| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Systematics |
2Department of Ecology and Evolutionary Biology and the Peabody Museum of Natural History, Yale University, P.O. Box 208105, New Haven, Connecticut 06520-8105 USA; 3Institut für Systematische Botanik, University of Zurich, Zollikerstrasse 107, CH-8008 Zurich, Switzerland
Received for publication October 7, 2004. Accepted for publication April 8, 2005.
ABSTRACT
The cacti are well-known desert plants, widely recognized by their specialized growth form and essentially leafless condition. Pereskia, a group of 17 species with regular leaf development and function, is generally viewed as representing the "ancestral cactus," although its placement within Cactaceae has remained uncertain. Here we present a new hypothesis of phylogenetic relationships at the base of the Cactaceae, inferred from DNA sequence data from five gene regions representing all three plant genomes. Our data support a basal split in Cactaceae between a clade of eight Pereskia species, centered around the Caribbean basin, and all other cacti. Two other Pereskia clades, distributed mostly in the southern half of South America, are part of a major clade comprising Maihuenia plus Cactoideae, and Opuntioideae. This result highlights several events in the early evolution of the cacti. First, during the transition to stem-based photosynthesis, the evolution of stem stomata and delayed bark formation preceded the evolution of the stem cortex into a specialized photosynthetic tissue system. Second, the basal split in cacti separates a northern from an initially southern cactus clade, and the major cactus lineages probably originated in southern or west-central South America.
Key Words: biogeography • Cactaceae • inferior ovary • Pereskia • phylogeny • phytochrome C • stem-based photosynthesis • trnK/ matK
The Cactaceae contain 1500–1800 species renowned for their remarkable morphological and physiological adaptations to drought (Barthlott and Hunt, 1993
). Although they are found in a range of environments, they are especially conspicuous components of the New World's arid regions and represent one of the world's most spectacular desert radiations. The great majority of cactus diversity is found in two major lineages, the Opuntioideae and Cactoideae. Most members of these groups are what might be regarded as "typical" cacti: they are stem succulents with only vestigial or ephemeral leaves and a well-developed photosynthetic stem cortex with CAM carbon metabolism, specialized "collapsable" xylem cells that aid in water storage (Schleiden, 1845
; Mauseth et al., 1995
), deeply recessed inferior ovaries, and specialized short shoots (areoles) with very reduced internodes that produce spines, new long shoots, glochids (in Opuntioideae), and flowers. The remaining cacti consist of two small genera, Pereskia and Maihuenia. These two were at one time united into a third subfamily, Pereskioideae, but this grouping is supported only by the shared absence of many of the typical cactus characters noted above, and in recent years Maihuenia has been placed in its own subfamily, Maihueniodeae. Multiple phylogenetic studies support the monophyly of the opuntioid and cactoid subfamilies, but their relationships to Maihuenia and Pereskia have remained unresolved. This is due both to limited sampling of Pereskia (Hershkovitz and Zimmer, 1997
; Applequist and Wallace, 2001
; Nyffeler, 2002
), as well as an inability of the molecular data to resolve basal cactus nodes (Wallace, 1995
). To date, no studies have confirmed or rejected the monophyly of Pereskia.
Pereskia species are often interpreted as "relictual cacti," and are used as a general model of the ancestral condition from which the highly specialized morphology and physiology of the core cacti arose (Rauh, 1979
; Gibson and Nobel, 1986
; Mauseth and Landrum, 1997
). Pereskia species are widely distributed in the Caribbean and Central and South America in a range of drier forest habitats. They have been described as having superior to inferior ovaries, broad, flattened leaves with C3 photosynthesis, areoles with leaf production, dense, fibrous wood, a simple cortex without cortical bundles, poorly developed stem epidermal and hypodermal layers, nonsucculent tissues, and as inhabiting relatively mesic environments (Mauseth and Landrum, 1997
). This generalized depiction of Pereskia species has led botanists to believe that the stem succulent cacti are derived from woody, nonsucculent trees with C3 photosynthesis, as opposed to other growth forms (e.g., an herbaceous, succulent CAM plant) (Griffith, 2004
).
While the "Pereskia model" has been useful, it also downplays some potentially important ways in which Pereskia species differ from one another. An alternative perspective of Pereskia is supported by other studies that have emphasized the substantial ecological, morphological, and anatomical diversity found within Pereskia (Schumann, 1898
; Berger, 1926
, 1929
; Boke, 1954
; Backeberg, 1958
; Bailey, 1962
; Bailey, 1963a
, b
, c
, d
; Boke, 1963
, 1966
, 1968
; Leuenberger, 1986
; Martin and Wallace, 2000
). In his monograph of the genus, Leuenberger (1986)
delineated several species groups within Pereskia that are marked by particular combinations of vegetative and reproductive features (Table 1), but gave no indication of how these subgroups might be related to one another, nor did he explicitly argue that Pereskia is monophyletic. Our ability to infer early events in the evolutionary history of the cacti rests squarely on resolving two outstanding problems in cactus phylogeny: (1) whether Pereskia is monophyletic, and (2) how Pereskia species are related to the rest of the cacti.
|
MATERIALS AND METHODS
Sequence generation
Total genomic DNA was isolated from fresh or dried leaf or cortical tissue using a slightly modified procedure of the DNeasy Plant Mini kit (Qiagen, Valencia, California USA) as described in Nyffeler (2002)
. In some instances, extractions were so mucilaginous that the precipitation step was repeated.
Five distinct gene regions representing all three plant genomes were amplified and sequenced using standard primers found in the literature (rbcL: Bousquet et al., 1992
; trnK/matK: Johnson and Soltis, 1994
; Nyffeler, 2002
; psbA-trnH IGS: Sang et al., 1997
; phyC: Mathews and Donoghue, 1999
; cox3: Duminil et al., 2002
) with the exception of phyC, for which two new internal sequencing primers were designed (phyC454R: 5' GSACCCATGTTTGCC 3' and phyC701F: 5' GGATTCAAACTGTGC 3'). Fragments were amplified in 25 µL reactions (1 µL genomic DNA, 0.2 µL AmpliTaq polymerase [Applied Biosystems, Foster City, California USA], 2.5 µL, 2.5 mmol dNTP, 2.5 µL 10x buffer, 2.5 µL 10 mmol MgCl2, 2.5 µL 10 mmol primer, 1–2 µL BSA, 9.5–11.5 µL ddH20) using an automated thermal cycler and standard PCR protocols, with PCR cycling times as follows for the different gene regions: rbcL: 95°C for 5 min, 40 cycles of 94°C for 1 min, 48–52°C for 1 min, 72°C for 2 min, final extension of 72°C for 10 min; psbA-trnH: 95°C for 2 min, 35 cycles of 94°C for 30 s, 50°C for 30 s, 72°C for 1 min, final extension of 72°C for 7 min; cox3: 98°C for 5 min, 40 cycles of 95°C for 1 min, 56°C for 1 min, 72°C for 1 min 30 s, final extension of 72°C for 5 min; phyC: 94°C for 5 min, then a "touchdown" protocol with a progressive drop in annealing temperature every 3 cycles (94°C for 1 min, 68°C–65°C–62°C–59°C–56°C for 2 min, 72°C for 2 min), 25 cycles of 94°C for 1 min, 53°C for 1 min, 72°C for 1 min, final extension of 72°C for 15 min; trnK/matK: 95°C for 4 min, 40 cycles of 94°C for 30 s, 48–52°C for 1 min, 72°C for 1 min 30 s, final extension of 72°C for 7 min. PCR products were cleaned using a PCR Purification kit (Qiagen) and directly sequenced, with the exception of the phytochrome C region. PhyC PCR product was cloned using the Invitrogen TOPO TA Cloning Kit for Sequencing (Carlsbad, California USA), and 2–6 clones of appropriate size (
1.1 kb) per taxon were sequenced. Except as noted later, clones from each taxon were nearly identical and formed monophyletic groups in a parsimony analysis of all clones (tree not shown), so we conclude that there is one copy of phyC and that slight differences between clones were due to PCR error or multiple alleles. One clone for each species was randomly chosen for inclusion in the phylogenetic analysis. Two distinct copies of phyC were recovered from Pereskiopsis and Quiabentia, but in each case the additional copy was easily distinguished by its great divergence from the other phyC sequences.
Dye terminator cycle sequencing reactions for all regions were performed in 20 µL volumes (1.5 µL 5x buffer, 1–3 µL PCR product, 2 µL BigDye v.3, 0.5 pmol primer, 13.3–15.3 µL ddH2O) using the following protocol: 96°C for 30 s, 26 cycles of 96°C for 10 s, 50°C for 5 s, 60°C for 4 min. Reactions were then cleaned using EdgeBiosystem plates (Gaithersburg, Maryland, USA) and run on either an MJResearch BaseStation51 (MJ Research, South San Francisco, California, USA) or ABIPrism 3700 automated sequencer (Applied Biosystems).
Contiguous sequences were constructed and edited using Sequencher v.4.2 (Gene Codes Corp., Ann Arbor, Michigan, USA), and homologous sequences for each region were aligned in MacClade v. 4.06 (Maddison and Maddison, 2003
). In most cases, alignment was unambiguous. However, the psbA-trnH IGS region varied considerably in length between taxa, and attempts to align outgroup sequences with Cactaceae sequences for this region gave dubious results. We therefore coded outgroup taxa as missing for this region. Aligned data matrices for all analyses are available from the first author upon request, and are also available from TreeBASE (http://www.treebase.org).
Phylogenetic analyses
We arranged the data into two major partitions, nuclear (phyC) and chloroplast/mitochondrial (rbcL, psbA-trnH IGS, trnK/ matK, cox3), and analyzed them separately and in combination. We combined the cpDNA and mtDNA markers into one partition as both of these organelles are typically maternally inherited. Also, there is little variation in the mtDNA marker (13 informative sites; see Table 2), and each informative site supports a bipartition that is also supported by the cpDNA markers. We tested for incongruence between the partitions using a partition homogeneity test (the ILD test of Farris et al., 1994
), as implemented in PAUP* version 4.b10 (Swofford, 2002
). We then performed parsimony (MP), maximum likelihood (ML), and Bayesian phylogenetic analyses on all three data sets (nuclear, cpDNA/mtDNA, combined) using similar methods for each data set. All MP and ML analyses were performed using PAUP* version 4.b10; Bayesian analyses used MrBayes v. 3.0b4 (Huelsenbeck and Ronquist, 2001
).
|
ML analyses
Models of molecular evolution were chosen for each data set using Modeltest version 3.05b (Posada and Crandall, 1998
). Heuristic searches were performed using a starting tree built by stepwise addition with 100 random addition replicates and TBR branch swapping, and 500 bootstrap replicates with five random addition replicates per bootstrap were performed to estimate support values for clades. In the combined data set, only 350 bootstrap replicates were performed.
Bayesian analyses
Models of molecular evolution were chosen for each data set and each gene region using Modeltest version 3.05b (Posada and Crandall, 1998
). For any data set containing more than one gene region (i.e., all analyses with the exception of the nuclear data set), we ran a mixed-model analysis, allowing each gene region to evolve under its own best-fit model. All Bayesian analyses were performed assuming default prior parameter distributions set in MrBayes v.3.04b. Posterior probabilities of trees were approximated using the MCMC algorithm with four incrementally heated chains (T = 0.2) for 10 000 000 generations and sampling trees every 1000 generations. To estimate appropriate burn-ins, posterior parameter distributions were viewed using Tracer v.1.01 (Rambaut and Drummond, 2003
). Discarding the first 100 trees (100 000 generations) was adequate in ensuring stationarity.
Hypothesis testing and sensitivity analyses
To test for alternative positions of the root of Cactaceae, two parametric bootstrap tests were performed (Goldman et al., 2000
). Our first test constrained Pereskia to be monophyletic, and in this case, our outgroups rooted along the branch to the Opuntioideae. In our second test, we constrained the position of the root along the branch to the Andean and southern South American Pereskia clades (see Discussion, Pereskia paraphyly and ovary position). This scenario is consistent with one first proposed by Berger (1926)
, in which his Eupereskia, a subgenus corresponding roughly to our Andean and southern South American Pereskia clades, is sister to all other cacti.
In each case, a ML heuristic search was performed on the original combined data set with tree topology T0 constrained to the alternative hypothesis, and the test statistic 
= LML – LT was calculated. Using SeqGen v.1.2.7 (Rambaut and Grassly, 1997
), we simulated 100 replicate data sets using the best-fit model of evolution for the original data set and tree topology T0 with its branch lengths. For each simulated data set Di, two ML heuristic searches were performed, one unconstrained (LML) and one constrained to topology T0(LT), and the test statistic i = LiML – LiT was calculated. The alternative hypothesis T0 was rejected if the original fell outside the 95th percentile of the distribution of i.
In performing these analyses, we were concerned that our original data set contained missing data, while our simulated data sets did not; the simulated data sets might therefore contain more phylogenetic information than our own. For this reason, we trimmed our combined data set into a smaller matrix with minimal missing data: in all, we removed six taxa (Ceraria fruiticulosa, Portulacaria afra, Pereskiopsis porteri, Quiabentia verticillata, Pereskia diaz-romeroana, and Pereskia bahiensis), the psbA-trnH IGS region, and the 5' trnK intron. This resulted in a data set of length 5060 characters, 450 of which are informative. Modeltest chose the same model (GTR + G + I) with slight differences in base frequencies, the substitution rate matrix, and the gamma distribution. The ML tree for this data set is topologically identical to the ML tree for the larger data set. With this smaller data set, we re-ran the parametric bootstrap tests as described.
Outgroup jackknifing
To assess the degree to which outgroup taxon sampling might affect the position of the root of Cactaceae, we generated a set of random combinations of our seven outgroups (both in the number and identity of taxa) and ran MP analyses using each set. To generate the outgroup combinations, a small Perl script was written that is executable in Mac OSX terminal window or Linux (available from the first author or as a free download called "hotgroup" at http://www.yphy.org/phycom). For this analysis, we generated 100 random outgroup replicates and ran each under a MP heuristic search with five random addition replicates and TBR branchswapping. Best trees from each search were saved to a tree file, and outgroup number, identity, and root placement was recorded for all trees.
RESULTS
Taxon sampling and missing data
While our combined data set includes 38 taxa, not all taxa were sequenced for all five gene regions (Table 2; Appendix). We were unable to amplify the phyC region from two of our outgroup taxa, Ceraria fruticulosa and Portulacaria afra. In the trnK/matK data set, 36 of 38 taxa are sequenced for the matK gene and flanking 3' trnK intron, while a subset of those taxa are also sequenced for the flanking 5' trnK intron. Also, two Pereskia species (Pereskia bahiensis and Pereskia diaz-romeroana) were only sequenced for the trnK/matK region; this region provided substantial information regarding the placement of these taxa within two strongly supported Pereskia clades. Finally, two taxa in our data set are chimeric: Cereus comprises C. alacriportanus (trnK/matK) and C. fernambucensis (all other regions), and Pereskiopsis A comprises P. diguetii (trnK/ matK) and P. aquosa (all other regions).
Data partition combinability
The ILD test rejected the null hypothesis that our two data partitions were derived from the same data pool (P = 0.045). Visual inspection of trees from analyses of the individual partitions revealed two distinct areas of incongruence (Fig. 1), and removal of either the representatives of the Anacampseroteae (= Grahamia bracteata, Grahamia coahuilensis, and Anacampseros telephiastrum) outgroup taxa or the clade comprising P. zinniiflora, P. portulacifolia, P. quisqueyana, and P. marcanoi from the ILD analysis resulted in P > 0.05. As accurate resolution of either tree region is not the primary purpose of this study, we did not investigate partition congruence issues further, and pooled all markers together for a combined analysis. To be certain that incongruence in one part of the tree was not affecting our analysis in other parts of the tree, we removed the aforementioned taxa from the data set and re-ran our MP analyses. The resulting trees for the remaining taxa were topologically identical to those recovered from analyses of the full data set.
|
cpDNA/mtDNA markers
The 1000 MP heuristic searches resulted in 26 shortest trees with length 1173 and CI = 0.83 (CI excluding autapomorphies = 0.69). The 100 heuristic searches using maximum likelihood and the HKY85 model of evolution resulted in a single tree with score –lnL = 14937.27192 (Fig. 2). Bootstrap support was weak for deeper nodes within the tree using either MP or ML (unresolved in Fig.1, support values <50%). There is, however, strong MP, ML, and Bayesian support for Cactaceae, Cactoideae, and Opuntioideae, as well as several clades of Pereskia that correspond well with Leuenberger's (1986)
species groups (Fig. 1). Pereskia guamacho and P. aureiflora are sister to one another, as indicated by Leuenberger. A group of five Pereskia species (P. grandifolia, P. sacharosa, P. nemorosa, P. stenantha, and P. bahiensis), distributed in the cerrhado, caatinga, and chaco woodland ecosystems of southern South America, is well supported and is hereafter referred to as the SSA (= southern South America) clade (Fig. 3). There is also strong support for uniting the four Pereskia species restricted to the Caribbean islands of Hispaniola and Cuba. The cpDNA/ mtDNA data set strongly suggests that the species endemic to Hispaniola (P. quisqueyana, P. portulacifolia, and P. marcanoi) are not monophyletic; instead, P. zinniiflora, an endemic to the island of Cuba, is the sister species of P. portulacifolia. The Caribbean island endemics are subtended by P. bleo, which has been described as a riparian species, and is distributed in forests of Panama and Colombia. We will refer to (Pereskia bleo (P. quisqueyana, P. marcanoi, P. portulacifolia, P. zinniiflora)) as the Caribbean clade (Fig. 3). Pereskia aculeata, a geographically widespread and ecologically diverse species with a semi-scandent habit, is placed as sister to a group of three morphologically similar species (P. horrida, P. diaz-romeroana, and P. weberiana) that live in the dry Inter-Andean valley regions of Peru and Bolivia. Leuenberger united the three Andean species but could not place P. aculeata. We will refer to (P. aculeata (P. horrida, P. diaz-romeroana, and P. weberiana)) as the Andean clade (Fig. 3).
|
|
. Our MP heuristic search recovered 48 trees with length 665 and CI = 0.75 (CI excluding autapomorphies = 0.63). ML recovered a single best tree with –lnL = 5428.33231 (Fig. 2). MP, ML, and Bayesian analyses recover all of the aforementioned clades supported by the chloroplast–mitochondrial data set. Additionally, phyC resolves a basal split in the Cactaceae between one clade of Pereskia and all other cacti. The basal Pereskia clade comprises P. lychnidiflora, P. guamacho + P. aureiflora, and the Caribbean clade. This basal split is strongly supported by ML bootstraps and Bayesian posterior probabilities, but less so by parsimony (Fig. 1). This clade comprises all of the Pereskia species with distributions centered loosely around the Caribbean basin. We will refer to this newly identified Pereskia lineage as the Northern Pereskia clade (Fig. 3).
The Andean and SSA Pereskia clades are united with the opuntioid, cactoid, and Maihuenia lineages. Maihuenia and Cactoideae, in turn, are placed in a sister relationship. Within the Cactoideae, Blossfeldia liliputana is placed as sister to the other cactoids included in this study. This supports the controversial finding of Nyffeler (2002)
that B. liliputana, a diminutive, globular cactus of northern Argentina and Bolivia, is the basal member of Cactoideae. While the sampling of Cactoideae in this study is clearly too limited to hypothesize major cactoid relationships, it is encouraging that phyC, a nuclear gene, is so far in agreement with the results of Nyffeler (2002)
, which was based on only chloroplast markers.
Two areas of incongruence between our nuclear and cpDNA/mtDNA data sets are worth noting (Fig. 1). The first concerns the arrangement of taxa in the Portulaca/Talinum/ Anacampseroteae outgroups, with the chloroplast sequences suggesting a Cactaceae-Anacampseroteae sister relationship, and phyC suggesting that Portulaca is the nearest living relative of Cactaceae. Because Portulacaria and Ceraria were not included in our phyC analysis, we thought this may be a result of differential outgroup taxon sampling. Re-analyzing our cpDNA/mtDNA data set without Portulacaria and Ceraria, however, did not change the branching order of these outgroups.
The second incongruence lies within the well-supported Pereskia clade restricted to the Caribbean islands of Cuba and Hispaniola. Our chloroplast data places the Cuban endemic P. zinniiflora within a clade comprising the Hispaniolan endemics P. quisqueyana, P. portulacifolia, and P. marcanoi. The nuclear data, on the other hand, firmly resolve the Hispaniolan endemics as a monophyletic group sharing a sister relationship with the Cuban P. zinniiflora. Relationships among these four taxa deserve additional attention. The three Hispaniolan species currently exist as a handful of small, rather isolated populations, though it is not impossible to imagine gene flow occurring between them.
Combined analyses
Combining all gene regions into one data set yielded a well-supported hypothesis of basal cactus phylogeny (Figs. 2, 3). The 1000 heuristic MP searches resulted in 12 trees of length 1846 and CI = 0.80 (CI excluding autapomorphies = 0.66). The 100 heuristic ML searches found a single ML tree with –lnL = 20373.64. MP, ML, and Bayesian analyses all resulted in an identical tree topology, consisting of 3 major clades: the Northern Pereskia clade, the Opuntioideae, and Maihuenia + Cactoideae. Pereskia is paraphyletic, with the Northern Pereskia clade placed as the sister of all remaining cacti. The Andean and SSA Pereskia lineages in turn form a weakly supported clade that appears to share a sister relationship with the "core cacti" (Fig. 3), consisting of the opuntioid, cactoid, and Maihuenia lineages. We will refer to ((Andean, SSA)(Opuntioideae (Maihuenia, Cactoideae))) as the caulocactus clade (caulo = stem, referring to their specialized stems, see Discussion, early events in the evolution of stem-based photosynthesis).
For the most part, the phyC and combined trees are in agreement, with the major exception being the formation of the core cacti lineage in the combined analysis. Bayesian support for the core cacti is strong, ML bootstrap support is less so, and the MP bootstrap value is less than 50%.
Testing for alternative rootings
Our conclusion of a paraphyletic Pereskia is dependent upon the assumption that the outgroup taxa are attaching along the correct branch in Cactaceae. Problems with incorrect outgroup attachment become more likely when chosen outgroups are distantly related to the ingroup (Sanderson and Doyle, 2001
; Graham et al., 2002
; Soltis and Soltis, 2004
). While we feel that our sample of outgroup taxa probably includes the closest living relatives of the cacti (Hershkovitz and Zimmer, 1997
; Applequist and Wallace, 2001
; R. Nyffeler, unpublished data), there is considerable sequence divergence among the outgroups, as evidenced by the long branch lengths in Fig. 2. As described earlier, we investigated this potential problem in two ways. First, we used parametric bootstrapping to compare two alternative rooting hypotheses. Second, we conducted an outgroup jackknifing experiment to see how different combinations of available outgroups might alter the placement of the root.
Results from our parametric bootstrapping tests strongly reject a monophyletic Pereskia (original data set, P < 0.01, reduced data set, P < 0.01), as well as a rooting along the branch to the Andean + SSA clade (original, P < 0.01, reduced, P < 0.01). In addition, results from our outgroup jackknifing experiment provide some evidence that rooting along the Northern Pereskia clade is not an artifact of outgroup taxon sampling. Of 100 random replicate outgroup samplings, six outgroup combinations resulted in alternative rootings along either the opuntioid or cactoid branches. In each of these six cases, three or fewer outgroup taxa were present, and Ceraria fruticulosa and/or Portulacaria afra were always included. These two taxa are the most distantly related outgroups in our analysis (Hershkovitz and Zimmer, 1997
; Applequist and Wallace, 2001
; R. Nyffeler, unpublished data). Within the Talinum/Portulaca/Anacampseroteae lineages, nearly all taxa rooted along the branch to the Northern Pereskia clade, whether they were alone or present in any other combination. The exception was Anacampseros telephiastrum, which rooted to the Northern Pereskia clade when it was the only outgroup, but rooted along the opuntioid or cactoid branches when it was present with C. fruticulosa or P. afra.
DISCUSSION
Pereskia paraphyly and ovary position
The tree shown in Fig. 3, in which Pereskia is paraphyletic, is strongly supported by our analyses. However, most of the basal structure of our tree is contributed by the phyC gene, so it may be more appropriate at this stage to consider this to be a gene tree rather than the cactus species tree. While the implications of Pereskia paraphyly should be considered at this stage, we would like to see our results confirmed with additional genes before embarking on a new nomenclature for Cactaceae.
The first hypothesis of Pereskia paraphyly dates back to Berger (1926
, "Schema 1"), who split the genus into Eupereskia and Rhodocactus and united Rhodocactus with Opuntioideae, Cactoideae, and Maihuenia. His decision to split Pereskia in this way was based on his interpretation of the Pereskia ovary, with Eupereskia having a distinctly superior ovary with basal placentation and Rhodocactus having parietal placentation and a more developed receptacle surrounding the ovary. From later work, it appears that Berger had access to five Pereskia species (P. sacharosa, P. aculeata, P.grandifolia, P. bleo, and P. lychnidiflora) (Berger, 1929
). His two Pereskia genera correspond with the Northern (= Rhodocactus) and Andean + SSA (= Eupereskia) lineages, with the exception of P. grandifolia, which he united with the Northern species. However, his idea of family level relationships (Eupereskia (Rhodocactus, Cactoideae, Opuntioideae, Maihuenia)) was rejected in our parametric bootstrap tests.
Subsequent Pereskia classifications followed on this theme, though they have proposed slight rearrangements of taxa and have differed in the assignment of genus or subgenus rank (Backeberg and Knuth, 1936
; Backeberg, 1958
). All these schemes have been based primarily on differences in ovary position because it was generally agreed that the variability of the Pereskia gynoecium held the key to understanding the evolution of the "sunken ovary" of the core cacti. In spite of this interest, however, the Pereskia gynoecium did not receive careful study until much later (Boke, 1963
, 1964
, 1966
, 1968
; Leins and Schwitalla, 1988
), and earlier misinterpretations are likely responsible for the confusing history of Pereskia classification. Leuenberger (1986)
stressed the complex nature of the ovary in Pereskia, noting that "the transitional nature of ovary and ovule position, subject to distortions during ontogeny of flower and fruit, makes it an unreliable character" (p. 51). Pereskia sacharosa, for example, has been described as having "fully superior" and "half inferior" ovaries, with both "basal" and "basal-parietal" placentation, depending on the author. Leuenberger concluded that most Pereskia species could be considered half superior or half inferior, with truly inferior ovaries found only in the Caribbean clade (Fig. 4A). It appears that the evolutionary transition to the "cactus-type" inferior ovary that botanists had long hoped to find in Pereskia does exist, but as an independent event within the Caribbean clade. Careful comparative studies within the Northern Pereskia clade might be helpful in understanding this transition, and it would also be useful to assess the degree of similarity of the Caribbean ovary to the inferior ovary of the core cacti.
|
). This is reminiscent of the short-shoot systems of many desert plants, which are utilized to produce quick flushes of leaves without first investing in new stem tissue. Because some Pereskia species and all core cacti lack this trait, a reasonable scenario in the transition to leaflessness would be first to lose the ability to flush leaves from the areole. While this may still be the case (as likely in the Cylindropuntieae; see Wallace and Dickie, 2002
), extant Pereskia species provide us with no clear record of it. Areole leaf production appears to have a complicated history, with potentially two losses (in the caulocactus and Caribbean clades) and then two subsequent gains (within the Caribbean and SSA clades), or three gains (in the Northern, Caribbean, and SSA clades) and one loss (within the Caribbean clade) (Fig. 4B).
In contrast, there are two stem characters that clearly unite the Andean and SSA Pereskia clades with the core cacti. Most members of the caulocactus clade produce stomata on their stem epidermis and have the delayed bark formation characteristic of the stem photosynthesizing cacti (Fig. 4C, D). A notable exception is Maihuenia. While Mauseth (1999)
discussed stomata on fragmented epidermis in the areolar pits of both Maihuenia species, they lacked stomata in all other stem regions, so we interpret Maihuenia stems here to be essentially astomatous (Leuenberger, 1997
, also describes them as lacking stem stomata). The placement of Maihuenia in our trees implies that both stem stomata and delayed bark formation have been lost in this lineage. Maihuenia today consists of two highly specialized cushion plant species living in cold, dry regions of Patagonia and south-central Chile (Leuenberger, 1997
; Mauseth, 1999
). It is not immediately clear why these traits have been lost, though their cushion plant growth habit (often forming very dense stands of stems with leaves crowded at tips) seems unlikely to promote the use of the stem as a photosynthetic organ.
All members of the Northern Pereskia clade lack stem stomata and delayed bark formation, though stomata are very rarely found on the stems of P. portulacifolia and P. guamacho (E. Edwards, personal observation). It is currently impossible to ascertain whether these traits were lost in the Northern clade or gained in the caulocactus clade because we are still uncertain as to the true sister taxa to Cactaceae, and there is little mention of these traits in the Portulacaceae literature (Carolin, 1987
, 1993
; Eggli and Ford-Werntz, 2001
; but see Hershkovitz, 1993
, for comment on stem stomata in Talinum). Nevertheless, we can say that stem stomata and delayed bark formation were present in the cactus lineage prior to the divergence of opuntioid and cactoid lineages and that the evolution of these two characters appears to have been correlated.
It is worth considering how the presence of these traits might influence the functioning of stems and leaves in Pereskia. Several greenhouse studies have investigated the photosynthetic behavior of a variety of Pereskia species (Rayder and Ting, 1981
; Nobel and Hartsock, 1986
, 1987
; Martin and Wallace, 2000
) and minimal C3 stem photosynthesis has been recorded only in P. horrida (Martin and Wallace, 2000
). It is difficult to extrapolate these findings to what might happen in a natural environment, however, and ecological observations of wild Pereskia species are currently quite limited in scope. Projects characterizing the ecological physiology of the different Pereskia lineages are currently underway (Edwards et al., 2004
) and should be helpful in discerning any functional differences associated with these anatomical changes.
In the meantime, however, it seems that further modifications of the cortical tissue are needed to make much photosynthetic use of the stem. In a survey of 28 cacti, including seven Pereskia species representing the Northern, Andean, and SSA clades, Sajeva and Mauseth (1991)
emphasized what they considered to be key developments in the evolution of the cactus stem. These include a hypodermal layer comprised of collenchymatous cells with thickened walls, substantial intercellular air space to aid in CO2 diffusion, and the arrangement of cortical chlorenchyma tissue into "spongy mesophyll" and "palisade" layers that appear similar to those in photosynthetic leaves. While their sampling did not include any opuntioids, other researchers have found similar cortical structure in Opuntioideae (Nobel, 1988
; North et al., 1995
). No Pereskia species included in the Sajeva and Mauseth (1991)
study had their cortical tissue differentiated into such layers, and further, all of their stems had relatively small volumes of intercellular air space. It appears from our phylogenetic results that the evolution of stem stomata and delayed bark formation preceded the key cortex modifications that allowed the stem to function as an efficient photosynthetic organ (cf. "developmental enablers" in Donoghue, 2005
). This order of events may not be unique to cacti: in a recent survey of noncactus plants with stem-based photosynthesis, Mauseth (2004)
found that the majority of investigated species with delayed bark formation, potentially photosynthetic stems, and photosynthetic leaves did not have a specialized cortex, while nearly all of the aphyllous species with stems as their primary photosynthetic organ also had a well-developed palisade cell layer in their outer cortical tissue.
Historical biogeography of the cacti
There have been several hypotheses regarding the geographical origin of the cacti, all largely based on where the presumably basal members of Cactaceae currently reside. Buxbaum (1969)
cited both the Caribbean and central South America as likely areas, due to the presence of Pereskia and of opuntioid and cactoid lineages that he considered "ancestral." Leuenberger (1986)
concluded that northwestern South America was a more reasonable location, primarily because he imagined a late Cretaceous origin of the group and because centering its origin as far away from Africa as possible might explain the poor representation of cacti there. Our data set, as well as other recent molecular phylogenetic studies (Hershkovitz and Zimmer, 1997
; Nyffeler, 2002
), suggest that the cacti are not that old, however, because sequence divergence between the major cactus clades is limited. More recently, Wallace and Gibson (2002)
hypothesized a central Andean origin for the family. They assumed that the Andean Pereskia, certain Opuntioideae, and the cactoid Calymmanthium are basal cactus lineages, and all reside in Peru, Bolivia, and northern Argentina.
Our ability to infer the geographic origin of Cactaceae is limited by insufficient knowledge of its closest relatives, but within the cacti there is a striking geographic structure to our hypothesis of basal relationships (Fig. 5). The basal split in Cactaceae suggests an initial separation within the cacti into a primarily northern and a primarily southern clade. The Northern Pereskia clade is comprised of species scattered in Central and northern South America, Cuba, and Hispaniola, with the exception of P. aureiflora of eastern Brazil. Within the caulocactus clade, Maihuenia is restricted to central/southern Chile and Argentina, the Andean Pereskia clade (with the exception of P. aculeata, the only geographically widespread Pereskia species) is found only in Peru and Bolivia, and the SSA clade is scattered throughout drier forest regions of eastern Brazil, Paraguay, Uruguay, Argentina, and Bolivia. Blossfeldia is confirmed here as a basal cactoid lineage, and grows only in northern Argentina and Bolivia. While our sampling and resolution are insufficient within Opuntioideae to make inferences about the geographic distribution of its basal members, recent work of others has placed its earliest diverging lineages in Chile and Argentina (Griffith, 2004
).
|
; Burnham and Graham, 1999
). Early uplift in the central Andean region (25–20 mya) is hypothesized to have occurred under a fluctuating arid/semi-arid climate regime (Hartley, 2003
), which presents a likely scenario for early cactus diversification. The placement of Cacteae, an exclusively North American lineage with its center of diversity in Mexico, as being among the earliest diverging lineages of Cactoideae (represented here with Echinocactus, but confirmed with broader sampling in Nyffeler, 2002
), suggests relatively early movements out of this region and across the continent.
Conclusions
Pereskia, long thought to represent the "ancestral" cactus, appears to be paraphyletic, with the Andean and SSA Pereskia clades more closely related to the core cacti. This result allows us to make several new inferences regarding early cactus evolution. First, it appears that the specialized "sunken" inferior ovary of the cacti has evolved twice independently, once in the core cacti, and once in the Caribbean Pereskia clade. Second, most members of the two Pereskia clades that are united with the core cacti delay bark formation and produce stem stomata. As these Pereskia clades lack other stem specializations found in the core cacti (e.g., differentiation of cortical tissue into specialized cell types, large intercellular air spaces, well developed hypodermis), we infer that the expression of stem stomata and delayed bark formation were among the first changes associated with the transition to stem-based photosynthesis. Finally, our phylogenetic hypothesis implies an initial north–south geographic split during early cactus evolution and that both Opuntioideae and Cactoideae originated in southern or west central South America.
While these results are promising, they also point to areas that need work. First and foremost, we would like to see our results tested with additional genes before reclassifying the Cactaceae. Clearly, the true position of the cacti within the portulacaceous alliance needs to be settled, as this is critical to interpretations of early cactus character evolution. Perhaps most important, however, is the need for careful study of Pereskia ecology. The specialized physiology and water use strategies of the core cacti have been well characterized (Nobel, 1988
), while virtually no studies have been conducted on Pereskia in the wild (but see Diaz, 1984
; Diaz and Medina, 1984
). The cactus life form represents one of the more extreme cases of ecological and morphological specialization in plants, and knowledge that (and precisely how) Pereskia is paraphyletic provides a unique opportunity to examine the earliest steps in its evolution. Discovering more about how members of the different Pereskia lineages function in their environments would complement what we already know about their anatomy and morphology, allowing for an integrated approach to understanding this remarkable evolutionary transition.
Appendix
Taxa used in this study, GenBank accession numbers for the five regions studied, source (wild, or if cultivated, location and accession number), and voucher information. The following abbreviations are used for herbaria and botanical gardens: YU = Yale University, B = Berlin Botanical Garden, ZSS = Sukkulenten-Sammlung, Zürich, Z = Zürich Botanical Garden, NTG = National Tropical Garden-Kampong, Miami, JBN = Jardín Botánico Nacional, Dominican Republic, DBG = Desert Botanical Garden, Phoenix, ISC = University of Iowa.
Taxon; GenBank accession: phyC, rbcL, cox3, psbA-trnH, trnK/matK; Source; Voucher specimen.
Anacampseros telephiastrum D.C.; AY875311, AY875247, AY875270,—,—; Cult. ZSS 90 1682/10; Lavranos & Bleck s.n., South Africa, ZSS.
Anacampseros telephiastrum D.C.;—,—,—,—, AY875373; Cult. ZSS 90 16523; Lavranos & Bleck s.n., South Africa, ZSS.
Austrocylindropuntia subulata (Muehlenpf.) Backbg.; AY875305, AY875235, AY875281, AY875346, AY875364; Cult. B 153-19-74-80; Cubr 37075, garden, B.
Blossfeldia liluputana Werderm.; AY875301, AY875232, AY875282, AY875348, AY875366; Cult. B 160-16-86-20; Leuenberger & Arroyo 3579, Argentina, B.
Brasilopuntia brasiliensis (Willd.) A. Berger; AY875304, AY875234, AY875278, AY875343, AY875370; Cult. B 153-76-74-80; Cubr 29521, 38452, garden, B.
Calymmanthium substerile Ritter; AY875314, AY875230, AY875250, AY875320, AY015291; Cult. ZSS 89 3442; Anon. 1437, garden, ZSS.
Ceraria fruticulosa Pearson & Stephens;—, AY875218, AY875266,—, AY875371; Cult. YU; Edwards 96, garden, YU.
Cereus alacriportanus Pfeiff.;—,—,—,—, AY015313; Cult. ZSS 941313; Eggli et al. 2493, Brazil, Z.
Cereus fernambucensis Lemaire; AY875293, AY875240, AY875272, AY875328,—; Cult. B 166-88-83-10; Leuenberger et al. 3107, Brazil, B.
Echinocactus platyacanthus Link & Otto;—, AY875215, AY875257, AY875327, AY015287; Cult. ZSS 92–16–86; Anon., Mexico, ZSS.
Echinocactus platyacanthus Link & Otto; AY875294,—,—,—,—; Cult. ZSS 90 2985/0; Lüthy 015, Mexico, ZSS.
Grahamia bracteata Gill.;—,—,—,—, AY015273; Cult. ZSS 94 1326; Leuenberger & Eggli 4184, Argentina, ZSS.
Grahamia bracteata Gill.; AY875308, AY875217, AY875273,—,—; Cult. B 142-32-94-10; Leuenberger & Eggli 4230b, Argentina, B.
Grahamia coahuilensis (S. Watson) G.D. Rowley; AY875310, AY875246, AY875280,—,—; Cult. B 262-01-94-40; Lautner L92/22, Mexico (Cult.), B.
Grahamia coahuilensis (S. Watson) G.D. Rowley;—,—,—,—, AY875374; Cult. ZSS 90 1259; Glas & Foster 1934, Mexico, ZSS.
Maihuenia patagonica (Phil.) Britton & Rose; AY875303, AY875245, AY875277, AY875342, AY015281; Cult. B 030-30-88-10; Leuenberger & Arroyo 3850, Argentina, B.
Maihuenia poeppigii (Pfeiff.) K. Schum.; AY875309, AY875216, AY875269, AY875329, AY015282; Cult. B 048-15-93-10; Leuenberger & Arroyo 4180, Chile, B.
Opuntia dillenii (Ker-Gawl.) Haw.; AY875302, AY875233, AY875283, AY875341, AY875369; Cult. B 304-11-99-10; Greuter s.n. (4 December 1999), Dominican Republic, B.
Pereskia aculeata Miller; AY875312, AY875229, AY875260, AY875323,—; Cult. NTG; Edwards 5, garden, YU.
Pereskia aculeata Miller;—,—,—,—, AY875355; Cult. B 376-02-86-20; Zanoni 36108, Dominican Republic, B.
Pereskia aureiflora Ritter; AY875297, AY875231, AY875261, AY875322, AY875354; Cult. B 166-54-83-20; Leuenberger et al. 3054, Brazil, B.
Pereskia bahiensis Gürke;—,—,—,—, AY875351; Cult. B 166-86-83-10; Leuenberger et al. 3054, Brazil, B.
Pereskia bleo (Kunth) D.C.; AY875289, AY875227, AY875265, AY875339,—; Cult. JBN; Edwards 13, garden, YU.
Pereskia bleo (Kunth) D.C.;—,—,—,—, AY875359; Cult. B 277-01-88-80; Schwerdtfeger 12678, garden, B.
Pereskia diaz-romeroana Cardenas;—,—,—,—, AY875353; Cult. B 039-02-77-30; Rauh 40627, Bolivia, B.
Pereskia grandifolia Howarth var. grandifolia; AY875298, AY875228, AY875253, AY875325,—; Cult. NTG; Edwards 2, garden, YU.
Pereskia grandifolia Howarth var. grandifolia;—,—,—,—, AY875362; Cult. B 038-04-77-80; Schwerdtfeger 12489, garden, B.
Pereskia guamacho F.A.C. Weber; AY875291, AY875242, AY875254, AY875335,—; Wild; Edwards 15, Venezuela, YU.
Pereskia guamacho F.A.C. Weber;—,—,—,—, AY015275; Cult. B 001-16-74-70; Schwerdtfeger 13066, garden, B.
Pereskia horrida (Kunth) D.C. var. horrida; AY875287, AY875224, AY875258, AY875332, AY875356; Cult. B 256-01-82-30; Schwerdtfeger 13066, garden, B.
Pereskia lychnidiflora D.C.; AY875286,—,—, AY875330, AY875358; Cult. B 003-04-78-10; Leuenberger & Schiers 2508, Guatemala, B.
Pereskia lychnidiflora D.C.;—, AY875238, AY875255,—,—; Cult. DBG 1990054502; Zimmerman 2603, Honduras, DBG.
Pereskia marcanoi Areces; AY875288, AY875221, AY875267, AY875337,—; Wild; Edwards 9, Dominican Republic, YU.
Pereskia marcanoi Areces;—,—,—,—, AY875360; Cult. B 259-02-82-30; Marcano & Cicero, Dominican Republic, B.
Pereskia nemorosa Rojas Acosta; AY875296, AY875241, AY875276, AY875334,—; Cult. B 305-01-80-70; Cubr 23639b, garden, B.
Pereskia nemorosa Rojas Acosta;—,—,—,—, AY875350; Cult. B 039-05-77-30; Schwerdtfeger 7085a, 15650, garden, B.
Pereskia portulacifolia (L.) D.C.; AY875315, AY875226, AY875264, AY875338,—; Wild; Edwards 11, Dominican Republic, YU.
Pereskia portulacifolia (L.) D.C.;—,—,—,—, AY875361; Cult. B 376-01-86-10; Zanoni 35204, Haiti, B.
Pereskia quisqueyana Liogier; AY875292, AY875220, AY875263, AY875336,—; Wild; Edwards 7, Dominican Republic, YU.
Pereskia quisqueyana Liogier;—,—,—,—, AY875352; Cult. B 259-05-82-33; Cicero s.n., Dominican Republic, B.
Pereskia sacharosa Grisebach; AY875299, AY875222, AY875256, AY875326, AY875363; Cult. B 133-10-82-30; Rente s.n., Bolivia, B.
Pereskia stenantha Ritter; AY875295, AY875244, AY875271, AY875333,—; Cult. B 166-81-83-20; Leuenberger 3081, Brazil, B.
Pereskia stenantha Ritter;—,—,—,—, AY015276; Cult. ZSS 86 4200; Horst & Uebelmann 759, Brazil, ZSS.
Pereskia weberiana K. Schumann; AY875313, AY875223, AY875259, AY875331, AY875357; Cult. B 046-03-80-70; Schwerdtfeger 10206, garden, B.
Pereskia zinniiflora D.C.; AY875290, AY875237, AY875262, AY875321, AY015277; Cult. ZSS 84 2526; Anon. 19537, Cuba (cult.), ZSS.
Pereskiopsis aquosa (F.A.C. Weber) Britton & Rose; AY875300, AY875225, AY875251, AY875324,—; Cult. YU; Edwards 98, garden, YU.
Pereskiopsis diguettii (F.A.C. Weber) Britton & Rose;—,—,—,—, AY015280; Cult. ZSS 94 2160; Lomeli et al. s.n., Mexico, ZSS.
Pereskiopsis porteri (Brandegee ex F.A.C. Weber) Britton & Rose; AY875306, AY875243, AY875275, AY875344,—; Cult. B 169-03-84-30; Hohmann s.n., Mexico, B.
Portulaca oleracea L.; AY875317, AY875249, AY875284,—,—; Unk.; Applequist 7, Unk., ISC.
Portulaca oleracea L.;—,—,—,—, AY875349; Wild; Nyffeler s.n., Switzerland, Z.
Portulacaria afra Jacq.;—, AY875219, AY875268,—, AY875368; Cult. YU; Edwards 97, garden, YU.
Quiabentia verticillata (Vaupel) A. Berger; AY875319, AY875239, AY875279, AY875340,—; Cult. B 154-12-74-80; Schwerdtfeger 22507, garden, B.
Quiabentia zehntneri Britton & Rose; AY875307, AY875236, AY875274, AY875345, AY875372; Cult. B 163-09-88-30; Leuenberger et al. 3078, Brazil, B.
Talinum paniculatum (Jacq.) Gaertn.; AY875316, AY875214, AY875252,—,—; Cult. YU; Edwards 6, garden, YU.
Talinum paniculatum (Jacq.) Gaertn.;—,—,—,—, AY015274; Cult. ZSS 93 1952; Martinez & Eggli 203, Mexico, ZSS.
Tephrocactus articulatus (Pfeiff.) Backbg.; AY875318, AY875248, AY875285, AY875347, AY875367; Cult. YU; Edwards 99, garden, YU.
FOOTNOTES
1 The authors thank Wendy Applequist, Miriam Diaz, Urs Eggli, Beat Leuenberger, the Desert Botanical Garden (Phoenix, Arizona USA), the Berlin Botanical Garden (Berlin-Dahlem, Germany), the Sukkulenten-Sammlung (Zü rich, Switzerland), and the Jardín Botánico Nacional (Santo Domingo, Dominican Republic) for tissue or DNA samples and/or logistical support in the field. We owe special thanks to Dianella Howarth and Philippe Reuge for their help and advice in the lab and to Casey Dunn, Nico Cellinese, and the Donoghue Lab for helpful discussions. Beat Leuenberger and two anonymous reviewers provided comments that greatly improved this manuscript. This work was funded in part from a National Science Foundation Graduate Research Fellowship and a Deland Award for Student Research from the Arnold Arboretum of Harvard University to E. J. E, and by a grant from the Swiss National Science Foundation (823A-056624) to R. N. 
4 Author for correspondence (e-mail: erika.edwards{at}yale.edu
)
LITERATURE CITED
Applequist W. L. R. S. Wallace 2001 Phylogeny of the portulacaceous cohort based on ndhF sequence data. Systematic Botany 26: 406-419[ISI]
Backeberg C. 1958 Die Cactaceae. Handbuch der Kakteenkunde. G. Fischer, Jena, Germany
Backeberg C. F. M. Knuth 1936 Kaktus-ABC. Nordisk Forlag, Copenhagen, Denmark
Bailey I. W. 1962 The xylem of Pereskia sacharosa and Pereskia aculeata. Journal of the Arnold Arboretum 43: 376-388
Bailey I. 1963a The xylem of Pereskia grandifolia and Pereskia bleo. Journal of the Arnold Arboretum 44: 222-231
Bailey I. 1963b The xylem of pereskias from Peru and Bolivia. Journal of the Arnold Arboretum 44: 127-137
Bailey I. 1963c The xylem of pereskias from southern Mexico and Central America. Journal of the Arnold Arboretum 44: 211-221
Bailey I. 1963d The xylem of Pereskia colombiana, Pereskia guamacho, Pereskia cubensis, and Pereskia portulacifolia. Journal of the Arnold Arboretum 44: 390-401
Barthlott W. D. R. Hunt 1993 Cactaceae. In K. Kubitzki, J. G. Rohwer, and V. Bittrich [eds.], The families and genera of vascular plants, 161–197. Springer Verlag, Berlin, Germany
Berger A. 1926 Die entwicklungslinien der Kakteen. G. Fisher, Jena, Germany
Berger A. 1929 Illustrierte handbucher sukkulenter pflanzen: Kakteen. Ulmer, Stuttgart, Germany
Boke N. H. 1954 Organogenesis of the vegetative shoot in Pereskia. American Journal of Botany 41: 619-637[CrossRef][ISI]
Boke N. H. 1963 Anatomy and development of the flower and fruit of Pereskia pititache. American Journal of Botany 50: 843-858[CrossRef][ISI]
Boke N. H. 1964 The cactus gynoecium: a new interpretation. American Journal of Botany 51: 598-610[CrossRef][ISI]
Boke N. H. 1966 Ontogeny and structure of the flower and fruit of Pereskia aculeata. American Journal of Botany 53: 534-542[CrossRef][ISI]
Boke N. H. 1968 Structure and development of the flower and fruit of Pereskia diaz-romeroana. American Journal of Botany 55: 1254-1260[CrossRef][ISI]
Bousquet J. S. H. Strauss P. Li 1992 Complete congruence between morphological and rbcL-based molecular phylogenies in birches and related species (Betulaceae). Molecular Biology and Evolution 9: 1076-1088[Abstract]
Burnham R. J. A. Graham 1999 The history of neotropical vegetation: new developments and status. Annals of the Missouri Botanical Garden 86: 546-589[CrossRef][ISI]
