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 Google Scholar Google Scholar Articles by Scharaschkin, T. Articles by Doyle, J. A. Search for Related Content PubMed Articles by Scharaschkin, T. Articles by Doyle, J. A. Agricola Articles by Scharaschkin, T. Articles by Doyle, J. A.

Character evolution in Anaxagorea (Annonaceae)¹

²Section of Evolution and Ecology, University of California, Davis, California 95616 USA

Received for publication April 10, 2005. Accepted for publication October 10, 2005.

ABSTRACT

Anaxagorea is a critical genus for understanding morphological evolution in Annonaceae because it shares a variety of features with other Magnoliales that have been interpreted as primitive relative to other Annonaceae. We present a detailed discussion of morphological characters used in a combined morphological and molecular phylogenetic analysis of Anaxagorea, along with implications of the analysis for character evolution in the genus. In spite of a high level of homoplasy in stamen and leaf venation characters, their removal results in loss of resolution in the trees obtained. The distributions of characters on trees confirm assumptions that several distinctive similarities between Anaxagorea and other Magnoliales are primitive retentions (e.g., the presence of an adaxial plate of xylem in the midrib, nonpeltate stamen connectives, inner staminodes, and several leaf architectural characters). However, lateral extensions of the "laminar" stamens, though possibly ancestral in Anaxagorea, are convergent with those in other Magnoliales. A number of morphological synapomorphies have been identified for a clade containing most Central American species and another comprising all Asian species (e.g., conical bud shape and reduced inner petals for the Central American clade, and adaxial cuticular striations and capitate stigma shape for the Asian clade).

Key Words: adaxial plate • Anaxagorea • Annonaceae • leaf architecture • morphology • stamens

There have been several morphological and molecular cladistic studies of the angiosperm family Annonaceae, all of which have indicated that Anaxagorea is the sister group of the remaining taxa (Doyle and Le Thomas, 1994 , 1996 ; van Zuilen, 1996 ; Doyle et al., 2000 ; Sauquet et al., 2003 ; Richardson et al., 2004 ). In the morphological analyses, the basal position of Anaxagorea was supported by its possession of several distinctive character states that also occur in other Magnoliales. However, these outgroup states are not seen in all species of Anaxagorea. In almost all such cases, it was assumed that the conditions shared with the outgroups were ancestral in Anaxagorea, and the genus (which was treated as a single taxon) was scored accordingly.

Some such characters concern the androecium. For example, most species of Anaxagorea possess inner staminodes, a trait also found in some other Magnoliales, namely Eupomatiaceae (Eupomatia), Degeneriaceae (Degeneria), and Himantandraceae (Galbulimima). However, two Anaxagorea species lack inner staminodes. Previous analyses scored the genus as having inner staminodes, and as a result this feature was one of several synapomorphies linking Annonaceae with Eupomatia, Degeneria, and Galbulimima, to the exclusion of Magnoliaceae, in the morphological and molecular analyses of Doyle and Endress (2000) and Sauquet et al. (2003) . This scoring assumed that inner staminodes were originally present in Anaxagorea and were lost independently within Anaxagorea and in other Annonaceae. Another possibility is that Anaxagorea is paraphyletic and loss of inner staminodes is a synapomorphy of some Anaxagorea species and the rest of the family. Similarly, the stamens of most Anaxagorea species are more or less laminar, as in other Magnoliales, and thereby differ from the peltate stamens (with a cap-like connective extension) found in most other Annonaceae. However, some Anaxagorea species have stamens that approach the peltate type. Again, it was assumed that the laminar type was ancestral and deviations from it occurred independently in Anaxagorea and other Annonaceae.

Similar patterns are seen in leaf architecture and anatomy. Most species of Anaxagorea have brochidodromous leaf venation, but some are eucamptodromous. Eucamptodromous venation is also found in the majority of the other Annonaceae, whereas most other Magnoliales have a brochidodromous pattern, often of the festooned type (cf. Hickey and Wolfe, 1975 ). A feature found throughout other Magnoliales is an adaxial plate of xylem in the midrib. This plate seems to have been lost in Annonaceae, with the exception of most (though not all) species of Anaxagorea. Again, Doyle and Le Thomas (1996) assumed that the adaxial plate in Anaxagorea was a primitive feature retained from the ancestors of Annonaceae, which was lost independently within Anaxagorea and in other Annonaceae.

To test these assumptions requires an understanding of phylogenetic relationships in Anaxagorea and an examination of the variation within the genus in this framework. To address this problem and to evaluate biogeographic scenarios for the unique Asian-neotropical distribution of Anaxagorea, we conducted separate and combined analyses of morphological characters and plastid DNA sequences, the results of which are reported elsewhere (Scharaschkin and Doyle, 2005 ). The main objectives of the present article are (1) to provide a detailed justification of the morphological characters used in that study, in particular those showing variation in Anaxagorea, (2) to assess whether the similarities between Anaxagorea and the other Magnoliales were primitively retained or secondarily derived within Anaxagorea, and (3) to examine the evolution of morphological characters, in particular those pertaining to leaf architecture and stamen morphology, and their utility in determining phylogenetic relationships in the genus, using trees obtained from our combined morphological and molecular analysis.

MATERIALS AND METHODS

Taxon sampling
The 25 Anaxagorea species in our data set represent all currently recognized species of the genus except A. radiata, which may actually belong to A. javanica (Maas and Westra, 1984 , 1985 ; Maas et al., 1986 ; Berry et al., 1999 ; Scharaschkin and Doyle, 2005 ). The 16 outgroups include the other families of Magnoliales, namely Myristicaceae, Magnoliaceae, Himantandraceae (Galbulimima), Degeneriaceae (Degeneria), and Eupomatiaceae (Eupomatia), and the same exemplars of the principal clades making up the rest of the Annonaceae that were included in a broad-scale analysis of Magnoliales by Sauquet et al. (2003) (Tables 1 and 2). Doyle and Le Thomas (1996) treated Annona and Asimina as members of the "Annona group" (Annona, Anonidium, Raimondia, Rollinia, and Rolliniopsis) and the "Asimina group" (Asimina and Deeringothamnus). Because most other members of the Annona group are probably nested within Annona, which was therefore used to score most characters, and because Deeringothamnus does not differ from Asimina in characters used in this analysis, we replaced these groups with Annona and Asimina, respectively. Recent molecular analyses indicate that Anonidium is related to Neostenanthera rather than Annona (Richardson et al., 2004 ), but its exclusion would affect scoring of only one character in the present data set (raphe shape: see below Results and Discussion, Seeds).

We made an effort to define characters that were not logically correlated and seemed to vary independently across taxa. For some character systems, such as leaf architecture, many other characters might be formulated, but most of these would be correlated with existing characters, and their inclusion would lead to overweighting of what are really single characters and might therefore result in spurious relationships or misleadingly high support for clades.

Approximately 10 characters are more or less quantitative. Use of such characters in phylogenetic analysis has been criticized by some authors (e.g., Stevens, 1991 ) but defended by others based on simulation studies and other empirical evidence (Thiele, 1993 ; Poe and Wiens, 2000 ). In an attempt to define systematically valid states in quantitative characters, we plotted data in histograms and looked for apparent natural breaks in the distribution of measures.

A problem frequently encountered in our study, as in many others that mix characters from broader-scale analyses with new characters that appear to be useful at the species level, is that many of the latter characters are excessively variable at higher levels. We found this was especially true for some quantitative characters. Also, in several quantitative characters, we were able to recognize breaks between states in Anaxagorea, but the same breaks did not appear to be valid in outgroup taxa. In such cases, we scored Anaxagorea for these characters but scored either all outgroups or other families of Magnoliales as unknown.

The resulting data set consists of 75 parsimony-informative characters, of which 49 are binary and 26 are multistate (Table 1). Five quantitative characters that were ordered in Doyle and Le Thomas (1996) were retained as such, while all the remaining characters are unordered. Variation in characters within taxa is scored as uncertainty (e.g., 0/1).

Leaves from all available species of Anaxagorea were sectioned to examine the midrib anatomy in detail, as were leaves of the outgroups, to verify previously published observations (Table 2). Leaves from herbarium specimens were soaked in water overnight. They were then freehand sectioned and stained with phloroglucinol for 1–2 min before examination under a light microscope.

Leaves from herbarium specimens were X-rayed using a cabinet-style analytical X-ray machine at the School of Veterinary Medicine, University of California, Davis. The basic procedure follows Wing (1992) ; the specific settings used were 11–12 kV, 3 mA, tube distance = 50.8 cm with a 2.5-min exposure time. X-radiographs were then examined under a dissecting scope for characters such as variation in higher-order venation. Voucher information for specimens X-rayed is listed in Table 2.

Stamen and staminode morphology was examined in detail using both light microscopy and scanning electron microscopy (SEM). Voucher information for specimens examined using SEM is listed in Table 2.

Morphology of the floral receptacle was studied by making longitudinal sections of preserved and boiled herbarium material. Mesotestal fibers were examined by making freehand sections of the seed coat with a razor blade and observing the fiber orientation under a dissecting microscope. A list of species sectioned and vouchers used to examine these characters is presented in Table 2.

Trees
Trees used in this study are derived from Scharaschkin and Doyle (2005) . The molecular data consisted of DNA sequences from the atpB-rbcL, psbA-trnH, and trnL-trnF spacer regions and the trnL intron. In the analyses considered here we constrained outgroup relationships to the topology found by Sauquet et al. (2003) , which is consistent with the arrangement of Annonaceae in Richardson et al. (2004) . Relationships within Anaxagorea were unresolved in the strict consensus of more than 50 000 most parsimonious trees derived from the molecular data, but the use of agreement subtrees (Finden and Gordon, 1985 ) showed that relationships amongst a subset of taxa remained stable. These relationships included the nested position of two clades that contain all Asian and most Central American species, a result that was critical for biogeographic scenarios. Similarly, only a few clades occurred in the consensus of 60 most parsimonious trees based on morphology. However, relationships were better resolved in the four trees found in the combined analysis (Fig. 1a). The rooting of these trees is identical; the differences amongst them lie in the placement of the South American species A. pachypetala and the positions of A. floribunda, A. macrantha, A. rufa, and A. phaeocarpa in a large clade of South American taxa. This large clade, which contains the most recent common ancestor of A. floribunda, A. acuminata, and A. dolichocarpa and all of its descendants, will be referred to as the "core South American clade." In analyses of all taxa, bootstrap and decay values within Anaxagorea were generally low (Fig. 1a). However, bootstrap values from an analysis of just those taxa found in one of the molecular agreement subtrees were much higher (Fig. 1b), suggesting there is a fairly robust "backbone" of relationships in the genus.

View larger version (18K): [in this window] [in a new window]   Fig. 1. Relationships in Anaxagorea inferred from parsimony analyses of combined morphological and molecular data (Scharaschkin and Doyle, 2005 ). (a) Strict consensus of four most parsimonious trees from an analysis of all species, with bootstrap and decay values, and (b) most parsimonious tree from an analysis of the species found in an agreement subtree based on the molecular data, with bootstrap values from an analysis of the combined data for this subset of species

We examined morphological evolution on two of the four trees produced by the combined analyses, which differ only in the placement of A. pachypetala, using MacClade (Maddison and Maddison, 2001 ). It is the variation in position of this species that has the greatest effect on inferred patterns of character evolution. These two trees will be referred to throughout as tree A and tree B. In tree A, A. pachypetala is the sister group of the core South American clade alone, whereas in tree B it is the sister group of the core South American clade plus the Central American clade. The other two most parsimonious trees differ in the arrangement of species in a subgroup of the core South American clade and therefore have less effect on evolutionary scenarios across the genus. The two trees used were chosen because they contain the only relationship in this clade (of A. brevipedicellata and A. gigantophylla) with substantial bootstrap support (Scharaschkin and Doyle, 2005 ).

The practice of reconstructing morphological evolution by optimizing characters on a tree based wholly or in part on morphological data has been criticized as involving circular reasoning (e.g., Hedges and Maxson, 1996 ). However, the concept that such reasoning is circular has itself been rejected on logical grounds (e.g., Lee, 1997 ; de Queiroz, 2000 ). Any such tree is based not only on the character whose evolution is being discussed at a given time, but also on all the other characters included in the analysis, which together provide a best available estimate of phylogenetic relationships. Even if the circularity argument is accepted, it would only require removing one character at a time from the analysis, not all morphological characters. In any case, the alternative procedure of optimizing morphological characters on a molecular tree would be a futile exercise in the present study, because well-resolved trees were obtained only by combining molecular and morphological data.

RESULTS AND DISCUSSION

Although definitions of characters and their states are presented in the Appendix, most of the practical and conceptual problems encountered and the arguments used in defining characters are treated in this section, because we view our analysis of morphological characters as one of the main results of this study. Results and discussion for sets of related characters are treated together in order to avoid unnecessary duplication.

Midrib histology
Midrib histology (character 2) is an aspect of leaf anatomy that has drawn much attention in Magnoliales, as most families of the order differ from other "basal" angiosperms in possessing an adaxial plate of xylem in the midrib. This feature was inferred to be a synapomorphy of Magnoliales by Doyle and Endress (2000) . However, the adaxial plate is absent in most Annonaceae, with the conspicuous exception of some, though not all, Anaxagorea species (Fig. 2a–f). Doyle and Le Thomas (1996) scored the Annona group as 0/1 based on what they interpreted as an adaxial plate in Annona in fig. 23 of van Setten and Koek-Noorman (1986) . However, this structure was labeled as sclerenchyma, and our observations on several Annona species confirm this interpretation, so we have rescored Annona as lacking an adaxial plate.

View larger version (85K): [in this window] [in a new window]   Fig. 2. Midrib anatomy. (a) Galbulimima (Himantandraceae): arc plus fused plate, protuberances absent/inconspicuous. (b) Asimina (Annonaceae): simple arc. (c) Anaxagorea brachycarpa: arc plus unfused plate, protuberances absent/inconspicuous. (d) A. borneensis: simple arc. (e) A. crassipetala: arc plus fused plate, prominent protuberances. (f) A. rheophytica: arc plus fused plate, protuberances absent/inconspicuous. Scale bar = 200 µm

In Anaxagorea, some species have elaborate protuberances on the adaxial surface of the adaxial plate, whereas others have very simple plates or plates with inconspicuous projections, as in other Magnoliales (Fig. 2, contrast e and f). This was coded as an additional character (3), for which those taxa lacking a plate were scored as unknown (?).

Serial sections were made along the length of the midrib to assess whether the degree of fusion changed as a function of distance from the petiole. It was easiest to observe the adaxial plate in the lower half of the blade, although if a distinct plate was present in this area, it usually continued as a recognizable structure through the rest of the blade. The only exception to this observation was in Eupomatia, in which the adaxial plate was actually more clearly visible in the middle to upper third of the blade. Sugiyama (1976a , b , 1979) did an extensive study on the node-leaf vascular anatomy in Magnoliales, including Eupomatia, and we consulted her observations closely in order to understand midrib histology. As a result of our observations on Eupomatia, which Sugiyama (1976b , 1979) described as having an adaxial plate, all those taxa that we had scored as lacking an adaxial plate were resectioned throughout the length of the blade to rule out the presence of an adaxial plate. Sections were also made through the petiole, as those taxa having an adaxial plate also have additional vascular bundles adaxial to the main arc of petiole bundles. Since our observations on the vascular bundle arrangement in the petiole were consistently correlated with status of the adaxial plate in the blade, we did not treat petiole anatomy as an additional character.

The presence of an adaxial plate has been cited as support for the basal position of Anaxagorea (Doyle and Le Thomas, 1996 ). This assumes that presence of this feature is ancestral in Anaxagorea and its absence in some species is due to loss. The members of the genus that lack an adaxial plate are the Asian clade (A. borneensis, A. javanica, and A. luzonensis) and the Brazilian species A. silvatica. This has been considered evidence that the Asian species and A. silvatica may be related (Maas and Westra, 1984 ). Maas et al. (1986) claimed that A. rheophytica also lacks an adaxial plate, but our observations clearly show that a plate is present (Fig. 2f).

The phylogenetic results support the assumption that the presence of an adaxial plate is ancestral in Anaxagorea and that the plate was lost independently within Anaxagorea and in the remaining Annonaceae (Fig. 3a). The strong bootstrap support for Anaxagorea (Fig. 1) rules out the possibility that the genus is paraphyletic, with the species without a plate linked with other Annonaceae (as in some morphological trees with outgroups unconstrained; Scharaschkin and Doyle, 2005 ). Because the Asian clade and A. silvatica form two adjacent lines rather than a clade (it is two steps less parsimonious to link the two groups), it is equivocal whether the plate was lost independently in these two lines or was lost once in their common ancestor and reappeared in the combined Central American and core South American clade. Most Anaxagorea species have an adaxial plate fused with the vascular arc, and this is homologous with the situation seen in the other Magnoliales. An unfused plate arose independently in three species.

View larger version (44K): [in this window] [in a new window]   Fig. 3. Tree A, showing inferred evolution of midrib histology (a) and protuberances on the adaxial plate (b) in Anaxagorea and outgroups

Leaf architecture
Appearance of the midrib on the surface of the leaf (character 4) was defined as having two states, because variation within taxa necessitated treating concave and flat together as one state as opposed to convex, whereas for the secondary veins (character 5) flat and convex were usually associated. The midrib was originally concave-flat in Magnoliales and became raised in A. brevipedicellata and two lines in other Annonaceae (Asimina, Isolona), as well as in several genera not included in the present data set (cf. Doyle and Le Thomas, 1996 ), without uniting any of these taxa. The secondary veins are primitively convex-flat in Myristicaceae and Magnoliaceae, but the four other families of Magnoliales are united by a shift to flat or convex secondaries (Fig. 4a). A shift back to concave secondaries is a synapomorphy of the core South American clade, with three reversals to flat or convex. These observations clearly show that the topography of the midrib and that of the secondaries are not correlated.

View larger version (42K): [in this window] [in a new window]   Fig. 4. Tree A, showing inferred evolution of the appearance of secondary veins on the upper leaf surface (a) and revolute leaf margins (b) in Anaxagorea and outgroups

Another potentially informative feature of some Anaxagorea species is distinctly revolute leaf margins (character 8). Revolute margins arose independently in A. rufa, A. brevipedicellata, and either once or twice in A. silvatica and the A. rheophytica-A. inundata clade (Fig. 4b).

Doyle and Le Thomas (1996) treated secondary venation as one character. However, close examination of the variation in Anaxagorea revealed that the type of venation (character 9), as defined by Hickey (1973) , and the degree of vein curvature (character 10) are not always correlated. Doyle and Le Thomas (1996) concluded that their results supported the view that there was "increasing regularity of successively higher vein orders" (increase in "rank"; Hickey and Wolfe, 1975 ) within the family. Most Anaxagorea species have relatively irregular, "low-rank" brochidodromous venation (Fig. 5a), but some species have a pattern that we describe here as approaching eucamptodromous (Fig. 5b), i.e., with the secondary veins thinning as they approach the leaf margin and some of them not forming distinct brochidodromous loops. Many other Annonaceae show this transitional eucamptodromous venation, while some or all members of the other families of Magnoliales have pronounced festooned brochidodromous venation, with many successive secondary loops towards the margin of leaf (Fig. 5c).

View larger version (110K): [in this window] [in a new window]   Fig. 5. Leaf architectural characters. (a) Anaxagorea brachycarpa: brochidodromous, straight secondary veins, branched intersecondaries, reticulate tertiary venation. (b) A. borneensis: approaching eucamptodromous, moderately curved secondary veins, intersecondaries absent or infrequent, weakly percurrent tertiary venation. (c) Eupomatia laurina: festooned brochidodromous, straight secondary veins, branched intersecondaries, reticulate tertiary venation. Scale bar = 1 cm

View larger version (47K): [in this window] [in a new window]   Fig. 6. Tree B (a), showing inferred evolution of secondary vein type, and tree A (b), showing inferred evolution of secondary vein curvature in Anaxagorea and outgroups

Secondary vein spacing (character 11) and angle (character 12) were considered too variable within taxa for inclusion in Doyle and Le Thomas (1996) , but for the purposes of our finer-scale study such information appeared to be potentially informative. Irregularly spaced secondary veins (Fig. 5a, b) are ancestral in Anaxagorea, a symplesiomorphy with some of the outgroups to Annonaceae (Fig. 5c). Other Annonaceae are united by a shift to uniform spacing, although this result needs confirmation with broader taxon sampling. In most Anaxagorea species (Fig. 5b), the angle of the secondary veins decreases towards the base of the leaf, and this is the ancestral state for the genus, with some lines independently changing to uniform or irregular.

The number of secondary veins (character 13) is another character not included in any previous study of Annonaceae, and in the present analysis it was recorded only in Anaxagorea, because it appeared to be too variable in the outgroups. Our counts were based on larger, mature leaves on a given branch, excluding small basal leaves and incompletely developed leaves. In plotting the average number of secondary veins on each side of the leaf, we found a clear break in the distribution, with most species having either more than 10 or fewer than 10 veins per side. The ancestral condition in Anaxagorea is less than 10 secondary veins per side (Fig. 7). The increase to more than 10 veins on each side is either a synapomorphy of the core South American clade and the Central American clade, with an independent origin in A. borneensis in the Asian clade, or a synapomorphy of all these groups, with a reversal in the Asian species A. javanica and A. luzonensis.

View larger version (37K): [in this window] [in a new window]   Fig. 7. Tree A, showing inferred evolution of number of secondary veins on each side of the leaf in Anaxagorea and outgroups

View larger version (44K): [in this window] [in a new window]   Fig. 8. Tree A, showing inferred evolution of intersecondary vein type (a) and tertiary venation (b) in Anaxagorea and outgroups

We examined several other leaf architectural characters but concluded they were too variable within taxa for inclusion in a phylogenetic study (e.g., leaf shape, leaf size, distance between margin of the leaf and secondary loops, and length-to-width ratio of intercostal areas).

It is significant that if leaf architectural characters are removed and the resulting combined data set is reanalyzed, there is an increase in the number of equally parsimonious trees produced—from four trees with the complete character set to 4839. A strict consensus of these 4839 trees shows a marked decrease in resolution in Anaxagorea. Only the Central American clade (with the same internal relationships) and the clade consisting of A. angustifolia and A. rheophytica are still present. This decrease in resolution is indicative of the utility of leaf architectural characters in phylogenetic studies. There is also a slight decrease in the consistency index and retention index after removal of leaf architectural characters, from CI = 0.695 to 0.692 and from RI = 0.705 to 0.700, but this change is probably too small to be meaningful.

Leaf and wood anatomy
Doyle and Le Thomas (1996) treated foliar hairs as a single character with two states: (0) simple or absent, (1) stellate or peltate. However, Maas and Westra (1984) , Koek-Noorman and Berendsen (1985) , and Maas et al. (1986) described Anaxagorea as having simple, 2– 4 branched (peltate), and stellate trichomes, with some species having only one type but others having more than one. For this reason, we have treated hairs as three separate characters. The presence of simple trichomes (character 16) is ancestral in Anaxagorea, and most species have simple hairs, except for a few losses (Fig. 9a). The ancestral state for peltate trichomes (character 17) is equivocal (Fig. 9b). Their absence is either a synapomorphy or a plesiomorphy of the Asian Anaxagorea species, but it is an unequivocal synapomorphy for one clade in the core South American group. Stellate trichomes (character 18) are ancestrally absent in Anaxagorea, and their origin is an important synapomorphy of the core South American clade, with only a single loss in A. brachycarpa.

View larger version (42K): [in this window] [in a new window]   Fig. 9. Tree A, showing inferred evolution of simple trichomes (a) and peltate trichomes (b) in Anaxagorea and outgroups

View larger version (45K): [in this window] [in a new window]   Fig. 10. Tree A, showing inferred evolution of sclereids (a) and oil cells in the leaves (b) in Anaxagorea and outgroups

The shape of anticlinal cell walls in unspecialized epidermal cells on the abaxial leaf surface (character 22) and the presence of striations on the adaxial surface (character 23) are two new characters added to this study, as fairly extensive information is available for Anaxagorea (Maas and Westra, 1984 ) and in the case of the anticlinal walls for other Annonaceae as well (Roth, 1981 ). Straight to slightly curved anticlinal walls and adaxial striations are important synapomorphies of the Asian clade, to which these states are restricted.

Our two wood anatomical characters were taken from Doyle and Le Thomas (1996) , with the addition of data for Anaxagorea species. Vessel density (character 24) is not informative within Anaxagorea, as all species studied have a density of 10–40 vessels/mm², except for A. inundata, which straddles the limit between this state and <10 vessels/mm². Similarly, A. inundata is the only Anaxagorea species known to have narrow rays (character 25). Because of the presence of narrow rays in A. inundata, the lack of data for other near-basal species of Anaxagorea, and the scattered distribution of ray width in other Annonaceae, the ancestral condition in Anaxagorea is equivocal, rather than wide, as assumed by Doyle and Le Thomas (1996) . As a result, wide rays may be either a synapomorphy or a symplesiomorphy for the Asian clade, the core South American clade, and A. pachypetala. The ancestral state for the family is also equivocal, as in Doyle and Le Thomas (1996) . Both these characters are rather homoplastic for the family as a whole, as noted by Doyle and Le Thomas (1996) .

Inflorescences
The inferred ancestral state for inflorescence position (character 26) in Anaxagorea is axillary, with terminal inflorescences originating independently in A. silvatica and the subgroup of the Asian clade consisting of A. javanica and A. luzonensis.

Annonaceae are often described as having either solitary flowers or cymes (rhipidia), but as discussed by Doyle and Le Thomas (1996) this distinction (character 27) is problematic. As pointed out by L. Chatrou (Utrecht University, personal communication), all Annonaceae except the Cymbopetalum group (Bocageeae: Johnson and Murray, 1995 ) have bracts on the pedicel that are capable of producing additional flowers from their axils and often do so occasionally even when they usually have solitary flowers. Hence Doyle and Le Thomas defined their character to distinguish taxa that usually have only one flower at a time from those that have several. As such, this character is more quantitative than qualitative. Under this definition, all Anaxagorea species bear flowers in cymes, but given the present sampling of Annonaceae, the ancestral condition in the family is equivocal, rather than cymes as concluded by Doyle and Le Thomas (1996) . More extensive taxon sampling and quantitative evaluation of variation in this character are needed.

Persistence of the upper bract on the pedicel (character 28) is an informative character for Anaxagorea, in which persistent bracts are ancestral. In tree B, caducous bracts are a synapomorphy for the core South American clade, although a reversal to persistent bracts unites A. brevipedicellata and A. gigantophylla (Fig. 11). However, whether the transition to caducous bracts occurred at the base of this clade or below the node uniting it with A. pachypetala is equivocal in tree A, as the condition in A. pachypetala is not known.

View larger version (34K): [in this window] [in a new window]   Fig. 11. Tree B, showing inferred evolution of persistence of the upper bract in Anaxagorea and outgroups

The petal whorl character (30) is a combination of information from Doyle and Le Thomas (1996) and Doyle and Endress (2000) , with the addition of the state "inner whorl absent," found in A. borneensis and a variety of A. javanica, both Asian taxa. The presence of two whorls of petals is ancestral for Anaxagorea, as for Annonaceae as a whole, but the inner whorl was lost independently in A. borneensis and A. javanica.

All Anaxagorea species except A. silvatica have valvate petal estivation (character 31). With A. silvatica nested within Anaxagorea, valvate petals are ancestral in Annonaceae as a whole, as inferred by Doyle and Le Thomas (1996) . Doyle and Le Thomas also concluded that the valvate state was a synapomorphy of the family, but because the sister group of Annonaceae appears to be Eupomatia, which has no petals, its point of origin is now equivocal.

Relative length of petals (character 32) was treated by Doyle and Le Thomas (1996) in terms of two characters, for inner and outer petals (their characters 34 and 35), in which the states were subequal vs. reduced or absent, but here the presence or absence of petal whorls has been taken into account in character 30. Some Anaxagorea species have inner petals that are conspicuously shorter than the outer petals, and this character was therefore introduced. The presence of shorter inner petals is a conspicuous synapomorphy of the Central American clade, along with conical bud shape, but the two characters do not appear to be intrinsically correlated, as A. macrantha, the only South American species in which buds became conical, does not have shorter inner petals.

Petal shape (character 33), as defined by Doyle and Le Thomas (1996) , is not informative for relationships in Anaxagorea, in which all species have oval petals, and this aspect of petal morphology is symplesiomorphic with the rest of the family. However, the petals of all Anaxagorea species are unusual in being relatively fleshy. In many species the petals also have a conspicuous keel, but others have bowl-shaped petals without any protrusions.

Two separate characters for inner and outer petals were included in this study, because the presence or absence of a keel varies independently in the two whorls. Because they appear to be correlated with thick petals, these characters were not scored outside Anaxagorea. The keel on the outer petals (character 34) is a rather homoplastic character that arose independently in several lines and may have undergone one or two reversals to the ancestral condition of no keel (Fig. 12a). The keel on the inner petals (character 35) is equally homoplastic, but its presence is ancestral rather than derived. Reconstruction of its subsequent history depends on the position of A. pachypetala. In tree A, one loss of the keel accounts for its absence in A. silvatica, the Asian clade, and most species of the Central American clade (Fig. 12b), and its reappearance is a synapomorphy of A. pachypetala and the core South American clade. In tree B, with A. pachypetala interpolated between the Asian and Central American clades, it is equivocal whether the keel was lost several times or lost once and regained several times between the A. angustifolia-A. inundata clade and the core South American clade.

View larger version (40K): [in this window] [in a new window]   Fig. 12. Tree A, showing inferred evolution of a keel on the outer petals (a) and the inner petals (b) in Anaxagorea and outgroups

Xylopia also has inner staminodes, but the tree implies that these arose independently. Because Xylopia and Anaxagorea were also distantly separated in previous analyses, we did not score Xylopia for subsequent inner staminode characters. To do so would run the risk of the Maddison effect (Maddison, 1994 ), where characteristics of a structure in one taxon affect the inferred ancestral state in another taxon in which the structure originated independently.

In Anaxagorea, some species have a large number of inner staminodes (character 38), and the distribution of numbers shows a clear break around 20, allowing the definition of two states. The presence of a large number of inner staminodes is a prominent synapomorphy of the clade containing the most recent common ancestor of A. brachycarpa and A. phaeocarpa and all its descendants, which makes up the larger part of the core South American group. The lack of homoplasy in this character illustrates the systematic value of such quantitative characters.

The shape of the tip of the inner staminodes (character 39) was a difficult character to break into states, because of the great amount of variation within taxa. A number of different options were examined; the current definition appears to be the best way to handle the polymorphisms, since taxa with truncate staminodes often have rounded ones too, but other taxa have strictly rounded staminodes. Some taxa have glandular areas on the edge of the inner staminodes (character 40), whereas others do not. Glandular areas, or mats of unicellular hairs secreting a sticky substance, have also been described in Degeneria, Galbulimima, and Eupomatia (Endress, 1984 , 1994b ; Endress and Hufford, 1989 ), and these were scored for this character. Figure 13 shows some of the diversity in the shape of the apex of the inner staminodes and the presence or absence of glandular areas in Anaxagorea.

View larger version (98K): [in this window] [in a new window]   Fig. 13. Staminode morphology in Anaxagorea. (a) A. prinoides: pointed, lacking glandular areas. (b) A. petiolata: round, lacking glandular margin. (c) A. brachycarpa: round, glandular area present. Scale bar = 250 µm

View larger version (44K): [in this window] [in a new window]   Fig. 14. Tree A, showing inferred evolution of inner staminode shape (a) and glandular areas on inner staminodes (b) in Anaxagorea and outgroups

Three aspects of stamen form (Fig. 15) appear to vary independently and were therefore treated as separate characters. Although stamens in Anaxagorea are more or less laminar, resembling those of most other Magnoliales (Doyle and Le Thomas, 1996 ), they vary considerably in the shape of the apex (character 42), as well as in whether the apex of the connective is bent and if so, whether it is bent towards the thecae (abaxially) or away from them (adaxially) (character 43). The third character (44) accounts for whether the connective extends laterally beyond the thecae, resulting in the most strongly laminar stamens, which were considered primitive by Walker (1971) . We introduced an additional character (45) to express the fact that in some Anaxagorea species those stamens close to the inner staminodes differ markedly from the outer stamens, whereas in other species there is no such differentiation.

View larger version (61K): [in this window] [in a new window]   Fig. 15. Stamen morphology in Anaxagorea. (a) A. brachycarpa: pointed, apex bent away from thecae, lateral extensions of connective absent. (b) A. dolichocarpa: truncate, apex bent towards thecae, lateral extensions of connective absent. (c) A. prinoides: pointed, apex not bent, lateral extensions of connective absent. (d) A. rufa: rounded; apex not bent, lateral extensions of connective absent. (e) A. acuminata: pointed, apex bent away from thecae, lateral extensions of connective present. Scale bar = 250 µm

View larger version (42K): [in this window] [in a new window]   Fig. 16. Tree A, showing inferred evolution of stamen shape (a) and stamen differentiation (b) in Anaxagorea and outgroups

View larger version (42K): [in this window] [in a new window]   Fig. 17. Inferred evolution of lateral connective extensions on tree B (a) and tree A (b) in Anaxagorea and outgroups

A striking but problematic character at the level of Annonaceae is the distinction between normal and locellate anthers (character 46). Doyle and Le Thomas (1996) defined this character as referring to mature anthers, but they recognized that septa may be absent at maturity but present early in development, as emphasized by Tsou and Johnson (2003) . Tsou and Johnson (2003) also showed that the septa in locellate anthers are of two different types. These are not serious problems for the present study, because Anaxagorea does not have locellate anthers, and the two taxa that do have them (Mkilua, Xylopia) have the same type. However, they indicate that the character needs reassessment in future broader-scale analyses.

As with leaf architectural characters, when stamen characters are removed from the data set, there is an increase in the number of equally parsimonious trees produced, from four to 362. Here, however, there is essentially no change in the consistency and retention indices, from CI = 0.695 to 0.694 and from RI = 0.705 to 0.704. In the strict consensus, Anaxagorea is still monophyletic, but the only other clades that remain are the Central American clade, the Asian clade (with internal relationships unresolved), and the clade consisting of A. angustifolia and A. rheophytica.

Pollen
All pollen characters are taken from Doyle and Le Thomas (1996) , without any changes in states or the scoring of outgroups, and with those Anaxagorea species scored that have been examined in various studies. As concluded by Doyle and Le Thomas (1996) , most pollen characters of Anaxagorea have states that are ancestral for Annonaceae and are shared with their closest outgroups in Magnoliales (boat-shaped, monosulcate, single grains with an imperforate tectum and granular infratectum). However, more recent broader-scale analyses indicate that some of these features, most notably granular infratectal sculpture, are derived relative to those of Magnoliaceae, Myristicaceae, and more basal angiosperms (Doyle and Endress, 2000 ; Sauquet et al., 2003 ).

Receptacle and carpels
As noted by Doyle and Le Thomas (1996) , shape of the stamen-bearing portion of the receptacle (character 56) and shape of its carpel-bearing apex (character 57) vary independently of each other. We found no variation in either character in Anaxagorea, but we included them to assess their status in Anaxagorea. As in Doyle and Le Thomas (1996) , the flat or convex receptacle apex seen in Anaxagorea is a synapomorphy of the Annonaceae, whereas the flat or conical shape of the stamen-bearing portion was retained from lower in Magnoliales.

Detailed information on carpel number (character 58) is available for nearly all species of Anaxagorea (Maas and Westra, 1984 , 1985 ); however, there did not seem to be any natural break in the distribution of numbers within the genus. We retained the break at 10 that Doyle and Le Thomas (1996) recognized as potentially useful across Annonaceae, but all species of Anaxagorea have more than 10 carpels, which is also the basic state for the rest of the family.

Doyle and Le Thomas (1996) scored stigma shape (character 59) in Anaxagorea as sessile, but close examination showed that although most species are indeed sessile, a few are capitate. The ancestral condition in Anaxagorea is sessile, and this is a symplesiomorphy shared with other Magnoliales. Capitate stigma shape is an important synapomorphy of the Asian clade but also arose independently in A. inundata. Doyle and Le Thomas (1996) inferred that the ancestral state for other Annonaceae was also sessile, but based on our results this state is equivocal. This is because more recent phylogenetic studies (Doyle et al., 2000 ; Sauquet et al., 2003 ; Richardson et al., 2004 ) have shown that Cananga (with a capitate stigma) is related to Ambavia and other ambavioids and not to Xylopia as found by Doyle and Le Thomas (1996) .

Ovule number (character 60) does not vary in Anaxagorea, but it is an informative character in the family, and the presence of two basal ovules is a conspicuous synapomorphy of Anaxagorea.

Fruit
Our treatment of fruit characters follows Doyle and Le Thomas (1996) . None of these characters vary in Anaxagorea. Recent work indicates that pseudosyncarpy (character 61) arose by different pathways within Annonaceae (Briechle-Mäck, 1994 ; Svoma, 1998 ; Chatrou and He, 1999 ), but because Annona is the only pseudosyncarpous member of the family sampled here, this is inconsequential for the present study. We scored the base of the monocarps (character 62) in Anaxagorea as stipitate, although the condition differs anatomically from that found in other Annonaceae (Doyle and Le Thomas, 1996 ). Given present outgroup relationships, it is equivocal whether the stipitate condition is homologous in Anaxagorea and other Annonaceae. Ventral fruit dehiscence (character 64) is a synapomorphy of Anaxagorea, not a primitive retention as sometimes thought, whereas the indehiscent state of other Annonaceae is homologous with that in related Magnoliales. The original thickness of the fruit wall in Annonaceae (character 65) was equivocal in Doyle and Le Thomas (1996) , but our results indicate that a thin fruit wall was ancestral, as in Anaxagorea, due to the different position of Cananga.

Seeds
We found no variation in seed characters in Anaxagorea, but we retained them in order to consider Anaxagorea in the context of other Annonaceae. Character definitions follow Doyle and Le Thomas (1996) , with two changes. Doyle and Le Thomas (1996) scored flat and grooved raphe (character 66) as separate states, but with our reduced sampling of Annonaceae we decided to merge the two states, as Artabotrys is the only taxon showing a grooved raphe, and it is polymorphic in being flat as well. Changing Annona from uncertain (flat or raised), as the Annona group was scored by Doyle and Le Thomas (1996) , to flat, as a result of exclusion of Anonidium, would not affect these results, since Annona is nested in a clade with a basically grooved or flat raphe. For similar reasons, in the aril character (70), absent and rudimentary were originally treated as two separate states but have been merged in this study. Johnson and Murray (1995) argued that the bilobed arils of Bocageeae (represented by Mkilua) are not homologous with the arils of Cananga and Xylopia. This is not a problem for the present constrained analyses, in which these three groups are separated, but it means that this character should be reevaluated in future broader-scale analyses.

A micropylar plug (character 67) is absent in Anaxagorea, but the presence of a large plug is ancestral in the rest of the family; in the absence of information on outgroups, the polarity of this character is uncertain. The presence of idioblasts in the inner integument (character 71) is a symplesiomorphy of Anaxagorea and Eupomatia, while the loss of idioblasts is a synapomorphy of the remaining Annonaceae.

The mesotestal fiber character (72) seemed potentially informative in Anaxagorea, because Christmann (1986) had reported that some species have longitudinal fibers and others have crossed fibers. The genus was therefore scored as polymorphic by Doyle and Le Thomas (1996) . However, we found the crossed arrangement in all Anaxagorea species examined. An explanation for this discrepancy is suggested by the fact that we saw only longitudinal fibers in some localized sections of the seed coat, but we observed the crossed arrangement after sectioning in several areas around the seed. Doyle and Le Thomas (1996) concluded that crossed mesotestal fibers were a synapomorphy of the Annonaceae and not homologous with the longitudinal fibers of Eupomatia, but with Eupomatia as the sister group of Annonaceae mesotestal fibers may be homologous of the two taxa. However, with the present character definition this is equivocal, since the next outgroups have no fibers, and the state in the common ancestor of Eupomatia and Annonaceae could be either lack of fibers, longitudinal fibers, or crossed fibers.

Endosperm ruminations of the irregular type (character 73), as in Anaxagorea, Ambavia, and Cananga, are ancestral in Annonaceae and homologous with those in the three closest outgroups. The structure of the ruminations (character 74) and seed shape (character 75) are two new characters in this study. Anaxagorea is unusual in having hollow, papery ruminations (Corner, 1949 ), as opposed to the more common solid ruminations, and this is a synapomorphy for the genus. Similarly, Anaxagorea is united by having two markedly asymmetrical seeds, with flat facing surfaces, whereas the seeds in the other Annonaceae are symmetrical.

Conclusion
This study has shown that many of the distinctive features that set Anaxagorea apart from other Annonaceae are indeed ancestral and homologous with similar features in other Magnoliales, particularly aspects of leaf architecture, midrib histology, inner staminodes, and stamen morphology. Deviations from these conditions occurred independently within Anaxagorea and in other Annonaceae. Some of these homoplastic changes appear to be systematically correlated, such as loss of the adaxial plate in the midrib and a shift from brochidodromous to eucamptodromous venation in A. silvatica and the Asian clade and a change from sessile to capitate stigma in the Asian clade. However, these apparent correlations may be misleading, since eucamptodromous venation also originated in several other lines within Anaxagorea, and a capitate stigma also arose in A. inundata. Similarly, a shift to truncate stamens and loss of inner staminodes both occurred in the Asian clade, but in different species (A. borneensis, A. luzonensis), and the same changes are seen in different species elsewhere in the genus.

We also identified a greater range of systematically useful variation in some characters than previously documented, although in a few cases apparently promising characters turned out to be uninformative (e.g., the reported variation in mesotestal fiber orientation was contradicted by the finding that all specimens examined had the same state). Many morphological synapomorphies for important clades have also been identified, such as capitate stigma and adaxial cuticle striations in the Asian clade, conical buds and shorter inner petals in the Central American clade, concave secondary veins and stellate trichomes in the core South American clade, foliar astrosclereids in the combined Central American and core South American clade, and numerous inner staminodes with glandular areas in a subgroup of the core South American clade.

Habit

 1. Habit (0) tree, (1) liana (D&L1). All Anaxagorea species are trees, but this character is variable among other Annonaceae and was retained for consistency in character inclusion.

Midrib histology

 2. Midrib histology (0) simple arc, (1) arc with unfused plate, (2) arc with fused plate (D&L5, D&E32).

 3. Protuberances of vascular tissue from adaxial midrib plate (0) absent or inconspicuous, (1) prominent.

Sources: Sugiyama (1976a <