurs preferentially in the interspace created by the first, second, or third sepal primordium, because these primordia create the largest gaps between the petals on the floral apex.

The increased width between petal primordia might facilitate the anomalous growth patterns within the floral apex, but identifying this change in development does not address the actual mechanism through which doubling of stamen primordia is attained. In P. siamica, the bifurcation of the stamen initials occurs only after each stamen primordium has emerged from the floral apex (Figs. 26, 27). Certain primordia become wider than others, and shortly thereafter the widened primordia cleave into distinctly smaller units (Figs. 27–30). In P. siamica, the division of the wider primordia can best be described as dédoublement in its most conservative sense as applied by Ronse Decraene and Smets (1993) . By their definition, dédoublement refers to the division of a primary primordium into two or more units, in which all of the primary primordial tissue is utilized in the formation of the daughter, or secondary, primordia. The doubling of stamen primordia in P. siamica appears to adhere to this definition; the primary wider region begins to cleave into two secondary primordia (Fig. 28), and these in turn develop into two fully functional stamens (Figs. 31, 32).

However, in P. moluccana var. glaberrima, the doubling occurs much earlier in development. Several of the stamen primordia emerge from the floral apex at half the size of normal primordia (Fig. 39). Dédoublement is not as readily applied to P. moluccana var. glaberrima. In this case, the hypothesis of congenital dédoublement, a subset of the more strict definition provided by Ronse Decraene and Smets (1993) , might provide a possible explanation. Given the phylogenetic proximity of these two taxa to one another, it is reasonable to expect that transitional forms of the ontogenetic pathways might be observed. The hypothesis of congenital dédoublement would require that either of two changes has occurred in P. moluccana var. glaberrima: either the widened primary primordial stage observed in P. siamica is lost entirely, or the development of P. moluccana var. glaberrima has accelerated such that it occurs before the stamen primordia emerge from the floral apex. A third explanation for the condition observed in P. moluccana var. glaberrima is that the extra stamens are derived through an ontogenetic pathway that is entirely unrelated to that of P. siamica. However, if the results of the ontogenetic study are viewed within the context of the phylogenetic analysis, the most parsimonious explanation is that the anomalous pathways are historically homologous at some level and that the relaxation of strict regulation in the numbers of reproductive organs is a shared synapomorphy for P. moluccana var. glaberrima, P. tonkinensis, and P. siamica.

Gynoecium
The development of the gynoecium is equally as variable as the androecium in the two anomalous species of Passiflora examined here. In P. siamica, when five stamens develop, only three carpels are produced. In six-stamen flowers, either three or four carpels arise. In the seven-stamen form, only three carpels were observed. In P. moluccana var. glaberrima, all possible combinations were observed between the seven and eight-stamen forms with four and five carpels. Based on these results, it is difficult to draw any conclusion about the relationship between stamen and carpel number, though it seems likely that a connection between the two whorls does exist. It is interesting to note that very few cases of either infrageneric or infraspecific syncarpous gynoecial variability have been reported (for examples see Tisserat et al., 1990 , and Rohrer et al., 1991 ). The lack of carpel number variability within individual flowers might be due to several factors, including the limited remaining space allocated to carpel initiation within the floral apex, the late timing of carpel initiation on the apex, or decreased fecundity due to problems in the construction of a functional compitum, the intercarpel transmitting tissue of the pollen tube, that allows for pollen tube transmission (Endress, 1987 , 1994 ). In cases where carpel number is increased, the ovary becomes somewhat deformed, and the carpels are arranged in parallel rows (Endress, 1994 ). This flattened arrangement is observed in the five-carpellate P. moluccana var. glaberrima buds (Figs. 53, 54). In the anomalous Passiflora species, reproductive functionality of those flowers with higher carpel number could not be assessed. Based on the results of the current study, it is not possible to determine a distinct relationship between stamen and carpel number in either P. siamica or P. moluccana var. glaberrima.

Relevance of the anomalous Passiflora to Passifloraceae
Passifloraceae consists of two extremely speciose genera, Passiflora and Adenia, as well as 15 smaller genera. This situation makes it difficult to compare the structures characteristic of the larger genera to the smaller genera. Additionally, without a well-supported phylogeny, no context exists for examining the transformations among the genera. For example, one of the central problems in understanding the evolution of Passifloraceae has been the determination of the primitive androecial and gynoecial conditions for the family. Members of the Passifloraceae vary greatly in stamen and carpel number (Table 1). The tribe Passifloreae is highly variable in numbers of floral parts; tribe Paropsieae is somewhat less variable: Barteria, Viridivia, and Smeathmannia are polyandrous, with at least five, but often more than 20 stamens, and range from three to six carpels. Androsiphonia and Paropsia have five stamens and two to six carpels. Outgroup comparisons are problematic due to a lack of information on relationships among Passifloraceae and its allies.

View this table: [in this window] [in a new window]   Table 1. Variation in floral morphology in the genera of Passifloraceae, based on descriptions provided by Hutchinson (1967) and De Wilde (1971 , 1974 )

Bernhard (1999) correctly pointed out that no members of those families that are believed to be closely related to the Passifloraceae (e.g., Malesherbiaceae, Turneraceae, Flacourtiaceae s.l.) have triplostemonous androecia, as would be expected if this condition were primitive for the family. Additionally, Bernhard lists several genera within the Passifloraceae whose development is not congruent with a triplostemonous hypothesis. While Basananthe triloba (Bolus ex Schinz) W.J.J.O. De Wilde does possess ridges that alternate with the stamens, the initiation of these ridges occurs much later than that of the stamens. In Crossostemma, there are two alternating whorls of fleshy teeth around the stamens, but again, the initiation of these teeth occurs much later than that of the fertile stamens (Bernhard, 1999 ). The polyandrous androecium in Smeathmannia is clearly derived from a single whorl, where each member emerges simultaneously within a clearly defined zone (Fig. 8F–H in Bernhard, 1999 ). Similar to Smeathmannia, the Asian taxa derive the androecium from a single, clearly defined peripheral zone within the floral apex (Figs. 3, 12, 22, 39–43). If it were true that the extra stamens in the Asian species represented relictual members of inner staminal whorls, a widened, more irregular zone of initiation would be expected. The clear dédoublement observed among members of the same whorl in P. siamica further refutes De Wilde's (1974) hypothesis of triplostemony. Based on his examination of several genera in Passifloraceae, Bernhard (1999) concluded that the ancestral androecium in Passifloraceae was most likely haplostemonous. The haplostemonous condition would be more likely to facilitate cases of one-whorled polyandry, such as those observed in several genera within the Passifloraceae (e.g., Viridivia, Barteria, and Smeathmannia).

In the closest relatives to Passifloraceae, the Malesherbiaceae and Turneraceae, stamen number is constant, and both families have a single whorl of fertile stamens. Interpretation of the significance of the stamen condition in Malesherbiaceae and Turneraceae is limited by the absence of a strongly supported phylogeny for these families. Both the molecular phylogenetic analysis of Soltis et al. (2000) using 18S ribosomal DNA, rbcL, and atpB, and the Chase et al. (2002) analysis using only rbcL data resolved Malesherbiaceae and Turneraceae as sister families. If this hypothesis is correct, the ancestor to Malesherbiaceae-Turneraceae and Passifloraceae could have been either haplostemonous or polyandrous. However, the recent ndhF analysis of Alford (2005) and Krosnick et al. (2005) , with much denser sampling within closely allied families, resolves Malesherbiaceae as sister to a clade containing Turneraceae and Passifloraceae. If the latter hypothesis continues to be supported, the primitive androecial condition for the Passifloraceae may have been a single pentamerous whorl. Until phylogenetic analysis resolves the relationships among Passifloraceae, Malesherbiaceae, Turneraceae, and Flacourtiaceae, it will be difficult to confidently determine if a pentamerous or a one-whorled polyandrous androecium was the ancestral condition in Passifloraceae.

Gynoecium
The relevance of the gynoecial variation observed in the anomalous Asian species to the rest of the Passifloraceae is less straightforward. According to De Wilde's (1971) account, the minimum carpel number observed in Passifloraceae is two, and the greatest is six. Even if the Paropsieae were to be supported as sister to the rest of Passifloraceae, this would do little to help polarize the evolution of the gynoecium for the family given its extreme variability. Unfortunately, character polarization is not facilitated by looking to the Malesherbiaceae or Turneraceae. Malesherbiaceae ranges between three and four carpels, while Turneraceae is consistently three carpellate. If the Malesherbiaceae-Turneraceae clade is supported (Soltis et al., 2000 ; Chase et al., 2002 ), the primitive gynoecial condition for Passifloraceae is ambiguous. If Malesherbiaceae is sister to a clade containing Turneraceae and Passifloraceae (Alford, 2005 ; Krosnick et al., 2005 ), the ancestral gynoecium for Passifloraceae would have been either three or four-carpellate and is thus also ambiguous. It seems that the primitive gynoecial condition in Passifloraceae will remain enigmatic until a phylogeny of the genera in Passifloraceae and closely related taxa is confidently resolved.

In the normal developmental pathway for Passiflora, five stamens emerge alternately with the petals, later followed by three carpels; of these, one carpel is alternate with the stamens and two are opposite. This atypical arrangement might suggest that there were five carpels at one time and a reduction of two carpels occurred, resulting in the three observed. No evidence of such a reduction exists in early ontogenetic stages of either normal or anomalous Passiflora species, so this conclusion is difficult to support. The carpel number for P. siamica and P. moluccana var. glaberrima does not exceed five, which also might support a five-carpel hypothesis. Spatial allocation of the floral ground plan might be limited such that even anomalous species would have an upper limit on the number of members in each whorl. Given the three-carpellate condition in Turneraceae and the three to four carpels observed in Malesherbiaceae, it seems most likely that the ancestral condition was trimerous. The variation observed in stamen and carpel number in Passifloraceae, if achieved through dédoublement in a similar manner as the Asian species of Passiflora, could be considered characteristic of the family.

Occurrences of doubling in the angiosperms
It is notable that while doubling of primordia within floral whorls occurs broadly throughout the angiosperms, doubling events most often occur at the border between two different floral whorls, rather than within a particular whorl. Petals and stamens are often doubled in the Magnoliid complex, the Alismatales, and the Caryophyllids (Endress, 1987 , 1994 ). Frequently, additional stamens are inserted at the border between the petal and stamen whorl, and occur in taxa that have many free stamens and carpels. However, almost all of the cases of stamen doubling within a single whorl are found in the Flacourtiaceae s.l. (Endress, 1994 ; Bernhard and Endress, 1999 ), Capparaceae (Karrer, 1991 ), Brassicaceae (Endress, 1992 ), Zygophyllaceae (Ronse Decraene and Smets, 1991 ), Loasaceae (Hufford, 1988 ), Begoniaceae (Ronse Decraene and Smets, 1990 ), and Dilleniaceae (Endress, 1997 ).

Of particular relevance to the taxa considered here is the close phylogenetic relationship of Flacourtiaceae s.l. to Passifloraceae. There is a great deal of variability in petal, stamen, and carpel number in this group, and it is worthwhile to examine the nature of the variability to determine if there is a developmental commonality applicable to the Asian Passiflora. The Flacourtiaceae s.l. has been dissolved, and species are now placed in three unrelated families, the Achariaceae, Samydaceae, and Salicaceae (Chase et al. 2002 ; Alford, 2005 ). While familial definitions have changed, phylogenetic analysis consistently places the newly created families near the Passifloraceae, Malesherbiaceae, Turneraceae, Goupiaceae, Violaceae, Lacistemataceae, and Scyphostegiaceae (Chase et al., 2002 ; Sosa et al., 2003 ; Davis and Wurdack, 2004 ; Davis et al., 2005 ).

Petal variation
Species in the former Flacourtiaceae vary greatly in petal number. In Caloncoba Gilg, dédoublement results in the insertion of additional petals within a single whorl (Endress, 1994 ). Lindackeria dentata (Oliv.) Gilg flowers vary between six and 10 petals, Camptostylus ovalis (Oliv.) Chipp has double and triple positions within the petal whorl, and C. mannii (Oliv.) Gilg also varies between six and seven petals through double or triple positions (Bernhard and Endress, 1999 ). In the Achariaceae, variability in petal number is a synapomorphy that unites several tribes, including the Lindackerieae and the Erythrospermeae (Chase et al., 2002 ).

Stamen variation
Many genera of the former Flacourtiaceae are polyandrous and often vary in both stamen number and initiation pattern (Bernhard and Endress, 1999 ). Several genera in the newly circumscribed Samydaceae vary in stamen number within a single whorl (Sleumer, 1980 ). In the genus Lunania Hook., stamens range from 6–12 within a single whorl across the genus, but in L. parviflora Spruce ex Benth., the stamens range from 10 to 12 within an individual plant. Similar levels of variation are observed in the genera Samyda Jacq., in which stamens vary from 6–13 in a single whorl in certain species, and Casearia Jacq., which has 6–10 stamens across the genus. Stamen number is also quite variable in the newly defined Achariaceae. Kiggelaria africana L. is a dioecious species in which male flowers have 10–12 stamens (Bernhard and Endress, 1999 ). In this species, at least one of the stamen initials appears to undergo a bifurcation similar to the doubling observed in the Asian Passiflora. In Camptostylus mannii, C. ovalis, Lindakeria dentata, and Caloncoba echinata (Oliv.) Gilg, the stamens in male flowers are polyandrous and are arranged in an irregular pattern on the floral apex, suggesting a similar relaxation in genetic regulation of floral organ number in these taxa.

Carpel variation
As mentioned previously, examples of angiosperms with syncarpous gynoecia that vary in carpel number within the flowers of an individual plant are uncommon. In Achariaceae, however, several species are variable in carpel number. In particular, Bernhard and Endress (1999) noted that female flowers of Camptostylus mannii range between four and five carpels, while Lindackeria dentata and Caloncoba echinata both have three to four carpels. Carpel number in these species may be affected by the irregular arrangement of the androecium in early stages of development, which would in turn influence the available space and orientation of the later emerging carpel whorl.

The similarity of floral developmental patterns in the Achariaceae and Samydaceae to the conditions observed in Passiflora is notable. The propensity for doubling or even tripling of unit positions within particular whorls appears to be relatively common in these families. Furthermore, doubling is present in the petal, stamen, and carpel whorls in both Samydaceae and Achariaceae. This might suggest that a common genetic mechanism is present within this lineage that facilitates doubling events. Malesherbiaceae is variable in carpel number as well, and further investigation may reveal other cases of doubling in other closely related taxa. It seems reasonable to conclude that the dédoublement of stamens and carpels observed in the Asian Passiflora is homologous to the doubling observed in the Samydaceae, Achariaceae, and Malesherbiaceae in that all the taxa discussed here have a relaxation of genetic constraints within the floral apex.

Important questions remain regarding the evolutionary significance of the doubling events in the Asian Passiflora. For example, the orientation of doubled stamens or carpels might influence the efficacy of pollination in these species. Another implication of doubling might be the creation of new reproductive systems within a species, because the situation might facilitate sexual selection and promote outcrossing. The position of individual anomalous flowers within the inflorescence may be of importance as well. In Achariaceae, for example, terminal flowers in the cymose inflorescences of the monoecious species Ceratiosicyos laevis (Thunb.) A. Meeuse are female, but secondary branches produce only male flowers. Such arrangements are common in Ruta graveolens L. and Saxifraga florulenta Moretti (P. K. Endress, University of Zurich, personal communication). Similar situations may be present within the Asian Passiflora, and examination of mature flowering material will provide an opportunity to address some of these issues.

Future considerations
The analysis presented here has identified the ontogenetic changes leading to the anomalous floral condition observed in the Asian Passiflora. Viewing this change in the context of phylogenetic relationships in supersection Disemma suggests that the anomalous pattern is due to a partial or complete loss of regulatory control on the number of stamens and carpels within the floral apex. The similar type of variation in stamen and carpel number characteristic of other genera in Passifloraceae and the closely related Achariaceae and Samydaceae is interesting and merits further investigation. In order to determine if similar processes are responsible for stamen and carpel polymorphism in the remainder of Passifloraceae, further sampling is required, although this is difficult in that most genera are depauperate and often quite rare. Additionally, to confidently assess the basal androecial and gynoecial conditions for the family, we need a well-resolved phylogeny that includes sufficient sampling within the Paropsieae, Passifloreae, Malesherbiaceae, Turneraceae, Achariaceae, and Samydaceae. The great level of diversity within the Passifloraceae must be viewed within a broad phylogenetic context in order to fully appreciate the many transformations that have occurred, both morphologically and developmentally.

Subgenus

Supersection—Species; Collector no. (location); GenBank accession numbers: trnL-F and intergenic spacer; ITS1, 5.8s, ITS2.

Decaloba (DC.) Rchb.

Auriculata J.M. MacDougal & Feuillet—Passiflora auriculata Kunth; S. Krosnick 350 (OS); DQ284534; DQ284532.

Bryonoides (Harms) J.M. MacDougal & Feuillet—P. adenopoda DC.; S. Krosnick 258 (OS); AY632727; AY632702. P. morifolia Mast.; S. Krosnick 311 (OS); DQ284535; DQ284533.

Cieca (Medic.) J.M. MacDougal & Feuillet—P. coriacea Juss.; S. Krosnick 20 (OS); DQ087429; DQ087420. P. tenuiloba Engelm.; D. Goldman 1770 (BH); AY