PHYLOGENY BASED ON CHLOROPLAST DNA SEQUENCES

CHARLES D. BELL,1, 2 ERIKA J. EDWARDS,1 SANG-TAE KIM,1 AND MICHAEL J. DONOGHUE 1

Abstract. Eight new rbcL DNA sequences and 15 new sequences from the 5' end of the chloroplast ndhF gene were obtained from representative Dipsacales and outgroup taxa. These were analyzed in combination with pre- viously published sequences for both regions. In addition, sequence data from the entire ndhF gene, the trnL-F intergenic spacer region,the trnL intron,the matK region, and the rbcL-atpB intergenic spacer region were col- lected for 30 taxa within Dipsacales. Phylogenetic relationships were inferred using maximum parsimony and maximum likelihood methods. Inferred tree topologies are in strong agreement with previous results from sep- arate and combined analyses of rbcL and morpholo gy, and confidence in most major clades is now very high. Concerning controversial issues, we conclude that Dipsacales in the traditional sense is a monophyletic group and that is more closely related to than it is to . is only weakly supported as the sister group of the Caprifolieae (within which relationships remain largely unresolved), and the exact position of Diervilleae is uncertain. Within Morinaceae, is the sister group of Morina plus Cryptothladia. Dipsacales now provides excellent opportunities for comparative studies, but it will be important to check the congruence of chloroplast results with those based on data from other genomes. Keywords: Dipsacales, , , Morinaceae, Dipsacaceae, Valerianaceae, phylogeny, chloroplast DNA.

The Dipsacales has traditionally included the ). It excludes and its relatives, as C ap ri foliaceae (s e n s u l at o, i . e. , i n cl u d i n g well as Dipsacaceae, M o ri n a c e a e, a n d and ), Adoxaceae, Dip- Valerianaceae. We reject this traditional cir- sacaceae, and Valerianaceae (e.g., Cronquist, cumscription of Caprifoliaceae on the grounds 1 9 8 8 ) , and sometimes segregates such as of non-monophyly. According to all recent phy- Morinaceae and Triplostegiaceae. In contrast, logenetic studies (e.g., Donoghue, 1983, 1985; the classification of the Angiosperm Phylogeny D o n oghue et al., 1992; Ju dd et al., 1 9 9 4 ; Group (APG, 1998) included the bulk of the Backlund and Donoghue, 1996; Backlund and C ap ri foliaceae in Dipsacales, along with Bremer, 1997; Kim et al., 1999; Olmstead et Dipsacaceae, Valerianaceae, and Morinaceae, al., 2000; Donoghue et al., 2001), Sambucus but exc luded an expanded Ad o xaceae (inclu d i n g and Vibu r nu m a re more cl o s e ly re l ated to Viburnum, Sambucus, and Adoxa), which they A d ox a and its re l at ives than they are to left unassigned to any order within . C ap ri fo l i e a e, D i e rv i l l e a e, and Linnaeeae, The molecular phylogenetic studies reported wh i ch are instead more cl o s e ly re l ated to he r e concern Dipsacales in the traditional sense. Morinaceae, Dipsacaceae, and Valerianaceae. The circu m s c r iption of the Capri f oliaceae has A more restricted usage of Caprifoliaceae also become confusing, and to avoid misunder- ex cludes S a m bu c u s and Vi bu rnu m, on the standings we need to clarify our usage of the grounds that these are probably more closely name. Caprifoliaceae in the traditional sense related to Adoxa and its relatives, but retains i n cludes S a m bu c u s, Vibu r nu m, C ap ri fo l i e a e C ap ri fo l i e a e, D i e rv i l l e a e, and Linnaeeae. (, Lonicera, , and Again, we reject the use of Caprifoliaceae in Tri o s t e u m, and possibly H ep t a c o d i u m) , this restricted sense because it is not mono- D i e rvilleae (D i e rv i l l a and We i ge l a) , a n d phyletic. Specifically, all recent analyses (refer- Linnaeeae (Abelia, Dipelta, Kolkwitzia, and ences above) indicate that the Linnaeeae is

We are grateful to Torsten Eriksson, Pat Reeves, and Dick Olmstead for generating a number of the sequences used in our analyses and for allowing us to refer to their manuscript with MJD. We also acknowledge Kris Peterson for her help with sequencing. This study has most recently been supported by an NSF Biotic Surveys and Inventories grant to collect in the Hengduan Mountain region of China (DEB-9705795). 1 Department of Ecology and Evolutionary Biology, Yale University, New Haven, Connecticut,06511-8106, U.S.A. 2 Author for correspondence. E-mail: [email protected].

Harvard Papers in , Vol. 6,No. 2, 2001, pp. 481–499. © President and Fellows of Harvard College, 2001. 482 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 m o re cl o s l ey re l ated to Mori n a c e a e, D i p- Adoxa clade, and the remainder of the tradi- s a c a c e a e, and Va l e ri a n a c e a e, than to the tional Cap ri foliaceae are parap hy l e t i c. Caprifolieae or Diervilleae. Specifically, the Linnaeeae appears to be more In view of the phylogenetic results, and wish- closely related to Morinaceae, Valerianaceae, ing to retain monophyly, Backlund and Bremer and Dipsacaceae than it is to either Caprifolieae (1998) suggested another strat egy, wh i ch or Dierv i l l e a e. Howeve r, s u p p o rt for these Backlund and Pyck (1998) followed with the major clades varies considerably among analy- naming of two new families. They restricted the ses, and the exact placement of several key taxa name Cap ri foliaceae to the fo rmer Cap ri- ( e. g. , H ep t a c o d i u m, Tri p l o s t egi a) re m a i n s fo l i e a e, and then elevated Diervilleae to unresolved. Moreover, there is disagreement Diervillaceae and Linnaeeae to Linnaeaceae. concerning the placement of several other taxa, This usage was adopted in the Angiosperm es p e c i a l l y Li n n a e a and the Diervi l l e a e , as well as Phylogeny Group ordinal classification (APG, relationships within the Caprifolieae. Finally, 1998). We reject this strategy on the grounds s eve ral lineages have re c e ived insuffi c i e n t that it causes instability and confusion to sub- attention. For instance, we still know little stitute the names Caprifoliaceae, Diervillaceae, about relationships among Morina, Acantho - and Linnaeaceae for the already well estab- calyx, and Cryptothladia of the Morinaceae. lished names Cap ri fo l i e a e, D i e rv i l l e a e, a n d Here we present broad phylogenetic analyses Linnaeeae solely for the sake of adjusting taxo- including 8 new rbcL sequences and 15 new nomic ranks. These names refer to the same DNA sequences from the 5' end of the ndhF entities, and rank assignments are arbitrary gene. For more detailed studies within Dip- (APG, 1998). sacales we have added sequences of several Our usage of both the names Adoxaceae and other ch l o roplast coding and non-coding Caprifoliaceae follows the initial suggestion of regions: matK, the trnL intron, and the inter- Judd et al. (1994) and the node-based phyloge- genic spacer (IGS) regions of trnL-F and rbcL- netic definitions of Donoghue et al. (2001). The atpB. Our hope was that separate and combined mapping of these names onto a phylogenetic analyses of this much-expanded chloroplast hypothesis is shown in Fig. 1, adapted from d ataset would further cl a rify Dipsacales Donoghue et al. (2001). Adoxaceae here refers phylogeny. to the clade that includes Viburnum, Sambucus, and Adoxa and its relatives. Caprifoliaceae MATERIALS AND METHODS includes Caprifolieae, Diervilleae, Linnaeeae, Samples Morinaceae, Dipsacaceae, and Valerianaceae. We carried out analyses on two different The unusual and most obviously beneficial fea- datasets. Our 46-taxon dataset included rbcL ture of this nomenclature is that all of the tradi- and ndhF sequences from 32 representatives of tional names are retained (so long as these refer major lineages within Dipsacales and 14 asterid to clades). The potentially confusing aspect of outgroup taxa. Our 30-taxon dataset consisted this treatment is that, in addition to including of Dipsacales taxa alone scored for all six Caprifolieae, Diervilleae, and Linnaeeae, the chloroplast gene regions. Information on the C ap ri foliaceae also includes Mori n a c e a e, source of materials and on GenBank Dipsacaceae, and Valerianaceae. We believe accession num b e r s is provided in Tables 1 and 2. that the benefits of this solution far exceed any potential confusion. Donoghue et al. (2001) DNA Extraction and PCR Protocols provide additional discussion of this approach. Total DNAs were extracted using the CTAB Previous analyses of Dipsacales phylogeny method of Doyle and Doyle (1987). The DNA have been based on morphological characters, extracts were further purified with the Prep-A- molecular data, and a combination of the two Gene DNA Purification Kit (Bio-Rad). Double- (e.g., Donoghue, 1983; Donoghue et al., 1992; stranded copies of all regions were amplified Judd et al., 1994; Backlund and Donoghue, using standard Po ly m e rase Chain Reaction 1996; Backlund and Bremer, 1997; Pyck et al., (PCR) in 25- to 50-µL reactions. All reactions 1999; Pyck and Smets, 2000; Donoghue et al., were heated at 94 C for 3 min. The reactions 2001). As mentioned above, these studies have entailed 35 cycles consisting of 94 C for 1.5 cl e a rly established that traditional Cap ri- min, 48–56 C for 2 min, and 72 C for 3 min. foliaceae do not form a monophyletic group. Dye terminator cycle sequencing followed the Viburnum is closely related to a Sambucus- protocol specified by the ABI PRISM Dye 2001 BELL ET AL., DIPSACALES PHYLOGENY 483

FIG U R E 1. Application of names according to the phylogenetic of Donoghue et al. (2001).

Primer Cycle Sequencing Ready Reaction Kit nal primers (see Young et al., 1999). The com- (Revision B, August 1995, Perkin-Elmer) and plete coding region was sequenced from 29 was visualized using an ABI 377 automated taxa. From Valerianellalocosta we were able to DNA sequencer. sequence only the first 507 base pairs. This par- rbcL. rbcL was amplified and sequenced tial sequence was included in our phylogenetic using standard primers (see, e.g., Olmstead et analyses, with undetermined nucleotide posi- al., 1992, 1993). tions scored as missing data (“?”) in PAUP* ndhF. A portion of the 5' end of the chloro- (Swofford, 2001). plast gene ndhF was amplified with primers trnL-F IGS and trnL intron. The trnL-F IGS ndhF-1 and ndhF-1318R (Olmstead and region and trnL intron were amplified and Sweere, 1994). Specifically, we sequenced the sequenced using the universal primers trnl-c, first 1320 base pairs of the coding region, trnl-d, trnl-e, and trnl-f of Taberlet et al. (1991). which corresponds to the region studied by rbcL-atpB IGS. The rbcL-atpB IGS region P y ck et al. (1999). A m p l i fi c ation pri m e rs , was amplified with the primers rbcL-atpBR along with ndhF-274, ndhF-274R, ndhF-803, and atpB-F of Manen et al. (1994). Sequencing and ndhF-803R (Olmstead and Sweere, 1994) was accomplished using the amplifi c at i o n were then used to sequence each corresponding prim e r s along with two additional prim e r s we region. The entire ndhF molecule was also designed specific a l l y for Dipsacales (atp B - s q K 1 : sequenced for these and additional taxa (see 5 ' - C ATAT M N TAT G G C G C A A ACC-3'; at p B - Tables 1 and 2). sq K 1 R : 5' - G G T T G C G C C A TAK A TATG - 3 ' ) . matK. The matK region, as well as a portion of the flanking region at the 3' end, was ampli- Sequence Alignment fied and sequenced with the primers used by Contiguous sequences were assembled using Young et al. (1999). To obtain the entire coding Sequencher 2.1 (Gene Codes Corp., Madison, region, a variety of primer pairs were employed Wis.). MacClade (Maddison and Maddison, to amplify small fragments of the molecule, 1992) was used to translate DNA sequences which were sequenced using appropriate inter- into protein sequences to aid in the alignment 484 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 of r b cL , n d hF, and m atK sequences. A l l datasets. Heuristic search methods were used sequences were readily aligned by eye, with with TBR branch swapping and collapse of few ambiguities. The resulting data matrices zero-length branches. Analyses were repeated are available in TreeBASE (www.herbaria.har- 100 times with the “random addition” option. vard.edu/treebase) or upon request from the Bootstrap tests were performed using 1000 first author. replicates with nearest neighbor interchange (NNI) branch swapping. Parameters for boot- Datasets strap tests were fixed to values estimated from The sequence data wer e partitioned into var i- the maximum likelihood tree. ous datasets differ ing in the number of taxa and cha ra c t e r s. One set of analyses focused on a RESULTS b roader dat a s e t , i n cluding rep re s e n t at ive 46-Taxon rbcL dataset Dipsacales and potentially rel a ted groups of This data matrix consisted of 1428 aligned As t e ri d a e . This consisted of 46 taxa with data base pairs, of which 434 were variable and 250 fr om rb c L and the fir st 1320 base pairs of the parsimony informative. Maximum parsimony nd h F gen e . A second set of sepa r ate and com- analyses resulted in 102 trees of 1040 steps bined analyses focused on an expanded sample (CI= 0.538, 0.425 excluding invariant charac- of Dipsacales and consisted of 30 taxa scored for ters; RI= 0.619). The strict consensus of these rb c L, the entire nd h F gen e , tr nL- I G S , tr nL trees is presented in Fig. 2, which marks those in t ro n , ma tK, and atp B- I G S . On the basis of the branches with bootstrap support of 50% or br oader analys e s , and on previous molecular and h i g h e r. Our maximum likelihood search mo rp h o l o gical studies, the resulting Dipsacales resulted in a single tree with a -lnL value of tr ees wer e rooted along the bran c h connecting 7846.4885 (1041 steps under parsimony). Ad o xaceae to Capri f oliaceae (see below). The traditional Dipsacales for m a clade in all tr ees; however , bo o t s t r ap support for this con- Phylogenetic Analyses clusion is rather low (60%). Even less certa i n All analyses were conducted using PAUP* ar e conclusions about possible close rel at i ves of ( S wo ffo rd, 2001). Maximum pars i m o ny Dipsacales. Within Dipsacales there is stron g searches were conducted using heuristic search su p p o r t for a basal split between Ad o xaceae and methods with tree bisection re c o n n e c t i o n eve rything else. A d oxaceae here incl u d e s (TBR) bra n ch swap p i n g, c o l l apse of ze ro - Vibu r num , Sa m bu c u s , and a stron g l y supporte d length branches, and equal weighting of all clade consisting of Si n a d ox a , Tet ra d ox a , an d characters. The analyses were repeated 100 Ad ox a . Tet ra d ox a and Ad ox a ar e more clo s l e y times with the “random addition” option to rel a ted to one another than either is to Si n a d ox a . minimize problems of multiple islands of most The remaining Dipsacales constitute the parsimonious trees. Sets of equally most parsi- Caprifoliaceae sensu Judd et al. (1994) and monious trees were summarized by a strict con- Donoghue et al. (2001), including traditional sensus tree. To assess confidence in clades, Caprifolieae, Diervilleae, Linnaeeae, Morin- bootstrap tests (Felsenstein, 1985) were per- aceae, Valerianaceae, and Dipsacaceae. Within fo rmed using 300 rep l i c ates with heuri s t i c this well-supported clade, basal relationships search settings identical to those of the original are poorly resolved, with only weak support for search. the idea that Caprifolieae plus Heptacodium A series of likelihood ratio tests was per- for m an early bran c h (Pyck and Smets, 2000). As formed (on a variety of tree topologies) to sh o wn in Fig . 2, rel a tionships within Capri fo l i e a e determine which model of sequence evolution (i.e., among Leycesteria, Lonicera, Symphori - best fit the data using the program PORN* carpos, and ) are highly uncertain. (Bell, 2001). A variety of “best fitting” models Within the remainder of the Caprifoliaceae were found, depending on the taxa and data there are a few well supported clades, including p a rtition being examined (see Table 3). the traditional Morinaceae, Valerianaceae, and Maximum likelihood searches were carried out Dipsacaceae, but all other relationships are in PAUP* using the ap p ro p ri ate model. uncertain in this dataset. As noted previously Parameters for each search were simultane- by Donoghue et al. (2001), a particular anom- ously estimated via maximum likelihood for all a ly in r b cL analyses is the sep a ration of 2001 BELL ET AL., DIPSACALES PHYLOGENY 485

FIG U R E 2. Strict consensus trees from analyses of the 46-taxon rbcL and ndhF datasets (see text). Numbers above the branches are bootstrap values greater than 50%. 486 HARVARD PAPERS IN BOTANY Vol. 6, No. 2

Linnaea from the other Linnaeeae. In addition, Valerianaceae, and Dipsacaceae. Within this the remaining Linnaeeae appear in rbcL trees group we see some support for uniting to be paraphyletic, with Abelia as the sister Triplostegia with Dipsacaceae, rather than with group of the Diervilleae. However, bootstrap Va l e rianaceae as in prior analyses (e. g. , support for these groupings is uniformly low Backlund and Donoghue, 1996; Backlund and (usually <50%). Bremer, 1997). Our maximum likelihood result also supports Our ndhF maximum likelihood tree is in the monophyly of Dipsacales, which is linked strong agreement with those recovered using to a clade consisting of C o l u m e l l i a a n d p a rs i m o ny, ex c ept for the placement of Desfontainia. However, support for these rela- Diervilleae. In parsimony trees (Fig. 2) the tionships is weak (bootstrap value <50%). Diervilleae is united with Caprifolieae plus Bootstrap support for clades within Dipsacales Heptacodium, whereas in the maximum likeli- is ge n e ra l ly low, with the ex c eption of hood tree it is the sister group of all other Caprifolieae (93%), Morinaceae (100%), and Caprifoliaceae. In general, however, our ndhF Dipsacaceae (100%). results match previously published trees based The maximum likelihood and pars i m o ny on ndhF sequnces (Pyck et al., 1999; Pyck and t rees are consistent with one another. Smets, 2000) and on the combination of rbcL Furthermore, there is general agreement with and morp h o l ogy (Backlund and Donog h u e, previous analyses based on rbcL sequences for 1996; Donoghue et al., 2001). Dipsacales (Donoghue et al., 1992; Backlund and Donoghue, 1996; Backlund and Bremer, 46-Taxon Combined dataset 1997) and for Asteridae (e.g., Olmstead et al., Our combined rbcL and ndhF data matrix 1992), as well as on morphological characters consisted of 2667 aligned base pairs, of which (Donoghue, 1983; Judd et al., 1994; Backlund 940 were variable and 556 parsimony informa- and Donoghue, 1996). tive. Maximum parsimony analyses resulted in 48 most parsimonious trees of 2382 steps (CI = 46-Taxon ndhF dataset 0.5460, 0.4432 excluding invariant characters; This data matrix consisted of 1239 aligned RI = 0.6612). Our maximum likelihood search base pairs, of which 506 were variable and 306 resulted in a single tree with a -lnL value of parsimony informative. Maximum parsimony 16846.62023 (2383 steps under parsimony). analyses resulted in six most parsimonious This tree is presented in Fig. 3, which shows trees of 1313 steps (CI = 0.565, 0.469 exclud- branches with bootstrap values of 80% or more. ing invariant characters; RI = 0.705). The strict Our pars i m o ny results for the combined consensus is shown in Fi g. 2, again with dataset are similar to those found using ndhF branches marked that are supported at 50% or alone. We see improved support for Dipsacales more. Our maximum likelihood search resulted as a clade (70%),but again weak support (54%) in a single topology with a -lnL value of for a link between Dipsacales and . 8700.22647 (1318 steps under parsimony). Adoxaceae splits from Caprifoliaceae at the As with rbcL, maximum parsimony searches base of the Dipsacales. Within A d ox a c e a e, recover a monophyletic Dipsacales, which in most relationships are now supported at near this case is weakly linked with an Apiales 100%, whereas relationships at the base of the cl a d e. A ga i n , Vi bu rnu m and S a m bu c u s a re Caprifoliaceae and within the major clades united with Adoxa and its relatives in a clade (e.g., Caprifolieae, Linnaeeae), continue to be that is sister to the Caprifoliaceae. Within the poorly supported. Interestingly, Heptacodium Caprifoliaceae, members of the Caprifolieae and the Caprifolieae together form a basal are strongly united, but support for relation- clade, and Diervilleae is united with the rest of ships within this clade is mostly poor. The link the Caprifoliaceae with a bootstrap value of b e t ween Cap ri folieae and H ep t a c o d i u m i s 53%. In agreement with the ndhF sequences weak, as is the connection to Diervilleae. alone, Triplostegia is united with Dipsacaceae In contrast to rbcL analyses, the ndhF data rather than with Valerianaceae. strongly support the monophyly of Linnaeeae Results from the maximum likelihood anayl s e s in the traditional sense, with Linnaea firmly (Fi g . 3) agree with those from parsi m o n y. Per h ap s positioned at the base of this clade. Sister to the most import a n t ly, these results even more Linnaeeae is a clade consisting of Morinaceae, st ro n g l y unite the Dipsacales (88% bootstrap 2001 BELL ET AL., DIPSACALES PHYLOGENY 487

FIG U R E 3. Maximum likelihood tree from the 46-taxon combined analysis of rbcL and ndhF (see text). Numbers above the branches are bootstrap values greater than 80% (support for the clade indicated by * was 66%; support for the clade indicated by # was 46%). 488 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 su p p o r t) and for tify the placement of Li n n a e a Backlund and Donoghue, 1996; Backlund and with the rest of the Linnaeeae (100%) and of Bremer, 1997), the APG (1998) classification Trip l o s t e gia with the Dipsacaceae (80%). did not assign A d oxaceae to their more restricted Dipsacales. However, our analyses 30-Taxon datasets support the monophyly of the Dipsacales as tra- Summary statistics for our parsimony and ditionally circumscribed (see Fig. 2–3). Trees maximum likelihood analyses of separate and recovered when ndhF and rbcL are analyzed combined datasets are summarized in Table 3. separately unite Adoxaceae with Caprifoliaceae Trees obtained from separate searches using (Fig. 2), though with weak bootstrap support. each of the additional chloroplast markers (not Support for this conclusion is, however, reason- shown) were generally in strong agreement ably strong (70% for parsimony, 88% for max- with one another as well as with those from i mum likelihood) in combined r b cL /n d hF analyses of the combined data. The Dipsacales analyses (Fig. 3). Importantly, our results do tree recovered from our maximum likelihood not support the inclusion of Columellia and analysis of the combined dataset is presented in Desfontainia within Dipsacales, as suggested Fig . 4. On the basis of our broader analyses of previously (Backlund and Donoghue, 1996; rb c L and nd h F (see abo ve) , this is rooted along Backlund and Bremer, 1997). the bra n ch connecting A d oxaceae and The basal split within Dipsacales separates C ap ri fo l i a c e a e. For each 30-taxon analy s i s , the Adoxaceae from the Caprifoliaceae (sensu bo o t s t r ap values are given in Table 4 for the Judd et al., 1994; Donoghue et al., 2001). As in major clades defined by Donoghue et al. (2001), previous analyses, Viburnum is strongly linked whi c h are indicated on the tree in Fig . 1. with the compound-leaved A d ox o i d e a e Focusing on the results from our combined (Sambucus and Adoxa and its relatives). Within analyses, and specifically on the maximum Adoxoideae, our analyses strongly support the likelihood tree in Fig. 4, it is especially note- herbaceous Adoxina clade of Donoghue et al. worthy that bootstrap support for most clades (2001), with as the sister group of (16 of 27) is now at 100% and that four other plus Adoxa (contra Liang, 1997, who branches are supported above 95%. These well- united Sinadoxa and Adoxa). s u p p o rted re l ationships include seve ral fo r- R e l ationships at the base of the Cap ri- merly controversial links, most notably the foliaceae remain somewh at uncertain. Our connection between Tri p l o s t egi a a n d combined results place Diervilleae ( D i p s a c a c e a e. Within the Cap ri fo l i a c e a e, plus ) as the sister group of the remain- Diervilleae is identified as the sister group of ing taxa, but support for this position is weak, the rest, which form a clade supported at 84%. and the alternative that Caprifolieae is sister to The remaining problems, with bootstrap values the rest of the Caprifoliaceae cannot be rejected between 60% and 70%, are the relationships (1) with confi d e n c e. Unfo rt u n at e ly, this uncer- between Heptacodium and the Caprifolieae, (2) tainty limits our ability to infer the basal condi- among the major branches within Caprifolieae, tion for several key morphological features, (3) among Linnaeeae, M o ri n a c e a e, a n d especially carpel number (2, 3, or 5) and Valerianaceae plus Dipsacaceae, and (4) within type (capsule or ). the core Valerianaceae. Traditionally, the Caprifolieae has included L ey c e s t e ri a, L o n i c e ra, S y m p h o ri c a rp o s, a n d DISCUSSION AND CONCLUSIONS Triosteum. The monophyly of this group is very Our sep a rate and combined analyses of strongly supported, but relationships among the Dipsacales chloroplast DNA sequences yield major lineages remain obscure. The position of trees that are broadly consistent with previ- Heptacodium also remains uncertain. Although ously published results. And, in every case, the our combined analyses support its placement as p hy l ogenetic cl a s s i fi c ation proposed by the sister group of the Caprifolieae (as in Pyck Donoghue et al. (2001) is upheld (Fig. 1). The and Smets, 2000), confidence in this arrange- present study is exceptional, however, in pro- ment is limited in the individual and combined viding far gre ater confidence in the major analyses. Understanding the evolution of sev- clades within the Dipsacales; our combined eral characters, especially archi- analyses support most of the major clades with tecture, depends on the resolution of this issue. bootstrap values of 100% (Fig. 3–4). Our combined analyses strongly support the Because Dipsacales monophyly was called monophyly of Linnaeeae, despite the separa- into question by some previous studies (e.g., tion of Linnaea in rbcL analyses. Furthermore, 2001 BELL ET AL., DIPSACALES PHYLOGENY 489

FIG U R E 4. Maximum likelihood tree from the combined chloroplast DNA dataset for 30 taxa (Table 3). Numbers above the branches are bootstrap values greater than 80% (support for the clade indicated by * was 60%; support for the clade indicated by # was 67%). 490 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 relationships within Linnaeeae now seem well calyx of Morinaceae originated independently e s t abl i s h e d, with L i n n a e a b e i n g the sister of the 8-ribbed condition found in Dipsacaceae group of the rest, and Abelia and Dipelta more and the inner ep i c a lyx of Tri p l o s t egi a. closely related to one another than either is to Alternatively, the epicalyx may have originated Ko l k w i t z i a. These fi n d i n g s , along with the in the ancestor of the Valerina clade and then d e s c ription of the fossil D i p l o d i p e l t a been lost in the Valerianaceae. The winged (Manchester and Donoghue, 1995), provide a of , situated at the base of the context for the more-detailed developmental Valerianaceae, may represent an independent studies that are needed to establish homologies fusion of bracts or a stage in the loss of the epi- within the inflorescence. c a lyx; additional developmental studies are As suggested previously (e.g., Judd et al., needed to evaluate these possibilities. 1994; Backlund and Donog h u e, 1 9 9 6 ), th e R e l ationships remain uncertain within Linnaeeae is strongly united in our analyses Dipsacaceae, in part owing to the limited sam- with the Valerina clade of Donoghue et al. pling of taxa. Our 46-taxon ndhF analysis and ( 2 0 0 1 ) , wh i ch includes Mori n a c e a e, our 30-taxon analyses indicate that Dipsacaceae, and Valerianaceae. This supports and Pterocephalis are more closely related to the view that the characteristic abortion of two one another than either is to . In the carpels (Wilkinson, 1949) and the origin of combined 30-taxon analysis, the bootstrap sup- achene fruits (Fukuoka, 1972) occurred in the port for Pterocephalis plus Dipsacus reaches common ancestor of Linnaeeae and Valerina. 99%. However, this is contradicted by our 46- Valerina is monophyletic in all of our analyses taxon rbcL analysis and (weakly) by our com- ex c ept for the sep a rate analysis of m atK bined rbcL and ndhF analysis,as well as by the sequences, though bootstrap support remains m o rp h o l ogical analysis of Caputo and weak. This clade is marked by a number of morphological changes, including a shift to Cozzolino (1994). herbaceous habit. Finally, within Valerianaceae the conclusion Within Va l e ri n a , M o rinaceae is the sister that Patrinia and are basal lin- group of a clade including the Dipsacaceae, eages is now very well established. This sug- Trip l o s t e gia , and Val e ri a n a c e a e . Support for gests that the group initially diversified within Mo r inaceae is ver y stron g , and within this cla d e Asia, probably within the eastern Himalayas. Mo ri n a and Cry p t o t h l a d i a ar e more clo s e l y Relationships within the core Valerianaceae are re l ated to one another than either is to much less clear, aside from the strong connec- Ac a n t h o c a ly x . These results are supported by a tion between Fedia and . We are var iety of morph o l o gical cha ra c t e r s (Cannon encouraged by the weak union of and Cannon 1984; Caputo and Cozzo l i n o , with , which both have characteristic 1994). Specific a l l y, Mo ri n a and Cry p t o t h l a d i a spurs, but further resolution will require ar e united by possession of who r led , a additional taxonomic sampling and molecular tw o-lipped calyx , reduction from four to two markers. Of special interest is the origin and functional stamens, lobed flo r al nectarie s , an d spectacular dive rs i fi c ation of the South pollen with ext ra o rd i n a r y equato r ial prot r usions. American , which are not represented in One of the most interesting results of the pre- this dataset. sent analysis is the strong support obtained for Having now achieved a very high level of the linkage of Tri p l o s t egi a with the c o n fidence in the backbone phy l oge ny of D i p s a c a c e a e, as opposed to with the Dipsacales, the stage is now set for continued Valerianaceae (e.g., Backlund and Donoghue, resolution of relationships within major clades, 1996; Donoghue et al. 2001), or with especially Viburnum, Lonicera, Dipsacaceae, M o rinaceae plus Dipsacaceae (Peng et al., and core Valerianaceae. Dipsacales also now 1995). This result is consistent with some pre- provide an excellent system for studies of char- vious hypotheses concerning morp h o l ogi c a l acter evolution, diversification rate, and histor- evolution. For example, it supports the view ical biogeography. We caution, however, that ( e. g. , Hofmann and Gottmann, 1 9 9 0 ; our results are based solely on chloroplast DNA Manchester and Donoghue, 1995; Roels and data. Although chloroplast trees are consistent Smets, 1996) that the epicalyx in Morinaceae, with those based on morph o l o gy, we look for wa r d Triplostegia, and Dipsacaceae was derived by to the addition of nuclear and/or mitochondrial fusion of the supernumerary bracts seen in sequences to solidify our understanding of Linnaeeae. It is possible that the 12-ribbed epi- Dipsacales phylogeny. 2001 BELL ET AL., DIPSACALES PHYLOGENY 491 492 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 2001 BELL ET AL., DIPSACALES PHYLOGENY 493 494 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 2001 BELL ET AL., DIPSACALES PHYLOGENY 495 496 HARVARD PAPERS IN BOTANY Vol. 6, No. 2 2001 BELL ET AL., DIPSACALES PHYLOGENY 497 498 HARVARD PAPERS IN BOTANY Vol. 6, No. 2

LITERATURE CITED

ANGIOSPERM PHYLOGENY GROUP (APG). 1998. An FELSENSTEIN, J. 1985. Confidence limits on phyloge- ordinal classification for the families of flowering nies: An approach using the bootstrap. Evolution plants. Ann. Missouri Bot. Gard. 85: 531–553. 39: 783–791. BACKLUND, A., AND K. BREMER. 1997. Phylogeny of FU K U O K A, N. 1972. Taxonomic study of the the Asteridae s. str. based on rbcL sequences, with Caprifoliaceae. Mem. Fac. Sci. Kyoto Univ. Ser. particular reference to the Dipsacales. Plant Syst. Biol. 6: 15–58. Evol. 207: 225–254. GUS TA F S S O N , M. H. G. , A. BAC K L U N D, A N D B. —— — . 1998. To be or not to be: Principles of clas- BREMER. 1996. Phylogeny of the sensu sification and monotypic plant families. Taxon 47: lato based on rbcL sequences with particular ref- 391–400. erence to the Goodeniaceae. Plant Syst. Evol. 199: BAC K L U N D, A . , A N D M. J. DO N O G H U E. 1996. 217-242. M o rp h o l ogy and phy l oge ny of the ord e r HOFMANN, U., AND J. GOTTMANN. 1990. Morina L. Dipsacales. Pages 1–55 in A. BACKLUND, ED., and Triplostegia Wall. Ex DC. Im Vergleich mi Phylogeny of the Dipsacales. Part 4. Department Valerianaceae und Dipsacaceae. Bot. Jarb. Syst. of Systematic Botany, Uppsala Unive rs i t y, 111: 499–553. Uppsala, Sweden. JUDD, W. S., R. W. SANDERS, AND M. J. DONOGHUE. BACK L U N D , A. , AN D N. PYC K . 1998. Diervillaceae and Linnaeaceae, two new families of caprifolioids. 1994. Angiosperm family pairs—preliminary phy- Taxon 47: 657–661. logenetic analyses. Harvard Pap. Bot. 5: 1–51. KIM, K. J., AND R. K. JANSEN. 1995. ndhF sequence BELL, C. D. 2001. PORN:* A hierarchical likelihood ratio calculator for LINUX. Yale Unive rs i t y, evolution and the major clades in the sunflower Ne w Haven , Connecticut. http://pantheon.yal e. e d u / fa m i ly. Pro c. Natl. A c a d. Sci. U. S.A. 92: ~cdb29/pornstar.html 10379–10383. BREMER, B., R. G., OLMSTEAD, L. STRUWE, AND J. A. KIM, T., B.-Y. SUN, C. W. PARK, AND Y. SUH. 1999. SWEERE. 1994. rbcL sequences support exclusion Phylogenetic implications of matK sequences in of Retzia, Desfontainia, and Nicodemia from the Caprifoliaceae. Amer. Jour. Bot. (Suppl.) Abstract . Plant Syst. Evol. 190: 213–230. 86: 3202. CANNON, M. J., AND J. F. M. CANNON. 1984. A revi- LIANG, H. 1997. A comparative study of floral vas- sion of Morinaceae (Magnoliophyta-Dipsacales). culature in Adoxaceae. Acta Bot. Yunnanica 19: Bull. Brit. Mus. Nat. Hist. (Bot.) 12: 1–35. 260–264. CAPUTO, P., AND S. COZZOLINO. 1994. A cladistic LIN, C. M., Z.-Q. LIU, AND S. D. KUNG. 1986. analysis of Dipsacaceae (Dipsacales). Plant Syst. N i c o t i a n a ch l o roplast ge n o m e : X. Corre l at i o n Evol. 189: 41–61. between the DNA sequences and the isoelectric CRO N QU I S T, A. 1988. The Evolution and focusing patterns of the LS of rubisco. Plant Mol. Classification of Flowering Plants. New York Biol. 6: 81–87. Botanical Garden, Bronx, New York. MADDISON, W. P., AND D. R. MADDISON. 1992. DONOGHUE, M. J. 1983. The phyogenetic relation- MacClade: Analysis of Phylogeny and Character ships of Vi bu rnu m. Pages 143–166 in N. I. E volution. Ve rsion 3.0. Sinauer A s s o c. , PLAT N I C K A N D V. A. FU N K, E D., A dvances in Sunderland, Massachusetts. Cladistics. Vol. 2. Columbia University Press, MANCHESTER, S. R, AND M. J. DONOGHUE. 1995. New York. Winged fruits of Linnaeeae (Caprifoliaceae) in the —— — . 1985. Pollen diversity and exine evolution Tertiary of Western North America: Diplodipelta in Viburnum and the Caprifoliaceae sensu lato. gen. nov. Int. Jour. Plant Sci. 156: 709–722. Jour. Arnold Arb. 66: 421–469. MANEN, J.-F., A. NATALI, AND F. EHRENDORFER. DONOGHUE, M. J., T. ERIKSSON, P. A. REEVES, AND R. 1994. Phylogeny of -Rubieae inferred G. OLMSTEAD. Phylogeny and phylogenetic taxon- o my of Dipsacales, with special re fe rence to from the sequence of a cpDNA intergenic region. Sinadoxa and Tetradoxa (Adoxaceae). Harvard Plant Syst. Evol. 190: 195–211. Pap. Bot. 6: 459–479. MO R G A N, D. R., A N D D. E. SO LT I S. 1 9 9 3 . DONOGHUE, M. J., R. G. OLMSTEAD, J. F. SMITH, AND P hy l ogenetic re l ationships among members of J. D. PALMER. 1992. Phylogenetic relationships of Saxifragaceae sensu lato based on rbcL sequence Dipsacales based on r b cL sequences. A n n . data. Ann. Missouri Bot. Gard. 80: 631–660. Missouri Bot. Gard. 79: 333–345. OL M S T E A D, R. G. , A N D J. A. SW E E R E. 1 9 9 4 . DOYLE, J. J., AND J. L. DOYLE. 1987. A rapid isola- Combining data in phylogenetic systematics: An tion procedure for small quantities of fresh tis- empirical approach using three molecular datasets sue. Phytochem. Bull. 19: 11–15. in the Solanaceae. Syst. Biol. 43: 467–481. 2001 BELL ET AL., DIPSACALES PHYLOGENY 499

OLMSTEAD, R. G., B. BREMER, K. M. SCOTT, AND J. ROELS, P., AND E. SMETS. 1996. A floral ontogenetic D. PALMER. 1993. A parsimony analysis of the study in the Dipsacales. Int. J. Plant Sci. 157: Asteridae sensu lato based on rbcL sequences. 203–218. Ann. Missouri Bot. Gard. 80: 700–722. SWOF F O R D , D. L. 2001. PAUP*. Phy l o genetic An a ly s i s OLMSTEAD, R. G., K. J. KIM, R. K. JANSEN, AND S. J. Using Pars i m o n y (*and Other Methods). Ver sion 4. WAGSTAFF. 2000. The phylogeny of the Asteridae Sinauer As s o c. , Su n d e rl a n d , Ma s s a ch u s e t t s . s e n s u l at o based on ch l o roplast n d hF ge n e TABERLET, P., L. GIELLY, G. PAUTOU, AND J. BOUVET. sequences. Mol. Phylo. Evol. 16: 96–112. 1991. Universal primers for amplification of three OLMSTEAD, R. G.,H. J. MICHAELS, K. M. SCOTT, AND non-coding regions of chloroplast DNA. Plant J. D. PALMER. 1992. Monophyly of the Asteridae and identification of their major lineages inferred Mol. Biol. 17: 1105–1109. from DNA sequences of rbcL. Ann. Missouri Bot. WILKINSON, A. M. 1949. Floral anatomy and mor- Gard. 79: 249–265. phology of Triosteum and of the Caprifoliaceae in PENG, C.-I., H. TOBE, AND M. TAKAHASHI. 1995. general. Amer. Jour. Bot. 36: 481–489. Reproductive morphology and relationships of XIANG, Q.-Y., D. E. SOLTIS, D. R. MORGAN, AND P. S. Triplostegia (Dipsacales). Bot. Jahrb. Syst. 116: SOLT I S . 1993. Phyl o genetic rel a tionships of Co r nus 505–516. L. se n s ul at o and pu t at i ve rel at i ve si n fe rred fr om rb c L PYCK, N., AND E. SMETS. 2000. A search for the phy- sequence data. Ann. Missouri Bot. Gard. 80: 72 3 – 7 3 4 . l ogenetic position of the seven-son fl owe r YO U N G, N. D. , ST E I N E R, K. E., A N D C. W. (Heptacodium, Dipsacales): Combining molecular DEPAMPHILIS. 1999. The evolution of parasitism in and morphological evidence. Plant. Syst. Evol. S c ro p h u l a ri a c e a e / O ro b a n ch a c e a e : Plastid ge n e 225: 185–199. sequences refute an evolutionary transition series. PYCK, N., P. ROELS, AND E. SMETS. 1999. Tribal rela- Ann. Missouri Bot. Gard. 86: 876–893. tionships in Cap ri fo l i a c e a e : Evidence from a cladistic analysis using ndhF sequences. Syst. Geogr. Pl. 69: 145–159.