ANALYSIS OF GENERIC RELATIONSHIPS IN ANACARDIACEAE

Cladistic analyses were undertaken of Anacardiaceae using non-sequence data (30 genera and 81 characters from morphology, anatomy, palynology and chemotaxonomy), sequence data (26 genera – rbcL) and a combined dataset of 16 genera. All analyses supported a group of genera which can be recognised at the subfamily level: Anacardioideae. Sequence data and combined datasets supported the recognition of a second subfamily: Spondiadioideae Kunth emend. Wannan. Both datasets also suggested that Buchanania lies outside both subfamily groups.


INTRODUCTION
The Anacardiaceae is a well recognised world-wide family of mostly tropical trees which has historically been placed in the Sapindales or Rutales (Bentham & Hooker, 1862;Takhtajan, 1980;Dahlgren 1980Dahlgren , 1983Dahlgren , 1989;;Cronquist, 1981;Angiosperm Phylogeny Group, 1998;Judd et al., 1999).Cronquist (1981) placed it with the Julianiaceae and Burseraceae, the three being the only families in the Sapindales with biflavonyls and vertical intercellular secretory canals in the primary and secondary phloem.The close relationships of the Anacardiaceae and Burseraceae have been recently reiterated by analysis of rbcL and atpB sequence data (Gadek et al., 1996;Savolainen et al., 2000a).The Burseraceae are distinguished by having two epitropous ovules per locule, in contrast to the one apotropous ovule in the Anacardiaceae.The Burseraceae also frequently possess lobed cotyledons in contrast to entire cotyledons in the Anacardiaceae.Amphipterygium and Orthopterygium, once a separate family (Julianiaceae), are now considered part of the Anacardiaceae based on molecular and non-molecular data (Peterson & Fairbrothers, 1983;Wannan & Quinn, 1988, 1990, 1991;Angiosperm Phylogeny Group, 1998;Judd et al., 1999, Savolainen et al., 2000b).
The Anacardiaceae is generally considered to constitute about 70 genera and 600 species which are concentrated in the tropics of Africa, Asia and America with a smaller number of species occurring in subtropical and temperate areas.A number of infrafamilial classifications have been proposed in the Anacardiaceae (Bentham & Hooker, 1862;Marchand, 1869Marchand, , 1874;;Engler, 1883Engler, , 1892)), but the most widely used for the last 100 years has been the five tribes of Engler (1883Engler ( , 1892Engler ( , 1897) ) which are based on floral characters and leaf dissection.A recent division of the family into five subfamilies by Takhtajan (1987) appears not to have been well accepted.
A range of systematic studies have tested the applicability of Engler's tribal classifi cation, the most comprehensive having used stem anatomy (Jadin, 1894) or wood anatomy (Heimsch, 1942;Dadswell & Ingle, 1948;Kryn, 1953).These found support, with some reservations, for Engler's tribes.Young (1976) looked at the wood flavonoids of 16 genera of the Anacardiaceae (including Julianiaceae), in the tribes Anacardieae, Rhoeae and Spondiadeae, as well as representatives of the Burseraceae, Rutaceae, Simaroubaceae, Cneoraceae, Meliaceae, Sapindaceae, Aceraceae, Hippocastanaceae and Juglandaceae.He found that there was a range of 5-deoxyflavonoids which was restricted to the Anacardiaceae (including Julianiaceae), but there were no clear patterns at a tribal level.
More recently, Wannan & Quinn (1990, 1991) described the pericarp and floral morphology in 30 genera sampling all tribes in the family.They found that the distribution of reproductive, vegetative and secondary product character states did not closely reflect the subfamily taxonomies of either Engler (1883Engler ( , 1892) ) or Takhtajan (1987).Rather, they found support for two informal groups, but suggested that these required further study to confirm their status.The first (Group A) included Engler's tribes Anacardieae (without Buchanania), Rhoeae (without Pentaspadon and Campnosperma), Dobineeae and Semecarpeae, and the second (Group B) included the Spondiadeae but with the addition of Buchanania, Pentaspadon and Campnosperma.Work by Von Teichman and associates has confirmed the importance of pericarp structure for illucidating generic affinities in the family (Von Teichman & Robbertse, 1986a, b;Von Teichman, 1987, 1990, 1991, 1992, 1993;Von Teichman & Van Wyck, 1988, 1994).Recent studies of seed anatomy (Pienaar & Von Teichman, 1998) and wood anatomy (Dong & Baas, 1993) have also provided support for Wannan & Quinn's (1991) groups.Support for Wannan & Quinn's (1991) two informal groups has also been provided by an unpublished analysis of anatomical, morphological and rbcL sequence data across 17 genera (Terrazas & Chase, 1996).Their conference abstract reported two clades, broadly corresponding to Wannan & Quinn's groups.Some of their sequence data were included in a molecular analysis of the Sapindales (Gadek et al., 1996) which used 7 genera from the Anacardiaceae and three from the Burseraceae.This analysis showed the Anacardiaceae and Burseraceae as sister groups, and two main clades in the Anacardiaceae corresponding to Group A and B but with Buchanania diverging prior to both.Other sequence data (rbcL: Chayamarit, 1997) from an analysis of 16 Thai genera has also provided some support for the informal groups of Wannan & Quinn (1991), but the absence of bootstrap or decay analysis made it impossible to assess the strength of support for their clades.Recent sequence data from the internal transcribed spacer region from the ribosomal DNA has provided an indication of relationships amongst genera referred to Engler's Rhoeae or Wannan & Quinn's (1991) subgroup A2 (ITS: Miller et al., 2001).American species of Rhus s.s.(subgenera: Lobadium and Rhus) were shown to be closely related and more distant from other genera of the Rhoeae including Actinocheita, Cotinus, Malosma, Schinus, Searsia and Toxicodendron.
The relationships of some genera of the Rhoeae were also investigated by Aguilar-Ortigoza et al. ( 2004) using non-sequence data.Their main focus was on the 6 species of Pseudosmodingium from Mexico which were shown to be most closely related to Bonetiella, also from Mexico.However, the larger clade, corresponding to genera of the Rhoeae, included Smodingium (Africa) and Mexican representatives of Cardena siodendron, Cotinus, Rhus and Toxicodendron.A subsequent paper by a similar team (Aguilar-Ortigoza & Sosa, 2004) combined sequence (rbcL) and non-sequence data for 22 genera of Anacardiaceae.Both separate datasets show good support for Wannan & Quinn's Group A (18 genera, including Anacardium and Mangifera from Engler's tribe Anacardieae) but less support for Group B (3 genera).In both analyses Buchanania is placed as a sister taxon to genera in Group A, but outside the clade with Group B genera.Interestingly, the paper compares the Anacardiaceae clade with a clade of hemipteran insects (Calophya spp.) which feed on the family; the later shows closely related species feeding on Spondias and Buchanania.A conference abstract (Pell & Urbatsch, 2001) describing analyses of sequence data from the chloroplast genome (matK, trnL and the intergenic spacer between the trnL exon and trnF) has strongly supported the two groups of Wannan & Quinn (1991).Pell & Urbatsch (2001) reported two major clades in the family: one with members of the tribes Rhoeae, Semecarpeae, Dobineeae and Anacardieae, and the other with members of the Spondiadeae and a few members of the Rhoeae.They also reported that the Anacardiaceae proved to be monophyletic.Thus, there are some data which support the proposed infrageneric classification of Wannan & Quinn (1991).As yet, however, there has not been any broad analysis of generic relationships using the range characters which are known for the family.This paper analyses the available morphological, anatomical, chemotaxonomical, cytological, palynological characters and the available sequence data (rbcL) and aims to test support for the two informal subfamily groups proposed by Wannan & Quinn (1991) and identify key data gaps in the family.

Non-sequence data
The terminal taxa used in this analysis are genera (Table 1) with characters scored from usually more than one species.The characters used are listed in Table 2 and Table 3a, b is the data matrix.A list of the autapomorphies is provided in Table 4.Many descriptions of character states were reviewed for each taxon.In a few cases where differing character states were argued in the literature, these were scored as multiple states with each source cited.In most cases, however, a single reliable authority has been cited in Table 3a, b following a critical analysis of the literature.Some unpublished data are included and are supported by vouchers in Appendices A and B.

Sequence data
Sequences for the chloroplast encoded rbcL gene were obtained for a subset of taxa either from GenBank or from the sources cited in Table 5.Sequences were aligned in PAUP* (Version 4.0b10;Swofford, 2002).

Analyses
Heuristic parsimony analyses were performed in PAUP* set for tree bisection reconnection branch swapping on the best trees.Multistate characters were treated as polymorphisms.Multiple replicates of random taxon addition were employed to search for multiple islands of trees, and the CONDENSE option was employed to delete duplicate trees.Support for clades was inferred using the bootstrap option in PAUP* (Felsenstein, 1985) with 500 replicates, and also by decay values (Bremer, 1988;Donoghue et al., 1992).Decay command files were created in MacClade version 4.05 (Maddison & Maddison, 2002) and executed in PAUP* using 10 replicates of random taxon addition on each constraint tree.Output trees from PAUP* were transferred to MacClade and manipulated to test other topologies and explore character state evolution.2) Wannan & Quinn, 1991.
Only unambiguous character state changes were recorded on the branches in the final figures.

Choice of taxa
Ideally terminal taxa should be species, so that both generic concepts and intergeneric relationships could be tested in the cladistic analyses.Unfortunately, nonmolecular data are not available for most of the species in the family.In fact, the full range of characters has yet to be scored for a single species.The limited data that are available have been assembled piecemeal by many workers as suitable material serendipitously has come to hand.Hence, in order to obtain a preliminary estimate of the phylogenetic signal in the available data, genera are used as the terminal taxa in the nonmolecular analysis, with the data drawn from one or more species.Genera were chosen primarily on availability of data, preferably from a number of authors.Thirty genera of Anacardiaceae are included representing all five tribes (Engler, 1892) and both informal subfamily groups (Wannan & Quinn, 1991).The outgroup for the analysis comprised three genera from the Burseraceae (Bursera, Canarium and Garuga) representing the three tribes (Leenhouts, 1955;Forman et al., 1994).The details of the genera used are provided in Table 1.Sequence data were available for representative species of only 16 of the genera included in the nonmolecular analysis (Table 5).Exemplars of three outgroups, Burseraceae, Rutaceae and Sapindaceae, were included, with trees being rooted on the last two.
Additional information is provided below for some characters (Table 2) (5) Leaves -The outgroup and most Anacardiaceae have imparipinnate leaves (b).Fewer genera have simple leaves (a), though in some genera there are species with both.Paripinnate leaves (c) are very uncommon, occurring in some species of genera that possess mostly imparipinnate leaves.Bipinnate leaves occur only as polymorphic character states in Spondias (S. bipinnata; Airy Shaw & Forman, 1967) and Bursera (B.bipinnata; Porter, 1970) and have not been scored.
(10) Flower sex -Most genera of the Anacardiaceae, and probably Burseraceae, have unisexual flowers (b), though frequently with the aborted remnants of the other sex present.Fewer have bisexual (a) or polygamous (coded a & b) flowers.Botanists have frequently confused the sex of flowers and it is likely that the flowers of some genera which have been recorded as polygamous, and which have not been closely studied, will be found to be unisexual.In many cases the male flowers are clearly unisexual (with an aborted smaller ovary) while the female flowers appear to be bisexual but the stamens have aborted anthers which are apparent only after sectioning (Wannan & Quinn, 1992).In genera with very strongly dimorphic unisexual flowers (i.e.parts absent in one sex) the number of parts has been scored from the sex in which they are present (e.g.Dobinea).In Amphipterygium the female flowers have no perianth.The male flowers are recorded as apetalous with a 5-8-partite calyx and a similar number of alternisepalous stamens (Hemsley, 1908;Hutchinson, 1959;Cronquist, 1981).Given the absence of alternisepalous stamens, and the frequent occurrence of antesepalous stamens elsewhere in the family, the perianth of the male flowers of Amphipterygium is here interpreted as representing a corolla.The Burseraceae are recorded with 3-, 4-or 5-partite calyces.
(12) Corolla number -Most genera of the Anacardiaceae have a 5-partite corolla (b), but in a few it is 6-partite (a), 4-partite (c) or 3-partite (d).Many genera are polymorphic.The Burseraceae are recorded as 3-, 4-or 5-partite.Pistacia has been recorded as having no corolla (-).In male flowers of Pistacia there is a single perianth whorl of mostly 4 (rarely 5) segments (pers.obs. on P. chinensis).The stamens are equal in number to, and stand opposite, the perianth segments suggesting that, as in other genera of the Anacardiaceae (e.g.Cotinus, Semecarpus), there is a single antesepalous whorl of stamens (and hence no corolla).This conclusion is supported by Bentham & Hooker (1862), Baillon (1878), Standley & Steyermark (1949), Rechinger (1969) and Siddiqi (1978).Payer (1857) and Eichler (1878) also considered that only the calyx is present in female flowers.The reporting of alternitepalous stamens in P. malayana (Hou, 1978) suggests that this character needs examination across all species attributed to this genus.In genera with strongly dimorphic unisexual flowers (i.e.parts absent in one sex) the number of parts has been scored from the flower in which parts are present (e.g.Dobinea).
(17) Nectariferous disc -Most of the outgroup and Anacardiaceae have an intrastaminal disc (a).A small number of genera have an extrastaminal disc (b) or no sign of any disc at all (c).This character can be difficult to interpret in unisexual male flowers where there are no carpellodes.However, in most cases the disc is still apparent, or the character can be scored from assessment of the sterile stamens in the female flower.
(18) Floral axis -The outgroup and most Anacardiaceae have an undeveloped floral axis with an hypogynous flower (a).In these, the nectariferous disc encircles the ovary with the stamens inserted on its outer edge (i.e.intrastaminal).Some genera in both families show a tendency towards perigyny (b), where the lower parts of the perianth are fused and the stamens are connate, or adnate to the fused perianth or to the disc, forming a circular column around the ovary (e.g.Canarium, Garuga, Melanochyla, Thyrsodium).In the Anacardiaceae, true epigyny (inferior ovary) is restricted to Drimy carpus and Holigarna, which are not included in this analysis.
(19) Occurrence of androgynophore -These are absent (a) from most Anacardiaceae and Burseraceae.An androgynophore (b) occurs in some species of Mangifera, Swin tonia and Dobinea (and also Gluta which is not included in this analysis).The occurrence of a gynophore has not been scored as it occurs only in Garuga.
(20) Carpellode number -Evidence of carpellodes varies considerably from welldeveloped carpellodes (with small locules) through to very small bulges of remnant carpellode tissue (Wannan & Quinn, 1991).Common states are those which reflect the number of carpels in the female flowers, i.e. 5 (a), 3 (c) or 1 carpellode (e).In some cases there appears to have been reduction of the vestigial carpels so much that none are evident (Amphipterygium, Astronium, Pistacia, Rhodosphaera, Semecarpus); this condition is scored as inapplicable (-).The character is also not applicable in bisexual flowers.This character has not been widely scored in the outgroup.
(21) Carpellode position -The outgroup and most Anacardiaceae often have their carpellodes reflecting the position of carpels in the female flowers: either standing opposite the petals (a), three carpellodes in a flower with a 5-partite perianth (b), or standing opposite the sepals (c).This character is not applicable (-) in bisexual flowers and in those where none are evident (as above).
( 22) Carpel number -Ranges in the Anacardiaceae from 13 in Pleiogynium (an autapomorphy not coded for in this analysis) to a single carpel (e).Abortion of carpels leading to pseudomonomery is widespread in the families.This character has been inferred from the number of styles or stigmas, often with anatomical support from vascular bundles in the ovary wall and/or abortive locules (Wannan & Quinn, 1991).
(23) Position of fertile carpel -One of the features of the outgroup and many Anacardiaceae is that, even in multicarpellary flowers, there is only one fertile seed produced.
The remaining carpels abort following anthesis or were never properly formed (Wannan & Quinn, 1991).In the outgroup and many Anacardiaceae the fertile carpel is antepetalous (a).In other Anacardiaceae the fertile carpel is antesepalous (b).
(24) Number of locules at anthesis -This character scores locules whether or not they are fertile and provides some indication of the degree of carpel abortion occurring during development.The greatest number of locules is in Pleiogynium, i.e. up to 12 (an autapomorphy not coded in this analysis); other genera have 1-5 locules.
(25) Carpel connation -There is a varying degree of carpel connation across the Anacardiaceae from apocarpus gynoecia in Buchanania (a), to genera in which the ovaries and parts of the styles are fused (c).Most genera in the Anacardiaceae have gynoecia in which the ovaries are connate and the styles free (b).The Burseraceae have almost complete connation with even the stigmas partly fused (d).
(26) Carpel definition -This character refers to the furrows between the individual carpels in the ovary (a) and is a feature of Buchanania, Dracontomelon and Spondias.
Although not present (b) in most Anacardiaceae or Burseraceae it does occur widely in more distant outgroups such as the Rutaceae and Sapindaceae.
(28) Stigma morphology -There is a wide variety of stigma morphologies in the family but capitate or clavate stigmas (b) are the most common condition in the Anacardiaceae.
More restricted morphologies include: -an oval opening at the top of each carpel as in Dracontomelon (a), which is generally characteristic of those genera in which there is good carpel definition but where the style gradually merges with the ovary, and -spathulate (c), which appears to be a development of the capitate condition (e.g.

Amphipterygium, Pistacia).
There are a number of autapomorphs which have been coded as inapplicable (-, see Table 4).The stigma morphologies of the Burseraceae are not well known.Garuga has a Dracontomelon-type of stigma (a).There has been no anatomical investigation of the crown-like angular stigma in Canarium that appears to be derived from the Dracontomelon-type.It may occur more widely in the Burseraceae.
(31) Ovule insertion - Robbertse et al. (1986) have suggested that the differing positions of funicle insertion on the locule wall may be related to the abortion of carpels alongside the fertile carpel (they cite Mangifera).While this may be true in some genera, it appears not to be the case in Pistacia, where on occasions there are two fertile carpels produced both with basally attached ovules (fig.6d in Grundwag, 1976), nor in Astro nium, where there is only a single fertile carpel (and two sterile carpels as evidenced from two other styles and stigmas) with an apically attached ovule.
(32) Number of ovule integuments -This character has not been widely scored.The outgroup and most Anacardiaceae have two ovule integuments (a).Much less common is one (b), and very rarely two on the outside and one on the inside (Pistacia).There are suggestions that a single integument has been derived from the bitegmic condition, but there is little agreement on how this may have occurred.Robbertse et al. (1986), working on Mangifera, have suggested that the single integument condition may be a neotonic form, but Copeland (1962) indicated that the single integument in Anacardium exhibited features that indicated it was the fusion product of two integuments, and Von Teichman (1990) suggested that in Tapirira there has been reduction of the inner integument.Hence, the single integument state may not be homologous in all taxa.
(33) Micropyle orientation -Micropyle orientation is upwards or superior (a) in the outgroup and most Anacardiaceae, but an inferior micropyle (b) occurs in some genera with a basally attached funicle.In Pistacia the micropyle is initially inferior but becomes superior during development (Marchand, 1869;Copeland, 1955).Both states were coded in this genus.
(34) Winged fruit -In the analysis winged fruit (b) occur only in Loxopterygium and Schinopsis.The ciliate margins in the fruit of Blepharocarya and the membranous margin in Dobinea have been interpreted as unwinged.
(35) Postanthetic growth of hypocarp -Occurs only in Anacardium and Semecarpus (b).The outgroup and most Anacardiaceae have no significant postanthetic growth of the perianth or floral axis (a).Postanthetic growth of the calyx occurs in Astronium (and also Parishia, not in this dataset).Postanthetic growth of the corolla occurs in Swintonia (and also Gluta, not in this dataset).These latter two conditions (calyx and corolla) were not coded as they are autapomorphs (see Table 4).
(36) Number of seeds in fruit -Most Anacardiaceae have a 1-seeded fruit (e).Fewer genera have 5 (a), 4 (b), 3 (c) or 2 (d) seeds.Pleiogynium sometimes has 12 seeds, which is double the number of parts in either perianth whorl.This autapomorphy has not been coded in this analysis.
(37) Operculum -This character refers to the small cap or lid which covers an opening to the locule in some of the larger heavily lignified fruits of the Anacardiaceae and Burseraceae, and which is usually dislodged during germination (Hill, 1933(Hill, , 1937)).Most Anacardiaceae have no operculum (a).The outgroup and some Anacardiaceae have an operculum (b) consisting of either a single piece (Canarium, Cyrtocarpa, Dracontomelon, Lannea p.p., Pleiogynium, Spondias) or two pieces (Harpephyllum, Lannea p.p.).This feature requires further study, to establish homologies especially in the Burseraceae, where there appear to be differences in splitting of the fruit at germination (Ng, 1975).
(50) Testa consistency -The outgroup and most Anacardiaceae have a membranous testa (a).Only some species of Semecarpus have a coriaceous testa that is scored as non-membranous (b).This character is uninformative for the genera in this analysis, but was scored to cite its occurrence across the family.
(51) Testa connection to endocarp -The outgroup and most Anacardiaceae have a testa which is free from the endocarp (a), but Swintonia (and species of Bouea, Drimycarpus, Melanochyla, Parishia, Rhus which are not in this study) have an adherent testa (b) and some Mangifera species have a labrynthine testa (c).This character is uninformative for the genera in this analysis, but was scored to cite its occurrence across the family.
(58) Resin canals in the phloem -Vertical intercellular secretory canals occur in the primary and secondary phloem (a) of all Anacardiaceae and Burseraceae.They are apparently absent (b) from all other members of the Sapindales/ Rutales.This character was scored in Table 3 to confirm its occurrence across all of the genera.
(69) Chromosome number -There is a wide variety of chromosome number (2n) in both the Anacardiaceae and Burseraceae.The diploid numbers are scored as there has been very little analysis of base numbers in the family.
(70 -81) Occurrence of flavonoids and biflavonoids -These have been recorded across both Anacardiaceae and Burseraceae.A range of 5-deoxyflavonoids occur in the heartwood of some Anacardiaceae and Burseraceae (Young, 1976).Biflavonoids have also been reported from the leaves of some Anacardiaceae and Burseraceae (Wannan & Quinn, 1991;Graham, unpubl.).

Autapomorphies
Table 4 lists autapomorphies that were not included in the analysis.Some of these autapomorphies may be synapomorphies in the broader context of the family.3a.Character matrix.Character states as in Table 2; -inapplicable; ?unknown; polymorphisms within parentheses.

Genera Autapomorphies
Amphipterygium -no perianth in female flowers female flowers and fruits enclosed in an involucre (dissimilar to Blepharo carya) Astronium -postanthetic growth of calyx Blepharocarya -small stigmatic area on ventral side of the stylar apex ciliate margins of fruit female flowers and fruits enclosed in an involucral inflorescence (dissimilar to Amphipterygium) Buchanania -fertile carpel in Buchanania has no stigma, therefore the number of stigmas is equal to one less than the number of carpels /styles.Campnosperma -discoidal stigma with irregular lobes formed seedless second locule in fruit which is not evident in the ovary at anthesis Canarium -diffuse parenchyma (presumably apotracheal) in wood of some species Cotinus -postanthetic pedicel development of aborted flowers (giving rise to smokey appearance) Dobinea -no perianth in female flowers stigma is small area on the ventral surface of the style extending downwards from the tip membranous margin on the fruit Garuga -gynophore Mangifera -stigma is a reduced area at the tip of the style labyrinthine testa in some species Pentaspadon -globose stigma Pleiogynium -carpel number twice the number of corolla segments (i.e. up to 12 locules) Semecarpus -postanthetic growth of hypocarp Swintonia -postanthetic growth of corolla adherent testa

RESULTS AND DISCUSSION
Nonsequence data A heuristic analysis with 500 replicates of random taxon addition gave a single island of 24 equally parsimonious trees of 510 steps with a rescaled consistency index of 0.53 and a retention index of 0.64.The majority rule consensus is shown in Fig. 1 together with the results of a decay analysis.Character evolution was investigated in MacClade and is shown in Fig. 2.
Towards the base of the tree there are 10 genera (Buchanania, Campnosperma, Cyrto carpa, Dracontomelon, Harpephyllum, Lannea, Pentaspadon, Pleiogynium, Spondias, Tapirira) referred by Wannan & Quinn (1991) to Group B which corresponds to Engler's Spondiadeae with the addition of Campnosperma, Buchanania and Pentaspadon.Relationships between them are not fully resolved and there is no clustering of the lineages.

Characters
Of the 81 characters used, 2 were constant (16, 58), 15 were non-informative (3,4,6,8,11,12,18,19,27,35,50,51,53,70,80) and the remaining 64 were informative.Indications from the remaining unscored genera in the family suggest that a number of the uninformative characters will be informative in a larger dataset or when the character is scored for all 30 taxa in the dataset.Overall, the breadth of the analysis was limited by absence of data for many genera.Although taxa for the analysis were chosen on the basis of their characters being known, there were still gaps in certain characters.In other cases, there were disagreements in the scoring of character states between authors, which have been generally coded as polymorphisms.There is little doubt that the scoring of currently absent states and the checking of apparently poly-morphic taxa will improve the phylogenetic signal.Two other problems related to character scoring may be assisted by a wider analysis of species or genera.The first was the need to code polymorphisms for pericarp characters where a discrete cell layer was sometimes variably present or variably lignified (e.g.character 45).The second was the occurrence of character states which were apparently unique synapomorphies for clades except for a polymorphic occurrence in a genus outside the clade.Most of these isolated polymorphic occurrences were checked as closely as possible, but the following could not be checked and appear questionable based on the overall phylogenetic signal for these taxa: 3 carpellodes (20c) and 1 locule at anthesis (24e) in Cyrtocarpa, and presence of Fustin in heartwood (75a) in Amphipterygium.2.

Sequence data
The 27 taxa for which rbcL sequences were available (Table 5) include species of 16 genera of Anacardiaceae, representing both Group A and B, and three of the five tribes, three genera of Burseraceae, and seven outgroup genera from Sapindaceae and Rutaceae.Wannan & Quinn, 1991.4) Leenhouts, 1955;Forman et al., 1994.
The first 28 base pairs (bp) of the aligned database were excluded from the analysis because they included one of the primer sites or were missing in many taxa.Combined data An analysis of combined sequence and non-sequence data was undertaken for 16 taxa using Bursera and Canarium as outgroups.A heuristic analysis of 500 replicates of random taxon addition gave a single tree of 593 steps (CI = 0.71 excluding uninformative characters, RI = 0.64 and RC = 0.53) as shown in Fig. 4, together with the results of bootstrap and decay analyses.This combined tree shows a strong clade (BS 100%, decay +11) recognising the family Anacardiaceae.The ingroup comprises two weakly supported clades, one with 10 genera of Group A and Buchanania (Group B), and another with the remaining three genera of Group B. The larger clade contains a wellsupported subclade (100%, +10) of Group A genera with Buchanania as sister taxon.Within this Group A clade there are subclades comprising Astronium and Schinopsis (100%, +12), Amphipterygium and Pistacia, (93%, +3), Cotinus and Toxicodendron (75%, +3), and Anacardium and Mangifera (94%, +4).The second Anacardiaceae clade is a weakly supported clade (67%, +2) comprising three Group B genera from Engler's Spondiadeae.In this clade, Spondias is again a sister group to a weakly supported subclade (65%, +2) with Tapirira and Cyrtocarpa.

Taxonomic implications
The rbcL data indicates that Burseraceae is closely related to the Anacardiaceae, suggesting that it lies within the ingroup which includes genera of Anacardiaceae.However, the combined data suggest that Burseraceae is well differentiated from the Anacardiaceae.Sequence data from other regions of the chloroplast genome (Pell & Urbatsch, 2001) also support a monophyletic Anacardiaceae, distinct from its sister group the Burseraceae.
Non-sequence and sequence data provide some support for the subfamily groups of Wannan & Quinn (1991)   Group A by Wannan & Quinn (1991).This clade includes Blepharocarya as well as genera previously referred to the Julianiaceae.Both datasets suggest that Buchanania is a weakly linked sister taxon to Group A rather than to Group B (Wannan & Quinn, 1991).In the non-sequence data the Group A clade also includes genera from the tribes Dobineeae and Semecarpeae, which were not sampled by the sequence data.The non-sequence data suggested that there are some closer relationships amongst Group A genera, some of which are supported by rbcL data (Astronium and Schinop sis, Amphipterygium and Pistacia, Anacardium and Mangifera).Closer relationships within Group A may be further resolved by sequence data from other areas (e.g.internal transcribed spacer region; Miller et al., 2001).Neither dataset provides strong support for the recognition of a clade comprising taxa from Wannan & Quinn's (1991) Group B (Fig. 3, 4).In the non-sequence data these genera occur as a basal polyotomy and in the sequence and combined data there is a weakly supported clade with Cyrtocarpa, Spondias and Tapirira, from Group B. However, support for Group B is provided by sequence data from other areas of the chloroplast genome (Pell & Urbatsch, 2001).Sequence and non-sequence data do not support Buchanania within Group B as suggested by Wannan & Quinn (1991).Rather, Buchanania appears as a sister taxon weakly clustered with Group A in all datasets.However, there is no strong suggestion from any dataset that Buchanania has any close affinities with genera of the Anacardieae where it was placed by Engler.The recognition of subclades amongst the genera of Wannan & Quinn's (1991) subgroup A1 will require a more widespread survey of the three tribes Anacardieae, Dobineeae and Semecarpeae.Initial indications, however, that Gluta and Bouea, with apotracheal wood parenchyma (Dadswell & Ingle, 1948;Kryn, 1953) and similar pericarp anatomy and floral anatomy (Wannan & Quinn, 1990, 1991) may be close to Mangifera and Swintonia.This suggestion is supported by the analysis of rbcL data by Chayamarit (1997), whose sequences were not made available for inclusion in this analysis.
Both data sources support the recognition of a large clade which can be recognised at the subfamily level and includes genera from Group A (or Engler's Anacardieae, Dobineeae, Rhoeae and Semecarpeae).This subfamily, here designated Anacardioideae, appears to be defined by the possession of a single unique synapomorphy (23b antesepalous fertile carpel), although this character is yet to be scored in some genera.Many members of this subfamily also possess a stratified endocarp.The combined data show weak support for the recognition of a second subfamily group based on Group B (without Buchanania) or on Engler's Spondiadeae that is here designated as Spondiadioideae Kunth emend.Wannan.This subfamily is also supported by sequence data from other regions of the chloroplast genome (Pell & Urbatsch, 2001).In this analysis however, there are no nonsequence synapomorphies that would define this second subfamily; instead the subfamily appears paraphyletic.The subfamily can be recognised, however, by the retention of the plesiomorphic (nonstratified) pericarp condition in the family.There are suggestions that the study of the operculae in the Anacardiaceae and Burseraceae may provide some synapomorphic support for Group B. Initial observations suggest that the operculae in the Anacardiaceae and Burseraceae may not be homologous, but hard data in this area are lacking.Some of Engler's tribes may be recognisable in modified form within the subfamily taxonomy proposed above.The nonsequence data appear to support Engler's tribes Rhoeae (without Campnosperma and Pentaspadon) and Anacardieae (without Buchanania).However, the absence of support for these from rbcL data, and from the combined dataset, suggests that any tribal boundaries will need to be determined from a wider survey of taxa and characters.There are as yet no indications of tribes within the subfamily Spondiadioideae.There remain several genera (Buchanania, Campnosperma, Pentaspadon) whose subfamily position remains unclear.Placed as a sister taxon to the Anacardioideae in all analyses, Buchanania may constitute a separate subfamily.Further sequence data for all three genera would appear to be crucially important as all three possess morphological and anatomical characters which appear intermediate between the two subfamilies.An analysis of rbcL data by Chayamarit (1997) showed Buchanania in a clade with Pentaspadon and Rhus, but the degree of support for this clade was not shown.

Fig. 3 .
Fig. 3.One of the eight equally parsimonious trees of 541 steps in a single island found from a heuristic search of the rbcL database with branch lengths proportional to the amount of change; consistency index = 0.54 excluding uninformative characters; retention index = 0.69; rescaled consistency index = 0.48.Arrows indicate branches collapsing in the strict consensus.Bootstrap percentages > 50% shown above the branches; decay values > 1 shown below shown below branches.Also showing Wannan & Quinn's (1991) subfamily groups, Engler's tribes.Abbreviations as in Fig.1.

Fig. 4 .
Fig. 4. Single most parsimonious tree of 593 steps found from heuristic searching of the combined database with branch lengths proportional to the amount of change; consistency index = 0.71 excluding uninformative characters; retention index = 0.64; rescaled consistency index = 0.53.Bootstrap percentages > 50% shown above the branches; decay values > 1 shown below shown below branches.Also showing Wannan & Quinn's (1991) subfamily groups and Engler's tribes.Abbreviations as in Fig.1.

Table 5 .
Taxa included in the molecular analysis.
and Engler's tribes.Both datasets support placement of most genera from Engler's tribes Anacardieae and Rhoeae in a single group recognised as