Etoluriotr.52(2).1 991{p. p. .113560 2

BUTTERFLIES AND PLANTS: A PHYLOGENETIC STUDY

Nrxl,qs J,qNzrR Nn SOnENN yr-rN Departntenrt tf Zoologt',U nit,ersityo J Stockholn,1 06 91 StockholmS, wedert IE -ntail : niklas.jan z.@z .oolog i. su.s e

Abstract.-A database on host plant records frorn 437 ingroup taxa has been used to test a number of hypotheses on the interaction between and their host plants using phylogenetic methods (sirnple character optinrization. concentrated changes test, and independent contrasts test). The phylogeny was assembled from various s()urces irnd host plant clades were identified according to Chase et al.'s rbcJ--basedp hylogeny. The ancestral host plant appears to be associated within a highly derived rosid clade, including the family Fabaceae. As fossil data suggest that this clade is older than the butterflies, they must have colonized already diversilied plants. Previous studies also suggest that the patterns of association in most -plant interirctions are more shaped by host shifts, through colonization and specialization. than by cospeciation. Consequently, we have focused explicitly on the mechanisms behind host shilis. Our results confirm, in the light of new phylogenetic evidence, the pattern reported by Ehrlich and Raven that related butterflies feed on related plants. We show that host shifts have generally been rnore comrnon between closely related plants than between nore distantly related plants. This finding. together with the possibility ofa highertendency of recolonizing ancestral hosts, helps to explain the apparent large-scale conservation in the patterns of association between and their host plants, patterns which at the same tinle are more flexible on a more detailed level. Plant growth form '*'as an even more conservative aspect of the interaction between butterflies and their host plants than plant phylogeny. However, this is largely explained by a higher probability of colonizations and host shifts while 1'eedingo n trees than on other growth fonns.

Ke\, words.-Coevolution, host shifts, insect-host plant interactions, . Papilionoidea. specialization.

Received March 22, 1996. Accepted November 2.6, 1991

Few systems have played such an important role in our butterfly fossil dates back to 48 M.Y.B.P and the diversifi- understanding of how interactions evolve as butter- cation of the butterfly families probably took place at the end flies and their host plants. To a large extent this is the result of the Cretaceous,a bout 66 M.Y.B.P (Emmel et al. 1992). of a single influential paper by Ehrlich and Raven (196.1). At least some families even in the most recently derived of Their essay inspired a flood of paperi on difl'erent aspectso f the plant clades used in this analysis date to this time, such this association, and a number of related hypotheses on the as, Urticaceae( a member of Rosid I in Chase et al., 1993): evolution of insect-plant interactions have emerged. How- 90 M.Y.B.P.,R utaceae( Rosid 2): 52 M.Y.B.P.,A piaceae( As- ever, there have been few attempts to exploit the large da- terid 2): 52 M.Y.B.P., Apocynaceae( Asterid l): 60 M.Y.B.P. tabase on butterfly-host plant affiliations to test such hy- (dates fiom Eriksson and Bremer 1992). Therefore, the clades potheses using phylogenetic methods (Mitter and Brooks themselves must be even older. It is reasonable to regard the 1983; Miller 1987a). A major reason for this is that well- evolution ofcurrent associationsa s arising generallyt hrough supported phylogenies. fbr both butterflies and seed plants, butterfly colonization of already-diversified hosts, and that have been unavailable.H owever, this is slowly changing,a nd is the approach we shall take. This is not to say that coevo- today it is possible to put together reasonably robust phy- lution is an unimportant process in the interaction between logenies for both groups. Chase et al. (1993) have recently butterflies and their host plants. only that evidence fbr it published a molecular phylogeny for all seed plants, which should be sought at other levels of resolution. is probably the best estimate of large-scale angiosperm phy- There are two fundamentally different approachest o com- logeny to date. Butterfly phylogenies are also emerging and parative analyses using phylogenetic data. One approach we have constructed a plausible phylogeny across the but- seeks to find and explain general ecological or evolutionary terflies by combining these published estimates. correlations( Felsenstein1 985; Grafen 1989; Harvey and Pa- Ehrlich and Raven ( I 964) argued that the patterns of host gel 1991: Pagel 1992), while the other seekst o reconstruct plant association that we observe today have been shaped by and explain particularh istoricale ventso r sequenceso f events a stepwise coevolutionary process in which plants evolve along branches in a phylogeny (Mitter and Brooks 1983: defensesa gainst natural enemies,a nd these enemies in turn Coddington l98U; Sill6n-Tullberg 1988; Maddison 1990, evolve new capacitiest o cope with thesed efenses.P lantst hat Brooks and Mclennan l99l). These approachesa re com- escapef iom herbivores can diversify in the absenceo l'en- plernentary (Coddington 1994; Nylin and Wedell 199,1;P agel emies.I nsectst hat eventually managet o colonize one ol'these 1994) and we have in the presentp aper used both, depending plants will enter a new adaptivez one and can in turn diversify on the prcblem. onto the relatives ol' this plant, becauset hey will be chem- The question of ancestral host associationsi s a clearly ically similar. Ehrlich and Raven argued that thesep rocesses historical problem. On what plant did the first butterfly feed'/ have led to the main pattern they had observed,n amely that This question is interesting in its own right, but answering related butterflies tend to f'eed on related groups of plants. it also provides necessary information for any phylogenetic Most or all plant diversificatir)nu p to the level of resolution tests regarding direction of evolution of host plant associa- used in our analysis had probribly already taken place at the tions. Ehrlich and Raven (1964) regardedi t most likely that time the butterflies started to diversify. The oldest known the ancestral host plant family was Aristolochiaceae. ln con- 486

1 - il 9 9 l J T h c -S o c i c t l ,l i r r t h c S t r " r dov l ' E , r o l u t i o n .A l l r i g h t s r c s c r v c d B U T T E R F L I E S A N D P L A N T S 487

trast, Scott (1986) noted that Fabaceae were eaten by the Are patternso f butterflyh ost plant utilizationn onrandoms o most basal branches of several butterfly families and sug- that relatedb utterfliesf eed on relatedp lants,a s suggestedb y gested that the ancestral host probably was a legume. More Ehrlich and Raven?( 2) What was the ancestralh ost plant recently, Ackery (1991) suggestedt hat Malvales may instead associationa, nd has this associationc onstrainedh ost plant be lhe anceslralh ost association. utilizationi n butterflies?(3 ) Are hosts hiftsi nvolving closely Ehrlich and Raven (1964) noted several factors influencing relatedp lant speciesm ore common than shifts to more dis- the association between butterflies and their host plants, but tantly related plants'?( 4) Are there idenrifiableg roups of particularly stressedt he importance of the plant's secondary unrelatedp lants that often occur togethera s hosts?( 5) Is metabolic substances.T hey noted that many higher taxa of plant phylogenya morec onservativea specto fbutterfly-plant plants are characterized by distinctive secondary chemistry, associationsth an plant growth fbrm, or vice versa?( 6) Are and cited a number of examples in which related butterflies major host shiftsm ore commoni n woody-plant-feedingth an feed on related plants. They also cited examples where related in herb-f'eedinglin eages?( 7) Are tree-feedingb utterflyt axa butterflies are t'eeding on unrelated plants with chemical sim- associatedw ith a largern umbero f host plant cladest han are ilarities. These observations are consistent with, though not herb-feedingt axa? sufficient to demonstrate, the importance of plant chemistry. However, nobody has tried to determine whether and at what MErnoos taxonomic scales the tendency to feed on related plants is Unlesso therwises tated,a ll analysesh aveb eenc arriedo ut statistically demonstrable for butterflies as a whole, as op- usingt hec omputerp rogramM acClade( vers. posed to selected examples. 3.05,M addison and Maddison 1992). Although plant chemistry has been viewed as the prime factor governing the evolution of butterfly-host plant asso- Phylogenies ciations (e.g. Feeny 1915, 1976, l99l; Jermy 1916, 1984 Scriber and Slansky 1981: Berenbaum 1983; Zangerl and The plant phylogenyu sedi n this study follows the rbcl- Berenbaum 1993; Fiedler 1995b), other aspects of the host, baseda nalysiso f seed plant relationshipsb y Chase et al. not necessarily well correlated with phylogeny, might also (1993).T hey perfbrmedt wo diff'erents earchesu sings lightly have a large eff'ect (Benson et al. 1975; Smiley 1978; Price different taxon sampling and weighting procedures.T hese et al. 1980; Courtney 1984; Bernays and Graham 1988; An- searchesp roducedv ery similart rees.W e havef br the present derson 1993). One example is host growth form. Different analysisu sedt he treep roducedb y their search2 (or treeB ), growth forms can dominate in different habitat types, which which they judged to be the most reliable.C hasee t al. sum- have distinct combinations of microclimate, enemies, etc. marizedt heir findingsi n a simplifiedc ladogram( their {ig. 2) These require specialized adaptations, making it more dif- in which most terminal taxa were given informal names,r e- ficult for a butterfly to colonize a new growth form and habitat flecting their approximatec orrespondenceto groupingsi n than to colonize a new host plant with a different chemical previous classificationsT. his summary phylogeny is pre- composition in the same habitat. Herbaceous host plants may sentedi n a modified fbrm in figure 2 (Chasee t al. 1993). also require a different search behavior by ovipositing fe- With minor modificationsto be noted,w e haveu sedt heC hase males than do arboreal hosts. Thus, host growth fbrm might et al. terminalc ladesa ndn omenclaturea st he charactesr tates be more conserved than host clade membership, when both in our analysisa nd discussiono f butterflyh ost associations. are traced on the butterfly phylogeny. For better resolutionw ithin their "Rosid l" clade, excep- We might also expect that the propensity to shift between tionally important as butterfly-hostp lants, we have recog- host taxa would differ between butterflies feeding on diff'erent nized six subcladesf,o llowing the branchingp atterno f their growth forms. For example, Feeny (1916) hypothesized that "tree B," which we labeled "Rosid lA . . . F," Unlesso th- herbaceous plants are def'endedb y diverse "qualitative" tox- erwise specified,t he term "plant clade" in this paperr efers ins that require corresponding diverse physiological and be- to these terminal cladesa nd subcladesi n the Chasee t al. havioral adaptations in the attacking insect, while trees are phylogeny.T here are problems with this analysis,m ainly characterized by "quantitative" def'ensec onsisting of a lim- arising from the computationald ifficulties of analyzing a ited number of digestion-reducing agents such as tannins, dataseot f this size.I n a critiqueo f the analysis,B aum (1994) which do not require specialized detoxification tactics. Under notes that although it is very likely that Chasee t al. have this hypothesis, trees make up a chemically more homoge- not found the most parsimonioust ree, the final phylogeny nous group. Host shifts should thus be easier and more com- includem anyh igherl evel groupingss uggestedb y traditional mon between trees than between herbs. This would influence systematistsa,n d that event he unconventionapl lacementso f the relationship between phylogenetic and growth form con- somet axao fien fit surprisinglyw ell with morphologicadl ata. servatism, especially if tree feeding is common among but- In any case,i t is the most comprehensivea ttempts o far to terflies. Similar arguments have been advanced to explain an reconstructa phylogenyf or the seedp lants as a whole, and apparent association between tree feeding and polyphagy should be a better estimateo f the true phylogeny than one (Futuyma 1976; Fiedler 1995a), but these hypotheses has inferred fiom previoust axonomy.F or most of our analyses never been tested using phylogenetic methods. we haveo nly usedt he terminalc ladesi n this phylogeny,n ot In this paper, we have performed phylogenetic analyses of the deeperb ranchings,w hich may be lessr eliable.I n a few the interaction between butterflies and their host plants to casesw here host plant families were not included in the address the lbllowing questions regarding the patterns and analysisb y Chasee t al., their positionsw ere inferredf rom causes of host shifts, colonizations and specialization: (1) the classificationo f Cronquist( 1981). Thesep lant families 4 8 8 N. JANZ AND S. NYLIN

Trapezitina(e1 ) t Appendix 2, (the subfamilyl evel relationshipsa re showni n Hesperiina(e1 ) F i g . 1 ) . Megathymina(e1 ) Hesperidea There are perhapse venm ore uncertaintiesin the butterfly Pyrginae(1 ) (outgroup) phylogenyt hani n thep lantp hylogeny.P artso f thep hylogeny are poorly resolved. This is particularly true for the large Pynhopygina(e1 ) ^ ^ ^ l ^ t i ^ ^ ^ / i \ family Lycaenidaeb, ut alsot br Pieridae,w herew e havec ho- v u c i l d u i l r a s r / \ I sen to collapseb ranchesa bout which the phenetica nalyses Baroniina(e1 ) by Geiger (1980) and Ehrlich and Ehrlich (1961) were in con- if'- Parnassiina(e4 ) Papilionidaeflict. The basal structure of Nymphalidae is left unresolved Papilioniina(e5 5) whereH arvey (1991)c onflictsw ith Scotta ndW right (1990), Dismorphiina(e1 ) but the nymphalid taxonomic groupingsi n our phylogeny Coliadinae(1 ' 1) Pieridae follow thet axonomyo f Harvey ( 1991).W hereM iller ( 1987b) Pierinae(2 8) conflicts with Hancock (1983) on the resolutiono f species Riodininae(9 ) Riodinidae groupsi n Papilionini,w e havef ollowed Miller's morer ecent I Polyommatina(e6 8) study. We have also, when possible,t ried to estimatet he phylogenetic Theclinae(6 7) effect of the reconstructiono n our results. The choiceo f outgroupt o the butterfliesis somewhapt rob- Lycaeninae(1 Lycaenidae ) lematic.W hen tracingc haractere volutiono n a phylogenetic (1) Poritiinae tree,o ne shouldi deally use severalo utgroups,i ncluding the Curetinae(1 ) sisterg roup, to correctly reconstructa ncestralc haracters tates Libytheinae( 1) (Maddisone t al. 1984).I n our caset his criterion is difficult Apaturinae(1 5 ) to fulfill as the phylogeny of higher Lepidopterai s almost Danainae(1 1 ) completelyu nknown (Nielsen 1989).T he skippers( Hesper- Morphinae(5 ) iidae) are strong candidatesa s the sister taxon to the true Brassolina(e1 ) butterflies,b ut it has been suggestedth at the wholly South American group Hedyloideas hould be positionedb etween Satyrinae(4 ) Nymphalidae the true butterfliesa nd the skippers( Scoble 1986).H edylo- Rosid1 B Calinagina(e1 ) -i idea is poorly known (including its host plant affiliations) f: ilr:r::: ::::::::.*r:x.'. Charaxinae (18 ) No . l and its position is controversial.T he most recent analyses Heliconiina(e4 5) I actually f'avorH esperiidaea s sisterg roup to the Papilionoidea ! ves .'- Limenitina(e4 5) (Weller :1 and Pashley1 995;d e Jonge t al. 1996;S coble1 996). NN Equivocal ::::.., Nymphalina(e4 4) For this reasonw e have chosent o excludeH edyloideaf iom the study, but we test the effect that this may have on the FIc. 1. Simplified version of the phylogeny of Papilionoideaused in the analyses, showing the major relationships within butterflies. reconstructiono f the ancestrahl ost plant association. Total number of taxa in the phylogeny actually used in the analyses is given within parentheses behind each taxon name. Utilization of Host Plant Datct plant (Cannabidaceae, the clade Rosid 1B Fabaceae, Krameriaceae, Data host plant Moraceae, Polygalaceae, Rhamnaceae, Rosaceae, Ulmaceae, Urti- on utilization were collectedf rom several caceae, and Zygophyllaceae) is traced on the phyiogeny. The as- sources( Ehrlich and Raven 1964;C ommon and Waterhouse signment of states to branches follows optirnization on the complete 1972;L arsen1 974 , l99l1'S mart1 975;J ohnstona ndJ ohnston phylogeny. 1980;H igginsa ndH argreaves1 983;S cott1 986;d e la Maza Ramires 1987; DeVries 1987; Migdoll 1987; Miller 1987a1. Ackery 1988; Parsons1 991; Corbet and Pendlebury1 992; were Salicaceae( placedi n Rosid lA), Plantaginacea(eA s- Ebert 1993).B ecausei t is dif{icult to evaluatet he validity terid l), and Cactaceae(H amamelidl ). of literature data on host plant affiliations, false records cer- There has not yet been any comparablea ttempt to perform tainly exist. We have thereforet ried to be conservativeb, y a combined phylogenetic analysis of the butterflies as a excludinga necdotalh ost plant records.W e included a host whole. For this reasonw e haves ynthesizedre sultsf rom sev- plant record only if it was corroboratedb y two independent eral smaller studiesi nto a single phylogeny for all butterflies sources,i f the plant was recordedf or more than one species (Fig. I, Appendix 2). Sourcesf or the different parts of the in the butterfly ,i f there were recordso f more than one phylogeny,a ndt he type of evidencet hey presenteda, rel isted host plant genusf rom the samep lant family, or if it was the in Table 1. only plant recorded for this butterfly group. Atypical hosts The phylogenyh as beenr esolvedt o genericl evel in most with little supportw ere excluded.T he risk we therebyr un groups,t he exceptionsb eing groupsw ith little or no variation of erroneously excluding host plant data that are correct is, in host use( e.g.,S atyrinaew, hich hasb eenr esolvedt o tribal we believe, outweighedb y the advantageo f excluding a level) and groups with large variation in host use and for greater number of records that are incorrect. The host plant which a detailedp hylogenyi s available( e.g.,P apilio, which databaseis providedi n Appendix 3. has been resolvedt o speciesl evel). The level of resolution is likely to have some effect on the reportedp atterns( Sill6n- Character Coding Tullberg1 993,s eeR esults)T. he completep hylogenyc onsists Optimizing a complex characters uch as host plant utili- of 431 ingroup taxa, and is given in parentheticalf ormat in zation is problematic.O ne major issue is to treat multiple BUTTERFLIES AND PLANTS 489

T,q.el-El . Sources for phylogenetic information regarding the butterflies. Morph, morphological data; Behav, behavioral data; Tax, ,"-"""-t. tr""t'

' I a x o n / t u x a Level of resolution used by us Typc 01 data Referenccs Papilionoidea Butterfly families and choice of outgroups Morph Kristensen1 976 rRNA Martin and Pashley1 992 Morph * mtDNA Weller and Pashley1 995 Morph de Jong et al. 1996 Hesperioidea Subfamilies Morph * Behav Scott and Wright 1990 Papilionidae Subfarnilies, trrbes Morph Miller 1987b Parnassiini Genera Morph Hancock1 983 Graphini, Troidini Genera, species Morph Milier 1987b Species Morph Munroe 1961 Grctphiurn Species Morph Saigusae t al. 1982 Papilionini Genera and species groups Morph Munroe 1961 Morph Hancock 1983 Morph Miller 1987b Pctpilio Species lnTDNA Sperling 1992,1993 Pieridae Subfarnilies Morph Ehrlich and Ehrlich 1967 Allozymes Geiger 1980 Tribes, genera Morph (Tax) Smart 1975 Riodinidae Genera Morph (Tax) Smart 1975 Lycaenidae Subfamilies Morph Ackery 1984 Tribes, genera Morph Eliot 1973 Morph (Tax) Srnart1 975 Nymphalidae Subfamilies, tribes Morph Scott and Wright 1990 Tribes, genera Morph (Tax) Harvey 199I Genera Morph (Tax) Smart 1975 Danainae Genera Morph Ackery 1984 Genera Morph Ackery and Vane-Wright 1984 Acreini Species groups Morph Pierre 1987

associationsA. lthough mostb utterflyt axai n our analysisa re hosts. In the following tests we have chosen different ap- restrictedt o one plant clade, 36Vou se plants belonging to proaches to this dilemma, depending on the problem. more than one clade.T here is no method fbr characterc oding capableo f handlingt his problemi n a completelys atisfactory Ancestral Host Plant Association way. However, several approachesc an be taken to work aroundt he problem, all with different problemsa nd advan- To estimatet he phylogenetics equenceo f butterfly-host tages. plant associationws e createda multistatec haracterw ith mul- One method is to simply optimize plant clade use as an tiple host use coded as polymorphisms.T o control for the unorderedm ultistate characterw ith multiple associations possibilityo f erroneousa ncestrasl tatea ssignmendt ue to the treateda s ambiguity.T his makesi t easyt o interpretp atterns fact that polymorphisms are only allowed at terminals in of multiple host use,b ut leadst o loss of infbrmation.I t does MacClade3 , we also optimizedt he samed ataa s a matrix of not help to code multiple associationsa s polymorphismsa, s binary characters. MacClade3 doesn ot allow polymorphics tatest o be assigned Therei s somec ontroversyo ver thed egreeo f certaintyw ith to internal nodesi n the phylogeny.A n alternativei s to op- which a characters tatec an be ascribedt o an internal node timize each plant clade as an independentb inary character. of a phylogeny (e.9., Frumhotf and Reeve 1994). Optimi- This allows easyi nvestigationo f specifich ost associations, zation of a charactert hat changest oo quickly relativet o the but makesi t difficult to interpretm ultiple host use and can branchingp atterno f the phylogenyw ill not resulti n a plau- leadt o falser econstructionosf nodesa sh avingn o association sible historicalr econstructionT. o addresst his problem, we at all. Currently the only way to correctly handle multiple performeds everalr andomizationsto test to what extento ur associationsw, hile allowing ancestralp olymorphisms,is to reconstructiono f the ancestrasl tatei s dependenot n the fre- codeh ost usea s a multistatec haractequ singa separates tate quencya ndd istributiono f characters tatesa monge xtantt axa for each host plant associationa nd, in addition, a separate and on the structureo f the butterfly phylogeny.T heset ests statef br each possiblec ombinationo f plant clade associa- fbllow the samel ogic and use the samen ull hypothesisa s tions, and then assignt ransfbrmationw eights using a step the "permutationt ail probability" test (Archie 1989; Faith matrix (seeA ppendix 4 and Maddisona nd Maddison 1992, and Cranston1 991). p. 83). A difficulty with this approachi s that the numbero f First, if host plant utilization changest oo fast relative to statesg rows exponentiallyw ith the numbero f plant groups. speciationt here will be no phylogenetics ignal in this char- Limitation on how many statest he currentv ersion( 3.05) of acter and ancestralh osts cannot be reconstructedb y opti- MacCladec an handler estrictsa nalysist o a maximumo f four mization on the butterfly phylogeny (Frumhoff and Reeve plant groupss imultaneouslyT. he methodc an thereforeo nly 1994).R econstructiona t the ancestranl ode will in that case be usedo n eithera broad scaleo r on subsetso f the butterfly be a function of the relative frequencieso f currenth ost as- 490 N . J A N Z A N D S . N Y L I N sociations.I f our reconstructionis an artifacto f a host plant groups.H owever,u sing the numbero f plant families in our associationb eing commonb ecauseit is for somer easone asy Appendix I as a crude measureo f diversity,t his assumption to colonize,t hen this associations houldb e more or lessr an- seemsr easonablet:h erea re5 4 rosid families.4 7 asteridf am- domly distributeda mong butterfliest oday.I t follows that if ilies, and 51 familieso f others eedp lantsa mongt he butterfly the reconstructiono f the ancestranl ode usingt he actuald is- hosts. tribution of character states among living taxa differs sig- As there is always variation betweenr esolutionsw e first nificantly from that under a random distribution, given the neededt o know whethert he found differencesw ere consis- samef requencieso f characters tates,t hen the ancestrahl ost tent over the resolutionsI.n this test and in similar testst hat assignmenti s more likely to reflect the actuale volutionary follow, we have used an approacht o the problem of irres- history. To addresst his question,a randomizationt est was olution similar to Losos( 1994)a nd Martins (1996).F or each carriedo ut by reassigningth e observeds tates1 000t imes at randomr esolutionw e calculatedt he pairwised ifferencesb e- random and reconstructingt he ancestraln ode eacht ime. The tween numbero f colonizationso f rosids.a sterids,a nd other stateso f all extantt axa were randomized,i ncludingt he out- seedp lantsf rom the ancestracl ladeR osid 1B. For instance, group. if therew ere 35 colonizationso f otherr osids.2 3 of asterids. Second,a s the butterflyp hylogenyi s uncertaino n several and 17 of others eedp lantsi n a given resolution,t hep airwise points, we testedh ow sensitiveo ur reconstructiono f the an- diff'erenceb etweenr osids and asteridsw ould be +12. be- cestralh ost wast o the treet opologyb y partiallyr andomizing tweenr osidsa nd others eedp lants* 18 , andb etweena sterids tree structure.F irst, we completelyr andomizedt he structure and other seedp lants +6. The distribution of thesed iffer- within eachf amily, only retainingt he structureb etweenf am- encesw ere examinedt o get an estimationo f the numbero f ilies. We thenr epeatedth e randomizationso f branchesw ithin resolutionsw heret heh ypothesisw asc orroboratedo rrefuted. families,w hile retainingt hem ostb asalb ranchi n eachf amily. ln our examplea ll differencesw ere in accordancew ith the In eachc ase,t he ancestracl haracters tatesw erer econstructed hypothesisa, s they werea ll positive.N ote that this procedure for 1000 randomizations. only gives an estimateo f the consistencyo f the found dif- Afier the above analysish ad indicatedt he ancestralh ost ferenceo ver the examinedp hylogeniest,h e magnitudeo f the plant clade,w e further separatedth is cladei nto smalleru nits diff'erencem ay still be small enought o be a resulto f chance. in an attempt to determinet he ancestralp lant association We therefore performed chi-square tests of fit to the null more closely. hypothesism entioneda bove,b oth on the averagen umbers of colonizationso ver the 100 resolutionsa nd on each indi- Colonization and PhvlogeneticD istance vidual resolution. ln the other analysisw e investigatedc olonizationsw ithin To test the prediction that host shifts and colonizationsa re and betweent he two large sisterg roups "rosids" and "as- more common between closely related host taxa we per- terids." We comparedt he number of colonizationst o and formed two different analyses.F irst we analyzeds hifls from from plantsb elongingt o the sameg roup (rosidso r asterids) the presumablya ncestralc lade (Rosid 18) to three major with the numbero f colonizationsb etweent heseg roups,m ak- plant groups.T his test was carried out on very broad plant ing no assumptiona bout the ancestralh ost plant. To make categoriesn, amely colonizationso r shifts from Rosid lB to the test as conservativea s possible,w e excludeda ll changes the threem ajor plant groups" rosids" (othert han Rosid 1B), within the terminal cladeso f the Chasee t al. (1993) phy- "asterids" and "other plants" (Chasee t al. 1993),m aking logeny (Rosid 1-3 and Asterid 1-5), countingo nly changes the step matrix coding describeda bove in CharacterC oding betweent heser atherl argec ladesa s changes" within rosids" appropriate.E ach of these groups and each possiblec om- and "within asterids."T he numberso f unambiguousc olo- bination of them were treateda s separates tates,w ith 15 in nizationsw ere testedb y goodness-of-Iita, gainstt he null hy- all. Gains and lossesw ere given equalw eight,t hat is, a gain pothesist hat the numbero f changesw ithin asteridso r rosids and a loss of one of thesep lant groupsb oth carried a cost andb etweent hem shouldb e equal.T he rationalei s that since of one (Appendix 4). Becauses tep matricesc annotb e used we assumet hat the two groups are approximatelye qually with unresolvedt rees,t he test was carried out on 100 phy- diverse.i f colonizationsa re random, half should be to the logeniesw ith randomly resolvedp olytomies. otherm ajor clade.A s the total numbero f colonizationsc ould The test hypothesisp redicts that shifts from the ancestral be summedu sing binary characterst,h ere was no need for host plant should most commonly be to plants in the same stepm atrix coding and hencen o needt o resolvep olytomies. major clade (rosids)a nd leastc ommonly to plantsm ost dis- To identify groups of unrelatedp lants that often occur to- tantly related to the original host (other seedp lants).T he gethera s butterfly host plants, we constructeda seedp lant numbero f unambiguousc hangesf rom feedingo n Rosid lB phylogenyu sing only presenceo r absenceo f butterfly taxa to feeding on the three groups were counted using the step as charactersu, sing a methods imilar to what is often used matrix and the "chart statec hanges" option in MacClade. in cospeciations tudies( e.g.,P atersone t al. 1993).T hesew ere We testedt hesen umbersa gainstt he null hypothesist hat col- analyzedu sing PAUP 3.1.1 (Swofford 1991) with detault onizations should be equally distributed among the three "factory" settings( simple addition sequenceo, ne tree held groups (one-third to each), assumingt hey are of approxi- at each step during stepwisea ddition,t ree bisection-recon- mately equal diversity and thus provide equally sized "tar- nection[ TBR] swappinga lgorithm,M ULPARS option in ef- gets" for random colonization. Accurate and meaningful fect, no topological constraints).W e compared the plant measureso f diversityf or theseg roupsa rev ery hardt o obtain, groups suggestedb y this analysisw ith the groups in the as thev do not easilv translatet o the traditionalt axonomical Chasee t al. (1993) phylogeny.T he butterfly charactersa nd BUTTERFLIES AND PLANTS 491 plant taxa fbllow the matrix showni n Appendix 3. We used were included in the analysis( i.e., when a colonizationi s a hypotheticala ncestorw ith zero (no butterfly associations) followedb y specializationo n then ovel plant),w e only count- as outgroup,a s the origin of angiospermsp redatest he but- ed a shift if no speciesin the butterflyt axon hasr etainedt he terflies,a nd we did not want to put a constrainto n what plant old associationS. hiftsw erec ountedu sings tepm atrix coding groupsc ould be united by the analysis.W ell-definedg roups of plant usea nd the "chart all changes"o ptioni n MacClade. in this analysist hat are not supportedb y the Chasee t al. The concentrated-changetes st can only be used on com- phylogenyw ere interpreteda s host shift tendenciesn ot cor- pletely resolvedp hylogenies,s o the test was iteratedo ver respondingt o plant phylogeny. 100 butterfly phylogeniesw ith randomly resolvedp olyto- mies.M acClade( Maddisona nd Maddison 1992)u sesa sim- Plant Ph1'logen\v ersusG ron-thF ornt ulation algorithmt o calculatet he statisticso n largep hylog- enies and our calculationsw ere basedo n 1000 such simu- To assessw hetherp lant phylogenyo r growth form is the lations for each randomly resolvedt ree, using the default more evolutionarilyc onservativea specto fbutterfly hostu se, settingso f the program( allowing either statet o be ancestral we comparedt he number of stepsn eededt o trace each on and using actualc hanges). the butterflyp hylogeny,u sing a coding that assignsa n equal As an alternativet esto f the associationb etweent ree-feed- number of statest o both features.T he categoriesw e used ing and tendencyt o host shift, we counted the number of were for plant groups, "rosids," "asterids," and "other terminalp lant cladesf rom the Chasee t al. (1993)p hylogeny plants," and for growth forms, "herbs," "vines," and "trees that were useda s hostsb y the terminal taxa in our butterfly and shrubs." Plant group and growth form were coded as phylogeny.I f this measureo f "host used iversity" is treated multistate charactersw ith one state for each category and a as a continuousc haracteri,t s correlationw ith tree feeding separates tatef or eachc ombinationo fcategories( sevens tates can be assessedb y the approacho f independentc ontrasts in total). Transformationw eights were assignedu sing step (e.g. Fefsenstein1 985; Pagel 1992; Purvis and Rambaut matrices( seea bovea nd Appendix 3). The numberso f steps 1995).A s the butterfly terminal taxa are most often genera, neededt o trace the two charactersw ere averagedo ver 100 taxa with multiple host plant associationms ay represenct ol- phylogeniesw ith randomly resolvedp olytomies,a nd tested lectionso f speciess pecializedo n differentp lants,p olyphagy by goodness-of'-faitg ainstt he null hypothesiso f equaln um- of individual specieso, r both. In any case," host used iver- bers,t hati s, thatp lantg rowth form andp hylogenya ree qually sity" should approximatet he numbero f host colonizations conservativea spectso f the association.T o investigatet he that havet akenp lacew ithin that taxon,r egardlesso f whether consistencyo f the diffbrencesin the numbero f statesn eeded theseh ave led to increasesin host rangeo r to divergencein to tracet he two characterst.h eir distributionsw eree xamined host plant usea mongs pecies. over the randomr esolutions( seea bove).T his analysisc ould Thesed ataw erea nalyzedw ith ComparativeA nalysisu sing potentiallyb e influencedn ot only by the diversityo fthe plant IndependentC ontrasts( CAIC; Purvis and Rambaut 1995), clades( seea bove),b ut alsob y the "availability" ofdiffbrent using the "brunch" algorithm, which is designedt o test if growth forms for colonization. However, at least on the fam- changesi n a discretec haracter( like tree feeding)a re asso- ily level, the distributiono f growth forms does not seemt o ciated with changesi n a continuousc haracter( like host be alarminglyu nequal,a mongt he familiesl istedi n the Cron- range).T his test differs fiom the concentratedc hangest est quist systemo n the Flowering Plant Gatewayw ebsite,2 82 in simply testingw hetherc hangesin two charactersa rec or- containt reeso r shrubs,9l vines,a nd 208 herbs( Watsona nd related,w ithout identifying one charactera s logically inde- Dallawitz 1992). pendento r causative.O ne advantageo f CAIC is that it has The characters tatesa re very unequali n frequency,w ith an algorithmf or handlingp olytomies.D etailso f the testsc an strongb ias toward feedingo n the plant group "rosids" and be fbund in Pagel (1992) and Purvis and Rambaut( 1995). the growth form "trees." The questiont herefbrea riseso f The contrastsg eneratedb y CAIC were testedb oth qualita- whethert here is a different tendencyt o shift betweenp lant tively with a sign test or quantitativelyw ith a one-samplet- groups while feeding on trees or to shift betweeng rowth test (seeH oglund and Sill6n-Tullberg1 994). forms while feedingo n rosids. CAIC usese xplicit assumptionsa boutb ranchl engthsa nd We usedt wo procedurest o test this. First, the possibility offbrs two default alternatives:a ll branch lengthsa ssumed that changesi n plant clade use are more likely to occur on equal (correspondingto a punctuationaml odel of evolution) a certain growth form (and vice versa)w as testedw ith the or ageso f taxa assumedp roportionalt o the number of in- concentrated-changetes st (Maddison 1990; Maddison and cluded species( correspondingto a gradualm odel of evolu- Maddison 1992).S pecificallyw e testedi f major host shifts tion), using an algorithm by Grafen (1989). As we had no were concentratedto branchesr econstructeda s woody-plant informationo n branchl eneths.w e usedb oth. 1-eedersa,n d if shifls betweeng rowth fbrms are concentrated to branchesr econstructeda s feedingo n rosids.T hus the test Rgst,'L.ls was repeatedtw ice, using "plant group" and "growth form" Patterns in turn as the dependentv ariable.A s the test cannoth andle o.fH ost Plant Utilization multistatec haractersth e charactersw ere recodeda s binary. Butterfliesu se almost all major seedp lant families, and All taxa that use rosids were coded as rosid feeders( regard- even a few nonseedp lants, some speciesd o not even 1-eed lesso f what elset hey feed on). The others tatet husr epresent on plants( Fig. 2). Neverthelessa,s w asp ointedo ut by Ehrlich butterfly taxa that do not feed on rosids. Growth forms were andR aven( 1964),t he useo fplant cladesa ppearsn onrandom. treatedt he same way. To ensuret hat only completes hifts Somef amilies are heavily utilized by many butterfly groups 492 N . J A N Z A N D S . N Y L I N

Cycads (4) Luthrodesi n Polyommatinae)O. ther examplesi nclude As- Ginkgo (0) teraceaew, hich, consideringit s size,i s also relativelyr arely Pinaceae( 3) usedb y butterflies. Furthers upportingE hrlich and Raven's( 1964)c ontention otherc onifers( 1) thatr elatedb utterfliest endt o feedo n relatedg roupso fplants, Gnetales( 0) the Chasee t al. (1993) phylogenys uggeststh at someo sten- Ceratophyllum (0) sibly disparateh ost plant assemblageasr e more phylogenet- ically homogenoust han previoust axonomy suggestedF. or paleoherb' l (16) example,u nder the classificationo f Cronquist (1981), the Laurales( 29) diverseh ost list of the nymphalidt ribe Nymphalini includes monocots( 36) as dominant themes three families in the order Urticales Magnoliales(2 0) (Hamamelidae)t,w o in Rosales( Rosidae),o ne in Salicales (Dillenidae),a nd two in Fagales( Hamamelidae)I.n addition paleoherb2 (0) therea re speciesf eedingo n Ericales( Dillenidae),A sterales. ranunculids(8 ) (Asteridae)R hamnales(R osidae)M, alvales( Dillenidae),a nd hamamelid1 (9) Liliales( Liliidae).I n the Chasee t al. (1993)a nalysish, ow- ever, most groups; (Rosales), hamamelid2 (0) the tiequently used Rosaceae Urticales, Salicales,F agales and Rhamnales,f ormerly in Gunnera (0) three subclassesa.l l fell within the Rosid I subclade.w hile caryophyllid(s4 4) Malvales belongedt o the sisterg roup Rosid 2 and Grossu- asterid5 (3) lariaceae( Rosales)t o Rosid 3. Thus, the nine plant orders of main nymphalidh osts,p reviouslys catteredo ver five sub- asterid4 (2) classesp robablyh ave relativelyc lose affinities. asterid3 (38) asterid2 (32) Ancestral Host Plant Association asterid 1 (90) The optimizationo f host use as a multistatec haractera nd rosid3 (13) asb inary charactersb oth suggest hat the ancestrahl ostp lant r o s i d2 ( 1 1 0 ) was located within the clade we have called "Rosid lB" rosid1 F (0) (Fig. 1, seeA ppendix I for a list of includedf amilies).T his was the only cladet hat evenc amec loset o being drawnb ack rosid1 E (22) to the root of the butterfly phylogeny.P lant taxa such as rosid1 D (4) Rosid 2 and Asterid I (seeA ppendix l), also usedb y many rosid1 C (32) butterflies( Fig. 2), are apically distributedo n the butterfly phylogeny. rosid1 B (170)o Hesperioidea,t he most likely sister group, was used as rosid1 A (92) outgroupf or the optimizations.T his group is mainly asso- Ftc;. 2. Phylogenetic relationships among seed pltrnts fiom Chase ciatedw ith Fabaceae(R osid lB) and monocotyledonsI.f the et al. (1993). The names Ros 1A to E are assigned by us to subclades sisterg roup to Papilionoideain steadt urns out to be Hedy- identified by Chase et al. (1993). Numbers of associated taxa in the loidea,t he reconstructionw ill be somewhatm ore uncertain. butterfly phylogeny are indicated after plant clade names. The dot denotes presur.neda ncestral host plant clade. Host plant data on Hedyloideaa re scarceb ut indicate that Sterculiaceae( Rosid 2) is the most important host plant group. Sterculiaceaies a membero f Malvales,s uggestedb y (e.g.,F abaceae)w, hile othersa re dorninanth ostsf br partic- Ackery (1991) to be the ancestrahl ost group for butterflies. ular subsetso f butterflies.I n Papilionidae,A ristolochiaceae However,e veni n this caseo ur optimizationss uggesth at the (Paleoherb1 ) and ( Rosid2 ) arec learlyd ominating colonizationso f Rosid 2 by Hedylidae,s ome Hesperiidae, themes,w hile Pieridaea re typically associatedw ith plants and someP apilionidaea re independenet volutionarye vents in Brassicaceaaen d relatedf amilies, such as Capparidaceae and that Rosid lB is the most probablea ncestrahl ost plant (Rosid 2). Fabaceae( Rosid 1B) are common hosts in Rio- group. The same is true if Hedyloideai s used as the only dinidae and Lycaenidae,a nd perhapsU rticales (Rosid 1B) outgroupo r if no externalo utgroupi s useda t all. togetherw ith Passifloraceaaen d relatedf amilies (Rosid 1A) The reconstructeda ncestraln ode appearst o be signifi- could be said to be predominanth osts in Nymphalidae,a l- cantly different from reconstructionsu nder random character thought his is a very largeg eneralizationO. ther,e quallyc on- statea ssignmentR. osid lB f-eedingw as ascribedt o the an- spicuousp lant groups.h owever,a reo nly rarely useda sh osts cestraln ode in only 49 of 1000r andomizations(P : 0.049). by butterfliesT. he very largef amily Orchidaceaeis only used The relatively high proportion of Rosid 1B feeding on the by a few generai n Lycaenidae( Hypolycaenaa nd Chliaria in phylogeny (39c/o)is thereforen ot a sufficient explanationf or Theclinae)a nd Nymphalidae (Faunisi n Amathusiinae).L ike- its reconstructiona s the ancestrasl tate.M oreover,a s the dis- wise, gymnospermsa re usedb y only a handful of speciesin tributiono f Rosid 1B f'eedersin the phylogenyi s nonrandom, Pieridae( Neophasiai n Pierinae)a nd Lycaenidae( Callophrys, it is unlikely that the widespreadu se of this group (and the Eumaeus. and Stn'mon in Theclinae. and Theclinesthesa nd reconstructiono f the ancestrals tate)s hould simplv fbllow BUTTERFLIES AND PLANTS 493

120 50 1 0 0 a 4 0 c @ .F c o BO . N ^ ^ { t l , N 6 " " C o o 60 o o o n O 40 c o) > 1 0 20

RosidsA sterids Other Within Between plants

Ftrt. 3. Association between likelihood of colonizations and phylogenetic distance. (a) Mean number of colonizations from the ancestral clade Rosid lB to three large plant groups of approximately equal diversity, averaged over 100 phylogenies with randomly resolved polytomies (see text). (b) Number of colonizations of plants belonging to the same major clade, that is, from a rosid to a rosid or an asterid to an asterid, compared wrth the number of coionizations liom an asterid to a rosid or vice versa. from the fact that Rosid lB might fbr some reason be easy In only fbur resolutionso ut of 100 there were more colo- to colonize. A historical explanation of the utilization of at nizationso f "other plants" than of asterids.T hus, we con- least this plant group is therefbre more probable. clude that the differencesw ere consistento ver the phylog- It follows from this result alone that the phylogenetic struc- eniesw ith randomlyr esolvedp hylogenies. ture must be more important fbr the reconstruction of the The numbero f colonizationso f otherr osidsa veraged3 5.0 ancestral association than simply frequency of use of each (range:2 5-49) over 100p hylogerriesw ith randomlyr esolved plant taxon. A thousand random resolutions of all branches polytomies.C olonizationso f asteridss, isterg roup to the ros- within the butterfly families (keeping only the structure be- ids, averaged2 3.2 (16-33), while colonizationso f any plant tween families) generated only 30 cases where utilization of outside the rosids and asteridso nly averagedL 7.4 (8-23). Rosid lB was ascribed to the butterfly ancestor (P : 0.03), These averagen umbers depart significantly from the null indicating that this reconstruction is a very improbable out- hypothesiso f equal proportions( Xr : 6.3S, df : 2, P : come if we regard the phylogenetic relationships within but- 0.041). Howeveq among the 100 actual resolutions,t here terfly families as completely unknown. Howeveq it is enough were 39 where the diffbrencesi n colonizationsp rovedn on- to retain the most basal branch in each family and randomly significant.T hus we cannot safely concludet hat coloniza- resolve all branches above to make Rosid lB the most likely tions are not evenly distributeda mong groupsi n the "true" ancestral association. This means that the reconstruction of resolution,e ven though they are consistenti n direction. the basal branches within each family is critical for the re- Comparisonso f shilisj ust betweenr osidsa nda steridsg ave construction of the ancestral butterfly-host plant association. strongers upportt o theh ypothesis( Fig. 3b). The total number Atler optimizing a multistate character where Rosid lB of colonizationsb etweent heseg roups accountedf or 62 of had been further divided into two subclades (1B1: Fabaceae 173u nambiguousc hangeso n the phylogeny,o r 35.87ow, hile and Polygalaceae; 182: remaining families, see Appendix l) 11I shifts occurreda mong plant cladesw ithin rosids or as- indicated that the ancestral host plant family most likely was terids( 1r - 13.88,d f : I, P < 0.001).T his was a very Fabaceae (as Polygalaceae is a very minor host). The result conservativet est, as shifts within the terminal plant clades of this more detailed analysis was not used further, and for of Chasee t al. (1993)w eren ot counted. this reason it was not put through the randomization tests Ehrlich and Raven (1964) noted that some groupso f ge- described above. nealogicallyu nrelatedp lantso fteno ccurredt ogethera sh osts, suggestinga n underlyingc hemicalc onvergenceO. ur analysis Colonization cmd Phylogenetic Distance using the butterfly cladesa s characterss uggestsa dditional A comparison of the number of unambiguous changesf iom such convergencese xtendingE hrlich and Raven'so bserva- what we suggest to be the original host plant clade (Rosid tion. The most notableg roupingo f unrelatedp lantsb y but- lB) to the three large groups of roughly equal size, "other terflies was utilization of plants in Asterid I togetherw ith rosids," "asterids," and "other seed plants," gives some plants in Rosid I and 2. Twenty-sevenb utterfly taxa in our support to the hypothesis that the probability of a host shift analysisu se plants in all thesec ladesa s hosts,r epresenting is related to the phylogenetic distance between the plant between l8 and 23 independente volutionary events.T he groups involved (Fig. 3a). There was no single resolution plant families usedw ere most often Rosaceaea nd Ulmaceae where the number of colonizations of rosids was fewer than in Rosid 1, Rutaceae,T iliaceae,a nd Sapindaceae/Sterculi- the number of colonizations of asterids and "other plants." aceaei n Rosid 2, and OleaceaeR, ubiaceaea, nd Verbenaceae 494 N . J A N Z A N D S . N Y L I N

24 40

l l J C @ 1 8 a 30 a G E C @ t c 25 6 c c t t o 20 E 'od c C) I o t c c) E 6 C 1 0 3 5 0 0 Onw oody Onh erbaceous Positive plants plants b Ftr;. 4. Association between feeding on woody plants and host shifts or colonizations. (a) Mean number of shifts between rosids and other plants when feeding on woody versus herbaceous plants, averaged over 100 phylogenies with randomly resolved polytomies. (b) Independent contrasts between tree- and nontree-feeding lineages. Positive contrasts are those in which the tree-feecling iineage uses more_h osts clades per butterfly taxon. Black bars represent results using scaled branch lengths, and shaded bars represeni resulLsu sing equal branch lengths (see text).

in Asterid l Other, less strongly supported groupings in our Not surprisingly,t herew as no evidencea t all for the com- analysis that differed from those of Chase et al. (1993) in- plementaryh ypothesis,t hat feeding on rosid plants should clude Rosid 3 united with Asterid 3 (5-6 independent evenrs), discourage shifts between different growth forms, as in no and Asterid 2 united with ranunculids (3 independent events). randomizationw as the P-valueb elow 0.1. The independenct ontrastste sts howeda stronga ssociation Plant Phy-log€n\t urrtu"- Growth Form betweent ree feedinga nd useo f an increasedn umbero f host plant cladesb y butterfly taxa, giving further supportt o the The mean number of steps needed to trace the character hypothesist hat colonizationso f new host plants are facili- "major plant groups" (rosids, asterids and other plants) on tated by f'eedingo n trees (Fig. 4b, exemplifiedi n Fig. 5). 100 phylogenies with randomly resolved polytomies was This relationshipw as highly significantu nderb oth rhe sign 217.1 (range 210-226), while it was 168.6 (range 163-174) test and the one-sample/ -test( Table2 ). The trend was con- fbr the character "growth form." This result departs signif- sistento ver all butterflyf amilies,a lthoughn ot significanti n icantly from the null hypothesis that changes in growth form all families,p robablyd ue to small numberso f contrastsT. he and plant clade are equally common (X2 : 0.10, df : l, P two algorithmsf or estimatingb ranchl engthsg avev ery sim- : 0.014). The difference is consistent as the distributions of ilar results. numbers of steps for the two charactersa crossr andomizations A possiblep roblemw ith theset estsi s that diversityo fhost do not overlap. This indicates that, even at this low level of usea nd taxon size arec onfoundedO. ne can not entirelyr ule plant group resolution, growth form may in fact be a more out the possibility that tree-feedingt axa in generalc ontain evolutionarily conservative aspect of butterfly-host plant as- more speciest hat herb-feedingta xa. sociations than plant phylogeny. There is, howeveq one plant clade and one growth fbrm DrscussroN that are much more widely utilized than the others. Rosids are used as host plants by 69c/oo f the butterfly taxa, while Ehrlich andR aven's( 1964)m ain observationst,h atr elated asterids and other plants are used by only 30o/o and 35Vo, butterflieso ften feed on relatedh ost plants,a nd that some respectively. Likewise, 73% of the butterfly taxa feed on trees plant groups are commonly used by butterfliesw hile other and/or shrubs, whrl,e 2lo/o f'eed on vines and 31Vo feed on largep lant groupsa re not, are upheldb y our reanalysis( Fig. herbs. Thus, a higher tendency to shift between plant clades 2). As others have noted (e.g., Jermy 1976, 1984), these when feeding on trees would increase the apparent average patternsc an be explainedb y sequentiacl olonizationo f re- rate of host plant shifi. Likewise, a lower tendency to shift lated plant groupsw ithout the coevolutionaryt wist that the between growth forms when f'eedingo n rosids could decrease plantsh ave escapeda nd radiatedi n the absenceo f butterfly the apparent rate of shifts among growth forms. f'eeding.U nder a coevolutionaryi nterpretation,u nderutili- Interpretation of the concentrated changes test is compli- zationo f somep lant groupsb y butterfliesw ould be ascribed cated, as the results fiom the 100 random resolutions poly- to chemical det'encese volved to exclude butterflies fiom tomies vary substantially (Fig. 4a). For a majority of reso- feeding.l n the light of our results,h owever,i t is not likely lutions (61), the number of major host shifts taking place on that such groups,w hich are also much older than the but- branchesc haracterizedb y feeding on trees and shrubs is high- terflies,h ave ever been usedb y butterflies.S ome may have er than expected by chance (P < 0.05). However, the P-values evolvedc hemicald efensesa gainsto ther herbivorest hat are ranged from 0.002 to 0.349. equallye ffectivea gainstb utterflies.O thers,e speciallyt hose BUTTERFLIES AND PLANTS 495

Battus TeeLn 2. Relationship between tree l'eeding and number of plant Pharmacophagus clades used by butterfly taxa, calculated using independent con- -. Euryades trasts. Two different algorithrns have been used to estimate branch -. Cressida lengths: (l) all branch lengths equal; or (2) branch lengths scaled so that the ages of taxa are proportional -. Pachilopta to the number of species t . : r c o n t a i n * LOSana -. Trogonoptera a-lest Q i r r n t e c r '. Troides B ranch Prob- -. Panosmia pos/rrcg Proh- Taxon Ic n g lh \ / df ability conrrasrs ability -. Atropaneura - Parides Papilionoidea Equal 4.23 62 < 0.001 34/8 < 0.001 Scaled 3.77 62 < 0.001 < '. Papilioi ndra 36/6 0.001 Papilionidae Equal 1.27 '. brevicauda 5 0.2s9 2/r l 000 Scaled 1.62 5 0.1 66 3/O 0.248 -. Papiliom achaon Pieridae Equai l 4l 7 * Papiilop olyxenes 0.201 2/o 0.480 Scaled |.24 7 N 0.2s5 3/O 0.248 Papilioz elicaon * * 0 / 0 * -. Riodinidae Equal 0 Papiliox uthusg roup Scaled * 0 * l/0 * s Papiliop rincepsg roup Lycaenidae -. Equal 2.91 t 9 0.009 tt/3 0.061 Papilio drurya group Scaled 2.29 1 9 0.034 -' It/3 0.061 Papilio demolion group Nymphalidae Equal 2.93 26 0.007 t]/4 0.009 rc Papilio'fuscusg' roup Scaled 2.88 26 0.008 18/3 0.002 -. Papilio'protenorg' roup 'i Could not be calculated. ru Papilio'polytesg' roup - Papilioh eraclidesg roup -- Papilioc hilasag roup - Papilioe lepponeg roup fects butterfly host selection.A nother prediction,t hat may ru Papilio pyrrhostictag roup be easiert o test,i s that if thesep lantsd o shares ucha trait, -. Papiliot roilus they should also be linked in the diets of other groups of -. Papiliop ilumnus phytophagousin sects. -. Papilio palamedes Consideringt he variationi n dominatingh ost taxa among - Papilioc anadensis butterfly families, it is somewhatr emarkablet hat the most - Papiliog laucus basal branchesi n each family feed on plants in Rosid lB, s Papiliom ulticaudatus such as FabaceaeU, rticaceae,U lmaceae,o r RosaceaeT. he N Papilioe urymedon Rosid I B cladei s alsob y f ar the most likely ro havei ncluded r Papilior utulus the ancestralb utterfly host. The characters tater andomiza- FIc. 5. Association between tree feeding and propensity fclr host tions strongly suggestt hat this pattern does reflect evolu- shifts and colonizations exemplified by Troidini and Papilio in Pa' tionary history, not this plant group being fbr some reason pilionidae. White branches indicate herb-feeding, black branches unusuallye asyt o colonize.E ven if Rosid 1B were easiert o woody-plnnt-feeding lineages. Number of plant clades from Chase colonize,t herec ould still be a historical et al. (1993) used as hosts by each butterfly taxon are indicated explanationa: ll but- below butterfly names. One of the two to three shifts between rosids terfliesm ay be literally preadaptedto the chemicala nd other and other plants, taking place on herb-feeding branches, and three traitso f thesep lants,s imply becauseth eir ancestorsfe d upon of the 13 32 such shitis taking place on woody plant-feeding plantsc ontainingt hem.F urthers ubdivisiono f Rosid I B cor- branches, are shown in this figure (bars on branches). The figure roboratesS cott's (1986) suggestiont hat Fabaceaew as the also shows three of the ,12 independent contrasts between herb- and tree-feeding taxa. In all 3 contrasts shown the tree-1'eedingt axa are most likely ancestrahl ost plant family. associated with higher numbers of plant clades than are the herb- The strongc onservatismo f butterfly associationw ith ma- feeding taxa. jor plant cladesd oesn ot precludef requents hifts amongr e- lated host species.I ndeed,m any butterfliesf eed on several specieso r generaw ithin the samep lant family, suggesting very distantlyr elatedt o the original butterflyh ost,m ay sim- that thereh aveb eenm any colonizatione ventsb etweenp lants ply not have been "discovered" yet by butterflies,a s such too closely relatedt o be distinguishedb y our analysis. distant colonizationsa ppear to have been rare. Given the Restrictiono f most colonizationst o relatedp lants could current stateo f knowledge,i t is impossiblet o give any def- reflectc onstraintso n geneticv ariationi n the capacityt o feed inite explanationt o thesep atterns. on novel hostp lants,m aking a shift to an ancestrahl ostp lant Our analysisa lso extendsE hrlich andR aven'so bservation morel ikely thant o a completelyn ovelp lant (Futuymal 99l). that some groupso f unrelatedp lantsr epeatedlyc o-occuri n There is some evidencef or this in chrysomelidl eaf beetles the host lists of butterfly taxa, which they interpreteda s re- (Futuymae t al. 1993,1 994, 1995)a nd in burterflies( N. Janz flectingc hemicalo r other underlyings imilarity.T he clearest and S. Nylin, unpubl.).I n fact, one of the examplesE hrlich predictionl iom our analysisi s that plantsi n Rosid I (most and Raven (1964) give in their papero n colonizationo f re- importantlyR osaceaea nd Ulmaceae),R osid 2 (most impor- latedp lant groupsb y relatedb utterfliesc ould,i n the lighr of tantly Rutaceae,T iliaceae and Sapindaceae/Sterculiaceaet)h, e new plant phylogeny,b e betteru nderstooda s a recolon- and Asterid I (most importantly Oleaceae,R ubiaceaea, nd ization of the ancestralh ost plant clade: the switch of one Verbenaceae)s hare some chemical or other feature that af- genus in Parnassiini( Htpermnestrd) from the dominant Ar- 496 N . J A N Z A N D S . N Y L I N istolochiaceae (Pal l) and Rutaceae (Rosid 2) theme to feed AsnnnsoN, R. S. 1993. Weevilsa nd plants phylogeneticv ersus ecologicalr nediationo f evolution of host plant associationsin on Zygophyllaceae (Rosid 1B), which Ehrlich and Raven Curculioninae( ColeopteraC, urculionidae)M. em. Entornol.S oc. claimed to be closely related to Rutaceae. If such recoloni- Can. 165:197-232. zations are common, it means that a high number of host AncHtn,J . W. 1989. A randomizationt est for phyiogenetici nfor- shifts could go unnoticed in a phylogenetic study, because rnationi n systematicd ata.S yst.Z ool. 38:239-252. they tend to shift back to the original host, thus making the Brur,t, D. 1994. rbcl- and seed-planpt hylogeny.T rendsE col. Evol. 9 : 3 9 - 4 1 . pattern it is on a finer overall look more conservative than BnNsoNW, . W., K. S. Bnows,.txn L. E,.G rr-er:nr.[ 975. Coevo- level. Or, put in another way, an opportunistic pattern of host lution of plantsa nd herbivores:p assionl lower butterfliesE. vo- plant utilization on a microevolutionary scale may well result lution 29:659-680. in a conservative pattern on a macroevolutionary scale. BrnsNseur,t,M . R. 1983. Coumarinsa nd caterpillars:a casef or coevolution.E volution 3'7:163 119. The fact that plant growth form was the more conservative Br.nNevs,E . A., eiru M. Gneuen. 1988. On the evolutiono f host aspect of host associationsi n our analysis suggestst hat other specificityi n phytophagousa rthropodsE. cology 69:886 892. factors than plant chemistry, such as habitat or community Bnoors, D. R., eNo D. H. Mcl-ulrN,qr. 1991. Phylogeny,e cology, structure, play an important role in shaping the large-scale and behavior.U niv. of ChicagoP ress,C hicago. patterns of butterfly-host plant association. A similar con- Cuesp,M . W., D. E. Sor-rrsR, . G. Or-rrsrneoD, . MonceN,D . H. LEs, B. D. MrsHr-ln,M . R. Drrv,c.r-rR-,. A. Pnrca,H . G. Hrlrs. clusion was reached in a recent phylogenetic study ofweevils Y. Qru, K. A. KnoN,J . H. Rnrrrc, E. Coxrr, J. D. Per.uen.J . and their host plants (Anderson 1993). The influence of R. MeNuenr, K. J. SvrsNl,qH, . J. Mrcuells, W. J. Knuss, K. growth fbrm is accentuatedb y the elevated rate of host shifts G. Kepor.,W . D. Cr-.a,nMr,. HnonEu,B . S. Grur, R. K. JaNsnr, in lineages feeding on trees, the most frequently used plant K. KrrrrC, . F WrraperJ, . F Smrlrr,G . R. FunNrenF, H. Slneuss, Xr,+Nc,G . M. Pr LrNKErrP, . S. Sor-rrs,S . M. SrvnNsr-NS,. E. growth form among butterflies, while changes among Q. Wrr,r-r,qr,rPs., A . Geoex, C. J. QurNN,L . E. Er;t-'r,cRruE,. Go- growths form appear to have occurred independently of the LnNSERGG,. H. LnapN,J n, S. W. Gn,rHr:v,S . C. H. Bar

press, eds. Coevolution of and plants. Univ. of Texas lution of two binary characters: are gair.rso r losses concentrated Austin. on certain branches of a phylogenetic tree? Evolution 14:539' 1976. Plant apparency and chemical def'ence. Rec. Adv. 55'7. Phytochem. 10:1-40. MenorsoNW, .P .,e ro D. R. MenorsoN1. 992M. acClndea:n alysis I 991 . Chemical constraints on the evolution of swallowtail ofphylogeny and charactere volution.V ers.3 .05.S inauer.S un- butterflies. Pp.315-340 in P. W Price, T. M. Lewinsohn, G. W. derland. MA. Fernandes, and W. W. Benson, eds. Plant- interactions: M.roorsoN,W . P., M. J. DoNoc;nr_'rA:,N DD . R. MeoorsoN. 19g4. evolutionary ecology in tropical and temperate regions. Wiley, Outgroupa nalysisa nd parsimony.S yst.Z ool. 33:83-103. New York. Menrm, J. A., eNn D. P. Pasrrr-Ev.1 992. Molecular svstematic FnlslNsrnr:-, J. 1985. Phylogenies and the comparative method. analysiso f butterfly family and some subfamily relarionships A m . N a t . 1 2 5 : 1 - 1 5 . (LepidopteraP: apilionoidea)A. nn. Entomol. Soc.A m. 85:127- FIr:olEn. K. 1995a. Lycaenid butterflies and plants: hostplant re- I 39. lationships, tropical versus temperate. Ecotropica I :5 1,58. MenrrNs,E . P. 1996.C onductingp hytogeneticc omparatives tudies 1995b. Lycaenid butterflies and plants: is myrmecophily when the phylogenyi s not known. Evolution 50:12-22. associated with particular hostplant preferences? Ethol. Ecol. Mtcoor-r-,I . 1987.F ield guidet o the butterflieso f southernA lrica. E v o l . 7 : 1 0 7 1 3 2 . C. Struik PublishersC, ape Town, South Africa. Fnur.tror,r-, P C., ,cNn H. K. Rnr.vu. 1994. Using phylogenies to MtLrsn, J. S. 1987a. Host-plantr elationshipsin the papilionidae test hypotheses of adaptation: a critique of some current pro- (Lepidoptera)p: arallelc ladogenesiso r colonization'l p o s a l s . Cladistics E v o l u t i o n 4 8 : 1 7 2 - 1 8 0 . 3 : 1 0 5 - l2 0 . FL:r 'uyr,r,qD, . J. 1976. Food plant specialization and environmental 1987b. Phylogenerics tudiesi n the papilioninae (Lepi- predictability in Lepidoptera. Am. Nat. l1O:285-292. doptera:P apilionidae).P h.D. diss. Arn. Mus. Nat. Hist.. Niw 1991. Evolution of host specificity in herbivorous insects: York. genetic, ecological, and phylogenetic aspects. Pp.431-454 inp. Mrrrnn., C., 'rNo D. R. Bnoors. 1983. Phylogenetica spectso f W. Price, T. M. Lewinsohn, G. W. Fernandes, and W. W. Benson. coevolution.P p. 65-98 in D. J. Futuym:ra nd M. Slatkin, eds. eds. Plant-anirnal interactions: evolutionary ecology in tropical Coevolution.S inauegS underlandM, A. and temperate regions. Wiley, New York. Mr-,NRoEE, . 1961. The classificationo f the papilionidae( Lepi- Fr-rrriyMA, D. J., M. C. Knlsr:, ANr) S. J. Scrrupr.r_p.1 993. Genetic doptera)C. an.E ntomol.S uppl.l 7:1-51. constraints and the phylogeny of insect-plant associations-re- Ntnt.srN,E . S. 1989. Phylogenyo f major lepidopterang roups.p p. sponses of Ophraella contmuna (Coleoptera, Chrysomelidae) to 281-294 ir B. Fernholm,K . Bremeq host plants of its congeners. Evolution 47:888-905. and H. Jornvall,e ds.T ie hierarchy of lif'e. Elsevier, Amsterdam, The Netherlands. F u l u y r v r . r ,D . J . , J . S . W e r - s H , T . M o n l o r , D . J . F u N r , e N o M . C . Nvr-rN,S ., .tNu N. WEollr-. 1994. Knusr. 1994. Genetic variation in a phylogenetrc context-re- Sexual size dimorohism and comparativem ethods.P p. 253 280 inP. sponses of two specialized leaf beetles (Coleoprera, Chryso- Eggleton R. V"ne- Wright, eds.P hylogenetics "npdr ess, nelidae) to host plants of their congeners. J. Evol. Biol. j:l2j- and ecology.A cadernic Lon- 1 4 6 . oon. Prcnl, M. D. 1992. F u r u v l r e , D . J . , M . C . K l l s n , e N o D . J . F u N x 1 9 9 5 . G e n e t i c A methodf or the analysiso f comparatived ata. J. Theor. constratnts on macroevolution: the evolution of host alfiliation Biol. 156:431-442. p. in the leaf beetle genus Ophroella. Evolution 49:79'l-809. 1994. The adaptationiswt ager.P p. 29-52 in Eggleton Getcrn, H. J. 1980. Enzyme electrophoretic studies on the genetic and R. Vane-Wright,e ds.P hylogeneticsa nd ecology.A cademic relationships of Pierid butterflies (Lepidoptera: Pieridae). I. Eu- Press,L ondon. r o p e a n t a x a . J . R e s . L e p i d . 1 9 : 1 8 1 1 9 5 . PensoNs,M . 1991. Butterflieso f the Bulolo-WauV allev. Bishoo GnepnN, A. 1989. The phylogenetic regression. Philos. Trans. R. Museum Press,H onolulu, HI. S o c . L o n d . B B i o l . S c i . 3 2 6 : 1 1 9 - 1 5 6 . Petrnsox,A . M., R. D. Gnev,,rNoG . P.W ellrs. 1993.p arasites. HaNcocx, D. L. 1983. Classification of the Papilionidae (Lepi- petrelsa nd penguins:d oes louse presencer eflect seabirdp hy- doptera): a phylogenetic approach. Smithersia 2:l-48. Iogeny?I nt. J. Parasitol.2 3:515-526. HARVEy, D. J. 1991. Higher classification of the Nymphalidae, PIennn,J . 1987. Syst6matiquec ladistiquec hez les Acraea (Lepi- Appendix B. Pp. 255-273 in H. F Nijhout. The developmenr doptera,N ymphalidae).A nn. Soc. Enrornol.F r. 23 ll-2j. and evolution of butterfly wing patterns. Smithsonian Institution Pr

sematic butterfly larvae: a phylogenetic analysis. Evolution 42: Asterid 3: Diapensiaceae, Ebenaceae, Epacridaceae, Ericaceae, 293-305. Myrsinaceae, Primulaceae, Sapotaceae, Symplocaceae, Theaceae. of biased inclusion of taxa on the cor- 1993. The etfect Asterid 4: Alangiaceae, Araliaceae, Cornaceae (Conlus), Nyssa- in phylogenetic trees. Evo- relation between discrete characters ceae, Hydrangeaceae. l u t i o n4 7 : 1 1 8 2 - l l 9 l . Sr,rent, P. 1975. The illustrated encyclopedia of the butterfly world. Asterid 5: Dilleniaceae. Vitidaceae. Salamander Books. London. Ranunculids: Berberidaceae, Fumariaceae, Menispermaceire, Pa- Svrr.s.r, J. 1978. Plant chemistry and the evolution of host spec- paveraceae, Ranunculaceae. ifrcity: new evidence frotn Heliconius and Passiflora. Science 2 O I : 1 4 5 - 7 4 7 . Paleoherbs 1: Aristolochiaceae, Piperacetre. Spr:nr-rNc.F , A. H. 1992. Mitochondrial DNA variation and Hal- Monocots: Arecaceae, Bromeliaceae, Commelinaceae, Cyperaceae, dane's ruie in the Papilio glaucus and P. troilus species groups. Dioscoreaceae, Heliconiaceae, Liiiaceae, Musaceae, Orchidaceae, Heredity 1 l:227 -233 . Poaceae, Smilacaceae, Zingiberaceae. 1993. Mitochondrial DNA phylogeny of the Papilio ma- Laurales: Hernandiaceae, Lauraceae, Monimiaceae. clicon species group (Lepidoptera: Papilionidae). Mern. Ento- nol. Soc. Can. 165:233-242. Magnoliales: Annonaceae, Cannelaceae, Magnoliaceae, Wintera- Sworr-ono, D. L. 1991. PAUP: phylogenetic analysis using par- simony. Vers. 3.1.1. Illinois Natural History Survey, Champaign, Paleoherbs 2: Chloranthaceae, Illiciaceae, Nympheaceae. I L . Wersox, L., eNo M. J. D,qr-r-ewtr z. l99l . The families of flowering Hamamelid 1: Platanaceae, Proteaceae, Sabiaceae, Aizoaceae, plants: descriptions, illustrations, identification, and information Amaranthaceae. Cactaceae? retrieval. Vers. 16. March 1997. URL http://www.keh.ukans.edu/ Caryophyllids : Caryophyllaceae, Chenopodiaceae, Nyctaginaceae, delta. In: The flowering plant gateway, URL hrrp:l/ Olacaceae, Plumbaginaceae, Polygonacetre, Portulaceae, Santala- www. isc.t amu.e du/FLORA/cronang. htm. ceae. Viscaceae/Loranthaceae. WnllEn, S. J., euo D. P. PesrtlEv. 1995. In search of butterfly origins. Mol. Phylogenet. Evol. 4:235-246. Pinaceae. Wnr-r-r:n, S. J., D. P P,qsHI-r,v,A ND J. A. M,cR'ttN. 1996. Reas- (other) conifers: Cupressaceae, Podocarpaceae, Taxaceae, Taxo- sessment of butterfly family relationships using independent diaceae. genes and rnorphology. Ann. Entomol. Soc. Am. 89:184-192. Cycads: Cycadaceae, Zamtaceae. Zrscpnl, A. R., lNo M. R. BEIrpNerurl. 1993. Plant chemistry, insect adaptations to plant chemistry, and host plant utilization AppnNorx 2 patterns. Ecology 74 :47 -54. Description of the butterfly phylogeny used in this study. Clades parantheses, refer to taxon # in Appendix Corresponding Editor: C. Mitter are identified by numbers 3 The phylogeny can be obtained from the authors in electronic form upon request. (((440,(438,43e)),(44r.(442,414,3()()s),,((3( ,(2,4)))1,8( (,1(e ),(20. Appr,Notx 1 (2t ,(22,((s3, 36),((2(62,3 ,(2s,24))),((32,(33,31()()3,(1 2,3 70,) (,2 8, List of important host families fbr butterflies included in the clades 2 e ) ) ) ) ) ) ) ) ) ) , ( 1 7 , ( ( 6 , ( 7 , ( ( ( e , 8 ) , ( 1 1 , 1 0 ) ) , ( ( 1 2 , 1 3 ) , ( 1 6 , 1 4 , 1 5 ) ) ) ) ) , ( ( ( 5 0 , recognized by Chase et al. (1993). Question marks denote families (.(4e,46),(478) ,) ),Q e,((4s.44),(40,(43,(4t,42)()()6)0)(),3, 73, 8)), that were not included in the analysis by Chase et al., and for which (5e(,( s6(,5 8,s7)),((s2I ),5,( 55,(s4,s3)))))))))))1),(((6e 0,Ie ), (e3, positions have been inferred lrom the classification of Cronquist e2),94,95,(96,97)1,e080,,(9 9)),((868,8 7,8 e),(80,2(1,' s7, (84,8s)), ( 1 e 8 1 ) . 82,8 3 ,( (8 1 ,7 1 ,7 O )(,1 8 ,73 ,'79 ,'7'76,7,7 4 )),( 6 7,6 9,6 8 ) ,6 s,6 6 ,( 63,6 4), 6 2 ) ) ) )(,(1 0 91, 0 2(,1 0 11, 0 3 )(,1 0 81, 0 71, 0 61, 0 51, 0 4 ) )(,( (l 8 0 ,( ( 18 1 , (A): Lin- Rosid l: Celastraceae, Erythroxylaceae, Euphorbiaceae, 18 2),((Q2r,222,223),1(29 ,220,225,224,226|) 3),,2(2 29,(227,228)), aceae, Malphigiaceae, Ochnaceae, Passifloraceae,S alicaceae?,V io- 202,(1( 95 ,1 9 6 ),( 19 8 ,1 9 7)),1 8 3,(244,247,r2,233 O,232,2334,2,233s , laceae; (B): Cannabidaceae, Fabaceae, Krameriaceae, Moraceae, 236 ,23'72,3 82, 392, 40,24r, 242.2432,4 6,2 45) ,( (2I 6,2 I s) ,2 | 8,2| 4, Polygalaceae, Rhamnaceae, Rosaceae, Ulmaceae, Urticaceae, Zy- 2r1) , (212,( 209,2 rO),2 l 1) , (206,2 05,( 207,2 08)(, 204,2 03))(, 201 , gophyllaceae; (C): Begoniaceae, Betulaceae, Casuarinaceae, Cu- ( 1 e 9 , 2 0 0 )()1,9 0(,1 9 1 , ( 1 8 e(,1 8 6 , 1 8 51, 8 71, 8 8 ) )1, 9 3 , (D) 1 9 2 ) ) , 1 8 4 , curbitaceae, Fagaceae,J uglandaceae,M yricaceae; : Oxalidaceae 1 6 , 1l s ) , ( 11 2 , (1 4 r, 1 3 ) ) ) , ( ( 1 5 8 , ( 1 s 6l,s145,13 5, 51, s 7 ) ) , (E): 1 9 4 ) ) ) , ( ( ( 1 (Oxalis): Combretaceae, Melastomaceae, Myrtaceae, Punica- (1 76 , 1 70 , 1 72 , 16 3 , 1 6 41,75 ,1 6 1 . 137, r 11 ,1 7 4 ,1 6 8 ,1 6 5 ,1 6 9 ,1 6 6 ,1 2 0 , ceae. 162.167,r2r6,(01,1 781,7 7) )),(I 59,(I 28,1 2 7)),(1(2 9,I 31 ,I 30),1( 32, Rosid 2: Aceraceae, Anacardiaceae, Bataceae, Bornbacaceae,B ras- 13 3 ,l 3 5 , 13 4 ) , ( (1(3 6 , 13 8 , 1 3 73, 19 , 14 0 ) , 114) , ( (1( 4 2 , 1 4 3I) 5, (l , 1 5 2 , sicaceae,B urseraceae, Capparaceae,C aricaceae, Geraniaceae,H ip- l 50)) ,( 14 4,( (1 4 9,1 48)(,t 47,1 ,16I ,4 5) ))))),( (t zs,( | 24,( r22,r 23))), pocastanaceae,M alvaceae, Oxalidaceae (H y-pserocharis),R eseda- 12 6,( 1 1 8 ,( l l 7,l l e))))I, 7 e),I I 1,1 I 0),( 437(,( 4 364, 3s(, ((433,434), ceae, Rutaceae, Sapindaceae, Simaroubaceae, Sterculiaceae, Tili- (43r,432)()(, 426,421()4, 28(,4 304, 2e))))),42(s(,( 4I 8,4 I 9),( 41 s , aceae, Tropaoleaceae. (4r 7,4 1 6 )) ),( 420(,4 22,4r2, 423,424))()(3, 9 3 ,( 3 e 9 ,3 e s ,3 9 4 ,39 6 ,39 7, 398)),(832 ,(38 3,38 4,38 s)),(3I9,3 90,3e2),8(73, 38 6),38 8,38 9),(I4 4 , Rosid 3: Crassulaceae, Grossulariuceae (Ribes . . .), Hamamelida- ( 404,4 03)(,4 0r , 4O24, 00)4, 05,4 10 ,4 09,( 408,4 O14, 06),( 41 | , 4 12), ceae, Saxilrag rceae ( Sari fra ga). 4t3),( ( ( 267Q, 66,259,260.264,263.262t,,22665 ')(,( 257,2s6,2s5), 2s 8 ,2 5 4 ,253 ,25 2 ,2sr ,2s O ),248()2 , 49(,( 268(,( 2692, 7O )(,( 2'7t, (272 , Asterid 1: Acanthaceae, Apocynaceae, Asclepiadaceae, Bignoni- (Aucuba'), 273)),((247,( 21s ,2'67) ),(27.'(72 78 ,279 ))))))(,2 80(,2 8I ,( (282,(284, aceae, Boraginaeae, Convolvulaceae, Cornaceae Gen- (2e2(,( 28s(,2 86, (288(, 28e(,2 90,29I (337, Lamiaceae, Logania- 283)), 287)), ))))))))))), tianaceae, Gesneriaceae, Hydrophyllaceae, (33 (35 (350,349),34(83)4 , 0,(344,343'),(34r,342)),339,345 Plantaginaceae'J,R ubiaceae, Scrophulariaceae, So- 8, l . ,346, ceae, Oleaceae, 347)(, 381(, 3803, 79,3 78))(,( (3583, s9),( 357(, 3523, 5s,3 s6,3 s3, lanaceae. Verbenaceae. 3s 4)) ,3 6 4(, (3 6 03, 6 1 ) ,( 36 2,3 63)) ) ,3 69(,3 6 8 ,3 6 7,366,63s ),(321, 37O , Asterid 2: Apiaceae, Aquifoliaceae, Araliaceae, Asteraceae, Cam- 311 ,3'67, 37s ,374 ,373 ,31t ))),(( ( ( 295 ,( 296 ,298 )),249, 297), 2 933, 00, panulaceae, Caprifoliaceae, Cornaceae (Corokia), Cornaceae (Gri- 2 9 93, 0 1 , 3 0 2( 3, 0 3(,3 0 4 , 3 0 s )()()3,I 6 , 3I 3 ,3 I 7 , 31 s ,3 r 4 , 3 1 2 , ( 3 0 9 , se linio), Cornaceae (.He I w i ng ia ), Dipsacaceae, Menyanthaceae, Pit- 308)3, 073, 06(,3 1 1 ,3 1 0 )),( ((3233,2 2,32r3. 20)3, 27(,3 283, 26,32s, tosDoraceae.V alerianaceae. 324),l39 ),31 8,(332,13, 33 33,3353,43, 3-r 0,3-r6 ,329))))))))))); BUTTERFLIES AND PLANTS 499

o\

O

E

-c 6

€

-

a

a)

t . , - o x .!-. € r t O O - . 1 - , t r € r - C a - a l - j 6 € r € l C - d 6 f n € r 4 l C - . r - J r @ r - 5 O - N - + h € r ! 6 O - d - i r @ r € 5 O 6 3 a 3 c - - c - o o : : : - : = = = = - - - * : 1 1 : : !

d * q j a s : i l

-: : a - i-t

'' 'c

, o

? ;=: ---= 3 = 1P- E 1 . - - a E 3 ? s a E i t ? - : ^ . - : 7 a . ' . ja = =72 ' =r i = = - = \ ! ; : i . - a i i r - : 4 _ U = i G t i i = : 1=;t + : i i , Z i 4= - 7 i : a : : i r i r i Z O . : i'ti / r-: i7=u; ; l ; z+ Z Ei li 1= ;= ;i i t Ei ; z2 ?i i =-it Eiz i 27 i 12 i 3 j ; f ( ; i * : - : J = - ; ; j > ; d E : i 5 : : . . . . , , , 2 2 . ; : ' ' , . . , . . :

; = = o d F L ; : . : - c F U A :'

Q g . = a . O e O * ^ ; = X x € ! o ; : s r ; < o c - d -i a n € r € o a - d - + n € r € 6 o - d - t n € r € o c - d - $ n € r r 3 cr !? t t * + i + + i + J s00 N . J A N Z A N D S . N Y L I N

.b

1a ; a - L I ^ L . Y 2 i ^ i - t - a = z E ! - + F i E + L : ; i ; : : = . 1 . - = _ . n , - : . 4 : - - ! E t ; : ! i E = E : =_ a . . - ! = - > , ! 1 2 1 1 i ? 7 - . ; t ' - == e t i a i . 4 1 - t ? = ! - = i : , , ; - , = = - e" =F = ; ; : r - - 8 . - = ^ : ; . : , . E i 2 ; j 2 . ) - t E E A { . a 2 i : a d : t j - i - : i i ; : i ; a e : : : i : r ,

. ! r

E

E :' E r

O

:x z a

. g . _ 3 ' :

= l - ' a o u i l l : ; : c d - : ; a i a n < : J U d

a

a i d

3 E q

- c r - i n G r - 3 C + a l - ! 3 0 - a r - + n € r € t a - . t - , 1 6 € r { 6 c - r l a f n € r - ] - o - - + - i t c - a r - J 6 € r F o - - . t - t 6 J J r n n 6 n € € @ c o @ @ € € o r r r r r r r r r r c o 3 3 5 0 BLITTERFLIES AND PI-ANTS 5 0 1

:. = - - d ! l g e 1 - ! i = = i i ' : z - r = : j = i = = ! : - E = = - . - ; : i . : : , : = - - t - ; ; t . ! ' ij ? e = = -= t ; ; ? = : 4 , d = . = = = - i= Q ; , - - :i z i = 7 = - i : = a = j = ' = v v = 7 - = v = F = ' =rE- c -5a3 ' - s - z L < ., a - E r 1 n n - z. i i : - i : : i : 6 i l 1" Z: =- i- :\ E, ar rr - il

; d . : E : u d !

"o = !:

U L a € r r a - - . r - t n € r - 5 c - c r - a r @ r € o - - . r - - * . r - 4 6 € r { o a - t n € r r O - - . 1 - J ' a € r - t i t * 1 1 1 + + + L . L . L . ' . @ o c @ G c c € € € r r o - z r r a.

=Z E _ . r _ . 1 F : - a t D l a E o - = - E = . 1 - . r . r r i t r r - . r - - - + + , ^ t = = = = , ! = - i = - r ; = , : : , = Z f - ' = = - - Z = - . = = = = . = = = 2 a y F , . ; i E a = = . = ! Z a 1 : i : = - = . ' t : 4 ! - - = L -' 12 1=- Z + Z a = ' ! . e = z l t , Y - ' ! = - . i i i 1 . : , i . z : . = = ! i : , i E - a ? ?++ t ; ; : i i = t = = = - a = a = : a z L : . e f r < 4 < < < < < < < - < - < - < , < 4 , 4<. < 4 4 < < . > E a > n a < : 4 Z Y d - t L : . L > e Y = 4 , : > > d A , -

E ; = E E ? Z - z ' t Z

E

L

- . r - , f n € r - o a - . r - 4 6 € r - 3 C - d - f n € r - O O - d - / 1 6 e r € O C - d - f n C r - € € c c € € c c € r d d d d d . t d d d N d a r d d d 502 N . J A N Z A N D S . N Y L I N

APPL,NDIX4 Descriptions of step matrices used in this paper. Step matrix "a" was used to describe transformation costs be- tween the l5 possible combinations of "Rosid 1B," "other rosids" (all rosids except Rosid lB), "asterids," and "other seed plants." State numbers translate as follows: 0, Rosid lB; I, other rosids; 2, asterids; 3, other seed plants; 4, Rosid 1B * other rosids; 5, Rosid 1B * asterids; 6, Rosid lB * other seed plants; 7, other rosids f asterids; 8, other rosids * other seed plants; 9, asterids + other seed plants; A, Rosid 1B f other rosids f asterids; B, Rosid 1B + other rosids * other seed plants; C, Rosid 18 + asterids * other seed plants; D, other rosids i asterids + other seed plants; E, all groups. Step matrix "b" was used to describe transfbrmation costs be- tween the seven possible combinations of the plant groups "rosids," "asterids," and "other seed plants." The same step matrix was also used for the seven possible combinations of the growth forms "herbs," "vines," and "trees and shrubs." State numbers translate as follows: 0, rosids (herbs); I, asterids (vines);2, other seed plants (trees and shrubs); 3, rosids * asterids (herbs f vines); 4, rosids * other seed plants (herbs + trees and shrubs); 5, asterids + other seed plants ( vines * trees and shrubs); 6, all groups. = -

= ; c c - . : I t r 1 = = = = : t o : u From: 0 0 1 1 1 2 2 2 2 2 3 3 3 3 1 0 2 t 1 1 3 3 2 2 ,| 0 2 2 J I 3 2 2 3 2 0 2 '| 3 1 3 2 A 2

2 2 0 3 3 I 3 2 0 :: 5 2 1 1 3 3 0 2 2 J J '| 6 1 3 0 2 2 4 2 3 1 :E T U r € o - - d - a 6 o r - 3 a - . r - i r , € r € t a - d - 7 2 I 3 3 I 0 4 2 3 + = + 1 1 3 ! t t ? ? ! ! T i i T T ? ? + T T + + + + 'I 8 2 J 1 r; 4 0 2 2 3 1 /. 9 3 1 0 2 3 3 1 c 'I A ,l Z A 3 1 0 3 3 1 r B J 2 2 I I 3 3 r 3 0 'I J I 3 2 0

o 3 2 3 I I J 2 2 0

E J 2 3 2 0

T o : 0 1 2 3 4 5 6

From: 0 0 I 1 3 1 1 0 2 1 1 0 I 1 'I 1 1 0 1 1 1 : 1 : - ! - . 2 : t E 5 J 1 1 0 1 - i : . + ; + 2 i = j i + E - . r : ; i 6 2 2 I I 1 0

a t

2

E

c - - . r - a r € r € l a - . 1 - i r € r € o - - d - J ' . € - o - * ------+ t f - t i + t f J a t t a + + +