Amer. J. Bot. 73(3): 387-397. 1986.

CLADISTIC TESTS OF HYPOTHESES CONCERNING EVOLUTION OF XEROPHYTES AND MESOPHYTES WITHIN SUBG. PHYTARRHIZA ()l

AMY JEAN GILMARTIN AND GREGORY K. BROWN2 OwnbeyHerbarium and Departmentof , WashingtonState University, Pullman, Washington99164, and 2Departmentof Botany, University of Wyoming, Laramie,Wyoming 82071

ABSTRACT TillandsiaL. Subg.Phytarrhiza (Visiani) Baker (Bromeliaceae) is a distinctivegroup of about 35 epiphytic . These exhibit a range of habits from xeric to mesic. The evolutionary relationshipsof the contrastingadaptations need to be establishedhere as well as in the subfamily as a whole. Relations between the subgenus and other tillandsioids are problematicaland phylogeneticreconstruction of its member-specieswould be facilitated by identification of Phytarrhiza'srelative (sistertaxon) sharingthe same most recent common ancestorwith Phy- tarrhiza.This paperexamines the two most likely sistertaxa, Subg.Pseudo-Catopsis Baker and Subg.Diaphoranthema (Beer) Baker. Diaphoranthema is rejectedas sister taxon. The accepted evolutionarytree, rooted by Pseudo-Catopsis,indicates that most habitalevolutionary changes in Phytarrhizahave been between mesic and semi-mesic forms and from mesic to xeric forms. Methods developed for testing specificevolutionary hypotheses are broadlyapplicable.

THERE IS CONSIDERABLEinterest in the direction ofxeric specializations of otherwise mesic taxa; of evolution of species within Phytarrhiza as for example, coreaceous, linear leafed species well as for the entire and subfamily be- of Lycium (Solanaceae). Such xeric adaptations cause of questions raised by Pittendrigh (1948), might seem to be difficult to reverse, i.e., un- Medina (1974), and Benzing and Renfrow likely to evolve to a more mesic state. Stebbins (1971 a, b), and discussed by Benzing, Givnish (1974) pointed out, however, that any belief and Bermudes (1985), and Gilmartin (1983), that xerophytes, in general, are irreversibly regarding relationships between mesic, tank specialized has no foundation in fact. We are forms and more strees-adapted "atmospheric left with both courses being essentially equally bromeliads." Benzing et al. (1985) examined probable. the question: are extreme xeric tillandsioids Geographic areas that might be paricularly derived from mesic forms- Schimper's (1888) prone to harbor evolutionary sequences of a interpretation?; or are mesic, epiphytic til- given lineage in both directions, would be re- landsioids derived from xeric precursors-Pit- gions where pluvial and arid climates have al- tendrigh's (1948) interpretation? Benzing also ternated. During the Pleistocene, pulsating, cli- considered a third interpretation, proposed by matic changes occurred at least in some parts Medina (1974) who examined carbon path- of South America (Haffer, 1969; Sarmiento, ways. Medina proposed that both xeric and 1975; Simpson, 1975; Solbrig, 1976; Gilmar- mesic tillandsioids arose from precursors tin, 1983). adapted to conditions of high light intensity With these ideas in mind, and using phy- and humidity. logenetic reconstruction methods, we exam- There is little support for the often held no- ined the possible direction(s) of evolution tion that xeric adaptations in general, are more among members of a closely related group of frequently acquired from mesic progenitors tillandsioid species. Tillandsia is a large bro- than the reverse. Some reasons for the notion meliad genus of mixed habit with its center of are the many examples in the world's deserts distribution in South America. The study group members are all epiphytic or saxicolous, in either case, nutritionally independent from the ' Received for publication 6 May 1985; revision ac- cepted 4 October 1985. substrate (Pittendrigh, 1948) and include me- Research supported by NSF Grant BSR 84407573 to sic, semi-mesic and xeric species. the authors. The following individuals provided extremely Species of Phytarrhiza with thin, flexible leaf- helpful critical reviews: David Benzing, Phil Cantino, Vicki blades 2.0 cm or more in width are considered Funk, Loren Rieseberg, Karen Simmons, Douglas Soltis, mesic termed and John Utley. Robin Lesher provided technical assis- as (Fig. la). Schimper (1908) tance and illustrations were rendered by Sheila Gil- mesic forms with leaf-rosettes capable of con- martin. taining water as tank epiphytes. Tank bro- 387 388 AMERICAN JOURNAL OF BOTANY [Vol. 73

N .=-"~~~~-~~---., .8

Fi. )Tyicliladsoifescaitb,em-m - . c

C K41W? 811'a

Fig. 1. a) Typical tillandsioid of mesic habit; b) semi-mesic habit; c) xeric habit. meliads retain a reservoirof water trappedby seek the answer in taxonomic revisions, and the leaf-rosetteand usually occupy moist areas by constructing cladograms and examining with mean annual precipitationof 200 to 400 patterns of morphological character-state cm, and in areaswhere no months receive less changes. than about 4%of the annualprecipitation (Gil- Phytarrhizainitially appearedto be mono- martin, 1973, 1983). Semi-mesic species have phyletic,unlike other Tillandsiasubgenera, e.g., a poorlydeveloped tank. Leaf-bladesare thick- Allardtia and Tillandsia, that could not be er and usually less than 2.0 cm wide (Fig. lb). shown to be natural groups (Gardner, 1982). They may flourishwhere annual precipitation Tillandsia Subg. Phytarrhizaspecies all share exceeds about 100 cm. Xeric species (Fig. lc) the uniquely derived petal-character,blades have no tank; leaf-bladesare very narrowand broadand conspicuous(Gilmartin, 1983), and succulent. Xeric tillandsioids can be expected is the only bromeliad group with conspicuous to occur in areas with a dry season and about petal-lamina. The three subgenera Diaphor- 30 to 100 cm annual precipitation (Rumley, anthema, Pseudo-Catopsis, and Phytarrhiza 1965; Gilmartin, 1972, 1973, 1983). all have very short styles that are includedwith We ask the question:within this group,which the stamens within the corolla. Togetherthese direction(s)of evolutionary change, if any, oc- three subgeneraconstitute a putatively mono- curred most frequently:xerism toward mesic phyletic group. adaptations, or mesic forms toward xero- The relationshipof Phytarrhizato other til- phytes? In this first of a series of papers we landsioids has been a difficult question given March, 1986] GILMARTIN AND BROWN-TILLANDSIA. XEROPHYTES AND MESOPHYTES 389 previously available data and traditional an- forming the groups: leaf-blade width and de- alytical tools. This has not impeded specula- gree of succulence, and presence and devel- tion. The tools of modem phylogeneticrecon- opment of a tank. This was by design to avoid struction are brought to bear here on the circularity,so that resultingcladograms would question of relations of Phytarrhiza to other be based on other data than the very characters tillandsioids with focus on two of the most for which we hoped to discern polarity, i.e., likelycontenders for sistertaxon status.A sister direction of evolution. (The groups did not taxon is that evolutionaryunit sharingwith the changewhen the analyses included these char- groupunder study the same most recent com- acters.) mon ancestor. Its identification is one of the most effectivemeans to determinedirection of MATERIALSAND METHODS-Datawere as- evolution of characters(Hennig, 1966; Wiley, sembled from the Monograph by Smith and 1981), i.e., to establish which states are an- Downs (1977), and methods of phylogenetic central, which are derived. reconstructionwere applied under the princi- Strongtendencies toward parallelevolution ple of maximum parsimony (Wiley, 1981). are evident within every major group of or- Maximum parsimony has the goal of recon- ganisms. Among traits showing parallel evo- structing phylogeny with the fewest possible lution,those involvingless complexityare more character-statechanges for each character.The likely to exhibit reversals,i.e., change in more directions of change (character-statepolari- thana singledirection, than arethose involving ties), wereestablished using the two most likely greatercomplexity (Futuyma, 1979) and while alternative sister taxa, Subg. Pseudo-Catopsis both parallelsand reversals may occur, often and Subg.Diaphoranthema. This approachhas we expect more of the formerand fewer of the been described as the outgroup substitution latter. Stebbins' (1974) concepts of paths of method by Donoghue and Cantino (1984). least resistance may help to explain why par- These authors suggested using multiple out- allel evolutionary change occurs as frequently groups alone and in combination to identify a as it does among related taxa. consensustree. Our objective is to identify the Parallelevolution is very common in flow- most likely hypothetical sister taxon, because ering (Cronquist, 1968; Funk, 1981), our central concern is the direction(s) of evo- and bromeliads are no exception (Benzing, lution relative to xerophytesand mesophytes. 1980; Gardner, 1982). For example phyloge- The study group is analyzed with the two al- netically isolated bromeliad taxa have several ternative sister taxa that prior research sug- times independentlyacquired scales on gested. the ventral petal surface (Smith and Downs, Deployment of Diaphoranthemaas the sis- 1974, 1977, 1979). Parallel charactersshould ter taxon to Phytarrhizais implicit in Smith's not be removed however, before constructing remarks (1934) regardingthe apparent close- treesbecause they may be useful at branchtips ness of these two taxa. Characterslinking Phy- (Funk, 1981). In this workwe retaincharacters tarrhizaand Diaphoranthemainclude exserted regardlessof subsequent evidence of parallel petal blades, presence of floral tube, anthers changes. By parallel character-statechanges, equalling or exceeding stigma with both sets we mean recurrentchanges of the same char- of organs included within the corolla (Smith acter in the same direction whether it is a re- and Downs, 1977). However, it is not clear currentgain or loss. whetherDiaphoranthema was thought to be a Certainly,reversals occur frequently.Own- closely related taxon at the same level as Phy- bey and Aase (1955) pointed out followingtheir tarrhiza,i.e., what would be termeda sistertax- researchwith Allium, that some reversalsoccur on or thought to have evolved from one or especially readily. Loss of a structureis par- several phytarrhizanspecies, i.e., represent a ticularlylikely to recurwhen the original evo- part of the study group or ingroup in current lution of the trait involved several genes, and terms. The latterwould mean that Phytarrhiza mutation of any one of these might prevent is not monophyletic,but paraphyletic,and this expression of the trait (Ownbey and Aase, question is examined. We are often forced to 1955). Relative to the outgroup, these are re- deal with paraphyletic groups during early curringforward changes or recurringreversals. stages of analyses-and analyses may reveal However,within a closelyrelated group of taxa, study groups to be paraphyletic(Platnick and such recurringlosses would constituteparallels Funk, 1983). within the group under study. Priorresearch on the subgeneraof Tillandsia Three of the most conspicuous, morpholog- and of Vriesea (Gilmartin, 1983) implicated ical determinates of mesic versus xeric habit Subg. Pseudo-Catopsisas an alternative sister (Fig. 1) were not included in the data when taxon to Phytarrhiza. Characterslinking all 390 AMERICAN JOURNAL OF BOTANY [Vol. 73

three groups (Diaphoranthema, Pseudo-Ca- bracts, elliptic and ecarinate, stems long and topsis,and Phytarrhiza)the short style includ- conspicuous, and inflorescence compound. ed stigmas and stamens. Monographic re- Employingrooting by P-C, character-states1A search,previous phenetic analysis and pattern and 3B would representderived states;relative cladistics,all pointed to one ofthese two groups to D, character-state7A has the derived state as the possible sister taxon to the study group, while for both putative sister taxa, 2B is an- Subg. Phytarrhiza. Hypothetical trees which cestral.Table 1 lists all 15 charactersand their use Pseudo-Catopsisas the sister taxon (here- states. Figure 2 shows the character-stateval- after termed P-C trees) were compared with ues of all putative evolutionaryunits including trees based on Diaphoranthemaas the sister the two putative sister taxa to Phytarrhiza taxon (D trees). (P-C, Diaph 2). Every taxon in Pseudo-Catopsis and Dia- Monophylesis of Diaphoranthematogether phoranthemainitially was includedin all anal- with three species of Phytarrhizais supported yses. Subsequently,a single species of P-C, out by the derived character-states, distichous of the toal of about 48 was excluded. Tillandsia leaves and few scape-bracts.The three phytar- pugiformisL. B. Smith, may have stems, a trait rhizan species that share these states are T. that sets it apart from other species of P-C. bandensis Baker, T. mallemontii Glaziou ex Five of 17 species of Diaphoranthema had Mez, and T. crocata (E. Morren) Baker. leaves that were polystichousin contrastto the Our hypotheses concerningwhich direction typicaldistichous habit. These tend to be poly- of change occurs more readily are indicated in ploids (Till, 1984). The putative sister taxon Table 1 by an asterisk adjacent to the state that was used for the D trees representsonly fromwhich a changeis presumedto most easily distichous leaved members of this subgenus. originate. We hypothesize, for example, that Phylogenetic reconstructionwas done with an evolutionary change in characterfive from PAUP version 2.0 (Swofford, 1983) on the A, scape-bractsseveral or many to B, none to WSU Amdahl IV. Data were input in standard three, will occur more frequentlythan change format. Among the program's available op- from no or few scape-bractsto several scape- tions, the followingwere used: 1) "closest"de- bracts.This is not to say that we are presuming termined the sequence in which evolutionary polaritiesfor our studygroup. On the contrary, units would be added to the tree; 2) "global we will test our hypotheses against the objec- branchswapping" identified the shortesttrees; tively generatedtrees and the polaritiesof char- 3) "mulpars" searched for multiple equally acterswill be determinedby the effects of out- parsimonioustrees using global branch swap- groups on the ensuing trees. ping (see above); 4) "apolist" listed the apo- It was only afteralliances were identified and morphies, and 5) "chglist" listed changes in analyses were run that each alliance was clas- each characterand the branch(es)where these sified as mesic or xeric. As it turned out, using appeared.Trees were rooted in two ways, the leaf-blade width as the indicator, each group Lundberg(1972) method with "outstates,"i.e., was uniform in this regardalthough the char- states of putative outgroupsister taxon estab- acter was not part of the matrix (Table 1), nor lishing direction of changes, and by including did we considerit when drawingup the groups. the outgroupin the analyses. These groups depended solely on the selected charactersin Table 1. RESULTS-Preliminary to generating trees, Charactereight (foliar scales spreading or 13 phytarrhizanspecies-groups (alliances) and adpressed)may have a bearingon mesic versus single species were identified based upon as- xeric habits and habitats. It is well document- sembled data (Appendix). Several groups ap- ed, e.g., that the spreadingfoliar scales of xeric pearedto be monophyletic using either P-C or tillandsioids are ideally suited for water ab- D, or both as the sistergroup. The groupswere sorption (Tomlinson, 1969; Benzing, 1980). selectedprior to the computeranalyses for phy- Foliarscales are present in all tillandsioidsfrom logeneticreconstruction. Phytarrhiza Group 1, xeric sites, but species from mesic habitatsmay (Phyt I) for example, sharesthree uniquely de- also have a dense covering of scales, e.g., T. rived (apomorphic)states relativeto sistertax- tetranthavar. aurantiaca(Griseb.) L. B. Smith on P-C. The character-statedata are: state 5B- (Gilmartin,1972), T. magnusianaWittm., and scape bracts none to three, 6B-leaves disti- T. argentea Griseb. (Gardner, 1982). Dense chous, 12A-leaf-blades terete (Fig. 2). Some foliar scales and narrow,thick leaves of T. bul- putative evolutionaryunits (EUs) are set apart bosa Hooker give this species the appearance by combinations of traits, e.g., T. humilis Pres of a strong-xerophyte;yet, apparentlyit grows together with group VI has the unique set of equallywell in shady, moist habitats (Benzing, character-states:1B, 2B, 3A, 7B, i.e., floral 1980). There is very imperfectconcordance of March, 1986] GILMARTIN AND BROWN-TILLANDSIA XEROPHYTES AND MESOPHYTES 391

P C 1 A B B B A A B B A A A B B A B TABLE 1. Charactersand character-statesused to estab- lish alliances of species in Tillandsia Subg. Phytar- DIAPH1 A B A B B A A A B B A A A ? ? rhizaandforphylogenetic reconstructions. An asterisk (*) indicatesthat that state is hypothesizedto be easier DIAPH2 A B A B B B A A B B A A A A ? to changethan the alternatestate. Theseare not char- acterpolarities. Polarities wereobtained with two al- PHYT2 A B B A A A A B ? B ? E B B B ternativeoutgroups

PHYT1 A B A A B B A A B B A A A B A 1. Floralbract shape: A = ovate to triangular;B = elliptic or lanceolate. DURATII A B A A A A B A A B A B A B A 2. Floral bracts keel: A = keeled; B = ecarinate. 3. Stems:A = stems elongate,well developed;*B = stem PHYT3 B A B A A A A B A B B B B B A compact, apparentlylacking. 4. Petal blades: A = distinct and conspicuous; * B = CACPURP B A B A A A B A B B B B B ? ? indistinct. 5. Scape-bracts: * A = several to many; B = none to PHYT4 A A B A A A A B A B B B B B ? three at the most. 6. Leaf-arrangement: *A = polystichous; B = disti- PHYT5 A A A A A A B B A B B B B B ? chous. 7. Inflorescence: A = simple; * B = compound. PHYT6 B B A A A B A B B A B B BA 8. Leaf-bladeindumentum: A = spreadingscales; B = scales adpressed,minute. * = = AUREA B B B A B B A B B 9. Sepal union: A free; B connate at least in part. A A A A B B 10. Sepal symmetry:A = asymmetric;* B = symmetric. 11. Sepal length:A = mostly not over 12 mm; B = 12 PHYT7 B B B A A A B B A B B B B B ? mm or longer. 12. Leaf-blades:A = terete;B = ligulateto triangular,not PHYT8 B B B A A A B A A B A B A B ? terete. 13. Presence of roots at maturity: A = present; * B = HUMULIS B B A A A A B A A B B B B A ? absent. 14. Sepal indumentum:A = lepidote for the most part; CAERULEA B B A A A A A A B B A B B B A B = glabrousor subglabrous. A = = Fig. 2. Data matrix (charactersare columns, rows are 15. Petal color: blue or violet; B white and/or putative evolutionaryunits) for 13 phytarrhizangroups yellow. and putative sister taxa. A question mark designatesan unscorablecharacter. Trees using Diaphoranthemaas the outgroupused Diaph 2. acter-statechanges, only the directionof change. Treesusing the 2 sistertaxa were equal in length, trichome form-density with xeric or mesic 24 character-statechanges. Therefore, ways nature. were sought to evaluate trees of equal length. Habit types, mesic (m), semi-mesic (s), and A new method was employed to gauge trees xeric (x) could be determinedfor the terminal by their stability, i.e., number of equal length- groups on trees, but at internal nodes more ened trees and by resolution of evolutionary than one possibilityexists. Possible habits have units.A set of fewer,equally parsimonious trees been entered at the nodes in Fig. 3 and 4. that are more fully resolved is preferredbe- In Fig. 3a, the inferredancestor for the entire causethey aremore preciseand more falsifiable group could only be mesic (m) or semi-mesic (i.e., testable with additional data) than are a (s) as indicated.Change from xeric (x) to mesic larger number of equally short trees that are could have occurred once on the P-C tree at less resolved. This is an extension of the par- the node subtendingunit VII. The alternative, simony principle (Occam's Razor) which cla- no changefrom x, is also possible at this node. distics applies usually to the number of steps In addition, several possible changes between of character-statechanges in trees. mesic and semi-mesic are evident on the four Total parallelism, i.e., convergence and possible P-C trees. On the seven D trees (Fig. reversals, homoplasy, is indicated by the CI, 4a-g), the inferredancestor is xeric.At the node consistency index, (number of parallels and subtending T. humilis there is change from x reversalsdivided by numberof character-state to m. Two possible changes from an internal changes), and PAUP treats equally parallels node with m to s at a terminalgroup are evident and reversals. For instance, a tree produced on the D trees. with 10 -characters and 20 character-state Phylogenetic reconstructions with PAUP changeshas a CI of 0.50. Some changes must produce undirectedtrees, that is, positions of be parallels or reversals in this case, but the the root do not alter the total number of char- relative contributionsto homoplasy of paral- 392 AMERICAN JOURNAL OF BOTANY [Vol. 73

f (x)Y (x)

(m,x) (m,s) (in,x (m,x) VII(m) VII(m) 1s II(s) Ii(s) Vll(m)

(MS' IIIV(m) (s ) (s) (S) lIi(s)* c V(m) IV(s)

b) c) d)

< _- I** I(x) ,/~~**V(S

i(S) / 13LAL ~~~~~~~~~Daphe(x;)

/ H Vlll(x) -~~~~~~~~(x

a) Fig. 3. Phylogeneticreconstructions for 11 alliancesof TillandsiaSubg. Phytarrhiza, plus Diaphoranthemarooted with Pseudo-Catopsis(P-C trees). The numberof character-statechanges is proportionalto branchlengths (0 to 5): a) entiretree with consensuscircumscribed by a brokenline; b) throughd) representall of the variants.Synapomorphies are indicatedin a). Unresolved nodes are indicatedby two dots. lels and reversalsare not indicated by the CI. peared in all four trees rooted with P-C out- Trees maximizingparsimony (24 character- states or rooted by P-C when it was included state changes)had the same consistency index in the study group. of 0.58. Figures 3a and 4a are the two most There were seven trees that were rooted by frequent,shortest trees for 11 evolutionaryunits Diaphoranthema or by its outstates. This is of Subg.Phytarrhiza. These are rootedin Fig. 3 three more than the set of P-C trees. Their with outstates of Pseudo-Catopsis (P-C) and consensus incorporatessix phytarrhizanunits, in Fig. 4 with Diaphoranthema(D). The bro- I, VI, VIII, T. humilis, T. aurea and T. caerulea ken line in each case indicates the concensus, and Diaphoranthema(broken line Fig. 4). The i.e., all EUs within the broken line uniformly same configurationsoccurred when P-C and D appeared exactly as shown in every equally were used together as an outgroup, though short tree. The three classes of habits are in- such rooting could not produce a tree because dicated: m (mesic), s (semi-mesic), x (xeric). the ingroup was not monophyletic when both All variants of the two basic configurations of these were used as the outgroup. areshown in Fig. 3 (a-d), and Fig. 4 (a-g). These Equallyparsimonious trees (phylogenetic in- variantsinvolve the branchsupporting groups ferences)may differ in at least four ways. P-C II, III, IV, V and VII in Fig. 4 and only II, III, and D treesdiffered in: 1)total numberof trees, IV and V in Fig. 3. The concensus tree for the 2) number of parallels and total reversals, 3) cladogramsgenerated with P-C as the outgroup the number of resolved groups,and 4) the par- (Fig. 3) includes seven phytarrhizanEUs: I, ticularcharacters that underwentchanges (Ta- VI, VII, VIII, T. humilis, T. aurea and T. ble 2). In fegardto 1), there were four possible caerulea,plus Diaphoranthema. The exactsa 'me P-C trees and seven possible D-trees. Regard- configurationof these evolutionary u-nitsap- ing 2), each of threepossible types of character- March, 1986] GILMARTIN AND BROWVN-TILLANDSIA XEROPHYTES AND MESOPHYTES 393

11 e 11I(s) (I(s) (s) 1il(s) (M,S) V(s) (i) IV(s)ms) V (m) V(m) (s) VII(m) - (i) VII 11I(s) b) c)

(m,s) 1 s IVs) \s (ins) 1VI(i) (i) Ill(s)

(m)I_,Vl@(m) l / @ @ I l (s)

e)lm)1sm Il(s) _s) 1 (s) V( () VI_(m)- (in,s) -_(*VII(m)

s) ,Diaph (x) L_3(s) AIll(s) U A g) (M's VI~~~~ (in,s) l(M') ) (in,)Vm d) e)~~- ils / MSms)Vin*. ~~~~~~~~aurea (x) ,- \ l . . ~~~~~caerulea(x) ,

IssV(s)

Fig. 4. Phylogenetic reconstructions for 11 alliances of Tillandsia Subg. Phytarrhizarooted with Subg. Diaphor- anthema (D trees). Number of character-state changes is indicated by branch lengths (0 to 5). Synapomorphies are indicated in a). state changes occurred:a) a single change on resolved in the consensus trees for both sets of the tree, without any parallelisms,b) parallel- trees (Fig. 3, 4). ism, i.e., two or more changes in the same Regardingour hypothesesabout frequencies characterin the same direction,c) reversal,i.e., of character-statechanges, three of the changes one or more changes in one direction AND in character3 (stem elongation)on the P-C tree one or more changesin the opposite direction. and two on the D tree supportedour hypothesis Table 2a identifies the number of character- about frequencyof character-statechange (Ta- state changes for the two principle D and P-C ble 1, 2b). The D tree had a single change in trees. Total numbers of each type (single, par- the opposite direction as well. Character4 on allel and reversal) were 13, 7, 4 on the P-C the D tree agreedwith our hypothesis, but dis- tree, and 14, 4, 6 on the D tree. There were agreed on the P-C tree. Our hypothesis about two more reversals and three more parallels character7, inflorescencesimple or compound on the D than the P-C trees. The ratio of par- was not supported by either tree. Characters allels to reversals on the P-C tree was 7:4 = 5, 6, 9, and 13 on the P-C tree consistently 1.75;on the D treeit was 4:6 = 0.66. Regarding showed change in concordance with our hy- 3), the number of resolved groups of Phytar- potheses. These hypotheses must be rejected rhiza on D trees, 5-7 groups are supportedby by the changes on the D tree. one or more uniquely derived character-states (Fig. 4). On P-C trees, 6-7 phytarrhizanunit groups are supported (Fig. 3). Conversely, 4- DISCUSSION-Twosets oftrees of equal length 6 and 4-5 groups on D and P-C trees respec- (24 character-state changes, consistency index tively, were unsupported.Three nodes are un- 0.58) were generated. One set polarized char- 394 AMERICAN JOURNAL OF BOTANY [Vol. 73

TABLE 2a. Numbersof character-statechanges of actual TABLE 2b. Numbers of character-statechanges on the treesand the hypotheticaloptimal tree.All trees have principaltwo trees, the D-tree rootedby outstates of 24 character-statechanges. Optimal or besttrees have T. Diaphoranthema and P-C tree rootedby outstates the hypotheticallybest allocationto the tree assuming of T. Pseudo-Catopsis. S = single change, P = one that the most likelycharacter-state changes are single parallelchange (i.e., in additionto the single change), changesand parallels and reversalsare progressively R = reversal less likely. Differencesin numberof changes between the best tree and the D (Diaphoranthema)and P-C D tree P-C tree (Pseudo-Catopsis)trees are in the last column Character From To S P R From To S P R

Hypo- 1 A B A B theti- Differ- Differ A B A B cally ences ences best tree between between B A 11 1 B A 1 I 1 having P-C D and 2 B A 1 B A 1 changes P-C and best D best 3 B A B A No. single changes 14 13 1 14 0 B A B A No. parallels 10 7 3 4 6 A B 11 1 B A 1 2 0 No. reversals 0 4 4 6 6 4 B A 1 A B 1 Total 8 12 5 B A 1 A B 1 6 B A 1 A B 1 7 A B A B A B B A acterswith Pseudo-Catopsis(P-C) and the other B A 1 2 1 B A 2 2 8 A B 1 B A 1 set used Diaphoranthema(D). These two sets 9 B A 1 A B 1 were evaluated by relative stabilities and dif- 10 (invariant within the study group) ferencesin resolution,i.e., the numberof zero- 11 A B 1 B A 1 lengthedand/or trichotomous branches. In this 12 A B 1 B A 1 way Diaphoranthema (D) was rejected and 13 A B B A (P-C) was acceptedas the sis- B A 1 1 B A 11 0 Pseudo-Catopsis 14 A B B A ter groupto Phytarrhiza.The P-C treewas then B A 1 1 B A 11 0 used to test hypothesized directions of evo- 15 A B A B lution of particularcharacters, and to test our B A 1 0 1 B A 1 0 1 hypothesizedgreater frequency of parallelsthan Totals 14 4 6 13 7 4 reversals. Stability-These criteriaare used to evaluate the phylogenetic reconstructions. Coombs, polaritymay producebetter data than another Donoghue,and McKinley(198 1) suggestedthat polarity even though the character-statesre- only very good data (meaning accurate as to main the same. polaritiesand with little homoplasy) are likely Of the two sets of trees generated with the to yield highly stable or consistent cladograms. two alternative polarities, trees using P-C to A principalreason for stablecladograms is good polarize characterswere more fully resolved. data. When data are inaccurateas to polarities There were four P-C trees but seven D-trees. or there is little support at nodes or there is More groups on the P-C trees were resolved much homoplasy, cladogramstend to be un- than the D trees. The fewer trees and slightly stable, meaningthey changegreatly with small improved resolution of the P-C trees implies changes in the data (Coombs et al., 1981). It greater simplicity and testability of these re- follows that sets of equally parsimonioustrees constructions. Furthermore,results suggest a may be evaluatedby differencesin relative sta- need to examine I, IV, VIII, and T. caerulea, bility (consistency).Those with more stability that were unresolved in nearly every tree re- must be preferredas being more likely to be gardlessof which of the two hypothesizedsister based on good data. taxa were used to polarize characters. It re- Of two or more equallyparsimonious clado- mains to be seen if any other characterswould grams, i.e., having the same number of char- support monophylesis of the problematical acter-state changes, we prefer the one(s) ap- groups. parentlybased on better data. We use degree On the shortest trees to these eleven alli- of stability,meaning numbers of different,equal ances, out of 24 changes, the D tree included length trees and number of resolved nodes as 6 reversals,among 6 charactersand 4 parallels the pointerto help identify trees that are based among 3 characters(Table 2b). The P-C tree on the best data. The charactersand character- had four reversalsamong three charactersand states are the same but a polarity change is seven parallelsamong five characters.We might consideredto resultin differentdata. Thus, one ask how our numbersof parallelsand reversals March, 1986] GILMARTIN AND BROWN-TILLANDSIA XEROPHYTES AND MESOPHYTES 395

comparewith a hypotheticallybest tree of the (1984) in species of Diaphoranthema,further same length. supporting Diaphoranthema as a monophy- We hypothesizedbefore the analysisthat the letic group. best tree for these taxa is one having the fewest possible parallels,here this would be 10 among Questionabletaxa - The taxa Tillandsiacac- the 24 changes and no reversals (Table 2a) in ticola, T. purpurea, and T. duratii were in- the 14 characters.The differencesbetween P-C cluded initially. Although these three taxa are and D trees and the hypotheticaltree are 8 and considered within Phytarrhiza,they were not 12 changes, respectively. The P-C tree which includedin the finalanalysis. Tillandsiaduratii we have selected on other grounds supports is particularlyproblematical as its inclusion in our hypothesis of minimal reversal. initial analyses resulted in greatly increased The ratio of number of parallelsto reversals variation in topology regardlessof polarities. for four problematicalcharacters (3, 7, 13, 14) Tillandsia duratii is a floater, appearingin a that showed homoplasy and differedin num- numberof differentpositions on various trees. bers of parallelsand reversals,is 6:2 = 3.0 for Such variable placement is consistent with a the P-C trees and 3:4 = 0.75 for the D trees. hybrid origin and perhaps this taxon, a par- Our prior hypothesis about the ratio of par- ticularlylarge, robust member, is a polyploid allels to reversalsis more in concordancewith hybrid. This is being checked currently.It was the P-C tree than the D tree, rejectedon other omitted from these analyses whose principal grounds. goal was to identify the sistertaxon to the study Character4, relative expansion of lamina of group. petal blades, showed a single change on the P- In most of the P-C trees maximizing par- C tree from A to B at the node subtending simony, T. cacticola and T. purpureadirectly Diaphoranthemaand a single change from B subtendedthe consensus branch. Inclusion of to A at the base of the D tree. As a result of this pair awaits more information about T. these analyses,we came to realize that the pet- purpurea'sputative relative, T. straminea HBK. als of Diaphoranthema are much more like Tillandsia straminea initially was placed by those of Phytarrhizathan of Pseudo-Catopsis, Smith in Subg.Allardtia, moved by Gilmartin though not nearly so conspicuous as in Phy- (1972) to Phytarrhiza, and subsequently re- tarrhiza. Petals of Diaphoranthema species duced to synonomy with T. purpureain Phy- typicallybecome somewhat narrowbelow the tarrhizaby Smith and Downs (1977). blade, whereas petals of Pseudo-Catopsisare ligulate without any apparent constriction or The sister taxon-Among traits of the ac- readily identifiable blade. The cladistic anal- cepted sister taxon, Pseudo-Catopsis,are the yses helped to focus our attention on the sim- following: floral bracts ovate to triangularin ilarity in petals of the two subgenera,Phytar- shape, ecarinate;plant with no apparentstem rhiza and Diaphoranthema. at maturity; scape and scape-bractswell de- veloped; leaves polystichous; inflorescence Diaphoranthemaand the studygroup-Trees compound, plants without roots at maturity; rooted by outstates of P-C that did not include petals ligulatewith blades not at all or scarcely Diaphoranthemawere shorter but had more distinct; petal blades inconspicuous and homoplasy than trees that did include Dia- scarcelyor not at all exsertedbeyond the calyx; phoranthema. Homoplasy in characters2, 8, pistil mostly exceeding anthers. 1 1 and 15 occurredwhen Diaphoranthemawas not included but not when Diaphoranthema CONCLUSIONS-Relationshipbetween xero- was included. Thus, althoughthe length of the phytesand mesophytes within Phytarrhiza were trees was 22 when Diaphoranthemawas omit- establishedwhen we were able to reject the D ted (shorter than the 24-step trees including tree hypothesis on the basis of the number of this subgenus)homoplasy was reduced(higher monophyletic groups and greater stability consistencyindex) when Diaphoranthemawas (consistency)of the P-C trees. Based on either included. Overall homoplasy as given by the or both criteria,we must accept the hypothesis consistency index was 0.58 with Diaphoran- that of these two most likely candidates, Subg. thema and 0.54 when Diaphoranthema was Pseudo-Catopsisis the sister taxon, not Dia- omitted. phoranthema.Progenitors are mesic or semi- We conclude as a result of these analyses mesic and derivatives tend to be xeric as shown that Phytarrhizasensu Smith and Downs is on P-C trees. The xeric members of the con- paraphyletic. Diaphoranthema is monophy- census group are derived from a mesic or xeric letic. W. Till (pers. comm.) found consistency ancestor shared with group VII on the accepted of the stigma types of Brown and Gilmartin P-C trees. We conclude that xerophytism with- 396 AMERICAN JOURNAL OF BOTANY [Vol. 73 in Phytarrhizaevolved from mesic evolution- by the P-C tree. The P-C tree did not sustain ary units. our hypothesis that character4 changes more Evolutionarychange between mesic and semi readily from B (indistinct petal blades) to A mesic adaptationsmay have occurredin both (distinct, and conspicuousblades) than the re- directions.But we have no evidence of change verse. We thereforereject this hypothesis and from semi-mesic to xeric adaptations. Thus, acceptan alternativehypothesis. This research in these taxa the semi-mesic adaptation does has suggested to us that character 4 should not appearto be intermediatebetween the ple- consist of threerather than two states, one state siomorphic, mesic and apomorphic, xeric ad- beingligulate without any distinctivepetal limb, aptations. a second state being a recognizablebut small Changesrelative to habit on the rejected D limb, and the third state, very distinct and tree would also lend some credenceto the no- conspicuous petal limbs. Results with char- tion that the mesic habit changed in Phytar- acter 7 (inflorescence compound or simple) rhizamore frequentlytoward xerophytism than showed nearlyequal frequencyin either direc- the reverse.While the inferredancestor for the tion and we reject our hypothesis concerning rejectedD tree is xeric;within the study group, this character. several changesare from mesic to semi-mesic, Establishingcorrect character polarities of and there is but a single possible change from alliances of species, of Subg. Phytarrhiza is xeric to mesic. extending our understandingof evolution of We have been able to reject the hypothesis the mesic and xeric habits in bromeliads, an of Subg. Diaphoranthema as sister taxon to active areaof investigation(e.g., Benzinget al., Tillandsia Subg. Phytarrhiza in spite of its 1985). Once a species is channeledinto one or readilyapparent close relationshipwith the lat- the other of the habit-types,further speciation ter. Indeed, these analyses support the inclu- tends to remain within the confines of this sion of Subg.Diaphoranthema within a mono- channel. This seems to be true, in particular phyleticgroup consisting of alliancesof species for the xeric end of the spectrumwithin Phy- assignedto Phytarrhiza.This seems to call for tarrhiza. However, speciation may involve an eventual changein subgenericcircumscrip- evolutionaryjumps from the mesic evolution- tions. Ultimately,with additionaldata on floral ary channelto less mesic or to xeric modes and architecture,particularly that ofthe gynoecium apparentlythis has occurredat least twice in and androecium, chromosome cytology, and Phytarrhiza, from mesic ancestors of the al- stable carbon isotopes, phylogenieswill be in- liances V and VII. ferred and with some modifications, these The utility of these applications of cladistic groupsmay be recognizedas sections. Research methodologyunder assumptions of maximum to obtain required data is in progress (e.g., parsimony,lies in theirpower to distill patterns Brown and Gilmartin, 1984; Brown et al., of similarities and differencesdown to a few 1984). hypothesized reconstructions,principally two The 11 species groups are working-groups in this research.It enabled rejection of one of and more data is requiredbefore these should these in favor of the other by relative degrees be formalized as Sections. In every P-C tree of resolution and by consistency (the relative (Fig. 3) alliances III and IV share a mesic or numbers of equally parsimious trees). semi-mesic common ancestorwith a mesic al- Phylogeneticanalysis can incorporatea de- liance, either V or VII, and usually the three sirabledegree of objectivity, and can help test alliances, III, IV, and V do so. Uniformly, the hypothesesabout specificcharacters. With the xeric concensus group (I, VI, VII, VIII, T. au- effortsof the early cladists and those who have rea, T. caerulea,T. humilis)has a xeric or mesic been developingthe methods up to the present, most recent common ancestor with the mesic the potentialpower of the method is significant alliance VII. and realizable. Our hypothesis regardingexpected relative frequencies of directions of change for char- LITERATURE CITED acter 3, presenceor absence of stems, was sup- BENZING, D. H. 1980. The biology of the bromeliads. portedby everycladogram including those with Mad River Press, Eureka, CA. the two alternateroots. Changefrom compact ,T. GIVNISH,ANDD. BERMUDES.1985. Absorptive to elongate stems, the direction that we hy- trichomes in Brocchiniareducta (Bromeliaceae) and pothesized, occurredfive out of six times (Ta- their evolutionary and systematic significance. Syst. ble other charac- Bot. 10: 81-91. 2b). Hypotheses concerning , AND A. RENFROW. 197 1a. The significance of ters were not sustained by both sets of photosynthetic efficiency to habitat preference and cladograms. However, hypothesized direc- phylogeny among tillandsioid bromeliads. Bot. Gaz. tions for characters5, 6, 9, 13 were sustained 132: 19-30. March, 1986] GILMARTIN AND BROWN-TILLANDSIA XEROPHYTES AND MESOPHYTES 397

,AND . 1971b. Significance ofthe patterns SMITH, L. B. 1934. Geographicalevidence on the lines of CO2 exchange to the ecology and phylogeny of the of evolution in the Bromeliaceae.Bot. Jahr.66: 446- Tillandsioidae (Bromeliaceae). Bull. Torrey Bot. Club 468. 98: 322-327. ,ANDR. J. DowNs. 1974. FloraNeotropica Mono- BROWN,G. K., AND A. J. GILMARTIN. 1984. Stigma struc- graph.No. 14, part 1. HafnerPress, N.Y. ture and variation in Bromeliaceae-neglected taxo- , AND . 1977. FloraNeotropica Monograph. nomic characters. Brittonia 36: 364-374. No. 14, part 2. HafnerPress, N.Y. , G. S. VARADARAJAN, A. J. GILMARTIN, AND H. ,AND . 1979. FloraNeotropicaMonograph. LUTHER. 1984. Chromosome reports, Bromeliaceae. No. 14, part 3. HafnerPress, N.Y. Taxon,33: 758-759. SOLBRIG, 0. 1976. The origin and floristic affinities of COOMBS, E., M. DONOGHUE, AND R. McGINLEY. 1981. the SouthAmerican temperate desert and semi desert Characters, computers and cladograms: A review of region. In D. Goodall [ed.], Evolution of deset biota. the Berkeley cladistics workshop. Syst. Bot. 6: 359- Univ. Tex. Press, London. 372. STEBBINS,G. L. 1974. Floweringplants, evolution above CRONQUIST, A. 1968. The evolution and classificationof the species level. BelknapPress, Cambridge. flowering plants. Houghton Mifflin Co., N.Y. SWOFFORD,D. 1983. PAUP, Phylogeneticanalysis using DONOGHUE, M., AND P. CANTINO. 1984. The logic and parsimony.III. Nat. Hist. Survey. limitations of the outgroup substitution approach to TILL, VON WALTER. 1984. SippendiffenenzierungInner- cladistic analysis. Syst. Bot. 9: 192-202. halb Tillandsia Subgenus Diaphoranthema. Ph.D. FUNK, V. A. 1981. Special concerns in estimating plant dissertation,Univ. Wien, Austria. phylogenies. In V. A. Funk and D. R. Broods [eds.], TOMLINSON, G. 1969. Bromeliaceae.In C. R. Metcalfe Advances in cladistics, pp. 73-86. N.Y. Bot. Gard., [ed.], Anatomy of the ,Vol. 3. Clar- N.Y. endon Press, Oxford. FUTUYMA, D. 1979. Evolutionary biology. Sinauer, Sun- WILEY, E. 0. 1981. Phylogenetics:The theory and prac- derland, Mass. tice of phylogeneticsystematics. John Wileyand Sons, GARDNER, S. 1982. A systematic study of Tillandsia subg. N.Y. Tillandsia. Ph.D. thesis. Texas A&M Univ., College Station. GILMARTIN, A. J. 1972. The bromeliaceaeof Ecuador. Monogr. Phaner. Vol V. J. Cramer, Lehre, Germany. APPENDIX 225 pp., 18 pls. Groups(alliances) of speciesand one-speciesalliances in 1973. Transandean distributions of Bromeli- Tillandsia Subg. Phytarrhiza, based on states of the fol- aceae in Ecuador. Ecology 54: 1389-1393. lowing 14 characters:floral bract shape, and presence or 1983. Evolution of mesic and xeric habits in absence of a keel; presence or absence of stems; relative Tillandsia and Vriesea (Bromeliaceae). Syst. Bot. 8: size of petal blade;scape-bract number; leaf arrangement; 233-242. inflorescencetype; leaf-blade indumentum;sepal union, HAFFER, J. 1969. Speciation in amazonian forest birds. and length;leaf-blade terete or not; rootspresent or not at Science 165: 131-137. maturity;sepal indumentum,and petal color HENNIG, W. 1966. Phylogenetic systematics. Univ. Ill. Press, Urbana. Group I: T. bandensis Baker, T. mallemontii LUNDBERG, J. G. 1972. Wagner networks and ancestors. Glaziou ex Mez, T. crocata (E. Mor- Syst. Zool. 21: 398-413. ren) Baker. MEDINA, E. 1974. Dark CO2 fixations, habit preference Group II: T. acosta-solisii Gilmartin, T. triglochi- and evolution within the Bromeliaceae. Evolution 28: noides Presl, T. cornuta Mez and So- 677-686. diro, and T. dodsonii L. B. Smith. OWNBEY,M., AND H. AASE. 1955. Cytotaxonomicstudies Group III: T. lindenii Regel, T. cyanea Linden ex in Allium. I. The Allium canadense alliance. Research K. Koch, T. umbellata Andre and T. Studies State College Wash. Monogr. Suppl. 1: 1-106. narthecioidesPresl. PITTENDRIGH, C. S. 1948. The bromeliad-Anopheles- Group IV: T. anceps Loddiges, T. monadelpha (E. malaria complex in Trinidad. I. The bromeliad flora. Morren) Baker, T. pretiosa Mez, and Evolution 2: 58-89. T. scaligera Mez and Sodiro. PLATNICK, N., AND V. A. FUNK. 1983. Summary. In Group V: T. dyeriana Andre, T. hamaleana E. Advances in cladistics. 2: 199-200. Columbia Univ. Morren, T. nubis Gilmartin, T. wag- Press, N.Y. neriana L. B. Smith. RUMLEY, G. 1965. An investigation of the distribution Group VI: T. arhiza Mez, T. reichenbachii Baker, of rainfall with elevation for selected stations in Ec- T. streptocarpaBaker. uador. In J. Griffiths [ed.], Research on tropical rain- Group VII: T. laxissima Mez, T. platyrhachis Mez, fall patterns, pp.24-89. Dept. Army, Project No. IVO- and T. venusta Mez and Werckle. 14501-B53A-09-04. Group VIII: T. paleacea Presl, T. peiranoi Castel- SARMIENTO, G. 1975. The dry plant formationof South lanos, and T. schunkei L. B. Smith. America. J. Biogeogr. 2: 233-251. T. aurea Mez. SCHIMPER, A. F. W. 1888. Die epiphytische vegetation T. caeruleaHBK. Amerikas. Bot. Mitt. Tropen. II. T. humilis Presl. 1908. Plant geography upon a physiological basis T. duratiiVisini. (transl. W. R. Fisher). Clarendon Press, Oxford. T. purpurea Ruiz and Pavon and T. SIMPSON, B. 1975. Pleistocene changes in the flora of the cacticola L. B. Smith. high tropical Andes. Paleobiol. 1: 273-294.