American Journal of Botany 92(4): 653±666. 2005.

VIBURNUM PHYLOGENY BASED ON COMBINED MOLECULAR DATA: IMPLICATIONS FOR AND BIOGEOGRAPHY1

RICHARD C. WINKWORTH2 AND MICHAEL J. DONOGHUE

Department of Ecology and Evolutionary Biology, Yale University, P.O. Box 208106, New Haven, Connecticut 06520-8106 USA

We investigated phylogeny using separate and combined analyses of DNA sequence data from two chloroplast and three nuclear loci. Separate analyses of nuclear and chloroplast data sets resulted in gene that were generally congruent with one another and with trees from two previous analyses. Our gene trees do differ in the position of section Pseudotinus, as well as in species relationships within sections Pseudotinus and Lentago. However, tests for incongruence indicate that differences between the nuclear and chloroplast data are not signi®cant. Furthermore, gene trees from combined analyses were highly similar to those found in separate analyses, suggesting that these localized differences do not affect other parts of the . Our analyses provide convincing support for numerous relationships, although there is still uncertainty at the base of the tree. To facilitate future study, we propose informal names for 12 well-supported species groups, as well as for several higher-level clades. We also discuss the biogeographic implications of our phylogeny, focusing on repeated, although apparently temporally incongruent, patterns of disjunction between the Old and New Worlds.

Key words: biogeography; data combination; phylogeny; taxonomy; Viburnum.

Viburnum (, ) is a clade of ca. 160 nized four subsections within Megalotinus, and Hara (1983) species of and small trees. Although broadly distributed proposed subsections within sections Odontotinus, Viburnum, around the Northern Hemisphere, Viburnum has centers of and Solenotinus. Regional treatments have dealt with the Asian species diversity in eastern Asia and Latin America (Rehder, (Hara, 1983; Rehder, 1908; Hsu et al., 1988), Malesian (Kern, 1908; Hara, 1983; Donoghue, 1983a), and its range extends 1951; Kern and van Steenis, 1951), South American (Killip into the Southern Hemisphere in the mountains of southeast and Smith, 1929, 1931), Mexican and Central American (Mor- Asia (Kern, 1951) and South America (Killip and Smith, ton, 1933; Donoghue, 1982), and North American species 1931). The group displays relatively little variability in ¯ower (McAtee, 1956). These have varied in scope, but mostly con- and morphology, but conspicuous diversity in numerous tain only keys and descriptions; species relationships within or other characters; variation in endocarp shape, fruit color, between regions have seldom been considered. morphology, morphology, and in¯orescence form is par- The monophyly and phylogenetic placement of Viburnum ticularly striking. are now clearly established (e.g., Wilkinson, 1948; Donoghue, Linnaeus (1753) originally erected Viburnum. A later revi- 1983b; Pyck et al., 1999; Bell et al., 2001; Donoghue et al., sion by Oersted (1861) clearly identi®ed the principal variable 2001, 2003). In contrast, despite an earlier morphological characters within the group, although his proposed subdivision study (Donoghue, 1983a) and several unpublished molecular into ®ve genera was never widely adopted. Since Oersted's data sets (see abstracts by Donoghue and Sytsma, 1993; Don- (1861) treatment, few botanists have considered the classi®- oghue and Baldwin, 1993; see also Baldwin et al., 1995) evo- cation of Viburnum as a whole. One exception was Rehder lutionary relationships within Viburnum have remained largely (1940), who, although concentrating on the Asian species, rec- unresolved. However, two recent molecular phylogenetic stud- ognized nine sections. More recently, Hara (1983) greatly clar- iesÐDonoghue et al. (2004) using chloroplast trnK intron and i®ed the subgeneric nomenclature with his review of the sec- nuclear ribosomal ITS DNA sequences, and Winkworth and tion-level classi®cation. This treatment recognized 10 sections Donoghue (2004) using sequences from the low-copy number (Table 1)Ðthe nine suggested by Rehder (1940) plus section nuclear gene GBSSIÐhave improved our understanding of re- Oreinotinus to accommodate the Latin American species, for lationships within Viburnum. In broad terms, they con®rm the which Morton (1933) had proposed 10 sections. There have monophyly of many section-level taxa. Section Odontotinus been few section level revisions. The exceptions include a was the most striking exception, being divided into (1) a pur- thorough treatment of the eastern North American species of ple-fruited North American lineage that also includes the Latin section Lentago (Jones, 1983) and a monograph of several species complexes in Mexico and Central America (Dono- American section Oreinotinus, and (2) a mostly red-fruited ghue, 1982). In other studies, Hsu (1975) segregated a new Asian lineage that also contains the North American purple- section (Platyphylla) from Megalotinus, Kern (1951) recog- fruited species, V. acerifolium. Sections Lentago, Solenotinus, and Viburnum were also found to be nonmonophyletic. How- 1 Manuscript received 5 June 2004; revision accepted 7 January 2005. ever, for each of these groups, a core clade contains all but The authors thank Bob Cook and the staff of the Arnold Arboretum for one representative of the section (see below). allowing us to make extensive collections on the grounds and for their general Perhaps more importantly, the studies of Donoghue et al. support. Lab groups at Harvard and Yale provided valuable feedback. We (2004) and Winkworth and Donoghue (2004) provided con- thank ValeÂry MaleÂcot for input on species numbers. We also thank Richard Olmstead and an anonymous reviewer for helpful comments on an earlier sistent resolution of relationships among the sectional-level version. lineages, identifying three well-supported suprasectional 2 (E-mail: [email protected]); Telephone: 203-436-4992. clades. The largest suprasectional lineage contains subclades 653 654 AMERICAN JOURNAL OF BOTANY [Vol. 92

TABLE 1. Traditionally recognized sections of Viburnum (e.g., Hara, 1983).

Approximate number Section of species Generalized geographic range Lentago 7 Eastern , except V. elatum in Mexico Megalotinus 18 Southeast Asia, extending west to India and south to Indonesia Odontotinus 37 Temperate Asia and Eastern North America, except V. orientale in the Caucasus Mountains Opulus 5 Circumboreal Oreinotinus 38 Mexico, Caribbean, Central and South America Pseudotinus 4 Asia, except V. lantanoides in Eastern North America Solenotinus 26 Asia, extending west to India and south to Indonesia Tinus 7 Asia, except V. tinus in Europe Tomentosa 2 China, Japan Viburnum 14 Asia, except V. lantana in Europe

corresponding to (1) section Opulus, (2) section Tinus, (3) the lationships within Viburnum by combining the data sets of New World Odontotinus plus Oreinotinus, and (4) the Old Donoghue et al. (2004) and Winkworth and Donoghue (2004) World Odontotinus±acerifolium lineage plus V. cylindricum of with sequences from an additional chloroplast marker, the section Megalotinus. Each of the two remaining suprasectional psbA-trnH intergenic spacer (IGS). For our analyses, we lineages consists of a well-supported pair of section level formed new data partitions. Speci®cally, a chloroplast data set clades. In one of these, V. plicatum (section Tomentosa)is that included trnK intron sequences from Donoghue et al. sister to the core of section Solenotinus; in both studies this (2004) and our psbA-trnH IGS sequences, and a nuclear par- grouping is weakly associated with V. urceolatum (tradition- tition that consisted of nrITS (Donoghue et al., 2004) and ally of section Viburnum). The third suprasectional lineage GBSSI (Winkworth and Donoghue, 2004) sequences. We used contains the core species groups of sections Lentago and Vi- gene trees from analyses of chloroplast, nuclear, and combined burnum. This lineage also includes V. nudum (traditionally of data sets to test phylogenetic hypotheses suggested by earlier section Lentago), although its exact relationship to these sec- studies, to investigate differences between the data sets, and tion-level clades remains unclear. to clarify relationships left unresolved in previous analyses. The most striking difference between the gene trees inferred The results provide a ®rm basis for reclassi®cation of Vibur- by Donoghue et al. (2004) and those of Winkworth and Don- num and for a preliminary biogeographic analysis. oghue (2004) relates to the duplication of the GBSSI paralogs in several Viburnum lineages. Speci®cally, V. erubescens, V. MATERIALS AND METHODS odoratissimum, and V sieboldii of section Solenotinus and the Oreinotinus±dentatum clade were represented twice on GBSSI Taxon samplingÐFor this study, we included the same set of taxa (i.e., 41 gene trees. These lineage-speci®c duplications of the GBSSI Viburnum species plus an outgroup, Sambucus canadensis) that were used by Winkworth and Donoghue (2004). This sample includes approximately one loci are consistent with karyological evidence for polyploidy quarter of all Viburnum species, but more importantly it includes represen- in these groups (Winkworth and Donoghue, 2004). Although tatives of all traditionally recognized sections and major species complexes. in these two lineages the placement of GBSSI homoeologues Although we broadly covered Viburnum diversity, our sampling within certain tends to favor an allopolyploid origin, such inferences are clades was limited. In particular, section Megalotinus was represented by just complicated by (1) loss or failure to identify homoeologues in one of the 18 recognized species. certain taxa and (2) placement of related diploid species. As noted by Winkworth and Donoghue (2004), data from addi- Molecular methodsÐIn general, molecular methods (i.e., genomic DNA tional nuclear loci are needed to con®rm inferences about the isolation, PCR ampli®cation, fragment puri®cation, and sequencing) followed origins of polyploidy in Viburnum. the protocols described by Winkworth and Donoghue (2004). For PCR, we In addition to differences in locus copy number, there were used the oligonucleotide primers psbAF and trnHR (Sang et al., 1997) and a several localized regions of uncertainty or con¯ict between thermocycling pro®le that incorporated a 53ЊC annealing temperature and a gene trees recovered in the two studies. In most cases, the 1-min extension time. For some accessions, direct sequencing of PCR frag- alternative arrangements seem to re¯ect limited phylogenetic ments consistently failed due to the presence of long adenosine or thymine signal rather than differences in evolutionary history. How- strings. In such cases, ampli®cation products were cloned, screened, and se- ever, one possible exception involves relationships within the quenced as described by Winkworth and Donoghue (2004). Voucher infor- core of section Lentago. Here, differences between the rela- mation for the accessions used in this study is as reported in Winkworth and tionships suggested by nuclear and chloroplast markers are Donoghue (2004); accession details and GenBank numbers for our psbA-trnH consistent with an earlier hypothesis suggesting homoploid hy- IGS sequences are available in the Appendix (see Supplemental Data accom- brid speciation (Donoghue et al., 2004). Perhaps the most crit- panying the online version of this article). ical outstanding issue involves relationships at the base of Vi- Sequence alignments and data matricesÐPreliminary multiple sequence burnum. Although the analyses of Donoghue et al. (2004) and alignments for the psbA-trnH IGS DNA sequences were performed using Winkworth and Donoghue (2004) have resolved many major ClustalX (Thompson et al., 1994), followed by visual inspection. Prior to clades, relationships near the base remain uncertain. Speci®- phylogenetic analyses, all ambiguous positions were excluded from the data cally, (1) the placement of section Pseudotinus, (2) relation- matrix. Furthermore, to remain consistent with the data matrices of Donoghue ships among the suprasectional lineages, V. urceolatum, and et al. (2004) and Winkworth and Donoghue (2004), we excluded all gapped V. clemensiae, and (3) the position of the root all remain un- positions from our psbA-trnH IGS data set. We discuss the compatibility of resolved. several parsimony informative gaps with the topology based on nucleotide In the present study, we further investigate phylogenetic re- substitutions in the combined chloroplast data set. April 2005] WINKWORTH AND DONOGHUEÐVIBURNUM PHYLOGENY 655

For phylogenetic analyses, we prepared three data matrices, each containing cates with NNI branch swapping; parameters values were estimated from the the same taxa as used by Winkworth and Donoghue (2004). A chloroplast optimal ML tree. Our trees and data matrices are available in TreeBASE data set was compiled by combining the trnK intron matrix from Donoghue (study accession number S1153, matrix accession numbers M1980±1982). et al. (2004) and our new psbA-trnH IGS sequences. All taxa were represented by both loci in this matrix. A nuclear data set was prepared by combining the Biogeographic reconstructionsÐWe investigated biogeographic disjunc- nrITS data matrix (Donoghue et al., 2004) with the GBSSI matrices from tions between the Old World and the New World using dispersal-vicariance Winkworth and Donoghue (2004). In most species of Viburnum, GBSSI is analysis (Ronquist, 1997) as implemented in the program DIVA version 1.1 present as two paralogues; Winkworth and Donoghue (2004) characterized (Ronquist, 1996). This method reconstructs historical biogeography by min- the region between exons 8 and 12 for both of these loci. In addition to these imizing the number of vicariance, dispersal, and extinction events required to main copies, a pair of homoeologues were identi®ed from several species. We explain speci®c distribution patterns. For this analysis, we used the maximum considered several methods for combining the GBSSI matrices with the nrITS likelihood topology from our analysis of the full combined data set, with one data set, in which each species is represented by a single sequence (e.g., modi®cation. Molecular analyses do not strongly support the placement of uninode coding, Simmons et al., 2000). However, to account for the GBSSI the eastern North American species V. nudum; based on morphological evi- homoeologues in our combined data set, either nrITS or GBSSI sequences dence (Donoghue, 1983a; Jones, 1983), we resolved this taxon as sister to the would need to be duplicated, thereby increasing the in¯uence of that partition remainder of section Lentago. We used the default cost matrix, which assigns on the resulting gene tree. We opted instead to include in the nuclear matrix a cost of 1 to dispersal and extinction events, and 0 to vicariance. those GBSSI homoeologues that were topologically congruent with the nrITS phylogeny of Donoghue et al. (2004). Speci®cally, we used the Winkworth RESULTS and Donoghue (2004) B homoeologues from the Oreinotinus±dentatum clade and the D homoeologues from V. erubescens, V. odoratissimum, and V. sie- Aligned psbA-trnH IGS DNA sequencesÐWe were unable boldii. Winkworth and Donoghue (2004) did not recover GBSSI sequences to directly sequence the psbA-trnH IGS for three Viburnum for several taxaÐWAXY1 sequences are missing for V. odoratissimum, as are species, speci®cally, V. clemensiae, V. furcatum, and V. sar- WAXY2 sequences for V. dentatum, V. edule, section Tinus, and Sambucus gentii. For these, we cloned PCR fragments, obtaining DNA canadensis. These sequences are represented as missing data in our nuclear sequences from four or ®ve clones. Preliminary analyses in- matrix. We prepared a combined data set by concatenating our chloroplast dicated that clones from a given accession formed a clade in and nuclear matrices. the Viburnum phylogeny. Therefore, we selected a single rep- resentative of each species for inclusion in the ®nal matrix. ILD and Templeton testsÐWe tested for incongruence between data par- Our psbA-trnH IGS DNA sequences for 41 Viburnum ac- titions within the chloroplast and nuclear data matrices using the incongruence cessions and Sambucus canadensis ranged in length from 410 length difference (ILD) test as implemented in PAUP* 4.0b10 (Swofford, to 444 nucleotides, resulting in an aligned data matrix of 525 2002). For the chloroplast matrix, this required a single pairwise comparison; nucleotides positions. Although this alignment contained sev- for the nuclear data set, we made all pairwise tests and a simultaneous com- eral gaps, the high degree of nucleotide similarity resulted in parison of all three markers. a matrix with little alignment ambiguity. There were only two We also tested for incongruence between the chloroplast and nuclear ma- regions where alignment proved problematic. In one case, this trices. Based on previous studies and our preliminary analyses, we identi®ed several potential sources of con¯ict between these two data setsÐspeci®cally, was due to several overlapping gaps (i.e., alignment positions relationships within sections Lentago and Pseudotinus and the placement of 351±377), in the other it resulted from the variable length of Pseudotinus. To test these differences, we conducted a series of ILD tests to a string of adenosine residues (i.e., alignment positions 455± compare levels of incongruence in the complete data set with those in matrices 461). Following exclusion of all gapped and ambiguous sites, containing different subsets of taxa from sections Lentago and Pseudotinus. the ®nal data set contained 330 nucleotide positions; this in- We used this approach to evaluate the relative contribution of the different cluded 62 varied positions, of which 24 were parsimony in- areas of con¯ict to the overall level of incongruence. We also examined dif- formative. ferences between the chloroplast and nuclear data using the signi®cantly less parsimonious test (SLP test; Templeton, 1983). In these tests, we used the Combined analyses of chloroplast sequencesÐThe chlo- 70% majority rule consensus of MP trees as our ``test tree'' and constructed roplast data matrix contained a total of 1461 nucleotides po- a ``rival tree'' by constraining the node of interest to re¯ect the topology sitions; 1131 from the trnK intron and 330 from the psbA- suggested by the other data set (``test'' and ``rival'' are used as in Mason- trnH IGS. This included 178 varied positions, of which 68 Gamer and Kellogg, 1996). For example, the rival topology for evaluating the were parsimony informative. The ILD test indicated no sig- placement of section Pseudotinus suggested by the chloroplast data was iden- ni®cant con¯ict between these two partitions (P ϭ 0.681). A tical to the test tree except that Pseudotinus was placed sister to the Soleno- MP analysis of the combined chloroplast data set recovered tinus±Tomentosa clade. Tests were performed as implemented in PAUP* 10 optimal trees, each 222 steps long with a consistency index 4.0b10. (CI) of 0.892 (without autapomorphies, CI ϭ 0.774) and a retention index (RI) of 0.930. These trees differed only in (1) Phylogenetic analysesÐWe inferred phylogenetic relationships using both the monophyly and inferred relationships of section Tinus, and the maximum parsimony (MP) and maximum likelihood (ML) optimality cri- (2) the placement of V. odoratissimum relative to the remain- teria as implemented in PAUP* 4.0b10. MP analyses used heuristic search ing core species of section Solenotinus. The ML search used methods, tree-bisection-reconnection (TBR) branch swapping, zero-length a GTR ϩ G substitution model and resulted in a single to- branches collapsed, and all characters equally weighted. Analyses were re- Ϫ ϭ peated 100 times with RANDOM ADDITION. Parsimony bootstrap analyses pology ( lnL 3468.22102; Fig. 1) that was highly similar were performed with 1000 replicates and nearest neighbor interchange (NNI) to those recovered using parsimony. Bootstrapping indicated branch swapping. For ML analyses, the best-®tting evolutionary model was moderate (i.e., greater than 65%) to strong (i.e., greater than found using a hierarchical series of likelihood ratio tests as implemented in 80%) support for many of the major clades. PORN* (Bell, 2001). Subsequent heuristic searches used the best-®t model As described by Donoghue et al. (2004), the alignment of (model parameters were simultaneously estimated using ML), TBR branch trnK intron sequences contained a single gap that was parsi- swapping, and collapsed zero-length branches. Analyses were repeated 100 mony informative with respect to ingroup relationshipsÐthis times with RANDOM ADDITION. ML bootstrap analyses used 1000 repli- deletion, which marks the Pseudotinus clade, is also consistent 656 AMERICAN JOURNAL OF BOTANY [Vol. 92

Fig. 1. Single optimal maximum likelihood tree obtained from the analysis of combined trnK intron and psbA-trnH IGS sequences for Viburnum, showing bootstrap values over 50% (1000 replicates). Names of traditionally recognized sections are noted at the right. with our combined chloroplast analyses based on nucleotide signi®cant con¯ict was detectedÐP values ranged from 0.763 substitutions. As noted earlier, the alignment of psbA-trnH IGS for the nrITS/WAXY2 comparison to 0.067 for the three-mark- sequences contained several gaps. Unique and unambiguous er comparison. All taxa were included for phylogenetic anal- indels support the core species group of section Solenotinus ysis, with absences coded as missing data; missing cells ac- (two insertions and a deletion) and the Oreinotinus±dentatum counted for approximately 6% of the data matrix. The MP clade (one insertion). search recovered 432 equally parsimonious trees, each 876 steps long with CI ϭ 0.731 (without autapomorphies, CI ϭ Combined analyses of nuclear sequencesÐThe nuclear 0.568) and RI ϭ 0.812. The optimal MP trees differed in (1) matrix contained 2148 nucleotide positions; 556 from nrITS, the placement of Sambucus canadensis (and hence the root) 807 from WAXY1, and 785 from WAXY2. In this matrix, 141 and V. nudum, and (2) species relationships within the core positions were varied, of which 69 were parsimony informa- clade of section Solenotinus, within the Old World Odontotin- tive. ILD tests of the nuclear data matrix included only those us clade, and within both the Oreinotinus±dentatum and ra®- taxa that were represented in all three data sets. In all pairwise nesquianum species complexes. For the ML search a GTR ϩ comparisons and in a test that included all three markers, no I ϩ G substitution model was used, resulting in a single op- April 2005] WINKWORTH AND DONOGHUEÐVIBURNUM PHYLOGENY 657

Fig. 2. Single optimal maximum likelihood tree obtained from the analysis of combined nrITS, WAXY1, and WAXY2 sequences for Viburnum, showing bootstrap values over 50% (1000 replicates). Names of traditionally recognized sections are noted at the right. timal topology (ϪlnL ϭ 8446.05975; Fig. 2). This topology difolium and V. furcatum is inconsistent with our nuclear anal- was very similar to those found using parsimony, and many yses. of the inferred relationships were well supported by bootstrap analyses. Winkworth and Donoghue (2004) identi®ed several gaps in Tests for incongruence between the chloroplast and nu- the GBSSI alignments that were consistent with clades in their clear data setsÐVisual inspection of gene trees from our sep- gene trees based on nucleotide substitutions. In all but three arate analyses of chloroplast and nuclear data suggest several cases, the indels reported by these authors were also consistent areas of incongruence between these data sets (Figs. 1 and 2). with clades found in our combined analyses of GBSSI and Speci®cally, the chloroplast and nuclear trees differ with re- nrITS sequences. Two insertions in WAXY1Ðone uniting V. spect to (1) the position of Pseudotinus, (2) species relation- dilatatum, V. erosum, and V. melanocarpum, and one shared ships within Pseudotinus, and (3) species relationships within by V. edule and V. trilobumÐare incompatible with our nu- Lentago. We explored these possible con¯icts using the ILD clear trees; similarly an insertion in WAXY2 that unites V. cor- and SLP tests. 658 AMERICAN JOURNAL OF BOTANY [Vol. 92

TABLE 2. Incongruence length difference test scores from comparisons of chloroplast and nuclear data partitions in a series of Viburnum data matrices.

Data matrix NP Full taxa set 1 0.015 Single exemplars of both Lentago and Pseudotinus 15 0.078±0.271 Single exemplar of Lentago/all of Pseudotinus 5 0.019±0.044 All of Lentago/single exemplar of Pseudotinus 3 0.040±0.053 Single exemplar of Lentago and all Pseudotinus removed 5 0.882±0.968 All Lentago removed and single exemplar Pseudotinus 3 0.239±0.335 All Lentago included and all Pseudotinus removed 1 0.556 All Lentago removed and all Pseudotinus included 1 0.054 Both Lentago and Pseudotinus removed 1 0.959 Note: N ϭ number of tests.

ILD testsÐWe tested for incongruence between the nuclear the nuclear rival tree (P ϭ 0.0253), the reciprocal test was not and chloroplast partitions in the full combined data set (i.e., all statistically signi®cant (i.e., the number of characters that re- taxa for which all markers were represented were included) and quired different numbers of steps on the test and rival tree was a series of matrices that contained differing subsets of taxa. We not Ͼ5). Similarly, tests that used the chloroplast data and used this approach as a heuristic to examine the effect of taxa nuclear rival tree to evaluate within section differences were or clades on con¯ict between the chloroplast and nuclear data not signi®cant because of insuf®cient character differences be- sets. Results of these tests are summarized in Table 2. tween the test and rival trees. Within Lentago, the chloroplast An ILD test of the full combined data set is consistent with topology resulted in a signi®cantly worse ®t of the nuclear our observations of topological differences between chloro- data (P ϭ 0.0067) than the test tree. The same test of rela- plast and nuclear gene trees. That is, the P value (P ϭ 0.015) tionships within Pseudotinus could not reject the null hypoth- for this test indicates some con¯ict between these two data esis (P ϭ 0.1797). partitions. Furthermore, our tests of reduced data matrices sug- gest that the three areas of topological con¯ict we have iden- Combined analyses of chloroplast and nuclear DNA se- ti®ed do not contribute equally to the overall level of incon- quencesÐOur separate analyses and statistical tests suggested gruence. Speci®cally, differences between P values for data highly localized incongruence between the chloroplast and nu- sets that exclude Pseudotinus and matrices that contain one or clear data. In particular, our tests suggest that differences in more exemplars of this section are greater than between tests the placement of Pseudotinus make a substantial contribution that include all representatives of a section and tests that con- to the overall level of incongruence. To further explore the tain single exemplars. This suggests that topological con¯ict effect of this difference, we conducted a pair of combined within Pseudotinus and Lentago contributes less to the overall analyses, one using the full set of taxa and the other with level of incongruence than do differences in the overall place- representatives of section Pseudotinus removed. In both cases, ment of Pseudotinus. missing cells accounted for approximately 4% of the total cells in the data matrix. SLP testsÐThe SLP test provides a useful method for eval- uating incongruence between data matrices because it can be Full taxon setÐParsimony tree searches resulted in 288 applied at various topological levels (Seelanan et al., 1997). trees of 1109 steps, with CI ϭ 0.756 (without autapomorphies, In our tests, we directly compared alternative arrangements for CI ϭ 0.591) and RI ϭ 0.830. Optimal trees in this search the three incongruent regions noted above. Results of these differed with respect to (1) the relationships among the three tests are summarized in Table 3. suprasectional lineages, Pseudotinus, and V. clemensiae, and Tests on the placement of Pseudotinus are not conclusive. (2) species relationships within several clades (i.e., Pseudotin- Although the chloroplast data were a signi®cantly worse ®t to us, core Solenotinus, Old World Odontotinus, Oreinotinus± dentatum, and the ra®nesquianum species complex). A GTR ϩ I ϩ G substitution model was used in the ML search and TABLE 3. Signi®cantly less parsimonious test scores from comparisons resulted in a single optimal topology (ϪlnL ϭ 12163.30184; of chloroplast and nuclear data matrices for Viburnum. Fig. 3). This tree is generally consistent with those from sep-

Test data and tree Rival constraint NParate analyses of nuclear and chloroplast data sets, combining well-supported clades from each. Bootstrap tests provided Placement of section Pseudotinus moderate to strong support for many relationships within Vi- Chloroplast Nuclear 5 0.0253 burnum. Nuclear Chloroplast 4 0.0455 With respect to relationships that differed in separate chlo- Species relationships within section Pseudotinus roplast and nuclear analyses, the combined tree generally fa- Chloroplast Nuclear 3 0.0833 vored the nuclear resolution. In particular, the relationships Nuclear Chloroplast 5 0.1797 inferred within sections Pseudotinus and Lentago are identical Species relationships within section Lentago to those recovered in nuclear analyses. The combined tree also Chloroplast Nuclear 2 0.1573 favored a direct relationship between Pseudotinus and the Nuclear Chloroplast 11 0.0067 clade containing Solenotinus, Tomentosa, and V. urceolatum. Note: N ϭ the number of characters requiring a different number of In this case, however, the exact placement of Pseudotinus dif- steps to be reconstructed on the test and rival tree. fers between the nuclear and combined analyses, although April 2005] WINKWORTH AND DONOGHUEÐVIBURNUM PHYLOGENY 659

Fig. 3. Single optimal maximum likelihood tree obtained from the combined analysis of chloroplast and nuclear sequences for Viburnum, showing bootstrap values over 50% (1000 replicates). Names of traditionally recognized sections are noted at the right. bootstrap support for these relationships is weak. In any case, lationships inferred were well resolved and supported, except the inclusion of Pseudotinus in the combined analysis had little in those lineages where maximum parsimony trees differed. effect on the overall structure and support for clades elsewhere The ML tree was also largely consistent with those recovered in the tree. in separate analyses of chloroplast and nuclear data sets, as well as our full combined data set. In Lentago, where separate Minus PseudotinusÐThe MP search resulted in 144 trees of analyses suggested differing resolutions, the relationships sup- 1038 steps, with CI ϭ 0.761 (without autapomorphies, CI ϭ ported by the nuclear data were recovered in the combined 0.591) and RI ϭ 0.835. These trees differed slightly with re- analysis. spect to the relationships inferred within the core Solenotinus clade, the Old World Odontotinus clade, the Oreinotinus±den- Biogeographic reconstructionsÐDispersal-vicariance anal- tatum clade, and the ra®nesquianum species complex. ML ysis (Fig. 5b) suggests an initial diversi®cation of Viburnum analyses used a GTR ϩ I ϩ G substitution model and resulted in the Old World, with a number of transitions into the New in a single optimal topology (ϪlnL ϭ 11662.62886). This tree World. In four cases, the movement is unequivocally recon- was very similar to those recovered using parsimony; the re- structed as from the Old World to the New World. Speci®cally, 660 AMERICAN JOURNAL OF BOTANY [Vol. 92

V. acerifolium, V. lantanoides, Lentago, and the clade con- where the Old World Odontotinus species form a polytomy taining the New World Odontotinus and Oreinotinus all appear with V. cylindricum (Megalotinus). For the three remaining to be derived from Old World ancestors. Transitions between nonmonophyletic sections, our gene trees also support the pre- the Old and New Worlds are also inferred within Opulus, al- vious analyses, placing a single species outside of a mono- though in this case the directionality is uncertain. DIVA re- phyletic core group. For section Lentago, the core clade is constructions suggest either (1) that the New World taxa V. strongly supported (99±100% bootstrap support), but the edule and V. trilobum are independently derived from Old placement of V. nudum remained uncertain. In chloroplast World ancestors, or (2) there was a transition to the New World analyses, this species formed a polytomy with the core clades at the base of the clade and the Old World species, V. sargen- of sections Lentago and Viburnum; in nuclear and combined tii, represents a transition back to the Old World. trees, it was sister to a clade containing these two lineages (60 and 65% bootstrap support, respectively). In the case of sec- DISCUSSION tion Solenotinus, ®ve of the six representatives form a strongly supported clade (98±100% bootstrap support). In all of our When multiple data sets are available, a common question analyses, V. clemensiae was well separated from Solenotinus is whether or not to combine the individual sources into a and all other species groups; consistent with previous studies, single analysis. The issue of data combination has received our chloroplast and combined analyses suggest that this rare considerable attention in the literature and views on the most Bornean species is sister to the remainder of Viburnum (88 appropriate approach to this problem have varied from always and 73% bootstrap support, respectively). For section Vibur- to never combine (see de Queiroz et al., 1995; Johnson and num, the Taiwanese±Japanese species, V. urceolatum, is well Soltis, 1998). This lack of consensus re¯ects the knowledge separated from the core clade that is supported by bootstrap that different data sets (i.e., different gene loci or data types) values of 83±100%. In all of our optimal ML trees, V. urceo- may differ in their underlying evolutionary histories and, latum is consistently placed sister to a clade containing the therefore, provide alternative estimates of the organismal phy- core Solenotinus plus V. plicatum of section Tomentosa, albeit logeny. The issue is further confounded because it is clear that with limited support (see next section). there are several possible causes of topological incongruence, Only single representatives of sections Megalotinus and To- at least some of which are artifactual. Consequently, it has mentosa were included in our analyses. Additional sampling become important to assess whether the differences between will be required to assess the monophyly of these groups. data sets are of potential biological signi®cance. In this study, we have used several data sets to investigate phylogenetic relationships within Viburnum. Generally, our Suprasectional lineagesÐDonoghue et al. (2004) and Wink- gene trees are very similar to each other and also to those worth and Donoghue (2004) identi®ed three major subclades reported by Donoghue et al. (2004) and Winkworth and Don- within Viburnum. These groups were also recovered in our oghue (2004). However, our inclusion of additional chloroplast separate and combined analyses, generally with improved sequences, separate and combined analyses of the chloroplast bootstrap support. Further, within these lineages, relationships and nuclear partitions, and statistical testing of apparent to- among section-level clades were well resolved and supported. pological con¯icts provide several new insights. In the follow- The largest suprasectional clade contains the traditional sec- ing sections, we describe the key features of our trees and tions Megalotinus, Odontotinus, Opulus, Oreinotinus, and Tin- insights from the statistical tests. We compare our results to us (71±97% bootstrap support). This clade is further divided trees from earlier studies and discuss implications for our un- into four well-supported subclades (92±100% bootstrap sup- derstanding of Viburnum taxonomy and biogeography. port in the combined analysis), corresponding to (1) section Opulus, (2) section Tinus, (3) the New World Odontotinus and Gene treesÐIn this section, we discuss the impact of our section Oreinotinus, and (4) an Old World Odontotinus±ac- gene trees and statistical tests on our understanding of Vibur- erifolium lineage plus V. cylindricum of Megalotinus. Rela- num phylogeny. Given the overall similarity of our combined tionships among these lineages were unresolved in chloroplast chloroplast and nuclear analyses, we discuss results from anal- analyses, but based mainly on resolution provided by GBSSI, yses of the full taxa set, unless otherwise noted. our nuclear and combined analyses provide moderate to strong support for relationships among these clades. Section Opulus The monophyly of traditional sectionsÐOur analyses pro- appears to be sister to the other groups (75 and 93% bootstrap vide strong support for the monophyly of sections Opulus, support in nuclear and combined analyses, respectively). In Pseudotinus, and Tinus. Bootstrap support for these clades is turn, the Old World Odontotinus±acerifolium±cylindricum 87±100%, except in chloroplast analyses where Tinus receives clade is sister to one containing section Tinus and the New less than 50% bootstrap support. We also consider our results World Odontotinus±Oreinotinus lineage. In the ®rst clade, our to be consistent with the monophyly of section Oreinotinus analyses placed V. cylindricum at the base (78 and 70% boot- (c.f. Winkworth and Donoghue, 2004), although in this case strap support, respectively), with V. acerifolium and V. kan- the result is not well supported. suense united in a clade (97% bootstrap support in both anal- As in previous molecular studies, four of the traditional sec- yses) that is sister to the remainder of the Old World Odon- tions were found not to be monophyletic, speci®cally, Lentago, totinus (87% bootstrap support in both analyses). Within the Odontotinus, Solenotinus, and Viburnum. In the case of Odon- second group, Tinus is sister to an exclusively New World totinus, our analyses again suggest two major lineages, a pre- lineage (76 and 100% bootstrap support, respectively). This dominantly red-fruited Asian clade and a purple-fruited North New World group is divided into two clades, one containing American clade that also includes section Oreinotinus. Gen- V. ellipticum, V. molle, and V. ra®nesquianum (100% boot- erally, these two lineages are well supported (78±100% boot- strap support in both analyses) and the other containing V. strap support); the exception is in the chloroplast analyses dentatum of Eastern North America plus the Latin American April 2005] WINKWORTH AND DONOGHUEÐVIBURNUM PHYLOGENY 661 section Oreinotinus (99 and 100% bootstrap support, respec- help because including and its relatives would simply tively). add another exceptionally long branch to the problem. In our trees, the two remaining suprasectional lineages each contain a pair of section-level lineages. In one, the core clade Relationships within sectionsÐRelationships within the sec- of section Viburnum (i.e., excluding V. urceolatum) is directly tion-level clades were generally unresolved in our chloroplast linked to section Lentago (87±99% bootstrap support). This analyses. In contrast, the nuclear and combined data sets pro- grouping includes V. nudum, although its exact position is un- vided much greater resolution and support for these relation- certain. The third suprasectional lineage consists of core So- ships. In general, inferred relationships are highly similar to lenotinus (i.e., with the exception of V. clemensiae) and V. those reported by Winkworth and Donoghue (2004) based on plicatum of section Tomentosa (96±100% bootstrap support). GBSSI sequences alone. Here we focus on the two cases where This group also appears to include the morphologically dis- the chloroplast and nuclear data sets suggest different phylo- tinctive V. urceolatum. As in previous molecular analyses, our genetic historiesÐthe core of Lentago and Pseudotinus. gene trees consistently place this species sister to the Soleno- Our separate analyses of nuclear and chloroplast data sets tinus±Tomentosa clade (60±68% bootstrap support). This re- suggest relationships within core Lentago that are consistent lationship is most strongly supported in our combined analyses with those recovered by Donoghue et al. (2004). In chloroplast that excluded section Pseudotinus. In this case, the link be- analyses, V. lentago, V. prunifolium, and V. ru®dulum of east- tween V. urceolatum and the Solenotinus±Tomentosa clade ern North America are united to the exclusion of the Mexican was supported by a bootstrap value of 81%. species, V. elatum (66% bootstrap value); within the eastern Although there is now considerable support for the presence North American clade, V. lentago is more closely related V. of three main groups within Viburnum, the placement of sec- prunifolium (62% bootstrap value). In contrast, analyses of tion Pseudotinus relative to the two smaller suprasectional nuclear data place V. lentago sister (98% bootstrap support) clades remains unclear. In the two previous molecular phylo- to a clade containing V. elatum, V. prunifolium, and V. ru®- genetic studies, these relationships were unresolved, and it was dulum, within which V. prunifolium and V. ru®dulum are unit- assumed that resolution of relationships simply required more ed with 77% bootstrap support. Although not conclusive, these data. However, our separate analyses of nuclear and chloro- results are consistent with the hypothesis that V. prunifolium plast data matrices suggest con¯icting placements of Pseudo- originated following hybridization between V. ru®dulum and tinus. Speci®cally, chloroplast analyses favor a link with the V. lentago (Brumbaugh and Guard, 1956; Rader, 1976; Don- Lentago±Viburnum clade (81% bootstrap support), whereas oghue et al., 2004). When assessed with the ILD test, these the nuclear data suggest a closer relationship with the Solen- differences do not, in themselves, result in a signi®cant level otinus±Tomentosa±urceolatum clade (60% bootstrap support). of incongruence. However, a SLP test showed that the nuclear Our statistical tests are also consistent with con¯ict concerning tree is a signi®cantly better ®t to these data than the rival the placement of Pseudotinus. In particular, ILD tests indicate chloroplast topology. In the reciprocal test, there were insuf- that differences in the placement of this lineage make a sub- ®cient sites for a statistically signi®cant result; most likely re- stantial contribution to the overall level of incongruence be- ¯ecting the lack of resolution for species level relationships in tween data sets. To con®dently place this lineage within the the chloroplast data set. phylogeny will require more data. However, our analyses high- Previous molecular analyses left relationships within Pseu- light the possibility that hybridization and/or lineage sorting dotinus uncertain. Only WAXY1 sequences had provided clear events may have played a role in the early evolution of Pseu- resolution, uniting the Japanese V. furcatum and North Amer- dotinus. ican V. lantanoides (99% bootstrap support) to the exclusion of V. cordifolium from the Himalayas (Winkworth and Don- Relationships among suprasectional lineages and the root oghue, 2004). This relationship was also recovered in analyses positionÐPrevious studies have not clearly resolved relation- of our nuclear matrix (88% bootstrap support). In contrast, ships at the base of Viburnum. Unfortunately, even with our analyses of the chloroplast data strongly suggest a direct link larger combined data matrices support for basal relationships between V. cordifolium and V. furcatum (95% bootstrap sup- remains weak. Both chloroplast and combined analyses sug- port). We also note that this relationship is supported by a gest that the root falls along the branch separating V. clemen- unique insertion in WAXY2 (Winkworth and Donoghue, 2004). siae from a clade containing the remainder of Viburnum (Figs. Use of the ILD test to evaluate this con¯ict is confounded by 1 and 3). In contrast, analyses of the nuclear data set place the differences in the placement of section Pseudotinus inferred root along the edge separating the Lentago±Viburnum clade from chloroplast and nuclear partitions. However, a SLP test from everything else. Beyond this, there is uncertainty about using the nuclear matrix indicates that the chloroplast rival tree the relationships among the major clades. Speci®cally, it is is not a signi®cantly worse partitioning of these data; there unclear whether Viburnum is divided into two clades as in our were insuf®cient sites that differed in length between the to- combined analysis (Fig. 3) or whether relationships at the base pologies for the reciprocal test to have statistical signi®cance. of the tree are a grade relative to a large clade containing the Clearly ®nal resolution of relationships within both Lentago traditional Megalotinus, Odontotinus, Opulus, Oreinotinus, and Pseudotinus requires additional sampling and markers. and Tinus as in our separate analyses (e.g., Figs. 1 and 2). The great genetic distance to extant relatives and the rela- Informal clade namesÐMolecular phylogenetic analyses tively short branches close to the base of Viburnum suggests (e.g., Donoghue et al., 2004; Winkworth and Donoghue, 2004; that understanding basal relationships will be an ongoing chal- this study) have greatly improved our understanding of Vi- lenge. To this end, it may be useful to expand sampling of the burnum phylogeny. Given the general congruence between Sambucus±Adoxa clade, especially if GBSSI can be routinely these studies and the strong support for a number of major recovered from this group. However, previous results (e.g., clades, it seems appropriate to begin to reclassify Viburnum. Bell et al., 2001; Donoghue et al., 2001) suggest this may not Although we recognize the importance of providing formal 662 AMERICAN JOURNAL OF BOTANY [Vol. 92

Fig. 4. Schematic tree diagram with informal phylogenetic nomenclature for some key Viburnum lineages. Triangles represent species groups and are approximately scaled to species number. Species groups are coded: A, Lobata;C,Lantana;D,Oreinodontotinus;H,Tomentosa;L,Lentago;J,Succodontotinus; M, Megalotinus;O,Opulus;P,Pseudotinus;R,Mollodontotinus;S,Solenotinus;T,Tinus. phylogenetic de®nitions for taxonomic names, at present we the three other clades, we have used Odontotinus as the basis merely wish to provide a system of informal names to facilitate for our new names, in each case adding a pre®x that re¯ects discussion and future work. We leave the formal naming of some feature of the included . The sister group of Lobata clades until we have achieved a better understanding of spe- contains the remaining Asian members of the traditional sec- cies-level relationships. tion Odontotinus. For this clade, we coin the name Succodon- Molecular and morphological studies have provided clear totinus, a reference to the juicy produced by most of evidence for the monophyly of some, and the nonmonophyly these species. The two remaining species groups contain New of other, traditionally recognized sections. It would be useful World taxa that were placed by Hara (1983) in Odontotinus to have names that can be consistently applied to the newly subsection Dentata. For the clade containing V. ellipticum, V. identi®ed lineages. Speci®cally, we will provide names for the molle, and V. ra®nesquianum, we suggest Mollodontotinus, 12 major species groups and several of the more inclusive which refers to the soft pubescence on the in some clades (Fig. 4). In choosing names, we have attempted to min- members of this group. The ®nal group consists of V. dentatum imize potential confusion and to maintain continuity with the and all representatives of the Latin American section Orein- taxonomic literature (e.g., Hara, 1983). To this end, we have otinus. To highlight the inclusion of the large Latin American retained previous section names when these correspond to clade, we refer to this lineage as Oreinodontotinus. clades that contain all, or almost all, of the traditionally in- Our phylogenetic analyses have also identi®ed several well- cluded species. Speci®cally, we have maintained Lentago, supported higher-level clades. Two of these contain represen- Opulus, Pseudotinus, Solenotinus, and Tinus. Megalotinus and tatives of the traditional section Odontotinus. Most members Tomentosa are also retained; these groups have been repre- of Lobata and Succodontotinus produce red fruits (V. aceri- sented in our studies by single exemplars, but it seems rea- folium of Lobata and V. melanocarpum of Succodontotinus are sonable to use these names until the sampling is expanded. exceptions), and we therefore suggest the name Erythrodon- The exception to our strategy of retaining sectional names is totinus for the clade that includes these two lineages. Similarly, the name Viburnum itself. Although this group, with the ex- in reference to their purple fruits, we propose Porphyrodon- ception of V. urceolatum, does correspond to a well-supported totinus for the clade comprised of Oreinodontotinus and Mol- clade, the name Viburnum is most often applied to the genus lodontotinus. Erythrodontotinus and Porphyrodontotinus as a whole. To minimize confusion, we will refer to this group along with Opulus and Tinus form the largest of the suprasec- as the Lantana lineage, after the widely cultivated European tional clades identi®ed in our molecular analyses. We refer to species, V. lantana. this entire clade as Imbricotinus because two pairs of imbricate Each of the four remaining species groups (Fig. 4) contains bud scales generally characterize these plants. This contrasts taxa traditionally assigned to section Odontotinus, which mo- with Valvatotinus, the name we apply to the suprasectional lecular and morphological analyses have repeatedly shown to clade that contains Lentago and Lantana. These plants produce be nonmonophyletic. For the clade represented by V. aceri- a single pair of valvate bud scales in Lentago and naked folium and V. kansuense, we propose the name Lobata,in with non-overlapping involute vernation in Lantana. reference to their tri-lobed leaves (also found in Opulus). This clade appears to correspond to taxon that Hara (1983) recog- Biogeographic implicationsÐThe wide distribution of Vi- nized as subsection Lobata within section Odontotinus. For burnum makes it an excellent model for investigating Northern April 2005] WINKWORTH AND DONOGHUEÐVIBURNUM PHYLOGENY 663

Fig. 5. Biogeographic patterns within Viburnum. (A) Map illustrating six broadly de®ned regions of Viburnum diversity and the distribution of our 12 species groups across these areas. For each clade, a line connects the regions in which it is present; the number of species present in each of the regions is also given. Species groups are coded: A, Lobata;C,Lantana;D,Oreinodontotinus;H,Tomentosa,L,Lentago;J,Succodontotinus;M,Megalotinus;O,Opulus;P, Pseudotinus;R,Mollodontotinus;S,Solenotinus;T,Tinus. The partially dotted line for Pseudotinus indicates that no species are present in Europe. The formula associated with each region indicates the number of clades. Lineages represented by V. clemensiae and V. urceolatum in our analyses fall outside the 12 major clades but are included in our estimates of regional species diversity. (B) Ancestral area reconstruction of Old World±New World disjunctions in Viburnum from DIVA (Ronquist, 1996). Unshaded ϭ Old World; shaded ϭ New World; striped ϭ equivocal. Species groups denoted as for (A).

Hemisphere biogeography. In particular, it provides an oppor- personal communication), a review of the taxonomic and ¯o- tunity to investigate classic patterns of disjunction between the ristic literature, and on M. Donoghue's (unpublished data) as- Old and New Worlds (e.g., Wen, 1999, 2001; Xiang et al., sessment of species limits in several clades. With the exception 2000; Donoghue et al., 2001; Manos and Donoghue, 2001; of V. clemensiae and V. urceolatum, the remaining 156 can be SanmartõÂn et al., 2001; Xiang and Soltis, 2001; Donoghue and assigned to one of our species groups (Fig. 5a). For most Moore, 2003; Donoghue and Smith, 2004). For the purposes clades, we expect the estimated species numbers to be rela- of providing a biogeographic overview, we have estimated that tively stable, although considerable taxonomic uncertainty re- there are 158 species of Viburnum worldwide. This number is mains in Megalotinus and Oreinotinus. In general, however, based on a list of speci®c and infraspeci®c taxa kindly pro- future species-level taxonomic changes are unlikely to alter vided by V. MaleÂcot (Institut National d'Horticulture, France; our general conclusions. 664 AMERICAN JOURNAL OF BOTANY [Vol. 92

For our preliminary biogeographic analysis, we recognized In addition to disjunctions within the 12 species groups, our six broadly de®ned geographic regionsÐwestern North Amer- phylogeny indicates two Old World-New World disjunctions ica, eastern North America, and Latin America in the New between these lineages. Speci®cally, in Valvatotinus the Eur- World, and temperate eastern Asia, southeastern Asia, and Eu- asian Lantana clade is sister to Lentago of eastern North rope in the Old World (Fig. 5a). Using this scheme, temperate America and Mexico. Likewise, the Eurasian Tinus is sister to eastern Asia, with ca. 82 species, is the most species rich, the exclusively New World Porphyrodontotinus clade. followed by Latin America (with ca. 39 spp.), southeastern Our ancestral area reconstructions using DIVA (Fig. 5b), Asia (ca. 19 spp.), eastern North America (ca. 13 spp.), Europe together with recent molecular age estimates for Dipsacales (ca. 4 spp.), and western North America (ca. 3 spp.). At the (Bell and Donoghue, 2005), provide a framework for making extremes of regional species richness, there is a clear relation- preliminary inferences about the historical biogeography of ship between the total of species and clade diversity. Speci®- Viburnum. Despite uncertainty at the base of the tree, these cally, temperate eastern Asia has the greatest diversity of both analyses suggest that the Viburnum stem lineage most likely species and major clades (i.e., 82 species in nine clades), arose in Asia during the Late Cretaceous or Early Tertiary (ca. whereas western North America is the least diverse in both 60±80 mya). It also appears that the initial diversi®cation of species and clades (i.e., three species in two clades). However, the crown clade occurred within Asia. Indeed, several species this relationship is not straightforward in the remaining re- groups may never have extended beyond Asia (i.e., Megalo- gions. Although Latin America ranks second in total species tinus, Solenotinus, Succodontotinus, and Tomentosa). How- richness, these species represent just two clades. Indeed, all ever, our analyses indicate transitions between the Old World but one of the 39 species in Latin America belong to Orei- and New World in several lineages. In four cases, DIVA anal- nodontotinus; the exception being the Mexican V. elatum of yses unequivocally reconstruct the movement as being from Lentago. In stark contrast, each of the four European species the Old World to the New World (Fig. 5b). For the Tinus± represents a different and distantly related clade: Lantana, Porphyrodontotinus and the Lantana±Lentago splits, arrival in Lobata, Opulus, and Tinus. Species from these two regions the New World appears to have occurred relatively early. Ab- represent the extremes of dispersion across the phylogeny solute dates are unclear, because we have yet to con®dently (Webb et al., 2002). That is, the distribution of the Latin Amer- date the crown clade, but relatively early splits are suggested ican species is highly localized in the phylogeny (i.e., their by substantial morphological and molecular disparity within closest relatives are found within the same region), whereas and between the subclades, as well as by interregional dis- the four European species are hyper-dispersed (i.e., they are junctions within the subclades (i.e., between Asia and Europe distantly related to one another, and each is more closely re- in Lantana and Tinus, and between eastern North America and lated to a species outside of the region). The two remaining Latin America in Lentago and Oreinodontotinus). Uncertainty regions fall between these two endpoints. Six clades occur in about the placement of V. nudum leaves open the possibility Eastern North America. Three of these are represented by a of an additional Old World-New World disjunction within Val- single species (Lobata, Oreinodontotinus, and Pseudotinus), vatotinus. In contrast to disjunctions between these major spe- two by a pair of species (Mollodontotinus and Opulus), and cies groups, the disjunctions within Lobata and Pseudotinus one by six species (Lentago). Similarly, the 19 species found appear to have arisen much more recently. This is suggested in southeastern Asia belong to three clades; one represented by limited morphological and molecular differentiation be- by a single species (Succodontotinus), one by four species (So- tween the Old and New World species, but it is also consistent lenotinus), and the third by 14 species (Megalotinus). with the low species diversity in each of these clades. Specif- Our phylogenetic analyses also highlight patterns of dis- ically, in Pseudotinus the eastern North American species, V. junction within Viburnum. With the exception of the Tomen- lantanoides, is morphologically very similar to the three Asian tosa clade, which contains two species con®ned to temperate species, in particular to V. furcatum of Japan and northeastern eastern Asia, each of the other 11 species groups occupies China. Similarly, the three species of Lobata are similar in more than one region (Fig. 5a). Most commonly, these line- appearance, especially the eastern North American V. aceri- ages are present in neighboring regions. Three are shared be- folium and the European V. orientale. tween the two Asian regions (Megalotinus, Solenotinus, and Our analyses imply at least one additional Old World±New Succodontotinus), two between eastern North America and World disjunction within the circumboreal Opulus clade. How- Latin America (Lentago and Oreinodontotinus), and one be- ever, both the exact number and the directionality of events in tween eastern and western North America (Mollodontotinus). this clade remain uncertain because V. opulus and V. koreanum The three remaining two-region clades are more widely dis- are missing from our analyses. Gene trees are consistent with junct. Most species of Tinus and Lantana occur in temperate either two independent shifts from the Old World to the New eastern Asia, but each of these clades includes a single Eu- World or, alternatively, an early transition to the New World ropean species. Pseudotinus is disjunct between the Old and followed by re-entry into the Old World. Despite this uncer- New Worlds, with three species in temperate eastern Asia and tainty, the morphological similarity and high latitude distri- one in eastern North America. The two remaining species bution of the ®ve recognized species suggests that the circum- groups, Lobata and Opulus, are found in three and four re- boreal distribution of the Opulus lineage arose quite recently. gions, respectively. Single representatives of Lobata occur in Phylogenetic analyses of Viburnum are consistent with sim- temperate eastern Asia, Europe, and eastern North America ilar studies on a broad range of taxa and with our current (based on preliminary data, we include V. orientale of the understanding of the historical biogeography of northern tem- Caucusus mountains in this clade; R. Winkworth and M. Don- perate forests. This study highlights the ®nding that patterns oghue, unpublished data). In the circumboreal Opulus clade, of disjunction are reiterated over time (Donoghue and Moore, ®ve species are shared between four areas; a single species in 2003). Consistent with other molecular studies (e.g., Dono- Europe, two in temperate eastern Asia, and two that occur in ghue et al., 2001), multiple, temporally incongruent disjunc- both eastern and western North America. tions appear to have arisen within Viburnum during the Ter- April 2005] WINKWORTH AND DONOGHUEÐVIBURNUM PHYLOGENY 665 tiary; often these involved movements from the Old World to DE QUEIROZ, A., M. J. DONOGHUE, AND J. KIM. 1995. Separate versus com- the New World, presumably via Beringia (Donoghue and bined analysis of phylogenetic evidence. Annual Review of Ecology and Moore, 2003; Donoghue and Smith, 2004). That these events Systematics 26: 657±681. DONOGHUE, M. J. 1982. Systematic studies in the genus Viburnum. Ph.D. appear to have occurred at different times suggests that the dissertation, Harvard University, Cambridge, Massachusetts, USA. contemporary community has been assembled over an extend- DONOGHUE, M. J. 1983a. A preliminary analysis of phylogenetic relation- ed period and that as a result, interactions between lineages ships in Viburnum (Caprifoliaceae s.l.). Systematic Botany 8: 45±58. will have existed for differing amounts of time. In the case of DONOGHUE, M. J. 1983b. The phylogenetic relationships of Viburnum. In N. Viburnum, for example, it would appear that Lentago and Por- Platnick and V. Funk, [eds.], Advances in cladistics, vol. 2, 143±166. phyrodontotinus were present in the New World before the Columbia University Press, New York, New York, USA. entry of the Lobata, Opulus,orPseudotinus lineages. A more DONOGHUE,M.J.,AND B. G. BALDWIN. 1993. Phylogenetic analysis of Vi- burnum based on ribosomal DNA sequences from the internal transcribed detailed account of this dynamic history must await more pre- spacer regions. American Journal of Botany 80 (Supplement): 146 (Ab- cise estimates for the absolute timing of events and more de- stract). tailed phylogenetic work within the major clades. DONOGHUE, M. J., B. G. BALDWIN,J.LI, AND R. C. WINKWORTH. 2004. Viburnum phylogeny based on chloroplast trnK intron and nuclear ri- Future directionsÐMolecular systematic studies have pro- bosomal ITS DNA sequences. Systematic Botany 29: 188±198. vided a clear and consistent picture of Viburnum phylogeny, DONOGHUE, M. J., C. D. BELL, AND J. LI. 2001. Phylogenetic patterns in 12 well-supported species groups have been identi®ed and re- northern hemisphere plant geography. International Journal of Plant Sci- lationships among most of these lineages have been resolved ences 162 (Supplement): S41±S52. DONOGHUE, M. J., C. D. BELL, AND R. C. WINKWORTH. 2003. The evolution with con®dence. However, despite this progress, relationships of reproductive characters in Dipsacales. International Journal of Plant at the base of the phylogeny remain uncertain. Speci®cally, Sciences 164 (Supplement): S453±S464. there is little support for relationships among the three major DONOGHUE, M. J., T. ERIKSSON,P.A.REEVES, AND R. G. OLMSTEAD. 2001. lineages, Pseudotinus, V. urceolatum, and V. clemensiae,or Phylogeny and phylogenetic taxonomy of Dipsacales, with special ref- for the exact position of the root. The large genetic distance erence to Sinadoxa and Tetradoxa (Adoxaceae). Harvard Papers in Bot- separating Viburnum from its closest living relatives and the any 6: 459±479. structure of the phylogeny itself (i.e., short branches close to DONOGHUE,M.J.,AND B. R. MOORE. 2003. Toward an integrative historical biogeography. Integrative and Comparative Biology 43: 261±270. the base of the tree) suggest that the resolution of this problem DONOGHUE,M.J.,AND S. A. SMITH. 2004. Patterns in the assembly of tem- will require much more data. Although such uncertainty limits perate forests around the Northern Hemisphere. Philosophical Transac- our understanding of some broad scale evolutionary patterns tions: Biological Sciences 359: 1633±1644. within Viburnum, the current results provide an excellent DONOGHUE,M.J.,AND K. SYTSMA. 1993. Phylogenetic analysis of Viburnum framework for further research on character evolution, rates of based on chloroplast DNA restriction-site data. American Journal of Bot- diversi®cation, and historical biogeography. any 80 (Supplement): 146 (Abstract). Future studies will need to expand taxon sampling, espe- HARA, H. 1983. A revision of the Caprifoliaceae of Japan with reference to allied plants in other districts and the Adoxaceae. Academia Scienti®c cially for clades represented by one or a few species in the Books Inc., Tokyo, Japan. current analyses. The sample used in our recent analyses has HSU, P.-S. 1975. Notes on genus Viburnum of China. Acta Phytotaxonomica yielded some quite unexpected results. Most importantly, the Sinica 13: 111±128. inclusion of V. clemensiae from Borneo revealed the nonmon- HSU, P.-S., J.-Q. HU, H.-J. WANG. 1988. Caprifoliaceae. In P.-S. Hsu [ed.], ophyly of section Solenotinus and suggested a novel position Flora reipublicae popularis sinicae, vol. 72. Science Press, Beijing, Chi- for the root of Viburnum. We expect, however, that adding na. more species will not greatly change the overall picture. In JOHNSON,L.A.,AND D. E. SOLTIS. 1998. Assessing congruence: empirical examples from molecular data. In D. E. Soltis, P. S. Soltis, and J. J. most cases, we anticipate that additional species will fall with- Doyle [eds.], Molecular systematics of plants II, DNA sequencing, 296± in the clades that we have identi®ed, although species assigned 343. Kluwer Academic Press, Dordrecht, Netherlands. to the morphologically diverse Megalotinus could provide JONES, T. H. 1983. A revision of the genus Viburnum section Lentago (Ca- some important surprises. Another important goal in the future prifoliaceae). Ph.D. dissertation, North Carolina State University, Ra- will be to obtain more precise estimates of divergence times leigh, North Carolina, USA. within Viburnum through the careful study and the inclusion KERN, J. H. 1951. The genus Viburnum (Caprifoliaceae) in Malaysia. Rein- fossils. wardtia 1: 107±170. KERN,J.H.,AND C. G. G. J. VAN STEENIS. 1951. In C. S. Sargent [ed.], Trees and shrubs, vol. II, part II, 105±116. Houghton Mif¯in, Boston, LITERATURE CITED Massachusetts, USA. ILLIP AND MITH BALDWIN, B. G., M. J. SANDERSON,J.M.PORTER,M.F.WOJCIECHOWSKI, K ,E.P., A. C. S . 1929. The genus Viburnum in northwestern C. S. CAMPBELL, AND M. J. DONOGHUE. 1995. The ITS region of nu- South America. Bulletin of the Torrey Botanical Club 56: 265±274. clear ribosomal DNA: a valuable source of evidence on angiosperm phy- KILLIP,E.P.,AND A. C. SMITH. 1931. The South American species of Vi- logeny. Annals of the Missouri Botanical Garden 82: 247±277. burnum. Bulletin of the Torrey Botanical Club 57: 245±258. BRUMBAUGH,J.H.,AND A. T. GUARD. 1956. A study of evidences for in- LINNAEUS, C. 1753. Species plantarum. Laurentius Salvius, Stockholm, Swe- trogression among Viburnum lentago, V. prunifolium, and V. ru®dulum den. based on leaf characteristics. Proceedings of the Indiana Academy of MANOS,P.S.,AND M. J. DONOGHUE. 2001. Progress in northern hemisphere Sciences 66: 300. phytogeography: an introduction. International Journal of Plant Sciences BELL, C. D. 2001. PORN*: a hierarchical likelihood ratio calculator for LI- 162 (Supplement): S1±S2. NUX. [Computer program] Distributed by the author, available at web- MASON-GAMER,R.J.,AND E. A. KELLOGG. 1996. Testing for phylogenetic site, http://www.phylodiversity.net/pornstar. con¯ict among molecular data sets in the tribe Triticeae (Gramineae). BELL,C.D.,AND M. J. DONOGHUE. 2005. Dating the diversi®cation of Dip- Systematic Biology 45: 524±545. sacales: comparing models, genes, and evolutionary implications. Amer- MCATEE, W. L. 1956. A review of nearctic Viburnum. Privately printed, ican Journal of Botany 92: 284±314. Chapel Hill, North Carolina, USA. BELL, C. D., E. J. EDWARDS, S.-T. KIM, AND M. J. DONOGHUE. 2001. Dip- MORTON, C. V. 1933. The Mexican and Central American species of Vibur- sacales phylogeny based on chloroplast DNA sequences. Harvard Papers num. Contributions of the U.S. National Herbarium 26: 339±366. in Botany 6: 481±499. OERSTED, A. S. 1861. Til belysning af slaegten Viburnum. Videnskabelige 666 AMERICAN JOURNAL OF BOTANY [Vol. 92

Meddelelser fra Dansk Naturhistorisk Forening I Kjobenhavn 13: 267± (*and other methods), version 4. Sinauer Associates, Sunderland, Mas- 305. sachusetts, USA. PYCK, N., P. ROELS, AND E. SMETS. 1999. Tribal relationships in Caprifoli- TEMPLETON, A. R. 1983. Phylogenetic inference from restriction endonucle- aceae: evidence from a cladistic analysis using ndhF sequences. System- ase cleavage site maps with particular reference to the evolution of hu- atics and Geography of Plants 69: 145±159. mans and the apes. Evolution 37: 221±244. RADER, L. L. 1976. A biosystematic study of and V. THOMPSON, J. D., D. G. HIGGINS, AND T. J. GIBSON. 1994. CLUSTAL W: ru®dulum (Caprifoliaceae). M.S. thesis, University of Tennessee, Knox- improving the sensitivity of progressive sequence alignment through se- ville, Tennessee, USA. quence weighting, position speci®c gap penalties and weight matrix REHDER, A. 1908. The of eastern Asia. In C. S. Sargent [ed.], choice. Nucleic Acids Research 22: 4673±4680. Trees and shrubs, vol. II, part II, 105±116. Houghton Mif¯in, Boston, WEBB, C. O., D. D. ACKERLY,M.MCPEEK, AND M. J. DONOGHUE. 2002. Massachusetts, USA. Phylogenies and community ecology. Annual Review of Ecology and REHDER, A. 1940. Manual of cultivated trees and shrubs. Macmillian, New Systematics 33: 475±505. York, New York, USA. WEN, J. 1999. Evolution of eastern Asian and eastern North American dis- RONQUIST, F. 1996. DIVA, version 1.1. [Computer program and manual] junct distributions in ¯owering plants. Annual Review of Ecology and available by anonymous FTP from Uppsala University at website, Systematics 30: 421±455. http://www.ebc.uu.se/systzoo/research/diva/diva.html. WEN, J. 2001. Evolution of eastern Asian±eastern North American biogeo- RONQUIST, F. 1997. Dispersal-vicariance analysis: a new approach to the graphic disjunctions: a few additional issues. International Journal of quanti®cation of historical biogeography. Systematic Biology 46: 195± Plant Sciences 162 (Supplement): S117±S122. 203. WILKINSON, A. M. 1948. Floral anatomy and morphology of some species SANG, T., D. J. CRAWFORD, AND T. F. S TUESSY. 1997. Chloroplast DNA of the genus Viburnum of the Caprifoliaceae. American Journal of Bot- phylogeny, reticulate evolution, and biogeography of Paeonia (Paeoni- any 35: 455±465. aceae). American Journal of Botany 84: 1120±1136. WINKWORTH,R.C.,AND M. J. DONOGHUE. 2004. Viburnum phylogeny: ev- SANMARTIÂN, I., H. ENGHOF, AND F. R ONQUIST. 2001. Patterns of animal dis- idence from the duplicated nuclear gene GBSSI. Molecular Phylogenet- persal, vicariance and diversi®cation in the Holarctic. Biological Journal ics and Evolution 33: 109±126. of the Linnaen Society 73: 345±390. XIANG, Q.-Y., AND D. E. SOLTIS. 2001. Dispersal-vicariance analyses of in- SEELANAN, T., A. SCHNABEL, AND J. F. WENDEL. 1997. Congruence and tercontinental disjuncts: historical biogeographical implications for an- consensus in the cotton tribe (Malvaceae). Systematic Botany 22: 259± giosperms in the Northern Hemisphere. International Journal of Plant 290. Sciences 162 (Supplement): S29±S39. SIMMONS, M. P., C. D. BAILEY, AND K. C. NIXON. 2000. Phylogeny recon- XIANG, Q.-Y., D. E. SOLTIS,P.S.SOLTIS,S.R.MANCHESTER, AND D. J. struction using duplicate genes. Molecular Biology and Evolution 17: CRAWFORD. 2000. Timing the eastern Asian±eastern North American 469±473. ¯oristic disjunction: molecular clock corroborates palentological esti- SWOFFORD, D. L. 2002. PAUP*: phylogenetic analysis using parsimony mates. Molecular Phylogenetics and Evolution 15: 462±472.