| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Systematics and Phytogeography |
2Department of Botany, University of Wisconsin, Madison, Wisconsin 53706 USA, 3Molecular Systematics Section, Royal Botanic Gardens, Kew, UK
Received for publication November 15, 2001. Accepted for publication April 18, 2002.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Key Words: Cannabaceae • Cecropiaceae • ndhF • phylogenetics • rbcL • Rosales • trnL-F • Urticales
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
; Dahlgren, 1989
) is a distinctive yet controversial assemblage of up to seven families and about 2600 species. The group has included Cannabaceae, Cecropiaceae, Celtidaceae, Moraceae, Ulmaceae, and Urticaceae (Thorne, 1968
, 1992
; Berg, 1977
, 1989
; Dahlgren, 1980
, 1983
; Takhtajan, 1997
). Barbeyaceae were included only by Dahlgren (1989)
and Cronquist (1981)
. The Eucommiaceae and Rhoipteleaceae have at times been placed with these urticalean families, but all recent morphological (Berg, 1989
) and molecular (Chen et al., 1998
) evidence has argued against such relationships. As presently known from the fossil record, at least Ulmaceae and Celtidaceae evolved and diversified during the later Cretaceous and early Tertiary (Tiffney, 1986
; Manchester, 1989a
, b
). The other families have a comparatively poor fossil record (Collinson, 1989
), whereas monotypic Barbeyaceae are unknown in fossil form (Dickison and Sweitzer, 1970
). The largely tropical families Moraceae and Urticaceae make up 90% of the diversity in the order and have been separated into five tribes each (Berg, 1989
; Friis, 1989
, 1993
; Humphries and Blackmore, 1989
; Rohwer, 1993
). Two smaller families, Cannabaceae and Cecropiaceae, have typically been associated with Moraceae and Urticaceae (Kubitzki, 1993a
, b
). A wealth of morphological, cytological, chemical, and molecular data now support the separation of Ulmaceae and Celtidaceae (Zavada and Kim, 1996
; Ueda, Kosuge, and Tobe, 1997
; Wiegrefe, Sytsma, and Guries, 1998
).
Variation in biogeography and morphology is tremendous within the urticalean families and undoubtedly has contributed to controversies surrounding circumscription of the group, interfamilial relationships, and higher level relationships (Berg, 1989
). A wide range of growth forms is represented, including leptocaul to pachycaul, evergreen or deciduous trees, shrubs, climbers, hemi-epiphytes, subshrubs, and herbs, the latter comprising several kinds of succulents or even annuals. The leaves vary markedly, often with the peculiar brochidodromous or palmi-pinnate venation, a venation pattern correlated with subsequent lobing or compounding of the blade. Mucilage cells and canals are common, with latex production in Moraceae, Urticaceae, and Cecropiaceae. Flowers are either bisexual or more commonly unisexual and either insect or wind pollinated. The tremendous variation in structure of the gynoecium and stamen configuration and mode of pollen release is outlined in Berg (1989)
. Similarly great variation occurs in inflorescence structure and fruit/seed dispersal. Wind dispersal is common in Ulmaceae and Urticaceae; water dispersal is present in some Moraceae, Urticaceae, and Cecropiaceae; and animal dispersal is characteristic of Celtidaceae, Moraceae, Urticaceae, and Cecropiaceae. Explosive fruit dispersal (autochory), known from Moraceae and Urticaceae, may represent the first step towards animal dispersal (Berg, 1989
).
Based largely on pollen, floral, and anatomical features, the major families of the urticalean lineage have often been allied to amentiferous groups and thus have been linked to Hamamelidae (sensu Cronquist, 1988
) and specifically the higher hamamelids (Takhtajan, 1969
; Sweitzer, 1971
; Cronquist, 1981
; for a definition of lower and higher hamamelids, see Walker and Doyle [1975
]). Wolfe (1974)
further emphasized similarities in leaf venation in Ulmaceae and many other higher hamamelids, especially Fagaceae and Betulaceae. On the other hand, similarities in some features of the flowers, leaves, and vascular tissue have been used to link the urticalean families with Malvales (Thorne, 1973
, 1983
; Berg, 1977
; Dahlgren, 1983
; Wolfe, 1989
; Judd, Sanders, and Donoghue, 1994
; Takhtajan, 1997
). Thorne (1983)
specifically placed his Urticales between Malvales and Rhamnales in Malviflorae. A relationship with a rhamnalean lineage is also evident in Lindley's (1853)
placement of Ulmaceae within Rhamnales. However, based on wood anatomy, Sweitzer (1971)
specifically negated the placement of the urticalean families near either Malvales (sensu Thorne) or Rhamnales (sensu Lindley). Thorne (1973)
suggested that Euphorbiaceae was close to the urticalean families based on shared features of apetaly, unisexual flowers, latex formation, and varying degrees of palmate venation.
A wealth of molecular data, however, now place the urticalean families within a well-defined and supported Rosales (sensu APG, 1998
; including Barbeyaceae, Dirachmaceae, Elaeagnaceae, Rhamnaceae, and Rosaceae). Although sampling of urticalean families has been spotty and thus giving ambiguous results in previous rbcL or 18S rDNA analyses (Chase et al., 1993
; Gunter, Kochert, and Giannasi, 1994
; Soltis et al., 1997
; Qiu et al., 1998
), the broad angiosperm analyses using two (Savolainen et al., 2000a
) or three genes (Soltis et al., 2000
) and the recent eudicot-wide analysis using rbcL (Savolainen et al., 2000b
) recognized a monophyletic urticalean lineage imbedded within the Rosales (we will hereafter refer to these as urticalean rosids). Relationships of the urticalean rosids to other members of the Rosales have been ambiguous in some studies (Thulin et al., 1998
; Richardson et al., 2000
). However, Soltis et al. (2000)
placed Barbeyaceae with Elaeagnaceae with low support and not with the well-supported clade of urticalean rosids.
The few cladistic analyses based on morphological characters performed to understand the circumscription of the urticalean lineage and relationships of the families have had poor taxonomic sampling and contradictory results (Fig. 1). Berg (1989)
presented tentative schemes showing relationships of and within the group, but these were only intuitive. Humphries and Blackmore (1989)
cladistically examined 12 taxa within the urticalean families (Barbeyaceae as functional outgroup) using 15 morphological characters. Ulmaceae, Celtidaceae, and Cannabaceae were positioned as early diverging lineages in the most parsimonious tree, but Moraceae and Cecropiaceae were both shown to be broadly paraphyletic (Fig. 1a). Judd, Sanders, and Donoghue (1994)
conducted a preliminary cladistic analysis of 12 representative genera of the urticalean assemblage with Tilia (Malvales) as outgroup using 17 morphological characters. They recognized Ulmaceae, Celtidaceae, and a large Urticaceae encompassing Cannabaceae, Cecropiaceae, and Moraceae (Fig. 1b). Zavada and Kim (1996)
examined relationships within Ulmaceae and Celtidaceae (18 taxa) along with single representatives for Moraceae and Cannabaceae using 33 morphological, chemical, and cytological features. Using Barbeyaceae as outgroup, they concluded that Ulmaceae (except for Ampelocera) were monophyletic and sister to a broadly paraphyletic Celtidaceae containing the other families examined (Fig. 1c). Gunter, Kochert, and Gianasi (1994)
placed the urticalean families within a higher hamamelid clade based on a morphological analysis, although the lineage itself was not monophyletic.
|
). Relationships among the other families were weakly supported, although strong and novel support was seen for the origin of Cannabaceae from within an otherwise monophyletic Celtidaceae. A preliminary rbcL analysis of the Ulmaceae and Celtidaceae (11 genera) and four genera representative of other urticalean families provided strong support for a monophyletic Ulmaceae sister to a broadly paraphyletic Celtidaceae including Urticaceae, Moraceae, and Cannabaceae (Ueda, Kosuge, and Tobe, 1997
). The three-gene analysis of Soltis et al. (2000)
also provided strong support for the separation of Ulmaceae and Celtidaceae. A recent but smaller scale matK analysis of Celtidaceae (five genera of Celtidaceae and eight other genera of urticalean rosids) placed Cannabaceae solidly within a portion of the Celtidaceae (Song et al., 2001)
. Several studies have begun to examine relationships within Ficus (Herre et al., 1996
; Weiblen, 2000
).
The broad analysis of urticalean rosids and relatives presented here using rbcL, trnL-F, and ndhF sequence variation follows and extends greatly in both taxa and gene sampling previous molecular studies. Analysis of ndhF DNA sequences provides considerably more informative characters than rbcL (Kim and Jansen, 1995
; Alverson et al., 1999
) and trnL-F has proven effective within and among families in Rosales (e.g., Richardson et al., 2000)
. Thus, combining information from all three plastid regions should provide considerable resolution within and among the urticalean rosids (Soltis et al., 1998
, 2000
). Specific questions we ask here with a far broader sampling of the urticalean rosids include (1) How are the urticalean rosids related to other members of Rosales? (2) Are Ulmaceae sister to the remaining urticalean rosids as suggested by other studies? (3) Do Celtidaceae form a broadly paraphyletic group that includes Moraceae, Urticaceae, and Cannabaceae? (4) Are Cannabaceae aligned with Moraceae or with Celtidaceae as suggested by plastid restriction site and matK data? (5) Are Moraceae paraphyletic? (6) Are Cecropiaceae closely related to Urticaceae or Moraceae? These data also permit us to begin the reevaluation of evolutionary change in some critical morphological, cytological, and chemical characters, a process that will hopefully lead to the development of a stronger and more accurate morphological data set for this difficult group.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
; Humphries and Blackmore, 1989
; Rohwer, 1993
): Moreae, Ficeae, Castilleae, Dorstenieae, and Artocarpeae. The ten taxa sampled for Urticaceae encompass the four large tribes (Friis, 1989
): Urticeae, Parietarieae, Boehmerieae, and Elatostemeae (only the small Forsskaoleae was not sampled). The results of a broader molecular analyses (Nandi, Chase, and Endress, 1998
; Qiu et al., 1998
; Savolainen et al., 2000a
, b
; Soltis et al., 2000
) guided the choice of a fairly extensive set of outgroup taxa from among putatively closely related rosids in the eurosids I clade (APG, 1998
). These included 19 species of other Rosales (Dirachmaceae, Barbeyaceae, Rhamnaceae, Elaeagnaceae, and Rosaceae), five genera of Fagales, and seven genera each of Fabales and Cucurbitales. Oxalis of the Oxalidales, more distantly related in eurosids I (Soltis, Soltis, and Chase, 1999
; Savolainen et al., 2000a
, b
), was used as the ultimate outgroup to permit simultaneous resolution of the ingroup and these selected outgroups (Maddison, Donoghue, and Maddison, 1984
). Taxon sampling with trnL-F and ndhF was more limited and did not attempt to address issues of outgroup relationships. Twenty-five representatives of urticalean rosids, a subset of the 47 sampled with rbcL, were examined with ndhF and 26 taxa with trnL-F (Coussapoa of Cecropiaceae additionally done with trnL-F). Based on the more extensive rbcL analyses, Ceanothus and Rhamnus (Rhamnaceae) were used as outgroups for urticalean rosids in the trnL-F and ndhF analyses. All sequences except for the outgroups are new.
DNA extraction, amplification, and sequencing
Most of the DNA was extracted using a modified 6% cetyltrimethylammonium bromide (CTAB) extraction (protocol D of Smith et al., 1991
) or with DNeasy Plant Mini kit (Qiagen, Valencia, California, USA). In the former method, DNA was precipitated with ethanol and either sodium chloride or ammonium acetate, as these cause less polysaccharides to co-precipitate with the DNA than is the case with isopropanol or with ethanol in the presence of sodium acetate (Bult, Källersjo, and Suh, 1992
). Particularly in the case of Ulmaceae (Wiegrefe, Sytsma, and Guries, 1994
) and to a lesser extent for other urticalean rosids, subsequent salt washes (Sytsma, 1994
) were required in order to consistently amplify the DNA. Polymerase chain reaction amplification and cycle-sequencing followed the methods described elsewhere (Conti, Fischbach, and Sytsma, 1993
; Conti, Litt, and Sytsma, 1996
; Qiu et al., 1998
; Givnish et al., 2000
). Sequencing product was precipitated in ethanol and sodium acetate to remove excess dye terminators before being run out on an ABI Prism 377 DNA sequencer. Contiguous alignments were edited using Sequencher vs. 3.0 (Gene Codes, Ann Arbor, Michigan, USA).
Overlapping sequence fragments of rbcL were obtained from both strands using up to ten primers (see Conti et al., 1997
; Qiu et al., 1998
for details). Overlapping sequence fragments of the 3' end of ndhF were obtained from both strands using primers 5–14 (see Olmstead, Sweere, and Wolfe, 1993
). Sequences of ndhF, for which indels existed, were easily aligned visually in Se-Al version 2.0a6 (Rambaut, 2001
). Amplification and sequencing primers for the trnL-F region (comprising the trnL intron between the 5' and 3' exons and the intergenic spacer between the trnL 3' exon and trnF) used the universal primers c, d, e, and f from Taberlet et al. (1991)
to get coverage of both strands. Ambiguous alignment regions of trnL-F in Se-Al were excluded by command. Indels in both trnL-F and ndhF were coded using the guidelines of Baum, Sytsma, and Hoch (1994)
, but were not added as extra characters to their respective data sets. They were subsequently examined to ascertain if they further supported groups based solely on base pair (bp) changes.
Phylogenetic analysis
Variation in rbcL, trnL-F, and ndhF sequences, singly and in various combinations, was used to reconstruct phylogenetic relationships using PAUP* (Swofford, 2000
) on a Macintosh G4. To explore the possibility of the presence of multiple islands of most parsimonious trees (Maddison, 1991
), 1000 random addition sequences with MulTrees (save multiple trees) and TBR (tree bisection and reconnection) branch swapping were used to search under Fitch (1971)
parsimony. Bootstrap (Felsenstein, 1985
) and Bremer support values (Bremer, 1988
) were obtained to explore the relative degree of support for specific relationships. Bootstrap analyses using 1000 full heuristic runs (Simple addition sequence, TBR branch swapping, MulTrees) were done only on informative characters for the smaller rbcL, trnL-F, and ndhF data sets. Because full heuristic bootstrap analysis was not feasible on the 85 taxa rbcL data set, the same searching strategy as described for the smaller data sets was employed with the restriction of retaining only 1000 trees during searching in each replicate. To examine Bremer support values for individual branches, trees up to ten steps longer than the most parsimonious (depending on numbers of trees obtained and memory available) were obtained using the same heuristic algorithm employed in the Fitch analysis. Bremer support values for all branches still retained in the strict consensus of these extra-step trees were obtained using reverse topological constraints with 100 random addition sequences for each search, as suggested by Swofford (1993)
and implemented by Baum, Sytsma, and Hoch (1994)
. Support for the placement of specific lineages within increasingly more inclusive clades (e.g., Cannabaceae within Celtidaceae; Cecropiaceae within Urticaceae) was determined by an iterative, sequential removal of taxa in the larger clade, and examination of Bremer support and bootstrap support for each of the new branches (see RESULTS for example and explanation of this type of sensitivity analysis). The number of extra steps required to force taxa together based on previous systematic hypotheses was obtained by enforcing topological constraints with 100 random addition sequences. Additional tree topologies, suggested by previous studies, were examined for length using either topological constraint commands in PAUP* with 100 random addition sequences or with the tree editor in MacClade 3.05 (Maddison and Maddison, 1992
). The significance of differences between constrained and unconstrained trees was tested using the Templeton (Wilcoxon signed-ranks; Templeton, 1983
) nonparametric Tree Score option in PAUP*. For combined data set analyses involving rbcL, only the taxa sampled for trnL-F and ndhF were used. In all these analyses, Rhamnus lycioides (rbcL and trnL-F)/R. davurica (ndhF), Ceanothus sanguineus (rbcL and ndhF)/C. coeruleus (trnL-F), and Maclura pomifera (rbcL and ndhF)/M. cochinchinensis (trnL-F) were treated as single taxa. Additionally, two taxa lacked sequences for one of the three DNA regions (Cecropia and Coussapoa were not sequenced for ndhF or rbcL, respectively) and were simply given N's for their missing sequence in the combined analyses.
Character-state mapping
The lack of even a moderately complete morphological data set, relative to the sampling of urticalean rosids and outgroup taxa in this study, precludes for now a thorough morphological cladistic analysis, comparisons of molecules and morphology, and combined data set analyses. Similar difficulties are discussed in Humphries and Blackmore (1989
, p. 270). A common practice thus has been to reduce taxon sampling or to use family placeholders (see Humphries and Blackmore, 1989
; Judd, Sanders, and Donoghue, 1994
; and Thulin et al., 1998
for examples), but this is not appropriate here as many of the families that show marked variation in characters previously used may well be either paraphyletic (as suggested by these previous authors using morphological characters) or even polyphyletic (e.g., Cecropiaceae; Berg, 1989
). Thus, patterns of morphological and cytological evolution were assessed with selected characters by overlaying character states onto a reduced cladogram obtained from the single tree of the combined rbcL, trnL-F, and ndhF data set. The morphological and cytological characters chosen were obtained from previous cladistic studies on urticalean rosids (e.g., Humphries and Blackmore, 1989
; Gunter, Kochert, and Giannasi, 1994
; Judd, Sanders, and Donoghue, 1994
; Zavada and Kim, 1996
[note that the character states for Chaetoptelea and Chaetachme were switched in their Table 1]) and, when necessary, from primary literature (Bechtel, 1921
; Tippo, 1938
; Kuprianova, 1962
; Grudzinskaya, 1967
; Sweitzer, 1971
; Mehra and Gill, 1974
; Berg, 1977
, 1989
; Giannasi, 1978
, 1986
; Cronquist, 1981
; Manchester, 1989b
; Oginuma, Raven, and Tobe, 1990
; Takaso and Tobe, 1990
; Terabayashi, 1991
; Friis, 1993
; Kubitzki, 1993a
, b
; Rohwer, 1993
; Todzia, 1993
; Tobe and Takaso, 1996
; Judd et al., 1999
). Characters examined included (1) flower sexuality, bisexual or unisexual; (2) floral hypanthium, present or absent; (3) embryo position, straight or curved; (4) pollen pore number, 2–3 or 4–6; (5) secondary leaf veins, teeth not terminal or teeth terminal; (6) ovule position, apical or basal; (7) anther shape, standard, inflexed, or inflexed and explosive; (8) chromosome number, x = 14, 13, 10, or 8; (9) laticifers, absent, throughout, or bark only; and (10) fruit types, drupe, winged, or achene.
|
, 1999
). DISCRETE employs a Markov model to examine the evolution of pairs of binary characters on phylogenetic trees, taking branch length into account and weighting gains and losses equally. A log-likelihood ratio was calculated to see if the observed rates of evolution fit better with a model of dependent versus independent character evolution, relative to the results of 500 Monte Carlo simulations in which fruit (drupe, not drupe) and habitat (tropical, not tropical) states were assigned randomly and independently to branch tips. |
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
and Dahlgren, 1989
) requires seven additional steps. This longer tree is not however significantly different from the most parsimonious tree with the Templeton test (P > 0.05). The Rosaceae form the next most immediate sister group to this larger clade of urticalean rosids and relatives, followed by the orders Fagales, Fabales, and Cucurbitales. Urticalean rosids are well supported as a monophyletic group in the large rbcL analysis (Fig. 2; 89% bootstrap). The individual families Ulmaceae, Cannabaceae, Moraceae, and Urticaceae (with Cecropiaceae) are each strongly monophyletic (>89%), whereas Celtidaceae (with Cannabaceae) have little support. Relationships among urticalean families seen in the larger rbcL analysis were consistent with results of all analyses where the number and selection of outgroup orders were varied. However, relationships among lineages within the clade comprising Dirachmaceae, Barbeyaceae, Rhamnaceae, Elaeagnaceae, and Rosaceae and between this clade and urticalean rosids varied depending on the subset selection of outgroup taxa. Specifically, the lineage of Barbeya, Dirachma, and Rhamnus and Berchemia from Rhamnaceae was often placed as the sister group to urticalean rosids.
|
|
|
; Zavada and Kim, 1996
) and a less taxon-dense rbcL analysis (Ueda, Kosuge, and Tobe, 1997
) that provided little or no support for recognition of either Moraceae or Urticaceae. Consistent with all previous morphological and molecular studies, however, is the relatively weak support for the monophyly of Celtidaceae (with or without Cannabaceae included). The expanded family has no bootstrap support with rbcL or trnL-F (Fig. 2; Table 1). Aphananthe can swing to the base of the Moraceae + Urticaceae clade in trees that are one step longer in the rbcL analyses. The shortest trnL-F tree places Aphananthe as sister to all other Celtidaceae, Urticaceae, and Moraceae. Only in the ndhF and combined analyses does the placement of Aphananthe stabilize as sister to all other Celtidaceae and Cannabaceae (bootstrap of >80%; Fig. 3; Table 1). The rbcL, trnL-F, ndhF, and combined analyses all support Cannabis and Humulus as being within Celtidaceae and usually as sister to Pteroceltis (Table 1; up to 91% bootstrap support). Sensitivity analyses conducted on the combined data set to test the placement of Cannabaceae within Celtidaceae by iteratively excluding one or more genera of the Celtidaceae but leaving Pteroceltis and the early diverging Aphananthe increased bootstrap support (98%) for the placement of Cannabis and Humulus with Pteroceltis. Additionally, trees where Cannabaceae are not included within Celtidaceae are 19 steps longer; these trees are significantly different (P < 0.0177) with the Templeton test. Thus, there is strong support for this placement of Cannabaceae within Celtidaceae, although their exact placement as sister to Pteroceltis is less strongly supported.
With minor exceptions (e.g., Moreae), tribes within Moraceae and Urticaceae are monophyletic with our current sampling, although these are large families and taxon sampling is not dense. Unlike Celtidaceae, group support within these two families is generally strong (Figs. 2–3; Table 1). Tribe Moreae (if including Artocarpeae, see below) appears to be a basal grade or even polyphyletic within Moraceae. The basal split in Moraceae is weak, with either Maclura of Moreae (in either rbcL analysis) or other Moreae + Artocarpeae (in trnL-F, ndhF, and combined analyses) representing the sister clade to all other Moraceae. The molecular results unequivocally support the placement of Cecropiaceae with Urticaceae rather than with Moraceae. Cecropiaceae are polyphyletic with Poikilospermum quite unrelated to other Cecropiaceae (+7, +44, +46, and +154 extra steps required to make the family monophyletic in the rbcL, trnL-F, ndhF, and combined analyses, respectively; P < 0.066–0.0001 in the Templeton tests). The position of Poikilospermum as sister to Pellionia (Urticaceae) is further supported by a 3-bp indel in the ndhF data. Excluding Poikilospermum, all analyses except the larger rbcL survey (Fig. 2b) still strongly place the remainder of Cecropiaceae as imbedded within Urticaceae (Figs. 2–3; Table 1). There is strong bootstrap support (98–100%) in the trnL-F, ndhF, and combined analyses for the subclade within Urticaceae comprising Cecropiaceae + Parietaria + Boehmeria (Table 1). An extra 18 steps are required to force Coussapoa and Cecropia from the Urticaceae which is significant at P < 0.0009 in the Templeton test.
Character-state mapping
Three characters support the basal split within the urticalean rosids. All urticalean rosids except Ulmaceae possess the synapomorphic condition of flowers strictly unisexual, curved embryo, and a lack of hypanthium (Fig. 4A, B). Additionally, the Ulmaceae are recognized as a monophyletic clade by 4–6 pored pollen and secondary leaf veins ending in teeth, with a reversal of the latter character in Ampelocera (Fig. 4C). Aneuploid chromosome reduction to n = 10 in Cannabis and subsequent reduction to n = 8 in Humulus supports the placement of Cannabaceae within a subclade of Celtidaceae (Fig. 4D). Basal ovules is a synapomorphy supporting the placement of Cecropiaceae with Urticaceae rather than with Moraceae (Fig. 4E). The inflexed, explosively dehiscent anthers defines Urticaceae but are lost in Cecropiaceae, although Poikilospermum still retains the inflexed nature of the anthers (Fig. 4F). Due to variation in this stamen character in Moraceae, especially the tribe Moreae, the evolutionary history of the peculiar urticaceous stamen type (Fig. 4F) is unclear (see DISCUSSION). The evolution of laticifers in these families (Fig. 4G) also appears to have a complicated history (see DISCUSSION). Drupes or drupe-like fruits are plesiomorphic in urticalean rosids and their retention is clearly associated (P < 0.011) with the tropical forest habitat (Fig. 4H). Achene fruits have originated independently in Cannabaceae and Urticaceae (Fig. 4H).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Rosid ancestry of urticalean rosids
Relationships of urticalean rosids to other rosids based on the large rbcL survey, although weakly supported by bootstrap and Bremer support analyses (Fig. 2), are consistent with and extend the results seen in the Qiu et al. (1998)
and Thulin et al. (1988)
survey. The relationships among the outgroup rosid orders differ slightly from those outlined in the system of APG (1998)
or seen in the combined analysis of rbcL and nonmolecular characters for angiosperms (Nandi, Chase, and Endress, 1998
); however, this portion of the rosid tree is weakly supported based on our taxon sampling (Fig. 2) and an eudicot-wide analysis (Savolainen et al., 2000b
). Urticalean rosids are not closely related to the higher hamamelids, Malvales, or Euphorbiaceae based on these and previous molecular analyses (Qiu et al., 1998
; Savolainen et al., 2000a
, b
; Soltis et al., 2000
), as both urticalean rosids and higher hamamelids are part of the eurosid I group (Fig. 2; APG, 1998
). These results support Berg's (1989)
observations that similarities between urticalean rosids and Euphorbiaceae might be viewed as convergences due to occupation of lowland habitats and tendencies to floral reduction and mixed entomophily and anemophily. Similarly, the relationship between urticalean rosids and higher hamamelids (e.g., Gunter, Kochert, and Giannasi, 1994
) must be interpreted as more recent convergence within the eurosid I group. Specifically, the reduction of floral phyllomes from bicyclic to unicyclic and the development of monoecy are evolutionary changes associated with the amentiferous syndrome (Gunter, Kochert, and Giannasi, 1994
). The strongest, nonmolecular argument made for excluding urticalean rosids from these groups was Behnke's (1973
, 1989
) observation, based on sieve-element plastids, that urticalean rosids are not directly related to either the higher hamamelids or Malvales.
A complex of Rhamnaceae, Elaeagnaceae, the monotypic Barbeyaceae (Arabia and northeastern Africa; see Bouman and Boesewinkel, 1997
; Thulin et al., 1998
), and the enigmatic, monotypic Dirachma (Socotra, south of Yemen; previously placed in Geraniaceae but more recently in its own family; see Boesewinkel and Bouman, 1997
) comprises, at least in part, the sister group to the urticalean clade. The placement of Barbeya outside urticalean rosids and in a complex of taxa including Dirachma, Elaeagnaceae, and Rhamnaceae is consistent with recent molecular analyses (Thulin et al., 1998
; Richardson et al., 2000
; Savolainen et al., 2000a
, b
; Soltis et al., 2000
) and with the suggestion of Tobe and Takahashi (1990)
, based on trichome and pollen morphology, that Barbeyaceae are sufficiently different to be removed from the urticalean clade. Likewise, this placement is consistent with Berg's (1989)
conclusion that Barbeyaceae are not closely related to any of the urticalean families and that their placement with the urticalean rosids has persisted not because of shared characters, but rather due to the lack of any clear alternative placement. However, Bouman and Boesewinkel (1997)
argued that seed characteristics of Barbeya indicate an undisputed urticalean affinity. Of the other members of the sister group to urticalean rosids, Elaeagnaceae and Rhamnaceae have been suggested previously as putative relatives (Hallier, 1905
; Thorne, 1968
). As discussed in Qiu et al. (1998)
, the relationship of urticalean rosids to these taxa is supported in part (with some reversals or parallelisms) by a mitochondrial intron feature, polyembryony, tannin production, anomocytic stomata, calcium oxalate crystals in parenchymatous tissue, and dense silver tomentum formed by the curly, unicellular trichomes on the abaxial leaf surface. To ensure monophyletic groupings, it is thus best to consider families of urticalean rosids and their close relatives as members of a larger order Rosales (sensu APG, 1998
). Within the Rosales this monophyletic group can be diagnosed by cystoliths of calcium carbonate within specialized cells (lithocysts), reduced and inconspicuous flowers with five or fewer stamens, and two-carpellate, unilocular ovaries with a single apical to basal ovule (Judd et al., 1999
).
The lack of well-resolved relationships of the urticalean lineage and various other clades within the Rosales to each other prevents at the present time detailed biogeographic analysis of urticalean rosids. To understand the biogeographic origin and radiation within this group, one must assess biogeographical patterns from its sister group within Rosales and patterns at the base of urticalean rosids. The sister relationship of the urticalean lineage to a complex of families including Rhamnaceae, Elaeagnaceae, Barbeyaceae, and Dirachmaceae indicated in Fig. 2 is very weak. Biogeographical patterns in this latter group are also complex: Rhamnaceae are cosmopolitan, especially common in the tropics; Elaeagnaceae occur mainly in the temperate and warm northern hemisphere with extensions to the Old World tropics; and the monotypic Barbeyaceae and Dirachmaceae are confined to the arid Middle East and Socrota. Likewise, the early splitting lineages within urticalean rosids exhibit a widespread geographical pattern. The first split in the Ulmaceae involves (1) Holoptelea from the paleotropics and Ampelocera from the neotropics and (2) other genera from both areas of the Northern Hemisphere. The first several splits within Celtidaceae are genera restricted to either the neotropics or paleotropics. Considering these patterns within urticalean rosids and their sister group in the Rosales and that both the more derived Moraceae and Urticaceae are widespread in tropical (and to some extent temperate) regions of the world, it seems reasonable to propose that the urticalean lineage at least initially diversified in very warm temperate or more likely tropical regions, but it is difficult to ascertain with any confidence whether this lineage arose in the New World or the Old World. Based on a calibration of nodes within the angiosperm tree using a three-gene data set, Wikström, Savolainen, and Chase (2001)
estimated the origin of the urticalean clade at 65–67 million years ago (myr) and the separation of Ulmaceae vs. other urticalean families at 55–57 myr. However, fossil evidence points to earlier origins of urticalean rosids (reviewed in Manchester, 1989b
) with pollen combining both ulmoid and celtoid features seen at the beginning of the Turonian stage of the Upper Cretaceous (ca. 90 myr), Celtis-like pollen from the late Turonian, and ulmoid-like leaves from the Santonian stage of the Upper Cretaceous (ca. 85 myr). The greater ability for movement among continents at these times (Raven and Axelrod, 1974
) might explain the difficulty in ascertaining biogeographical origins of urticalean rosids or their subclades.
Circumscription of Ulmaceae and Celtidaceae
Consistent with previous cpDNA restriction site and matK analyses of urticalean rosids (Wiegrefe, Sytsma, and Guries, 1998
; Song et al., 2001
) is the very strong support for Ulmaceae, their sister status to the remainder of the urticalean clade, and thus the recognition of the two families Ulmaceae and Celtidaceae (Wiegrefe, Sytsma, and Guries, 1998
). The distinctness of Ulmaceae in relation to all other families is evident in a number of characters. Strictly unisexual flowers (Fig. 4A), curved embryos and loss of hypanthium (Fig. 4B) are synapomorphies for all urticalean rosids minus Ulmaceae. Although strictly bisexual flowers are retained in some of the genera of Ulmaceae, other genera have mixed bisexual and unisexual flowers (Fig. 4A). In addition, the shift from 2–3 pored pollen to 4–6 pored pollen and from secondary leaf veins not ending in teeth to secondary leaf veins ending in teeth defines Ulmaceae (Fig. 4C). The inclusion of the problematic Ampelocera within Ulmaceae corroborates the results of both Wiegrefe, Sytsma, and Guries (1998)
and Ueda, Kosuge, and Tobe (1997)
; Ampelocera was not sampled in the matK study of Song et al. (2001)
. However, this placement is in striking disagreement with the morphological cladistic analysis of Zavada and Kim (1996)
in which Ampelocera was sister to the rest of Celtidaceae. The use of fruit type in once separating Ulmoideae and Celtoideae within Ulmaceae sensu lato is clearly suspect as the derived, winged fruits (and other modified water-dispersed fruits) have originated at least three times from ancestral fleshy fruits in these two taxa (Fig. 4H).
As mirrored in Wiegrefe, Sytsma, and Guries (1998)
and Song et al. (2001)
, but not in the less sampled rbcL study of Ueda, Kosuge, and Tobe (1997)
, Celtidaceae (with Cannabaceae) are monophyletic but not strongly so and appear to consist of several distinct lineages. All genera suggested to belong to the family (and likewise to Ulmaceae) based on cpDNA restriction sites (Wiegrefe, Sytsma, and Guries, 1998
), matK sequences (Song et al., 2001)
, or morphology (Wiegrefe, Sytsma, and Guries, 1998
), do belong to Celtidaceae. A larger clade of Celtis, Pteroceltis, Trema, Parasponia, and Lozanella is supported by pollen morphology (Takahashi, 1989
), seed coat morphology (Takaso and Tobe, 1990
), vernation (Terabayashi, 1991
), and n = 10 chromosome number (Oginuma, Raven, and Tobe, 1990
). Although Lozanella does not form a clade with these other four genera in the larger rbcL analysis (Fig. 2), this is a weak region of the tree and this larger clade is recovered in one step longer trees. The combined data analysis (Fig. 3), however, strongly supports (91% bootstrap, Bremer support of 5) the placement of Lozanella within this clade.
The problematic Aphananthe, Chaetachme, and Gironniera are largely tropical or subtropical genera from Asia and Africa and, as argued by Wiegrefe, Sytsma, and Guries (1998)
, are critical taxa within Celtidaceae to fully understand the evolution and biogeography of the family. Aphananthe is clearly allied with the n = 10 clade of Celtidaceae based on pollen structure (Kuprianova, 1962
; Takahashi, 1989
), vernation (Terabayashi, 1991
), and gynoecial vasculature (Omori and Terabayashi, 1993
), but differs in seed coat morphology (Takaso and Tobe, 1990
), ovule anatomy (Takaso, 1987
), presence of flavonols rather than glycoflavones (Giannasi, 1978
), and the plesiomorphic chromosome number of n = 14 (Oginuma, Raven, and Tobe, 1990
). Chaetachme is also isolated from the n = 10 clade of Celtidaceae based on pollen (Takahashi, 1989
), vernation (Terabayashi, 1991
), possibly seed coat morphology (Takaso and Tobe, 1990
), and a chromosome number of n = 15 (Todzia, 1993
). The phylogenetic position of Gironniera has been even more problematic. Gironniera possesses glycoflavones typical of Celtidaceae, but flavonols as in Ulmaceae and Aphananthe (Giannasi, 1978
). The placement of Gironniera near the base of Celtidaceae, as argued by Giannasi (1978)
, is indicated by its unique type of pollen (Takahashi, 1989
), vernation (Terabayashi, 1991
), seed coat morphology (Takaso and Tobe, 1990
), and n = 14 chromosome number (Oginuma, Raven, and Tobe, 1990
).
Thus the results presented here, with the problematic Aphananthe, Chaetachme, and Gironniera all shown to be early diverging lineages within Celtidaceae (Fig. 2), are consistent with these morphological (also see Fig. 4A–C), cytological, and chemical features, a preliminary and less resolved morphological cladogram of Ulmaceae and Celtidaceae (Zavada and Kim, 1996
), and the less sampled matK analysis of Song et al. (2001)
in which only five genera of Celtidaceae were examined. Moreover, these molecular results support the scenario that n = 14 is plesiomorphic within Celtidaceae and possibly urticalean rosids as a whole, with parallel aneuploid increase and reduction occurring in the lineages leading to Chaetachme (to n = 15) and to the n = 10 clade of some Celtidaceae plus Cannabaceae (Fig. 4D). The biogeographical origin of Celtidaceae is most likely tropical, with Aphananthe sister to the rest of the family and one of the remaining two clades (Gironniera, Chaetachme, and Lozanella) strictly tropical (Fig. 2).
Relationships of Cannabaceae
Cannabaceae either have been linked with or included in Moraceae (e.g., Engler, 1889
; Judd, Sanders, and Donoghue, 1994
) or placed in an unresolved position in the urticalean lineage (e.g., Berg, 1989
; Humphries and Blackmore, 1989
). The unclear relationships of Cannabaceae to other families are likely due in part to its small size and unusual morphological features (Berg, 1989
). The three species of Cannabaceae share with Moraceae, Ulmaceae, and Celtidaceae an ovary with two styles (sometimes reduced) and an apical, pendulous, and anatropous ovule vs. the pseudomonomerous ovary with one style and a basal, erect, and orthotropous ovule as seen in Urticaceae and Cecropiaceae (Fig. 4E). Cannabaceae share with Ulmaceae and Celtidaceae the lack of milky latex (although laticifers are present) but differ from these two woody families in being herbs or herbaceous vines. Laticifers throughout all plant parts define Moraceae and Cannabaceae (Fig. 4G), but these may well be of different types. Cannabaceae are often considered to lack laticifers, although Judd, Sanders, and Donoghue (1994)
disagreed based on Metcalfe (1966)
. In any case, the laticifers present in Cannabaceae lack the milky latex seen in Moraceae. Cannabaceae also share with Ulmaceae and Celtidaceae the apparently plesiomorphic feature of polyembryony, also seen in Rhamnaceae, but lost (and thus synapomorphic) in Moraceae, Urticaceae, and Cecropiaceae (Dahlgren, 1991
; Qiu et al., 1998
).
The plastid DNA results presented here provide support for the origin of the Cannabaceae (Cannabis and Humulus) from within the n = 10 clade of Celtidaceae (Figs. 2–3, 4D), a result first suggested based on a cpDNA restriction site analysis (Wiegrefe, Sytsma, and Guries, 1998
), recently confirmed with plastid matK (Song et al., 2001)
, and with preliminary nuclear 26S rDNA (unpublished data). Although Cannabaceae are strongly imbedded in the Celtidaceae (98% bootstrap by iterative exclusion of some Celtidaceae; see RESULTS above), their exact sister lineage is uncertain and changes with matK (Song et al., 2001)
, rbcL (Fig. 2), trnL-F (tree not shown; available at http://ajbsupp.botany.org/v89/), ndhF (tree not shown; available at http://ajbsupp.botany.org/v89/), and combined data sets (Fig. 3). However, at least Pteroceltis, singly or with other genera, is implicated in all analyses. Considering that portions of Celtidaceae and Cannabaceae alone in urticalean rosids share the derived base chromosome number x = 10 (n = 10 for Cannabis, n = 8 for Humulus; Mehra and Gill, 1974
), this relationship has independent support. In addition, Cannabis, Humulus, and at least Pteroceltis of the n = 10 clade in Celtidaceae share a distinctive S-type sieve-element plastid (Fig. 6.5 in Behnke, 1989
), a remarkable similarity not indicated by Behnke (1989)
perhaps because of the novelty of such a relationship at that time. The micropapillate/smooth surface sculpturing of the nonglandular trichomes in Cannabaceae is also similar to that seen in the n = 10 clade of Celtidaceae, a combination of sculpturing seen only sporadically elsewhere in isolated genera of Urticaceae (Tobe and Takaso, 1996
). The many morphological features examined in the past for addressing the issue of familial status of Ulmaceae and Celtidaceae (reviewed in Wiegrefe, Sytsma, and Guries, 1998
) need now to be reexamined based on the clear finding that Cannabaceae are derived within Celtidaceae. As indicated by Wiegrefe, Sytsma, and Guries (1998)
and echoed by Song et al. (2001)
, Cannabaceae should be combined with Celtidaceae (see below for implications on the name of this family).
Relationships of Moraceae, Urticaceae, and Cecropiaceae
In sharp contrast, however, to the results of three previous cladistic analyses of urticalean rosids based on morphology (Humphries and Blackmore, 1989
; Judd, Sanders, and Donoghue, 1994
; Zavada and Kim, 1996
), the plastid DNA results provide strong support for the monophyly of Moraceae and of Urticaceae including Cecropiaceae (Figs. 2–3; Table 1). These previous morphological analyses (Fig. 1) generated conflicting results but uniformly demonstrated that Moraceae are not monophyletic and that Moraceae and Urticaceae/Cecropiaceae are often nested within a clade including Celtidaceae (with or without Cannabaceae). These studies have thus argued for either a more inclusive family delimitation (e.g., Urticaceae to include Moraceae, Cannabaceae, and Cecropiaceae but not Celtidaceae: Judd, Sanders, and Donoghue, 1994
; and Celtidaceae to include all urticalean rosids except for Ulmaceae: Zavada and Kim, 1996
) or the break up of Moraceae into smaller monophyletic groups (Humphries and Blackmore, 1989
). Thus, the molecular results do not support these previous findings including the derivation of Urticaceae from within a paraphyletic Moraceae as argued by Judd, Sanders, and Donoghue (1994)
.
Previous morphological comparisons and these plastid DNA results (Figs. 2–3; Table 1) indicate that Cecropiaceae in the broad sense are (1) more closely related to Urticaceae than to Moraceae, (2) derived from within Urticaceae, and (3) polyphyletic with the Asian-Australian Poikilospermum separately derived from within Urticaceae. Five of the six genera of Cecropiaceae are distinct morphologically from both Moraceae and Urticaceae (see Berg, 1978
; reviewed in Berg, 1989
; Berg, Akkermans, van Heusden, 1990
; and Setoguchi et al., 1993
) based on a suite of characters that includes stamen inflexation, gynoecium, fruit type, inflorescence, laticifers, wood, and cystoliths. Berg (1978)
considered Cecropiaceae intermediate between Moraceae, with which they share possession of straight stamens, lack of elongated cystoliths, and tendency to woody habit, and Urticaceae, with which they share orthotropous and sub-basal or basal ovules, a single stigma, and reduced laticifers (Fig. 4E, G; Table 3). Thus these molecular analyses indicate minimally that the single stigma, basal and orthotropous ovule, and laticifers restricted to bark are synapomorphies for Cecropiaceae and Urticaceae.
