Phylogenetic Relationships in the : I. A. Cladistic Analysis of Morphological Data Author(s): Timothy M. Evans, Robert B. Faden, Michael G. Simpson, Kenneth J. Sytsma Source: Systematic Botany, Vol. 25, No. 4 (Oct. - Dec., 2000), pp. 668-691 Published by: American Society of Taxonomists Stable URL: http://www.jstor.org/stable/2666727 Accessed: 10/09/2010 13:35

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=aspt.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

American Society of Plant Taxonomists is collaborating with JSTOR to digitize, preserve and extend access to Systematic Botany.

http://www.jstor.org SystemnaticBotany (2000), 25(4): pp. 668-691 ? Copyright2000 by the AmericanSociety of Plant Taxonomists

PhylogeneticRelationships in the Commelinaceae: I. A Cladistic Analysis of Morphological Data

TIMOTHY M. EVANS1 Departmentof Botany,University of Wisconsin,430 LincolnDrive, Madison, Wisconsin 53706 Presentaddress, author for correspondence: Department of Biology,Hope College,35 East 12thStreet, Holland,Michigan 49423-9000

ROBERT B. FADEN Departmentof Botany,NHB 166,National Museum of NaturalHistory, Smithsonian Institution, Washington,DC 20560

MICHAEL G. SIMPSON Departmentof Biology,San Diego StateUniversity, San Diego, California92182

KENNETHJ. SYTSMA Departmentof Botany,University of Wisconsin,430 LincolnDrive, Madison, Wisconsin 53706

CommunicatingEditor: Richard Jensen

ABSTRACT. The plantfamily Commelinaceae displays a wide rangeof variation in vegetative,floral, and inflorescencemorphology. This high degree of variation,particularly among charactersoperating under strongand similarselective pressures (i.e., ), has made the assessmentof homology among morpho- logical charactersdifficult, and has resultedin severaldiscordant classification schemes for the family. Phy- logeneticrelationships among 40 of the 41 generain the familywere evaluatedusing cladisticanalyses of morphologicaldata. The resultingphylogeny shows some similarityto themost recent classification, but with some notabledifferences. Cartonemna (subfamily Cartonematoideae) was placed basal to therest of the family. Triceratella(subfamily Cartonematoideae), however, was placed amonggenera within tribe Tradescantieae of subfamilyCommelinoideae. Likewise, the circumscriptionsof tribesCommelineae and Tradescantieaewere in disagreementwith the mostrecent classification. The discordancebetween the phylogenyand the most recentclassification is attributedto a high degreeof convergencein variousmorphological characters, par- ticularlythose relating to the androeciumand the .Anatomical characters (i.e., stomatal struc- ture),on theother hand, show promisefor resolving phylogenetic relationships within the Commelinaceae, based upon theiragreement with the most recent classification.

The Commelinaceae,a well definedfamily of 41 neae is foundmainly in Africaand Asia. Only six genera and about 650 species,is characterizedby genera (,Buforrestia, Commelina, Floscopa, several featuresincluding a distinctclosed leaf Murdannia,and )have indigenousspecies in sheath,a succulentleaf blade, and three-merous bothhemispheres. flowerswith distinctpetals and (Cronquist Older classificationsof the Commelinaceae relied 1981; Faden 1985;Faden and Hunt 1991).The gen- heavilyon floralfeatures (see Fadenand Hunt1991, era are largelytropical and subtropical,but several fora review).Woodson (1942) firstused inflores- extendinto temperateregions. The greatestdiver- cence charactersfor higher-level classification, but sity is in ,where, along with Madagascar, he was largelyunfamiliar with the Old Worldgen- nearlyhalf the genera and about40% ofthe species era. Pichon(1946) employedanatomical characters are found(Faden 1983a). to separatethe Cartonemaas the familyCar- There is a naturaldivision of taxa betweenthe tonemataceae,but he returnedto more traditional Old and the New World.The seven subtribesof charactersto definehis ten tribesof Commelina- tribeTradescantieae (sensu Faden and Hunt 1991) ceae. Brenan(1966) used a varietyof morphological are each whollyconfined to eitherthe Easternor charactersto define15 informal"groups" of gen- Westernhemisphere, whereas the tribeCommeli- era. These have served as the basis of surveysof 668 2000] EVANS ET AL.: COMMELINACEAE 669

anatomical(Tomlinson 1966, 1969) and cytological molecularand morphologicalstudies are similarto (Jonesand Jopling1972) charactersin the family, the molecularphylogenies (e.g., Chase et al. 1995; but theyhave littletaxonomic significance. Linderand Kellogg1995). Recent work by Givnish, The mostrecent classification, which will be used Evans,Pires, and Sytsma(Givnish et al. 1995,1999) in thisstudy, was put forwardby Faden and Hunt generallysupports these findings. It also places the (1991).In theirclassification, the Commelinaceaeis Hanguanaceae as the sisterfamily of the Comme- divided into two unequal subfamilies(Table 1). linaceae and the Xyridaceaeand Eriocaulaceaein SubfamilyCartonematoideae contains the tribes the grass/sedgeclade. Cartonemateaeand Triceratelleae,each comprising Certainmorphological characters also lend sup- a single genus. SubfamilyCommelinoideae con- port to the separationof the Commelinaceaefrom tainsthe remaining38 generathat are arrangedin otherfamilies of the originalCommelinales (sensu two tribes,Tradescantieae and Commelineae,the Cronquist1981, or Dahlgrenet al. 1985). As dis- formerof whichis dividedinto seven subtribes. cussed above,Dahlgren et al. (1985) separatedthe Evans (1995)conducted a cladisticanalysis of the Commelinaceaefrom the othercommelinoid fami- Commelinaceaeusing both morphological and mo- lies in partby the presenceof raphides,which are lecularcharacters. This paper representsa modifi- absentin theMayacaceae, Rapateaceae, Xyridaceae, cation and expansionof the morphologicalcom- and Eriocaulaceae (Dahlgren and Clifford1982). ponentsof thatstudy. Raphides are widespread,however, in the Ponted- Relationshipsof Commelinaceaeto otherMono- eriaceae,Philydraceae, and Haemodoraceae(Dahl- cot Families. The familyCommelinaceae is the gren et al. 1985). Additionally,whereas the Com- namesake forthe order .Cronquist melinaceaewere separatedfrom other families in (1981) definedthe orderby thepresence of perfect the Commelinalesby thepresence of an amoeboid flowersthat are adapted for generalinsect polli- tapetum,the genus Pontederiaand all investigated nation,having showy thatare differentiatedmembers of the Haemodoraceae also have an amoe- fromthe sepals. He includedthree other families in boid tapetum(Dahlgren and Clifford1982; Simp- this order:Xyridaceae, Rapateaceae, and Mayaca- son 1988, 1990). Finally,flowers in the Commeli- ceae. He separated the Commelinaceaefrom the naceae show a strongtendency toward zygomor- other three families by their well-definedand phy,whereas they are generallyactinomorphic in closed leaf sheath,and thesucculent blade. the other membersof the Commelinalessensu Dahlgrenet al. (1985)included the same families Cronquist(1981) or Dahlgrenet al. (1985). Zygo- in Commelinalesas Cronquist,but with the addi- morphicflowers are common,however, in thePon- tionof theEriocaulaceae. Eriocaulaceae was placed tederiaceae,Philydraceae, and Haemodoraceae. in itsown orderby Cronquist,who maintainedthat Althoughnumerous taxonomic treatments have it was mostlikely derived within Xyridaceae. Dahl- been produced forthe Commelinaceae,there has grenet al. (1985) distinguishedthe Commelinaceae been littleconsensus as to whichcharacters should fromthe othercommelinoid families by the pres- be used to definerelationships among the genera. ence of raphides,an amoeboid tapetum,and the The earliersystems relied heavily upon featuresof leaf charactersused by Cronquist (closed leaf the highlyvariable androecium, but theyignored sheathand succulentblade). charactersof the inflorescence.The classification Although cladistic analyses of morphological producedby Faden and Hunt (1991),which will be data supporta monophyleticCommelinales sensu used throughoutthis study,incorporated a broad Dahlgrenet al. (1985;Stevenson and Loconte1995), rangeof information,such as charactersfrom mor- recentlypublished moleculardata suggest other- phology,anatomy, and palynology.None of these wise. Nucleotidesequence data fromthe chloroplast classifications,however, is rootedin an evolutionary gene rbcL(Chase et al. 1993;Clark et al. 1993;Du- framework.The purpose of this studywas to ex- vall et al. 1993) place the Commelinaceaenear the amine the intergenericrelationships in the Com- Pontederiaceae,Philydraceae, and Haemodoraceae, melinaceae using morphologicaland anatomical all of which are part of a largerclade containing charactersin a cladisticanalysis. thefamilies of the order Zingiberales. The Rapatea- ceae (theonly other representative of Commelinales MATERIALS AND METHODS included in the previous molecular studies) is placed in a differentclade that contains,among OperationalTaxonomic Units (OTU's). A criti- otherfamilies, Poaceae and Cyperaceae.Combined cal step in any cladisticanalysis is the selectionof 670 SYSTEMATICBOTANY [Volume 25

TABLE 1. Classificationof the Commelinaceae (as presentedby Fadenand Hunt1991). *indicates Old Worldsubtribe, **indicatesNew Worldsubtribe.

Family:Commelinaceae R. Br. Subfamily: Cartonematoideae(Pichon) Faden & D. Hunt Tribe: Cartonemateae(Pichon) Faden & D. Hunt CartonemaR. Brown Tribe: TriceratelleaeFaden & D. Hunt TriceratellaBrenan Subfamily: Commelinoideae(Bruckner) Faden & D. Hunt Tribe: Tradescantieae(Meisner) Faden & D. Hunt Subtribe: *PalisotinaeFaden & D. Hunt PalisotaReichb. Subtribe: *StreptoliriinaeFaden & D. Hunt StreptolirionEdgew. SpatholirionRidley AetheolirionForman Subtribe: *Cyanotinae(Pichon) Faden & D. Hunt CyanotisD. Don (Including AmischophacelusR. Rao & Kamm.) BelosynapsisHassk. Subtribe: *ColeotrypinaeFaden & D. Hunt ColeotrypeC. B. Clarke PorandraD. H. Hong AmischotolypeHassk. Subtribe: **Dichorisandrinae(Pichon) Faden & D. Hunt DichorisandraMikan SiderasisRaf. GeogenanthusUle CochliosteinaLem. Undescribedgenus Subtribe: **ThyrsantheminaeFaden & D. Hunt ThyrsanthemumnPichon GibasoidesD. Hunt TinantiaScheidw. ElasisD. Hunt MatudauithusD. Hunt WeldeniaSchult. f. Subtribe: **TradescantiinaeRohw. GibasisRaf. TradescantiaL. (IncludingSetcreasea Schumann & Sydow,Separotheca Waterf., Cymn- bispathaPichon, Campelia Rich., Rhoeo Hance, Zebrina Schnizl.) CallisiaLoefl. (Including Hadrodemnas H. Moore, Cuthbertia Small, AploleiaRaf., Leiandra Raf., Phyodina Raf., Leptorhoeo C. B. Clarkeex Hemsley) TripogandraRaf. SatIvalleaWright Tribe: Commelineae(Meisner) Faden & D. Hunt StanfieldiellaBrenan FloscopaLour. BuforrestiaC. B. Clarke MurdanniaRoyle AnthericopsisEngl. TricarpelemaJ.K. Morton PseudoparisH. Perrier PolyspathaBenth. DictyospermumWight PolliaThunb. AneilemnaR. Brown RhopalephoraHassk. CommelinaL. (IncludingPhaeosphaerion Hassk., Commelinopsis Pichon) 2000] EVANS ET AL.: COMMELINACEAE 671 evolutionaryunits, or operationaltaxonomic units son 1990; see also Bininda-Emondset al. 1998). (OTU's). It is assumed thatthe unitsbeing exam- Some of the difficultiesof addressingthe firstop- ined have all descendedfrom a most recentcom- tion(monophyly) are discussedabove. The theoret- mon ancestor,and representdistinct evolutionary ical and practicaldifficulties in dividingall poly- entities.If the OTU's are polyphyletic,then inac- morphicOTU's into monomorphicunits is dis- curate("misleading") reconstructions of theirevo- cussed in some detailin Donoghue (1994). lutionaryhistory may be produced(Nixon and Da- Therefore,in this analysis,a single species was vis 1991;Donoghue 1994). scored for each genus where possible (Table 2). Of the 41 generain the Commelinaceae,ten are Data forindividual species of four genera (Porandra, monotypic(Aetheolirion, Anthericopsis, Elasis, Giba- Pseudoparis,Rhopalephora, and an undescribedge- soides,Matudanthus, Sauvallea, Streptolirion, Tapheocar- nus) were unavailable,so theywere leftas poly- pa Conran,Triceratella, and Weldenia),and thereis morphicto reflectvariation within each genus.An no difficultyin treatingeach as an OTU. For each attemptwas made,whenever possible, to selectthe of the remaining31 genera,however, claims of same species being used in a complementarymo- monophylymight be misleading.The genus Calli- lecular cladistic analysis of the Commelinaceae sia, for example,as circumscribedby Faden and (Evans et al., in prep.)so thatthe resultsfrom the Hunt (1991), contains20 species in seven sections two data sets would be moredirectly comparable. (Hunt 1986b). There is no single morphological On the basis of the recentmolecular studies on characterthat defines the genus,and its question- monocots(Chase et al. 1993;Clark et al. 1993;Du- ably monophyleticstatus (Hunt 1986b)might pro- vall et al. 1993;Givnish et al. 1999),representatives duce misleadingrelationships in a cladisticanaly- of the familiesPontederiaceae (Heteranthera) and SiS. Haemodoraceae (Haemodorum)were used as out- Likewise,the genus Tradescantiasensu Fadenand groups.These two taxa were selecteddue to their Hunt (1991) is large (containingabout 70 species) putativelybasal positionin theirrespective families and morphologicallydiverse. Some membersof this (Simpson1990; Graham et al. 1998)and availability genusuntil recently have been placed intoas many of morphologicaldata. Fortyingroup and two out- as six othergenera (Setcreasea, Separotheca, Cymbis- group genera or exemplarswere includedin this patha,Campelia, Rhoeo, and Zebrina).Hunt (1975, study(Appendix 1, 2). The recentlydescribed ge- 1980,1986a) has graduallylumped all ofthese gen- nus TapheocarpaConran (1994) was not included era into ,but the monophylyof the ge- due to insufficientdata. nus is yetto be demonstrated. Character Selection. Characters were scored Althoughit is clearthat large genera such as Cal- fromboth living and pressedplants housed at the lisia and Tradescantiamust be treatedcautiously in US National Herbarium(US). Observationswere a cladisticanalysis, the same difficultiesmay also made on living plant material growing in the arise for smaller,more "clearly defined"genera. SmithsonianInstitution Botany Research Green- Rhopalephora,for example, is distinguishedfrom houses and in the field.Where living or pressed Aneilemaby a combinationof characters:inflores- were not available for examination,infor- cence axis veryshort, filaments fused ba- mationwas takenfrom published literature (Table sally, and denselycovered by hook 2). Fortyseven characters were scored. hairs,and dorsal capsule valve deciduous (Faden PhylogeneticAnalysis. Data were enteredinto 1977).All ofthese characters occur within Aneilema, a matrixusing the computerprogram MacClade butnot in thiscombination, and no singlemorpho- version3.05 (Maddison and Maddison 1992), and logicalcharacter is unique to Rhopalephora.The im- the phylogeneticanalyses were performedwith plied hierarchicalrelationships between the two PAUP*version 4.Ob2a (Swofford1999). A total of generasuggests one or the otherof themmay be 3.7 percentof thecells ofthe matrix were scored as paraphyletic. "missing,"either because thedata wereunavailable To avoid the problemsdiscussed above, each orbecause particular characters were not applicable OTU should be treatedin one of threeways: 1) it to certaintaxa. shouldbe clearlydemonstrated to be monophyletic; In an initialanalysis, all characterswere treated 2) if polymorphicfor any character,it should be as unordered.A multiple-islandsapproach was split into smaller groups that are monomorphic used to findthe most parsimonious trees (modified (Nixon and Davis 1991); or 3) it should be repre- fromOlmstead et al. 1993;see also Maddison1991; sentedin the analysisby exemplarspecies (Simp- Olmstead and Palmer1994). To evaluatethe sup- 672 SYSTEMATICBOTANY [Volume 25

TABLE 2. Primaryliterature sources for information at thespecies level used in a cladisticanalysis of morphological charactersin theCommelinaceae.

Species Source Aetheolirionstenolobiutn Forman Forman1962 Ainischotolypehispida (Lesson & A. Rich.)D. Y. Hong Clarke1881 Anithericopsissepalosa (C. B. Clarke)Engl. Brenan1966; Faden, unpubl. data Aneilemnacalceolus Brenan Faden 1991 Belosyinapsiskewensis Hassk. Clarke1881; Faden, unpubl. data Buforrestiamaunii C. B. Clarke Brenan1960; Faden, unpubl. data Callisiaelegans Alexander ex. H. E. Moore Moore1958 Cartonemaphilydroides E Muell. Brenan1966; Faden, unpubl. data Cochliosternaodoratissirntm Lemaire Read 1965;Faden, unpubl. data Coleotrypenatalensis C. B. Clarke Obermeyerand Faden 1985 Commneliniabenghalensis L. Obermeyerand Faden 1985 Cyanotisbarbatus D. Don Brenan1966 Dichorisandrahexandra (Aubl.) Standl. Matuda 1955 Dictyospermumrontacmtm Wight Morton1966; Faden, unpubl. data Elasishirstita Hunt Hunt 1978;Faden, unpubl. data Floscopaafricania (P. Beauv.)C. B. Clarke Berhaut1988 Geogenanthuspoeppigii (Miq.) Faden Moore 1954 Gibasisgeniculata (Jacq.) Rohweder Hunt 1985 Gibasoideslaxiflora (C. B. Clarke)Hunt Hunt 1978;Faden, unpubl. data Mattudanthusnanius (Martens & Gal.) Hunt Hunt 1978;Faden, unpubl. data Murdanniaedutlis (Stokes) Faden Faden 1980 Palisotaambigua (P. Beauv.)C. B. Clarke Clarke1881 Pollicmannii C. B. Clarke Brenan1966 Polyspathapaniculcta Benth. Clarke1881; Faden, unpubl. data Porandra Hong 1974;Faden, unpubl. data Pseudoparis Faden,unpubl. data Rhopalephora Faden 1975,1977 Sauvallea Faden,unpubl. data Siderasisftuscata (Lodd.) H. E. Moore Clarke1881 (as Pyrrhema);Faden, unpubl. data Spatholirionlongifoliumn (Gagnep.) Dunn Faden 1985 Stanfieldiellaimperforata (C. B. Clarke)Brenan Brenan1960 Streptolirionvoltubile Edgew. Forman1962 Thyrsanthentinufloribundum (Martens & Galeotti)Pichon Hunt 1993 Tinantitierecta (Jacq.) Schltdl. Hunt 1993 Tradescantiavirgiiiacma L. Andersonand Woodson1935 Tricarpeleinagigantea (Hassk.) Hara Morton1966; Faden, unpubl. data Triceratelladrurnniondii Brenan Brenan1961 Tripogandraamplexicatulis (Klotzsch) Woodson Hunt 1993 Weldeniacatndida Schultes s. Hunt 1993;Faden, unpubl. data UndescribedGenus (Faden,in prep.) Faden,unpubl. data

port foreach node, 100 bootstrapreplicates were in PAUP) based on theirfit to the treesproduced conducted.Due to the high numberof treespro- in the unorderedanalysis. The multi-stepanalysis duced, a maximumof 5000 treesper replicatewas describedabove was thenrepeated using the re- saved. weightedcharacters. This process was repeatedun- As a measureof whichcharacters are morephy- til the same set of most parsimonioustrees was logeneticallyinformative and whichare morehom- producedin two consecutiveruns. PAUP allows the oplasious (with respect to the topologiesof the characterto be re-weightedbased on consistency treesproduced), as well as to arriveat a morestable index (CI), retentionindex (RI), or rescaledconsis- topology,the characterswere re-weighteda poster- tencyindex (RC). The maximal value of each of iori(using the "REWEIGHT CHARACTERS" command theseindices was used in threeseparate analyses. 2000] EVANS ET AL.: COMMELINACEAE 673

In additionto theunordered analysis, an analysis deredanalysis. The CI- and RC-basedanalyses pro- was performedin which threecharacter transfor- duced identicalsets of 15 equally mostparsimoni- mationswere ordered( locule number[Char- ous trees(not shown) thatdiffered from the unor- acter35] and seed numberper locule [Characters dered topologyin the lineages sisterto the zygo- 37 and 38]). The same proceduredescribed above morphicclade. In the unorderedtree (Fig. 1), the was used in this analysisto findmultiple islands zygomorphicclade has a sistergroup consisting of of mostparsimonious trees. Stanfieldiella,Pollia, , and .The For purposesof characterstate mapping, one of CI- and RI-basedanalyses produced a set ofnested the mostparsimonious trees was arbitrarilyselect- relationshipsfrom the more basal Siderasisto, se- ed forillustration. However, due to thelack of res- quentially,Pollia, Stanfieldiella, Anthericopsis, Dischor- olutionof several clades in thestrict consensus tree, isandra,and the zygomorphicclade. distributionswere examined on all ofthe most par- Ordered CharacterAnalysis. When assump- simonioustrees. tions about the sequence of evolutionarysteps in characters35, 37, and 38 were incorporated,115 RESULTS equally most parsimonioustrees were produced (notshown). This topology is nearlyidentical to the UnorderedAnalysis. In theunordered analysis, unorderedtopology, except that it showed greater 154 equally most parsimonioustrees were pro- resolutionin thezygomorphic clade. The undescri- duced,with a length(excluding autapomorphies) of bed genus and Cochliostemawere united in one 239 steps and a consistencyindex (CI) of 0.43 clade and Polyspatha,Aneilema, and Commelina (length= 272 and CI = 0.44 whenautapomorphies formedan unresolvedclade. are included).Figure 1 illustratesthe tree chosen forcharacter state reconstructions. DISCUSSION Cartonemais sisterto the restof Commelinaceae. The remaininggenera form four lineages (Fig. 1): The discordantclassifications that have been pro- Callisia,which is sisterto the remaininggenera; posed forthe Commelinaceaereflect the high de- Weldenia;and two largelyunresolved clades. Within gree of uncertaintyin homologyamong morpho- the clade thatis sisterto Anthericopsis,15 genera logical charactersin the family.Likewise, the low (Buforrestia,Floscopa, Pseudoparis, Tricarpelemna, Polys- CI values formany characters in this study(Table patha,Aneilema, Commelina, Rhopalephora, Dictyosper- 3) demonstratea highamount of homoplasyin the mum,Palisota, Tinantia, , Aetheolirion, family. The cladisticanalysis presentedhere, al- ,and an undescribedgenus) forma thoughsupporting particular elements of some pre- clade thatis supportedby two characters,antese- vious classifications,does not agree closelywith palous staminalfilament lengths (Character 26), any one of them.The zygomorphicclade (Fig. 1) and the presenceof a zygomorphicandroecium includes all but three genera (Pollia,Stanfieldiella, (Character31; note:Floscopa produces an asymmet- and Murdannia)of Faden and Hunt's (1991) tribe ric androecium).For ease of reference,this clade Commelineae.However, several membersof Tra- will be referredto as the"zygomorphic clade." The descantieaeare also included in this clade. Like- remaininggenera (Murdannia, Gibasis, Elasis, Thyr- wise, fewelements of Brenan's(1966) classification santhemumn,Gibasoides, Matudanthus, Coleotrype, Por- are supported.Of the15 informal"groups" put for- andra,, , , Tricera- ward by Brenan,one, "Group VI" (= subtribeCy- tella,Streptolirion, Spatholirion, Sauvallea, Tradescantia, anotinae), is monophyletichere. Relatively narrow and Tripogandra)form a large clade that is sup- circumscriptionsof severalof Brenan'sgroups (i.e., ported by the presence of bearded antesepalous the inclusionof onlyone or two generaper group) staminalfilaments (Characters 22, 23). For ease of make a directcomparison of relationshipsamong reference,this clade will be referredto as "Clade those generadifficult. Pichon's (1946) classification 1." included ten tribes,five of which containonly a A posteriori Re-weighting.Each of the re- singlegenus. Of the remainingfive tribes, not one weightedanalyses (CI-, RI-, and RC-based;Table 3) is supportedby thisanalysis. The largenumber of producedtopologies that were similarto the un- taxonomicgroups with only a single genus in Pi- orderedtree. The RI-based analysis produced 45 chon'sand Brenan'ssystems, as well as lack of sta- equally most parsimonioustrees (not shown) that tisticalsupport for most clades in thisanalysis, sug- were a subset of the 154 treesfound in the unor- gest thatthere is a high degree of uncertaintyin 674 SYSTEMATICBOTANY [Volume 25

Anthericopsis Buforrestia Floscopa Pseudoparis

4 ||0j C | Commelineae

60*%O Aneilema r"| |Commelina g||1[ j_Rhopalephora mu. Palisota I Palisotinae Tinantia I Thyrsantheminae _.. | Undescribedgenus |||||||||||||||||||| |_Cochlostema Dichorisandrinae Tradescantieae IHU| | | | | | Aetheolirion StreptoliriinaeS Geogenanthus Dichorisandra Dichorisandrinae Siderasis Pollia I Commelineae Stanfieldiella Weldenia I Thyrsantheminae I Tradescantieae Murdannia I Commelineae

Gibasis ITradescant nae F/asis j I Thyrsanthemum| 1 Gibasoides ~~~~Thyrsantheminae Matudanthus Coleotrype *Tradescantieae

Porandra Coleotrypinae Amischotolype Cyanotis Cn a Be/os ynapsis Caoiu Triceratella I Triceratelleae I (lade Streptolirion Streptoliriinae 80% Spatholirion

I ~~~~SauvalleaSauvallea I Tradescantieae Tradescantia Tradescantiinae Tripogandra || 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~Ca//lisia| ~ ~~~ Cartonema | Cartonemateae E Haemodorum Heteranthera

FIG. 1. A representativeof the 154 equallymost parsimonious trees found in theunordered analysis of 47 morpho- logicalcharacters in Commelinaceaeusing Haeniodorurin and Heteranztheraas outgroups. Dashed linesrepresent branches thatcollapse in the strictconsensus of mostparsimonious trees. Bootstrap values greaterthan 50% are shown above branches.Subtribal and tribalaffinities are shownto rightof genericnames. "Clade 1" and the"zygomorphic clade" discussedin the textare indicated. assessinghomology of structures among genera of Cartonema.Callisia is a memberof subtribeTra- Commelinaceae. descantiinae,a well definedgroup based on fea- Basal Lineages in the Commelinaceae. Each tures of the inflorescence(cincinni contracted, analysisdivided the Commelinaceae into several fused in bifacialpairs; Faden and Hunt 1991).The mainlineages, with Cartonema sister to therest of New Worlddistribution of this entiresubtribe, as thefamily and Callisiasister to everythingexcept well as its putativelyrecent origin (Faden and 2000] EVANS ET AL.: COMMELINACEAE 675

TABLE 3. Characterstatistics from each analysis.CI = consistencyindex; RI = retentionindex; RC = rescaled consistencyindex; Wc, = characterweight after re-scaling according to the CI; WRI = characterweight after rescaling accordingto theRI; WRC = characterweight after re-scaling according to theRC.

Weight

Character States Steps CI RI RC WCI WRI WRC 1 01 3 .333 .500 .167 333 286 167 2 01234 13 .308 .308 .095 307 357 94 3 01 9 .111 .429 .048 100 400 36 4 012 6 .333 .556 .185 286 500 127 5 01 2 .500 0.00 0.00 500 0 0 6 012 7 .286 .444 .127 286 500 127 7 01 5 .200 .636 .127 200 583 127 8 01 4 .250 .400 .100 250 286 100 9 012 7 .286 .375 .107 286 375 107 10 01 3 .333 .600 .200 333 800 200 11 01 4 .250 0.00 0.00 250 0 0 12 01 5 .200 .429 .086 200 375 86 13 01 3 .333 .333 .111 333 667 111 14 01 2 .500 .500 .250 500 333 250 15 01 2 .500 .500 .250 500 333 250 16 01 3 .333 .500 .167 333 500 167 17 01 1 1.00 - - 1000 1000 1000 18 01 1 1.00 1.00 1.00 500 1000 417 19 01 2 .500 .500 .250 1000 1000 1000 20 01 3 .333 .333 .111 333 333 111 21 01 3 .333 .778 .259 250 778 167 22 01 3 .333 .882 .294 500 889 471 23 01 3 .333 .882 .294 500 889 471 24 01 1 1.00 1.00 1.00 1000 1000 1000 25 01 3 .333 .889 .296 250 842 208 26 012 2 1.00 1.00 1.00 1000 1000 1000 27 012 2 1.00 1.00 1.00 1000 1000 1000 28 012 7 .286 .167 .048 333 333 111 29 012 9 .222 .222 .049 222 333 49 30 01 2 .500 0.00 0.00 500 0 0 31 012 5 .400 .812 .325 500 813 438 32 012 3 .667 0.00 0.00 667 0 0 33 01 7 .143 0.00 0.00 143 0 0 34 012 3 .333 .667 .222 333 667 222 35 012 4 .500 .333 .167 500 333 167 36 01 6 .167 .444 .074 125 444 28 37 0123 7 .286 .737 .211 286 609 211 38 0123 8 .250 .667 .167 250 545 167 39 0123 11 .273 .500 .136 273 421 136 40 01 6 .167 .500 .083 200 455 120 41 01 2 .500 .500 .250 500 500 250 42 01 5 .200 .750 .150 167 750 115 43 0123 9 .333 .667 .222 375 722 271 44 01 3 .333 .600 .200 333 800 200 45 01 2 .500 .667 .333 500 667 333 46 012 2 1.00 1.00 1.00 1000 1000 1000 47 01 1 1.00 1.00 1.00 1000 1000 1000 676 SYSTEMATICBOTANY [Volume 25

Hunt 1991) shed doubt on the basal position of used anatomicalcharacters to suggestthat the ge- Callisia. nus may forma linkbetween Cartonema, which he The positionof Cartonemasister to therest of the placed in a separatefamily, and the Commelina- familyis consistentwith otherdata, particularly ceae. He indicatedthat it mightbe treatedas a sep- rbcLsequences (Evans 1995;Evans et al., in prep.). aratesubfamily of Commelinaceae. Faden and Hunt On the basis of various morphologicalcharacters (1991) placed Triceratellaand Cartonemain a sub- (e.g.,actinomorphic flowers with six fertilestamens, familywithin the Commelinaceae, with each genus the absenceof raphidecanals), Cartonema has been in a separatetribe. postulatedto be one of the more primitivemem- The morphologicaldata presentedhere place Tri- bers of the family(Pichon 1946; Hutchinson1959; ceratellaas a sistergroup of subtribeCyanotinae Brenan1966; Faden and Hunt 1991). (Cyanotisand Belosynapsis),nested well withinthe Relationshipsin the Tradescantieae.With the Tradescantieae.The clade containingBelosynapsis, exceptionsof Murdannia(tribe Commelineae) and Cyanotis,Sauvallea, Spatholirion, Streptolirion, and Tri- Triceratella(subfamily Cartonemateae), Clade 1 con- ceratellais supportedby a singlecharacter (sessile sistsexclusively of members of Tradescantieae (Fig. flowers).The presenceof sessile flowersis labile 1). Several membersof the tribe,however, are withinCommelinaceae, having been lost once and placed in other parts of the tree. As discussed gained at least threetimes (Fig. 2). Althougha re- above,Callisia, for example, is basal to all othergen- lationshipbetween Triceratellaand the Streptoliri- era except Cartonema.Likewise, eight genera are inae has some support,it needs furtherstudy. Until placed eithersister to the zygomorphicclade (Di- less homoplasiouscharacters are foundto support chorisandraand Siderasis)or withinit (Palisota,Tin- such a relationship,it will remainsuspect. antia,undescribed genus, Cochliostema, Aetheolirion, Of the seven subtribescircumscribed by Faden and Geogenanthus).The position of Dichorisandra and Hunt (1991),only two, Coleotrypinae and Cy- and Siderasisas thesister group of Pollia (tribe Com- anotinae,are supported as monophyleticin this melineae)is supportedby thepresence of biseriate study. The remainingsubtribes are polyphyletic seeds (Character36). This character,however, is and largelyunresolved. Two genera in subtribe highlyvariable within the family.Within the zy- Thyrsantheminae(Thyrsanthemum and Elasis) con- gomorphicclade and its sisterclade, biseriate seed tributeto a polytomyin Clade 1. Thyrsantheminae, arrangementhas arisena minimumof threetimes as definedby Faden and Hunt (1991), is based (Fig. 2). This characteris evenmore variable in oth- largelyupon the absenceof charactersor thepres- er most parsimonioustrees, some of which force ence ofputatively primitive characters, and maybe Aetheolirionand Geogenanthusto gain biseriate seeds an unnaturalassemblage (R. Faden,unpubl. data). independently.Likewise, relative filament length Recentmolecular data also suggestthat this sub- withinthe antepetalousstaminal whorl (Character tribeis polyphyletic(Evans et al., in prep.),and the 27) is the single characterthat supports the clade lack ofresolution and weak supportin thisanalysis containingAetheolirion, Cochliostema, Geogenanthus, illustrate the difficultyin findingtaxonomic char- Tinantia,and the undescribedgenus as the sister acters that will unambiguouslyunite these taxa groupof thezygomorphic clade. This charactertoo withother genera. is variable within the zygomorphicclade (not Relationshipsin the Commelineae. Ten of the shown). thirteengenera in tribeCommelineae were placed The taxonomicplacement of the genus Triceratella within the zygomorphic clade (Fig. 1). Of thethree has been problematic.When Triceratellawas origi- remainingCommelineae genera, two (Pollia and nally described,a loose affinitywith StanfieldiellaStanfieldiella) are part of the sistergroup of the zy- was suggested (Brenan 1961). Tomlinson(1964) gomorphicclade; Murdannia is nestedwithin Clade

FIG. 2. Distributionof characterson the representativetree (see Fig. 1) fromthe unordered analysis of 47 morpho- logicalcharacters in Commelinaceae.Left: Seed arrangement.Biseriate seed arrangementhas arisenat leastthree times withinthe zygomorphicclade and its sisterclade. One additionalstep is requiredin some otherequally most parsi- monioustrees due to shiftsin positionof Aetheolirion and Geogenanthus.Right: Pedicel. The presenceof pedicellate flowers is homoplasiouswithin Commelinaceae. The distributionof thischaracter is notaltered in anyother most parsimonious trees. 2000] EVANS ET AL.: COMMELINACEAE 677

U) it-o-5

U~~~~~?~~ ~ ~ ~ ~~ - US Z -,CZ E -8 : U U.~~~~~~~~~-~~~~ U CU Q Q) QE ~ E E EE E E EE ) EE E E )E E E E

U-~~~~~~~~~~~~~~~~~~~~~~~~~~~j.UU ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~~~~~~Q 678 SYSTEMATICBOTANY [Volume 25

1. The only synapomorphythat supports the zy- weightedon the basis of threedifferent measures gomorphicclade is the presenceof a zygomorphic of theirinternal consistency (CI, RI, and RC). Each androecium(Character 31), a characterthat is var- of these re-weightingschemes starts with the un- iable withinthe familyas a whole. However,all orderedtopology, and thenre-evaluates the fitof membersof thisclade, except Palisota, share a com- the charactersto thattopology. The CI, a measure mon stomataltype consistingof six cells, with of the fitof the charactersto thetree, is derivedas small terminalsubsidiary cells (Figs. 3 and 4). Al- the ratioof thenumber of statechanges for a char- thoughthis stomatal type is restrictedto generabe- acterin the data to thenumber of statechanges for longingto the tribeCommelineae, it is not a syna- the characteron the hypothesizedtree (Kluge and pomorphyfor the zygomorphicgroup, as it is also Farris1969). A potentialproblem with the consis- foundin Pollia,Anthericopsis, and Murdannia.This tencyindex, however, is thatit onlymeasures how anatomicalcharacter is in agreementwith a molec- well a characterfits onto a particulartopology; it ular phylogenyfor Commelinaceae(Evans 1995; does not evaluatehow well thatcharacter actually Evans et al. 2000),and the distributionof stomatal supportsthe topology (i.e., if it is synapomorphicfor typeshere raises questions regarding the placement any clade). Farris(1989) developedthe retention in- of Murdanniawithin Clade 1, and calls for a re- dex (RI) and rescaledconsistency index (RC) in an evaluationof the homology of the filament bearding attemptto incorporatethe internalconsistency of betweenMurdannia and the othergenera in that each characterwith respectto the worstand best clade. possiblefit to a tree.The RI and RC both incorpo- Althoughthe unorderedanalysis produced a ratethe minimum number of steps a charactermust highly unresolved zygomorphicclade, the re- undergoon any tree,as well as the actual amount weightinganalyses each yielded a smaller clade of synapomorphieson thehypothesized tree. withinthe zygomorphicgroup containingPolyspa- By using these values as bases forre-weighting tha,Aneilema, Rhopalephora, Commelina, Dictyosper- characters,slightly different properties of thechar- mum,and Tricarpelema.This clade is supportedby actersare being selected.When CI is used, forex- a reductionin seed numberin each locule (Tricar- ample,each characteris re-weightedaccording to pelemais polymorphicfor this character).As rela- the numberof steps it undergoeswhen mapped tionshipsin the tribeCommelineae have been dif- onto the tree.However, this schemedoes not take ficultto evaluatedue to a lack of informativechar- intoaccount whether the character supports any of acters(Faden and Hunt 1991),the formation of dis- the clades in the tree.Incorporating RI and RC, tinct clades based on fruit characteristicsmay however,does take this into account.A good ex- providesome insightas to where such characters ample of the differencesin thesemeasures can be may be found. seen withCharacter #5 (number of flowers per cin- A Comparisonof Different Methods of Parsimony cinnus).This characterhas a CI of 0.5 (Table 3). Analysis. Several differentmethods of character However,it does not uniteany clades (Fig. 4), so it weightingand orderingwere exploredwith this lends no supportto the topology.Because it does data set,and thequestion arises as to whichone(s) not supporta clade,it has an RI and rC value of are the best approximationof the true historyof zero (Table 3). Therefore,the CI re-weighting the family.The unorderedanalysis containsthe scheme assigned a relativelyhigh weightto that fewesta prioriassumptions about characterevolu- characterwhereas the RI and RC decreased its tion,as it incorporatesfew assumptions about char- weight. acterweight or directionof transformation. The ab- While each of the re-weightingschemes pro- sence of additionalassumptions, however, imposes duced phylogeniessimilar to the unorderedphy- a penaltyin termsof the potentialloss of phylo- logeny(RC- and Cl-basedanalyses produced a sub- geneticinformation. For example,the numberof set of theunordered trees), they all showedgreater locules in the fruit(Character 35) may representa resolutionthan the unorderedanalysis. Whereas series,in which fruitswith threeunequal locules the unweightedanalysis produced a nearlycom- (e.g., one locule is reduced) are intermediatebe- pletelyunresolved zygomorphic clade, each of the tweenfruits with three locules and fruitswith two re-weightinganalyses produced a highlyresolved locules. By coding such charactersas ordered,hy- zygomorphicclade. The re-weightinganalyses dif- potheses about evolutionarytrends may be incor- feredfrom each othermainly in the relationships poratedin estimatesof phylogeneticrelationships. of the sistertaxa to the zygomorphicclade. These In additionto ordering,characters were also re- differencesreflect biases toward differentcharac- 2000] EVANS ET AL.: COMMELINACEAE 679

t.m.c. )O 07 ~~g.m.c. not )Q 0 _- Xinfrequent I~ la

2-celled (e.g., Cuthbertia)

0 \ o0 (e R

0 %~~~~~

% % ~~~a

0~~~~~~~~~~~~~~~~ 0~~~~~~

6-cel~~~~~~~le I-e led

(Geogenanthus)~ ~ ~ ~ (e.g.,Comm elin ) '4 ~~~~~~L.s.c.(2) 2.

FIG. 3. Diagramsdepicting developmental pathways for the four stomatal types found in theCommelinaceae: sto- mata with two subsidiarycells, four subsidiary cells, and two typeswith six subsidiarycells. Abbreviations:t.m.c., terminalsubsidiary cell mother cell; g.m.c., guard cell mother cell; l.m.c., lateral subsidiary cell mother cell; t.s.c., terminal subsidiarycell; ls.c. (1), innermostlateral subsidiary cell; ls.c. (2), outermostlateral subsidiary cell (fromTomlinson 1966). 680 SYSTEMATICBOTANY [Volume 25

S t c x 4- x ? c -s- S -e c- So o -= g 0

0 ~ ~ 0

. ? < g 0 Qe 0 U CL

? E0~~~~ 3 'm 42n~~~~~~~~~~~~~c c

0-52

, * E1z s]o

-v 0 _ , i-

0I El. = 2000] EVANS ET AL.: COMMELINACEAE 681 ters,depending on how well theysupport the ini- plant material.This work was supportedin part by a tial (unordered)phylogeny (e.g., Character 5, num- GraduateStudent Research Grant from the American So- ber of flowersper cincinnus,as discussedabove). cietyof Plant Taxonomists,and by a ShortTerm Visitor It is unclearwhich of these values (CI, RI, or RC) Grantand GraduateStudent Fellowship from the Smith- sonianInstitution. will providea moreaccurate reflection of the suit- abilityof each characterin reconstructingphylog- eny,but theydo provideinformation about which LITERATURE CITED charactersare morehomoplasious and/or more in- ANDERSON, E. and R. E. WOODSON. 1935. The species of formativewith respect to a particulardata set and Tradescacmtiaindigenous to the United States. Contri- topology.It is evidentthat the numberof flowers butions from the Arnold Arboretum of Harvard Uni- per cincinnus,for example, does not contributeto versity9: 1-132. thephylogenetic interpretation provided by therest BININDA-EMONDS, 0. R. P., H. N. BRYANT, and A. P. Rus- of the data. Furthermore,if the data set as a whole SELL. 1998. Supraspecific taxa as terminals in cladistic is misleading,then re-weightingthe charactersa analysis; implicit assumptions of monophyly and a posteriorimay reinforcethe incorrectparts of the comparison of methods. Biological Journal of the Lin- phylogenyand reduce the supportfor some more nean Society64: 101-133. accurateareas. BERHAUT, J.1988. Pp. 130-131in FloreIllustree du Senegal: The Commelinaceaeare quite diversemorpho- Moniocotyledonieset Pteridophytes. Dakar: Clairafrique. BRENAN, J.P. M. 1960. Notes on African Commelinaceae: logically,and, as has been demonstratedin this II: The genus BiiforrestiaC. B. Cl. and a new related study,many of themorphological characters in the genus,Stanifieldiella Brenan. Kew Bulletin14: 280-286. familyare highlyhomoplasious (e.g., characters re- . 1961. Triceratella,a new genus of Commelinaceae lating to pollination).Many classificationsof the from southern Rhodesia. Kirkia 1: 14-19. familyhave been based in largepart on thesehom- . 1966. The classificationof Commelinaceae. Botan- oplasiouscharacters, thereby creating unnatural as- ical Journalof the Linnean Society 59: 349-370. semblages(an obvious exampleof thisis Brenan's CHASE, M. W, D. E. SOLTIS, R. G. OLMSTEAD, D. MORGAN, [1966] "Group I", a diverse and artificialgroup D. H. LES, B. D. MISHLER, M. R. DUVALL, R. A. PRICE, unitedmainly by inflorescencecharacters and a de- H. G. HILLS, Y-L. Qiu, K. A. KRON, J. H. RETTIG, E. hiscentfruit). Although relatively few anatomical CONTI, J. D. PALMER, J. R. MANHART, K. J. SYTSMA, H. charactershave been examinedin greatenough de- J. MICHAELS, W J. KRESS, K. G. KAROL, W D. CLARK, M. HEDREN, B. S. GAUT, R. K. JANSEN, K-J. KIM, C. F tail for a comprehensiveanatomy-based cladistic WIMPEE, J. F SMITH, G. R. FURNIER, S. H. STRAUSS, Q- analysis,anatomical characters (e.g., stomatal char- Y XIANG, G. M. PLUNKETT, P. S. SOLTIS, S. M. SWEN- acters;Fig. 3) appear to hold greatpromise for eval- SEN, S. E. WILLIAMS, P. A. GADEK, C. J. QUINN, L. E. uatingrelationships at thetribal and, possibly,sub- EGUIARTE, E. GOLENBERG, G. H. LEARN, S. W GRA- triballevels. HAM, S. C. H. BARRETT, S. DAYANANDAN, and V. A. ALBERT. 1993. Phylogenetics of seed plants: an anal- ACKNOWLEDGEMENTS. This paperrepresents a portion ysis of nucleotide sequences from the plastid gene of I Evans' Doctoralthesis in theUniversity of Wisconsin rbcL. Annals of the Missouri Botanical Garden 80: Departmentof Botany. We wish to thankcommittee mem- 528-580. bers Tom Givnish,Bob Kowal, David Spooner,and Ted , D. W STEVENSON, P. WILKIN, and P. J. RUDALL. Garland forreading an earlierdraft of the manuscript. 1995. Monocot systematics: a combined analysis. Pp. Additionally,we thankKandis Elliot for preparation of the 685-730 in Moniocotyledonis:systemnatics anid evolutioni, figures.The UnitedStates National Herbarium and Smith- eds. P. J. Rudall, P. J. Cribb, D. F Cutler, and C. J. sonianInstitution Botany Research Greenhouses provided Humphries. Kew: Royal Botanic Garden.

FIG. 4. Distributionof characterson representativetree (see Fig. 1) fromthe unordered analysis of 47 morphological charactersin Commelinaceae.Left: Stomatal structure. This anatomicalcharacter may be usefulfor higher level rela- tionshipswithin the family. Although it is homoplasioushere, its distributionis congruentwith recent molecular data (Evans 1995,in prep.).Note thatthe four-celledstomatal type is one step shorterin equally mostparsimonious trees thatunite Palisota with the clade containingCochliosteta, Titnoatia, and theundescribed genus. The numberof steps for theother stomatal types is notaltered in otherequally most parsimonious trees. Right: Number of flowers per cincinnus. Althoughthis character possesses a relativelyhigh consistency index (CI = 0.75),it does notunite any clades and thus has a retentionindex of zero. The distributionof thischaracter is not alteredin any othermost parsimonious trees. 682 SYSTEMATICBOTANY [Volume 25

CLARK,W. D., B. S. GAUT,M. R. DUVALL,and M. T. CLEGG. A.E.T.EA.T.,Twelfth Plenary Meeting, 4-10 Septem- 1993.Phylogenetic relationships of the Bromeliflorae- ber,1988. Hamburg. P. 58. Commeliniflorae-Zingiberifloraecomplex of mono- . 1991.The morphologyand taxonomyof Anieilema cotsbased on rbcLsequence comparisons. Annals of R. Brown (Commelinaceae).Smithsonian Contribu- theMissouri Botanical Garden 80: 987-998. tionsto Botany76: 1-166. CLARKE, C. B. 1881.Commelinaceae. Pp. 113-324in Mon- and D. R. HUNT. 1991. The classificationof the ographiaephaneroganiarinm, 3, eds. A. De Candolleand Commelinaceae.Taxon 40: 19-31. C. De Candolle.Paris: G. Masson. and K. E. INMAN. 1996.Leaf anatomyof theAfri- CONRAN, J.G. 1994. Tapheocarpa(Commelinaceae), a new can generaof Commelinaceae: Anthericopsis and Muir- Australiangenus with hypogeousfruits. Australian dannia.Pp. 464-471in Thebiodiversity ofAfrican plants, SystematicBotany 7: 585-589. eds. L. J.G. van der Maesen, X. M. van der Burgt, CRONQUIST, A. 1981.An integratedsystem of classification and J.M. van Madenbachde Rooy.Procedings of the of floweringplants. New York:Columbia University XlVth A.E.T.EA.T. Congress,22-27 August, 1994, Press. Wageningen,The Netherlands.London: KluwerAc- DAHLGREN, R., and H. T. CLIFFORD. 1982. Theinonocotyle- ademicPublishers. dons:a comparativestudy. New York:Academic Press. FARRIS,J. S. 1989. The retentionindex and rescaledcon- , E N. RASMUSSEN, and P. E YEO. 1985.The families sistencyindex. Cladistics 5: 417-419. of the :structure, function, and tax- FORMAN,L. L. 1962. Aetheolirion,a new genus of Com- onomy.Berlin: Springer Verlag. melinaceaefrom Thailand, with notes on allied gen- DONOGHUE, M. J.1994. Progress and prospectsin recon- era. Kew Bulletin16: 209-221. structingplant phylogeny. Annals of the Missouri Bo- GIVNISH,T. J.,T. M. EVANS,and K. J.SYTSMA. 1995. The tanicalGarden 81: 405-418. Commelinales:an extremeexample of convergent DUVALL, M. R., M. T. CLEGG, M. W CHASE, W. D. CLARK, evolutionin monocots.American Journal of Botany W. J.KRESS, H. G. HILLS, L. E. EGUIARTE, J. F SMITH, 82(6,Supplement): 132. B. S. GAUT, E. A. ZIMMER, and G. H. LEARN. 1993. 1 , J.C. PIRES and K. J.SYTSMA. 1999. Poly- Phylogenetichypotheses for the monocotyledons con- phyly and convergentmorphological evolution in structedfrom rbcL sequence data. Annalsof theMis- Commelinalesand Commelinidae:evidence from souriBotanical Garden 80: 607-619. rbcLsequence data. Molecular Phylogenetics and Evo- EVANS, T. M. 1995.A phylogeneticanalysis of the Com- lution12: 360-385. melinaceaebased on morphologicaland molecular HONG, D. 1974. Revisio CommelinacearumSinacarum. data. Ph.D. thesis,University of Wisconsin, Madison. Acta PhytotaxonomicaSinica 12: 459-488. I R. B. FADEN, and K. J.SYTSMA. 2000. Homoplasy GRAHAM,S. W, J.R. KOHN,B. R. MORTON,J. E. ECKEN- in theCommelinaceae: a comparisonof different clas- WALDER,and S. C. H. BARRETT.1998. Phylogenetic ses of morphologicalcharacters. Pp. 557-566in Pro- congruenceand discordanceamong one morpholog- ceedingsof theSecond International Conference on the ical and threemolecular data sets fromPontederi- ComparativeBiology of the Monocots, Sydney, September aceae. SystematicBiology 47: 545-567. 1998, eds. K. L. Wilson and D. A. Morrison.Mel- HUNT,D. R. 1975.The reunionof Setcreasea and Separotheca bourne:CSIRO. with Tradescantia.American Commelinaceae: I. Kew FADEN, R. B. 1975. A biosystematicstudy of the genus Bulletin30: 443-458. AneilemnaR. Br. (Commelinaceae). Ph.D. thesis,Wash- . 1978. Threenew generain Commelinaceae.Kew ingtonUniversity, St. Louis. Bulletin33: 331-334. . 1977. The genus RhopalephoraHaask. (Commeli- . 1980.Sections and seriesin Tradescantia.American naceae). Phytologia37: 479-481. Commelinaceae:IX. Kew Bulletin35: 437-442. . 1980. The taxonomyand nomenclatureof some . 1985.A revisionof GibasisRafin.: American Com- Asiatic species of Murdannia(Commelinaceae): the melinaceaeXII. Kew Bulletin41: 107-129. identityof Commelina medica Lour. and Comninelinatuib- . 1986a. Camnpelia,Rheo, and Zebrinaunited with erosaLour. Taxon 29: 71-83. Tradescantia.American Commelinaceae:XIII. Kew . 1983a. Phytogeographyof AfricanCommelina- Bulletin41: 401-405. ceae. Bothalia14: 553-557. . 1986b. Amplificationof CallisiaLoefl. American . 1983b.Floral dimorphism, dimorphism and Commelinaceae:IX. Kew Bulletin41: 407-412. floralbiology in a WestAfrican Commelinaceae, Pal- . 1993. Commelinaceae.Pp. 130-201in FloraNovo- isotahi-suta (Thunb.) K. Schum.American Journal of Galiciana,Vol. 13, Limnocharitaceae to Typhaceae,ed. Botany70: 113. R. McVaugh.Ann Arbor:University of Michigan Her- . 1985. Commelinaceae.Pp. 381-387in Thefamilies barium. of themonocotyledons, eds. R. M. T. Dahlgren,H. T. HUTCHINSON, J. 1959. Commelinaceae,Cartonemataceae, Clifford,and P. F Yeo. Berlin:Springer. in Familiesof flozvering plants, Second Edition.Volume . 1988.Reproductive biology of PalisotaReichb., an 2, pp. 561-566.Oxford: Oxford University Press. Africanendemic genus of Commelinaceae. Abstracts, JONES, K. and C. JOPLINH. 1972. Chromosomesand the 2000] EVANS ET AL.: COMMELINACEAE 683

classificationof the Commelinaceae. Botanical Journal SWOFFORD,D. L. 1999.Phylogenetic analysis usinig parsitmoniy of the LinneanSociety 65: 129-162. (* atndother miethods) Version 4.0. Sunderland,Massa- KLUGE, A. G. and J.S. FARRIS. 1969.Quantitative phyletics chusetts:Sinauer Associates. and theevolution of anurans. Systematic Zoology 18: TOMLINSON,P. B. 1964.Notes on theanatomy of Triceratella 1-32. (Commelinaceae).Kirkia 4: 207-212. LEE, R. E. 1961.Pollen dimorphism in Tripoganidragrandi- . 1966.Anatomical data in theclassification of Com- flora.Baileya 9: 53-56. melinaceae.Botanical Journal of the LinneanSociety LINDER, H. P. and E. A. KELLOGG. 1995.Phylogenetic pat- 59: 371-395. ternsin the Commelinidclade. Pp. 473-496in Mono- . 1969.Commelinaceae. Pp. 12-63 in Aniatomyof the cotyledons:systeinatics and evolution, eds. P. J.Rudall, P. nmoniocotyledonis,3. Ed. C. R. Metcalfe,Oxford: Clar- J.Cribb, D. F Cutler,and C. J.Humphries. Kew: Royal endonPress. BotanicGarden. WOODSON, R. E. 1942.Commentary on theNorth Ameri- MADDISON, D. R. 1991. The discoveryand importanceof can generaof Commelinaceae. Annals of the Missouri multipleislands of most-parsimonious trees. System- BotanicalGarden 29: 141-154. aticZoology 40: 315-348. MADDISON, W. P. and D. R. MADDISON. 1992. MacClade: APPENDIX 1 Analysisof phylogeny and evolution.Version 3.0. Sun- derland,Massachusetts: Sinauer Associates. Descriptionand rationalefor characters and cod- MATUDA, E. 1955.Las CommelinaceaeMexicanas. Sobre- ing used in this analysis. A briefjustification for tirode los Anales del Institutode Biologia26: 303- 432. the rationalebehind the orderingscheme for char- MOORE, H. E. 1954. The Seersuckerplant-Geogenanthlus acters 35 (fruitlocule number), 37 (seed numberper undatus.Baileya 2: 41-45. dorsallocule), and 38 (seed numberper ventralloc- . 1958.Callisia eleganis, a new species,with notes on ule) is included.Although the characterswere not thegenus. Baileya 6: 135-147. polarized a priori,they were scored so thatthose MORTON, J.K. 1966.A revisionof the genus AnZeilem1IaR. statesthat have traditionallybeen believed to be Brown (Commelinaceae)with a cytotaxonomicac- primitivein the family(based on observedtrends countof the WestAfrican species. Botanical Journal in monocotsas a whole) were givena stateof "0", of the LinneanSociety 59: 431-479. and those statesthat have been consideredmore NIXON, K. C. and J.I. DAVIS. 1991.Polymorphic taxa, miss- derivedwere scoredas "1" or higher. ing values and cladisticanalysis. Cladistics 7: 233- 241. CHARACTER 1. INFLORESCENCEPEDUNCLE OBERMEYER, A. and R. FADEN. 1985. Commelinaceae.Pp. DEVELOPMENT. The inflorescencemay be pe- 23-60 in Floraof souithern Africa, Vol. 4, Pt. 2, ed. 0. A. dunculateor sessile to subsessile.The inflorescenc- Leistner.Cape Town:Government Printer. es ofHaemodorum and Heterantheraare pedunculate. OLMSTEAD, R. G, B. BREMER, K. M. SCOTT, and J.D. PALM- This characteris dividedinto two states:0 = inflo- ER. 1993.A molecularsystematic analysis of theAs- rescencepeduncle well developed; 1 = inflores- teridaesensu lato based on rbcLsequences. Annals of cencepeduncle absent or nearlyso. the MissouriBotanical Garden 80: 700-722. CHARACTER 2. CINCINNUS BRACT DEVEL- and J.D. PALMER. 1994.Chloroplast DNA system- OPMENT. The bract of the cincinnusmay be atics:a reviewof methodsand data analysis.Amer- large,small or lacking.If present,it maybe persis- ican Journalof Botany81: 1205-1224. tentor cauducous.In Commelina,the cincinnus bract PICHON, M. 1946. Sur les Commelinacees.Notulae syste- maticae.Museum National d'histoire Naturelle. Pha- enclosesthe inflorescenceand is spathe-like.In all nerogamie.Paris 12: 217-242. threegenera of the subtribeStreptoliriinae sensu READ, R. W 1965.Cochliosteina velutinuin (Commelinaceae): Faden and Hunt (Streptolirion,Spatholirion, and Ae- a new speciesfrom Colombia. Baileya 13: 8-15. theolirion),bracts may be large and foliaceous, SIMPSON, M. G. 1988.Embryological development of Lach- small, or lacking.In Stanfieldiellaand Floscopa,a nanthescaroliniana (Haemodoraceae). American Jour- clear gradationfrom large foliaceousbracts at the nal of Botany75: 1394-1408. lowerregion of theinflorescence, to smallbracts at . 1990.Phylogeny and classificationof the Haemo- thedistal regions often can be foundwithin a single doraceae.Annals of the MissouriBotanical Garden inflorescence. 77: 722-784. The developmentof the cincinnusbract is vari- STEVENSON, D. W and H. LOCONTE. 1995. Cladisticanal- ysis of monocotfamilies. Pp. 543-578in Monocotyle- able in theoutgroups. The bractsin Haemodorumare dons:systematics anid evolutioni, eds. P. J. Rudall, P. J. small and cauducousor absent,and in Heteranthera Cribb,D. F. Cutler,and C. J.Humphries. Kew: Royal theyare large and foliaceous. BotanicGarden. This characteris dividedinto four different char- 684 SYSTEMATICBOTANY [Volume 25 acterstates: 0 = cincinnusbracts small and persis- cinnus,implying the presenceof more than one tent;1 = cincinnusbracts small and cauducousor flowerper primaryinflorescence. Belosynapsis, Un- absent;2 = cincinnusbracts all largeand foliaceous describedGenus, Sauvallea,Murdannia, and Aneile- or spathaceous;3 = some cincinnusbracts large ma,as well as Cartonema,each containspecies in- and foliaceousand some small or lacking;4 = cin- terpretedhere as havingcincinni that, through re- cinnusbracts grading from large and foliaceousat duction,consist of a singleflower. theproximal region of theinflorescence to smallor Severalflowers per cincinnusare foundin thein- lackingat the distalregions. florescencesof Haemodorumand Heteranthera. As it is difficultto makeany reliable assumptions This characteris divided into two states:0 = regardingthe orderof transformationof thischar- sometimesmore than one flowerper cincinnus;1 acter,it is leftunordered in the second analysis. = always onlyone flowerper cincinnus. CHARACTER 3. CINCINNUS PEDUNCLE DE- CHARACTER 6. TYPES PRESENT. VELOPMENT. The individualcincinni may be ei- While Commelinaceaetypically have perfectflow- therpedunculate or sessile (or nearlyso). In some ers,two differenttypes of andromonoecy have aris- genera (see Character7) the cincinnusis reduced en withinthe family. In sevengenera (Palisota, Tin- to a singleflower, making it problematicto distin- antia,Tricarpelema, Polyspatha, Aneilema, Rhopalephora, guish between the cincinnuspeduncle and the and some speciesof Dichorisandra),perfect and sta- flowerpedicel (Character7). In these cases, it is minateflowers may be foundin randompositions possibleto distinguishbetween the two structures on the same plant.Members of the subtribeStrep- by theirpositional relationship to thefloral bracte- toliriinae,as well as Pseudoparisand Commelina,pos- ole (see Character4). Whena floralbracteole is pre- sess both perfectand staminateflowers that are sent,any stalkthat is proximalto thebracteole (i.e., largelyseparate from one another.Haemodorum and betweenthe bracteole and themain axis) is consid- Heterantheraproduce only perfect flowers. ered to be the cincinnuspeduncle, while any stalk This characteris divided into threestates: 0 = thatis distal to the bracteoleis interpretedas the flowersall perfect;1 = flowersperfect and stami- flowerpedicel. nate,distribution random; 2 = flowersperfect and The cincinniof Haemodorumand Heterantheraare staminate,types largely separate. distinctlypedunculate. It maybe a safeassumption that the perfect flow- This characteris dividedinto two states:0 = cin- er type is primitivein the family,but thereis no cinnuspeduncle well developed;1 = cincinnuspe- way to discern which type of andromonoecyis duncleabsent or nearlyso. morederived than the other,or whetherthey both CHARACTER 4. BRACTEOLEDEVELOPMENT. arose independentlyfrom the perfectflower state. Each flowerin the cincinnusmay be opposed by a Thereforethis character is leftunordered in thesec- small bracteole.Most commonly,the bracteoleis ond analysis. persistentand not perfoliate,but in two genera, CHARACTER 7. PEDICEL PRESENCE/AB- Aneilemaand Rhopalephora,as well as in some Mur- SENCE. Flowersin the Commelinaceaemay be ei- danniaspecies, it is usuallyperfoliate. Several gen- therpedicillate or sessile to subsessile (see Char- era (Palisota,Streptolirion, Spatholirion, Siderasis, Coch- acter3 fora discussionabout the identificationof liostema,Sauvallea, Murdannia, Tricarpelema, Dictyos- the flowerpedicel and the cincinnuspeduncle in permum,and Commelina)contain species in which one-floweredcincinni). All ofthe genera in thetribe the bracteoleis cauducousor absent.The bracteole Commelineaepossess pedicellateflowers, whereas is absentin Heterantheraand persistent,not perfo- approximatelyhalf of the generain theTradescan- liatein Haemodorum. tieae,as well as the genus Triceratella,have sessile This characteris divided into threestates: 0 = or subsessile flowers.The flowersin Haemodorum bracteolepersistent, not perfoliate;1 = bracteole are pedicellateand in Heterantheraare sessile. persistent,perfoliate; 2 = bracteolecauducous or This characteris divided into two states:0 = absent. flowerspedicellate; 1 = flowerssessile or nearlyso. No clearlyordered series is apparent for this CHARACTER 8. COLOR. In mostgen- character,so it is leftunordered in thesecond anal- era, membersof the outerperianth whorl are dis- ysis. tinctlysepal-like, although they may be slightlycol- CHARACTER 5. NUMBER OF FLOWERS PER ored. The sepals of Palisotaand thethree genera of CINCINNUS. The basic unit of the inflorescence the subtribeStreptoliriinae are -likeand col- in the Commelinaceaeis thehelicoid cyme, or cin- ored thesame as thepetals. This stateis also found 2000] EVANS ET AL.: COMMELINACEAE 685

in some species of Porandraand Amischotolype,and Os has arisen(almost certainly independently) in Pollia as well. Both outgroupgenera produce tepals,so the "sepals" were scoredas petal-like. = This characteris dividedinto two states:0 out- QASO er perianthwhorl distinctlysepal-like (although sometimescolored); 1 = outerperianth whorl sim- 0 ilar to thepetals. 3-- CHARACTER 9. SEPAL (OUTER PERIANTH WHORL) WIDTH. While sepal width has not been used as a taxonomiccharacter in muchof the family,it has been usefulin definingthe subtribe Streptoliriinae.In mostgenera in thefamily, the se- pals are narrowerthan the inner (or lateral)petals. In the Streptoliriinae,as well as in some species of Callisiaand Floscopa,the sepals are noticeablywider than the innerpetals. Additionally,several genera possess sepals thatare nearlyequal in widthto the innerpetals. Both outgroups produce tepals, so the "petals" and "sepals" were scored as equal in Inflorescence Axis width. This characteris divided into threestates: 0 FIG. 5. Floraldiagram illustrating the relative position sepals narrowerthan the innerpetals; 1 = sepals of eac floralunit in the Commelinaceae.OS = outerse- equal in widthto theinner petals; 2 = sepals wider pal; IS = innersepal; OP = outerpetal; IP = innerpetal; = = thanthe inner petals. OAS onterantesepalous stamen; IAS innerantese- OAP = outer IAP = CHARACTER 10. PETAL SIZE AND FORM. palous stamen; antepetalousstamen; innerantepetalous stamen. The "outer" and "inner"po- The corollain the Commelinaceaeis composedof sitionsrefer to therelative location of ea structreoin the threepetals and may be eitheractinomorphic or developingbud. zygomorphic.The zygomorphicstate is expressed throughone or both of the followingfeatures: 1) enlargementor reductionin size of the outerpetal slight (particularly when compared to the strongly (the terms"inner" and "outer" are meantto rep- zygomorphic corollas found in the Commelineae), resentthe positions of thepetals in thebud, where or is due primarilyto the androecium, these genera the outerpetal is the singlepetal thatslightly en- are treated here as having actinomorphic corollas. velops the two lateral,or innerpetals [Fig.5]; they The tepals in Haemodorumand Heterantheraare all do notnecessarily correspond to theposition of the equal. petals relativeto themain axis ofthe inflorescence, This characteris divided into two states: 0 = pet- as theflower may sometimes be rotated60 degrees; als all equal or subequal (corolla mostly actino- Faden1975); or 2) shape ofthe outer petal (theouter morphic);1 = one petal stronglydifferentiated petal may be saccate and/or stronglyreflexed in fromthe othertwo (corollazygomorphic). the zygomorphiccorollas). A zygomorphiccorolla CHARACTER 11. PETAL FUSION. The petals is foundprimarily in some generaof the tribe Com- maybe eitherfree or fused,as inCyaom otis, Coleoty- melineae(Floscopa, Buforrestia, Tricarpelema [in some pe, Weldenia,and some species of Tradescantia.The species], Polyspatha,Aneilema, Rhopalephora, and degree of fusion is variable, ranging from the rel- Commelina).Strongly zygomorphic flowers are rare atively short corolla tubes in Cyanotisto the long in tribeTradescantieae but occasionallyoccur (e.g., corolla tube in Weldenia. Tinantiaanomala [Torrey] C. B. Clarke).It shouldbe The tepalsin Haemodorumare distinctand in Het- noted,however, that several genera in thetribe Tra- erantheraare connate. Because of the uncertaintyof descantieae(e.g., Cochliostema) express a slightten- homology between petal connation and tepal con- dencytoward zygomorphic corollas (the outer petal nation, this character was scored as "unknown"' in is subequal to the inner petals), particularlyin Heteranthera.This characteris divided into two flowersthat are orientedhorizontally instead of states:0 = petalsdistinct; 1 = petalsat leastbasally vertically.Because zygomorphyin these generais connate. 686 SYSTEMATICBOTANY [Volume 25

CHARACTER 12. PETAL FORM. The shape of character);and 2) it allows forthe consistentrec- thepetals is variablein thatsome specieshave dis- ognitionof homologousstamens, regardless of the tinctlyclawed petals and othersdo not.Because the orientationof theflower to themain axis of thein- outerpetal may evolve independentlyof the two florescence(in some genera,the entire flower is ro- innerpetals, it is possible to have a situationin tated60 degrees). whichtwo ofthe petals are clawed and thethird is CHARACTERS 14-17. STAMEN PRESENCE/ not (Faden 1991). The petals in Haemodorumand ABSENCE. As was noted above, the basic (and Heterantheraare not clawed. putativelyprimitive) form of the androeciumin This characteris dividedinto two states:0 = pet- monocotsis the presenceof two whorls of three als not clawed; 1 = at least one petal clawed. stamenseach. One, two,or all threestamens may CHARACTER 13. PETAL FRINGING. Al- be absentor vestigialin eitherwhorl. For example, thoughthe petal marginis usually not fringedin the outerantesepalous stamen (OAS) is absentin the Commelinaceae,fringed margins are foundin Dictyospermum,as well as some species of Palisota, threegenera within the subtribeDichorisandrinae Dichorisandra,Floscopa, Murdannia, Pollia, Aneilema, (Geogenanthus,Undescribed Genus, and Cochlioste- and Rhopalephora.Both outgroup genera lack the an- ma). The petals are not fringedin eitheroutgroup tesepalouswhorl of . genus. Each evolutionarycomponent in theandroecium Thischaracter is dividedinto two states:0 = pet- is scoredfor its presence or absence:Character 14- al marginsentire; 1 = petal marginsfringed. outerantesepalous stamen; Character 15-inner an- Characters14 through33 representthe androe- tesepalousstamens; Character 16-outer antepetal- cial characters.Because these characterscomprise ous stamen;Character 17-inner antepetaloussta- such a large proportionof the those used in this mens.Each of thesecharacters is divided intotwo study,and because of the complexevolutionary states:0 = stamen(s)present; 1 = stamen(s)absent. trendsfound in the androecium,it is necessaryto CHARACTERS 18-21. STAMEN FERTILITY firstdiscuss the mannerin whichthese characters The stamensmay be eitherfertile or sterile.While were distinguishedfrom each otherbefore describ- mostof the sterile stamens produce no pollen,a few ing each one in detail. species possess antherodes(staminode anthers) The simplesttype of androeciumin the Com- that produce sterilepollen (e.g., the antepetalous melinaceaeis composedof two whorls of three sta- stamen in Tripogandragrandiflora (Donnell-Smith) menseach, all equal and fertile(Fig. 5). Stamensof Woodson[Lee 1961],the outer antepetalous stamen the outerwhorl are antesepalous,and thoseof the in Aneilemahockii DeWild. [Faden1991], and thein- innerwhorl are antepetalous.Within each whorl, ner antepetalousstamens in Palisota[Faden 1983b, one stamenmay evolveindependently of theother 1988]).It shouldbe noted thatalthough the sterile two. In orderto accuratelyscore the stamens,the anthersthat produce no pollen (or in whichno an- androeciumwas divided into fourdifferent com- theris produced)and thosethat do producesterile ponents.Each whorlwas divided intotwo units,a pollen were both scored the same (i.e., "sterile"), single outerstamen, and two innerstamens. The thehomology of thesedifferent types of sterilityis terms"outer" and "inner"here do not referto the suspect.The presenceof anthersthat produce ster- locationof the stamensrelative to the axis of the ile pollen is rare,however, and is almostcertainly inflorescence,but ratherto the sepal or petal in autapomorphicfor those taxa thatexpress it. If a frontof whichthey are attached.For example,the particularstamen or whorlis absentor vestigialin outerantesepalous stamen (the outer stamen of the a genus (see Characters14-17), then stamen fertil- outerwhorl) is the stamenthat is attachedin front ityis not applicable,and it was scored as "?". The of the single sepal thatslightly encloses the other antesepalouswhorl of stamensis absent,and the two sepals in bud. Likewise,the inner antesepalous antepetalouswhorl is fertilein Haemodorumand stamensarise in frontof the two innersepals in the Heteranthera. bud. The same conventionis used forthe antepe- Each unit of the androecium,then, is scoredfor talous stamens.Although it may appear unneces- staminalfertility: Character 18-outer antesepalous sarilycomplex, this systemfor naming the mem- stamen;Character 19-inner antesepalousstamens; bersof the androecium accomplishes two purposes: Character 20-outer antepetalous stamen; and 1) it divides the androecium into individually Character21-inner antepetalousstamens. These evolvingunits (as opposed to artificiallytreating charactersare divided into two states: 0 = sta- each whorl,or the entireandroecium, as a single men(s) fertile;1 = stamen(s)sterile. 2000] EVANS ET AL.: COMMELINACEAE 687

CHARACTERS 22-25. STAMINAL FILAMENT filamentsbetween whorls, it is usefulto distinguish BEARDING. The filamentsmay be eitherglabrous betweena "top" (sometimescalled "posterior"or or bearded,often with distinctlymoniliform hairs. "/posticous"in horizontalflowers) and a "bottom" Althoughseveral differenttypes and degrees of (or "anterior"or "anticous")of theflower (see Fig. beardingmay be found(even when comparingdif- 5). The "top" was earlierdefined as thehalf of the ferentfilaments from the same flower),information flowercontaining the outer antesepalous and inner is lackingregarding homology of thedifferent hair antepetalousstamens, while the "bottom" of the types.Therefore, the plantsare scoredon the sim- flowercontains the remainingthree. Because the ple basis of thepresence or absenceof any typeof outgroupgenera lack theantesepalous whorl, these hairs. characterswere not applicible,and were scored as The stamenfilaments of Haemodorumand Heter- antheraare all glabrous. Relativefilament length between whorls, then, is Each unit of the androeciumis scored forthe scored as two characters:Character 28-relative presence/absence of staminal hairs: Character 22- lengthof outerantesepalous (OAS) and inneran- outer antesepalousfilament; Character 23-inner tepetalous(IAP) filaments("top"); and Character antesepalousfilaments; Character 24-outer ante- 29-relative lengthof inner antesepalous (IAS) and petalous filament;and Character25-inner ante- outer antepetalous (OAP) filaments("bottom"). petalous filaments.These charactersare divided They are divided into threestates: 0 = filaments into two states:0 = staminalfilament(s) glabrous; equal; 1 = antesepalousfilament(s) longer, ante- 1 = staminalfilament(s) bearded. petalousfilament(s) shorter; 2 = antesepalousfila- CHARACTERS 26-27. RELATIVE FILAMENT ment(s)shorter, antepetalous filament(s) longer. LENGTH WITHIN A WHORL. A considerable There is no obvious evolutionarysequence amountof variationin staminallength is foundin among these characterstates, so theyare leftun- the Commelinaceae,particularly within the tribe orderedin the second analysis. Commelineae.While absolutefilament lengths are CHARACTER 30. FILAMENT ADNATION TO not informativedue to variationin overallflower THE COROLLA. Epipetalousfilaments are found size withinthe family,relative length among sta- in Coleotrypeand Weldenia,as well as some species minalfilaments is directlycomparable, even among of Tradescantia.The filamentsof Haemodorumare flowersof varying sizes. The lengthof each filament epipetalousand thoseof Heterantheraare free.This relativeto the otherswas determined(both within characteris divided into two states:0 = filaments a whorl and betweenwhorls; see also Characters free;1 = filamentsepipetalous. 28-29) by firstcomparing the lengthof the outer CHARACTER 31. ANDROECIAL SYMMETRY. filamentwith the inner filaments in thesame whorl, While the androecialcharacters above provide a and thencomparing each outerfilament with the fairlycomplete description of the variationfound two innerfilaments of the otherwhorl (Characters eitherwithin a single stamen(or pair of stamens 28-29). Thus, all of thelength variation may be in- when consideringthe lateralones) or betweendif- cludedwithout any unnecessary redundancy in the ferentindividual componentsof the androecium, charactercoding. theydo notcompletely describe the arrangement of The antepetalousstamens are equal in Haemodo- the androeciumas a whole.As was mentionedear- rumand Heteranthera. lier,the putativelyprimitive state for the androeci- The filamentlength comparisons within a whorl um in theCommelinaceae is a radiallysymmetrical are scored as follows:Character 26-relative fila- arrangementof six fertilestamens. Almost half of mentlength in the antesepalouswhorl; and Char- the genera,however, possess species with a zygo- acter27-relative filamentlength in the antepetal- morphic(bilaterally symmetrical) androecium. Ad- ous whorl.These charactersare divided intothree ditionally,the androecium in Cochliostemaand states:0 = filamentsequal; 1 = outerfilament lon- manyspecies of Murdannia is neitheractinomorphic ger, inner filamentsshorter; 2 = outer filament norperfectly zygomorphic, but asymmetric because shorter,inner filaments longer. one or moreof the stamens curve to one side ofthe There is no obvious evolutionarysequence flower.In many Murdanniaspecies, the stamens among these characterstates, so theyare leftun- show a mirror-imagesymmetry between flowers, orderedin the second analysis. similarto enantiostyly,but due to the curvatureof CHARACTERS 28-29. RELATIVE FILAMENT the stamensinstead of the style (Faden, unpubl. LENGTH BETWEENWHORLS. In comparingthe data). Althoughthe homology of the asymmetrical 688 SYSTEMATICBOTANY [Volume 25

androeciain these generais suspect,it cannotbe the ,which may have threeequal locules (ra- assumedthat they arose independently, particularly diallysymmetrical), reduction in size of one locule, fora familyin whichsuch heavy emphasis has been formingthree unequal locules (in Siderasis,Rhopa- placed on androecialcharacters. The androeciumis lephora,and some species of Coleotrype,Aneilema, actinomorphicin both outgroupgenera. This char- and Commelina),or eventhe complete loss ofone or acteris divided into threestates: 0 = androecium two locules (in Floscopaand Polyspatha,as well as actinomorphicat anthesis;1 = androeciumzygo- some species of Callisia,Aneilema, and Commelina). morphic(bilaterally symmetrical) at anthesis;2 Althoughit mightbe argued thatthere may be a androeciumasymmetrical at anthesis. correlationbetween the degree of zygomorphyof CHARACTER 32. ANTHER DEHISCENCE. otherparts of the flower(i.e., the corolla or thean- Antherdehiscence in theCommelinaceae is almost droecium)with the numberof locules in the fruit always throughlongitudinal slits. Porandra, Dichor- (therebymaking this characterredundant), it isandra,and some species of Coleotrypeexhibit por- should be noted thatflowers of some genera are icidal antherdehiscence through terminal pores or radiallysymmetrical for one partof theflower and slits.In Cyanotisanthers, dehiscence occurs basally. zygomorphicfor another. For example, both the co- Bothoutgroup genera exhibit terminal longitudinal rolla and the androeciumin Siderasisare regular, antherdehiscence. This characteris divided into while the fruitis composedof two largerand one threestates: 0 = antherdehiscence longitudinal; 1 smallerlocules, and Palisotahas a zygomorphican- = antherdehiscence poricidal, by terminalpores or droeciumand a berrycomposed of three equal loc- slits;2 = antherdehiscence poricidal basally. ules. It is apparentthat the evolutionof the zygo- There is no obvious evolutionarysequence morphicstate may occurindependently within dif- among these characterstates, so this characteris ferentparts of the flower. leftunordered in the second analysis. The fruitsof both outgroup genera contain three CHARACTER 33. ANTHER CONNECTIVE equal locules. WIDTH. The width of the connectiverelative to This fruitcharacter is dividedinto three different the anthersacs is variablein the Commelinaceae. states:0 = fruitconsisting of threeequal locules;1 In mostgenera, the connective is narrowerthan the = fruitconsisting of three unequal (to two)locules; anthersacs. Nine genera,however, possess species 2 = fruitconsisting of two (or occasionallyone) withbroad connectives. Both outgroup genera have locules.Because the statesin thischaracter form a narrowconnectives. This characteris divided into continuousseries, they were orderedin thesecond two states:0 = connectivenarrower than the anther analysisas follows:0 <- 1 X- 2. sacs; 1 = connectivebroader than the anthersacs. CHARACTER 36. SEED ARRANGEMENT. CHARACTER 34. STYLE CURVATURE. The Aboutone thirdof thegenera contain species with singlestyle in a flowermay be centralin theflower, biseriateseed arrangement,while the remaining or it may have a distinctcurve to one side. The generahave uniseriateseed arrangement. threegenera in subtribeColeotrypinae (Coleotrype, Haemodorumand Heterantherahave biseriateseed Porandra,and Amischotolype),as well as Gibasoides, arrangement. Tinantia,and some species of Aneilemnaand Coin- This characteris divided into two states:0 = melinahave curvedstyles. Additionally, some spe- seed arrangementuniseriate; 1 = seed arrangement cies ofMurdannia exhibit enantiostyly, in whichthe biseriate. directionof curvature of the style is predictable(i.e., CHARACTERS 37-38. SEED NUMBER PER thereare "left"and "righthanded" flowers;Faden, LOCULE. The locule facingthe outersepal (dor- unpubl.data). The stylein Haemodorumis regularly sal locule) evolvesindependently of the locules fac- curved (enantiostylous)and in Heterantherais ing the innersepals (ventrallocules). In Aneilema, straight.This characteris dividedinto three states: forexample, the dorsallocule, when developed,al- 0 = style always central;1 = style irregularly ways containsfewer seeds thaneach ventrallocule. curved;2 = styleregularly curved (enantiostylous). The fruitof Dictyospermum,in contrast, always con- CHARACTER 35. FRUIT LOCULE NUMBER. tainsa singleseed in each of thethree locules. Ad- The fruitin the Commelinaceaeis typicallya loc- ditionally,some generaproduce fruits with the pu- ulicidalcapsule (withthe two notableexceptions of tativelyprimitive state of more than two seeds in Palisota,which produces a berry,and Pollia,which each locule (e.g.,Cochliostema and Weldenia).Finally, produces an indehiscent,berry-like, crustaceous some fruitscontain all threelocules, but one or two fruit).Various degrees of zygomorphy are foundin of themdo not produceseeds. 2000] EVANS ET AL.: COMMELINACEAE 689

The fruitsin Haemnodorumconsist of three locules, morphic for Dichorisandra.Note, however,that each of which produces two seeds. The fruitsof Amischotolypehispida (the species used in thisanal- Heterantheraconsist of threelocules, each of which ysis) does produce arillateseeds. Both outgroup producesnumerous seeds. generaproduce exarillateseeds. This characteris The numberof seeds per locule,then, is treated divided into two states:0 = seeds exarillate;1 = as two independentcharacters: Character 37-seed seeds arillate. numberper dorsallocule; Character 38-seed num- CHARACTER 42. PTYXIS. Ptyxisrefers to the ber per ventrallocule. These charactersare divided foldingof the leaves in bud. In theCommelinaceae, intofour different states: 0 = morethan two seeds/ the developingleaf is typicallyin one of two posi- locule; 1 = two seeds/locule; 2 = one seed/ locule; tionsas it emergesfrom the bud: its marginsmay 3 = zero seeds/locule(if the locule is entirelyab- be rolledinto two distinctcurls on the upper sur- sent,then it is scoredas "unknown"). face,scroll-like (involute) or theymay be rolledinto Because the statesin this characterform a con- a singlecurl (convolute;or "supervolute"in Dahl- tinuousseries, they are orderedin thesecond anal- grenand Clifford1982). Occasionally they are sim- ysis as follows:0 <- 1 <- 2 <- 3 (note thatthese ply folded(conduplicate). Ptyxis in Haemodorumis numbersindicate the characterstate, not the actual convolute,and informationis unavailablefor Het- numberof seeds/locule). eranthera,so it is scoredas unknown.This character CHARACTER 39. EMBRYOTEGA POSITION. is divided into two states:0 = ptyxisconvolute or All theseeds in theCommelinaceae possess a small conduplicate;1 = ptyxisinvolute. embryotega,or operculum,a cap-like structure CHARACTER 43. STOMATAL STRUCTURE. over the embryo.The embryotegamay be dorsal, While most classificationsfor the Commelinaceae semi-dorsal,semi-lateral, lateral, or terminalac- have relied heavily upon traditionalmacro-mor- cordingto the genus. The outgroupsdo not pro- phologicalcharacters, attempts to incorporateana- an duce embryotega,so they are scored as un- tomicalcharacters have shownpromise (e.g., Pichon known. 1946; Tomlinson 1966, 1969; Faden and Inman This characteris dividedinto four states: 0 = em- 1996). However,genus-specific data formost ana- bryotegadorsal; 1 = embryotegasemi-dorsal to tomical charactersare not readily available.This semi-lateral;2 = embryotegalateral; 3 = embryote- studyincorporates four different anatomical char- ga terminal.Although it is easy to envisiona trans- acters(Characters 43-46). formationseries in which the positionof the em- Of the differenttypes of anatomicaldata that bryotegahas moved aroundthe seed (e.g.,starting have been studied, the most useful is stomatal fromthe dorsal position and movingthrough semi- structure.Four distinct types of stomatahave been dorsal and semi-lateralto the lateral),there are no describedin theCommelinaceae (Fig. 3; Tomlinson intermediatesbetween the terminalposition and 1966, 1969), the simplestof whichhave only two any of the others.Without further developmental informationto documentthe orderof transforma- subsidiarycells positionedlaterally (found in Tri- tion,this character is leftunordered in the second ceratella,some Cartonemaspecies, and both out- analysis. group genera).Most membersof the tribeTrades- CHARACTER 40. HILUM SHAPE. The funic- cantieaehave four-celledstomata, with two lateral ular scar is most commonlylinear, but it is punc- and two terminalsubsidiary cells. The stomataof tiformin Cartonema,Triceratella, Palisota, Belosynap- Streptolirion,Spatholirion, and Aetheolirion(subtribe sis, Weldenia,Callisia, and Pollia,as well as a few Streptoliriinae),as well as Geogenanthus,are six- scatteredspecies of othergenera. Haemodorum pro- celled,with four lateral and two terminalsubsidi- duces a punctiformhilum, and informationwas un- ary cells. In these threegenera, the terminalcells available for Heteranthera,so it is treatedas un- are relativelylarge, extending to the outeredge of known.This characteris divided intotwo states:0 thelateral cells. Finally, members of the tribe Com- = hilumpunctiform to elliptic;1 = hilumoblong melineaepossess stomatawith four lateral and two to linear. terminalsubsidiary cells, but theterminal cells are CHARACTER 41. ARIL. The genus Dichorisan- smaller,not extendingto theouter edge ofthe out- dra,as well as some species of Amischotolypeand ermostlateral cells. Stomata in Haemodorumand Porandraproduce arillate seeds. Depending on Heterantheraare two-celled.This characteris divid- whichspecies are chosento representAmischotolype ed into fourstates: 0 = stomatatwo-celled; 1 = and Porandra,this charactercould appear autapo- stomatafour-celled; 2 = stomatasix-celled, termi- 690 SYSTEMATICBOTANY [Volume 25 nal cellslarge; 3 = stomatasix-celled, terminal cells in Murdannia,but theyare abundantin Aneilema. small. Due to the difficultyof collectingdata on hair dis- AlthoughTomlinson (1966) presentsinformation tribution,as well as assessinghomology on such on the developmentof each of thesestomatal types data,however, this character is simplyscored for its (Fig. 3), he also mentionsthat the sequenceof cell presenceor absence. divisionsis variable.Until more about the devel- Glandularhairs are presentin some Haemodor- opmentof thestomatal apparatus is understood,it aceae and somePontederiaceae (Simpson 1990), but is difficultto make a priorihypotheses regarding theyare largelyrestricted to the outersurfaces of the orderof characterstate transformations, so this flowersand inflorescencebranches. As the homol- characteris unorderedin the second analysis. ogy ofthese hairs with the vegetative glandular mi- CHARACTER 44. HOOK HAIRS. Tomlinson crohairsis suspect, both outgroupgenera were (1966) describedtwo main categoriesof hairs for scored as lackingglandular microhairs. theCommelinaceae: "microhairs", which show lim- This characteris divided into two states:0 = ited variabilityin structure,and "macrohairs", glandularmicrohairs absent; 1 = glandularmicro- which are quite diverse.He recognizedeight dif- hairspresent. ferenttypes of macro-hairs,ranging from single- CHARACTER 46. RAPHIDE CANALS. Raph- celledpapillae to multicellularuniseriate hairs. One ide canals are presentin all generaof Commelina- particulartype of macro-hairwhich has been ex- ceae exceptCartonema. Indeed, the absence of raph- aminedmore closely by one of the authors(R. Fa- ide canals has been one of themost compelling ar- den, unpubl. data) is the "hook hair".These two- gumentsfor the separationof Cartonemainto its celled,hook shaped hairs have been observedon own family(Pichon 1946; Hutchinson 1959; Tomlin- the leaves of Polyspatha,Dictyospermum, Pollia, Rho- son 1966).Of the remaininggenera, only in Tricer- atellaare the raphide canals located next to the palephora,Aneilema, and some species of Commelina, veins,instead of being evenlydistributed through- and are notablyabsent in the rest of the family out theleaf blade. (Tomlinson1966; R. Faden, unpubl. data). Hook Raphide canals are absent in Haemodorumand hairsare absentin bothoutgroup genera. This char- Heteranthera. acteris dividedinto two states:0 = hook hairsab- This characteris divided into threestates: 0 = = sent;1 hook hairspresent. raphidecanals absent;1 = raphidecanals present CHARACTER 45. GLANDULAR MICRO- nextto the veins; 2 = raphidecanals present,not HAIRS. Three-celledglandular microhairs(or exclusivelynext to theveins. simply"microhairs" in Tomlinson1966) are pre- CHARACTER 47. CLOSED LEAF SHEATH. sent on the leaves of everygenus in the Commeli- All membersof the Commelinaceae possess distinct naceae exceptCartonema and Triceratella.The distri- closed sheathingleaf bases (Cronquist1981; Dahl- butionof thesehairs on the leaf,stem, and flower gren et al. 1985). Althoughboth outgroupgenera is quitevariable, and Tomlinsonhas suggestedthat possess a sheathingleaf base, it is open. This char- theirdistribution might be useful for diagnosing acter was divided into two states: 0 = closed genera.As an example,he points out thatfew or sheathingleaf bases absent;1 = closed sheathing no microhairsare foundon theadaxial leafsurface leafbases present. 2000] EVANS ET AL.: COMMELINACEAE 691

APPENDIX 2 Data matrixof morphologicalcharacters used in cladisticanalysis of Commelinaceae.For alignment purposes,characters that are polymorphicin a taxonare representedby a letter:A = 0/1; B = 0 /2; C = 1/2; D = 0/1/2; E = 2/3.

11111111112222222222333333333344444444 Character 12345678901234567890123456789012345678901234567 Cartonema OO1O1O100000000000000OOOOOOOOOOOOOOOCCOOOOOOOO1 Triceratella 02000010?00010000000011110000000000100000?00011 Palisota 00020101100000000110111002121010000000000110121 Streptolirion 03020211200000000000011110000000100011110120121 Spatholirion 03020211200000000000011110000000000100010020121 Aetheolirion 0300021120000001000?010?02?0?010000100210?20121 Cyanotis ?2100010001100000000011110000002000011300010121 Belosynapsis ABAOAOAOOOOOOOOOOOO0011110000000000011300010121 Coleotrype 10100010001100000000011110000100010OCC210010121 Porandra 1010OOlA10000000000011110000001010011211110121 Amischotolype A010OOlA10000000000011110000000010011211110121 Dichorisandra AOOOOAOOOOOOOOOOOOOOOOOOOOOO0110001OOlA1110121 Siderasis OOOBOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO11EO??0110121 Geogenanthus 0010000000001001000?010?12?0?010100100210120121 Undescribed Genus OOOOAOOOOOOO1OA10011OAA?Al?2?01000?021210110121 Cochliostema 020200000000100000110110112220201?0100010110121 Thyrsanthemum 00000000000000000000011110000000000011110010121 Gibasoides 00000000000000000000011110000000110011210010121 Tinantia 00000100000000000000011012100010010000210110121 Elasis 0?0000000000000000000111100000000?0011010010121 Matudanthus 1?1?00000000000000000111100000000?0011???010121 Weldenia 121000000011000000000000000001000?0000C00010121 Gibasis 00000000000000000000011110000000?00011000110121 Tradescantia 02100000000000000000011110000000100011010010121 Callisia 00100010000000000000000000000000100011000010121 Tripogandra 00100000000000000000011000021010A00011010010121 Sauvallea 12121010000000000000011110000000??????????10121 Stanfieldiella ?4000000100000000000000000000000000000010130121 Floscopa 040000002100000000000000021000100020?2010130121 Buforrestia 021?0000010000000000000002002010000000210130121 Murdannia 000000000000000000011111100110000BOO11B00030121 Anthericopsis 00100000000000000001100000011000000000210030121 Tricarpelema 0102010001OA00000100100002100010AOOODDC10030121 Pseudoparis 000002000000001001?010?00?10?010000000?10?30121 Polyspatha 021?0100010100000100100002100010002??2110131121 Dictyospermum 01020000000001001?001?00????1010000?22210131121 Pollia 00000000000100000000000000000000000100000131121 Aneilema 000101000101OA000100100002121010111021210131121 Rhopalephora 000101000001OA000100100002122010000?22210131121 Commelina 02020200010100000100100002101010011021210131121 Haemodorum 00000001100001100??00??00?0??100020111?00000000 Heteranthera 0202001110?001100??00??00?0??000000100??0?00000