| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Systematics |
2Jodrell Laboratory, Royal Botanic Gardens, Kew, Richmond, TW9 3DS UK; 3Missouri Botanical Garden, P.O. Box 299, St. Louis, Missouri 63166 USA; 4Institut de Biotechnologie des Plantes, Batiment 630, Université de Paris XI, F-91405 Orsay, Cedex, France; and 5Department of Botany, Universidade Federal do Rio Grande do Sul, Avenida Bento Goncalves, 9500, Porto Alegre RS, Brazil
Received for publication September 28, 2000. Accepted for publication March 27, 2001.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Key Words: incongruence • Iridaceae • rbcL • rps4 • trnL • trnL-F
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
60 genera (Goldblatt, 1990, 1991
), representing one of the largest families of the superorder Lilianae (sensu Dahlgren, Clifford, and Yeo, 1985
). Members of Iridaceae are typically characterized by the possession of isobilateral, equitant leaves, styloid crystals, inferior ovaries, and flowers with three stamens. Although worldwide in distribution, the family is particularly diverse in Africa where there are some 1000 species, most of which are restricted to southern Africa.
Rigorous and multi-disciplinary studies by many authors have, to date, failed to produce a consensus subfamilial classification. The first phylogenetic analysis of Iridaceae, using modern cladistic techniques (Goldblatt, 1990
), formed the basis of the most recent classification of the family. This analysis used 52 characters from phytochemistry, cytology, pollen structure, anatomy, and morphology to identify four major clades. Given subfamily status (Goldblatt, 1991
), these were designated Isophysidoideae, Nivenioideae, Iridoideae, and Ixioideae. In turn, Iridoideae comprised tribes Mariceae, Tigrideae, Iridineae, and Sisyrinchieae, and subfamily Ixioideae comprised tribes Pillansieae, Watsonieae, and Ixieae.
In a subsequent cladistic analysis of Iridaceae, Rudall (1994)
used 33 characters that included more anatomical characters than were used by Goldblatt (1990)
. This analysis recognized the four subfamilies and seven tribes sensu Goldblatt. However, the relationships among the subfamilies found in the two separate analyses were not identical. The principal areas of conflict concerned the relationship of Ixioideae to the rest of the family and the placement of Isophysis. In Goldblatt's scheme, Ixioideae were the most derived clade, whereas Rudall's analysis placed them sister to the rest of the family. Also, Goldblatt placed Isophysis sister to the rest of the family whereas Rudall placed it sister to Nivenioideae. In Rudall's analysis Isophysis together with Nivenioideae then formed the most derived clade.
The most recent phylogenetic representation of Iridaceae is that of Souza-Chies et al. (1997)
using molecular data derived from the region coding for protein 4 of the plastid small ribosomal subunit (rps4). This tree, inferred by the interpretation of a relatively small number of molecular characters (600 base pairs, of which only 18% were potentially parsimony informative), placed Isophysis as the sister taxon to the rest of the family. Subfamily Ixioideae formed a well-supported clade, although there was little resolution within them, and subfamily Nivenioideae did not form a monophyletic group but rather a paraphyletic grade with Ixioideae as the terminal clade. In that analysis the monophyly of subfamily Iridoideae was not supported, but rps4 alone provided insufficient evidence to refute their monophyly.
Few nonmolecular characters remain to be studied that could resolve the conflicts among the phylogenetic interpretations of Iridaceae. Our study includes molecular characters from three additional plastid regions as a source of phylogenetic information and combines these data with the supplemented rps4 data of Souza-Chies et al. (1997)
into a single matrix. The three plastid DNA regions sequenced were the trnL (UAA) intron, the trnL-trnF (GAA) intergenic spacer (these two collectively termed the trnL-F region), and the gene for the large subunit of ribulose 1,5 bisphosphate carboxylase/oxygenase (rbcL). The aim of this analysis was to enhance the current understanding of Iridaceae phylogeny and elucidate some presently unresolved key questions, among which the following are the most pertinent. (1) The relationships among the four subfamilies (Goldblatt, 1990
), which includes the proper placement of Isophysis, a Tasmanian endemic lacking one synapomorphy often assumed for Iridaceae, the inferior ovary. Some earlier treatments have assigned Isophysis to its own family (Bentham and Hooker, 1883)
. (2) The familial and tribal position of the Madagascan achlorophyllous mycoparasite ("saprophyte") Geosiris, which in the past has been referred to Burmanniaceae, assigned to its own family Geosiridaceae (Jonker, 1939
), or placed in subfamily Nivenioideae (Goldblatt et al., 1987
; Goldblatt, 1990
). (3) Delimitation of Iridoideae, including the proper status of tribe Sisyrinchieae within this subfamily; of particular interest is the placement of the African genus Bobartia in Sisyrinchieae (Goldblatt and Rudall, 1992
) as most other members of this tribe are American or Australian. (4) Generic and tribal relationships within Iridoideae, Ixioideae, and Nivenioideae.
In addition to these taxonomic issues, a second concern was the topic of congruence of separate matrices and how incongruence can best be determined. We were particularly concerned about how to separate mere sampling error from truly incongruent patterns, as reviewed by Huelsenbeck, Bull, and Cunningham (1996)
.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
. Herbarium material of Klattia flava was extracted using a modified 2x CTAB method (Fay et al., 1998
) with propan-2-ol instead of ethanol for precipitation of the DNA and a 2-wk precipitation period at –20°C. All DNA extracts were purified by cesium-chloride ethidium-bromide equilibrium density gradients (1.55 g/mL). Purified total DNAs were dialysed in 1x TE buffer and stored at –80°C.
|
were amplified to achieve conformity between the four data sets.
Twenty to fifty nanograms of total genomic DNA were used as a template for Taq-mediated amplification. The 100-µL reactions contained Promega magnesium-free thermophilic buffer (50 mmol/L KCl, 10 mmol/L Tris-HCl, 0.1% Triton X-100), 3 mmol/L MgCl2, 0.004% BSA (Savolainen et al., 1995
), 0.2 mmol/L each dNTP, 100 ng of each primer, and 2.5 U Taq polymerase (Promega, Ltd., East Croydon, UK). Thirty cycles of DNA amplification were carried out in a Perkin Elmer (Warrington, UK) DNA thermal cycler using the following program for each of the three plastid regions: denaturation, 94°C, 1 min; annealing, 50°C, 30 sec; extension, 72°C, 1 min.
Amplification of the rbcL gene was carried out using a forward primer that matched the first 20 bp (base pair) of the exon and a reverse primer beginning at either position 1360R (5'-CTTCACAAGCAGCAGCTAGTTC-3') or 1368R (5'-CTTTCCAAATTTCACAAGCAGCA-3') on the complementary strand. Amplification using these primers produced a 1388 or 1391 bp fragment of the rbcL exon. In some cases amplification of the complete gene was not possible due to degradation of the genomic DNA. In these cases the gene was amplified in two parts using internal primers 636F and 724R (Muasya et al., 1998
); this reverse primer is monocot specific. For rbcL, four sequencing reactions per taxon were required with primers 1F, 636F, 724R, and 1360R/1368R to achieve >80% overlap.
Primers "c" and "f" (Taberlet et al., 1991
) were used to amplify the intron and intergene spacer region between the trnL and trnF exons. The amplified fragment varied in length from 650 to 900 bp and resulted in an aligned matrix of 1311 bp. For cases in which amplification of the "c" to "f" region failed, internal primers "d" and "e" (Taberlet et al., 1991
) were used to amplify the gene in two nonoverlapping segments, as these primers are direct complements. Only two sequencing reactions, with primers "c" and "f," were required in cases for which amplification of the whole trnL-F region was successful. Greater than 80% overlap was achieved in most cases. All trnL-F sequences were easily aligned by eye. Four discrete gaps were coded as present/absent (A/T) characters and added to the end of the trnL-F matrix; otherwise gaps were coded as missing.
A fragment including the rps4 gene, an intergene spacer and the ser-tRNA gene were amplified using primers that annealed to the 5' end of rps4 (rps5') and to trnS (tRNAs) downstream from rps4 (Souza-Chies et al., 1997
). The resulting amplified fragment was 800 base pairs in length. Only the 600 bp rps4 exon was used in this analysis because the 3' spacer sequence between rps4 (200 bp) was not available for all taxa. As for trnL-F, all rps4 sequences were aligned by eye. Gaps were coded as missing.
Amplified double-stranded DNA fragments were purified using "Wizard" mini columns (Promega) and directly sequenced on an ABI 373A or 377 automated sequencer using standard dye-terminator chemistry following manufacturer's protocols (Applied Biosystems, Warrington, UK). For editing and assembly of the complementary strands "Sequence Navigator" and "Autoassembler" (Applied Biosystems) were used. Each base position was individually checked for agreement of the complementary strands.
Phylogenetic analyses
All cladistic analyses were performed using the parsimony algorithm of the software package PAUP* version 4.02b; (Phylogenetic analysis using parsimony; Swofford, 1998
) on a Power Macintosh 7200/90 with 16 MB RAM. Although trnL-F was treated as one region in the analyses, we recognize that it is composed of two distinct and potentially functionally different regions. However, the low levels of variation detected (despite its noncoding nature) made combining the entire region into a "noncoding" data set the most practical approach. The data matrices for each of the three (rbcL, rps4, and the complete trnL-F region) plastid DNA regions and a combined data matrix of all three were analyzed using 1000 replicates of random taxon-addition order to find islands of equally most parsimonious trees (Maddison, 1991
), tree bisection-reconnection (TBR) branch swapping, with MULPARS (keeping all equally most parsimonious trees) on, and all character transformations treated as equally likely (Fitch parsimony; Fitch, 1971
). To minimize the time spent searching large numbers of trees, a limit of ten trees was set for each replicate. After completing the replicates, all trees found were then used as starting trees for another round of swapping with a tree limit of 5000. We then used successive approximations weighting (SW; Farris, 1969
). Characters were reweighted according to the rescaled consistency index based on the best tree(s), and after each round of reweighting a heuristic search (ten random replicates) was performed. As described above, all optimal trees were then collected by using the trees from the ten replicates as starting trees to collect all shortest trees. When the tree length remained the same in two successive rounds, these were the successive approximation weighted (SW) trees. Internal support was assessed with 1000 bootstrap replicates (Felsenstein, 1985
) with Fitch weights. Only those groups of >50% frequency were reported. Other tree and character manipulations were performed using MacClade version 3.05 (Maddison and Maddison, 1992
). To calculate the number and performance (consistency, CI, and retention indices, RI) of transitions (ts) and transversions (tv) in each region, we used a step matrix to weight transversions to zero. From the ts number and its CI and RI, we could then calculate the tv number and its statistics.
To evaluate relative signal strength in each of the three matrices (treating the trnL-F intron and spacer as one), starting-tree length distributions were examined using 5000 replicates of random taxon addition with no swapping and Fitch weights (Chase and Cox, 1998
). These were then plotted as percentage longer than the shortest tree lengths found in the original searches described above. We assume that the proximity of starting tree length distribution to the shortest tree length is due to the extent the calculations performed as taxa are added (before swapping begins) reflect the optimum tree length found. The clearer the signal, the more closely the calculation of starting tree length reflects the most parsimonious topology.
In addition, to assess congruence among pairwise combinations of the three data sets and for all data sets combined we carried out 100 replicates of the partition homogeneity test (Farris et al., 1995
), with a full heuristic search and random taxon addition.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
|
Analysis of trnL-F
Of the 1313 sites included in the analysis, 521 (51%) were variable and 297 (23%) were potentially parsimony informative. Separate analysis of the trnL intron and trnL-F spacer gave similar but highly unresolved and weakly supported patterns (not shown). All members of subfamily Ixioideae are characterized by an indel 270 bp in length, which occurs in the trnL-F intergenic spacer. It was not possible to code this indel due to slight length variation among species, but the subfamily is already well supported without adding this deletion to the matrix. Four other, well-defined discrete indels unique within Ixioideae were coded on the basis that they might increase resolution within the subfamily. Further discrete indels were not coded because these only marked groups for which bootstrap support was already high. Analysis of the entire trnL-F matrix, including the four indels, with Fitch weights gave more than 5000 equally parsimonious trees of 1023 steps with CI = 0.66 and RI = 0.80. Applying SW resulted in >5000 equally parsimonious trees of length 506 570 with CI = 0.89 and RI = 0.94 (Fitch length = 1023 steps). Seven nodes observed in the single tree illustrated (Fig. 2) were not recovered in the strict consensus of all the weighted trees obtained; 15 nodes were not recovered in the strict consensus of the Fitch trees. Again several polytomies are present within Ixioideae in each individual tree. In comparison to the rbcL SW topology, less resolution is provided by trnL-F, but this topology did place Watsonia and Lapeirousia within Ixioideae. As with rbcL, Ixioideae formed the most derived clade within a paraphyletic Nivenioideae, and Witsenia, Klattia, and Nivenia also formed a monophyletic group. The same tribal groupings recovered by rbcL within Iridoideae are recovered by trnL-F, again excluding Bobartia and Diplarrhena from tribe Sisyrinchieae with Bobartia placed within tribe Irideae and Diplarrhena positioned as sister to the rest of Iridoideae. Isophysis is again positioned as the sister taxon to the remainder of the family.
|
|
; Sheahan and Chase, 2000
). We found no strongly supported (>85%) and incongruent clades, and thus we have combined all three matrices directly in spite of the differences in tree topologies (see above) and the negative results of the partition homogeneity test (see below). Parsimony analysis with Fitch weights produced 244 trees of 2498 steps with CI = 0.58 and a RI = 0.73. Applying SW resulted in three trees of 1 021 582 steps with CI = 0.86 and RI = 0.90. These trees were again a subset of the Fitch trees. One of the three equally parsimonious SW trees is illustrated in Fig. 4. Although polytomies are present, the combined tree provides some resolution within Ixioideae, with several clades marked by high bootstrap support, notably the alliance of two Watsonieae genera, Thereianthus and Micranthus with the monogeneric tribe Pillansieae. Watsonia and Lapeirousia (Watsonieae) formed a consistent clade, as did the three Ixieae genera Schizostylis, Hesperantha, and Geissorhiza. Freesia and Anomatheca (Ixieae) were also well supported as sister taxa. In terms of other major groupings, the combined tree was as described for trnL-F.
|
|
|
Partition homogeneity tests (Farris et al., 1995
) for all combinations of the data matrices indicated that the individual data sets are highly incongruent. In particular, the rbcL/rps4 and trnL-F/rps4 pairs were much more incongruent than rbcL/trnL-F. This "incongruence" is reflected in the numbers of steps missed by each of the individual analyses relative to the combined tree; rps4 deviated the most, missing 56, and rbcL missed the fewest (three). The starting-tree length distributions also indicate that the pairwise and the three-way combinations of the matrices are congruent; none of the combined starting-tree distributions falls farther away from the shortest trees than those of the separate matrices.
Molecular evolution
In the protein-coding genes, the CI was lowest for third positions in rps4, but in rbcL the CI of third positions was no lower than that for first positions (Table 4). Retention indices for third positions were higher than for either first or second positions. As expected, more third positions were variable than either first or second positions.
|
Table 5 shows the number of steps contributed individually by transitions and transversions in each region. As expected, more steps are contributed in each region by transistions than transversions; however, RIs are consistently higher for ts than tv for all three regions. We observed the following ts : tv ratios, rbcL = 1.5 and rps4 = 1.7. Substitution bias for the trnL-F region is better expressed in terms of the ratios for the intron and intergene spacer individually. The spacer has a slight excess of ts with a ratio of 1.1, and the intron has a greater number of ts with a ratio of 1.8.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Weighting based upon estimated frequency of change is a common practice (Albert, Chase, and Mishler, 1993
); however, comparison within data sets analyzed here indicate that there is no basis for discriminating and down-weighting certain classes of characters (as was also found by Olmstead, Reeves, and Yen, 1998
). For example, CIs were equally high and RIs higher for those characters that changed more frequently, i.e., ts vs. tv and third positions (for the protein-coding genes). This phenomenon was also observed for rbcL within monocots (Chase et al., 1995
). We take the high RI to indicate that variable third positions are not saturated (Naylor and Brown, 1998
) and that ts represent equally reliable characters as tv. This is contrary to assumptions for many weighting schemes that down-weight more frequently changing characters. Instead we applied successive weighting (SW) to weight all characters based upon their performance (RC) on the best tree, regardless of their frequency of change. In our matrices to down-weight transitions and third positions as whole categories could only lead to an underestimate of the signal possessed by these data. Higher substitution rates and worse performance are not correlated in these plastid regions. Furthermore, simulation studies (Hillis, 1998
) demonstrated that high substitution rates per se need not adversely affect character performance in phylogeny estimation.
There are reasonable arguments on both sides of the debate as to whether combining data sets to enhance phylogenetic inference is legitimate and/or useful (summarized in Huelsenbeck, Bull, and Cunningham, 1996
). The premise that separate analyses should be performed on subsets of the available data, however, first relies on the existence and delimitation of different classes of evidence with respect to phylogeny estimation (de Queiroz, Donoghue, and Kim, 1995
). Kluge and Wolfe (1993)
questioned whether such classes of data exist and concluded that there is no reason to believe that many putative subdivisions of evidence have discoverable boundaries. The partitions that we have used here are at least functionally independent, although all four belong to a genome that is inherited uniparentally in an intact manner.
Miyamoto and Fitch (1995)
have also argued that different genes/spacers/introns have real discoverable boundaries because they have distinct locations in the genome. However, we believe that the case presented here is an example of a situation in which only analyzing data sets separately does not do justice to the evidence contained in the data as a whole. Of course the issue of incongruence and the abilty to discern whether data arising from different sources, in this case different plastid regions, are truly in disagreement with one another in terms of their phylogenetic signal is pivotal. Our approach thus has been to analyze each data set separately as a means of exploring possible disagreements among data sets. In this case, the trees arising from separate analyses were apparently incongruent, but they only disagreed in relationships that received <65% from the bootstrap. We would argue as did Sheahan and Chase (2000)
that this does not represent a case of true incongruence among data sets; the "conflict" between analyses is caused simply by of a lack of phylogenetic signal (i.e., sampling error) in the separate data sets, rather than real differences among data sets with respect to some property that affects phylogeny estimation. However, the partition homogeneity tests (Farris et al., 1995
) carried out on all combinations of the data matrices overwhelmingly rejected the null hypothesis that the data sets contained congruent phylogenetic signal. As stated by Huelsenbeck, Bull, and Cunningham (1996)
, the major problem in assessing incongruence is separating sampling error from true incongruence.
A secondary problem identified by Weins (1998) is that incongruence is not likely to be a characteristic that affects placements of all taxa simultaneously, and so it is best assessed on a node-by-node basis. We agree that local incongruence should be identified so that the relative strength of the phylogenetic signal associated with the specific, incongruent relationships can be evaluated (Seelanan, Schnabel, and Wendel, 1997
). Whole-matrix approaches such as data decisiveness (Goloboff, 1991
) and partition homogeneity (Farris et al., 1995
) incongruence tests are less rigorous indicators of congruence because they are averaged over the whole topology and thus do not allow for the fact that the signal contained in a data set may support portions of the topology to different degrees. Whole-matrix assessments therefore suffer from two problems: (1) artificiality of estimates of matrix boundaries leading to unreasonable estimates of within and between matrix incongruence (within matrix incongruence being assumed to be due to sampling error such that greater estimates of between matrix sampling error are assumed to be evidence of true incongruence); and (2) inability to assess patterns of support for specific nodes on which all of the data agree, whereas only one or few specific nodes may be incongruent. We thus favor direct combination and node-by-node comparison of patterns of internal support as the best measures of incongruence. Whole-matrix measures are simply too coarse to be useful, and as long as resolution is improved and bootstrap percentages elevated then the individual matrices must be considered congruent. In our study, this is in spite of differing placements of some taxa in the individual matrices, which must then be assumed to be due to sampling error.
Comparison of the clarity of phylogenetic signal from each plastid DNA region provides several interesting points of discussion. The clearest signal (Fig. 5) is demonstrated by the noncoding trnL-F region, and consequently it is trnL-F that produces more groups arranged as they are in the combined tree. It may be argued that with such clear signal present in trnL-F the other two matrices are superfluous. However, despite producing large numbers of supported groups, trnL-F is not the most accurate of the three data sets, if we can consider the combined tree as the most accurate (due to overall higher levels of bootstrap support). This becomes evident when the individual matrices are optimized onto the combined topology and the resulting tree length compared to the tree length obtained by analyzing the data sets separately. Calculated in this way, rbcL had only three undetected substitutions, whereas trnL-F and rps4 had 20 and 56, respectively. The rbcL topology presented here does not appear to be the most accurate due to the misplacement of the two ixioid genera Watsonia and Lapeirousia. However, this is the data set that misses the least number of steps when compared to the combined topology, and thus we may conclude that in fact rbcL contains a more accurate signal than either trnL-F or rps4 (i.e., that the rbcL matrix produced a pattern that differed least from the combined matrix). This is not the conclusion that we would have reached if we had looked only at levels of bootstrap support or starting tree length distributions, and it highlights corroboration as the most powerful indicator of accuracy.
The misplaced taxa (Watsonia and Lapeirousia; Fig. 1b) found in the rbcL tree (relative to the combined tree) fall into a clearer perspective once we know that the combined topology requires only three additional steps over rbcL analyzed alone. Thus, to move these two taxa to this presumed correct position involved such a slight loss of parsimony that we conclude there simply is no clear pattern in rbcL for the position of these two genera (as mentioned in the RESULTS section, we sequenced both accessions a second time to be certain of the accuracy of these sequences). To move the sister clade of Watsonia and Lapeirousia in the rbcL tree to a similar position in Ixioideae required 25 steps, thus further indicating that the lack of clear patterns is focused on just these two taxa. (The two other sequence regions, rps4 and trnL-F, were sequenced from the same DNA accessions so it is also clear that they were not misidentified or the tubes/polymerase chain reactions mislabelled.)
The objective of phylogenetics, to find the most accurate estimate of hierarchical relationships, does not necessarily equate to the most parsimonious solution if there are insufficient data (Savolainen et al., 2000
). A single gene, just because it is a definable functional unit, may not alone constitute a reliable source of phylogenetic information (Chase and Cox, 1998
). We applied the parsimony criterion as our likelihood model of evolution, but the most "accurate" tree (in our case, the combined tree) can demonstrate how individual analyses may be gross underestimates of homoplasy (as in the case of rps4). This analysis clearly shows that reliance on any single data set may produce misleading conclusions. Ultimately, directly combining data resolves more relationships with enhanced bootstrap support, and thus direct combination is preferred regardless of results from the combinability tests, which are inevitably unreliable and unsatisfactory for addressing the causes of disagreement between trees from separate analyses.
Phylogenetic relationships
Previous cladistic analyses of nonmolecular characters (Goldblatt, 1990
; Rudall, 1994
) identified four distinct groups within Iridaceae. To date, no evaluation has been performed to evaluate the robustness of these clades, and hence the two sets of trees represent conflicting hypotheses that must be equally accepted as possible explanations of Iridaceae phylogeny. In contrast the trees presented here have been evaluated for the level of support for each of the individual clades. The taxonomic implications of the combined tree are discussed below.
Position of Isophysis
Affinities of the monotypic genus Isophysis vary considerably among classification systems, and its inclusion in Iridaceae is controversial largely due to its possession of a superior ovary. Isophysis does, however, share with the rest of the family at least two synapomorphies: presence of styloid calcium oxalate crystals (Goldblatt, Henrich, and Rudall, 1984
) and flowers with three stamens. These characters have led recent authors to include Isophysis in Iridaceae: tribe Isophysideae (Hutchinson, 1934
) and subfamily Isophysidoideae (Goldblatt, Henrich, and Rudall, 1984
; Dahlgren, Clifford, and Yeo, 1985
). However, the placement of this genus within the family has also been disputed (Goldblatt, 1990
; Rudall, 1994
). The combined analysis indicates that Isophysis belongs in a position as sister to the remainder of Iridaceae, which is consistent with the cladistic analysis by Goldblatt (1990)
. Isophysis is unambiguously placed in all combined trees, and bootstrap support for Iridaceae excluding Isophysis is high (96%).
The placement of Isophysis as sister to Iridaceae in Goldblatt's (1990)
analysis may be explained somewhat by his choice of outgroups, which all possess a superior ovary. If Isophysis is the earliest diverging genus of Iridaceae, this could imply that the superior ovary is the ancestral state for the family. Subsequent to Goldblatt's analysis, the rbcL monocot analysis (Chase et al., 1995
) and others (Rudall et al., 1997
; Chase et al., 2000b
; Fay et al., 2000
) placed Iridaceae within the lower asparagoids in a clade including Doryanthaceae and Ixiolirionaceae, with Tecophilaeaceae (including Cyanastraceae) as their sister group. These closest relatives, as implied by DNA analyses, all possess an inferior ovary and are the families from which outgroup taxa have been chosen for this molecular analysis. Since both ingroup and outgroup taxa possess an inferior ovary, the superior ovary of Isophysis must be regarded as an autapomorphy and thus uninformative. Therefore Goldblatt (1990)
placed Isophysis in the correct position but for the wrong reasons.
Nivenioideae
The six genera included in Nivenioideae sensu Goldblatt (1990)
are all represented in this analysis. The monophyly of Nivenioideae is not supported in any of the molecular analyses; instead the subfamily comprises a paraphyletic grade that collectively forms a clade with subfamily Ixioideae. Within paraphyletic Nivenioideae, the three shrubby Cape genera Witsenia, Klattia, and Nivenia form a well-supported clade (100%) in the combined analysis. The inclusion of the Madagascan saprophyte Geosiris in the Nivenioideae-Ixioideae clade is consistent in all of the molecular analyses and confirms its proper status within, and as a member of, the family (Goldblatt et al., 1987
). In the combined tree, the Australian genus Patersonia represents the sister group of the Nivenioideae-Ixioideae clade.
Coherence of Nivenioideae has been questioned by previous authors (Goldblatt, 1990
) because, in addition to its broad geographical distribution, only three nonmolecular characters define the subfamily: bipinnate rhipidia, a blue perianth, and a fugacious flower. The latter two are also found in Iridoideae and cannot be assessed robustly by outgroup comparison since they are homoplasious both within Iridaceae and outgroups. The inflorescence type (rhipidium) in particular is unique to Iridaceae, and thus it is impossible to independently assess which condition is plesiomorphic (Isophysis has single flowers and is thus of no assistance). Despite both the difficulties in providing a strong argument for the monophyly of Nivenioideae on morphological grounds and their nonmonophyly in the molecular phylogeny, constraining the subfamily to be monophyletic resulted only in a slight loss of parsimony (<1%). The paraphyly of this subfamily is thus suspicious because of extremely high levels of sequence divergence in the nivenioids and low levels in Ixioideae.
Ixioideae
Delimitation of the largely African subfamily Ixioideae is in accordance with most systems of classification of Iridaceae that have consistently accepted them as a distinct group (tribe Ixieae of Bentham and Hooker, 1883
; Diels, 1930
). Ixioideae are well defined by both morphological and anatomical characters. However, relationships within this subfamily remain ambiguous due to the lack of divergence in these plastid DNA regions. Several groupings do emerge in the combined analysis. Tribe Watsonieae are split into two well-supported groups, with the exception of Savannosiphon, which falls into an unsupported position within Ixieae. One of these groups, composed of Therianthus and Micranthus, appears to be associated with the monogeneric tribe Pillansieae. These three taxa are embedded in the partly unresolved tribe Ixieae.
Sampling for Ixioideae may be improved, but it appears that the plastid regions used to reconstruct this phylogeny do not exhibit enough variation to resolve the relationships within this apparently rapidly radiating subfamily. Improved resolution may be achieved by sequencing more variable nuclear ribosomal DNA (nrDNA) regions, for example the nuclear internal transcribed spacer region (ITS). However, within many genera of Iridaceae ITS rDNA appears to exist in a series of highly divergent, paralogous repeats at different chromosomal locations (Chase et al., unpublished data), which makes the ITS region difficult to use for phylogenetic reconstruction.
Iridoideae
Subfamily Iridoideae sensu Goldblatt (1990)
emerges as a monophyletic group in the combined analysis with bootstrap of 99%. Many tribal groupings are also well supported and mostly in accordance with those outlined by Goldblatt (1990)
. The exceptions to the tribal groupings sensu Goldblatt (1990)
are the placement of two members of tribe Sisyrinchieae: Bobartia and Diplarrhena. In Goldblatt's scheme, Bobartia is the only South African member of the tribe, but the molecular data place Bobartia in a clade with representatives of tribe Irideae (Europe/Africa), a position that has been considered but not supported by any prior cladistic analysis (Goldblatt and Rudall, 1992
). Diplarrhena is unusual within Iridaceae because it possesses only two stamens, whereas three stamens are otherwise uniform in the family. The combined molecular analysis positions Diplarrhena as sister to Iridoideae, in which it is probably best accommodated as a monogeneric tribe. The effects of combination of data sources are particularly apparent within this subfamily. No individual data set alone adequately resolves or provides support for the monophyly of the subfamily and its tribal groupings. The topology of the combined tree may be seen as representing the additive signals of all three data sets. A similar pattern for the placement of both Bobartia and Diplarrhena was obtained by Donato, Leach, and Conran (2000)
using ITS2 nrDNA sequences, although the position of the latter cannot be properly assessed because it was assigned as the outgroup.
In conclusion, this study strongly supports directly combining data for systematic inference when more than one data set is available even if "combinability" tests indicate incongruence. Combining consensus trees would not resolve the positions of Diplarrhena and Patersonia, for example. As pointed out by Huelsenbeck, Bull, and Cunningham (1996)
, separating sampling error from incongruence is difficult, and thus we advocate a node-by-node approach to this issue, in agreement with the conclusions of Wiens (1998)
. Further work on Iridaceae should include the combination of nonmolecular characters with additional DNA sequence data, which may provide greater resolution within Ixioideae in particular.
|
|
|
|
FOOTNOTES |
|---|
6 Author for reprint requests (m.chase{at}rbgkew.org.uk
).
|
LITERATURE CITED |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Bentham G. J. D. Hooker 1883 Iridaceae, Genera Plantarum 3(2). Lovell, Reeve and Company, London, UK
Brummitt R. K. H. Banks M. A. T. Johnson K. A. Doherty K. Jones M. W. Chase P. J. Rudall 1998 Taxonomy of Cyanastroideae (Tecophilaeaceae): a multidisciplinary approach. Kew Bulletin 53: 769-803[CrossRef]
Chase M. W. A. de Bruijn G. Reeves A. V. Cox P. J. Rudall M. A. T. Johnson L. E. Eguiarte 2000a Phylogenetics of Asphodelaceae (Asparagales): an analysis of plastid rbcL and trnL-F DNA sequences. Annals of Botany (London) 86: 935-956
Chase M. W. A. V. Cox 1998 Gene sequences, collaboration, and analysis of large data sets. Australian Systematic Botany 11: 215-229[CrossRef][ISI]
Chase M. W. M. R. Duvall H. G. Hills J. G. Conran A. V. Cox L. E. Eguiarte J. Hartwell M. F. Fay L. R. Caddick K. M. Cameron S. Hoot 1995 Molecular phylogenetics of Lilianae. In P. Rudall, P. J. Cribb, D. F. Cutler, and C. J. Humphries [eds.], Monocotyledons: systematics and evolution, 109–137. Royal Botanic Gardens, Kew, London, UK
Chase M. W. D. E. Soltis P. S. Soltis P. J. Rudall M. F. Fay W. H. Hahn S. Sullivan J. Joseph T. Givnish K. J. Sytsma J. C. Pires 2000b Higher-level systematics of the monocotyledons: an assessment of current knowledge and a new classification. In K. L. Wilson and D. A. Morrison [eds.], Monocots: systematics and evolution, 3–16. CSIRO Publishing, Collingwood, Victoria, Australia
Dahlgren R. M. T. H. T. Clifford P. F. Yeo 1985 The families of the monocotyledons: structure, evolution and taxonomy. Springer, Berlin, Germany
de Queiroz A. M. J. Donoghue J. Kim 1995 Separate versus combined analysis of phylogenetic evidence. Annual Review of Ecology and Systematics 26: 657-681[ISI]
Diels D. 1930 Iridaceae. In A. Engler and K. Prantl [eds.], Die Naturlichten Pflanzenfamilien, 2nd ed., Part 15a, 469–505. Engelmann, Leipzig, Germany
Donato R. C. Leach J. G. Conran 2000 Relationships of Dietes (Iridaceae) inferred from ITS2 sequences. In K. L. Wilson and D. A. Morrison [eds.], Monocots: systematics and evolution, 407–413. CSIRO Publishing, Collingwood, Victoria, Australia
Doyle J. J. J. L. Doyle 1987 A rapid DNA isolation procedure for small quantities of fresh leaf tissue. Phytochemistry Bulletin 19: 11-15
Duvall M. R. M. T. Clegg M. W. Chase W. D. Lark W. J. Kress H. G. Hills L. E. Eguiarte J. F. Smith B. S. Gaut E. A. Zimmer G. H. Learn Jr 1993 Phylogenetic hypotheses for the monocotyledons constructed from rbcL sequences. Annals of the Missouri Botanical Garden 80: 607-619[CrossRef][ISI]
Farris J. S. 1969 A successive approximations approach to character weighting. Systematic Zoology 18: 274-385
Farris J. S. M. Källersjö A. G. Kluge C. Bult 1995 Testing significance of incongruence. Cladistics 10: 315-319[CrossRef][ISI]
Fay M. F. R. G. Olmstead J. E. Richardson E. Santiago G. T. Prance M. W. Chase 1998 Molecular data support the inclusion of Duckeodendron cestroides in Solanaceae. Kew Bulletin 53: 203-212[CrossRef]
Fay M. F. P. J. Rudall S. Sullivan K. L. Stobart A. Y. de Bruijn G. Reeves F. Qamaruz-Zaman W.-P. Hong J. Joseph W. J. Hahn J. G. Conran M. W. Chase 2000 Phylogenetic studies of Asparagales based on four plastid DNA loci. In K. L. Wilson and D. A. Morrison [eds.], Monocots: systematics and evolution, 360–371. CSIRO Publishing, Collingwood, Victoria, Australia
Felsenstein J. 1985 Confidence limits on phylogenies: an approach using the bootstrap. Evolution 39: 783-791[CrossRef][ISI]
Fitch W. M. 1971 Toward defining the course of evolution: minimum change for a specific tree topology. Systematic Zoology 20: 406-416[CrossRef][ISI]
Goldblatt P. 1990 Phylogeny and classification of the Iridaceae. Annals of the Missouri Botanical Garden 77: 607-627[CrossRef][ISI]
Goldblatt P. 1991 An overview of the systematics, phylogeny and biology of the southern African Iridaceae. Contributions from the Bolus Herbarium 13: 1-74
Goldblatt P. J. E. Henrich P. J. Rudall 1984 Occurence of crystals in Iridaceae and allied families and their phylogenetic significance. Annals of the Missouri Botanical Garden 71: 1013-1020[CrossRef][ISI]
Goldblatt P. P. J. Rudall 1992 Relationships of Bobartia. South Africa Journal of Botany 58: 304-309
Goldblatt P. P. J. Rudall V. I. Cheadle L. J. Dorr C. A. Williams 1987 Affinities of the Madagascan endemic Geosiris, Iridaceae or Geosiridaceae. Adansonia 9: 239-248
Goloboff P. A. 1991 Homoplasy and the choice among cladograms. Cladistics 7: 215-232
Hillis D. M. 1998 Taxonomic sampling, phylogenetic accuracy, and investigator bias. Systematic Biology 47: 3-8
Huelsenbeck J. P. J. J. Bull C. W. Cunningham 1996 Combining data in phylogenetic analysis. Trends in Ecology and Evolution 11: 152-158[CrossRef]
Hutchinson J. 1934 The families of flowering plants. Oxford University Press, Oxford, UK
Jonker F. P. 1939 Les Geosiridacees, une nouvelle famille de Madagascar. Recueil des Travaux Botaniques Néerlandais 36: 473-479
Kluge A. G. A. G. Wolf 1993 Cladistics: what's in a word?. Cladistics 9: 183-199[CrossRef][ISI]
Maddison D. 1991 The discovery and importance of multiple islands of most-parsimonious trees. Systemtic Zoology 40: 315-328
Maddison W. P. D. R. Maddison 1992 MacClade: analysis of phylogeny and character evolution. Sinauer, Sunderland, Massachusetts, USA
Miyamoto M. M. W. M. Fitch 1995 Testing species phylogenies and phylogenetic methods with congruence. Systematic Biology 44: 64-76
Muasya A. D. A. Simpson A. Culham M. W. Chase 1998 An assessment of suprageneric phylogeny in Cyperaceae using rbcL DNA sequences. Plant Systematics and Evolution 211: 257-271[CrossRef][ISI]
Naylor G. J. P. W. M. Brown 1998 Amphioxus mitochondrial DNA, chordate phylogeny, and the limits of inference based on comparisons of sequences. Systematic Biology 47: 61-76[CrossRef][ISI][Medline]
Olmstead R. G. P. A. Reeves A. C. Yen 1998 Patterns of sequence evolution and implications for parsimony analysis of chloroplast DNA. In D. E. Soltis, P. S. Soltis, and J. J. Doyle [eds.], Molecular systematics of plants II, 164–187. Kluwer Academic, New York, New York, USA
Rudall P. J. 1994 Anatomy and systematics of Iridaceae. Botanical Journal of the Linnean Society 114: 1-21[CrossRef]
Rudall P. J. C. A. Furness M. W. Chase M. F. Fay 1997 Microsporogenesis and pollen sulcus type in Asparagales (Lilianae). Canadian Journal of Botany 75: 408-430
Savolainen V. M. W. Chase C. M. Morton S. B. Hoot D. E. Soltis C. Bayer M. F. Fay A. de Bruijn S. Sullivan Y.-L. Qiu 2000 Phylogenetics of flowering plants based upon a combined analysis of plastid atpB and rbcL gene sequences. Systematic Biology 49: 306-362[CrossRef][ISI][Medline]
Savolainen V. P. Cuenoud R. Spichiger M. D. P. Martinez M. Crevecoeur J.-F. Manen 1995 The use of herbarium specimens in DNA phylogenetics: evaluation and improvement. Plant Systematics and Evolution 197: 87-98[CrossRef][ISI]
Seelanan T. A. T. A. Schnabel J. F. Wendel 1997 Congruence and consensus in the cotton tribe (Malvaceae). Systematic Botany 22: 259-290[CrossRef][ISI]
Sheahan M. C. M. W. Chase 2000 Phylogenetic relationships within Zygophyllaceae based on DNA sequences of three plastid regions, with special emphasis on Zygophylloideae. Systematic Botany 25: 371-384[CrossRef][ISI]
Souza-Chies T. T. G. Bittar S. Nadot L. Carter E. Besin B. Lejeune 1997 Phylogenetic analysis of Iridaceae with parsimony and distance methods using the plastid gene rps4. Plant Systematics and Evolution 204: 109-123[CrossRef][ISI]
Swofford D. L. 1998 PAUP* 4.0: phylogenetic analysis using parsimony (* and other methods). Sinauer, Sunderland, Massachusetts, USA
Taberlet P. L. Gielly G. Pautou J. Bouvet 1991 Universal primers for amplification of three non-coding regions of chloroplast DNA. Plant Molecular Biology 17: 1105-1109[CrossRef][ISI][Medline]
Wiens J. J. 1998 Combining data sets with different phylogenetic histories. Systematic Biology 47: 568-581[CrossRef][ISI][Medline]
This article has been cited by other articles:
|
|
 |
|
  M. D. Pirie, M. P. B. Vargas, M. Botermans, F. T. Bakker, and L. W. Chatrou Ancient paralogy in the cpDNA trnL-F region in Annonaceae: implications for plant molecular systematics Am. J. Botany, June 1, 2007; 94(6): 1003 - 1016. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Duangjai, B. Wallnofer, R. Samuel, J. Munzinger, and M. W. Chase Generic delimitation and relationships in Ebenaceae sensu lato: evidence from six plastid DNA regions Am. J. Botany, December 1, 2006; 93(12): 1808 - 1827. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Samuel, W. Gutermann, T. F. Stuessy, C. F. Ruas, H.-W. Lack, K. Tremetsberger, S. Talavera, B. Hermanowski, and F. Ehrendorfer Molecular phylogenetics revea |
