Dioecy and Its Correlates in the Flowering Plants

AmericanJournal of Botany 82(5): 596-606. 1995.

DIOECY AND ITS CORRELATES IN THE FLOWERING PLANTS

SUSANNE S. RENNER2 AND ROBERT E. RICKLEFS Instituteof SystematicBotany, University of Mainz, Bentzel-Weg 2, D-55099 Mainz,Germany; and Departmentof Biology, University of Pennsylvania, Philadelphia, Pennsylvania 19104-6018

Considerableeffort has beenspent documenting correlations between dioecy and variousecological and morphological traitsfor the purpose of testing hypotheses about conditions that favor dioecy. The dataanalyzed in thesestudies, with few exceptions,come from local floras,within which it was possibleto contrastthe subsets of dioecious and nondioecioustaxa withregard to thetraits in question.However, if there is a strongphylogenetic component to thepresence or absenceof dioecy,regional sampling may result in spuriousassociations. Here, we reportresults of a categoricalmultivariate analysis ofthe strengths ofvarious associations of dioecy with other traits over all floweringplants. Families were scored for presence ofabsence of monoecy or dioecy,systematic position, numbers of speciesand genera,growth forms, modes of pollination and dispersal,geographic distribution, and trophicstatus. Seven percent of angiospermgenera (959 of 13,500)contain at least some dioeciousspecies, and ;6% of angiospermspecies (14,620 of 240,000) are dioecious.The mostconsistent associationsin thedata setrelate the presence of dioecyto monoecy,wind or waterpollination, and climbinggrowth. At boththe family and thegenus level, insect pollination is underrepresentedamong dioecious plants. At thefamily level, a positivecorrelation between dioecy and woodygrowth results primarily from the association between dioecy and climbing growth(whether woody or herbaceous)because neither the tree nor the shrub growth forms alone are consistently correlated witha family'stendency to includedioecious members. Dioecy appears to have evolvedmost frequently via monoecy, perhapsthrough divergent adjustments of floralsex ratiosbetween individual plants. Monoecy itself is relatedto abiotic pollinationand climbinggrowth as revealedby multivariateanalysis. Dioecy and monoecyare concentratedin theless advancedsuperorders of Thorne (1992) and subclassesof Cronquist (1988). The frequencyof dioecy found in a local flora thereforereflects the level of dioecy in itsparticular pool offamilies as muchas, or morethan, local selectivefactors. The positiveassociations of dioecy with abiotic pollination and monoecyare related to floraldevelopmental and morphological attributes,as is thenegative association with bird and bat pollination;the positive association of dioecywith climbing growthis tentativelyexplained in termsof differential selection for optimal resource allocation to sexualfunction. If rapid upwardgrowth is at a premiumin climbersand iffruit set at leasttemporarily inhibits growth or requiresthe production ofthicker, more slowly growing stems to supportheavy fruits, it mightbe advantageousto postponefemaleness. Ifthe effect is strong,this may favor male plants.

Over thepast 15 yrconsiderable effort has been spent fruitsand frugivore-dispersedseeds (Bawa, 1980; Giv- documentingcorrelations between dioecy and various nish,1980; Steiner, 1988; Ibarra-Manriquez and Oyama, morphologicaland ecologicaltraits for the purposeof 1992;but see Fox, 1985;Muenchow, 1987; Thomson and testinghypotheses about conditionsthat favor dioecy. Brunet,1990), small flowers (Bawa and Opler,1975; Fox, Ecologicalcorrelates attracted attention in particularbe- 1985;Ibarra-Manriquez and Oyama,1992), white to yel- cause dioecyhas provendifficult to explainsimply as an low or greenishflowers (Bawa and Opler,1975; Muen- outbreedingmechanism in taxa lackingself-incompati- chow,1987), water pollination (Cox, 1988),and so-called bility(for discussionsof this issue, see Charlesworth unspecializedinsect pollinators (Bawa and Opler,1975; and Charlesworth,1978; Thomsonand Barrett,1981; Bawa, 1994; contraRenner and Feil, 1993). Charlesworth,1985; Thomson and Brunet,1990). At dif- While older reportstested the associationsbetween ferentspatial and hierarchicalscales, dioecyhas been dioecyand particularattributes separately, more recent associatedwith, among other attributes, wind pollination oneshave used multivariateanalysis, taking into account (Darwin, 1876; Kerner,1895; Sporne,1949; Stebbins, correlationsamong some of the explanatoryvariables 1951; Kaplanand Mulcahy,197 1; Freeman,Harper, and themselves(Fox, 1985; Muenchow,1987; Steiner,1988; Ostler,1980; Muenchow, 1987; Steiner,1988), monoecy Ibarra-Manriquezand Oyama,1992). Some authors have (Yampolskyand Yampolsky,1922; Lewis, 1942; Wes- attemptedto incorporatephylogenetic information in tergaard,1958; McComb,1966; Lloyd,1982), perennial analysesof the association of dioecy with other traits by growthand lignified secondary xylem, or woodiness (Dar- lookingat thespecies, genus, and familylevels separately win, 1876; Stebbins,1950; Baker, 1959; Croat, 1979; or by restrictinganalyses to monophyleticgroups (Hart, Conn, Wentworth,and Blum, 1980; Freeman,Harper, 1985; Donoghue,1989; Cox, 1990, 1993; Lahav-Ginott and Ostler,1980; Fox, 1985;but see Steiner,1988), fleshy and Cronk,1993). Thedata analyzed in these studies, with a fewexceptions 1 Manuscriptreceived 19 May 1994;revision accepted 23 November suchas Cox's (1990, 1993)phylogenetic analyses, all come 1994. fromlocal floras.Such locallycircumscribed data allow The authorsthank D. Charlesworth,P. Cox, P. Endress,C. Freeman, one to contrastthe subsets of dioecious and nondioecious S. Heard,J. Kohn,-J. Thomson,and K. Bawa forcomments on the taxawith regard to particularecological and morpholog- manuscript;and P. Berry(Onagraceae), H.-J. Esser (Euphorbiaceae), S. Knapp(Solanaceae), J. Kuijt (parasitic angiosperms), C. Ott(Menisper- ical traitsof interest.For example,the relationships be- maceae),and J.Rohwer (Lauraceae) for information on dioecyin par- tweena particularfloral color or size and dioecycan be ticulargroups. analyzedonly if floral colors or sizes forthe universe of 2 Authorfor correspondence. nondioeciousspecies are known.At present,this is pos- 596 May 1995] RENNER AND RICKLEFS-DIOECY IN THE FLOWERING PLANTS 597 sible onlyon a local basis. No studyhas yetaddressed (1988) and the 28 superordersof Thorne(1992); total the correlatesof dioecyin the floweringplants at large, numberof genera (in thefamily data set)and numberof and some apparentassociations among dioecy and par- dioeciousspecies (in thegenus data set);growth forms; ticulartraits may, therefore, be theresult of regionalef- modesof pollinationand dispersal;geographic distribu- fects,particularly if thereis a strongphylogenetic com- tion; and trophicstatus. Genera accepted are those of ponentto thepresence or absenceof dioecy.As longas Brummitt(1992); familiesare those of Cronquist (1988), suchsampling problems remain unrecognized they may exceptthat in 18 cases morerecent, broader family con- resultin theintroduction of unnecessary ad hoc hypoth- ceptswere accepted. To look at theeffects of differences eses to explainsupposed correlations that did notplay a in familialconcepts, the data setswere also brokendown causal rolein theevolution of dioecy. intothe more narrowly circumscribed families of Thorne Here we reportresults of a studyin whichwe have (1992). determinedthe relative strengths of various associations Dioecyand monoecywere scored as presentin a family betweendioecy and particulartraits across all flowering whenat least one speciesin thatfamily is dioeciousor plants.Because we use a multivariateapproach, the pos- monoecious.Particularly in thecase offamilies with few sibilityof fortuitous associations among the traits them- dioecioustaxa great care was taken to searchthe literature selvesis reduced.Our analysis, to somedegree, also takes forindependent confirmation of theirdioecious status. care of phylogeneticbias becauseit is not restrictedto Numeroustaxa (:20 familiesand ;:200 genera)listed as particularfamilies that happen to dominatelocal floras includingdioecious species by Yampolskyand Yampol- (see Steiner,1988, for a particularlystriking example of sky(1922) had to be removedfrom the data set,either thiseffect in theCape floraof SouthAfrica). It is based becausethe supposedly dioecious plants have turnedout on a newcompilation of dioecious angiosperm genera and to be sex-changing,polygamous, monoecious, or gyno- families,the onlyprevious such compilationby Yam- dioecious,or because of nomenclaturaland taxonomic polskyand Yampolsky(1922) beingoutdated. Systema- changes.Yampolsky and Yampolskybased theircom- tists'views of the phylogeneticrelationships within the pilationon Englerand Prantl's NatiirlichePflanzenfamili- angiospermsand the circumscriptionand systematic en and thevolumes of Engler'sPflanzenreich published placementof numerous genera have changedduring the by 1912. Theyconcluded that out ofthe total of 10,113 past 70 yr,new cases of dioecyhave been reported,and angiospermgenera then recognized, 972 (9.6%)comprised manyothers mentioned by Yampolskyand Yampolsky dioeciousand/or polygamodioecious species. However, have turnedout to be in error.In anotherpaper, the same theirlist includesat least 90 generasince mergedwith data set willbe used to studythe distribution of dioecy othergenera; some, such as Streblus,are listed eight times in theframework of recently published angiosperm phy- underdifferent names. There are also 43 unpublished logenies(Renner, unpublished data). names(nomina nuda), the identity of whichis obscure. Specificallywe analyzewhether at the family level dioe- Severalother genera, from a modernsystematic view- cyis associatedwith monoecy, particular geographic dis- point,are completely misplaced as to family.Conversely, tributions,wind or watervs. animalpollination, fleshy 170 additionaldioecious genera and manyadditional fruitsand, therefore, animal-dispersed vs. abioticallydis- familieswere added eitherbecause of newlydiscovered persedseeds, or particular growth forms. Because we lacked instancesof dioecy or because of taxonomic changes, such data forall angiospermswe wereunable to addressas- as narrowermodern family concepts. Our data set con- sociationsbetween dioecy and small,greenish flowers tains959 generathat include dioecious species. (Bawa and Opler, 1975; Fox, 1985; Muenchow,1987; Mostdioecious genera comprise only dioecious species; Ibarra-Manriquezand Oyama,1992) and between dioecy however,325 ofthe 959 generahave more than one mat- and pollinationby small, supposed generalistinsects ing system,and forthem numbers of dioeciousspecies (Bawa, 1994). The formermay be a correlateof theas- had to be estimatedfrom figures given in recentliterature, sociationdemonstrated below between wind pollination usuallyfloras. Statements such as "rarelydioecious" or anddioecy, because wind-pollinated flowers are often small "occasionallydioecious" were scored conservatively as and lackattractive coloration; the second is based on the at least one dioeciousspecies, while "commonlydioe- erroneousidea thatpollinator smallness is correlatedwith cious" was scoredas 50% dioecious.Species numbers are unspecificpollinator-flower relationships and inefficient notused in theanalyses. (in termsof outcrossing)pollination (Renner and Feil, Growthforms were taken from floras and accountsof 1993). particularmodes of growth,such as Gentry(1991) for climbers,while geographic distributions are fromMab- MATERIALS AND METHODS berley(1987), exceptwhere more recent systematic work was available.Growth form was tabulatedas presenceor This studyis based on two data sets,one a data base absenceoftrees, shrubs, herbs, or climbers (whether woody of dioeciousangiosperm genera compiled by searching or herbaceous). theliterature and by correspondingwith specialists (see Pollinationand dispersalmodes wereassessed from Acknowledgments),the other a data base of all families descriptionsof the flowersand fruitsof each taxonin offlowering plants. For detailsof the second compilation variousfloristic and monographicsources (for details, see see Ricklefsand Renner(1994). Thegenus and the family Ricklefsand Renner,1994). In bothdata sets,the cate- data bases containthe followinginformation for each gorieswere 1) abiotic,2) biotic,and 3) variable,the last taxon: presence or absence of dioecy, presence or meaningboth types of pollination or dispersalpresent in absenceof monoecy(in thefamily data set);position in differentspecies within the genus or family. In thegeneric theangiosperms in terms ofthe 11 subclassesof Cronquist data set,pollination mode was additionallyscored as 1) 598 AMERICAN JOURNAL OF BoTANY [Vol. 82 wind,2) water,3) insects,4) bats,5) birds,6) variable, repeatedthe analyses using only small families (less than or7) unknown.Data on thepollination of many dioecious fivegenera per family), which heightens the likelihood of taxa had alreadybeen compiledin connectionwith an co-occurrenceof the variables in thesame genusor spe- earlierreview of the literature(Renner and Feil, 1993). cies; second,we also examinedassociations at thegenus A fewadditional autecological studies have sincebeen levelwhere possible. published(e.g., Kato, 1993), and thereare severalsys- The statisticalsignificance of each of the variablesis tematicreviews (Dobat and Peikert-Holle,1985; Cox, evaluatedby a x2 withits associatedprobability. Each 1988; Kubitzki,Rohwer, and Bittrich,1993; Webster, model has a likelihoodratio (LR) which,if significant, 1994a,b). For manyof the remaining genera, pollination indicatesadditional structure inthe data set not accounted by eitherwind or insectscould be assessedfrom floral forby the model. Initial models included no interactions. structure,with particular weight being given to thepres- Nonsignificantvariables were deleted from the model and ence or absenceof nectaries.All fleshyfruits, as well as when LR x2 values were significant,interactions were indehiscentlarge dry fruits, were scored as beingadapted addedby trialand errorto arriveat a modelwith a non- foranimal dispersal. Some smalldry fruits scored as abi- significantLR x2. Because monoecyis a matingsystem oticallydispersed may have seedsinitially dispersed bal- thatmay facilitatethe evolutionof dioecy(see below), listicallyand thenbiotically (by ants); it is unlikely,how- thesemodels were run both with (+M) andwithout (-M) ever,that this group introduces a large error into the data. monoecyas a variableto isolatethe effects of ecological Wheredescriptions of floral structures and/or fruits were interactionson dioecy. insufficientto assess reproductivemodes, the condition was leftunscored in thegeneric data set; all familiescould RESULTS be scored. For the family-levelanalysis, geographic distribution Out ofthe total of 13,479genera (Brummitt, 1992), at was coded as presentor absentin tropicallatitudes and least 959 (7.1%) containdioecious species; if Thorne's presentor absentin temperatelatitudes. The category (1992) estimateof 12,650genera for the angiosperms is "tropical"includes all familieswhose distributionsin- accepted,that percentage is 7.6%.A conservativeestimate clude eitherlow or highelevations in thetropics; "tem- ofthe number of dioecious species is 14,620or 6% ofan perate"includes all familieswhose distributions include assumedtotal of 240,000 speciesof floweringplants. Of temnperateorboreal zones-generally, zones with frost. The a conservativelycircumscribed 365 familiesof angio- distributionsof the dioecious genera were scored in con- sperms(essentially the familiesof Cronquist,1988; see siderabledetail for future analysis because many genera Materialsand Methods),157 (43%) containdioecious withvariable mating systems also have theirdioecious members.Of Thorne's (1992) morenarrowly defined 437 speciesconfined to a particularregion. For example,in families,167 (38%) have dioeciousmembers. Asperula(Rubiaceae), the 16 Australianspecies are di- In thefollowing we firstpresent results of analyses per- oecious whilethe Mediterraneanspecies are hermaph- formedusing the family-level data set(Tables 1-12); this roditic. is followedby resultsbased on genus-leveldata (Tables Each variablewas scoredas 0 (absent)or 1 (present). 13 and 14). Withineach of thecategories of variables(distribution, Contingencyanalyses relating each variableindividu- matingsystem, pollination, dispersal, growth form, tro- allyto dioecyare presented in Table 1. Variablesthat are phic status)each familyhad to have a scoreof 1 forat significantlyand stronglyassociated with dioecy are, in leastone of thevariables but could also have a scoreof orderof the strength of their correlation, monoecy, climb- 1 forall of them.None could have scoresof 0 forall of inggrowth form, biotic dispersal, abiotic pollination, shrub them.For some of the analyses,the familieswere also growthform, and tropicaldistribution. dividedinto two groups based on numberof genera: size Categoricalmodels are presentedin Table 2. The = 0, fewerthan 5 genera; size = 1, 5 or more genera. strongesteffects in an initialmodel (no interactions)were In orderto examinethe relationshipof dioecyto all monoecy,climbing growth form, biotic dispersal, abiotic variablessimultaneously, categorical multivariate con- pollination,and tropicaldistribution. The individualas- tingencyanalyses (SAS Institute,Inc., 1988, CATMOD sociationof shrubbygrowth with dioecy evidently was procedure)were performed in whichthe presence or ab- subsumedin thismodel through correlation with other senceof dioecywas relatedto presenceor absenceof all variables.In thisanalysis, the likelihoodratio was sig- the othervariables in the data set. Categoricalmodels nificant,indicating additional structure in thedata setin take into accountassociations among the independent theform of interactions. A seriesof models with various (predictor)variables and revealthe unique statistical re- two-wayinteraction terms were then run. The onlysig- lationshipsbetween dioecy and other variables in the data nificantinteraction was abioticpollination x monoecy. set.When applied at thetaxonomic level of family,this When thisinteraction was includedand nonsignificant approachidentifies the associationof traitswithin taxa effectswere deleted from the model, the likelihood ratio but does not implythat a pair of associatedtraits nec- becamenonsignificant (Table 2, model +M). Thus,sev- essarilyoccurs in the same genus or species. Indeed dioecy eral factorscontribute independently to thepresence of andmonoecy cannot co-occur in a singlespecies and hence dioecywithin a family,in descending order ofimportance: theirassociation could only be identifiedby a higher-level monoecy,tropical distribution, abiotic pollination, shrub analysis.A difficultywith this approach is thattwo traits growthform, and climbing growth form. In addition,how- maybe associatedbecause each is morelikely to be rep- ever,monoecy interacts with abiotic pollination in con- resentedin familieswith large numbers of species.We tributingto dioecy(Table 3). When monoecywas ex- triedto deal withthis difficultyin two ways:first, we cludedfrom the categorical analysis (Table 2, model-M), May 1995] RENNER AND RiCKLEFS- DIOECY IN THE FLOWERING PLANTS 599

TABLE 1. Family-level relationshipof presence (in 157 families) or TABLE 3. Contingencytable of theeffects of abioticpollination and absence (in 208 families)of dioecy to each of the variables. bioticdispersal on presenceof dioecy. All familiesare includedin theanalysis. Ratios are numbers of families with dioecy over total Number of families numberof families.a Variable absent Variable present Monoecy Categoryand Dioecy Dioecy Dioecy Dioecy variable absent present absent present x2a p Absent Present Total Monoecy 179 54 29 103 107.6 0.000 Abioticpollination Distribution Absent 33/198 71/83 104/281 Tropical 41 12 167 145 11.2 0.001 (17%) (86%) (37%) 53/84 111 80 97 77 0.2 0.648 Present 21/35 32/49 Temperate (60%o) (65%) (63%) Pollination Total 54/233 103/132 157/365 Abiotic 177 104 31 53 17.9 0.000 (23%) (78%) (43%) Biotic 24 33 184 124 6.0 0.014 a The relationshipbetween monoecy and dioecyin the absenceof Dispersal abioticpollination had LR X2= 123.4,P < 0.001,and in thepresence Abiotic 55 59 153 98 5.1 0.023 ofabiotic pollination, LR X2 = 0.2, P = 0.62. Biotic 117 46 91 111 26.8 0.000 Growth form Herb 103 87 105 70 1.2 0.264 and bioticdispersal (Table 4, butsee below).In addition, Shrub 119 59 89 98 13.9 0.000 dioecyis lesslikely in familiescontaining herbs and more Tree 118 72 90 85 4.2 0.039 Climber 157 72 51 85 33.8 0.000 likelyin familiescontaining shrubs and trees.A categorical modelexcluding nonsignificant effects and including Parasite 203 149 5 8 1.9 0.172 theinteraction of monoecywith abiotic pollination had Size of family 130 53 78 104 30.0 0.000 a nonsignificantlikelihood-ratio and revealed abiotic pol- a X2 is the likelihood ratio x2 value. lination,biotic dispersal, monoecy, and climbinggrowth to be the onlysignificant effects (Table 5, Model +M). Muchof the information contributed by monoecy is con- abioticpollination and climbinggrowth form remained tainedin themonoecy x abioticpollination interaction strongcorrelates of dioecy. (Table 6). Whenmonoecy was excludedfrom the analysis Because,on average,large families contain more di- (Table 5, Model -M), we could not finda whollysatis- oecious taxa thando small families(Table 1), the data factorymodel (LR, P = 0.034), but the same variables weresplit into two setshaving nearly equal numbersof (abioticpollination, biotic dispersal, climbing growth) re- familiesbased on numberof genera per family (less than mainedsignificant. five,and fiveor more).Among small families, the stron- Amonglarge families (Table 7), the strongesteffects gestindividual effects are monoecyand the pollination weremonoecy, biotic dispersal, climbing and shrub growth and dispersalvariables, with dioecy less oftenpresent in forms,and tropicaldistribution, with abiotic dispersal familieswith biotic pollination and abioticdispersal and exertinga weak negative effect. A categorical model showed more oftenpresent in familieswith abiotic pollination monoecyand tropicaldistribution to be theonly signif-

TABLE 2. Categoricalmodels of therelationship at thefamily level between dioecy and theindependent variables in thisanalysis.

Initial model Model +M Model -M

Variable X2 P X2 P X2 P

Monoecy 52.1 0.000 27.6 0.000 -a Tropicaldistribution 4.8 0.029 11.7 0.001 5.2 0.023 Temperatedistribution 0.5 0.500 Abioticpollination 5.7 0.017 8.5 0.004 39.9 0.000 Bioticpolliantion 0.1 0.710 Abioticdispersal 0.0 0.912 Bioticdispersal 6.1 0.014 Herbgrowth form 0.4 0.539 Shrubgrowth form 2.0 0.157 10.7 0.001 Treegrowth form 0.0 0.874 Climbinggrowth form 10.1 0.002 8.3 0.004 36.8 0.000 Abioticpollination x monoecy interaction 26.1 0.000 - - Abioticpollination x bioticdispersal 6.3 0.012 Tropicaldistribution x bioticdispersal 6.9 0.008 Likelihoodratio 192.2 0.004 19.5 0.493 11.4 0.249 (143)b (20) (10) a Variablenot included in analysis. b Degreesof freedom for likelihood ratio. 600 AMERICAN JOURNAL OF BOTANY [Vol. 82

TABLE 4. Relationshipof presence (in 53 families)or absence(in 130 TABLE 6. Contingencytable of theeffects of abioticpollination and families)of dioecyto each of thevariables independently among bioticdispersal on presenceof dioecy.Only smallfamilies (<5 smallfamilies (<5 genera). genera)are included in theanalysis. Ratios are number of dioecious familiesover total number of families.a Number of families Variable absent Variable present Monoecy Categoryand Dioecy Dioecy Dioecy Dioecy Absent Present Total variable absent present absent present x2 P Abioticpollination Monoecy 113 31 17 22 16.8 0.000 Absent 13/114 7/13 20/127 Distribution (11%) (54%) (16%) Tropical 30 9 100 44 0.9 0.35 Present 18/30 15/26 33/56 Temperate 74 28 56 25 0.3 0.61 (60%) (58%) (59%) Total 31/144 22/39 53/183 Pollination (22%) (56%) (29%) Abiotic 107 20 23 33 33.8 0.000 Biotic 20 23 110 30 15.4 0.000 a The relationshipbetween monoecy and dioecyin the absenceof abioticpollination had LR x2 = 11.8,P < 0.001,and in thepresence Dispersal of abiotic pollination,LR x2 = 0.0, P = 0.86. Abiotic 46 30 84 23 7.0 0.009 Biotic 76 21 54 32 5.4 0.020 Growthform sistentassociations in thedata setrelate the presence of Herb 75 40 55 13 5.3 0.021 dioecywithin a familyto monoecy,climbing growth, abi- Shrub 80 25 50 28 3.2 0.076 oticpollination, and biotic dispersal; tropical distribution Tree 71 20 59 33 4.3 0.038 is associatedwith dioecy only in largefamilies. Climber 117 43 13 10 2.5 0.112 Becausemonoecy was suchan importantfactor in as- a x2 is thelikelihood ratio x2. sociationwith dioecy, we examinedthe relationshipof monoecyitself to thedistribution of othervariables (ex- cludingdioecy) within families (Table 11). Severalfactors icanteffects, but therewas a stronginteraction between had significantassociations with monoecy:climbing bioticdispersal and abioticpollination (Table 8, Model growth(+), abioticpollination (+), bioticpollination (-), +Ma). Climbinggrowth form was not significant. bioticdispersal (+), and temperatedistribution (+). A The interactioneffect of abiotic pollination and biotic categoricalmodel of the dependence of monoecy on other dispersalin largefamilies can be visualizedby examining variablesproduced the followingresults: abiotic polli- thecontingency table relating presence or absence of dioe- nation(x2 = 46.8, P = 0.000), climbinggrowth (36.9, cy to combinationsof presenceand absenceof each of 0.000), temperatedistribution (3.9, 0.049), abioticpol- thesevariables (Table 9). Bioticdispersal and abioticpol- linationx bioticdispersal interaction (11.2, 0.001), like- linationwere not individuallysignificant effects. How- lihoodratio (12 df,23.4,0.025). Among families including ever,families with neither of theseattributes included taxawith abiotic pollination, presence or absence of biotic fewerdioecious taxa than families with either one orboth dispersalhad no influenceon monoecy(60.9% vs. 55.3%, of them.The same is trueof the interactionbetween abioticpollination and monoecy(Table 10), whichalso makesa suitablecategorical model (Table 8, model + Mb). TABLE 7. Relationshipamong large families (?5 genera)of presence Whenmonoecy was deletedfrom the categorical model (in 104 families)or absence(in 78 families)of dioecyto each of forlarge families, abiotic pollination, tropical distribu- thevariables. tion,and climbinggrowth form were significant effects, x Number of families alongwith the abiotic pollination bioticdispersal in- Variable absent Variable teraction(Table 8, model -M). present Categoryand Dioecy Dioecy Dioecy Dioecy To summarizethe categorical analyses, the most con- variable absent present absent present x2 P Monoecy 66 23 12 81 75.3 0.000 Distribution TABLE 5. Categoricalmodels for small families(<5 genera)of the relationshipbetween dioecy and theindependent variables in this Tropical 11 3 67 101 8.1 0.005 analysis. Temperate 37 52 41 52 0.1 0.73 Pollination Model +M Model -M Abiotic 70 84 8 20 2.9 0.09 Variable X2 P x2 P Biotic 4 10 74 94 1.3 0.25 Monoecy 6.2 0.013 - - Dispersal Abioticpollination 13.4 0.000 12.3 0.000 Abiotic 9 29 69 75 7.6 0.006 Bioticdispersal 7.8 0.005 9.8 0.002 Biotic 41 25 37 79 15.7 0.000 Climbinggrowth form 4.3 0.039 9.5 0.002 Abioticpollination x Growthform monoecy interaction 4.8 0.028 - - Herb 28 47 50 57 1.6 0.21 Abioticpollination x Shrub 39 34 39 70 5.6 0.018 climbinggrowth form 4.0 0.046 Tree 47 52 31 52 1.9 0.17 Likelihoodratio 15.6 0.111 10.4 0.034 Climber 40 29 38 75 10.4 0.001 (df) (10) (4) a x2 iS thelikelihood ratio x2. May 1995] RENNER AND RICKLEFS-DIOECY IN THE FLOWERING PLANTS 601

TABLE 8. Categoricalmodels for large families of the relationship be- TABLE 10. Contingencytable of theeffects of abiotic pollination and tweendioecy and theindependent variables in thisanalysis. monoecyon presenceof dioecy.Only large families (25 genera) areincluded in theanalysis. Ratios are numbers of dioecious fam- Model +Ma Model +Mb Model -M iliesover total number of families.a

Variable x2 P x2 p x2 p Monoecy Monoecy 48.9 0.000 20.9 0.000 - - Absent Present Total Tropicaldistribution 4.2 0.040 12.4 0.000 7.6 0.006 Abioticpollination 9.7 0.002 Abioticpollination Climbinggrowth form 5.8 0.017 Absent 20/84 64/70 84/154 Abioticpollination x (24%) (91%) (55%) Bioticdispersal 14.7 0.000 19.9 0.000 Present 3/5 17/23 20/28 Abioticpollination x (60%) (74%) (7 1%) Monoecy interaction 7.1 0.008 - - Total 23/89 81/93 104/182 13.3 0.360 3.5 0.360 9.8 0.455 Likelihoodratio (26%) (87%) (57%) (df) (12) (5) (10) a The relationshipbetween monecy and dioecyin theabsence of abioticpollination had LR x2 = 79.1, P < 0.001, and in thepresence of abioticpollination, LR x2 = 0.4, P = 0.54. x2 = 0.3, P = 0.60). Amongfamilies lacking abiotic pollination,presence of bioticdispersal increased the likelihood of monoecy(39.0% vs. 16.2%,x2 = 17.9, P < Dioecy and phylogeny-Dioecy is slightlymore com- 0.001). mon among dicot genera than among monocot genera Climbinggrowth, whether herbaceous or woody,also (8.2% vs. 5.1%; x2 = 30, P < 0.001). It occurs in 24 of was an important,and unexpected,factor in association the 28 superordersof Thorne (Table 13) and in each of withdioecy, and so we examinedthe relationship of this the six dicot and five monocot subclasses of Cronquist growthform to thedistribution of other variables within (1988). Because some ofCronquist's subclasses are poorly families(Table 12). Significantassociations were: dioecy definedand probablypolyphyletic, the followinganalysis (+), tropicaldistribution (+), monoecy(+), bioticdis- concentrateson Thorne's superorders,which because of persal(+), shrubgrowth form (+), bioticpollination (+), theirnarrower circumscription stand a betterchance of and abioticpollination (-). A categoricalmodel of the being monophyletic. dependenceof climbinggrowth on othervariables pro- Among the 14 dicot superordershaving > 100 genera duced the followingresults: shrubby growth (X2 = 26.5, the distributionof dioecious genera is extremelyhetero- P = 0.000), herbaceousgrowth (21.0, 0.000), bioticdis- geneous (X213 = 909, P < 0.0001). In absolute numbers, persal(8.3, 0.004), monoecy(8.1, 0.004), abioticpolli- most dioecious genera are found among Thorne's Mal- nation(5.2, 0.022), likelihood-ratio(22 df,24.9, 0.30). vanae (208), his Magnolianae (134), Violanae (90), Ru- In addition,there were two stronginteractions: herb x tanae (71), Commelinanae (62), Theanae (58), and Gen- shrubgrowth form (19.0, 0.000) and abioticpollination tiananae (49). In relativeterms, dioecy is concentratedin x monoecy(6.8, 0.009). monocotyledonous Pandananae (100%), Alismatanae The followingsections present results based on thege- (47%), and Triuridanae (25%) and the dicotyledonous nus-leveldioecy data set. Detailed global comparisons Rafflesianae(80%), Magnolianae (29%), and Malvanae betweenthe dioeciousgenera and the remainderof the (28%). The Magnolianae and Malvanae are thus among angiospermgenera were not feasable. However, because the most importantdioecious clades, in both relativeand thisis thefirst compilation of dioecy in theangiosperms absolute terms. This is mainly due to a few dioecy-rich since Yampolskyand Yampolsky(1922) and Charles- families, such as the Lauraceae, Menispermaceae, My- worth(1985) andbecause the data are suggestive of several risticaceae,and Monimiaceae in the firstsuperorder and trendsamong dioecious taxa, such as thelikely overre- the Euphorbiaceae, Moraceae, and Urticaceae in the sec- presentationof heterotrophism (see below),we hopethat presentingthese data herewill stimulate further data col- TABLE 11. Family-levelrelationships of the presence (in 132 families) lectionand eventualmultivariate analyses on thegeneric or absence(in 233 families)of monoecy to eachof the other vari- level. ables.

Variable absent Variable present TABLE 9. Contingencytable of theeffects of abioticpollination and Mon- Mon- Mon- Mon- bioticdispersal on presenceof dioecy.Only large families (25 oecy oecy oecy oecy Variable absent present absent present x2. P genera)are included in the analysis. Ratios are numbers of dioecious familiesover total number of families. Tropicaldistribution 40 13 193 119 3.8 0.057 Temperatedistribution 132 59 101 73 4.8 0.028 Biotic dispersal Abioticpollination 198 83 35 49 22.5 0.000 Absent Present Total Bioticpollination 26 31 207 101 9.4 0.002 Abioticdispersal 70 44 163 88 0.4 0.516 Abioticpollination Bioticdispersal 116 47 117 85 6.9 0.009 Absent 14/53 70/101 84/154 Herbgrowth form 130 60 103 72 3.6 0.057 (26%) (69%) (55%) Shrubgrowth form 119 59 114 73 1.4 0.241 Present 11/13 9/15 20/28 Treegrowth form 123 67 110 65 0.1 0.709 (85%) (60%) (7 1%) Climbinggrowth form 168 61 65 71 23.9 0.000 Total 25/66 79/116 104/182 Parasitehabit 226 126 7 6 0.6 0.452 (57%) a (38%) (68%) X2 is thelikelihood ratio x2 value. 602 AMERICAN JOURNAL OF BOTANY [Vol. 82

TABLE 12. Family-levelrelationships of the presence (in 136 families) TABLE13. Distributionof dioecy in Thorne's (1992) superorders. Total orabsence (in 229 families)of climbing growth form to eachof the numbersof genera in thistable are from Thorne, while the families othervariables. are circumscribedas in Cronquist(1988; see Methods).

Variable absent Variable present Number of families Total Dioecious genera Climb- Climb- Climb- Climb- number ers ers ers ers Dio- Mono- of Num- Variable absent present absent present x2. P Superorder Total ecious ecious genera ber Percent Tropicaldistribution 48 5 181 131 24.4 0.000 Dicots Temperatedistribution 115 76 114 60 1.1 0.294 1 Magnolianae 34 12 14 466 134 29 Monoecy 168 65 61 71 23.9 0.000 2 Nymphaeanae 2 0 0 81 0 0 Dioecy 157 51 72 85 33.8 0.000 3 Rafflesianae 3 1 1 10 8 80 Abioticpollination 163 118 66 18 12.4 0.000 4 Caryophyllanae 12 9 7 563 30 5 Bioticpollination 47 10 182 126 12.4 0.000 5 Theanae 40 14 10 540 58 19 Abioticdispersal 75 39 154 97 0.7 0.415 6 Celastranae 2 1 1 60 7 12 Bioticdispersal 124 39 105 97 23.0 0.000 7 Malvanae 20 17 12 769 208 27 Herbgrowth form 128 62 101 74 3.6 0.057 8 Violanae 26 13 11 725 90 12 Shrubgrowth form 133 45 96 91 21.7 0.000 9 Santalanae 9 8 6 162 37 22 Treegrowth form 118 72 111 64 0.1 0.794 10 Geranianae 16 4 3 189 7 4 Parasitehabit 220 132 9 4 0.2 0.617 11 Rutanae 23 8 8 1,129 71 6 12 Proteanae 1 1 1 75 2 2 a X2 is the likelihood ratio x2value. 13 Rosanae 39 15 14 381 36 9 14 Comnanae 23 16 11 650 44 7 ond superorder.Families with the highest concentrations 15 Asteranae 6 3 2 1,251 28 2 of dioeciousgenera in bothabsolute and relativeterms 16 Solananae 8 2 1 296 4 1 are the of 17 Loasanae 1 0 0 13 0 0 Menispermaceae(100% the generaare dioe- 18 Myrtanae 13b 3 2 463 3 1 cious),Myristicaceae (78%), Moraceae (62%), Urticaceae 19 Gentiananae 23 5 6 2,135 49 2 (52%), Anacardiaceae(50%), Monimiaceae(47%), Eu- Total dicots 305c 132 110 9,958 817c 8 phorbiaceae(39%), and Cucurbitaceae(32%). 20 Lilianae 15 8 2 1,184 16 1 Amongthe subclasses,Cronquist (1988, pp. 263-265 21 Hydatellanae 1 0 0 2 0 0 and pp. 451-457) considersthe Magnoliidae, Hamamel- 22 Triuridanae 1 1 1 8 2 25 idae, and Caryophyllidaeas less advancedthan the Dil- 23 Alismatanae 11 7 6 57 27 47 leniidae,Rosidae, and Asteridaeamong dicots, and the 24 Aranae 2 1 2 108 1 1 Alismatidaeand Arecidaeas relativelymore primitive 25 Cyclanthanae 1 0 1 11 0 0 thatthe Commelinidae, Zingiberidae, and Liliidae among 26 Pandananae 1 1 0 3a 3 100 27 Arecanae 1 1 1 200 31 15 monocots.Dioecy is stronglyconcentrated in theless ad- 28 Commelinanae 26 6 9 1,192 62 5 vancedsubclasses. Also Thorne(1992, p. 244, fig.1) tries Total monocots 60 25 22 2,765 142 5 to arrangehis superordersso that"The positionof these Totalflowering plants 365 157 132 12,723 959 7.5 superordinalstem, cross-sections of the phyleticshrub a indicatesas closelyas possiblemy interpretation oftheir Some Pandanaceaeare monecioussex-changers rather than consis- tentdioecists (Cox, 1982). relationshipsand theirrelative degree of specialization b This numberis inflatedby Cronquist'srecognition of Punicaceae fromtheir more primitive, archaic ancestors." We have and Sonneratiaceaeas distinctfrom Lythraceae. numberedthe superorders in hissequence (Table 13) and c Includingfour of Cronquist's families that are unassigned by Thorne: dividedthe dicot superordersroughly into two groups, Aextoxicaceae,Barbeyaceae, Corynocarpaceae, and Pandaceae; ofthese, one relativelyless advanced (numbers1-9), the other Aextoxiconand Barbeyaare dioecious. relativelymore advanced (numbers 10-19). As withCron- quist'ssubclasses, dioecy is concentratedin theless advancedsuperorders ofdicots (genera: x2 = 519,P < 0.0001; Old World tropicsthan in the New as suggestedby Tho- families:x2 = 5.4, P < 0.025). Monoeciousfamilies have mas and LaFrankie (1993), due primarilyto the highOld a paralleldistribution to dioeciousfamilies (Table 13)and World diversityof such dioecy-richfamilies as the Eu- exhibita similarconcentration in less advanced dicot phorbiaceae, Menispermaceae, and Restionaceae. superorders(X2 = 4.4, P < 0.05). Possiblereasons for the concentrationof dioecyin the less advancedclades are Growthforms ofdioecious angiosperm genera -We have givenin thediscussion. informationon thegrowth forms of 818 of the 959 genera with dioecy. Of the 818, 198 (24%) comprise only tree Geographicdistribution of dioeciousangiosperm gen- species, 157 (19%) only herbs, 128 (16%) only climbers, era-Of the959 dioeciousgenera, 217 occurin theneo- and 94 (11%) only shrubs.Tree and shrub species occur tropics,402 in thepaleotropics, 86 are pantropical,and togetherin 195 (24%) genera with dioecy. Other combi- 149 are foundexclusively in thetemperate zone. Thirty nations of growthforms, such as trees,shrubs, and herbs generahave temperateand tropicalmembers. The -re- (seven genera),shrubs and herbs (19 genera),shrubs and maining75 generaare mostlyrestricted to Hawaii,New climbers (ten genera), and so on are rare. Of the genera Zealand,Australia, New Caledonia,or New SouthWales. with dioecy, :z551 (67%) are woody (namely those with Becausethe total numbers ofgenera in the large geographic trees and shrubs or both growthforms, plus half those regionshave notbeen tabulated, it is notpossible to cal- with only climbers).In the angiospermsas a whole, 234 culatethe different relative frequencies of dioecy,but it familiesincluding tree and shrubgrowth forms have 7,929 seemslikely that dioecy is indeedmore frequent in the genera, or 59% of the total 13,500 genera; 182 families May 1995] RENNER AND RICKLEFS-DIOECY IN THE FLOWERING PLANTS 603

TABLE 14. Genus-levelanalysis of the association between dioecy and withbird-pollinated species (Porsch, 1931; and personal pollinators. estimate)in therest of the flowering plants (Table 14) it appearsthat insect and vertebratepollination are both Number of genera underrepresentedamong dioecious genera. This was also Pollinated by With dioecy Lacking dioecy notedby, for example, Richards (1986); belowwe discuss Insects 550 (5.1%) 10,317a possiblereasons for the nearabsence of bat pollination Windor water 232 (21.0%) 874 amongdioecists. Bats or birds 4 (0.4%) 750 a This numberof nondioeciousgenera pollinated by insectswas ar- Dispersalmodes ofdioecious angiosperm genera-Of the rivedat by subtractingfrom the totalnumber of angiospermgenera 814 dioeciousgenera for which there are sufficientdata (13,500) all thedioecious genera (959), 750 bird-and bat-pollinated on mode ofdispersal, 526 (65%) are entirelyor predom- genera,874 generain familieswith solely wind- or water-dispersed members,and an estimated600 generain families with various dispersal inantlyanimal-dispersed, 254 (31%)are dispersed by wind, modes.There are ;250 bat-pollinatedgenera (Dobat and Peikert-Holle, 28 arewater-dispersed, and sixhave bothbiotic and abi- 1985) and 500 bird-pollinatedgenera (Porsch, 1931 and our own es- oticmodes of dispersal. We haveno data on thedispersal timate). modesof 145 genera.In thefamily-level data set, 1,782 generabelong to 114 familieswith wholly biotic dispersal and 5,362genera belong to 88 familieswith both abiotic havingonly shrubs or treeshave 3,185genera, or 24% of and bioticdispersal. Taking half the latter plus the former thetotal angiosperm genera. Thus, between 24% and 59% gives an estimateof 4,463 animal-dispersedgenera, or ofall angiospermgenera may be woody,and so woodiness ;33% of theangiosperm total. As in thecase of woody wouldappear to be overrepresentedamong dioecious gen- growth,animal dispersal appears to be overrepresented era.In thefamily-level analysis, woodiness per se wasnot amongdioecious genera, even though this was nota par- a factorbecause the treegrowth form had no influence ticularlystrong factor in thefamily-level analysis. on thepresence or absenceof dioecy within a familyand theshrub growth form had an effectonly in some ofthe DISCUSSION analyses(but see Materialsand Methodsfor a discussion ofthe limitations of a multivariateapproach at thefamily As has oftenbeen pointed out, the scattered systematic level). distributionof dioecyin the angiospermssuggests that thismating system has evolved independently many times Dioecy and trophiccondition-Of the 959 dioecious and possiblyfor different reasons (Lewis, 1942; van der genera,at least43 (4%) are holoparasites,hemiparasites, Pijl, 1978;Lloyd, 1982). This is also suggestedby the fact or saprophytes.Dioecious parasitesoccur in the Bala- thatin some lineages,such as the Caryophyllaceaeand nophoraceae,Eremolepidaceae, Loranthaceae, Misoden- Chenopodiaceae,sexual expression varies almost contin- draceae,Opiliaceae, Rafflesiaceae, Santalaceae, and Vis- ually,with ready suppression of male or femalefunction caceae. Amongthe remainder of theangiosperms, para- in responseto environmentalconditions (Heslop-Harri- siticgenera are concentrated in theConvolvulaceae (Cus- son, 1957; Freeman,McArthur, and Harper,1984; Free- cutoideae),Cynomoriaceae, Hydnoraceae, Krameriaceae, man et al., 1993), whilein otherlineages, such as the Lennoaceae, Loranthaceae, Olacaceae, Santalaceae, Monimiaceae,Euphorbiaceae, and Cucurbitaceae, dicliny Scrophulariaceae(Rhinanthoideae, Orobanchoideae), and is basicallyfixed. In thedicotyledonous angiosperms as Rafflesiaceae(Mitrastemonoideae), which together may a whole,monoecy and dioecyare concentrated in theless comprisesome 135 heterotrophicgenera (1.1% of all advancedsuperorders of Thorne(1992). A consequence 13,500angiosperm genera). Dioecy therefore may be ov- ofthis nonrandom distribution is that frequencies of dioe- errepresentedamong heterotrophic plants. This was also cy in particularfloras (made muchof by manyauthors) suggestedby a species-levelanalysis of dioecy in theflora likelyreflect the level of dioecyin a particularpool of ofthe southeastern United States (Conn, Wentworth, and familiesas muchas, 'ormore than, local selectivefactors. Blum,1980); t 17% of all theCarolina heterotrophs are Therefore,the questionregarding which of the various dioecious. ecomorphologicalfeatures associated with dioecy play a causal role,and underwhich circumstances, will have to Pollinationmodes of dioecious angiosperm genera-Of be answeredfor a numberof individuallineages before the799 generawith dioecious species for which there are we can extrapolatefrom common features to common sufficientdata on themode of pollination, 232 (30%) are causes. wind-or water-pollinated,550 (68%) are predominantly To controlfor phylogeneticresemblances between insect-pollinated,and 14 (2%) have windand insectpol- membersof the same lineage, we are currentlyanalyzing linationin closelyrelated species and sometimeswithin ourdioecy data setwithin the framework of recently pub- a species.Three genera, the Liliaceae Collospermum and lishedmolecular phylogenies (Renner, unpublished data). Astelia(Dobat and Peikert-Holle,1985) and the Bala- However,as suggestedby D. Charlesworth(letter of 16 nophoraceaeDactylanthus (Webb and Kelly,1993), are March 1994) the differencesbetween and similarities pollinatedby bats. Another bat-pollinated but trioecious withinlineages should tend to obscuregeneral patterns; speciesis Pachycereuspringlei(Cactaceae; Flemming, per- so, wheregeneral patterns are neverthelessapparent we sonalcommunication). Freycinetia (Pandanaceae) is pol- can accordthem some weight. linatedby birds and bats(Cox, 1982).When these figures At thefamily level, the correlation between dioecy and are comparedto the ,:250 generawith bat-pollinated fleshyfruits (biotic dispersal) is significantbut does not species(Dobat and Peikert-Holle,1985) and 500 genera accountfor a largeproportion of the variationamong 604 AMERICAN JOURNAL OF BOTANY [Vol. 82

familiesin presenceor absenceof dioecy;at thegeneric ceae, Campanulaceae,Goodeniaceae, Lobeliaceae, and level,biotic dispersal appears to be overrepresentedamong otherfamilies in the Asteridaeexhibit a complexsyn- dioecists.The associationof dioecy with fleshy fruits found organizationof male and female organs, fQr example, with in particularfloras (see the introduction)likely reflects pollenfirst being deposited on and thenpresented by the the combinedeffects of local samplingand multipleas- style(secondary pollen presentation) or withcongenital sociationsamong the correlates themselves. Thus, for ex- and postgenitalfusion of carpels, petals, and stamens.A ample,across all angiospermsfleshy fruits are strongly similarsynorganization of male and femalestructures correlatedwith the tree growth form (Ricklefs and Renner, characterizesmany Fabaceae (forfurther examples, see in press),which in turnpredominates in certaintropical van der Pijl, 1978). All thisshould make the ontogenic families. suppressionof male or femaleorgans, without radical Tree growthalone is notassociated with dioecy at the alterationsin the structuralorganization of the flower, familylevel, whilethe shrubgrowth form has a weak moredifficult in the moreadvanced groups than in the positiveeffect only in some analyses;climbing growth, less advancedones. whetherwoody or herbaceous,by contrasthas a strong Modem familiesthat have managedto escape from enhancingeffect on thepresence of dioecy (possibly mech- suchconstraints are, e.g., the Apiaceae, Araliaceae, and anismsfor this association are suggested below). The sup- Asteraceae(the latterin spiteof secondarypollen pre- posed correlationbetween woodiness per se and dioecy sentation),all ofwhich have small flowers aggregated into has led variousauthors (e.g., Lloyd, 1982) to hypothesize inflorescencesand all ofwhich are capable of suppressing abouta particularneed for obligatory outbreeding in long- sexualfunction in some flowers.They have andromon- lived,woody plants due to geneticloads. Anothersug- oeciousand gynomonoeciousspecies, yet relatively little gestionrelates woodiness and dioecyvia themore intense dioecy(six of403 generaand :z34 of3,100 species in the selectionfor heterosis in theclimax habitats where these Apiaceae,four of 47 generaand ;:69 of 800 speciesin plantsoccur (Bell, 1982). The presentanalysis provides the Araliaceae,and 26 of 1,509 genera,and ;:526 of littleempirical justification for these hypotheses. 20,000 speciesin theAsteraceae). Dioecy is stronglyassociated with monoecy, abiotic A secondcorrelate of dioecy(and monoecy)is polli- pollination,and climbinggrowth. We will now address nationby wind or water. Several causative factors for the possiblemechanisms underlying these three associations. associationbetween dioecy and abioticpollination have The singlemost important predictor of a group'stendency beensuggested (Darwin, 1876; Kerner, 1895; van der Pijl, to acquiredioecy is thepresence of monoecy in thegroup 1978; Freeman,Harper, and Ostler,1980). In wind-pol- (as also foundby Yampolsky and Yampolsky,1922, and linated plants,pollen release may requirependulous, Lewis,1942). As pointedout by Westergaard (1958), the shakingstamens whereas pollen capture may favor large, generaldirection of evolutionarychange is likelyto be erect,free stigmas. Similarly, hydrophilous plants have frommonoecy to dioecy;clearly, however, phylogenetic floralmechanisms for the capture of water-borne pollen analysesof individualexamples are neededto support thatphysically interfere with mechanisms for the dispersal thisassumption. The routefrom monoecy to dioecymay of pollen(Cox, 1988, 1991). Also, theimprecise move- be especiallycommon because once a lineagehas the mentof pollen by wind or water is likelyto causefrequent physiologicaland morphologicalability to suppressmale selfing(in theabsence of self-incompatibility)in an her- or femalefunction in some flowers,the subsequent step maphrodite.Therefore, abiotic pollination should favor to dioecymay occurthrough divergent adjustments of unisexualflowers, the latterthen apparently facilitating floralsex ratiosbetween individual plants (Lloyd, 1972; theevolution of dioecy from monoecy. Support for such McArthuret al., 1992).It maybe easierfor strictly mon- an interpretationcomes fromrelatively few studies of oecious populationsto evolve dioecy(since mutations changesin breedingsystem in connectionwith changes affectingpollen and ovuleproduction must already have inpollination mode. InAcer(Hesse, 1979) and Thalictrum occurredin theirunisexual flowers) than for populations (Kaplan and Mulcahy,1971) a switchin some species thatproduce bisexual flowers, be theygynomonoecious, frominsect to windpollination accompanied the acquisi- andromonoecious,polygamo-monoecious, gynodio- tionof dicliny, polygamodioecy, and dioecy;while Hesse ecious,androdioecious, or trioecious.Indeed, how fre- interpretsanemophily to have precededdicliny in Acer, quentlygynodioecy has led towarddioecy is unclear Kaplanand Mulcahyfavor the opposite interpretation in (Charlesworthand Charlesworth, 1978; Gouyon and Cou- Thalictrum.Phylogenetic studies may eventually permit vet, 1987; Wellerand Sakai, 1991; Barrett,1992). us to establishthe evolutionary sequence of events in both Whyare dioecyand monoecyconcentrated in theless genera.In tropical,entomophilous Fagaceae, dicliny and advanceddicots in spiteof thefact that very few of the monoovulymay have beenpreadaptive to a lateracqui- mostprimitive extant Magnoliidae are wind-pollinated sitionof anemophily as suggestedby van derPijl (1978). (Endress,1990a)? We believethat the reason has to do In stillother groups, the transition from bisexual to uni- withfloral morphological constraints, or, rather, their ab- sexualflowers was notassociated with a switchfrom in- sence.In primitiveangiosperms, floral phyllotaxis is rel- sectsto wind,and in thesetaxa, insect pollination may ativelyplastic, and thereis littleintegration between male have somethingto do withthe evolutionof dioecy,al- and femalefloral parts (Endress, 1987, 1990b). In the thoughthe mechanism for such an effect(increased selfing more advanced groups,by contrast,numbers of floral ratesin partsof the rangewith different pollinator be- partsoften are low and fixed,and structureshave evolved havior?)is unclear. to directpollinator movement or to concealnectar (usu- Thirdly,dioecy is stronglyassociated with climbing allyvia perigynyor epigyny, sympetaly, and zygomorphy). growth,irrespective of whether the plant is a woodyliana Apocynaceae,Asclepiadaceae, Brunoniaceae, Calycera- oran herbaceousvine. Families with climbers, as a group, May 1995] RENNER AND RICKLEFS-DIOECY IN THE FLOWERING PLANTS 605

seemto have a combinationof features predisposing them LITERATURE CITED todioecy, namely high incidence ofmonoecy, fleshy fruits, and tropicaldistribution. However, the positive effect of theclimbing habit on a family'slikelihood of comprising BAKER, H. 1959. Reproductive methods as factorsin speciation in dioeciousspecies is independentof these other variables floweringplants. Cold Spring Harbor Symposium in Quantitative Biology 24: 177-19 1. (Table 2). We suspectdifferential selection for optimal BARR=r, S. C. H. 1992. Gender variationand theevolution of dioecy resourceallocation to sexualfunction to be responsible in Wurmbeadioica (Liliaceae). Journalof EvolutionaryBiology 5: forthis effect. An increasein femalefloral allocation with 423-444. increasedplant size has been found in severalmonoecious BAWA,K. S. 1980. Evolution of dioecy in floweringplants. Annual climbers(Bickel and Freeman,1993). This impliesre- Review of Ecology and Systematics11: 15-39. sourcelimitations to fruitset below a certainthreshold, 1994. Pollination of tropical dioecious angiosperms:a reas- andin is sessment?No, not yet.American Journal of Botany 81: 456-460. Cucurbitapeposubsequent growth indeedstrong- , AND P. A. OPLER. 1975. Dioecism in tropical foresttrees. lyinhibited after the first fruit is produced(Nitsch et al., Evolution 29: 167-179. 1952).If rapid upward growth is at a premiumin climbers BELL, G. 1982. The masterpieceof nature.Croom Helm, London. and if fruitset at least temporarilyinhibits growth or BICKEL,A. M., AND D. C. FREEMAN. 1993. Effectsof pollen vectorand requiresthe production of a thick,slowly growing stem plant geometryon floralsex ratio in monoecious plants.American to supportheavy fruits, it might be advantageousto post- Midland Naturalist 130: 239-247. BRummITr, R. K. 1992. Vascular plant families and genera. Royal pone femaleness.If the effectis strong,this may favor Botanic Gardens, Kew. male plants.Structural necessities may also favorsepa- CHARLESWORTH, B., AND D. CHARLESWORTH.1978. A model forthe rationof the sexes, with flowers bearing fruit better placed evolutionof dioecy and gynodioecy.American Naturalist 112: 975- on thickerstems and pollen-dispensing flowers on thinner, 997. moreexposed modules. CHARLEsWORTH, D. 1985. Distributionof dioecy and self-incompat- The near absenceof birdand bat pollinationamong ibilityin angiosperms.In P. J. Greenwood and M. Slatkin [eds.], Evolution: essays in honour of John Maynard Smith, 237-268. dioeciousplants (Table 14) maybe explainedby the fact Cambridge UniversityPress, Cambridge. thatvertebrate pollination requires adaptations of floral CoNN,J. S., T. R. WENTWORTH, AND U. BLUM. 1980. Patternsof morphology,which seems to almostpreclude the evo- dioecism in theflora of the Carolinas.American Midland Naturalist lutionof dicliny.First, pollination by large-bodiedani- 103: 310-315. mals,such as birdsand bats,requires large quantities of Cox, P. A. 1982. Vertebratepollination and the maintenanceof dioe- nectar(Opler, 1983), which can be offeredadequately only cism in Freycinetia.American Naturalist 120: 65-80. - . 1988. Hydrophilous pollination. Annual Review of Ecology in cups,tubes, or spurstypically formed by concave floral and Systematics19: 216-280. receptacles(often with inferior ovaries) and/or fused pet- 1990. Pollination and the evolution of breedingsystems in als (sympetaly).Secondly, pollination by thesevectors Pandanaceae. Annals of the Missouri Botanical Garden 77: 816- requiresrelatively large quantities of pollen,which are 840. oftenproduced by numerousstamens (secondary poly- 1991. Abiotic pollination:an evolutionaryescape foranimal- andry).Inferior ovaries, sympetaly, and secondarypoly- pollinated angiosperms. Philosophical Transactions of the Royal Society of London B 333: 217-224. andryare attributesof the relatively advanced subclasses 1993. Hydrophilouspollinationandbreedingsystemevolution Dilleniidae,Rosidae, and Asteridae,but nearly absent in in seagrasses:a phylogeneticapproach to the evolutionaryecology theMagnoliidae and Hamamelidae (Cronquist, 1988; En- of Cymodoceaceae. Botanical Journalof the Linnean Society 113: dress,1990b). Thus thereis a strongmorphological con- 217-226. strainton vertebratepollination. Indeed, of the circa 250 CROAT, T. B. 1979. The sexualityof the Barro Colorado Island flora generawith bat-pollinated species known in the angio- (Panama). Phytologia42: 319-348. CRoNQuIsT, A. 1988. The evolution and classificationof flowering sperms(Dobat and Peikert-Holle,1985), none belong to plants, 2d ed. Columbia UniversityPress, New York, NY. theMagnoliidae and Hamamelidae,while 28% arein the DARWIN,C. 1876. The effectsof cross and self fertilisationin the Rosidae, 26% in the Asteridae,25% in the Dilleniidae, vegetable kingdom.John Murray,London. 13% in themonocots, and 8% in theCaryophyllidae. The DOBAT, K., AND T. PEIKERT-HoLLE. 1985. Blutenund Fledermaiuse. distributionsof bat pollinationand dioecyin theangio- Verlag von Waldemar Kramer, Frankfurt. spermsare thusalmost mutually exclusive. DONOGHUE,M. J. 1989. Phylogeniesand the analysis of evolutionary sequences, with examples from seed plants. Evolution 43: 1137- Whilethe environmental conditions under which pop- 1156. ulationschange breeding systems will have to be ad- ENDRESS, P. K. 1987. Floral phyllotaxisand floralevolution. Botan- dressedexperimentally in allianceswith relatively labile ische JahrbiicherflirSystematik 108: 417-438. sex expression,such as Begonia, Atriplex,or Wurmbea 1990a. Evolution of reproductivestructures and functionsin (Barrett,1992; McArthur et al.,, 1992), the global analysis primitiveangiosperms (Magnoliidae). Memoirs of the New York performedhere suggests that across all angiospermfam- Botanical Garden 55: 5-34. 1990b. Patternsof floralconstruction in ontogenyand phy- iliesmonoecy, wind pollination, and climbing growth con- logeny.Biological Journalof the Linnean Society 39: 153-175. sistentlyfavor the evolution of dioecy. Other traits, such Fox, J. F. 1985. Incidence of dioecy in relationto growthform, pol- as heavy,fleshy fruits, may play a rolein certainlineages. lination and dispersal. Oecologia 67: 244-249. Wherethe ontogeneticfloral adjustments required for FREEMAN, D. C., K. T. HARPER, AND W. K. OSTLER. 1980. Ecologyof unisexualflowers are possiblewithout jeopardizing pol- plantdioecy in theintermountain region of Western North America lination,specialization for one sexual functionappears and California. Oecologia 44: 410-417. , E. D. McARTHuR, AND K. T. HARPER. 1984. The adaptive favoredby selection under a widerange of environmental significanceof sexual labilityin plants using Atriplexcanescens as conditionsas is apparentfrom our finding that geographic a principal example. Annals of the Missouri Botanical Garden 71: distributionhas no strongeffect on presenceor absence 265-277. of dioecy. S. C. SANDERSON, AND A. R. TIEDEMANN. 1993. The 606 AMERIcAN JOURNAL OF BOTANY [Vol. 82

influenceof topographyon male and femalefitness components of NrrscH, J. P., E. B. KURz, J. L. LIVERMAN, AND F. W. WENT. 1952. Atriplexcanescens. Oecologia 93: 538-547. The development of sex expressionin cucurbitflowers. American GENmy, A. 1991. The distributionand evolution of climbingplants. Journalof Botany 39: 32-43. In F. E. Putz and H. A. Mooney [eds.], The biology of vines, 3- OPLER,P. A. 1983. Nectar productionin a tropicaLecosystem.In B. 49. Cambridge UniversityPress, Cambridge. Bentleyand T. Elias [eds.], The biology of nectaries,30-79. Co- GIVNISH,T. J. 1980. Ecological constraintson theevolution ofbreeding lumbia UniversityPress, New York, NY. systemsin seed plants:dioecy and dispersalin gymnosperms.Evo- PoRscH, 0. 1931. Grellrotals Vogelblumenfarbe.Biologia Generalis lution 34: 959-972. 7: 647-674. GOUYON,P.-H., AND D. COUVET. 1987. A conflictbetween two sexes, RENNER,S. S., AND J. P. FEIL. 1993. Pollinatorsof tropicaldioecious females and hermaphrodites.In S. C. Stearns [ed.], The evolution angiosperms.American Journalof Botany 80: 1100-1107. of sex and its consequences. BirkhauserVerlag, Basel. RIcHARDs, A. J. 1986. Plant breeding systems. Allen and Unwin, HART, J. A. 1985. Evolution of dioecism in Lepechinia Willd. sect. Winchester,MA. Parviflorae(Lamiaceae). SystematicBotany 10: 147-154. RICKLEFS,R. E., AND S. S. RENNER. 1994. Species richness within HEsLoP-HARRsoN, J. 1957. The experimentalmodification of sex familiesof floweringplants. Evolution 48, in press. expressionin floweringplants. Biological Reviewsof the Cambridge SAS INSTITUrE,INc. 1988. SAS/STAT (TM) user's guide,Release 6.03 PhilosophicalSociety 32: 38-90. ed. SAS Institute,Inc., Cary, NC. HESSE, H. 1979. Ultrastrukturund Verteilungdes Pollenkittsin der SpoRNiE,K. R. 1949. A new approach to the problemof the primitive insekten-und windbliutigenGattung Acer (Aceraceae). Plant Sys- flower.New Phytolologist48: 259-276. tematicsand Evolution131: 277-289. STEBBINS, G. L. 1950. Variation and evolution in plants. Columbia IBARRA-MANRiQuEz,G., AND K. OYAMA. 1992. Ecological correlates UniversityPress, New York, NY. ofreproductivetraits of Mexican rainforest trees. American Journal 1951. Natural selectionand the differentiationof angiosperm of Botany 79: 383-394. families.Evolution 5: 299-324. KAPLAN, S. M., AND D. L. MuLcAHY. 1971. Mode of pollinationand STENmER,K. E. 1988. Dioecism and its correlatesin the Cape floraof floralsexuality in Thalictrum.Evolution 25: 659-668. South Africa.American Journal of Botany 75: 1742-1754. KATO, M. 1993. Floral biology of Nepenthesgracilis (Nepenthaceae) THOMAS, S. C., AND J.V. LAFRANKIE. 1993. Sex,size, and interyear in Sumatra. American Journalof Botany 80: 924-927. variation in floweringamong dioecious trees of the Malayan rain KERNER VON MARIAUN, A. 1895. Pflanzenleben,vol. 2. Bibliogra- forest.Ecology 74: 1529-1537. phisches Institut,Leipzig. THOMSON, J.D., ANDS. C. H. BARRErr. 1981. Selectionfor outcrossing, KUBIrzKI, K., J. G. ROHWER, AND V. BrTRIucH[EDS.]. 1993. The sexual selection,and the evolution of dioecy in plants. American familiesand generaof vascular plants,vol. 2. Springer-Verlag,Ber- Naturalist 118: 443-449. lin. , AND J.BRUNET. 1990. Hypothesesfor the evolution of dioecy LAHAv-GINoTT,S., AND Q. C. B. CRoNK. 1993. The mating system in seed plants. Trendsin Ecology and Evolution 5: 11-16. ofElatostema (Urticaceae) in relationto morphology:a comparative THORNE,R. F. 1992. Classificationand geographyof the flowering study.Plant Systematics and Evolution186: 135-145. plants. Botanical Review 58: 225-348. LEWIS,D. 1942. The evolution of sex in floweringplants. Biological VAN DER PuL, L. 1978. Reproductive integrationand sexual dishar- Reviewof the Cambridge Philosophical Society 17: 46-67. mony in floralfunctions. In A. J. Richards [ed.], The pollination LIOYD, D. G. 1972. Breedingsystems in Cotula L. (Compositae, An- of flowersby insects,79-88. Academic Press, London. themideae). New Phytologist71: 1181-1194. WEBB, C. J.,AND D. KELLY. 1993. The reproductivebiology of the 1982. Selection of combined versus separate sexes in seed New Zealand flora. Trends in Ecology and Evolution 8: 442-447. plants. The American Naturalist 120: 571-585. WEBSTER, G. L. 1994a. Classificationof the Euphorbiaceae.Annals of MABBERLEY, D. J. 1987. The plant-book.Cambridge University Press, the Missouri Botanical Garden 81: 3-32. Cambridge. 1994b. Synopsis of the genera and supragenerictaxa of Eu- McARTHuR,E. D., D. C. FREEMAN, L. S. LUCKINBILL,S. C. SANDERSON, phorbiaceae. Annals of theMissouri Botanical Garden 81: 33-144. ANDG. L. NOLLER. 1992. Are trioecyand sexual labilityinAtriplex WELLER, S. G., AND A. K. SAKII. 1991. The genetic basis of male canescens geneticallybased?: evidence from clonal studies. Evo- sterilityin Schiedea (Caryophyllaceae),an endemicHawaiian genus. lution46: 1708-1721. Heredity67: 265-273. McCoMB, J. A. 1966. The sex forms of species in the flora of the WESTERGAARD, M. 1958. The mechanismof sex determinationin south-westof westernAustralia. Australian Journal of Botany 14: dioecious floweringplants. Advances in Genetics9: 217-281. 303-316. YAMPOLSKY,C., AND H. YAMPOLSKY. 1922. Distributionof sex forms MuENcHow, G. E. 1987. Is dioecy associated withfleshy fruit? Amer- in the phanerogamicflora. Bibliotheca Genetica 3: 1-62. icanJournal of Botany 74: 287-293.