TRIBES AND CLADES WITHIN APIACEAE SUBFAMILY APIOIDEAE : THE CONTRIBUTION OF MOLECULAR DATA

Phylogenetic analyses of chloroplast gene (rbcL, matK), intron (rpl16, rps16, rpoC1) and nuclear ribosomal DNA internal transcribed spacer (ITS) sequences and chloroplast DNA restriction sites, with supplementary data from variation in size of the chloroplast genome inverted repeat, have been used to elucidate major clades within Apiaceae (Umbelliferae) subfamily Apioideae Drude. This paper summarizes the results of previously published molecular cladistic analyses and presents a provisional classification of the subfamily based on taxonomic congruence among the data sets. Ten tribes (Aciphylleae M. F. Watson & S. R. Downie, Bupleureae Spreng., Careae Baill., Echinophoreae Benth., Heteromorpheae M. F. Watson & S. R. Downie, Oenantheae Dumort., Pleurospermeae M. F. Watson & S. R. Downie, Pyramidoptereae Boiss., Scandiceae Spreng. and Smyrnieae Spreng.) are erected or confirmed as monophyletic, with Scandiceae comprising subtribes Daucinae Dumort., Scandicinae Tausch and Torilidinae Dumort. Seven additional clades are also recognized but have yet to be treated formally, and at least 23 genera examined to date are of dubious tribal or clade placement. The utility of these different molecular markers for phylogenetic inference in Apioideae is compared based on maximum parsimony analyses of subsets of previously published molecular data sets. Of the six loci sequenced, the ITS region is seen to be evolving most rapidly and rbcL is the most conservative. Intermediate in rate of evolution are matK and the three chloroplast introns; with rpl16 and rps16 evolving slightly faster than matK or rpoC1. The analysis of restriction sites, however, provided 2–4 times more parsimony informative characters than any single DNA locus sequenced, with estimates of divergence just slightly lower than that of the ITS region. The trees obtained from separate analyses of these reduced data sets are consistent with regard to the major clades inferred and the relationships among them. Similar phylogenies are obtained by combining data or combining trees, representing the supermatrix and supertree approaches to phylogenetic analysis, respectively. The inferred relationship among the tribes and informally recognized major clades within Apioideae is presented.

Tausch and Torilidinae Dumort.Seven additional clades are also recognized but have yet to be treated formally, and at least 23 genera examined to date are of dubious tribal or clade placement.The utility of these different molecular markers for phylogenetic inference in Apioideae is compared based on maximum parsimony analyses of subsets of previously published molecular data sets.Of the six loci sequenced, the ITS region is seen to be evolving most rapidly and rbcL is the most conservative.Intermediate in rate of evolution are matK and the three chloroplast introns; with rpl16 and rps16 evolving slightly faster than matK or rpoC1.The analysis of restriction sites, however, provided 2-4 times more parsimony informative characters than any single DNA locus sequenced, with estimates of divergence just slightly lower than that of the ITS region.The trees obtained from separate analyses of these reduced data sets are consistent with regard to the major clades inferred and the relationships among them.Similar phylogenies are obtained by combining data or combining trees, representing the supermatrix and supertree approaches to phylogenetic analysis, respectively.The inferred relationship among the tribes and informally recognized major clades within Apioideae is presented.

I
'Although Umbelliferae were the first family of flowering plants to achieve general recognition, after nearly three and a quarter centuries of successive and multinational effort, considerable disagreement still exists as to the proper delimitation of the family and even more uncertainty prevails as to its natural subdivisions and the criteria on which they should be erected.Clearly continued acquisition of new information and re-examination, refinement, and re-evaluation of accumulated evidence are urgently to be welcomed.'L. Constance (1971) The higher-level relationships within Apiaceae (Umbelliferae) have been traditionally difficult to resolve, particularly within its largest subfamily Apioideae Drude.
Comparison of the accounts of, for example, Koch (1824), de Candolle (1830), Bentham (1867), Boissier (1872), Drude (1898), Calestani (1905) and Koso-Poljansky (1916), which were erected largely on the basis of fruit morphology and anatomy, shows widely diverging opinions on the definition and composition of its tribes and subtribes.While such focus on fruit structure has been rejected by many (Heywood, 1971b(Heywood, , 1978a;;Theobald, 1971;Davis, 1972;Cronquist, 1982;Hedge et al., 1987;Shneyer et al., 1992Shneyer et al., , 1995)), the highly criticized century-old system of Drude ( 1898), or some modification thereof (Pimenov & Leonov, 1993), remains the most commonly used treatment.As such, many of the authors who published papers as a result of two international symposia on the family (Heywood, 1971a;Cauwet-Marc & Carbonnier, 1982) presented their findings in the framework of Drude's system.Drude (1898) divided Apiaceae into three subfamilies (Apioideae, Hydrocotyloideae Link and Saniculoideae Burnett), recognizing 8 tribes and 10 subtribes within Apioideae.Molecular phylogenetic studies, however, while confirming the monophyly of subfamily Apioideae, have shown that many of its tribes and subtribes are not monophyletic (Downie & Katz-Downie, 1996;Kondo et al., 1996;Plunkett et al., 1996aPlunkett et al., ,b, 1997;;Downie et al., 1996Downie et al., , 1998Downie et al., , 2000a,b,c;,b,c;Valiejo-Roman et al., 1998;Katz-Downie et al., 1999;Plunkett & Downie, 1999).The apioid umbellifers display a remarkable array of morphological and anatomical modifications of their fruits, some of which are likely adaptations for various modes of seed dispersal (Jury, 1986).Not surprisingly, these characters are prone to convergence and their almost exclusive use to delimit suprageneric taxa has confounded attempts to identify monophyletic groups.Our goal over the past six years has been to resolve the higher-level relationships within subfamily Apioideae.This is necessary in order to provide the framework for lower-level revisions, as well as to interpret patterns in the evolution of the many phytochemical, anatomical, cytological, morphological and palynological characters available for the group (such as those summarized in Heywood, 1971a, andCauwet-Marc &Carbonnier, 1982).Although these characters have been surveyed widely, trends in their evolution and their reliability in demarcating taxonomic groups have rarely been considered outside of the framework of Drude's system (for exceptions see Plunkett et al., 1996b, andKatz-Downie et al., 1999).Eventually, the synthesis of these newly acquired molecular data with a re-evaluation of existing evidence will culminate in the production of a modern classification for the subfamily.
This paper has three major objectives.First, we summarize the results of a variety of previously published macromolecular analyses as they apply to clarifying the evolutionary history of subfamily Apioideae.To date, major clades within the subfamily have been elucidated on the basis of phylogenetic analyses of chloroplast gene (rbcL, matK ), intron (rpl16, rps16, rpoC1) and nuclear ribosomal DNA ITS sequences and chloroplast DNA (cpDNA) restriction sites, with supplementary data obtained from variation in size and position of the chloroplast genome inverted repeat.Based on taxonomic congruence among these previously published data sets, a provisional classification of the subfamily is proposed.Second, by focusing on subsets of these molecular data, we compare the utility of different molecular markers for phylogenetic inference within the subfamily.Third, based on separate and combined analyses of these subset data (or the trees derived from them) and the results of our earlier publications using non-reduced data sets, the phylogenetic relationships among the major clades comprising Apioideae are inferred.While our studies are still very much in progress, the results obtained to date provide the necessary framework and explicit phylogenetic hypotheses from which future revisions and evolutionary studies can proceed.Authorities for names at rank genus and below are given in Tables 2-5; authorities for suprageneric taxa and others not in these tables are given when first used.

S  P M A
The earliest comparative studies of nucleic acids in Apiaceae were the DNA-DNA hybridization experiments of Valiejo-Roman et al. (1979, 1982) and Antonov et al. (1988; Table 1).The groups obtained on the basis of these hybridization data, while supporting the distinctiveness of subfamily Apioideae and its separation from subfamily Saniculoideae, stood in stark contrast to any existing system of classification available for Apioideae.DNA distances among the resolved apioid groups, however, were sufficiently low and the sampling too sparse to propose an alternative tribal treatment.Relationships within the family were also inferred using immunological comparisons of seed storage proteins.The earliest of these studies (Pickering & Fairbrothers, 1970) supported Drude's division of the family into three subfamilies; it also indicated that Apioideae are more similar serologically to Saniculoideae than to Hydrocotyloideae.Further analyses by Pickering & Fairbrothers (1971) confirmed five serological groupings for nine genera within Apioideae, corresponding to Drude's    et al. (1991, 1992, 1995), also based on immunochemical reactions of seed storage proteins but with greater sampling in tribes Peucedaneae and Smyrnieae Spreng., clearly indicated that many of the tribes and subtribes recognized by Drude do not form monophyletic groups.Subtribes Angelicinae and Peucedaninae were determined to be serologically not very well separated, whereas subtribe Tordyliinae was maintained as distinct.
The past several years has seen a flurry of molecular systematic activity on subfamily Apioideae (Table 1).Many of these studies considered DNA sequences from both coding and non-coding loci of nuclear and chloroplast genomes, with fewer studies incorporating cpDNA RFLP or structural rearrangement data.Considering only GenBank accessions of nucleotide sequence data, and excluding Apium graveolens, Daucus carota and Petroselinum crispum (because much of the data available for these economically important apioid species are the result of non-systematic studies), there has been a fivefold increase in the number of submitted sequences in over just the past two and a half years (from c.300 in March 1998to c.1500in September 2000), primarily the result of these comparative phylogenetic analyses.Considering all molecular phylogenetic studies cited in Table 1, a total of 195 apioid genera and 450 species have been examined to date, representing almost half of the 404 genera within the subfamily recognized by Pimenov & Leonov (1993).

Tribes of Apioideae
Within Apiaceae subfamily Apioideae, ten tribes and three subtribes have been erected or supported as monophyletic as a result of previously published phylogenetic analyses of molecular data (Table 2).These include tribes Aciphylleae M. F. Watson & S. R. Downie, Bupleureae Spreng., Careae Baill., Echinophoreae Benth., Heteromorpheae M. F. Watson & S. R. Downie, Oenantheae Dumort., Pleurospermeae M. F. Watson & S. R. Downie, Pyramidoptereae Boiss., Scandiceae TABLE 2. Generic composition of tribes and subtribes within Apiaceae subfamily Apioideae supported or erected on the basis of phylogenetic analyses of molecular data.Previously designated names for these clades (according to Downie et al., 1998, 2000a,b,c, and Katz-Downie et al., 1999) et al., 2000c).e With the addition of Pastinaca, this group coincides with Drude's Peucedaneae subtribe Tordyliinae (or tribe Tordylieae W. D. J. Koch).While ITS studies support strongly the monophyly of the group, some studies of plastid DNA remove Tordylium from the clade.f Members of this clade are allied closely to tribe Pleurospermeae and the genus Erigenia ( Valiejo-Roman et al., 1998;Katz-Downie et al., 1999;Pimenov et al., 1999).As Pleurospermum is taxonomically complex and surrounded by many small genera of dubious affinity, we refrain from naming this group until these genera are examined.If Erigenia is indeed to be included in the Komarovia clade (and if the latter retains its present complement of genera), the tribal name Erigenieae Rydb.would apply.g If further investigation supports this group, and if it is to be recognized at the tribal level, the earliest name Pimpinelleae Spreng.should be applied, as the type of Pimpinella (P.saxifraga L.) allies with this group.Registaniella is included here based on its morphological similarity to Psammogeton.

Other major clades within Apioideae
In addition to the aforementioned monophyletic tribes, seven other major clades have been recognized within subfamily Apioideae on the basis of phylogenetic analysis of molecular data (Downie et al., 1998(Downie et al., , 2000a,b,c),b,c).These include the Angelica, Apium, Arracacia, Conioselinum, Heracleum, Komarovia and Pimpinella clades, and the generic composition of each is provided in Table 3.The Angelica and Apium clades were first established on the basis of cpDNA evidence (Plunkett et al., 1996b;Downie et al., 1998), but increased sampling has since obscured their boundaries.Moreover, the results of the ITS-based investigations have never been fully compatible with those of the cpDNA studies, with members of the Apium clade forming as many as five separate lineages basal to the Angelica clade (Downie et al., 1998;Katz-Downie et al., 1999).Therefore, to reconcile these groups with recent data, a more restricted view of these clades has been put forth (Downie et al., 2000c).As such, congruence is achieved in the composition of each of these clades and the bootstrap values supporting them are much higher.The phylogenetic position of the Arracacia clade is not clear, for in several studies it arises from within the Angelica clade (Plunkett et al., 1996b;Downie et al., 1998) and in some trees its monophyly is not very well supported (Downie et al., 1998).The Arracacia clade may eventually be subsumed within the Angelica clade.The Heracleum clade coincides with Drude's Peucedaneae subtribe Tordyliinae (= Tordylieae W. D. J. Koch) and its separation from other peucedanoid taxa was confirmed previously using seed protein immunochemistry (Shneyer et al., 1995).A strong serological similarity is also apparent between Komarovia and Parasilaus, two members of the Komarovia clade (Pimenov et al., 1999).
Apioid tribes Careae, Echinophoreae and Pyramidoptereae, along with clades Angelica, Apium, Arracacia, Heracleum and Pimpinella (plus a number of genera of uncertain clade placement, see below) comprise a large assemblage of taxa termed the 'apioid superclade' (Plunkett & Downie, 1999).This superclade, confirmed as monophyletic in all analyses of molecular data to date and often supported by bootstrap values >90%, comprises the previously designated Aegopodium, Angelica, Apium and Crithmum clades (Plunkett et al., 1996b;Downie et al., 1998).The Aegopodium and Crithmum clades are now recognized as tribes Careae and Pyramidoptereae, respectively (Downie et al., 2000c).Previously, both Careae and Pyramidoptereae were considered part of an expanded Apium clade (Plunkett et al., 1996b) or Aegopodium clade (Plunkett & Downie, 1999;Downie et al., 2000a).While the apioid superclade is strongly supported as monophyletic, resolution of relationships among its constituent taxa is poor.However, a specific class of chloroplast genome structural rearrangement has been uncovered within the superclade and serves to strengthen support for several lineages delimited on the basis of DNA sequence data.Here, at least one expansion and seven different contractions of the junction between the large single-copy and inverted repeat (IR) regions of the chloroplast genome have been detected, each ranging in size from 1 to 16kb pairs (Plunkett & Downie, 2000).The frequency and large size of these IR junction shifts are unprecedented among angiosperms.Continued investigations on the phylogenetic utility of this molecular character in subfamily Apioideae are in order.

Genera of uncertain phylogenetic placement
As a consequence of the narrower circumscriptions of the Aegopodium, Angelica and Apium clades (Downie et al., 2000c), some genera previously placed within each are left without assignment to a specific group (yet they are still retained within the apioid superclade).These include Bifora, Conium, Coriandrum, Ferula, Levisticum and Prangos (and other genera identified by asterisks in Table 4).Their inclusion would result primarily in the blurring of the boundaries of the Angelica and Apium clades.The genera Cnidium, Conioselinum, Ferula, Ligusticum and Trachyspermum, however, are each not monophyletic and have some members assigned to specific tribes (Table 2) or informally recognized clades (Table 3).The Angelica clade is a large group that contains many genera of exceptional taxonomic difficulty (such as Angelica, Lomatium, Peucedanum, Selinum and Seseli ).Given that many large apioid genera are not monophyletic (Pimenov & Leonov, 1993;Valiejo-Roman et al., 1998;Downie et al., 2000c), the placement of any one genus into the Angelica clade (or, for that matter, any other clade heretofore recognized) must be treated as provisional until such genus is studied in detail.Considering those genera basal to the apioid superclade, Annesorhiza falls adjacent to Heteromorpheae, and Diplolophium and Erigenia fall adjacent to Komarovia (Downie & Katz-Downie, 1999).

P U  M M
We compare the utility of seven molecular data sets used for phylogenetic inference in subfamily Apioideae: restriction sites, ITS, rpl16 intron, rps16 intron, rpoC1 intron, matK, and rbcL.While comparisons between different molecular data sets were considered in earlier papers (e.g.Plunkett et al., 1997;Downie et al., 1998Downie et al., , 2000b;;Plunkett & Downie, 1999;Lee & Downie, 2000), this study is the first to offer a comparison among all seven markers.The 24-33 species considered in each data set (Table 5) represent many of the major lineages described above and, as far as possible, were included in multiple studies.However, it wasn't possible to make the data sets equitable in species composition.The rbcL matrix was constructed from four separate studies (Olmstead et al., 1992;Chase et al., 1993;Kondo et al., 1996;Plunkett et al., 1996a), whereas the matK study (Plunkett et al., 1996b) was carried out independently from those of the introns, ITS, and restriction sites.Of the 33 species included in each of the three intron studies, 30 species were shared.Here, Ferula, Anthriscus, and Aralia L. were each represented by one of two species; each species pair, however, is monophyletic as revealed by more comprehensive analyses (Downie et al., 2000a,b,c).Ten species (18 genera) were shared among the three intron and rbcL matrices, and 10 species (17 genera) were shared among the intron and matK data sets.Sixteen species (19 genera) were common to both rbcL and matK analyses.The ITS study (24 species) excluded members of Araliaceae and Apiaceae subfamilies Hydrocotyloideae and Saniculoideae, as well as the basal apioids Anginon, Bupleurum and Heteromorpha, owing to the difficulty in aligning many of their sequences with those of other Apioideae (Downie & Katz-Downie, 1996;Downie et al., 2000c) Pimenov & Leonov (1993) and the results of previously published molecular phylogenetic studies (Tables 2, 3).The symbol '???' denotes species of uncertain tribe or clade assignment ( a cpDNA restriction site study (Plunkett & Downie, 1999).b ITS study (Downie & Katz-Downie, 1996;Downie et al., 1998Downie et al., , 2000c)).c rpl16 intron study (Downie et al., 2000b).d rps16 intron study (Downie et al., 2000c).e rpoC1 intron study (Downie et al., 1998(Downie et al., , 2000b)).f matK study (Plunkett et al., 1996b).g rbcL study (Olmstead et al., 1992;Chase et al., 1993;Kondo et al., 1996;Plunkett et al., 1996a).
1999); Sanicula canadensis L. was also excluded but replaced with the closely allied Eryngium cervantesii F. Delaroche.The three intron, ITS, and restriction site data matrices shared 17 species and 20 genera (the groups of species representing Oenanthe and Bupleurum are also each monophyletic; Downie & Katz-Downie, 1999, and unpublished data), but upon consideration of rbcL and matK, the number of shared taxa among the seven data sets decreased substantially.Analyses of combined data therefore excluded data sets rbcL and matK.To reduce the number of terminals in the phylogenetic analysis of combined data or combined trees (discussed below), monophyletic groups of species belonging to the same genus (i.e.Ferula, Anthriscus, Aralia, Oenanthe and Bupleurum) were scored as a single taxon.

Experimental
Details of methodology and information on source of material, voucher deposition, and GenBank accession numbers are provided elsewhere (Plunkett et al., 1996a,b;Downie & Katz-Downie, 1996, 1999;Downie et al., 1996Downie et al., , 1998Downie et al., , 2000b,c;,c;Kondo et al., 1996;Plunkett & Downie, 1999).For the ITS study, complete ITS1 and ITS2 regions were used.Each of the three intron studies included data from the entire intron region, whereas only three quarters of the matK gene was sequenced (Plunkett et al., 1996b).Due to missing data at both 5∞ and 3∞ ends of rbcL for some species ( Kondo et al., 1996), only 1382 of a total of 1428 positions were included in the multiple alignment of these sequences.For the restriction site study, data were obtained for 14 enzymes and only variable sites were scored (Plunkett & Downie, 1999).The data sets analysed herein are available upon request.

Phylogenetic analysis
All phylogenetic analyses were carried out using maximum parsimony, implemented using PAUP* version 4.0b4a (Swofford, 1998).Shortest trees for each of the eight data sets (including that of combined rpl16, rps16, and rpoC1 intron data) were obtained using 100 random addition replicate searches, with tree bisection-reconnection (TBR) branch swapping and ACCTRAN optimization.Bootstrap values (Felsenstein, 1985) were calculated from 100 replicate analyses, simple addition sequence of taxa, and TBR branch swapping.Pair-wise nucleotide differences of unambiguously aligned nucleotide positions were determined using the distance matrix option in PAUP* (with no provision made to account for multiple hits), with alignment gaps in any one sequence treated as missing for all taxa.To examine the extent of conflict among the three intron data sets prior to combining data, we conducted the incongruence length difference (ILD) test of Farris et al. (1995), as implemented by PAUP* using simple addition sequence and TBR branch swapping.Five hundred replicates were considered for each partition.The results of this partition homogeneity test revealed that the three intron matrices do not yield significantly different phylogenetic estimates and, as such, an analysis of combined data is appropriate.All trees constructed from plastid DNA data were rooted with Aralia (Araliaceae), with the exception of the matK trees, which were rooted with the closely allied araliad genus Kalopanax Miq.(Aralia was unavailable for analysis).ITS trees were rooted with the apioid genus Physospermum, as sequences from more distantly related taxa were difficult to align.Relationships among those genera included in the three intron, ITS, and restriction site data matrices were further explored upon simultaneous consideration of all data (the 'supermatrix' or 'total evidence' approach; Kluge & Wolf, 1993;Wiens & Reeder, 1995).Here, five data matrices were combined into a single larger matrix and analysed using maximum parsimony, with data for those taxa not included in any single analysis scored as missing.An alternative approach for simultaneous consideration of all data (as well as unique taxa) is the construction of supertrees, where the resultant trees from each of the separate analyses are combined rather than the original data used to construct them (Sanderson et al., 1998).Using the method of matrix representation with parsimony (MRP; Ragan, 1992), the topologies of each of the five source (individual or consensus) trees obtained from maximum parsimony analyses of the five original data sets were translated manually into data matrices of binary characters, with question marks used for taxa not included in a given tree (Baum, 1992;Sanderson et al., 1998).The binary characters from all source trees were combined into one matrix and analysed using maximum parsimony, as detailed above for the separate analyses.Bootstrap values, however, were calculated for 1000 replicate analyses; decay values were also obtained (Bremer, 1988).The resultant maximally parsimonious trees, or supertrees, were rooted with Aralia, having all of its states scored as '0'.We reiterate that the original data matrices or source trees do not share identical termini and that monophyletic groups of species in the same genus were treated as single terminals in these combined analyses.Therefore, while we present a strict consensus of supertrees, it is not a 'strict supertree' (Sanderson et al., 1998) in the sense that it agrees with all the trees from which it was derived.

Comparative utility of molecular data
Characteristics of the eight data sets are presented in Table 6 and the single or strict consensus tree resulting from maximum parsimony analysis of each data set is presented in Fig. 2. In almost all instances, the phylogenetic results obtained from each of these reduced analyses are consistent with those of our previously published results when many more species were considered.The few areas of discord between the reduced and original analyses -most notable with regard to the rpl16 intron -are generally characterized by poorly supported nodes and, when collapsed, yield topologies of a similar nature.It is known that differences in species sampling may have an influence on the resulting phylogenetic hypotheses (Lecointre et al., 1993;Hillis, 1996Hillis, , 1998)), and our results support this conclusion.
Of the six coding (rbcL and matK ) and non-coding (ITS and introns) loci sequenced, the ITS region is evolving most rapidly, as evidenced by the greater percentage of sites that are potentially parsimony informative and its higher rate of sequence change (that in some pair-wise comparisons exceeds 30% of nucleotides; Table 6).This high rate of nucleotide substitution, however, precludes the ITS region for deep-level analysis, as available sequence data for Hydrocotyloideae, Saniculoideae, and some basal Apioideae could not be aligned unambiguously with data from other apioid taxa.In contrast, rbcL is evolving most conservatively, with TABLE 6. Characteristics of eight data sets (A-H ) used in maximum parsimony analyses of selected representatives of Apiaceae subfamily Apioideae and relatives.Only variable characters were included in the restriction site study.The single or strict consensus tree resulting from each of these analyses is presented in Figs.2A-H only 8% of all alignment positions parsimony informative and maximum pair-wise sequence divergence estimates just over 5% of nucleotides.Clearly, rbcL is best suited for deep-level comparisons.Intermediate in rate of evolution is matK and the three chloroplast introns, with the introns rpl16 and rps16 evolving slightly faster than matK or the intron rpoC1.The latter two regions appear to contribute similar information to the phylogenetic analysis (Downie et al., 1998).The most notable difference, however, between coding and non-coding regions is the prevalence of insertions and deletions (indels) in the latter.While indels can create problems in alignment interpretation, unambiguously aligned gaps can provide an important source of phylogenetic information (Downie et al., 1998;2000b;Downie & Katz-Downie, 1999).The analysis of cpDNA restriction sites provided 2-4 times more informative characters than any single DNA region sequenced, with levels of restriction site divergence, as assessed by mean character differences, similar to that inferred for the ITS region.Only in the analysis of combined intron data did the numbers of informative characters approach those obtained through restriction site mapping, although this was at the cost of sequencing over 3kb of intron DNA for each species compared.The continued acquisition of ITS, rps16 intron, and rpl16 intron sequences for molecular systematic study in Apioideae seems worthwhile, given their higher rates of sequence evolution over other examined loci and the ease and rapidity by which these sequence data can be procured.

Phylogenetic resolution
The results of the phylogenetic analysis of each of the seven reduced data sets and that of combined intron data reveal a group of species previously identified as the 'apioid superclade' (AS), and with the exception of the rbcL tree (Fig. 2G), bootstrap support for this clade is high (90-100%).Other shared clades include those recognized previously as tribes Scandiceae (S; bootstrap values 77-99%), Oenantheae (O; 63-100%), and Heteromorpheae (H; 99-100%).Data from the rpl16 intron (Fig. 2C ), however, do not resolve a monophyletic Scandiceae.Moreover, in those cpDNAderived trees in which it was included, Ligusticum (represented by its type L. scoticum) falls within tribe Scandiceae.In some cases, this is an artifact of the reduced sample however, because each group of species is monophyletic (as revealed by our earlier, more comprehensive analyses), data for each genus were combined and treated as a single exemplar.Data for genera not included in any separate analysis were scored as missing data.Complete species names are indicated in Table 5.The bracketed tribes and clades are those listed in Tables 2 and 3; the abbreviations AS, S, O and H are the same as outlined in Fig. 2.  et al., 1996;Downie et al., 1998); its position here, however, cannot be reconciled on the basis of any other kind of evidence.The single minimal length tree derived from maximum parsimony analysis of combined DNA sequence (rpl16 intron, rps16 intron, rpoC1 intron, ITS) and restriction site data is presented in Fig. 3.This tree has a length of 3929 steps, a consistency index (CI; excluding uninformative characters) of 0.482, and a retention index (RI ) of 0.639.Here, all clades of two or more taxa are strongly supported as monophyletic, as assessed by high bootstrap and decay values.The monophyletic Scandiceae is sister to Ligusticum scoticum.The strict consensus supertree of 16 minimal length 138-step trees constructed by the MRP method using five source trees (trees A-E in Fig. 2) and analysed using maximum parsimony is presented in Fig. 4 (CI, excluding FIG. 4. Strict consensus of 16 minimal length 138-step supertrees constructed by the MRP (matrix representation with parsimony) method using five source trees (Fig. 2, trees A-E) and analysed using maximum parsimony (CI, excluding uninformative characters=0.866;RI= 0.958).Numbers above nodes are bootstrap estimates for 1000 replicate analyses; numbers below nodes are decay values for trees up to 20 steps longer than those most parsimonious.The Ferula, Anthriscus, Oenanthe, Bupleurum and Aralia species sampled in each of the source trees were not equivalent, thus to reduce the number of terminals in the supertree these monophyletic groups of species were each scored as a single taxon.Taxa not included in any single source tree were coded as missing data.Complete species names are indicated in Table 5.The bracketed tribes and clades are those listed in Tables 2 and 3; the abbreviations AS, S, O and H are the same as outlined in Fig. 2. uninformative characters=0.866;RI=0.958).This tree is highly consistent to that derived from the analysis of combined data (Fig. 3) and upon collapse of those nodes supported by bootstrap values less than 55% the trees are almost identical, with the only difference between them being the position of Ligusticum.

P R W A
The putative relationship among the tribes and clades so far recognized in Apiaceae subfamily Apioideae is illustrated in Fig. 5.This phylogeny was inferred based on the results of combined data and trees described above (Figs 3, 4) as well as the results of previously published molecular analyses (Plunkett et al., 1996b;Downie & Katz-Downie, 1999;Plunkett & Downie, 1999;Downie et al., 1998Downie et al., , 2000a,b,c),b,c).We emphasize that at least 23 genera included in molecular studies to date (Table 4) do not comfortably find a home within any of these clades, although the majority FIG. 5. Summary of relationships among the tribes (Table 2) and informally recognized clades (Table 3) of Apiaceae subfamily Apioideae as revealed by phylogenetic analyses of molecular data (this study; Downie & Katz-Downie, 1999;Downie et al., 1998Downie et al., , 2000a,b,c),b,c).Not all genera examined to date have been assigned to a particular group (Table 4); many of these genera, however, are contained within the apioid superclade. of these taxa fall within the apioid superclade.Their inclusion would result primarily in the blurring of the boundaries of the Angelica and Apium clades.We also point out the provisional nature of these relationships, given that only half of the approximately 400 genera recognized in Apioideae have been considered in molecular study.Many of the remaining genera, however, are ascribed to the large tribe Apieae (Pimenov & Leonov, 1993), and while it is impossible to know their affinity until they are examined, a great many will likely fall within the apioid superclade.
Heteromorpheae are sister group to all remaining Apioideae, in line with their woody habit, primitive wood anatomy, southern African origin, and microfossil evidence (Plunkett et al., 1996b;Downie & Katz-Downie, 1999)

C
Phylogenetic analyses of molecular data provide very little support for Drude's (1898) often cited system of classification of Apioideae or for other subfamilial treatments that are based largely on morphology and anatomy.Drude's tribes Apieae, Coriandreae, Dauceae, Laserpitieae, Peucedaneae, Scandiceae, and Smyrnieae are each not monophyletic.Drude's tribes Dauceae, Laserpitieae, and Scandiceae, however, collectively form a monophyletic group (i.e.tribe Scandiceae).Of Drude's eight tribes, only the small morphologically distinct Echinophoreae is retained as monophyletic.Of Drude's ten subtribes and of those where sampling has been extensive (therefore, excluding the monotypic subtribe Silerinae Tausch and the bitypic subtribe Elaeoselininae Drude, both of tribe Laserpitieae, where material has so far been unavailable for analysis), only Scandiceae subtribe Scandicinae and Peucedaneae subtribe Tordyliinae show some resemblance to groups inferred by molecular study.Even so, the generic compositions of these clades are not identical to those proposed by Drude, with Molopospermum, Grammosciadium and Rhabdosciadium now removed from Scandicinae and Pastinaca added to Tordyliinae.
The results of these molecular phylogenetic analyses have helped to more accurately delimit the family Apiaceae as well as to clarify the natural subdivisions within its largest subfamily, Apioideae.Apiaceae comprise three major lineages, coinciding with subfamilies Apioideae and Saniculoideae and the Azorella clade, the latter representing some but not all members of Drude's subfamily Hydrocotyloideae.Based on taxonomic congruence among molecular data sets, ten major clades are recognized within Apioideae and treated at the rank of tribe, with Scandiceae comprising three subtribes.Each of these tribes and subtribes is strongly supported as monophyletic in many separate analyses of molecular data and, occasionally, by other lines of evidence.In naming these clades, we followed priority of validly published generic names (Pimenov & Constance, 1985) as regulated by the ICBN (Greuter et al., 2000).Seven additional clades are identified, with formal recognition pending further study and confirmation of monophyly.Of the 195 genera examined, 23 have yet to be assigned (or are assigned only in part) to a specific group.Many of these genera fall within the apioid superclade, thus additional phylogenetic study of this large clade is warranted.
While the relationships inferred on the basis of various types of molecular markers are generally congruent, it has not been possible to delimit many of these newly circumscribed tribes and clades using obvious morphological or anatomical synapomorphies (Downie et al., 2000a,b).It is hoped that future studies will add to this information, but if not, we would have to accept that the task of reclassifying Apioideae at the suprageneric level is to be accomplished on the basis of molecular data rather than morphology.We realize that this is a contentious issue; in the absence of evident and diagnostic morphological characters, several of the tribes recognized herein are difficult to identify or distinguish from one another.Attempts to construct a phylogeny for the entire subfamily using traditional (i.e.nonmolecular) characters have proven difficult, for while cladistic analyses of morphological and anatomical data offer much insight into the phylogeny of Scandiceae (Spalik & Downie, 2001;Spalik et al., 2001, and unpublished;Lee et al., 2001), Oenantheae (Downie & Spalik, unpublished), and basal Apioideae (Roux et al., 1978;Van Wyk et al., 1997;Van Wyk, 2001), these data appear to be of limited value at higher taxonomic levels (Downie et al., 2000a;Spalik & Downie, 2001).Such a conclusion is not surprising, given the common dissatisfaction among systematic botanists with traditional diagnostic characters used in the higher-level classification of Apioideae and the lack of consensus of relationships apparent among existing taxonomic treatments.Simply, many of the tribes and subtribes now recognized in Apioideae cannot be delimited unambiguously using morphological data and it may be unrealistic to think that morphological synapomorphies can be found for those tribes and other major clades supported by molecular studies.
Molecular systematic research on Apioideae has been collaborative, with participation international.These studies will continue, with emphasis on the placement of the 200 or so genera yet to be examined (including those heretofore refractory to proper phylogenetic placement) and the re-examination and refinement of the various tribes and clades recognized herein.Current and future revision of each of these major groups, incorporating both molecular and morphological data, will ultimately lead to the production of a modern classification for the subfamily.
. The placement of Heteromorpha alongside Anethum and Apium in the ITS study of Valiejo-Roman et al. (1998) must be considered spurious.Successively more distally branching lineages include Bupleureae, Pleurospermeae, the Komarovia clade, and Oenantheae.The relationships among many of the remaining clades are equivocal.While the apioid superclade and tribe Scandiceae can each be unambiguously circumscribed on the basis of molecular data, the relationship of these two clades to Aciphylleae, Smyrnieae, and the Conioselinum clade is not clear.

TRIBES AND CLADES WITHIN APIACEAE SUBFAMILY APIOIDEAE: THE CONTRIBUTION OF MOLECULAR DATA
Phylogenetic analyses of chloroplast gene(rbcL, matK ), intron (rpl16, rps16, rpoC1)and nuclear ribosomal DNA internal transcribed spacer (ITS) sequences and chloroplast DNA restriction sites, with supplementary data from variation in size of the chloroplast genome inverted repeat, have been used to elucidate major clades within Apiaceae (Umbelliferae) subfamily Apioideae Drude.This paper summarizes the results of previously published molecular cladistic analyses and presents a provisional classification of the subfamily based on taxonomic congruence among the data sets.

TABLE 1 .
A summary of systematic studies of Apiaceae subfamily Apioideae incorporating macromolecular data

TABLE 3 .
are provided in footnotes.Genera assigned to more than one suprageneric category may not necessarily be monophyletic within that group (see original papers for discussion) Major clades within Apiaceae subfamily Apioideae that have yet to be treated formally.Genera assigned to a particular clade may not necessarily be monophyletic within that clade (see original papers for discussion) The Apium clade sensu stricto(Downie et al., 2000c).This clade, if recognized at the tribal level, will be called Apieae [ined.].c We name this clade after the largest and earliest described genus, Arracacia.d Comprising two species each of Conioselinum and Ligusticum but excluding the types of each of these genera (C.tataricum Hoffm.and L.

TABLE 4 .
Genera of Apiaceae subfamily Apioideae included in molecular systematic studies but of uncertain tribal or clade placement.Asterisks (*) denote genera belonging to the apioid superclade.Genera designated as 'in part' have some members assigned to tribes (Table2) or major clades (Table3)

Table 4
FIG. 2. Single (treeA) and strict consensus (trees B-H ) trees derived from maximum parsimony analysis of cpDNA restriction sites (A) or DNA sequence data (B-H ).Tree H represents a combined analysis of the three intron data sets; the monophyletic genera Ferula, Anthriscus, and Aralia, each represented by one of two species in the separate analyses, were each scored as single exemplars.Tree summary statistics, including tree length, number of minimal length trees, and consistency and retention indices, are presented in Table6.Numbers above nodes are bootstrap estimates for 100 replicate analyses; only values >50% are indicated.Complete species names are indicated in Table5.Within subfamily Apioideae, the following major clades of two or more species are indicated: AS, apioid superclade; S, Scandiceae; O, Oenantheae; H, Heteromorpheae.FIG.3.Single minimal length 3929-step tree derived from maximum parsimony analysis of combined DNA sequence (rpl16 intron, rps16 intron, rpoC1 intron, ITS) and restriction site data (CI, excluding uninformative characters=0.482; RI=0.639).Numbers above nodes are bootstrap estimates for 1000 replicate analyses; numbers below nodes are decay values for trees up to 20 steps longer than those most parsimonious.The Ferula, Anthriscus, Oenanthe, Bupleurum and Aralia species examined in each of the separate analyses were not identical;